Exploring Musical Spaces

OXFORD STUDIES IN MUSIC THEORY

Series Editor Steven Rings

Studies in Music with Text, David Lewin

Metric Manipulations in Haydn and Mozart: Chamber Music for Strings, 1787–​1791, Danuta Mirka
Songs in Motion: Rhythm and Meter in the German Lied, Yonatan Malin
A Geometry of Music: Harmony and Counterpoint in the Extended Common Practice, Dmitri Tymoczko
In the Process of Becoming: Analytic and Philosophical Perspectives on Form in Early Nineteenth-​Century Music, Janet Schmalfeldt
Tonality and Transformation, Steven Rings
Audacious Euphony: Chromatic Harmony and the Triad’s Second Nature, Richard Cohn
Music as Discourse: Semiotic Adventures in Romantic Music, Kofi Agawu
Beating Time and Measuring Music in the Early Modern Era, Roger Mathew Grant
Mahler’s Symphonic Sonatas, Seth Monahan
Pieces of Tradition: An Analysis of Contemporary Tonal Music, Daniel Harrison
Music at Hand: Instruments, Bodies, and Cognition, Jonathan De Souza
Foundations of Musical Grammar, Lawrence M. Zbikowski
Organized Time: Rhythm, Tonality, and Form, Jason Yust
Flow: The Rhythmic Voice in Rap Music, Mitchell Ohriner
Performing Knowledge: Twentieth-​Century Music in Analysis and Performance, Daphne Leong
Enacting Musical Time: The Bodily Experience of New Music, Mariusz Kozak
Hearing Homophony: Tonal Expectation at the Turn of the Seventeenth Century, Megan Kaes Long
Form as Harmony in Rock Music, Drew Nobile
Desire in Chromatic Harmony: A Psychodynamic Exploration of Fin de Siècle Tonality, Kenneth M. Smith
A Blaze of Light in Every Word: Analyzing the Popular Singing Voice, Victoria Malawey
Sweet Thing: The History and Musical Structure of a Shared American Vernacular Form, Nicholas Stoia
Hypermetric Manipulations in Haydn and Mozart: Chamber Music for Strings, 1787–​1791, Danuta Mirka
Exploring Musical Spaces: A Synthesis of Mathematical Approaches, Julian Hook
 EXPLORING
MUSICAL SPACES
A Synthesis of Mathematical Approaches

JULIAN HOOK

1
 1
Oxford University Press is a department of the University of Oxford. It furthers
the University’s objective of excellence in research, scholarship, and education
by publishing worldwide. Oxford is a registered trade mark of Oxford University
Press in the UK and certain other countries.
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America.
© Oxford University Press 2023
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by license, or under terms agreed with the appropriate reproduction
rights organization. Inquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above.
You must not circulate this work in any other form
and you must impose this same condition on any acquirer.
CIP data is on file at the Library of Congress
ISBN 978–​0–​19–​024601–​3
DOI: 10.1093/​oso/​9780190246013.001.0001
1 3 5 7 9 8 6 4 2
Printed by Sheridan Books, Inc., United States of America
 Contents

Preface ix 2.4 Relations 58

Acknowledgments xvii 2.5 Modular arithmetic 65
2.6 Relationships among modular spaces 69
Notes 73
PART ONE FOUNDATIONS OF
Suggested reading 78
MATHEMATICAL MUSIC THEORY:
SPACES, SETS, GRAPHS, AND
GROUPS 1
3. Graphs 79
3.1 Graphs 79
1. Spaces I: Pitch and Pitch-​Class 3.2 Isomorphism of graphs 87
Spaces 3 3.3 Loops, multiple edges, and infinite
graphs 90
1.1 Pitch spaces 4
3.4 Directed graphs 96
1.2 Pitch-​class spaces 12
3.5 Transformation graphs and networks 106
1.3 Spaces generated by fifths and thirds 18
Notes 114
1.4 Tonnetz spaces 22
Suggested reading 118
Notes 33
Suggested reading 38
4. Spaces II: Chordal, Tonal, and Serial
2. Sets, Functions, and Relations 39
Spaces 119
4.1 Double-​circle spaces and related
2.1 Sets 39
constructions 119
2.2 Ordered sets and multisets 43
4.2 Tonnetz-​related chordal and tonal
2.3 Functions 45
spaces 127
4.3 Generic and diatonic chordal spaces 138 8. Transformations I: Triadic
4.4 Some additional models 143 Transformations 300
4.5 Analytical examples 147
Notes 161 8.1 Uniform triadic transformations 300
Suggested reading 166 8.2 Riemannian UTTs and neo-​Riemannian
analysis 306
5. Groups I: Interval Groups and 8.3 Other topics in triadic transformation
Transformation Groups 167 theory 322
Notes 334
5.1 The interval and transposition groups of Suggested reading 338
pitch space 168
5.2 Definition of a group; additive, modular, 9. Transformations II: Transformation
and multiplicative groups 173 Graphs and Networks; Serial
5.3 Abstract groups; further properties of Transformations 339
groups 178
5.4 Interval groups and interval spaces 183 9.1 Transformation graphs and networks:
5.5 Transformation groups and group basic properties 339
actions 189 9.2 Consistency properties 343
5.6 The relation between intervals and 9.3 Isomorphism and isography 347
transformations 195 9.4 Klumpenhouwer networks 357
Notes 201 9.5 Serial transformations and UTTs 364
Suggested reading 205 9.6 Transformations of pitch classes and order
numbers 371
Notes 380
PART TWO TRANSFORMATION Suggested reading 384
THEORY: INTERVALS AND
TRANSFORMATIONS, INCLUDING
NEO-​R IEMANNIAN THEORY 207 PART THREE GEOMETRIC MUSIC
THEORY: THE OPTIC VOICE-​L EADING
6. Groups II: Permutations, SPACES 385
Isomorphisms, and Other Topics in
Group Theory 209 10. Spaces III: Introduction to
Voice-​Leading Spaces 387
6.1 Permutation groups 209
6.2 Group tables and Cayley diagrams 217 10.1 The hexatonic triad graph as a continuous
6.3 Isomorphism of groups 224 voice-​leading space 388
6.4 Direct-​product groups 229 10.2 A larger space of three-​voice
6.5 Groups, equivalence relations, and chords 394
symmetry 233 10.3 The OPTIC relations 404
6.6 Quotient groups; considerations with non- 10.4 Normal forms in OPTIC spaces 415
commutative groups 240 Notes 421
Notes 247 Suggested reading 424
Suggested reading 251
11. Spaces IV: The Geometry of OPTIC
7. Intervals 252 Spaces 425
7.1 Label functions for interval spaces 252 11.1 Manifolds and orbifolds; one-​voice
7.2 Homomorphisms and isomorphisms of spaces 426
interval spaces 254 11.2 Two-​voice spaces 427
7.3 Direct products of interval spaces 263 11.3 Three-​voice OP-​space 439
7.4 Quotients of interval spaces 268 11.4 Three-​voice T-​, PT-​, PTI-​, OPT-​, and
7.5 Transposition operators and OPTI-​space 447
interval-​preserving mappings 273 11.5 Four-​voice OP-​space 460
7.6 Inversion operators and interval-​reversing 11.6 Four-​voice T-​, OPT-​, and
mappings 286 OPTI-​space 468
Notes 296 Notes 476
Suggested reading 299 Suggested reading 481

vi • C ontents
12. Distances 482 13.2 Diatonic scales in chromatic space 545
13.3 Signature transformations 550
12.1 Interval functions and measures of 13.4 Genus and species 568
distance 483 Notes 577
12.2 Distance functions; real and modular Suggested reading 581
interval spaces as distance spaces 487
12.3 Distance functions defined by graphs or 14. Scales II: Beyond the Diatonic 582
groups 493
12.4 Distance functions on product spaces 500 14.1 Seven-​note scales and spelled
12.5 Distance functions on quotient spaces; heptachords 583
OPTIC spaces as distance spaces 506 14.2 Maximal evenness and the geometry of
Notes 524 scales 597
Suggested reading 528 14.3 Beyond the chromatic: other specific
cardinalities 606
Notes 616
PART FOUR THEORY OF SCALES:
Suggested reading 619
DIATONIC AND BEYOND 529
13. Scales I: Diatonic Spaces 531 Appendix 1 List of Musical Spaces 621
Appendix 2 List of Sets and Groups 627
13.1 Diatonic and generic scales as musical References 631
spaces 531 Index 643

C ontents • vii
 Preface

MUSIC UNFOLDS on a vast playing field with a held biennially. It is not an exaggeration to say
mathematical structure of great complexity and that mathematical music theory has blossomed
beauty. This book is a study of that structure. into a sizable industry.
The price of admission to the field, how-
MATHEMATICS AS A TOOL ever, can seem exorbitant. A solid foundation
FOR THE STUDY OF MUSIC in mathematics, traditionally expected of stu-
dents in physics, engineering, and other dis-
Music theory has in recent decades been ciplines that use mathematics, is rare among
enriched by an infusion of ideas from many students of music. Important work in the field
other disciplines, among which mathematics has is scattered through countless articles and an
earned a particular preeminence. Mathematical armload of books, terminology and notation are
techniques enhance our understanding of not consistent, and much of this work is highly
numerous aspects of music, from scale con- technical, not readily accessible to those with
struction and rhythmic organization to triadic limited mathematical training, no matter their
harmony and atonal pitch structure, some- musical expertise. David Lewin’s Generalized
times building upon established music-​analyti- Musical Intervals and Transformations (Lewin
cal traditions but sometimes providing entirely [1987] 2007), the pathbreaking volume that
new methodologies and paradigms. Today’s singlehandedly established the field of trans-
music theory journals publish many articles formational music theory, has been a definitive
at least partly mathematical in nature; the resource for more than three decades, but it also
Journal of Mathematics and Music began publica- remains a famously difficult read for music theo-
tion in 2007, and international conferences on rists who must contend from its first pages not
“Mathematics and Computation in Music” are only with abstract concepts such as semigroup
congruences and homomorphisms but also challenging the explanatory power of theoretical
with an uncompromisingly formal Definition-​ models, mathematical or otherwise. In musical
Theorem-​Proof-​Corollary-​Remark style charac- analysis, mathematically informed descriptions
teristic of advanced mathematics texts.1 may work in tandem with other approaches,
For these reasons, while most professional sometimes to striking effect, producing insights
music theorists have at least a passing acquain- more revealing than would be possible by either
tance with some of the mathematical work being means alone.
done in their field, expansive swaths of this work Mathematics does not tell physicists how
remain mostly the province of specialists, terra they must do physics, and it does not prescribe
incognita to a large part of the music theory how music theorists should do their work either.
community. There is a long-​standing need for It provides a conceptual framework, a vocabu-
a book capable of serving as an initiation into lary, and a generously stocked toolbox, and while
the subject for practitioners of more traditional there are some constraints on how the tools may
strains of music theory and for a new generation be used, there are few instructions dictating
of students. This book aims to answer that need how they must be used or what one is supposed
by presenting mathematical techniques as fun- to build with them.2 Some uses of mathematics
damental tools that should be available to every may objectively be called incorrect, but the via-
working music theorist. ble paths one may choose to follow are bound-
Mathematical structure in music often less. The decision to apply a mathematical model
operates at what might be called a deep level, entails interpretive choices from the start about
seemingly far removed from the surface of any what to represent mathematically and how to
musical composition—​the level of the abstract represent it, questions that rarely have only a
structure of the spaces in this book’s title. One single good answer. This carte-​blanche quality
could make the argument that mathematical has sometimes led to criticism of mathematical
structure at this level is as fundamental to music approaches as lacking in coherent methodology,
as it is to physics, and—​at the risk of inviting inefficiently requiring that each analysis write
charges of “physics envy”—​I believe that the its own set of rules—​but it pays dividends in cre-
parallels between music and physics are instruc- ative potential, in the varied insights the tech-
tive. It should go without saying that valuable niques can produce, and in a casual disregard
work in music theory can be done without regard for stylistic boundaries and other traditional
to this deep level of structure, but so too can we limitations.
study observable physical phenomena with no
consideration of the atomic or subatomic scale.
In either field, things observed in the wild may
THE SPACE METAPHOR
be more complex, messy, and imperfect than Central to this book is the concept of musi-
the simplified situations most readily described cal spaces. Mathematicians work with spaces
mathematically: real-​ world conditions do not of many kinds, known by names such as vector
involve the perfect vacuums and frictionless spaces or topological spaces. In the context of this
planes encountered in classroom physics exer- book, the term space may be understood loosely
cises, and only occasionally does real music to refer to a collection of musically meaningful
display the patterned sequential behavior and entities in some visual or geometric arrange-
flawless symmetry that mathematical models ment. Musical spaces are to some degree meta-
account for most elegantly. But in either field, an phors—​musical notes, it could be argued, do not
understanding of the principles at work in the really form themselves into lines and planes and
idealized settings can inform our understanding circles—​but a premise of this book is that such
of whatever else we may observe, and the differ- metaphors are useful, and that musical spaces
ences between the real and the ideal are them- lend themselves to general abstract study much
selves illuminating. as mathematical spaces do. We shall not attempt
The music-​ physics analogy is strained, to present a more complete or precise definition
though, insofar as the rules taught in music of the term space in all its generality, but we will
theory classes hardly have the same status as formulate careful definitions of many particular
physical principles: a composer who is bound by spaces, some of which will be revisited frequently
the laws of gravity may choose to violate laws and become unifying threads in our exploration
of voice leading at will. Composerly creativ- of diverse topics. We will also develop a more or
ity demands analytical creativity, sometimes less systematic family of names and labels for

x • P reface
the most important spaces and the elements space is not an analysis as music theorists under-
within them, facilitating their study as objects of stand the term, but it can serve as a background
interest in their own right. or pre-​background for analysis. A spatial analy-
The idea of musical spaces is of course not orig- sis may entail tracing the way a musical example
inal to this book,3 and many readers will already moves about within a space.4 Such an analysis is
be acquainted with at least a few of the spaces inevitably reductive, as any given space is only a
we will study, such as pitch space and pitch-​class simplified projection of a complex musical real-
space. Even these conceptually simple examples ity;5 it may nevertheless be clarifying, in the
take a remarkable variety of forms. A pitch space same way that a flat map of airline routes can
may be discrete (think of the pitches playable be informative even though planes fly in three
on a piano) or continuous (pitches playable on dimensions. Spaces can serve not only to show
a trombone); it may be chromatic or diatonic or how music moves but also to lay out possibilities
perhaps something else. Pitch space is theoreti- for how it could move, allowing us to view musi-
cally infinite, while the twelve-​note pitch-​class cal events in relation to other things that might
space encountered by students of post-​tonal the- have happened instead; in this sense a space may
ory is finite and modular. Some spaces comprise be seen as a compositional resource—​or a com-
concrete things (physical pitches or frequen- positional constraint.6 The proliferation of avail-
cies), while others are abstract formal construc- able spaces facilitates a diversity of payoffs: we
tions (think of a collection of symbols standing may draw not just one map but many, viewing
for musical objects—​a space not of notes but the same musical edifice from multiple vantage
of possible names for notes). There are multidi- points and obtaining a prismatic, multilayered
mensional spaces whose elements are not single picture.7 If a space proves inadequate to capture
notes but chords, or ordered collections of notes a musical intuition, the inadequacy may reveal
such as twelve-​tone rows. Other spaces have as something about the space, the music, or both,
their elements time points, durations, or other and can motivate and inform the search for a
rhythmic entities. Some spaces date back con- more suitable spatial perspective.
ceptually to ancient Greece, some a few centu-
ries, some only a few years, and many spaces are THIS BOOK AND HOW
presented here for the first time. TO USE IT
Spatial map-​making is a creative act, guided
by analytical objectives, musical intuition, and This book’s focus on musical spaces brings with
aesthetics. Sometimes more than one geometric it a visual orientation, evident in the abundance
arrangement may be imposed on the same col- of figures. Other distinguishing features include
lection of objects—​the twelve pitch classes, for a unifying theoretical framework embracing
example, may be arranged in a chromatically spaces of many different kinds; the integration
organized clock diagram, a circle of fifths, or a of transformation theory with geometric theory,
neo-​Riemannian tonnetz—​and we may compare diatonic theory, and other techniques; the atten-
the musical relationships suggested by the vari- tion given to continuous spaces, in a departure
ous arrangements. Often two or more spaces, from the discrete mathematics that has more
though differing in genesis and content, turn often been used to describe musical structure;
out to have similar mathematical properties, the prominent status accorded to mod-​7 (generic)
and in the spirit of generalization that pervades structure on a nearly equal footing with mod-​
higher mathematics, we can learn about many 12, with related emphasis on formal aspects of
spaces at once by studying that shared struc- note spelling, the line of fifths, and enharmonic
ture. In addition to defining particular spaces, equivalence; and the hundreds of exercises. The
therefore, we will also define families of spaces book draws heavily on the work of numerous
sharing certain characteristics, among them scholars and includes expository treatment of
interval spaces (Lewin’s generalized interval sys- many concepts that have appeared elsewhere;
tems), distance spaces (known to mathematicians in addition to the sometimes copious endnotes,
as metric spaces), and voice-​leading spaces (which each chapter concludes with a short section list-
include the so-​called OPTIC spaces, which display ing suggested associated readings. But every
an assortment of intricate geometric and topo- chapter also contains new ideas, approaches,
logical structures). and examples, from which I hope that even those
Spaces are the worlds in which music lives, already well versed in mathematical music the-
the warp and weft of the musical universe. A ory will benefit.

P reface • xi
 In the hope of dismantling the barriers that while others illustrate mathematical concepts;
have discouraged many musicians from pursuing there are also transformation graphs, networks,
mathematical work, I have tried to keep math- annotated scores, and other analytical illustra-
ematical prerequisites to a minimum, assum- tions. Readers may benefit from scanning each
ing little more than fluency with basic algebraic chapter’s figures before reading the chapter, and
notations and manipulations. More advanced I encourage them to think about how the figures
mathematical concepts are introduced as have been constructed, why they take the forms
needed, with musical motivations kept close at they do, what the various annotations signify,
hand as much as possible. On the musical side of and—​as there are nearly always other options—​
the ledger, the book assumes acquaintance with how the figures might be improved.
the fundamentals of tonal theory, pitch-​class set In a nod to mathematical tradition, the exer-
theory, and twelve-​ tone theory.8 Schenkerian cises sprinkled through the text are integral
ideas and other advanced topics make only spo- to the fabric of the book, vital to the develop-
radic and largely tangential appearances. ment of its ideas. While I fondly imagine an
Watching music theorists and students grap- ideal reader assiduously working every exercise
ple with mathematical challenges, I have repeat- in sequence, the quantity of exercises will ren-
edly been impressed with the keenness of their der this unrealistic for most readers. Even so, all
intuitions about abstract mathematical concepts are encouraged to read and think about every
and structures. But an ability to communicate exercise and to work out the details of as many
these intuitions correctly and logically, with as they can. The exercises span a wide range of
proper terminology and notation, is sometimes difficulty levels, from the nearly trivial to the
lacking, even in published work.9 I hope that this abstract and open-​ended. Solvers may be asked
book may assist those who wish to develop these to produce their own figures, consult scores of
skills. There is a need for formalism, and I will musical works, or make calculations of various
neither suppress nor apologize for the relevant kinds. Some exercises involve proofs, perhaps
mathematics, but I can at least try to help read- disguised by words such as explain rather than
ers negotiate it and feel less threatened by it. prove; readers with proof-​ writing experience
Readers should be prepared to pay close atten- should be able to produce rigorous proofs, while
tion to definitions (what is required for an equiv- others may interpret the instructions to mean
alence relation?), to the identities and properties “explain to your satisfaction.” Hints for some
of the objects with which they are working (are exercises will be found in the endnotes (suf-
these notes pitches or pitch classes?), to niceties ficiently inconvenient, I hope, to avoid undue
of logic (if the transformations PR and T3 both temptation). The text sometimes refers back to
map a C-​major triad to an E♭-​major triad, can we previous exercises, giving away an occasional
conclude that PR =​ T3?), and to the assumptions answer in the process.
underlying seemingly simple questions (are F♯ Musical examples are numerous, even though
and G♭ the same note or not?).10 the book as a whole is more theoretical than
The simultaneous quest for readability and analytical. Many short examples are used to
mathematical rigor necessitates some com- illustrate concepts and techniques more than to
promises. A mathematician who picks up this offer analytical insights, though I hope to reward
book might wish for fewer elementary explana- readers with some of the latter along the way. A
tions, more proofs, and, in places, more precise handful of more substantial analytical mini-​
language.11 While no style of exposition can be essays, numbered separately as “examples,” go
ideally suited for all possible readers, I hope further, but readers should not expect to find
that the balance I have struck can provide use- comprehensive musical analyses. Figures, exer-
ful models for others setting out to work in the cises, and numbered examples, along with the
field of mathematical music theory and to write occasional table and mathematical theorem, are
about it. numbered in a single continuous stream within
The interplay between language and image in each section, a strategy that eliminates number-
musical analysis can be rich and complex, and ing conflicts and should make each item easy to
diagrams often take pride of place. If the fig- find by number.
ures in this book do not entirely tell the story by It is worth noting a few other things that
themselves, they at least shoulder a large part of this book is not. There is no corpus-​based study
the burden. Many figures depict musical spaces here, and no discussion of computational mod-
or relationships among elements within a space, eling or computer simulation. Though many

xii • P reface
mathematical techniques can trace a lineage generic spaces and the generic/​diatonic distinc-
through several decades or centuries of scholar- tion, the line of fifths, and the concept of spelled
ship (and I have tried to supply citations where pitch classes, all introduced in Chapter 1; the dis-
appropriate), I make no claim that the meth- cussion of the orthography problem for compo-
ods presented here represent historicist modes sition of functions in Section 2.3 (whose upshot
of musical thought. I generally refrain from is that chronological, left-​to-​right orthography
describing musical phenomena in perceptual or is preferred throughout the book); the functions
cognitive terms, though in many cases I suspect on modular spaces newly defined in Section 2.6;
that readers will find that they can hear in the and the distinction between finite and infinite
music what the theory describes, if only they will tonnetz graphs in Section 3.3.
try. There is little discussion of the ontological or Readers may choose to read the remainder of
epistemological status of spaces and other math- the book more selectively, and they have a vari-
ematical objects. Interactions of all of these ideas ety of options for doing so, as the choice of chap-
with the material in this book could be pursued, ters from the successively shorter Parts Two,
but doing so would strain both the boundaries of Three, and Four may be shaped by the reader’s
my own expertise and the boundaries of what is interests. Part Two further develops group-​theo-
already a very large volume.12 retic, transformational, and intervallic concepts.
Those munificent dimensions make a brief In Part Three the mathematics takes a topologi-
guide to navigating the book more than ordi- cal and geometric turn with the construction of
narily useful. If I were asked what I wish all multidimensional voice-​ leading spaces, while
music theorists knew about mathematics, the the focus in Part Four is on scales.
five chapters of Part One would be a good first Those who proceed straight ahead from Part
approximation to an answer. The material in One into Part Two will find that the level of
these chapters is fundamental to what follows. abstraction bumps up a notch in parts of Chapter
Chapters 1 and 4 in particular introduce many 6, and perhaps another notch in Chapter 7. The
spaces that will reappear throughout the book, group theory in Chapter 6 is more advanced
including an assortment of pitch and pitch-​class than that in Chapter 5, and some of the densest
spaces and tonnetz spaces; a series of analytical reading in the book is in Chapter 7, which probes
vignettes at the end of Chapter 4 illustrate some deeply into the theory of interval spaces (though
ways to view musical examples through these some of the complexity is notational more than
spatial lenses. The other three chapters in Part conceptual). Readers are exhorted not to be dis-
One offer primers in important mathematical couraged, as they will find themselves on gentler
techniques. Chapter 2 covers mathematical set terrain with Chapter 8’s triadic transformations
theory, including functions, equivalence rela- and neo-​Riemannian theory, and also in Chapter
tions, and modular arithmetic. Graph theory is 9, which combines a more detailed look at trans-
the subject of Chapter 3, and Chapter 5 intro- formation graphs and networks with topics
duces group theory. These chapters include short in the theory of serial transformations. These
introductions to several topics explored in greater chapters may, I believe, be read out of sequence.
depth later on, including neo-​Riemannian trans- Chapters 8 and 9 include references to mate-
formations, first appearing in Chapter 2; trans- rial from Chapters 6 and 7 on which they logi-
formation graphs and networks in Chapter 3; cally depend—​for example, the group-​theoretic
and interval spaces in Chapter 5, culminating in duality in Section 7.5 finds applications in both
Lewin’s insight about the relationship between triadic and serial contexts—​ but readers who
intervallic and transformational thinking. choose to skip Chapters 6 and 7 may still read
I hope that these opening chapters and the Chapters 8 and 9 profitably, referring back to the
exercises within them will provide valuable skipped chapters only as needed.
practice in thinking mathematically. There are The shadow of Lewin hangs large over parts
occasional pointers about logical reasoning, the of Chapters 7 and 9, as it does over the end of
organization of proofs, and the precise use of Chapter 5, but the spatial orientation gives
terminology. Some readers will come to the book Lewinian concepts a different flavor and (I hope)
well acquainted with much of what these chap- makes them more approachable, and many
ters have to offer; I encourage them to scan the of Lewin’s observations are developed in new
chapters to acquaint themselves with the musical ways, sometimes with simpler proofs. The neo-​
spaces and notational conventions. Distinctive Riemannian portion of Chapter 8 owes much
features to which attention may be called include to Richard Cohn. These chapters also include

P reface • xiii
reworked treatment of material I have published late 1970s, thus predating much of the mate-
previously on cross-​type transformations, map- rial in Parts Two and Three. Readers who have
pings of interval spaces, uniform triadic trans- worked through the thorniest portions of
formations, and the path consistency property Chapters 7, 11, and 12 will undoubtedly find the
of transformation graphs, as well as some new going in Part Four comfortably easier, and those
perspectives on serial theory. with an interest in scales may be able to negoti-
While many musical intuitions may be effec- ate Part Four without having read much of Parts
tively modeled using transformations, many Two and Three at all. These chapters are placed at
others cannot. The geometric spaces constructed the end of the book because of their many points
in Part Three offer many examples of the latter of contact with both the transformational and
kind. Though their compatibility with transfor- the geometric realms. To reap maximum benefit
mational methods is limited, these spaces are from this material, therefore, readers who have
general enough to accommodate a wide range arrived here via shortcut may find themselves
of musical phenomena, notably those related making occasional detours into the chapters they
to voice leading. Part Three depends on Part have passed over. Among several new features in
Two only in a minimal way (most importantly Chapter 13, my prior work on signature trans-
on some of the group theory from Chapter 6), formations is thoroughly reimagined; a new con-
and readers eager for geometry may skip directly cept of enharmonic coordinates is introduced;
from Part One to Part Three if they wish. The genus and species are defined using OPTIC
three chapters of Part Three should, however, be relations; and signature transformations are
read in order. Whether readers find Part Three employed in a new and general formulation and
easier or harder than Part Two may depend on proof of “cardinality equals variety.” In Chapter
whether they are more at home with geometry 14, spelled heptachords offer alternative lenses
or algebra. through which to view some chromatic har-
These voice-​ leading spaces are the OPTIC mony, and the final section is a highly selective
spaces, first described by Callender, Quinn, and exploration of some microtonal structures.
Tymoczko. One way to describe their role is to
say that they add a geometric component to the —​—​—​
study of pitch-​class sets. They are topological
spaces of a kind called orbifolds, with sometimes Few people would deny that music can be beautiful,
counterintuitive, even exotic, features (one but many might be surprised by the notion that
example is a Möbius strip). Chapter 10 provides mathematics can be beautiful. I hope that readers
motivation for these spaces, a careful formula- of this book will come to share the realization that
tion of the OPTIC equivalence relations from a powerful, complex, multifaceted mathematical
which they are derived, and, in Section 10.4, a apparatus assembled as a means of understanding
general system of normal forms for elements musical structure can be not only a tool but also
of OPTIC spaces, new in this book. Chapter 11 a thing of beauty in itself. The non-​planarity of
works in detail through the construction of the the finite tonnetz (Section 3.3), symmetries codi-
most important OPTIC spaces in as many as four fied by group structure (Section 6.5), the fantastic
dimensions, so it should come as no surprise topology of orbifolds (Chapter 11), the startling
that some of these spaces can be challenging to elegance of “cardinality equals variety” (Section
visualize. Chapter 12’s distance functions offer 13.4)—​these are abstract mathematical notions,
a sort of geometric rejoinder to Lewinian inter- and they can captivate our brains and tickle our
val functions suitable for the more general set- senses every bit as much as any music that we may
ting of the OPTIC spaces, many of which do not use them to study.
admit interval structures. Many examples and Walt Whitman, in “When I Heard the Learn’d
some challenging exercises in Chapters 11 and Astronomer,” describes being left “tired and
12 delve into structural details of OPTIC spaces sick” by the figures, charts, and calculations
at a level of granularity not found elsewhere. in an astronomy lecture, whereupon he went
Several of the fundamental concepts from outside and “look’d up in perfect silence at the
diatonic theory and scale theory in Part Four, stars.” Reading this poem, I cannot escape the
including the genus-​ species distinction, the sense that it is not the lecturer who has missed
principle “cardinality equals variety,” and maxi- the point, as Whitman suggests, but Whitman
mal evenness, stem from the work of John himself. To the astronomer, an understanding of
Clough and his collaborators beginning in the celestial mechanics, spectroscopy, or the physics

xiv • P reface
of star formation does not make the night sky 8. Of the many available texts in tonal theory, Laitz
any less beautiful; it makes it more beautiful. In 2016 is admired for its comprehensiveness and high
the same way, my fondest hope is that readers intellectual standards, Burstein and Straus 2020 for
of this book will come to recognize that a math- its concision. Both of those texts include condensed
ematical conception of the musical universe can introductions to pitch-​class set theory and twelve-​
only enrich their musical experience. tone theory as well; for a more detailed presentation
of these subjects see Straus 2016.
NOTES 9. Complaints about lax mathematical standards in
musical writing are not new; see, for instance, Regener
1. Incomprehension can verge on hostility. After 1974, which criticizes imprecision and logical omis-
Lewin’s first publication (Lewin 1959), a letter to sions in Forte 1973.
the editor of the Journal of Music Theory (Swift 1960) 10. The answer to the penultimate question is no:
accused Lewin of trying to intimidate readers with PR and T3 are different transformations. (See Section
methods akin to “killing a gnat with a pile-​driver.” 2.3.) This book’s answer to the last question is that
Decades later, a less caustic letter in the same journal there are some musical spaces in which F♯ and G♭ are
(Kaarela 1992), probably provoked in large part by the same note, and other spaces in which they are not.
Clough and Douthett 1991, lamented that the journal (Compare Figures 1.2.3 and 1.3.3.)
was “turning … into a Journal of Mathematics which, 11. A surprising number of terms have well-​estab-
really, is of little long term significance.” lished but conflicting meanings in music theory and
2. Schenkerian theory, in contrast, may be likened in mathematics, among them set, function, graph,
to a more specialized toolbox with somewhat more group, sequence, series, transposition, inversion, line,
explicit instructions. Buchler (2016, 41), invoking ter- composition, analysis, continuous, harmonic (and har-
minology from Lerdahl and Jackendoff 1983, remarks monic series), and metric (and metric space). The name
that transformational music theory offers well-​ Riemann could claim a place in this list, the math-
formedness rules but no preference rules. For related ematician Bernhard Riemann (1826–​66) and musi-
commentary see Hook 2007b, 165–​68. cologist Hugo Riemann (1849–​1919) both ranking
3. Alluding to the voice-​leading spaces studied in Part high among their fields’ eminent figures. In present-
Three of this book, Straus (2011, 46) observes that music ing mathematical concepts, I have generally followed
theory has entered “what might be called a new space age.” accepted mathematical usage, with exceptions in cases
4. Compare Westergaard 1996, 9: “What’s depicted where the risk of confusion seems high (for instance,
here is not music, but rather the space within which I avoid the group-​theoretic sense of transposition and
music moves.” the standard mathematical term metric space because
5. In a similar vein, Roeder (2009, par. 12.2) concedes of possible confusion with musical transposition and
that “transformational analyses are at best imperfect musical meter). I orient pitch-​class circles clockwise
and incomplete signifiers of musical experience.” from the top as music theorists do, rather than coun-
6. The idea of spaces as compositional resources terclockwise from the positive x axis as mathemati-
figures prominently in the work of Robert Morris cians prefer. Topological concepts are sometimes
(e.g., Morris 1987 and 1995) and occasionally in the approached with less rigor than algebraic ones; in par-
writings of Lewin and Tymoczko, among others. ticular, mathematicians may recoil at the discussion of
7. The word prismatic is borrowed from Rings 2011b, “continuous spaces,” a squishy term for which I do not
37–​38: “If any given passage of music admits of an supply a precise definition.
indeterminate number of … apperceptions, it of course 12. Some important topics in mathematical music
becomes impossible to execute anything approaching a theory are absent from this book. I particularly regret
complete analysis of a given work. Instead, the [trans- the omission of musical applications of the discrete
formational] apparatus encourages one to adopt an ana- Fourier transform, cryptically described long ago in
lytical technique that we might call prismatic, in which Lewin 1959, rediscovered by Ian Quinn (2006 and
phenomenologically rich local passages are refracted and 2007), and developed further by Emmanuel Amiot,
explored from multiple perspectives.” Jason Yust, and others.

P reface • xv
 Acknowledgments

Writing a book can feel like the most solitary insights, some of the most powerful and provoc-
of pursuits, never more palpably than in the ative in our field. I am unable to imagine what
coronavirus-​ shadowed world in which I find this book could have been, or if it could have been
myself setting down these words. But no book is anything at all, had not so much fertile ground
the product of a single mind—​certainly not this been broken by trailblazers such as these.
one, which has drawn ideas and inspiration from By great fortune, the first two editors of
countless other people. Oxford University Press’s Studies in Music
Of the scholars who have influenced my work Theory series have been important scholars in
the most, special mention must be made of a few areas that intersect with topics in this book. Rick
whose ideas permeate the book in ways that go Cohn, wearing his editorial hat, supplied gentle
beyond their many citations here. Two of them prodding as the book started to take shape, and
left us too soon: I never had the opportunity to his successor Steven Rings has been generous
know David Lewin, but I am fortunate to have with encouragement and forbearance as my work
had a few years in which to share ideas with John has progressed at its sometimes laggardly pace
Clough, whose enthusiasm for my first music-​ and outstripped its originally projected size.
theoretical work provided a much-​needed infu- I have taught topics related to this book
sion of confidence. Around the same time, I was in classes and seminars dating back almost
finding Richard Cohn’s then-​ new ideas about two decades, first at Penn State University but
neo-​Riemannian transformations, the tonnetz, mostly at Indiana University, as well as in a grad-
and parsimonious voice leading hugely stimulat- uate student workshop at the Society for Music
ing—​a description that applies as well to Rick’s Theory’s annual conference in 2011. Students
more recent work on diverse topics. Then there in recent iterations of my classes have read par-
is Dmitri Tymoczko, an inexhaustible fount of tial drafts of the book and offered many helpful
suggestions. I have benefited from discussions corrections), has re-​ energized me more than
and exchanges of ideas with numerous other once when progress was flagging. I have learned
people as well—​established scholars, students, that on the rare occasion when Leah finds some-
students who are now established scholars—​ thing difficult, it means I need to be clearer.
learning new things (and new music) from all of Indiana University, my scholarly and musical
them. All of this makes for a long list of deserv- home for a large fraction of my life, is also home
ing thankees and a strong likelihood that I will to many treasured colleagues, teachers, and role
overlook someone unjustly. I am particularly models. I am grateful to Gwyn Richards, long-
grateful to Jack Douthett, Nathan Lam, Robert time dean of the Jacobs School of Music, and
Peck, and Joseph Straus for many stimulating to all my current and former colleagues in the
conversations and other encouragement, and to department of music theory for the congenial
Catherine Losada and her seminar students at and intellectually invigorating environment they
the University of Cincinnati for their comments have unfailingly provided. Interactions with Kyle
on portions of the manuscript. For other con- Adams, Robert Hatten, Gretchen Horlacher, Eric
tributions large and small, I thank Sara Bakker, Isaacson, Roman Ivanovitch, Blair Johnston,
Breighan Brown Boeskool, Clifton Callender, Marianne Kielian-​ Gilbert, Andy Mead, Gary
David Clampitt, Thomas Fiore, Kyle Fyr, Rachel Potter, Frank Samarotto, and Daphne Tan
Hall, Daniel Harrison, Michael McClimon, have shaped my own musical thought and this
Brian Moseley, Jinny Park, Ian Quinn, Ramon book, and I cherish Mary Wennerstrom’s many
Satyendra, Paul Sherrill, Charles Smith, Nathan years of uniquely cheerful support and men-
Smith, James Sullivan, Ryan Taycher, Jason torship. I have worked on the book during por-
Yust, and Marek Žabka. tions of two IU sabbaticals, the first of which
My debts to Leah Frederick surpass anything was also supported by a Sabbatical Fellowship
I could express by including her name in a list. from the American Philosophical Society; an IU
Leah’s expertise in both the musical and math- Presidential Arts and Humanities Production
ematical realms makes her an astute reader and Grant has helped to defray publication costs. A
critic of this book, and the alacrity with which final word of appreciation goes to Mary Hrovat,
she has devoured each new chapter, some in whose abiding friendship has made this compli-
multiple drafts, working exercises and supply- cated game of solitaire a little less solitary. Now,
ing spot-​on comments (and more than a few Mary, finish that novel.

xviii • A cknowledgments
 PART ONE
Foundations of Mathematical
Music Theory
Spaces, Sets, Graphs, and Groups
 1
Spaces I
Pitch and Pitch-​Class Spaces

TO SOMEONE who has never considered the somehow be visualized: if some sort of diagram-
question, there may be no obvious reason why matic representation can be devised showing,
music’s underlying structures should be repre- if only metaphorically, the important elements
sented graphically or geometrically. Music is an of the conceptual framework and their relation-
auditory phenomenon, consisting of vibrations ships with each other. Even if these elements
transmitted through the air as sound waves, exist in sound and time rather than in light and
received by our ears and processed by our brains space, such a representation—​a cross-​domain
as acoustic data. Visual aspects of music—​the mapping between the domain of musical events
arrangement of notes on the page or musicians and the physical space in which the representa-
on the stage, the gyrations of the conductor, the tion is constructed—​may help us get our bear-
shine of the piano—​are seemingly of secondary ings and interpret what we hear.1
importance in our appreciation and understand- One example of such a visual representation
ing of a musical work. is familiar to every musician: the musical score
Most human beings, however, are visually ori- itself. Centuries of evolution in music notation
ented. More of our cerebral cortex is devoted to have fashioned a system remarkable for its preci-
vision than to all other senses combined. Visual sion and efficiency—​attributes shared by many
and spatial metaphors pervade our language. We of the other varieties of graphic representa-
rely on our eyes more than our ears or any other tions that we will encounter in this book (many
sense organs in finding our way around, iden- of them conceptually much simpler than many
tifying other people and objects, detecting pat- scores). The score remains only a representa-
terns, and learning new information. Because tion—​few would argue that a score is the piece
of this dependence on what we see, most people of music it represents—​but it is so essential a
find abstract concepts easier to grasp if they can representation that no musician in the Western

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0001
classical tradition, in possession of no matter spaces’ geometric properties and their topological
how proficient a musical ear, would think of properties.
learning a new piece without a score.
Loosely speaking, a score is a sort of graph, 1.1 PITCH SPACES
one whose vertical axis depicts pitch and whose
horizontal axis depicts time. As such, it is a A particularly simple representation of pitch
two-​dimensional representation, in which the space takes the form of a one-​dimensional con-
dimensions depict two different domains, or tinuum in which musical pitches are identified
spaces. The structures of both the pitch and with points on the real number line, resulting
time domains are more complex than they may in something like an abstract sort of keyboard.
appear from this simple description.2 The pitch Figure 1.1.1 depicts several octaves of this space,
domain will receive the lion’s share of the atten- and the inset shows greater detail in the octave
tion throughout this book, and this opening from C4 (middle C) to C5. The figure also shows
chapter will survey a number of manifestations pitch numbers that we shall use occasionally, mea-
of pitch space. What we mean by space will be sured in semitones above or below middle C.3 We
clarified by the numerous examples here and can easily enough conceive of pitches higher or
in succeeding chapters, but we may understand lower than any that we can physically hear; for
the term loosely to mean a collection of musically theoretical purposes it is convenient to assume
meaningful entities in some visual or geometric that pitch space, like the number line, extends
arrangement. The preliminary discussion here infinitely in both directions (suggested by the
will provide informal introductions to a num- way the line tails off in an ellipsis at each end
ber of important concepts, and the elementary of the figure), even if only a finite portion of it
examples of musical spaces presented in this can be drawn or is musically useful.4 The names
chapter will be central to all that follows. pitch and cpitch shown in Figure 1.1.1 will be
The discussion here is less mathematically explained below.
rigorous than that in later chapters, and some of Diagrams recognizably similar to Figure 1.1.1
the exercises may be answered in a rather loose have been in use at least since the time of ancient
and informal way. Readers are encouraged to pay Greece, and these representations have taken on
particular attention to several important dis- many variations. Before examining some of the
tinctions that differentiate some of the spaces more substantive variants, let us briefly consider
and other concepts presented here. For example, the simple ones described here (readers may con-
we distinguish between discrete and continu- struct the diagrams themselves if so inclined):
ous spaces; between chromatic and diatonic or
generic spaces (and we also make a distinction (a) like Figure 1.1.1, but with the note names
between diatonic and generic); between finite printed in a different typeface
and infinite spaces; between spaces in different (b) like Figure 1.1.1, but drawn on a larger
numbers of dimensions; in some cases, between sheet of paper at a larger scale, so that notes an
conformed and unconformed spaces; and between octave apart are separated by six inches

FIGURE 1.1.1 Pitch space, either discrete (pitch) or continuous (cpitch)

4 • E x p l o r i n g M u s i c a l S p a ces
 (c) like Figure 1.1.1, but showing pitches Variant (g) differs from the others in that it
from C3 to C7 rather than from C2 to C6 involves changing the labels by which some ele-
(d) like Figure 1.1.1, but in mirror image, ments of the space are identified. While there
with high notes to the left and low notes to may at times be good reasons to distinguish C♯
the right from D♭, in the most familiar interpretation of
(e) like Figure 1.1.1, but oriented vertically, pitch space these are two different names for
with high notes at the top and low notes at the same note, so on this basis we may say that
the bottom (g) continues to represent the same space as the
(f) like (e) but with high notes at the bottom original figure. In Figure 1.1.1 the sharp names
and low notes at the top have arbitrarily been chosen for all the black
(g) like Figure 1.1.1, but with the black keys keys, but other conventions are equally viable. A
labeled D♭4, E♭4, … rather than C♯4, D♯4, … labeling of a space may be regarded as a mapping
between the elements of the space and a collec-
Readers will probably agree readily that vari- tion of possible labels; to say that a labeling need
ants (a) and (b) are trivial; the diagram remains not be unique is to say that this mapping need
“the same” in all meaningful respects when the not be a one-​to-​one correspondence (a term to
font or the graphic scale is changed. The altera- be defined precisely in Section 2.3).
tion in (c) shifts the window that the figure Our formalization of musical spaces will
provides into the space, but this change too is ultimately consider that all of the above vari-
inconsequential as far as the structure of the ants indeed represent the same space. Several
space is concerned: high or low notes not physi- important points emerge from this discussion.
cally shown in a particular rendering are concep- First, a graphical transformation of a space does
tually still present in all the diagrams. not necessarily alter the space in any structural
Whether versions (d), (e), and (f) should be way, or in any way that matters to us at the
considered “the same” as the original may be less moment. But second, even if two representa-
clear. On the one hand, these alternatives are tions convey identical intrinsic information
all merely reflections or rotations of the origi- about a space, there may be legitimate visual,
nal picture, as if the same structure were being perceptual, conventional, or other extrinsic rea-
viewed from different angles; they all depict the sons for preferring one to the other. This second
same underlying pitch space (the same pitches point illustrates a third, broader principle that
are included in all of them) and convey funda- will be invoked frequently throughout this book:
mentally identical information about that space we should not allow our work to be constrained
(the order relations among the pitches are the by the limitations of a particular space. While we
same). On the other hand, these transforma- may be working primarily within the confines of
tions alter our perception of the space in ways a space whose structure is narrowly defined, we
that might be considered significant. The like- are always free to step out of the system momen-
ness of the space to a keyboard is sacrificed in tarily, basing musical, analytical, graphical, or
the vertical representations (e) and (f), and pedagogical decisions on considerations exter-
musicians may find the orientations of (d), with nal to that space.
high notes at the left, and (f), with high notes at Now consider Figures 1.1.2 and 1.1.3. In
the bottom, counterintuitive. These perceptions these diagrams the pitches in the underlying
are based on learned conventions that are not space remain the same as before, but the geom-
intrinsic to the structure of the space: there is etry has been altered in fundamental ways.
no musical reason why C5 must lie to the right In Figure 1.1.2 the lower register has been
of C4 rather than above, below, or to the left of stretched and the upper register compressed,
it. But learned conventions can be powerful, to so that for example the C2–​C3 octave appears
the point that an arrangement such as (d) may much larger than the C5–​C6 octave. In 1.1.3, the
be genuinely confusing in some situations, and reverse is the case, the higher octaves now being
one might reasonably choose to avoid using it on stretched. At first blush these diagrams may
that basis. appear oddly contrived, but there are cogent

FIGURE 1.1.2 A string-​length–​based representation of pitch space

Pitch and Pitch-Class Spaces • 5

FIGURE 1.1.3 A frequency-​based representation of pitch space

reasons for conceiving of pitch space in either of The discussion above illustrates the difference
these ways. Each note in Figure 1.1.2 is plotted between two branches of mathematics, geom-
at a point marking the length of a vibrating string etry and topology. The topological properties of
that will produce the given pitch. The small open a space are those that are preserved even if the
circle at the right end of the line, to which the space is stretched, compressed, bent, twisted, or
highest pitches converge, is not a point of the deformed in various ways (but without tearing
space as it does not represent an actual pitch, it, punching holes, or pasting together parts that
nor do any points of the space lie to its right. were not originally connected). Because Figures
This point corresponds to the theoretical loca- 1.1.1–​1.1.3 can be transformed into each other
tion of the string’s bridge; the greater the dis- by such deformations, the spaces are topologi-
tance from the bridge, the lower the note.5 In cally identical; topologically, they are all simply
1.1.3, meanwhile, each note is plotted in pro- lines.6 From the linear topology we may deduce
portion to its frequency, measured now from the something about the adjacency or ordering of the
open circle at the left side of the figure, which elements of the spaces, but not about distances
corresponds to a frequency of zero and again between those elements—​and it is the different
does not represent a pitch at all. A reciprocal ways of measuring distances and intervals that
relationship between the two representations is distinguish our three spaces. Measurements
visually evident in the dense clustering of notes that change when spaces are deformed are geo-
toward the right side of one figure and the left metric, not topological, properties.
side of the other. Even as interval spaces, however, the three
Figures 1.1.1–​ 1.1.3 may be described as figures are in some sense not very different.
three maps of the same space, but the differ- There are straightforward procedures for con-
ences among them are significant. It might be verting (or “mapping”) interval measurements
said that Figure 1.1.1 is a pianist’s view of pitch from one figure into another, and any one of
space, 1.1.2 a violinist’s, and 1.1.3 an acousti- the figures contains in itself enough informa-
cian’s. Although all three spaces comprise the tion to reconstruct the others, including inter-
same musical notes, each figure suggests its val structure. This situation may be summarized
own interpretation of distances between notes. by saying that the three interval spaces, while
If we wish to consider semitones in all registers not identical, are isomorphic. We will encounter
to be the same size, as they appear (approxi- several different manifestations of the concept
mately) on a piano keyboard, then Figure 1.1.1 of isomorphism in this book; isomorphism of
is an appropriate picture. If we want semitones interval spaces will be made precise in Chapter 7.
to get smaller as the register ascends, as they The representation in which all octaves are
do under the fingers of a string player, we may the same apparent size corresponds most closely
choose 1.1.2; or if we want them to get larger, with the way the pitch dimension is usually
as frequency values imply, we may choose 1.1.3. conceived in music theory, and of our three pre-
Distance and interval measurements will be dis- liminary depictions of pitch space, Figure 1.1.1
cussed in a general setting beginning in Chapter is the one that we will most often invoke. At
5, where we will define structures called interval least two important questions about this space
spaces that contain information not only about have been left unaddressed to this point, how-
membership in a space but also about intervals ever, and both have broad implications. First, is
between members. It will become apparent then the space of Figure 1.1.1 discrete or continuous?
that the spaces depicted in Figures 1.1.1–​1.1.3, That is, does the figure represent nothing more
though identical as abstract spaces (sets), are dif- than a chromatic scale, in which, as on the piano
ferent as interval spaces. keyboard, the next note after C is C♯ and there
is nothing in between? In that case (the discrete
Exercise 1.1.4 What note falls midway interpretation), the line drawn in Figure 1.1.1 is
between C4 and C5 in Figure 1.1.1? In 1.1.2? a fiction, a graphic convention to hold the figure
In  
1.1.3? together visually; the only real elements of the

6 • E x p l o r i n g M u s i c a l S p a ces
space are the dots. Alternatively (the continuous enharmonic equivalence is apparently implicit in
interpretation), we may imagine that the line the diagram, as the labeling of the black keys uses
segment between C and C♯ is infinitely divisible, only note names with sharps, evidently standing
packed full of intermediate pitches. To a pianist in for the enharmonically equivalent flats.
these intermediate pitches are unplayable, and Equal temperament and enharmonic equiva-
there is no standard way of naming or notating lence are usual assumptions in pitch-​class set
them, but they are acoustically real, may be iden- theory and some other related approaches to
tified numerically, and are physically available to musical structure. One way to answer our tun-
a violinist, trombonist, or singer. ing question would be to assume equal tempera-
Our answer to this first question is “both.” ment for the most part, and postulate a further
That is, Figure 1.1.1 represents two distinct multiplicity of spaces if needed to accommodate
spaces, discrete (or chromatic) pitch space and other tunings. Figure 1.1.1 could be taken first
continuous pitch space. The discrete space is a to represent an equal-​tempered pitch space; then
subset (or subspace) of the continuous, and the the same diagram, with some barely perceptible
two are plainly not isomorphic to each other. adjustments in the positions of the notes, could
Likewise, Figures 1.1.2 and 1.1.3 each exist in represent a just-​intonation pitch space, and so
both discrete and continuous versions, and the on through various other tunings (theoretically
same can be said of many other spaces to be in great numbers, because the non-​equal tunings
studied in this and the following chapters. Music are different for every key). This proposed solu-
theorists usually work in discrete spaces, dis- tion is less than satisfactory, partly because of
regarding the notes in the cracks. Continuous the unwieldy proliferation of spaces, but more
spaces will be of frequent interest in this book, particularly because it seems to force us to dis-
however, and we will encounter situations in tinguish among spaces whose differences are of
which a proper understanding of a discrete space no importance in most of our work: most music
depends upon an awareness of a continuous theory, after all, is carried out with no refer-
space of which it is a part.7 ence to tuning. We therefore adopt a different
In this book we will be referring to many dif- strategy.
ferent spaces regularly, including both versions A discrete pitch space, as we understand it
of Figure 1.1.1, so it is convenient to give them here, does not imply a particular tuning; rather, a
identifying labels: we refer to discrete chromatic given space may be tuned in a variety of different
space as pitch and continuous pitch space as ways. To make such an interpretation possible,
cpitch. In general we employ small capitals in we regard the space of Figure 1.1.1 in a formal,
labeling spaces; the initial c signals a continuous syntactic way, and we consider that the elements
space. Spaces pictured in figures will regularly be of this space are not physically real pitches (the
identified via legends like the one in Figure 1.1.1, name of the space notwithstanding) but abstract
with the space labels boxed. Not every space entities representing pitches. We may then study
mentioned in the book will be given such a label, formal relationships among these entities (their
but important spaces to which we will often syntax) without regard to the question of how
return will be labeled; Appendix 1 lists many of the pitches are tuned. In principle, they may be
these spaces for reference. tuned in any way at all, subject to certain mini-
The second unresolved question involves the mal constraints inferred from the diagram; the
tuning of the pitches in pitch space (particularly different tunings possible for a particular note
the discrete version). This question is more tech- are categorically equivalent as far as the struc-
nical than the first and will not be a primary con- ture of pitch space is concerned.8 In the discrete
cern of this book, but because of its fundamental pitch space pitch of Figure 1.1.1, the “abstract
nature it deserves to be considered briefly at this entities”—​the elements of the space—​are the
point and occasionally hereafter. Moreover, the note names …, B3, C4, C♯4, D4, …, and the “mini-
tactic by which we answer this question reveals mal constraints inferred from the diagram”—​
some important points about the nature of pitch the fundamental properties of the structure of
space and our conception of it. The equal spacing the space—​amount simply to the assertion that
of the notes in Figure 1.1.1, to the extent that it the notes are arranged in the illustrated order.
may be judged in a graphic portrayal of limited Though the diagram may suggest something
resolution, perhaps suggests that all the semi- about the relative sizes of acoustic intervals,
tones are of the same acoustical size, and there- that information is external to the structure of
fore that equal temperament is presumed. Also, the space, comparable in that regard to the size

Pitch and Pitch-Class Spaces • 7

or orientation of the figure as discussed above. this book will be in the realm of formal/​syntac-
In particular, notes appearing equally spaced in tic spaces rather than that of physical/​acoustic
the diagram need not actually be equally spaced spaces. We will make implicit assumptions about
in any physical sense. (We continue to refer to tuning when necessary, and when dealing with
the space of Figure 1.1.1 as pitch space in defer- continuous spaces, we generally assume that the
ence to tradition, but an alternate locution such notes are associated with specific frequencies,
as note-​name space might be technically more the integer-​valued points being equal-​tempered.
accurate in view of this syntactic interpretation.) But the approach just described allows us to con-
Construing spaces in this way gives us the sider other possibilities for discrete spaces. We
freedom to disregard matters of tuning most of will refer explicitly to acoustic spaces on occa-
the time, leaving open the possibility that (for sion, and we will revisit the tuning-​as-​mapping
instance) a perfect fifth may be a pure 3:2 inter- idea in Exercises 2.3.24–​2.3.26.
val in one application but equal-​tempered in Regarding pitch space as a formal rather
another, or that the notes of a discrete space may than physical construct pays further dividends
be tuned differently from the similarly named when spaces like the generic pitch space gpitch
notes in a continuous space. This understand- of Figure 1.1.6 are considered. Formally, the ele-
ing will also enable us to respect enharmonic ments of gpitch are the note names …, A3, B3,
distinctions on occasion: beginning with Figure C4, D4, E4, …, each consisting of a letter name and
1.1.11 below, we will construct spaces in which a register designation (hence the alternate name
F♯ and G♭ are two different elements that may or registral letter space). The space is generic—​the
may not represent the same acoustical pitch—​a term is borrowed from the scale theory litera-
distinction that becomes ever more significant ture—​in the sense that these note names do not
when we study relationships between diatonic include sharp or flat symbols; each octave of this
and chromatic spaces in Part Four of this book. space contains only seven notes, not twelve.9 In
If for some reason we wish to specify a partic- our space labels, generic spaces will regularly be
ular tuning, this may be accomplished with the signaled by the initial g.
aid of the frequency space shown in Figure 1.1.5. Having noted the evident likeness between
Frequency space, denoted freq, is a continuous Figure 1.1.6 and the C-​major scale, the reader
space whose elements are physical sound fre- may wonder why the space is called generic rather
quencies rather than abstract formal entities. A than diatonic. A diatonic scale, of course, includes
tuning may then be defined not as a space but as sharps or flats appropriate to its key (which may
a mapping from the syntactic space pitch to the mean none at all if the key is C major). The point
physical space freq, a mapping that assigns a of a generic space, in contrast, is that the pres-
physical frequency to every note of the syntactic ence or absence of sharps or flats is left unspeci-
space. Equal temperament at A4 =​440 Hz gen- fied. The elements of a generic space may be
erates one such mapping; the images under this called unsigned letter names—​the seven formal
mapping of the notes from the scale of Figure letter names without regard to accidentals. An
1.1.1 are shown beneath Figure 1.1.5, where unsigned letter does not automatically imply a
they form a discrete subspace of frequency space natural, but is taken to be an object of a categori-
that not surprisingly looks just like Figure 1.1.3. cally different kind, one that may be regarded
A different tuning would give rise to a different as standing in for any note named by the cor-
subspace of frequency space; that is, one syntac- responding letter, with or without an accidental
tic space may have multiple realizations in the attached.
physical space. By choosing to read every unsigned B as B♭, for
For reasons that are probably suggested by example, we may interpret generic pitch space in
the above discussion, the majority of our work in the context of a standard one-​flat key signature,

FIGURE 1.1.5 Frequency space freq

8 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 1.1.6 Generic pitch space gpitch (registral letter space)

FIGURE 1.1.7 Two diatonic pitch spaces

just as in ordinary musical notation a B on a staff concrete spaces are now signed: they include the
represents B♭ if the key signature demands it; in sharp and flat symbols required to distinguish
this interpretation the notes of the space form the various scales. The one-​flat space of 1.1.7a
not a C-​major scale but an F-​major scale. Or we thus includes the note B♭ in every octave. While it
may choose simultaneously to read every B as B♭ is tempting to identify this space as “the F-​major
and every C as C♯, resulting in a scale recogniz- scale,” it should be remembered that even a dia-
able as D harmonic minor—​not strictly a diatonic tonic space contains no information distinguish-
scale at all. Other interpretations would lead to ing one note as a tonic, so the more neutral term
structures even farther removed from diatonic “one-​flat diatonic pitch space” is preferable as a
scales, for which there are no common names, description of the space.12 We identify diatonic
even including the possibility that notes with the pitch spaces by the name dpitch (d for diatonic)
same name in different octaves may bear different together with an index number corresponding
accidentals.10 Intervals such as G♮–​B♮, G♮–​B♭, G♯–​B♭, to the number of sharps or flats in the diatonic
and G♭–​B♯, while of four different sizes in the chro- scale, counting sharps as positive and flats as
matic world, all reduce to the generic third G–​B; negative; the two spaces in Figure 1.1.7 are thus
we may say that they are generically equivalent.11 dpitch(−1) and dpitch(+​4). This definition does
(It would not make sense to call them “diatoni- not account for enharmonic equivalence: the six-​
cally equivalent”; the last two do not occur in any sharp and six-​flat diatonic collections dpitch(+​6)
diatonic collection at all.) In short, generic space and dpitch(−6) are two different spaces, and
admits many interpretations; while it may rep- more extreme spaces such as dpitch(+​8) or
resent a diatonic scale, it does not represent one dpitch(−15) are theoretically possible.
fixed diatonic scale, and it may also represent The generic space gpitch, and the distinction
pitch collections that are not diatonic at all. In between generic and diatonic spaces, may seem
this way, generic pitch space may be regarded as abstract. Conceptually, however, this distinc-
a sort of background system, conceptually prior tion underlies some of the most familiar topics
to the definition of specific accidentals (as well as in music theory. Generic space accommodates
tonal centers and other details), allowing various our understanding that the notes G♮, G♯, and G♭
diatonic scales and other scales with seven notes are all some flavor of “G,” and whenever we refer
per octave to be considered as realizations of the to interval sizes or chord types such as “third,”
same abstract space. “ninth,” “triad,” or “seventh chord,” we are implic-
More concretely, we may consider each dia- itly gauging pitch relationships in a generic space.
tonic scale as its own diatonic pitch space; two On a treble-​clef staff, we recognize that the inter-
examples are shown in Figure 1.1.7. The note val from middle C to any top-​space E is always
names that serve as the elements of these a tenth, regardless of accidentals that may be

Pitch and Pitch-Class Spaces • 9

appended to either note; this is a generic interval Generic space and all the diatonic spaces
measurement. Figure 1.1.8 distinguishes this way are by their very nature discrete, not continu-
of conceiving intervals in gpitch from the inter- ous: there are no letter names between C and
val reckoning in two other spaces. Identification D, nor do any notes lie between C and D in any
of the interval from C to E♭ as a “minor tenth” is diatonic scale (whether the letters bear acciden-
not possible in gpitch, but requires reference to tals or not). The notes in Figures 1.1.6 and 1.1.7
the chromatic space pitch.13 If a musical context are drawn at equal spacing. When we consider
of A♭ major is suggested, then we may refer to the generic pitch space as an interval space, we will
interval from C to E♭ as a “diatonic tenth,” indicat- define the generic interval between successive
ing something about an interval in the particular notes of this space to be 1 throughout the scale.
signed space dpitch(−4). The scale serves as its own yardstick for measur-
Interval measurements in these and other ing intervals; the interval between two notes of a
spaces will be discussed formally in later chapters, scale is calculated simply by counting steps of the
but the relationship between generic interval scale, much as a subway rider might measure dis-
sizes and the familiar ordinal names for inter- tances in stops rather than miles. In particular,
vals (such as “tenth”) merits brief attention to a generic interval measurement implies noth-
alleviate its potential for confusion. In chromatic ing about acoustic interval sizes; diatonic scales
space it is generally understood that the size of a contain steps of two different acoustic sizes, but
perfect fifth is seven semitones—​the number of both kinds of step are generic interval 1.
steps of a chromatic scale separating the notes in Notwithstanding the important distinctions
question—​and so we say that the chromatic, or just discussed, the one generic and the various
specific, interval between those two notes is 7. diatonic pitch spaces are all isomorphic. Loosely
Measuring similarly within a diatonic scale, notes speaking, two spaces are isomorphic if a picture of
a fifth apart are separated by four steps, and so one can be converted to a picture of the other by
although we call the interval a “fifth,” the generic renaming the elements, and in this case the only
interval between the notes in question is 4, not 5. renaming needed is the addition or removal of
Similarly the interval called a “second” has generic accidentals. This discussion suggests the idea (to
size 1, not 2, and the generic interval correspond- be developed in Chapter 13) that a key signature,
ing to the tenth in Figure 1.1.8a is 9, not 10. All or something like it, may be seen as a transforma-
the usual ordinal interval sizes—​second, third, tion, a means of converting generic space to dia-
fourth, …—​are off by one. This state of affairs tonic, or perhaps of converting one diatonic pitch
is a consequence of our way of naming intervals space to another.
(“fifth”) by counting notes (5) rather than steps
(4)—​an ancient and ineradicable tradition with Exercise 1.1.9
many awkward consequences.14 In particular, (a) Although they are isomorphic, there is a
basic arithmetic does not work in the familiar difference between the generic pitch space
terminology: compare “3rd +​3rd =​5th” with the gpitch of Figure 1.1.6 and the diatonic
corresponding generic interval equation 2 +​2 =​4. pitch space dpitch(0) with no sharps or
Although we use familiar interval names for con- flats (“C major”). Describe this difference as
venience in this book, whenever any calculation precisely as possible. Is the difference visible
is required, the generic interval sizes, obtained by in diagrams of the spaces?
counting steps, are always to be preferred.15 (b) Now consider the relationship between
gpitch, the generic pitch space of Figure
1.1.6, and pitch, the discrete chromatic
pitch space of 1.1.1. Explain how gpitch
may be identified with a subspace of pitch.
(c) Also explain why gpitch and pitch are
isomorphic. (An informal explanation will
suffice for now; this isomorphism will be
made more precise in Exercise 3.3.9 and in
  Section 7.2.)

An important and possibly counterintuitive

FIGURE 1.1.8 Interval measurements in three point, illustrated by parts (b) and (c) of the above
spaces exercise, bears mentioning: when infinite spaces

10 • E x p l o r i n g M u s i c a l S p a ces
are involved, it is possible for a space to be iso- and so on. Though the graphic arrangement
morphic to one of its own subspaces. may be unfamiliar, this space is in effect noth-
Diatonic scales are not the only subsets of ing more than the space of all possible standard
chromatic pitch space that can be viewed as note names—​a syntactic space, not a physical
spaces in their own right. We have already noted one. It differs from some previous figures in
that generic space gpitch may be realized in a that it makes no assumption about enharmonic
wide variety of non-​diatonic scales. One can also equivalence: F♯4 and G♭4 occupy two differ-
construct spaces with something other than ent locations in this space. As such, this space
seven notes per octave: pentatonic pitch spaces, enables us to rectify one shortcoming of the
whole-​tone pitch spaces, and octatonic pitch spaces, conception of tunings described previously: by
for example. While we will not explore every defining a tuning as a mapping from srl into
possibility, we will invoke spaces of these kinds the frequency space freq of Figure 1.1.5, we
on occasion. may construct a tuning in which F♯ and G♭ are
different pitches, something not possible for
Exercise 1.1.10 Explain why, when a mapping defined on the ordinary chromatic
constructing an octatonic pitch space, it is pitch space pitch.17
necessary to use signed rather than unsigned
letter names. Is the same thing true of a whole-​ Exercise 1.1.12 On a copy of Figure 1.1.11,
tone
   pitch space? draw curving lines to divide the diagram
into regions consisting of notes that are
The space called signed registral letter space enharmonically equivalent in the usual sense
(srl) in Figure 1.1.11 is the first figure we have of the term. For example, the notes E♯4, F4, and
encountered whose representation is funda- G♭  
♭4 should be in the same region.
mentally two-​dimensional rather than linear.
The note names appearing here are the signed Exercise 1.1.13 All of the diatonic pitch spaces
versions: those across the central row of the fig- may be regarded as subspaces of srl. Identify,
ure lack sharps or flats and may be understood in Figure 1.1.11, the note names (including
in this context as natural, while the same note accidentals) belonging to the spaces dpitch(−3)
names with sharps and flats appear above and and dpitch(+​4). Notice the nature of the visual
below the central row. Double sharps and flats relationship between these two sets of notes as
appear in the next higher and lower rows, and they appear in the figure—​a relationship that
in principle the diagram continues above and always holds between signed diatonic pitch
below with even more sharps and flats.16 The spaces whose indices differ by 7. Now identify
rows of the figure, then, may be identified with the note names for the space dpitch(+​11).
particular diatonic pitch spaces: dpitch(0) in If dpitch(−3) is E♭ major and dpitch(+​4)
the central row, dpitch(+​7) and dpitch(−7) is E major, what is dpitch(+​11)? How
above and below, followed by the progressively would you write the 11-​sharp key signature
more extreme dpitch(+​14) and dpitch(−14) theoretically required for this key? What more

FIGURE 1.1.11 Signed registral letter space srl

Pitch and Pitch-Class Spaces • 11

familiar diatonic pitch space is enharmonically
equivalent to dpitch(+​11)? What is the
difference between the index numbers of these
two
  enharmonically equivalent spaces?

Exercise 1.1.14 Triple (and higher) sharps

and flats are mainly an abstract theoretical
construction. They are assumed to be present
in Figure 1.1.11 for completeness, to allow
for the possibility that any note can always be
raised or lowered chromatically, no matter how
many sharps or flats it already possesses. (The
reasoning is similar to the understanding that
ordinary pitch space includes impractically
high and low registers: no matter how high or
low a note, some other note is always higher
or lower.) Nevertheless, one can construct
theoretically “correct” uses of triple sharps and FIGURE 1.2.1 A spiral representation of pitch
flats, even in familiar keys. space

(a) Write a short musical fragment consisting

of a few chords and tones of figuration in
F♯ major, in which the note F♯ ♯ ♯ appears
as a chromatic lower neighbor to scale
degree ♯2̂.
(b) Devise a similar use for the note B♭ ♭ ♭, in the
key of G♭ major.
(c) Play the notes F♯ ♯ ♯ and B♭ ♭ ♭ on a piano. What
  do you notice about them?18

1.2 PITCH-​C LASS SPACES

The pitch-​ space diagrams presented to this
point include no indication of specific relation-
ships between widely separated pitches. Indeed,
the linear topology of Figures 1.1.1–​1.1.3 does
not by itself accommodate this sort of informa-
tion. In a particular graphic rendering of a space,
however, the creative use of geometry may effec-
tively indicate such relationships.
Figures 1.2.1 and 1.2.2 illustrate two ways
in which the relation of octave equivalence
may be conveyed in a diagram of pitch space.
Topologically, these figures are lines like Figure
1.1.1. But in each case the line of pitch space has FIGURE 1.2.2 A helical representation of pitch
been curved into one coil per octave, bringing space
octave-​related pitches in close proximity to each
other. Figure 1.2.1 takes the form of a plane spiral
in which octave-​related pitches lie on spokes radi- the visual indication of the octave relationship.
ating from the center, while 1.2.2 depicts a helix If the coils are greatly stretched, either figure
in three-​dimensional space, with octave-​related becomes a long, slightly wavy line, not much
pitches aligned vertically.19 In either of these different from Figure 1.1.1, and octave relation-
figures, we may imagine tightening or loosening ships again become difficult to perceive. At the
the spread of the coils to strengthen or weaken other extreme, the coils may be pressed tightly

12 • E x p l o r i n g M u s i c a l S p a ces
 twelfth pitch. (Here we use the word identify in
the sense meaning “consider to be identical”; we
might also speak of “gluing together” or “con-
joining” every twelfth pitch.) As such, pc con-
tains only twelve different elements and is our
first example of a finite musical space. Continuous
pc space cpc, like its pitch counterpart cpitch,
includes an entire continuum of notes filling in
the space between C and C♯, but now each of
these intermediate notes is a pitch class rather
than a pitch. The remarks in Section 1.1 about
our abstract syntactic interpretation of pitch
space apply to pc space as well: although the
construction of the 12-​note space pc presup-
poses enharmonic equivalence, we nevertheless
understand that even this space may in principle
be tuned in a variety of ways.
FIGURE 1.2.3 Pitch-​class space, either discrete We have drawn Figure 1.2.3 with the pitch
(pc) or continuous (cpc) class C at the twelve-​o’clock position and other
pitch classes ascending in the clockwise direc-
together and taken to a mathematical limit in tion from there, matching the usual orienta-
which they all fuse into a single loop. At this tion encountered in pitch-​class set theory. In
point the octave relationship is so strong that many contexts, the pitch-​class numbers 0–​11,
also shown in the figure, are used in lieu of note
names. The identification of C with 0, C♯ with 1,
octave-​related pitches can no longer be distin-
guished from one another—​octave equivalence
is completely realized—​ and the entire space, and so on is conventional but arbitrary; other
whether originally conceived as spiral or helical, mappings between the notes and numbers, and
collapses into a circle, shown in Figure 1.2.3.20 other orientations of the diagram, are evidently
Circular models of musical pitch have been possible. When mathematicians work with
drawn at least since the seventeenth century.21 circles, in fact, they are more likely to employ a
The circle of Figure 1.2.3 is known in some per- counterclockwise orientation starting from the
ception studies as the chroma circle, but is proba- positive x axis (the three-​o’clock position).
bly familiar to most readers as the clock diagram
of pitch-​ class set theory.22 Indeed, because Exercise 1.2.4 In order to obtain Figure 1.2.3
octave-​related pitches are no longer distinct, the from Figure 1.2.2 as described above, must we
elements of this space are not pitches at all but view the helix from above or from below? How
pitch classes; the point labeled “C” in Figure 1.2.3 would the resulting picture of pitch-​class space
stands for an entire class of notes …, C2, C3, C4, differ if we viewed the helix from the opposite
C5, C6, … in any of the previous spaces. In termi- direction?
  
nology to be formalized in Section 2.4, a pitch
class (sometimes abbreviated pc) is an equiva- The final step in the construction of pc space,
lence class of the octave equivalence relation. The the conjoining of octave-​related pitches, is not a
points in Figure 1.2.3 represent these classes, so deformation of the sort that preserves topologi-
this space is pitch-​class space, or pc space.23 cal properties. Continuous pitch space cpitch
Its simple appearance notwithstanding, pitch-​ and continuous pc space cpc are topologically
class space shares some of the unsuspected distinct: the topology of pitch space is a line,
subtleties of pitch space and introduces a few while the topology of pitch-​class space is a cir-
more of its own. First, like the earlier diagrams, cle. If you start moving along a circle and keep
the circle of Figure 1.2.3 may represent either a going, you return eventually to your starting
discrete or a continuous space. Pitch-​class set point, something that never happens in a line.
theory traditionally works in discrete (or chro- Consequently, while in pitch space one note may
matic) pc space, which we label pc to distinguish be described as higher or lower than another,
it from discrete pitch space pitch. The space pc those terms are meaningless in pitch-​class space:
can be obtained from pitch by identifying every we cannot answer the question “Is D higher or

Pitch and Pitch-Class Spaces • 13

lower than G?” without knowing which D and neighborhood, the term used in topology) about
which G are intended. one pitch class in cpc, the continuous circle of
The terms ascending and descending, however, Figure 1.2.3—​say a quarter-​tone on either side
may be used without ambiguity in describing of D—​looks just like a corresponding neigh-
the two directions of motion possible in pitch-​ borhood about a pitch in cpitch, the line of
class space. We have already observed that the Figure 1.1.1. As we will discuss further below,
clockwise direction in Figure 1.2.3 corresponds the curvature of Figure 1.2.3, even if it can be
to ascending motion, in which notes of the chro- detected visually in the small neighborhood
matic scale are encountered in the order C, C♯, D, being examined, is not an intrinsic property
… . This state of affairs—​ascending and descend- of the space. Mathematicians refer to both the
ing are well-​defined concepts, but high and low line and the circle as one-​dimensional manifolds;
are not—​is not as paradoxical as it may appear, technically a one-​dimensional manifold is any
and one must understand it in order to make space whose local topology in a small neighbor-
sense of concepts such as voice leading in pitch-​ hood about any point is indistinguishable from
class space, to be studied at length later in this a line. A picture like Figure 1.2.3 is technically
book. A voice leading is generally defined not sim- an embedding of a one-​dimensional manifold
ply as a succession of elements of a space (such as (the space cpc) in a two-​dimensional space (the
notes or chords), but as a path through a space—​ plane in which the picture is drawn).
a way of getting from one place to another. In We will have much more to say about mani-
pitch space there is essentially only one way folds of one or more dimensions in our discus-
of getting from one place to another without sion of voice-​leading spaces in Part Three. The
reversing direction, but in pitch-​ class space relationship between pitch space and pitch-​class
that is not the case. In Figure 1.2.3 we can trace space is itself something that we will revisit
either an ascending (clockwise) or a descending several times in different contexts, not only to
(counterclockwise) path from G to D. Both paths deepen our understanding of these familiar
start at the same point and both end at the same spaces but also as a handy illustration of vari-
point, and it is precisely because these two paths ous mathematical concepts of which we will
are possible that we cannot say that one of the also be studying more complex manifestations.
two points is higher or lower than the other. Our study of sets and equivalence relations in
Chapter 2 will enable us to describe pc space as
Exercise 1.2.5 It may be said that there are a quotient set of pitch space. The construction
not just two ways of moving from G to D in will be encountered again in Chapters 5–​6 as we
pitch-​class space without reversing direction, study the algebraic structures underlying these
but infinitely many. Explain this statement, and spaces, in Chapter 7 when we study their prop-
give a musical description of the various implied erties as interval spaces, and in Chapters 10–​11
   leadings.24
voice as we generalize them to other voice-​leading
spaces.
The circular representation of pitch-​ class
space can sometimes lead to a mistaken idea Exercise 1.2.6 If Figure 1.2.3 is an embedding
of the dimensionality of this space.25 A circle of a one-​dimensional manifold in a two-​
drawn in the plane is a two-​dimensional figure, dimensional space, what is Figure 1.2.2? Is your
but pc space, like pitch space, is fundamentally answer affected when you consider that this
a one-​dimensional space. A point on a circle figure, like all the other figures in this book, is
may be determined by specifying a single num-    on a flat sheet of paper?26
drawn
ber, for instance an angle measurement from
0 to 360 degrees. Points of the plane that do The foregoing discussion might suggest a
not lie on the circle are not points of the space. question with epistemological undertones: is
Locally, at any point in pc space, there is only pitch-​class space really a circle? It is accurate to
one dimension in which one can move. (There say that pc space has the topology of a circle, and
are two opposing directions in which to move it is traditional to draw it that way, but there is
within that dimension—​ the ascending and no real justification for saying that pc space has
descending directions discussed above.) In fact, the shape of a circle. As observed at the outset,
although the global topologies of pitch space all of our diagrams of musical spaces are to some
and pc space are different, their topologies are degree metaphorical, and we have already noted
locally indistinguishable. A small region (or that the circle metaphor in particular can be

14 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 1.2.7 Two alternative representations of pitch-​class space

misleading because of unintended consequences

of its geometry. A circle is a plane figure with a
certain curvature; it divides the plane into two
regions, an inside and an outside. But pc space
has no curvature and no inside or outside. It
makes no musical sense to ask by how many
degrees pc space curves between the notes F and
A, or what lies at the center of Figure 1.2.3.
To reinforce this point, consider the two
alternative pictures of pitch-​class space shown
in Figure 1.2.7, neither of which curves at all.
The line in (a) at first glance resembles pitch
space, but the numerical register designations FIGURE 1.2.8 Generic pitch-​class space gpc
are absent. We are to understand, therefore, (letter space)
that the points of this space are pitch classes,
each of which appears many times along the
line. The three Cs visible in the figure all repre- may therefore be understood as a representa-
sent the same point of the space; the figure is tion of the entire space.
periodic in the sense that the same pattern of As depictions of pc space, both of the alterna-
elements repeats over and over. (Readers should tives in Figure 1.2.7 are just as valid as the circle
make peace with the idea that a single point of of Figure 1.2.3. They both share the topology of
a space may appear more than once in a picture a circle but not its shape. While the circle is easy
of that space; we will see many figures with this enough to grasp and avoids some of the disad-
property and will offer further commentary vantages of these alternative representations
about it as needed.)27 Figure 1.2.7b is not a line (the redundancy of 1.2.7a, and the arbitrary
but a line segment; it contains only one copy of placement of the cut marking the ends of the
the repeating module of 1.2.7a, and therefore module in 1.2.7b), it is important to remember
only a single appearance of each point of the that the circle has its own drawbacks, and there
space. The continuation symbols ≀ at each end are some advantages to the repeating-​module
of 1.2.7b express our understanding that the conception implied by the other diagrams. When
two ends are identified, so that we may move to we later consider analogous situations in higher
the right from pitch class B and arrive at C. It dimensions, we will find that modular represen-
is from attempting to realize this identification tations corresponding to the “straight circles”
in a physical way that the curvature and circular of Figure 1.2.7 are often easier to draw and to
shape of Figure 1.2.3 arise—​but these are arti- understand than representations more closely
facts of the higher-​dimensional physical space akin to the circular geometry.
in which the gluing is carried out, and of the Octave equivalence may be applied not only to
geometry of the resulting fabrication. In termi- continuous and chromatic spaces but to generic
nology that we will introduce more formally in and diatonic spaces as well, as illustrated in
Chapter 10, the line segment of 1.2.7b may be Figures 1.2.8–​1.2.10. First is generic pitch-​class
called a fundamental region for pitch-​class space; space (or gpc space), labeled gpc and shown in
the idea is that a fundamental region contains Figure 1.2.8; this is an abstract seven-​note space
one appearance of every point in the space and that may alternatively be called letter space, as its

Pitch and Pitch-Class Spaces • 15

FIGURE 1.2.9 Two diatonic pitch-​class spaces

scale may be modeled by mapping the appro-

priate space dpc(n) into scale-​degree space; for
example, we may interpret dpc(−1) as “C mix-
olydian” by mapping the note C to scale degree 1̂.
Enharmonic equivalence of diatonic spaces may
be deduced by mapping the spaces into pitch-​class
space; the spaces dpc(−6) and dpc(+​6) are differ-
ent, but if their notes are interpreted as ordinary
pitch classes, they define identical subsets of pc.
Enharmonic equivalence will be addressed in
other ways in several spaces introduced later in
this chapter, and formalized further in Section
FIGURE 1.2.10 Scale-​degree space sd 2.4 and in Chapter 13, where enharmonic ver-
sions of the spaces dpc(n) will be defined.
Figure 1.2.11 shows three representations
elements are simply the unsigned letter names C, of signed letter space (sl). This is essentially the
D, …, B. The generic pitch-​class numbers (gpc num- space of all traditional note names without reg-
bers) 0–​6 may substitute for the letter names. ister information—​ that is, assuming octave
The structure of this generic space is evidently equivalence. As such, signed letter space arises
similar to that of the discrete pitch-​class space pc from omitting the register designations from the
of Figure 1.2.3, but with only seven notes rather signed registral letter space srl (Figure 1.1.11);
than twelve.28 Generic pc space may be obtained this relationship is clear when the earlier figure
from generic pitch space gpitch (Figure 1.1.6) is compared with the rectilinear array of 1.2.11a.
by conjoining every seventh note. In gpc, as in Alternatively, signed letter space may be seen as
gpitch, the letter names may be taken to stand an extension of gpc (1.2.8), expanded to incor-
in for any notes named by those letters, with or porate all possible sharp/​flat signs; this deriva-
without accidentals—​but the register informa- tion is clearer in 1.2.11b, where the sharp notes
tion present in gpitch is missing in gpc. have been added in concentric rings outside the
A family of diatonic pitch-​class spaces is avail- original unsigned notes and the flat notes inside.
able when particular diatonic scales including The rendering in 1.2.11a shows a repeating mod-
appropriate accidentals are needed. Two exam- ule—​both C’s in the figure represent the same
ples, dpc(−1) and dpc(+​4), corresponding to the point of the space—​and is conceptually similar to
similarly named pitch spaces dpitch(−1) and the depiction of pitch-​class space in 1.2.7a. If we
dpitch(+​4) of Figure 1.1.7, are shown in Figure wrap this figure around on itself in the horizontal
1.2.9. Scale-​degree space sd, in Figure 1.2.10, is direction so that the identically named notes are
similar to the numerical representation in 1.2.8 conjoined, we obtain the picture in 1.2.11c, which
but uses the traditional circumflexed scale-​degree is not merely a circle but a cylinder of theoretically
numbers 1̂–​7̂ rather than 0–​ 6.29 The spaces infinite height. Notes named with the letters B, C,
dpc(n), like dpitch(n), account neither for tonal D, and E are visible on the front of the cylinder,
centers nor for enharmonic equivalence. The while F, G, and A are hidden on the back. One can
designation of a tonic note within a diatonic visualize Figure 1.2.11b as a distorted form of this

16 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 1.2.11 Three representations of signed letter space sl

cylinder, flattened into a plane, in which the top you transfer that drawing onto Figure 1.2.11a
of the cylinder (consisting of the note names with without any other changes, do the regions
sharps) has been stretched outward in ever-​larger accurately depict enharmonic equivalence in
circles, and the bottom (note names with flats) is this space? Sketch these regions in 1.2.11b
compressed inward, converging to a point at the and 1.2.11c as well. Theoretically, how many
center of the picture (not a point of the space). All different regions are there in Figure 1.1.11,
three versions of the figure have the same topol- and how many different regions are there in
ogy, which mathematicians describe as “the topol-    1.2.11?31
Figure
ogy of a cylinder.”30 While it is useful to be able to
conceptualize the cylindrical topology in any of The groupings of signed letters formed by
these ways, the modular rectilinear representation the regions in the preceding exercise corre-
of 1.2.11a is the easiest to work with in practice. spond to ordinary pitch classes; the exercise
essentially defines a mapping from signed letter
Exercise 1.2.12 In Exercise 1.1.12 you space into pitch-​class space. The mapping is not
partitioned the space srl of Figure 1.1.11 into one-​to-​one, because enharmonically equivalent
regions to show enharmonic equivalence. If spellings such as D♯ and E♭ are different signed

Pitch and Pitch-Class Spaces • 17

letters but represent the same pitch class. A its own generating interval. In the case of pc, the
signed letter may be called a spelled pitch class generator is a semitone; fifth is generated by
(spc), a pitch class for which a particular spell- the perfect fifth.33 As with the chromatic circle,
ing has been specified, and signed letter space some other graphic aspects of the circle of fifths
may be called spelled pitch-​class space—​a way of are purely matters of convention: Figure 1.3.1
describing pitch-​class space while respecting the follows usual practice in placing C at the top and
distinctions between enharmonically equivalent plotting ascending fifths (or descending fourths)
spellings—​though we shall generally reserve the in the clockwise direction.34
name spelled pitch-​class space for a different geo-
metric representation of the signed letters, to be Exercise 1.3.2
considered shortly (Figure 1.3.3).32 Similarly, the (a) Referring to the pitch-​class numbers in
space srl of Figure 1.1.11, which adds register Figure 1.3.1, describe mathematically the
distinctions to signed letter space, may alterna- relationship between adjacent numbers.
tively be called spelled pitch space. We will invoke (b) In Figure 1.2.3, if you connect pitch classes
spelled pitch classes often in this book, most in circle-​of-​fifths order by drawing straight
importantly in the study of scales in Part Four. lines across the circle from C to G, G to D,
Although spelled pitches and pitch classes are not and so on, what shape do you get? What if
a part of traditional theoretical nomenclature, you connect pitch classes in chromatic order
the concepts encapsulate ideas that are encoun-   in Figure 1.3.1?
tered routinely in the study of tonal music,
where enharmonic distinctions—​ between F♯ The relationship between the circles of pc and
and G♭, between an augmented second and a fifth is a deep one both mathematically and
minor third, or between a dominant seventh and musically, and we will revisit it several times.
a German augmented sixth, for example—​are Because different notes are adjacent in the two
generally considered indispensable. diagrams, one cannot be transformed into the
other by something as simple as a rotation or
1.3 SPACES GENERATED BY reflection. In other words, the difference between
FIFTHS AND THIRDS the chromatic circle and the circle of fifths runs
deeper than the simple graphic conventions men-
There are two well-​known circular representa- tioned previously. The two circles suggest differ-
tions of 12-​note pitch-​class space. One is the ent notions of the distance between two pitch
chromatically arranged clock face from Figure classes: we might say that the chromatic circle
1.2.3; the other is fifths space (fifth), better depicts a melodic relationship (in which sense C
known as the circle of fifths, shown in Figure is close to C♯ but distant from G), while the circle
1.3.1. The same twelve pitch classes form the of fifths expresses tonal relationships (in which
underlying space of both circles, but the arrange- the reverse is the case: the key of C is closer to
ment is different: each picture is characterized by the key of G than to the key of C♯). Indeed, it is
in the study of key relationships that the circle of
fifths is often encountered—​but of course Figure
1.3.1 is an accurate depiction of fifth relationships
between single notes as well.35
Despite the differences between their inter-
val structures, the obvious similarities between
the two circles suggest a close affinity between
the two, and we will confirm in Chapter 7 that
even as interval spaces they are isomorphic. Our
conclusions about chromatic pitch-​class space
and fifths space may be summarized as follows:
(1) they have the same underlying space; but (2)
they impose two different interval structures on
that space; and yet (3) those interval structures
are isomorphic. One other difference between
the two is noteworthy. Chromatic pc space pc is
FIGURE 1.3.1 Enharmonically conformed fifths a subspace of a continuous space cpc, but fifths
space fifth (the chromatic circle of fifths) space has no continuous counterpart: it does not

18 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 1.3.3 Spelled pitch-​class space spc (unconformed fifths space, or the line of fifths)

make sense to speak of notes lying between C to chromatic intervals. In spelling pitches we
and G on the circle of fifths. often prefer spellings that keep nearby events
As we have seen, one way to construct the close together in fifths space; this is clearly the
chromatic pitch-​class space of Figure 1.2.3 is to case in tonal music (for instance, in the vicinity
conjoin every twelfth note in the chromatic pitch of the notes E and B, the spelling C♯ is likely to be
space of 1.1.1. The circle of fifths may be simi- preferred to D♭) and perhaps holds in atonal con-
larly obtained by identifying every twelfth note texts as well (twelve-​tone rows are often spelled
in a linear version of fifths space, the line of fifths using twelve consecutive notes of fifths space).36
shown in Figure 1.3.3. The elements depicted in The seven notes of any diatonic scale are clus-
this figure are precisely the same spelled pitch tered together in fifths space, and in tonal music
classes present in Figure 1.2.11, and we take the the distribution along the line of fifths of the
line of fifths as our standard representation of notes in a musical passage provides many clues
spelled pitch-​class space (spc). This space orga- to the tonality.37
nizes all standard note names (without register While the process of constructing fifth
designations) into a single fifth-​generated line, (Figure 1.3.1) from spc (1.3.3) is closely analo-
“sharper” notes appearing in one direction, “flat- gous to the construction of pc (1.2.3) from
ter” in the other. Notes separated by twelve posi- pitch (1.1.1), two different musical phenom-
tions on the line of fifths, such as C and B♯, are ena are at work in these constructions. The ele-
enharmonically equivalent, so imposing enhar- ments of pitch are pitches including register
monic equivalence upon fifths space amounts designations, and pc converts these pitches to
to wrapping this line into a circle, producing the pitch classes by identifying octave-​equivalent
usual circle of fifths. The process of rolling up a pitches such as C4 and C5. The elements of spc
space in order to identify enharmonically equiv- are not pitches but signed letter names; fifth
alent points is sometimes referred to as con- is obtained by identifying enharmonically equiva-
formance; thus the circle of fifths may be called lent note names such as F♯ and G♭. Enharmonic
(enharmonically) conformed fifths space, while the distinctions do not come into play in pitch and
line of fifths is unconformed fifths space. We will pc, where spelling is not a concern, and octave
encounter more examples of conformed and differences are likewise irrelevant in spc and
unconformed spaces in Section 1.4. Meanwhile, fifth, because notes in these spaces contain no
another kind of equivalence is also apparent in register information. Octave equivalence and
the line of fifths: notes separated by seven posi- enharmonic equivalence are two familiar musi-
tions, such as C and C♯, share the same letter cal examples of equivalence relations, a topic to be
name, differing only in their accidentals, and are examined more thoroughly in Chapter 2.
therefore generically equivalent. It may seem surprising that the elements
Fifths space suggests its own notion of the of spc (Figure 1.3.3) are the same spelled pitch
distance between notes, very different from dis- classes (signed letter names) as in sl (1.2.11).
tance in pitch space but related in revealing ways Figure 1.2.11, in any of its three versions, spans
to familiar aspects of intervals. Adjacent notes a two-​dimensional surface, but 1.3.3 somehow
on the line of fifths (notes at a distance of 1) are manages to fit all the same notes into a single
separated by perfect intervals (fourths or fifths). dimension. Each version of Figure 1.2.11 con-
Notes at distances of 2, 3, 4, or 5 form major sists of seven different infinite columns of notes
or minor intervals, while a distance of 6 always (seven spokes in the case of part (b)), while 1.3.3
represents an augmented fourth or diminished comprises but a single infinite row, so it may
fifth, and distances of 7 and above correspond seem intuitively that sl should have seven times

Pitch and Pitch-Class Spaces • 19

as many elements. This is another indication In the circle of fifths, any seven adjacent notes
that infinite sets do not always behave like finite form a diatonic collection, but different diatonic
sets. A finite set is always measurably larger than collections may require different spellings for
any of its subsets. An infinite set, in contrast, some pitch classes. In the line of fifths, in contrast,
may be “the same size” as one of its subsets, in any seven-​note segment forms a properly spelled
the sense that their elements may be placed in diatonic collection. The white-​note diatonic col-
one-​to-​one correspondence. In this sense, lin- lection, called dpc(0) in the terminology intro-
ear fifths space is the same size as one column duced in Section 1.2, is centered on D, spc 0, and
of Figure 1.2.11a—​and simultaneously the same more generally the segment corresponding to the
size as all of Figure 1.2.11a. diatonic collection dpc(n) always extends from
We will begin exploring some of the mathe- spc n − 3 to spc n +​3, centered on spc n. The note
matical relationships among chromatic, generic, spc n represents the tonic of the dorian mode of
and spelled pitch-​class spaces (pc, gpc, and spc) the collection dpc(n) (dorian being the only dia-
in Chapter 2, and in Chapter 13 we will find tonic mode that is symmetrical about its tonic);
these relationships to be of considerable impor- the tonic of the corresponding major scale is spc
tance in comparing chromatic and diatonic n − 2, two places to the left.
structure. The spc numbers shown in Figure 1.3.3
will be useful in these applications. While we Exercise 1.3.6
have followed the usual convention of label- (a) Determine the complete correspondence
ing the note C with the number 0 in pc space, between the seven numbers from n − 3 to
in spc space there are advantages to identifying n +​3 and the modal tonics of the diatonic
0 with D instead: D is centrally located among collection dpc(n).
the seven unsigned notes (white keys) on the (b) Theoretically, what is the tonic of the major
line of fifths, so taking this note as the origin scale with 101 sharps in its key signature?
for spc numbers results in a system that exactly Observe that 101 =​14 ∙ 7 +​3 and also
balances sharps and flats. Spc numbers are 101 =​8 ∙ 12 +​5; how does your answer
theoretically unbounded, with multiple sharps   reflect these properties?
associated with large positive spc numbers and
multiple flats with large negative spc numbers, Exercise 1.3.7
but traditional note names involving no more (a) A signed scale degree (ssd) is a scale degree
than a double sharp or flat have spc numbers number (1̂ through 7̂) with an accidental
restricted to the range from −17 to 17. attached; for example, ♭6̂, ♮1̂, and ♯ ♯4̂ are
signed scale degrees.38 Construct pictures
Exercise 1.3.4 Use your theoretical of signed scale-​degree space (ssd space)
understanding of the line of fifths to answer analogous to the three arrangements of
the following questions without actually signed letter space in Figure 1.2.11.
counting all the intermediate notes: (b) Also construct a fifth-​generated picture
of ssd space analogous to Figure 1.3.3.
(a) In Figure 1.3.3, what note falls seven places Assume that the seven ♮ ssds are those
to the right of G♭? occurring in a major scale. Ssds ♮7̂ and ♯4̂
(b) What note falls fourteen places to the should be adjacent in this arrangement.
right of G♭? (c) Assign signed scale-​degree numbers to the
(c) What note (not visible in Figure 1.3.3) falls ssds in your picture from (b). If you assign
seven places to the left of G♭? the ssd number 0 to the ssd ♮2̂, you should
(d) What note falls twelve places to the find that the seven ♮ ssds are those with
right of G♭? numbers from −3 to 3. If you align your
(e) What note theoretically falls twelve places picture with spc space (Figure 1.3.3) so
to the left of G♭? that corresponding numbers line up, every
(f) What note has the spc number 17? spelled pitch class should align with its
(g)  What is the spc number of the note F♭ ♭? corresponding scale-​degree interpretation
in the C-​major scale; for example, spc A♭
Exercise 1.3.5 Use the line of fifths to should align with ssd ♭6̂.
explain informally why the Neapolitan triad ♭II (d) By shifting your picture of ssd space to the
occurs more frequently in minor keys than in left or right with respect to spc space, any
major
   keys. note can be made to line up with ssd ♮1̂,

20 • E x p l o r i n g M u s i c a l S p a ces
 and all other spcs should then align with Exercise 1.3.9
their corresponding scale-​degree numbers (a) Construct diagrams for the diatonic fifths
relative to that tonic. For example, the spaces corresponding to the keys of F♯ major
picture should show that relative to the and G♭ major. Construct both diagrams by
tonic E♭, ♭6̂ is the note C♭. Can you find a adding accidentals to Figure 1.3.8, keeping
simple formula relating the spc and ssd the letter name C at the top of the circle in
numbers? That is, if spc m corresponds to each case.
ssd k when the tonic (♮1̂) is spc n, can you (b) Describe the numerical relationships that
  find a formula that relates m, n, and k?39 hold between adjacent numbers in generic
fifths space, and compare the results with
Each seven-​ note diatonic segment of the   Exercise 1.3.2a.
line of fifths includes each of the seven letter
names exactly once. Rather than identifying Exercise 1.3.10 Construct a spiral of fifths
every twelfth note via enharmonic equivalence by bending the line of fifths into a spiral
as before, we may instead choose to identify resembling Figure 1.2.1. There are at least two
every seventh note, bringing together the generi- musically suggestive ways to do this: either
cally equivalent notes that share the same letter every seventh note or every twelfth note can
name. The result is generic fifths space gfifth, a align in successive coils of the spiral. When the
fifths-​generated arrangement of generic pitch-​ spirals are collapsed into circles, which figures
class space, shown in Figure 1.3.8. In principle,    from the two alternatives?40
result
a family of diatonic fifths spaces (diatonic circles
of fifths), analogous to the diatonic pitch-​class Exercise 1.3.11
spaces of Figure 1.2.9, could be obtained by (a) Write the following progression of
adding accidentals to Figure 1.3.8 appropriate root-​position diatonic seventh chords in
to particular diatonic scales; a fifth-​generated D major:
arrangement of scale-​degree space could also be
constructed. In any diatonic fifths space, one of      I7 –​IV7 –​viiø7 –​iii7 –​vi7 –​ii7 –​V7 –​I7
the fifths is not a perfect fifth but a diminished
fifth. Making that distinction, however, requires Use four-​part harmony and good voice
recognition of how the notes are situated in chro- leading. (Every other seventh chord may
matic space. Reckoned generically, all fifths are be incomplete, and every chordal seventh
the same, just as all generic steps are the same should resolve down by step.) Write a
in generic pitch-​class space. Many readers will key signature, so that no accidentals
recognize that although the term circle of fifths is are needed. Observe that the bass line
most commonly taken to refer to the chromatic circumnavigates two-​sharp diatonic
picture of Figure 1.3.1, sequential motion by fifths space, and note the location of the
fifth in tonal music more often follows a diatonic diminished fifth (or augmented fourth).
circle of fifths, a solitary diminished fifth taking Each of the other voices traces a path
its place among the perfect fifths. through which other space?
(b) Write a chromatic descending-​fifths
progression of eight dominant seventh
chords, beginning D7 –​G7 –​C7 –​···. The
bass line now should move exclusively by
perfect intervals. Use accidentals, not a key
signature. You should be able to do this in
such a way that your progressions in (a) and
(b) are generically indistinguishable—​that
is, if the key signature in (a) and accidentals
in (b) are disregarded, effectively allowing
us to view both progressions in generic
space, the two appear completely identical.
The voices in the chromatic progression
FIGURE 1.3.8 Generic fifths space gfifth (the trace paths through which space(s)? What is
generic circle of fifths)   the eighth chord?

Pitch and Pitch-Class Spaces • 21

 Exercise 1.3.13 In contrast to the three
circular representations of generic pitch-​class
space, we have seen only two fundamentally
different circular pictures of chromatic pitch-​
class space: pc (Figure 1.2.3), generated by
semitones or major sevenths, and fifth (1.3.1),
generated by perfect fourths or fifths. In
chromatic space there are many other intervals
from which to choose, but no other interval
generates a picture of the entire space. Why
not? Why is there no such thing as a chromatic
circle of thirds? What properties of the numbers
FIGURE 1.3.12 Generic thirds space gthird 7 and 12 ensure that every generic interval
(the generic circle of thirds) (except the unison) generates generic pitch-​
class space, but not every chromatic interval
generates
   chromatic pitch-​class space?42
Figure 1.3.12 shows yet another map of
generic pitch-​class space: generic thirds space
(gthird) or the generic circle of thirds. This 1.4 TONNETZ SPACES
third-​generated space is relatively unappreci-
ated, certainly less celebrated than the step-​ Chromatic and diatonic pitch-​class spaces are
and fifth-​generated spaces gpc and gfifth. small sets, so it should come as no surprise that
Sequences ascending or (especially) descend- there are few ways to generate these spaces by
ing diatonically by thirds are common, and a single interval. The spaces may, however, be
their progress may be traced on this circle. represented in other ways. In particular, an
Another aspect of its significance lies in the assortment of diagrams commonly known by
observation that the most common chords in a the German name Tonnetz (“tone network”) have
diatonic context—​triads and seventh chords—​ proven extremely useful in depicting various
comprise consecutive notes from the circle of kinds of pitch-​class relationships. A tonnetz is
thirds. This circle therefore suggests a harmonic a two-​dimensional array in which two different
interpretation of the distance between two intervals are used as generators, as illustrated in
notes, different from both the melodic under- Figure 1.4.1.43
standing of distance implicit in the stepwise Starting from C in the center of this figure,
space and the tonal distance of fifths space. The other notes are arranged by minor thirds in the
notes A and C, adjacent in thirds space, are har- horizontal direction and major thirds vertically,
monically close in that they appear together in
many of the same chords (triads on F and A,
and seventh chords on D, F, and A), and in that
chords rooted on A and on C will share multiple
common pitch classes. We shall see in Section
4.3 that a simple principle of functional har-
mony may be conveniently displayed in the
generic circle of thirds.41
Taken together, the three circular arrange-
ments of generic pitch-​ class space account
for every possible interval generator: seconds
or sevenths in gpc (Figure 1.2.8), fourths or
fifths in gfifth (1.3.8), and thirds or sixths in
gthird (1.3.12). Any two generic pitch classes,
that is, are adjacent on exactly one of the three
circles. These three diagrams (and rotations
and reflections thereof) are the only possible
ways to represent generic or diatonic pitch-​
class space by means of a single generating FIGURE 1.4.1 A tonnetz generated by major
interval. and minor thirds

22 • E x p l o r i n g M u s i c a l S p a ces
as indicated by the solid horizontal and vertical pairs of enharmonically equivalent spellings
lines. As a byproduct of this arrangement, other are shown. If the diagram were expanded in any
intervals appear systematically at various other direction, more such equivalents would appear,
angles throughout the diagram. A northeasterly quickly extending into the confusing realm of
move corresponds to an ascending perfect triple sharps and flats.
fifth—​the sum of a major third and a minor Adopting enharmonic equivalence results in
third. A northwesterly move is a semitone, the much-​simplified picture of Figure 1.4.2, an
the combination of an ascending major third enharmonically conformed tonnetz, in which an
and a descending minor third. Dashed lines in apparently larger tract of the space may be read-
the figure show that still other intervals occur ily viewed.44 The word apparently is significant,
on secondary diagonals corresponding to for the space of Figure 1.4.2 is not actually large
knights’ moves in chess. The lines in the dia- at all: the notes here are ordinary pitch classes,
gram converge on the central C, but the same of which there are only twelve. The figure is
pattern of lines could be shifted to originate divided by dashed lines into three-​by-​four rect-
from any other note, as the interval relation- angular modules, all of them identical and each
ships are consistent throughout the figure. In encompassing all twelve notes. The modules may
fact, for any given configuration of notes in the appropriately be called tiles because of the way
diagram, a geometric translation in any direc- they fit together to fill the plane. One such tile is
tion—​sliding the configuration as a group right, shown in Figure 1.4.3.
left, up, or down—​corresponds to the musical In principle, one tile suffices for the entire
operation of transposition. For example, starting tonnetz, with the understanding that whenever
on C, a move traces a C-​major triad, C–​E–​G. one reaches a boundary (by, for example, mov-
If the same shape is translated one space to the ing to the right from E♭), one re-​emerges at the
right, an E♭-​major triad E♭–​G–​B♭ is the result, opposite side of the tile (at F♯). The arrowheads
and translation to any other starting note will on the boundaries of the tile are a standard
produce another major triad. mathematical convention for conveying this
The notes appearing in Figure 1.4.1 are spelled understanding: the single arrowheads show
pitch classes (signed letters); enharmonic equiv- that the top and bottom edges of the tile are
alence is not assumed. Each minor-​third-​gener- conjoined, and the double arrowheads show
ated row in the tonnetz comprises the notes of that the left and right edges are conjoined. These
a diminished seventh chord, but two different edges serve as boundaries of the figure but are
spellings appear at the outer ends of the row for not boundaries at all in a topological sense; they
one of the notes of that chord. Likewise, each may be crossed freely and have no musical sig-
column is an augmented triad, in which two nificance. In fact, it would be a simple matter to

FIGURE 1.4.2 Tonnetz space (tonn): an enharmonically conformed tonnetz

Pitch and Pitch-Class Spaces • 23

FIGURE 1.4.3 One tile of the conformed tonnetz FIGURE 1.4.4 A doughnut-​shaped tonnetz

construct the tonnetz in such a way that there vertical and horizontal dimensions. This pro-
were no tile boundaries at these locations at all cess is illustrated in Figure 1.4.4, which we may
(imagine shifting all the dashed lines in Figure imagine to be formed by stretching and bending
1.4.2 one position up and to the right, for exam- a highly flexible copy of the rectangle of 1.4.3.
ple). We refer to such edges as false boundaries; First the top and bottom boundaries are glued
we will see many examples of false boundaries in together to form a cylinder; then the ends of the
our study of voice-​leading spaces in Part Three. cylinder are joined to form the final shape. In
Borrowing metaphors from physics or sci- Figure 1.4.4 the edges have been left unattached
ence fiction, we may imagine that in moving to clarify the construction; the gaps visible here,
from E♭ to F♯ across the false boundary we “tele- corresponding to 1.4.3’s false boundaries, van-
port” or pass through a “wormhole” from one ish when the gluing is complete. Another way to
side of the figure to the other—​but such imag- construct this surface is to sweep a small circle
ery, while possibly helpful in visualizing a path around a path that traces out a larger circle; this
in the tonnetz, is misleading in its suggestion construction suggests that a torus is a sort of
of preternatural behavior. Movement across a “product” of two circles, a notion that can be
false boundary is musically no different from made topologically precise.45
movement anywhere else in the space, and there This doughnut tonnetz is elegant, but several
should be no more mystery in this aspect of the caveats are in order. First, the torus is the surface
tonnetz than in the observation that round-​the-​ of the doughnut, not a solid body. In the music
world travelers, somewhere in the middle of the theory literature one can find the torus described
Pacific Ocean, suddenly find themselves on the as three-​or even four-​dimensional, but topologi-
opposite side of their map. cally, a torus is a two-​dimensional manifold, just
The infinite plane of Figure 1.4.2 is redun- as Figures 1.4.2 and 1.4.3 are two-​dimensional.
dant in the same way that the infinite line of (Structures analogous to tori in other dimensions
1.2.7a is redundant, but in two dimensions can be constructed, and will prove useful in Part
rather than one; the tile of Figure 1.4.3 stands Three; when it is necessary to distinguish among
in the same relation to the infinite plane as the them, the torus described here may be called 𝕋2,
modular line segment of 1.2.7b to the infinite the 2-​torus.46) Second, the twelve-​note musical
line. Figure 1.4.3, like 1.2.7b, may be called a space being depicted is only a discrete subspace
fundamental region for the space it represents. of the torus, not the entire continuous surface.
Figure 1.2.7a was described earlier as periodic; Finally, the geometry of the doughnut shape, like
Figure 1.4.2 may be called doubly periodic in light the geometry of the circle, is extraneous and
of the repeating patterns that appear both hori- potentially misleading as far as the structure of
zontally and vertically. the musical space is concerned. Indeed, it could
Figures 1.2.7a and 1.2.7b were alternate rep- even be described as a distortion of that space.
resentations of the pitch-​class circle from Figure Like the circle, the doughnut has its own curva-
1.2.3. What, then, is the tonnetz equivalent of ture, inside, and outside—​all of them more com-
that circle? The appropriate topological con- plicated than in the case of the circle, and musically
struction is known as a torus (with plural tori and meaningless. In Figure 1.4.4 the circles depicting
adjective toroidal). The torus is often described augmented triads appear to be much smaller in
as the surface of a doughnut shape, obtained by diameter than those depicting diminished seventh
wrapping a plane around on itself in both the chords, and in fact the three diminished seventh

24 • E x p l o r i n g M u s i c a l S p a ces
 expand our idea of what the word tonnetz can
mean) and Chapter 8.
The diagonal lines representing perfect fifths
slope from southwest to northeast . The other
diagonals , which are not drawn in the tri-
angulated tonnetz of Figure 1.4.5, join pitch
classes related by semitone. Semitones and per-
fect fifths, it will be recalled, are the generating
intervals of the two circular maps of pitch-​class
space, the spaces we have called pc (Figure 1.2.3)
and fifth (1.3.1). Both of these circles, there-
fore, may be said to be present in the tonnetz,
lurking in its diagonals. Moving twelve places in
one direction diagonally is equivalent to travel-
ing once around the corresponding circle.
FIGURE 1.4.5 A triangulated tonnetz (the neo-​
Riemannian tonnetz) Exercise 1.4.6
(a) In the tonnetz of Figure 1.4.2, what does a
whole-​tone scale look like? You should find
chords are depicted by circles that differ slightly in that a whole-​tone scale may be represented
size—​but those circumstances are also mere arti- by a straight line in either of two different
facts of the construction.47 Both the topology of orientations, or possibly by a zigzag line
the torus and the musical structure of the tonnetz combining segments from both of the
are perfectly well represented by a “flat torus” tak- others. How do these varied depictions
ing the form of either the infinitely repeating grid of one scale in the conformed tonnetz
of Figure 1.4.2 or the single tile of 1.4.3, and as a differ from each other in the unconformed
practical matter both of those representations are diagram of Figure 1.4.1? As subsets of
much simpler to draw, to visualize, and to work pitch-​class space, there are two different
with than the doughnut shape.48 whole-​tone collections. What do these
When we speak of “the tonnetz” or tonnetz collections look like in the tile of Figure
space (tonn), therefore, it is generally some- 1.4.3? On the doughnut of 1.4.4?
thing like Figure 1.4.2 or 1.4.3 that we have in (b) What does an octatonic scale look like in
mind—​or Figure 1.4.5, in which the tonnetz Figure 1.4.2? What do the three octatonic
from 1.4.2 has been triangulated, or divided   collections look like in 1.4.3 or 1.4.4?
into triangles, by the inclusion of horizontal,
vertical, and diagonal line segments connect-
ing notes related by minor thirds, major thirds, Exercise 1.4.7
and perfect fifths, respectively. The elements of (a) Any augmented triad and any diminished
tonnetz space are pitch classes, and the trian- seventh chord always share exactly one
gulation is appealing because each triangle com- common pitch class. Use Figure 1.4.3 to
prises the pitch classes of a familiar consonant explain why this is so.49
triad. Triangles of one orientation represent (b) What can be said about the number of
major triads such as C–​E–​G, while triangles of common tones between a whole-​tone
the opposite orientation are minor triads collection and an octatonic collection?
such as C–​E♭–​G. In both cases the root of the What sort of pitch-​class set is formed by
triad lies at the lower left corner of the triangle these common tones?
and the fifth at the upper right. As pitch-​class (c) The pitch classes in any two adjacent
sets, major and minor triads are related to each columns of Figure 1.4.3 combine to form a
other by inversion and share the prime form hexatonic collection, a set with prime form
037, and all 24 sets of that prime form appear 014589. How many different hexatonic
in the tonnetz. The capability of the tonnetz to collections are there? What can be said
depict triadic relationships has led to its wide- about the number of common tones
spread use in neo-​Riemannian theory, to be dis- between a hexatonic and an octatonic
cussed in Chapters 3–​4 (where we will further collection, and about the type of pitch-​class

Pitch and Pitch-Class Spaces • 25

 set they form? What about common tones Even more than was the case with the pre-
between a hexatonic and a whole-​tone vious diagrams, drawing a tonnetz requires
collection? Between a hexatonic collection making a number of decisions about graphic
  and a diminished seventh chord?50 conventions. Many different versions of the
tonnetz appear in the literature, differing
Exercise 1.4.8 What is the most compact among themselves in a number of more or less
possible tonnetz representation for a dominant arbitrary ways.51 Figure 1.4.10 presents three
seventh chord? A half-​diminished seventh versions with slightly varied geometry. The
chord? A minor seventh chord? Dominant and form in (a) is obtained from 1.4.5 by a subtle
half-​diminished seventh chords are related to adjustment of the spacing between rows, so
each other by inversion, while minor seventh that the triangles, rather than having two
chords are inversionally symmetrical. How are equal legs, have sides in the proportion 3:4:5,
these
   observations visible in the tonnetz? corresponding to the interval classes formed
by the notes in each triad.52 In Figure 1.4.10b
Exercise 1.4.9 Suppose we triangulate the the triangles are equilateral. This figure may
tonnetz of Figure 1.4.2 in a different way, be obtained from Figure 1.4.5 (or 1.4.10a) by
to include the diagonals rather than a shearing transformation, moving each row
(along with the verticals and horizontals). rightward relative to the row above (so that,
Triangles in this new configuration no longer for example, C falls below a point midway
represent major and minor triads. As pitch-​ between E and G rather than aligning directly
class sets, what is their prime form? Do the beneath E), then adjusting the row spacing.
two orientations of triangles represent The shearing may be taken further, so that C
inversionally
   related sets? aligns below G, as in 1.4.10c. The original rec-
tilinear configuration and right triangles are
recovered, but the diagonals point in the oppo-
site direction from before; major thirds now
appear diagonally and perfect fifths vertically
rather than the other way around.
In Figure 1.4.10c it may appear that the “gen-
erating intervals” are minor thirds and perfect
fifths (interval classes 3 and 5), rather than
major and minor thirds (ic 3 and 4) as in the
earlier figures. This is a distinction of little struc-
tural import, for the lines connect the same pitch
classes as before, and the triangles comprise the
same three intervals (ic 3, 4, and 5). Any two of
these intervals may be taken as “generators” and
may be oriented along any two non-​parallel axes;
it will then always be possible to form a triangu-
lation in which the third interval appears on the
third side of each triangle, as the sum or differ-
ence of the first two intervals. All tonnetzes con-
structed in this way are structurally equivalent,
in the sense that any of them may be obtained
from Figure 1.4.5 by some combination of rota-
tion, reflection, and shearing. Indeed, while our
usual representations of tonnetz space tonn
will be the rectilinear arrangements of Figures
1.4.2, 1.4.3, and 1.4.5, the relationships shown
in all of the various pictures in Figures 1.4.2–​
1.4.10 are sufficiently similar that we may regard
all of these figures as alternate depictions of this
FIGURE 1.4.10 Three different tonnetz same space. Exercise 1.4.9, however, showed
geometries that two tonnetzes generated by ic 3 and 4 may

26 • E x p l o r i n g M u s i c a l S p a ces
convey different relationships entirely if they are The tonnetz has a long and distinguished
triangulated differently. When identifying trian- history, which may be traced to the illustri-
gulated tonnetzes, therefore, a name based on ous eighteenth-​ century Swiss mathematician
two generators such as “ic 3-​4 tonnetz” is poten- Leonhard Euler.55 It was particularly favored by,
tially misleading; it is preferable to name all and has become closely associated with, German
three intervals explicitly (“3-​4-​5 tonnetz”), or to theorists of later generations. Its relationship to
identify the set class of the represented chords Gottfried Weber’s “table of tonal relations” will
(“037 tonnetz”).53 be examined in Chapter 4. Illustrations more
closely resembling the figures in this chapter
Exercise 1.4.11 appeared in works of Arthur von Oettingen and
(a) Figure 1.4.12 shows a tonnetz appearing Hugo Riemann. Oettingen’s and Riemann’s ton-
in an influential article by Gerald Balzano netzes are based on the same essential organi-
(1980, 73). Apart from the substitution of zational principles; recognizing celebrity over
pitch-​class numbers for note names, how is historical priority, we will follow other authors
this figure related to Figure 1.4.5? Can one in calling this figure the Riemannian tonnetz.56
be obtained from the other by rotation, or is Figure 1.4.14 shows the tonnetz as Riemann
a reflection required? constructed it. This orientation, with major and
(b) The lattice of triangles in Figure 1.4.12 minor thirds deployed on two diagonals and
connects the notes of the C-​major scale. fifths horizontally, approximates a 60-​ degree
There are seven triads in a diatonic scale, clockwise rotation of 1.4.10b. But there is a more
but only six triangles are visible in the substantive difference, consideration of which
  figure. Where is the missing triad? demands that we again confront matters of tun-
ing. Oettingen and Riemann supposed their dia-
Exercise 1.4.13 Viewed as a subset of grams to be generated by purely tuned intervals.
chromatic space, a diatonic collection is This assumption requires distinguishing not only
symmetrical about its dorian tonic (D in the between different spellings of enharmonically
case of the white-​note collection). Consider the equivalent notes, but even between some notes
appearance of a diatonic collection in each of sharing the same signed letter name. Working
the figures listed below and discuss how this from C in the central row of Figure 1.4.14, for
symmetry is apparent.54 example, we can move four places to the right—​
ascending through four perfect fifths—​ and
(a) Chromatic pitch-​class space pc arrive at E. There is another E, however, immedi-
(Figure 1.2.3) ately above and to the right of the original C, in
(b) The circle of fifths fifth (Figure 1.3.1) what Riemann called the “1st row of upper 3rds”
(c) The line of fifths spc (Figure 1.3.3) (Oberterzen). This E is not four fifths above C,
(d) Signed scale-​degree space (from but a pure major third above C, and allowing for
Exercise 1.3.7) octave adjustments, these two pitches are not
(e)  Balzano’s tonnetz (Figure 1.4.12) the same. The third-​generated E is lower than
the fifth-​generated E by a syntonic comma, a fre-
quency ratio of 81/​80.57 Oettingen and Riemann
indicated comma differences by underscores and
overscores; the underscored E is one syntonic
comma lower than the undecorated E. If the
“2nd row of upper 3rds” were extended one place
to the left, still another E would appear; this E
would be another comma lower, and like all its
rowmates it would bear a double underscore. The
Riemannian tonnetz is theoretically infinite, not
periodic in any direction; no two of its notes are
tuned identically, and considering the under-​
and overscores, no two are labeled identically.
Modern neo-​Riemannian theorists have gener-
FIGURE 1.4.12 A numerical tonnetz (after ally disregarded these tuning differences, result-
Balzano) ing in a distinction between the unconformed

Pitch and Pitch-Class Spaces • 27

FIGURE 1.4.14 The Riemannian tonnetz

Riemannian tonnetz of Figure 1.4.14 and the con- therefore triads—​ trichords of prime form
formed neo-​Riemannian tonnetz of 1.4.5, which 037. It is possible to construct tonnetz-​like
assumes enharmonic equivalence and includes diagrams using different intervals entirely.61
only twelve different notes.58 One example was suggested in Exercise 1.4.9.
For another example, one can construct a ton-
Exercise 1.4.15 Figure 1.4.2, like 1.4.5, netz whose axes join pitch classes separated by
provides only twelve pitch classes and does intervals of 1, 2, and 3 semitones (one of the
not capture the pure-​interval conception intervals must be the sum of the other two),
of Oettingen and Riemann. Explain why and whose triangles therefore represent 013
Figure 1.4.1, even though it does account trichords. David Lewin (1998) uses such a dia-
for enharmonic spelling distinctions, does gram in an analysis of the opening measures
not capture their conception either. While of the F♯-​minor fugue from Book I of Bach’s
Figure 1.4.14 is an unconformed tonnetz, 1.4.1 Well-​Tempered Clavier, whose subject strings
may be called a “partially conformed tonnetz.” together several 013 trichords.62
In what way is it conformed, and in what way
is it not? Topologically, Figure 1.4.14 is a plane Exercise 1.4.17 Draw a large portion of an 013
and 1.4.2 is a torus. What topology is implied tonnetz as described above, assuming equal
by  Figure 1.4.1?59 temperament. Can you find a repeating tile
containing
   all twelve pitch classes?63
Exercise 1.4.16 Consider the chord
progression I–​IV–​ii–​V–​I in G major. Plot The tonnetz in Figure 1.4.18, generated verti-
the movement of the chord roots in this cally by semitones and horizontally by tritones,
progression in the Riemannian tonnetz of has some interesting topological properties. The
Figure 1.4.14, assuming that roots of successive repeating six-​by-​two tiles are marked by dashed
chords are always related by pure intervals. lines in part (a), and one copy of this tile is
If you do this correctly, the progression will shown in (b). As before, the edges of the tiles are
end at a different G from the initial one. Is only false boundaries. These tiles, however, are
the final pitch higher or lower? Discuss the arranged in a distinctive way: horizontally adja-
possible implications of this phenomenon for cent tiles appear identical, but vertically adjacent
performance
   by an a cappella choir.60 tiles are left-​right mirror images of each other.
The arrowhead convention in 1.4.18b conveys
The tonnetzes we have considered spotlight this reversal: the two vertical double arrowheads
major and minor thirds and perfect fifths, and point in the same direction, indicating that

28 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 1.4.18 A tonnetz generated by semitones and tritones

these two edges are conjoined in the usual way, a single edge of the Möbius strip, which cir-
but the two horizontal arrowheads and are cumnavigates the loop twice as it traverses the
oppositely oriented, indicating that those two entire chromatic scale.64
boundaries are understood to be glued together
in reverse, so that for example the note above F Exercise 1.4.19 The tonnetz in Figure 1.4.18
is F♯, not C. To realize this gluing physically, we is not triangulated. It could be triangulated
would have to give the rectangular strip a half-​ in either of two ways, using either or
twist before gluing its ends together, resulting diagonals. Verify that the same three-​note
in the famous figure known as a Möbius strip, as pitch-​class sets arise either way. Recall from
suggested in 1.4.18c. Exercise 1.4.9 that these two triangulations
The musical property that necessitates the gave different results for the usual tonnetz
twist is the fact that when a tritone is transposed generated by major and minor thirds. What
by tritone, the original tritone returns with its special property of the semitone, the tritone,
two pitch classes interchanged. One could say, or the combination of the two is responsible for
though, that the image of the twisted strip, the  invariance of the trichords in this case?65
like the curvature of the circle of pitch-​class
space, is essentially fictitious, merely an aid to Example 1.4.20: Lutosławski, Musique
visualizing the musical space in a physical way; funèbre The opening measures of the
the twist is what we must do in order to con- Prologue to Witold Lutosławski’s Musique
struct a space of the proper topology within our funèbre (1958) offer a simple illustration
familiar Euclidean world. The Möbius strip itself of the application of a particular graphic
does not accomplish the identification of the representation of pitch-​class space in mapping
two vertical boundaries of Figure 1.4.18b—​it musical motion, and also of the benefits
implies end-​to-​end tiling, not side-​to-​side—​but of modifying a general diagram to suit the
as far as the discrete musical space is concerned, features of a specific example.66 The score of
side-​to-​side identification is unnecessary any- mm. 1–​5 is presented in Figure 1.4.21. The
how. Gluing those boundaries together would, two solo cellos play in imitation at the tritone.
for example, allow us to move rightward from B♭ Within each part, motion is exclusively by
to E—​but there is no need to do so, because we tritone and semitone. In the tonnetz tile of
can already accomplish the same move leftward Figure 1.4.18b, cello I begins at F at the upper
from B♭ to E within the strip. The two original left corner, then moves right to B, down to
vertical boundaries, in fact, are subsumed in B♭, left to E, down to E♭, and so on, tracing

Pitch and Pitch-Class Spaces • 29

FIGURE 1.4.21 Lutosławski, Musique funèbre, Prologue, mm. 1–​5

a methodical pattern . Cello II follows the

same path in mirror image , beginning B–​F –​
E–​B♭–​A. But because leftward and rightward
motion from any point amount to the same
thing in this space, cello II’s path may also be
represented in Figure 1.4.18a by a path of the
same shape as the original, not reflected but
translated to a different starting point. The
semitone motion is at first strictly descending,
so both voices work their way down the
diagram. Tritones, in contrast, may either
ascend or descend; the left-​right dimension in FIGURE 1.4.22 Ladder graph for analysis of
the figure does not correlate with register. Musique funèbre
By m. 4, both voices have traversed all twelve
pitch classes and reached the bottom row of
the tile. At m. 5 cello I moves from C to B, and reversing direction whenever—​and only
disappearing off the bottom of the tile and whenever—​the top or bottom rung is reached.
reappearing at the top, on the opposite side; a Each voice engages all twelve pitch classes
beat later cello II likewise moves from G♭ to F. At on its way down the ladder and again, in a
this point, however, the motion reverses. After different order, on the way back up, forming an
playing the B–​F tritone, cello I ascends, back ordered 24-​pc succession containing every pitch
across the seam from F to F♯. Cello II similarly class twice, which repeats in cyclic fashion.
reverses direction, and both voices work their Although this description of motion through
way back up through the picture. the ladder emphasizes pairs of tritone-​related
The B–​F tritone serves as an impenetrable pitch classes, the rhythm and phrase indications
boundary throughout the movement: whenever more often highlight 016 subsets—​the only
it is reached, whether from above or below, set type formed by three-​note segments of the
the next semitone motion is in the opposite pattern or by triangles of either orientation in
direction. This observation suggests that a the ladder and in the corresponding tonnetz
simpler picture may suffice for the analysis of (recall Exercise 1.4.19 above).
Lutosławski’s Prologue. Figure 1.4.22 abandons Because of the short time interval of the
the Möbius strip topology, replacing it with imitation, the two voices play a game of tag,
an uncomplicated ladder consisting of two occasionally merging, always remaining in
vertical rails generated by semitones, and close proximity to one another on the ladder,
seven (not six) horizontal rungs representing and never straying more than one rung apart.
tritones. The B–​F tritone appears twice, as both When they are a rung apart, they are always
the top and the bottom rungs of the ladder. on opposite rails, forming between them the
These two rungs theoretically represent the interval of a perfect fourth or fifth, never a
same region of the space of Figure 1.4.18, but semitone. The only vertical intervals to appear
they are distinct in the customized space of in these opening measures are unisons/​octaves,
1.4.22 because of the composer’s systematic tritones, and perfect fourths/​fifths.
differentiation between B–​F-​as-​top-​boundary Figure 1.4.23 illustrates the beginning of this
and B–​F-​as-​bottom-​boundary. The voices narrative, tracing the ladder positions of the
trace systematic paths up and down the two voices through mm. 1–​2. It is worth taking
ladder, alternately traversing rails and rungs, a moment to consider how the time dimension

30 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 1.4.23 Musique funèbre, mm. 1–​2 in the semitone-​tritone ladder

has been depicted in this figure. Time, like pitch, sometimes two rungs apart—​but initially
forms a space in its own right, in principle a only on the same rail, so major seconds are
one-​dimensional continuum naturally modeled possible here, but major thirds are not available
by the real numbers (a technique familiar to until more violins have entered in m. 22. The
physics students). This space is often called the clustering of voices in the space of Figure
time domain; we will refer to it as temporal space 1.4.22, in other words, remains apparent, but
and label it simply time. Although time is a the clusters gradually expand to fill more of the
continuous space, musical time, like musical space. By the time the second group of double
pitch, is commonly segmented into chunks—​ basses enters at m. 29, all possible intervals in
beats—​making a discrete representation of time pitch-​class space have been realized as vertical
suitable for many purposes. In the Lutosławski simultaneities. Even here, however, all ten
excerpt, motion occurs only on the half-​note voices remain within a span of four adjacent
beats, so this beat space provides the resolution rungs on the ladder (the sounding notes at
necessary to follow the temporal progress of the downbeat of m. 29 are D–​A♭–​E♭–​B♭–​F), so
the piece. The eight ladder snapshots in Figure that for example the major third B♭–​D (two
1.4.23 depict pitches at the first eight points rungs apart) is a possible vertical interval, but
of this half-​note beat space—​or more precisely C–​E (four rungs apart) is not. The exceptional
at eight of the first ten points, as there is no nature of the B–​F tritone is spotlighted in the
motion on the final beat of either m. 1 or m. 2. climactic passage in mm. 30–​32 and two similar
Beat space thus adds a dimension to the pitch-​ sections later, in which all voices converge on
based analysis, operating in conjunction with these
   notes.
the two-​dimensional ladder to produce a three-​
dimensional time-​space narrative.67 Exercise 1.4.24 As an alternative model for
As the piece continues and the polyphonic the analysis of this excerpt from Musique
texture expands, the relationships between funèbre, start with the pitch-​class circle (Figure
voices remain under careful control. The third 1.2.3) and draw six diameters across the circle,
voice to enter is a solo viola, in m. 6. The viola connecting tritone-​related notes. What sort of
follows a course like those of the cellos, and path does each voice follow on this map of the
all three remain within one rung of each other space? Can you modify the construction slightly
on the ladder, but semitones start appearing to account for the special boundary behavior of
vertically, as two voices now sometimes occupy the B–​F tritone? Can you see how this picture
adjacent positions on the same rail. With the can be topologically deformed into the ladder of
entry of violins in mm. 16–​17, voices are now Figure
   1.4.22?

Pitch and Pitch-Class Spaces • 31

Exercise 1.4.25 Return to the alternate a parallelogram is topologically equivalent to
tonnetz triangulation described in Exercise a rectangle—​if an appropriate shearing trans-
1.4.9. Use this figure to analyze overlapping formation were applied to the figure, the tiles
three-​note fragments in mm. 1–​3 and m. 19 of would become rectangles69—​and Figure 1.4.27
“Nacht”
   from Schoenberg’s Pierrot lunaire.68 retains the toroidal topology of the earlier fig-
ures. If the letter names C–​B in this figure are
Exercise 1.4.26 In the well-​known fugal replaced by scale degrees 1̂–​7̂, the resulting pic-
first movement of Bartók’s Music for String ture of scale-​degree space matches that of Fred
Instruments, Percussion, and Celesta, the Lerdahl (2001, 100).
entries of the fugue subject form two Figure 1.4.28 shows two different trian-
concurrent chains. One chain consists of gulations of the generic tonnetz. In part (a),
entries on the notes A–​E–​B–​…, ascending by the triangles form generic triads such as C–​
perfect fifths, while the other chain proceeds E–​G, while in (b), trichords such as C–​E–​F are
A–​D–​G–​…, descending by perfect fifths. formed instead. Judging by the importance
(The initial entry on A inaugurates both of triads in the neo-​Riemannian tonnetz, the
chains.) The two chains converge on E♭ at the reader may suppose that the former arrange-
movement’s climax and on A at its conclusion. ment is more useful, but it is the second ver-
Choose an appropriate figure from this sion to which we shall return more often. The
chapter on which to map the two chains of triangulation in (b) has the attractive property
entries, and construct such a map as a series that it provides simple representations of all
of  
snapshots similar to Figure 1.4.23. generic intervals: in this triangulation, unlike
the other, lines connect C to every other
Tonnetz diagrams are possible for generic generic pitch class. The lines at the three dif-
and diatonic as well as chromatic spaces. Figure ferent slopes, as they wrap around the torus,
1.4.27 illustrates one such construction, a trace the three circular arrangements of
generic tonnetz generated by thirds (horizon- generic pitch-​class space encountered previ-
tally) and fifths (vertically). Like other generic ously in the spaces gpc (Figure 1.2.8, arranged
spaces we have constructed, this generic ton- by step), gfifth (1.3.8, arranged by fourth/​
netz can easily be made into a diatonic ton- fifth), and gthird (1.3.12, arranged by third).
netz by applying appropriate accidentals. The Because of this property, the tonnetz of Figure
tiles shown in the figure are parallelograms. 1.4.28b enables appealing depictions of many
Other configurations of a generic tonnetz and aspects of generic or diatonic structure; we
its modular tiles are possible, subject to the shall revisit it as we investigate chordal and
constraint that a tile must contain each of the tonal spaces in Chapter 4 and generic spaces
seven letter names exactly once. In any case, in Chapter 13.

FIGURE 1.4.27 A generic tonnetz

32 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 1.4.28 Two triangulations of the generic tonnetz

Exercise 1.4.29 of those numbers” (Tufte 2001, 9). Some scientists

(a) The generic tonnetz in Figure 1.4.27 have proposed that map-​making was as important
(without any shearing) cannot be tiled with as language in the evolutionary development of the
identical rectangular modules whose sides human brain and the differentiation of humans from
run vertically and horizontally, meeting at other species (Dawkins 1998, 299). On the other
the corners. How do you know this? hand, the privileged position of visual metaphors
(b) In Figure 1.4.28a, every generic triad in human thought, sometimes known as ocularcen-
such as C–​E–​G appears in triangles of trism, has not gone unchallenged; see, for example,
two different orientations . In the Kavanagh 2004.
alternate triangulation of 1.4.28b, however, 2. For commentary on the inadequacies of the
a trichord such as C–​E–​F appears in a “Euclidean/​Cartesian score-​plane,” see Lewin [1986]
triangle of one orientation but not the 2006b, 81.
other . What accounts for the difference? 3. Pitch numbers are used in music theory less
(c) Considering triangles of both orientations, consistently than the pitch-​class numbers to be intro-
how many different generic triads are duced in Section 1.2. The assignment of the number
represented in Figure 1.4.28a? How many 0 to middle C is, of course, arbitrary. An alternative
different generic trichords are represented to the system shown in Figure 1.1.1, well known in
in 1.4.28b? some circles, is MIDI note-​numbering, which assigns
(d) Suppose the tiles from Figure 1.4.27 are to middle C the number 60.
superimposed upon Figures 1.4.28a and 4. The same convention is routinely observed in
1.4.28b. In (a), how many copies of each physics, when an idealized infinite space-​time contin-
triad are contained in each tile? In (b), how uum is assumed in calculations arising in the study of
many copies of each trichord are contained everyday physical phenomena.
in each tile? 5. From antiquity through the Middle Ages, a
(e) An alternative tiling of Figure 1.4.28a is monochord—​ a single string—​ served as a primary
possible, forming a Möbius strip rather basis for understanding musical pitch and motion.
than a torus (that is, the tiles are connected Some ancient Greek writings on music used string-​
end-​to-​end only, with a twist required as in length–​based models of pitch space comparable to
1.4.18), so that each tile contains just one Figure 1.1.2, but vertically oriented with the lowest
copy of each generic triad. Can you find pitches at the top. See, for example, the Sectio canonis,
  such a tiling?70 attributed inconclusively to Euclid and translated in
Barker 1989, 190–​ 208, especially the diagram on
page 208. Barker (134, note 43) also observes that
NOTES “the metaphors of ‘up’ and ‘down,’ and of movement
1. The benefits of visual representation are not, between high and low, are not built into ordinary
of course, limited to musical phenomena. As Edward Greek in the way that those of tension and relaxation
Tufte has written, “often the most effective way to are.” On the question of whether the traditional use
describe, explore, and summarize a set of num- of the spatial terms up, down, high, and low in refer-
bers—​even a very large set—​is to look at pictures ence to pitch is anything more than metaphorical, see

Pitch and Pitch-Class Spaces • 33

Zbikowski 2002, 65–​77. Cellists, who produce higher (15 semitones) from a major tenth (16), but cannot
pitches by moving their fingers down, may have a dis- distinguish a minor tenth from an augmented ninth
tinctive perspective on the last question. (also 15 semitones). The name “minor tenth” relies
6. Figures 1.1.2 and 1.1.3 are drawn as open half-​ on information from both chromatic and generic
lines, bounded on one end by a point that is not an spaces.
element of the space. But the topology of an open half-​ 14. A particularly ungainly consequence of this
line is the same as the topology of a line unbounded convention is that the unison, which is after all an
at either end. interval between two notes separated by no distance
7. Discrete spaces may be defined in a topologically at all, is somehow considered to have a positive size—​
precise way, based on the idea that a space is discrete as if one were to claim to have walked a mile while
if every point in the space is surrounded by some standing still, or to declare a baby to be a year old on
small region containing no other points of the space. the day she is born. It is perhaps relevant in this con-
Mathematicians do not generally speak of “continuous text that the Greeks, to whom this convention can be
spaces,” reserving the notion of continuity for func- traced, had no concept of the number zero.
tions rather than sets. Our understanding of “con- 15. The awkwardness of the usual interval-​naming
tinuous spaces” will remain fairly informal, though convention has been noted by several previous authors,
it corresponds loosely with the topological concept including Lewin [1987] 2007, 16–​ 17, and Taneyev
of path-​connectedness. A space is path-​connected if any [1909] 1962, 25–​28. Taneyev devotes the first several
two of its points can be joined by a path—​a continuous pages of the first chapter of his counterpoint treatise
mapping from an interval in the real number line into to explaining his numbering based on step-​counting,
the space in question. Continuous pitch space is path-​ which he calls “a more accurate method of indicating
connected, while discrete pitch space is not. There intervals than that in general use” (25). See Segall 2014
exist spaces for which neither discrete nor continuous for discussion of Taneyev’s theory of counterpoint.
is an appropriate description, for instance a space of 16. It is traditional, of course, to write 𝄪 rather than
pitches corresponding to the rational numbers within ♯ ♯ for double sharps. We will use the notation ♯ ♯ when
the real number line. we wish to emphasize the position of double sharps
8. The unnotated inflections associated with the within a systematic larger pattern (a pattern that may
medieval and Renaissance practice of musica ficta offer also include double flats, triple sharps, and so on).
another situation in which one note—​a single ele- 17. Equal temperament and enharmonic equiva-
ment of an abstract space—​may be associated with lence, while they often go hand in hand, are not the
more than one physical pitch. same thing. A form of enharmonic equivalence always
9. See Section 13.4 for more discussion of the holds on the piano—​F♯ and G♭ are always the same
related terms genus, generic, species, and specific. Our pitch—​even though a piano may be tuned in some-
usage of these words can be traced to the first para- thing other than equal temperament. Thus enharmonic
graph of Clough and Myerson 1985, where the authors equivalence does not imply equal temperament. While
explain that “triad” is a generic label for B–​D–​F, while the converse (equal temperament implies enharmonic
“diminished triad” is a specific label for B♮–​D♮–​F♮. Many equivalence) is valid in the acoustical sense, there is
subsequent authors have used the words in similar (if a syntactic sense in which enharmonic equivalence
not always identical) ways. need not be assumed even in equal-​tempered spaces.
10. An example of such a scale would be the Indeed, traditional tonal theory observes enharmonic
scale of the ninth-​century Enchiriadis treatises (see distinctions, as when we say that the leading tone in
Anonymous 1995), notorious for its augmented G major is correctly spelled as F♯, not G♭, even if played
octaves. on an equal-​tempered piano. The conception of tun-
11. This sense of generic equivalence will be invoked ings as mappings facilitates the study of such distinc-
occasionally in this book, for instance in the discus- tions by allowing two different elements of srl to map
sion of the line of fifths in Section 1.3 and in Exercise onto the same point of frequency space.
2.4.5. The interplay between specific and generic 18. Though extremely rare in musical scores, triple
structure will be a primary topic of study in Part Four. sharps and flats are not unknown. Readers who enjoy
12. Similar comments are often made about nam- such things will find triple sharps in Valentin Alkan’s
ing diatonic collections in post-​tonal contexts; see for Etude, Op. 39, No. 10 (1857), m. 291, and Max Reger’s
example Straus 2016, 244–​46, where notations such Sonata for Clarinet and Piano, Op. 49, No. 2 (1900),
as DIA1♭ are introduced. IV, m. 91. Bazayev 2013 notes examples of triple
13. Technically the interval in pitch separating sharps in the music of Nikolai Roslavets, and several
C from E♭ in Figure 1.1.8b is not a “minor tenth” but triple flats occur in Roslavets’s Piano Sonata No. 1
“15.” Intervals in pitch can distinguish a minor tenth (1914), mm. 152–​53. Extreme accidentals have also

34 • E x p l o r i n g M u s i c a l S p a ces
been introduced occasionally in theoretical contexts. the blanks in the following sentence: “Figure 1.2.2 is
For example, Lewin ([1984] 2006, 195–​96) reads a D♭-​ a _​_​_-dimensional
​​ projection of an embedding of
major triad in Parsifal as E♭ ♭ ♭ major, and Keller (1976, a _​_​_​​-dimensional manifold in a _​_​_​-dimensional
79) determines that the modulations in the opening space.”
section of Mozart’s Fantasy in C Minor, K. 475, lead 27. Some world maps show the same regions at
not back to the original C minor but to F♭ ♭ ♭ ♭ ♭ minor. both ends, most often in far eastern Russia and west-
For an extraordinary example see Dumitrescu 2007, ern Alaska; Cohn (2012, 28) accordingly calls the
a reconstruction of a 16th-​century modulating canon inclusion of the same element on both sides of a graph
in whose concluding section two of the voices are “a Bering Strait flaw.” Tufte (2001, 98–​99), however,
notated in a fifteen-​flat (!) key signature (one triple considers the graphical redundancy a virtue and advo-
flat and six double flats). cates the practice.
19. A spiral similar to Figure 1.2.1 appears in 28. Twelve-​and seven-​note pitch-​class spaces have
Westergaard 1975, 13. Helical pitch representations the most obvious musical applications, but math-
such as Figure 1.2.2 are often associated with the work ematically there is no hindrance to constructing a
of Roger Shepard; see, for example, Shepard 1982, 353. pitch-​class space with any desired number of notes.
Precedents for this model, however, can be found as Spaces of other sizes are sometimes considered in the
early as the mid-​nineteenth century (Marvin 1987, 66). study of microtonal pitch systems with more than
20. Musical reasons for choosing the octave as the twelve notes per octave (a microtone is any interval
interval governing the construction of Figures 1.2.1–​ smaller than the usual semitone), a few examples of
1.2.3 are probably self-​evident. As far as the linear which will be studied in Chapter 14.
structure of pitch space is concerned, however, the 29. Fred Lerdahl (2001, 99ff) works extensively
octave enjoys no special privileges. Representations with scale-​degree space, though he never explicitly
analogous to all three of these figures could be con- presents its circular diagram.
structed by aligning pitches separated by perfect fifths 30. Because signed letter space is discrete rather
or any other desired interval. than continuous, it is not quite accurate to say that the
21. Circular pitch diagrams appeared, for example, space itself has the topology of a cylinder; rather, we are
in Lippius [1612] 1977, Descartes [1653] 2013 (the modeling the space as a discrete subspace of a space of
diagrams apparently having been added by 1635 to cylindrical topology. Figure 1.2.11b may be considered
the original 1618 manuscript), and Newton [1665] to be drawn not in a plane but in a punctured plane from
2013. See also Wardhaugh 2008, 29–​58. The question which the central point has been removed; a punctured
of whether musical pitch should be regarded as dis- plane has the topology of a cylinder.
crete or continuous was of great importance to many 31. Hint for Exercise 1.2.12: In some order, the
of the same scholars. last two questions have two of the following three
22. The chroma circle appears, for example, in answers: 7, 12, infinitely many. A diagram of the
Shepard 1982, 360. For the use of clock diagrams in enharmonic equivalence regions in signed letter space
pitch-​class set theory, see Straus 2016, 46ff or Roig-​ may be found in Hook 2007c, 104.
Francolí 2008, 71ff. 32. Note carefully the difference between signed
23. Though both spaces are conceptually much and spelled. Signs (accidentals) are applied to letters
older, Morris 1987 may have been the first to for- in order to distinguish different notes with the same
malize the terms pitch space and pitch-​class space and letter name, such as D♭, D, and D♯. Spelling applies to
the distinction between them. In the stylistic con- pitches or to pitch classes, and is used to distinguish
vention followed here, the noun pitch class contains different names for enharmonically equivalent notes,
no hyphen, but the compound modifier in pitch-​class such as B♯ ♯, C♯, and D♭. This is why we refer to signed
space is hyphenated. letters but to spelled pitch classes.
24. Hint for Exercise 1.2.5: In a circle, a path may 33. It is traditional to consider the generating
pass through the same point more than once. interval of pc to be the perfect fifth, but because the
25. The word dimension is defined in several differ- elements of the space are pitch classes, not pitches,
ent ways in mathematics; the dimension of a vector there is no way to distinguish fourths from fifths, and
space is defined differently from the dimension of a indeed, the perfect fourth would serve as a generator
manifold, for example. We will not give a precise defi- just as well. The conventional privileging of the fifth
nition here; as long as readers are reasonably familiar over the fourth in this and many other contexts (for
with Euclidean space in one, two, and three dimen- example, “descending-​fifths sequences”) is presum-
sions they should be able to follow the discussion. ably motivated by its greater harmonic stability.
26. Hint for Exercise 1.2.6: A correct answer may 34. Some jazz textbooks (for example, Coker 1975,
be obtained by inserting three different numbers in 37) orient the circle of fifths with ascending fifths

Pitch and Pitch-Class Spaces • 35

proceeding counterclockwise. Levine 1995 calls the submediant harmony serving to prolong a tonic. For
two orientations the “circle of fifths” and the “circle of a synopsis of double emploi see Lester 1992, 133–​35.
fourths”—​a confusing nomenclature, inasmuch as the For Hauptmann’s triad of keys and the development
distinction between fourths and fifths in such a figure of Riemann’s function theory see Harrison 1994,
depends not only on the arrangement of the figure but 224–​25 and 265–​92 respectively. See also Tymoczko
also on the direction in which one traverses it and on 2011, 231–​32.
whether the interval from one pitch class to another 42. The properties described in Exercise 1.3.13 will
is measured in the ascending or descending direction. be explored more fully in Chapters 5 and 13–​14; see in
35. Tonal spaces whose elements are key areas particular Exercise 5.3.7c.
rather than notes will be explored in Chapter 4. For 43. Tonnetz is one of several terms from German
an entertaining survey of some early depictions of the harmonic theory to have infiltrated English-​language
circle of fifths, see Westergaard 1996, 12–​13. scholarship; others include Schritt, Wechsel, and
36. As a guideline for note-​spelling, compactness Leittonwechsel. Many authors, respecting the Teutonic
along the line of fifths is not the only criterion and origins of these terms, italicize and capitalize them
is sometimes superseded by others. For example, at each appearance and dutifully employ the German
the line-​of-​fifths rule favors the spelling C–​C♯–​E over plurals (Tonnetze, Schritte, Wechsel, Leittonwechsel).
C–​D♭–​E, but the latter is likely to be preferred in the These terms now appear in the literature—​and will
context of a passage in F minor. For another example appear throughout this book—​with such a frequency
see Exercise 14.1.3f. that this convention is in danger of becoming a dis-
37. Theoretical study of the line of fifths can be traction, so I have chosen to treat them as ordinary
traced at least to Rameau, who took an interest in “triple English nouns.
progressions” of the form 1–​3–​9–​27–​···, to which he 44. The terms conformed tonnetz and unconformed
was apparently introduced by Louis-​ Bertrand Castel tonnetz have their origins in Harrison 2002.
(Christensen 1993, 178–​79). The ratio of 3 between suc- 45. Some readers may suppose that wrapping a
cessive terms in this geometric progression corresponds plane both horizontally and vertically should result in
to the interval of a twelfth in pitch space, or a fifth when a sphere, which is topologically distinct from a torus.
octave differences are disregarded. Organization by fifth To see one reason why a sphere is not the appropri-
is central to the conception of pitch in Regener 1973. ate topology for a tonnetz, notice that in a sphere, a
Temperley 2000 presents basic applications of the line north-​south path and an east-​west path intersect in
of fifths, and Temperley 2001, 115–​36 and 183–​87, two distinct points on opposite sides of the sphere.
offers more detailed discussion, including ramifications In the tonnetz of Figure 1.4.3, a north-​south path
for note-​spelling and key-​finding. Temperley refers to and an east-​west path always intersect in only a sin-
signed letter names as tonal pitch classes (comparable to gle point.
spelled pitch classes as used here). 46. The one-​dimensional analog of the torus is the
38. The concept of signed scale degrees and some circle, which may be called 𝕋1. The observation that the
other ideas in this exercise were suggested by Nathan torus is a product of two circles may be summarized
Lam (2020). by saying that 𝕋2 is topologically equivalent to 𝕋1 × 𝕋1.
39. Hint for Exercise 1.3.7d: In the example men- 47. Mathematically, Figure 1.4.4 depicts a non-​
tioned, the spc number of the tonic E♭ is n =​−5, the isometric embedding of a two-​dimensional space in
spc number of C♭ is m =​−9, and the ssd number of ♭6̂ is a three-​dimensional space: equidistant points in the
k =​−6. Check a few other examples. torus may fail to be equidistant in the embedding.
40. There is a third alternative: Elaine Chew’s spiral 48. Tymoczko 2012, 38–​ 39, comments on the
array, to be studied in Section 4.4, is constructed by question of whether toroidal structure is intrinsic to
aligning every fourth note of the line of fifths. the tonnetz, and more generally on the relationship
41. Functional affinities between third-​ related between discrete tonnetz structure and continuous
chords, attributable to common tones and effi- spaces.
cient voice leading, are implicit or explicit in several 49. The property mentioned in Exercise 1.4.7a was
historically significant conceptions of harmonic noted by Carl Friedrich Weitzmann in his 1853 mono-
theory. Notable examples include the figured-​ bass graph on the augmented triad. See Cohn 2000, 90–​91.
tradition by which a C-​ major triad in first inver- 50. Hint for Exercise 1.4.7c: Remember that the left-
sion is regarded as a “chord on E”; Rameau’s double most and rightmost columns shown in Figure 1.4.3
emploi; Hauptmann’s “triad of keys” construction; are adjacent.
Riemann’s functional categories, which tend to group 51. Cohn 2011 offers further comparisons among
third-​related chords together; and the common phe- several roughly equivalent tonnetz geometries, includ-
nomenon in Schenkerian theory of a mediant or ing historical notes on the provenance of each version.

36 • E x p l o r i n g M u s i c a l S p a ces
 52. The use of 3:4:5 triangles in the tonnetz in Agmon 1995, 204), has become problematic
was suggested by Charles Smith (2003). The angle through its ever broader application, to the point
between the two shortest sides in a 3:4:5 triangle is that Cohn himself (2012, xiii–​xiv) largely eschews
a right angle. The possibility of depicting intervals in it. In this book I restrict the use of this term essen-
this way is fortuitous: major and minor triads are the tially to three situations for which I believe it is
only three-​note set class whose constituent inter- appropriate and for which no other accepted term
val classes can form right triangles, and one of only is available: the neo-​Riemannian tonnetz described
three set classes for which they can form triangles here; the triadic transformations P, L, and R and
at all. For a different approach to conveying pitch their compounds, which will first be encountered in
distances by tonnetz geometry, see Brower 2008, Section 2.3 and which are generally agreed to occupy
64–​65, Figures 15 and 16. a central position in neo-​Riemannian theory; and
53. For a tonnetz with major thirds and perfect fifths a space to be introduced in Section 4.2 called neo-​
oriented at right angles—​an ic 4-​5 tonnetz, but still an Riemannian triad space, which is intimately related
037 tonnetz—​see Longuet-​Higgins [1962] 1987. both to the neo-​Riemannian tonnetz and to the PLR
54. Also, the relationship of accidentals to the family of transformations. In the latter two settings
white-​note diatonic collection is such that sharps we need not be quite so diligent about distinguish-
and flats (and double sharps and double flats, and so ing Riemannian from neo-​Riemannian—​indeed, P,
on) are symmetrically positioned about the note D in L, and R are Riemannian transformations in a tech-
many arrangements. This sharp-​flat symmetry about nical sense to be defined in Chapter 8, and we will
D can be seen if Figure 1.4.1 is redrawn with D at its sometimes refer to neo-​Riemannian triad space as
center, and also in the Riemannian tonnetz of Figure Riemann space for short. In the case of the two very
1.4.14 below. different tonnetzes of Figures 1.4.14 and 1.4.5, how-
55. One of the most preeminent and prolific math- ever, the distinction is indispensable. Rings 2011a
ematicians of his or any other era, Euler (pronounced contrasts “paleo-​Riemannian” and neo-​Riemannian
“Oiler”) also published five treatises on music. A dia- analytical methodologies.
gram resembling a tonnetz appears in Euler [1739] 59. The note names in Figure 1.4.1 are signed
1968, 147, and a picture essentially identical to letters (spelled pitch classes). The standard musi-
Figure 1.4.10c appears in Euler [1773] 1960, 584. cal notation to which these note names correspond
See Mooney 1996 for a wealth of detail on the his- may be said to be “partially conformed” in the same
tory of the tonnetz, Cohn 1998a for a brief synopsis, way as the figure: notation does not recognize the
and Cohn 2011 for further commentary on the finer syntonic comma (notes separated by a syntonic
distinctions among some of the various versions. comma have the same spelling), but it does recog-
56. See Riemann 1914–​15, 20, translated in Wason nize the Pythagorean comma between F♯ and G♭. The
and Marvin 1992, 102; and Oettingen 1866, 15. The Pythagorean comma is the difference between twelve
diagonal thirds axes of Figure 1.4.14 follow Riemann pure perfect fifths and seven pure octaves; its fre-
1914–​ 15; Oettingen’s tonnetz is rectilinear, with quency ratio is 312/​219, corresponding to an interval
major thirds oriented vertically. A tonnetz in Riemann of 0.235 equal-​tempered semitones, just a bit larger
1900 (906) follows the Oettingen orientation; in both than the syntonic comma.
of his versions Riemann’s underscores correspond to 60. Hint for Exercise 1.4.16: The second move, from
Oettingen’s overscores. The 1900 tonnetz covers a C to A, is a descending pure minor third, immediately
larger portion of the space than Figure 1.4.14, extend- up and to the left in the Riemannian tonnetz. The phe-
ing as far as B♭ ♭ ♭ (with a triple overscore) and F♯ ♯ ♯ (with nomenon described in this exercise, sometimes called
a triple underscore). syntonic drift, was apparently first noted in a letter
57. If the central C is assigned a frequency value of written by Giovanni Battista Benedetti to Cipriano de
1, the fifth-​generated E (the Pythagorean major third) Rore around 1563 (Palisca 1985, 262–​64). Syntonic
has frequency equal to ( 32 ) /22 = 64
4 81 . (The fraction on drift occurs also under the assumption that each
the left side of this equation can be interpreted as “up chord is purely tuned and that common tones are sus-
four perfect fifths and down two octaves.”) The E a tained from chord to chord. See Lindley 2001, 291–​92,
pure major third above C (the just major third), mean- and Sethares 2005, 170–​71.
while, has frequency 54 = 64 80
. The two therefore differ 61. In a use of the word generic unrelated to the
by a ratio of 81/​80, which corresponds to an interval generic spaces considered in this book, Cohn 1997
of 12 ⋅ log2(81/​80) ≈ 0.215 equal-​tempered semitones, introduces the term generic tonnetz for tonnetz-​
slightly more than one-​fifth of a semitone. type diagrams generated by any desired intervals, in
58. The adjective neo-​Riemannian, introduced in microtonal spaces as well as the usual twelve-​note
Cohn 1996 (although “Neo-​Riemannism” appeared pitch-​class space. Several tonnetzes of different

Pitch and Pitch-Class Spaces • 37

intervallic structures figure prominently in Brown 68. The opening of “Nacht” is another of the
2003. Tonnetzes for all twelve trichordal set classes examples in Peck 2003; see also analyses by Gillespie
may be found in the online appendices to Straus 2011. (1992), Brown (2003, 35–​38), and Straus (2016, 28–​
As with the 037 tonnetzes in this chapter, there is 31). There are ten 014 trichords in mm. 1–​3. Nine of
more than one way to draw any of these. them form a logical path through the appropriate ton-
62. Technically Lewin’s graph is not the 013 ton- netz; one is anomalous.
netz described here but its dual graph, in a sense that 69. See Figure 4.3.1b for a rectangular tile for the
will be explained in Chapter 3 and explored further in generic tonnetz.
Chapter 4. See Exercise 4.4.4. 70. Hints for Exercise 1.4.29: The property in (a) has
63. Hint for Exercise 1.4.17: There is more than one something to do with the number 7; that in (b) has to
way to lay out this space and its repeating modules. do with properties of the particular generic trichords
Depending on the arrangement you choose, you may involved. The tiles in (e) can be parallelograms two
find that rectangular tiles will not work but some columns wide, with two vertical sides and two slop-
other shape will. ing sides. We will revisit this generic Möbius strip in
64. If the identification of the vertical boundar- Chapter 4; see Exercise 4.4.1.
ies is performed (in similar orientation) in addition
to that of the horizontal boundaries (in opposite
orientation), the resulting surface is known as a
SUGGESTED READING
Klein bottle. The Klein bottle is a two-​dimensional Edward Tufte’s The Visual Display of Quantitative
manifold, but even in three-​dimensional space it Information (first published in 1983; 2nd edition,
is not possible to construct a Klein bottle without 2001) inaugurated a remarkable series of books,
making the surface cross through itself. (It is pos- themselves masterpieces of design, recommended
sible in four dimensions.) See Weeks 2020, 41–​54, to anyone interested in developing effective
for the topology of Möbius strips and Klein bottles, practices for communicating through figures and
and Peck 2003 for some advanced tonnetz applica- diagrams. Isaacson (forthcoming) applies Tufte’s
tions of Klein bottles. principles in the field of music visualization.
65. Hint for Exercise 1.4.19: What happens if you Zbikowski 2002 and Spitzer 2003 comment on the
construct a tonnetz generated by semitones and per- metaphorical nature of mappings between musi-
fect fifths? By whole tones and tritones? cal and physical spaces. Westergaard 1996 offers
66. Peck 2003 analyzes the opening of Musique a (literally) poetic survey of a few musical spaces.
funèbre, situating it within a fragment of a Klein-​bot- For an introduction to concepts of topology and
tle tonnetz. geometry that is simultaneously accessible, light-
67. Cohn (2012, 115) refers to a series of snap- hearted, and mathematically conscientious, see
shots depicting motion through a space as a “strobo- Weeks 2020. A more traditional presentation of
scopic portrait.” topology is Munkres 2018.

38 • E x p l o r i n g M u s i c a l S p a ces
 2
Sets, Functions, and Relations

MANY MUSIC theorists are familiar with set the subsequent applications. Even those already
theory in the context of small finite sets of conversant with this material may benefit from
pitch classes; most are probably aware that considering some of the examples and exercises,
the mathematical field of set theory is vastly some of which present useful principles that will
richer. All of the spaces studied in Chapter 1 be invoked in later chapters. Topics that may be
are, first and foremost, sets. Many of these new for many readers include multisets, intro-
spaces are theoretically infinite sets, but small duced at the end of Section 2.2, and the func-
finite sets such as scales and chords are also tions defined on spelled pitch-​ class space in
of great musical interest, and relationships Section 2.6.1
among these various sets will engage our atten-
tion many times. 2.1 SETS
These observations are the justification for
the modest survey of set theory in this chapter, We follow the nearly universal convention of
which also covers the essential concepts of func- listing the elements of a set in braces;2 thus the
tions (transformations), relations, equivalence set whose members are the numbers 1, 4, and 6
relations, and modular arithmetic. The examples is written {1, 4, 6}. If this set is denoted S, then
and exercises in this chapter introduce a variety 4 ∈ S is a shorthand for the true statement “4 is
of musically significant sets, functions, and rela- an element of S,” and 5 ∉ S for the equally true
tions that will reappear many times later. The statement “5 is not an element of S.” Another
mathematics here is completely standard, and common shorthand is the ellipsis: {0, 1, …, 11}
much of the terrain will be recognizable to many is readily understood as the set consisting of all
readers, but it is included to ensure that readers twelve integers from 0 through 11, while {…,
are well acquainted with it before delving into −4, −2, 0, 2, 4, …} is the infinite set of all even

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0002
integers (positive, negative, and 0). The brace The discussion in Chapter 1 suggests that
notation may also be employed more informally, continuous pitch space cpitch may be identified
with a description rather than a list inside the with ℝ, and that either discrete chromatic pitch
braces; for instance {all odd positive integers} space pitch or generic pitch space gpitch may
could denote the set whose elements are 1, 3, be identified with ℤ. Musical interpretations of
5, …, while {all prime numbers} has elements 2, ℚ are less obvious, but the rational representa-
3, 5, 7, 11, … .3 Sets are by definition unordered: tion of intervals in just intonation is one exam-
{1, 4, 6} and {4, 6, 1}, though written differently, ple, and in the time domain it may be noted that
are the same set. (Ordered sets, a different—​and conventional rhythmic notation, including ties
differently notated—​construction, will be dis- and tuplets, enables the representation of all
cussed in Section 2.2.) positive rational durations.
Several familiar and useful sets of numbers The elements of a set need not be numbers:
are conventionally denoted as follows: one can form a set whose elements are planets,
US presidents, or musical notes (or note names,
ℝ =​{all real numbers} or other symbols representing notes). Chromatic
ℤ =​{all integers} =​{… , −3, −2, −1, 0, 1, 2, 3, …} pitch-​class space may thus be written simply as
ℚ =​{all rational numbers} =​{a ∕ b | a, b ∈ ℤ, b ≠ 0} pc =​{C, C♯, …, B}, and chromatic pitch space as
pitch =​{…, A♯3, B3, C4, C♯4, D4, …}. The elements
These sets are listed for reference in Appendix of a set may themselves be sets; for example, we
2, along with many other sets and other stan- may consider a set of triads, each of which is a
dard mathematical structures used in this set of three notes. The elements of a set do not
book. The vertical bar in the last expression for all need to be the same kind of object: one can
ℚ may be read as “such that” or “where”: ℚ is contemplate a set such as {6, Jupiter, Grover
♯
the set of all numbers of the form a ∕ b such that Cleveland, C } (though it is difficult to imagine
a and b are elements of ℤ and b is not equal for what purpose).
to 0. Not all real numbers are rational; well-​ The number of elements in a set S is known
known examples of irrational numbers include as the cardinality of S and will be denoted #S.
2 and π. (Other notations sometimes encountered
On the other hand, all integers are rational, include |S| and card S.) Thus #{1, 4, 6} =​3. If
4
and all rational numbers are real. Therefore ℤ S is an infinite set we write #S =​∞. There is a
is a subset of ℚ, which is in turn a subset of ℝ, unique empty set of cardinality 0, containing no
as summarized by the symbols ℤ ⊂ ℚ ⊂ ℝ. The elements at all; the empty set may be notated
statement “S is a subset of T ” is true whenever as { }, but the symbol ∅ is traditionally used for
every element of S is also an element of T; for this purpose. The reader should understand why
example, {1, 4, 6} ⊂ {1, 4, 5, 6}, but {1, 4, 6} ⊄ 0, ∅, {0}, and {∅} are four different things: a
{1, 4, 5, 7}. Every set is a subset of itself by this number, a set of cardinality 0, and two different
definition, but we reserve the symbol ⊂ for the sets of cardinality 1.
proper subset relation, the situation in which In many situations, all elements under consid-
at least one element of the superset T does not eration belong to a single large set U, and there-
belong to the subset S. If we wish to allow for fore all sets under consideration are subsets of
the possibility of equality we may write S ⊆ T. U. Such a set U is called a universal set. In musical
Readers should understand the distinction contexts, any of the various spaces introduced in
between the relation symbols ∈ and ⊂: the state- Chapter 1 may play the role of a universal set: in
ments 1 ∈ {1, 4, 6} and {1} ⊂ {1, 4, 6} are true, pitch-​class set theory, for example, the universal
♯
but neither would be true if the relation sym- set is U =​ pc =​{C, C , …, B}, a set of cardinality 12.
bols were interchanged. The intersection of sets S and T, denoted S ∩ T,
Some subsets of ℝ, ℤ, and ℚ may conve- consists of all elements common to both S and
niently be identified by appending subscripts T. For example, {1, 4, 5, 6} ∩ {4, 6, 8, 9} =​{4, 6}.
to the set labels, as illustrated by the following Two sets whose intersection is empty are called
examples: disjoint sets; for example, {1, 5, 6} and {4, 8, 9}
are disjoint. The union of S and T, denoted S ∪ T,
ℝ>0 =​{all positive real numbers} =​{x ∈ ℝ | x > 0} consists of all elements belonging to either S or
ℤodd =​{all odd integers} =​{… , −5, −3, −1, 1, 3, 5, …} T or both. For example, {1, 4, 5, 6} ∪ {4, 6, 8, 9} =​
ℚ≠0 =​{all rational numbers except 0} {1, 4, 5, 6, 8, 9}. Even though 4 and 6 belong
=​{a ∕ b | a, b ∈ ℤ, a ≠ 0, b ≠ 0} to both sets, they are listed only once each in

40 • E x p l o r i n g M u s i c a l S p a ces
the union. In traditional set theory, either an Exercise 2.1.2
element belongs to a set or it doesn’t; it can- (a) A distributive law states that
not belong more than once. We shall loosen this
restriction in the discussion of multisets in the S ∪ (T ∩ V) =​(S ∪ T) ∩ (S ∪ V)
next section.
In the context of a universal set U, every set for any sets S, T, and V. Draw a Venn
S has a complement Sʹ (sometimes alternatively diagram showing three intersecting
written S or SC), consisting of all elements of U sets, shading the region indicated in the
that do not belong to S. Readers may be familiar equation. Verify that this law holds for the
with complements in pitch-​class space; for exam- following subsets of pitch-​class space: S is
ple, the complement of a diatonic collection is a the diminished seventh chord containing B;
pentatonic collection, and the complement of T is the black-​key pentatonic collection; V
a diminished seventh chord is an octatonic col- is the whole-​tone scale containing C. (You
lection. The cardinalities of complementary sets should work out both sides of the above
always sum to the cardinality of the universal set equation and arrive at the same set both
(12 in the above examples).5 ways, in this case a six-​note set.)6
Readers are also probably familiar with the (b) Complete the statement of the second
use of Venn diagrams as pictorial representations distributive law
of intersections, unions, and complements of
sets, as in Figure 2.1.1. Such pictures are intui- S ∩ (T ∪ V) =​ _​_​_​__​ ​_​_​_​_​_​_​_​__​ ​,
tively appealing, though their value is limited
because it is often difficult to construct a dia- analogous to the first law in part (a) above.
gram that adequately captures all the possibili- Construct an appropriate Venn diagram
ties in a given situation, especially when more and verify that the law holds for the
than two or three sets are involved. following sets: S is the whole-​tone scale
A great number of properties involving inter- containing C; T is a C-​major triad; V is an
sections, unions, and complements may be for-   F♯-​major triad.
mulated, a few of which are illustrated in the
following exercises involving subsets of pitch-​ Exercise 2.1.3
class space pc. (a) One of de Morgan’s laws states that “the
complement of an intersection is the union
of the complements”—​that is,

(S ∩ T)ʹ =​ Sʹ ∪ Tʹ.

Construct an appropriate Venn diagram

(which will need to show a universal set),
and verify this law in the case in which S
is the C-​major diatonic collection and T is
the acoustic scale on C (that is, the set {C,
D, E, F♯, G, A, B♭}, which could also be called
the ascending melodic minor scale on G).
Several of the sets that arise in the course
of working out this example have familiar
names. What are they?
(b) Complete the statement of the second
of de Morgan’s laws, “the complement of
a union is …,” and write it symbolically.
Construct an appropriate Venn diagram
and verify this law when the sets involved
are two different diminished seventh
  chords.

FIGURE 2.1.1 Venn diagrams showing intersec- Exercise 2.1.4 The cardinality of the union
tions, unions, and complements of sets of two finite sets may be calculated by adding

Sets, Functions, and Relations • 41

the cardinalities of the two sets and then Exercise 2.1.6 The symmetric difference of
subtracting the cardinality of the intersection two sets S and T, denoted S ∆ T, consists of all
(to avoid counting those elements twice). This is elements that belong to either S or T but not
the union-​cardinality law: both. Using the notation from the previous
exercise, S ∆ T =​(S\T) ∪ (T\S), or alternatively
#(S ∪ T) =​#S +​#T − #(S ∩ T). S ∆ T =​(S ∪ T) \ (S ∩ T). For example, {1, 4, 5, 6}
∆ {4, 6, 8, 9} =​{1, 5, 8, 9}.
(a) Verify the union-​cardinality law for the sets
S and T of Exercise 2.1.3a. (a) Calculate the symmetric difference
(b) Use the union-​cardinality law to explain wt0 ∆ bo7.
why any two diatonic collections (which are (b) Calculate the symmetric difference
seven-​note subsets of twelve-​note pitch-​ wt1 ∆ oct01.
class space) must always have at least two (c) Using your answers to (a) and (b) as an
pitch classes in common. example, write a general formula of the
(c) Recall from Exercise 1.4.7a that an form Sʹ ∆ Tʹ =​ _​__​ ​_​_​_​.
augmented triad and a diminished seventh (d) What is the symmetric difference of two
chord always intersect in exactly one note. different octatonic collections?
Use the union-​cardinality law to calculate (e) Modify the union-​cardinality law to
how many notes are in the union of these obtain a formula for the cardinality of the
two chords; then choose specific chords and symmetric difference of two sets, and verify
verify that your calculation is correct. your formula for the sets in parts (a), (b),
(d) An enneatonic collection is the complement   and (d).
of an augmented triad. Use one of de
Morgan’s laws and the result of part (c) Exercise 2.1.7 The distributive laws from
above to calculate how many notes are in Exercise 2.1.2 may be extended to larger
the intersection of an enneatonic collection collections of sets; for example, the equation
and an octatonic collection; then choose
specific collections and verify that your S ∩ (T1 ∪ T2 ∪ ⋯ ∪ T6)
  calculation is correct.7 =​ (S ∩ T1) ∪ (S ∩ T2) ∪ ⋯ ∪ (S ∩ T6)

Exercise 2.1.5 If S and T are two sets, the holds. Consider this equation in the case in
difference S\T is the set S ∩ Tʹ, which consists of which S is a whole-​tone collection and T1,
all elements that belong to S but not to T. For T2, …, T6 are pitch-​class sets of cardinality
example, if S =​{1, 4, 5, 6} and T =​{4, 6, 8, 9}, 2 representing interval classes 1, 2, …, 6
then S\T =​{1, 5} while T\S =​{8, 9}. respectively. (For example, T1 might be the set
{5, 6}, T2 might be {3, 5}, and so on.)
(a) Consider wt0, the whole-​tone collection
containing pitch class 0 (C), and bo7, the (a) What are the possible cardinalities of the
fully diminished seventh chord containing set S ∩ T1? What about S ∩ T2, …, S ∩ T6?
B, as pitch-​class sets. Calculate wt0\bo7 and (b) Use the above distributive law and your
bo7\wt0. observations from part (a) to explain why
(b) Consider wt1, the whole-​tone collection it is not possible to divide pitch-​class space
containing C♯, and oct01, the octatonic pc into six disjoint subsets of cardinality 2
collection containing pitch classes 0 and   representing the six interval classes.9
1 (C and C♯), as pitch-​class sets. Calculate
wt1\oct01 and oct01\wt1. Exercise 2.1.8
(c) The sets in wt1 and oct01 in part (b) are (a) Explain why, if S is any four-​note pitch-​class
the complements of the sets wt0 and bo7 in set, there must always be some diminished
(a). What do you notice about the answers seventh chord T such that #(S ∪ T) ≤ 6.
to those two parts of the question? Write a (b) Use the result of part (a) and one of de
general formula of the form Sʹ \Tʹ =​ _​__​ ​_​_​_​. Morgan’s laws to show that every eight-​
(d) Explain why, for all sets S, T, and V, note pitch-​class set has at least six notes in
  S\T ⊆ S\V ∪ V\T.8   common with some octatonic scale.10

42 • E x p l o r i n g M u s i c a l S p a ces
2.2 ORDERED SETS AND Exercise 2.2.1
MULTISETS (a) Write the 4! =​24 permutations of the set
{G3, C4, D4, E4} in musical notation.
Sets as described in Section 2.1 may suffice to (b) How many four-​note strings can be formed
describe musical structures such as a triad or from the notes of this set if repeated notes
an octatonic collection, but only if the order are allowed? (If one or more of the given
of the notes does not matter. To accommodate four notes are repeated within a four-​
situations in which order is important we use note string, then some other notes will
a different construction, ordered sets. While the necessarily not be used at all.)
terminology and the brace notation { } for unor- (c) Write a general formula for the number
dered sets are well standardized, a number of of ordered sets of cardinality m that
different terms and notations for ordered sets can be formed from an unordered set of
may be encountered. One common practice, cardinality n, allowing repetitions and
which we will sometimes follow, is to enclose   omissions.12
the elements of an ordered set in parentheses.
Thus while {G3, C4, D4, E4} is just one set regard- Ordered sets of cardinality 2, more com-
less of the order in which the four notes are monly known as ordered pairs, are useful for
listed inside the braces, the ordered sets (E4, C4, many purposes. Readers are probably familiar
D4, G3) and (G3, D4, E4, C4) represent two dif- with the use of ordered pairs of real numbers
ferent melodic lines—​two different permuta- (x, y) to represent points in a plane. While the
tions of the same four notes. Depending on the set ℝ of real numbers is naturally identified
context, ordered sets may also be referred to as with a one-​dimensional line, the use of ordered
strings, segments, lines, rows, series, lists, tuples pairs enables an entire plane to be represented
(“ordered 4-​tuples” in the above examples), or numerically, combining information in two
vectors (the usual interval-​class vector in pitch-​ dimensions, horizontal and vertical. The ordered
class set theory is an ordered set of six num- nature of the pairs is crucial, as for instance
bers).11 An alternate notational convention (1, −3) and (−3, 1) represent two different
that occasionally proves useful is to omit the points. The elements x and y are called the first
parentheses and link the elements with dashes, and second components, or coordinates, of the
as in E4–​C4–​D4–​G4. We will consider finite ordered pair (x, y).
ordered sets only. Mathematicians do not use A similar, if somewhat more abstract, con-
the element symbol ∈ with ordered sets; thus struction is possible for sets other than ℝ. If S
C4 ∈ {G3, C4, D4, E4} is a valid statement but C4 ∈ and T are any two sets, the Cartesian product of
(E4, C4, D4, G3) is not. S and T, denoted S × T (usually pronounced “S
The number of ordered sets that can be cross T ”), is the set of all ordered pairs (s, t) such
formed from the elements of a fixed universal that s ∈ S and t ∈ T. Figure 2.2.2 offers a sche-
set far exceeds the number of unordered sets. matic illustration of this construction, combin-
In general, a set of cardinality n may be ordered ing information from both sets in an abstract
in n! different ways, where n! denotes n factorial, two-​dimensional picture. If S and T are finite
the product 1 ∙ 2 ∙ 3 ∙∙∙ n of all the integers from 1 sets, the cardinality of the Cartesian product is
to n. Factorials grow very large even for modestly the product of the cardinalities of the two sets;
sized values of n. For example, the number of the schematic figure shows an example with
possible twelve-​tone rows is 12! =​479,001,600: #S =​5, #T =​4, and #(S × T) =​20. Intuitively, the
almost half a billion permutations of a single picture suggests that a Cartesian product may
twelve-​element set. be partitioned into a series of horizontal slices,
Moreover, in many situations repetitions are each of which resembles the set S, or into a series
allowed in ordered sets: C4–​D4–​E4–​C4 is a valid of vertical slices, each of which resembles T. The
four-​note string, even though its notes all belong horizontal and vertical directions are arbitrary
to the three-​note set {C4, D4, E4}. In such cases here, but it is true that if t0 is any fixed element
the word cardinality may be ambiguous; here we of T, the ordered pairs (s, t0) are in one-​to-​one
use the word to mean the length of the string, correspondence with the elements of S, and if s0
so the cardinality of the ordered set C4–​D4–​E4–​ is any fixed element of S, the ordered pairs (s0,
C4 is 4. t) are in one-​to-​one correspondence with the

Sets, Functions, and Relations • 43

 be unambiguously identified with an ordered
pair; for example, B♭ ♭ is (−1, −2). Determine
the coordinates of each of the following
notes: C, C, C, C, B♯, D♭ ♭. What note appears
at coordinates (12, 0)? At (8, 1)? At
(−4, −2)? It is generally true that any note
enharmonically equivalent to C (in the usual
equal temperament) has coordinates in this
tonnetz that may be obtained by plugging
some integers a and b into the ordered-​pair
formula (12a − 4b, b). Verify that all of the
above examples are consistent with this
  statement.13
FIGURE 2.2.2 The Cartesian product of two sets Various other musical spaces can be described
as Cartesian products. Points in the conformed
tonnetz tile of Figure 1.4.3 may be represented
elements of T. The condition that S and T may as ordered pairs in a Cartesian product of two
be “any two sets” is understood to allow for the small finite sets, an idea whose implications will
possibility that they are the same set. In fact, the be pursued in Section 6.4 when we examine the
usual ordered-​pair representation of the plane is interval structure of tonnetz space. The continu-
the Cartesian product ℝ × ℝ, which is sometimes ous surface of a torus such as Figure 1.4.4 may
abbreviated ℝ2. be represented as the Cartesian product of two
circles, a representation that will play a role in
Exercise 2.2.3 our study of voice-​leading spaces in Part Three.
Triads may be represented by ordered pairs such
as (0, +​) (C major) or (3, −) (E♭ minor); in this way
(a) Explain how signed registral letter space
(Figure 1.1.11) may be constructed as a
Cartesian product S × T, where S is generic the set of 24 major and minor triads may be iden-
pitch space (Figure 1.1.6) and T is the set tified with the Cartesian product S × T, where S is
of all possible accidentals, T =​{…, ♭ ♭ ♭, ♭ ♭, ♭, ♮, pitch-​class space pc and T is the two-​element set
♯, ♯ ♯, ♯ ♯ ♯, …}. {+​, −} (or, if one prefers, the set {1, −1}).
(b) Explain how signed letter space (Figure Cartesian products of more than two sets may
1.2.11) may be constructed as a Cartesian be constructed using ordered tuples larger than
  product of two sets. pairs. If S, T, and V are three sets, the Cartesian
product S × T × V consists of all ordered triples
(s, t, v) of elements s ∈ S, t ∈ T, and v ∈ V. The
Exercise 2.2.4 appropriate picture for such a set, analogous to
(a) Ordered pairs may be used to define a Figure 2.2.2, is now three-​dimensional. Points in
coordinate system for labeling points on three-​dimensional space may be represented by
any of the tonnetz diagrams presented in coordinates (x, y, z) in ℝ × ℝ × ℝ, or ℝ3. The same
Chapter 1. In the conformed tonnetz of principles apply in higher dimensions, so points
Figure 1.4.2, for example, we may assign in n-​dimensional space may be identified with
the ordered pair (0, 0) to C in the center ordered n-​tuples (x1, x2, …, xn) in ℝn. Spaces of four
of the diagram, (1, 0) to the E♭ to its right, or more dimensions can be difficult to visualize in
and (0, 1) to E directly above. In the central our three-​dimensional surroundings, and these
tile, what are the coordinates of G? Of D? higher-​ dimensional spaces seem strange and
Besides (0, 0), what are the coordinates of exotic to many people, but there are no barriers
the various other Cs in this tonnetz? What to constructing them abstractly, and with practice
pitch class is represented by the ordered one can begin to develop intuitions about them.14
pair (−4, 2)? By (5, −3)? By (101, 46)? Chords in a four-​voice musical texture may
(b) In the Riemannian tonnetz of Figure 1.4.14, be represented by points in pitch4—​that is,
consider a coordinate system in which by ordered 4-​tuples of notes in chromatic pitch
C is (0, 0), G is (1, 0), and E is (0, 1). The space. In this construction, each voice in the
coordinate axes so defined do not form a texture is given its own dimension in which
right angle, but every point may nevertheless to move. Adopting the convention that the

44 • E x p l o r i n g M u s i c a l S p a ces
voices in the 4-​tuple appear in the order (bass, C-​major triad with a doubled root.” The ordinary
tenor, alto, soprano), we are able to distinguish pitch-​class set {C, E, G} cannot account for the
between chords such as (C3, C4, E4, G4) and (C3, doubling. An ordered set may list C twice, but
E4, C4, G4), the latter of which includes a voice only by imposing an order on the four notes. The
crossing between the two middle voices. With order may be of interest, as it was in some previ-
pitch numbers assigned to notes as in Figure ous examples, but it may be irrelevant if our only
1.1.1, these two chords appear in ℤ4 at the concern is how many times each note appears in
points (−12, 0, 4, 7) and (−12, 4, 0, 7). This strat- the chord.
egy may be adapted to model chords in generic The appropriate construction in the latter
or diatonic space rather than chromatic, or in pc situation is a multiset, which in this case may be
space rather than pitch space. Taking ℤ to rep- written {C, C, E, G}. Multisets are a generalized
resent generic pitch space gpitch rather than version of unordered sets in which repeated ele-
pitch, for example, the first of the above chords ments are allowed. Some aspects of set theory
is now (−7, 0, 2, 4). do not generalize well to multisets (for example,
A chord progression is an ordered set of functions, discussed in the next section, cannot
chords, and therefore a tuple of tuples. For exam- easily be defined on multisets), but multisets are
ple, a progression of six chords in four voices is nevertheless useful in many circumstances. For
an ordered 6-​tuple, each of whose elements is the multiset {C, C, E, G}, we say that the cardi-
an ordered 4-​tuple. We may associate this pro- nality is 4, counting all repetitions. The note C
gression with a series of six points in 4-​dimen- is an element of the multiset with multiplicity 2.
sional space ℤ4, or possibly with a single point in E and G are both elements of multiplicity 1; all
24-​dimensional space ℤ24. This rather elaborate other pitch classes may be said to have multiplic-
mathematical representation of a short chord ity 0 in this multiset, which means that they do
progression begins to suggest that even simple not belong to it at all. An ordinary set is a mul-
musical phenomena may be more complex than tiset of a particular type, all of whose elements
we sometimes imagine. The voice-​leading spaces have multiplicity 1; in informal parlance, how-
to be examined in Part Three, in which we will ever, the use of the word multiset is sometimes
model voice leadings as paths from one point to taken to imply that repeated elements are pres-
another in higher-​dimensional spaces, will offer ent. Every multiset may be reduced to an ordi-
opportunities to develop some of these ideas. nary set of equal or lesser cardinality, containing
all the same elements but always with multiplic-
Exercise 2.2.5 Recall the chord progressions ity 1; for example, the reduction of {C, C, E, G}
from Exercise 1.3.11a (a descending-​fifths is {C, E, G}. Of course, some information—​the
sequence of diatonic seventh chords in D original multiplicities of the elements—​is lost in
major) and 1.3.11b (a chromatic descending-​ the reduction process.15
fifths sequence of dominant seventh chords).
Transcribe all chords of both progressions into
ordered 4-​tuples in ℤ4 (representing pitch4) as
2.3 FUNCTIONS
described above. Also transcribe the diatonic The concept of function is central in all branches
progression into ordered 4-​tuples in ℤ4, where of mathematics. The relevance of functions to
now ℤ represents dpitch(+​2) (D-​major diatonic musical structure has been clear at least since
pitch space) and 0 stands for the note C♯4. All David Lewin’s development of transformation
together you will have three different numerical theory (though in fact some functions were
representations of chord progressions. If you widely used in music theory well before then),
arrange the 4-​tuples in columns, so that each and functions will figure prominently through-
progression becomes an array of four rows of out this book. We regard the words function,
numbers, you should see that two of the three mapping, and transformation (and, in many cases,
follow systematic numerical patterns, but the operator or operation) as essential synonyms,
other
   appears slightly erratic. Why is this? though one or another of these terms may be
favored in certain contexts.16
Musical applications of sets and ordered sets Intuitively, if S and T are two sets, a function f
are numerous. We will encounter a few situa- from S to T is a rule that assigns to each element
tions, however, for which neither of these con- x of S some element y of T. It is usual to write y =​
cepts is precisely appropriate. Consider the idea f(x); the right side of this equation is pronounced
of pitch-​class sets with doublings, for example “a “f of x.” Another traditional notation is f: S → T,

Sets, Functions, and Relations • 45

as shorthand for “f is a function from S to T.” workings of the black box (gears? microchips?
Several variants of this arrow notation are useful elves?) are irrelevant to the definition of the
at times, such as S → 
f
T , x → f(x), or x →
f
y. function; the only thing that matters is the asso-
Notice that the arrows in these shorthand nota- ciation between the input and output.18
tions may link either sets (S and T) or elements This last point has important implications for
of those sets (x and y).17 In many important situ- the sometimes deceptive notion of equality of
ations S and T are the same set; when we refer to functions. Consider the following two functions
a transformation on S we mean a function f: S → S. f and g from ℝ to ℝ (that is, S and T are both
To define a function properly, one must equal to ℝ, the set of real numbers): for every
specify the sets S and T, and one must explain real number x, f(x) is defined by the equation
how f(x) is determined for each x in S. The set S f(x) =​2x +​2, while g(x) is defined by the equation
is called the domain of the function. A function g(x) =​2(x +​1). In other words, we calculate f by
is required to be defined at every point in its doubling a number and then adding 2, and we
domain; for every x in S, there must be one and calculate g by first adding 1 and then doubling
only one y in T for which y =​ f(x). This y is called the result. Basic algebra tells us that f(x) =​ g(x)
the value of the function at the argument x; y may for all possible inputs x. Whichever of the two
also be called the image of x under the function calculations we perform, the result will be the
f. Functions must be deterministic; that is, if the same; the black boxes for f and g cannot be distin-
same function f is applied to the same argument guished by their behavior. It follows, therefore,
x on two different occasions, the result f(x) must that f and g are the same function. The equation f =​
be the same both times, determined only by f g is a valid statement about the functions them-
and x. The calculation of a function may not rely selves. Two functions may be defined differently,
on nondeterministic processes such as rolling a and different algorithms may be followed in cal-
die or consulting the stock market. culating them, but if they always give the same
Figure 2.3.1a depicts a function as an arrow results, the functions are in fact equal—​and
pointing from a set S to a set T. This figure if two things are equal, they are not really two
encourages us to think of a function as a map- things: they are one thing. (Do not overlook the
ping, a dynamic process that leads from one place word always; two functions that give the same
to another; every x in S is mapped by this arrow results for some inputs but not for others are not
to a point y =​ f(x) somewhere in T. Figure 2.3.1b the same function at all.) The point is a simple
shows a different metaphor, that of a function but important one, and a frequent source of con-
as a “black box”: the element x is fed into the fusion in musical applications.19
box as an input, and the output f(x) somehow The role of the letter f in the above discussion
emerges. This imagery reinforces the idea of deserves brief scrutiny. Informally we may speak
a function as a transformation, a process that of “the function f(x) =​2x +​2”; more precisely, the
transforms x into f(x). Both of these schematic equation f(x) =​2x +​2 is a definition, a formula used
pictures, the dynamic arrow and the transformer to define the function that is being denoted by the
box, are valuable ways to think about functions. name f. The letter f is a variable commonly used
Like other metaphors, however, they can mis- to denote functions, just as the letter x is a vari-
lead, and can sometimes suggest specious asso- able that is more often used to denote objects of
ciations. In particular, the contents and inner other kinds.20 These letters may stand for different

FIGURE 2.3.1 (a) A function f: S → T; (b) a function as a “black box”

46 • E x p l o r i n g M u s i c a l S p a ces
things in different situations, and indeed, many graphs simultaneously in a way that is space-​effi-
different functions in this book will be given the cient, precise, and readily comprehensible.21
name f. Conversely, the function given by the The function graph in Figure 2.3.2 is concep-
above equation could perfectly well be called g, F, tually similar to a player-​piano roll. The domain
or ξ; the name is irrelevant to the identity of the of the function f—​the horizontal dimension of
function. In fact, it is sometimes possible to refer the figure and the long dimension of the roll—​is
to functions without naming them at all, by means temporal space time, a continuous space, and
of expressions such as “the function x → 2x +​2.” the graph shows that the function is defined at
Readers are likely to be familiar with the pro- all points of this space (or at least the segment of
cess of graphing functions defined on the real the space shown here). The function is not, how-
numbers. If f: ℝ → ℝ, the graph of f is a picture ever, a continuous function as the term is defined
of all the ordered pairs (x, y) for which y =​ f(x), in mathematics, because of the way it jumps
a subset of ℝ × ℝ =​ ℝ2. The graph of the func- abruptly to a new value each time the melody
tion f (or g) in the preceding paragraphs is a moves to a new note.22 The excerpt begins at
straight line of slope 2 passing through the time 0 (the first beat of the first measure), and
point (0, 2). Although the physical construction f(0) is the pitch G4, pitch number 7 in our usual
of such a graph presupposes that the function’s numbering. But it is not just f(0) that is equal
arguments and values are real numbers, the asso- to 7; in fact f(t) =​7 for all real numbers (time
ciation of a function with a set of ordered pairs points) t with 0 ≤ t < 2. Then at t =​2 (the third
is possible for any function at all: a function f: beat), the value of f changes abruptly: for all t
S → T may be identified with a subset of S × T, with 2 ≤ t < 5, f(t) =​8 (the pitch A♭4).23 When no
specifically the set of all ordered pairs (x, y) for note values smaller than quarter notes are pres-
which y =​ f(x). ent, continuous temporal space may effectively
Graphs of the sort just mentioned will be of be replaced by discrete beat space, because all
limited interest in this book; other uses of the essential information about the function is cap-
word graph, to be introduced in Chapter 3, will tured by its values at integer time points: f(0) =​
receive more attention. It is worth remarking, f(1) =​7, f(2) =​ f(3) =​ f(4) =​8, f(5) =​0, and so on.
however, that a musical score closely resembles
such a graph, or more accurately a combina- Exercise 2.3.3
tion of many such graphs. A melodic line may (a) In the manner just described, the subject
be thought of as a function, mapping from the of the E-​major fugue from Book II of The
time domain into pitch space. Figure 2.3.2 shows Well-​Tempered Clavier may be modeled
a melody and its representation as a function by the function f(0) =​ f(1) =​−8, f(2) =​−6,

The time signature is , and we are here

graph. Musical notation is remarkable for its abil- f(3) =​−3, f(4) =​−4, f(5) =​−6, f(6) =​−8.
ity to consolidate information from many such

FIGURE 2.3.2 A melody as the graph of a function

Sets, Functions, and Relations • 47

 measuring time in half notes. The values In the definition of a function f: S → T, the
are negative because the pitches are below sets S and T do not play symmetrical roles. The
middle C. Draw the graph of f, and based definition implies that for each x in S there is
on this description, write the subject in exactly one y in T for which y =​ f(x), but there
musical notation. is no corresponding requirement that for each y
(b) At time 6 (the third beat of measure 2), there be exactly one x. It is entirely possible that
a real answer enters, an exact imitation for some values of y there will be many different
of the subject a perfect fifth higher. This x’s such that f(x) =​ y, while for some other y there
answer may be represented by a function may be no such x at all. The function graphed in
g, starting with g(6) =​−1 (g(t) is undefined Figure 2.3.2 provides an example, as each pitch
for values of t smaller than 6). Write the (y) occurs at more than one point (x) in the
values of g from g(6) to g(12) and draw continuous time domain, the pitch D♭4 occurs a
its graph on the same set of axes with the second time after other intervening notes have
graph of f. sounded, and some other pitches never occur at
(c) Write a formula that shows how to calculate all. For another example, consider the function
g(t) from the values of the function f. f: ℤ → ℤ defined (for integers only) by f(x) =​ x2.
(d) More generally, suppose functions f and g If y =​4, there are two different values of x for
represent two melodic lines, and suppose which f(x) =​ y, namely x =​2 and x =​−2. But if
that g imitates f, at a time interval of y =​5, there is no such x at all, because there is no
b beats (later) and a pitch interval of s integer whose square is 5.
semitones (higher). Write a general formula These observations lead to a series of impor-
that expresses the relationship between the tant definitions. Some functions f: S → T have the
  functions f and g.24 property that different x’s always map to different
y’s. That is, whenever x1 and x2 are elements of S
The graphs in Figure 2.3.4 highlight aspects and x1 ≠ x2, it follows that f(x1) ≠ f(x2)—​or, equiva-
of chromatic and diatonic structure that are lently, the only way that f(x1) =​ f(x2) can be true is
concealed by traditional musical notation. Each if x1 =​ x2. A function f satisfying this property is
part of the figure shows the three voices in some said to be one-​to-​one, or an injection. The function
form of an ascending 5–​6 sequence. In (a) the shown schematically in Figure 2.3.6a is not one-​
sequence is the familiar diatonic variety—​one of to-​one, because the figure shows two different
the most common diatonic sequence patterns—​ elements of S mapped to the same element of T.
and the voices are graphed in C-​major diatonic Some other functions f: S → T have the prop-
pitch space dpitch(0). Part (b) analogously erty that every y is f(x) for some x. That is, if y is
graphs a chromatic version of the sequence in any element of T, there is guaranteed to be at
chromatic pitch space.25 Finally, (c) graphs the least one x in S such that f(x) =​ y. If f is such a
diatonic sequence in chromatic space. This third function, we say that f maps S onto T. The prepo-
graph appears less orderly than the other two, sitions to and into may be used for any func-
because the rules governing the sequential tion at all, but onto is reserved for this special
motion operate in a different pitch space from situation. In fact, onto is routinely pressed into
the space depicted in the graph. The step sizes ungainly service as an adjective; it is common to
in each voice, perfectly regular in both (a) and say “f is onto.” Onto functions are also called sur-
(b), appear somewhat uneven in (c), and the jections. The function in Figure 2.3.6b apparently
inconsistent interval sizes between adjacent fails to be onto, because one element of T is not
voices—​irregularly alternating major and minor the image of any element of S.
thirds—​are visually apparent. The function f: ℤ → ℤ defined by f(n) =​2n is a
simple example of a function that is one-​to-​one
Exercise 2.3.5 Make a series of three graphs, but not onto. If m and n are two different inte-
similar to Figure 2.3.4, for the two chord gers, then 2m and 2n are different as well—​but
progressions from Exercises 1.3.11 and 2.2.5. the function does not map onto ℤ because an
Your graphs should show four voices, but odd integer such as 3 is not 2n for any integer
depending on how you initially wrote the n. This function may be converted to an onto
progressions, there may be unison doublings function, however, by a simple modification. All
in some chords. Are irregularities visible even integers occur as values of the function,
when the diatonic progression is graphed in so if we define f: ℤ → ℤeven rather than f: ℤ → ℤ,
chromatic
   space? the function is now both one-​to-​one and onto.

48 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 2.3.4 Graphs of ascending 5–​6 sequences: (a) a diatonic sequence graphed in diatonic pitch
space; (b) a chromatic sequence graphed in chromatic pitch space; (c) a diatonic sequence graphed in
chromatic pitch space

In general, the set of values actually taken by a originally specified, but by definition every func-
function is called the range of the function. That tion maps onto its range.26 This example also
is, the range of a function f: S → T is the set {f(x) | shows other ways in which the sets involved in
x ∈ S}; this may be a smaller set than the set T the definition of a function are a crucial part of

Sets, Functions, and Relations • 49

FIGURE 2.3.6 (a) A function that is not one-​to-​one; (b) a function that is not onto

the definition. If the same formula f(x) =​2x is is the function g: ℤ → ℤ defined by g(n) =​−n. A
taken as the definition of a function from ℝ to one-​to-​one, onto function f: S → T relates the
ℝ rather than from ℤ to ℤ, this new function is sets S and T symmetrically, because in this case
both one-​to-​one and onto. Some other sets—​the for each y in T there is one and only one x in S for
set of US presidents, for example—​could not be which f(x) =​ y. Such a function matches values of
the domain of a function defined by this equa- x exactly with values of y, and the function itself
tion at all. may be called a one-​to-​one correspondence; such
For an example of a function that is onto functions are also known as bijections, or bijec-
but not one-​to-​one, let T be the set {0, 1}, tive functions.
and consider the function g: ℤ → T defined as A bijection can be thought of as an “isomor-
follows:27 phism of sets.” A bijective function f: S → T can
exist only if the sets S and T are of equal cardinal-
ity, and in fact the construction of a one-​to-​one
0 if n is even,
g ( n) =  correspondence is a standard way to demon-
1 if n is odd. strate that two sets are the same size. When we
later give precise definitions of isomorphism for
Obviously this g maps many integers to the interval spaces and other types of structures, a
number 0 and many others to 1. bijection of sets will always be the first require-
Suppose S and T are finite sets and f: S → T is a ment for the isomorphism, supplemented by
one-​to-​one function. Because different elements other conditions involving intervals or other rel-
of S always map to different elements of T, it fol- evant structural details.
lows that the set T must be at least as large as S. For a bijective function, it is possible to work
That is, a one-​to-​one function f: S → T can exist backward, recovering the unique value of x
only if #T ≥ #S. corresponding to any given y, as illustrated in
Figure 2.3.9. If f: S → T is a bijection, we may
Exercise 2.3.7 Complete the following define a new function g: T → S so that for each
statement, analogous to the above: “If S and y in T, g(y) is the unique x in S such that f(x) =​
T are finite sets, an onto function f: S → T can y. This function g is called the inverse function
exist
   only if ….” of f and denoted f −1 (pronounced “f-​inverse”).
Inverse functions have the characteristic prop-
Exercise 2.3.8 Our first example of a function erty that when combined with the original func-
that is one-​to-​one but not onto was the tion, they map back to the starting point; that
function f: ℤ → ℤ defined by f(n) =​2n. The set is, f −1(f(x)) =​ x for every x in S, and f(f −1(y)) =​ y
ℤ in that example is an infinite set. Consider for every y in T. Note carefully the arrangements
12-​note chromatic pitch-​class space pc. Is it of functions and arguments in these two equa-
possible to construct a function f: pc → pc that tions: the function f is applied to the argument x
is one-​to-​one but not onto? Onto but not one-​ (an element of S) but not to y (an element of T),
to-​one? Formulate a general principle about while f −1 is applied to y but not to x.
mappings
   f: S → S on a finite set S. If f is a bijection and a formula for f(x) is
given, it may be possible to solve for x, obtaining
Many functions are both one-​ to-​
one and a formula for f −1(y). In the case of the bijection
onto. A simple example mentioned above is the f: ℝ → ℝ, f(x) =​2x, solving the equation y =​2x
function f: ℝ → ℝ defined by f(x) =​2x; another for x gives x =​ y ∕ 2, so the inverse function is given

50 • E x p l o r i n g M u s i c a l S p a ces
 Exercise 2.3.11 An inverse function cannot
be defined unless the original function is both
one-​to-​one and onto. Reconsider the two non-​
bijective functions below (both introduced
previously), and describe the problems you
encounter if you attempt to define an inverse
for either of them:
FIGURE 2.3.9 The inverse of a bijective function
(a) f: ℤ → ℤ defined by f(n) =​2n.
0 if n is even,
by f −1(y) =​ y ∕ 2. This should make intuitive sense: (b) g: ℤ → {0, 1} defined by g (n) = 
the inverse of multiplying by 2 is dividing by 2.  1 if n is odd.
For the bijection g: ℤ → ℤ, g(n) =​−n, setting   
m =​ g(n) =​−n and solving for n yields n =​−m, so Exponential and logarithmic functions are
g−1(m) =​−m. The functions g and g−1 are both among the most important functions in math-
defined on the set of integers ℤ, and the formu- ematics. Exponential functions are defined by
las g(n) =​−n and g−1(m) =​−m show that g and g−1 formulas such as f(x) =​2x (2 raised to the xth
are the same function in this case: to undo tak- power), f(x) =​10x, or more generally f(x) =​ bx,
ing the negative of a number, you simply take where b is a positive constant, the base of the
the negative again. Functions that are their exponential. (Negative bases are problematic
own inverses are called involutions. An involu- when the exponent is not an integer, so we con-
tion can act as a toggle: successive applications sider exponential functions only for positive
of the function map back and forth repeatedly values of b.) The argument x may be any real
between the same two elements (such as n and number (positive, negative, or zero), and the
−n). Involutions are common in musical applica- value bx is always positive; in fact, every expo-
tions, and we will see many examples. nential function is a bijection from ℝ to ℝ>0.
The inverse function of the exponential function
Exercise 2.3.10 Seven functions are listed f(x) =​ bx is the logarithmic function g(y) =​logb(y);
below. Determine whether each function is that is, logb(y) (the logarithm of the number y
one-​to-​one, onto, both, or neither. (Each of to the base b) is defined to be the unique x such
the four possibilities occurs at least once in the that bx =​ y.
list.) If a function is not onto, what is its range? Standard texts in pre-​ calculus mathemat-
If a function is a bijection, what is its inverse ics, calculus, and mathematical analysis detail
function? numerous important properties of exponential
and logarithmic functions. Although our need
(a) f: ℝ → ℝ, f(x) =​ x +​3. for them will be limited, these functions deserve
(b) f: ℝ≠0 → ℝ≠0, f(x) =​1 ∕ x. mention here because they describe in a precise
(c) f: ℝ → ℝ, f(x) =​|x|. (The notation |x| way the relationship between pitch space and
denotes the absolute value of the number x. If frequency space, noted informally in Chapter 1.
x ≥ 0, then |x| is the same as x. But if x < 0, Suppose x is a note in pitch space (either discrete
then |x| is −x, the corresponding positive pitch space pitch or continuous pitch space
number. For example, |3| =​|−3| =​3.) cpitch), numbered from C4 =​0 as in Figure
(d) f: ℝ → ℤ, f(x) =​ ⎣x⎦. (The notation ⎣x⎦ 1.1.1. Tuned in A-​440 equal temperament, the
denotes the greatest integer in x, that is, frequency associated with the note x is given by
the largest integer n such that n ≤ x. For  x −9 
example, ⎣π⎦ =​3, and ⎣−3.68⎦ =​−4.) 
y = f ( x ) = 440 ⋅ 2 12  .


(e) f: ℝ>0 → ℝ>0, f(x) =​ x .

(f) f: ℤ>0 → ℤ>0, f(n) =​the nth prime number. Conversely, if y is a frequency in frequency space
(For example, f(1) =​2, f(2) =​3, f(3) =​5, ….) freq (Figure 1.1.5), then the corresponding
(g) f: ℤ → {0, 1, …, 6}, f(n) =​the remainder pitch number in continuous pitch space is
when n is divided by 7. (For example,
f(16) =​2, f(35) =​0, and f(−11) =​ 3—​the
last because −11, being equal to −2 · 7 +​3, ( ) + 9.
x = g ( y ) = 12 ⋅ log 2
y
440
  exceeds a multiple of 7 by 3.)

Sets, Functions, and Relations • 51

FIGURE 2.3.12 Relationships between pitch and frequency: (a) frequency as an exponential function
of pitch; (b) pitch as a logarithmic function of frequency

Graphs of the functions f and g are shown in g(f(x)) =​ x for every x in pitch, and that
Figure 2.3.12. As is characteristic of inverse func-   f(g(y)) =​ y for every y in freq.
tions, the two graphs are mirror images of each
other, flipped about a diagonal line. The shape of The above discussion of inverse functions
the exponential/​logarithmic curve makes pos- included expressions such as f −1(f(x)), in which
sible the familiar phenomenon already noted in the output of one function (f ) becomes the input
Chapter 1: notes that are equally spaced in pitch for another (f −1). This illustrates an important
space, such as the octave-​related Cs shown in process known as composition of functions, illus-
Figure 2.3.12a, become more widely spaced in trated in Figure 2.3.14. In the most general set-
the upper registers of frequency space.28 ting, composition of functions involves three
sets, S, T, and V, and two functions, f: S → T and
Exercise 2.3.13 Use the above formulas to g: T → V. The first function must map into the
answer the following questions. A calculator will set that is the domain of the second function (or
be needed for (c) and (d).29 a subset of that domain). In this case, if x ∈ S,
then y =​ f(x) is an element of T, and therefore z =​
(a) Verify that the formula for f(x) gives the g(y) =​ g(f(x)) (“g of f of x”) belongs to V. So we can
frequency 440 Hz for the note A4, and that define a single function h: S → V by the equation
the formula for g(y) gives the note A4 for the h(x) =​ g(f(x)). This h combines the action of both
frequency y =​440. f and g into a single function, called a composite
(b) What is the frequency of the note A3? function, or the composition of the original two
(c) What is the frequency of middle C? functions.
(d) A note on the piano has frequency 1396.91 Composition of functions is conceptually sim-
Hz. What note is this? ple enough, but it leads to a thorny notational
(e) The numbers 2, 9, 12, and 440 all appear conundrum known as the orthography problem. If
in both of the above formulas. Explain the h(x) =​ g(f(x)), many mathematicians call h “the
musical relevance of each of these numbers, composition of g and f,” commonly symbolized
as well as of the quantities (x − 9) ∕ 12 and h =​ g ○ f. In this notation g appears before f, as
y ∕ 440 appearing in the formulas. it does in the expression g(f(x)). But in the set-
(f) Use the above formulas and the algebra ting this notation describes, f is conceptually
of exponential and logarithmic functions prior to g: one must calculate f(x) before apply-
to verify that the functions f and g are ing g to that result. Consequently when reading a
inverses. That is, you should show that notation such as g ○ f or g(f(x)), we find ourselves

52 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 2.3.14 Composition of functions

reading from right to left in order to follow the that if f: S → T is a bijective function and f −1:
chronology of the operations. Accordingly, such T → S is its inverse function, then f ● f −1 =​ E. To
notations are referred to as right-​to-​left functional be precise, we should note that each set has its
orthography (or sometimes just left orthography). own identity function, so we should distinguish
In some musical applications a number of ES, the identity function on the set S, from ET,
transformations are composed in succession, the identity function on T. The composite func-
and in such situations the incorporation of a tion f ● f −1 maps from S to S, ultimately mapping
string of symbols in right-​to-​left orthography each element of S to itself, so this composition
within a larger left-​to-​right narrative becomes is ES, not ET.
ever more confusing. For this reason we will
usually adopt a different convention, using the Exercise 2.3.15 Assume as above that f: S → T
symbol ● to denote composition of functions is a bijective function and f −1: T → S is its
in left-​to-​right functional orthography (or right inverse function; let ES and ET denote the
orthography), which follows the conceptual identity functions on the two sets. Three pairs
chronology of the composition. That is, if h(x) =​ of statements are presented below. In each case
g(f(x)), we write h =​ f ● g. The solid black circle ● one of the two statements is correct; identify
may be pronounced as the word “then”; that is, the correct statements.
f ● g may be read as “f-​then-​g” or as “f followed
by g.” The composite function that we symbolize (a) Which is correct: f −1 ● f =​ ES or f −1 ● f =​ ET?
by f ● g is precisely the same function that others (b) Which is correct: ES ● f =​ f or ET ● f =​ f ?
may write as g ○ f; only the orthography is differ- (c)  Which is correct: f ● ES =​ f or f ● ET =​ f ?
ent. Often it will be convenient to suppress the
operation symbol entirely, writing simply fg for f In everyday life we are accustomed to the
● g. When no symbol for the composition of func-
idea that the order in which we do things mat-
tions is present, we follow the convention that ters. The instruction “go straight three blocks,
left-​to-​right orthography is always understood: fg then turn right” will take you to a different place
always means “f-​then-​g,” never “g-​then-​f.” This from “turn right, then go straight three blocks,”
convention too can lead to some awkwardness, and it is surely important to take the lid off the
as in equations such as z =​(fg)(x) =​ g(f(x)), but pot before you pour in the soup ingredients. On
the confusion can often be averted by recasting the other hand, whether you add the potatoes
the statement in strictly left-​to-​right fashion, or the carrots first may not matter much, and
for example via the arrow notation x fg → z. 30 the instruction “go three blocks north, then two
If S is any set, we can define a function E: blocks east” will often take you to the same place
S → S by the equation E(x) =​ x. This E is called as “go two blocks east, then three blocks north.”
the identity function on S.31 The identity func- In the realm of functions, compositions f ● g and
tion doesn’t actually do anything, as it merely g ● f are generally different, but they may be the
maps every element to itself. Identity functions same in some cases. If they are the same, we
are useful in somewhat the same way that the say that f and g commute. The following exercise
number 0 is useful: they arise in calculations and illustrates the point that some pairs of functions
are needed in order to complete a logical system. commute while others do not, along with an
The above discussion of inverse functions shows additional important observation: to determine

Sets, Functions, and Relations • 53

whether two functions are equal it may not be Exercise 2.3.18
sufficient to check their values at just one or two (a) Let f be the mapping from pitch (chromatic
arguments. pitch space) to pc (chromatic pitch-​
class space) that maps every pitch to its
Exercise 2.3.16 corresponding pitch class. For example,
(a) Let f: ℝ → ℝ and g: ℝ → ℝ be defined by f(E5) =​E. Is f one-​to-​one? Is it onto?
f(x) =​ x2 and g(x) =​1 − x. Let h denote the (b) Let g be the mapping from pc to pitch that
composite function f ● g, and let k denote maps every pitch class to the corresponding
the composite function g ● f. Calculate pitch in the fourth (middle C) octave. For
the following values of these composite example, g(E) =​E4. Is g one-​to-​one? Is
functions: h(0), k(0), h(1), k(1), h(2), k(2). it onto?
Are h and k the same function? That is, do f (c) Is g the inverse function of f ? Is it true that
and g commute? If possible, write formulas x fg→ x for every pitch x in pitch? Is it
true that y  → y for every pitch class y in
gf
for h(x) and k(x), and verify your answer to
the previous question algebraically. pc? (For f and g to be proper inverses, both
(b) Repeat part (a) for the functions f(x) =​4x − 6 statements must be true.)
  and g(x) =​5x − 8. (d) Is it possible to construct a bijection
  between the sets pitch and pc?
Exercises 2.3.17–​2.3.19 below introduce sev-
eral musical transformations. All will be studied Exercise 2.3.19
at greater length later, but these exercises offer (a) Let triad denote the set of 24 major and
quick overviews of some of their properties as minor triads. Let P denote the parallel
mathematical functions. transformation on triad, the function from
triad to triad that maps every triad to the
Exercise 2.3.17 In chromatic pitch-​class triad of opposite mode with the same root.
space pc, T3 denotes upward transposition by For example, P maps C major to C minor,
3 semitones, so for example T3(C) =​E♭. Also, and also maps C minor to C major. Is P
I denotes inversion about C, so for example one-​to-​one? Is it onto? What is the inverse
I(D) =​B♭. As functions from pc to pc, are function of P?
these transformations one-​to-​one? Are they (b) Let R denote the relative transformation,
onto? Calculate the results of applying the which maps every triad to the triad of
composite transformation I ● T3 (“I-​then-​T3”) opposite mode representing a key with the
to the pitch classes C and D, and also the same key signature. For example, R maps C
composite transformation T3 ● I. Are there any major to A minor, and maps C minor to E♭
pitch classes for which I ● T3 and T3 ● I both major. Is R one-​to-​one? Is it onto? What is
yield
   the same result? the inverse function of R?
(c) P and R are transformations of great
The notation in the previous exercise is likely importance in neo-​Riemannian theory.
to look somewhat familiar to readers who have A third fundamental neo-​Riemannian
studied pitch-​class set theory. Note, however, transformation, L, takes its name from
that I ● T3 is the transformation that is com- Riemann’s word Leittonwechsel, there
monly written “T3I” in set theory, in right-​to-​left being no comparable term in common
orthography. The latter notation contradicts the use in English.32 Applied to any major
left-​to-​right convention followed here, whereby triad, L yields a minor triad whose root
“T3I” means T3 ● I. It should also be noted that is a major third higher than the original
T0, “transposition by 0 semitones,” is simply triad; applied to a minor triad, L yields a
another name for the identity function on pitch-​ major triad whose root is a major third
class space, a function that could also be denoted lower. For example, L maps C major to E
Epc. When we are working in pitch-​class space, T0 minor, and maps C minor to A♭ major. Is L
will be our usual name for this identity function. one-​to-​one? Is it onto? What is the inverse
Transposition and inversion operators can also function of L?
be defined in pitch space and many other musi- (d) What is the result of applying the composite
cal spaces; in pitch space the same notation T0 function PR (that is, P ● R, or “P-​then-​R”) to
denotes the identity function Epitch. a C-​major triad? To a C-​minor triad? What

54 • E x p l o r i n g M u s i c a l S p a ces
 is the result of applying RP to each of these functions to ordered sets, as when transposing a
triads? Are PR and RP the same function? twelve-​tone row by T3.
What is the inverse function of PR? Answer
the same questions for PL and LP, and for Exercise 2.3.20
  RL and LR. (a) By the means just described, transposition
operators may be applied to triads. For
The importance of P, L, and R in neo-​ example, T3 maps the C-​major triad {C, E,
Riemannian theory springs from a combination G} to the E♭-​major triad {E♭, G, B♭}, and also
of many remarkable properties to be explored maps C minor to E♭ minor. As a function
later, most completely in Chapter 8. For exam- defined on the set triad, is T3 one-​to-​one?
ple: each of these transformations always maps Is it onto? What is the inverse function of
a triad to another triad with which it shares T3? Is T3 the same function as PR?
two common tones, and in fact P, L, and R are (b) What is the result of applying PT3 to a
the only ways to move from triad to triad while C-​major triad? To a C-​minor triad? What
preserving two common tones. Some of these is the result of applying T3P to each of
properties are related in fundamental ways to these triads? Are PT3 and T3P the same
Riemann’s ideas of harmonic dualism, to which function?33 Answer the same questions for
P, L, and R trace their heritage; we can justifiably   RT3 and T3R, and for LT3 and T3L.
call them Riemannian transformations (a term to
be given a precise definition in Chapter 8), not Now suppose that f: S → T is a function and
feeling an obligation to burden them with the B is a subset of T. It is common to write f −1(B)
more cumbersome adjective neo-​Riemannian at for the set {x ∈ S | f(x) ∈ B}. This set, a subset of
every appearance. This terminology should not S, consists of all elements of S that get mapped
be taken to imply, however, that the modern by the function f into the subset B; it is called
conception of these transformations is indis- the inverse image of B under f. For example, let
tinguishable from Riemann’s own conception; f: pitch → pc be the function from Exercise
indeed, Riemann’s practices differ from con- 2.3.18a, which maps every pitch to its corre-
temporary neo-​Riemannian theory and analysis sponding pitch class. If B is the pitch-​class set
in a number of important ways, some of which {E♭, E, G}, then f −1(B) is the pitch set {…, E♭3, E3,
were suggested in the discussion of tonnetzes in G3, E♭4, E4, G4, E♭5, E5, G5, …}, consisting of all rep-
Section 1.4 (where we were more insistent on dis- resentatives of the three pitch classes in B in all
tinguishing Riemannian from neo-​Riemannian). registers.
Suppose that f: S → T is a function and A is
a subset of S. In this situation it is common to Exercise 2.3.21
write f(A) for the set {f(x) | x ∈ A}. This set, a sub- (a) In the example just considered, f −1(B) is
set of T, consists of all the values taken by the a much larger set than B (in fact, f −1(B)
function f on elements of A; it is called the image is an infinite set while B is finite); this is
of A under f. To write f(A) for this set is a slight possible because the function f is not one-​
abuse of notation—​the function f is technically to-​one. It may also happen that f −1(B) is a
applied not to the object A (which is not an ele- much smaller set than B. Give an example,
ment of the domain of f) but to elements of the using the function g: pc → pitch from
set A, which are then collected to form another Exercise 2.3.18b. Also give an example of a
set—​but it rarely leads to confusion and is often nonempty set B for which g−1(B) =​ ∅.
very convenient. In music, it is this convention (b) If f: S → T is a function and B ⊆ T, the set
that allows us to suppose that the same transpo- f −1(B) may always be defined as above,
sition operators that are defined on pitch classes even if the inverse function f −1 does not
also act on pitch-​class sets. Knowing, for exam- exist. (Recall that f −1 exists only if f is
♭ ♯
ple, that T3(E ) =​F , T3(E) =​G, and T3(G) =​B , we♭ both one-​to-​one and onto.) If f −1: T → S
♭ ♯ ♭
may therefore write T3({E , E, G}) =​{F , G, B }. By does exist, however, we now have two
this ruse we understand that operators such as different meanings for the notation f −1(B):
T3 are well-​defined functions not only on pitch-​ it may refer either to (1) the inverse image
class space pc but also on the set of all subsets of of the set B under the function f, or to (2)
that space—​a set we may call pcset, the set of the image of the set B under the function
all pitch-​class sets. In a similar way we may apply f −1. Fortunately, in this situation sets (1)

Sets, Functions, and Relations • 55

 and (2) are always the same. Illustrate on musical spaces. The interval function on
this property using the function f =​ T3 on chromatic pitch space is a function int: pitch ×
pitch-​class space and the pitch-​class set pitch → ℤ that measures intervals in semitones;
  B =​{E♭, E, G}.34 for example, int(C4, E♭4) =​3.
In Section 1.1 we remarked that a labeling of
Occasionally we will encounter functions that a space may be regarded as a mapping between
take two or more arguments, written in a form the space and a set of possible labels for that
such as f(x, y) =​ z rather than simply f(x) =​ y. Such space. In some situations it is convenient to con-
a “function of two variables” is readily accom- sider functions from the space S to a set of labels
modated in the theory of functions outlined L, while in other circumstances a function from
above, by regarding the ordered pair (x, y) as a L to S may be more intuitive. We refer to a map-
single entity. Specifically, (x, y) is an element of ping of either kind as a label function for S. As an
a Cartesian product S × T, where S and T are sets example, review the picture of pitch-​class space
such that x ∈ S and y ∈ T. We may therefore take pc from Figure 1.2.3. This figure shows two sets
S × T (or a subset of some such Cartesian prod- of labels for the elements of pc, one using pitch-​
uct) as the domain of the function, and write f: class numbers and the other using note names.
S × T → V for some appropriate set V. Functions In order to understand the relationship between
of more than two variables may likewise be the two sets of labels it is important to maintain
defined on Cartesian products of a correspond- a careful distinction between the labels and the
ingly larger number of sets; for example, the objects they represent: either a number or a note
action of a function f: S1 × S2 × S3 × S4 → T may be name can stand for a pitch class, but neither one,
written f(x1, x2, x3, x4) =​ y. strictly speaking, is a pitch class.
Examples of functions of two variables The numerical labels may be said to define a
include the standard binary operations of arith- label function f: L → S from the set L =​{0, 1, …,
metic. Addition, for example, may be regarded 11} (the integers mod 12) to the space S =​ pc.
as a function f: ℝ × ℝ → ℝ via the equation (The integers mod 12 will be described in Section
f(x, y) =​ x +​ y. The operation symbol +​signi- 2.5, where we will make the point that integers
fies the function here, even though it appears mod 12 are not precisely the same thing as the
between the two arguments rather than before ordinary integers 0, 1, … , 11—​but this dis-
them. The binary operations of subtraction, mul- tinction need not concern us at the moment.)
tiplication, and division are also functions of two Carefully observing the distinction between
variables. The symbol −, however, also has a sec- objects and labels, we may say, for example, that
ond usage, syntactically distinct from the binary this f maps the number 3 to the pitch class 3. This
operation of subtraction (f(x, y) =​ x − y): the same function is a bijection between L and S, so it has
symbol is used also in the usual notation for an inverse function f −1: S → L, which maps the
negative numbers, where it denotes a function pitch class 3 to the number 3.
of a single variable (f(x) =​−x)—​a unary operation. The note-​name labels imply a different label
The impossibility of dividing by 0 means that function g: Lʹ → S, which maps from a different
the equation f(x, y) =​ x ∕ y does not define a func- set of labels, Lʹ =​ sl (signed letter space, from
tion on the entire Cartesian product ℝ × ℝ as the Figure 1.2.11), into the same space S =​ pc. This
other binary operations do, but on the slightly label function g is not one-​to-​one, because for
smaller domain ℝ × ℝ≠0. example the signed letters D♯ and E♭ are mapped
An example of a binary operation whose argu- to the same pitch class (3) in pc. Indeed, because
ments are not numbers is afforded by composi- Lʹ =​ sl contains more than twelve elements (it
tion of functions. If f and g are functions from a is theoretically infinite), no one-​to-​one function
set S to the same set S, then f ● g is another such from Lʹ to the space S =​ pc is possible. Because
function. In this way we may regard ● as a binary it is not one-​to-​one, g has no inverse function.
operation defined on the set of functions from Figure 1.2.3 shows only one of the possible note-​
S to S—​a function whose arguments are them- name labels for each note, so the figure does not
selves functions. This operation will be ubiqui- describe the label function completely.
tous in our study of musical transformations. Many different label functions are used in the
Another important musical application of figures in this book, but we will rarely mention
functions of two variables will emerge in the them explicitly. Such functions will be of impor-
study of intervals. Beginning in Chapter 5, we tance in Chapter 7, however, where we shall see
will define an assortment of interval functions that each element of an interval space gives rise

56 • E x p l o r i n g M u s i c a l S p a ces
to a particular label function (in this case a func- is precisely 440 ∙ 21 ⁄ 12 Hz, and h(A5) =​ f(21) (one
tion from S to L), and in Chapter 11, in which it octave higher) is precisely 440 ∙ 2 =​880 Hz. As
will sometimes be convenient to apply multiple a function from signed registral letter space to
labels to the points in a voice-​leading space. the positive real numbers, is h one-​to-​one? Is it
onto? If it is not onto, describe the range of h
Exercise 2.3.22 mathematically.
  
(a) For the function g: Lʹ → S just described,
where Lʹ =​ sl and S =​ pc, what is the value Exercise 2.3.25
of g(F♯ ♯)? What is g(A♭ ♭ ♭)? What is g(C♯ ♯ ♯)? (a) To define Pythagorean tuning precisely it
Express each answer as a pitch-​class is helpful to consider a different picture
number. of signed registral letter space—​a tonnetz
(b) Suppose we use Figure 1.2.3 to define a generated by octaves and fifths. Start with
function gʹ: S → Lʹ, where S =​ pc and Lʹ =​ sl. the line of fifths (Figure 1.3.3). Assign a
For example, gʹ maps pitch class 3 to the register to each note so that A4 is in the
signed letter D♯, not E♭, because D♯ is the center and every interval along the line is a
label that appears in the figure. Explain why perfect fifth; the line of fifths now reads …,
gʹ is not an inverse function for g. (Some C3, G3, D4, A4, E5, B5, F♯6, … . Add a second,
related functions will be described in more vertical dimension to the figure, arranged
  detail in Sections 2.4 and 2.6.)35 in octaves, so that the central A4 is in the
middle of a column that reads (from bottom
Exercise 2.3.23 Each of the Figures 1.1.6 (the to top) …, A2, A3, A4, A5, A6, …, and every
space gpitch), 1.2.8 (gpc), and 1.3.3 (spc) other note of the original line of fifths
includes two sets of labels. Each labeling may be is similarly surrounded in the vertical
considered as a label function of the form f: L → dimension by its octave relatives. This
S for some set L of possible labels. Describe the octave-​and fifth-​generated arrangement of
domain (L) of each of these label functions. Are signed registral letter space may be called
all  
of these functions one-​to-​one? Pythagorean pitch space.
(b) Every note name of signed registral letter
It was mentioned in Section 1.1 that a tun- space appears exactly once in Pythagorean
ing may be modeled as a mapping from a syntac- pitch space. If A4 is at coordinates (0, 0),
tic space into a physical space, a function that then E5 is at (1, 0) and A5 is at (0, 1). What
assigns a physical frequency to every note name. are the coordinates of B♭4? What are the
Such tuning functions take the form f: L → freq; coordinates of A♯4?
they may be considered a special case of label (c) In Pythagorean tuning the frequency ratio
functions, mapping into frequency space rather of an octave is 2, and the frequency ratio of a
than into some other space. The final three perfect fifth is 3 2 . If we take f(A4) again to be
rather abstract exercises of this section investi- 440 Hz, then f(A5) will be exactly 440 · 2 =​
gate three tunings defined in this way. 880 Hz, and f(E5) will be exactly 440 · 3 2 =​
660 Hz. In general, the note at coordinates
Exercise 2.3.24 Equal-​tempered tuning is (x, y) will have a frequency of exactly 440 ·
described in one sense by the function f that (3 2 )x · 2y Hz. What is the frequency of B♭4?
was graphed in Figure 2.3.12a. This function What is the frequency of A♯4? Calculate the
is defined on all of continuous pitch space interval between B♭4 and A♯4 as a frequency
cpitch, but in particular it is defined on the ratio, and estimate its size in pitch space.
discrete space pitch. A tuning function defined (d) As a function from srl to ℝ>0, is f one-​to-​
on the note names of signed registral letter one? Is it onto? If it is not onto, describe
space srl (Figure 1.1.11), however, is also of   the range of f.36
interest. There is a function g: srl → pitch that
describes the usual mapping from note names
to pitch numbers (g(C4) =​0, g(D♯4) =​ g(E♭4) =​3, Exercise 2.3.26 Just tuning adds the pure
and so on); the desired tuning function on major third (frequency ratio 5 4 ) to the list of
srl is then the composition h =​ g ● f, where f: privileged intervals, alongside the octave and
pitch → freq is as above. This definition takes perfect fifth. Signed registral letter space is
h(A4) =​ f(g(A4)) =​ f(9) =​440 Hz as the basis for not sufficient for defining just tuning, because
the tuning; h(B♭4) =​ f(10) (one semitone higher) just intervals can lead to two different tunings

Sets, Functions, and Relations • 57

for the same note name (as was observed are not numbers; they are statements that may
previously in the discussion of the Riemannian be either true or false. A binary relation may be
tonnetz in Chapter 1). The appropriate space thought of as a property of elements in pairs—​a
for defining just tuning—​just pitch space—​is condition that any pair of objects may or may
a three-​dimensional tonnetz, generated by not satisfy. A relation symbol such as =​or > func-
octaves, perfect fifths, and major thirds. tions syntactically as the verb of the statement
One can form such a space by starting with in which it appears.
the Riemannian tonnetz (Figure 1.4.14) and The symbol ~ is commonly used to stand
supplying a third dimension for the octaves. for an abstract binary relation. If ~ is a binary
Define an xy coordinate system for the relation on a set S, then for any elements x and
Riemannian tonnetz as in Exercise 2.2.4b, but y in S the statement x ~ y at least makes sense;
with its origin (0, 0) at A4. The x coordinate whether it is true or false depends on the spe-
measures distance in perfect fifths away cific elements x and y. (The statement x ~ y
from A4, while y counts major thirds. Add a z may be pronounced “x is related to y,” though
coordinate in the third dimension to account many people informally pronounce the relation
for octaves. Thus E5 is at (1, 0, 0), C♯5 (with an symbol, saying for example “x twiddle y.”) Well-​
underscore) at (0, 1, 0), and A5 at (0, 0, 1). It is known examples of binary relations defined on
difficult to draw this three-​dimensional tonnetz numbers (ℝ, ℚ, or ℤ) include =​, ≠, <, >, ≤, and
clearly, but try to draw a small part of it. Every ≥. One can define many other sorts of relations,
note name of signed registral letter space, in however, based on virtually any property involv-
any octave and altered with any desired number ing two numerical variables. For example, we
of underscores or overscores, appears exactly could define x ~ y to mean that x2 − 9 ≤ y. In that
once somewhere in this tonnetz. If we take case the statements 1 ~ 3, 1 ~ 5, and 3 ~ 1 are all
f(A4) =​440 Hz as usual, then the frequency of the true, but 5 ~ 1 is false (which may be symbolized
note at coordinates (x, y, z) in just pitch space by writing 5 ≁ 1).
is 440 · (3 2 )x · (5 4 )y · 2z Hz. The frequencies of E5, Relations and functions are different things,
C♯5 , and A5 are therefore exactly 440 · 3 2 =​660 but some connections between them are appar-
Hz, 440 · 5 4 =​550 Hz, and 440 · 2 =​880 Hz, ent. A statement y =​ f(x) describes a particular
respectively. property of x and y, so in that sense a function
Calculate the coordinates, and the frequency, may be considered a special kind of relation. In
of each of the notes C♯5 (without an underscore), general a binary relation ~ may be identified
D♭5, and D♭5 in just pitch space. All should be with the set of all ordered pairs (x, y) of ele-
close to 550 Hz, the frequency calculated above ments for which the statement x ~ y is true. The
for C♯5, but each one should be slightly different.
last sentence of the preceding paragraph says, in
this interpretation, that the set identified with
The notation in the Riemannian tonnetz
indicates that C♯5 should be one syntonic
the relation ~ contains the ordered pairs (1, 3),
comma lower than C♯5. Estimate the musical
(1, 5), and (3, 1), but does not contain (5, 1). We
remarked in the previous section that a func-
size of the syntonic comma. Is the new tuning tion f may also be identified with a set of ordered
function f (a function from just pitch space pairs (x, y), namely those for which y =​ f(x). The
to frequency space) one-​to-​one? Is it onto? distinguishing feature of a function is that a
Describe
   the range of f.37 single x cannot be paired with two different y’s;
in a function, that is, no two ordered pairs may
have the same first component. There is no such
restriction on relations, as the ordered pairs
2.4 RELATIONS (1, 3) and (1, 5) in the above example illustrate.
In expressions such as 3 +​7 =​10 or 9 > 2 ∙ 4, Relations, like functions, can often be graphed.
the symbols +​and ∙ indicate binary operations, The ordered-​pair restriction for functions typi-
as discussed in the previous section. The sym- cally results in graphs composed of lines or
bols =​and >, on the other hand, are not opera- curves, but the graph of a relation such as >
tions. These symbols represent binary relations. frequently includes a large area of two-​dimen-
Operations are functions and map objects to sional space.
other objects: if x and y are numbers, then x +​ Various generalizations of the concept of rela-
y and x ∙ y are also numbers. But x =​ y and x > y tion are possible. There are some relations for

58 • E x p l o r i n g M u s i c a l S p a ces
which the two objects being related need not be definitions, give two more examples of the
elements of the same set. A familiar example is action of each operation, and for each relation
the element relation ∈ itself. In the statement x give two examples of pairs of elements for
∈ Y, x and Y are objects of two different kinds, which the relation is true and two examples for
one an element and the other a set. The most which it is false.
general binary relation, therefore, is defined not
on a single set S but on two possibly different (a) If x and y are notes in chromatic pitch-​class
sets S and T; as a set of ordered pairs, the rela- space pc, x ⊛ y means that x and y belong
tion is a subset of the Cartesian product S × T. In to the same whole-​tone scale. For example,
the case of the element relation ∈, the set T must E♭ ⊛ B is a true statement.
be a set of sets. (b) If x and y are notes in pc, x ⊛ y denotes
In a situation often encountered in musical the distance between x and y in fifths
applications, we may wish to define a transfor- space—​the number of steps separating the
mation—​ a function—​ only to find that some two notes on the circle of fifths, measured
technical detail prevents us from doing so, and whichever way around the circle results in
therefore a relation is the only option. For exam- the shorter path. For example, E♭ ⊛ B is 4.
ple, suppose we wish to define a transformation (c) If x and y are notes in continuous pitch space
f so that f(X) =​ Y is true whenever X is a fully cpitch, x ⊛ y is the note that lies exactly
diminished seventh chord and Y is the minor midway between x and y. For example, E♭4
triad to which X resolves by viio7–​i motion; ⊛ B3 is C♯4. Why is it necessary to specify
that is, we want f(bo7) =​c, f(c♯o7) =​d, and so on. continuous pitch space in this definition?
If chords are defined as pitch-​class sets, then (d) If x and y are notes in cpitch, x ⊛ y means
apparently f should be a function from the set S that the interval between x and y is one
of all diminished seventh chords to the set T of whole tone or less. For example, C♯4 ⊛ B3 is
all minor triads. The fact that each diminished   a true statement.
seventh chord has multiple possible chords of
resolution, however, means that no function Exercise 2.4.2 In the examples below, assume
with the specified behavior is possible: the state- first that the chords X and Y are understood
ments f(bo7) =​c and f(do7) =​e♭ cannot both be to be pitch-​class sets. Four attempts to define
true because bo7 and do7 are the same element of transformations are presented. Two of these are
S, while c and e♭ are two different elements of T. valid functions; two are only relations. Which are
The viio7–​i motion describes a relation on the sets the functions? For each function, what exactly is
S and T, but not a function. The distinction is an the domain? For the non-​functions, can you find
important one, because many established tech- some way to modify the definition to capture the
niques of transformation theory are not appli- essential information in a function?
cable to general relations. Sometimes it may be
possible to rectify the problem by redefining the (a) f(X) =​ Y is true whenever X is a minor triad,
objects, in this case considering the chords X and Y is a fully diminished seventh chord,
Y to be not simply sets of pitch classes but sets and Y resolves to X via viio7–​i motion. For
of spelled pitch-classes, as introduced in Section example, f(c) =​bo7. (This is like the example
1.2. The chords bo7, do7, fo7, and g♯o7 are identical discussed above, but with the roles of X and
as pitch-class sets but different as spelled pitch-​ Y interchanged.)
class sets—​but this modification may introduce (b) f(X) =​ Y is true whenever X is an augmented
an undesirable complexity (because, for instance, triad and Y is the major triad that results from
spc space is an infinite set). The moral is that it lowering the fifth of X. For example, f(C+​) =​C.
is important to be precise in defining the objects (c) f(X) =​ Y is true whenever X is an
with which we are working, to distinguish func- augmented triad and Y is the whole-​tone
tions from relations carefully, and to specify the scale containing all the notes of X. For
domains of functions clearly.38 example, f(C+​) =​{C, D, E, F♯, G♯, A♯}.
(d) f(X) =​ Y is true whenever X and Y are
Exercise 2.4.1 Four musical definitions for the dominant seventh chords and Y may be
symbol ⊛ are presented below. Two of these obtained from X by holding one common
are binary operations (functions) and two are tone, lowering two notes by one semitone,
binary relations. Which are which? In addition and lowering one note by two semitones.
to any examples that may be given with the   For example, f(G7) =​C7.

Sets, Functions, and Relations • 59

 Relations need not be binary: like functions, slice—​the same equivalence class. Likewise w
they may take any number of arguments. If A, B, and z are equivalent. But y and z are not equiva-
and C are points in Euclidean space, the state- lent; the fact that they appear close together in
ment “points A, B, and C lie on a line” defines the diagram is irrelevant. An equivalence class
a ternary relation, a relation among three vari- may contain many elements; we may say infor-
ables. Relations of a single variable—​unary rela- mally that all the elements in one class are equiv-
tions—​are possible as well. A property such as “n alent, but such a statement means technically
is a perfect square” defines a unary relation on that any two elements of the class are equivalent,
the integers, a statement that is true for some because equivalence relations are always offi-
values of n but false for others. cially binary relations.
An important category of binary relations, It is convenient to introduce the notation x
known as equivalence relations, offers a frame- as shorthand for “the equivalence class contain-
work for formalizing various concepts that go ing x.” One equivalence class may have multiple
by the name equivalence, useful in situations labels using this shorthand: in Figure 2.4.3, the
in which objects may be “the same” in some equivalence class labeled x could just as well be
meaningful way without being identical in all called y instead. In general, a statement x = ​y
respects. Familiar musical constructions such as (a statement about equality of equivalence classes)
octave and enharmonic equivalence, which come is true if and only if the statement x ~ y (a state-
into play in many of the spaces described in ment about equivalence of elements) is true. So,
Chapter 1, are examples of equivalence relations. in the figure, x = ​y , and likewise w = ​z , but y ≠ z.
Equivalence relations provide another perspec- The set of all equivalence classes is called the quo-
tive on these spaces and a general mathematical tient set of S modulo the relation ~, denoted S ∕ ~.
apparatus for describing the equivalence itself. That is, S ∕ ~ =​{x | x ∈ S}. In the figure, S ∕ ~ is a set
An equivalence relation ~ on a set S is shown of cardinality 4. The word modulo derives from
in schematic fashion in Figure 2.4.3. An equiva- modular arithmetic, an important special case
lence relation has the effect of breaking up the discussed in the next section.
set S into smaller sets, called equivalence classes, We may define a function that maps every ele-
in such a way that two objects in the same equiv- ment x in S to its equivalence class x in S ∕ ~. This
alence class are always “equivalent”—​ that is, function is called the quotient map on S modulo ~,
related to each other in whatever way the rela- denoted Q~. That is, Q~: S → S ∕ ~ is defined by the
tion specifies. In the figure, the set S has been equation Q~(x) = ​x . Because every equivalence
sliced into four equivalence classes. Two ele- class contains at least one element, the quotient
ments of S are related by ~ if and only if they map always maps S onto S ∕ ~. Quotient maps are
are in the same equivalence class.39 Equivalence not generally one-​to-​one, however; from the dis-
classes are always disjoint subsets of S (an ele- cussion in the preceding paragraph, Q~(x) =​ Q~(y)
ment cannot be in two of them), and their union if and only if x ~ y.
must be all of S (every element belongs to one Let us examine how these concepts apply in
of them). In the picture, the elements labeled x the musically familiar case of octave equivalence.
and y are equivalent because they are in the same In formalizing octave equivalence, we may con-
sider the set S to be either discrete or continu-
ous pitch space (pitch or cpitch), or possibly
generic pitch space gpitch or one of its diatonic
variants dpitch(n). We denote octave equiva-
lence by the notation ~O. The statements C3 ~O C4
and C3 ~O C5 are therefore true. In fact, all Cs, in
all registers, are related by ~O and form a single
equivalence class—​which is, of course, precisely
what we know as the pitch class C. Every equiva-
lence class of the relation ~O is such a pitch class,
consisting of pitches separated by whole num-
bers of octaves.
If we choose pitch as our original space, there
are twelve equivalence classes altogether; if we
FIGURE 2.4.3 An equivalence relation start instead with cpitch, there are infinitely

60 • E x p l o r i n g M u s i c a l S p a ces
many, and if we start with gpitch there are only pitch classes, and the quotient space is simply
seven. In each case, the equivalence classes may another representation of 12-​note pc space pc.
be identified with pitch classes of the appro- The quotient map QE: sl → pc is the label func-
priate type—​ elements of the corresponding tion called g in Exercise 2.3.22 and the preceding
pc space (pc, cpc, or gpc). The construction of discussion. The information discarded by this
pc space from pitch space, which was visual- quotient map is spelling information: QE maps
ized in Chapter 1 by bending the line of Figure the differently spelled signed letters G♯ and A♭ in
1.1.1 to form the circle of Figure 1.2.3, is there- sl to same pitch class in pc, the pitch class com-
fore an example of the formation of a quotient monly denoted “8” in pc set theory.
set (or quotient space). Using the notation just The same space pc therefore arises as a quo-
described, we may write pc =​ pitch ∕~O, cpc =​ tient space in two apparently quite different
cpitch ∕~O, and gpc =​ gpitch ∕~O. Each of these ways: as a quotient of chromatic pitch space
quotient spaces has a corresponding quotient pitch by octave equivalence, or as a quotient
map QO. For example, QO(G♯5) =​G♯; that is, QO of signed letter space sl by enharmonic equiva-
maps the pitch G♯5 in pitch to the pitch class G♯ lence. To see why this is so, it is useful to con-
in pc—​and of course it also maps every other sider srl, a space in which both octave and
G♯ in pitch to the same pitch class. (This quo- enharmonic distinctions may be made. In srl,
tient map is precisely the function f studied in signed notes are octave-​equivalent if they are
Exercise 2.3.18a.) In effect, QO discards register indistinguishable when octave transpositions
information, identifying octave-​ related notes. are ignored (for example, G♯4 and G♯6); signed
Generally a quotient map may be described as notes are enharmonically equivalent if they are
discarding any information that could be used indistinguishable when enharmonic spelling dif-
to differentiate elements within one equivalence ference are ignored (that is, if they represent the
class, thereby rendering those elements indistin- same note on the piano—​for example, G♯4 and
guishable. This is an important point: the value A♭4). The notes G♯4 and G♯6 are not enharmoni-
of the quotient construction lies in its ability to cally equivalent (they are not the same note on
group like objects together, but the information the piano), and G♯4 and A♭4 are not octave-​equiv-
that differentiates those objects is lost in the alent (octave transposition cannot make them
process. For this reason we should not imagine the same, because G♯ and A♭ are two different
that we must replace a set with a quotient set at signed letters).42 When octave equivalence ~O
every opportunity; as in the case of pitch space is applied to srl, the resulting quotient space is
and pitch-​class space, it is often desirable to have sl, but if enharmonic equivalence ~E is applied
both the original set and the quotient set avail- to srl, the quotient space is pitch. The resulting
able for appropriate uses.40 assortment of quotient spaces is listed below,
Enharmonic equivalence ~E differs from with one example of the action of the quotient
octave equivalence in its dependence on nota- map shown for each quotient space:
tion. Perhaps for this reason, formal aspects of
QO(G♯5) =​ G♯
enharmonic equivalence as an equivalence rela-
tion have received comparatively little scholarly srl ∕ ~O =​ sl ∈ sl
attention.41 Enharmonic equivalence is defined sl ∕ ~E =​ pc QE(G♯) =​ 8 ∈ pc
not on pitch space but on signed registral letter srl ∕ ~E =​ pitch QE(G♯5) =​ 20 ∈ pitch
space srl (Figure 1.1.11), signed letter space sl pitch ∕ ~O =​ pc QO(20) =​ 8 ∈ pc
(Figure 1.2.11), or spelled pitch-​class space spc
(the line of fifths from Figure 1.3.3, a reconfigura-
tion of sl). Exercises 1.1.12 and 1.2.12 suggested Regarding the last two sample actions above,
depictions of the equivalence classes in srl and when the pitches of pitch are numbered as in
sl. Enharmonic equivalence classes appear more Figure 1.1.1, the number associated with G♯5 is
complex in structure than octave equivalence 20, because G♯5 lies 20 semitones above middle
classes in these spaces, because octave equiva- C. Then QO maps the pitch 20 in pitch to the
lence is based on octaves of uniform size while pitch class 8 in pc via a simple calculation in
enharmonic equivalence depends upon the pat- mod-​12 arithmetic, to be explained in the next
tern of unequal step sizes in the diatonic scale. section.
The quotient space of signed letter space modulo These actions are summarized in Figure 2.4.4.
~E, however, is something familiar: the twelve The upshot of the figure, and of the discussion
equivalence classes may again be identified with above, is that pc is derived from srl by applying

Sets, Functions, and Relations • 61

FIGURE 2.4.4 Actions of the quotient maps QO and QE on an element of signed registral letter space

both equivalence relations ~O and ~E, and they (b) If you take the quotient space from part
may be applied in either order, resulting in the (a) and then apply octave equivalence,
two intermediate spaces sl and pitch. In fact, which space from Chapter 1 is the result?
we can obtain pc directly from srl by means of What element of this space is the result
a single composite equivalence relation ~OE. Two of applying the quotient map QO to the
notes in srl are related by ~OE whenever they note G5?
may be related by octave and enharmonic equiva- (c) Verify that the same quotient space from
lence in any combination; for example, G♯4 ~OE A♭6 (b) may be obtained from srl by applying
is true because G♯4 ~O G♯6 and G♯6 ~E A♭6. The the equivalence relations ~G and ~O in the
quotient srl ∕ ~OE is the space pc. The quotient reverse order. Make a diagram similar to
map QOE associated with the composite equiva- Figure 2.4.4 showing the actions of the
lence relation is precisely the composite function quotient maps QO and QG on the element
QO ● QE. The figure illustrates the fact that QO G♯5 of srl.
and QE commute: QO ● QE =​ QE ● QO.43 Whenever (d) The space obtained in parts (b) and (c)
two or more equivalence relations are defined on may be constructed directly from srl by
the same space, composite equivalence relations applying a composite equivalence relation
of this sort can always be formed, and the order ~OG. Describe what it means for two notes
in which the original equivalence relations are in srl to be related by ~OG.
applied is of no consequence. This principle will (e) Having considered composite equivalence
be of some importance in our study of voice-​lead- relations ~OE and ~OG, you may wonder
ing spaces in Part Three, in which several equiva- about the possibility of a composite relation
lence relations will be in play simultaneously. ~GE, or even a triple composite ~OGE.
Explain why a combination of generic and
Exercise 2.4.5 Another equivalence relation, enharmonic equivalence is not musically
mentioned briefly in Chapter 1, is generic useful.44
equivalence ~G. Two notes in srl are generically (f) In the tonnetz of Figure 1.4.1, what are the
equivalent if they are the same when accidentals   generic equivalence classes?
are ignored; the quotient map QG simply
strips all sharps or flats from a note name. For Exercise 2.4.6 Consider spelled pitch-​class
example, C3 ~G C♯3, D♭ ♭5 ~G D𝄪5, and QG(G♯5) =​G5. space spc (the line of fifths, Figure 1.3.3).
Which space from Chapter 1 arises as the
(a) Which space from Chapter 1 arises as the quotient space spc ∕ ~E? Recall that spc is
quotient space srl ∕ ~G? the same set as sl in a different geometric

62 • E x p l o r i n g M u s i c a l S p a ces
arrangement. Is the quotient space the same set (b) On the set S × T, define two equivalence
as  
sl ∕ ~E in a different arrangement? relations ~S (S equivalence) and ~T (T
equivalence) as follows. Two ordered pairs
Equality, always denoted by the symbol =​, is are S-​equivalent if they have the same first
automatically an equivalence relation on any set component (S component); two ordered
whatsoever. It is, in fact, an equivalence relation pairs are T-​equivalent if they have the
of a special kind, because every element is in an same second component (T component).
equivalence class by itself. (If x and y are two dif- For example, the statements (C, +​) ~S
ferent elements, then of course the statement (C, −) and (C, +​) ~T (A♭, +​) are both correct.
x =​ y is never true.) It follows that that there What are the ~S equivalence classes, and
are as many equivalence classes as there are ele- what are the ~T equivalence classes? How
ments in the set, and that the quotient set is as many equivalence classes of each type are
large as the original set. Another way of saying there? Verify that the elements of (S × T) ∕ ~S
this is to say that the quotient map Q=​ is a bijec- are in one-​to-​one correspondence with
tion. Quotient maps are always onto, but equal- the elements of S, and the elements of
ity is the only equivalence relation for which the (S × T) ∕ ~T are in one-​to-​one correspondence
quotient map is one-​to-​one. with the elements of T.
At the other extreme, the minimum possible (c) Suppose we identify the elements of (S × T) ∕ ~S
number of equivalence classes is 1, which hap- with the elements of S as described above.
pens in the case of universal equivalence, defined The quotient map QS associated with the
so that x ~ y is true for all x and y in a set. In this equivalence relation ~S may then be viewed
case the quotient set contains only a single ele- as a function QS: S × T → S. What element
ment. Equality is the finest possible equivalence of S is QS(C, −)? What is QS(E, +​)? With QT
relation on any set, in the sense that it distin- defined in a corresponding way, what are
guishes all elements from each other; universal QT(C, −) and QT(E, +​)? (In this situation QS
equivalence is the coarsest, as it recognizes no and QT are sometimes called the projections
distinctions at all. In general, fine equivalence from the Cartesian product S × T onto the
relations give rise to large quotient sets, and   original sets S and T, respectively.)
coarse equivalence relations to small quotient
sets. The most interesting examples of equiva- Several other equivalence relations are
lence relations fall between these extremes, so important in music theory besides those already
that some pairs of elements are related while mentioned. At least three different equivalence
others are not. relations on the collection of all pitch-​class sets
are commonly recognized: two sets may be con-
Exercise 2.4.7 In Section 2.2 we defined a sidered equivalent if they are related by trans-
kind of product of sets, the Cartesian product. position (T equivalence), or if they are related by
We have now defined a quotient of sets, via an any combination of transposition and/​or inver-
equivalence relation. The terminology suggests sion (TI equivalence), or if they have the same
that these two concepts are somehow related to interval-​class vector (interval-​content equiva-
multiplication and division, and are inversely lence).45 T equivalence and TI equivalence will be
related to each other. In fact, the sets involved considered in more general contexts in Chapters
in a Cartesian product can be recovered from 6 and 10.
the product by forming quotient sets. This
exercise shows how these relationships work in Exercise 2.4.8 Of the three relations T
a simple situation. equivalence, TI equivalence, and interval-​
content equivalence, which is the finest?
(a) Let S be the set {C, E, A♭} (a set of three    is the coarsest?46
Which
pitch classes), and let T be the set {+​, −},
interpreted as the triad qualities major (+​) Not all binary relations are equivalence rela-
and minor (−). An element of the Cartesian tions. The relation <, for example, is not an
product S × T is an ordered pair such as equivalence relation: there is no way to break
(E, +​), which may be identified with an E-​ down the integers, or the real numbers, into
major triad. List the six triads in S × T in an smaller equivalence classes in such a way that all
appropriately organized 3-​by-​2 array. (These elements within one class are related by < but ele-
six triads form a hexatonic system, to be ments in different classes are not. In general, for
explored in Section 3.1.) a binary relation ~ on a set S to be an equivalence

Sets, Functions, and Relations • 63

relation, the following three conditions must be To show that a relation is not an equivalence
satisfied for all elements x, y, and z in S: relation, one need only give a single example to
show that one of the conditions (R), (S), or (T)
(R) Reflexive property: x ~ x. fails. For instance, simply because the statement
(S) Symmetric property: if x ~ y, then y ~ x. 2 < 2 is false, we see that the relation < is not
(T) Transitive property: if x ~ y and y ~ z, reflexive, which by itself tells us that < cannot be
then x ~ z. an equivalence relation; nothing else needs to be
checked. For the record, though, symmetry also
Reflexivity, symmetry, and transitivity are proper- fails for <, because 2 < 3 is true while 3 < 2 is
ties of the relation ~, not properties of the set S or false. The relation < is transitive (if x < y and y
of specific elements x, y, and z. Using symmetry as < z, it does follow that x < z), but transitivity in
an illustration, either a relation is symmetric or it is the absence of reflexivity and symmetry means
not; if there is even a single pair of elements x and little; the relation has still failed two of the three
y for which x ~ y but y ≁ x, then the relation ~ is not required conditions for equivalence relation
symmetric. From the above discussion of equiva- status.
lence classes, and from Figure 2.4.3, it should be
clear why an equivalence relation must satisfy con- Exercise 2.4.10 Four binary relations are
ditions (R), (S), and (T). Symmetry, for example, described below. Two of them are equivalence
requires that if x and y are in the same equivalence relations and two are not. Which ones are the
class, then y and x are in the same equivalence equivalence relations? For the equivalence
class—​which is exactly the same thing. relations, verify conditions (R), (S), and (T),
What may not be so obvious is that condi- and describe the equivalence classes. For the
tions (R), (S), and (T), taken together, are suf- other relations, determine which (if any) of the
ficient to ensure that a binary relation ~ is an three conditions are satisfied and which are not,
equivalence relation. That is, if a relation ~ on a and give specific examples of the conditions
set S is reflexive, symmetric, and transitive, then that fail.
~ automatically partitions S into equivalence
classes as described above. For this reason, in (a) On the set ℝ of real numbers, the relation ≤
the mathematical literature an equivalence rela- (“less than or equal to”).
tion is often formally defined as any binary rela- (b) On the set ℤ of integers, the relation ≠ (“is
tion satisfying conditions (R), (S), and (T), from not equal to”).
which the properties of equivalence classes are (c) Let S be some collection of finite sets. If
then derived as consequences.47 X and Y are sets in S, define X ~ Y to be
To determine whether a given binary relation true whenever #X =​#Y—​that is, whenever
is or is not an equivalence relation, therefore, X and Y contain the same number of
it suffices to check conditions (R), (S), and (T). elements.
These checks are rarely difficult and sometimes (d) Let S be the set of all lines in the plane. If L1
obvious. For octave equivalence the appropriate and L2 are lines in S, define L1 ~ L2 to be true
statements are: whenever the lines L1 and L2 are parallel.
  Consider every line to be parallel to itself.
(R) Every pitch is octave-​equivalent to itself.
(S) If pitch x is octave-​equivalent to y, then y is Exercise 2.4.11 Five binary relations in
octave-​equivalent to x. musical contexts are described below. At
(T) If pitch x is octave-​equivalent to y and y least one of these is an equivalence relation,
is octave-​equivalent to z, then x is octave-​ while some others are not. Answer the same
equivalent to z. questions as in the previous exercise.

Considering that “octave-​ equivalent” means (a) In continuous pitch space cpitch, define x
“separated by a whole number of octaves,” the ~ y to be true whenever the pitches x and y
truth of all three statements should be clear. are separated by an interval of 3 semitones
or less.
Exercise 2.4.9 Write out statements (R), (b) In chromatic pc space pc, define x ~ y to
(S), and (T) for the relation of enharmonic be true whenever there is some dominant
equivalence on signed registral letter space, and seventh chord containing both pitch classes
verify
   that all three statements are true. x and y.

64 • E x p l o r i n g M u s i c a l S p a ces
(c) In pc, define x ~ y to be true whenever may all be understood to be defined on S. These
there is some diminished seventh chord equivalence relations, known collectively as the
containing both x and y.48 OPTIC relations because of their initial letters,
(d) This relation is defined on the set triad will be of great importance in our study of voice-​
consisting of the 24 major and minor triads    spaces in Part Three.49
leading
in pitch-​class space. For triads X and Y,
define X ~ Y to be true whenever the triads Exercise 2.4.14 Listed below are five pairs of
X and Y share at least one common tone. pitch tuples. The two tuples in each pair are
(e) Let S be the collection of all pitch-​class sets. equivalent in at least one of the above senses.
For pitch-​class sets X and Y in S, define X ~ Identify which of the five equivalence relations
Y to be true whenever X and Y are related by (~O , ~P , ~T, ~TI, and/​or ~C) is satisfied by each
  inversion only (not allowing transposition). pair of ordered sets.

Exercise 2.4.12 Define a relation ~ on the (a) (C4, E4, G4) and (D5, F♯5, A5)
set ℝ>0 of positive real numbers as follows: (b) (C4, E4, G4) and (E4, G4, C4)
x ~ y is true whenever x ∕ y =​2n for some integer (c) (C4, E4, G4) and (C4, C4, E4, G4)
n (that is, the ratio of one number to the other (d) (C4, E4, G4) and (G4, E♭4, C4)
is a power of 2). For example, 163 ~ 6 because (e)  (C4, E4, G4) and (C4, E5, G4)
( 163 )/6 = 321 = 2−5. Verify that ~ is an equivalence
relation by checking conditions (R), (S), and (T). Exercise 2.4.15 Two other equivalence
Also describe a possible musical interpretation of relations introduced above, enharmonic
the relation ~. In which of the spaces described in equivalence and generic equivalence, do not
Chapter
   1 is this relation applicable? make sense as relations on the collection
of pitch tuples. Why not? Can you devise a
Two additional equivalence relations, permu- space of musical objects in which all of the
tational equivalence and cardinality equivalence, relations ~O , ~P , ~T, ~TI, ~C, ~E , and ~G are
will be of interest in later chapters. The notion    ​defined?50
well-
of permutational equivalence was implicit at the
beginning of Section 2.2, where we observed
that (E4, C4, D4, G3) and (G3, D4, E4, C4) are 2.5 MODULAR ARITHMETIC
two different ordered sets but that one may be
obtained from the other by permutation—​by Readers versed in pitch-​class set theory prob-
rearranging the same elements in a different ably have some familiarity with mod-​12 arith-
order. We say that these ordered sets are per- metic. Central to the study of arithmetic mod
mutationally equivalent, and we write (E4, C4, D4, 12 (or, more generally, arithmetic mod n) is an
G3) ~P (G3, D4, E4, C4). For cardinality equivalence, equivalence relation, congruence mod n. Because
let S be a collection of multisets. For multisets X acquaintance with these concepts will be
and Y in S, define X ~C Y to be true whenever X assumed frequently, and because they will later
and Y have the same reductions as sets (eliminat- be extended in musically relevant ways, it is use-
ing duplicate elements and reducing each ele- ful to review them briefly here.
ment to multiplicity 1, as was also discussed in If n is an integer ≥ 2, then integers x and y are
Section 2.2). For example, {C, C, E, G} ~C {C, E, G, congruent modulo n, usually abbreviated to mod n,
G, G} is a true statement about pitch-​class mul- if the difference x − y is divisible by n—​that is, if
tisets, because both of these multisets reduce to x − y =​ cn for some integer c. This congruence is
the set {C, E, G}. commonly notated x ≡ y (mod n), or occasionally
x ≡n y. For example, the congruences 15 ≡ 3 (mod
Exercise 2.4.13 Verify that conditions (R), (S), 12), 19 ≡ −5 (mod 12), and 50 ≡ 1 (mod 7) are
and (T) hold for the relations ~P and ~C. all true statements. The number n is the modu-
lus associated with the congruence. If x and y are
Let S be the collection of all pitch tuples, or positive integers, x ≡ y (mod n) means that x and
ordered pitch sets—​ordered sets of elements of y both yield the same remainder when divided
chromatic pitch space, of any length, possibly by n. Care must be taken when extending this
with duplications. The relations of octave equiv- interpretation to negative numbers: it may seem
alence, T equivalence, TI equivalence, permuta- counterintuitive to say that 19 and −5 yield the
tional equivalence, and cardinality equivalence same remainder when divided by 12, but it is

Sets, Functions, and Relations • 65

true that both 19 and −5 exceed the next smaller In terminology from Section 2.4, ℤ12 is ℤ ∕ ≡12,
multiple of 12 by the same amount (7). the quotient set of the integers ℤ by the equiva-
To verify that congruence mod n is an equiv- lence relation ≡ (mod 12). According to this
alence relation, we check conditions (R), (S), definition, an integer mod 12 is technically not
and (T): a number but a set of numbers, a subset of ℤ.
To say that each element of ℤ12 is identified
(R) For every integer x, x ≡ x (mod n). This is with an entire class of congruent numbers is
true because x − x =​0 =​0 ∙ n. analogous to saying that a pitch class is not just
(S) If x ≡ y (mod n), then y ≡ x (mod n). This is one note but an entire class of octave-​equiva-
true because y − x is the negative of x − y; if lent notes.
one of those differences is divisible by n, so Readers are probably accustomed to seeing
is the other. mod-​12 integers written 0, 1, …, 11, as if they
(T) If x ≡ y (mod n) and y ≡ z (mod n), then x ≡ z were ordinary integers, not sets, and indeed we
(mod n). This is true because x − z =​(x − y) +​ shall write them that way most of the time. The
(y − z); if the latter two differences are divis- more formal approach described above offers a
ible by n, so is x − z. worthwhile reminder that integers mod 12 are
not quite the same things as the ordinary inte-
For reasons that are probably obvious by now, gers bearing those names, just as the pitch class
the most important moduli for musical applica- C is not the same thing as the note C4. In par-
tions are 12 and 7. If pitch numbers are assigned ticular, ℤ12 is not a subset of ℤ; it is a quotient
to the elements of chromatic pitch space as in set of ℤ. The ordinary integers 0, 1, …, 11 form
Figure 1.1, then congruence mod 12 is the same a subset of ℤ; this subset, which contains one
thing as octave equivalence. In generic or dia- representative of each congruence class, may be
tonic pitch space, it is congruence mod 7 that called a module for the quotient set ℤ12. In some-
corresponds to octave equivalence.51 what the same way, in Chapter 1 we described
The equivalence classes of the relation of mod-​ the line segment of Figure 1.2.7b as a module for
n congruence may be called congruence classes; pitch-​class space, and the tile in Figure 1.4.3 as
the names remainder classes and residue classes a module for the conformed tonnetz of Figure
are also used. The congruence classes mod 12 are 1.4.2, as it contains one representative of each
note that appears in the full tonnetz.
0 =​{…, −24, −12, 0, 12, 24, …} The order in which the congruence classes
(numbers ≡ 0 mod 12), 0–​11 are listed above is such that if 1 is added
1 =​{…, −23, −11, 1, 13, 25, …} to a number in one class, the resulting number
(numbers ≡ 1 mod 12), belongs to the next class in the list. If 1 is added
2 =​{…, −22, −10, 2, 14, 26, …} to a number in the class 11, the resulting num-
(numbers ≡ 2 mod 12), ber belongs to the class 0; thus 0 follows 11 in
⋮
the same way that 1 follows 0 and so on. For this
reason the integers mod 12, like the chromatic
11 =​{…, −13, −1, 11, 23, 35, …}
pitch classes they are often used to represent,
(numbers ≡ 11 mod 12).
may logically be arranged in a circular clock
diagram.
The class 0 consists of all the multiples of 12,
The addition mentioned in the preceding
and the uniform spacing of 12 between succes-
sive elements of the class is maintained through paragraph is ordinary addition of integers (ele-
the other eleven congruence classes as well. ments of ℤ), not addition of integers mod 12
Every integer belongs to exactly one of these (congruence classes). In order to perform arith-
twelve classes; for example, 21 ∈ 9 and −19 ∈ 5. metical operations on the integers mod 12, we
The equivalence class 9 could also be called 21, must first consider how those operations are
and 5 could also be called −19; the classes are defined. For example, what is the mod-​12 sum
identified in the list above using the numerically 5+9 ​ ? Because 5 +​9 =​14 as a sum of ordinary
simplest name for each of them, from 0 to 11. integers in ℤ, and because 14 belongs to the
We define ℤ12, the integers mod 12, to be the congruence class 2, it seems logical to suppose
set of the twelve congruence classes: that 5 + 9​ =2​ . The clock circle seems to confirm
this conclusion: starting at 5 and counting 9
places clockwise, we arrive at 2. If we interpret
ℤ12 = { 0 , 1, . . . , 11 }. 5 and 9 as interval sizes, then an interval of

66 • E x p l o r i n g M u s i c a l S p a ces
5 +​9 semitones is exactly 2 semitones greater At this point we may abandon the formal-
than an octave; a musician comfortable with ism of the approach to modular arithmetic
mathematical language might even say “a per- outlined above, and agree to write the inte-
fect fourth and a major sixth add up to a major gers mod 12 as 0, 1, …, 11. We will usually
second, modulo the octave.” also write =​rather than ≡ for modular congru-
There is a potential pitfall in this reasoning. ence, though the qualifying “mod 12” should
The congruence classes 5 and 9 contain many be retained whenever ambiguity is possible.
other numbers besides 5 and 9. If we were to So rather than writing 5 + 9 ​ =​ 2, we write
choose different representatives of the same 5 +​9 =​2 (mod 12). We will sometimes refer
two congruence classes, add them, and get a to such a formula as an “equation”; technically,
number that belongs to some congruence class it may be regarded either as an equation (=​)
other than 2, our definition of addition would be of congruence classes or as a congruence (≡)
inconsistent. Fortunately, a little experimenta- of numbers. At times when it is important to
tion should convince the reader—​and it is not remember that we are referring to a mod-​12
difficult to prove rigorously—​that this problem integer rather than to an ordinary integer, the
never arises. For example, choosing representa- subscript notation 0mod 12, 1mod 12, …, 11mod 12 is
tives 29, 17, and −7 from the class 5 , and adding another useful alternative.
them respectively to representatives −3, 21, and As a practical matter, to perform arithmetic
−15 from the class 9, we calculate that 29 +​(−3) =​ (addition, subtraction, and multiplication) in
26, 17 +​21 =​38, and (−7) +​(−15) =​−22, and the modular world without worrying about the
observe that the sum in every case belongs to the full apparatus of congruence classes, one can
same congruence class 2 that we calculated the perform the operations in the normal way, with
first time. Our definition of addition is consis- the understanding that whenever the calculation
tent after all; addition mod 12 is well-​defined as a yields a number larger than 11 or smaller than
binary operation on ℤ12. Formally, the definition 0, we may add or subtract 12 as often as neces-
may be stated in the form m − + ​n− = (m + n ): the sary to put it back in range. The circle diagram
sum of two congruence classes m − and n− is the is often a helpful aid in visualizing mod-​12 addi-
congruence class of the sum m +​ n of representa- tion (clockwise motion around the circle) and
tives of those classes.52 subtraction (counterclockwise). Some additional
Similar considerations are at play in the defi- examples:
nitions of subtraction and multiplication mod
12. For example, 5 − 9 = 8 ​ and 5 · 9 = 9​ , as the 4 − 6 =​10 (mod 12): normally 4 − 6 is −2,
reader can verify by choosing any representa- so add 12 to this result;
tives of the congruence classes 5 and 9, perform- 5 · 5 =​1 (mod 12): normally 5 · 5 is 25,
ing the appropriate operations, and checking to so subtract 12 twice; and
which congruence class the result belongs. But it 9 · 8 =​0 (mod 12): normally 9 · 8 is 72;
is important to realize that division mod n is not subtract 12 six times.
defined. This should come as no surprise; after
all, division cannot be defined as an operation The last equation illustrates another way in
on the integers ℤ to begin with (fractions being which modular arithmetic differs from ordinary
unavailable in ℤ). But the difficulty in ℤ12 runs arithmetic: it is possible to multiply two mod-​
deeper, as the equation 5 · 9 = 9 ​ illustrates. If it 12 integers, neither of which is 0, and get 0 as
were possible to perform division mod 12, we a product.
should surely expect that 9 ÷ 9 should equal 1—​
for the equation 1 · 9 = 9 ​ is certainly true. But Exercise 2.5.2 Perform the indicated mod-​12
5·9=9 ​ as well, which suggests that 9 ÷ 9 should calculations. Write each answer in the simplest
also equal 5. This curious circumstance—​that possible form (that is, as an integer from 0
two different integers mod 12 can satisfy the to 11).
condition demanded of the same quotient—​is
something that never arises in the world of ordi- (a) 10 +​10 (mod 12)
nary numbers. (b) 10 − 2 (mod 12)
(c) 5 − 11 (mod 12)
Exercise 2.5.1 Actually there are three mod-​12 (d) 3 · 5 (mod 12)
integers x satisfying x · 9 = 9
​ . Two of them are (e) 11 · 11 (mod 12)
1 and
   5 . What is the third? (f)  (9 +​10) · (4 − 7) (mod 12)53

Sets, Functions, and Relations • 67

 The quotient map Q12: ℤ → ℤ12 associated (e) 1 − 2 (mod 4)
with the equivalence relation ≡ (mod 12) maps (f)  2 · 2 (mod 4)
any integer n to its congruence class n in ℤ12. For
example, Q12(19) =​7 . We have just decided, how- The fact that 7 is a prime number is responsi-
ever, to call this congruence class 7 rather than 7, ble for some distinctive features of mod-​7 arith-
so we can write Q12(19) =​7. The number 7 that metic, unlike anything in the mod-​12 world. For
we have chosen to name this class is numerically example, while the equation 3 · 4 =​0 (mod 12)
the simplest of the various integers that are con- is true, the equation x · y =​0 (mod 7) cannot be
gruent to 19 (mod 12)—​the only such number satisfied unless either x or y is 0. Also, we saw in
in the range 0–​11. For this reason, calculation Exercise 2.5.1 that three different mod-​12 inte-
of Q12 may be regarded as a process of reducing gers x satisfy 9x =​9 (mod 12), but nothing of the
a number mod 12. Although it is called a “quo- sort is possible mod 7. In fact, if a and b are inte-
tient map,” Q12 may intuitively be thought of as gers mod 7 and a ≠ 0 (mod 7), it turns out that
a “remainder map” instead, in the sense that 7 the equation ax =​ b (mod 7) always has exactly
is the remainder when 19 is divided by 12. As one solution x. In effect, this means that in the
before, care must be taken when applying this mod-​7 world it is indeed possible to “divide” by
interpretation to negative numbers.54 any nonzero mod-​7 integer, solving the equation
Most of the principles of mod-​12 arithmetic ax =​ b for x =​ b ∕ a. (This property is a manifesta-
described above apply equally well to mod-​7 arith- tion of some principles of group theory to be dis-
metic, or for that matter to mod-​n arithmetic for cussed in Chapter 5—​see in particular Exercise
any integer n ≥ 2. The integers mod n are the con- 5.2.7b—​ and has important consequences
gruence classes of the congruence relation mod n, involving musical behavior in diatonic spaces, to
and the set of these integers mod n, a set of car- be explored in Chapter 13.)
dinality n, is denoted ℤn. The reader may verify
the following equations in ℤ7: 6 +​6 =​5 (mod 7); Exercise 2.5.4 In parts (a)–​(d), determine the
3 − 4 =​6 (mod 7); 6 · 5 =​2 (mod 7). The function unique solution (mod 7) of each equation:
f described in Exercise 2.3.10g was precisely the
quotient map Q7. (a) 2x =​1;
We have described musical interpretations (b) 4x =​3;
of ℤ12 and ℤ7 in the formation of chromatic and (c) 5x =​2;
diatonic pitch-​class spaces. A musical application (d) 6x =​4.
of other moduli besides 12 and 7 arises in the (e) Give examples to show that the special
construction of beat-​class spaces, derivable from property of mod-​7 arithmetic described
the beat space mentioned in Example 1.4.20. In above fails in mod-​4 arithmetic. Specifically,
a  time signature, for example, beats occupying give an example of mod-​4 integers a and b,
corresponding positions in two different mea- with a ≠ 0 (mod 4), for which the equation
sures have beat numbers that are congruent mod ax =​ b (mod 4) has no solutions, and
4. Congruence mod 4 converts beat space beat another example for which there are at least
into  beat-​class space bc(4), in the same way two different solutions.
that congruence mod 12 converts pitch space (f) The equations 7 +​6 =​1 (mod 12) and
pitch into 12-​ note pitch-​
class space pc. The 7 − 6 =​1 (mod 12) are identical except for
beats traditionally numbered 1–​4 are identified the operation signs (+​in one case, − in
with the mod-​4 integers 0–​3 in bc(4), and any the other), and both equations are true.
subset of ℤ4 may be realized as a beat-​class set, Is an analogous pair of mod-​7 equations
analogous in many ways to a pitch-​class set.55 possible? That is, are there any mod-​7
integers a and b for which a +​ b =​ a − b
Exercise 2.5.3 Perform the indicated (mod 7), apart from the trivial case in
calculations. Write each answer in the simplest   which b =​0 (mod 7)?56
possible form, as an integer from 0 to n − 1,
where n is the appropriate modulus. Exercise 2.5.5
(a) Suppose you know that x is an integer such
(a) 3 +​5 (mod 7) that x =​7 (mod 12). What, if anything, can be
(b) 3 − 5 (mod 7) said about x mod 3? About x mod 4? About
(c) 3 · 5 (mod 7) x mod 5? What special properties of the
(d) 3 +​1 (mod 4) numbers 3, 4, 5, and 12 are relevant here?57

68 • E x p l o r i n g M u s i c a l S p a ces
(b) If a is an integer mod 3, the mod-​12 pitch-​ on several later occasions; in particular, they will
class numbers x such that x =​ a (mod 3) figure into our more thorough exploration of
form what sort of pitch-​class set? If b is generic spaces and applications to chromatic har-
an integer mod 4, the mod-​12 pitch-​class mony in Part Four. They also afford some oppor-
numbers x such that x =​ b (mod 4) form tunities to practice some moderately abstract
what sort of pitch-​class set? reasoning involving modular arithmetic.60
(c) Suppose m and n are coprime positive A central portion of spelled pitch-​class space
integers, which means that m and n spc appears in Table 2.6.1. Immediately below
have no common factors larger than each note name is the associated spc number
1. Suppose a is an integer mod m and from Figure 1.3.3. Recall that by convention,
b is an integer mod n, and suppose spc numbers assign the number 0 to the note
you are interested in finding a single D, unlike pc numbers, which assign 0 to C. Spc
number x for which the congruences x numbers are integers, theoretically unbounded,
=​ a (mod m) and x =​ b (mod n) are both but common note names with no more than two
true. It is a consequence of a theorem sharps or flats correspond to the spc numbers
of number theory, sometimes called the shown in the table, ranging from −17 to 17.
Chinese remainder theorem, that these The remaining three lines of the table display
two congruences always have a common numbers that can be calculated as functions of
solution, which is unique mod mn (that the spc number n; we use the Greek letters π (pi),
is, if x1 and x2 are both solutions of the γ (gamma), and α (alpha) to denote these three
two given congruences, then x1 =​ x2 (mod functions. The mod-​12 pitch-​class number cor-
mn)). If m =​3, n =​4, a =​1, and b =​2, responding to spc n, π(n), is called the projection
what is the unique value of x mod 12?58 of n in pc space. (The letter pi may be considered
(d) Use the property from (c) to give an to stand for either projection or pc.) In mapping
alternative explanation for the fact noted in spelled pitch classes to ordinary pitch classes, π
Exercise 1.4.7a: a diminished seventh chord effectively accomplishes the same thing as the
and an augmented triad always have exactly function QE: sl → pc, the quotient map of the
  one pitch class in common. enharmonic equivalence relation on signed let-
ter space described in Section 2.4. Unlike QE,
While modular arithmetic typically involves however, π is defined on spelled pitch-​class num-
integers only, constructions such as the real bers rather than on note names. That is, whereas
numbers mod 12, denoted ℝ12, are possible as QE is a function from sl to pc, π is a function
well.59 The congruence relation ≡ (mod 12) is from ℤ to ℤ12; an equation such as π(−6) =​8 cap-
defined for non-​integers in the same way as for tures in a completely numerical way the under-
integers; for example, 4.25 ≡ 16.25 (mod 12). standing that the note name A♭ (spc number −6)
The quotient set ℝ12 is the appropriate math- represents pitch class 8. We will determine a way
ematical model for continuous pitch-​class space to calculate the value of π(n) shortly.
cpc. Here there are infinitely many congruence Similarly, the generic projection of an spc n is
classes rather than merely twelve—​the infinitely γ(n), the associated generic pitch-​class number
many points of the continuous circle of cpc. As (or gpc number; gamma stands for generic). Gpc
numerical representatives of these classes we numbers are integers mod 7, based on C =​0. The
may take all the real numbers from 0 to 11.999... . function γ: ℤ → ℤ7 is therefore the numerical
References to integers mod 12 (elements of ℤ12) representation of the quotient map QG: sl → gpc
are common throughout this book; on the rarer explored in Exercise 2.4.5.
occasions when we refer to “numbers mod 12,” Finally, α(n) is the accidental index of the spc
elements of ℝ12 are what we have in mind. n (alpha for accidental), an integer that simply
counts the sharps or flats in a note name, count-
2.6 RELATIONSHIPS AMONG ing sharps positive and flats negative. For exam-
MODULAR SPACES ple, the accidental index of G♭ ♭ is −2, while the
accidental index of F♯ ♯ ♯ is +​3 (F♯ ♯ ♯ is spc 18, lying
We conclude this chapter with a discussion of sev- just beyond the range shown in Table 2.6.1). As
eral important aspects of the relationships among the table shows, seven consecutive notes along
chromatic, generic, and spelled pitch-​class spaces, the line of fifths share the same accidental index;
which hinge on modular structure. These proper- α always advances from one value to the next
ties, though not widely recognized, will be useful between the letter names B and F.

Sets, Functions, and Relations • 69

 Of the three functions π, γ, and α, it is α If m and n are two different spcs with the
that is easiest to calculate numerically: for same accidental index (α(m) =​ α(n)), then we
every spc n, know that m and n must lie fewer than seven
places apart along the line of fifths—​that is,
α(n) =​ n/​7 rounded to the nearest integer. |m − n| ≤ 6. Therefore m and n cannot possibly
be separated by as many as twelve places, or even
For example, α(11) =​+​2, because the integer as many as seven, so it follows from the preced-
closest to 11 ∕ 7 is 2, and α(−2) =​0, because the ing paragraph that π(m) ≠ π(n) and γ(m) ≠ γ(n).
integer closest to −2 ∕ 7 is 0. This simple formula This means that two different spcs cannot share
is one reason for choosing D =​0 rather than the same value of α and also the same value of
C =​0 as the origin for spc numbers; the latter either π or γ. In other words, if we know the
convention would necessitate a more compli- value of both α(n) and π(n), or if we know the
cated formula for the accidental index. value of both α(n) and γ(n), then n is uniquely
To determine a formula for π(n), observe in determined.
the table that each time n increases by 1, π(n)
increases by 7 (mod 12); this is because spc Exercise 2.6.3 Answer the following questions,
numbers are organized along the line of fifths, preferably without consulting Table 2.6.1.
and a perfect fifth is seven semitones. Because
π(0) =​2, it follows that for every n, (a) If α(n) =​+​1 and π(n) =​10, what is n? (You
are looking for the spc number of a note
representing pitch class 10 whose spelling
π ( n) = 7n + 2 ( mod 12).
includes one sharp.)
(b) If α(n) =​−3 and π(n) =​2, what is n?
For example, π(11) =​7 ∙ 11 +​2 =​79 =​7mod 12, and (c) If α(n) =​−2 and γ(n) =​5, what is n? (The
π(−2) =​7 ∙ (−2) +​2 =​−12 =​0mod 12. The number 2 note name should be obvious: there is only
that appears in the formula for π(n) is attribut- one note whose spelling includes two flats
able to the two-​semitone discrepancy between C and the letter name A.)
and D, the notes from which pc and spc numbers (d) Is there any n for which α(n) =​+​1 and π(n) =​
are reckoned.61   4? What does this mean musically?

Exercise 2.6.2 The above discussion leaves open the ques-

(a) Check that the formulas given above for tion of whether n can be determined from
α(n) and π(n) hold for the notes E, G♭ ♭, knowing both π(n) and γ(n). A little thought
and F♯ ♯ ♯. shows that the answer must be no: there are
(b) Using reasoning similar to the above only twelve possibilities for π(n) (an integer
discussion of π(n), obtain a formula for γ(n). mod 12) and only seven possibilities for γ(n)
This will be a mod-​7 formula, not mod-​12. (an integer mod 7), so there are only 7 ∙ 12 =​
Check that your formula works for the same 84 possible combinations of π(n) and γ(n),
  three notes you considered in part (a). whereas the number of integers n is infinite.
Nevertheless, there is a practical sense in which
The functions π, γ, and α map ℤ onto ℤ12, ℤ7, if we know both the pc-​space and the gpc-​space
and ℤ, respectively, but none of these functions projections of a note—​both its pitch-​class num-
is one-​to-​one. We have already observed that α ber and its letter name—​then we can deduce
takes the same value on seven successive spcs. what the note is.
The formula π(n) =​7n +​2 (mod 12) implies that To begin with, consider again the formula
if m and n are two spcs with m ≡ n (mod 12),
then π(m) =​ π(n), which is just another way of
saying that every twelfth note along the line π ( n) = 7n + 2 ( mod 12).
of fifths is enharmonically equivalent. The for-
mula for γ(n) from the preceding exercise like- Let us try to solve this equation for n. Multiplying
wise implies that if m ≡ n (mod 7), then γ(m) =​ both sides by 7 gives
γ(n)—​that is, every seventh note along the line
of fifths is generically equivalent (has the same
letter name). 7π ( n) = 49n + 14 ( mod 12).

70 • E x p l o r i n g M u s i c a l S p a ces
Table 2.6.1. Mappings π, γ, and α defined on spelled pitch-​class space

NOTE NAME F♭♭ C ♭♭ G ♭♭ D ♭♭ A♭♭ E♭♭ B♭♭ F♭ C♭ G♭ D♭ A♭ E♭ B♭

Spelled pitch class n ∈ℤ −17 −16 −15 −14 −13 −12 −11 −10 −9 −8 −7 −6 −5 −4
Pitch class π(n) ∈ ℤ12 3 10 5 0 7 2 9 4 11 6 1 8 3 10
Generic pitch class γ(n) ∈ ℤ7 3 0 4 1 5 2 6 3 0 4 1 5 2 6
Accidental index α(n) ∈ℤ −2 −2 −2 −2 −2 −2 −2 −1 −1 −1 −1 −1 −1 −1

F C G D A E B F♯ C♯ G♯ D♯ A♯ E♯ B♯ F ♯♯ C ♯♯ G ♯♯ D ♯♯ A ♯♯ E ♯♯ B ♯♯
n −3 −2 −1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
π(n) 5 0 7 2 9 4 11 6 1 8 3 10 5 0 7 2 9 4 11 6 1
γ(n) 3 0 4 1 5 2 6 3 0 4 1 5 2 6 3 0 4 1 5 2 6
α(n) 0 0 0 0 0 0 0 +​1 +​1 +​1 +​1 +​1 +​1 +​1 +​2 +​2 +​2 +​2 +​2 +​2 +​2

In a mod-​12 formula we can replace 49 by 1 and two congruences always have a common solu-
replace 14 by 2, because 49 =​1 (mod 12) and tion, and that solution is unique mod 84. So
14 =​2 (mod 12). Therefore while the integer n itself cannot be deduced
from π(n) and γ(n), we can determine its congru-
ence class mod 84. This is more than sufficient to
7π ( n) = n + 2 ( mod 12),
determine the common notation for any musical
note: recall that all the notes with no more than
or equivalently two sharps or flats satisfy −17 ≤ n ≤ 17, and the
numbers in that range are all different mod 84.
n = 7π ( n) − 2 ( mod 12). Exercise 2.6.5 Questions (a)–​(c) below
can be answered on the basis of a musical
This formula tells us how to recover n from understanding of π(n) and γ(n). The following
π(n), except that it tells us only the congruence exercise will develop an explicit formula, but
class of n mod 12, not the actual value of the that formula is not needed here. In parts (a) and
integer n. (b), assume that the note in question is spelled
with no more than two sharps or flats. The note
Exercise 2.6.4 Using reasoning similar to in (c) may require three.
the above, take your formula for γ(n) from
Exercise 2.6.2b and solve it for n, obtaining the (a) If π(n) =​1 and γ(n) =​1, what is n?
congruence
   class of n mod 7. (b) If π(n) =​9 and γ(n) =​4, what is n?
(c) If π(n) =​1 and γ(n) =​2, what is n?
If we know both π(n) and γ(n), therefore, we (d) The rather extreme spc numbers m =​−42
can determine the congruence class of n mod 12 and n =​42 are congruent mod 84. Determine
and also the congruence class of n mod 7. Let us the note names corresponding to m and
suppose n ≡ a (mod 12) and n ≡ b (mod 7). The n. Verify that π(m) =​ π(n) and γ(m) =​ γ(n).
numbers 12 and 7 are coprime, so by the Chinese (Fortunately this is not something that
remainder theorem (see Exercise 2.5.5c), these   arises in practical situations!)

Sets, Functions, and Relations • 71

Exercise 2.6.6 This exercise, which requires The spc transposition τk is defined on spc
a bit of algebra, outlines a way to deduce an numbers n by the formula τk(n) =​ n +​ k. In other
explicit formula for n (mod 84), given the values words, τk “transposes” spelled pitch classes
of π(n) (mod 12) and γ(n) (mod 7). along the line of fifths spc in exactly the same
way that Tk transposes notes in chromatic
(a) Suppose first that we know n ≡ a (mod 12) pitch space pitch.62 In fact, as functions from
and n ≡ b (mod 7). Let k be the number ℤ to ℤ, τk and Tk are the same function. Because
49a +​36b. Show that the congruences k ≡ n the same numbers represent different notes in
(mod 12) and k ≡ n (mod 7) are both true. the spaces spc and pitch, however, the musi-
What properties of the numbers 49 and 36 cal interpretation is different and the alternate
make both of these congruences work? It notation is useful. Moreover, via the associa-
follows that k − n is divisible by both 12 and tion of spelled pitch classes with signed letters,
7, and hence divisible by 84—​which means it is convenient to assume that τk is defined on
that k ≡ n (mod 84). signed letters as well as on spc numbers. For
(b) Above we showed that the formula for instance, we may write τ3(2) =​5 or equivalently
π(n) may be solved for n in the form n =​ τ3(E) =​C♯, indicating that C♯ lies three places to
7π(n) − 2 (mod 12). So 7π(n) − 2 plays the the right of E on the line of fifths; T3(E) is of
role of the number a in part (a) above. course something else entirely. The fact that B♭
In Exercise 2.6.4 you showed that the lies five places to the left of A may be expressed
formula for γ(n) may be solved for n (mod by the equation τ−5(1) =​−4, or equivalently
7); the result of that exercise is a formula τ−5(A) =​B♭.
that may be used for the number b. A fact that will prove useful several times is
Making these two substitutions, write the that in a mod-​12 sense, either the spc trans-
number k =​49a +​36b and simplify it mod position τj or the pitch-​class transposition Tk
84. If you do this correctly, you should may be converted to the other by multiplying
arrive at the formula the index number by 7. If spelled pitch class m
occurs immediately to the left of n on the line
n = 7π ( n) − 12γ ( n) − 2 (mod 84). of fifths, then n =​ τ1(m) and the corresponding
pitch classes are fifth-​related, which is to say
(c) Verify that the formula from (b) holds for
the notes D, C♯, and B♭ ♭.
π(n) =​ T7(π(m)). More generally, if n =​ τj(m), then
π(n) =​ T7j(π(m)); thus the pitch-​class transposi-
(d) Verify that your answers to parts (a)–​(c)
tion corresponding to τj is T7j. (Like the index
of Exercise 2.6.5 are consistent with the
numbers of all transposition operators in pc, the
  formula from part (b) of this exercise.
number 7j may be reduced mod 12.) Conversely,
if we know that π(n) =​ Tk(π(m)), for what value of
We have referred several times to the trans-
j can it be true that n =​ τj(m)? There will be more
position operators Tj on pitch-​ class space.
than one such j since many notes on the line of
Two other families of functions that will
fifths are enharmonically equivalent. But what-
prove useful are generic transposition, denoted
ever j is, we just established that the equation k =​
by the lowercase letter t, and spc transposi-
7j (mod 12) must be true. Multiplying both sides
tion, denoted by the Greek letter τ (tau). The
of that congruence by 7 and using the fact that
generic transposition operator tj transposes
7 ∙ 7 =​1 (mod 12), we conclude that j =​7k (mod
generic pitch classes by j steps in generic pitch-​
12): up to congruence mod 12, the spc transposi-
class space gpc; thus the formula tj(n) =​ n +​ j
tion corresponding to Tk is τ7k. The congruence
(mod 7) holds for all mod-​7 gpc numbers n. For
7 ∙ 7 =​1 (mod 12) is essentially responsible for
example, t5(E) =​C, or equivalently t5(2) =​0,
the fact that the same multiplier 7 works in both
because the gpc numbers corresponding to
directions.
E and C are the mod-​7 integers 2 and 0, and
2 +​5 =​0 (mod 7). Musically, this means that Exercise 2.6.7
C is five generic steps—​a sixth—​above E. The (a) Determine j and k for which τj(F) =​B♭ and
small t distinguishes generic transpositions tj Tk(F) =​B♭. Verify that k =​7j (mod 12) and
from the more familiar chromatic transposi- j =​7k (mod 12). Do the same for the notes
tions Tj, which work analogously in mod-​12 D and B, and for the notes E and A♭.
chromatic space pc, as described by the equa- (b) The generic projections of spelled pitch
tion Tj(n) =​ n +​ j (mod 12). classes 2 (E) and 5 (C♯) are γ(2) =​2 (E) and

72 • E x p l o r i n g M u s i c a l S p a ces
 γ(5) =​0 (C). The spelled pitch classes E and refer to a row of notes in a specified order—​a usage
C♯ satisfy the equation τ3(E) =​C♯, while the not consistent with standard mathematical terminol-
corresponding generic pitch classes E and ogy, in which no fixed order of the elements of a set
C satisfy t5(E) =​C. In general, if two spcs is assumed. David Lewin opens his book Generalized
are related by τj, for what value of k will the Musical Intervals and Transformations (Lewin [1987]
corresponding gpcs be related by tk? That 2007, 1) by explaining that he will avoid the word set for
is, if τj(m) =​ n, for what value of k will it be this reason, but many recent music scholars have used
true that tk(γ(m)) =​ γ(n)? Write a mod-​7 the word in the standard mathematical sense, with no
  formula for k in terms of j.63 order implied.
2. It would perhaps be more accurate to say that
Exercise 2.6.8 The sets A, X, Y, and Z the use of braces to denote sets is “nearly univer-
appearing in the following questions are spc sal except in music theory.” Allen Forte flouted this
sets—​sets of spelled pitch classes. Spc sets convention by writing sets in square brackets, most
may be written either in note names or in spc famously in Forte 1973 but previously in Forte 1964,
numbers: for instance, the C-​minor triad is 139, where an endnote ascribes the notation to now-​
the spc set {C, E♭, G} =​{−5, −2, −1}. Answer obsolete “typographical reasons.”
the questions numerically, but also think 3. Modern mathematicians recognize the first
about musical implications. You will need to prime number as 2, not 1. Uniquely among positive
recall the notation for images f(A) and inverse integers, 1 is neither prime nor composite. It has many
images f −1(B) from Section 2.3. distinctive properties not shared by the primes (for
example, it is its own multiplicative inverse). If 1 were
(a) If A is any set of twelve or more consecutive admitted to the list of primes, one of the most impor-
integers (spc numbers), what is π(A)? tant properties of prime numbers, the uniqueness of
(b) If A is any set of seven or more consecutive prime factorization, would be sacrificed, as in the mul-
integers, what is γ(A)? tiple factorizations 6 =​3 ∙ 2 =​3 ∙ 2 ∙ 1 =​3 ∙ 2 ∙ 1 ∙ 1.
(c) If y is any integer, what can you say about 4. Set theory distinguishes many different infinite
α−1({y})? cardinalities; it can be shown, for instance, that #ℤ
(d) Let X, Y, and Z be the following spc sets: and #ℚ are the same but #ℝ is larger. Such abstrac-
tions will not concern us here; interested readers may
X =​{−3, −1, 0, 3} =​{G, B, D, F} see Stoll 1979, Chapter 2.
Y =​{−6, −4, −3, 0} =​{B♭, D, F, A♭} 5. The prime symbol ʹ does not always imply com-
Z =​{−4, −3, 0, 6} =​{B♭, D, F, G♯} plementation; we shall employ it for other purposes
from time to time.
Calculate the images π(X), π(Y), and π(Z) 6. Hint for Exercise 2.1.2a: To be as general as pos-
(these are pc sets, or subsets of ℤ12), and also sible, a Venn diagram for three sets should divide the
γ(X), γ(Y), and γ(Z) (these are gpc sets, or plane into eight regions, because there are eight pos-
subsets of ℤ7). Which two of the three sets X, Y, sible combinations of sets to which any given element
and Z have the same projection in pc space (the may belong (including the possibility that it belongs to
same image under π)? Which two have the same none of the three sets).
projection in gpc space (the same image under 7. Hint for Exercise 2.1.4d: The question asks for the
γ)? Which two of the three sets X, Y, and Z are cardinality of a certain intersection S ∩ T. Perhaps you
related by spc transposition? Which sets have can determine the cardinality of (S ∩ T)ʹ more easily.
projections that are related by T-​transposition 8. Hint for Exercise 2.1.5d: If x ∈ S\T, that does not
pc space?64
in   tell us whether x ∈ V or not. But certainly either x ∈ V
or x ∉ V; what can you say about x in each case?
9. Hint for Exercise 2.1.7: If the union T1 ∪ T2 ∪ ⋯
∪ T6 is all of pc, then the intersection of this union
NOTES with S is all of S. If T1, T2, …, T6 are disjoint, then the
1. The mathematical study of sets developed out sets S ∩ T1, S ∩ T2, …, S ∩ T6 are also disjoint, so the
of a nineteenth-​century interest in the theory of infi- cardinality of their union is the sum of their cardinali-
nite series. The words collection, family, and class are ties. Your observations from (a) should show that this
used more or less interchangeably with set (although sum cannot be 6.
mathematicians make some technical distinctions, 10. Hints for Exercise 2.1.8: You may be able to see
particularly involving the word class). In early writings intuitively why (a) must be true. For a formal proof, let
on twelve-​tone music, the word set was often used to T1, T2, and T3 be the three diminished seventh chords.

Sets, Functions, and Relations • 73

By applying the distributive law from Exercise 2.1.7 to writing S → T to show a mapping from one set to
the set S ∩ (T1 ∪ T2 ∪ T3), you should be able to show another and x ↦ y to show a mapping of elements.
that S must intersect one of the diminished seventh 18. Instead of supposing that a function trans-
chords in at least two notes; then apply the union-​car- forms one object into another, it may be preferable
dinality law from Exercise 2.1.4 to determine some- to imagine that the first object is being replaced by, is
thing about the union of S with that chord. For (b), let moving to, or is simply being associated with the sec-
S be the complement of the given eight-​note set and ond object. The word function has a variety of uses in
apply the result of (a). everyday English, some of which resemble its use in
11. In music theory, a row or a series is typically an mathematics. To say “the president’s approval rating
ordered set of pitch classes. Mathematicians sometimes is a function of economic conditions,” for example, is
use the word sequence for an ordered set of numbers but effectively to make a statement of the form y =​ f(x),
are unlikely to use the word series in this way, reserving postulating a hypothetical black box (f ) that takes
series for the sum of the numbers in a sequence. economic conditions (x) as input and calculates the
12. Hints for Exercise 2.2.1: You will not want to write president’s approval rating (y). Other common uses
out all the strings in part (b), because in this case the of the word seem to bear little relation to the mathe-
number of ordered sets allowing repetitions is more than matical definition, as when function is used to describe
ten times the number without repetitions. The answer to the purpose or role of something in a general sense
(c) is a single simple formula, valid regardless of whether (“What is the function of this odd gizmo?”). The term
m < n, m =​ n, or m > n. See Hook 2007d for these and “harmonic function” in music theory is an uncertain
other techniques of counting pitch-​class sets, set classes, case. My understanding of the term is closer to the
ordered sets, and various other musical structures. “purpose or role” definition, more or less independent
13. The integers a and b in this formula correspond of mathematical functions. Brian Hyer (2011, 112),
to the two types of “conformance” considered in on evidence that he admits is “largely circumstantial,”
Exercise 1.4.15. Enharmonic respelling along the line suggests that Hugo Riemann may have had the math-
of fifths (conformance of the Pythagorean comma) is ematical theory of functions in mind as he developed
represented by a (as can be checked by setting a =​1 his own Funktionstheorie, but Trevor Pearce (2008)
and b =​0), while b accounts for conformance of the explores other possible origins of Riemann’s use of
syntonic comma (set a =​0 and b =​1). the term.
14. The commonly encountered dictum “time 19. Neo-​Riemannian theory presents numerous
is the fourth dimension” has misled many people. opportunities for confusion of this kind because of
Mathematicians conceive freely of spaces with four or the many ways in which triadic transformations may
more spatial dimensions. Time may be given its own be defined and calculated. For example, we will see in
dimension in a multidimensional construction, but so Section 4.2 that the composite transformations LPL
can essentially anything else that can be quantified: and PLP, though they may be calculated by differ-
temperature, population, money, and musical pitch ent sequences of steps, always yield the same results
are just a few examples. when applied to any major or minor triad, so they
15. Although discussions of sets with multiplici- are in fact the same function (defined on the space
ties can be traced back many centuries, such construc- of major and minor triads). The composite LP, on the
tions are not accommodated in classical set theory, other hand, is not the same function as the transposi-
and a theory of multisets was not formalized until the tion T4 defined on triads, because while LP and T4 give
1980s (Blizard 1989). Robinson 2009 studies applica- identical results when applied to any major triad, they
tions of multisets in pitch-​class set theory. disagree when applied to a minor triad. In Section 8.3
16. Lewin ([1987] 2007, 3) adopts a more restric- we will discuss the diatonic mediant transformation
tive definition of transformation, requiring that a M, which acts like the neo-​Riemannian R on major
transformation be a function from a set S to the same triads but like L on minor triads, and which is there-
set S, what we are here calling a transformation on S. He fore (as a function) not equal to either of them. See
also uses the term operation specifically for bijective Exercises 2.3.20, 4.2.12, and 6.2.16 for some related
transformations. Neither of these usages is standard examples.
in mathematics, and they will not be followed here. In 20. Variables may be contrasted with constants,
the music theory literature one sometimes encounters which always stand for the same objects. The symbols
so-​called transformations that are not actually well-​ 3, −1∕8 , and π are numerical constants, always stand-
defined mathematical functions, a use of the word ing for the same fixed numbers; we would never dare
that is discouraged here. write “let 3 be a set.” Similarly the symbols P and T3
17. Mathematicians sometimes use two different are functional constants, traditionally represent-
styles of arrows in these two situations, for instance ing fixed functions, the neo-​Riemannian parallel

74 • E x p l o r i n g M u s i c a l S p a ces
transformation and pitch-​ class transposition by 3 may be encountered elsewhere, however: some texts
semitones respectively. refer to T as the range, and the smaller set of values
21. A single graph such as Figure 2.3.2 may appear taken by the function as its image.
to be of little analytical value, in part because in an 27. A definition by cases, like the definition of this
abstract way it fairly closely resembles the score itself. function g, is perfectly permissible. Although we infor-
Nevertheless, many analyses resembling graphs of mally described a function as a “rule,” there need not
pitch as a function of time, more elaborate than this be a single formula that prescribes the behavior of a
one, may be found in the music theory literature. See, function in every case. This definition does not violate
for example, Robert Cogan and Pozzi Escot’s graph the requirement that functions be deterministic: two
of the pitch structure of the Introduction to Elliott choices are apparently available, but the applicable
Carter’s Second String Quartet (1976, 64–​66); several choice is completely determined by the number n.
figures in Jonathan Bernard’s study of the music of 28. Physicists and acousticians accustomed to
Ligeti (1999); Miguel A. Roig-​Francolí’s “spatial reduc- working with frequencies sometimes refer to pitch
tion” of the first movement of Lutosławski’s Jeux véni- space as log-​frequency space. The logarithmic relation-
tiens (2008, 290); or Richard Hermann’s analysis of ship between physical frequency and musical pitch is
Berio’s 1956 Quartetto per archi (2009, 124–​27). among the best-​known illustrations of a more general,
22. Loosely speaking, a function from ℝ to ℝ is largely empirical psychophysical principle known as
continuous if its graph can be drawn in a continuous Fechner’s law after the German experimental psychol-
motion without taking one’s pen off the paper. Curves ogist Gustav Theodor Fechner (1801–​87). Fechner’s
and angles are permissible in the graph of a continu- law states that the subjective sensation produced by
ous function, but skips like those in Figure 2.3.2 are a physical stimulus is proportional to the logarithm of
not. The formal definition of continuous functions, the intensity of the stimulus. Other musically relevant
important in the study of calculus, is rather technical, forms of Fechner’s law apply in matters of loudness
depending in the most general setting on topological (decibels are a logarithmic measure of sound inten-
features of the spaces involved. sity) and, at least approximately, tempo (a difference
23. The light vertical dotted line segments in Figure in tempo between 40 and 45 beats per minute is per-
2.3.2 serve only to tie the segments of the graph ceived as a greater difference than that between 160
together visually; they are not technically part of the and 165, just as log2(45) − log2(40) is greater than
graph of the function. How exactly the function should log2(165) − log2(160)). For the history of Fechner’s law
be defined at x =​2, the instant at which the melody see Masin, Zudini, and Antonelli 2009.
changes notes, could be a matter of some contention. 29. Hint for Exercise 2.3.13: Many calculators offer
The description given here assumes that the function functions called “log” (the common logarithm to the
takes its new value at that instant rather than retain- base 10) and “ln” (the natural logarithm to the base
ing its old one. The function therefore takes a constant e =​2.71828..., a function of importance in calculus).
value on a “half-​closed, half-​open interval” denoted Base-​2 logarithms may be calculated using the for-
[0, 2) by mathematicians, who might convey this mula log2(x) =​log10(x) ∕ log10(2), or log2(x) =​ln(x) ∕ ln(2).
behavior in the graph by drawing a small solid circle 30. Lewin ([1987] 2007, 2) briefly discusses the
at the left end of each of the horizontal line segments orthography problem. Though Lewin settles on right-​
and a small open circle at the right end. Physically and to-​left orthography for most purposes, much recent
musically, however, this distinction is inconsequen- work in transformation theory has employed left-​to-​
tial, perhaps even meaningless, because a well-​defined right orthography. Ultimately the problem originates
pitch cannot be produced in an infinitesimal point in in the notation f(x) itself, in which the conceptually
time (or in any span of time shorter than the period prior x is written to the right of the conceptually sub-
of vibration of the corresponding sound frequency). sequent f. A notation such as (x)f, though it looks
24. Hint for Exercise 2.3.3d: One and only one of peculiar to anyone accustomed to the traditional nota-
the following formulas is correct: g(t) =​ f(t +​ b) − s; tion and would require various associated linguistic
g(t) =​ f(t +​ s) − b; g(t) =​ f(t − b) +​ s; or g(t) =​ f(t − s) +​ b. adaptations, would alleviate orthographic difficulties
25. This chromatic version of the ascending 5–​6 considerably.
sequence, though less familiar than the diatonic, is by 31. The use of the letter E for identity functions is
no means rare. An extended example, rising chromati- consistent with the common use of e for the identity
cally through seven semitones, occurs in mm. 18–​32 element in group theory, to be discussed in Chapter 5.
of the chorus “Si ridesta” in Act I, Scene 4 of Verdi’s Identity functions are sometimes denoted by the let-
La traviata. ter I, but this notation seems inadvisable in musical
26. The possibly larger set T is sometimes called contexts in which confusion with inversion operators
the codomain of the function. Different terminology is likely.

Sets, Functions, and Relations • 75

 32. While the English-​ speaking world has bor- if, for example, all but one of the notes of one chord
rowed the German word Leittonwechsel, the English are T5 transpositions of notes in the other chord, while
parallel contradicts German usage, in which the the notes in the remaining voice are related instead by
Parallelklang corresponds to the English relative (and T4 or T6. This *T5 is a relation, not a function: if A is a pc
the English parallel is known as Variant). In modern set, the set *T5(A) cannot be uniquely defined, because
neo-​Riemannian theory the letters P and R always several different sets may be related to A in this way.
denote the English parallel and relative, respec- Nevertheless, the notations for the functions Tn and
tively. Jeffrey Prater, in his translation of Diether relations *Tn are deliberately similar, and the two have
de la Motte’s Harmonielehre (de la Motte [1976] been mingled productively in analytical applications;
1991, 215), introduces in place of leittonwechsel the see Straus 2003 and Straus 2005b.
term counter-​relative, a translation of de la Motte’s 39. The expression “if and only if” is common in
Gegenparallelklang—​a logical name but one that has mathematical writing as a way of expressing the logi-
never gained currency. cal equivalence of two statements. Suppose p and q
33. Hint for Exercise 2.3.20: This exercise has are statements. “If p, then q,” sometimes symbolized
included four yes-​no questions to this point, three in p ⇒ q, means that if p is true, q must also be true. This
part (a) and one in (b). Of these four questions, exactly statement rules out the possibility that p is true while
three should be answered yes. q is false, but leaves open the possibility that q may be
34. Hint for Exercise 2.3.21b: Set (1) consists of all true while p is false. The assertion “p if and only if q,”
pitch classes x such that T3(x) is E♭, E, or G. To calculate symbolized p ⇔ q, rules out the latter possibility also,
set (2), you must determine what the function T3−1 is, because it asserts p ⇒ q and q ⇒ p simultaneously,
and apply this function to the notes E♭, E, and G. thereby implying that p and q are either both true
35. Hint for Exercise 2.3.22b: Recall, as in Exercise or both false. To say that “two elements are related
2.3.18c, that for g and gʹ to be proper inverse func- by ~ if and only if they are in the same equivalence
tions, it would have to be the case that gʹ(g(x)) =​ x for class,” therefore, means that two elements related by
every x in sl, and also that g(gʹ(y)) =​ y for every y in ~ must be in the same equivalence class, and also that
pc. One of these two equations is true, but the other two elements in the same equivalence class must be
is not. related by ~.
36. Hints for Exercise 2.3.25: The difference between 40. It is tempting to suppose that a quotient
the frequencies of B♭4 and A♯4 is the Pythagorean set is inherently more abstract than the set from
comma, mentioned in the notes to Section 1.4. Of the which it was constructed. Our formulation of octave
three questions in part (d), one should have the same equivalence effectively regards pitches as indivisible
answer as the corresponding question in Exercise gestalts and defines pitch classes more abstractly,
2.3.24, but the others should have different answers. as equivalence classes of pitches. But this assump-
Pythagorean pitch space is sometimes known as 3-​ tion can be misleading: musical pitch is already an
limit pitch space, as all its frequency ratios are rational abstraction, and a case can be made that a pitch such
numbers generated by multiplying and dividing fac- as A4 is itself an equivalence class comprising many
tors no larger than 3. different acoustical signals, produced by various
37. By analogy with the 3-​limit pitch space in the instruments or other means, that can be recognized
preceding exercise, just pitch space may be called as the pitch A4.
5-​limit pitch space: all its frequency ratios may be 41. Aspects of the relationship between enhar-
derived from factors no larger than 5. The space monic and octave equivalence were presented in
of the Riemannian tonnetz itself, which does not Hook 2015.
include register information, is 5-​limit pitch-​class 42. For other illustrations of the point that
space. Because even-​numbered harmonics duplicate one kind of equivalence does not necessarily imply
lower harmonics at the octave, n-​limit pitch spaces another, see Cohn 2012, 102–​03.
are of interest only for odd values of n. Spaces similar 43. We are being slightly careless in asserting
to those described in Exercises 2.3.25 and 2.3.26 can that QO and QE commute, for technically Figure 2.4.4
be constructed in higher dimensions for larger values shows two different functions QO (one from srl to sl
of n; 7-​limit spaces are implicit in Karg-​Elert [1930] and one from pitch to pc), and likewise two differ-
2004, 53–​56, and constructed explicitly in Johnston ent functions QE. But these identically named func-
[1976–​77] 2006, 72. tions are not independent: the action of QO on pitch
38. An example of a relation that is sometimes and that of QE on sl are determined by the actions of
presented in quasi-​transformational language is fuzzy the correspondingly named quotient maps on srl.
transposition, also called near-​transposition. Two pitch-​ Intuitively, there is a clear sense in which both of the
class sets may be said to be related by *T5, “fuzzy T5,” functions QO do the same thing (they both discard

76 • E x p l o r i n g M u s i c a l S p a ces
register information, thereby merging octave-​related and we use mod-​7 arithmetic to know that the 25th
notes) and both of the functions QE do the same day of the month falls on the same day of the week
thing (they both discard spelling information, merg- as the 11th.
ing enharmonically equivalent notes), and those two 52. To be even more formal, we could adopt a nota-
things may be done in either order with the same tion such as +​12 for addition mod 12, calling attention
result. to the distinction between this binary operation on
44. Hint for Exercise 2.4.5e: Keeping in mind that ℤ12 and ordinary addition +​on ℤ. The definition of +​12
any number or sharps or flats is allowed in srl, you is then m +12 n = (m + n ). The symbol +​on the right
should be able to show, for example, that C4 ~GE G4. side of this equation is ordinary addition of integers,
Are there any pairs of notes in srl that are not related and the symbol =​appears rather than ≡ (mod 12);
by ~GE? the relation here is equality of sets, not congruence
45. In the early stages of his formalization of of numbers.
pitch-​ class set theory, Allen Forte (1964) worked 53. Hints for Exercise 2.5.2: The sums and differ-
with interval-​ content equivalence. His later, more ences in parts (a)–​(c) should be visualized in the
definitive presentation of the subject (1973) pro- clock diagram. In a mod-​ 12 calculation involving
ceeded from the basis of TI equivalence. The familiar several steps, 12 may be freely added or subtracted
set classes are based on TI equivalence, but vestiges at any point along the way. An exercise like (f) may
of interval-​content equivalence remain discernible in therefore be completed in a variety of ways: inter-
various details of Forte’s theory such as the ordering mediate results may be simplified if desired, or the
of the “Forte numbers.” entire calculation may be performed first in ordinary
46. In general, given two equivalence relations ~1 arithmetic, with the mod-​12 reduction taking place
and ~2, it is not always possible to classify one as finer only at the end. In (e), it may simplify matters to
or coarser than the other. It may happen, for example, observe that 11 =​−1 (mod 12), so 11 · 11 should be
that there are elements x and y such that x ~1 y but x the same as (−1) · (−1).
≁2 y, and also other elements w and z such that w ~2 z 54. An alternate terminology, encountered espe-
but w ≁1 z. This situation does not arise, however, for cially in some programming languages, treats mod as a
the three relations in Exercise 2.4.8, all of which are binary operation on ℤ equivalent to the quotient map
comparable on the fine-​coarse scale. Q12; thus one may write “19 mod 12 =​7.” This is not
47. For a formalization of equivalence relations as the standard mathematical syntax, and it will not be
described here, see Stoll 1979, 29–​31. employed here.
48. Hint for Exercise 2.4.11b–​c: Your answers to (b) 55. For musical meter conceived as equivalence
and (c) should be different. What properties of domi- classes of time points (possibly at several levels), see
nant and diminished seventh chords make it possible Benjamin 1984, 375. Beat-​class set analysis has been
for chords of one of these types to be equivalence developed by Cohn (1992b), Roeder (2003), and others.
classes, but not for chords of the other type? 56. Hint for Exercise 2.5.4f: By performing some
49. There are some subtleties in the definitions of simple algebraic manipulations on the equation a +​ b
these relations in the most general case, discussion =​ a − b (mod 7), see if you can arrive at an equation
of which will be deferred to Chapter 10. Cardinality that can be solved for b, in the manner of parts (a)–​(d).
equivalence in particular must be defined with care; 57. Hint for Exercise 2.5.5a: The answer to one of
the definition we have presented here is incomplete. the first three questions is “nothing at all.”
Also, at present we have not defined I equivalence in 58. The Chinese remainder theorem has taken
the absence of T equivalence, for reasons suggested by many forms, the oldest of which appeared in a Chinese
Exercise 2.4.11e. In Chapter 10 we will define I equiva- text dated between 280 and 473 ce. See Davis and
lence, which turns out to be only of limited interest by Hersh 1981, 187–​95, for an informal discussion. The
itself; a satisfactory form of inversional equivalence will form given here is a special case of Theorem 121 in
still require that transposition be considered also. Hardy and Wright 2008, 121.
50. Hint for Exercise 2.4.15: The last question has 59. Observe carefully that 12 is a subscript in the
no simple answer. Using Exercise 2.3.25 as a clue, you notations ℤ12 and ℝ12. A notation such as ℝ12, with a
may be able to devise a space in which most of the numerical superscript, would denote a twelve-​dimen-
listed relations are easily defined but in which the defi- sional space, by analogy with our previous use of ℝ3
nitions of ~T and ~TI are rather complicated. for the threefold Cartesian product ℝ × ℝ × ℝ.
51. Coincidentally, we all acquire some facility with 60. Some of the formalism described in Section 2.6
both mod-​12 and mod-​7 arithmetic through our reck- was introduced in Hook 2011; the techniques relate
oning of time and the calendar. We use mod-​12 arith- also to some constructions in Regener 1973. Regener
metic to calculate that five hours after 9:00 is 2:00, refers to the line of fifths as the quint group and to

Sets, Functions, and Relations • 77

diatonic steps as diatones. The origin of his system is may also be derived from the formula for γ from
the note F rather than D or C. Exercise 2.6.2b.
61. If the same note were used as the origin for 64. Hint for Exercise 2.6.8d: The last four questions
the numbering of both spc space and pc space, the have four different answers. The theory of spc sets will
formula for π(n) would be simply π(n) =​7n (mod 12). be developed at greater length in Part Four.
As we have noted, however, setting D =​0 has decided
advantages in spc space, and the slight awkwardness
associated with using two different origins seems less
SUGGESTED READING
objectionable than violating the nearly universal con- Numerous texts on mathematical set theory are
vention of C =​0 in pc space. available, at various levels of detail and sophis-
62. As we will explore in more detail in Chapter 5, tication. Most of the concepts presented in this
the notation Tk may be used for a transposition opera- chapter are covered in more depth in the opening
tor either in pitch space or in pitch-​class space. The chapters of Stoll 1979. For the basics of pitch-​class
transposition τk is called the quint operation Qk in set theory see Straus 2016. Modular arithmetic is
Bazayev 2013. part of the branch of mathematics known as num-
63. Hint for Exercise 2.6.7b: You may be able to ber theory; a classic text on the subject, last revised
come up with this formula by experimenting with seventy years after its first publication, is Hardy
the spc and gpc numbers in Table 2.6.1. The formula and Wright 2008.

78 • E x p l o r i n g M u s i c a l S p a ces
 3
Graphs

IN STUDYING musical spaces such as those pre- of this chapter introduces transformation graphs
sented in Chapter 1, it is important to distinguish and networks, concepts of great importance in
between the spaces themselves and the diagrams the chapters to follow.
we draw of them. While the set theory introduced Graph-​ theoretic terminology varies from
in Chapter 2 is an appropriate tool for studying source to source; the discussion on the follow-
some structural aspects of musical spaces and their ing pages will introduce terms appropriate for
elements, many of the diagrams are more properly our purposes and, whenever possible, consistent
in the province of graph theory, the subject of this with the terminology used in the relevant music
chapter. We have seen already that a single space theory literature. Graph theory is fundamentally
may often be diagrammed in several strikingly a branch of discrete mathematics, and the empha-
different ways. In some cases graph theory itself sis in this chapter is accordingly on discrete
may not actually account for such distinctions; structures (chromatic rather than continuous pc
for example, there is a technical sense in which space, for example).
the three representations of signed letter space in
Figure 1.2.11 are “the same graph.” Nevertheless,
there are often musical or visual reasons for pre-
3.1 GRAPHS
ferring one arrangement over another, and we We shall use the diagram in Figure 3.1.1 to
will examine some of these considerations here illustrate several basic concepts of graph the-
as well. Graphs appear regularly in the literature ory. This diagram shows voice-​leading relation-
in the area known as neo-​Riemannian theory, and ships among eight triads: major triads on the
our exploration of graph theory will provide an roots C, E, and A♭, indicated by capital letters;
opportunity to introduce a few neo-​Riemannian minor triads on the same roots, indicated by
concepts as musical motivation. The last section lowercase c, e, and g♯; and augmented triads

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0003
B+​ and C+​.1 These triads form a hexatonic system, The musical properties depicted in the hexa-
in that they contain a total of only six differ- tonic triad graph are of some interest, and we
ent pitch classes, the elements of the hexatonic shall return to them. First, however, we turn our
pitch-​class collection {11, 0, 3, 4, 7, 8}. Because attention to the diagram itself, shown in Figure
we will refer to Figure 3.1.1 and related graphs 3.1.2 stripped of its musical content. This more
frequently, for convenience we call it a hexatonic abstract diagram is indeed a graph as the term
triad graph. For the purposes of this diagram is used in graph theory; we call it the underlying
triads should be considered as nothing more graph of Figure 3.1.1. (This is an entirely differ-
than three-​note pitch-​class sets; chord inver- ent usage of the word from the graph of a func-
sions and enharmonic spellings, for instance, tion, mentioned in Section 2.3. The word graph
are of no import.2 in this book will be used most often in the sense
Inspection of the diagram shows that two of graph theory, but we reserve the right to use
triads connected by a line always share two the word informally in other ways from time
common tones, differing only by a single-​semi- to time.) This particular graph consists of eight
tone displacement in the remaining note. For vertices (also called nodes or points) and twelve
example, to convert B augmented to E minor, edges (or lines). The graph is an undirected graph,
one need only raise the pitch class D♯ to E, leav- because the edges have no arrowheads attached
ing the other two notes (B and G) unchanged. to indicate direction; we will consider directed
In contrast, the C-​minor and C-​augmented tri- graphs later in this chapter. Two vertices joined
ads share only a single common tone, so no line by an edge are said to be adjacent to each other,
joins them. and each edge is incident with its two end verti-
ces. The degree of a vertex is the number of edges
incident with it. If all vertices in a graph have the
same degree, the graph is regular. The graph in
Figure 3.1.2 is therefore regular, as all its verti-
ces are of degree 3.
The hexatonic triad graph in Figure 3.1.1 may
be called a vertex labeling of Figure 3.1.2. The label
attached to a vertex is sometimes also called the
contents of the vertex. In other situations it is
useful to label the edges of a graph, or both the
vertices and the edges. Most of the graphs we
encounter in musical applications will be labeled
in some way. Because every edge in a graph is
incident with exactly two vertices, graphs are
a particularly convenient means of depicting
binary relations: the vertices of a graph may be
labeled with the elements of the set on which
FIGURE 3.1.1 A hexatonic triad graph the relation is defined, and an edge is provided
for each pair of related elements. The graph in
Figure 3.1.1 indeed depicts a certain binary
relation on its eight triads: the single-​semitone
voice-​leading property described above.
Several subgraphs of the hexatonic triad graph
are shown in Figure 3.1.3. The defining property
of a subgraph is that its vertices form a subset
of the original vertices and its edges form a sub-
set of the original edges. The geometric arrange-
ments of the vertices in these subgraphs may not
match the original figure; more on that momen-
tarily. The subgraph in 3.1.3a is a path—​that is,
a series of alternating vertices and edges.3 This
picture is arranged in a musically suggestive way,
so that ascending voice leading corresponds with
FIGURE 3.1.2 The underlying graph of Figure 3.1.1 upward movement in the graph: as one traverses

80 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 3.1.3 Five subgraphs of the hexatonic triad graph

the path from bottom to top, starting with the all are of some musical interest. The 4-​cycle in
B-​augmented triad, each of the three voices slips (b) consists of the four chords from the origi-
up in succession until the C-​augmented triad is nal graph that contain the pitch class B, and the
reached. edges from the original graph joining them. The
In graph theory, the length of a path refers to 6-​cycle in (c), known as a hexatonic cycle, com-
the number of its edges, so the length of the path prises the six major and minor triads from the
in Figure 3.1.3a is 3 (the number of voice-​leading original graph, omitting the augmented triads.
moves), not 4 (the number of triads).4 A quick A widely cited passage in the first movement of
glance at Figure 3.1.1 (or 3.1.2) shows that every Brahms’s Double Concerto, Op. 102, mm. 270–​
pair of vertices is joined by some path—​one can 78, circumnavigates this hexatonic cycle slightly
get from any vertex to any other by traversing more than once around, presenting chords in the
edges of the graph—​a property summarized by order A♭ –​g♯ –​E –​e –​C –​c –​A♭ –​g♯ –​E.5 Hexatonic
saying that Figure 3.1.1 is a connected graph. The cycles are a characteristic construction in neo-​
distance from one vertex to another in a graph is Riemannian theory, and we will revisit them sev-
the length of the shortest path joining them. A eral times. Finally, the 8-​cycle in Figure 3.1.3d is
path of length 1 is a single edge, so the distance notable because it includes every vertex of the
between two adjacent vertices is 1; the distance original graph (but not every edge); such span-
from any vertex to itself is 0. No paths from B+​ to ning cycles are possible in some graphs but not
C+​ in Figure 3.1.1 have length shorter than 3, so in others.6
the distance from B+​ to C+​ is exactly 3. Each of these cycles may be interpreted as
a series of harmonies, connected by smooth
Exercise 3.1.4 chord-​ to-​
chord voice leading and returning
(a) How many different paths of length 3 are eventually to its starting point. One can con-
there from B+​ to C+​ in Figure 3.1.1? struct a cyclic graph depicting the sequence of
(b) Identify another pair of vertices in Figure key areas appearing in a tonal work that begins
3.1.1, other than B+​ and C+​, whose distance and ends in the same key. Cyclic graphs have
from each other is 3. many other uses in music theory, and in fact
(c) Is there any pair of vertices in Figure 3.1.1 several of the diagrams in Chapter 1 are cycles.
whose distance from each other is greater Chromatic pitch-​class space pc and the circle of
  than 3? fifths fifth are 12-​cycles, while gpc, gfifth,
and gthird (generic pitch-​ class space, fifths
The subgraphs in parts (b)–​(d) of Figure 3.1.3 space, and thirds space) are 7-​cycles. Each row
are cycles (or cyclic graphs). A cycle is a path that of a conformed tonnetz such as Figure 1.4.2
returns to its starting vertex. These three cycles may be diagrammed as a 4-​cycle, and each col-
are of three different lengths—​4, 6, and 8—​and umn as a 3-​cycle (which, as a graph, is nothing

Graphs • 81
more than a triangle). All of these examples are different visual arrangements as manifestations of
interval cycles, cycles of pitch classes generated “the same graph,” and to consider the diagrams in
by a single interval, but no such restriction is Figure 3.1.3 to be subgraphs of 3.1.1 even though
necessary when forming a cyclic graph. One can their vertices are arranged differently.
arrange the notes of an octatonic collection as an The statement quoted above is true as far as
8-​cycle even though they are not generated by abstract graph structure is concerned, but it may
a single interval—​indeed, one can form a cycle be qualified by remarking that some arrange-
using the notes of any pitch-​class set at all. Beat-​ ments may be more effective than others in com-
class spaces are well-​known examples of cyclic municating certain aspects of that structure. We
structure in the time domain. will expand on this observation by presenting
Figure 3.1.3e may look like four graphs rather three additional figures (Figures 3.1.5, 3.1.7,
than one, but the general definition of a graph and 3.1.10), all of them “the same graph” as the
allows us to regard the figure as a single discon- hexatonic triad graph from 3.1.1, and all convey-
nected graph with four connected components. ing the same basic information. That is, all these
Each component consists of just two vertices graphs have vertices representing the same eight
joined by an edge; the musical interest in this triads, and edges join the same pairs of vertices
example lies in the fact that the same B–​C voice-​ in all of them. Each picture, however, highlights
leading move links all four pairs of triads. The a different aspect of the musical relationships
triads containing the pitch class B (the same tri- among the triads, and each is appealing in its
ads appearing in 3.1.3b) appear in the lower row own way.
of (e), with the corresponding triads containing Figure 3.1.5, taking its cue from the path in
C arranged directly above them. 3.1.3a, arranges all eight vertices so that ascend-
Graph theory texts often define graphs ing voice leading correlates with upward motion
abstractly, saying, for example, that a graph is through the graph. The vertices accordingly
an ordered triple (V, E, i) consisting of a set V of arrange themselves in four tiers, with the B-​aug-
undefined objects called vertices, a set E of objects mented triad alone on the lowest tier, the three
called edges, and an incidence function i defined on minor triads on the second tier, the three major
the domain E, which maps every edge to a 2-​ele- triads on the third tier, and the C-​augmented
ment subset of V (defining which two vertices are triad alone at the top. Pitch-​class numbers for
incident with the edge). Such a definition takes no each triad have been added to this graph, with
account of how the graph may be arranged on the the three voices displayed in an ordered-​triple
page. One leading text goes so far as to say, “there format in which the first component is always
is no single correct way to draw a graph; the rela- 11 or 0 (that is, either the note B or C), the sec-
tive positions of points representing vertices and ond component is always 3 or 4 (E♭ or E), and the
the shapes of lines representing edges usually third is always 7 or 8 (G or A♭).
have no significance” (Bondy and Murty 2008, 3). Each edge joins triads on two successive tiers
It is this understanding that allows us to consider and is labeled with a notation for the appropriate

FIGURE 3.1.5 A second representation of the hexatonic triad graph

82 • E x p l o r i n g M u s i c a l S p a ces
voice-​leading move: for example, “11–​0” on the Figure 3.1.5 may remind the viewer of an
edge from E to C+​ indicates that the “lower” E-​ oblique view of a three-​ dimensional box. In
major triad is transformed to the “higher” C-​ fact, as an abstract graph, the vertices and edges
augmented triad via the semitonal motion B –​C. of this figure correspond exactly to the frame
Each pair of adjacent triads (joined by an edge) of a cube. The original hexatonic triad graph in
consists of an upper triad and a lower triad in this Figure 3.1.1 also may be thought of as a cube, in
way, linked by one moving voice. (This upper/​ a tilted perspective view, looking directly at one
lower nomenclature depends on the understand- face and seeing the opposite face receding into
ing that pitch class 11 can ascend to pitch class the distance. Figure 3.1.7 makes this cube-​frame
0; recall the discussion in Section 1.2 of ascend- structure more explicit. The figure has been
ing and descending motion in pitch-​class space.) reoriented so that the edges of the graph align
The sum class of a triad (or any pitch-​class set) with the coordinate axes in a representation of
is the mod-​12 sum of its pitch-​class numbers. three-​dimensional space ℝ3. Each of the three
For example, the E-​minor triad belongs to sum dimensions represents one of the three voices
class 10, because 11 +​4 +​7 =​10 (mod 12). In in the triads: the horizontal x axis represents
every pair of adjacent triads, the sum class of the the 11–​0 voice (pitch classes B and C), the y axis
upper triad exceeds the sum class of the lower (which should be imagined pointing back into
triad by 1. The sum classes of the triads shown, the page, away from the viewer) represents the
which are all 9, 10, 11, or 0, correspond exactly 3–​4 voice (E♭ and E), and the vertically oriented
to the four tiers in the graph, as shown at right z axis represents the 7–​8 voice (G and A♭). The
in the diagram. Sum classes are thus an impor- ordered triple designating any triad is precisely
tant part of the organizing principle for the fig- the usual coordinate representation of the cor-
ure—​a principle that will become increasingly responding point in the three-​dimensional space
significant when larger voice-​leading spaces are defined by these axes. In this way, we may see the
considered in Part Three. eight triads, or the eight vertices of the graph, as
elements of a three-​dimensional Cartesian prod-
Exercise 3.1.6 Suppose all the triads in Figure uct, {11, 0} × {3, 4} × {7, 8}. This representation
3.1.5 are transposed up by one semitone, so that suggests the possibility that the hexatonic triad
the triad on the lowest tier is C+​ and that on graph might be expanded to form a much larger
the highest tier is D♭+​. The resulting figure is the three-​dimensional Cartesian product containing
graph of a different hexatonic system, but it has all possible three-​note chords—​an idea we will
the same graphical appearance as Figure 3.1.5. pursue in Part Three.
What are the new sum classes? What can you The four triads bordering any one face of the
say in general about the sum classes of major cube all share one coordinate. Figure 3.1.8 illus-
triads?
   Of minor triads? Of augmented triads? trates by highlighting the left face, whose four

FIGURE 3.1.7 A third representation of the hexatonic triad graph

Graphs • 83
 this way, but this representation calls atten-
tion to an aspect of the graph structure that
was not evident in any of the earlier pictures.
Specifically, the vertices have been partitioned
into two subsets—​the top row and the bottom
row—​in such a way that every edge joins one
subset to the other; no edge joins two vertices
within the same subset. A graph whose verti-
ces may be partitioned in this way is called a
bipartite graph. The hexatonic triad graph was
FIGURE 3.1.8 Two faces of the cube of Figure bipartite all along, but Figure 3.1.10 makes this
3.1.7, with the associated pitch-​
class numbers property explicit in a way that the previous rep-
11 and 0 resentations did not. The bipartite structure,
moreover, helps to explain some other features
and musical implications of the graph, as the
following exercise demonstrates.

Exercise 3.1.11
(a) The subgraphs of Figure 3.1.1 shown in
Figure 3.1.3 included a 4-​cycle, a 6-​cycle,
and an 8-​cycle, but not a 3-​cycle (triangle),
5-​cycle, or 7-​cycle. Explain why a bipartite
graph can never include any cycles of odd
FIGURE 3.1.10 A fourth representation of the length.
hexatonic triad graph (b) Suppose a composer writes a succession
of triads from the hexatonic triad graph,
starting with C major on the downbeat
triads share pitch class 11 (B), and the right face, in a  meter. Suppose the chords move
whose triads share pitch class 0 (C). The left face consistently in quarter notes, and that each
is exactly the subgraph from Figure 3.1.3b, and triad is connected to the next by single-​
the four edges oriented horizontally, joining the semitone motion corresponding to an edge
two shaded faces, form the subgraph in 3.1.3e. in the graph. On which beats of a measure is
it possible to return to C major? On which
Exercise 3.1.9 A cube has six faces. Besides the beats is it possible for an E-​minor triad to
two faces highlighted in Figure 3.1.8, identify   occur?7
the others, which form two more pairs of
opposing faces. Determine which pitch class is The original representation of the hexatonic
common to the four triads on each face. These triad graph in Figure 3.1.1 possesses one nota-
faces are also present in the original hexatonic ble characteristic not shared by any of the other
triad graph of Figure 3.1.1. At a glance, Figure alternatives. All of our diagrams have, of course,
3.1.1 may appear to consist of only five four-​ been drawn in the two-​dimensional plane of the
sided
   faces. Where is the sixth? paper, but only in Figure 3.1.1 has the drawing
been accomplished in such a way that no two
A final recasting of the hexatonic triad graph edges cross each other. There are two crossings
is shown in Figure 3.1.10. The arrangement is each in Figures 3.1.5 and 3.1.7, and many cross-
related to that of Figure 3.1.5, but there are ings in Figure 3.1.10. (The crossing points are
only two tiers instead of four: the B-​augmented not vertices of the graphs.) A representation
triad has been moved into the row with the without edges crossing is a planar embedding of
major triads, and C-​augmented into the row a graph. A graph for which a planar embedding
with the minor triads. In terms of sum classes, is possible is a planar graph. Some graphs with
the chords with odd sum classes (9 and 11) are many edges are not planar, which is to say that
on the top row, those with even sum classes (10 a planar embedding is not possible; we will see
and 0) on the bottom. Musically, it may be hard some examples of non-​planar graphs later in this
to see any advantages to drawing the graph chapter.8

84 • E x p l o r i n g M u s i c a l S p a ces
 Planarity and embeddings are important top-
ics in graph theory. Embeddings of graphs into
surfaces or spaces other than the plane are also
of interest, forming a link between graph theory
and topology. The cube of Figure 3.1.7 may be
viewed as an embedding in three-​dimensional
space ℝ3 rather than in a two-​dimensional plane
ℝ2. Alternatively, by curving the edges slightly
one can conceive of the cube as a spherical embed-
ding, an embedding of the same graph on the
surface of a sphere (a two-​dimensional surface,
itself embedded in three-​ dimensional space).
Planar embeddings and spherical embeddings
essentially amount to the same thing, at least
for finite graphs like those we have considered
so far: readers may be able to visualize ways to
convert any planar embedding to a spherical
embedding or vice versa, demonstrating that a
graph has a planar embedding if and only if it has FIGURE 3.1.13 The dual graph of the hexatonic
a spherical embedding. triad graph
Above we discussed the six faces of the cube
of Figure 3.1.7. Faces are, in fact, well-​defined
for any planar (or spherical) embedding of a associated with one face. Figure 3.1.13 repro-
graph: they are simply the regions in the plane duces the original hexatonic triad graph in the
(or sphere) bounded by the graph’s edges.9 The background, and labels each face with the corre-
unbounded region exterior to the graph counts sponding pitch-​class number (bold and circled).
as one face. Thus the graph of Figure 3.1.1 has The number 7, representing the pitch class G,
six faces, corresponding to the faces of the cube is the label for the exterior face of the original
of Figure 3.1.7. While there may be many differ- graph. Whenever two faces share an edge in the
ent possible ways to embed a graph in a plane, original graph, the corresponding pitch-​ class
the various embeddings will always have the numbers are connected by a new line, which
same number of faces. In fact, the number of crosses the original edge. The result is a new
faces is easy to calculate, using a classical theo- graph, shown in bold lines in the figure. (The
rem called Euler’s formula: if the numbers of ver- edges of a graph need not always be drawn as
tices, edges, and faces in a planar embedding of straight lines!) This new graph is called the dual
a connected graph are denoted v, e, and f respec- graph of the original graph.
tively, then v +​ f =​ e +​2. The dual graph is another planar graph, its
vertices corresponding to the original faces
Exercise 3.1.12 (labeled with pitch-​class numbers), and its faces
(a) Verify that Euler’s formula holds for the corresponding to the original vertices (labeled
graph of Figure 3.1.1 and for each of the with triad names). While the original hexatonic
connected subgraphs in Figure 3.1.3a–​d. triad graph has eight vertices and six faces, its
(b) The statement of Euler’s formula given dual graph has six vertices and eight faces. The
above requires the graph to be connected. number of edges (12) is unchanged; each edge
Verify that the formula fails for the of the dual graph crosses exactly one edge of
disconnected graph of Figure 3.1.3e. If c is the original graph. Because only the vertices
the number of components of a graph, can and faces are exchanged, and because v and f
you find a generalization of Euler’s formula, are added together in Euler’s formula, the sums
involving v, e, f, and c, that holds for all in that formula work out the same for both
  planar graphs?10 graphs. A further aspect of the duality is that in
the original graph the faces are four-​sided, and
As noted above, each face of the hexatonic three edges meet at each vertex, while in the
triad graph is bounded by four triads shar- dual graph the faces are three-​sided, and four
ing a common pitch class, and each pitch class edges meet at each vertex. If we start with the
of the hexatonic collection {11, 0, 3, 4, 7, 8} is dual graph and perform the same construction

Graphs • 85
 With only eight vertices, the hexatonic triad
graph is fairly simple, but a variety of more
complex graphs can be constructed based on
similar principles; a few are suggested in the
two exercises below. Such graphs may be con-
strued as resources for a composer seeking to
write nonfunctional progressions characterized
by smooth voice leading: from the chord at any
vertex in one of these graphs, the adjacent ver-
tices show other chords that might potentially
be available.

FIGURE 3.1.14 The octahedral dual graph of the Exercise 3.1.17 Construct voice-​leading graphs
cube of Figure 3.1.7 similar to the hexatonic triad graph as described
below. Try to arrange the vertices of each graph
in a way that is as simple, symmetrical, and
again, we will simply exchange the vertices and visually appealing as possible.
faces a second time, thereby recreating the origi-
nal hexatonic triad graph: the dual of the dual is (a) Construct an octatonic triad graph with
the original graph.11 eight vertices, labeled with the major and
A dual graph may be constructed for any minor triads on the roots C, E♭, F♯, and A.
planar embedding of a graph, or indeed for any The pitch classes of these eight chords form
embedding of a graph into any two-​dimensional an octatonic collection. Two vertices are
surface. For a graph with edge crossings such as joined by an edge if the triads are related by
Figure 3.1.10, faces are not well-​defined and a a semitone or whole-​tone displacement in a
dual graph cannot be constructed. For the cube single voice. Thus the C-​major and C-​minor
of Figure 3.1.7, regarded as a spherical embed- triads are adjacent, because they are related
ding, the construction of the dual graph is illus- by the semitonal displacement E –​E♭. But C
trated in Figure 3.1.14. A new vertex is placed at major is also adjacent to A minor, related by
the center of each of the cube’s six square faces. the whole tone G –​A.
When the new edges are drawn, the resulting (b) Construct an enneatonic seventh-​chord graph
dual graph is a three-​dimensional figure called an with nine vertices, labeled with the major-​
octahedron, with eight triangular faces inscribed minor, minor-​minor, and half-​diminished
within the cube in an arrangement resembling a seventh chords on the roots C, E, and A♭.
pyramid atop its mirror reflection. The cube and The pitch classes of these chords form the
the octahedron are two of the five Platonic sol- enneatonic collection enn234 =​{0, 2, 3, 4, 6,
ids (or regular polyhedra), known since antiquity, 7, 8, 10, 11}. Two vertices are joined by an
and their graphs are dual to each other.12 edge if the seventh chords are related by a
semitone or whole-​tone displacement in a
Exercise 3.1.15 Figures 3.1.13 and 3.1.14 both single voice. Thus the C7 chord is adjacent
depict dual graphs of the hexatonic triad graph. both to Cmm7 (via the semitone E –​E♭) and
Are these two duals “the same graph”? That is, to eø7 (via the whole tone C –​D).
do edges join all of the same pairs of labeled (c) Construct an octatonic seventh-​chord graph
vertices
   in both graphs? with twelve vertices, labeled with the major-​
minor, minor-​minor, and half-​diminished
Exercise 3.1.16 Carefully redraw the seventh chords on the roots C, E♭, F♯, and
octahedron of Figure 3.1.14 at a larger size. A. The pitch classes of these chords form
Label each face of the octahedron with the an octatonic collection. Two vertices are
name of the triad whose pitch classes occupy joined by an edge if the seventh chords
the surrounding vertices. Construct the dual are related by a semitone (not whole-​tone)
graph, which will now be a three-​dimensional displacement in a single voice. Thus the
figure inscribed within the octahedron. Verify Cmm7 chord is adjacent to four other
that the resulting graph is “the same graph” as chords: cø7 (via G –​G♭), C7 (via E♭ –​E), E♭7
the  original cube.   (via C –​D♭), and aø7 (via B♭ –​A).13

86 • E x p l o r i n g M u s i c a l S p a ces
Exercise 3.1.18
(a) Just as the voices in three-​note chords
may be modeled by the dimensions in a
three-​dimensional space, the voices in
four-​note chords may be modeled by the
dimensions in a four-​dimensional space.
The octatonic seventh-​chord graph from
Exercise 3.1.17c may be constructed as
part of a four-​dimensional hypercube whose
vertices are identified with elements of the
four-​dimensional Cartesian product {0, 1} ×
{3, 4} × {6, 7} × {9, 10}. Construct a graph of FIGURE 3.2.1 Two isomorphic graphs
this hypercube.
(b) How many vertices did the original
octatonic seventh-​chord graph have? How denote pitch classes and each edge represents the
many vertices are in the hypercube from interval of a whole step. Two of the same vertex
part (a)? What chords are represented by labels appear in both graphs, but they represent
the additional vertices? chords in one graph and pitch classes in the other;
(c) The hypercube has 32 edges, each of which in one case (C) the matching labels appear at the
joins two chords sharing three common same location in both graphs, but in the other case
tones, such as Cmm7 and cø7. The hypercube (E) they do not. So it perhaps does not seem accu-
also has 24 two-​dimensional faces and rate to call these two pictures “the same graph.”
eight three-​dimensional “hyperfaces” Nevertheless, it is clear that something about the
more properly known as facets. Give structure of the two graphs is the same. Indeed,
musical descriptions of the sets of chords both are 6-​cycles, and it seems reasonable to say
  represented by the faces and facets.14 that cycles of the same length have the same
graph-​theoretic structure. This phenomenon is
what the concept of isomorphism encapsulates.
3.2 ISOMORPHISM OF Small labels v1–​v6 and v1ʹ–​v6ʹ have also been
GRAPHS attached to the vertices in the two graphs. We
may think of these as names for the vertices,
We used the expression “the same graph” sev- distinguished from the content labels indicat-
eral times in the previous section. When the ing the triads and pitch classes. We may define a
vertices in one diagram have the same labels as function f from the vertices of the first graph to
the vertices in another, it is a simple matter to the vertices of the second, mapping each vertex
check whether edges in the two diagrams join to the vertex in the corresponding location: that
corresponding vertices, and if so we may deem is, f(v1) =​ v1ʹ, f(v2) =​ v2ʹ, …, f(v6) =​ v6ʹ. This func-
the two diagrams to represent “the same graph” tion has the following properties:
even if the vertices and edges are arranged dif-
ferently on the page. In some situations, how- (a) f is one-​to-​one: different vertices in Figure
ever, two graphs with different vertex labels (or 3.2.1a always map to different vertices in 3.2.1b.
none at all) may share an underlying structure in (b) f is onto: every vertex in 3.2.1b is the
a way that may not be apparent from the way the image under f of some vertex in 3.2.1a.
graphs are drawn. To handle such cases requires (c) f preserves adjacency: two vertices vi and
defining exactly what it means for two graphs to vj in Figure 3.2.1a are adjacent (joined by an
be isomorphic. edge) in that graph if and only if the correspond-
As a simple illustration, consider the two ing vertices f(vi) and f(vj) are adjacent in 3.2.1b.
graphs in Figure 3.2.1. Part (a) reproduces the For example, an edge joins v3 and v4, and corre-
hexatonic cycle from 3.1.3c, while (b) is a dia- spondingly an edge joins v3ʹ and v4ʹ. There is no
gram of a whole-​tone scale. Musically, the two edge joining v2 and v6, and correspondingly there
graphs represent two very different things: in the is no edge joining v2ʹ and v6ʹ.
first graph the vertices represent chords and the
edges indicate single-​semitone voice-​leading con- In terminology from Chapter 2, properties (a)
nections, while in the second graph the vertices and (b) mean that f is a bijection—​a one-​to-​one

Graphs • 87
correspondence between the vertices of the two Exercise 3.2.2 When two graphs are
graphs. Property (c) then says something more: isomorphic, there is often more than one
it ensures that corresponding vertices under possible isomorphism between them. The above
this bijection are in the same structural rela- discussion of the two 6-​cycles of Figure 3.2.1
tionships to each other. Specifically, the edges in referred to the specific isomorphism f(v1) =​
both graphs always join corresponding pairs of v1ʹ, f(v2) =​ v2ʹ, …, f(v6) =​ v6ʹ. This isomorphism
vertices. is the most obvious one, both because of the
The above properties are precisely what is way the graphs are drawn and because of the
required of an isomorphism of graphs. To be way the vertices are numbered. It is not the
explicit, suppose A and Aʹ are two graphs, with only possible isomorphism of the two 6-​cycles,
sets of vertices V and Vʹ respectively. An isomor- however. Construct a different isomorphism
phism from A to Aʹ is a bijective function f: V → Vʹ from Figure 3.2.1a to 3.2.1b. How many
that preserves adjacency. That is, if v1 and v2 are different isomorphisms are possible altogether?
two elements of V (vertices of the graph A), then How many different isomorphisms are possible
f(v1) and f(v2) are elements of Vʹ (vertices of Aʹ), from chromatic pitch-​class space to the circle
and it is required that f(v1) and f(v2) be adjacent of fifths? Also construct a bijection from the
(joined by an edge) in Aʹ if and only if v1 and v2 vertices of 3.2.1a to those of 3.2.1b that is not
are adjacent in A. If an isomorphism from A to Aʹ an  isomorphism.15
exists, we say that the graphs A and Aʹ are isomor-
phic, written A ≅ Aʹ. The power of isomorphisms is most evident
Figures 3.1.1, 3.1.5, 3.1.7, and 3.1.10, our four in the comparison of graphs whose vertices do
representations of the hexatonic triad graph, are not bear the same labels, like the two 6-​cycles
all isomorphic. The isomorphisms in this case of Figure 3.2.1. For another example consider
are apparent because of the way the vertices in Figure 3.2.3, a graph whose vertices represent all
the graphs are labeled: the eight vertices bear the the forms of the twelve-​tone row appearing in
same triadic labels in all four graphs, and map- Webern’s Three Songs, Op. 23. The prime form
ping the vertices of one graph to the correspond- P0 for Op. 23 is the row form in the voice at the
ingly labeled vertices in another graph (B+​ → B+​, beginning of the third song, “Herr Jesus mein.”
c → c, and so on) defines an isomorphism in each As is often the case in Webern’s serial works, the
case. It is the identical vertex labels that jus- row forms used here follow a particularly system-
tify calling these four graphs “the same graph” atic scheme. Besides P0, the other rows include I0
rather than just “isomorphic graphs” like the (the inverted form beginning on the same pitch
pair in Figure 3.2.1. The isomorphism, however, class as P0); R0 and RI0 (the retrogrades of P0
conveys the essential similarity of graph struc- and I0); and the tritone transpositions of those
ture; whether two identically labeled graphs are four rows, P6, I6, R6, and RI6. In serial analy-
actually “the same” or merely “isomorphic” is a ses in this book, we will follow a convention of
distinction that we need not concern ourselves using bold letters for row forms, which enables
with too much. us to distinguish bold I and R, representing row
It is important to be aware that even if the forms, from italic I and R, which represent trans-
vertices of two graphs bear the same labels, and formations applied to row forms.16
even if the graphs are isomorphic, mapping cor- The graph shows one way of conceiving serial
respondingly labeled vertices onto each other relationships among these eight row forms.
will not always necessarily produce an isomor- Two row forms are joined by an edge if they are
phism. As an illustration, recall the diagrams related by one of the three transformations I
in Chapter 1 of chromatic pitch-​class space pc (inversion about the initial pitch class of P0), R
(Figure 1.2.3) and fifths space fifth (Figure (retrograde),17 or T6 (transposition by tritone).
1.3.1). These two graphs are isomorphic: both The graph takes the form of a cube. Edges ori-
are 12-​cycles. But in this case the mapping C → ented horizontally, along what we previously
C, C♯ → C♯, D → D, … is not an isomorphism. That called the x axis, join row forms related by inver-
mapping is a bijection, but it does not preserve sion; edges oriented along the y axis join row
adjacency: C is adjacent to C♯ in pc but not in forms related by T6; and edges oriented verti-
fifth. The easiest way to construct an isomor- cally along the z axis join retrograde-​related row
phism between these two graphs is to associate forms. It is clear from this arrangement that
vertices in corresponding places on the circles, Figure 3.2.3 is isomorphic to Figure 3.1.7, the
mapping C → C, C♯ → G, and so on. cubic version of the hexatonic triad graph. This

88 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 3.2.3 A graph showing relationships among some forms of a twelve-​tone row

isomorphism is valid even though the musical (T) Finally, transitivity requires that if A ≅ Aʹ
phenomena depicted in the two graphs—​voice-​ and Aʹ ≅ Aʹʹ, then A ≅ Aʹʹ. If f: V → Vʹ and g:
leading connections among triads and serial Vʹ → Vʹʹ are isomorphisms, the desired
operations on twelve-​ tone rows—​ may seem isomorphism from A to Aʹʹ is the composite
unrelated. Of course, the cubic graph is a rela- function f ● g.
tively simple structure, and as with other simple
graphs such as paths or cycles, it should not sur- It is transitivity that allows us to conclude that
prise us to see such a structure arising in a vari- Figures 3.1.1 and 3.2.3 are isomorphic, by this
ety of contexts.18 logic: 3.1.1 ≅ 3.1.7 and 3.1.7 ≅ 3.2.3; therefore
If Figure 3.2.3 is compared to the original 3.1.1 ≅ 3.2.3. In fact, transitivity shows explic-
hexatonic triad graph in Figure 3.1.1, it may not itly how to construct an isomorphism from 3.1.1
be at all apparent that they are isomorphic, as to 3.2.3, once the two isomorphisms with 3.1.7
these two graphs have different vertex labels are known. For example, C in 3.1.1 corresponds
as well as different visual appearances. They with C in 3.1.7, which is at the same location
must be isomorphic, however, because we have as I6 in 3.2.3; the composite isomorphism f ● g
now established that both of them are isomor- therefore maps the vertex labeled C in 3.1.1 to
phic to Figure 3.1.7. Drawing this conclusion the vertex labeled I6 in 3.2.3.
depends on the fact that graph isomorphism is an
equivalence relation. As detailed in Section 2.4, Exercise 3.2.4 One of the three graphs you
this means that the relation of isomorphism is constructed in Exercise 3.1.17 is isomorphic to
reflexive, symmetric, and transitive. It is instruc- the graph in one of the figures in Section 3.1.
tive to consider briefly why each of these proper- Which
   two graphs are these?
ties holds:
Isomorphic graphs automatically share basic
(R) Reflexivity means that A ≅ A for every graph graph-​theoretic properties. If A ≅ Aʹ, the iso-
A: every graph is isomorphic to itself. In morphism guarantees that A and Aʹ have the
fact, the identity function on the vertices same number of vertices and the same number
of any graph (f(v) =​ v for every vertex v) is of edges, and corresponding vertices in the two
automatically an isomorphism. graphs have the same degrees. If A is regular, so
(S) The symmetric property states that if is Aʹ; the same is true if connected, bipartite, or
A ≅ Aʹ, then Aʹ ≅ A. This is true because if planar is substituted for regular. It is possible for
a function f: V → Vʹ from the vertices of a planar embedding, such as Figure 3.1.1, to be
A to the vertices of Aʹ is an isomorphism, isomorphic to a graph that is not drawn as a pla-
then it is a bijection, so it has an inverse nar embedding, such as Figure 3.1.7. Recall, how-
function f −1: Vʹ → V. Because f preserves ever, that for a graph to be considered planar we
adjacency, f −1 does as well, so f −1 is an iso- require only that a planar embedding be possible,
morphism from Aʹ to A. so the hexatonic triad graph is planar no matter

Graphs • 89
 isomorphism from X to Xʹ is a structure-​preserving
bijection. An isomorphism must first be a bijec-
tion—​a one-​to-​one correspondence between
the elements of the two structures. But more is
generally required for a function f to be an iso-
morphism: f must also preserve structure in the
sense that if elements x1, x2, … of X are in some
FIGURE 3.2.5 Two non-​isomorphic graphs shar- structural relationship in X, then their images
ing many features f(x1), f(x2), … are in the same relationship in Xʹ.
The language about “some structural relation-
ship” is necessarily vague, because we have
how it is drawn. If two planar embeddings are not specified the nature of the structures X
isomorphic, they have the same number of faces. and Xʹ. If the structures are graphs, then the
If two isomorphic graphs are not connected, structural relationships involve adjacency of
they have the same number of components. vertices, but other types of structures entail
Isomorphism implies many things, but one different sorts of relationships among the ele-
must be careful not to assume that some of these ments, which the appropriate kind of isomor-
things, even in combination, imply isomorphism. phism must preserve.
Consider the two simple graphs in Figure 3.2.5.
Each of these graphs has the same number of 3.3 LOOPS, MULTIPLE
vertices (4) and edges (4); both graphs are pla- EDGES, AND INFINITE
nar embeddings, both have the same number of
faces (2), and both are connected. But they are
GRAPHS
not isomorphic, as can be seen by noting that the The graphs presented in the preceding sections
degrees of the vertices do not match: the cycle in are the simplest sort of graphs studied in graph
(a) is regular, with all its vertices having degree theory: finite undirected graphs without loops
2, while the graph in (b) has one vertex of degree or multiple edges. We will have occasion to waive
1 and one of degree 3. No adjacency-​preserving all of these restrictions from time to time; in the
bijection could map one of the degree-​2 vertices remainder of this chapter we examine a few of
of (a) onto the degree-​1 vertex of (b). these extensions of the graph concept.20
Figure 3.3.1, based on a graph of Joseph
Exercise 3.2.6 The following assertion once Straus, is a voice-​leading graph of trichordal set-​
appeared in a music theory article: If two graphs class space, depicting semitonal voice-​ leading
have (a) the same number of vertices, (b) the connections among the twelve three-​note set
same number of edges, (c) the same distribution classes, here identified by prime forms.21 The
of degrees among the vertices, and (d) the edge connecting 025 to 036, for example, shows
same number of components, then they are that a set of type 025 may be converted into a set
isomorphic. Construct an example to show that of type 036 by moving one voice by semitone (for
this assertion is false. That is, construct two instance {0, 2, 5} → {11, 2, 5}). This graph con-
graphs that are not isomorphic even though tains four loops—​edges joining a vertex to itself.
conditions
   (a)–​(d) are all true.19 For example, there is a loop on the vertex 025,
because any set of this type may be converted
We used the term isomorphic as early as into another set of the same type by a semitonal
Chapter 1, where several of the spaces we exam- displacement (for instance {0, 2, 5} → {0, 3, 5}).
ined were said to be isomorphic. Those isomor- There are no loops on most vertices, because
phisms sometimes involve graphs, but they for most set classes single-​semitone motion to
often involve additional elements of structure another set of the same type is not possible.
besides the graph. In later chapters we will give The arrangement of the vertices in this fig-
explicit definitions of isomorphism for sev- ure deserves notice. After the initial 0, common
eral other kinds of structures, including groups to all the prime forms, the second and third
(Section 6.3), interval spaces (Section 7.2), and numbers are arranged in a straightforward two-​
transformation graphs (Section 9.3). dimensional coordinate system, with the second
A common strategy underlies the various number (which ranges from 1 to 4) graphed ver-
types of isomorphisms. In general, if X and Xʹ tically and the third number (from 2 to 8) hori-
are mathematical structures of some sort, an zontally. The graph is a planar embedding, and

90 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 3.3.1 A voice-​leading graph of trichordal set-​class space (after Straus)

many of the edges seem to be organized in a sys- property that every set may be converted
tematic arrangement. Nevertheless, the graph to another set of the same type by single-​
as a whole has an oddly asymmetrical appear- semitone voice leading in two different ways.
ance, due in part to the constraint provided by Which
   set class is this?
the set classes themselves: there is no vertex in
the 017 position, for example, because there is Exercise 3.3.3 What is the greatest distance
no set class with that prime form (the set {0, 1, between any two vertices in Figure 3.3.1?
7} being inversionally related to {0, 1, 6}). Also Which pair(s) of set classes are separated from
there is no obvious pattern to the positions of each other by that maximum distance? Choose
the four loops. such a pair of maximally distant vertices—​call
In contemplating the appearance of Figure them set classes X and Y—​and write, in musical
3.3.1, it is worth remarking on an important dif- notation, a series of trichords, each related to
ference between this graph and some of those the next by a one-​semitone displacement in
presented earlier, such as the hexatonic triad one voice, that moves from a chord of type X
graph of Figure 3.1.1. The vertices in 3.1.1 rep- to a chord of type Y as efficiently as possible,
resented specific pitch-​ class sets, which are following a path through the graph. (Recall that
fairly concrete musical objects (if perhaps not “distance” refers to the number of edges in the
as concrete as pitch sets would be). The vertices shortest possible path joining two vertices, so
in 3.3.1, on the other hand, represent objects the distance will be the number of semitone
at a higher level of abstraction—​ set classes, moves in your chord progression, not the
which are equivalence classes of pitch-​class sets number of chords.) Must one voice sustain a
modulo an equivalence relation (TI equivalence, common tone through all the chords of such a
described in Section 2.4). As the discussion in progression?
  
Section 3.1 suggests and as we will see in detail
in Part Three, the hexatonic triad graph is a sub- Exercise 3.3.4 Construct a voice-​leading
set of a larger space comprising all three-​note graph of dyadic set-​class space, analogous to
pitch-​class sets. Because set classes are depen- Figure 3.3.1 but using two-​note rather than
dent upon an additional equivalence relation,    note set classes.22
three-​
however, Figure 3.3.1 is not something as simple
as a subset of that space; it is a quotient space. Exercise 3.3.5 It is tempting to suppose
A closer inspection of this quotient space in that the following statement about Figure
Section 11.4 will shed light both on the general 3.3.1 is true: If we start with any trichord and
shape of Figure 3.3.1 and on the placement of displace one voice at a time by semitone, any
the loops in the graph. sequence of sonorities we can produce will trace
a path through the graph. This statement is
Exercise 3.3.2 Provide examples of voice not accurate as it stands, however, because a
leadings for all four of the loops in Figure semitonal displacement such as {0, 1, 4} → {0,
3.3.1. One of these set classes has the 0, 4} produces a sonority that is not a trichord

Graphs • 91
at all in the sense of pitch-​class set theory Exercise 3.3.8
and therefore does not appear anywhere in (a) Construct common-​tone graphs, in the
the figure. Musically, {0, 0, 4} is a dyad with manner of Figure 3.3.6, corresponding to
a doubled note; mathematically it is not a the graphs you constructed in parts (a) and
three-​note set but a multiset as defined in (b) of Exercise 3.1.17.
Section 2.2. (b) A common-​tone graph corresponding to the
This observation suggests that a more octatonic seventh-​chord graph of Exercise
complete picture of trichordal voice-​leading 3.1.17c is probably too unwieldy to draw.
space might include multiset classes such Counting all the multiple edges, calculate
as 004. Construct this larger graph. When   how many edges such a graph would have.25
you are finished, you should have a graph
for which the italicized statement in the The pitch spaces studied in Chapter 1 are infi-
preceding
   paragraph is true.23 nite sets, and several of the diagrams presented
there may be considered infinite graphs.26 The
Figure 3.3.6 introduces another new feature: diagram of discrete chromatic pitch space pitch
double edges. This graph shows the same major in Figure 1.1.1, for example, may be regarded as
and minor triads as in the hexatonic cycle of a graph with infinitely many vertices and edges.
Figure 3.1.3c, but models a different relation In Section 1.1 we called the line in this figure
between these triads. Specifically, each edge in a “fiction” as far as the representation of a dis-
this graph represents one common tone between crete space was concerned—​but now we may
two triads. The graph displays a regular pattern: consider the line to depict the edges joining
each triad in the hexatonic cycle shares two com- successive vertices. An infinite graph cannot be
mon tones with its two immediate neighbors in drawn in its entirety, but the infinite sets of ver-
the cycle, one common tone with the other two tices and edges and the adjacency relation can be
triads of the same mode, and no common tones described mathematically. Many infinite graphs
at all with the triad on the opposite side of the feature a simple repeating pattern that can be
cycle (called its hexatonic pole).24 conveyed by drawing a small part of the graph,
as the examples from Chapter 1 illustrate.
Exercise 3.3.7 Each vertex in Figure 1.1.1 is adjacent to
(a) The graph in Figure 3.3.6 is not a planar exactly two other vertices; for example, C4 is
embedding, as there are six crossings adjacent to B3 and C♯4. In other words, every
of the single edges in the interior of the vertex is of degree 2, and the graph is regular
hexagon. This graph is actually planar, as defined in Section 3.1. Other terms, such as
however. Redraw the graph to show a planar connected, bipartite, and planar also carry over
embedding. to infinite graphs. Two infinite graphs may
(b) If the double edges in Figure 3.3.6 are be shown to be isomorphic by constructing an
replaced by single edges, the resulting graph adjacency-​preserving bijection between them as
is isomorphic to a graph in a figure earlier in before, the only difference being that the func-
  this chapter. Which graph is this? tions are now defined on infinite sets.

Exercise 3.3.9
(a) Is the infinite graph of Figure 1.1.1
connected? Bipartite? Planar?
(b) Exercise 1.1.9c asked for an informal
explanation why discrete chromatic pitch
space pitch and generic pitch space gpitch
are isomorphic. Make this explanation
more precise by showing how to construct
an isomorphism between the two graphs.
If the isomorphism f: pitch → gpitch is
defined so that f(C4) =​C4 and f(C♯4) =​D4,
  what is f(C5)?

FIGURE 3.3.6 A common-​tone graph of a hexa- The special nature of tonnetz graphs merits
tonic cycle some attention. The triangulated tonnetz of

92 • E x p l o r i n g M u s i c a l S p a ces
Figure 1.4.5, shown again in Figure 3.3.10a, and Exercise 3.3.11
the three alternate representations from Figure (a) The edge that slopes southwesterly from
1.4.10 may all be regarded as infinite graphs. All D in Figure 3.3.10b arrives at what other
four of these graphs are isomorphic: adjacency-​ vertex?
preserving bijections from the vertices of one (b) Define a function from the vertices of
graph to identically labeled vertices in another Figure 3.3.10a to the vertices of 3.3.10b in
graph should be fairly easy to visualize. While such a way that each vertex of (a) is mapped
these graphs are infinite, we must remember to the unique vertex of (b) bearing the
that the space they represent—​discrete pitch-​ same label. Is this function one-​to-​one? Is
class space pc—​is finite. The conformed prop- it onto? Does it preserve adjacency? Is it
erty of the tonnetz implies that the four vertices an isomorphism? Does it have an inverse
labeled “G” in Figure 3.3.10a all represent the function? Is it possible to construct a one-​
same note in pitch-​class space, and the entire to-​one adjacency-​preserving function from
infinite graph shows only twelve different pitch the vertices of 3.3.10b to the vertices of
classes.   3.3.10a?27
As discussed in Chapter 1, there are advan-
tages and disadvantages to such a representa- The conformed-​ tonnetz interpretation of
tion. In any case, a more efficient graph of the pitch-​class space is thus represented by two dif-
tonnetz is at hand, in the modular tile from ferent graphs, one infinite and the other finite.
Figure 1.4.3, redrawn in Figure 3.3.10b with This point warrants elaboration, because these
the addition of the triangulating diagonals. This two graphs and the relationship between them
graph has only twelve vertices, not infinitely are occasional sources of misunderstandings in
many, and they are in one-​to-​one correspon- neo-​Riemannian theory. One possibly confusing
dence with the notes of pc. The broken edges point is the fact that although the finite graph
at the boundaries of the picture must be read of Figure 3.3.10b looks like a part of the infinite
with the understanding of the toroidal topology graph of 3.3.10a, it is not technically a subgraph.
implied by the arrowheads on the bounding rect- To be sure, one may identify the vertices of (b)
angle: when an edge encounters the boundary with correspondingly labeled vertices in a cen-
on one side of the diagram, it is understood to tral block of (a), and the unbroken edges in the
continue at the opposite side. For example, the central part of (b) then correspond to edges of
edge sloping northeasterly from the note B (a). But the broken edges in (b) are a problem:
near the lower-​right corner of the figure arrives subgraph status would require that all edges of
at F♯ on the left side. Diagonals in this tonnetz (b) be literally present within the specified sub-
represent perfect fifths, so of course this edge set of (a), while an edge such as B–​F♯ in (b) joins
from B must arrive at F♯: B and F♯, despite their two vertices that are not adjacent in (a). Rather
distant positions in the picture, are adjacent ver- than a subgraph, 3.3.10b might more accurately
tices in the graph. be called a quotient graph of 3.3.10a. The finite

FIGURE 3.3.10 The triangulated tonnetz as (a) an infinite graph; (b) a finite graph

Graphs • 93
graph can be constructed from the infinite one
not by isolating a subset, but instead by conjoin-
ing identically labeled vertices—​an equivalence
relation.
Confusion also sometimes arises about
whether the tonnetz is a planar graph. This ques-
tion is subtler than it may appear. The infinite
graph of Figure 3.3.10a is planar; to be precise,
that figure shows a finite portion of a planar
embedding of an infinite graph. The alternate FIGURE 3.3.13 K5, the complete graph on five
representations of infinite tonnetz graphs in vertices
Figure 1.4.10 are also planar embeddings.
The subtlety emerges when the finite tonnetz
graph of Figure 3.3.10b is considered. This fig- drawn with fewer than the five crossings visible
ure is drawn in a plane, and no edge crossings in Figure 3.3.13, but it cannot be drawn with no
are visible, so it is tempting to conclude that this crossings at all. See Exercise 3.3.15 below for a
graph too is planar. But this reasoning is falla- proof that K5 is not planar. Here we will use this
cious. In the context of planar embeddings, the result to show that the finite tonnetz graph can-
broken edges are tantamount to cheating: after not be planar because a copy of K5 lurks within it.
all, edge crossings in any graph can be eliminated A planar graph cannot have a non-​planar sub-
by breaking some edges in half. If you try to graph: a planar embedding of the larger graph, if
extend the ends of the broken edges in the plane it exists, must include a planar embedding of the
of 3.3.10b so that each meets its partner in a smaller one. If we can find a subgraph of Figure
continuous curve, you will find yourself unable 3.3.10b isomorphic to the non-​planar K5, there-
to do so without introducing some edge cross- fore, that will prove that the finite tonnetz graph
ings, for the finite tonnetz graph is not a planar is non-​planar. Figure 3.3.10b does not actually
graph. To depict the tonnetz as a finite graph, we have a subgraph isomorphic to K5, but it has a
must sacrifice planarity: the options are an infi- subgraph that works just as well for this purpose.
nite planar graph or a finite non-​planar graph. This is the subgraph in Figure 3.3.14a, compris-
We present the non-​planarity of the finite ton- ing ten vertices and fifteen edges from the ton-
netz formally in Theorem 3.3.12 below, followed netz. Figure 3.3.14a is isomorphic to Figure
by a proof. The proof depends on a standard 3.3.14b; the vertex labeling shows the isomor-
result in graph theory, the non-​planarity of the phism explicitly. Graph (b) is not isomorphic to
graph known as K5, a proof of which is outlined K5 (it has too many vertices), but its resemblance
in Exercise 3.3.15. Some readers may choose to to K5 is clear. Figure 3.3.14b is called a subdivi-
skip these proofs, but it seems worthwhile to sion of K5: it may be obtained from K5 by adding
include them here because the results are not new vertices in the middle of some of the edges.
widely known in music theory.28 Additionally, In the figure, the five vertices not part of K5 are
the proofs themselves may be instructive, as labeled in smaller type than the others. If a pla-
they introduce a few new ideas from graph the- nar embedding of 3.3.14b were possible, these
ory in combination with other concepts already extra vertices could be eliminated and a planar
presented, and Exercise 3.3.15 provides an illus- embedding of K5 would result. Because we know
trative model of a proof by contradiction. that is impossible, we may conclude that Figure
3.3.14b is not planar, so the isomorphic graph in
Theorem 3.3.12: Non-​planarity of the finite (a) is not planar either, nor is the finite tonnetz
tonnetz graph The finite tonnetz (Figure graph of which 3.3.14a is a subgraph.
3.3.10b)
   is not a planar graph.
Exercise 3.3.15 This exercise outlines a
The proof of the theorem makes use of the proof that K5 is not planar.29 We employ the
graph in Figure 3.3.13. This graph is called the deductive technique of proof by contradiction.
complete graph on five vertices, commonly denoted That is, we start by assuming the opposite of
K5; the word complete indicates that each vertex what we are trying to prove, namely that a
is adjacent to every other. K5 is known as one of planar embedding of K5 does exist; the aim is
the simplest non-​planar graphs. Edge crossings then to derive some logical contradiction from
are unavoidable in drawing this graph: it can be this assumption. If we can do this, we will be

94 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 3.3.14 A subgraph of the finite tonnetz, isomorphic to a subdivision of K5

justified in concluding that the assumption this contradiction completes the proof that
was false, and therefore that K5 is not a   K5 is not a planar graph.
planar graph.
Although the finite tonnetz cannot be embed-
(a) Suppose a planar embedding of K5 exists. ded in a plane, it can be embedded in a torus.
Such an embedding, like any planar Indeed, it is because of the non-​planarity of the
embedding of a graph, will have well-​ tonnetz that we must tolerate the sometimes
defined faces. Some of the faces may be perplexing aspects of the toroidal representa-
triangles with three sides, while others tion: if the finite graph could be embedded in a
may have more than three sides. Any faces plane, that embedding would surely be more con-
with more than three sides can always be venient. The convention of depicting the torus
triangulated—​divided into triangles by by a rectangle with opposite sides identified
the addition of new edges to the graph. allows us to regard the broken edges in Figure
For example, a face with four sides can be 3.3.10b as whole edges, without crossings. The
divided into two triangles by drawing one doughnut-​shaped tonnetz of Figure 1.4.4 is also
new edge. Suppose that all the faces of the revealing in this context. The graph inhabits the
embedding of K5 (including the exterior surface of the doughnut, and this surface corre-
face, if necessary) are divided into triangles sponds to the rectangular tile of Figure 3.3.10b.
in this way, and call the resulting fully The edges corresponding to 3.3.10b’s diagonals
triangulated graph Kʹ. Suppose Kʹ has v were not shown in Figure 1.4.4, but they could
vertices, e edges, and f faces. Because every be added, within the surface of the doughnut,
face of Kʹ is bounded by exactly three edges, with no crossings.
it is tempting to conclude that 3f =​ e, but
that is not correct. Explain why this formula Exercise 3.3.16
is wrong, and make a simple modification (a) How many vertices and edges are in the
so that it becomes correct. Rewrite the finite tonnetz? How many faces are in the
formula in the form f =​ _​_​_​, where the toroidal embedding of Figure 3.3.10b?31
right side of the equation is an expression (b) The basic statement of Euler’s formula, v +​
involving e but not f.30 f =​ e +​2, holds only for connected graphs
(b) Recall that by Euler’s formula, v +​ f =​ e +​2. embedded in a plane (or a sphere), not in a
We know that v =​5, because the vertices torus. Your answers to (a) should confirm
of Kʹ are the vertices of the graph K5 that that this formula fails for the finite tonnetz
we started with. Plugging v =​5 and the graph. A more general version of Euler’s
result of part (a) into Euler’s formula, you formula is applicable, however. The general
will get a formula involving e and no other statement is v +​ f =​ e +​ χ, where χ (the Greek
variables. Use basic algebra to solve this letter chi) is a number called the Euler
equation for e. characteristic of the surface in which the
(c) Noting that Kʹ, by the nature of its graph is embedded. The Euler characteristic
construction, has at least as many edges as of a sphere is 2, but the Euler characteristic
K5, explain why the result of part (b) is a of a torus is 0, so Euler’s formula for a
contradiction. By the logic described above, toroidally embedded connected graph is

Graphs • 95
 simply v +​ f =​ e. Verify that the numbers you for the continuous space, and this discrete grid
  obtained in (a) satisfy this formula.32 may help us to understand the structure of the
more complete continuous space. In Part Three
Because the tonnetz graph is embeddable in we will see more complex examples of continu-
a torus and a subdivision of K5 occurs within it, ous spaces that can be most easily visualized by
we may conclude that K5 itself is embeddable drawing a discrete subset in this way.
in a torus. In fact, more is true: K6 and K7, the
complete graphs on six and seven vertices, can 3.4 DIRECTED GRAPHS
each be embedded in a torus. We will encounter a
musical application of K7 in Section 4.3 for which A number of music theory textbooks include
the toroidal embedding will prove convenient. In diagrams resembling Figure 3.4.1, attempting to
the meantime, readers may enjoy the challenge show patterns formed by the most common func-
of trying to find this embedding on their own, in tional diatonic progressions in a major key.33 This
the way suggested in the following exercise. figure is not simply a graph like those discussed
above, but a directed graph: each edge is deco-
Exercise 3.3.17 The generic tonnetz of Figure rated with an arrowhead to indicate its direction.
1.4.27 shows a placement of seven vertices The figure shows, for instance, that motion from
in a torus. By joining these vertices with vi to ii is possible while the oppositely directed
carefully chosen edges, construct a toroidal ii → vi is not. The diagram resembles a flowchart
embedding
   of K7. for a computer program or decision-​making pro-
cedure, showing in a compact way which options
In Section 1.4 we distinguished the conformed are available at any moment. The arrowheads are
neo-​Riemannian tonnetz, which we have dis- an essential part of the information conveyed;
cussed further in this section, from the uncon- indeed, the figure would be meaningless without
formed Riemannian tonnetz. The latter was them, as every vertex is joined to every other
drawn without edges in Figure 1.4.14, but trian- vertex by an arrow in at least one direction. It
gulating edges could readily by added, yielding is possible for two vertices to be joined by a pair
an infinite graph isomorphic to Figure 3.3.10a. of opposing arrows ⇄, as is the case with the I
The preceding sentence should be read with cau- and V chords in the figure. A single bidirectional
tion, for while the two graphs are isomorphic, arrow ↔ may be used as a shorthand for the pair
the spaces they represent are not. Conformed of opposing arrows in this situation.
tonnetz space is finite; Riemannian tonnetz space Directed graphs entail some modest modi-
is infinite. The unconformed Riemannian ton- fications to the terminology introduced previ-
34
netz does not wrap around on itself as the torus ously for undirected graphs. The term directed
♯
does; the various vertices labeled C, B , and D♭ ♭ graph is sometimes shortened to digraph. It is
in Figure 1.4.14 all represent different notes. traditional to speak of arrows rather than edges.
The finite graph in Figure 3.3.10b has no coun- Whereas the degree of a vertex in an undirected
terpart for the unconformed tonnetz: while it is graph was defined simply as the number of
possible to represent a finite space using an infi- edges incident with it, in digraphs the in-​degree
nite graph as in 3.3.10a, an infinite space such as of a vertex (the number of arrows pointing into
Riemannian tonnetz space can never be depicted it) must be distinguished from its out-​degree
accurately by a finite graph. (the number of arrows pointing out). In Figure
Whether finite or infinite, planar or non-​ 3.4.1, the vertex labeled I has in-​degree 1 and
planar, graphs are by their very nature discrete
structures. If Figure 1.1.1 is viewed as continu-
ous pitch space cpitch, it is not a graph at all in
the sense of this chapter. Nevertheless, it is clear
that our diagrams of continuous spaces such
as cpitch (Figure 1.1.1) or cpc (the continu-
ous pitch-​class space of Figure 1.2.3) bear some
relation to the graphs of the corresponding dis-
crete spaces. A discrete graph may be effectively
converted into a continuous space by regarding
every edge as an infinite set of new points of the FIGURE 3.4.1 A directed graph showing com-
space. The discrete graph becomes a sort of grid mon functional diatonic progressions

96 • E x p l o r i n g M u s i c a l S p a ces
out-​degree 4, suggesting that a larger variety of relation y ~ x. An undirected graph therefore
chords may follow a tonic triad than precede it. can depict only relations that are symmetric as
The opposite is true of V, which has in-​degree defined in Section 2.4—​relations with the prop-
4 and out-​degree 1, implying that while many erty that whenever x ~ y is true, then y ~ x is
things can precede a dominant, what happens also true. With directed graphs there is no such
next is much more predictable. While the precise restriction. The arrows in Figure 3.4.1 show that
counts of in-​degrees and out-​degrees here may on the set of diatonic triads, the relation “may
oversimplify a complex situation, these general be followed by” is not symmetric: vi may be fol-
conclusions are hard to dispute. lowed by ii, but not vice versa.
A directed path is a path in which all arrows To be sure, sometimes the information com-
are traversed in the same direction. In Figure municated by the arrowheads in a digraph
3.4.1, we may trace a directed path I → vi → may be conveyed by a different means entirely.
ii → V. Ignoring arrowheads would enable us also Figure 3.4.3 recasts the hexatonic triad graph
to trace a path I –​ii –​vi –​V, but that path is not from Figure 3.1.1 two more times, now as
directed: it cannot be traversed, either from I to digraphs. The arrangement of the vertices in
V or in reverse from V to I, without traversing at 3.4.3a matches that of Figure 3.1.1, while 3.4.3b
least one arrow backward. Likewise a cycle may matches Figure 3.1.5. The arrows in both graphs
be either directed or undirected. point in the direction of ascending voice leading.
This same relation was already depicted in Figure
Exercise 3.4.2 3.1.5 by the relative vertical positions of the ver-
(a) Although the in-​degree and out-​degree of a tices; hence in 3.4.3b all the arrows point upward.
vertex may differ, the sum of the in-​degrees This figure therefore includes some redundancy,
of all the vertices in a directed graph must as the relation represented by the arrows may
always match the sum of all out-​degrees. already be inferred from the visual arrangement.
Verify this property in Figure 3.4.1, and In 3.4.3a, on the other hand, the arrows commu-
explain why it must always hold. nicate something that is not otherwise apparent:
(b) Identify all the directed 3-​cycles (cycles they do not all point in the same direction, and
containing three vertices), 4-​cycles, and 5-​ the “highest” chord in the voice-​leading sense,
  cycles in Figure 3.4.1. C+​, does not appear at the top of the figure. Figure
3.4.3b, like 3.1.5, includes edge crossings, while
The use of arrows offers additional precision 3.4.3a, like 3.1.1, is a planar embedding. Only by
and flexibility to graphs. It was remarked in means of arrows is Figure 3.4.3a able to capture
Section 3.1 that graphs are commonly used to the voice-​ leading relationship shown in 3.1.5
depict binary relations, but undirected graphs while retaining the planar configuration of 3.1.1.
tacitly impose a limitation on these relations. If The highest and lowest vertices in Figure
two vertices in a graph represent objects x and 3.4.3b, and their counterparts in 3.4.3a, have
y, an edge without an arrowhead cannot distin- special properties: B+​ has in-​degree 0, and C+​
guish between a relation x ~ y and the opposite has out-​degree 0. B+​ is not accessible by directed

FIGURE 3.4.3 Two digraph representations of the hexatonic triad graph

Graphs • 97
path from any other vertex, nor does a directed It is not always possible to arrange the verti-
path lead from C+​ to any other vertex. A vertex of ces of a directed graph so that all arrows point
in-​degree 0 is called a source, and a vertex of out-​ upward. The possibility of such an arrangement,
degree 0 is a sink. (Of course these properties as in Figure 3.4.3b, indicates a rather special
arise only relative to a given graph. To say that property of a graph. To describe this property
C+​ is a sink in this context does not imply any requires some additional terminology. If v1 and
special musical property of that triad; it simply v2 are vertices in a digraph, we say that v1 pre-
means that if any voice of the C-​augmented triad cedes v2, notated v1 ⇢ v2, if there is some directed
is raised by semitone, the resulting sonority is path from v1 to v2. This includes the case in
something not in this graph.) A directed graph is which there is an arrow directly from v1 to v2, but
strongly connected if every vertex is accessible by also the case in which the two vertices are joined
directed path from every other vertex. Viewed only by a longer path. We also understand by
as undirected graphs, Figures 3.4.1, 3.4.3a, and this definition that every vertex automatically
3.4.3b are all connected. As directed graphs, precedes itself; this is justifiable on the grounds
however, Figure 3.4.1 is strongly connected, but that a vertex by itself may be considered to be a
the graphs of Figure 3.4.3 are not—​because of path of length 0, starting and ending at the same
the absence of a directed path from C+​ to B+​, for point. If v1 ⇢ v2, then v1 is called a predecessor of
example. v2, and v2 a successor of v1. This precedence rela-
It should come as no surprise that the two tion effectively extends the relation depicted by
graphs of Figure 3.4.3 are isomorphic. The defi- the arrows to accommodate pairs of vertices sep-
nition of isomorphism, however, also requires arated by greater distances; in Figure 3.4.3, for
adjustment in the case of directed graphs. As example, e precedes C+​, even though those two
before, an isomorphism is first of all a bijection vertices are not directly adjacent, because there
from the vertices of one graph to the vertices of is a directed path e → C → C+​.
the other, but it is no longer quite sufficient to Because every vertex precedes itself, the pre-
require that the bijection preserve adjacency; it cedence relation ⇢ is reflexive by the definition
must preserve the direction of each arrow. That from Section 2.4: the statement v ⇢ v is true for
is, if an arrow points from v1 to v2 in one graph every vertex v. Precedence is also automatically
and f is an isomorphism, then there must be an transitive: if v1 ⇢ v2 and v2 ⇢ v3 are true, then a
arrow from f(v1) to f(v2) in the second graph. The path from v1 to v2 can be combined with a path
isomorphism from Figure 3.4.3a to 3.4.3b maps from v2 to v3 to obtain a path from v1 to v3, so
each vertex to the correspondingly labeled ver- v1 ⇢ v3 is true as well.36 The relation ⇢ is not nec-
tex, and it is easy to check that arrows connect essarily symmetric, however, and therefore is not
all the same pairs of vertices, always in the same generally an equivalence relation. If both v1 ⇢ v2
directions, in both figures. and v2 ⇢ v1 are true for some pair of vertices v1
and v2, the two paths can be combined to form a
Exercise 3.4.4 directed cycle. In a graph without directed cycles,
(a) What is the length of the longest directed the precedence relation is antisymmetric, which
path in Figure 3.4.3a or 3.4.3b? Are there means that one vertex can never be both prede-
any directed cycles in this graph? cessor and successor to another—​the relations
(b) The 4-​cycle containing the triads C, C+​, E, v1 ⇢ v2 and v2 ⇢ v1 cannot both be true unless v1
and e in Figure 3.4.3a is not a directed cycle. and v2 are the same vertex.
This subgraph may, however, be viewed as A binary relation that is reflexive, antisym-
a directed graph in its own right. Find a metric, and transitive is called a partial order.37
subgraph of Figure 3.4.1 that is isomorphic As noted above, the precedence relation in a
to this subgraph of 3.4.3a. directed graph is always reflexive and transitive.
(c) Suppose we define the arrow relation in In Figure 3.4.3 (either (a) or (b)) the precedence
the hexatonic triad graph in the opposite relation is also antisymmetric, and is therefore
way from what was described above, so a partial order. The antisymmetry property, or
that the arrows represent descending equivalently the absence of directed cycles, is
motion by semitone. This change amounts what makes the upward-​pointing arrangement
to reversing the direction of every arrow of 3.4.3b possible: it ensures that no contradic-
in Figure 3.4.3a. Show that the resulting tion can arise in the process of placing succes-
directed graph is isomorphic to the sors above their predecessors. The word order
  original Figure 3.4.3a.35 in this context refers to the way the precedence

98 • E x p l o r i n g M u s i c a l S p a ces
relation arranges the vertices from predecessors Subset relations between sets that differ by two
to successors, or from lowest to highest in the or more elements (such as {1} and {0, 1, 3}) are
graph (indeed, we have already used the terms not shown directly by edges, but may be inferred
“high” and “low” in reference to Figure 3.4.3b). from the edges present and the transitivity of
The order relation in this figure is partial (not the subset relation ({1} ⊆ {0, 1} and {0, 1} ⊆ {0,
total) because the graph shows some pairs of 1, 3}, therefore {1} ⊆ {0, 1, 3}). The original set
vertices that are not in the precedence relation {0, 1, 3, 6} appears at the top of the lattice, and
to each other at all: neither E nor C, for example, the empty set ∅ at the bottom. No arrowheads
precedes the other. are shown; because the vertical dimension corre-
sponds to the precedence relation as in Example
Exercise 3.4.5 3.4.3b above, the arrowheads are superfluous
(a) Is the precedence relation in the digraph and may be omitted with no loss of information.
of Figure 3.4.1 antisymmetric? Does your Nevertheless, the antisymmetric precedence
answer change if the arrow from I to V is relation remains implicit in such diagrams, and
removed? upward-​pointing arrowheads on each edge may
(b) The most common mathematical use of the be inferred from the configuration.
word order refers to the ordering of the real The abstract subset relation on set classes is
numbers along the number line, described also a partial order. A set class X is an abstract
by the relation < (“less than”) or ≤ (“less subset of a set class Y if some set in the class X
than or equal to”). Verify that the relation is a subset of some set in Y. For example, rep-
≤, defined on the set ℝ, is a partial order as resenting set classes by their prime forms, the
defined above. (Actually ≤ is a total order, dyadic set class 03 is an abstract subset of 025
because given any two real numbers x and because any set of type 025, such as the set {0,
y, either x ≤ y or y ≤ x must be true. A total 2, 5}, has a subset of type 03, in this case the
  order is a special case of a partial order.) set {2, 5}. The abstract subset “lattice” of the set
class 0136 is shown in Figure 3.4.6b. The two
Partial order relations have many uses in graphs in Figure 3.4.6 are not isomorphic. They
mathematics, and often give rise to lattice-​like do not even have the same number of vertices:
graphs depicting hierarchical relationships such the set {0, 1, 3, 6} has six two-​note subsets and
as those in Figure 3.4.6. The subsets of any set four one-​note subsets, while the set class 0136
are partially ordered by the subset relation ⊆. has only five different abstract two-​note subsets
The graph in (a) is the subset lattice for the pitch-​ and (of course) only a single abstract one-​note
class set {0, 1, 3, 6}. Two subsets are joined by subset. This difference between the literal and
an edge in this graph only when the larger sub- abstract subset relations is a byproduct of the
set is obtained from the smaller by adjoining a TI equivalence relation that defines set classes.
single element (for example, {0, 1} and {0, 1, 3}). Whenever two different subsets such as {0, 3}

FIGURE 3.4.6 (a) A subset lattice for the set {0, 1, 3, 6}; (b) an abstract subset “lattice” for the set
class 0136

Graphs • 99
and {3, 6} belong to the same set class, they will Abstract subset “lattices,” for this reason, are
appear separately in 3.4.6a but will be merged not true lattices. We will continue to call them
into a single vertex in (b). The abstract subset “lattices,” however, because of their intuitively
“lattice” of (b) is less regular in its construction lattice-​like appearance and because they have
than the subset lattice in (a): in (a) all vertices been known by that name in the music theory
on the same level (sets of the same cardinality) literature. (In Part Three we will also use the
have the same in-​degree (edges connecting from word lattice in other contexts.)39
below) and the same out-​degree (edges connect-
ing from above), but this is not the case in (b), Exercise 3.4.7 Verify that the graph of Figure
where, for instance, only two abstract two-​note 3.4.3b is a lattice. Identify glb(C, A♭), lub(C, A♭),
subsets of set class 036 appear, compared with glb(C, g♯), and lub(C, g♯) in this lattice. If an
three abstract two-​note subsets for every other arrow were added from g♯ to C, explain why the
three-​note set class. graph
   would then fail to be a lattice.
The word “lattice” appears in quotation
marks in reference to Figure 3.4.6b for a reason.
Exercise 3.4.8
The word has a technical mathematical defini-
(a) Construct a subset lattice for the pitch-​
tion, which 3.4.6a satisfies but 3.4.6b does not.
class set {0, 1, 4, 7}, and an abstract subset
Specifically, a lattice is a digraph whose prece-
“lattice” for the set class 0147.
dence relation is a partial order and in which
(b) Construct a subset lattice for the pitch-​
every pair of vertices has both a latest common
class set {0, 3, 6, 9}, and an abstract subset
predecessor (more usually called a greatest lower
“lattice” for the set class 0369.
bound, abbreviated glb) and an earliest common
(c) Compare your graphs in (a) and (b) with
successor (least upper bound, or lub).38 In the sub-
each other and with the two graphs in
set lattice of 3.4.6a, the greatest lower bound of
Figure 3.4.6. Are any of the subset lattices
any two subsets is their intersection, the larg-
isomorphic? What about the abstract subset
est common subset of both of them; similarly,
“lattices”?
the least upper bound of any two subsets is
(d) One of the abstract subset “lattices” from
their union, or smallest common superset. For
  part (a) or (b) is actually a lattice. Which one?
example,
Exercise 3.4.9
glb({0, 3}, {1, 3, 6}) =​{0, 3} ∩ {1, 3, 6} =​{3}
(a) Explain why the subset relation ⊆, defined
on the subsets of any set, is always reflexive,
and antisymmetric, and transitive.
(b) Explain why the abstract subset relation on
lub({0, 3}, {1, 3, 6}) =​{0, 3} ∪ {1, 3, 6} set classes is reflexive, antisymmetric, and
=​{0, 1, 3, 6}. transitive.
(c) Any lattice whose partial order is the usual
The subsets of any set always form a lattice in subset relation ⊆ may be called an inclusion
this way. The abstract subset relation, on the lattice. Give an example of an inclusion
other hand, is not so well-​ behaved in mat- lattice that does not consist of all subsets of
ters of greatest lower and least upper bounds.   some set, but only of some of them.40
In 3.4.6b, some pairs of set classes have well-​
defined greatest lower bounds: for example,
Lattices have also been used to depict metric
glb(02, 05) is the one-​note set class 0, while
relationships. Figure 3.4.10 is a version of what
lub(02, 05) is 025, the only three-​ note set
Richard Cohn has called a ski-​hill graph.41 The
class in the graph of which 02 and 05 are both
vertices are labeled with duration values rang-
abstract subsets. But set classes 02 and 03 have
ing from small subdivisions of a beat to larger
no least upper bound. Two different three-​note
values that typically span several measures.
set classes, 013 and 025, are common upper

triplet, while 3♪. indicates a duration equal

bounds (abstract supersets) of 02 and 03, and A notation such as indicates an eighth-​note
neither of these is a subset of the other, so nei-
ther is a least upper bound. In effect, it is not to 3 dotted eighth notes, or 9 sixteenth notes
generally possible to define intersections and (a duration that otherwise cannot be written
unions of set classes as it is for actual sets. without ties). The edges indicate subdivisions:

100 • E x p l o r i n g M u s i c a l S p a ces
 (such as the relation between 
duple subdivisions in the case of edges in the Following Cohn, we shall call this infinite lattice

and  or between ♪. and .), and triple subdivi-

orientation the ski-​hill lattice.42

sions for the edges (such as between  and . Exercise 3.4.11

and ♪).
(a) Suppose a quarter note is assigned value 1,
or between
so that the value of an eighth note is 1 2 ,
If we imagine an upward-​ pointing arrow
a dotted half note is 3, and so on. Redraw
on each edge, then each arrow points from the
Figure 3.4.10 using numbers rather than
smaller unit to the larger—​from the subdivision
note values.
to its multiple. The precedence relation in Figure
(b) All the numbers appearing in your
tion. The duration ♪ is a subdivision of . (there
3.4.10 is, in other words, the subdivision rela-
redrawn figure in (a) are positive rational
numbers. Not all positive rational
♪ precedes . in this graph because there is a
are 12 eighth notes in a dotted whole note), and
numbers appear here, however, even in
directed path from ♪ to . in the graph. (In fact the theoretically infinite ski-​hill lattice.
there are three such paths; more on that shortly.) Of the six numbers 1 81 , 4 5 , 32 27 , 22 7 , 14, and
108, only three will appear in the lattice.
horizontal row (such as  and .), because even
No edges connect different durations in the same
Which ones are these? By looking at a
though one is a shorter duration than the other, rational number, how can you determine
neither is a subdivision of the other. quickly whether or not it belongs in this
This precedence relation is a partial order. lattice?43
(c) Determine an arithmetical formulation of
pairs of vertices such as 3. and 2. on the top
Technically, Figure 3.4.10 is not a lattice because
the precedence relation in this lattice. That
row have no common successor (multiple) in the is, if x and y are two rational numbers in
graph, and also because pairs of vertices such this lattice, how can you easily determine,
as  and
from the numbers only, whether or not “x
on the bottom row have no com- precedes y” is a true statement?
mon predecessor (subdivision). One can easily (d) Arithmetically, if x and y are two rational
imagine extending the graph to include these numbers in the lattice, what is glb(x, y)?
multiple of 3. and 2. is 6., while a common
larger and smaller durations, however: a common   What is lub(x, y)?44

subdivision of  and is . In principle the Figure 3.4.12 presents six paths within the
graph may be regarded as an infinite lattice, ski-​hill lattice. Reading the edges as upward-​
extended in all directions by multiplying and pointing arrows as discussed above, we may
dividing note values by 2 and 3 as appropriate. regard the paths in (a)–​(e) as directed paths. Each

FIGURE 3.4.10 A portion of a ski-​hill lattice of metric durations (after Cohn)

Graphs • 101
of these directed paths corresponds to what Brahms experiments extensively with alter-
Cohn calls a metric state: a consonant hierar- native metric states, clearly presenting states
chy of pulses, each of which is a subdivision of (b), (c), (d), and (e) at various points in the
all longer durations and an integer multiple of movement—​ (b) and (c) many times, (d) and
all smaller durations appearing in the path. The (e) less often. State (b) regroups the six eighth
graph in (f), in contrast, is not a directed path notes within a measure in threes, suggesting a
 meter in opposition to the prevailing . State
as neither . nor  is a subdivision of the other,
and does not represent a metric state, inasmuch
(c) is a hemiola pattern common in many triple-​
and neither precedes the other in the precedence meter compositions; the grouping of the quarter
relation of the lattice. notes in twos could be notated in , or in three-​
A metric state may be used to depict the met- bar hypermeasures in . Alternative states may
ric organization of a musical passage, and the be projected not only by the actual note dura-
ski-​hill lattice facilitates comparisons between tions in the score, but also by phenomena such
metric states and the study of metric conflicts.45 as harmonic rhythm, accents, articulation, pat-
In fact, the metric states in Figure 3.4.12a–​e are tern repetition, and phrasing.
all realized in the first movement of Brahms’s Occasionally, two or more states are present
Symphony No. 2. State (a) is the normative simultaneously, in different layers within the
metric state for the  time signature, and the orchestral texture. Figure 3.4.13a shows a climac-

measure-​long . pulse is triply subdivided into ,

prevailing state in most of the movement. The tic passage from the movement’s development

which in turn is duply subdivided into ; moving imposed. The  pulse is common to these three
section in which states from 3.4.12a–​c are super-

upward in the hierarchy, the value . indicates metric states; the instruments projecting states
the common hypermetric grouping into two-​ (a) and (c) are not actually playing eighth notes,
measure units. At times four-​bar hypermeasures but their notes align with groupings of eighth

ric . pulse, equal in duration to twelve eighth

are projected, indicated parenthetically in the notes in the concurrent state (b). The hypermet-
topmost segment of (a).

FIGURE 3.4.12 Five metric states in the ski-​hill lattice, and one impostor

102 • E x p l o r i n g M u s i c a l S p a ces
notes, is also common to all three states. The anchor the melody in the prevailing state (a).
The  time signature applies accurately to both,
ric state to another (each state projects either  or
intervening levels, however, vary from one met-

., and either  or .). In the lattice, states (a)–​(c)

illustrating the fact that a metric state often con-
veys more detailed information about more lev-

from  to ., reflecting the three different ways

correspond to the three possible directed paths els of a metric hierarchy than an ordinary time
signature. State (e), illustrated in Figure 3.4.13c,
that the prime factors of the number 12 may be occurs only rarely in this movement. Because

lacks the . pulse common to all the others; only

arranged: (a) represents the factorization 2 × 3 × the quarter notes are grouped in fours, this state

the higher 2. grouping can serve as a common

2 reading from bottom to top, while (b) is 3 × 2 × 2
and (c) is 2 × 2 × 3. It is this multiplicity of paths,
read instead from top to bottom, that motivates multiple.

poised at ., selecting any of the three available

Cohn’s ski-​hill metaphor: we may imagine a skier Comparison of states (c) and (e) in Figure
3.4.12 shows that although different durations
routes down to the  vertex. appear in these paths, the paths have the same
State (d) is the only state in Figure 3.4.12 to shape. Musically this means the durations in
incorporate triple subdivisions at two different these two states are in the same proportions,

vision of the . measure. (Another metric state

levels: its triplet eighths are a nine-​part subdi- related by exact augmentation or diminution.
Any other durations in the same 1:2:4:12 pro-
with this property combines the pulses , ., portions from the ski-​hill lattice (for instance,
, , ♩ , and .) would form another path with

 time signature.) Figure 3.4.13b shows accom-
and 3., and is most often represented by the
this shape. In principle, music notated using
panimental figuration in state (d) in the flute; the note values from one of these paths could
the eighth notes in the strings, meanwhile, always be renotated using the note values from

FIGURE 3.4.13 Metric states from Figure 3.4.12 in Brahms, Symphony No. 2 in D Major, Op. 73, I: (a)
mm. 246–​49; (b) mm. 156–​58; (c) mm. 28–​31

Graphs • 103
 Exercise 3.4.16 Study the scores of the
following works by Brahms, and identify at least
three or four clearly projected metric states
in each of them. You may need to consider
hypermetric structure.

(a) Symphony No. 3 in F Major, Op. 90, I

(b) String Quartet in C Minor, Op. 51, No. 1, I
(c) String Quartet in A Minor, Op. 51, No. 2, IV
(d) String Quartet in B♭ Major, Op. 67, I47
(e) Capriccio in C Major, Op. 76, No. 848
FIGURE 3.4.14 The state class common to met-
(f) Intermezzo in E Major, Op. 116, No. 4 (do
ric states (c) and (e) of Figure 3.4.12
not overlook mm. 30–​31)
(g) Piano
   Concerto No. 2 in B♭ Major, Op. 83, I
the other. The shape common to the three paths
is presented abstractly in Figure 3.4.14. Note Three more state classes are shown in Figure
values have been removed here; in their absence, 3.4.17. All of these are found in tonal music, but
notations ×2 and ×3 have been added to clarify somewhat rarely. State class (a) includes triple
the metric relationships. We call this figure a subdivisions at three consecutive levels; (b) and
state class. A state class may be realized by any (c) include triple subdivisions at two non-​con-
of the various metric states sharing its shape.46 secutive levels.

Exercise 3.4.15 Exercise 3.4.18 Each of the thirteen excerpts

(a) Verify that the relation “having the same listed below illustrates one of the unusual state
shape” is an equivalence relation on metric classes in Figure 3.4.17. Consulting scores
states, and that the equivalence classes are of these works, determine the state class
the state classes. applicable in each case, and supply the note
(b) States (c) and (e) from Figure 3.4.12 values to complete the metric state. In some
are paths of length 3, each representing cases you will need to consider hypermetric
durations at four different levels in the structure. Taking the actual note values into
metric hierarchy. If we consider paths of account, you should find twelve different metric
length 2 (representing durations at three states representing the three state classes
levels), common shapes become easier to from Figure 3.4.17. Which two examples use
find. Identify three different state classes precisely the same metric state?
of length 2, each of which appears as a
subgraph of at least three different metric (a) Bach, Goldberg Variations, Variation 26 (the
states from Figure 3.4.12. voice notated in 18 16)
(c) As directed graphs (assuming arrows (b) Beethoven, Sonata in C Minor, Op. 111, II,
pointing upward), the paths in 3.4.12c mm. 33–​48 (the L’istesso tempo variation
and 3.4.12e are isomorphic. But the in “6 16 ”—​but is Beethoven’s time signature
shape equivalence relation on metric correct?)
states, from part (a) of this exercise, is (c) Same piece as (b), the following variation
not generally the same thing as digraph (mm. 49–​64, L’istesso tempo in “12 32”—​but is
isomorphism. Give examples to illustrate Beethoven’s time signature correct?)
the difference. One of the two statements (d) Same piece as (b) and (c), the following
“digraph isomorphism implies shape variation (mm. 65ff)
equivalence” and “shape equivalence (e) Beethoven, Symphony No. 9, II,
implies digraph isomorphism” is true, mm. 177–​9449
and the other is false. Which is which? (In (f) Brahms, Quintet in G Major, Op. 111, I,
Section 3.5 we will describe a property mm. 57–​68
stronger than digraph isomorphism that (g) Brahms, Vier ernste Gesänge, Op. 121,
adequately represents shape equivalence. No. 1 (“Denn es gehet dem Menschen”),
  See Exercise 3.5.12.) mm. 90–​95

104 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 3.4.17 Three rare state classes

FIGURE 3.4.20 Bach, Chorale No. 356, “Jesu, meine Freude,” mm. 1–​2: (a) score; (b) directed graph of
bass line; (c) Schenkerian graph

(h) Fauré, Piano Quartet in C Minor, Op. 15, II, pulse), while the other string parts play the

( ), 4, 8, and 16. Construct a ski-​hill graph

mm. 1–​18 (and elsewhere) same rhythm in augmentation by factors of 2
(i) Mendelssohn, Song Without Words in E
Major, Op. 38, No. 3, prevailing pattern showing all the pulses projected in this texture.
(j) Same piece as (i), mm. 38–​39 and 67–​68 The tubular bell, meanwhile, sounds a regular
(k) Rachmaninoff, Rhapsody on a Theme of pattern of attacks separated by 12, 12, and 24
Paganini, Op. 43, Variation XIV, mm. 24–​25 quarter notes. How does the bell pattern fit into
(the last two measures before rehearsal 38), the
  lattice you constructed for the string parts?
in piano, flutes, oboes, and brasses
(l) Same piece as (k), Variation XIV, mm. 28–​33 As a final example of a directed graph in this
(m) Tchaikovsky, Symphony No. 4, I, mm. 116–​33 section, consider Figure 3.4.20b, in which the
and 295–​312 (disregard subdivisions smaller vertices are the notes in the bass line of the Bach
  than eighth notes) Chorale phrase given in (a). The arrows resemble
slurs, directed to indicate structural priority. The
Exercise 3.4.19 In Arvo Pärt’s Cantus in various root-​ position and first-​inversion ton-

repeatedly play the rhythm   (outlining a .

Memory of Benjamin Britten, the first violins ics are analyzed as prolongations of (and hence
subordinate to) the initial tonic, so arrows point

Graphs • 105
backward from these notes to the initial D, while occasion a well-​constructed and musically sensi-
other nearby notes are shown as subordinate to tive directed graph may imply relationships that
the notes to which they resolve. The entire graph violate Schenkerian principles; Exercise 3.5.2 in
takes on a suggestive quasi-​Schenkerian appear- the next section will offer an example.
ance; a more traditional Schenkerian sketch is
shown for comparison in (c).50 Exercise 3.4.21 Is the precedence relation in
Several aspects of this comparison merit com- the graph of Figure 3.4.20b a partial order?
ment. First, while the digraph and the Schenkerian Is the graph a lattice? Why might the name
graph are similar in appearance and in content, semilattice be appropriate for such a graph?
the information they convey is not identical. The Discuss the implications of this semilattice
well-​developed apparatus of Schenkerian theory property for the graph’s depiction of musical
and notation includes various elements that the structure.
  
digraph cannot directly accommodate. In par-
ticular, the passing motion D–​E–​F, neighboring 3.5 TRANSFORMATION
motion D–​C♯–​D, and Bassbrechung D–​A–​D are all GRAPHS AND NETWORKS
in effect ternary rather than binary relationships,
joining three elements by a single notation in a Figure 3.5.1a shows a series of notes proceed-
way that is impossible in a digraph; just to con- ing alternately by descending major thirds and
nect all the notes, Figure 3.4.20b requires more ascending perfect fourths. Such a series could be
arrows than 3.4.20c has slurs. Such distinctions interpreted as the pattern of chord roots in the
as beams vs. slurs and open vs. filled noteheads chromatic ascending 5–​6 sequence encountered
are Schenkerian conventions for clarifying struc- previously in Figure 2.3.4b. We are interested
tural priority; these conventions are somewhat in these notes as pitch classes independent of
redundant, however, as the hierarchical relation- register. In the directed graph of Figure 3.5.1b,
ships can usually be deduced without them, and the vertices are labeled with pitch classes cor-
in any case some similar devices could be intro- responding to the notes of the series, and the
duced in the digraph notation. arrows are labeled with transposition operators
Perhaps more importantly, while the digraph on pitch-​class space. The first diagonal arrow in
enables the representation of these structural the diagram communicates the simple fact that
relationships, it provides no guidance in deter- the transposition operator T8, applied to the
mining them. Bach’s bass notes may be joined pitch class C, produces the pitch class A♭; the first
by arrows in many different ways, most of them arrow in the top row shows similarly that T1(C) =​
musically nonsensical. Constructing a mean- C♯. The arrows in the graph form a repeating pat-
ingful representation such as 3.4.20b requires tern, a graphic representation of the sequential
external knowledge, an understanding of con- two-​note pattern. The notes of the original series
cepts such as stability, prolongation, and resolu- proceed across the graph from left to right, fol-
tion—​in short, something like the criteria used lowing the zigzag path traced by alternating T8
in Schenkerian analysis. Accordingly, while the and T5 arrows. The reasons for including the T1
digraph of 3.4.20b is hardly an adequate sub- arrows will be discussed below.52
stitute for a Schenkerian analysis, it perhaps Figure 3.5.1b is an example of a transforma-
reveals something about the structural relation- tion network, one of many important concepts
ships implicit in such an analysis; the analysis, whose analytical effectiveness was demon-
while not Schenkerian in any explicit sense, is strated by David Lewin. This section offers an
nevertheless at least compatible with the basic introduction to transformation networks and
tenets of Schenkerian theory. In a similar vein, the related transformation graphs; Chapter 9 will
transformational analyses, of which we will explore some of their more technical aspects. A
see many in later chapters, can be compatible transformation network is a directed graph with
with prolongational structure, and Schenkerian a particular kind of labeling: the vertices are
concepts such as expansion, progression, and labeled with musical objects of some sort, and
unfolding may be considered transformational the arrows are labeled with functions relating
in spirit; transformation and prolongation are those objects. A transformation network may be
not inherently antithetical. Many scholars thought of as a graphic extension of the func-
f
have commented on questions like these, and tion notation x →  y introduced in Chapter 2,
many hybrid transformational/​ prolongational combining many objects and functions in a
analyses have appeared.51 At the same time, on single diagram—​ or, conversely, that function

106 • E x p l o r i n g M u s i c a l S p a ces
 relationships between two-​note groups and out-
line a chromatic ascent that is probably a clearly
audible feature of any musical passage in which
this series occurs.

Exercise 3.5.2 Add arrows to Figure 3.5.1a

corresponding to the arrows of 3.5.1b. If the
arrows are regarded as slurs, would all of these
slurs be permissible in a proper Schenkerian
graph? Does this observation make you want
FIGURE 3.5.1 (a) A pitch-​class series; (b) a T-​net
to remove any arrows? Which ones, if any,
representation
would you omit? Is any important information
sacrificed
   in doing so?54
notation may itself be regarded as a particu-
larly simple transformation network. Because Although arrows in a transformation network
all of the transformations in Figure 3.5.1b are may be included or omitted at the discretion of
transposition operators, this figure is a network the analyst, one firm rule must be observed.
of an especially simple but useful type, called a We call this rule the label consistency requirement;
f
transposition network, or more briefly a T-​net. We it demands that if an arrow x →  y appears in a
will soon see many examples of transformation network, then it must be the case that f(x) =​ y.
networks involving other kinds of objects and Simply put, a transformation network must tell
mappings.53 the truth. The arrow from C to C♯ in Figure 3.5.1b
Some of the arrows in Figure 3.5.1b—​those is labeled T1, and indeed T1(C) =​C♯ is a true state-
labeled T8 and T5—​relate events that are con- ment. If this arrow were labeled, say, T9 instead,
secutive on the musical surface of (a). The T1 the network would be asserting that T9(C) =​C♯;
arrows, in contrast, bypass intervening notes, such a falsehood violates label consistency and is
relating events that are not temporally adja- disallowed in a proper transformation network.
cent. There are no general rules of construc- If we for some reason decide to add an arrow
tion for transformation networks to specify from B♭ to C, the only transposition operator
which objects should or should not be joined by that can permissibly be used to label that arrow
arrows. We could, for instance, draw a new arrow is T2. (Other kinds of functions could map B♭ to
pointing from B♭ back to C in 3.5.1b, labeled T2, C, such as inversion about B, or the function x →
without violating any rules of network construc- 7x +​2 (mod 12) on pitch-​class numbers. These
tion—​even though B♭ follows C by some distance functions may appear in transformation net-
in the music. There seems no good reason to add works—​but such a network would no longer be
this arrow, however, unless we are prepared to a T-​net.)
assert some structurally or perceptually signifi- Figure 3.5.3 shows several variants on the
cant relationship between these two notes or network from 3.5.1b, introducing some refine-
chords. Perhaps more sensibly, if the musical ments to the notation that are useful on occa-
sequence ends at the note E, we might choose to sion (though these conventions are not observed
include an arrow labeled T4 from C to E, showing consistently in the transformational literature).
the major-​third relationship between the start- Figure 3.5.3a distinguishes between what may
ing and ending points. The choice of arrows to be called direct and indirect transformations.
be included in a transformation network reflects Direct transformations are those relating imme-
a series of conscious decisions on the part of the diately adjacent musical events; these are the
analyst, and as such becomes an important part T8 and T5 arrows, unchanged from the previous
of the analytical process. The arrows in 3.5.1b network. Indirect transformations, here the T1
are oriented to indicate chronological order, transpositions, are shown by dashed arrows,
always pointing from earlier events to later ones, signaling connections between events not liter-
and the vertices in the graph have been arranged ally adjacent in the music. Directness and indi-
from left to right in this order also—​but these rectness are not properties of transformations
are deliberate choices, not necessarily appropri- as functions, but rather characteristics of the
ate in every network. The T1 arrows are included way transformations model events in a musical
because, even though consecutive notes are not context. The distinction is not necessarily always
related by T1, these transformations clarify the of analytical interest, nor is it always clearly

Graphs • 107
defined: we will see many transformation net- to larger groupings of objects as an efficient
works that do not reflect temporal organization way of showing that one extended passage is a
at all. (Also, we reserve the right to use dashed transposition (or some other transformation)
arrows for other purposes from time to time.) of another, in a less cluttered way than would
Nevertheless, whenever the difference is mean- be required to show all the arrows between indi-
ingful, the graphic convention of 3.5.3a may be vidual objects. Drawing the bubbled groupings
useful: one can immediately perceive the order may require decisions about segmentation, as a
of events in the music by tracing the path of comparison of the alternative versions in (b) and
the solid arrows, while the inclusion of indirect (c) illustrates, and an incomplete bubble may
transformations preserves the ability to make sometimes appear at the beginning or end of a
analytical observations involving widely sepa- pattern. The T5 arrows have been omitted from
rated events. (b) and the T8 arrows from (c), but these could be
Parts (b) and (c) of Figure 3.5.3 group the added (at the risk of reintroducing some visual
two-​note segments within bubbles, and show complexity to the figures), not as arrows acting
the T1 transpositions acting not on individual on entire bubbles but as arrows pointing from
notes but on the bubbles—​that is, on two notes one note in one bubble to a different note in the
at a time. This notation captures our intuition next.55
that the chromatic ascent operates at the level of Networks incorporating several transforma-
two-​note groups. Each interbubble T1 arrow may tions typically reveal something about how the
be interpreted as a shorthand for two separate transformations interact. We will study the alge-
→ C ♯ and A ♭ 
T1 T1
T1 arrows (C  → a in the case braic properties of various transformations at
of the first T1 arrow in (b)), each of which maps a length in later chapters, but one straightforward
pitch class in one bubble to the correspondingly property of transpositions on pitch-​class space is
positioned pitch class in the other. Other trans- relevant here, namely the equation TmTn =​ Tm +​ n,
formations now appear as arrows within bubbles: where the sum m +​ n in the subscript is calcu-
T8 in (b) and T5 in (c). The pattern of these intra- lated mod 12. (As always when combining func-
bubble arrows is the same in every bubble in a tions, TmTn denotes composition in left-​to-​right
network. The bubble technique may be applied order, Tm ● Tn—​though in this case the order
does not matter because the transpositions com-
mute.) For example, T8T5 =​ T1, because 8 +​5 =​
1 (mod 12). This equation is implicit in the first
triangle in Figure 3.5.1b, which incorporates two
different directed paths from C to C♯: a single
arrow labeled T1, and a path of length 2 labeled
with the transpositions T8 and T5. Because T8T5
and T1 are the same function, these two paths,
starting at the same point, are guaranteed to
arrive at the same pitch class. The T1 arrow is in
some sense redundant, as the transposition it
represents may be deduced from the other two
arrows—​but this redundancy is not necessar-
ily undesirable if one wishes to call attention to
multiple relationships in a single graph.
The bubble notation in Figure 3.5.3b–​ c
also reflects—​ indeed, depends on—​ certain
transformational properties. As noted above,
→ C♯
T1
the first T1 in (b) implies both C 
and A ♭ 
T1→ a. It follows, then, that there are
two different transformational paths from
→ C♯ 
T1 T8
C to A: a path C  → a and a second
path C  T8
→ A ♭ T1→ a with the same two tran­

spositions but in the opposite order. Of course,

T1T8 and T8T1 are the same function: both are
FIGURE 3.5.3 Alternate notations for the T-​net equal to T9. The fact that T1 and T8 commute,
of Figure 3.5.1b that is, is implicit in the bubble notation. For the

108 • E x p l o r i n g M u s i c a l S p a ces
bubble technique to work as it does here, each
interbubble transformation must commute with
all intrabubble transformations in the bubbles on
which it acts. Pitch-​class transpositions always
commute with each other, so bubble notation
may be used freely in T-​nets. In other situations,
however, this notation may be problematic, for
we shall see many examples of non-​commut-
ing transformations (even transpositions fail to
commute with some other transformations).56

Exercise 3.5.4 How is the commutativity of

T5 and T8 apparent in the original T-​net of
Figure
   3.5.1b?

Figure 3.5.3d shows one more variation

on the T-​net of 3.5.1b. The notes are arranged
as before, but the arrows have been reversed.
Correspondingly, each transformation has been
replaced by its inverse: the original T8, T5, and
T1 are now T4, T7, and T11 respectively. This sub- FIGURE 3.5.5 Two transformation networks
stitution is consistent with the logic of inverse and their underlying transformation graph
f
functions: if x → y and g is the inverse g
func-
tion of f, then it is also true that y  → x . It
would be possible to rearrange the vertices so are unlabeled. Because of the absence of vertex
that the arrows once again point from left to labels, Figure 3.5.5c stands in relation to (a)
right, but the arrangement shown here, which and (b) as the abstract graph of 3.1.2 stands in
preserves the original position of each pitch relation to the hexatonic triad graph of 3.1.1,
class, shows the derivation of the new network or as the state class of 3.4.14 stands in relation
from the old more clearly. The network in 3.5.3d to metric states (c) and (e) of 3.4.12. A digraph
may be called the reverse of the T-​net of 3.5.1b. such as 3.5.5c, whose arrows bear transforma-
If the original note series was given in a tempo- tional labels but whose vertices are unlabeled, is
ral ordering, then the arrows in 3.5.3d trace that a transformation graph.
ordering in retrograde (we refrain from using The careful distinction between transforma-
the word retrograde for reversed networks gen- tion network and transformation graph was drawn
erally, because not every network represents a by Lewin. This distinction has been retained by
temporal series). Because some functions have most recent practitioners of transformation
no inverses, some transformation networks theory, and we will observe it throughout this
cannot be reversed. If every transformation in book (although the word graph by itself may
a network has an inverse, we call the network refer to a transformation network, which after
reversible. Transposition operators always have all is still a kind of graph). A transformation
inverses, so every T-​net is reversible. graph, like Figure 3.5.5c, is the more abstract of
Figure 3.5.5a reproduces the T-​net from the two constructions: it shows relationships
3.5.1b, and 3.5.5b shows a very similar T-​net. among objects without specifying precisely
Both T-​nets have the same configuration of ver- what those objects are. A transformation graph
tices and arrows, and even the same transfor- may be likened to a network that has forgotten
mational labels on the arrows. The pitch classes its vertex labels, remembering only the arrow
appearing as vertex labels, however, are differ- labels. Abstract transformation graphs are use-
ent. Musically, the series in (b) is simply a trans- ful because the relationships are sometimes of
position of that in (a); one may imagine drawing greater or more general interest than the objects
large bubbles around both networks and con- on which they act, and because it is common (as
necting the two bubbles by an arrow labeled T7. in Figure 3.5.5a–​b) for two or more networks
The structure that the two T-​nets share to share the same graph structure. A trans-
is shown in Figure 3.5.5c. This figure is not a formation network, like (a) or (b), is more con-
transformation network, because its vertices crete, bringing the graph to life in a real musical

Graphs • 109
setting. We call (c) the underlying transformation descending-​fifths sequence; this sequence pat-
graph of the networks in (a) and (b), and say tern is also sometimes called a “descending-​seconds
that each of those networks realizes the graph. sequence” based on the relationship between
If the transformation labels are removed from a two-​chord segments, indicated here by the T10
transformation graph, the result is a still more arrows. Some structural likenesses between
abstract structure, an unlabeled directed graph; Figure 3.5.6 and the previous networks are obvi-
we call this the underlying digraph of both the ous, and it is evident that similar networks could
transformation graph and the transformation be constructed for other sequence patterns.
network, both of which realize the digraph. Nevertheless, as a transformation network,
Following the discussion of isomorphism in Figure 3.5.6 is not strongly isographic to 3.5.5a,
Section 3.2, the reader may suspect that the two because of the different transformational labels.
transformation networks of Figure 3.5.5a–​b are These two networks are not even isographic in the
isomorphic. Indeed they are—​but isomorphism more general group-​theoretic sense to be defined
of transformation graphs and networks is a com- in Chapter 9, which would require, loosely speak-
plex subject, as several different types of isomor- ing, that corresponding transformations play
phism and the related concept of isography may corresponding roles in the group structure.
be defined for these structures. Some of these Because the arrows labeled T5 in Figure 3.5.6 cor-
concepts require an understanding of group-​the- respond to arrows bearing two different labels in
oretic properties of transformations, so a general 3.5.5a, no bijective mapping between the trans-
discussion of this subject is deferred until Section formations involved in the two networks is possi-
9.3. The two networks of Figure 3.5.5a–​b, how- ble. The strongest statement that can be made in
ever, exhibit a particularly close relationship, comparing the graph structures of Figures 3.5.5a
one that is easy to describe. These two T-​nets and 3.5.6 is that the underlying digraphs of the
are strongly isographic, which means simply that two networks are isomorphic.
they share the same underlying transformation
graph, including the arrow labels. More explic- Exercise 3.5.7 Each series of notes below
itly, the underlying digraphs are isomorphic (as is the series of chord roots in a chromatic
defined previously for directed graphs), and cor- sequence. Construct a T-​net representing each
responding arrows in the two digraphs bear iden- sequence. Make multiple versions of each
tical transformational labels. (To say that the graph, experimenting with direct and indirect
underlying transformation graphs of strongly transformations and bubble notation. What
isographic networks must be “identical” is poten- equation of functional composition, analogous
tially misleading, because in keeping with the to T8T5 =​ T1 in Figure 3.5.1b, is applicable in
general notion of graph isomorphism, it is pos- each case? Are any of these networks strongly
sible that the visual arrangements of the vertices isographic to any from Figures 3.5.1–​3.5.6?58
in the two graphs may be very different—​as we
observed previously with the various versions of (a) Beethoven, String Quartet in B♭ Major,
the hexatonic triad graph.)57 Op. 18, No. 6, IV, mm. 21–​25:
Figure 3.5.6 shows another T-​net resem- E–​F♯–​B–​C♯–​F♯–​G♯–​C♯–​E♭–​A♭
bling those of Figure 3.5.5, but with some dif- (b) Liszt, Etudes d’exécution transcendante,
ferent transformations. The pitch-​ class series No. 8, “Wilde Jagd,” mm. 178–​84:
here descends by perfect fifths, as indicated by E–​A♭–​E♭–​G–​D–​F♯–​D♭–​F–​C
the T5 arrows. If these are chord roots in a har- (c) Dvořák, Symphony No. 7 in D Minor,
monic sequence, it is the familiar chromatic Op. 70, IV, mm. 233–​41:
F–​B–​B♭–​E–​E♭–​A–​A♭–​D–​C♯
(d) Rimsky-​Korsakov, Scheherazade, I, mm. 30–​39:
  C♯–​E–​B♭–​E♭–​G♭–​C–​F–​A♭–​D–​G–​B♭–​E–​A

Many of the graphs in the remainder of this

book will be transformation graphs and net-
works. Moreover, it is not unusual for a graph to
have the implicit structure of a transformation
graph or network even if no transformations
FIGURE 3.5.6 Another T-​net are explicitly indicated. To illustrate this point,

110 • E x p l o r i n g M u s i c a l S p a ces
Figure 3.5.8 revisits four graphs introduced ear- we may imagine the diagram extended to form an
lier in this chapter and recasts them in the form infinite lattice, generated by repeating the same
of transformation networks, demonstrating that pattern of transformations. Finally, Figure 3.5.8d
transformational structure was effectively pres- is the cubic graph of row forms from 3.2.3. The
ent in these graphs all along. Part (a) reproduces transformations T6, I, and R were already shown
the finite tonnetz from Figure 3.3.10b with the in 3.2.3, but they have now been incorporated
addition of transposition labels T3, T4, and T7 into the network as labels on the arrows.
on the triangulating edges; the tonnetz is now
a T-​net. The arrowheads, not present in previous Exercise 3.5.9 Show how discrete chromatic
tonnetz figures, are needed here in order to indi- pitch-​class space pc (Figure 1.2.3) and the circle
cate which way the transpositions operate. of fifths (1.3.1) may both be regarded as simple
Figure 3.5.8b is the hexatonic cycle from transformation networks. (One transposition
3.1.3c. Adjacent triads in this cycle are related operator, repeated many times, suffices for each
by the transformations P (parallel) and L (leit- network.)
  
tonwechsel) in alternation. Both P and L are
involutions (they are their own inverses), so the Exercise 3.5.10
transformational relationships are appropriately (a) The T6 arrows in Figure 3.5.8d are
indicated with bidirectional arrows; as an alter- bidirectional, but the T arrows in most
native, the arrowheads could be omitted entirely. other transformation networks we have
Figure 3.5.8c reworks a portion of the ski-​hill encountered were not (and could not have
lattice from 3.4.10. The note values labeling been). Why? What is the only Tn other than
the vertices have been replaced with numerical T6 that can be represented this way?
equivalents, taking the quarter note to have value (b) All four transformation networks in
1, and the transformations shown are multiplica- Figure 3.5.8 are reversible. Construct the
tion by 2 (labeled ×2) and by 3 (×3). As before, reverse of each network. In two cases,

FIGURE 3.5.8 Four transformation networks: (a) a tonnetz; (b) a hexatonic cycle; (c) a ski-​hill lattice;
(d) a network showing relationships among twelve-​tone rows

Graphs • 111
 the reverse is identical to the original been generated instead by octaves and fifths. In
network. What special property of these fact, the following exercise shows that the ski-​
  two networks makes this possible? hill graph, regarded as a pitch space in this way,
is essentially equivalent to a pitch space con-
Exercise 3.5.11 structed in a previous exercise.59
(a) The hexatonic cycle of Figures 3.1.3c and
3.5.8b was originally a subgraph of the Exercise 3.5.13 Redraw Figure 3.5.8c using
hexatonic triad graph of Figure 3.1.1. There pitch names—​more precisely, signed registral
is not an obvious way to interpret the entire letter names (as in Figure 1.1.11). If the vertex
hexatonic triad graph as a transformation labeled 1 in 3.5.8c is relabeled C4, then 2 will be
network, because it includes augmented C5 (an octave higher), and 1 3 will be F2 (a perfect
triads, and no standard transformations twelfth lower). Compare this diagram of signed
relate major or minor triads to augmented registral letter space with the Pythagorean pitch
triads. With some machinations, however, space
   you constructed in Exercise 2.3.25.
it is possible to recast this graph as a
transformation network. Define a function Example 3.5.14: Webern, “Herr Jesus mein”
Aug on the domain of major and minor Figure 3.5.8d, in its original incarnation as
triads as follows: applied to a major triad, Figure 3.2.3, was described as comprising
Aug raises the fifth of the chord by a the row forms in Webern’s Op. 23 songs. This
semitone, and applied to a minor triad, Aug abstract cube, however, tells us little about the
lowers the root by a semitone. Using Aug events of those songs. We conclude this chapter
as well as P and L, redraw Figure 3.1.1 as a by examining the row forms in the opening
transformation network. Every edge should measures of the third song from Op. 23, “Herr
be an arrow, either one-​or two-​headed, Jesus mein,” with the aim of suggesting some
with a transformational label. ways in which a transformation network may be
(b) Is the function Aug one-​to-​one? Does it laid out to capture the narrative structure of a
have an inverse function? Is the network musical composition.
you constructed in part (a) reversible? Figure 3.5.15a shows the first two row
How can you see the answers to all of these forms in the voice in “Herr Jesus mein” and
questions immediately by looking at the the first six in the piano, together occupying
  network? approximately mm. 1–​7 (about the first quarter
of the song). Rows are presented linearly
Exercise 3.5.12 In Exercise 3.4.15c you in both the voice and the piano. The piano
observed that isomorphism of directed states RI6 before the voice enters, then plays
graphs does not capture the relation of shape RI0 and P0 during the vocal statement of P0.
equivalence on metric states in the ski-​hill Two statements of P0 are therefore heard
lattice. Does strong isography of transformation concurrently; the last seven notes of this
networks capture that equivalence? That is, is it row form are stated at the same pitch levels
true that two metric states, as subnetworks of in both the voice and the piano, and in close
Figure 3.5.8c, have the same shape if and only if temporal proximity. The second vocal row,
they
   are strongly isographic? R6, is accompanied by two rows in the piano,
R0 and P6, and the beginning of another,
Occasionally a single graph may have two I6. The figure lays out these eight row forms
substantially different interpretations. Figure in score order, showing transformational
3.5.8c was presented as a network of durations, relationships between the two rows in the
a construction in the time domain—​ but the voice, between successive rows in the piano,
same graph may be interpreted instead as a pitch and between simultaneous rows in the two
space, by reading the vertex labels as frequencies parts. Recall the notational convention by
(based, for example, on middle C =​1) rather than which we distinguish boldface row form
durations. In effect the ski-​hill graph becomes a labels from italic transformational labels. The
kind of tonnetz, one whose generating intervals transformations appearing here are those in
are pure octaves (ratio 2) and perfect twelfths Figure 3.5.8d: T6, I, and R, and combinations of
(ratio 3). The perfect fifth (3 2 ) arises through those. In fact, 3.5.8d may be used to deduce the
combination of a twelfth and an octave, so the transformational relationship between any two
same frequencies may be considered to have of the eight row forms appearing there. These

112 • E x p l o r i n g M u s i c a l S p a ces
transformations all commute with each other,60 combinations of these. Antipodal corners of the
so the composite transformations labeled RT6 cube are related by the triple composition RIT6,
and RI in 3.5.15a could equally well have been but no such transformation occurs between
labeled T6R and IR. immediately successive or simultaneous rows in
Figure 3.5.15b maps the same passage in these opening measures. (Only once in the song
a different way. The layout here matches the does the voice negotiate this transformation,
cube of 3.5.8d. The chain of arrows depicting leaping cross-​cube to an antipodal corner. This
the piano’s row forms begins at RI6, traverses is at the very end: the last two vocal rows in the
an edge of the cube to the adjacent RI0, then song are P0 and RI6, matching the first row in the
crosses a face diagonally to arrive at P0, the voice and the first in the piano. The piano ends
vertex at the antipodal corner of the cube from the song with this same pair of antipodal rows.)
the RI6 starting point, and a location by now Several details of the two networks of Figure
already staked out by the voice. The voice and 3.5.15 are instructive, as are the relationships
piano then diverge again, moving away from P0 between the two. Both the event network of (a)
to two transpositionally related R forms. The and the spatial network of (b) have their own
last three row forms in the piano form a pattern strengths and shortcomings, and both strategies
that neatly reverses the first three, moving again were of great interest to David Lewin.61 The
from one corner of the cube (R0) to the antipodal vertices of (a) correspond to events (rows) in
corner (I6), this time traversing first a face the music, and the left-​to-​right chronology of
diagonal and then an edge. The concurrent row in this figure displays the narrative structure of
the voice, R6, is directly adjacent to the first two the piece; it is a blow-​by-​blow account, showing
of these three piano rows. The five arrows traced with some precision what happens and when,
by the piano form a symmetrical pattern, edge–​ even (with the aid of a background grid of
diagonal–​edge–​diagonal–​edge. Transformations measures, and boxes indicating the extent of
that trace edges of the cube are the simple row statements) the temporal duration of each
twelve-​tone operations T6, R, and I (each of row and brief overlaps between rows in the
which occurs exactly once in this passage), piano. The spatial network in (b) shows these
while the face diagonals correspond to pairwise chronological aspects less clearly, instead laying

FIGURE 3.5.15 Two transformation networks for Webern, “Herr Jesus mein,” Op. 23, No. 3, mm.
1–​7: (a) an event network; (b) a spatial network

Graphs • 113
out an abstract space of row forms—​the same using this convention with note names, although we
space as in Figures 3.2.3 and 3.5.8d. This space, will sometimes use it with pitch-​class numbers (so
which we may call the underlying space of Figure that 0+​and 0− may in some contexts stand for C-​major
3.5.15b’s network, functions as a sort of game and C-​minor triads).
board on which the music moves about; vertices 2. Hexatonic systems, explored in Cohn 1996,
represent not events but points in this abstract play important roles in neo-​Riemannian theory, to be
space. detailed in Chapter 8. In Chapter 10 we will use exten-
If Figure 3.5.15a shows what happens sions of the hexatonic triad graph to motivate our
in Webern’s song, 3.5.15b offers a more study of voice-​leading spaces.
comprehensive picture of what might have 3. Graph theory texts sometimes distinguish
happened, and of the choices that Webern between paths and walks: a walk is allowed to revisit
made from the alternatives available to him. the same vertex of the original graph more than once,
This second figure encourages us to attend to while all the vertices in a path must be distinct. We
structural relationships that might otherwise shall not observe this distinction here.
have escaped our notice; moreover, the relative 4. Measuring the length of a path by counting
positions of row forms in the cube perhaps edges rather than vertices is conceptually similar to
make the transformational relationships more measuring the size of an interval by counting steps
palpable, less abstract than the arrow labels in rather than notes, as described in Section 1.1.
(a). A glance at (b) shows that one row form, P0, 5. The Brahms example is discussed in Cohn
appears in both the voice and the piano in these 1996, 13–​15.
opening measures, and that another potentially 6. Spanning cycles are also known as Hamiltonian
available row form, I0, has yet to appear at cycles, after Sir William Rowan Hamilton (1805–​
all. Another asset of (b) is that the same 1865), an Irish physicist, astronomer, and mathema-
spatial organization may be used in analyzing tician who studied the surprisingly difficult general
other pieces, as long as all the row forms are problem of determining whether such a cycle exists in
related by T6, I, and R. These advantages could, a graph or not.
however, become drawbacks as the analysis is 7. Hint for Exercise 3.1.11b: Imagine the chord
extended. It is clear how (a) could be expanded progression successively traversing edges of Figure
to a larger network that encompasses the entire 3.1.10, starting at C major. It is often useful to imag-
piece, but the scheme of (b) quickly becomes ine that the vertices of a bipartite graph are colored
unworkable as the same row forms are visited using two different colors.
multiple times; a series of such diagrams, or an 8. Finding a simple criterion for determining
animation, would be required to show even this whether a graph is planar was a tantalizing unsolved
brief song in its entirety.62 problem in graph theory until Kazimierz Kuratowski
provided such a criterion in 1930. For Kuratowski’s
Exercise 3.5.16 Study the score of “Herr theorem and related results see Chapter 11 of Harary
Jesus mein.” Following the suggestions in the 1969 or Chapter 10 of Bondy and Murty 2008.
preceding paragraph, construct a single event 9. The term face is not defined in general for graphs
network, and a series of spatial networks, that are not embedded in a plane, although it is pos-
spanning
   the entire song.63 sible to define faces for some particular graphs in
higher dimensions, such as the hypercube of Exercise
The recent music theory literature includes 3.1.18 below.
many transformation networks and other graphs 10. Hint for Exercise 3.1.12b: For a connected graph,
displaying an enormous variety of organiza- c =​1. For the graph of Figure 3.1.3e, c =​4 and f =​1. To
tional precepts and graphic techniques. At this modify the formula v +​ f =​ e +​2 so that it works for
juncture the reader may enjoy browsing through connected graphs, one need only replace the number 2
the figures in the later chapters of this book to by a simple expression involving c.
sample the flavors of some of these graphs. 11. In mathematics, the term dual appears in a
variety of settings, in reference to symmetrical rela-
tionships obtaining between two structures of which
NOTES one is formed from the other by interchanging certain
1. In this book, when appended as a superscript to kinds of opposed elements (such as the vertices and
a note name, the symbol +​represents an augmented faces of a graph). In set theory, the two de Morgan
triad. Some authors use the notation C+​ for a C-​major laws (see Exercise 2.1.3) illustrate a kind of duality
triad, in opposition to the minor triad C−. We will avoid between the union and intersection operations. Other

114 • E x p l o r i n g M u s i c a l S p a ces
manifestations of mathematical duality appear in 15. Hint for Exercise 3.2.2: In constructing an iso-
logic, Boolean algebra, projective geometry, and group morphism from 3.2.1a to 3.2.1b, one can choose to
theory; a form of group-​theoretic duality will emerge map vertex v1 to any of the six vertices of 3.2.1b. That
in our study of interval spaces (see Theorem 7.5.24). choice does not completely determine an isomor-
Duality also plays important roles in modern phys- phism, however, so the number of possible isomor-
ics, especially the wave-​particle duality of quantum phisms is greater than 6. Generally, the number of
mechanics. The most famous duality in music theory possible isomorphisms is a measure of the symmetry
is the harmonic function theory of Hugo Riemann of a graph. When asymmetrical graphs are isomor-
(inspired by Moritz Hauptmann and others), in which phic, only a single isomorphism may be possible, but
the dominant function is dual to the subdominant, for symmetrical graphs there may be a large number.
and major and minor chords, scales, and keys are seen We shall see in Chapters 5–​6 that group theory plays a
as mirror images of each other. We shall touch upon role in graph isomorphisms, shedding particular light
some aspects of Riemann’s theories in later chapters; on questions involving symmetry.
for a thorough examination of its dualistic aspects see 16. The row numbering here is that of Bailey 1991,
Harrison 1994. Duality has also been linked to the 339, 372–​73 (where a detailed row analysis of the
ideas of philosophers such as Hegel, who contended three songs may be found). The symmetrical distribu-
that all processes result from the conflict of opposites. tion of Webern’s row choices ensures that the same
12. The other three Platonic polyhedra are the tet- eight row form designations would appear in the
rahedron (4 vertices, 6 edges, 4 triangular faces), the graph if any of the other seven row forms appearing
dodecahedron (20 vertices, 30 edges, 12 pentagonal here were called P0. Bailey’s row numbering is based
faces), and the icosahedron (12 vertices, 30 edges, 20 on a movable-​0 system; the first note of P0 is A♭, not C.
triangular faces). The dodecahedron and icosahedron In a fixed-​0 system, the row designated P0 here would
are dual to each other in the same way as the cube and be P8, and the graph would consist of row forms bear-
octahedron, while the tetrahedron is dual to itself. For ing the index numbers 8 and 2 rather than 0 and 6,
graphs of all five, see Bondy and Murty 2008, 21. but the relationships described in the graph would be
13. Hint for Exercise 3.1.17: The graphs in parts unaffected. For a general discussion of how row names
(a) and (b) are fairly simple; (c) is somewhat more are altered by such changes of “coordinate system,” see
complex. In (c), try arranging the four minor-​minor Section 9.5.
seventh chords in a square, positioning the dominant 17. Here R denotes the retrograde transformation,
seventh chords outside the square and the half-​dimin- applicable to twelve-​tone rows or other ordered sets;
ished seventh chords inside; ideally, the shape of the elsewhere we use the same symbol for the relative
graph should have fourfold rotational symmetry, with transformation on triads. Because the domains on
T3-​related chords positioned 90 degrees apart. All of which these two functions are defined do not inter-
the graphs in this exercise, and the hexatonic triad sect, there should be little danger of confusion.
graph itself, are subgraphs of graphs presented in 18. Both Figure 3.1.7 and Figure 3.2.3 are transfor-
Douthett and Steinbach 1998, some of which we will mation networks generated by groups of transforma-
encounter in later chapters. tions; the graphs are isomorphic because the groups are
14. Hints for Exercise 3.1.18: You can find the octa- isomorphic. See Exercises 6.4.8 and 6.4.9.
tonic seventh-​chord hypercube by looking ahead to 19. Hint for Exercise 3.2.6: There are many ways to
Figure 11.5.1b, but it is a good exercise to construct construct such a counterexample. The graphs in your
it yourself. As noted in Chapter 2, spaces of more example need not have any particular musical signifi-
than three dimensions are difficult to visualize and cance, but for a musical example, consider two different
impossible to construct physically. Nevertheless, it ways of partitioning chromatic pitch-​class space into
is not difficult to draw a graph of a four-​dimensional two cycles, not necessarily of the same size. To make the
hypercube on a two-​dimensional sheet of paper. One exercise a bit more challenging, see if you can construct
way to do this is to place graphs of two three-​dimen- an example in which both graphs are connected (so the
sional cubes similar to Figure 3.1.7 next to each other, number of components is 1). Harder still, construct an
and draw additional edges connecting corresponding example in which both graphs are connected and regular.
vertices of the two cubes. In part (b), you should find 20. Terminology for graphs with loops and multiple
that the chords represented by the hypercube include edges is not standardized. Harary 1969 reserves the
two additional chord types besides those of the origi- term graph for graphs without loops or multiple edges;
nal octatonic seventh-​chord graph. In (c), each face a multigraph may have multiple edges but no loops,
and facet may be obtained by fixing some voices while while a pseudograph may have either or both. In Bondy
allowing the others to vary by semitone. The facets are and Murty 2008, all of these are called graphs, and a
ordinary three-​dimensional cubes. graph with no loops or multiple edges is called simple.

Graphs • 115
None of these terms will be used in this book; most of 27. Hint for Exercise 3.3.11b: Of the six yes-​no
our graphs will have no loops or multiple edges, and questions in this exercise, exactly two should be
these will generally be noted when they appear. answered yes.
21. Straus’s graph, which does not include the 28. The non-​planarity of the finite tonnetz graph
loops, appears in Straus 2003, 337; Straus 2005b, 52; is alluded to in London 2002, 136, but a formal proof
and Straus 2016, 180. is not given.
22. Hint for Exercise 3.3.4: This graph is not very 29. Several different proofs of the non-​planarity of
interesting! Many published set-​class tables omit the K5 may be found in the graph theory literature. The
dyadic set classes, better known as interval classes 1–​ proof outlined in this exercise, using Euler’s formula,
6. As set classes, their prime forms are 01, 02, …, 06. is based on Harary 1969, 103–​04. For an entirely dif-
A form of this graph appears in Straus 2005b, 51. A ferent proof, see Bondy and Murty 2008, 245. K5 is
closely related space will be studied in Chapter 11; see one of two small “forbidden subgraphs” for planar
Figure 11.2.12c. graphs; the other is the complete bipartite graph K3,3,
23. Hints for Exercise 3.3.5: Seven multiset which may be defined as the graph on six vertices par-
classes must be added to the graph: six dyads with titioned into two sets u1, u2, u3 and v1, v2, v3, such that
doublings, and the multiset class 000, representing each u vertex is adjacent to each v vertex.
a three-​voice unison. The full graph should have 19 30. Hint for Exercise 3.3.15a: The formula you are
vertices and 41 edges, including five loops. The mul- seeking is valid in every fully triangulated planar
tiset classes can be seen along with the traditional graph. As an example, you may find it helpful to con-
set classes in Figure 11.4.12 of this book as well as sult the planar embedding you constructed in Exercise
in Figure 10 in Callender 2004, though they appear 3.3.7a. (Replace the double edges with single edges.)
in both cases as reference points in a continuous If you solved that exercise correctly, you should have
space rather than as vertices of a discrete graph. In a graph all of whose faces are triangles. Does the rela-
Section 10.4 we will develop a general theory of nor- tion 3f =​ e hold in that graph?
mal forms that justifies the use of labels such as 004 31. Hint for Exercise 3.3.15a: The vertices are
for multiset classes. easy to count; edges and faces are trickier. Each of
24. The common-​tone properties of the hexatonic the broken edges appears in Figure 3.3.10b in two
cycle depicted in Figure 3.3.6 are summarized in pieces, so if you count the pieces, you should count
Cohn 1996, 19. The hexatonic pole relation and other only one of each pair. As an alternative approach
aspects of hexatonic cycles will be explored in more to counting the edges, you may take advantage of
detail in Section 8.2. the fact that the tonnetz graph is regular, with six
25. Hints for Exercise 3.3.8: The common-​tone edges meeting at each vertex. If you multiply the
graph corresponding to Exercise 3.1.17a (the octa- number of vertices by 6, you will have counted every
tonic triad graph) is planar; draw a planar embed- edge twice, so the number of edges should be just
ding if you can. The graph corresponding to Exercise half of that number. See if you can find an analo-
3.1.17b (the enneatonic seventh-chord graph) is not gous method for counting the faces, based on the
planar. This graph will include some single, some fact that every face is bounded by three edges. Still
double, and some triple edges, because a pair of sev- another approach would be to count the musical
enth chords may share as many as three common objects represented by the vertices, edges, and faces
tones. Instead of drawing double and triple edges, of the graph; for instance, each edge corresponds to
you may choose to draw them all as single edges one instance of interval class 3, 4, or 5.
labeled with the number 1, 2, or 3 as appropriate. 32. For uses of the Euler characteristic in graph
Counting all the multiple edges, the total number theory and the general problem of embedding graphs
of edges in this graph should be 54. To calculate in surfaces, see Harary 1969, 116–​20 or Bondy and
the number of edges required for the common-​tone Murty 2008, 275–​81. For a more general discussion of
graph corresponding to Exercise 3.1.17c, figure the the Euler characteristic in topology, see Weeks 2020,
number of edges incident with one dominant sev- 157–​78.
enth chord and multiply by 4 (the number of domi- 33. Figure 3.4.1 may be compared with similar dia-
nant seventh chords in the graph); do the same for grams in Chapter 7 of Kostka, Payne, and Almén 2018;
a minor-​minor and a half-​diminished seventh chord; with Figure 8.1 in Gauldin 2004, 121; with the “table
add up these three numbers; and then divide by 2. of usual root progressions” in Piston 1987, 23; and
Why is the last step necessary? with the recent theories of Tymoczko (2011, 226–​30).
26. Some authors restrict the term graph to finite While pedagogically useful, such flowchart-​style maps
graphs only. For a survey article on infinite graphs see of functional harmony have been criticized on several
Thomassen 1983. grounds. For one thing, they are inevitably incomplete.

116 • E x p l o r i n g M u s i c a l S p a ces
To include even such common phenomena as iii and vii automatically becomes a subset of those pulses that
chords and plagal and deceptive motion would require subdivide it; for example, the . pulse is a subset of
so many additions to Figure 3.4.1 that the graph would both the . and  pulses. With this understanding the
become unduly complex, running a risk of suggest- ski-​hill lattice becomes another example of an inclu-
ing that “anything is possible.” Other objections run sion lattice similar to those discussed previously, but
deeper, pointing out, for instance, that in their focus on with the largest sets now appearing at the bottom of
chord-​to-​chord connections and root motion these dia- the graph rather than at the top.
grams disregard underlying prolongational and voice-​ 43. Hint for Exercise 3.4.11b: Start by factoring the
leading concerns. In presenting this figure as a first numerator and denominator of a rational number into
example of a directed graph, our interest is less with the products of prime numbers.
musical merits of the diagram than with its properties 44. The notions of divisibility, greatest common divi-
as a graph. For an alternative graphical model of func- sor, and least common multiple, normally considered
tional harmonic progressions see Figure 4.3.8. only for integers, can be extended to all rational num-
34. Lewin ([1987] 2007, 193) uses the term node/​ bers (but not to all real numbers).
arrow system rather than directed graph. 45. The metric conflicts whose analysis is enabled
35. Hint for Exercise 3.4.4b–​c: To specify an isomor- by metric states are grouping dissonances in the termi-
phism completely, you must indicate exactly how the nology of Harald Krebs (1999). Metric states do not
vertices of one graph are mapped to the vertices of the directly facilitate the analysis of Krebs’s displacement
other. That is, for every vertex v in one graph, which dissonances.
vertex in the other graph is f(v)? There may be more 46. Leong 2007 studies properties of metric states
than one possible isomorphism, and therefore more sharing the same shape, as well as states whose shapes
than one possible choice of f(v) for some vertices v, but are related by rotation or reflection such as those in
you need only define one isomorphism completely, Figure 3.4.12b–​e.
being sure to verify that your bijection preserves the 47. Hint for Exercise 3.4.16d: Because the . becomes
direction of all arrows. the  at the change to  (m. 58), the notation for the
36. The arrow relation in a directed graph need metric states could create misleading impressions about
not be transitive; the graphs of Figure 3.4.3 include the relationships among them. How might this problem
arrows e → C and C → C+​ but not e → C+​. The pre- be addressed? (See the movement’s coda for an idea.)
cedence relation may be called the transitive closure of 48. See Lewin 1981 for commentary on relation-
the arrow relation: it is the minimal transitive relation ships between metric and tonal structure in the
that includes all the arrows. C-​major Capriccio.
37. For a survey of the theory of order relations, 49. For a detailed discussion of many metrical
see Stoll 1979, 48–​55. aspects of the Scherzo of the Ninth Symphony see
38. There are alternative ways to define lattices Cohn 1992a. The passage in Exercise 3.4.18e is exam-
based purely on set theory or Boolean algebra with ined on pp. 190–​91.
no reference to graphs. See, for example, Stoll 1979, 50. The graph in Figure 3.4.20b resembles the “sec-
248–​54. The graph of a lattice is sometimes called a ondary notation” of Lerdahl and Jackendoff (1983,
Hasse diagram. 201–​03) more closely than it resembles a traditional
39. Abstract subset “lattices” are intimately con- Schenkerian graph. Lerdahl and Jackendoff’s slurs
nected with Allen Forte’s theory of K and Kh rela- bear no arrowheads, but the same notion of structural
tions. See Morris 1997 for discussion, examples, and priority is implicit in them (they are derived from
analytical applications. the branching pattern in a prolongational tree, which
40. Hint for Exercise 3.4.9c: Such a lattice can arise itself can be constructed as a directed graph).
as a sublattice of Figure 3.4.6a—​ for example, by 51. Sentiments consonant with those put forth here
regarding the elements 0 and 1 as an inseparable pair, are expressed in Lewin [1987] 2007, 217–​18, Cohn 1989,
and considering only subsets that contain both or nei- 57, and Hook 2007b, 167–​68; also see Rings 2011b, 35–​
ther of those elements. Be sure to verify that your lat- 40 and 144–​48 for perceptive commentary. Cook (1996,
tice satisfies the glb and lub properties. 144) points out that Lewin’s analysis of Debussy’s “Feux
41. Cohn 2001 introduces ski-​ hill graphs with d’artifice” (Lewin [1993] 2007, 149), or at least his ter-
applications to music by Brahms and Dvořák; Leong minology, displays an “increasingly Schenkerian” bent.
2007 and Murphy 2009 offer further applications. For Kopp 2002 presents many transformation networks in
some related graphs see London 2012, 25–​47. a quasi-​Schenkerian format resembling Figure 3.4.20b.
42. If a pulse is defined (as Cohn sometimes A special issue of the Journal of Schenkerian Studies
does) not as a single duration but as a set of time (vol. 2, 2007) is devoted entirely to the intersection of
points separated by equal durations, then each pulse Schenkerian and neo-​Riemannian methods.

Graphs • 117
 52. This example, in which the transformations are compare the two graphs. The sequence in (d) is based
defined on pitch classes rather than on chords, clearly on a three-​note pattern, requiring modifications to
has its limitations as an analysis of the 5–​6 harmonic the visual layout.
sequence, as it disregards chord quality, inversion, and 59. London (2002, 139) refers to the ski-​hill lattice
voice leading. Transformations defined on chords, as a zeitnetz by analogy with the tonnetz. He is quick
including transformational analysis of harmonic to point out, however, that many aspects of the pitch
sequences, will be studied more extensively in Chapter domain, such as the conformed property of the finite
4 and especially in Chapter 8. tonnetz, have no direct analogs in the time domain.
53. Lewin formalizes transformation networks 60. The commutativity of T6 and I will be estab-
(and transformation graphs, discussed below) in lished formally in Chapter 5 (see Theorem 5.5.6).
Chapter 9 of Lewin [1987] 2007, after presenting sev- Commutativity properties involving R will be noted in
eral examples informally in Chapters 7 and 8; most of Chapter 6 and explored in more detail in Chapter 9.
those in Chapter 7 are T-​nets. For an extensive appli- 61. Lewin’s most extensive commentary on the dif-
cation of T-​nets in analysis, see Sallmen 2016. ference between event networks and spatial networks
54. Hint for Exercise 3.5.2: The answer to the first appears in his analysis of Stockhausen’s Klavierstück
question in this exercise is no. See if you can make a III (Lewin [1993] 2007, 16–​67), especially in reference
convincing argument that the answer to the last ques- to his e­ xamples 2.4–​2.6. Borrowing terminology from
tion is yes. For a related example see Hook 2007b, 167. Bamberger 1986, Lewin refers to chronological and spa-
55. For a few examples in later chapters making tial network layouts as “figural” and “formal,” respec-
prominent use of bubble notation, the reader may tively. The terms event network and spatial network are
refer to Figures 8.3.20, 9.3.17, 9.5.12, and 13.3.24. from Roeder 2009. See also Rings 2011b, 140–​44; the
An early example of bubble notation appears in Lewin distinction between syntagmatics and paradigmatics
[1987] 2007, 205, Figure 9.8f–​g. Rather than regard- described in McCreless 1991 is also relevant.
ing the interbubble arrows as a shorthand for multiple 62. Apart from the graphic arrangement of the
arrows, Lewin instead describes the entire graph as a vertices, both of the diagrams in Figure 3.5.15 include
“network of networks,” in which the vertices of the some information that is not strictly part of the
main network represent smaller networks. This con- transformation network. This information includes
ception gains new importance in Lewin’s later study of the assignment of row forms to the voice or piano in
Klumpenhouwer networks (Lewin 1990, Lewin 1994), both graphs; the measure numbers and overlapping of
and we will revisit it in Section 9.4. simultaneous row forms in (a); and, in (b), the inclu-
56. In Chapter 9 we will consider generalized forms sion of several edges of the cube that do not corre-
of bubble notation in which this commutativity is not spond to transformations occurring in the music.
required. These observations are related to a network 63. Hint for Exercise 3.5.16: At least five or six
property to be called path consistency in Chapter 9; the spatial networks are recommended. The detailed row
equation T8T5 =​ T5T8 =​ T1 ensures that the networks analysis in Bailey 1991, 372–​73, may be useful.
in Figures 3.5.1 and 3.5.3 are path-​consistent. Path
consistency should not be confused with label consis-
tency, the truth-​telling property described previously.
SUGGESTED READING
While label consistency is required of all transforma- Among the wide selection of texts on graph theory,
tion networks, we (unlike Lewin) allow networks that Harary 1969 is regarded as a classic, while Bondy
are not path-​consistent. and Murty 2008 is now widely used. For the hexa-
57. Strong isography was first defined by Henry tonic systems that form the musical basis of the
Klumpenhouwer for the networks now known as hexatonic triad graph see Cohn 1996 and Chapter
Klumpenhouwer networks; see Chapter 9. 2 of Cohn 2012, as well as Section 8.2 of this book.
58. Hints for Exercise 3.5.7: Some pitch classes For considerations arising in analytical applications
occur more than once in some of these series, but of transformation networks, see Chapter 9 of this
each occurrence may be given its own vertex in the book, the four essays in Lewin [1993] 2007, Roeder
graph. For (c), in addition to constructing a T-​net as 2009, and Buchler 2016. Roeder and other authors
suggested in this exercise, you may find it instruc- in the same special issue of Music Theory Online
tive to graph the sequence in the tonnetz from Figure offer valuable commentary on strategies of network
1.4.18, generated by semitones and tritones, and to construction.

118 • E x p l o r i n g M u s i c a l S p a ces
 4
Spaces II
Chordal, Tonal, and Serial Spaces

THE SPACES studied in Chapter 1 are all some some of those encountered previously, may be
form of pitch or pitch-​class space, and their ele- construed as transformation networks, and the
ments are single notes. Higher-​ level musical clearest way to conceive of the structure of the
entities such as chords, key areas, and twelve-​ spaces is sometimes through an understanding
tone rows appear in these spaces only indirectly of transformational relationships. We will make
if at all, not as elements but as groupings of note here of the basic transformations under-
elements. Some of the diagrams in Chapter 3, lying the organization of each space, and some
however, incorporate chords or rows as their properties of these transformations, particu-
fundamental objects (vertex labels in a graph), larly those involving triads, will begin to emerge.
suggesting that it may be fruitful to construct The emphasis for now, however, remains on
musical spaces with such objects as their ele- the spaces themselves; more detailed study of
ments. This chapter is devoted to exploring the properties of the transformations will be
these possibilities. deferred to Part Two.
Many new considerations arise in conjunc-
tion with this more complex objective. The vari- 4.1 DOUBLE-​C IRCLE
ety in the way such spaces may be constructed is SPACES AND RELATED
vast, and this chapter only hints at what is pos-
sible; later chapters will present some further
CONSTRUCTIONS
strategies. The representations in this chapter, As a first attempt to construct a space of triads,
however, include some that have received signifi- we might start with a diagram such as Figure
cant attention in the theoretical and analytical 1.2.3 (chromatic pitch-​class space pc) or Figure
literature, as well as some that will find applica- 1.3.1 (fifth, the circle of fifths), and consider
tion in later chapters. Many of these spaces, like that each note name may stand not just for a

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0004
single note but for a chord or a key area—​that is, denotes inversion about the axis through C and
a triad whose root is the given note, or the key F♯ in circular pitch-​class space. In pitch-​ class
with that tonic triad. This strategy produces a numbers, I maps pitch class x to pitch class −x
logical arrangement of the twelve major triads or (mod 12). Applied to a C-​major triad, I yields
the twelve minor triads, but not both. A simple an F-​minor triad, as one can check by working
way to integrate both qualities of triads in one out the mapping of the three notes individu-
space—​a space we may call consonant triad space, ally: 0 → 0, 4 → 8, 7 → 5 (C → C, E → A♭, G →
written triad for short—​is to provide two con- F). As a triad ascends by semitone, its inversion
centric circles, one for major triads and the other about a fixed axis descends by semitone; thus I
for minor. Such double-​circle spaces take a variety applied to C major yields F minor, and I applied
of forms, several of which we introduce in this to D♭ major (a semitone higher) yields E minor (a
section. Some of these diagrams provide reveal- semitone lower). In this scheme, consequently,
ing glimpses into the transformational structure the two circles are oppositely oriented, with T1
of the spaces, in ways to be revealed more fully in corresponding to clockwise motion in the circle
Chapter 6, and we will see applications of other of major triads and counterclockwise motion in
similar pictures in Chapters 8 and 9.1 the circle of minor triads.
The double-​ circle strategy allows for mul- Every major triad is related to every minor
tiple possible arrangements of the 24 triads, triad by some inversion. The parallel transfor-
depending on whether pc or fifth is used as the mation P, appearing in Figure 4.1.1a, may be
starting point and on the alignment of the two regarded as a sort of inversion operator, but it
circles.2 Two arrangements based on the chro- does not correspond to inversion about a fixed
matic circle are shown in Figure 4.1.1. Perhaps axis as I does. Rather, P inverts any triad so that
the most obvious such scheme is 4.1.1a, in its root maps onto its fifth and vice versa. The
which both circles of triads appear in the same axis of inversion is determined by the context—​
orientation as in the usual arrangement of pc. by the position of the triad within pitch-​class
This arrangement is organized by the transfor- space. For this reason P is sometimes called a
mations T1 and P: the transposition operator T1 contextual (or contextually defined) inversion oper-
defines clockwise motion from any triad to the ator; triadic contextual inversions such as P are
next higher triad of the same mode, and paral- also known as wechsels (Hugo Riemann’s term).
lel major and minor triads in the two circles are While fixed inversion operators such as I may be
aligned radially. The placement of major triads more familiar to most readers through their use
in the outer circle and minor triads in the inner in pitch-​class set theory and serial theory, the
is an arbitrary decision of no real consequence; relationships described by contextual inversions
“inside” and “outside” have no meaning in the often correspond more intuitively with familiar
space depicted here. ways of thinking about chordal relationships,
Figure 4.1.1b substitutes the inversion opera- particularly in tonal contexts.3 The I-​related
tor I of pitch-​class set theory for P. As usual, I pairs of triads in 4.1.1b include C–​f (a pair of

FIGURE 4.1.1 Two chromatic arrangements of consonant triad space (triad)

120 • E x p l o r i n g M u s i c a l S p a ces
triads with one common tone and roots related (c) The construction of Figure 4.1.1a ensures
by fifth), E♭–​d (no common tones, roots related that P-​related triads always appear in pairs.
by semitone), and E–​ c♯ (two common tones, In your diagram for part (b) of this exercise,
roots related by minor third)—​a less consistent some P-​related triads (such as C major and
relationship than that exhibited by the P-​related C minor) are aligned, but others are not.
triads in 4.1.1a. We shall have more to say about How far apart are A major and A minor
the group-​theoretic properties of these transfor- in this diagram? (Recall from Section 3.1
mations and about fixed and contextual inver- the definition of the distance between two
sions later, but for the present the point is that vertices in a graph.) How many P-​related
the similar orientations of the two circles in (a) pairs are aligned? Are any P-​related pairs
are characteristic of the behavior of contextual aligned in Figure 4.1.1b? If you were to
inversions, while the opposite orientations in construct diagrams like 4.1.1b using each of
(b) are characteristic of fixed inversions. Fixed the twelve fixed inversion operators, in how
inversions reverse the orientation not only of many of those twelve diagrams would some
individual chords but of entire spaces, in a way   P-​related triads be aligned?4
that contextual inversions do not.
In either 4.1.1a or 4.1.1b, one of the two Exercise 4.1.3
circles could be rotated relative to the other, (a) The transposition operator T1 and the fixed
creating twelve different possible alignments inversion operator I do not commute; that
between major and minor triads. The twelve pos- is, IT1 is not the same function as T1I. Verify
sible alignments of (b) correspond to the twelve this by applying both IT1 and T1I to a C-​
fixed inversion operators In =​ ITn (that is, I ● Tn, major triad. How is this noncommutativity
“I-​then-​Tn,” in accordance with the left-​to-​right evident in Figure 4.1.1b?
orthography convention introduced in Section (b) Does T1 commute with P? How is your
2.3). Similarly, the twelve possible alignments of answer to this question evident in
(a) correspond to twelve different wechsels, to be   Figure 4.1.1a?5
studied further in Chapter 8.
The circle of fifths may be used as readily as
Exercise 4.1.2 the chromatic circle as the basis for a double-​
(a) Another example of a contextual inversion circle space. (Recall from Exercise 1.3.13 that
is the leittonwechsel transformation L. no other interval besides semitones and perfect
Redraw Figure 4.1.1a in an arrangement in fifths generates a coherent picture of twelve-​
which L-​related triads such as C major and note pitch-​class space.) In fact, Figure 4.1.4a,
E minor are aligned. based on the circle of fifths, is likely a more famil-
(b) Redraw Figure 4.1.1b in an arrangement in iar configuration than either of the diagrams in
which triads related by I7 (=​ IT7), such as C Figure 4.1.1, in which triads were arranged chro-
major and C minor, are aligned. matically. The arrangement in 4.1.4a is generated

FIGURE 4.1.4 A circle-​of-​fifths–​based arrangement of consonant triad space (triad)

Chordal, Tonal, and Serial Spaces • 121

by the transposition operator T7 and the rela- Section 4.2 we will compare it with two other
tive transformation R (a contextual inversion). distance measures. Closely related keys are sepa-
Rather than T7, transformation theorists have rated by a distance of 1; the distance between
sometimes worked with the so-​called dominant keys that are not closely related is the number
transformation D, which maps any triad to the of accidentals by which their signatures differ,
triad of the same mode a fifth lower. As a func- allowing for enharmonic respelling.
tion on triad space, D is equal to T5, the inverse
of T7; either T7 or T5 (=​ D) works equally well as a Exercise 4.1.5
generator for a double-​circle space.6 (a) What is the distance between F major and
When the vertices of Figure 4.1.4a are taken C minor in Figure 4.1.4b? What would this
to represent key areas rather than simply triads, distance have been if calculated using 4.1.4a
the diagram is the familiar circle-​of-​fifths model instead?
of key relationships. Each key is adjacent in the (b) What is the distance between E major and
graph to three other keys to which it is generally A♭ major in Figure 4.1.4b?
considered to be closely related: the two fifth-​ (c) What is the greatest distance between any
related triads of the same mode, and its relative   two keys in Figure 4.1.4b?
major or minor. We may define key space, denoted
key, as the set of the 24 major and minor keys. Exercise 4.1.6
The elements of key space are of course in one-​ (a) Construct a diagram of consonant triad
to-​one correspondence with the elements of space based on the circle of fifths and the
triad space, and any map of either space could fixed inversion operator I. (This diagram
theoretically be a map of the other, but our dif- could be a map of key space, but you will
fering conceptions of key and triad may make it need to think about triads in order to
appropriate to impose different additional layers understand the action of I.)
of structure on the two spaces. For example, the (b) Does T7 commute with R? Does T7 commute
large distance separating F and c♯ in Figure 4.1.4a with I? How are your answers to these
is consistent with the way we likely understand questions evident in Figure 4.1.4a and the
the relationship between F major and C♯ minor diagram you constructed in part (a) of this
as keys. But by some measures of triadic distance, exercise?
F major and C♯ minor are not so distant at all, (c) On the basis of Exercises 4.1.3 and 4.1.6b,
particularly if the distance measure is based on what general conclusions can you draw
voice leading, as a one-​semitone displacement about the commutativity of transposition
in each voice suffices to convert one triad to the operators with fixed or contextual inversion
other (these triads are in the hexatonic pole rela-   operators?
tionship). Several measures of triadic distance
are defined below; the voice-​leading distance will Exercise 4.1.7 As maps of key space, Figures
be introduced in Chapter 10, and distance func- 4.1.1 and 4.1.4 include only two modes, major
tions will be studied in a more general context in and minor. The goal of this exercise is to
Chapter 12.7 construct a more comprehensive mode space
The first theorist to arrange key space as in containing all seven diatonic modes on each of
Figure 4.1.4a was apparently David Kellner in the twelve tonic notes.
1737.8 A refinement in Figure 4.1.4b makes You may wish to experiment with several
the graph an even better model of distances ways to construct such a space, but a torus with
between keys: the added diagonal edges bring a 12-​by-​7 modular tile is suggested. In one
C major into adjacency with both E minor and dimension, arrange the twelve transpositions
D minor, in accordance with the usual concep- of the same mode in circle-​of-​fifths order (C
tion of closely related keys. Any given key is now ionian, G ionian, D ionian, …). In the other
adjacent to five other closely related keys, those dimension, arrange the seven modes of the
whose signatures differ from that of the starting same diatonic collection, also in circle-​of-​fifths
key by no more than one sharp or flat. One can order (C ionian, G mixolydian, D dorian, …).
use this graph to determine a distance between Construct the complete tile, which contains 84
any two keys, defined (as in Section 3.1) as the elements. This arrangement may be regarded as
number of edges in the shortest path joining the an extension of Figure 4.1.4a in which the radial
two vertices. We refer to this distance measure dimension is expanded to include seven modes
as the circle-​of-​fifths distance between keys; in rather than two. Consider cycling through the

122 • E x p l o r i n g M u s i c a l S p a ces
modes, starting with C ionian and raising one Computer animation can be a valuable tool for
note at a time by semitone to create a new mode: displaying longer stretches of music, but other
C ionian, C lydian, C♯ locrian, C♯ phrygian, … . strategies, such as the series of snapshots in
What path is traced through the torus by this Figure 1.4.23, can sometimes be devised.
progression
   of modes?9 In the case of the double circle of Figure 4.1.4,
a possible approach is to unwrap the circle of
The P and T1 relationships in Figure 4.1.1a fifths along the vertical axis of a graph whose
suggest specific voice-​ leading connections horizontal dimension is time; the two concentric
between triads: T1 is accomplished by sliding all circles of major and minor triads may be distin-
three voices up by semitone, while movement guished by using two different graphic symbols
to a parallel triad is accomplished by semitonal when plotting triads in the graph. The picture
motion in just one voice. In principle, however, that results has been called a tonality graph by
double-​circle spaces allow for any major triad to Stephen Jablonsky (2003); it is essentially the
be aligned with any minor triad, so there is no graph of a function that maps from the time
guarantee that smooth voice-​ leading connec- domain into fifths space. An example with some
tions will be present in a graph of this kind: a modifications to Jablonsky’s notation appears
graph might, for example, juxtapose C-​ major in Figure 4.1.8, a tonality-​graph analysis of the
and B♭-​minor triads, which are not connected by chorale-​like central section of Chopin’s Nocturne
any especially smooth voice leading. In general, in G Minor, Op. 37, No. 1. The excerpt is in E♭
therefore, double-​ circle graphs should not be major, so the horizontal line representing E♭
regarded as “voice-​leading graphs” but as some- major/​C minor is highlighted and taken as the
thing more abstract, perhaps as a kind of logical origin for the vertical circle-​of-​fifths axis. The
geometric catalog of all the major and minor tri- music proceeds in a regular quarter-​note har-
ads, a database that can be sorted in a variety of monic rhythm, four chords per measure. Black
ways. In Part Three we will examine spaces orga- circles represent major triads, white circles
nized in ways that more directly address voice-​ minor triads, and hybrid symbols are introduced
leading connections between chords. for the three dominant seventh chords (plotted
Chord progressions may be plotted in spaces in the same location as major triads on the same
such as those of Figures 4.1.1 and 4.1.4 simply by roots) and the excerpt’s one half-​diminished sev-
tracing the music from chord to chord through enth chord (eø7 in m. 54, plotted in the graph at
the graph, much as we traced the course of indi- the position corresponding to its G-​minor tri-
vidual notes in Lutosławski’s Musique funèbre adic subset).
in the ladder graph of Figures 1.4.22–​1.4.23, or The graph shows at a glance that the excerpt
Webern’s twelve-​tone rows in the cubic graph of opens in straightforward diatonic fashion: the
3.5.15b.10 As we noted in conjunction with those chords in mm. 41–​44, repeated exactly in mm.
earlier examples, however, plotting anything 45–​48, never stray more than one step from the
more than a short musical excerpt on one graph tonic, either in the dominant direction (“sharp-
becomes difficult when the music inevitably wise,” or upward in the graph) or the subdomi-
crosses or retraces portions of its earlier path. nant direction (“flatwise,” downward). Beginning

FIGURE 4.1.8 Tonality graph for Chopin, Nocturne in G Minor, Op. 37, No. 1, mm. 41–​57 (after
Jablonsky)

Chordal, Tonal, and Serial Spaces • 123

in m. 49, however, the tonal territory covered and half-​diminished seventh chords, arranged in
by the graph expands dramatically, with wider circle-​of-​fifths order (reading “C” to mean C7 and
swings from chord to chord reflecting increased “c” to mean cø7). In fact, the same structure could
chromaticism in the music. The A-​major chord represent a space comprising the 24 sets belong-
on the last beat of m. 54 is six vertical steps from ing to any asymmetrical set class, such as 014
home, fully halfway around the circle of fifths. trichords. Such trichords do not have a “root” or
The graph may call attention to a variety of other “quality” in the traditional sense, so it may not be
interesting patterns as well; for example, in this immediately clear which vertex should stand for
primarily major-​key excerpt, minor triads appear which trichord, but this is not much of a concern:
substantially more often on the second and third we may, for example, represent each trichord by
beats of a measure (twelve occurrences) than on the first note in its normal order, placing the 014-​
the first and last beats (five). type trichords in the outer circle and the inver-
Figure 4.1.8 demonstrates some of the appeal sionally related 034-​type trichords in the inner
of graphs of this type, but limitations are also circle (so that “C” would represent the set {C, C♯,
apparent. Chord inversion is ignored, and if the E} while “f” would represent {F, G♯, A}).12
strategies noted above for plotting dominant
and half-​ diminished seventh chords are not Exercise 4.1.11 The restriction to asymmetrical
completely satisfactory, more serious problems set classes in the preceding paragraph ensures
are posed by chords such as diminished triads, that a set class is represented by 24 different
diminished seventh chords, and augmented pc sets. Symmetrical set classes have fewer
sixths (none of which occurs in this excerpt). representatives and correspondingly give rise
Perhaps a more significant problem still is the to simpler spaces. For each of the set classes
fact that no particular affinity is visible in the listed below, construct a space comprising all
graph between parallel major and minor triads. the sets in the class. Each space should consist
As a consequence, while the close relationship of either one circle or two concentric circles, as
between a major triad and its dominant is evi- appropriate. Adjacent sets on each circle should
dent, a minor triad and its major dominant are be related by T1, but symmetry may result in
separated by four steps along the vertical axis, some circles having fewer than twelve elements.
which makes even a simple diatonic passage in a
minor key such as the tonicization of F minor in (a) 0358 (the minor-​minor seventh chord)
mm. 50–​52 look more extravagantly chromatic (b) 013679 (the Petrushka chord)
than it actually is. (c) 0268 (the French augmented-sixth chord)
(d)  0369 (the fully diminished seventh chord)13
Exercise 4.1.9
(a) Keeping in mind that the vertical axis Even the requirement that the sets in the two
of Figure 4.1.8 is a circle, describe the circles be related by inversion may be relaxed if
appearance of the tonality graph of a piece desired: if we encounter a musical example in
that modulates all the way around the circle which major triads alternate with half-​dimin-
of fifths. ished seventh chords, we can model that pro-
(b) The topology of a tonality graph such as gression in a space that combines a circle of
Figure 4.1.8 is neither a plane nor a torus. major triads with a circle of half-​ diminished
  What is it? seventh chords. In principle, we could go further
and combine three or more concentric circles to
Exercise 4.1.10 Construct a tonality graph accommodate a greater variety of chord types in
similar to Figure 4.1.8 for Chopin’s Prelude in E a single space. The drawback in these far-​flung
   Op. 28, No. 9.11
Major, representations is that the meaning of moving
from one chord to an adjacent chord on a dif-
The construction of Figures 4.1.1 and 4.1.4 ferent circle—​the radial dimension in the dia-
began by associating points in pitch-​class space gram—​becomes increasingly nebulous, as it can
with triads. We can equally well establish a one-​ no longer be described in terms of a single (fixed
to-​one mapping between pitch classes and other or contextual) inversion operator.
types of pitch-​ class sets, thereby assembling Serial theory, however, offers an appealing
double-​circle diagrams for a variety of different opportunity to combine four circles. In Figures
chords. Figure 4.1.4a could be interpreted, for 3.2.3 and 3.5.8d we constructed a cube con-
example, as a space consisting of all dominant sisting of eight row forms related by inversion,

124 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 4.1.12 Two three-​dimensional representations of twelve-​tone row space (some vertical edges
omitted for clarity)

retrograde, and transposition—​but only trans- in circular pitch-​class space; rather, J is a kind
position by tritone (T6). The T dimension of of contextual inversion. Figure 4.1.12b shows
that three-​dimensional figure may be extended an analogous picture using the more familiar
to include all twelve transpositions, replicat- fixed inversion operator I, inversion about C,
ing a square cross-​section of I-​ and R-​related here assumed to be the first note of P0. A check
rows not just twice but twelve times, adjacent of the numerical subscripts confirms that circles
slices related by T1. The slices must be circularly representing inversionally related row forms
arranged so that the initial slice is adjacent to its (P and I, or R and RI) are oriented similarly in
T11 transposition. The resulting space, shown in Figure 4.1.12a but oppositely in (b), in keeping
Figure 4.1.12a, consists of two concentric circles with our earlier observations about the effects
on each of two layers; its 48 vertices represent all of contextual and fixed inversion operators
the standard forms of a twelve-​tone row. in Figure 4.1.1. No properties of a particular
The row space of this figure, like the earlier twelve-​tone row were used in the construction
cube, is three-​dimensional: from any vertex one of either part of Figure 4.1.12, so both figures
can move circumferentially, vertically, or radi- are applicable to the forms of any row at all. In
ally. Circumferential movement is defined by fact, both figures describe the structure of trans-
transposition as just described. Vertical move- formation groups fundamental to twelve-​tone
ment is defined by retrograde: each row form in music, in ways to be explored in Section 6.2.
the top layer is the retrograde of the correspond-
ing row in the bottom layer. (The vertical edges Exercise 4.1.13 Some twelve-​tone rows are
on the inner shell of the figure, connecting I and symmetrical, so that some R or RI form of the
RI forms, have been omitted for clarity.) Finally, row is identical to P0. Construct examples of
radial movement is defined by an inversion twelve-​tone rows of these types. Explain how
labeled J. Specifically, J inverts any row form to the row space of Figure 4.1.12 may be simplified
preserve its index number. This is an awkward the case of a symmetrical row.14
in  
way to define a transformation—​in traditional
row numbering, it means inverting a P or I row Exercise 4.1.14 As described above, Figure
form about its first note and an R or RI form 4.1.12a may be regarded as an expansion of
about its last note, so that we must know which 3.5.8d. In the cube of the earlier figure, the
type of row form we are dealing with before inversion operator I represents inversion
applying the transformation—​but it accurately about a fixed pitch class, the first note of
describes the relationship between radially P0. Nevertheless, this inversion operator
adjacent row forms in the figure, designed to apparently relates row forms sharing the same
ensure alignment of all row forms with match- index number, such as P0 and I0. Explain why a
ing indices. fixed inversion operator suffices in that limited
This J is not one of the fixed inversion opera- situation but not in the greater generality of
tors In, which are inversions about specific axes    4.1.12a.15
Figure

Chordal, Tonal, and Serial Spaces • 125

Exercise 4.1.15 Construct a four-​circle graph a variety of alignments between its circles (the
like Figure 4.1.12a whose elements are chords rows of Figure 4.1.16). Alternate alignments
of four types: major triads, minor triads, may be exploited in analyzing combinatorial
dominant seventh chords, and half-​diminished relationships of the sort explored by Schoenberg
seventh chords. Each square cross section of the in many of his mature twelve-​tone works. The
space will consist of two parallel triads and the arrangement in Figure 4.1.18, in which P0 and
two seventh chords containing those triads as R0 are aligned with I5 and RI5, is customized for
subsets,
   such as C, c, C7, and aø7. Schoenberg’s Piano Piece, Op. 33a, although the
same graph would serve as well for a number of
The doughnut-​like appearance of the graph in other pieces by Schoenberg, inasmuch as this
Figure 4.1.12 perhaps suggests a torus. Each of P0–​I5 combinatoriality seems to have been his
these graphs is effectively a toroidal embedding, favorite type.
and like the tonnetz graphs of previous chapters In Op. 33a, Schoenberg consistently pairs
can be represented conveniently by a rectangular row forms P10 and I3 because of combinatorial
tile with opposite edges conjoined, as shown in considerations (the first hexachords of these two
Figure 4.1.16. The circles of 4.1.12a have become rows combine to form a twelve-​tone aggregate).
the horizontal rows of 4.1.16, and the square Similarly he uses R10 in conjunction with RI3.
cross-​sections have become the columns. A pos- David Lewin introduced the notion of twelve-​tone
sible drawback to this arrangement is that two areas to describe such a collection of row forms;
different musical relationships, inversion and here we write 𝒜10 for the twelve-​tone area {P10,
retrograde, are now conflated in a single verti- R10, I3, RI3}. Lewin suggested that Schoenberg’s
cal dimension, whereas in 4.1.12a J and R corre- use of twelve-​tone areas is analogous in some
sponded to movement in the radial and vertical respects to the use of key areas in tonal music.
dimensions respectively. The first 27 measures of Op. 33a are “in 𝒜10,” as
only the four row forms of 𝒜10 appear in these
Exercise 4.1.17 measures. At m. 28, however, Schoenberg “mod-
(a) Like the tonnetz, the graph of row space in ulates” to area 𝒜0 =​{P0, R0, I5, RI5}; he then pres-
Figure 4.1.12a or 4.1.16 can be embedded ents two row forms from 𝒜5 (P5 and I10) before
in a torus but not in a plane. Referring back returning to 𝒜10 in the last section of the piece.
to the proof of Theorem 3.3.12 as a model, As Figure 4.1.18 illustrates, each twelve-​tone
demonstrate that Figure 4.1.16 contains a area is one column of the modular row space (or
subdivision of the complete graph K5 and one cross-​section of a corresponding doughnut-​
therefore cannot be planar. shaped space).16
(b) As a graph embedded in a torus, how many
vertices, edges, and faces does Figure 4.1.12a Exercise 4.1.19 The above discussion of Op.
or 4.1.16 have? Verify that these numbers 33a assumes “fixed-​zero” row-​numbering;
satisfy v +​ f =​ e, the toroidal version of Euler’s row form P10 starts on B♭. In a “movable-​zero”
  formula from Exercise 3.3.16b. system this row might be called P0. Would the
map of row space in Figure 4.1.18 need to be
Like the double-​circle graphs considered ear- modified if the movable-​zero system were used?
lier, the four-​circle graph of row space admits Would
   the twelve-​tone areas change?

FIGURE 4.1.16 A modular two-​dimensional representation of twelve-​tone row space

126 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 4.1.18 A row space showing twelve-​tone areas in Schoenberg, Piano Piece, Op. 33a

4.2 TONNETZ-​R ELATED the equal division of an octave into four minor
CHORDAL AND thirds. Any attempt to depict P and R simultane-
TONAL SPACES ously must therefore introduce some additional
relationship as well if it is to transcend this small
A useful picture of key relationships emerged cycle and accommodate relationships among all
in Figure 4.1.4, the double circle of fifths align- keys (that is, if it is to be a connected graph of all
ing relative major and minor keys. A possible of triad space).
shortcoming of this figure, already mentioned
in conjunction with the related tonality graph of Exercise 4.2.2 Figure 4.2.1 is a PR-​cycle (or an
Figure 4.1.8, is that it pays no heed to the paral- RP-​cycle). The hexatonic cycle of Figure 3.5.8b
lel relationship. A double-​circle space may juxta- may similarly be called a PL-​ (or LP-​) cycle.
pose relative keys or parallel keys, but not both. Construct an RL-​ (or LR-​) cycle. Comment on
Consequently, and somewhat counterintuitively, the relationship between the RL-​cycle and the
Figure 4.1.4 places C major just as close to F♯ circle
   of fifths.
minor as it is to C minor: the circle-​of-​fifths dis-
tance from C major to either of those keys is 3. If the space is to model tonal relationships,
If we wish to allow both parallel and relative the perfect-​fifth relationship T7 (or its inverse,
moves in the same graph, we will soon realize the dominant transformation D =​ T5) seems
that the number of other keys accessible from a likely candidate for a way to expand the PR-​
a given key by combinations of P and R is lim- cycle of Figure 4.2.1 to other keys. If we unwrap
ited. Starting from C major, the possibilities the PR-​cycle in one dimension and the circle of
are exhausted by the eight keys in the PR-​cycle fifths in another, the result is the space shown in
shown in Figure 4.2.1. Graphically, this figure Figure 4.2.3. We call this arrangement Weber
is indeed a cycle as defined in Section 3.1; musi- space, or weber, recognizing its often-​ cited
cally, this particular cycle is a manifestation of appearance in the work of Gottfried Weber
around 1820, although similar diagrams were
published as much as half a century earlier.17
In any case, the appeal of this space has been
powerful and enduring: the same basic organi-
zation underlies the “chart of the regions” in
Schoenberg’s posthumously published Structural
Functions of Harmony, and also the “regional
space” of Fred Lerdahl.18 Experimental work
by Carol Krumhansl has confirmed the cogni-
tive salience of Weber space: through investiga-
tion of listeners’ responses to a variety of tonal
stimuli, Krumhansl and her colleagues obtained
FIGURE 4.2.1 A PR-​cycle of major and minor a measure of perceived distances between keys
triads and applied statistical techniques to convert

Chordal, Tonal, and Serial Spaces • 127

FIGURE 4.2.3 Weber space (weber)

these distances to a perceptual map, which Two of the three basic neo-​Riemannian
turned out to resemble a slightly distorted ver- operations, P and R, are explicitly present in the
sion of Figure 4.2.3.19 construction of Weber space, and the third, L,
The topology of Weber space is that of a torus, appears almost as handily, in the form of diago-
but no simple rectangular module (comparable to nal lines sloping downward from any minor key
the tile of Figure 1.4.3 for the tonnetz or Figure to the major key positioned to its lower right.
4.1.16 for row space) can be drawn in Figure 4.2.3 Figure 4.2.5a adds edges to Weber space corre-
to describe the toroidal topology. The parallelo- sponding to P, R, and L, effectively rendering the
grams defined by dashed lines in the figure may space a transformation network. Because each
be used instead, opposite edges of each module of these three transformations is an involution
being stitched together in the usual way; the (equal to its own inverse), every edge in this
process is topologically equivalent to the more graph may be considered to have arrowheads
familiar rectangular modules if less tidy geo- at both ends. Each face of the resulting graph is
metrically. Like other toroidal spaces we have bounded by six edges, two labeled with each of
studied, weber is finite—​there are only 24 major the three operations. In Figure 4.2.5b an adjust-
and minor keys, after all—​and one of these paral- ment to the geometry reshapes these faces into
lelogram tiles, which contains all 24 keys, theo- regular hexagons, which we will call cells, in a
retically suffices to model the entire space. As honeycomb-​like grid.22
before, however, it is often more convenient in We will use the term neo-​Riemannian triad
practice to imagine a theoretically infinite graph space, or nrtriad, to refer to either of the
and show at least some partial repetitions of the graphs of Figure 4.2.5, but especially the hex-
pattern. We must just keep in mind, when draw- agonal cellular arrangement in (b). This version
ing the space this way, that duplicate key names of the graph gives equal status to all three of the
represent the same point of the space.20 neo-​Riemannian operations while downplay-
ing the perfect-​fifth relations that were part of
Exercise 4.2.4 the original motivation for Figure 4.2.3; the less
(a) Parallelogram tiles for Weber space may be celebrated 4.2.5a has been included primarily
drawn in various ways other than the way to clarify the relation between neo-​Riemannian
shown in Figure 4.2.3, but rectangles with triad space and Weber space. Particularly when
horizontal and vertical edges will not work. contrasting it to Weber space, we will some-
What problem arises if one tries to use times shorten the name “neo-​Riemannian triad
rectangles, three rows by eight columns, as space” to Riemann space.23 Weber space and
tiles for the toroidal topology of this space?21 Riemann space are identical as sets (both con-
(b) As a finite graph embedded in a torus, how sist of the same 24 triads or keys) but differ in
many vertices, edges, and faces does Weber other respects, particularly in the edges that are
space have? Verify that these numbers understood (explicitly or implicitly) to be part of
satisfy v +​ f =​ e, the toroidal version of the graph-​theoretic structure of the spaces, the
  Euler’s formula. associated transformations, and the geometric

128 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 4.2.5 Two representations of neo-​Riemannian triad space (nrtriad)

representations: Weber space is organized by P, (c) Riemann space as described above is a

R, and T7 (or T5), while Riemann space is orga- directed graph with bidirectional arrows
nized by P, R, and L. exclusively. As an alternative approach, we
may define arrows to show the direction
Exercise 4.2.6 of ascending voice leading: for instance,
(a) Review, if necessary, the definition an arrow points from C major to A minor,
of bipartite graphs in Section 3.1. The not the other way around, because the
graph of Riemann space in Figure 4.2.5 ascending voice leading G → A transforms C
(either version) is bipartite in a musically major to A minor. Redraw Figure 4.2.5b as a
suggestive way. What are the two sets into directed graph following this convention.
which the vertices (triads) are divided? (d) In Riemann space each triad is adjacent
(b) As a finite graph embedded in a torus, to the three other triads with which it
how many vertices, edges, and faces does shares two common tones. How can you
Riemann space have? As in the previous tell by looking at Figure 4.2.5b which triads
exercise for Weber space, verify that these share one common tone with a given triad?
numbers satisfy the toroidal version of Add edges to a copy of this graph so that
Euler’s formula. triads with one or more common tones are

Chordal, Tonal, and Serial Spaces • 129

 adjacent. In this new graph, each vertex is of the triangulated tonnetz. Moreover, a fore-
adjacent to how many other vertices? What ground vertex, representing a triad, lies not in
is the largest distance between any two just any face of the background, but in the spe-
  triads in this graph? cific triangle whose three vertices are labeled
with the pitch classes of the foreground triad.
Exercise 4.2.7 Trace the pattern formed by a For instance, a vertex representing F major lies
PR-​cycle, an PL-​cycle, an RL-​cycle, and a circle-​of-​ at the center of the triangle whose vertices are
fifths pattern in both Figure 4.2.5a and 4.2.5b. labeled 5, 9, and 0, the pitch-​class numbers of
Which one of these types of cycle is a spanning the notes of the F-​major triad.24
cycle for the graph, as defined in Section 3.1? We can think of T3, T4, and T7 as ways to move
What does the orientation of the arrows in your either from vertex to vertex in the background
directed graph from Exercise 4.2.6c tell you or from cell to cell in the foreground. Similarly,
about
   the voice leading in these cycles? we can think of P, L, and R as ways to move
either from vertex to vertex in the foreground or
Each hexagonal cell in Figure 4.2.5b is sur- from triangle to triangle in the background. The
rounded by six other cells, and the keys appear- parallel transformation P, for instance, is a diag-
ing in these cells are systematically related. onal move in the foreground (indicated by a solid
Consider, for example, the highlighted cell com- line)—​but we may also imagine performing P in
prising the keys F–​a–​C–​c–​A♭–​f. Immediately the background layer by flipping a triangle over
to its right is another cell A♭–​c–​E♭–​d♯–​B–​g♯. its diagonal edge (a dotted line). Similar remarks
Geometrically, this second cell is a translation of apply to R and L. Two triangles related by P, L, or
the first; musically, it represents a T3 transposi- R always share two common vertices, signifying
tion of the first cell, each of these keys being a that each of these Riemannian transformations
minor third higher than the correspondingly preserves two common tones.
placed key in the first cell. Similarly, immediately
above and to the left of each cell is its T4 trans- Exercise 4.2.9
position, a major third higher, and immediately (a) Imagine a colored version of Figure 4.2.8
above and to the right is its T7 transposition, a in which the solid P lines and dotted T7
perfect fifth higher—​what we might call the lines are red, the R and T4 lines are blue,
“dominant” of the original cell. and the L and T3 lines are green. Every solid
These relationships may have a familiar ring. line then crosses a dotted line of the same
The transpositions T3, T4, and T7 are precisely color. While these crossings should not
the relationships that defined the triangulated be read too literally—​no musical meaning
tonnetz of Figure 3.5.8a as a transformation attaches to the locations in the figure
network. Figure 4.2.8 demonstrates that the where lines cross—​there is nevertheless
correspondence between the triangulated ton- some significance to the relation between
netz and neo-​Riemannian triad space goes well the two phenomena illustrated by a
beyond this vague similarity. This diagram, which pair of intersecting lines. In the parallel
synthesizes a number of different relationships transformation, what role is played by
we have considered to this point, is constructed T7 (or notes related by T7—​that is, notes
in two layers. The foreground layer is a reorien- separated by interval class 5)? Describe, in a
tation of the cellular Riemann space of Figure similar way, the connection between T4 and
4.2.5b. The background layer is a separate graph, R, and between T3 and L.
its edges drawn in light dotted lines. This back- (b) Every hexagonal cell in the foreground
ground graph is tonnetz space tonn, a triangu- layer has a single background pitch class
lated tonnetz identical to Figure 1.4.5 or 3.5.8a at its center—​the “nucleus” of the cell. For
except that note names have been replaced by example, the nucleus of the cell D–​f♯–​A–​
pitch-​class numbers to avoid confusion with the a–​F–​d is pitch class 9 (the note A). What
triad or key names in the foreground layer. property links that note to those six triads?
Correspondences between the two layers of (c) In Exercise 3.3.16a you determined the
Figure 4.2.8 are numerous. Each vertex of the numbers of vertices, edges, and faces in
foreground layer lies in the middle of one tri- the triangulated tonnetz (as a finite graph
angular face of the background layer. In the ter- embedded in a torus). In Exercise 4.2.6b
minology introduced in Section 3.1, the graph you determined corresponding numbers
of neo-​Riemannian triad space is the dual graph for neo-​Riemannian triad space. Verify

130 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 4.2.8 The triangulated tonnetz and neo-​Riemannian triad space as dual graphs

that these two sets of numbers are in of the triangulated tonnetz and Riemann space
the proper relationship for a pair of dual as in Figure 4.2.8; or occasionally even (5) to
graphs: that is, both graphs should have the the Weber space of Figure 4.2.3. In keeping with
same number of edges, and the numbers of this tradition, we shall use the word tonnetz here
  vertices and faces should be interchanged. with some flexibility, though we will refrain from
applying it to Weber space. Of course we retain
Because of its strong affinity with the trian- the option to use our other, more specific names
gulated tonnetz, neo-​Riemannian triad space for these diagrams when clarity demands.
is sometimes described as a kind of tonnetz in In any case, it is the ease of depicting rela-
its own right. In the neo-​Riemannian litera- tions between triads or key areas (as in Weber
ture, therefore, the word tonnetz may refer (1) or Riemann space), more than pitch-​class rela-
to the triangulated tonnetz of pitch classes; tions (as in the triangulated tonnetz), that
(2) to some other related graphs presented in drives much of the appeal of the tonnetz and
Chapter 1, such as the unconformed tonnetz of its widespread applications in neo-​Riemannian
Figure 1.4.1; (3) to the neo-​Riemannian triad theory and analysis. Both Weber and Riemann
space of Figure 4.2.5b; (4) to an amalgamation space have proven valuable as maps in which

Chordal, Tonal, and Serial Spaces • 131

 collection, and the tonnetz offers a concise visual
demonstration of why a PL-​cycle is also called a
hexatonic cycle. The tonnetz admits four different
pairings of two adjacent columns, corresponding
to the four transpositions of the hexatonic col-
lection. Following Straus (2016, 257), we label
these four sets hex01, hex12, hex23, and hex34,
indexing each set by the numerically smallest
pair of consecutive pitch-​class numbers appear-
ing within it;26 the Brahms progression inhabits
the collection hex34.

Exercise 4.2.11 Using reasoning similar to the

preceding paragraph, describe the appearance
FIGURE 4.2.10 PL-​ cycle in Brahms, Double
of the three octatonic collections oct01, oct12,
Concerto, Op. 102, I, mm. 270–​78
and oct23 in the tonnetz, and explain why a
  cycle may be called an octatonic cycle.
PR-​
to trace the progress of triadic progressions.25
Although our understanding of these triadic Starting from C major, the operations L and
relations will be further enhanced by our study R yield the two diatonic-​third–​related keys, E
of algebraic properties of Riemannian and minor and A minor respectively. In general, L
other transformations in Chapter 8, pausing relations are characterized by the interval of
now to explore a few such applications in the a major third between roots, R relations by a
tonnetz of Figure 4.2.8 will offer some new minor third. Of course E minor and A minor
ways to appreciate the behavior of P, L, and R, are also the diatonic mediant and submediant
among other musical rewards. keys of C major. The correspondence between L
The PL-​ cycle from the Brahms Double and R, on the one hand, and mediant and sub-
Concerto discussed in conjunction with the mediant, on the other, works one way for major
hexatonic triad graph in Section 3.1—​the tri- keys and the opposite way for minor keys. That
adic progression A♭–​g♯–​E–​e–​C–​c–​A♭–​g♯–​E—​may is, the leittonwechsel of a major key is its medi-
serve again as an illustration. In the tonnetz, ant and the relative is its submediant, while for
this progression takes on the appearance shown a minor key the relative is the mediant and the
in Figure 4.2.10. This representation shows just leittonwechsel is the submediant. In Chapter 8
one module of the tonnetz (every triad appears we will define mediant and submediant trans-
once), and exploits the drawing conventions of formations, M and M−1; these are not the same
the flat torus: moving downward through the functions as R and L, and we will see reasons why
graph, the path reaches the bottom boundary of it is not really accurate to apply any form of the
the figure, at which point it reappears at the top. adjective “Riemannian” to M and M−1.27
One could avoid the need for a discontinuity in When L or R is used in combination with
the path by using an infinite tonnetz graph such the mode shift characterizing the P transfor-
as Figure 3.3.10a, but such a depiction obscures mation, as in the PL-​ and PR-​cycles, chromatic
the fact that the progression’s opening chords third relations result, involving chords com-
reappear at the end. Indeed, the progression monly called “mixture” chords. Starting from C
circumnavigates the PL-​ cycle and then some, major, applying first R and then P yields A major:
overshooting the initial A♭-​major triad and con- C R → a P→ a , or, more compactly, C RP → a .
tinuing as far as E. (As usual, RP denotes R ● P, “R-​then-​P.”) Similarly
The pitch classes in the background layer the reader should be able to verify the follow-
that are engaged by this progression form two ing composite actions and visualize them on
adjacent columns in the triangulated tonnetz. the tonnetz: C PR → E ♭, C LP → E, C PL → a ♭.
The three pitch classes in any one column define Generally PL-​ cycles (hexatonic cycles) feature
an augmented triad; two adjacent columns major-​third root relationships, while PR-​cycles
comprise a total of six pitch classes, forming a (octatonic cycles) move by minor thirds. It is
hexatonic collection (prime form 014589) as the clear from the above actions that PR ≠ RP and
union of two adjacent augmented triads. A PL-​ PL ≠ LP: Riemannian transformations do not
cycle thus exhausts the notes of one hexatonic commute with each other.

132 • E x p l o r i n g M u s i c a l S p a ces
 Each of P, L, and R is a mode-​reversing opera- for every x. It turns out, however, that if f
tion; that is, each transformation converts a and g are composites of P, L, and R, and if f
major to a minor triad or vice versa. In the tri- and g yield the same result when applied to
angulated tonnetz, each application of P, L, or one triad, then they will automatically yield
R moves from a triangle in one orientation to the same result on every other triad as well,
an adjacent triangle in the opposite orienta- so that in this case f and g must actually
tion. Any composite of two of these operations be the same function. We will derive this
restores the original orientation and is therefore property in Section 8.2 as we study the
mode-​preserving, which is why the composites algebra of Riemannian transformations. For
in the previous paragraph, applied to C major, now, in light of the four duplicate results
always yielded some other major triad (A, E♭, E, mentioned above, this property means that
A♭). Any combination of three of P, L, and R is four algebraic identities must hold: PLP
again mode-​reversing; by adjoining an additional =​ LPL and three others. Can you explain
application of P to the composites PR and PL, we each of these identities visually, using
can obtain the chords of E♭ minor and A♭ minor, Figure 4.2.8? What kind of a path does PLP
sometimes called “double mixture” chords in describe through the tonnetz? What kind
relation to C major. of path does LPL describe, and why do they
The ready availability of all these relations in lead to the same place? Write the other
the tonnetz, coupled with the fact that progres- three algebraic identities, and visualize
sions involving chromatic thirds often prove   them in the tonnetz.
resistant to more traditional methods of tonal
analysis, makes such progressions particularly Exercise 4.2.13 Several of the following
amenable to analysis in the tonnetz. Perfect-​ triadic progressions or key successions
fifth relations may be described using compos- from the nineteenth-​century repertoire
ites of L and R (C LR
→ g, C RL → F) , but in the have been analyzed in the neo-​Riemannian
absence of clear third relations there is seldom literature.28 Trace each progression in (a)–​(f)
a good reason to conceive of the connection as a path through the foreground graph
between a tonic and its dominant or subdomi- of Figure 4.2.8, identify the Riemannian
nant in this way. For these and other reasons transformations involved, and answer the
tonnetz analysis is generally less likely to offer additional questions (g)–​(i).
revealing insights about functional diatonic
progressions than about progressions featuring (a) Beethoven, “Spring” Sonata for Violin and
chromatic third relationships. Piano, Op. 24, II, mm. 37–​54:

Exercise 4.2.12 B♭–​b♭–​G♭–​f♯–​D–​d–​B♭

(a) Without repeating one transformation
twice in succession (which would (b) Liszt, Consolation No. 3 in D♭ Major, mm.
accomplish nothing), twelve different triple 18–​43 (see comment in (g) below):
composites of P, L, and R may be formed:
PLP, PLR, PRP, PRL, and eight others. List D♭–​f–​F–​a–​A–​D♭
all twelve, calculate the result of applying
each of these triple composites to a C-​major (c) Smetana, Overture to The Bartered Bride,
triad, and trace these composite moves on mm. 128–​44:
the tonnetz.
(b) In part (a) you should have found that C–​c–​E♭–​e♭–​G♭–​f♯–​A–​a–​C
the composites PLP and LPL, applied to
a C-​major triad, lead to the same place. (d) Franck, Piano Quintet in F Minor, I,
You should also have found three other mm. 26–​33:
pairs of composites with the property
that the two composites in each pair give E♭–​c–​C–​a
identical results. Recall from the discussion
of equality of functions in Section 2.3 (e) Beethoven, Symphony No. 9, II, mm. 143–​71:
that if f(x) =​ g(x) for some x, we cannot
automatically conclude that f =​ g; that C–​a–​F–​d–​B♭–​g–​E♭–​c–​A♭–​f–​D♭–​b♭–​G♭–​e♭–
conclusion requires knowing that f(x) =​ g(x) ​C♭–​a♭–​E–​c♯–​A

Chordal, Tonal, and Serial Spaces • 133

(f) Chopin, Fantasy in F Minor, Op. 49, mm. every triad in the progression apparent in the
43–​60 (see comment in (g) below):29    31
graph?

f–​A♭–​c–​E♭–​g–​B♭–​e♭–​G♭–​b♭–​D♭–​f The musical examples in the preceding exer-

cises are uncommonly systematic in the repeti-
(g) The examples in parts (b) and (f) above tive patterns by which they move through the
include small irregularities in what is space of the tonnetz. In general we have no right
otherwise a regular pattern. What chord to expect music to be so well-​behaved. The fol-
seems to be missing in the Liszt example? lowing exercise shows that even less orderly
Which composite transformation (PL, LP, musical passages may be amenable to tonnetz
PR, RP, LR, or RL) might be invoked here? analysis.
The irregularity in the Chopin progression
cannot be explained as a single chord Exercise 4.2.15 The fourth movement of
missing from a pattern, but you should Schubert’s “Trout” Quintet is a series of six
have no trouble describing the harmonic variations on his song “Die Forelle.” The
motion here in a more traditional way. (It theme, Variations 1–​3 , and Variation 6 are
may be relevant to note that the B♭ chord in in D major. In this exercise we examine what
(f), uniquely among all the chords in (a)–​ happens during the comparatively brief
(f), is actually a dominant seventh chord.) time that the movement explores other
(h) Because of the irregularity just noted, the keys, and we analyze Schubert’s navigation
Chopin example does not constitute a through the tonnetz, away from D major and
complete cycle of the pattern it defines, back again.
though it begins and ends with the same triad. Variation 4 is in the parallel minor, and
The Beethoven example in (e), despite its Variation 5 begins with a sudden move to B♭
remarkable length, is also not a complete cycle. major, heard as VI of D minor. B♭ major proves
How many more chords would be needed to unstable; the second half of Variation 5 shifts to
bring this progression back to C major? Trace B♭ minor before cadencing in D♭ major (tonally
this complete cycle in Figure 4.2.8. the most remote point from D major—​but,
(i) The progressions in parts (a) and (b) incidentally, the key of the original song). At
exemplify the same basic cycle type as this point Schubert inserts a seven-​measure
the Brahms Double Concerto example transition that begins with a shift to D♭ minor
graphed in Figure 4.2.10. In the tonnetz, and then returns to D major by a series of
your analyses should show that two of descending diatonic thirds, d♭ =​c♯–​A–​f♯–​D. The
these three progressions trace paths in large-​scale tonal trajectory may therefore be
one direction while the other traces a path summarized as
in the opposite direction. What musical
features of the progressions explain this D–​d–​B♭–​b♭–​D♭–​d♭–​A–​f♯–​D.
observation? Compare your analyses in
  parts (c)–​(d) and (e)–​(f) in a similar way.30 Trace this path in the tonnetz of Figure 4.2.8,
and identify the Riemannian transformations
Exercise 4.2.14 Manrico’s aria “Ah sì, ben mio, involved. Also compare this path with those
coll’essere” from Part III of Verdi’s Il trovatore of the preceding exercises; you should find
begins with this series of modulations: fragments of several different patterns
combined
   here.
f–​A♭–​a♭–​F♭–​d♭–​D♭
Several questions may occur to readers
(The triads are spelled in the score as notated working through exercises like those above,
above.) Trace the path of this progression in especially when working in an infinite flat ton-
the tonnetz of Figure 4.2.8, and identify the netz such as Figure 4.2.8. What does it mean
Riemannian transformations involved. What when, for example, Beethoven’s progression
additional transformation could theoretically from Exercise 4.2.13a begins at one B♭-​major
lead from the ending key back to the starting triad but apparently ends at a “different” B♭-​
key, completing a cycle? (Verdi does not supply major triad in the tonnetz? Does this analysis
this transformation; the aria ends in D♭ major.) imply that the passage is somehow not tonally
How is the fact that the pitch class A♭ belongs to closed? The most satisfactory answer to this

134 • E x p l o r i n g M u s i c a l S p a ces
question depends on distinguishing the graph other tonicizations last several measures, unlike
from the space it depicts, and on recalling the the very brief tonicizations of D♭ minor, A major,
toroidal topology of the latter, which allows an and F♯ minor that figure crucially in our analysis
infinite graph to represent a finite space, each of the transition back to D major. A justification
of whose elements appears at many different for disregarding these other tonicizations might
places in the graph. It is therefore possible for be that, however salient they may seem in other
a progression to begin and end at two differ- ways, they do not participate in any broader
ent points of the graph while beginning and movement through tonal space: they lead in
ending at the same point of the space. The two each case back to the previous key rather than
B♭s, though they appear distinct in the plane heading in some new direction. In contrast, each
figure, are nevertheless the same triad, and the step in the descending-​thirds chain in the final
path in neo-​Riemannian triad space is a closed transition, though brief, is an essential part of
one. There is no reason to think that the com- that broader tonal movement: the modulation
posers of these progressions regarded them as back to D simply would not work (or at least
anything but tonally closed: hexatonic cycles would not work in the same way) if any of these
like that of Exercise 4.2.13a appeared in the keys were omitted.
literature a handful of times before 1800 and This distinction leads to a more general
with increasing frequency thereafter,32 almost point. It is in the nature of analysis to be reduc-
never with any apparent attempt to compen- tive, and the analyses suggested above are not
sate by moving back to the starting point in the necessarily more so than other methods with
tonnetz.33 which readers may be familiar. The kinds of
Readers who have consulted scores of the reductions produced by different methods,
works in the preceding exercises may find them- however, may differ. The tonnetz analysis of
selves questioning the criteria by which some of the Liszt progression in Exercise 4.2.13b calls
the chords enumerated above were chosen. In attention to the way that Liszt divides the
the initial Brahms example and the Beethoven octave symmetrically into three major thirds
excerpt in Exercise 4.2.13e, the analysis accounts and navigates through this cycle by a repeat-
for every chord (indeed, every note) in the music, ing pattern, alternating ascending thirds with
so there should be little room for debate. Some mode changes. A Schenkerian analysis of the
of the other examples, however—​ particularly same passage would necessarily highlight
4.2.13a–​b, 4.2.14, and 4.2.15—​may be consid- different features. The Schenkerian analyst
ered more problematic, as these excerpts contain could not reasonably include the bass notes
more chords than those given in the exercises. D♭–​F –​A–​D♭ without also including A♭, the struc-
Some of the listed chords are preceded by their tural dominant preceding the return of the D♭
own dominants, which our analysis disregards; tonic; F might be read as a third-​divider within
the Beethoven analysis in 4.2.13a ignores a com- the larger D♭–​A♭ span, and A =​B♭ ♭ as a more local
plete I–​IV–​V–​I progression in G♭ major. Are we event (♭VI). In short, the tonnetz analysis accen-
guilty of cherry-​picking only the chords that fit a tuates the symmetry of the pattern, while the
desired pattern? Schenkerian analysis subjugates that symmetry
The question is a legitimate one, and sev- to the norms of traditional diatonic tonality.34
eral points should be made in response. First, The reader can probably think of a number of
the objects in neo-​Riemannian triad space may reasons why tonnetz analysis could never be a
represent triads or key areas. In some cases the universal replacement for Schenkerian analysis.
tonnetz analysis may literally recount a progres- To be sure, the musical phenomena that can be
sion of triads; in other cases it may more accu- described in the tonnetz are limited in variety
rately depict a succession of tonal centers. In the in comparison with all that a good Schenkerian
potentially troublesome examples mentioned sketch may account for. Some readers may prefer
above, all of the listed chords are tonicized and to think of the triadic spaces described here as
may legitimately be considered key areas. In an abstract pre-​background level of structure, on
the “Trout” example of Exercise 4.2.15, it might which common-​practice tonality superimposes
then be objected that other keys are also toni- various kinds of conventions not intrinsic to the
cized in this movement: A major within the spaces themselves. Even as representations of
D-​major sections, F major within the D-​minor chords and key areas in tonal music, Weber and
variation, and F major again within the B♭-​major Riemann spaces have evident limitations: they
portion of Variation 5. Moreover, some of these cannot distinguish a triad from a seventh chord,

Chordal, Tonal, and Serial Spaces • 135

or a tonic triad in C major from a dominant in F In comparison with the double-​circle spaces of
major, and they generally pay no heed to diatonic Section 4.1, the spaces considered in this section
organization. The failure of tonnetz analysis offer somewhat different conceptions of the dis-
to recognize a tonal center might be seen as an tance between two key areas. Distances are still
advantage in some cases, such as the PL-​cycle in figured between vertices in a graph, as described
the Brahms Double Concerto, in which it is not more generally in Section 3.1, but the graph is
clear that any triad is more deserving than any now either the rectilinear grid of Weber space
other of being called “tonic” (or “dominant” or or the hexagonal grid of neo-​Riemannian triad
“subdominant”), but a drawback in other cases space. We call these two distance measures the
in which a clearer tonal orientation prevails. Weber distance and the neo-​Riemannian distance
In response to some of these shortcomings, (or PLR distance), respectively. The Weber dis-
Lerdahl refines his regional space (essentially tance between two triads is the smallest number
Weber space) into a more complex chordal-​ of P, R, and T7 (or T5) operations, in any combi-
regional space, presented in Section 4.3 below. nation, that will map one to the other; the neo-​
Transformation theory has devised various ways Riemannian distance is the smallest number of
to confront the other limitations, some of which P, L, and R operations. For example, the Weber
we will consider in later chapters. While tonnetz distance from C major to A♭ major is 3 because, at
analysis continues to find its most appealing appli- the very least, some combination of one vertical
cations in passages displaying certain characteris- and two horizontal moves is required to get from
tic patterns, and generally remains insufficient for C to A♭ in Figure 4.2.3 (for instance, C–​F–​f–​A♭).
detailed analysis of complete works, tonnetz anal- The neo-​Riemannian distance between the same
ysis and related methods can often be customized two triads is only 2, however, because the com-
for use in conjunction with other methods, some- bination of a single P and L leads from C to A ♭.
times in unexpectedly diverse musical situations Table 4.2.16 compares these distance mea-
and with compelling results. The last section of sures with the circle-​of-​fifths distance defined
this chapter will offer a few illustrative analyses. in Section 4.1, showing the distance from a
C-​major triad to every other triad by all three mea-
—​—​—​ sures. There are other alternatives: Krumhansl’s

Table 4.2.16. Circle-​of-​fifths distance, Weber distance, and neo-​Riemannian distance

from C major to every major and minor key

C IRC L E -​OF -​FIF T H S W E B E R DI S TA N C E F ROM NEO-​R IE M A NNI A N

DI S TA N C E F ROM C M A J OR T O … DI S TA N C E F ROM
C M A J OR T O … C M A JOR T O …

C major 0 C minor 3 C major 0 C minor 1 C major 0 C minor 1

D♭ major 5 C♯ minor 4 D♭ major 4 C♯ minor 4 D♭ major 4 C♯ minor 3
D major 2 D minor 1 D major 2 D minor 2 D major 4 D minor 3
E♭ major 3 D♯ minor 6 E♭ major 2 D♯ minor 3 E♭ major 2 D♯ minor 3
E major 4 E minor 1 E major 3 E minor 2 E major 2 E minor 1
F major 1 F minor 4 F major 1 F minor 2 F major 2 F minor 3
F♯ major 6 F♯ minor 3 F♯ major 4 F♯ minor 3 F♯ major 4 F♯ minor 3
G major 1 G minor 2 G major 1 G minor 2 G major 2 G minor 3
A♭ major 4 G♯ minor 5 A♭ major 3 G♯ minor 4 A♭ major 2 G♯ minor 3
A major 3 A minor 1 A major 2 A minor 1 A major 2 A minor 1
B♭ major 2 B♭ minor 5 B♭ major 2 B♭ minor 3 B♭ major 4 B♭ minor 5
B major 5 B minor 2 B major 4 B minor 3 B major 4 B minor 3

136 • E x p l o r i n g M u s i c a l S p a ces
empirical studies have already been mentioned, Weber distance from E♭ major to G minor;
Lerdahl (2001) has developed some consider- and the neo-​Riemannian distance from A♭
ably more elaborate distance measures between major to B minor.38
chords and keys, and we will study distance (e) In Table 4.2.16, the first column of
measurements in a more general setting in numbers, excluding the initial 0, forms a
Chapter 12. But the Weber distance in particu- symmetrical pattern (52341614325). Why
lar is appealingly simple and aligns well with is this? Does a similar pattern hold for the
many musicians’ and theorists’ understandings minor keys? Do the tables of Weber and
of key relationships. In particular, admitting the neo-​Riemannian distances have similar
parallel relationship as one way to achieve close- properties?39
ness rectifies the shortcomings of the circle-​of-​ (f) Because Table 4.2.16 gives only distances
fifths distance, noted at the beginning of this from C major, it cannot be used to calculate
section: by either the Weber distance or the distances between two minor keys.
neo-​Riemannian distance, C major is no longer Construct analogous tables giving the
as close to F♯ minor as it is to C minor. In light double-​circle, Weber, and neo-​Riemannian
of the central role of third relations in the con- distances from C minor to every other
struction of Weber and Riemann space, it should   minor key.
be unsurprising that third-​related keys, too, are
generally closer by either the Weber distance or Distantly related keys, like those consid-
the neo-​Riemannian distance than by the circle-​ ered in part (b) of the preceding exercise, have
of-​fifths distance—​a feature exploited often by another distinctive characteristic: in Weber or
Schubert, as analyses in Section 4.5 will show.35 Riemann space, such keys may often be joined
by several different paths of equal or nearly
Exercise 4.2.17 equal length, paths that may lead in different
(a) Compare the circle-​of-​fifths, Weber, and directions and pass through entirely different
neo-​Riemannian distances from C major regions of the space. For example, the March to
to D minor, from C major to E major, and the Scaffold from Berlioz’s Symphonie fantastique
from C major to F minor. Does one way includes a notorious juxtaposition of the dis-
of measuring distance seem to correlate tant keys of G minor and D♭ major. Figure 4.2.18
best with your intuitions about the tonal shows that in Weber space one can move from G
distances between these keys? Is your minor to D♭ major in three very different direc-
intuition different if you think of distances tions. The horizontal Path 1 locates the two keys
between chords rather than keys?
(b) What is the greatest circle-​of-​fifths distance
from C major to any other key? How
many keys are located at this maximum
distance from C major? Answer the
analogous questions for the Weber and neo-​
Riemannian distances.36
(c) The neo-​Riemannian distance from C major
to every major key is an even number, and
the neo-​Riemannian distance to every
minor key is an odd number. Why is this?
How is Exercise 4.2.6a related to this
property? Why does this property not hold
for the other two distance measures?37
(d) Table 4.2.16 may be used in conjunction
with transposition to determine distances
between other pairs of keys. For example,
the Weber distance from A major to F minor
may be found by transposing both keys up
three semitones to C major and G♯ minor,
and consulting the table for the answer (4). FIGURE 4.2.18 Ambiguous relationship between
Use this method to determine the circle-​of-​ G minor and D♭ major in Berlioz, Symphonie
fifths distance from F major to B major; the fantastique, IV, mm. 152–​58

Chordal, Tonal, and Serial Spaces • 137

three steps apart on a PR-​chain; that is, it identi- is not dependent on knowing which triads are of
fies D♭ major as “the relative of the parallel of the which qualities.
relative of G minor.” Path 2 is nearly vertical, and If the vertices of the generic tonnetz repre-
involves fifth relations: D♭ is now “the relative of sent triads, then the edges represent possible
the subdominant of the subdominant of the motions from one triad to another through
subdominant of G minor” (or if diagonal moves various intervals. As the legend beneath Figure
by L are admitted, it could be “the leittonwech- 4.3.1a shows, the graph accounts for all pos-
sel of the subdominant of the subdominant”). sible generic intervals, arranged along three
Finally, the northwesterly Path 3 describes D♭ axes: the diagonal seconds axis shows seconds
as “the major dominant of the leittonwechsel of and sevenths, the horizontal thirds axis shows
the major dominant of G minor.” The fact that all thirds and sixths, and the vertical fifths axis
of these descriptions are about equally complex shows fourths and fifths. Every generic triad is
could be said to contribute to a listener’s sense related to every other by one of these intervals,
of disorientation upon hearing this unortho- and therefore every one of the seven vertices in
dox pairing of keys: there are no musical clues the graph is adjacent to every other. The finite
to signal “which way we are going” when the graph of diatonic chordal space in (b) is actually
music lurches to D♭. When keys are separated a complete graph on seven vertices, which we called
by shorter distances there is no such ambiguity: K7 in Chapter 3. K7 is not a planar graph (recall
there is only one simple way to get from G minor from Exercise 3.3.15 that even the simpler graph
to D minor or to B♭ major. K5 cannot be embedded in a plane), but Figure
4.3.1b is a toroidal embedding of K7.40
4.3 GENERIC AND
DIATONIC CHORDAL Exercise 4.3.2
SPACES (a) The graph in Figure 4.3.1b is a complete
graph as described above. At a glance,
Fewer chords are available in generic or diatonic however, it might appear that this graph
space than in chromatic space, but notwith- cannot possibly be complete, because the
standing their limited numbers, generic chords diagonals are missing; only the opposing
form spaces in their own right. Figure 4.3.1a is diagonals are present. Explain this
the triangulated generic tonnetz from Figure apparent paradox.
1.4.28b. Originally the vertices in this graph rep- (b) In Figure 4.3.1b, trace the patterns formed
resented single notes—​points in generic pitch-​ by a generic (or diatonic) scale, circle
class space. The same graph can be repurposed of fifths, and circle of thirds. In other
as a chordal space by letting the letters stand for words, identify subgraphs of Figure 4.3.1b
triads constructed on those notes. There is only corresponding to Figures 1.2.8, 1.3.8, and
one triad on any note in generic space, so there is 1.3.12. Each graph should be a 7-​cycle.
no need to differentiate major and minor triads (c) As a graph embedded in a torus, how many
on the same roots, as we did in chromatic space. vertices, edges, and faces does Figure 4.3.1b
Generic triad space (gtriad) is therefore a space have? As in Exercises 4.1.17b, 4.2.4b, and
of only seven objects, isomorphic in at least some 4.2.6b, verify that these numbers satisfy
aspects of its structure to generic pitch-​class   the toroidal version of Euler’s formula.
space. As a finite tonnetz, the toroidal graph
in Figure 4.3.1a has only seven vertices, each The generic tonnetz, like its chromatic coun-
of which appears once in every parallelogram terpart, may be regarded as a transformation
tile. Figure 4.3.1b shows one of these tiles, after network. The appropriate transformations are
applying a shearing transformation that con- the generic transposition operators tk, first intro-
verts the tile into a rectangle (altering the slopes duced in Section 2.6 and shown in the legend
of the graph’s edges but preserving the adjacency beneath Figure 4.3.1a. Though often over-
relation). Like the generic spaces in Chapter 1, looked, generic (or diatonic) transpositions
Figure 4.3.1 may also readily represent a diatonic are extremely useful operations; the relation-
tonnetz, determined by a fixed assignment of ships they model are ubiquitous in virtually
sharps or flats to the seven generic pitch classes. all music in which diatonic structure plays a
In a diatonic context, three of the seven triads role.41 Generic tk, like chromatic Tn, may act
will be major, three minor, and one diminished, either on single notes (in this case generic pitch
but the organization of the generic chordal space classes) or on chords (gpc sets). The operator tk

138 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 4.3.1 (a) A generic tonnetz as a representation of generic triad space (gtriad); (b) one tile
of the generic tonnetz

transposes a note or chord up through k steps are related to each other by some tk. In chromatic
within its generic or diatonic space. For exam- space, triads differing in quality are not related
ple, in generic pc space gpc, t1(G) =​A, t5(A) =​ by Tn—​which is precisely why other transforma-
F, and t3 applied to an A triad (A–​C–​E) yields a tions such as P, L, and R are needed in order to
D triad (D–​F–​A). In three-​flat diatonic pc space describe the chromatic tonnetz as a transforma-
dpc(−3) (“the E♭-​major scale”), t1(G) =​A♭, t5(A♭) =​ tion network.
F, and t3 applied to an A♭-​major triad yields a D-​ As Figure 4.3.1 suggests, the transpositions
diminished triad. Mathematically, tk is a mod-​7 t1 and t6 are inverses (comparable to T1 and T11
operator, whose effect is to add k to any number in chromatic space), as are t2 and t5, and also t3
mod 7, just as Tn adds n mod 12. Using generic and t4. Inverse functions relate the same pairs of
pitch-​class numbers with C =​0, the relationship elements, exchanging their places: thus t2(E) =​G
t5(A) =​F is represented by the congruence 5 +​5 and t5(G) =​E. A transposition by t2 in a generic or
=​3 (mod 7): the first 5 is the gpc number for A, diatonic pitch-​class space may represent either
the second 5 is the interval of generic transposi- an ascending third or a descending sixth in pitch
tion, and 3 is the gpc number for F. (Recall from space, while t5 can be either an ascending sixth
Section 1.1 the relationship between generic or a descending third; similar remarks apply to
intervals and familiar interval names; t5 corre- the pairs t1/​t6 and t3/​t4. We shall have more to
sponds to transposition through the interval of say about ways of measuring intervals, and their
an ascending sixth, not a fifth.) relationships with transposition operators in
The idea that a diminished triad may be a general, in Chapter 7.
“transposition” of a major triad may strike some
readers as counterintuitive, but distinctions of Exercise 4.3.3 Express your answers to the
chord quality are external to the generic space following calculations both in musical terms
in which tk operates. Generic intervals are mea- and using generic pitch-​class numbers.
sured by counting steps within that space, not by
acoustic interval size. By this measure, all triads (a) In generic space gpc, calculate t4(F), t5(B),
in generic space have the same structure, and all and t6 applied to an E triad.

Chordal, Tonal, and Serial Spaces • 139

(b) In four-​sharp diatonic space dpc(+​4) Comparison of the sequence networks of
(“E major”), calculate t3(A), t5(D♯), and t6 Figure 4.3.4 with those of Figure 3.5.3 is instruc-
  applied to a C♯-​minor triad. tive. The present sequences are diatonic rather
than chromatic: the graph in 3.5.3a is a T-​net
Diatonic chord progressions may be mapped while the new graphs are t-​nets. The diatonic
in the generic tonnetz in much the same way sequences follow regular patterns in generic
that we traced some chromatic progressions in space but not in chromatic space: the specific
the tonnetz of Figure 4.2.8. Figure 4.3.4 shows intervals in 4.3.4a include one diminished fifth,
two common diatonic sequence patterns. The which must always be present in sufficiently
descending-​fifths sequence in (a) is generated long sequences of this type, and the t1 steps in
by repeated application of the generic transposi- 4.3.4b include both major and minor seconds. A
tion t3, a descending fifth or ascending fourth, difference in the manner of graphical presenta-
and accordingly traces a straight vertical path tion may also be noted: in terminology intro-
along the fifths axis of the generic tonnetz. duced previously in conjunction with Figure
The sequence in (b) is the type familiar from 3.5.15, Figure 3.5.3a is an event network (orga-
Pachelbel’s Canon, also known as the Romanesca nized chronologically from left to right), while
pattern; it consists of alternating t4 and t1 trans- the graphs in Figure 4.3.4 are spatial networks
positions, descending fourths and ascending (arranged like the diatonic tonnetz).
steps, together forming a westbound zigzag path The dashed arrows in Figure 4.3.4, like those
on the tonnetz. in 3.5.3a, represent indirect transformations

FIGURE 4.3.4 Two diatonic sequences in the generic tonnetz: (a) a descending-​fifths sequence (t3t3 =​ t6);
(b) a “Pachelbel” sequence (t4t1 =​ t5)

140 • E x p l o r i n g M u s i c a l S p a ces
arising as composites of two consecutive direct chords A♭–​D♭–​F–​b♭–​D♭–​G♭. These triads do
transformations. The combined effect of t4 and t1 not all belong to one diatonic space (the F-​
is t5 (that is, t4t1 =​ t5), because of course 4 +​1 =​5 major chord is a secondary dominant), but
(mod 7). The two-​chord blocks in the Pachelbel we may nevertheless analyze the sequence
sequence accordingly descend by thirds (t5)—​ in generic space as A–​D–​F–​B–​D–​G. Trace
this sequence is often called a “descending-​ this sequence in the generic tonnetz
thirds sequence”—​ and this explains why the and write the corresponding equation of
trajectory of the sequence as a whole parallels transposition operators.
the thirds axis of the tonnetz. The equation (c) Do the same with the sequence that opens
t4t1 =​ t5 may be compared with the notation Fauré’s Pavane, Op. 50: f♯–​D–​E–​c♯–​D–​b–​C♯.
D3(−4/​+​2) for the same sequence in Laitz 2016 (d) Unusual sequence types can sometimes be
(414–​16). Laitz’s notation may be read as “a described as alterations of more familiar
descending-​thirds sequence with root motions types, in which some chords are substituted
of descending fourths and ascending seconds.” for others a third away (with which they
The use of familiar ordinal interval numbers here therefore share two common tones). For
obscures the simplicity of the arithmetic, for −4 example, every other chord of a sequence
and +​2 do not add up to −3. Similarly, Laitz’s may be transposed by t2 or t5. Can either
notation for the descending-​fifths sequence in or both of the sequences in (b) and (c) be
(a) is D2(−5/​+​4), recognizing the fact that such described in relation to a more common
sequences are usually organized in two-​chord type in this way? How does the tonnetz
blocks descending by step; in the language of picture of the unusual sequence resemble
diatonic transposition operators this sequence that of the more familiar one?43
pattern is t3t3 =​ t6. The tonnetz analysis disre- (e) With your answers to (c) and (d) in mind,
gards the distinction between descending fifths analyze the following two passages from
and ascending fourths (both are t3), and also dis- Ravel’s Pavane pour une infante défunte:
regards chord inversions, which are sometimes
used to distinguish various subtypes of these mm. 3–​6: C–​a–​D–​G–​C–​f♯ø–​b
sequence patterns. 42
mm. 30–​34: C–​a–​b–​G–​a–​f♯ø–​b

Exercise 4.3.5 Both progressions harmonize the same

(a) Another common sequence type is the melody. Most of the chords are diatonic
“ascending 5–​6” sequence with chord roots seventh chords, but the sevenths may be
C–​A–​D–​B–​E–​C–​⋯, labeled A2(−3/​+​4) by   ignored for the purposes of this analysis.
Laitz. Trace this sequence as a path in the
generic tonnetz, and write the corresponding In a tonal context, the seven triads in the dia-
equation involving a composition of generic tonic tonnetz may be assigned roman-​numeral
transposition operators. (This sequence was labels. Figure 4.3.6a, otherwise arranged like
graphed in a different way in Figure 2.3.4a.) Figure 4.3.1a, displays the labels associated
(b) Schubert’s Scherzo in D♭ Major for piano, with a major tonic key, and 4.3.6b shows cor-
D. 593, No. 2, contains at mm. 5–​7 a responding labels for a minor key. The mode-​
sequence of a less common type, with neutral labeling in 4.3.6c is occasionally useful

FIGURE 4.3.6 (a) Diatonic triad space in a major key; (b) diatonic triad space in a minor key; (c) a
mode-​neutral representation of generic triad space

Chordal, Tonal, and Serial Spaces • 141

FIGURE 4.3.7 Chordal-​regional space (after Lerdahl)

in describing likenesses between passages that

share generic structure while differing in chro-
matic detail; we will encounter an illustration
in Figure 4.5.15 below. These diagrams corre-
spond to Lerdahl’s chordal space; one module
of such a tonnetz is what Lerdahl terms the
chordal core.44
This core forms an integral part of Lerdahl’s
chordal-​regional space, a portion of which is
shown in Figure 4.3.7. At the large-​scale level,
indicated by the boldface key names in the fig-
ure, chordal-​regional space is a map of key areas,
organized like Weber space (Figure 4.2.3). But
each key area is now not just a single point of
the space but a region with its own cluster of
satellites, a miniature representation of chordal FIGURE 4.3.8 Functional harmonic progres-
space in that key, corresponding to Figure 4.3.6a sions in the diatonic circle of thirds
or 4.3.6b. A point in chordal-​regional space may
be specified by naming one of the 24 keys and
one of the seven diatonic triads within that key. long as key areas can be identified unequivo-
Points of chordal-​regional space are therefore cally, chordal-​regional space enables more fine-​
in one-​to-​one correspondence with elements of grained analysis than Weber space or a tonnetz
the Cartesian product of Weber space and dia- by itself. Lerdahl supplies several such chordal-​
tonic triad space, a set of cardinality 24 ∙ 7 =​ regional analyses, and we will see an application
168. Lerdahl’s geometric arrangement, however, in Example 4.5.12.46
imposes an additional hierarchical organization Figure 4.3.8 diagrams the thirds axis of dia-
on chordal-​regional space not implied by its set-​ tonic triad space in circular form; the arrange-
theoretic description.45 ment is like Figure 1.3.12’s generic circle of
This space addresses a previously noted short- thirds but with roman numerals in place of the
coming of Weber space by differentiating among note names. The organization by thirds brings
appearances of the same triad in the context of together chords of the common functional cat-
different key areas: for example, a tonic triad in egories: iii and vi, adjacent to I, can in some
C major is indicated by the capital letter C at the circumstances extend or substitute for a tonic
center of its region, while a dominant triad in F harmony, while ii, IV, and vi commonly serve
major is V in the F-​major region and a mediant a pre-​dominant function and iii and viio, adja-
triad in A minor is III in the A-​minor region. As cent to V, can take on aspects of the dominant.

142 • E x p l o r i n g M u s i c a l S p a ces
T–​P–​D–​T motion is directed counterclockwise twisted, forming a Möbius strip. Why is it
in this circle, and in fact this figure offers a not necessary to conjoin the two vertical
refinement of the standard T–​P–​D–​T harmonic edges in this module? You can construct
paradigm. It is sometimes noted that func- this Möbius strip physically using a strip
tional harmonies tend to progress primarily by of paper or transparent plastic, labeling
three types of root motion: by descending fifth, the seven pitch classes along the edge of
by ascending second, or by descending third.47 the strip and drawing lines across the strip
In Figure 4.3.8, a descending third is one posi- to show the triangulation. The triangles
tion counterclockwise, a descending fifth is two represent diatonic triads; in what order do
positions counterclockwise, and an ascending they fall along the strip? A Möbius strip has
step is three positions counterclockwise. We only one edge; in what order do the pitch
may therefore say briefly that functional har- classes occur along this edge?49
monic progressions tend to move counterclock- (b) This Möbius strip may be used to model
wise in the circle of thirds—​a more detailed rule a chain of triads descending diatonically
than the coarser T–​P–​D–​T principle, but more by thirds. Each triad is represented by a
straightforward than elaborate networks of triangle, each note by a vertex; successive
arrows such as Figure 3.4.1. (As with the T–​P–​ triads share two common tones. If you
D–​T paradigm, the most common exceptions to attempt a similar construction in which
the counterclockwise rule occur in movement triangles represent the triads of a PL-​cycle
away from the tonic.)48 in chromatic space such as C–​c–​A♭–​g♯–​E–​
e–​C, you will get not a Möbius strip but
4.4 SOME ADDITIONAL an untwisted cylindrical loop. Why is this?
MODELS What determines whether a closed loop of
triangles will or will not require a twist?
This section consists mostly of a series of exer- Which do you get using the triads of a PR-​
cises, outlining a few other ways that tonal or   cycle? An RL-​cycle?
chordal structures may be depicted.
Exercise 4.4.2 Like many graphs, neo-​
Exercise 4.4.1 This exercise involves a Riemannian triad space can be drawn in
representation of generic space as a Möbius a number of strikingly different ways; the
strip. You may find it helpful to review the hexagonal grid of Figure 4.2.5b is only one of
construction of a different Möbius-​strip several viable representations. This exercise
tonnetz in Section 1.4. (We will see another explores two other possible ways to arrange 24
application of that tonnetz in Figure 4.5.8.) vertices representing triads or key areas and
to connect them with edges representing P,
(a) Start with the generic tonnetz from Figure L, and R. You may find it helpful to use three
1.4.28a, in which the note names represent different colors for the edges in these graphs, as
generic pitch classes and triads appear as suggested in Exercise 4.2.9.50
triangles. Apply a shearing transformation
to the columns, sliding each column up (a) Arrange the 24 triads or keys in a circular
slightly relative to the column to its left, so RL-​cycle, then add additional edges crossing
that each triangle has one vertical edge and the interior of the circle as needed for
two oppositely sloping edges. In this new P. Riemann space is not a planar graph,
configuration E, rather than appearing to so some of the P edges will need to cross
the right of C and to the left of G, should each other.
appear both to the left and to the right of (b) Another possibility is to arrange the
the midpoint of the vertical edge joining vertices of Riemann space in four concentric
C and G. In this figure you should be able circles, each of which is a PL-​cycle. Put
to draw rectangular modules, taller than adjacent cycles such as C–​E–​A♭ and D♭–​F–​A
they are wide, each containing every note next to each other. Additional edges will
name exactly once. Horizontally adjacent be needed for R; most of these edges will
modules should be identical, but vertically connect two adjacent circles, but some of
adjacent modules are mirror images of each them will have to connect the innermost
other, so the ends of the rectangle can be circle to the outermost, cutting across the
stitched together only if the rectangle is two intermediate circles. Align the vertices

Chordal, Tonal, and Serial Spaces • 143

 in the four circles in a way that maximizes as in the preceding exercise, and label the
the symmetry of the resulting graph and appropriate transformations P̂, L̂, and R̂.52
minimizes edge crossings. The ringlike (b) Although the 013 tonnetz bears a
appearance of this graph befits its status as superficial resemblance to the usual 037
a toroidal embedding: it can be seen as the tonnetz, it differs in important ways,
top view of an embedding of Riemann space because the pitch classes repeat in a
  in the surface of a doughnut. different pattern (as identified in Exercise
1.4.17). Trace a P̂L̂-​cycle, a P̂R̂-​cycle, and an
The next several exercises explore ways of L̂R̂-​cycle in this tonnetz and compare them
using tonnetzes to model relationships among with their triadic counterparts.
types of chords other than simple triads. (c) Trace the following cycle of 013 trichords in
the 013 tonnetz:
Exercise 4.4.3
(a) Review the tonnetz of 014 trichords from {0,1,3} → {0,2,3} → {2,3,5} → {2,4,5}
Exercise 1.4.9, based on an alternate → {1,2,4} → {1,3,4} → {0,1,3}
triangulation of the usual tonnetz.
Construct the dual graph of this 014 Analyze this trichord progression using P̂,
tonnetz, overlaid on the original graph in L̂, and R̂. Does the corresponding string of
the manner of Figure 4.2.8, and label each the usual triadic transformations P, L, and
vertex of the dual graph with an appropriate   R generate a cycle?
trichord.
(b) The relationships among the 014 trichords In the above examples, tonnetz structures
in this graph are in some ways similar to have consistently modeled spaces of three-​note
relationships among triads in the usual chords—​because, of course, a triangle has three
037 tonnetz. Label the edges of your dual vertices. It is reasonable to wonder whether
graph with three transformations P*, L*, similar diagrams can somehow model spaces of
and R* by analogy with P, L, and R. Choose chords of other cardinalities. The following exer-
the labels so that, for example, P*, like P, cises explore two tonnetz models for the space
represents motion across a diagonal edge of of 0258 tetrachords (dominant and half-​dimin-
the background graph. Trace a P*L*-​cycle, a ished seventh chords). One model is two-​dimen-
P*R*-​cycle, and an L*R*-​cycle in this tonnetz sional, the other three-​dimensional. Neither is
and compare them with their triadic entirely satisfactory as a representation of this
counterparts. Do these cycles have the same space, though the three-​dimensional model is in
lengths as those in the usual tonnetz? Do some ways more revealing. In Part Three we will
the P*L*-​ and P*R*-​cycles use the notes of consider techniques by which spaces of chords of
hexatonic and octatonic collections? any cardinality may be constructed.
(c) Though they appear similar in the tonnetz
and satisfy many of the same algebraic Exercise 4.4.5 If a trichord is represented
relationships, P*, L*, and R* differ from the by arranging three pitch classes in a triangle,
Riemannian triadic transformations P, L, one might suppose that a tetrachord may be
and R in their voice-​leading characteristics. represented by a square.
This difference should be apparent if you
write chord progressions corresponding to (a) Draw a square whose vertices (clockwise
the cycles you traced in (b). The usual P and from the lower left) are labeled 0, 7,
L move one voice by semitone while R moves 4, and 10—​the pitch-​class numbers
one voice by a whole tone. Only one of P*, L*, of the C7 chord. (Other arrangements
or R* is comparably smooth; which is it?51 of the same four pitch classes around
(d) In Exercise 1.4.25 you traced paths formed a square are possible; the indicated
by trichords in Schoenberg’s “Nacht” in the arrangement is recommended, but feel
014 tonnetz. Describe these paths using P*, free to experiment with others.) Add
  L*, and R*. other vertices surrounding these four
so that the four squares adjacent to this
Exercise 4.4.4 one represent the chords aø7, c♯ø7, eø7,
(a) Review the tonnetz of 013 trichords from and cø7, each of which shares at least two
Exercise 1.4.17, construct its dual graph pitch classes with C7. Arrange the added

144 • E x p l o r i n g M u s i c a l S p a ces
 vertices in the most consistent possible positioned parallel to layer 0 so that B♭ in layer 1
way (two different intervals will alternate lies above the center of the C–​E–​G triangle in
along the horizontal axis, and likewise layer 0. The C7 chord is thus represented not by a
along the vertical axis), and expand the polygon in the plane but by a tetrahedron, a solid
graph to include additional dominant and three-​dimensional figure with four vertices and
half-​diminished seventh chords, as many four triangular faces (the C–​E–​G base and three
as necessary so that the pattern becomes sloping faces). The distance between the two lay-
clear. The graph is a grid of squares, not ers may be adjusted so that the tetrahedron is
a triangulation, and the dual graph is regular, all its faces being equilateral triangles
another such grid. Label the vertices of the same size. All the other notes in layer 1
of the dual graph with the appropriate are correctly positioned to combine with the
chord names. major triads in layer 0 to form dominant seventh
(b) Discuss the shortcomings of this graph as chords in the same way.
a representation of 0258 tetrachord space. Consider the G-​minor triad represented by
How many other 0258 chords share at least the triangle G–​ B♭–​
D in layer 1. This triangle
two notes with C7? How many of these are perches atop the peaks of the tetrahedra rep-
adjacent to C7 in the dual graph, and how resenting the A7, C7, and E7 chords. The note
far away are the others? How many notes E belongs to all three of these chords, and the
are actually shared by the C7 and eø7 chords, E in layer 0, directly beneath the center of this
and in what way does the graph fail to triangle, combines with the three notes of the
  represent this? G-​minor triad to form an eø7 chord. Half-​dimin-
ished seventh chords are represented in this
A somewhat more satisfactory version of tonnetz by inverted tetrahedra, each comprising
0258 chordal space is a three-​dimensional ton- three vertices from layer 1 and one from layer 0,
netz first studied by Edward Gollin (1998).53 The as shown in Figure 4.4.6b. The layers of vertices
first stage in the construction of such a tonnetz in (b) are the same as those in (a); one should
is illustrated in Figure 4.4.6a. In layer 0 is a copy imagine the two figures superimposed, so that
of the triangulated tonnetz from Figure 1.4.10b, the inverted tetrahedra in (b) fit in between the
with equilateral triangles, now viewed obliquely. original tetrahedra in (a). The eø7 tetrahedron
Layer 1 contains a second copy of the same graph, shares two vertices and one edge with each of the

FIGURE 4.4.6 (a) Major-​minor seventh chords and (b) half-​diminished seventh chords in a three-​
dimensional tonnetz (after Gollin)

Chordal, Tonal, and Serial Spaces • 145

A7, C7, and E7 tetrahedra. All 0258 tetrachords Does another tetrahedron representing this
are represented as one type of tetrahedron or same chord adjoin the C7 tetrahedron at a
the other. (These tetrahedra do not fill up all the different vertex?
space between the two layers; three-​dimensional (d) Altogether, how many different tetrahedra
space cannot be “tiled” with regular tetrahedra representing dominant and half-​diminished
in the way that the plane is tiled by equilateral seventh chords meet at each vertex? List all
triangles.) 0258 chords that share one common tone
All dominant and half-​ diminished seventh with C7, and verify that all of these chords
chords appear between layers 0 and 1, but in are represented by tetrahedra that meet the
principle the construction can continue with C7 tetrahedron at one vertex.56
additional layers upward (layers 2, 3, …) and (e) The vertices in layer 1 are positioned above
downward (layers −1, −2, …), forming a pattern the centers of the triangles in layer 0, not
that repeats itself in all three dimensions. If the directly above the vertices in that layer. The
various vertices labeled with the same pitch class first layer above layer 0 whose vertices line
are conjoined, this conformed three-​ dimen- up precisely above the vertices of layer 0 is
sional tonnetz consists of many repetitions layer 3. What note in layer 3 falls directly
of one small finite block, a three-​dimensional above C in layer 0? What is the first layer
counterpart of the rectangular tile of the usual above layer 0 whose notes line up exactly
two-​dimensional tonnetz. The appropriate topo-   over the same notes in layer 0?57
logical conception is the 3-​torus 𝕋3, a solid analog
of the two-​dimensional surface that is familiar Elaine Chew has devised a model she calls the
to us as the 2-​torus 𝕋2. spiral array for representing an assortment of
tonal entities.58 As shown in Figure 4.4.8a, the
Exercise 4.4.7 This exercise requires visualizing spiral array may be constructed by bending the
relationships in the three-​dimensional tonnetz. line of fifths into a helix. The resulting configu-
You may find it helpful to make some additional ration resembles Figure 1.2.2, but the notes in
drawings, extending parts of Figure 4.4.6 to that figure were not arranged by fifths. The spi-
include other nearby points and tetrahedra, ral array is more closely akin to the two spirals
especially in layers 2 and −1. of fifths suggested in Exercise 1.3.10, but it is
not identical to either of those; the distinguish-
(a) For each of the six edges of the tetrahedron ing feature of this representation is that notes
representing the C7 chord, identify an related by major thirds—​ every fourth note
inverted tetrahedron representing a half-​ along the line of fifths—​are aligned, thus bring-
diminished seventh chord that shares ing the notes of any major or minor triad into
this edge, and therefore shares the proximity on two successive coils of the helix.
corresponding two notes with the C7 chord. As Figure 4.4.8b illustrates, the spiral array may
One of these tetrahedra is the eø7 chord also be obtained by wrapping a tonnetz around
shown in 4.4.6b, which shares the E–​B♭ edge. a cylinder in a certain way, conjoining the two
Two other such tetrahedra also lie between vertical boundaries of the figure.
layers 0 and 1, while the rest lie below layer The triangles representing F-​major, C-​major,
0, formed by notes in layers 0 and −1.54 and G-​major triads are shaded in Figure 4.4.8a.
(b) Among the half-​diminished seventh chords Chew calculates the positions of these triangles
that share two or more pitch classes with in three-​dimensional space, and shows how each
C7, one chord is not accounted for among triangle may be represented by a single point,
those in (a) above. Which chord is this? calculated as the centroid (center of gravity) of
Although no tetrahedron representing the positions of the three vertices. Because these
this chord shares an edge with the C7 three chords are the primary triads in the key of
tetrahedron, you should find that two C major and together include all the notes of the
different tetrahedra representing this chord C-​major scale, the key of C major is then repre-
meet the C7 tetrahedron at two different sented in turn as the centroid of the points repre-
vertices.55 senting the three chords—​a point in the interior
(c) One of the six chords you identified in of the spiral. The calculations for a minor key are
(a) has three common tones with C7, even more complex, involving weighted centroids of i,
though the tetrahedra share only two iv, IV, v, and V chords (all of which are required
common vertices. Which chord is this? in order to accommodate the notes in all forms

146 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 4.4.8 (a) The spiral array (after Chew); (b) construction of the spiral array by wrapping a ton-
netz around a cylinder

of the minor scale). Chew presents spiral-​array–​ as we move down. Under the assumption of
based algorithms for determining the key of a enharmonic equivalence, however, it is possible
musical passage, for identifying key boundaries to conform the space in this dimension as well.
(modulations), and for pitch-​spelling. Doing so requires wrapping the vertical axis and
cylindrical shell of the spiral array around on
Exercise 4.4.9 Dominant and half-​diminished themselves; this may be realized in a doughnut-​
seventh chords take the form of tetrahedra shaped torus resembling Figure 1.4.4. The line
in the spiral array, but not regular tetrahedra. of fifths becomes a circle of fifths, but a wavy
Draw the tetrahedron representing the C7 one, not perfectly circular in shape. Draw a
chord; observe that in three dimensions, the picture of this doughnut-​shaped space, showing
edge connecting B♭ and E is longer than the the positions of all twelve pitch classes and the
edge connecting C and E. Also draw several of circle
   of fifths.
the half-​diminished seventh chords that share
two pitch classes with C7, and the one half-​
diminished seventh chord that shares three
pitch classes with C7. Recall from Exercise
4.5 ANALYTICAL EXAMPLES
4.4.7c that in the three-​dimensional tonnetz, We conclude this chapter with a series of ana-
the tetrahedra representing chords with lytical examples, illustrating a variety of applica-
three common pitch classes did not share a tions of tonal, chordal, and serial spaces. Readers
triangular face, only an edge. Do the tetrahedra should note the manner in which different spa-
representing those chords in the spiral array tial models are chosen, and sometimes modified,
share
   a triangular face? for use in different musical situations. It is hoped
that these examples will demonstrate the poten-
Exercise 4.4.10 The spiral array as described tial for the tonnetz and related spaces to illu-
above is a partially conformed structure, minate aspects of musical structure that might
obtained by wrapping a tonnetz only in one otherwise have been overlooked. Of course, the
dimension. In the vertical dimension of brief discussions here should not be mistaken
Figure 4.4.8a, the spiral array may be left for complete analyses of the passages being
unconformed, like the original line of fifths, examined; readers are encouraged to consult the
so that ever-​sharper pitch classes appear as we scores and supplement the analyses with further
move up the helix and ever-​flatter pitch classes details obtained by other appropriate methods.

Chordal, Tonal, and Serial Spaces • 147

Example 4.5.1: Webern, String Quartet, an unordered set, that pitch-​class content is the
Op. 28, I chromatic tetrachord {n − 1, n, n +​1, n +​2}. For
Webern’s Op. 28 quartet offers an opportunity convenience we call n the root of the two BACH
to devise a customized variant of the row space forms n+​ and n−; of course no association with
of Figure 4.1.12a. The work is based on the other meanings of the word root is implied. The
“derived row” shown in Figure 4.5.2, comprising complete twelve-​tone row consists of the three
three successive forms of a B–​A–​C–​H tetrachord BACH forms 7+​, 11−, and 3+​. The successive T4
(to be called “BACH forms” for short). The transpositions of the roots, and therefore of the
quartet is composed entirely of BACH forms, chromatic tetrachords, ensure that the three
and for reasons explored below, an analysis BACH forms complete a twelve-​tone aggregate.
using BACH forms is in many ways more The 24 possible BACH forms are readily
straightforward than an analysis using complete organized in the double-​circle space shown in
twelve-​tone rows. To this end we shall construct Figure 4.5.3a. This space is generated by the
a space of BACH forms. semitone transposition T1, which produces
The BACH motive, like the entire row, has motion around the circles, and the retrograde
RI symmetry: any RI form of the motive is operation R, which toggles between circles.
identical to some P form. There are therefore Because of the RI symmetry of BACH forms,
only two types of BACH forms to consider: P R is in this case equivalent to a contextual
=​ RI forms such as B♭–​A–​C–​B♮, and R =​ I forms inversion, specifically the inversion that
such as B♮–​C–​A–​B♭. As a shorthand labeling for preserves the pitch-​class content of the
the 24 BACH forms, we write n+​ for the P form tetrachord.
whose first note is pitch class n, and n− for the In mm. 1–​12 of the quartet’s first movement,
retrograde of n+​, the R form whose last note Webern presents nine BACH forms in the order
is n. In this notation, B♭–​A–​C–​B♮ is 10+​, and
B♮–​C–​A–​B♭ is 10−. This notation ensures that the 7+​, 11−, 3+​, 7−, 11+​, 3−, 7+​, 11−, 3+​.
pitch-​class content of n+​ and n− is identical; as
The pitch-​class content of the passage (and the
remainder of the work) leaves little doubt about
the tetrachordal organization, although the
segmentation of the musical surface into BACH
forms is not aurally apparent, as each form is
divided between two instruments and the forms
sometimes overlap temporally.
FIGURE 4.5.2 Webern, String Quartet, Op. 28: Analyzed in terms of twelve-​tone rows,
twelve-​tone row this series of notes is problematic. The last

FIGURE 4.5.3 (a) A space of BACH forms for Webern, Op. 28; (b) an orbit of BACH forms generated
by the transformation U =​ T4R

148 • E x p l o r i n g M u s i c a l S p a ces
three BACH forms restate the first three and 9.5.13, when we revisit this analysis with a
therefore the row of Figure 4.5.2, but the three sharper focus on the algebraic structure of
BACH forms in the middle do not constitute the transformations involved. Also see Figure
any P, I, R, or RI form of that row. (There is a 9.6.11 for a graphic depiction of the symmetry
row form that starts with 7−, but it continues of the BACH tetrachord and of the entire row.
with 3+​ and 11−, not 11+​ and 3−.) Traditional
twelve-​tone analyses of the work grapple with Exercise 4.5.4 If the symbols n+​ and n− are
this problem in two ways, either by breaking reinterpreted to stand for major and minor
the passage into four rows, overlapping at the +​ triads respectively, then Figure 4.5.3a matches
BACH forms, or by supposing that the second Figure 4.1.1a. What does an orbit such as
row is the same as the first but with its internal {0+​, 0−, 4+​, 4−, 8+​, 8−} represent in the triadic
tetrachords retrograded.59 interpretation? We have encountered a cycle
As a progression of BACH forms, however, through the triads of this orbit previously; is
the pattern is a perfectly regular one, generated that cycle the same as the cycle generated by the
entirely by a transformation we shall call transformation
   U, or is it different?
U, whose action may be summarized “add 4
and change sign.” In terms of more familiar Example 4.5.5: Adams, Phrygian Gates
transformations, U =​ T4R. The transformations Diatonic collections dominate the pitch
T4 and R commute (this should be clear from material of Phrygian Gates, John Adams’s
Figure 4.5.3a; see also Exercise 4.1.6), so we self-​described “behemoth” for piano of 1977–​
could equally well write U =​ RT4. Figure 4.5.3b 78. The opening 113 measures use almost
shows how the pattern travels twice around the exclusively the notes of the four-​sharp diatonic
space of 4.5.3a, cycling through six different collection (the diatonic pitch-​class space
BACH forms, then returns to its starting point dpc(+​4)), the only exceptions being a few B♭s in
and begins retracing its earlier path. mm. 63–​73. At m. 114, the first of the “gates”
The six BACH forms highlighted in referred to in the title, there is an abrupt
Figure 4.5.3b form one orbit generated by shift to the one-​flat collection dpc(−1); this
the transformation U. The word orbit hints section is shorter than the first but its diatonic
suggestively at the route the BACH forms framework no less evident. Adams describes
trace around the figure, but is also a technical these opening sections as being in A lydian
term in group theory, to be defined precisely and A phrygian respectively, but while the
in Chapter 6. An orbit is a set; this particular diatonic collections are not in dispute, the tonal
orbit is the set {3+​, 3−, 7+​, 7−, 11+​, 11−} of six centers, and therefore the modes, are somewhat
BACH forms, three each of the +​and − types, ambiguous.60
whose roots (3, 7, and 11) differ by multiples The complete succession of diatonic
of 4 (mod 12). This orbit is invariant under T4: collections in Phrygian Gates is as follows:
the T4 transposition of any BACH form in the
orbit also belongs to the orbit. The orbit is also 4♯, 1♭, 5♯, 0♯/​♭, 6♯/​♭, 1♯, 5♭, 2♯, 4♭, 3♯, 3♭, 4♯, 2♭, 5♯
invariant under the retrograde R and under
the transformation U =​ T4R. If we apply U Of course, 0♯/​♭ denotes the white-​key
successively to a BACH form not among those collection. The section marked 6♯/​♭ is notated
listed above, such as 4−, a different orbit will in six sharps, enharmonically equivalent to six
result, such as {0+​, 0−, 4+​, 4−, 8+​, 8−}. This 0–​4–​8 flats. This synopsis slightly oversimplifies the
orbit is simply a T1 transposition of the 3–​7–​11 last part of the piece, in which the 3♭, 4♯, 2♭, and
orbit in Figure 4.5.3b, a rotation of the picture 5♯ collections each appear numerous times,
one position in the clockwise direction. Two sometimes in rapid alternation. Otherwise
other orbits, a 1–​5–​9 orbit and a 2–​6–​10 orbit, the pattern is straightforward: at the “gates”
round out the collection; successive application between sections, the diatonic collections
of U to any BACH form will always produce one alternately shift five places in the flatwise
of these four orbits. direction (as from 4♯ to 1♭) and six places in
The large majority of the pitch-​class content the sharpwise direction (as from 1♭ to 5♯). In
of Webern’s quartet arises through applications light of enharmonic equivalence, moving six
of U to BACH forms. Long strings of forms places sharpwise has the same effect as six
within one orbit therefore permeate the piece. places flatwise—​halfway around the circle of
More details will be presented in Example fifths—​and five places flatwise is equivalent

Chordal, Tonal, and Serial Spaces • 149

to seven places sharpwise. In his notation, classes. In terms of signature transformations,
Adams chooses between sharpwise and flatwise the analogous observation is s−5s6 =​ s1, reflected
moves for practical reasons, usually alternating in the fact that the 5♯ collection has one more
the two, in order to stay within the territory sharp than the 4♯ collection.
bounded by the extremes of 6♭ and 6♯. Figure 4.5.6 graphs this succession of diatonic
Transformational relationships among collections in a format similar to the tonality
diatonic collections will be examined in some graph of Figure 4.1.8, substituting diatonic
detail in Chapters 7 and 13. The collection shifts collections for the triads that appeared there on
in Phrygian Gates may be described using what the vertical axis. We have previously regarded
we will call signature transformations, a concept diatonic collections as musical spaces in their
applicable to a variety of diatonic structures. own right, so this vertical axis (topologically a
Specifically, the signature transformation s6 circle) may be thought of as a “space of spaces.”
adds six sharps to any key signature (deleting Because the 4♯ and 5♯ collections appear near
a flat counts as adding a sharp), and a five-​ the beginning of the work and also near the
place flatwise move is s−5. The same actions end, the graph is centered vertically on these
may be described in more familiar terms by two collections. The regularity of the pattern is
applying pitch-​class transpositions to diatonic apparent from the graph: a series of peaks and
collections. The five-​place flatwise move valleys, each peak-​valley pair one step higher
transposes a diatonic collection by T1 (think than the pair before, so that the last two valleys
of 4♯ → 1♭ as a transposition from the E-​major are at the same elevations as the first two peaks.
collection to the F-​major collection), and the The figure may be regarded as a transformation
six-​place move is T6 (1♭ → 5♯ as F major to B network of diatonic collections, using
major). Whichever formulation is preferred, the either transposition operators or signature
transformations involved here represent two of transformations.61
the most remote relationships possible between Adams describes Phrygian Gates as a “tour
diatonic collections, with the consequence that of half the cycle of keys.” All twelve diatonic
adjacent collections in the above list always collections appear in the piece, but the graph
share only two pitch classes—​the smallest establishes a pattern that is not completed: for
possible overlap between two seven-​note the peaks and valleys to return to their initial
collections (see Exercise 2.1.4b). The composite positions would require almost twice as many
of two successive operations, however, is sections, wrapping from the top of the graph
another matter. The transposition operators around to the bottom. Adams’s description
T1 and T6 combine to form T7 (that is, T1T6 =​ assigns all the peaks to the lydian mode and
T7), a transposition by perfect fifth, which is all the valleys to the phrygian; if we accept this
the closest possible relationship between two interpretation, then each lydian-​phrygian pair
different diatonic collections. Consequently, shares a single tonal center, and only seven
alternate collections in Adams’s chain, such as different tonics are heard, progressing by fifth,
4♯ and 5♯, always share six of their seven pitch sharpwise, from A to E♭/​D♯.

FIGURE 4.5.6 Graph of diatonic collections in Adams, Phrygian Gates

150 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 4.5.7 Revised graph showing durations

Adams also likens the structure of the piece

to a “modulating square wave,” an image more
suggestively approximated by Figure 4.5.7.
This graph refines the previous figure by taking
into account the durations of the sections,
calculated using the metronome indications
in the published score. This graph shows some
regularity in the time dimension as well. In
the first part of the piece the lydian peaks
progressively shrink in size while the phrygian
valleys widen, culminating in the ponderous
“System of Weights and Measures” in 3♯ (C♯
phrygian) at m. 640. The last large section of
the piece (mm. 809–​1092), in which each of
the final four diatonic collections appears many
times, has its own internal ternary structure.62
The consistent T1 and T6 relations between
successive diatonic collections suggest the
possibility of constructing a tonnetz generated
by those two transpositions. In fact, we have
already constructed such a tonnetz, in Figure
1.4.18. The elements in that tonnetz were single
pitch classes, but the twelve pitch classes may
be replaced with the twelve diatonic collections,
preserving the transpositional relationships
among them (for instance, replacing each pitch
class with the diatonic collection corresponding
to the major scale on that note). The resulting
tonnetz is shown in Figure 4.5.8a; the figure FIGURE 4.5.8 Diatonic collections in Phrygian
includes the 6-​by-​2 tiles, arrows showing Gates: (a) in a T1–​T6 tonnetz; (b) in a Möbius strip
the T1 and T6 relationships, and the path
traced by the collections in Phrygian Gates.
In terminology from Section 3.5, this figure the rectangle must therefore be twisted before
is a spatial network for Phrygian Gates, while its ends can be stitched together. This strip is
Figures 4.5.6 and 4.5.7 are event networks. shown in Figure 4.5.8b. The vertical edges of
This tonnetz, like its relative in Chapter the modules are not conjoined in the Möbius
1, is topologically a Möbius strip: vertically strip, but as before there is no real need to do
adjacent tiles are reflections of each other, and so. Each row of the tonnetz contains only two

Chordal, Tonal, and Serial Spaces • 151

different diatonic collections, so rather than Claggart, who had falsely accused him of
crossing a vertical edge we can always arrive at mutiny. (The scene concludes Act III of the
the same collection by moving in the opposite original four-​act opera of 1951; in the two-​act
direction. revision of 1961 it is Act II, Scene 2.) Following
In Chapter 1 we applied the T1–​T6 tonnetz in a dramatic monologue by Captain Vere, who
an analysis of Lutosławski’s Musique funèbre. must now tell Billy his fate (“I am the messenger
Adams’s characteristic path through the of death”), a remarkable orchestral postlude
space of Figure 4.5.8 alternates horizontal presents a succession of 34 assorted major and
and vertical moves in the same way as
Lutosławski’s pattern in the earlier graph.
In our Lutosławski analysis we ultimately
abandoned the Möbius strip topology, because
Lutosławski does not exploit it: his movement
through the tonnetz always reverses direction
upon reaching the B–​F tritone, so the simpler
ladder graph of Figure 1.4.22, bounded at each
end by that tritone, proved more appropriate
for analytical purposes. The Möbius strip is
better suited for Phrygian Gates. Alternately
traversing rails and rungs of the twisted
ladder as shown in Figure 4.5.8b, Adams
circumnavigates the strip, and rather than
reversing direction, pushes forward a bit
further, suggesting a pattern that could
continue. The ladder metaphor is compromised
in the circumnavigation: a Möbius strip,
because of the twist, has only one rail. Every
rung is traversed exactly once, along with
seven different rail segments.
Two different short paths are possible from
4♯ to 5♯ in the Möbius strip, namely
4 ♯  → 1 ♭  → 5♯ and 4♯  → 2 ♭  → 5♯.
T1 T6 T6 T1
♭
The path through 1 is taken first, but the twist
brings us back to 4♯ in a different orientation
the second time, so that 2♭ follows this time,
completing the tour of twelve diatonic collec-
tions just before the process stops upon reach-
ing 5♯ for the second time. A second circuit of
the strip, completing the cycle suggested in the
above discussion of Figure 4.5.6, would traverse
the rail segments that were missed previously,
and would traverse all the rungs in the opposite
direction from the first time.

Exercise 4.5.9 Study the score of China Gates,

the shorter companion piece to Phrygian Gates.
Construct a graph of the diatonic collections
appearing in the piece, similar in style to Figure
4.5.6 or 4.5.7. Compare your graph to Adams’s
own schematic diagram of the piece, printed in
the
  front of the published score.
FIGURE 4.5.11 Tonnetz representation of tri-
Example 4.5.10: Britten, Billy Budd ads in Britten, Billy Budd, Act II, Scene 2, rehearsal
In a memorable scene from Billy Budd, Billy 102–​104: (a) chords 1–​34; (b) chords 1–​17 only;
is sentenced to death for accidentally killing (c) chords 18–​34 only

152 • E x p l o r i n g M u s i c a l S p a ces
minor triads, each sustained a full measure in
broad tempo.
The kaleidoscopic variety of registers,
instrumental combinations, and dynamic
levels renders suspect any claims about “voice
leading” in this music. Indeed, after the first
six measures, no instrument so much as plays
even in two consecutive chords. Nevertheless,
tonnetz analysis proves revealing. Because
every triad harmonizes the note F, A, or C,
the triads cluster about the F-​major triad in
the tonnetz, as shown in Figure 4.5.11a. A
fragment of the dual graphs from Figure 4.2.8
appears here, just enough to show the complete
hexagonal cells centered on pitch classes 5, 9,
and 0. The picture shows not the order of the
chords but their frequency: the circles’ areas FIGURE 4.5.13 Modulations in Schubert, “Auf
are proportional to the number of times the dem Flusse,” mm. 1–​41, in chordal-​regional space
various triads appear among the 34 chords in
the passage.63 Thirteen different triads contain
at least one of the notes F, A, or C, and each of operations can often describe the connections.
those triads appears at least once, but the figure Even when no simple PLR compound is suitable,
shows a strong preponderance of F-​major and Schubert often uses the reduced distances to
C-​major triads, and a weaker preference for the third-​related keys available in the tonnetz or
other major triads over the minor. Weber space to good advantage, producing
Dividing the passage in half reveals a temporal harmonic trajectories that trace compact and
aspect of Britten’s organization, reflected in logical paths in those spaces.
Figures 4.5.11b–​c. The first seventeen measures, We begin with an application of Lerdahl’s
shown in (b), career unpredictably from chord chordal-​regional space to Winterreise No. 7, the
to chord, engaging all thirteen available triads much-​analyzed “Auf dem Flusse.”65 The status
with only four duplications. But of the last of E minor as tonal center is highlighted by a
seventeen chords, in (c), twelve are either F double circle in Figures 4.5.13 and 4.5.14, a
major or C major, and seven of the available technique we will adopt on a few occasions.66
triads do not occur at all. This narrowing of The song begins with two stanzas in E minor
tonal focus, together with a generally reduced followed by two in E major, then returns to E
dynamic (there are no tutti chords among the minor for the climactic fifth stanza. Movement
last seventeen, and the only dynamic above between the parallel keys is accomplished in the
pianissimo coincides with the only minor triad), most straightforward of ways, by pivoting on
suggests that Captain Vere is regaining his dominant triads (V/​e =​V/​E) at m. 22 and m. 38.
composure, moving from barely controlled pain E minor and E major are adjacent keys in Weber
and anger to subdued resignation and resolve—​ space and therefore adjacent regions in chordal-​
and also perhaps that the entire passage may be regional space. The pivot dominant is shown
considered an unorthodox composing-​out of an in Figure 4.5.13 by a double line suggesting a
F-​major triad.64 distended =​sign, connecting V/​e with V/​E and
labeled with the measure numbers where the
Example 4.5.12: Schubert, “Auf dem Flusse” pivoting occurs. The arrows joining diatonic
The final three analyses in this chapter involve triads in each key imply the functional harmony
music by Schubert. Schubert’s works are often that unfolds in those keys, but do not capture
attractive candidates for analysis in the tonnetz its every detail.
and related spaces: his sometimes startling The most striking tonal motion in this
excursions to remote tonal areas can pose opening segment of the song is not the shift
challenges to traditional analytical methods, between parallel keys but rather the eerie drop
but because mode mixture and chromatic-​ to D♯ minor for the second phrase of each of the
third relationships are so frequently involved, first two stanzas. D♯ minor acts as a shadow to E
appropriate combinations of the P, L, and R minor, its half-​step-​too-​low strangeness a signal

Chordal, Tonal, and Serial Spaces • 153

of the poet’s pained reproach to the stream dashed lines in the figure).69 Excursion 2 moves
(“You are fallen into silence … you lie there cold first to G major, a closely related key for a
and lifeless”).67 The movement to and from D♯ change, but the abrupt mid-​phrase modulation
minor is accomplished by a pivot through the (m. 56, two measures earlier than the slip to
very same B-​major triad (now V/​e =​VI/​d♯), its the shadow key that we have by now come to
sinister character enhanced by parallel octaves expect, and with no pivot chord other than
between voice and bass (mm. 8–​9 and 17–​18). the G-​major triad itself) and the fragmented
Each of these tonal slips leads to a cadence in vocal line make this move at least as jarring as
the shadow key (mm. 12 and 21, the first two those before. In any event, G is subsequently
authentic cadences in the song), followed by darkened by its own shadow tonality, F♯ minor,
another appearance of the B-​major pivot chord before E minor is restored. In Excursion 3 the
and a return to the original tonic. The harmonic mid-​phrase move is to the more remote G
progress of the piece may be traced in the figure minor, and the final return to E minor, only two
by consulting the measure numbers provided. bars later, is equally unprepared; there is no
The remoteness of the shadow tonality is pivot chord in either case.
graphically evident: as defined in Section 4.2, Figure 4.5.14 has an unkempt appearance, in
D♯ minor lies at a Weber distance of 4 from E keeping with the savage, seemingly haphazard
minor and 3 from E major (the neo-​Riemannian nature of the tonal shifts in the fifth stanza. A
distances are the same as the Weber distances grain of method may be discerned by viewing
in this case).68 the modulations through the lens of generic
The turbulent fifth stanza (“My heart, within space. Setting aside the shadow keys, the
this brooklet/​Do you your image know?”), principal keys in each of the three excursions—​
expanded by text repetitions to nearly the G♯ minor, G major, and G minor—​are all some
length of the first four combined, features the flavor of “G,” hence some flavor of “III.” In
more extravagant modulations diagrammed in the mode-​neutral generic space from Figure
Figure 4.5.14. Three times Schubert surges from 4.3.6c, therefore, all three excursions take the
E minor to a remote region and returns. The simple form shown in Figure 4.5.15. Parallels
first such journey, labeled “Excursion 1” in the among the three “G” passages are not difficult
figure, begins with the now-​familiar drop to the to detect; most obviously, the bass line states
shadow key of D♯ minor, but this D♯ becomes the the song’s opening motive in each of these
dominant of G♯ minor. The return to E minor keys, and they are the only keys other than E
is via its dominant as usual, but the presence minor in which this motive appears. This trio
of a seventh (m. 52), not diatonic to G♯ minor, of IIIs might be said to function collaboratively
deprives the B chord of the pivot-​chord status as “the secondary key area” for the song’s final
that a triad might have claimed (shown by stanza.70

FIGURE 4.5.14 Modulations in “Auf dem Flusse,” mm. 41–​74, in chordal-​regional space

154 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 4.5.15 Modulations in “Auf dem
Flusse,” mm. 41–​74, in generic chordal space FIGURE 4.5.17 Schubert, Impromptu in F
Minor, D. 935 (Op. 142), No. 1: association of
tonal functions with rows in Weber space
Example 4.5.16: Schubert, Impromptu in F
Minor, D. 935 (Op. 142), No. 1
The first piece in Schubert’s second set of four
Impromptus exhibits a more orderly tonal
plan than the song in the preceding example,
its grander scale notwithstanding. All the
important key areas in the work belong to a
single PR-​cycle, one row of Weber space—​

(D), d, F, f, A♭, a♭, C♭, c♭

—​and all of these keys except D major are

touched upon at least briefly. Because Weber
space is organized vertically by perfect fifths,
the dominants of all the keys in this “tonic”
row appear in the next row above, and the
subdominants in the row below, as shown in FIGURE 4.5.18 Graph of tonal centers in the
Figure 4.5.17. The three different rows of keys tonic PR-​cycle in D. 935, No. 1
that constitute Weber space—​the three disjoint
PR-​cycles—​therefore correspond to tonal
functions with great consistency throughout m. 21, swerves back leftward to F minor at m.
the piece. For example, the subdominant keys D♭ 34, touching off an unstable passage at the end
major and B♭ major are tonicized briefly at m. 36 of which A♭ is regained, after a somewhat greater
and m. 165, respectively, and the dominant keys effort, by m. 45. There follows an extraordinary
of E♭ major and C major at m. 110 and m. 223.71 dream scene, pushing ever further rightward:
Figure 4.5.18 traces the Impromptu’s path from A♭ major to A♭ minor (m. 69), C♭ major (m.
through the space of the tonic PR-​cycle. The 77), and a hint of C♭ minor (m. 81).
fact that tonal motion takes place within this Then the course reverses; the dream ends, and
essentially one-​dimensional space frees up the with it the exposition, in A♭ major. The secondary
vertical dimension of the figure for a temporal material in the recapitulatory rotation follows an
segmentation. The piece is an expansive sonata essentially identical course, transposed down a
form without development. The principal minor third to F major. This is a T9 transposition,
keys follow an entirely traditional scheme, the two places leftward in the PR-​cycle, indicated in
exposition moving from F minor to A♭ major Figure 4.5.18 by a version of the bubble notation
(one place rightward in the PR-​cycle) and the introduced in Chapter 3 (Figure 3.5.3). From
recapitulation from F minor to F major (one the more westerly home base of F major, the
place leftward). Lurking within the second-​ rightward probe in the recapitulation of the
theme groups, however, are exploratory tendrils dream scene now extends only as far as A♭ minor,
probing the distant reaches of the space. The but the initial leftward swerve touches upon the
exposition, having attained A♭ major as early as previously unexplored territory of D minor.

Chordal, Tonal, and Serial Spaces • 155

Exercise 4.5.19 Redraw the information from is to say that these are the six keys whose
Figure 4.5.18 in a style resembling Figure 4.5.7, tonic triads contain the note D.73 We call this
displaying time horizontally (in proportion to the tonic-​dominant cell, as it includes G major
measures in the score), but with the vertical and D major, the tonic and dominant keys of
axis organized by the PR-​cycle instead of the the movement, as well as their parallel minor
circle
   of fifths. keys. Those four harmonies, in fact, are stated
plainly at the outset: a G-​major triad sustains
Exercise 4.5.20 The first movement of for almost two measures, crescendo, inflects to
Tchaikovsky’s Symphony No. 4 features minor at the end of m. 2, and crashes into the
many key areas from the same PR-​cycle as gigantic G-​minor chord at m. 3, after which the
the Schubert Impromptu discussed above. very same thing happens on the dominant—​D
Study the tonal plan of this movement in the major and D minor—​in mm. 6–​8. (As if to
tonnetz or Weber space, also describing, when prove the equal status of major and minor, at
appropriate, relationships involving a few the recapitulation the modes exchange places;
significant key areas outside the tonic PR-​cycle. G major and G minor are juxtaposed also in
How are the quasi-​sequences in movement the movement’s final bars.) In the progression
II, mm. 54–​62 and 110–​16 (the two are not G–​g–​D–​d, the G–​g and D–​d moves could
identical) relevant to your analysis of the first of course be described as instances of the
movement?
   parallel transformation P; transformational
descriptions of the g–​D relationship will be
Example 4.5.21: Schubert, String Quartet in considered in Chapter 8. The principal keys
G Major, D. 887 (Op. 161), I of the exposition are G major and D major, as
Many adventurous tonal odysseys unfold we might expect, but G major in particular is
within the epic dimensions of Schubert’s last shaded continually with coloration from its
quartet. The keys range further afield than parallel key, G minor emerging more fully into
the single PR-​cycle of the Impromptu just the limelight in mm. 43–​46 and 98–​102. D
considered, but many of the motions through minor announces its presence briefly in mm.
tonal space exhibit great regularity, and 90–​91, 154–​56, and 158–​60.
tonnetz-​based analysis contributes insights The longest passages in the exposition not in
at times complementary to what others have G or D are in the other two keys of the tonic-​
written about this much-​studied work. While dominant cell, B minor and B♭ major. These are
the entire quartet rewards close study, the the diatonic mediants of G major and G minor
following discussion concentrates on the first-​ respectively, an L relationship in the first case
movement exposition and the beginning of the and R in the second; in relation to D major and
development.72 D minor they are diatonic submediants, and
The prominent keys of the exposition are the Riemannian links are reversed. The medial
deployed about one hexagonal cell in the caesura on F♯ at m. 63 seems to prepare either
tonnetz from Figure 4.2.5b—​the PLR-​c ycle B major or B minor. The subsequent sidelong
shown in Figure 4.5.22. This is the cell whose slip to V⁴₃ in D major for the start of the second
nucleus is pitch class 2 in the background layer theme is delectable, but if the G-​major material
of the dual tonnetz from Figure 4.2.8, which seemed constantly to teeter on the edge of
G minor, the D-​major sections seem just as
frequently on the verge of resolving into B
minor after all: see mm. 68 and 72–​74, several
parallel passages later on, and especially the
extended segment in B minor at mm. 129–​41.
The second theme, like the first, is treated
to a series of variations. The serene B♭-​major
variation at mm. 110–​22 completes the roster
of keys in the tonic-​dominant cell, introduced
by a slip from D to V⁴₃/​B♭ that exactly transposes
the F♯–​V⁴₃/​D motion mentioned above.74
FIGURE 4.5.22 The tonic-​dominant cell of tonal The most celebrated of Schubert’s
centers in Schubert, String Quartet in G Major, D. experiments in this movement is the chromatic
887, I, exposition sequence descending by whole steps that takes

156 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 4.5.23 (a) D. 887, I, mm. 15–​20; (b) tonnetz representation

shape within the primary theme, shown in from G major to F major is 2, from G major to E♭
Figure 4.5.23a. The I–​V6 in G major at mm. major 3—​but in this case the neo-​Riemannian
15–​16 repeats a step lower in F, in the manner distances are more faithful to tonal function, ♭VI
of the opening of the “Waldstein” Sonata.75 But typically having a more clearly defined function
Schubert pursues the sequential implications than ♭VII. Indeed, it is as ♭VI that E♭ functions at
one chord further than Beethoven, to E♭. Figure m. 19, moving down by step to the dominant of
4.5.23b shows this sequence speeding away G major, bringing the sequence back from the
from G major in an east-​northeasterly direction brink of otherworldly realms, and revealing it all
on the tonnetz. The dashed arrows labeled as part of an extraordinarily colorful rendition
“2m” show the two-​measure blocks descending of the time-​honored lament-​bass pattern. The
by whole step, from G to F to E♭, via a two-​ descending fourth in the cello in mm. 15–​20 is
chord chromatic sequence pattern that may fully chromatic, but as David Beach has observed,
be compared to the sequence patterns studied the sequential pattern and two-​bar groupings
previously in the generic tonnetz (Figure 4.3.4). prioritize the notes G–​F♮–​E♭–​D, the descending
F major is already outside the bounds of the tetrachord associated with the key of G minor.76
tonic-​dominant cell, and E♭ apparently still more Figure 4.5.23b resembles Figure 12 in Lewin
remote. The toroidal wrapping of the tonnetz, 1991, 125, which maps the same phrase in
however, actually brings E♭ major nearer to G a Riemann-​inspired tonnetz. According to
major than either of those keys is to F major—​a Lewin, the tonnetz reveals the passage to
neo-​Riemannian distance of 2 rather than 4—​as be a “metaphorical modulation” because
the dotted arrows at the lower left of Figure it arrives at a G-​major triad in a different
4.5.23b remind us. (E♭ major, L of G minor, location in the tonnetz from the one where
belongs not to the tonic-​dominant cell but to it began. Lewin does not claim that the two
the subdominant-​tonic cell, which also contains Gs are acoustically distinguishable, and his
C major and C minor, keys that figure in the tonnetz does not indicate syntonic comma
recapitulation, and E minor, the key of the slow differences as Riemann’s does. He contrasts
movement.) The relative distances would be this interpretation with a Schenkerian reading
reversed in Weber space—​the Weber distance and with Dahlhaus [1978] 1986, which both

Chordal, Tonal, and Serial Spaces • 157

contend that the Gs are the same. Here we dominant, which undergoes another such mode
would describe this circumstance by saying change, thus restating the G–​g–​D–​d progression
that the two Gs, even if represented by distinct from the opening of the movement. This sequence,
vertices in a graph, are nevertheless the same in fact, develops this pattern at greater length,
point in a musical space, and that the topology traces an entirely new path on the tonnetz, and has
of the space allows the phrase to return to a new goal, befitting its formal role as transition to
its starting point by a path that would not be the secondary key area. It is a chromatic ascending-​
possible in a space of simpler topology. fifths sequence, each bass note supporting a mode
This chromaticized descending-​tetrachord change; at the two-​measure (four-​chord) level,
theme with its internal whole-​step descent it steps not down by whole step like the original
is restated in varied form at m. 24, and again sequence but up by whole step, from G to A, then
in slower harmonic rhythm at m. 33. A third B, finally cadencing on the dominant of B at the
variation seems to start at m. 54, but as shown medial caesura mentioned above.77
in Figure 4.5.24a there are two important In Figure 4.5.24b it is the move from D
differences. First, the opening tonic triad gives way minor northward to A, rather than eastward
momentarily to G minor before moving on to the to F, that propels this sequence on a different

FIGURE 4.5.24 (a) D. 887, I, mm. 54–​59; (b) tonnetz representation

158 • E x p l o r i n g M u s i c a l S p a ces
course from the previous one. The chords D, F, step spans the same distance as a descending
and A share a hexagonal cell of their own, the whole step. Like the earlier sequence, this one
dominant–​secondary-​dominant cell. The choice traverses a broad swath of tonal space only to
of a different route across this cell gives this end up fairly near to its starting point; the last
sequence its north-​northeasterly trajectory, three stations on the sequential route (B major,
traversed at the same speed as the previous B minor, and F♯ major) share a cell with D
east-​northeasterly one. Neo-​Riemannian major, the key of the immediately forthcoming
distances are symmetric, so an ascending whole second theme.

FIGURE 4.5.25 (a) D. 887, I, mm. 168–​77; (b) tonnetz representation; (c) hierarchy of sequential
motion in pitch space

Chordal, Tonal, and Serial Spaces • 159

 The implications of the descending-​whole-​step to B, followed by another statement of the
pattern are taken to their boldest extremes in descending-​tetrachord theme on E.
the remarkable passage in Figure 4.5.25a, which As Carl Dahlhaus has noted, the “tonally
springs from the close of the exposition and extraterritorial” whole-​tone scale was “a
launches the development section. No longer perplexing phenomenon in the 1820s”; once such
confined by the interval of a perfect fourth or a a pattern has broken free of traditional harmonic
phrase of a sonata theme, the sequence spirals models Schubert apparently finds it sufficiently
into a bottomless void, descending from D fascinating that he must pursue its ramifications
through an octave and more. The whole-​step at great length.78 In the movement’s coda
pattern D–​C–​B♭–​⋯ clearly derives from the (mm. 415–​24), the long sequence is heard yet
model of mm. 15–​19, and the dotted arrows in again, but reintegrated with its functional tonal
Figure 4.5.25b show this whole-​step descent, surroundings: it descends this time from G to E♭,
comparable to that in 4.5.23b. But harmonic like the tetrachordal model from which it grew,
reinforcement is supplied only at every other and E♭ is once again ♭VI, its resolution to G major
whole-​step arrival, highlighting a larger major-​ intensified by the addition of an augmented
third pattern D–​B♭–​F♯–​⋯ at the two-​measure sixth (m. 426). While the movement as a whole
level, shown by solid arrows. This pattern recalls is a tonal sonata form conforming in its broad
three important harmonic guideposts from outlines to expected norms, the challenges
the second-​theme section in the exposition: its that these “purple” passages pose to traditional
main key area (D), the key of the one exceptional methods of tonal analysis only enhance the
variation (B♭), and the medial caesura (F♯). Here appeal of spatial descriptions like those given
the three are treated more democratically, but here.79
not as precise equals. Details differentiate the
arrivals on B♭ (m. 170) and D (m. 174) from Exercise 4.5.26 Study the sequence in mm.
those on F♯ (m. 172) and B♭ (m. 176)—​the 41–​49 of the first movement of Schubert’s
embellishing chords are V₅⁶ in the former cases, Piano Trio in B♭ Major, D. 898 (Op. 99) in
augmented sixths in the latter—​highlighting a the tonnetz. Which of the sequences in the
still larger pattern encompassing only the initial G-​major Quartet does this sequence most
D, the central F♯, and the final B♭. closely
   resemble?
The developing hierarchy of sequential units
at various levels is most clearly shown not in the Exercise 4.5.27 In the tonnetz, analyze the
tonnetz but in ordinary pitch space, as in Figure chromatic sequences in Schubert’s Quartettsatz,
4.5.25c. Every level of this hierarchy is articulated D. 703, mm. 105–​08 and mm. 173–​80. Each
in the score by one means or another, from the of these sequences resembles one of those
single semitone (the entire whole-​tone scale is discussed above in the G-​major Quartet; point
filled in chromatically in the cello, broken only   any significant differences.80
out
at the central F♯) to the entire compass of 16
semitones. This boundary interval is theoretically Exercise 4.5.28 Study the remainder of the
a doubly augmented ninth—​the sum of eight first movement of the G-​major Quartet, as well
major seconds—​but Schubert spells it as a major as the fourth movement, identifying sequential
tenth by notating the step from A♭ to F♯ as a patterns and other features amenable to analysis
diminished third, and also reduces it to a major in the tonnetz (or Weber space) similar to that in
third by an octave displacement after the same F♯. the preceding discussion. You should not expect
Even when B♭ is reached at m. 176, Schubert that all parts of the work will be illuminated by
is not satisfied. This B♭ becomes the dominant the tonnetz, but you may find several passages
of E♭, and the original descending-​tetrachord or  
tonal relationships to be of interest.
version of the primary theme is heard in
E♭ major, as if to prove that the whole-​tone Exercise 4.5.29 In a similar vein, study
sequence could have continued from m. 19 the tonal design of the first movement of
all along. Then, astonishingly, it all happens Schubert’s B♭-​Major Sonata, D. 960, in the
again. The endless descent from Figure 4.5.25 is tonnetz, being particularly on the lookout for
repeated whole, starting now on E♭ and tracing major-​third (PL-​c ycle) relationships. Compare
the opposite whole-​tone scale from the one your analysis with that of Cohn (1999) or
outlined moments before, its eight steps leading Lerdahl
   (2001, 114–​15).

160 • E x p l o r i n g M u s i c a l S p a ces
NOTES 6. Lewin ([1987] 2007, 176) introduced the domi-
nant transformation under the name DOM, subse-
1. See, for example, Figure 6.2.10b, which maps quently shortened to D by Hyer (1995) and many other
the algebraic structure of transposition and inversion writers. The motivation for calling D the “dominant”
operators; Figures 8.1.10, 8.1.11, and 8.2.3, describing transformation is the idea that D models what happens
actions of various transformations on triads; and Figure at a V–​I cadence, at least in a major key. Interactions
9.6.7, showing relationships among twelve-​tone rows. of D (and transpositions more generally) with other
2. The various arrangements of the same set of triads triadic transformations will be explored in Chapter 8.
in Figures 4.1.1 and 4.1.4 are all labeled with the name 7. See Figures 12.4.7 and 12.5.11 for triadic dis-
of the set, triad. It is only for practical reasons that we tance calculations directly relevant to this example.
do not give each configuration its own name, as we did The triadic distance measures introduced here may be
for pc and fifth: we will mention the set triad often, compared with those in Chapter 1 of Cohn 2012, one
but only rarely will we need to refer specifically to one of which corresponds closely to the voice-​leading dis-
of the arrangements in these figures. We will introduce tance described here.
special names for some particular arrangements of triad 8. Early attempts to incorporate major and minor
space, weber and nrtriad, later in this chapter. together in a circle of fifths were oddly problematic.
3. As a useful aid for conceptualizing the differ- In 1728 Johann David Heinichen arranged the 24 tri-
ence between contextual inversions such as P and ads in a single circle, in the order C–​a–​G–​e–​D–​b–​⋯.
fixed inversions such as I, a circular clock diagram of Two different kinds of relationships appear between
pitch-​class space may be drawn on one old-​fashioned adjacent triads, relative pairs such as C–​a alternating
overhead transparency and a triad on a second trans- with stepwise connections such as a–​G. A few years
parency, overlaid on the first so that it can be rotated later Johann Mattheson produced the arrangement
and reflected to transpose and invert the triad. The C–​G–​e–​b–​D–​A–​f♯–​c♯–​⋯, in which major and minor
axis of inversion for the operator I is fixed in pc space alternate not singly but in pairs, with the result that
and therefore may be drawn on the first transparency, while C major and G major are adjacent, two minor
but the axis of inversion for the contextual inversion triads separate G major from D major. The problem
P is instead drawn on the transparency with the triad, with both of these schemes lies in the attempt to
so that the axis moves as the triad is transformed. depict two different types of relationship—​T7 and
For P, the axis bisects the perfect-​fifth interval in R—​around the perimeter of a single circle. While
the triad, an intervallic relationship that remains Heinichen’s and Mattheson’s circles fundamentally
unchanged no matter how the triad is transposed or remain one-​ dimensional figures as described in
inverted. Chapter 1 (using the second dimension only to real-
4. Hint for Exercise 4.1.2c: You should find two cat- ize the closure of the circle), Kellner’s figure (4.1.4a),
egories of fixed inversion operators In, differentiated like the other double-​circle diagrams in this chapter,
by some simple property of the number n. is genuinely a two-​ dimensional space. The figures
5. Hint for Exercise 4.1.3b: The question is whether described here appeared in Heinichen [1728] 1969,
PT1 is the same function as T1P. Try applying both of 837; Mattheson [1735] 1980, 131; and Kellner [1737]
these composite functions to the same input triad, 1979, 60. All three are reproduced and discussed in
such as C major. If the results differ, even on a single Westergaard 1996, 12–​13.
input, then you know that PT1 and T1P are not the 9. For a picture of a mode space similar to the one
same function. If you think that they are the same suggested in Exercise 4.1.7, see Bates 2012, 39. The
function, then to conclude this definitively you must path through the modes described at the end of the
verify that they always give the same results; conceiv- exercise may be formalized using signature transforma-
ably they might match when applied to C major but tions, to be studied in Section 13.3.
not when applied to some other triad. (Recall the dis- 10. For a graph of tonal centers in the first move-
cussion of equality of functions in Section 2.3: “f =​ g” ment of Beethoven’s “Appassionata” Sonata in the
means that f(x) =​ g(x) for all possible inputs x, not just double circle of Figure 4.1.4, see Carpenter 1983, 19.
for some x.) In practice, checking a few examples may 11. The E-​Major Prelude is one of the examples of
be sufficient to reveal a pattern and to give you confi- tonality graphs in Jablonsky 2003, along with excerpts
dence that the pattern will always hold, and a graph by Bach, Mozart, Beethoven, and Debussy.
such as Figure 4.1.1a may also help you grasp such a 12. The generalized theory of normal forms to be
pattern. In the present situation, at the very least you introduced in Section 10.4 will clarify the status of
should certainly check the behavior of PT1 and T1P the descriptors 014 and 034 for the two inversionally
when applied to both major and minor triads. related trichord types (they are OPT normal forms).

Chordal, Tonal, and Serial Spaces • 161

 13. Hints for Exercise 4.1.11: Chords may be repre- 21. Hint for Exercise 4.2.4a: Parallelogram tiles can
sented as sets of pitch-​class numbers, though in some be converted to rectangles by a shearing transforma-
cases traditional names such as Cm7 may be more tion (see Figure 12.3.1b for such a picture of Weber
convenient. One vertex in the graph for (b), for exam- space), but with the triads arrayed as in Figure 4.2.3,
ple, might be labeled {0, 1, 4, 6, 7, 10} (or 01467t for rectangular tiles are not possible. For a tile to deter-
short). The Petrushka chord is the only set class that mine a toroidal topology, its boundaries must be glued
possesses transpositional but not inversional sym- together in such a way that any edge of the graph that
metry. The four graphs constructed for this exercise, reaches one of those boundaries must continue from
taken together, should have a total of 33 vertices. The the corresponding point on the opposite boundary.
graph for the French augmented-sixth chords may be 22. This graph is dubbed the chicken-​wire torus in
compared with a more elaborate space of the same Douthett and Steinbach 1998, 248.
chords in Lind and Roeder 2009. 23. The triadic space considered here is properly neo-​
14. Hint for Exercise 4.1.13: A very simple answer Riemannian rather than Riemannian, in the sense that
will suffice for the last question. Symmetrical rows will it corresponds (as detailed below) with the conformed
be examined further in Section 9.5; see Exercise 9.5.1 neo-​Riemannian tonnetz rather than the unconformed
for a few examples. Riemannian tonnetz. Some readers might therefore
15. Hint for Exercise 4.1.14: Consider the special object to the shorthand Riemann space, preferring
nature of the tritone transposition (compare with instead the slightly more cumbersome neo-​Riemann
Exercise 3.5.10a). space. Because we shall have no occasion to refer to an
16. Lewin (1967a and 1967b) introduced the idea unconformed version of this space, however, the use of
of twelve-​tone areas in analyzing passages from Moses the shorter term for the more frequently encountered
und Aron and Schoenberg’s Violin Fantasy, Op. 47, space is unlikely to lead to confusion.
and pursued the idea in several other articles written 24. A dual relationship between a chord-​ based
around that time. For analyses of Op. 33a including graph and a note-​ based graph was visible also in
twelve-​tone areas, see Straus 2016, 347–​53, and Roig-​ Figures 3.1.13 and 3.1.14, the hexatonic triad cube
Francolí 2008, 205–​12. and its octahedral dual. Tymoczko 2012 studies dual
17. François-​Guillaume Vial’s “genealogical tree of relationships of this kind in a more general context.
harmony” of c. 1767 (reproduced in Lester 1992, 230) 25. In addition to Lerdahl 2001, some analyses
displays the same underlying organization as Weber incorporating Weber space appear in Rings 2007. The
space. Weber’s diagram, actually a left-​right mirror image neo-​Riemannian literature is rich with analyses in
of Figure 4.2.3, appears in Weber 1817–​21, vol. II, 81. assorted variants of Riemann space; see in particular
18. For the chart of the regions see Schoenberg the many tonnetz analyses in Cohn 2012. For com-
[1954] 1969, 20; interestingly, elsewhere in the same parison of the analytical virtues of Weber and Riemann
volume (68–​75) Schoenberg presents a five-​part clas- space, see Cohn 2011, 327–​28. Cohn 2012, 131–​33,
sification of key relationships that bears only a vague offers a tonnetz analysis of the succession of tonal
relationship to his diagram. Lerdahl 2001 presents centers in Robert Schumann’s Dichterliebe, revising
Weber space on pp. 42–​43 and the similarly organized Lerdahl’s Weber-​space analysis (Lerdahl 2001, 138–​39).
regional space on pp. 63–​66. 26. Hexatonic collections, like other scales, can
19. Krumhansl’s map appears in Krumhansl 1983, be musical spaces in their own right; this is why our
55; it is reproduced as Figure 2.3c in Lerdahl 2001, 46. labels for such collections are typographically similar
See Krumhansl 1990 for related considerations, and to the labels for other spaces.
Krumhansl 1998 for further empirical results. 27. In keeping with the convention established for
20. An unconformed version of Weber space, orga- the dominant transformation D introduced in Section
nized by pure perfect fifths and minor thirds, is theo- 4.1, the mediant transformation M (first defined by
retically possible, but we will not have occasion to use Lewin) maps from the mediant to its tonic, not the
it. Identically named keys eight positions apart on the other way around. Thus M applied to C major yields A
same row of this unconformed space theoretically dif- minor, and M applied to C minor yields A♭ major. See
fer by the interval known as the greater diesis, the dif- Section 8.3.
ference between four 6:5 minor thirds and an octave; 28. The “Spring” Sonata example is mentioned in
this is a sizable interval, given by the ratio 648:625 Cohn 1996, 31 (where several additional examples of PL-​
and equal to about 63 cents (0.63 equal-​tempered cycles are also discussed), offered as an example in Cohn
semitones). The lesser diesis, also called the enharmonic 2012, 26–​28, and graphed in a tonnetz in Hook 2006.
diesis, is the difference between three 5:4 major thirds Liszt’s Consolation is analyzed in some detail in Kopp
and an octave, given by the ratio 128:125 and equal to 2002, 240–​46, and in Cohn 2012, 172–​73, 180–​82, and
about 41 cents. 186–​88. Lewin (1967b, 25) likens Liszt’s modulations

162 • E x p l o r i n g M u s i c a l S p a ces
by major third in this piece to the motions by minor 34. Rings 2007 offers further commentary on the
third through twelve-​tone areas in Schoenberg’s Violin tension between tonic-​ dominant relationships and
Fantasy, Op. 47. Extended PR-​ chains are consider- others typically foregrounded in transformational
ably rarer in musical practice; the Smetana excerpt is analyses.
analyzed in Goldenberg 2007, 71 and 75–​76, and the 35. Some comparisons between Weber and neo-​
Franck in Cook 2005, 113–​116. Cohn has written about Riemannian distances may be found in Spitzer 2003.
the sequence from the Ninth Symphony on several occa- Spitzer refers to the tonnetz as “Riemannian space,”
sions, most relevantly Cohn 1997, 34–​36, and about the and (somewhat confusingly) to Weber space as “pitch
Chopin Fantasy excerpt in Cohn 2012, 98–​100. space.”
29. A peculiar notation in Chopin’s manuscript 36. There is only one key at the maximum neo-​
of the F-​ Minor Fantasy, reproduced inconsistently Riemannian distance from C major. Keys separated
in published editions, highlights the smooth voice by this maximum neo-​Riemannian distance are called
leading of the transformations in Exercise 4.2.13f: an tonnetz poles (a name suggested by Scott Murphy).
apparently superfluous eighth-​note beam, linking the 37. A bonus question related to 4.2.17c: Explain
root of one triad to the fifth of the next, traces the why Figure 4.2.3 is a bipartite graph if considered an
moving voice at each new chord. infinite graph, but is not bipartite when regarded as a
30. Hint for Exercise 4.2.13i: There may be a tempta- finite graph.
tion to answer the first question by saying that some 38. The property mentioned in Exercise 4.2.17d
of the progressions start with P while others start holds because transposition is a distance-​preserving
with L, or perhaps that some are “PL-​chains” while operation in all three triadic spaces. That is, if Tn is a
others are “LP-​chains”—​but these are not satisfactory transposition operator, if x and y are triads, and if dist
explanations. Consider the possibility of removing the represents the distance function in double-​circle tri-
initial A♭-​major triad from the Brahms progression: adic space, Weber space, or Riemann space, then the
the resulting progression would now start with a dif- equation dist(Tn(x), Tn(y)) =​dist(x, y) always holds.
ferent operation, but it would still move in the same Distance-​preserving transformations will be studied
direction on the tonnetz. more generally in Chapter 12.
31. The Verdi example in Exercise 4.2.14 is dis- 39. Hints for Exercise 4.2.17e: This symmetry is
cussed in Cohn 1997, 43–​45, and in Cohn 2012, 116. related to the distance-​preserving property in (d): a
The first two figures in Riemann [1914–​15] 1992 call path from C major to D♭ major, for example, has the
attention to a family of six triads such as these; see same length as some path from what other key to C
Clark 2011b for related commentary. major? For paths linking keys opposite in mode, the
32. Cohn (2012, 25–​28) notes examples of hexa- same conclusion cannot necessarily be drawn, but you
tonic cycles in the finales of Mozart’s Symphony No. should nevertheless find some sort of symmetry in
39, K. 543, and Haydn’s Symphony No. 98. two of the three tables of distances from C major to
33. In unconformed Riemannian (as opposed to minor keys.
neo-​Riemannian) tonnetz space, the two B♭s would 40. For a slightly different representation of a
be different. Acoustically this difference is the diesis toroidal embedding of K7 see London 2002, 138.
between three major thirds and an octave; notation- 41. The first scholar to call attention to the impor-
ally it is betrayed by the enharmonic respelling that tance of transformational properties of diatonic trans-
must occur somewhere in every hexatonic cycle, with- position was John Clough; see particularly Clough
out which Beethoven’s progression, beginning in B♭ 1979–​80. Many keyboard exercises, such as the cel-
major, would end in C♭ ♭ major. But the acoustical dis- ebrated exercises of Charles-​Louis Hanon, consist
tinction is unrealizable in instrumental performance, largely of patterns transposed repeatedly by diatonic
making the conformed tonnetz model more appropri- t1. Vocal warm-​up exercises, in contrast, are often
ate in a practical sense. Some related considerations sung in successive chromatic transpositions by T1.
are discussed in the early neo-​Riemannian literature, 42. See Hook [2014] 2020 for a detailed investiga-
such as Cohn 1996, as well as in Chapter 1 of Cohn tion of generic sequence patterns and their represen-
2012, where a similar arrival on a supposed C♭ ♭ rather tation in the generic tonnetz, with many examples. A
than B♭ in the first movement of Schubert’s B♭-​major tonnetz of equilateral triangles is appealing for this
Sonata is initially characterized as “an error to repair” purpose, as it enables a symmetrical presentation of
before other interpretations are considered. (The the six nonzero generic intervals. The approach to
Verdi progression of Exercise 4.2.14, hypothetically diatonic sequences here may also be compared with
completed by a final return to F minor, arrives back at analyses in Ricci 2002, Clough 2008, Kochavi 2008,
its starting point even in the unconformed tonnetz, and later in this book (see, for example, Figures 8.3.17
and accordingly requires no respelling en route.) and 13.1.7).

Chordal, Tonal, and Serial Spaces • 163

 43. Third-​substitution figures prominently in the from WTC I takes place in the dual graph considered
model of diatonic harmony proposed in Chapter 7 of in this exercise. Reed and Bain 2007 pursue Lewin’s
Tymoczko 2011; for a discussion of sequences (includ- idea further, offering three-​dimensional animations
ing that of Fauré’s Pavane) see pp. 238–​45. of the same fugue in a space that is a hybrid of the
44. See Lerdahl 2001, 57. Lerdahl diagrams the original 013 tonnetz and the pitch helix of Figure
chordal core as a 3-​by-​3 array, the most compact rect- 1.2.2. Fitzpatrick 2011 comments on the merits of the
angular segment that includes every triad, although as two approaches.
seen in Figure 4.3.6 it redundantly includes two of the 53. Tymoczko (2012, 42–​ 46) discusses Gollin’s
seven triads twice. space in relation to other generalizations of the
45. The construction of chordal-​regional space is tonnetz.
related to the mathematical concept of tangent spaces, 54. Hint for Exercise 4.4.7a: The relationship
although the discrete nature of this space sets it apart between C7 and any of these half-​diminished seventh
from most applications of tangent spaces. Generally, chords may be modeled by inversion about the shared
each point x on a surface S may have its own tangent interval. For example, C7 and eø7 are related by the
space, a space that describes the possible directions in contextual inversion that inverts any dominant or
which one can pass through x while moving within S. half-​diminished seventh chord about its tritone.
46. See Lerdahl 2001, 96–​ 101, for a detailed 55. The chord mentioned in Exercise 4.4.7b,
chordal-​regional analysis of Chopin’s E-​Major Prelude; uniquely, has the property that it shares two notes with
briefer analyses of numerous other works appear else- C7 only if one of the chords is misspelled: as properly
where in the book. spelled dominant and half-​diminished seventh chords
47. This point is made, for example, in Laitz in spelled pitch-​class space, these two chords can have
2016, 375. no more than one common tone. In an unconformed
48. For diagrams similar to Figure 4.3.8 see Agmon three-​dimensional tonnetz, constructed without the
1995, 200–​01. A descending-​thirds–​based grammar assumption of enharmonic equivalence, the two tet-
of functional harmony is outlined in Tymoczko 2011, rahedra representing this chord, as described in the
226–​30. For related considerations see also White exercise, would represent two different chords. On the
and Quinn 2018, and the discussion of M-​chains in other hand, in the conformed finite tonnetz in which
Section 8.3 of this book. enharmonically equivalent pitch classes are conjoined,
49. Several versions of the diatonic Möbius strip these “two tetrahedra” are actually the same tetrahe-
appear in the literature, for example in Mazzola dron—​a tetrahedron with which the C7 tetrahedron
2002, 322; Brower 2008, 74; and Cohn 2012, 175–​79. shares two vertices but no edge! Tetrahedra in this
Brower (p. 91) points out an elusive reference to such relationship are geometrically impossible in ordinary
a Möbius strip in David Lewin’s analysis of a passage Euclidean space ℝ3 but possible in the 3-​torus 𝕋3.
from Parsifal (Lewin [1984] 2006, 196). 56. Hint for Exercise 4.4.7d: If you count all tetra-
50. The graphs described in Exercise 4.4.2 appear hedra adjacent to the C7 tetrahedron, including those
in Waller 1978. For further alternative pictures of that share an edge (two vertices) as well as those that
neo-​Riemannian triad space, see Figure 10.2.15 and share a single vertex, the total should be 22. Because
Exercise 10.2.16. of the duplications mentioned in (b) and (c) and sev-
51. The potential for smooth voice-​leading con- eral others, however, these 22 tetrahedra represent
nections with several different chords in the same set something fewer than 22 different chords.
class is a distinctive characteristic of the 037 triad. 57. Hint for Exercise 4.4.7e: The B♭ in layer 1 falls
Cohn (1997 and 2012) has remarked at length on over the center of the C–​E–​G triangle in layer 0, a
this property and its implications for voice leading point that is precisely one-​third of the way from G to A
in triadic music. Tymoczko (2006) observes that this in that layer. Some vertex in layer 2, which you should
property is attributable to the triad’s near-​symmetry, be able to identify, falls over a point exactly two-​thirds
as a single-​semitone displacement from the perfectly of the way from G to A, and some vertex in layer 3 falls
symmetrical augmented triad—​a point to be clarified directly above A.
in the discussion of voice-​leading spaces in Part Three. 58. The most complete presentation of the spiral
52. Hint for Exercise 4.4.4: Your labeling of the array is Chew 2014. Chew offers a variety of analyses
transformations P̂, L̂, and R̂ may depend on how you in the spiral array, including a study of musical humor
laid out your original 013 tonnetz. For the purposes and incongruity in P. D. Q. Bach’s Short-​Tempered
of this exercise, the correspondence with P, L, and R Clavier.
will be clearest if you organize the tonnetz horizon- 59. For an analysis in four overlapping rows see
tally by minor thirds and vertically by whole steps. Bailey 1991, 390–​91; for an analysis with retrograded
Lewin’s (1998) analysis of the F♯-​minor fugue subject tetrachords, see Straus 2016, 342–​47.

164 • E x p l o r i n g M u s i c a l S p a ces
 60. For Adams’s descriptions of Phrygian Gates and (2012, 122–​25) presents a transformational analysis
China Gates see http://​www.ear​box.com/​phryg​ian-​ of the song organized as an event network (his Figure
gates-​china-​gates/​; see also Adams 2008, 88–​90. 6.8), which may be contrasted with the spatial graphs
61. It is because T1 and s−5 yield the same results presented here.
when applied to the diatonic collections appearing 66. Rings 2011b uses double circles to denote root
in Figure 4.5.6 that the arrows are labeled with nota- nodes in his rooted graphs (graphs in which one ver-
tions such as “T1 or s−5.” The word or has been cho- tex is distinguished in some way from all the others);
sen carefully. The label does not say “T1 =​ s−5”—​and his root nodes often, but not always, represent tonal
indeed it should not, because T1 and s−5 are not equal centers. Lerdahl 2001 highlights tonics and occasion-
as mathematical functions. Applied to most subsets of ally other salient events with square boxes in some
a diatonic collection, the addition of five flats to a key graphs.
signature does not produce the effect of a T1 transpo- 67. The word shadow has been applied in a nota-
sition at all. bly similar context in reference to procedures of tonal
62. For an analysis of temporal proportions in displacement common in the music of Prokofiev (Bass
Phrygian Gates see Fyr 2011. At the tempo change at 1988). An example involving the same pair of keys—​
m. 402, Adams supplies a metric modulation instruc- D♯ minor shadowing E minor—​occurs in the opening
tion that is inconsistent with the given metronome measures of Prokofiev’s Cinderella.
markings; the calculations for Figure 4.5.7 are based 68. Figure 4.5.13 suggests a return to E minor
on the (slower) metronome indications. The calculated at m. 22 and an immediately subsequent move to E
durations add up to a total duration for the work of major. No E-​minor triad appears in this measure, so
just under 22 minutes; most recorded performances the analysis could plausibly show a pivot directly from
are a bit slower still, averaging between 24 and 25 VI/​d♯ to V/​E. The reading in the graph postulates that
minutes. the listener recognizes the B-major triad from its pre-
63. There are dangers in using two-​dimensional vious appearance at the parallel spot of m. 13, and,
areas to represent one-​dimensional data, as with expecting a corresponding resolution to E minor the
the circles in Figure 4.5.11. It is the areas, rather second time, hears it momentarily as V/​e.
than the diameters, of the circles that have been 69. G♯ minor lies at a Weber distance of 3 from
set in proportion to the incidence of the triads. E minor, but a neo-​Riemannian distance of 2 (being
The area of a circle is proportional to the square of reachable from E minor by PL). Some remarks of
its diameter. If diameters were used instead, the Lewin ([1986] 2006a, 113–​16) are relevant to this
eye, in perceiving the circles’ areas, would greatly difference and suggest a preference for the Weber dis-
overestimate the implied variation, as the F-​major tance. Lewin observes that G♯ minor projects a domi-
triad’s circle would have 64 times the area of the F-​ nant function in E minor, as it supports the structural
minor triad’s circle, even though the actual ratio in 5̂ and 7̂, particularly the insistent leading tones in the
frequency of appearance is only 8 to 1—​a misrep- vocal line. For this reason he argues that G♯ minor is
resentation exhibiting what Edward Tufte calls a lie more aptly regarded as the relative of B major rather
factor of 8. Using areas as the basis for proportions than as the leittonwechsel of E major.
is greatly preferable but runs the opposite risk, the 70. Similar observations may sometimes be
risk that variation may be underestimated instead advanced to support claims of unity among the
(Tufte 2001, 53–​77). movements of multi-​movement works. In the Piano
64. The last of the 34 chords is C major. In the orig- Quartet in E♭ Major, Op. 87, by Dvořák (a composer
inal version of the opera, this passage is followed by a whose tonal journeys often resemble Schubert’s), the
six-​measure codetta that closes the act in F major. The three movements in E♭ have the secondary keys G
codetta has been excised from the revised version, in major (first movement), G minor (third), and G♭ major
which the final C-​major triad leads attacca to the fol- (fourth; G♭ is also the main key of the second move-
lowing scene, which opens in F major. ment)—​all again some form of III. In light of the com-
65. Lewin’s extended analysis of “Auf dem Flusse” plete absence of B♭ as a significant tonal center, and
(Lewin [1986] 2006a) intersects the discussion pre- the many additional coloristic uses of G♭ throughout
sented here in a number of details, amid a wealth the work, it may not be far-​fetched to claim for this
of insights about text-​music relations. Lewin also shared III the status of “dominant” in the large-​scale
presents his own reductive analysis and a critique tonal organization of the quartet.
of Schenker’s in Der freie Satz (Schenker [1935] 71. Some readers will note the close likeness
1979, Fig. 40/​2). Clark (2011a, 77–​92) comments on between these PR-​ cycle tonal functions and the
Schenker’s and Lewin’s readings, particularly in rela- “axis system” postulated by Ernő Lendvai for the
tion to the G♯-​minor passage in the final stanza. Cohn music of Bartók (Lendvai 1971, 1–​16). Indeed, an

Chordal, Tonal, and Serial Spaces • 165

association of functions with keys in a minor-​third Schubert extends the pattern sequentially, at much
cycle such as a Lendvai axis or a row in Weber space greater length.
captures in a simple if imprecise way a number of 78. See Dahlhaus [1978] 1986, 10–​12. A claim that
intuitions about tonal functions in a wide range “Schubert used the whole-​tone scale” in the G-​Major
of repertoires. For instance, the tonic row appears Quartet, of course, invites justifiable criticism. That
above the subdominant row, in accordance with the the whole-​tone level of the hierarchy in Figure 4.5.25c
tonic’s capability to function as the dominant of the is articulated in the score is not in question, but its
subdominant; similarly, the appearance of subdomi- presence does not prove that the whole-​tone scale is
nants in the row above dominants is consistent with for Schubert a space in which to compose (as it is for
the subdominant-​like function of a chord such as Debussy on occasion) rather than, as seems more likely,
V/​V. We have remarked on functional affinities a particular subset of chromatic space that happens
between third-​ related chords several times previ- to arise as a byproduct of sequential patterning. The
ously, in the discussion of the generic circle of thirds same may be said of an earlier example of tonal cen-
in Section 1.3 and again in reference to Figure 4.3.8, ters moving systematically through a whole-​tone scale,
and also in Exercise 4.3.5d in the context of diatonic the endlessly rising canon from Bach’s Musical Offering.
sequences. In some situations, however, similarly Tymoczko (2002) raises similar questions about
functioning chords may be related by major thirds apparent octatonic organization in some examples by
rather than minor thirds. Thus Cohn (1999, 219) Stravinsky.
associates tonal functions with PL-​(rather than PR-​) 79. Cohn (1999, 232) observes that “diatonically
cycles in the first movement of Schubert’s B♭-​Major indeterminate” passages in the nineteenth century
Sonata, D. 960. (See Exercise 4.5.29 below.) Because often occupy a middleground structural level, sand-
there are four PL-​cycles rather than three, one cycle is wiched between more traditional diatonic procedures
left without an associated function. at both foreground and background levels.
72. Clark (2011a, 161–​82 and elsewhere) discusses 80. Tymoczko (2011, 282–​ 84) analyzes the
this movement at length, with many references to Quartettsatz sequences and plots them in a segment
analyses by other scholars. The analysis in Yust 2015 of a three-​dimensional chord space that we will study
has common ground with our discussion of sequen- in Chapter 11.
tial passages (but Yust’s use of the word generic differs
from the sense of the word in this book).
73. Cohn 2012 refers to these hexagonal cells as SUGGESTED READING
neighborhoods, and to the circumnavigation of such a
The analytical sketches in Section 4.5 have points
neighborhood as a pitch retention loop, coincidentally
of contact with many modes of analysis that may
abbreviated PRL.
be encountered in the literature. The tonnetz-​based
74. Clark (2011a, 175–​82) comments on several
analyses in Cohn 2012 are relevant to our work in
previous analyses of this second theme; see also Kopp
this and later chapters. The cognitively informed
2002, 197–​98 and 205–​06. Hyland 2016 examines the
analyses in Lerdahl 2001, often with an emphasis
interaction of sonata and variation principles in this
on prolongational functions, play out in several
movement.
different spaces, including Weber space and chordal-​
75. Although its most famous use is in the
regional space (as in our discussion of “Auf dem
“Waldstein,” variants of this whole-​step descent appear
Flusse” in Example 4.5.12).
in many other works, from Chopin (the Trio theme
from the second movement of the Cello Sonata, Op. 65) In the realm of Schubert scholarship, Clark 2011a
to Liszt (the final Stretto, m. 537, of the Piano Concerto offers not only original analytical insights but also
in A Major), Bruch (the main theme of the last move- a wealth of commentary about two centuries of
ment of the Scottish Fantasy), Richard Rodgers (“Climb analytical approaches to Schubert’s music. Lewin’s
Ev’ry Mountain” from The Sound of Music), and, at least extended essay on “Morgengruss,” written in 1974
in short sequential passages, Mozart (Sinfonia concer- for a never-​completed book of Schubert essays and
tante, K. 364, III, mm. 152–​56). unpublished for four decades (Bard-​Schwarz and
76. See Beach 1994, 10–​11. Beach relates this minor Cohn 2015), predates Lewin’s groundbreaking work
tetrachord G–​F♮–​E♭–​D to a less conspicuous major tet- in transformational theory, but the wide-​ranging
rachord G–​F♯–​E♮–​ D that spans from the end of the perspectives and rich detail are vintage Lewin.
primary-​theme group (m. 54) to the cadence on D at the Among other Schubert analyses with connections
end of the first statement of the second theme (m. 77). to those in this chapter are Cohn 1999 (on the B♭-​
77. In Beethoven’s “Waldstein,” the second state- Major Sonata, D. 960), Rings 2007 (on the E♭-​Major
ment of the opening theme also begins with a whole-​ Impromptu, D. 899, No. 2), and Yust 2015 (on the
step ascent, from C major to D minor—​but here too, G-​Major Quartet).

166 • E x p l o r i n g M u s i c a l S p a ces
 5
Groups I
Interval Groups and Transformation Groups

THE ABSTRACT algebraic structures known as studied here—​interval groups and transforma-
groups have played a central role in mathemati- tion groups—​and described in a novel way the
cal thought for almost two centuries. The foun- connections between them. Much of the power
dations of group theory are generally said to and elegance of these applications stems from
have been laid by Évariste Galois (1811–​32) just symmetry, which as noted above is always inher-
before his death in a duel at the age of twenty, ent in groups, but this symmetry also proves to
though the field’s origins can be traced through be a constraint in some contexts, for many musi-
an assortment of earlier work in number theory, cal constructions lack the symmetry that would
the theory of algebraic equations, and geometry. be required of a group.2
Groups are intimately connected to symmetry—​ The elements of group theory in this chap-
wherever symmetry appears, a group is probably ter, together with the somewhat more advanced
close behind—​and applications of group theory topics in Chapter 6, provide an introduction
abound in many branches of mathematics, phys- to musical applications of group theory and
ics, and chemistry.1 the group-​theoretic study of symmetry, which
In music, Schoenberg’s twelve-​tone system should be sufficient to enable the reader to nego-
and the operations of pitch-​ class set theory tiate the groups appearing in the remainder of
are inherently governed by group structures, this book. Some parts of these chapters may
as Milton Babbitt was among the first to rec- seem more suitable for a mathematics text than
ognize. David Lewin’s 1987 book Generalized for a book about music, but musical payoffs will
Musical Intervals and Transformations called accumulate later in the book, and in this chapter
unprecedented attention to the importance of I have endeavored to keep musical motivations
groups in musical contexts, identified the two close at hand by using intervals and transforma-
principal categories of musical groups to be tions (about both of which much more will be

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0005
said in later chapters) as prototypes for the more We begin by revisiting the discrete chromatic
general concepts of group theory. This chapter pitch space pitch from Figure 1.1.1. In Chapter
concludes with Lewin’s fundamental insight 1 and again in Section 2.1, we observed an appar-
relating intervals and transformations. ent link between this pitch space and ℤ, the set
of integers. We now make this connection more
5.1 THE INTERVAL AND explicit; we shall see that it depends on the prop-
TRANSPOSITION GROUPS erties of ℤ not only as a set but as a group. The
example serves as a preliminary illustration not
OF PITCH SPACE only of the group concept, but more specifically
The definition of a group, while not intrinsically of the applicability of groups in studying musical
difficult, is rather abstract, even more so than intervals and transformations.
that of a graph (Chapter 3). A group consists of Figure 5.1.1a shows a small collection of
a set and a binary operation satisfying certain notes in pitch—​C4, E4, and G4, the notes of a
precisely defined properties, or axioms, to be C-​major triad—​and some of the intervals they
detailed in Section 5.2 below. Mathematicians form. The number 4 on the arrow from C4 to E4
find it useful to define abstract structures in means that the interval from C4 to E4, measured
this axiomatic way whenever they find similar in semitones, is 4. As a shorthand for this state-
characteristics arising in a number of superfi- ment we write int(C4, E4) =​4. The notation “int”
cially different contexts. By defining a group to in this equation represents a function, called
be any set satisfying the specified properties, the interval function of discrete pitch space.
and then studying the logical consequences of The function int maps from pitch × pitch (the
those properties, we are able to deduce things set of all ordered pairs of notes in pitch space)
about many different groups at once, rather into ℤ (the set of integers), because intervals
than studying each individual group in isola- between notes in pitch are always integers.
tion and rehashing the same reasoning sepa- Using a notation introduced in Chapter 2, we
rately for each one.3 may write int: pitch × pitch → ℤ. In an equa-
To develop a good understanding of the defi- tion of the form int(x, y) =​ i, the interval i is a
nition, however, it is useful to have some specific different kind of object from x and y: x and y are
examples in mind from the start. Fortunately, a pitches (elements of the space pitch), while i is
number of musically relevant groups are already an integer (an element of ℤ).
at hand, lurking beneath the surface of some of The figure shows that int(C4, E4) =​4, int (E4,
the musical spaces we have studied in Chapters G4) =​3, and int(C4, G4) =​7. These numbers
1–​4. In this section we will identify two of these satisfy the equation 4 +​3 =​7—​that is, int (C4,
groups and note some of their properties, in E4) +​int(E4, G4) =​int(C4, G4). The general form
preparation for presenting the formal definition of this interval-​sum equation is int(x, y) +​int(y, z)
of a group and further examples in the following =​ int(x, z). This property may seem obvious, but
sections. there are subtleties. The interval function int is

FIGURE 5.1.1 Interval relationships in a C-​major triad in discrete pitch space

168 • E x p l o r i n g M u s i c a l S p a ces
defined to model directed intervals (sometimes We will refer to a set of objects and the inter-
also called ordered or oriented intervals). While vals they form as an interval configuration. This
int(C4, E4), the ascending interval from C4 to E4, definition will be made a bit more precise later
is 4, int(E4, C4), the descending interval from E4 on, but the above discussion involves a three-​
to C4, is not 4 but −4. Figure 5.1.1b shows why note interval configuration, the set {C4, E4, G4},
the interval function must be defined in this whose complete roster of intervals includes 4, 3,
way—​with ascending and descending intervals 7, −4, −3, −7, and three intervals of size 0. All
represented by numbers of opposite signs—​in of these intervals are shown in Figure 5.1.1d.
order for interval-​ sum equations to work in Clearly there is considerable redundancy in this
general. If we choose the pitches x, y, and z in list of intervals. The 0s are automatic, and the
the order x =​E4, y =​C4, and z =​G4, the interval-​ negative numbers represent the same intervals
sum equation becomes int(E4, C4) +​int(C4, G4) =​ as the positive numbers, oppositely directed.
int(E4, G4), or (−4) +​7 =​3. This equation is true, In fact, the first two intervals in the above list
but it would fail if int(E4, C4) were taken to be effectively tell us all we need to know about
4 rather than −4. When intervals are oriented the rest: given only that x, y, and z are three
differently, the orientation matters in the calcu- notes with int(x, y) =​4 and int(y, z) =​3, from
lations and must be accounted for. This is why the general properties of the interval func-
the interval function must be defined on ordered tion we may deduce that int(y, x) =​−4, that
pairs of pitches rather than on unordered sets of int(x, z) =​7, and indeed all of the other intervals
two pitches: int(x, y) and int(y, x) are two dif- in the configuration.
ferent things. In general, int(y, x) =​−int(x, y); This point is illustrated by part (f) of Exercise
we call int(y, x) the inverse of int(x, y). With int 5.1.2 above: from the fact that the successive
defined as a directed interval function in this notes in the set {D3, F♯3, A3} are related by the
way, the interval-​sum equation int(x, y) +​int(y, same two intervals 4 and 3 that we saw in the
z) =​int(x, z) always holds, for any three pitches set {C4, E4, G4}, we know immediately that the
x, y, and z at all. rest of the intervals in the configuration must
The interval from any note to itself is always be identical as well. The D-​major triad is, of
0. Figure 5.1.1c illustrates the interval-​ sum course, a transposition of the C-​major triad in
equation for the three notes C4, C4, and E4 (two pitch space. Interval configurations are trans-
of which are the same): 0 +​4 =​4. posable, and transposition is an interval-​preserv-
ing operation: the transposed configuration has
Exercise 5.1.2 Make sketches similar to Figure exactly the same intervals as the original, so the
5.1.1a–​c for each of the following combinations pictures of the two configurations look identical
of three notes. Assume that the notes are given except for the note names.4 Interval configura-
in the order x, y, z, and verify that the interval-​ tions are also invertible, as part (g) of the same
sum equation int(x, y) +​int(y, z) =​int(x, z) holds exercise shows: the configuration of the F-​minor
in each case. triad is generated from the pitch C4, not by inter-
vals 4 and 3 but by −4 and −3. Inversion is not
(a) C4, E4, G5 (remember that some intervals in an interval-​preserving operation but an interval-​
pitch space are larger than 12) reversing one.
(b) C4, E4, C4 The pictures in Figure 5.1.1 are directed
(c) E4, G4, C4 graphs as defined in Chapter 3. The complete
(d) E4, G4, C5 picture of a three-​note interval configuration,
(e) G4, E4, C4 as in 5.1.1d, has nine arrows, one for every
(f) D3, F♯3, A3 ordered pair of notes. The preceding observa-
(g) C4, A♭3, F3 tion about our ability to deduce all the intervals
(h)  C4, E♭4, G4 in the configuration from knowledge of two of
them is tantamount to saying that the complete
Exercise 5.1.3 Suppose we defined the interval information about a configuration’s intervals is
function to take positive values for descending determined as long as the graph is connected. A
intervals and negative values for ascending, graph with three vertices requires at least two
rather than the other way around, so that, for arrows in order to be connected, so at least two
instance, int(C4, E4) =​−4 and int(E4, C4) =​4. intervals must be specified in order to deduce
Does the interval-​sum equation work for this the complete interval content of a three-​note
upside-​
   down interval function? configuration.

Interval Groups and Transformation Groups • 169

Exercise 5.1.4 that addition is “a binary operation defined on
(a) The sets in parts (a) and (d) of Exercise ℤ” (that is, a function from ℤ × ℤ to ℤ); neverthe-
5.1.2, even though they are C-​major triads, less it is crucial to the structure of any group to
are not transpositions in pitch space of the ensure that the operation applied to group ele-
interval configuration {C4, E4, G4}. Why not? ments can never take us outside the group.
(b) List all nine intervals present in the F-​minor Two well-​known properties of addition are
triad from Exercise 5.1.2g. Compare this list the commutative property and the associative
with the intervals in the C-​major triad. What property. The commutative property states that
can you conclude about the interval content x +​ y =​ y +​ x for all integers x and y: elements of ℤ
  of inversionally related configurations? always commute. The associative property asserts
that (x +​ y) +​ z =​ x +​(y +​ z) for all integers x, y,
Exercise 5.1.5 A four-​note interval and z. While the commutative property involves
configuration contains a total of sixteen interchanging the order of the elements being
intervals: six positive intervals, their inverses, added, in the associative property it is not the
and four 0s. List the six positive intervals in order of the elements that matters but the way
each of the following four-​note sets: they are grouped, and therefore the order in
which the operations are performed. To calculate
(a) {C4, D♭4, E4, F♯4} (x +​ y) +​ z we first calculate x +​ y and then add z
(b) {C4, D♯4, E4, F♯4} to the result, while to calculate x +​(y +​ z) we first
(c) {C4, D♭4, E♭4, G4} calculate y +​ z and then add the result to x.
The commutative property of the interval
Successive notes in set (a) are related by group ℤ is illustrated by comparison of the C-​
intervals 1, 3, 2, while successive notes in (b) major triad discussed above and the C-​minor
are related by the same intervals in a different triad in Exercise 5.1.2h: int(C4, E4) +​int(E4, G4) =​
order, 3, 1, 2. In pitch-​class set theory, sets (a) 4 +​3, while int(C4, E♭4) +​int(E♭4, G4) =​3 +​4.
and (c), with prime forms 0146 and 0137, are Of course the two sums agree, since both must
Z-​related, sharing the same interval-​class vector. equal int(C4, G4).
Despite these observations, for this exercise
you should have found three different lists of Exercise 5.1.6 Illustrate the associative property
intervals for the three sets. Why does this not using the intervals x =​int(C4, G4), y =​int(G4, B♭4),
contradict
   the Z relation?5 and z =​int(B♭4, E4) (the last of these is a negative
number). Draw a graph showing these intervals,
The above discussion begins to show some of and
  calculate both (x +​ y) +​ z and x +​(y +​ z).
the important roles that ℤ, the set of integers,
plays in describing interval structures in pitch The number 0 occupies a privileged position
space, and suggests that addition of integers also in the group ℤ; it is known as the identity element
plays a role. In fact, the integers, with the binary of the group. The distinctive property of the
operation of addition, form a group, and this group identity element, observed previously in Figure
is the interval group of discrete pitch space pitch. 5.1.1c, is that x +​0 =​x (and also 0 +​x =​ x) for
Technically, to specify a group we must identify every integer x. In pitch space, 0 appears as the
both a set (in this case ℤ) and a binary operation interval from any note to itself.
defined on that set (in this case addition); there- We have also seen examples of the inverse of
fore the group of integers is sometimes written an integer in ℤ: the inverse of any number x is
as an ordered pair (ℤ, +​). In practice, however, the the number −x. This is true even if x is already a
operation is often obvious once the set is specified negative number, in which case its inverse −x is a
(in fact, the operation is often ordinary addition). positive number. For example, the inverse of 3 is
If we speak of “the group ℤ,” we understand implic- −3, and the inverse of −4 is −(−4) =​4. A number
itly that addition is the associated operation. and its inverse always sum to the identity ele-
Several familiar properties of ℤ, and more ment: the equations x +​(−x) =​0 and (−x) +​ x =​0
specifically of addition defined on ℤ, are relevant are true for every integer x. The identity ele-
to its status as a group. First, the sum of two ment 0, uniquely among the integers, is its own
integers is always an integer: if x ∈ ℤ and y ∈ ℤ, inverse (−0 =​0). In pitch space, we have seen
then x +​ y ∈ ℤ. This is the closure property; we say that inverses arise whenever the orientation of
that ℤ is closed under addition. This property is an interval is reversed: the equation int(v, u) =​
probably obvious, and it is implicit when we say −int(u, v) holds for all pitches u and v.

170 • E x p l o r i n g M u s i c a l S p a ces
 We observed some similar equations previ-
ously, for example in conjunction with Figure
3.5.1. In fact, Figure 5.1.7, like 3.5.1b, is a T-​net:
a transformation network all of whose trans-
formations are transpositions. One difference
is that the transpositions in Figure 5.1.7 act
on pitch while those of the earlier T-​net acted
FIGURE 5.1.7 Transposition relationships in a on pc. While in pc there are only twelve differ-
C-​major triad ent transposition operators T0, …, T11, in pitch
there are infinitely many: …, T−2, T−1, T0, T1,
T2, …, one transposition Tn for every integer n.
Figure 5.1.7 recasts Figure 5.1.1a using the Transposition operators with negative indices,
transposition operators T4, T3, and T7 rather than such as T−1 and T−2, are needed in order to trans-
the intervals 4, 3, and 7. It should be clear that pose downward. The functional composition
this figure conveys essentially the same infor- analogous to the interval-​sum equation (−4) +​7
mation as the earlier one, but the comparison =​3 associated with Figure 5.1.1b is T−4 ● T7 =​ T3.
illustrates the contrast that David Lewin drew
between intervallic and transformational con- Exercise 5.1.8 Why are transposition operators
ceptions of musical relationships. While the with negative indices not needed in pitch-​
interval labeled “4” in 5.1.1a tells us in a passive, class
   space?
static way how far the note E4 lies from C4 (four
semitones), the transformation labeled “T4” on Exercise 5.1.9 Choose three of the interval
the corresponding arrow in 5.1.7 replaces this configurations from Exercise 5.1.2 and redraw
conception with a more active, dynamic one, them as T-​nets in the style of Figure 5.1.7.
telling us what we must do to move from C4 to Do the same with one configuration from
E4 (go up four semitones). The conceptual shift Exercise
   5.1.5, using the positive intervals only.
from intervals to transformations is signaled in
the figures by a change of arrowhead style, a con- For reasons that will become clearer later, we
vention we will observe often, though it is not denote the infinite set of transposition opera-
possible to do so with complete consistency. We tors in pitch space {…, T−2, T−1, T0, T1, T2, …} by
will have more to say about these contrasting the name 𝒯∞ (the subscript is the symbol for
conceptions at the end of this chapter; at present infinity). If Tm and Tn are any two elements of
we are interested in comparing the mathematical 𝒯∞, then Tm ● Tn =​ Tm+​n is also an element of 𝒯∞;
behaviors of the transposition operators and the in other words, composition of functions ● is a
interval arithmetic.6 binary operation defined on the set 𝒯∞, and 𝒯∞ is
The transposition operators Tn are not inte- closed under that operation. The crucial observa-
gers, and cannot be added like the intervals in tion is that (𝒯∞, ●) turns out to be another exam-
Figure 5.1.1—​an expression like T4 +​ T3 does ple of a group, and the correspondence between
not make sense. The transposition operators intervallic and transformational thinking
can nevertheless be combined in a way corre- shows us that this transformation group (𝒯∞, ●)
sponding exactly to the addition of intervals. works in much the same way as the interval group
The appropriate operation for combining these (ℤ, +​) discussed above. In fact, the groups (ℤ, +​
operators is composition of functions, denoted ● ) and (𝒯∞, ●) are isomorphic in a sense to be made
in Chapter 2. Recall that T4 ● T3 means “T4-​then-​ precise in Chapter 6.
T3”—​that is, “go up four semitones, then go up In some respects this observation may seem
three semitones.” The net effect of this compos- surprising. The elements of the two groups are
ite function, clearly, is to go up seven semitones; entirely different sorts of objects (numbers in
thus T4 ● T3 =​ T7. More generally, the equation one case, functions in the other), and the binary
Tm ● Tn =​ Tm+​n holds for all integers m and n. This operations are also very different (addition of
property is closely analogous to the interval-​ numbers, composition of functions). On the
sum equation discussed previously, expressed other hand, comparison of Figures 5.1.1 and
in terms of transposition operators rather than 5.1.7 shows that both groups seem to describe
intervals; composition of the transposition oper- the same musical situations equally well, and
ators Tm and Tn corresponds to addition of the the above discussion shows that any statement
intervals m and n. about one of the groups (such as the integer

Interval Groups and Transformation Groups • 171

sum 4 +​3 =​7) can be readily translated to a cor- to assert this kind of commutativity means that
responding statement about the other group all pairs of elements of the set commute under
(the functional composition T4 ● T3 =​ T7). With the given operation. An operation such as com-
groups, as with graphs in Chapter 3, a great ben- position of functions may be commutative on
efit to the abstract mathematical approach is some sets (such as the set of transposition opera-
its generalizing power—​the ability that it gives tors) but not on others.
us to make an observation in one context, and The binary operation in the group (𝒯∞, ●) is
thereby to conclude, because of a shared abstract also associative. In this case a more general
structure, that the observation applies also in statement is true: composition of functions is
another, apparently very different, context. always associative. It can be somewhat confusing
The group (𝒯∞, ●), like (ℤ, +​), is commutative. to puzzle out what this means, but the upshot
This may come as a surprise also, because the is simple, as Figure 5.1.10 demonstrates. A gen-
reader may recall that composition of functions eral statement of associativity requires three
in general is not commutative: there are numer- separate functions f: S → T, g: T → U, and h:
ous examples of functions f and g for which U → V, linking four possibly different sets S, T, U,
f ● g ≠ g ● f (see Exercise 2.3.16). But transposi- and V. Keeping in mind that f ● g means f-​then-​
tion operators in pitch space always commute g, applying the composition (f ● g) ● h to an ele-
with each other. Formally this may be proved by ment x of S means that we first apply f-​then-​g to
the chain of equations x, then apply h to the result. Applying f ● (g ● h)
to x, meanwhile, means that we first apply f to
x, then apply g-​then-​h to the result. Either way,
Tm • Tn = Tm+ n = Tn+ m = Tn • Tm ,
clearly, the process amounts to applying f, g, and
h in that order, so the results are the same.
which shows that Tm and Tn commute precisely The identity element of the group (𝒯∞, ●) is the
because addition of the numbers m and n is transposition T0, transposition by zero semitones.
commutative. Applied to any pitch, this function leaves the pitch
As we observed in Section 2.3, some things unchanged. T0 is precisely the function that we
commute and others do not. It is important to called the identity function in Section 2.3, denoted
recognize that while we may speak of a pair of ele- E; more precisely it might be denoted Epitch, the
ments commuting, commutativity is not a prop- identity function on the space pitch. An identity
erty of an individual element. It does not make function ES may be defined on any set S; when the
sense to say “T4 is commutative,” because T4 com- set is a musical space on which transposition can
mutes with some other functions (such as T3 or be defined, we usually use the notation T0 rather
any other Tn) but not with others (T4 does not than ES. The functional equation analogous to
commute with the inversion operator I). Most an integer sum such as 3 +​0 =​3 is T3 ● T0 =​ T3.
often, commutativity is considered as a property More generally, if f: S → S is any function defined
of a binary operation defined on a particular set; on a set S, the equations f ● ES =​ f and ES ● f =​ f

FIGURE 5.1.10 The associative property for composition of functions

172 • E x p l o r i n g M u s i c a l S p a ces
always hold (see Exercise 2.3.15 for some slightly on G, which means that if x and y are any two
more general statements). elements of G, then x ⊗ y is defined; but ⊗ may
Just as the inverse of the integer n in the group be addition, multiplication, composition of func-
(ℤ, +​) is −n, the inverse of Tn in the group (𝒯∞, ●) tions, or something utterly different. The opera-
is T−n. Inverses in (ℤ, +​) satisfy the equations n tion ⊗ is called the group operation of the group
+​ (−n) =​0 and (−n) +​ n =​0; the analogous equa- (G, ⊗). As mentioned above, we may refer to the
tions in (𝒯∞, ●) are Tn ● T−n =​ T0 and T−n ● Tn =​ T0. set G itself as the group, but only if the opera-
In terminology from Section 2.3, these equations tion ⊗ is obvious or has already been specified;
mean that T−n is the inverse function of Tn, and in another common parlance, we say that “G is a
we could write T−n =​ Tn−1. Observe the notation group under the operation ⊗.”
carefully: the symbol − is part of the subscript in In order for (G, ⊗) to be a group, the following
T−n, and part of the superscript in Tn−1. Note also four properties—​the group axioms—​must hold:
that these uses of the words identity and inverse
for group elements are compatible with the ways • Closure. For all elements x and y of the set G,
in which the same words were defined previously x ⊗ y is an element of G.
for functions: in the transposition group on pitch • Associativity. For all elements x, y, and z of G,
space (and in transformation groups more gener- (x ⊗ y) ⊗ z =​ x ⊗ (y ⊗ z).
ally), the identity function is the identity element • Identity element. There exists an element e of G
of the group, and inverse functions are inverse such that x ⊗ e =​ e ⊗ x =​ x for every x in G.7
elements in the group. • Inverses. For every element x of G there exists
an element xʹ of G such that x ⊗ xʹ =​ xʹ ⊗ x =​ e.
Exercise 5.1.11 Listed below are four true
statements about the group (ℤ, +​). Convert each A group, by definition, consists of any set G and
statement to a corresponding statement about operation ⊗ satisfying the above four axioms.
the group (𝒯∞, ●). All of these axioms are familiar from the
examples in Section 5.1. As noted there, the
(a) −4 +​17 =​13. closure axiom might be considered superfluous,
(b) (2 +​(−6)) +​7 =​5 +​(9 +​(−11)). inasmuch as closure is implicit in the stipulation
(c) −(m +​ n) =​(−m) +​(−n) for all integers m that ⊗ be a binary operation defined on the set
and n. (Use the notation Tn−1 for inverse G. Closure is essential to group structure, how-
functions.) ever, and there are situations in which a lack of
(d) There is only one n in ℤ for which n =​−n, closure might easily be overlooked, so it is gener-
  namely n =​0. ally important to check that it is satisfied.
The reader will probably have noted that one
It is clear by now that the properties of the of the important properties discussed in Section
transposition group (𝒯∞, ●) are very similar to 5.1 is missing from the group axioms, namely
the properties of the interval group (ℤ, +​). In commutativity.8 In fact, groups are not required
fact, the properties that we have enumerated for to be commutative. If the operation ⊗ in a group
these two groups correspond very nearly with (G, ⊗) is commutative—​that is, if it is true that
the group axioms—​the properties that every x ⊗ y =​ y ⊗ x for all elements x and y of G—​then
group is required to satisfy. We are ready for the the group itself is called a commutative group.9
formal definition of a group. Both of the groups introduced in Section 5.1,
(ℤ, +​) and (𝒯∞, ●), are commutative, as are numer-
5.2 DEFINITION OF ous other familiar examples of groups, but many
A GROUP; ADDITIVE, other important groups are noncommutative.
MODULAR, AND We have previously noted some cases in which
MULTIPLICATIVE GROUPS commutativity fails (for example, transposition
operators do not generally commute with inver-
Formally, a group is defined as an ordered pair sion operators), and such behavior may often be
(G, ⊗), where G is a set and ⊗ is a binary opera- described using noncommutative group struc-
tion defined on the set G. This formulation tures. We will encounter our first explicit exam-
allows for great generality. It does not specify ples of noncommutative groups in Section 5.5.
what sort of a set G is: it may be a set of num- Many presentations of group theory use
bers or a set of functions or some other kind of multiplicative notation for the group opera-
set entirely. Meanwhile, ⊗ is a binary operation tion, writing x ∙ y, or even xy with no operation

Interval Groups and Transformation Groups • 173

symbol at all. It is also common to refer to x ∙ y of cpitch. If we consider Figure 1.1.1 to repre-
as the product of x and y regardless of what the sent this continuous space, the intervals that
operation ∙ actually is. We will lapse into these arise are no longer restricted to the integers but
habits occasionally, but multiplicative notation can be any real numbers at all. For instance, if
carries the risk that the reader will assume that C4+​0.5 indicates a pitch one quarter-​tone (0.5 semi-
the operation is multiplication, or something tones) above C4, and E4−0.1 indicates a pitch 0.1
much like multiplication, when it might in fact semitones below E4, then int(C4+​0.5, E4−0.1) =​ 3.4.
be something quite different. For the moment The group properties described in Section 5.1 for
we have deliberately chosen a symbol ⊗ that has the discrete space (and its group) hold as well for
no standard numerical interpretation, to avoid the continuous space (and its group).
preconceptions that may be associated with
multiplicative notation. Group theorists often Exercise 5.2.1 Verify that the interval-​sum
reserve the addition symbol +​for commutative equation int(u, v) +​int(v, w) =​int(u, w) holds if
operations; this practice will be followed here,   the pitch C4+​0.5, v is E4−0.1, and w is G4+​0.25.
u is
and indeed +​will usually denote ordinary addi-
tion or modular addition. The interval-​sum equation, it may be noted,
Likewise, it is common to write x−1 (pro- is not strictly a group property, inasmuch as it
nounced “x inverse”) for the inverse of an ele- refers to things that are not part of the group—​
ment x, the element called xʹ in the statement the notes u, v, and w and the interval function
of the inverse property above. This notation int. This equation is characteristic of interval
is familiar from ordinary multiplicative arith- spaces, to be defined formally in Section 5.4.
metic, where x−1 denotes the number 1 ∕ x, the It is also possible to form additive groups
reciprocal (multiplicative inverse) of x; but the smaller than ℤ. For example, the even integers
notation x−1 is commonly used in group theory …, −4, −2, 0, 2, 4, … form an additive group
whether the group operation is multiplication (ℤeven, +​). To be certain that (ℤeven, +​) is a group,
or not. In the additive group (ℤ, +​), x−1 is actu- we must check that the sum of two even integers
ally the number −x. is always an even integer (closure); that the iden-
(ℤ, +​) is one familiar example of a group tity element 0 is an element of ℤeven; and that
whose elements are ordinary numbers, but other the additive inverse −x of every even integer x is
such examples are close at hand. In the remain- even—​all of which are true, of course. It is not
der of this section we will build up our repertoire necessary to check the associative property in
of groups by surveying several more examples, this case: we already know that addition is asso-
introducing along the way a few more terms ciative on all of ℤ, so it is automatically associa-
and concepts from group theory. We will discuss tive on any subset of ℤ.
three main types of groups in this section: addi- If the even integers form an additive group,
tive, modular, and multiplicative groups. (We will what about the odd integers ℤodd =​{…, −3, −1, 1,
encounter many more types of groups later.) 3, …}? In fact, (ℤodd, +​) is not a group, for at least
Additive groups—​ groups such as (ℤ, +​) in two reasons. First, the closure property does not
which the elements are numbers and the group hold: the sum of two odd integers is even, not
operation is ordinary addition—​form our first odd. Second, there is no identity element in ℤodd,
major category. In the construction of (ℤ, +​) because the number 0 is even. We could try to
we could easily replace the integers ℤ with the solve the second problem by adjoining the iden-
real numbers ℝ. To check that (ℝ, +​) is a group tity element 0 to the set ℤodd, but that would not
requires verifying that the group axioms are address the lack of closure, and even a single
still satisfied for this larger set, which should be violation of a group axiom is enough to prevent
apparent: clearly the sum of two real numbers is something from being a group.
always a real number (closure); addition is asso- Suppose that (G, ⊗) is a group and H is a sub-
ciative whether the numbers being added are set of G. As the previous examples illustrate, it
integers or any other real numbers; 0 remains may or may not happen that H, when consid-
the identity element; and every real number x ered together with the group operation ⊗ that it
has an inverse −x. In a similar way, the rational inherits from G, is a group in its own right. If H
numbers ℚ form an additive group (ℚ, +​). is a group, then it is a subgroup of G. The above
The group (ℝ, +​) plays the same role in contin- examples show that the even integers form a
uous pitch space cpitch that (ℤ, +​) plays in the subgroup of the additive group of integers, but
discrete space pitch: (ℝ, +​) is the interval group the odd integers do not.

174 • E x p l o r i n g M u s i c a l S p a ces
 In general, to check that a subset H is a sub- could be the interval group of a musical space
group of G requires verifying three things: that is a superset of pitch and a subset of
cpitch. Two such groups are described in this
• Closure. H must be closed under the operation ⊗. exercise.
That is, for all elements x and y of H, x ⊗ y must
be an element of H. We know that x ⊗ y is some (a) Construct the smallest possible additive
element of G, because G is a group and therefore group that contains all integers and the
satisfies the closure axiom—​but for the subset number 1 2 . Give a musical interpretation of
H to be a subgroup, every product of elements this group as the interval group of a certain
of H must also belong to H. subspace of cpitch.
• Identity. The identity element e of G must be (b) Another group may be described in
an element of H. There is automatically an reference to the time domain. Consider a
identity element e in G, but it has to belong to quarter note to have value 1, and consider
H in order for H to be a subgroup. all durations that can be represented in
• Inverses. The inverse x−1 of every element x traditional notation, including quarter,
in H must be an element of H. Every x has eighth, sixteenth notes, and so on;
an inverse x−1 in G; if H is to be a subgroup, ties are allowed, but triplets and other
it must contain the inverses of all of its tuplets are not. By including positive and
elements. negative numbers and 0, we may consider
numbers to represent intervals from one
As noted above in the case of ℤeven, it is not time point to another, directed either
necessary to check associativity for a subgroup forward or backward in time. Describe
H: as long as G is known to be a group, then asso- this set mathematically and explain why it
ciativity holds for all combinations of elements forms an additive group. Can you see why
of G, including those in H. this group is sometimes called the ruler
The additive groups considered so far form a   group?11
chain of subgroups,
Consider the set ℤ>0 =​{1, 2, 3, …} of positive
(ℤeven, +) ⊂ (ℤ, +) ⊂ (ℚ, +) ⊂ (ℝ, +). integers. Under the operation of addition, ℤ>0
satisfies the closure axiom—​the sum of two pos-
itive integers is a positive integer—​and of course
Each group in this chain is a subgroup of the fol- the associativity axiom. ℤ>0 is not a group, how-
lowing one. ever, because the identity and inverse axioms
fail in the absence of 0 and the negative num-
Exercise 5.2.2 bers. Adding the number 0 to the set yields the
(a) Of what subset of pitch space pitch is set ℤ≥0 =​{0, 1, 2, 3, …} of nonnegative integers,
(ℤeven, +​) the interval group? There are two which has an identity element but still lacks
such subsets; make sure you understand inverses and therefore fails to be a group.
why ℤeven, not ℤodd, is the interval group for In general, a set S with a binary operation
both of them.10 satisfying the closure and associativity axioms
(b) The set ℤeven is sometimes denoted 2ℤ, is called a semigroup. A group is therefore a spe-
because it consists of all numbers of the cial case of a semigroup, namely a semigroup in
form 2x, where x ∈ ℤ. This notation may which the identity and inverse axioms hold. The
suggest a way in which other subgroups examples of ℤ>0 and ℤ≥0 suggest intuitively why
of ℤ could be constructed. Is the set 3ℤ the name semigroup is appropriate: each of these
(multiples of 3) a subgroup of (ℤ, +​)? What sets looks approximately like “half a group”—​
about nℤ for any positive integer n? Give but for a group the other half is equally essen-
examples of subsets of pitch space for tial. If a semigroup lacks an identity element,
which some of these groups serve as the one can always be added if desired. In the case of
  interval group. ℤ≥0, it is also a simple enough matter to expand
the semigroup to a group by including the nega-
Exercise 5.2.3 Some additive groups fall tive numbers, but that is not always the case
between ℤ and ℚ in the above chain of with semigroups; some semigroups of transfor-
subgroups: that is, there exist groups (H, +​) mations cannot reasonably be expanded to com-
such that (ℤ, +​) ⊂ (H, +​) ⊂ (ℚ, +​). Such a group plete group structures.

Interval Groups and Transformation Groups • 175

Exercise 5.2.4 Each of the following sets of The sole example of a finite additive group
numbers fails to be an additive group for one or is the simplest group possible: the set {0}. The
more reasons. Give specific numerical examples empty set is not a group, because every group
to show which group axioms fail in each case. must have at least an identity element—​but a
(For instance, to show that the closure axiom set with an identity element and nothing else is
fails you should exhibit two numbers in the set indeed a group, sometimes called a trivial group.
whose sum is not in the set.) Two of the five In the group {0}, the element 0 is of course the
sets are semigroups; identify them. identity and is also its own inverse; this trivial
group satisfies the closure axiom because the
(a) The set {−10, −9, −8, …, 8, 9, 10} consisting only thing to check is the equation 0 +​0 =​0.
of all integers from −10 to 10. If we attempt to include any other element
(b) The set ℝ>0 of all positive real numbers. besides 0 in an additive group, we are quickly led,
(c) The set ℚ≥0 of all rational numbers greater as suggested above, to the conclusion that the
than or equal to 0. group must be infinite. If, for example, we want
(d) The set ℝ≠5,−5 of all real numbers excluding our group to contain the number 3, then the clo-
the numbers 5 and −5. sure axiom implies that it must also contain 6
(e) The set {…, −13, −12, −11, −10, −91 2 , −9, (because 3 +​3 =​6), and therefore 9, 12, and so
−81 2 , −8, …, 8, 81 2 , 9, 91 2 , 10, 11, 12, 13, …} on; moreover, the inverse axiom implies that the
consisting of all integers, together with the negatives of these numbers must also be pres-
  half-​integers between −10 and 10 only. ent. So any additive group that contains 3 must
contain, at the very least, all multiples of 3. This
The non-​groups in the preceding exercise call is the set called 3ℤ in Exercise 5.2.2b above. It is
our attention to a few requirements for groups the smallest symmetrical, homogeneous set con-
that, though not stated directly in the group taining both 0 and 3 (in terminology to be made
axioms, are consequences of them. Generally, precise in Section 5.3, it is the group generated by
groups are symmetrical and homogeneous struc- the number 3)—​and it is an infinite set. Exactly
tures. The sets in (b) and (c) fail to be sym- the same reasoning applies if any other nonzero
metrical because the positive numbers are not integer is used in place of 3. In fact, the only non-
balanced by their negative counterparts as the trivial subgroups of ℤ are the infinite sets of the
inverse axiom would require. Sets (d) and (e) are form nℤ as described in Exercise 5.2.2b—​all the
symmetrical—​they are perfectly balanced about multiples of n for some integer n. Groups of this
0—​but the numbers in these sets are not distrib- type will be described more fully in Section 5.3.
uted uniformly, and this inhomogeneity makes This implication of infinity could be avoided,
the closure axiom impossible to satisfy. however, if a group could somehow cycle back on
It may be tempting to conclude from an itself. If repeated application of the group opera-
example such as that in Exercise 5.2.4a that a tion eventually produced elements that were
group must be an infinite set: the closure condi- already accounted for—​if, say, an equation such
tion might seem to imply that one can always as 3 +​3 +​3 +​3 =​0 could somehow be true—​
combine two group elements to generate a third then perhaps a finite group could be possible.
“larger” element, then a fourth, and so on ad infi- Such things never happen in ℤ or ℝ, but we know
nitum. This conclusion is (with one exception) another setting in which this very equation is
true about additive groups of numbers, but it true: mod-​12 arithmetic. This observation intro-
is not true about groups in general—​illustrat- duces our second main category of groups, mod-
ing the danger of leaping to general conclusions ular groups. The integers mod 12 form a group
based on only a few examples. In fact, there (ℤ12, +​), and more generally the integers mod n
are many important examples of finite groups, form a group (ℤn, +​) for every integer n ≥ 2. Just
some of them familiar in musical contexts (the as (ℤ, +​) is the interval group of discrete pitch
twelve transposition operators on pitch-​class space pitch, (ℤ12, +​) is the interval group of dis-
space form a group, for example, as do the 24 crete pitch-​class space pc. Similarly, (ℤ7, +​) is the
transposition and inversion operators), some of interval group of generic pitch-​class space gpc or
which we will begin to study later in this chap- of any diatonic pitch-​class space dpc(n).
ter. To see how this is possible, it may be help- In a sense modular groups such as (ℤ12, +​) are
ful to attempt to construct a small finite additive also “additive,” because the group operation is a
group, to see what goes wrong, and to see if it kind of addition, but it should be remembered
can be fixed. that the symbol +​in the group (ℤn, +​) represents

176 • E x p l o r i n g M u s i c a l S p a ces
addition mod n, which, as discussed in Chapter as demonstrated by the equations x ∙ 1 =​1 ∙ x
2, differs in important respects from ordinary =​ x. The multiplicative inverse of a number x
addition of integers or real numbers. In any case, must therefore be x−1 =​1/​x, the number which,
the modular groups (ℤn, +​) are some of the sim- when multiplied by x, yields the identity ele-
plest examples of finite groups. The cardinality ment 1; the characteristic equation for inverses,
of ℤn is the number n. It is traditional to use the the multiplicative counterpart of x +​(−x) =​0,
word order rather than cardinality for the num- is x ∙ x−1 =​1. A new wrinkle arises, however, if
ber of elements in a group, so we say that the x =​0, because in this case no such number x−1
integers mod n form a group of order n. exists. Because 0 has no multiplicative inverse,
The identity element of the group (ℤn, +​) is it follows that the number 0 can never be an ele-
the mod-​n integer 0, and the inverse of a mod-​n ment of a multiplicative group. Consequently
integer x is the mod-​n integer −x. For example, in familiar sets such as ℤ, ℚ, and ℝ do not form
ℤ12, the inverse of 1 is −1 =​11, and the inverse multiplicative groups at all.
of 8 is −8 =​4. We observed in Section 2.5 that We can turn either the real numbers ℝ or the
ℤn is technically not a subset of ℤ but a quotient rational numbers ℚ into a multiplicative group,
set. The same is true of the groups: (ℤn, +​) is not however, in the most straightforward way imag-
a subgroup of (ℤ, +​) but a quotient group, a con- inable: by omitting the number 0. To verify that
cept to be defined formally in Chapter 6. (Even (ℝ≠0, ∙) is a group requires recognizing that the
if we contrived to make ℤ12 into a subset of ℤ product of two nonzero real numbers is always
by identifying the mod-​ 12 integers with the a nonzero real number, and that if x is a non-
ordinary integers 0, …, 11, it would still not be zero real number, so is x−1 =​1/​x. The same state-
a subgroup because the group operations are dif- ments remain true if the word real is replaced
ferent: 8 +​7 =​3 is a true statement in ℤ12 but a by rational (if x =​ a/​b, then x−1 =​ b/​a), so (ℚ≠0, ∙)
false statement in ℤ.) Note also that (ℤ7, +​) is not is a multiplicative group as well. Multiplicative
a subgroup of (ℤ12, +​): even though {0, 1, …, 6} ⊂ groups can also be formed using positive real
{0, 1, …, 11} is a true statement about subsets of or rational numbers only; in fact these groups
ℤ, integers mod n are not ordinary integers, and (ℝ>0, ∙) and (ℚ>0, ∙) are sometimes more useful
integers mod 7 are not integers mod 12—​and, than the larger groups (ℝ≠0, ∙) and (ℚ≠0, ∙).12
again, the group operations are different. The integers, with or without 0, do not form
a group under multiplication, because the multi-
Exercise 5.2.5 In Exercise 5.1.11d we observed plicative inverse 1/​x of an integer x is not in gen-
that only one element of the group ℤ is its own eral an integer. The inverse and closure axioms
inverse, namely 0. That statement is not true of ensure that if a multiplicative group contains
every group. Which element of ℤ12, besides 0, is all the positive integers, then it must contain all
its  
own inverse? the positive rational numbers as well, since all
such numbers may be obtained by multiply-
At the end of the discussion of modular arith- ing positive integers and their inverses. There
metic in Section 2.5 we noted the possibility of is, however, one nontrivial example of a finite
constructing a quotient set ℝ12, the real num- multiplicative group consisting of integers: the
bers mod 12. In fact, (ℝ12, +​) forms a group, an group {1, −1}, a multiplicative group of order 2.
example of a continuous modular group. In ℝ12 the The reader should have little trouble seeing that
equation 8.6 +​4.7 =​1.3 is a true statement; this {1, −1} is closed under multiplication and that
group arises musically as the interval group of each of the two elements is its own inverse.
continuous pitch-​class space cpc. Simple as it is, this group turns up with some reg-
We have noted that multiplicative notation ularity in musical applications. It is sometimes
such as x ∙ y or xy is commonly used in group convenient to write its elements simply as signs +​
theory, but we have yet to give any examples of and − rather than the numbers 1 and −1, with the
true multiplicative groups—​groups whose opera- understanding that signs multiply by the rules
tion is ordinary multiplication. Multiplicative +​+​=​−− =​+​and +​− =​−+​=​−; we will occasionally
groups, the last main category of familiar refer to this group as 𝒮𝑖𝑔𝑛, the sign group.13
numerical groups to be discussed in this section,
are in some respects prototypical of groups in Exercise 5.2.6
general. Multiplication, like addition, is com- (a) Do the negative real numbers form a
mutative as well as associative. The identity ele- multiplicative group? In other words, is
ment of a multiplicative group is not 0 but 1, (ℝ<0, ∙) a group?

Interval Groups and Transformation Groups • 177

(b) Do the real numbers form a group under 12). But the number 4 has no inverse: 4 ∙ x is
subtraction? In other words, is (ℝ, −) a always 4, 8, or 0 for every mod-​12 integer x, and
group?14 can never be equal to 1 (the identity).
(c) Do the positive real numbers form a group In general, an integer k has a multiplicative
under division? In other words, is (ℝ>0, ÷) inverse in ℤn if and only if the numbers k and n
a group? are coprime, which (as defined in Exercise 2.5.5)
(d) Do the positive real numbers form a group means that they have no common factors larger
under exponentiation? In other words, is than 1. This is consistent with the observa-
  (ℝ>0, ^) a group, where x ^ y denotes xy? tions in the preceding paragraph: 5 and 12 are
coprime, but 4 and 12 are not. The mod-​n inte-
While the preceding exercise suggests that it gers that are coprime to n are called units mod
may not be appropriate to consider subtraction n, and the set of all units mod n is denoted ℤn×.
as a potential group operation in its own right, Because a group element must always have an
subtraction nevertheless arises naturally in addi- inverse, any multiplicative group that is a sub-
tive groups. This is because subtraction amounts set of ℤn must be a subset of ℤn×. It turns out
to adding one element to the additive inverse that ℤn× itself is always a group, the units group
of another: that is, x − y =​ x +​(−y). The group mod n, and is therefore the largest possible mul-
axioms therefore imply that an additive group is tiplicative group within ℤn. The superscript ×
closed under subtraction as well as under addi- reminds us that ℤn× is a multiplicative group,
tion. Likewise division arises in multiplicative not to be confused with the additive group ℤn.
groups. In a group, if we can add, we can auto- The number of units mod n is often denoted
matically subtract as well, and if we can multiply, φ(n), so ℤn× is a group of order φ(n); the func-
we can also divide. tion φ, is known as Euler’s phi function, has many
In Figure 1.1.5 and again in Section 2.3 we interesting properties. The following exercise
encountered frequency space freq, in which shows that φ(12) =​4 while φ(7) =​6, so that even
intervals are calculated not by counting semi- though ℤ7 is a smaller set than ℤ12, ℤ7 has the
tones but by dividing frequencies. Frequencies larger units group.
are always positive numbers, and the quotient of
two frequencies can be any positive number, so Exercise 5.2.7
the interval group of frequency space is exactly (a) The units mod 12 are 1, 5, 7, and 11; that
the multiplicative group (ℝ>0, ∙). While continu- is, ℤ12× =​{1, 5, 7, 11}, so the cardinality of
ous pitch space cpitch and frequency space ℤ12× is φ(12) =​4. What is the multiplicative
freq contain exactly the same notes, their inter- inverse of 7? Of 11? Verify that the
val groups (ℝ, +​) and (ℝ>0, ∙) may appear very product of any two elements of ℤ12× is
different: the first is an additive group while the an element of ℤ12×. Also verify that the
second is multiplicative. Structurally, however, numbers 8 and 9 have no inverses mod 12,
these two groups turn out to be isomorphic—​ and observe that each of these numbers
another example of the power of group theory to shares some factor (larger than 1) with the
reveal likenesses between apparently dissimilar number 12.15
situations. A formal definition of isomorphism (b) If p is a prime number, then every nonzero
of groups, and further discussion of the isomor- integer mod p is automatically coprime
phism between (ℝ, +​) and (ℝ>0, ∙), will appear in to p and hence a unit mod p; therefore
Section 6.3. ℤp× =​{1, 2, …, p − 1}, and φ(p) =​ p − 1.
Having now seen examples of additive Calculate the multiplicative inverse of each
groups, multiplicative groups, and additive mod- of the six elements of ℤ7×.
ular groups, the reader may be wondering about (c) What are the units mod 9? Calculate the
the possibility of multiplicative modular groups.   inverse of each of them.
Integers mod n can be multiplied; does ℤn, or
some subset of ℤn, form a group under multipli- 5.3 ABSTRACT GROUPS;
cation? The answer is yes, but the appropriate FURTHER PROPERTIES
subset is perhaps not obvious. In general, some
integers mod n have multiplicative inverses,
OF GROUPS
while others do not—​and the exceptions are not The elements of the groups considered in the
limited to the number 0. In ℤ12, for example, the preceding section are ordinary numbers or num-
number 5 is its own inverse, since 5 ∙ 5 =​1 (mod bers mod n, but the real power of group theory

178 • E x p l o r i n g M u s i c a l S p a ces
stems from the fact that group elements can be A property that merits a bit more discus-
many other kinds of objects besides numbers, sion is the solvability of algebraic equations
and that the same algebraic structures often in groups. In basic algebra we learn to solve
turn up in many different kinds of situations. equations such as x +​3 =​8 or 9x =​−36 for the
Harnessing this power requires adopting a more unknown x, by subtracting 3 from both sides
abstract notation than the numerical nota- of the first equation or dividing both sides of
tion that might be appropriate in some specific the second by 9. A similar technique works in
group. This is why group elements are generally any group. Suppose that (G, ⊗) is a group and
denoted in the abstract by letters such as a, b, … a and b are known elements of G, and suppose
or x, y, …: whether the letters stand for numbers we wish to find an element x of G satisfying the
or something else, we can deduce some things equation a ⊗ x =​ b. Intuitively we would like to
about their behavior that apply in any case. We “cancel” a somehow, obtaining an equation for
may even define groups whose elements are let- x, perhaps something like “b divided by a.” The
ters such as x, y, …: a set of abstract algebraic general way to do this is to use the operation
symbols can form a group as long as we give a ⊗ to multiply both sides of the given equation
precise definition of a binary operation on that by a−1, the inverse of the element we want to
set and verify that the group axioms are satis- “divide” by. If
fied. In this section, therefore, we forgo the con-
creteness of the preceding examples and adopt a⊗x=b
this higher level of abstraction as we study fur-
ther properties of groups. is true, then, multiplying both sides on the left
A number of elementary properties are direct by a−1, the equation
consequences of the group axioms and therefore
hold true in every group. A few of these proper-
ties are listed below without much elaboration. a–1 ⊗ (a ⊗ x) = a–1 ⊗ b
Interested readers may attempt to derive these
properties from the axioms, or may consult a must be true as well. By the associative property,
text such as Dummit and Foote 2004 for details: a−1 ⊗ (a ⊗ x) is equal to (a−1 ⊗ a) ⊗ x, so

• Uniqueness of the identity. The identity element (a–1 ⊗ a) ⊗ x = a–1 ⊗ b.

of a group is unique. In any group, if an equa-
tion x ⊗ y =​ x or y ⊗ x =​ x is true of any group But by the general property of inverses, we know
elements x and y, then y must be e, the iden- that a−1 ⊗ a is the identity e; therefore
tity element of the group.
• Uniqueness of inverses. The inverse of every
group element is unique. If an equation x ⊗ y e ⊗ x = a–1 ⊗ b.
=​ e or y ⊗ x =​ e holds for any group elements x
and y, then y must be x−1, the inverse of x. Finally, by the property of the identity element,
• Reciprocal property of inverses. If y is the e ⊗ x =​ x, so the equation reduces to
inverse of x, then x is the inverse of y. In other
words, (x−1)−1 is equal to x for every element x x = a–1 ⊗ b.
of every group. Inverses come in pairs—​except
when they come singly. The identity is always We have solved the original equation for x.
its own inverse, and recall from Exercise 5.2.5 The reasoning by which we derived this solu-
that other elements may have this property tion made use of all of the group axioms: associa-
as well. A non-​identity element that is its own tivity, inverses, and the identity were mentioned
inverse is called an involution. That is, an invo- explicitly, and closure was needed in order to
lution is a group element x such that x ≠ e but know that we could multiply by a−1 and still have
x ⊗ x =​ e.16 an element of the group. A group is therefore the
• Generalized associativity. For any elements x1, most general mathematical structure in which
x2, …, xk of a group, the product x1 ⊗ x2 ⊗ ⋯ this kind of algebraic calculation can be carried
⊗ xk is uniquely determined and does not out. In any group at all, an equation of the form
depend on the way the expression is brack- a ⊗ x =​ b always has a solution, and that solution
eted. Parentheses are never required when is unique: x =​ a−1 ⊗ b is the only x satisfying the
writing products of group elements. given equation.

Interval Groups and Transformation Groups • 179

 With practice one learns to perform such (c) Explain why the result of part (b) implies
manipulations quickly. Because groups in gen- the general property (a ⊗ b)−1 =​ b−1 ⊗ a−1 for
eral need not be commutative, however, care any elements a and b of any group. Note the
must be taken to multiply equations on the order reversal: the inverse of a ⊗ b is b−1 ⊗ a−1,
proper side. If in the above process we multiplied which in a noncommutative group may not
a ⊗ x by a−1 on the right rather than on the left,   be the same as a−1 ⊗ b−1.
we would obtain (a ⊗ x) ⊗ a−1. This expression,
often written more simply as axa−1, is called a Exercise 5.3.4 Clearly the identity element
conjugate of x; in a general group this conjugate e of any group satisfies e ⊗ e =​ e. Suppose a is
is not necessarily equal to x. (Conjugation will be some group element such that a ⊗ a =​ a. Explain
studied in more detail in Section 6.6.) The ele- why a =​ e must be true: the identity is the only
ments a and a−1 cannot be canceled unless they element
   satisfying this equation.18
are adjacent, and there is no way to make them
adjacent unless we know that at least one of Groups are often described in terms of one or
them commutes with x. The unique x that solves more generators. In the additive group (ℤ, +​), we
the equation a ⊗ x =​ b is x =​ a−1 ⊗ b, which is not have noted that the subgroup generated by the
necessarily the same thing as b ⊗ a−1. number 3 is the infinite group 3ℤ =​{…, −6, −3,
Similar reasoning, often combined with the 0, 3, 6, …} of all multiples of 3. Here “the group
generalized associative property, may be used generated by 3” means “the smallest possible
to solve more complex algebraic equations. For group containing the number 3”; as observed in
example, given the equation a−1 ⊗ b =​ x ⊗ c−1 with Section 5.2, the closure and inverse axioms force
known group elements a, b, and c and unknown any such group to include all other multiples of
x, we multiply on the right by c and cancel c−1 ⊗ c 3, both positive and negative. In the group ℤ12,
to obtain x on the right side of the equation, however, things are different: because 3 +​3 +​3
equal to a−1 ⊗ b ⊗ c on the left. Because of asso- +​3 =​0 in this group, the subgroup generated by
ciativity, parentheses are unnecessary when the element 3 is here a finite group, namely {0, 3,
writing such a product. 6, 9}. The reader should check that this is indeed
a subgroup of ℤ12. Musically, this group can be
Exercise 5.3.1 interpreted as the interval group of a dimin-
(a) Use reasoning similar to the above to show ished seventh chord in pitch-​class space (any of
how to solve the group equation x ⊗ a =​ b the three diminished seventh chords will do):
for the unknown x. (This equation differs this means that the directed pitch-​class interval
from the first example above in that x between any two notes of such a chord is always 0,
appears to the left of a rather than to the 3, 6, or 9. This group in some ways resembles the
right.) group ℤ4 of integers mod 4; in fact, {0, 3, 6, 9} and
(b) Solve the group equation a ⊗ b =​ c ⊗ x ⊗ d ℤ4 are another example of isomorphic groups.
  for the unknown x.
Exercise 5.3.5
Exercise 5.3.2 Solve the following equations in (a) What is the subgroup of ℤ12 generated by
the specific groups from Sections 5.1 and 5.2 as the number 2? By the number 4? By the
indicated. Also give musical interpretations of number 6? These are all finite groups, so
each equation.17 you should be able to list all their elements.
Which of these three groups is the smallest?
(a) In (ℤ, +​), find x satisfying 24 +​x =​5. The largest? These groups are the interval
(b) In (𝒯∞, ●), find Tn satisfying T−5 ● Tn =​ T4. groups of what subspaces of pc?
(c) In (ℤ12, +​), find x satisfying 6 +​x =​3. (b) What is the subgroup of ℤ12 generated by
(d) In (ℤ7, +​), find x satisfying 6 +​x =​3. the number 8? It is the same as one of the
(e) In (ℝ>0, ∙), find x satisfying 3 2 ∙ x =​2. subgroups from part (a) above, demonstrating
(f) In (ℤ12×, ∙), find x satisfying 7 ∙ x =​11. that it is possible for two different elements to
   (ℤ7×, ∙), find x satisfying 4 ∙ x =​3.
(g) In generate the same group.
(c) What is the subgroup of ℤ12 generated by
Exercise 5.3.3   the number 1? By the number 5?19
(a) Suppose a and b are elements of a group and
x =​ a ⊗ b. What is x ⊗ b−1? The concept of group generators extends to
(b) In the same situation, what is x ⊗ b−1 ⊗ a−1? general (not necessarily additive) groups, but

180 • E x p l o r i n g M u s i c a l S p a ces
some new notations are appropriate. Suppose group G, so it is possible that we started with
(G, ⊗) is a group and a ∈ G. It is common to write a cyclic group in the first place.) The modular
a2 for a ⊗ a, a3 for a ⊗ a ⊗ a, and so on, using group (ℤ12, +​) is an example of a cyclic group of
the power notation familiar from ordinary mul- order 12, generated additively by the number 1.
tiplication. Because of generalized associativity, More generally, a cyclic group of order n is often
no parentheses are needed when writing these denoted 𝒞n. In the situation just described,
expressions, and there is no ambiguity about therefore, H =​{e, a, a2, …, an − 1} =​ 𝒞n.
what an can mean. The closure axiom ensures If an is the smallest power of a that equals
that if a ∈ G, then every power an also belongs the identity, then we say that a is an element of
to G. We also write a−n for the powers of a−1: thus order n. For example, in the cyclic group (ℤ12, +​),
a−2 =​(a−1)2, a−3 =​(a−1)3, and so on. The notation a0 1 is an element of order 12 (because 1 must be
always means the identity element e. added to itself 12 times to produce the identity
0), while 3 is an element of order 4 (because 3 +​
Exercise 5.3.6 Each of the following 3 +​3 +​3 =​0 mod 12). If a is an element of order
expressions is equal to an for some n (which n in any group, then a automatically generates a
may be positive, negative, or 0). Identify the cyclic subgroup of order n: the order of the ele-
appropriate power of a in each case.20 ment a (defined in this way) is the same as the
order of the group that it generates (defined as
(a) (a2)−1 =​ _​_​_​_​ the cardinality of the group). In a finite group G,
(b) (a−2)−1 =​ _​__​ ​_​ the order of every element a must be less than or
(c) a3 ⊗ a2 =​ _​__​ ​_​ equal to the order of G: “equal” if a generates the
(d) (a3)2 =​ _​__​ ​_​ entire group G, “less than” if a generates a proper
(e) a5 ⊗ a−2 =​ _​_​_​_​ subgroup of G.
(f)  (a−2)3 ⊗ a6 =​ _​__​ ​_​
Exercise 5.3.7
If a is an element of the additive group (ℤ, +​), (a) In (ℤ12, +​), in light of the results of Exercise
then the element referred to above as “a2 ” is 5.3.5, what is the order of the element 2? Of
actually a +​ a, the integer commonly called 2a, the element 4? Of 5? Of 6?
and “an ” means na. If a is any nonzero element (b) In chromatic pitch-​class space, start with
of ℤ, then “a,” “a2,” “a3,” … (the numbers a, 2a, any pitch class of your choice and generate a
3a, …) are all different. In a modular group such cycle by moving repeatedly through interval
as ℤ12, however, the number of elements is 2 (a major second)—​for example, C, D, E, F♯,
finite, so duplications must eventually appear … . How many different pitch classes are in
in this chain. In (ℤn, +​), na will always equal 0 the complete cycle? What if the generating
no matter what a is. interval is 3, 4, 5, or 6 instead of 2? How are
In the general setting, therefore, if a is an ele- these questions related to those in (a)?
ment of a group G, it may or may not happen that (c) As noted in Exercise 5.3.5b, a cyclic group
some an is equal to the identity e. If so, it follows may have more than one possible generator;
that an + ​1 =​ a, an + ​2 =​ a2, and so on; the sequence that is, it is possible that 〈a〉 =​ 〈b〉 for two
of powers repeats exactly from that point on.21 different elements a and b. In fact, four
Also, because an−1 ⊗ a =​ an =​ e, it follows that a−1 different elements of ℤ12 are of order
=​ an−1, and likewise a−2 =​(a2)−1 =​ an−2, a−3 =​(a3)−1 12, which means that any one of them
=​ an−3, and so on. In this case, the set H =​{e, a, generates ℤ12. Which elements are these?
a2, …, an−1} contains all possible powers of a and How does this property explain the fact
their inverses, and the exponents 0, 1, 2, …, n − 1 that the only possible circular arrangements
may be regarded as integers mod n rather than of pitch-​class space generated by a single
ordinary integers. The set H is closed under the interval are the chromatic circle (Figure
group operation and the formation of inverses, 1.2.3) and the circle of fifths? (Compare
ensuring that H is a subgroup of G. H is called Exercises 1.3.13, 5.2.7, and 5.3.5c.)
the group generated by the element a; the angle-​ (d) Now consider ℤ7, the integers mod 7. What
bracket notation 〈a〉 is often used to denote the is the subgroup 〈2〉 of ℤ7 generated by the
group generated by a. A group generated by a element 2? The subgroup 〈3〉? What do
single element in this way is called a cyclic group. these results imply about cycles of generic
(It is possible, of course, that H =​ 〈a〉 is the entire pitch classes generated by the generic
group G; we made no assumptions about the intervals 2 (“third”) and 3 (“fourth”)?

Interval Groups and Transformation Groups • 181

(e) Recall the multiplicative group ℤ7× from group 𝒯∞, in other words, is a specific example of a
Exercise 5.2.7b. In ℤ7×, what is the subgroup group in the class of infinite cyclic groups, any of
〈2〉 generated by 2? The subgroup 〈3〉? The which might be denoted 𝒞∞; the additive group
subgroup 〈6〉? Is ℤ7× a cyclic group?22 ℤ is another specific group in that same class. By
(f) Recall the multiplicative group ℤ12× from the same token, we write 𝒞n for an abstract cyclic
Exercise 5.2.7a. In ℤ12×, are there any group of order n, but some such groups may have
  elements of order 4? Is ℤ12× a cyclic group?23 other names. One such group is ℤn, the integers
mod n; also, we will write 𝒯12 for the transposi-
In an infinite group, as we have noted, the ele- tion group {T0, T1, …, T11} on pitch-​class space,
ments a, a2, a3, … may all be different. In this case a group to be explored in Section 5.5. The group
a is said to be an element of infinite order, and the 𝒯12, like 𝒯∞, is generated by T1. The equations
subset {e, a, a2, a3, …} of G is an infinite set. This 𝒯∞ =​ 〈T1〉 and 𝒯12 =​ 〈T1〉 are both correct; they
set is closed under the group operation (the gen- do not conflict because the two T1s technically
eral equation am ⊗ an =​ am +​ n remains valid), but it represent different transposition operators, one
is not closed under the formation of inverses. (If acting on pitch space and the other on pitch-​
a−1 were equal to some an, then an +​1 would equal e, class space.
so the order of a would not be infinite after all.)
The set {e, a, a2, a3, …} is therefore a semigroup Exercise 5.3.8
but not a group in this case. It can be made into (a) In the transposition group 𝒯∞ =​{…, T−2,
a group, however, by including the inverses as T−1, T0, T1, T2, …} on pitch space, what is
well: the set H =​{…, a−2, a−1, e, a, a2, …} is a sub- the subgroup 〈T2〉 generated by T2? The
group of G. In this case, H is called an infinite subgroup 〈T3〉? The subgroup 〈T−1〉?
cyclic group; as before, H is said to be generated (b) Let a be any positive real number. Describe
by the element a, and may be denoted 〈a〉. The the subgroup 〈a〉 of (ℝ, +​) generated by a.
notation 〈a〉 by itself does not tell us whether Does any element a generate all of ℝ? In
a group is finite or infinite. The word cyclic may other words, is (ℝ, +​) a cyclic group?
seem counterintuitive in the infinite case, since (c) Generally, every subgroup of a cyclic group
an infinite cyclic group does not form a cycle in is also cyclic. Suppose you know that H is
the same way that a finite cyclic group does. In a subgroup of (ℤ, +​) and that the smallest
group theory, the most general definition of a positive integer in H is 17. What is H?
cyclic group is this: any group, finite or infinite, (d) Even in an infinite group, it is possible for a
is cyclic if it is generated by a single element. non-​identity element to be of finite order;
The group (ℤ, +​) is therefore an infinite cyclic that is, it is possible that a ≠ e but an =​ e for
group, generated by the number 1; we may write   some n. Give an example.24
ℤ =​ 〈1〉. More generally, an infinite cyclic group
may be denoted 𝒞∞. It is important to remember Exercise 5.3.9
that an infinite group “generated” by an element (a) List all the subgroups of (ℤ12, +​). Arrange
a includes not only the positive powers of a but them in a subgroup lattice similar in concept
also their inverses, the negative powers of a—​it to the subset lattice of Figure 3.4.6a, clearly
includes, that is, all of the elements that must be showing which groups are subgroups of
present in any group in which a is present. In the which others.
situation just described, H =​ 〈a〉 =​{…, a−2, a−1, e, (b) The subgroup lattice in part (a) closely
a, a2, …} =​𝒞∞. resembles a divisibility lattice for the number
In Section 5.1 we introduced the notation 12, in which the elements are the integer
𝒯∞ for the group of transposition operators Tn divisors of 12, and in which x precedes y if
in chromatic pitch space. The evident similar- and only if x is a divisor of y. Construct this
ity between the sets 𝒞∞ =​{…, a−2, a−1, e, a, a2, …} divisibility lattice, and compare it with the
and 𝒯∞ =​{…, T−2, T−1, T0, T1, T2, …} suggests that ski-​hill lattice from Section 3.4, particularly
perhaps these two groups share a common struc-   the numerical version in Exercise 3.4.11.
ture. In fact, 𝒯∞ is an infinite cyclic group exactly
as described above, generated by the element T1. The notion of a group generated by an ele-
The powers of T1 are T12 =​ T2, T13 =​ T3, …, T1−1 =​ ment may be extended to a group generated by
T−1, T1−2 =​ T−2, …, and everything we have said two or more elements. Let us consider, for exam-
about the abstract group generated by a applies ple, the subgroup of (ℤ, +​) generated by the ele-
in this group generated by T1. The transposition ments 3 and 4—​a subgroup that may be denoted

182 • E x p l o r i n g M u s i c a l S p a ces
〈3, 4〉. If H is a subgroup of ℤ, and if 3 and 4 are In Section 5.1 we presented (ℤ, +​) as the
both in H, what else must be in H? The impor- interval group of chromatic pitch space. Several
tant thing to notice is that 4 − 3 =​1 must be in H groups in Section 5.2 were then described some-
(remember that additive groups are closed under what more informally as the interval groups
subtraction). Because the element 1 generates associated with various other musical spaces.
the entire group ℤ, it follows that every integer In this section we make this concept both more
can be written in some way as a sum of 3s, 4s, general and more precise, defining a kind of
and their inverses. Even though neither 3 nor 4 musical space called an interval space, and we
by itself generates ℤ, 3 and 4 together generate ℤ; show how this concept applies in some of the
we may write 〈3, 4〉 =​ ℤ. A musical consequence examples considered previously. Intuitively, an
of this statement is that every interval in chro- interval space is a musical space in which a viable
matic pitch space may be obtained through some notion of “interval” can be defined; the intervals
combination of ascending and descending major in a proper interval space are required to form
and minor thirds. The general principle is that some sort of group.26
two integers together generate ℤ whenever they The formal definition of an interval space
are coprime. Still more generally, the subgroup is axiomatic, like the definition of a group.
of ℤ generated by positive integers m and n is Generally, an interval space (S, G, int) consists of
always equal to the cyclic subgroup generated
by k, where k is the greatest common divisor of • a nonempty set S, often referred to as the
m and n. space itself;
• a group (G, ⊕), called the interval group of the
Exercise 5.3.10 Describe the subgroups of space; and
(ℤ, +​) generated by the listed elements. • a function int: S × S → G, called the interval
function of the space.
(a) The subgroup 〈3, 5〉 generated by 3 and 5.
(b) The subgroup 〈6, 10〉. Elements of the interval group G may be
(c) The subgroup 〈6, 15〉. called simply intervals, or, more fully, “the inter-
(d) The subgroup 〈10, 15〉. vals of the space S.” The notation int: S × S → G
(e)  The subgroup 〈6, 10, 15〉.25 means that the interval function is applied to
ordered pairs of elements of the space S; for each
x and y in S, int(x, y) is an element of the group G,
5.4 INTERVAL GROUPS called “the interval from x to y.” This notation is
AND INTERVAL SPACES familiar from the special case int: pitch × pitch
→ ℤ considered in Section 5.1; see Figure 5.4.1
In Section 5.2 we distinguished among some for a general picture.
broad categories of groups—​additive, modu- The set S, group G, and function int are
lar, and multiplicative groups—​on the basis required to satisfy the following two properties,
of the nature of the objects they contain and the interval-​space axioms:
the binary operations performed on those
objects. But there are other useful ways to cat- • Homogeneity. For every element x in the space
egorize groups. In musical contexts, groups S and for every interval i in the interval group
arise in two particularly significant ways: G, there exists one and only one element y in S
interval groups and transformation groups. The such that int(x, y) =​ i.
close correspondence between the two, already
foreshadowed in the discussion of the inter-
val group (ℤ, +​) and the transformation group
(𝒯∞, ●) in Section 5.1, is central to the work of
David Lewin. In the remainder of this chapter
we develop the concepts of interval groups and
transformation groups in greater generality
and present Lewin’s insight about the relation-
ship between them. These sections offer only
brief introductions to the respective types of
groups; both types will be explored at greater
length in Part Two. FIGURE 5.4.1 An interval in an interval space

Interval Groups and Transformation Groups • 183

• Interval-​sum equation. For any three elements most of the interval spaces we will study are
x, y, and z in S, int(x, y) ⊕ int(y, z) =​int(x, z).
commutative.
Readers who have studied vectors may notice
In pitch space, we know that a question such that the arrangement of arrows in Figure 5.4.2b
as “What note lies 15 semitones above F3?” suggests vector addition: if two vectors in the
always has a unique answer: whatever the given locations of the arrows labeled int(x, y) and int(y,
note and interval, there is always exactly one z) are added, the vector labeled int(x, z) is the
note that answers the question. The homogene- result.27 On the other hand, some readers, think-
ity axiom requires every interval space to have ing of distances on the page when looking at the
this property. In Figure 5.4.2a, given a point x same figure, may think that int(x, z) should not
in the space S and an interval i in the group G, equal int(x, y) ⊕ int(y, z) at all but should instead
there must always be some element of the space be something smaller: surely the straight-​line
at the location of the small circle—​at interval i distance from x to z is shorter than the path
from x. Every interval is available at every point from x to y to z. The latter intuition does not cor-
of the space; there cannot be an interval space respond with the behavior of interval functions,
in which certain intervals can be formed from but will apply in the context of distance functions,
some notes but not from others. Moreover, an to be studied in Chapter 12. The triadic distance
interval in the interval group can never take us measurements introduced in Chapter 4—​ the
outside the space. circle-​of-​fifths distance, Weber distance, and neo-​
The interval-​sum equation is illustrated in Riemannian distance—​are distance functions of
Figure 5.4.2b. This equation is familiar from this kind; none of these functions makes the set
Section 5.1, where it was discussed in the con- of major and minor triads into an interval space.
text of pitch space and its interval group (ℤ, +​). The reader should have no difficulty providing
In the general statement of the axiom, intervals examples to show that both the homogeneity
are combined using the group operation ⊕ in the axiom and the interval-​sum equation fail for all
group G, whatever that operation may be. three of these functions. (For example, if distweber
Intuitively, intervals are combined by “add- denotes the Weber distance, then distweber(C, E) =​
ing” them, which is why we are using a symbol 3 and distweber(E, f) =​4, but distweber(C, f) =​2,
that suggests addition in the context of interval which is certainly not equal to the sum 3 +​4.
groups, and why we call the second interval-​ Homogeneity would require a unique triad at a
space axiom the interval-​sum equation. In keep- Weber distance of 3 from C major, but in fact
ing with this choice of notation, we will write 0 there are six such triads.) Some possible interval
for the identity element of an interval group and structures on the space of triads will be consid-
⊖x for the inverse of the interval x. Many of the ered shortly, and relationships between distance
most important interval spaces (of which we will functions and interval functions will be consid-
list a few below, and more in subsequent chap- ered more broadly in Chapter 12.
ters) have as their interval groups some of the Two other simple but important properties of
additive or modular groups from Section 5.2, in interval spaces, consequences of the above axi-
which the group operation is in fact some form oms, involve the identity element and inverses
of addition, either ordinary addition or addition in interval groups:
mod n, and in which the notation −x is there-
fore appropriate. If the interval group G is com- • Identity interval. For every element x in any
mutative, we call S a commutative interval space; interval space, int(x, x) =​0: the interval from

FIGURE 5.4.2 (a) The homogeneity axiom and (b) the interval-​sum axiom in an interval space

184 • E x p l o r i n g M u s i c a l S p a ces
 any point of the space to itself is the identity In Section 5.1 we also introduced what we
element of the interval group. Moreover, if called interval configurations, such as the C-​major
int(x, y) =​0, then x =​ y. triad {C4, E4, G4} from Figure 5.1.1a. We can now
• Inverse intervals. For all elements x and y in define an interval configuration formally as any
any interval space, int(y, x) =​ ⊖int(x, y): the subset of an interval space. The triad is a subset
interval from y to x is the inverse of the inter- of the interval space pitch. The triad offered our
val from x to y. first illustration of the interval-​sum equation, but
by itself it is not an interval space, because it does
Both of these properties can be derived from not satisfy the homogeneity axiom. The note G4
the interval-​space axioms. We give the proof of is an element of the set, and intervals 3, 4, and 7
the identity interval property here, and leave the appear in the triad, so homogeneity would require
proof of the inverse interval property as an exer- the existence of elements y, yʹ, and yʹʹ such that
cise. Let x be any element in an interval space S. int(G4, y) =​3, int(G4, yʹ ) =​4, and int(G4, yʹʹ ) =​7; of
Then int(x, x) is some interval in G, the interval course there are no such notes in the given triad.
group of S; say int(x, x) =​ i ∈ G. The interval-​sum Generally an interval space must contain enough
equation tells us that int(x, x) ⊕ int(x, x) =​int(x, elements to reflect the complete homogeneous
x). (The general statement of the interval-​sum and symmetrical structure of the interval group—​
equation involves three variables x, y, and z, but a property not satisfied by smaller configurations
remember that it must hold for every possible like the triad. In fact, as discussed at the end of
choice of x, y, and z. In particular, it holds when Section 5.3, the integers 3 and 4 together gener-
x, y, and z are all replaced by the same x.) That is, ate the entire group ℤ; any interval space contain-
i ⊕ i =​ i. By Exercise 5.3.4, the only element of ing C4, E4, and G4 must therefore contain intervals
any group that satisfies this property is the iden- of every integer value, and so must include all of
tity; this establishes the fact that int(x, x) =​0. By chromatic pitch space.30
the homogeneity axiom, for each x in S there is We noted in Section 5.1 that all the intervals
a unique y in S such that int(x, y) =​0, so x itself in a three-​note configuration can be deduced
must be that unique y. from knowing two of them. Such observations
can be formulated in the general setting of inter-
Exercise 5.4.3 Use the interval-​space axioms val spaces. Suppose, for example, that x, y, and
and the identity interval property to prove the z are three elements of an interval space (S, G,
   interval property.28
inverse int). Suppose we know the values of int(x, y) and
int(x, z); how can we determine int(y, z)? This sit-
The definition of an interval space, like that of uation is illustrated in Figure 5.4.4. The unknown
a group, is essentially a construction in pure math- interval is not the sum of the other two, but the
ematics, and could be developed at some length interval-​sum equation does tell us that int(x,
with no mention of music.29 (The same could be y) ⊕ int(y, z) =​int(x, z). If we write int(x, y) =​ a
said of some of Lewin’s other abstractions, such and int(x, z) =​ b for the two known intervals and
as transformation graphs and networks as well as int(y, z) =​ i for the unknown interval, this equa-
Lewin’s theorem, presented in Section 5.6 below.) tion becomes a ⊕ i =​ b. This is precisely an equa-
One could conceive, for instance, of an inter- tion of the form considered in Section 5.3, where
val space modeling spatial relationships among it was written a ⊗ x =​ b. The general solution
physical objects in three-​dimensional Euclidean derived there is x =​ a−1 ⊗ b, which translates to
space, with a variety of possible non-​ musical i =​ ⊖a ⊕ b in the present situation. (Intuitively, i
applications. Of course, musical interpretations
are our primary interest here. The interval space
discussed in Section 5.1 is chromatic pitch space
pitch, whose interval group is (ℤ, +​) and whose
interval function int is the directed pitch interval
obtained by counting semitones between notes,
considering them positive if in the ascending
direction and negative if in the descending direc-
tion. For example, int(C4, G4) =​7, int(G4, C4) =​−7,
and int(C4, G5) =​19. We may write this interval
function as intpitch if necessary to distinguish it
from interval functions in other spaces. FIGURE 5.4.4 Solving an interval equation

Interval Groups and Transformation Groups • 185

is obtained by subtracting a from b; in the case of the chromatic scale, taking as its values not just
a noncommutative group, however, i is ⊖a ⊕ b, integers but all real numbers.
not b ⊕ ⊖a.) Similar techniques may be used to Chromatic pitch space pitch is a subspace
solve other more elaborate interval equations; of the continuous space cpitch—​a stronger
the fact that intervals form a group ensures that statement than saying merely that it is a subset.
such equations are generally solvable. Generally, if (S, G, int) is an interval space and Sʹ
is a subset of S, then to say that Sʹ is a subspace
Exercise 5.4.5 The interval-​sum equation for of S is to say that Sʹ, with the interval function
elements y, x, and z (in that order) tells us that that it inherits from S, forms an interval space
int(y, x) ⊕ int(x, z) =​int(y, z). Use this equation, in its own right. In order for this to be true, two
together with the inverse interval property, conditions must be satisfied: all the intervals
to give an alternate derivation of the result in formed by pairs of elements of Sʹ must form a
the preceding paragraph: if int(x, y) =​ a and subgroup Gʹ of G; moreover, the homogeneity
   z) =​ b, then int(y, z) =​ ⊖a ⊕ b.
int(x, axiom must hold in Sʹ relative to its own inter-
val group Gʹ. The latter statement means that for
Exercise 5.4.6 In the questions below, assume every x in the subset Sʹ and for every interval i
that x, y, z, and w are notes in chromatic in the subgroup Gʹ, there exists an element y in
pitch space. First draw an appropriate picture Sʹ such that int(x, y) =​ i. A unique such y always
for each example and answer the question exists in S because the interval space S satisfies
using algebraic reasoning as in the preceding the homogeneity axiom, but for Sʹ to be a sub-
paragraph; then choose a specific set of pitches space, this y must belong to Sʹ.
satisfying the given conditions and verify that We have already observed that these condi-
your answer is correct for those notes. tions are not satisfied in the case of an interval
configuration such as a triad, which is a subset of
(a) Suppose int(x, y) =​9 and int(x, z) =​13. pitch space but not a subspace. But both condi-
What is int(y, z)? tions are satisfied in the case of the subset pitch
(b) Suppose int(x, z) =​−4 and int(y, z) =​−7. of cpitch: intervals in pitch form the subgroup
What is int(y, x)? ℤ of ℝ, and if i ∈ ℤ, then the element at the inter-
(c) Suppose int(x, y) =​11, int(z, w) =​−3, and val i from any element of pitch is also an element
int(x, w) =​6. What is int(y, z)? of pitch. Other subgroups of (ℝ, +​), including
(d) Suppose int(x, y) =​7, int(y, z) =​−8, and subgroups of (ℤ, +​), give rise to other subspaces,
int(z, w) =​1. What can you conclude about which are therefore additional examples of real
  x and w? interval spaces. For example, a subgroup such as
3ℤ =​{…, −6, −3, 0, 3, 6, …} could be the inter-
Some of the other groups studied in Section val group of a pitch space such as {…, F♯3, A3, C4,
5.2 were described as the interval groups associ- E♭4, F♯4, …}, an infinite diminished-​seventh space
ated with various musical spaces. Now that we consisting of every third note of chromatic pitch
have formally defined interval spaces, we revisit space. All intervals within this subspace are mul-
some of these examples briefly. Many spaces tiples of 3, so the interval function defined on
will be mentioned in the remainder of this sec- this subspace takes values in the subgroup 3ℤ.
tion; readers who have difficulty in keeping them A general principle applies here, summarized
straight may benefit from occasional reference in the following theorem.
to the list in Appendix 1.
Several categories of interval spaces may be Theorem 5.4.7: Interval subspace property
distinguished, corresponding to the categories Suppose that (S, G, int) is an interval space, x0
of groups studied in Section 5.2. An interval is a fixed element of S, and Gʹ is a subgroup of
space whose interval group is the additive group G. Define a subset Sʹ of S consisting of those
of real numbers (ℝ, +​) or any subgroup of (ℝ, +​) elements x in S for which int(x0, x) ∈ Gʹ. Then
will be called a real interval space. The two obvious (Sʹ, Gʹ, intʹ ) is an interval space, where intʹ:
examples of real interval spaces are chromatic Sʹ × Sʹ → Gʹ is the same as the function int, but
pitch space pitch, whose interval group (ℤ, +​) is    only on the elements of Sʹ.
defined
a subgroup of (ℝ, +​), and continuous pitch space
cpitch, whose interval group is all of (ℝ, +​). The The space Sʹ described in the theorem is called
interval function intcpitch simply extends intpitch “the subspace of S generated by the element x0
to the notes in the cracks between the notes of and the subgroup Gʹ.” In the above examples,

186 • E x p l o r i n g M u s i c a l S p a ces
the diminished-​seventh space is the subspace of diminished-​seventh space generated by C4 and
pitch generated by the note x0 =​C4 and the sub- 3ℤ. What aspect of the relationship between
group Gʹ =​3ℤ of ℤ, and pitch itself is the sub- the notes C4 and E♭4 explains this observation?
space of cpitch generated by the note C4 and the In general, if x0 and x1 are two notes of an
subgroup ℤ of ℝ. interval space (S, G, int) and Gʹ is a subgroup
of G, under what conditions will the subspace
Exercise 5.4.8 This exercise outlines the two generated by x0 and Gʹ be the same as the
crucial elements of the proof of Theorem 5.4.7. subspace
   generated by x1 and Gʹ?

(a) In the situation described in the theorem, Exercise 5.4.10

how do we know that intʹ: Sʹ × Sʹ → Gʹ ? (a) Give an example of a subspace of pitch
That is, how do we know that when the whose interval group is 4ℤ =​{…, −8, −4,
function int is restricted to elements of the 0, 4, 8, …}. How many different such
subset Sʹ, it will always take values in the subspaces are there?
subgroup Gʹ ?31 (b) Give an example of a subspace of pitch
(b) It is clear that any three elements of Sʹ whose interval group is 12ℤ =​{…, −24, −12,
satisfy the interval-​sum equation, because 0, 12, 24, …}. How many different such
that axiom holds in the larger space S. But subspaces are there?
how do we know that (Sʹ, Gʹ, intʹ ) satisfies (c) In Exercise 5.2.3a you described a subspace
the homogeneity axiom? That is, if x ∈ Sʹ of cpitch whose interval group is 1 2 ℤ =​
and i ∈ Gʹ, how do we know that there exists {…, −2, −11 2 , −1, −1 2 , 0, 1 2 , 1, 11 2 , 2, …}.
  a unique y in Sʹ such that intʹ(x, y) =​ i?32 There is only one such space S for which
pitch is a subspace of S. Are there any such
The diminished-​seventh space {…, F♯3, A3, C4, spaces of which pitch is not a subspace?
E 4, F♯4, …} was described above as the subspace
♭   How many?
of pitch generated by the element x0 =​C4 and
the subgroup Gʹ =​3ℤ. If we chose x0 =​C♯4 rather As an example of a real interval space that
than C4, the generated subspace would be {…, is not a subspace of cpitch, recall generic pitch
G3, B♭3, C♯4, E4, G4, …}, a different diminished-​ space gpitch (Figure 1.1.6). gpitch is an inter-
seventh space. If pitches are numbered from val space in which intervals are measured by
C4 =​0, then the pitch numbers corresponding counting generic steps; for example, intgpitch(C4,
to the subspace containing C4 are …, −6, −3, 0, D4) =​1 (not 2), and intgpitch(C4, C5) =​7 (not
3, 6, …; these are the same numbers as the ele- 12). Because intervals in gpitch differ from
ments of the group 3ℤ. The pitch numbers cor- intervals between the notes of the same names
responding to the subspace containing C♯4, on in the spaces pitch and cpitch, gpitch is not
the other hand, are …, −5, −2, 1, 4, 7, …, which a subspace of either of those spaces. The inter-
do not form a group at all. Here it is important val group of gpitch is (ℤ, +​), the same as that
to keep in mind the distinction between the of pitch. In Section 7.2 we will define what it
elements of the space and the elements of the means for interval spaces to be isomorphic; it
group. The pitch numbers representing elements turns out that two interval spaces with the same
of the space may or may not form a group, but interval group are always isomorphic. The inter-
whether they do is irrelevant to the status of val spaces pitch and gpitch are therefore iso-
the interval space, which depends on the fact morphic, and neither is a subspace of the other.
that the intervals between those pitches form a Though gpitch is not a subspace of pitch or
group. Moreover, pitch numbering is arbitrary: of cpitch, elements of gpitch may be identified
the structure of the interval space and interval with certain elements of pitch, namely the ele-
group would be unaffected if we chose to num- ments of dpitch(0), the white-​key diatonic set.
ber pitches from C♯4 or any other reference pitch But dpitch(0), though it may be regarded as a
instead of C4. Interval numbering, in contrast, subset of pitch, is not a subspace either; in fact,
is not arbitrary: the interval from C4 to E♭4 is 3 dpitch(0), with the interval function it inherits
regardless of how the pitches are numbered. from pitch, is not an interval space at all. This is
because a diatonic scale is not homogeneous in the
Exercise 5.4.9 The subspace of pitch sense of the chromatic interval function: for exam-
generated by the note E♭4 and the subgroup 3ℤ ple, the elements E4 and F4 satisfy int(E4, F4) =​1,
is {…, F♯3, A3, C4, E♭4, F♯4, …}, exactly the same but no y in dpitch(0) satisfies int(F4, y) =​1.

Interval Groups and Transformation Groups • 187

 Another example of an interval space whose notes of an E-​major triad, and also by the
interval group is ℤ, and which is therefore iso- notes of an A♭-​major triad. Are the sums
morphic to both pitch and gpitch, is spelled the same for both triads? Are the sums the
pitch-​class space spc, in the form of the line of   same as those in part (a)?
fifths (Figure 1.3.3). Intervals between spelled
pitch classes are calculated by counting “steps”—​ Exercise 5.4.12
but these are steps along the line of fifths rather (a) Although the interval functions intpc and
than steps in a traditional scale. For example, intfifth are different, they give the same
intspc(C, G) =​1, intspc(G, C) =​−1, intspc(C, C♯) =​ 7, values in some cases: for example,
and intspc(C, B♯) =​12. intpc(C, D) =​intfifth(C, D) =​2. In fact,
A second main category of interval spaces if you choose pitch classes x and y at
consists of modular interval spaces, spaces whose random, there is a 50 percent chance
interval groups are modular groups like those that intpc(x, y) =​intfifth(x, y). Why is
discussed in Section 5.2. The canonical example this? For which pairs of pitch classes is
of a modular interval space is chromatic pitch-​ this equation true? In the cases in which
class space pc, whose interval group is (ℤ12, +​), this equation is not true, how are the two
the integers mod 12. The interval function intpc interval measurements related?
is the usual directed interval measurement of (b) Can you think of a simple mod-​12 equation,
pitch-​class set theory, which counts semitones valid in all cases, for calculating intfifth(x, y)
in the clockwise direction around the pitch-​class   in terms of intpc(x, y) or vice versa?33
circle (Figure 1.2.3). For example, intpc(C, G) =​7;
intpc(G, C) =​5 (which is equal to −7, the inverse An example of a modular interval space that
of 7 in the group ℤ12); and intpc(C, B) =​11 (which is not isomorphic to pc and fifth is generic
is equal to −1). pitch-​class space gpc. The interval space of gpc
Enharmonically conformed fifths space, is (ℤ7, +​), the integers mod 7, and the interval
defined by the circle of fifths (Figure 1.3.1), offers function intgpc counts steps around the generic
another example of a modular interval space. circle of Figure 1.2.8, so that, for example,
This space, fifth, consists of the same pitch intgpc(C, D) =​1 and intgpc(C, G) =​4. Because pc
classes as the space pc—​as sets, that is, fifth =​ and gpc are finite sets of two different cardinali-
pc—​and its interval group (ℤ12, +​) also matches ties (12 and 7), these two spaces cannot be iso-
that of pc. The interval function, however—​the morphic: chromatic and generic pitch space are
mapping from the space to the group—​is differ- isomorphic as interval spaces, but chromatic and
ent, as intervals in fifth are measured around generic pitch-​class space are not.
the circle of fifths. For example, intfifth(C, G) The continuous modular group (ℝ12, +​) was
=​1, intfifth(G, C) =​11, intfifth(C, C♯) =​7, and mentioned briefly in Section 5.2. This is the
intfifth(C, D) =​2. The spaces pc and fifth dem- interval group of continuous pitch-​class space
onstrate that it is sometimes possible to define cpc, which is an example of an infinite modular
two different interval structures on the same set, interval space.
thus giving rise to two different interval spaces. The third category of groups introduced in
Because the interval groups of pc and fifth are Section 5.2, after additive and modular groups,
also the same, however, the two interval spaces consisted of multiplicative groups. We noted
are isomorphic. The interval function intfifth there that the multiplicative group (ℝ>0, ∙)
on fifth (conformed fifths space) is evidently serves as the interval group of frequency space
related to the interval function intspc on spc freq. In frequency space, int(x, y) is the ratio
(unconformed fifths space), described above; fy / f​ x between the frequencies of the notes y and x,
this connection will be studied in Chapter 7. so for example the intervals formed by the notes
C4 and C5 are int(C4, C5) =​2 and int(C5, C4) = 1​ 2 .
Exercise 5.4.11 Elements of freq are pitches of specific fre-
(a) In the space pc, verify that the interval-​sum quencies, and the exact value of an interval such
equation is satisfied by the notes E, G♯, and as int(C4, G4) in this space depends on tuning.
B (the notes of an E-​major triad), and also If the perfect fifth is purely tuned, this inter-
by the notes of an A♭-​major triad. Are the val is 3 2 ; if it is equal-​tempered, it is a number
sums the same for both triads? very close to 3 2 , namely 27/​12 =​ s7, where s =​21/​12
(b) In the space fifth, verify that the (the twelfth root of 2), the frequency ratio of an
interval-​sum equation is satisfied by the equal-​tempered semitone.

188 • E x p l o r i n g M u s i c a l S p a ces
Exercise 5.4.13 Because frequency space is a classes of musical groups (along with interval
multiplicative interval space, the “interval-​sum groups). As a prototypical example of a transfor-
equation” in the second interval-​space axiom is mation group, recall from Section 5.1 the infinite
really an interval-​product equation in this space, cyclic group (𝒯∞, ●) of transposition operators Tn
namely int(x, y) ∙ int(y, z) =​int(x, z). Verify that acting on chromatic pitch space. Underlying this
this equation holds when x, y, and z are: example is the space pitch. The group 𝒯∞ con-
sists not of elements of the space or intervals
(a) the notes C4, E4, and G4, purely tuned; in the space but of functions Tn: pitch → pitch
(b) the same notes in equal temperament; and defined on the space. If x is a note in pitch and
(c) the same notes in the order E4, G4, C4, Tn is an element of 𝒯∞ (a transposition opera-
  purely tuned.34 tor), then Tn(x) is again a note in pitch. If x is
identified by a pitch number (relative to C4 =​0),
Continuous pitch space cpitch and frequency then Tn(x) is given by the formula Tn(x) =​ x +​ n.
space freq offer another example of two differ- The group operation in 𝒯∞ is ●, composition of
ent interval spaces defined on the same under- functions. As usual we use left-​to-​right orthog-
lying set. As was the case with pc and fifth, raphy, so Tm ● Tn means “Tm-​then-​Tn,” although
cpitch and freq turn out to be isomorphic for transposition operators Tn the order does
interval spaces. In a way that will be described not matter, because the group 𝒯∞ is commuta-
more precisely in Chapter 6, this isomorphism tive. We noted in Section 5.1 the general formula
follows from the isomorphism between the Tm ● Tn =​ Tm+​n for calculating the composition of
interval groups (ℝ, +​) and (ℝ>0, ∙) mentioned in two transposition operators.
Section 5.2, and ultimately from the exponen- We now abstract these observations to a
tial/​logarithmic relationship between pitch and more general setting. Suppose S is a nonempty
frequency from Figure 2.3.12. set (possibly one of the musical spaces we have
Students working with frequencies for the considered). As defined in Chapter 2, a transfor-
first time often make the mistake of subtract- mation on S is simply a function f: S → S. If x is
ing frequencies when they should be dividing. an element of S and f is a transformation on S,
Frequencies can be subtracted—​frequency dif- then f(x) is also an element of S. Two such trans-
ferences come into play in the acoustical phe- formations f and g can always be composed, and
nomenon known as difference tones—​but it is their composition f ● g (“f-​then-​g”) is another
the quotient of frequencies, not the difference, transformation on S. If G is a set whose elements
that corresponds to the usual musical notion of are functions f: S → S, it may happen G forms a
interval. Moreover, if we try to define an interval group with the operation ●. In this case (G, ●) is
function using frequency differences, we quickly called a transformation group on S, or a group act-
discover that musical pitches do not form a ing on S, and S is called the space associated with
proper interval space. To see this, consider the the transformation group G.35
frequencies x =​300 Hz and y =​100 Hz. Our sup- For a set of functions to form a group, the
posed interval function gives int(x, y) =​−200 group axioms must be satisfied. Associativity
Hz, so evidently −200 Hz is an element of the is not a problem; as noted in Section 5.1, com-
interval group. But then the homogeneity axiom position of functions is always associative.
implies the existence of some note z such that Composition is not in general commutative, but
int(y, z) =​−200 Hz, which implies that z must that is not a problem either, as groups are not
have a frequency of −100 Hz. Such a note does required to be commutative. But the other three
not exist, even in a theoretical sense. With inter- axioms—​closure, the identity, and inverses—​
vals defined in this way, the most we can say is all require some rather special things to be true
that musical pitches form a subset of a not very about the functions in G if they are to form
useful interval space whose other elements are a group.
musically fictitious. The closure axiom demands that if f and g are
elements of a transformation group G, then their
5.5 TRANSFORMATION composition f ● g must also belong to G. Any trans-
GROUPS AND GROUP formation group on pitch space that contains
the transpositions T4 and T3, for example, must
ACTIONS also contain their composition T7. Of course the
In this section we formally introduce transfor- group 𝒯∞ satisfies this requirement: if Tm and Tn
mation groups, the second of the two important are in 𝒯∞, so is Tm ● Tn =​ Tm+​n.

Interval Groups and Transformation Groups • 189

 The function that plays the role of the iden- can be expanded to a group, and a semigroup
tity element of G is the identity function ES on that is not a group will lack desirable group fea-
the space S, defined by ES(x) =​ x for all x in S. The tures such as symmetry and the solvability of
equations f ● ES =​ f and ES ● f =​ f hold for every algebraic equations.
function f: S → S, and these are precisely the In any case, there is no shortage of examples
conditions demanded of the identity element of musical transformation groups satisfying
of a group. For G to be a group, ES must there- all the above constraints, including the inverse
fore belong to G: every transformation group must axiom. Besides the group 𝒯∞ acting on chromatic
contain the identity function on the space on which pitch space, we may also consider 𝒯12 =​{T0, T1,
it acts. In the group 𝒯∞, the identity function is …, T11} as a transformation group acting on
T0 (which could also be called Epitch, the identity chromatic pitch-​class space. As we have noted,
function on the space pitch). the transformations T0, T1, …, T11 in 𝒞12 are not
We also observed in Section 5.1 that the the same as the transformations with the same
inverse of a function f in a group is the same names in 𝒯∞, because they are functions defined
as the inverse function f −1 defined in Chapter 2. on a different set; if we wish to distinguish them
Many functions do not have inverses: as detailed we may write T0pitch, T1pitch, … and T0pc, T1pc, …,
in Section 2.3, f −1 exists if and only if f is one-​ for example. Recall from Section 5.3 that 𝒯∞ is an
to-​one and onto. Every element of a group, how- infinite cyclic group and 𝒯12 a finite cyclic group
ever, is required to have an inverse in the group. of order 12; both are generated by T1 (though
Therefore every transformation f: S → S in a trans- these are two different T1s). The equations
formation group G on S must be one-​to-​one and Tn−1 =​ T−n and Tm ● Tn =​ Tm+​n hold in both groups,
onto, and its inverse function f −1 must also belong but in 𝒯∞ the subscripts are ordinary integers
to G. In 𝒯∞, every transposition operator Tn is while in 𝒯12 the subscripts are integers mod 12.
one-​to-​one and onto as a mapping on pitch space Thus, for example, the equations T8−1 =​ T4 and T9
● T =​ T are valid in 𝒯
(for every note y in pitch space there is exactly 5 2 12 but not in 𝒯∞.
one x such that Tn(x) =​ y), and the inverse of Tn is
T−n, which also belongs to 𝒯∞. Exercise 5.5.1
The inverse axiom is the most stringent of the (a) Calculate T7 ● T4−1 ● T11 ● T2−1 ● T3 in 𝒯∞,
constraints on the functions in a transformation and calculate the same composition in
group. Suppose we have a set A of functions f: 𝒯12. (Because composition of functions is
S → S, and suppose we would like to have a trans- associative, no parentheses are needed in
formation group. Every set automatically has an this expression. For practice, however, you
identity function, so if the identity function ES is may try grouping the operations in several
not already in the set A, it is a simple matter to different ways and verify that the result is
expand the set to contain it. The set A may fail the same.)
to be closed under composition of functions, but (b) The techniques for solving equations from
any two functions from S to S can be composed, Section 5.3 are valid in transformation
and once again we can expand A if necessary to groups (as they are in any other kind of
include all possible compositions (and iterated group). In 𝒯∞, what transformation f
compositions, if necessary) of functions in A. But satisfies the equation T10 ● f =​ T3?
the requirement imposed by the inverse axiom (c) In 𝒯12, what transformation satisfies the
is not so easily fulfilled, because some functions   equation T7 ● f ● T8 =​ T2?
simply do not have inverses. If we were to define
a transformation on pitch that fails to be one-​ Besides transpositions, the other most
to-​one or onto (for instance, defining f on pitch familiar transformations on pitch space and
numbers by f(x) =​2x, or f(x) =​ x2, or f(x) =​ 0—​a pitch-​class space are the inversion operators.36
constant function that maps every note to mid- In pitch space, we define I as inversion about
dle C), such a function could never be an element middle C. With pitch numbers centered on C4 =​0
of any transformation group. The concept of as usual, therefore, I(x) =​−x for every x in pitch.
semigroup, defined in Section 5.2, is potentially Then, for every integer n, we define a transfor-
useful in this case; recall that the identity and mation In by the equation In =​ I ● Tn. The sub-
inverse axioms are not required of semigroups. script n is commonly called the index number
The above remarks imply that any set of func- of the inversion operator In (or just the index of
tions f: S → S may be expanded to form a semi- inversion). Because T0 is the identity, I0 =​ I ● T0
group with identity. But not every semigroup implies that I0 is the same as I. The composition

190 • E x p l o r i n g M u s i c a l S p a ces
In =​ I ● Tn means that In(x) =​−x +​ n for every that if T3 ● I rather than I ● T3 is applied to G4
x in pitch. The latter formula may be rewrit- (7), the result is not A♭3 (−4) at all but rather D3
ten in the form In(x) =​ n − x; note that x (the (−10). If the transposition is performed first, its
pitch number) is subtracted from n (the index effect is reversed by the subsequent inversion.
of inversion), not the other way around. If Figure 5.5.2c shows, however, that the origi-
In(x) =​ y, then the equation y =​ n − x can be nal result A♭3 (−4) may be obtained by applying
rewritten as x +​ y =​ n, which tells us that any T−3 ● I instead of T3 ● I. This illustrates a general
two notes related by an inversion operator In principle of interaction between transposition
always sum to n, the index number. and inversion operators: the composition I ● Tn
The same equations, I(x) =​−x and In =​ I ● Tn, is not equal to Tn ● I but to T−n ● I. Any time we
may be used to define inversion operators on want to exchange the positions of an adjacent
chromatic pitch-​class space pc. As with transpo- I and Tn, we may do so by reversing the sign of
sition operators, we may if necessary write Inpitch the transposition. This principle applies both in
and Inpc to distinguish the operators on the two pitch space and in pitch-​class space.
spaces. The pitch-​ class inversion operators, Figure 5.5.2d shows that the effect of the
because of their widespread use in pitch-​class composite transformation I3 =​ I ● T3 can be
set theory, are probably more familiar to most obtained in a single step by moving the fulcrum,
readers than the corresponding transformations in this case reflecting the number 7 about the
on pitch space. Many aspects of the behavior point 3 2 on the number line rather than about 0.
of these transformations are similar on both In general, In may be regarded as inversion about
spaces, however, and in some ways the operators n/​2; the number n/​2 forms the axis of inversion
on pitch may be considered more fundamental. (or center of inversion) of the operator In. The
Figure 5.5.2a illustrates the calculation center of inversion, that is, is always half of the
I3(G4) =​A♭3 in pitch space. First, I inverts G4 (pitch index number. The number 3 2 is not an integer,
number 7) about the fulcrum at middle C, pro- of course, and does not represent an element
ducing F3 (−7); then T3 maps F3 to A♭3 (−4). The of the discrete space pitch. In general n/​2 is an
starting and ending pitch numbers 7 and −4 integer if n is even but not if n is odd, creating
sum to 3, the index of inversion. a distinction between an even-​indexed In, which
Left-​to-​
right orthography applies in the inverts about a note of chromatic space, and an
equation In =​ I ● Tn. As we noted in Chapter 2, odd-​indexed In, which does not. For every even
authors who follow the convention of right-​to-​ integer n, the operator In has one fixed point
left orthography would write the same equation in the space pitch: there is exactly one note x
as In =​ Tn ○ I instead—​and indeed the right-​to-​ in pitch satisfying In(x) =​ x, namely the note
left notation “TnI ” is common in pitch-​ class x =​ n/​ 2 that defines the axis of inversion.
set theory. Whatever the notation, it is crucial Inversion operators of odd index have no fixed
that I be performed before T3, as these transfor- points in pitch, although the fixed point at n/​2
mations do not commute. Figure 5.5.2b shows exists within the continuous space cpitch.

FIGURE 5.5.2 Inversions in pitch space

Interval Groups and Transformation Groups • 191

Exercise 5.5.3 The midpoint, or average, to the important observation that the transpo-
between any two real numbers x and y is sition and inversion operators together form a
given by the formula (x +​ y)/​2. Use this group. In pitch space this is an infinite group,
formula to verify that the midpoint
between
   x and In(x) is n/​2. 𝒯ℐ∞ =​{…, T−2, T−1, T0, T1, T2, …;
…, I−2, I−1, I0, I1, I2, …},
The reader may be familiar with the analo-
gous properties in pitch-​ class space, where while in pitch-​class space it is a finite group,
instead of inverting about a single point at
n/​2, the inversion In is a reflection about an axis 𝒯ℐ12 =​{T0, T1, T2, …, T11; I0, I1, I2, …, I11}.
that passes through both n/​2 and n/​2 +​6 (mod
12), the latter number falling diametrically The semicolons in the lists of group elements
opposite n/​2 on the pitch-​class circle. In mod-​ are “unofficial,” serving only as separators to
12 arithmetic, both the numbers x =​ n/​2 and signal that two different types of elements are
x =​ n/​2 +​6 satisfy the equation 2x =​ n. In pitch-​ being listed—​a property also suggested by the
class space, therefore, In has two fixed points if label 𝒯ℐ that we are using for these groups.38
n is even, the two points on the axis of inver- Generally a 𝒯ℐ group is twice as large as the
sion, differing by 6 mod 12. If n is odd, then corresponding 𝒯 group (although of course 𝒯∞
In has no fixed points in the discrete space pc. is already an infinite group). In particular, 𝒯ℐ12
The difference between even and odd indices is a group of order 24, not 12: it is called 𝒯ℐ12
of inversion is illustrated in Figure 5.5.4, both because it is a transformation group acting on a
in pitch space and in pitch-​class space. For the 12-​element set, the space pc.
odd indices, fixed points on the axes of inver- As noted in Section 5.3, the transposition
sion exist as points of the continuous spaces groups 𝒯12 and 𝒯∞ are examples of cyclic groups,
cpitch and cpc, so the qualitative difference as each is generated by the transposition T1; we
between even and odd indices that is so appar- also introduced the notations 𝒞n and 𝒞∞ for more
ent in the discrete spaces largely evaporates in general cyclic groups. The groups 𝒯ℐ12 and 𝒯ℐ∞
continuous space.37 are not cyclic, as neither of them is generated by a
single element. Each group is, however, generated
Exercise 5.5.5 by T1 and the inversion I (=​ I0); that is, we could
(a) In pitch, calculate I6(4), I13(−7), I−7(13), write 𝒯ℐ12 =​ 〈T1, I〉 and 𝒯ℐ∞ =​ 〈T1, I〉. As was the
I−1(E♭4), and I−12(F♯3). case with the transposition groups, these equa-
(b) In pc, calculate I6(4), I1(5), I5(1), I11(E♭), and tions look the same, but T1 and I have different
I0(F♯). Compare your answers to (a) and (b). meanings in the two equations. The 𝒯ℐ groups are
(c) In pitch, what inversion operator In examples of structures known as dihedral groups,
satisfies In(C5) =​B♭3? What note is the fixed for which we will use the general notations 𝒟n and
point of this inversion? 𝒟∞. The dihedral group 𝒟n is sometimes defined
(d) In pc, what inversion operator In satisfies as the group of all rotations and reflections of a
In(F) =​A♭? Does this inversion have any regular n-​sided polygon; the transpositions Tk are
fixed points in pc? In continuous pitch-​class the rotations while the inversions Ik are the reflec-
space cpc? tions. The order of the group 𝒟n is therefore 2n,
(e) What note in pitch is the unique fixed not n.39 Because transpositions do not commute
point of I26? with inversions, the dihedral groups 𝒯ℐ12 (or 𝒟12)
(f) Which two notes in pc are fixed points of I4? and 𝒯ℐ∞ (or 𝒟∞) are our first musical examples of
(g) In continuous pitch space cpitch, the noncommutative groups.
note E3+​0.5 (halfway between E3 and To be sure that 𝒯ℐ12 and 𝒯ℐ∞ are groups,
F3) is the fixed point of what inversion we must check the requirements listed previ-
operator In? ously: each of these sets of transformations
(h) In continuous pitch-​class space cpc, D+​0.5 must contain the identity function and must be
and G♯+​0.5 are fixed points of what inversion closed under the formation of compositions and
  operator In? inverses. Of course, the identity in each group is
T0 as before. The relevant properties of composi-
Many properties of inversion operators and tions and inverses are summarized in the follow-
their interaction with transpositions are tied ing theorem.

192 • E x p l o r i n g M u s i c a l S p a ces
FIGURE 5.5.4 Even and odd indices of inversion in discrete pitch space and pitch-​class space

Theorem 5.5.6: Group properties of At first glance the formulas may look com-
transposition and inversion operators plex and arbitrary, but they possess a certain
Statements (a)–​(g) below hold for all symmetry and logic. Statements (a) and (f),
transposition operators Tm and Tn and inversion which deal with transposition operators only,
operators Im and In on pitch space pitch or pitch-​ are already familiar to us. Parts (b) and (c)
class space pc. In pitch the subscripts m and n tell us that a transposition composed with an
are integers; in pc they are integers mod 12. inversion, in either order, always produces an
inversion, and part (d) tells us that a compo-
(a) Tm ● Tn =​ Tm+n; sition of two inversions produces a transposi-
(b) Im ● Tn =​ Im+​n; tion—​effectively, the space has been reflected
(c) Tm ● In =​ In−m; twice, returning to its original orientation. The
(d) Im ● In =​ Tn−m; subscripts m and n are added (m +​ n) in parts
(e) Tm ● In =​ In ● T−m; (a) and (b), but m is subtracted (n − m) in parts
(f) Tn−1 =​ T−n; (c) and (d); these are the cases in which the
(g)  In−1 =​ In (that is, In is an involution). second operation is an inversion, which (as
in the example from Figure 5.5.2 above) has
This theorem, which will be invoked many the consequence of reversing the effect of the
times in subsequent chapters, effectively tells us first operation. Part (e) generalizes the above
how to perform any kind of calculation involv- observation about exchanging the positions of
ing compositions and inverses of transposition transposition and inversion operators: adjacent
and inversion operators. Parts (a)–​(d) cover all operators Tm and In (not just I =​ I0, but any In at
possible compositions of T and I operators, and all) may always be interchanged as long as the
tell us that every such composition is always sign of the transposition is reversed at the same
another T or I, so the set 𝒯ℐ∞ or 𝒯ℐ12 is closed time. Finally, statement (g), stating that In is an
under composition of functions. Parts (f) and (g) involution, should be intuitively clear: each In
show that inverses of T and I operators are also T is a reflection about a certain point (n/​2), and
or I operators, so the set is closed under inverses when the same action is performed a second
as well. This is exactly what we need to know to time the effect of the first is undone. This prop-
conclude that 𝒯ℐ∞ and 𝒯ℐ12 are groups. erty is apparent also from Figure 5.5.4, where

Interval Groups and Transformation Groups • 193

the bidirectional arrows imply that a second The formulas in Theorem 5.5.6 may be used
application of In reverses the first. to simplify complex expressions involving trans-
It is instructive to derive the formulas in position and inversion operators. Deliberately
Theorem 5.5.6 directly from the formulas Tn(x) =​ taking a somewhat unwieldy example for illus-
x +​ n and In(x) =​ n − x. We do this below for state- trative purposes, in the group 𝒯ℐ12, let us sim-
ments (b), (d), (e), and (g), leaving (c) as an exercise. plify the composite function
Statement (b) involves a composition of the
form Im ● Tn. We therefore calculate what hap- I3 • T11 −1 • T8 • I5 • I1 −1 • I10 • T6 .
pens when we apply an inversion Im to a note
x, then apply a transposition Tn to the result: In the following calculation we omit the sym-
bols ● for simplicity; left-​to-​right orthography
x I
m
→ m − x T
n
→ (m − x ) + n. is always assumed. The theorem provides an
arsenal of tools with which to simplify such an
expression: we can use parts (f) and (g) to elimi-
The outcome of this process, (m − x) +​ n, is the nate the inverses; combine adjacent T operators
same as (m +​ n) − x, which is what we would get if using (a); combine adjacent I operators using (d);
we inverted x via the operator Im+​n. Therefore the and interchange adjacent T and I operators using
functions Im ● Tn and Im+​n give the same outputs (e) so that more Ts will be adjacent. Because we
on all possible inputs. By the discussion of equal- are in pitch-​class space, the subscripts are inte-
ity of functions in Section 2.3, it follows that they gers mod 12, and so, for instance, T11−1 =​ T1.
are the same function: Im ● Tn =​ Im+​n, as (b) asserts. Here is one possible strategy, which proceeds by
For (d), we apply two inversion operators in first eliminating the inverses, then eliminating
succession: as many inversion operators as possible:

I3T11−1T8I5I1−1I10T6
x I
m
→ m − x I
n
→ n − (m − x ).
=​ I3T1T8I5I1I10T6    eliminating inverses
via (f) and (g)
The result, n − (m − x), is equal to x +​(n − m), =​I 3(T1T8)(I5I1)I10T6 grouping Ts and Is
the same as if we had transposed x by Tn − m. to combine
Therefore Im ● In =​ Tn − m. =​I 3T9T8I10T6      combining them via
Statement (e) compares two compositions, (a) and (d)
one of the form T-​then-​I and the other of the =​ I3T5I10T6      combining another
form I-​then-​T. We can calculate the former using pair of Ts
part (c) (proved in Exercise 5.5.7 below), and =​ (I3T5)I10T6      grouping elements
the latter using (b). The left side of (e) is Tm ● In, to exchange
which according to (c) is equal to In−m. The right =​(T7I3)I10T6      exchanging them via (e)
side of (e), meanwhile, is In ● T−m. This is the type =​ T7(I3I10)T6      regrouping
of composition appearing in (b) (I-​then-​T), but =​ T7T7T6      combining two Is
the subscripts are different, so to evaluate it we =​ T8        combining all Ts
must substitute n for m and −m for n in state-
ment (b). When we do this, the result is In ● T−m In the first “combining” step, T1T8 =​ T9 by (a),
=​ In+​(−m). This is equal to In − m, which as we just while I5I1 =​ T1 − 5 =​ T8 by (d) (the subtraction in
saw is the same as Tm ● In. Thus the truth of (e) the subscripts is mod 12). In the “exchanging”
follows from (b) and (c). step, T5 is replaced by its inverse T7 when it is
Statement (d) applies to any inversions Im and swapped with the adjacent I3. At the final step,
In. In particular, m and n could be the same. In that the three subscripts on the adjacent T operators
case the formula becomes In ● In =​ Tn − n =​ T0. In are added mod 12. Of course, the calculation
other words, In is its own inverse, as part (g) states. could have proceeded in various other ways, such
as by combining operations in strict left-​to-​right
Exercise 5.5.7 Using reasoning similar to the order. The original expression contains an even
above,
   derive statement (c) of Theorem 5.5.6. number of inversion operators, so we may see at
a glance that the final result must be a transposi-
Exercise 5.5.8 Do the inversion operators by tion, not an inversion; in the calculation above
themselves
   form a group? Why or why not? all inversions are ultimately eliminated in pairs.

194 • E x p l o r i n g M u s i c a l S p a ces
Exercise 5.5.9 Verify that the above calculation a main theme of the book. The basic point is that
is correct when the composite transformation an arrow can be interpreted in two conceptually
in question is applied to the pitch class F distinct ways: as an interval, a measure of the dis-
(5). That is, calculate the result of applying tance separating a point x from a point y, or as a
I3T11−1T8I5I1−1I10T6 to x =​5, proceeding left to transformation, an action that one must perform
right through the string of transformations, to get from x to y.
and verify that the result is the same as We noted this point previously, in Section
applying
   the single transformation T8. 5.1. Our knowledge of group theory now enables
us to grasp Lewin’s insight more completely
and to probe the connection at a deeper level,
Exercise 5.5.10 because groups lie at the core of the connection
(a) In the group 𝒯ℐ12, what is T4−1I7T3I10−1T2? between the intervallic and transformational
(b) What is the same composite function conceptions. Interval groups, it turns out, are
(T4−1I7T3I10−1T2) in the group 𝒯ℐ∞?40 always functionally equivalent to transforma-
(c) Explain why the equation T6In =​ InT6 is tion groups of a certain kind, and conversely, a
always correct in 𝒯ℐ12 but never correct transformation group of that kind may always
in 𝒯ℐ∞. be used to define an interval-​space structure on
(d) In 𝒯ℐ12, what transformation f satisfies the the space on which it acts.
equation I4 ● f ● T7 =​ T2 ● I3? The way to convert an interval group into a
(e) In 𝒯ℐ∞, what transformation f satisfies the transformation group is suggested in a straight-
  equation I7 ● f ● I7 =​ I3? forward way by reconsidering Figures 5.1.1a and
5.1.7, which showed intervallic and transposi-
Cyclic groups of transformation operators tional relationships among the notes of a triad.
and dihedral groups of transposition and inver- Given that the interval from E4 to G4 is 3, how
sion operators are some of the most important can we define a transformation that takes us
and widely used musical transformation groups, from E4 to G4? The appropriate transformation
but many other kinds of transformations can is T3, transposition through the given interval.
appear in groups as well. Familiar examples This process, it turns out, is completely general:
include the multiplication operators M5 and we can define a transposition operator corre-
M7 on pitch-​class space, the retrograde opera- sponding to any interval in any interval space,
tion in serial theory, and the neo-​Riemannian and the various transposition operators form a
transformations P, R, and L on triads. We will transformation group on the space—​not just a
study groups containing such transformations transformation group but a transposition group.
in Part Two. The broad scope of this statement is notewor-
thy. We are accustomed to thinking of transposi-
5.6 THE RELATION tion operators on familiar pitch and pitch-​class
BETWEEN INTERVALS spaces, but suddenly we are claiming that it is
somehow possible to define “transposition” on
AND TRANSFORMATIONS any interval space at all—​a space whose inter-
Figure 5.6.1, a simple arrow from a point x to val group may be any sort of group at all—​and
a point y, captures the essence of the first and that these generalized transpositions also form
most famous figure in David Lewin’s Generalized a group. To see how transposition operators may
Musical Intervals and Transformations. This be defined in such a general setting, suppose (S,
unpretentious diagram appears on the first page G, int) is an interval space and i is an interval in
of Lewin’s introduction41 and gives the first G; we want to define a transformation we will call
indication of the intimate connection between Transi on S that “transposes” any element of the
intervals and transformations that is to become space S through the interval i. In the case of the
familiar pitch and pitch-​class spaces, Transi will
be the same as Ti. We reserve the notations Ti,
ti, and τi for transpositions in particular spaces,
writing Transi in the more general case.42
If the transposition Transi behaves as we
should expect, then presumably the interval
FIGURE 5.6.1 An interval—​
or a transforma- from any x in S to its transposition Transi(x)
tion? (after Lewin) should be i, the interval of transposition.

Interval Groups and Transformation Groups • 195

That is, if y =​Transi(x), we should expect that 𝒯S. We still must verify that Transi ● Transj is
int(x, y) =​ i. Is there actually such a y, an element equal to Transi ⊕ j. If x ∈ S, what element do we
of S for which int(x, y) =​ i? In fact there is: the get by applying Transi ● Transj to x? We know
homogeneity axiom for interval spaces says that from the definition of Transi that Transi(x) is the
there is exactly one such i. So here is our defini- unique y in S satisfying int(x, y) =​ i; then Transj(y)
tion of generalized transposition operators: if is the unique z in S satisfying int(y, z) =​ j. So Transi
● Trans applied to x yields z, and by the interval-​
(S, G, int) is an interval space and i ∈ G, then j
Transi: S → S is the function defined so that, for sum equation in the interval space S, int(x, z) =​
every x in S, Transi(x) is the unique y in S such int(x, y) ⊕ int(y, z) =​ i ⊕ j. But by the definition
that int(x, y) =​ i. of Transi ⊕ j, this same z must also be equal to
This definition of Transi is straightforward Transi ⊕ j(x), which by definition is the unique z
enough, and it is clear that every such Transi is such that int(x, z) =​ i ⊕ j. It follows that Transi ●
a transformation on the space S, but it may be Transj and Transi⊕j are indeed the same function.
far from obvious that all of these transpositions Finally, we must check that if Transi is a trans-
form a group. Crucial to this conclusion is the fact position in 𝒯S, then its inverse Transi−1 is also
that G, the interval group of the space, is already in 𝒯S. Intuitively we might expect Transi−1 to
known to be a group (this is part of the defini- be Trans−i, but the general formula is Transi−1 =​
tion of an interval space). Let 𝒯S denote the set Trans⊖i, where, as in Section 5.4, ⊖i denotes the
{Transi | i ∈ G} consisting of all the transposition inverse of i in the group (G, ⊕). To verify this for-
operators Transi on S, for the various intervals i mula we must show that Transi ● Trans⊖i is equal
in G. We will see that the group structure on 𝒯S is to the identity Trans0. We just established that
closely related to the group structure on G. the equation Transi ● Transj =​Transi⊕j holds for
In 𝒯S, as in every transformation group, the all intervals i and j in G. Substituting ⊖i for j in
group operation is ●, composition of functions. this equation gives Transi ● Trans⊖i =​Transi ⊕ (⊖i),
To be sure that (𝒯S, ●) is a group, we must check which is equal to Trans0 because i ⊕ (⊖i) =​0 by
that the identity transformation is in 𝒯S and the general property of inverses.
that 𝒯S is closed under the formation of com- The proof that 𝒯S is a group is complete. It is
positions and inverses. The identity transfor- worth noting how every step of this argument
mation in 𝒯S, as expected, is Trans0—​but we about the transformation group 𝒯S relied directly
must be careful about what this means. Here 0 on the corresponding fact about the interval
is the identity element of G, which is not nec- group G. The intervals i in G, clearly, are in one-​
essarily the ordinary number 0. The identity to-​one correspondence with the transposition
interval property established in Section 5.4 tells operators Transi in 𝒯S, and all the required prop-
us that int(x, x) =​0 for every x in an interval erties of the latter were derived from the same
space. Therefore Trans0(x), which by definition properties of the former. The situation general-
is the unique y such that int(x, y) =​0, must be x izes the observations we made in Section 5.1
itself. This means that, regardless of what sort of about the groups (ℤ, +​) and (𝒯∞, ●): even though
object the element called 0 may be, Trans0 really G and 𝒯S may be very different in some respects
is the identity function on S, and so the identity (the group operation in G could be anything,
is indeed present in 𝒯S. while in 𝒯S it is always composition of functions),
To check that 𝒯S is closed under composition something about their structure is very similar.
of functions, consider two transpositions Transi In fact, G and 𝒯S are always isomorphic groups,
and Transj in 𝒯S. Is Transi ● Transj an element in the way that we will formalize in Chapter 6.
of 𝒯S? From our knowledge of transpositions in It may be instructive to consider how the
pitch and pitch-​class space, we might expect that relationship between intervals and transpo-
Transi ● Transj should be Transi + ​j. That is almost sitions works in the case of frequency space
right, but “i +​ j” in the subscript does not quite freq, which was studied in Sections 5.2 and 5.4
make sense: the intervals i and j are elements of as an example of a multiplicative interval space.
the group G and must be “added” using the group The interval group of freq is (ℝ>0, ∙). According
operation in G. In principle G could be any kind to our definition of the generalized transposi-
of group at all; as in the definition of interval tion operator Transi, the interval equations
spaces in Section 5.4, let us write ⊕ for the group int(C4, C5) =​2 and int(C5, C4) = 1​ 2 in this space
operation in G, whatever it may be. The “sum” translate to Trans2(C4) =​C5 and Trans1/​2(C5) =​
i ⊕ j is some interval in G, so Transi ⊕ j is a legiti- C4. In other words, “transposition by 2” is an
mate transposition operator and an element of ascending octave, while “transposition by 1 2 ” is

196 • E x p l o r i n g M u s i c a l S p a ces
a descending octave. Also, because 1 is the iden- by a single semitone and the group G contains
tity in the multiplicative interval group, Trans1 no odd-​numbered transpositions, no such f is
is the identity in the group of transpositions. available. This group G contains insufficient
(There is no Trans0 here; remember that the transformations—​insufficient data—​to define
number 0 can never be an element of a multi- a complete interval structure on the space on
plicative group.) While these equations may which it acts. (For a more extreme example, the
look strange and fail to align with our intuitive group {T0} consisting only of the identity trans-
picture of the way transposition works, they formation does not tell us how to define any
show how an abstract sort of “transposition” intervals at all!)
can be defined in a logical way even in interval A different problem arises when the transfor-
spaces that are very different from the most mation group 𝒯ℐ12 of transposition and inver-
familiar ones. sion operators acts on the space S =​ pc (chromatic
The correspondence between intervals and pitch-​class space). The notes x =​C and y =​E are
transformations works both ways. We have elements of S, the transposition T4 is an element
shown how an interval group gives rise to a of 𝒯ℐ12, and these elements satisfy T4(C) =​ E.
transformation group consisting of general- If an interval space associated with 𝒯ℐ12 can
ized transpositions—​for every interval i we can be constructed as described above, therefore,
define a transposition Transi—​so that an inter- int(C, E) must be the interval associated with
vallic conception of a musical space can always T4 (the interval IntT4 in the notation proposed
be effectively replaced by an equivalent trans- above). The problem is that T4 is not the only
formational conception. We now turn our atten- transformation in 𝒯ℐ12 that maps C to E. The
tion to the opposite process, the conversion equation I4(C) =​E is also true, so by the same
from a transformational structure to an interval reasoning, int(C, E) would have to be equal to
structure. the interval IntI4. Somehow two different trans-
If we try to follow the outline of the preced- formations would have to be associated with the
ing discussion in reverse, we might expect this same interval, which means that the intervals
conversion to work roughly as follows. Given and transformations would not be in one-​to-​one
a transformation group G acting on a space S, correspondence after all (and the groups could
we put the transformations in G in one-​to-​one not be isomorphic). This group, rather than con-
correspondence with some sort of “general- taining too few transformations to define an
ized intervals”: that is, we somehow define, for interval structure, contains too many.
each function f in G, something that we call an A reconsideration of Figure 5.6.1 at this point
“interval” Intf. We show that these generalized may make the situation clear. If we are given
intervals form a group Gʹ, probably very similar a transformation group and wish to use it to
in structure to G, and define an interval function determine the interval from x to y, we can do so
on the space S whose interval group is Gʹ, with as long as there is one and only one function in the
the property that for any elements x and y in S group that maps x to y, but we have a problem if
and for any transformation f in G, int(x, y) =​Intf there is no such function or if there are two or
if and only if f(x) =​ y. more such functions. In the example with even-​
A little reflection shows that the process numbered transpositions only, the group is too
described in the preceding paragraph cannot small for the space: for some elements x and y
work in general. Consider, as an example, the there is no f in the group such that f(x) =​ y. In the
transformation group G =​{…, T−4, T−2, T0, T2, example with 𝒯ℐ12, the group is too large: for
T4, …} acting on chromatic pitch space S =​ pitch. some x and y, there are two different functions f
This group is a subgroup of the cyclic group 𝒯∞ such that f(x) =​ y. For the idea outlined above to
of transposition operators on pitch, consist- succeed as a way of replacing a transformation
ing only of the even-​numbered operators (see group with an equivalent interval group, there
Exercise 5.3.8a). If we are to define an interval must always be exactly one such f.
structure on pitch using the transformations This condition has a name. Suppose G is a
in G according to the process just described, we transformation group acting on a set S. If for
must somehow determine all possible intervals every pair of elements x and y in S there exists
between pairs of notes in the space. To deter- at least one function f in G such that f(x) =​ y,
mine an interval such as int(C4, C♯4) would the action of G on S is said to be transitive.43 If
require identifying a function f in G such that for every x and y in S there exists exactly one
f(C4) =​C♯4—​but because C4 and C♯4 are separated function f in G such that f(x) =​ y, the action of G

Interval Groups and Transformation Groups • 197

on S is said to be simply transitive.44 The action properties from Theorem 5.5.6 to verify
of even-​numbered transposition operators on that G is a group (a subgroup of 𝒯ℐ12).
pitch space fails to be transitive because the (b) The group G has 12 elements, as does the
group is too small; the action of 𝒯ℐ12 on pitch-​ space pc. Nevertheless, show that the
class space is transitive but fails to be simply   action of G on pc is not simply transitive.46
transitive because the group is too large. In
order for there to be exactly one way to get Simple transitivity turns out to be exactly
from any element of the space to any other, a the condition that we need in order to convert a
simply transitive group must be just the right transformational structure into an interval struc-
size.45 ture. Suppose G is a transformation group whose
It is easy to say exactly what “just the right action on a space S is simply transitive. We wish
size” means. Consider a specific element x0 in S, to replace G with an equivalent interval group.
and consider what happens when the various Actually we don’t need to replace G with another
functions in the group G are applied to x0. If the group at all, for we can define an interval-​space
action of G is simply transitive, then for every structure on S whose interval group is G itself.
y in S there must be a function fy in G such The simple transitivity of the action of G on S
that fy(x0) =​ y; therefore there are at least as tells us exactly how the interval function int: S ×
many elements in the group G as in the space S → G must be defined: int(x, y) is defined to be
S. But there can be no more than that, because the unique function f in G such that f(x) =​ y.
every f in G must be fy for some y in S, namely There is no need to verify that this interval
y =​ f(x0). In other words, the mapping from y to group is a group, or that the transformation
fy is a one-​to-​one correspondence between the group and interval group are in one-​to-​one cor-
elements of S and the elements of G. The order respondence: they are the same group. We do
of a simply transitive group, therefore, must need to verify, however, that (S, G, int) is an
be exactly the same as the cardinality of the interval space, which means we must verify the
set on which it acts. In particular, the group homogeneity axiom and the interval-​sum equa-
𝒯ℐ12, whose order is 24, is too large to act tion. Homogeneity is a direct consequence of the
in simply transitive fashion on pc, a space of way we have defined the interval function: for
cardinality 12. Also, a simply transitive group any x in S and any “interval” f in G (remember
acting on a finite set must always be a finite that the elements of G are functions f: S → S),
group, and a simply transitive group acting on the unique y in S such that int(x, y) =​ f is y =​ f(x).
an infinite set must be an infinite group. (The The “interval-​sum equation” in this context is a
example involving the even-​numbered trans- statement about composition of functions (the
position operators shows, however, that an group operation in G), namely int(x, y) ● int(y, z)
infinite group acting on an infinite set can fail =​ int(x, z). To see why this is true, the reader may
to be simply transitive.) find it helpful to refer back to Figure 5.4.2b, now
reading the intervals as functions and replacing
Exercise 5.6.2 ⊕ with ●. Recall that int(x, y) is by definition a
(a) Is the action of 𝒯12 on pitch-​class space function in G, say f, for which f(x) =​ y, and int(y,
simply transitive? z) is similarly a function in G, say g, for which
(b) Is the action of 𝒯∞ on pitch space simply g(y) =​ z. The left side of the above “interval-​sum
transitive? equation” is therefore the composite function
(c) Is the action of 𝒯ℐ∞ on pitch space simply f ● g, which must also belong to G because of the
  transitive? closure property of the group G. Because f maps
x to y and g maps y to z, it follows that f ● g maps
Exercise 5.6.3 This exercise outlines an x to z. But by simple transitivity, there can be
example showing that even a finite group of only one function in G that maps x to z, and by
“just the right size”—​the same cardinality as the definition of the interval function, that func-
the space on which it acts—​can fail to be simply tion is the one we are calling int(x, z). Therefore
transitive. int(x, y) ● int(y, z) =​ f ● g =​int(x, z), as desired.
The process of redefining functions as “inter-
(a) Consider the set G =​{T0, T2, T4, T6, T8, vals” may seem artificial. Because the elements
T10; I0, I2, I4, I6, I8, I10} of even-​numbered of a transformation group are not numbers
transposition and inversion operators but functions, it may seem that these “inter-
acting on pitch-​class space. Use the vals” do not tell us anything about distances

198 • E x p l o r i n g M u s i c a l S p a ces
between elements of a space, which (we might is not simply transitive, the correspondence
suppose) is what intervals are supposed to do. between transformations in 𝒯ℐ12 and pitch
But this redefinition lies at the very heart of the classes is systematic in another way. This
structural equivalence between the two concep- action is doubly transitive, which means that
tions—​and remember that the abstract defini- for every pair of pitch classes x and y in pc
tion of interval spaces does not require that an there are exactly two transformations f in
interval group consist of numbers. (The distance 𝒯ℐ12 such that f(x) =​ y. If x and y are the
functions to be defined in Chapter 12 take exclu- pitch classes 1 (C♯) and 5 (F), what are the
sively numerical values.) two functions? What if x and y are B and F?
Another example involving the transforma- (b) Let S denote the collection of all six
tion group 𝒯ℐ12 may be revealing. As we have pc sets of prime form 0268 (French
observed, the action of 𝒯ℐ12 on pitch-​class augmented-​sixth chords). Is the action
space pc is not simply transitive. But 𝒯ℐ12 also of 𝒯12 on S doubly transitive?
acts on the space triad of 24 major and minor (c) The definition of simple and double
triads in pitch-​class space. Writing uppercase transitivity can be extended in an obvious
letters for major triads and lowercase for minor, way to n-​tuple transitivity for any positive
for example, T4(C) =​ T4({0, 4, 7}) =​{4, 8, 11} =​E, integer n. Give an example of a set S of
and I4(C) =​ I4({0, 4, 7}) =​{4, 0, 9} =​a. The group pitch-​class sets on which the action of 𝒯12
𝒯ℐ12 has 24 elements, just the right number for is triply transitive.
a simply transitive group acting on the set of (d) What can you say about the transitivity
24 triads, and indeed the action of 𝒯ℐ12 on of the action of 𝒯ℐ12 on the collection of
triad is simply transitive. Any two triads,   French augmented-​sixth chords?
major or minor, are related by exactly one
transposition or inversion operator, not both, Exercise 5.6.6 Re-​examine the construction of
and the process outlined above suggests that a transposition group 𝒯S =​{Transi | i ∈ G} from
we can consider that operator as a sort of an interval space (S, G, int) in the first part of
“interval” between the two triads. The “inter- this section, and explain why the action of 𝒯S
val” between the C-​major and A-​minor triads on  S must always be simply transitive.48
is the inversion operator I4—​the unique func-
tion in 𝒯ℐ12 that maps one to the other. This is The functional equivalence between inter-
an unconventional use of the term “interval,” val spaces and simply transitive group actions
but it is not hard to imagine that I4 provides is summarized in the following theorem,
information loosely analogous to an interval called Lewin’s theorem because of its centrality
about the relation between the two objects, in GMIT.
and that this information might be a useful
way of “measuring” consonant triad space—​a Theorem 5.6.7: Lewin’s theorem
way that ensures an interval-​space structure, (a) Suppose (S, G, int) is an interval space. For
unlike the triadic distance measures from every i in G, define Transi: S → S so that for
Chapter 4. Incidentally, because the group every x in S, Transi(x) is the unique y in S
𝒯ℐ12 is noncommutative, (triad, 𝒯ℐ12, int) is such that int(x, y) =​ i. Then 𝒯S =​{Transi |
our first example of a noncommutative inter- i ∈ G} is a transformation group acting on
val space.47 S, 𝒯S is isomorphic to G, and the action of
𝒯S on S is simply transitive.
Exercise 5.6.4 Find the unique f in 𝒯ℐ12 (b) Suppose G is a transformation group acting
satisfying each of the following equations in simply transitive fashion on a space S.
involving triads: Define a function int: S × S → G so that
for all x and y in S, int(x, y) is the unique
(a) f(B) =​F function f in G such that f(x) =​ y. Then
(b) f(G♭) =​e   (S, G, int) is an interval space.
(c) f(b) =​G
(d)  f(c) =​b♭ The above discussion traces a complete proof
of Lewin’s theorem except for the isomorphism
Exercise 5.6.5 in part (a), which will be proved in Section 6.3
(a) Although the action of the group 𝒯ℐ12 (of after we have formalized the concept of isomor-
order 24) on the space pc (of cardinality 12) phism of groups.

Interval Groups and Transformation Groups • 199

Exercise 5.6.8 This rather abstract exercise (b) of Lewin’s theorem, part (b) of Exercise
shows that the two processes described 5.6.8 may offer some solace, allowing us to
in Lewin’s theorem—​the conversion from substitute transpositions for intervals. We can
intervals to transformations and vice versa—​ use the “interval” f to define a transposition
are inversely related: if you convert from operator Transf ; this Transf is a now a function
intervals to transformations and back, or from rather than an interval, and in fact it is identi-
transformations to intervals and back, you cal to f—​so an alternative formulation of part
get essentially the same structure you started (b) of Lewin’s theorem asserts that given a
with. This shows that no information is lost group acting on a set in simply transitive fash-
in the conversion: we can always recover an ion, one can define an interval space whose
interval structure from a transformational transpositions are precisely the functions in
structure or vice versa. the given group.

(a) Suppose (S, G, int) is an interval space. —​—​—​

Define the generalized transposition
operators Transi: S → S and the In a celebrated passage in Chapter 7 of GMIT,
transformation group 𝒯S as in part (a) Lewin expands on the conceptual distinction
of Lewin’s theorem. Then, using this between intervallic and transformational think-
transformation group 𝒯S in the role of G ing at some length. When we work with inter-
in part (b) of the theorem, define a new vals, he writes,
interval function intʹ: S × S → 𝒯S, forming
a new interval space (S, 𝒯S, intʹ ) as part [W]‌e are very much under the influence of
(b) describes. (This new interval function Cartesian thinking… . We tend to conceive
was called int in Theorem 5.6.7b, but in the primary objects in our musical spaces
the present situation it is not the same as as atomic individual “elements” rather
the original function int, so we call it intʹ than contextually articulated phenomena
instead.) The two functions int and intʹ are like sets, melodic series, and the like. And
both interval functions on S, one taking we tend to imagine ourselves in the posi-
values in G, the other in 𝒯S. Show that, for tion of observers when we theorize about
all x and y in S, if int(x, y) =​ i, then intʹ(x, y) musical space; the space is “out there,”
=​ Transi. This shows that although the two away from our dancing bodies or singing
interval functions are technically different, voices. “The interval from s to t” is thereby
they always map to corresponding elements conceived as modeling a relation of exten-
of the two interval groups. (In a way to be sion, observed in that space external to
made precise in Chapter 7, not only are ourselves; we “see” it out there just as we
the groups G and 𝒯S isomorphic, but the see distances between holes in a flute, or
interval spaces (S, G, int) and (S, 𝒯S, intʹ ) points along a stretched string. …
are also isomorphic.)
(b) Suppose G is a transformation group acting In contrast, the transformational attitude
in simply transitive fashion on a space S. is much less Cartesian. Given locations s
Define int: S × S → G as in part (b) of Lewin’s and t in our space, this attitude does not
theorem, and then consider the interval space ask for some observed measure of exten-
(S, G, int) from the perspective of part (a) sion between reified “points”; rather it
of the theorem. For every f in G (originally asks: “If I am at s and wish to get to t, what
a transformation, now reinterpreted as an characteristic gesture … should I perform
interval), define a generalized transposition in order to arrive there?” … “If I want to
operator Transf : S → S as described in part change Gestalt 1 into Gestalt 2 (as regards
(a). Show that the function Transf is identical content, location, or anything else), what
to f; that is, Transf(x) =​ f(x) for every x in S. sorts of admissible transformations in my
All the transformations in the original group space … will do the best job?” … This atti-
G are therefore recoverable from the interval tude is by and large the attitude of some-
  structure.49 one inside the music, as idealized dancer
and/​or singer. No external observer (ana-
If some readers remain uncomfortable with lyst, listener) is needed. (Lewin [1987]
the idea of a function f as an “interval” in part 2007, 158–​59)

200 • E x p l o r i n g M u s i c a l S p a ces
 Many scholars who have written about musi- synthesizing the two approaches: only a page
cal transformation theory have seized on this later, he drives the point home, writing that “we
passage, underscoring the dichotomy between do not have to choose either interval-​language or
the “static,” “Cartesian,” “external” intervallic transposition-​language; the generalizing power
conception and the “dynamic,” “active,” “inter- of transformational theory enables us to con-
nal” transformational approach. The “transfor- sider them as two aspects of one phenomenon”
mational attitude” and associated “characteristic (Lewin [1987] 2007, 160). True to his word, he
gestures” are sometimes cited to bolster the idea regularly juxtaposes transformational and inter-
that transformation theory is a fundamentally vallic terminology throughout the remainder of
new approach to music theory with its own dis- the book.50
tinctive outlook.
While the distinction is certainly worth mak-
ing, I believe that it has sometimes been exag-
NOTES
gerated. Lewin’s theorem, as described above, 1. For a detailed history of group theory, see
shows an explicit equivalence between intervals Wussing 1984 or van der Waerden 1985. Another
and transformations: for any interval space one notable precursor of group theory, one with some
can define a simply transitive transformation musical interest, is found in the English tradition
group that captures all the information inher- of change ringing, the ringing of church bells in sys-
ent in the interval structure; conversely, given tematic permutations. It has even been suggested
a group with simply transitive action on a set, (White 1996) that Fabian Stedman (1640–​1713), the
one can define an interval space whose transpo- author of two publications on change ringing, perhaps
sitions are precisely the elements of the given deserves to be called the first group theorist.
group. In the paragraph of GMIT immediately 2. Babbitt’s dissertation on the twelve-​tone sys-
following the passage quoted above in which he tem (Babbitt 1992) was completed in 1946, though
contrasts the intervallic and transformational his degree was not awarded until 46 years later.
approaches, Lewin proceeds to reunify them: Group structure is noted in Babbitt’s seminal 1960
article “Twelve-​Tone Invariants as Compositional
Either [intervals or transformations] can Determinants” and in several of his other publica-
be generated formally from the charac- tions. Among other musical applications of group the-
teristic properties of the other. More sig- ory prior to or contemporaneous with Lewin’s GMIT,
nificant than this dichotomy, I believe, is attention may be called to Gamer 1967, Budden 1972
the generalizing power of the transforma- (a group theory text with a chapter on groups in
tional attitude: It enables us to subsume music), Starr 1978, several references in Rahn 1980,
the theory of [intervals] … into a broader Balzano 1980, and Morris 1987.
theory of transformations. This enables 3. The centrality of the axiomatic method moti-
us to consider intervals-​ between-​things vated Bertrand Russell’s tongue-​in-​cheek character-
and transpositional-​relations-​between-​ ization of mathematics: “Pure mathematics consists
Gestalts not as alternatives, but as the entirely of asseverations to the effect that, if such
same phenomenon manifested in different and such a proposition is true of anything, then such
ways. (Lewin [1987] 2007, 159) and such another proposition is true of that thing. It
is essential not to discuss whether the first thing is
It is only through an understanding of group really true, and not to mention what the anything is,
theory that we can grasp how this intervallic-​ of which it is supposed to be true. … Thus mathemat-
transformational rapprochement works. In the ics may be defined as the subject in which we never
case of simply transitive actions of transforma- know what we are talking about, nor whether what we
tion groups, Lewin’s theorem shows that the are saying is true” (Russell 1901, 83–​84). In the case of
equivalence is quite exact. Considered more group theory, in which not only objects but also opera-
broadly, transformation theory “subsumes” tions are defined axiomatically, Russell might have
interval theory, rather than the other way added that “we do not know what we are doing.”
around, because, as we have seen, some trans- 4. The transposability of interval configurations
formation groups are not simply transitive and and the interval-​preserving property of transpositions
therefore do not correspond to interval struc- always hold in pitch space, and in fact in all commuta-
tures; in such cases transformation theory may tive interval spaces (as defined later in this chapter).
be capable of making statements that have no We will see in Chapter 7 that these properties do not
intervallic equivalents. But Lewin persists in necessarily hold in noncommutative interval spaces.

Interval Groups and Transformation Groups • 201

 5. For the Z relation, see Straus 2016, 112–​15. In the context of addition, homogeneity means that
The use of the letter Z in this context, coined by Allen differences between group elements are uniform
Forte, is of course unrelated to the traditional use of ℤ throughout the group; in the context of multiplica-
to denote the integers. The Z relation is one instance tion it means that ratios between group elements
of a general and rather difficult mathematical problem are uniform throughout the group. In the latter
known as homometric sets; homometric structures are interpretation, there is a sense in which 0 is very
of some importance in crystallography. Homometric much unlike every other real number, and therefore
sets not related by transposition or inversion can be a sense in which the real numbers become homoge-
constructed not only in pitch-​class space but also in neous only when 0 is excluded. The sort of symmetry
pitch space: the reader may verify that the pitch sets required of a multiplicative group also differs from
{0, 1, 2, 6, 8, 11} and {0, 1, 6, 7, 9, 11} have the same that of an additive group: it is multiplicative symme-
pitch interval content. try about the number 1, in which every number is
6. The similarity of the words transformation and balanced by its reciprocal (rather than additive sym-
transposition occasionally leads to confusion. The let- metry about 0, in which every number is balanced by
ter T in Tn stands for transposition, but every trans- its negative).
position operator is a transformation, so Tn is both 13. The trivial group {1} and the order-​2 group
a transposition and a transformation. We shall not {1, −1} are the only finite subgroups of the multipli-
use the letter T for any transformations other than cative group (ℝ≠0, ∙). Readers who have studied com-
transpositions. plex numbers, however, may verify that {1, −1, i, −i}
7. The use of the letter e for the identity element is a group of order 4 (where i is the imaginary square
of an abstract group, widespread since the nineteenth root of −1), and may be able to construct other finite
century, apparently derives from the German Einheit. multiplicative groups of complex numbers. Peck 2010
The notation id. is also sometimes encountered. describes some (rather difficult) musical applications
Identity elements in particular groups may have other of these complex groups.
names, such as 0, 1, or T0. In this book we use capital 14. Hint for Exercise 5.2.6b: Do not overlook the
E for the identity function, the identity element in a obvious. What is (4 − 3) − 2, and what is 4 − (3 − 2)?
group whose elements are functions. 15. The numbers 1, 5, 7, and 11 play an assort-
8. The identity and inverse axioms include the ment of special roles in ℤ12. Though we have
statements x ⊗ e =​ e ⊗ x and x ⊗ xʹ =​ xʹ ⊗ x, implying defined units here in terms of their multiplicative
that certain pairs of elements must commute (every behavior, the units mod n are also precisely the ele-
element commutes with the identity and with its own ments that generate ℤn as an additive group. See
inverse), but these special cases do not imply commu- Exercise 5.3.7c.
tativity of the entire group. 16. We introduced the word involution in Section
9. We use the term commutative group here, but 2.3 for a function that is its own inverse; here it denotes
mathematicians often refer to commutative groups a group element that is its own inverse. The two usages
as abelian groups, after Niels Henrik Abel (1802–​29), are closely analogous and often overlap, because func-
a Norwegian mathematician who, like Galois, man- tions often form groups.
aged to produce during his short lifetime a remark- 17. Hints for Exercise 5.3.2: These groups are com-
able quantity of important work. Although Abel did mutative, so in manipulating these equations we can
not use the language of modern group theory, his add or multiply on either the left or the right with the
investigations into the solvability of algebraic equa- same effect. A musical interpretation for (a) might
tions depend in a crucial way on the commutativity of be “What pitch-​space interval x, combined with an
underlying group structures. ascending double-​octave (+​24 semitones), produces
10. Hint for Exercise 5.2.2a: Remember that the an ascending perfect fourth (5 semitones)?” In parts
interval group of a musical space consists of the inter- (f) and (g), the groups are small, so it is not hard to
vals between the notes of the space, not of numbers find the x that solves each equation by trial and error.
assigned to the notes themselves. You should also be able to determine x using the rea-
11. Hint for Exercise 5.2.3b: The numbers 1 2 , 1 4 , and soning given above, although this will require know-
3 belong to this group, but 1 , 1 , and 1 do not.
4 3 5 6 ing the inverses of some elements of ℤ12× and ℤ7×.
12. Omitting 0 may seem to leave a hole in the A musical interpretation for (f) might be “If you wish
real number line, in apparent violation of the homo- to start on C, transpose by a certain interval seven
geneity principle mentioned above as a general char- times in succession, and end on B, what interval
acteristic of groups. Homogeneity—​ for which we should you choose?” The fact that the numbers 7 and
did not give a precise definition—​must, however, be 11 belong to the multiplicative group ℤ12× ensures
understood in the context of the group operation. that this equation has a solution; if 7 is replaced with

202 • E x p l o r i n g M u s i c a l S p a ces
4, for example, the resulting equation has no solution name interval space is more consistent with the names
in the integers mod 12. typically given to abstract structures by mathemati-
18. Hint for Exercise 5.3.4: If a ⊗ a =​ a is true, then cians, and also with other nomenclature in this book.
x =​ a is a solution to the equation a ⊗ x =​ a—​but we Otherwise, the presentation here differs from Lewin’s
just derived a general way of solving equations of this primarily in matters of notation; in particular, Lewin
type. Note carefully the property proved in this exer- uses multiplicative notation for the interval group,
cise: it states that if a ⊗ a =​ a, then a =​ e. It does not while I have chosen to use a symbol that suggests
state that if a ⊗ a =​ e, then a =​ e; we have already noted addition.
the possibility of involutions, non-​identity elements 27. Vector addition is performed in algebraic struc-
satisfying a ⊗ a =​ e. tures called vector spaces; see any text in linear algebra
19. Hint for Exercise 5.3.5c: Do not be misled by for details. A vector space is, among other things, an
the word subgroup. Technically every set is a subset of additive group, and some interval groups are vector
itself, and every group is a subgroup of itself. spaces, so in some cases an interval-​sum equation can
20. Hint for Exercise 5.3.6: In every case the answer indeed be considered an example of vector addition.
is the same as it would be in ordinary multiplicative Not all interval groups are vector spaces, however; the
algebra. The proofs require nothing more than careful modular group ℤ12, for example, is not a vector space.
attention to the definitions of the notations involved. 28. Hint for Exercise 5.4.3: What is int(x, y) ⊕
The notation (a−2)−1, for example, means the inverse int(y, x)?
of a−2, and a−2 means a−1 ⊗ a−1, so in order to show 29. Interval spaces are equivalent to mathematical
that (a−2)−1 =​ a2, you must show that a−1 ⊗ a−1 ⊗ a2 =​ constructions known as principal homogeneous spaces.
e (of course we also know that a−1 ⊗ a =​ e). 30. Because of considerations like these, the notes
21. Expressions such as an +​ 2 and a2 technically rep- playable on any given instrument, such as the 88 keys
resent different products: a product of n +​ 2 a’s in the of the piano, typically do not form an interval space.
first case and only two a’s in the second. The equation Nevertheless, it is often not difficult to conceive of
an +​ 2 =​ a2 says that these two different expressions rep- a larger interval space containing them, so the play-
resent the same element of the group. able notes may be regarded as an interval configura-
22. Hint for Exercise 5.3.7e: Make sure you under- tion. For commentary on the nature of “instrumental
stand the difference between the questions in (d) and spaces” see De Souza 2017, especially 53–​63.
those in (e). The three subgroups in (e) are groups 31. Hint for Exercise 5.4.8a: It may be helpful to
of order 2, 3, and 6—​but not necessarily in that think about the specific example just discussed, where
sequence! Gʹ is the subgroup 3ℤ =​{…, −6, −3, 0, 3, 6, …} of ℤ, x0 is
23. Hint for Exercise 5.3.7f: The answer to the last C4, and Sʹ is a diminished-​seventh space. Suppose that
question in (f) should be different from the answer x and y are elements of Sʹ. By the definition of Sʹ, this
to the corresponding question in (e). The elements of means that int(x0, x) and int(x0, y) are elements of Gʹ.
ℤ12× (the units of ℤ12) are the same four generators of Why does it then follow that int(x, y) ∈ Gʹ ?
ℤ12 that you identified in (c). There are three equiva- 32. Hint for Exercise 5.4.8b: Because S is an interval
lent ways to characterize the units mod n: they are the space, we know that a y satisfying int(x, y) =​ i exists
mod-​n integers coprime to n, the possible generators in S and is unique; we must show that this unique y
of ℤn as an additive cyclic group, and the mod-​n inte- belongs to Sʹ, which means we must show that int(x0,
gers with multiplicative inverses. y) ∈ Gʹ. We also know that i ∈ Gʹ, that x ∈ Sʹ, and there-
24. Hint for Exercise 5.3.8d: There is such an ele- fore that int(x0, x) ∈ Gʹ.
ment in one of the multiplicative groups discussed in 33. Hint for Exercise 5.4.12b: There are seven semi-
Section 5.2. tones in a perfect fifth.
25. Hint for Exercise 5.3.10e: If you can write the 34. Hint for Exercise 5.4.13: The frequency ratios
number 1 in terms of 6, 10, and 15 using nothing but for the pure major third and minor third are 5 4 and 6 5 ,
addition and subtraction, then you can conclude that respectively. The frequency ratio for an interval of k
these three numbers together generate all of ℤ, even equal-​tempered semitones is sk, where s =​21/​12.
though the preceding parts of the exercise should tell 35. Algebra texts commonly present a more
you that no two of them in combination are sufficient abstract definition of the action of a group G on a set
to do so. S, in which G is not required to be defined as a set of
26. The notion of interval space defined here is functions from S to S. In this approach, an action is
functionally equivalent to Lewin’s generalized interval defined as a function α: S × G → S; the notation xg
system (Lewin [1987] 2007, 26); the reader may imag- may be used in place of α(x, g). An action is required
ine that Lewin’s abbreviation GIS stands for either to satisfy x(g1∙g2) =​(xg1)g2 and xe =​ x for all elements x
generalized interval system or general interval space. The in S, and g1 and g2 in G; here ∙ is the group operation

Interval Groups and Transformation Groups • 203

in G and e is the identity element. (More specifically, 44. If for every x and y in S there exists at most one
this is the definition of a right group action; left actions function f in G such that f(x) =​ y, the action of G on S
may be defined similarly, using a notation such as gx. is said to be free (also known as semiregular). Therefore
The notation for a right action is consistent with the a group action is simply transitive if and only if it is
left-​to-​right functional orthography preferred in this both transitive and free. These definitions are standard
book.) If an action is defined in this abstract way, then in group theory. The equivalence of simply transitive
for every g in G we may define a function fg: S → S by action with interval structure is Lewin’s insight, but see
the formula fg(x) =​ xg; it can be shown that the func- Vuza 1988 for some related mathematical structures.
tions fg, for the various elements g in G, define a trans- 45. It is common to speak of “simply transitive
formation group on S as we have defined the term, groups,” but technically it is the action of a group on
and that group is isomorphic to G. We are therefore a particular space that is (or is not) simply transitive.
not sacrificing any generality by adopting an explicitly The action of 𝒯ℐ12 on the space pc is not simply transi-
transformational definition of group actions. tive, but shortly we will show an action of 𝒯ℐ12 on a
36. The words inverse and inversion are of course different space which is simply transitive, demonstrat-
related, but in the context of musical transforma- ing that simple transitivity depends not only on the
tion groups the distinction between them is impor- group but also on the set on which it acts.
tant. Functions and group elements have inverses; the 46. Hint for Exercise 5.6.3b: You may show that this
mappings In on pitch space and pitch-​class space are action is not simply transitive either by exhibiting two
inversions. notes x and y in pc for which there is no function f in G
37. For a study of the characteristics of even and such that f(x) =​ y, or by exhibiting two notes x and y for
odd indices of inversion see Alegant 1999. which there are two or more different such functions.
38. Some authors use a label such as 𝒯/​ℐ for a In this case both alternatives are possible.
group containing transpositions and inversions. We 47. The group 𝒯ℐ12 acts in simply transitive fashion
avoid the /​notation here because of possible confu- not only on triad but on some other spaces as well,
sion with the similar notation for quotient groups to such as some spaces of 24 pitch-​class sets or of twelve-​
be introduced in Section 6.6. tone rows, and so can be used to formulate notions of
39. Sources differ on the labeling of dihedral intervals in those spaces. Moreover, 𝒯ℐ12 is not the
groups: the group called 𝒟n here, of order 2n, is called only simply transitive group of transformations acting
𝒟2n by some group theorists. The notation 𝒟n is on triad, and therefore not the only possible interval
consistent with the names we are adopting for other group that allows triad to be construed as an inter-
transformation groups, which are indexed not by the val space. We will encounter several other examples
order of the group but by the cardinality of the space of simply transitive groups acting on triad in Part
on which it acts. Two, notably including the Riemann group, or PLR
40. Hint for Exercise 5.5.10a–​b: Remember that in group, generated by P, L, and R. In the interval space
𝒯ℐ∞, the subscripts are ordinary integers, not inte- corresponding to the Riemann group, the “interval”
gers mod 12, so for example T4−1 is T−4, not T8. Your between the C-​major and A-​minor triads is the relative
answers to (a) and (b) should each be a single transfor- transformation R.
mation of the form Tn or In. The n’s in the answers to 48. Hint for Exercise 5.6.5: 𝒯S consists of transposi-
(a) and (b) are different, but they should be congruent tions Transi. For any elements x and y in S, you must
mod 12. show that there is exactly one transposition Transi
41. See Lewin [1987] 2007. The figure is Figure 0.1 such that Transi(x) =​ y. What is i?
on page xi of the original 1987 edition, page xxix 49. Hints for Exercise 5.6.8: These properties may
of the 2007 reprint. Lewin labels the points s and t appear forbiddingly abstract, but the proofs are short,
rather than x and y, consistent with the notation for requiring nothing more than careful consideration
points in musical spaces that he adopts in much of of the definitions involved. In part (a), if int(x, y) =​ i,
the book. what does that say about Transi, and therefore about
42. Lewin ([1987] 2007, 46) and many subsequent intʹ ? In part (b), if f(x) =​ y, what does that say about
authors have written Ti for transposition in a general int(x, y), and therefore about Transf ?
interval space. We will encounter a few situations in 50. For a more fully developed argument that the
which using a different notation in the general case distinction between the intervallic and transforma-
will aid clarity. tional approaches has sometimes been overblown,
43. Do not confuse transitivity of group actions see Hook 2007b, 172–​77, which responds in part to
with transitivity of binary relations (discussed in Klumpenhouwer’s (2006) claim that Lewin advocated
Section 2.4)—​two entirely different conditions with replacing intervallic thinking with transformational
the same name. thinking.

204 • E x p l o r i n g M u s i c a l S p a ces
SUGGESTED READING Lewin’s Generalized Musical Intervals and
Transformations (Lewin [1987] 2007), though
Details of the principles of group theory covered idiosyncratic, challenging for many readers, and
in Chapters 5 and 6 are available in many abstract by now showing its age, remains an indispensable
algebra texts, such as Dummit and Foote 2004. resource in transformation theory. For further
Grossman and Magnus 1964 and Carter 2009 pres- commentary on GMIT and the Lewinian enter-
ent visually oriented approaches to group theory prise more generally, see Satyendra 2004 (“an
that many readers may find appealing. A classic informal introduction to some formal concepts”),
work on symmetry is Weyl 1952; more recent Klumpenhouwer 2006 (on the “anti-​Cartesian”
presentations of symmetry include du Sautoy nature of the transformational approach), Rings
2008 (which includes an elementary discussion 2006 and Hook 2007b (reviews of the 2007 repub-
of symmetry in music) and Conway, Burgiel, and lication), and Attas 2009 (on cognitive aspects of
Goodman-​Strauss 2008 (which develops a new cat- transformation theory).
egorization of, and notation for, symmetry types).

Interval Groups and Transformation Groups • 205

 PART TWO
Transformation Theory
Intervals and Transformations, including
Neo-​Riemannian Theory
 6
Groups II
Permutations, Isomorphisms, and
Other Topics in Group Theory

PART TWO further develops the transforma- These concepts shed new light on various con-
tional topics introduced in piecemeal fashion structions from earlier chapters; it turns out,
in Part One. Some of this work reflects the for example, that several graphs from Chapters
continuing influence of David Lewin, espe- 3 and 4, as well as many spatial transformation
cially the advanced study of interval spaces in networks, are Cayley diagrams in disguise. The
Chapter 7 and the sections of Chapter 9 topics here are standard ones in group theory,
devoted to transformation graphs and net- but conceptually some of this material, espe-
works. Chapter 8 examines triadic transforma- cially in Sections 6.5 and 6.6, is more abstract
tions, including neo-​Riemannian theory. than that in Chapter 5. Readers who find them-
Group theory plays essential roles in all of selves getting bogged down in the formalism
these areas, sometimes going beyond our intro- should try to grasp the basic definitions and
duction to groups in Chapter 5, so the present examples but may choose to skip over some
chapter continues that study with a number details, referring back to this chapter later for
of additional topics. Among these are the con- reference.
ception of finite transformation groups as per-
mutation groups (Section 6.1); representations
of a group’s structure in tabular form or via a
6.1 PERMUTATION GROUPS
directed graph known as a Cayley diagram (6.2); Recall from Section 5.5 the cyclic transposition
the long-​ promised formalization of isomor- group 𝒯12 acting on chromatic pitch-​class space
phism of groups (6.3); the formation of product pc, and consider as an example the transposition
and quotient groups (6.4 and 6.6); and the appli- T3. The action of T3 may be conveniently tabu-
cation of groups in the study of symmetry (6.5). lated in a 2-​by-​12 matrix,

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0006
  0 1 2 3 4 5 6 7 8 9 10 11
within its cycle (so that, for instance, T3(0) =​3),
T3 =   , and also that each cycle wraps around from its
 3 4 5 6 7 8 9 10 11 0 1 2 
last element to its first (for instance, T3(9) =​0).
The cycles are disjoint—​no element appears in
where each element in the upper row is mapped more than one cycle—​and every element of ℤ12
by T3 to the element appearing directly below appears somewhere in one of the cycles.
it. The matrix representation encourages us to Cycle representations can provide reveal-
think of T3 as a permutation of the set ℤ12, a way ing information about the structure of permu-
of reordering the twelve pitch classes. The initial tations. As a permutation, T3 consists of three
ordering shown in the top row of the matrix is four-​element cycles, or 4-​cycles; we say that T3 is
transformed by T3 into a new ordering in the a permutation of cycle type 43. Musically each of
bottom row. The position initially occupied by pc the three 4-​cycles is a diminished seventh chord;
0 is occupied after the reordering by pc 3, and starting from any given note, repeated applica-
likewise with the other pitch classes. The twelve tion of T3 will cycle through the notes of one of
pitch classes in the top row of the matrix are these chords but will never lead to a note from a
listed here in numerical order, but in principle different chord.
they may be listed in any order at all, as long as The other two permutations considered above
the matrix pairs each pc x with T3(x). have rather different cycle structures:
Matrix representations for two other trans-
formations on pc (or ℤ12) are I = (0)(1, 11)(2, 10)(3, 9)(4, 8)(5, 7)(6);
M 7 = (0)(1, 7)(2)(3, 9)(4)(5, 11)(6)(8)(10).
 0 1 2 3 4 5 6 7 8 9 10 11
I =
 0 11 10 9 8 7 6 5 4 3 2 1 
Neither of these operators has any cycles as long
as the 4-​cycles we observed in T3. The cycle struc-
and ture of the inversion operator I includes seven
cycles altogether, five 2-​cycles and two 1-​cycles;
the cycle type of this permutation is therefore
 0 1 2 3 4 5 6 7 8 9 10 11 1225. The 1-​cycles are the fixed points of this
M7 =  .
 0 7 2 9 4 11 6 1 8 3 10 5  operator (0 and 6), discussed in Section 5.5. The
multiplication operator M7 has cycle type 1623,
Here I is the familiar inversion operator, and M7 with no fewer than nine cycles, including six
is mod-​12 multiplication by 7, defined by the fixed points (every even pc is fixed by M7) and
equation M7(x) =​7x (mod 12). (Below we will three 2-​cycles.
study some transformation groups containing The cycles in the cycle representation for
this multiplication operator.) any permutation may be listed in any order,
and within one cycle any element may be
Exercise 6.1.1 Write matrix representations listed first. By convention, we usually choose
for the pitch-​class transformations T2, T6, I3, the notation so that the first cycle starts with
and
  M5 (multiplication by 5). 0 and each subsequent cycle starts with the
smallest element not yet accounted for. Fixed
Another convenient representation for per- points may be omitted from a cycle representa-
mutations is the cycle representation. The cycle tion, so we may write I =​(1, 11)(2, 10)(3, 9)
representation for T3 is (4, 8)(5, 7) and M7 =​(1, 7)(3, 9)(5, 11), with
the understanding that all elements not shown
in such a representation are fixed. Apart from
T3 = (0, 3, 6, 9)(1, 4, 7, 10)(2, 5, 8, 11). rearrangements and omissions of these kinds,
the cycle representation of a permutation is
The four-​element groupings are the cycles in this uniquely determined.
representation. The cycle notation (0, 3, 6, 9)
may be regarded as a typographic simplifica- Exercise 6.1.2 Write cycle representations for
tion of the more explicit notation 0 → 3 → 6 → T2, T6, I3, and M5. What is the cycle type of each
9 → 0. That is, we understand that T3 maps each of these permutations? Which two have the
pitch class to the element immediately following same
   cycle type?

210 • E x plor i n g M u s i cal S paces

Exercise 6.1.3 representations of the same type for
(a) Among the twelve transposition operators I ● M7 and M7 ● I. Do I and M7 commute?
T0, T1, …, T11, how many different cycle Compare your representations for these
types are there? transformations with the results of
(b) Among the twelve inversion operators I0, Exercise 6.1.1 or 6.1.2. What do you
I1, …, I11, how many different cycle types   notice, and why does it happen?
  are there?1
Exercise 6.1.6 Two permutations p and
Any permutation on a finite set may be written q on ℤ12 are defined below. The matrix
in matrix form and also via a cycle representation. representation is given for p, and the cycle
In general, the matrix has two rows and n columns, representation for q:
where n is the cardinality of the set. A permuta-
tion must be a bijective function (one-​to-​one and  0 1 2 3 4 5 6 7 8 9 10 11
p= ;
onto), which ensures that every element of the set  3 8 1 6 11 4 9 2 7 0 5 10
appears exactly once in each row of the matrix.
Similar matrices can be constructed for non-​
bijective functions, in which case the bottom row q = (0, 5, 6, 11)(1, 10, 7, 4 )(2, 3, 8, 9) .
includes duplications. Such functions are not per-
mutations; the following exercise gives an example.
(a) Write the cycle representation for p and the
Exercise 6.1.4 matrix representation for q.
(a) The function M2 (multiplication by 2) is (b) Use the matrix representation for p to
a well-​defined function on ℤ12, but it is derive a matrix representation for p−1.
neither one-​to-​one nor onto. Construct a (c) Use the cycle representation for q to derive
matrix representation for M2, and observe a cycle representation for q−1.
that the bottom row of the matrix is not a (d) How are p and q related?
permutation of the elements of ℤ12. (e) Can you find a way to express p and q as
(b) What happens if you try to construct a compositions of transformations Tn, In,
  cycle representation for M2?   and/​or Mn?2

Either matrices or cycle representations may A single cycle such as c =​(0, 3, 6, 9) may be
be used to facilitate calculations involving per- regarded as a permutation of ℤ12 in its own
mutations. For instance, if p and q are permu- right, the elements not appearing in the cycle
tations on ℤ12 (or some other finite set), the being understood as fixed points. The term cyclic
composition of functions p ● q is defined, and we permutation is widely used, although technically
may calculate the action of p ● q by first consult- a cyclic permutation of a set S is a permutation
ing either the matrix or cycle representation for that collects all the elements of S into a single
p to determine y =​ p(x), then locating y in the cycle—​a permutation of cycle type n1, where n is
corresponding representation of q to determine the cardinality of S. The cycle c above, though it
z =​ q(y). As bijective functions, permutations may be considered a permutation of ℤ12, is not a
always have inverses; calculating the inverse of a cyclic permutation of ℤ12; c is a cyclic permuta-
permutation amounts to reading its matrix rep- tion only of the subset {0, 3, 6, 9}.
resentation from the bottom up, or the cycles of Any two permutations acting on ℤ12 may be
its cycle representation from right to left. composed, so we may consider products such
as p =​(0, 3, 6, 9)(0, 6). This expression writes
Exercise 6.1.5 p as a composition of two cycles, p =​ c ● cʹ, where
(a) Use the matrix representations for T3 and c is the 4-​cycle (0, 3, 6, 9) as before and cʹ is the
I as described above to construct a matrix 2-​cycle (0, 6). Although it is written as a prod-
representation for I ● T3, and also for T3 ● I uct of cycles, (0, 3, 6, 9)(0, 6) is not the standard
(which by Theorem 5.5.6e is the same as I ● T9). cycle representation for p, because the two cycles
(b) Use the cycle representations for T3 and I appearing here are not disjoint (the elements 0
to construct cycle representations for I ● T3 and 6 appear in both of them). To understand
and T3 ● I. how this product works, remember that func-
(c) Using either the matrix or cycle tions are composed from left to right. When p is
representations for I and M7, construct applied to the element 0, the 4-​cycle c maps 0 to 3,

Permutations, Isomorphisms, and Other Topics • 211

which is unaffected by the 2-​cycle cʹ, so p(0) =​3. broken down further into a product of 2-​cycles.
When p is applied to 3, c maps 3 to 6 and then (Generally a k-​cycle (x1, x2, …, xk) is equal to a
cʹ maps 6 to 0, so p(3) =​0. By similar reasoning, product of (k − 1) 2-​cycles of the form (x1, x2)
p(6) =​9 and p(9) =​6. The standard cycle repre- (x1, x3) ⋯ (x1, xk).) Consequently every permu-
sentation for p is therefore p =​(0, 3)(6, 9). tation may be written as a product of 2-​cycles.
The 2-​cycles in such a product are not necessar-
Exercise 6.1.7 ily disjoint, and Exercise 6.1.7e shows that the
(a) Verify that the compositions 2-​cycle representation of a permutation is not
(0, 3, 6, 9)(0, 6) and (0, 6)(0, 3, 6, 9) are unique: the same permutation p can be written
different, demonstrating that non-​disjoint as either (0, 3)(0, 6)(0, 9)(0, 6) or (0, 3)(6, 9).
cycles do not generally commute. In this case one representation involves four 2-​
(b) Verify that the compositions cycles and the other involves two. A theorem of
(0, 3, 6, 9)(1, 4) and (1, 4)(0, 3, 6, 9) permutation theory states that if a permutation
are the same, illustrating the fact that can be written as a product of an even number
disjoint cycles do commute. of 2-​cycles, then every 2-​cycle representation
(c) Verify that the composition of 2-​cycles of that permutation has an even number of
(0, 3)(0, 6)(0, 9) is equal to the 4-​cycle 2-​cycles, and likewise if one 2-​cycle representa-
(0, 3, 6, 9). tion is odd, then every such representation is
(d) Verify that the composition of 2-​cycles odd. Permutations of these two types are known
(0, 6)(0, 9)(0, 6) is equal to the 2-​cycle (6, 9). as even and odd permutations.3
(e) Using the result of either (c) or (d), The function p just considered, with repre-
conclude that the composition of 2-​cycles sentations consisting of two or four 2-​cycles,
(0, 3)(0, 6)(0, 9)(0, 6) is equal to the is an even permutation. The 4-​cycle (0, 3, 6,
permutation p =​(0, 3, 6, 9)(0, 6) =​ 9), according to the 2-​cycle representation in
  (0, 3)(6, 9) discussed above. Exercise 6.1.7c, is an odd permutation. In gen-
eral, a cycle of even length is an odd permutation
Multiplying the cycle c =​(0, 3, 6, 9) by itself and vice versa. When two permutations, each a
repeatedly, we obtain the permutations product of 2-​cycles, are composed, the num-
ber of 2-​cycles in the product is the sum of the
numbers of 2-​cycles in the two original permu-
c =(0, 3, 6, 9) , tations. It follows that the product of two even
c 2 = (0, 3, 6, 9)(0, 3, 6, 9) = (0, 6)(3, 9), permutations is even (because the sum of two
even numbers is even), the product of two odd
c 3 = (0, 3, 6, 9)(0, 3, 6, 9)(0, 3, 6, 9) = (0, 9, 6, 3), permutations is even, and the product of an even
c 4 = (0, 3, 6, 9)(0, 3, 6, 9)(0, 3, 6, 9)(0, 3, 6, 9) and an odd permutation, in either order, is odd.
= (0)(3)(6)(9) = e. These principles make it a straightforward
matter to determine the parity (evenness or odd-
ness) of any permutation from its cycle struc-
As mappings on the set {0, 3, 6, 9}, these per- ture. Whenever a permutation is written as a
mutations c, c2, c3, and c4 are simply the trans- product of cycles (whether disjoint or not), the
positions T3, T6, T9, and T0 (the identity), and cycles of odd length, being even permutations,
together they form a cyclic group of order 4. always contribute an even number of 2-​cycles
Because c4 =​ e, the 4-​cycle c is itself a group ele- to any 2-​cycle representation. The parity of the
ment of order 4, as defined in Section 5.3. The permutation, therefore, is determined entirely
connection between the cycle c =​(0, 3, 6, 9) and by the number cycles of even length (including 2-​
the cyclic group 〈c〉 =​{e, c, c2, c3} =​{T0, T3, T6, cycles themselves). If the total number of cycles
T9} generated by c is apparent. We recall from of even length is even, the permutation is even;
Chapter 5, though, that any element a of a group if the number of cycles of even length is odd, the
G generates a cyclic subgroup 〈a〉 of G, regardless permutation is odd.
of whether a is a cyclic permutation—​or, for that
matter, whether the elements of G are conceived Exercise 6.1.8
as permutations at all. (a) Using the above principles and the cycle
Every permutation may be written as a prod- representations of T3, I, and M7 presented
uct of cycles, and by means of the technique previously, show that all three of these
illustrated in Exercise 6.1.7c, every cycle may be transformations are odd permutations.

212 • E x plor i n g M u s i cal S paces

(b) Among the twelve transposition operators a permutation of the elements of S and there-
T0, T1, …, T11, which are even permutations fore as an element of 𝒮n, so any other transfor-
and which are odd? mation group on S is always a subgroup of 𝒮n.
(c) Among the twelve inversion operators I0, In particular, the cyclic group 𝒞n (or 𝒯n) and
I1, …, I11, which are even permutations and dihedral group 𝒟n (or 𝒯ℐn), both introduced in
which are odd? Chapter 5, are subgroups of 𝒮n. For convenience
(d) Is M5 an even or odd permutation? in the following discussion, we assume that the
(e) Of the four elements in the cyclic group elements of S are labeled 0, 1, 2, …, n − 1 (that is,
generated by the cycle (0, 3, 6, 9), which are we assume that S =​ ℤn), although elements with
  even permutations, and which are odd?4 other names could be used instead. The cyclic
group 𝒯n consists of the transposition opera-
Permutations arise in a variety of musical con- tors T0, T1, …, Tn−1 on ℤn. As a permutation, T1 is
texts. Twelve-​tone rows may be described as per- simply the n-​cycle (0, 1, …, n − 1), and the other
mutations of ℤ12; in Chapter 9 we will note that transpositions may be generated by repeated
row transformations may be described as permu- application of T1. The dihedral group 𝒯ℐn con-
tations of either pitch-​class numbers or order num- sists of these n transpositions Tk and also the
bers. More generally, the objects permuted by a n inversions Ik, where I =​ I0 is the permutation
permutation need not be numbers at all. The rela- (0)(1, n − 1)(2, n − 2)⋯ and Ik =​ I ● Tk. 𝒯n is a
tive transformation R of neo-​Riemannian theory, commutative group of order n; 𝒯ℐn is a group of
for example, may be regarded as a permutation of order 2n, noncommutative for all n ≥ 3.6
the set triad of 24 major and minor triads, with a A few other subgroups of 𝒮n are worth noting.
cycle structure comprising twelve 2-​cycles: The even permutations of S, defined above, form
a subgroup, known as the alternating group 𝒜n.
R = (C, a)(D♭, b♭)(D, b). . .(B, g♯). The order of 𝒜n is n! ∕ 2, because as it turns out,
for every n ≥ 2, exactly half of the permutations
R is thus an even permutation on triads; permuta- in 𝒮n are even. (More is true: every permutation
tion parity will inform our study of triadic trans- group G always consists either of even permuta-
formations in Section 8.3. In another application, tions exclusively—​if G is a subgroup of 𝒜n—​or of
to be explored in Example 6.1.13 below, the per- even and odd permutations in equal numbers.)
muted objects may be lines in a contrapuntal Multiplication operators such as M5 and M7
texture. The relation of permutational equivalence, do not generally belong to 𝒞n or 𝒟n, and Exercise
which holds between tuples such as (E4, C4, D4, 6.1.4 showed that for some a, the function Ma fails
G3) and (G3, D4, E4, C4) that are rearrangements to be one-​to-​one and onto, and therefore cannot
of the same elements, was introduced in Section be an element of any group. In Section 5.2, how-
2.4 and will play an important role in our study of ever, we introduced ℤn×, the group of units mod
voice-​leading spaces in Part Three.5 n, a multiplicative group of order φ(n) consisting
If S is a set of cardinality n, every permuta- of those mod-​n integers a that are coprime to n.
tion of the elements of S may be represented as The functions Ma where a ∈ ℤn× likewise form a
above by a 2-​by-​n matrix. When the n elements commutative group of order φ(n), a subgroup of
are arranged in some fixed order in the top row 𝒮n known as ℳn, the multiplication group on ℤn.
of this matrix, it becomes clear that the number Larger groups may be constructed by combining
of such permutations is just the number of ways multiplication and transposition operators. In
in which the elements may be ordered on the particular, the affine group 𝒜𝑓𝑓n is generated by
bottom row. We know from Section 2.2 that this all transpositions Tk on ℤn and all multiplication
number is n! (n factorial). We have shown above operators Ma with a coprime to n. This group will
how to calculate compositions and inverses of be considered further in Section 6.2.
permutations, and in fact the n! permutations
form a group, under the operation of composi- Exercise 6.1.9
tion of functions. This group is called the sym- (a) The units mod 12 are 1, 5, 7, and 11,
metric group on n objects, denoted 𝒮n. As long as so ℳ12 is the group {M1, M5, M7, M11}.
n ≥ 3, 𝒮n is a noncommutative group. By what other name(s) do we know the
𝒮n is the largest possible transformation multiplication operator M1? By what other
group on a set S of n elements: every one-​to-​ name do we know M11?
one, onto mapping f: S → S may be regarded as

Permutations, Isomorphisms, and Other Topics • 213

(b) Start with the discrete pitch-​class space pc larger and deeper permutational structure. The
of Figure 1.2.3, and replace every pitch class subset relations 𝒞n ⊆ 𝒟n ⊆ 𝒜𝑓𝑓n ⊆ 𝒮n generally
x with M7(x). What is the resulting picture? hold. All of the listed groups smaller than 𝒜n
If you apply M7 again to the resulting may contain odd permutations, so those groups
picture, what is the result and why? What are not generally subgroups of 𝒜n (although they
if you use M5 or M11 instead of M7? Can you may be for some values of n; see Exercise 6.1.12g
see why M5 and M7 are sometimes called below), and ℳn is not generally a subgroup of
“circle-​of-​fifths transformations”? 𝒞n or 𝒟n.
(c) The permutations T1 and M7 do not In Chapter 5 we reviewed some familiar prop-
commute, but T1 ● M7 is the same as erties of transpositions Tk and inversions Ik on
M7 ● Tk for some k. What is k? ℤ12, which can now be supplemented through
(d) Let G be any permutation group on ℤ12 our understanding of these transformations as
containing M5, M7, and T1. Explain why permutations. Some properties of the transfor-
G must contain all the transposition mations on ℤ12 carry over to the more general
and inversion operators as well as all the case of ℤn. If the elements of ℤn are arranged
  multiplication operators in ℳ12.7 around a circle like the usual pitch-​class clock,
then replacing every number x with Tk(x) is
The common groups described above are sum- equivalent to rotating the circle through k posi-
marized in Table 6.1.10. The groups are listed in tions, and replacing every x with Ik(x) is equiv-
order from smallest to largest (with some excep- alent to reflecting the circle through an axis
tions to this ordering for small values of n). As n, passing through k ∕ 2, just as in ℤ12. Other char-
the number of permuted objects, grows large, n! acteristics of these transformations Tk and Ik
quickly becomes enormous, and therefore 𝒮n and on ℤn, however, depend in various ways on the
𝒜n become much larger than all the other groups number n, or on the relationship between k and
listed. For example, the order of the group 𝒮n is n. This is particularly true when it comes to the
12! =​479,001,600. Most of these nearly half-​bil- cycle structure of Tk and Ik as permutations of
lion permutations are of little apparent musical ℤn. The following exercises explore a few of these
relevance as transformations, although any of sometimes confusing properties.
them could arise in the form of a twelve-​tone
row. In any case, it is useful to remember that Exercise 6.1.11
even 𝒯12 (or 𝒞12) and 𝒯ℐ12 (or 𝒟12), the groups (a) Write the cycle representations of all seven
most familiar in musical contexts, are part of a transposition operators and all seven

Table 6.1.10. Some permutation groups acting on ℤn (n ≥ 3)

NA M E OR DE R DE S C R IP T ION COMMU TAT I V E?

ℳn Multiplication φ(n) All invertible multiplication yes

group operators: Ma(x) =​ ax (mod n),
a ∈ ℤn×
𝒞n (𝒯n) Cyclic group n All transpositions: yes
Tk(x) =​ x +​ k (mod n)
𝒟n (𝒯ℐn) Dihedral group 2n All transpositions and no
inversions:
Tk(x) =​ x +​ k, Ik(x) =​ k − x
(mod n)
𝒜𝑓𝑓n Affine group n ∙ φ(n) All affine transformations: no
f(x) =​ ax +​ b (mod n), a ∈ ℤn×,
b ∈ ℤn
𝒜n Alternating group n!/​2 All even permutations of ℤn no
𝒮n Symmetric group n! All permutations of ℤn no

214 • E x plor i n g M u s i cal S paces

 inversion operators acting on ℤ7 (which A/​B/​C (“A over B over C”) denote their
may be identified with generic pitch-​ appearance in that registral ordering, from
class space gpc). Which of these are even highest voice to lowest. We will refer to such an
permutations, and which are odd? arrangement as a contrapuntal configuration, or
(b) Do the same for all the transposition and a configuration for short. In the B♭-​Major Fugue
inversion operators on ℤ9. from Book I of the Well-​Tempered Clavier, if A
(c)  Do the same for ℤ10. denotes the subject, B the first countersubject,
and C the second countersubject, the first
Exercise 6.1.12 Answer the following complete configuration is C/​B/​A at m. 9, but
questions based on your knowledge of the same materials appear in a variety of other
transpositions and inversions on ℤ7, ℤ9, ℤ10, registral orderings later in the piece.
and ℤ12 from the above discussion and the The symmetric group 𝒮3 comprises the
preceding exercise. In parts (b) and (f), fill in six (=​3!) possible permutations of the three-​
each blank with the word even or odd to make element set S =​{A, B, C}: the identity E =​
the statements true. (A)(B)(C), two 3-​cycles (A, B, C) and (A, C, B),
and three 2-​cycles (A, B), (A, C), and (B, C). Note
(a) Consider the transposition operator Tk on carefully the notational distinctions among
ℤn. Let d be the greatest common divisor of {A, B, C}, which is the set S; (A, B, C), which
k and n, and let j =​ n ∕ d. In terms of d and j, is a permutation (a function p: S → S); and
what is the cycle type of Tk? A/​B/​C, which is a contrapuntal configuration
(b) If n is even, then Tk is an _​_​_​_​_​ (a musical deployment of the elements of S). If
permutation of ℤn if k is even, and an _​_​_​_​_​ A, B, and C are replaced with the elements 0, 1,
permutation if k is odd. If n is odd, then and 2 of ℤ3, then E and the two 3-​cycles are the
every Tk is an _​_​_​_​_​ permutation of ℤn.8 transposition operators T0, T1, and T2, while the
(c) Now consider the inversion operator Ik on three 2-​cycles are the inversion operators I0, I1,
ℤn. Suppose n is an even number, say n =​2r. and I2. (The reader may construct or imagine
If k is also even, what is the cycle type of Ik? a three-​note clock diagram—​an equilateral
(d) If n =​2r and k is odd, what is the cycle triangle with vertices labeled A, B, and C—​
type of Ik? and visualize the 3-​cycles as rotations and
(e) If n is an odd number, say n =​2s +​1, what is the 2-​cycles as reflections.) The two 3-​cycles
the cycle type of Ik? are inverses of each other in the group
(f) If n =​0 (mod 4), then Ik is an _​_​_​_​_​ ((A, B, C)−1 =​(A, C, B)), and every 2-​c ycle is an
permutation of ℤn if k is even, and an _​_​_​_​_​ involution ((A, B)−1 =​(A, B)). The identity and
permutation if k is odd. If n =​1 (mod 4), the two 3-​cycles are even permutations, while
then every Ik is an _​_​_​_​_​ permutation. If the three 2-​cycles are odd permutations. In the
n =​2 (mod 4), then Ik is an _​_​_​_​_​ particular case n =​3, the group 𝒮3 is the same
permutation if k is even, and an _​__​ ​_​_​ as 𝒟3 (a group of order 6), and 𝒜3 the same as
permutation if k is odd. Finally, if n =​3 𝒞3 (a group of order 3); these relationships do
(mod 4), then every Ik is an _​_​_​_​_​ not hold for any larger values of n.
permutation.9 After the initial configuration C/​B/​A at m. 9,
(g) As noted above, every permutation group the three subjects subsequently appear in
always consists either of even permutations the configurations A/​C/​B (m. 13), B/​A/​C
exclusively or of even and odd permutations (m. 22), C/​B/​A again (m. 26), B/​A/​C again
in equal numbers. For which values of n (m. 35), A/​B/​C (m. 37), and a final appearance
does the cyclic group 𝒞n consist entirely of of B/​A/​C (m. 41). The transformation network
even permutations? For which values of n of Figure 6.1.14a charts this progression of
does the dihedral group 𝒟n consist entirely counterpoints, linking successive configurations
  of even permutations? with the permutations in 𝒮3 that relate them.
Crucial to the construction of this network is
Example 6.1.13: Bach, Fugue in B♭ Major the fact that the group of six permutations acts
(WTC I) on the space of six configurations in simply
Triple counterpoint provides a straightforward transitive fashion, as defined in Section 5.6: for
musical motivation for studying permutations any two configurations x and y, there is always
on a three-​element set.10 If the three subjects in exactly one permutation p in 𝒮3 such that p(x) =​
a contrapuntal texture are A, B, and C, let y, so there can never be any doubt about which

Permutations, Isomorphisms, and Other Topics • 215

permutation is being applied.11 From C/​B/​A to voices while leaving C in the bass, producing the
A/​C/​B, the permutation is the 3-​cycle (A, B, C); new configuration A/​B/​C. The same 2-​cycle is
that is, A is replaced in the texture by B, B by C, then repeated, yielding B/​A/​C once more, the
and C by A. This same cyclic permutation then final appearance of this triple counterpoint in
recurs twice more in succession, yielding a third the fugue. The recurrences of the configurations
configuration B/​A/​C before returning to the C/​B/​A and B/​A/​C are shown in Figure 6.1.14a
original C/​B/​A . The musical effect of this cycle by the identity E, the dashed arrows signaling
is that each of the three subjects moves one indirect transformations in the manner of
position lower in the texture at each iteration, Figure 3.5.3a and other previous examples.
rotating back to the top after reaching the In terminology from Section 3.5, Figure
bottom. 6.1.14a is an event network that displays
To this point the transformations have the various contrapuntal configurations in the
been completely systematic, but rather than order in which they are heard. This network
cycling repeatedly through these same three may be contrasted with the spatial network
configurations, Bach now begins to vary his of part (b) of the figure, which exhibits a
procedure. The first permutation other than (A, space of the six possible configurations and
B, C) to appear is (A, C, B), the other available shows how the fugue traces a route through
3-​cycle. (A, C, B) is the inverse of (A, B, C), four of them. The large triangle shows the
however, so even though the transformation three cyclically related configurations C/​B/​A ,
is new, it does not lead to a new configuration, A/​C/​B, and B/​A/​C, the first three configurations
but instead back to the configuration B/​A/​C presented. These are related among themselves
heard just before. To escape this set of three by the permutations (A, B, C), a clockwise
configurations Bach next introduces a new move around the triangle, and (A, C, B), the
permutation in the form of the 2-​cycle (A, B). corresponding counterclockwise move. These 3-​
This 2-​cycle, the first odd permutation to be cycles, together with the identity permutation
employed, interchanges A and B in the top two E, constitute the cyclic group 𝒞3 generated by

FIGURE 6.1.14 Two transformation networks showing permutations of triple counterpoint in Bach,
Fugue in B♭ Major (WTC I): (a) an event network; (b) a spatial network

216 • E x plor i n g M u s i cal S paces

(A, B, C); application of any element of that 6.2 GROUP TABLES AND
group to any configuration in the outer triangle CAYLEY DIAGRAMS
will never lead out of that triangle. In a sense
to be defined precisely in Section 6.5, these In this section we present two convenient ways
three configurations constitute one orbit of the to describe the behavior of finite groups, one
group 𝒞3. The other three configurations are involving a table, the other involving a directed
arranged in an inner triangle, each positioned graph. A group table generalizes the grade-​school
near the configuration in the outer triangle to multiplication table, showing how all possible
which it is related by the 2-​cycle (A, B). These pairs of elements combine under the group
inner configurations constitute another orbit operation. Two group tables are shown in Figure
of 𝒞3, but Bach does not exploit this property. 6.2.1: part (a) shows the table for the modu-
The only evidence of this inner orbit in the lar group (ℤ4, +​), while (b) shows the table for
fugue is the single appearance of A/​B/​C near the the permutation group (𝒮3, ●) (or equivalently
end, which both arises from and immediately (𝒟3, ●)), acting on the three-​element set {A, B, C},
returns to B/​A/​C via the (A, B) exchange. as discussed above in Example 6.1.13.
The group operation in ℤ4 is mod-​4 addi-
Exercise 6.1.15 Write a comparative discussion tion, while that in 𝒮3 is composition of func-
of the two networks of Figure 6.1.14, using tions, so these tables might more accurately be
as a model the comparison of event networks called an “addition table” and a “composition
and spatial networks for Webern’s “Herr Jesus table,” respectively—​but as noted in Chapter
mein”
   from Example 3.5.14. 5 it is common to use multiplicative terminol-
ogy in discussing general groups. In general, the
Exercise 6.1.16 Study the triple counterpoint “product” xy (or x ⊗ y, whatever the operation ⊗
in the following two works by Bach, producing may be) of group elements x and y appears in the
for each an event network and a spatial network group table at the intersection of the row labeled
similar to those in Figure 6.1.14: x and the column labeled y, so the circled cells in
the two tables show that 2 +​3 =​1 in ℤ4 and that
(a) Fugue in C Minor from WTC I12 the equation (B, C)(A, C, B) =​(A, C) is true of
(b) Three-​Part Sinfonia in F Minor, permutations in 𝒮3. Because ℤ4 is a commutative
  BWV 79513 group, 2 +​3 is the same as 3 +​2; this commuta-
tivity is visible in the fact that the table for ℤ4 is
Exercise 6.1.17 Use permutations to study the symmetrical about a diagonal line. The group 𝒮3
registral ordering of the chords played by the flute, is not commutative, so its table lacks this sym-
clarinet, and violin in mm. 1–​9 of “Eine blasse metry, and indeed confirms that (B, C)(A, C, B) is
Wäscherin” from Schoenberg’s Pierrot lunaire. different from (A, C, B)(B, C).
For example, the two chords in m. 1 illustrate From Figure 6.2.1a, the reader can probably
configurations that might be notated C/​V/​F imagine the appearance of the group table for a
(“clarinet
   over violin over flute”) and F/​V/​C.14 general modular group ℤn with little difficulty.

FIGURE 6.2.1 Two group tables

Permutations, Isomorphisms, and Other Topics • 217

The way in which the table in (b) generalizes Exercise 6.2.4 Verify that the row/​column
to larger permutation groups, however, is less permutation property holds for all of the group
apparent. Even the order in which the permuta- tables
   in Figure 6.2.1 and Exercise 6.2.3.
tions should be listed in the table is not obvious;
in principle, a group table may be constructed The directed graph associated with a group,
using any order at all. It is conventional to list known as the Cayley diagram for the group, is a
the identity element first, however, and the way of depicting group structure visually. We will
order chosen in (b), in which the two 3-​cycles see that in some cases, the Cayley diagram for a
follow immediately after the identity, has the group may closely resemble a diagram that we
further advantage that the three elements of have already constructed for a space on which the
the subgroup 𝒞3 (or equivalently 𝒜3) appear same group acts. In a connection that has per-
first, so that the group table for 𝒞3 appears as haps been underappreciated in transformation
the upper-​left quadrant of the larger table. theory, Cayley diagrams are also closely related to
transformation networks; in fact they are trans-
Exercise 6.2.2 Verify that the entries in all the formation networks of a particular kind.17
cells of the group table for 𝒮3 in Figure 6.2.1b Cayley diagrams corresponding to the tables
are  correct. in Figure 6.2.1 are shown in Figure 6.2.5a–​b. The
reader can probably see an intuitive connection
Exercise 6.2.3 Construct group tables for the between the graph in 6.2.5a and the cyclic group
following small finite groups: ℤ4, but it is important to understand exactly
how the diagram is constructed. The graph has
(a) The modular group ℤ6 vertices labeled 0, 1, 2, and 3, one vertex for each
(b) The dihedral group 𝒯ℐ4 =​{T0, T1, T2, T3, I0, element of the group. The graph also includes
I1, I2, I3} acting on the set ℤ4 (construct this four arrows, which, as the legend below the
table so that the upper-​left quadrant is the graph indicates, are associated with the element
table for 𝒯4)15 1. An arrow x → y appears in the graph whenever
(c) The multiplicative group {1, −1} the equation x +​1 =​y holds in the group; hence
(d)  The mod-​12 units group ℤ12× =​{1, 5, 7, 11} the four arrows are 0 → 1, 1 → 2, 2 → 3, and 3 →
0. In effect, the group element 1 is acting as a
In constructing a few group tables as in the transformation, namely the mod-​4 transposition
above exercise, one becomes aware of charac- T1(x) =​ x +​1, and the graph shows the action of
teristic patterns, one of which deserves special that transformation. For every x there is exactly
mention. In the table for a group G, the row one y such that x +​1 =​y, so every vertex in the
labeled with an element a (“row a” for short) graph has out-​degree 1 (exactly one arrow point-
contains all the products ax as x varies through ing out from it); likewise for every y there is
the elements of G. Recall from Section 5.3 that exactly one x satisfying this equation, so every
if b is any other element of G, the equation ax vertex also has in-​degree 1.
=​ b always holds for exactly one x in G. It fol- Because of the special role played by the ele-
lows that the element b appears exactly once in ment 1 in its construction, this graph may more
row a. This property holds for any a and b in any precisely be called “the Cayley diagram for ℤ4 on
finite group at all, and it is true for columns as the generator 1.” In principle we could include
well (every b occurs exactly once in column a, as additional arrows associated with adding 2 or 3
can be seen by considering the equation xa =​ b). to each element of the group, but these are not
That is, every element of the group appears exactly necessary: the element 1 generates the group
once in every row and column of the group table. (adding 2 is equivalent to adding 1 twice, and so
Each row and column is therefore a permuta- on), so the complete behavior of the group oper-
tion of the elements of the group. This prop- ation can be deduced from the arrows shown in
erty—​which we call the row/​column permutation the figure. Because of this generating property,
property for group tables—​is familiar to those the graph is strongly connected as defined in
who have worked with twelve-​tone matrices or Section 3.4—​that is, every vertex is accessible
Sudoku puzzles, and as in both of those settings, from every other vertex by a directed path.
can serve as a quick check on the accuracy of a
group table: if an element appears twice in a row Exercise 6.2.6
or column, or not at all, something is wrong with (a) Construct a Cayley diagram for ℤ4 on the
the table.16 generator 3. This means that an arrow x → y

218 • E x plor i n g M u s i cal S paces

FIGURE 6.2.5 Three Cayley diagrams: (a) (ℤ4, +​) on the generator 1; (b) (𝒮3, ●) on the generators x =​
(A, B, C) and y =​(A, B); (c) (ℤ, +​) on the generator 1

in the graph corresponds to the equation depict “multiplication” (composition) by the per-
x +​3 =​y in ℤ4. mutation x =​(A, B, C)—​that is, an arrow p → q
(b) The number 2 by itself does not generate ℤ4, appears whenever the permutation equation
and therefore cannot be used to construct p ● (A, B, C) =​q holds. For instance, the equation
a proper Cayley diagram for ℤ4. If you (A, B, C)(A, B, C) =​(A, C, B) is represented by
try to construct such a graph, which of the arrow (A, B, C) → (A, C, B). A second set of
the properties mentioned above will fail arrows, drawn with dashed lines, represents
  to hold? multiplication by y =​(A, B); for instance, the
arrow (A, B, C) ⇢ (B, C) represents the equation
Figure 6.2.5a is an unusually simple Cayley (A, B, C)(A, B) =​(B, C). The dashed arrows are
diagram, because the group ℤ4 is cyclic. When bidirectional because y is an involution. Each ver-
a group is not generated by a single element, a tex now has out-​degree 1 and in-​degree 1 relative
more elaborate diagram is needed, as Figure to both sets of arrows in the graph, and therefore
6.2.5b illustrates in the case of the permuta- out-​degree 2 and in-​degree 2 for the graph as a
tion group 𝒮3 acting on the set {A, B, C}. Again whole. The fact that x and y together generate the
there is one vertex for each of the elements of group is confirmed by the fact that the graph as
the group, now six in number. No single ele- a whole is strongly connected: if we can get from
ment generates 𝒮3, so a single cycle of six arrows any vertex to any other by following some chain
does not suffice to convey the structure of this of arrows representing x and y, that means that
group. In such cases the construction of a Cayley every element in the group can be expressed as
diagram requires choosing a small set of generat- some combination of these two. The subgraph
ing elements for the group. The graph in 6.2.5b formed by the solid arrows alone, however, is a
is derived from the fact that the 3-​cycle x =​(A, disconnected graph, as is the subgraph formed by
B, C) and the 2-​cycle y =​(A, B) together generate the dashed arrows alone (the former graph has
𝒮3; that is, the figure is the Cayley diagram for 𝒮3 two components, the latter three), because nei-
on the generators x and y. (If the fact that x and y ther element generates the group by itself.
generate 𝒮3 is not obvious, we shall see that the As before, we could in principle add arrows
graph itself confirms this property.) Solid arrows showing the results of multiplying by other

Permutations, Isomorphisms, and Other Topics • 219

group elements, but once again this is unnec- (a) Show how the graph illustrates the equation
essary: with practice, any product of group ele- yx2y =​ x.
ments can be deduced quickly from the graph in (b) What simpler expression is xyx equal to?
Figure 6.2.5b. Consider, for example, the prod- (c) What simpler expression is yxy equal to?
uct (A, C, B)(A, C). To use the graph to multiply (d) What simpler expression is xyx2y equal to?
by (A, C), first look for a directed path from the (e) The expression xyxy represents a path that
identity E to (A, C). One such path consists of begins and ends at the same vertex of the
one dashed arrow followed by one solid arrow, graph. What does this mean about the
E ⇢ (A, B) → (A, C), which tells us that (A, C)   group element xyxy?
is equal to the product of the generators repre-
sented by those arrows, yx =​(A, B)(A, B, C). To Theoretically even an infinite group can
multiply any other element by (A, C), simply fol- have a Cayley diagram; such a diagram will be
low that same dashed-​then-​solid chain of arrows. an infinite graph as described in Section 3.3.
From a starting point at (A, C, B), the only possible A portion of a Cayley diagram for the infinite
dashed-​then-​solid path takes the form (A, C, B) ⇢ group (ℤ, +​) on the generator 1 is shown in
(A, C) → (B, C), ending with (B, C), so we may Figure 6.2.5c.
conclude that (A, C, B)(A, C) =​(B, C), as examina- We know from Exercise 6.2.6 that there
tion of the permutations (or consultation of the may be more than one way to construct the
table in Figure 6.2.1b) will confirm. The fact that Cayley diagram for a group. Constructing such
a dashed-​then-​solid arrow chain leads to a differ- a diagram requires first choosing a generating
ent place from a solid-​then-​dashed chain tells us set—​ a set of group elements that together
that the generators x and y do not commute, and generate the group. In the case of 𝒮3 we chose
therefore that the group is noncommutative. the generators (A, B, C) and (A, B), but various
The reader may have noted that there is more other pairs of elements could have been cho-
than one possible path in the graph from E to sen instead (see Exercise 6.2.8 below). Perhaps
(A, C). Rather than following the just-​described even more important, we must decide how to
path E ⇢ (A, B) → (A, C), we could choose arrange the vertices on the page; changing the
instead to traverse two solid arrows followed by layout can dramatically affect the appearance
a dashed arrow: E → (A, B, C) → (A, C, B) ⇢ (A, C). of a graph. Symmetrical arrangements are gen-
Alternatively, we could consider traversing a solid erally preferred—​the symmetry of the group
arrow in reverse followed by a dashed arrow, structure is an important part of what the dia-
which might be represented E ← (A, C, B) ⇢ gram is supposed to convey—​but larger groups
(A, C). Each of these paths describes a way to often exhibit complex kinds of symmetry that
express the permutation (A, C) in terms of the are difficult to capture adequately in a two-​
generators x and y. Because a solid arrow rep- dimensional picture.
resents x and a dashed arrow represents y, the
three expressions are yx, x2y, and x−1y. (Because Exercise 6.2.8
a solid arrow traversed normally indicates multi- (a) Construct a Cayley diagram for the
plication by x, a solid arrow traversed backward permutation group 𝒮3 (acting, as above,
indicates multiplication by x−1.) The reader may on the set {A, B, C}) on the generators
verify that when x and y are the permutations (A, C, B) and (B, C). At first, leave the
(A, B, C) and (A, B), respectively, the products yx, vertices positioned and labeled as they
x2y, and x−1y are indeed all equal to (A, C). These are in Figure 6.2.5b; change only the
observations begin to suggest a way in which arrows. Then consider the possibility of
Cayley diagrams may be useful in deducing alge- repositioning the vertices so that the
braic relations among the generators of a group. arrows shown in your new graph form
The following exercise explores this idea. a more symmetrical configuration. As
a directed graph, is your new graph
Exercise 6.2.7 Use the Cayley diagram in isomorphic to the graph in 6.2.5b?
Figure 6.2.5b to answer the following questions (b) Construct another Cayley diagram for 𝒮3,
about the group 𝒮3 and its generators x =​(A, B, this time on the generators (A, C) and (B, C).
C) and y =​(A, B). Confirm each algebraic result Is this graph isomorphic to Figure 6.2.5b?
by an appropriate permutation calculation or by (c) Is it possible to construct a Cayley diagram
referring to the group table in Figure 6.2.1b.   for 𝒮3 on (A, B, C) and (A, C, B)?

220 • E x plor i n g M u s i cal S paces

Exercise 6.2.9 The symmetry inherent in the and is also connected to the row/​column permu-
structure of a group ensures that the same tation property of group tables.18
configurations of arrows occur everywhere in The vertices of a Cayley diagram for a group
the graph. Once a Cayley diagram has been G are traditionally considered to represent ele-
constructed, therefore, any element of the ments of G itself, and the action of G on itself, as
group may occupy any vertex of the graph; detailed in the above discussion of Cayley’s theo-
as soon as one element has been assigned to rem, is always simply transitive. When a group
a vertex, the other vertex labels can then be G acts on some other set S in simply transitive
determined from the generators. Illustrate this fashion, the vertices may be labeled instead with
idea by relabeling Figure 6.2.5b so that the elements of S, and the resulting graph becomes a
identity E appears at the vertex that was labeled transformation network depicting the action of
(A, C) in the original figure. Leave all arrows G on S. This is the case in Figure 6.1.14b, where
unchanged. The resulting graph is still a Cayley the simple transitivity of the action of 𝒮3 on the
diagram for 𝒮3 on the generators (A, B, C) and space of contrapuntal configurations ensures a
(A, B), but each group element now occupies a one-​to-​one correspondence between elements
different
   location in the graph. of the group and elements of the space; the net-
work showing the action of G on S describes the
The reader may have noticed a strong likeness group structure as completely as the original
between the Cayley diagram in Figure 6.2.5b and Cayley diagram labeled with the elements of G.
the spatial transformation network of Figure Moreover, because Lewin’s theorem (Theorem
6.1.14b, which involved the same permuta- 5.6.7) asserts an equivalence between simply
tion group. The arrows in 6.1.14b represented transitive group actions and interval spaces,
transformations in the specific musical example the same diagram may be regarded as a picture
at hand, but we could easily image all of Figure of an interval structure on S, described, like the
6.2.5b’s arrows transferred onto the earlier original Cayley diagram, in terms of generators
graph. A more significant difference between of the group.
the two graphs is that in 6.1.14b the vertices Cayley diagrams for the transformation
represented contrapuntal configurations, while groups 𝒯12 (or 𝒞12) and 𝒯ℐ12 (or 𝒟12) are shown
in 6.2.5b the vertices represent group elements in Figure 6.2.10. For groups of these sizes (order
(permutations). But the action of the group 12 and 24, respectively), Cayley diagrams are
works essentially the same way in both cases: a simpler in appearance than group tables, as one
permutation in 𝒮3 maps one contrapuntal con- can quickly confirm by starting to construct a
figuration to another, but also maps one group 24-​by-​24 table for 𝒯ℐ12. These graphs show pat-
element to another by multiplication within the terns characteristic of cyclic and dihedral groups,
group. This observation is known as Cayley’s the- and in fact they are discernibly similar to the
orem: every element in a group may be regarded graphs in Figure 6.2.5a–​b (ℤ4 is cyclic, and recall
as a transformation acting on the group by that 𝒮3 is the same group as 𝒟3). The graph for
multiplication. 𝒯ℐ12 in 6.2.10b reintroduces a convention noted
More explicitly, every element a of a group in some examples in Chapters 3 and 4: because
G determines a function Multa: G → G defined the inversion operator I is an involution, we have
by the equation Multa(x) =​ xa for all x in G. dispensed with the arrowheads on the dashed
This function may be called right-​multiplication lines representing I, with the understanding
by a, and is conceptually similar to multiplica- that every such arrow points both ways. The sub-
tion operators such as M7 (though in a general scripts on the T and I operators in the vertices of
group the operation might be something other this graph proceed in opposite directions around
than ordinary multiplication). By identifying the the two circles, as the inversion aligned with Tk
group element a with the transformation Multa, is Tk ● I, which is the same as I ● T−k, or I−k. The
every group may be regarded as a permutation arrows in the two circles are oppositely directed
group acting on itself, and therefore as a sub- as well, reflecting the fact that T1I is equal not
group of a symmetric group, even if the original to IT1 but to IT1−1. (Tracing a solid-​then-​dashed
conception of the group did not involve trans- arrow chain, that is, does not produce the same
formations or permutations at all. This theorem result as tracing a dashed-​then-​solid chain—​but
is the principle underlying the interpretation of it does if the final solid arrow is traversed in
Cayley diagrams as transformation networks, reverse.) The reverse alignment of the subscripts

Permutations, Isomorphisms, and Other Topics • 221

FIGURE 6.2.10 Cayley diagrams for two transformation groups on pitch-​class space: (a) the cyclic
group (𝒯12, ●) on the generator T1; (b) the dihedral group (𝒯ℐ12, ●) on the generators T1 and I =​ I0

could be “fixed” by labeling the inversion opera- relation we have observed between the depic-
tors differently, using for instance Jk =​ TkI rather tions of 𝒮3 in the transformation network of
than Ik =​ ITk. But the oppositely directed arrows 6.1.14b and the Cayley diagram of 6.2.5b.
are intrinsic to the structure of the group and Several other graphs from Chapter 4 are
will not be altered by any relabeling.19 Cayley diagrams in disguise as well. Figure 4.1.1a
is effectively the Cayley diagram for a different
Exercise 6.2.11 Construct Cayley diagrams group acting on triad, the group 〈T1, P〉 gen-
for the cyclic and dihedral groups 𝒯7 and 𝒯ℐ7 erated by T1 and the parallel transformation P.
acting on ℤ7 (which may be identified with The structure of this group will be examined in
generic
   pitch-​class space gpc). Section 6.4, but it may be noted here that the
arrows in both circles point in the same direc-
The close resemblance between the Cayley dia- tion, illustrating the fact that the generators T1
gram for 𝒯12 and the structure of a space such as and P commute (unlike T1 and I in 𝒯ℐ12).
pc on which this group acts should be apparent. In a similar way, the three-​dimensional graphs
The vertices in Figure 6.2.10a could be labeled in Figure 4.1.12 suggest Cayley diagrams for two
with pitch-​class numbers 0, …, 11 rather than different groups of order 48 acting (in simply
transpositions T0, …, T11; in this way the Cayley transitive fashion) on the space of all 48 forms
diagram shows the simply transitive action of of a twelve-​tone row. Each of these groups is gen-
𝒯12 on pc. It may come as more of a surprise to erated by three generators: the transposition T1,
realize that we have essentially encountered the the retrograde operator R, and either the fixed
Cayley diagram for 𝒯ℐ12 in another guise pre- inversion operator I (in 4.1.12b) or the contextual
viously as well, in the double-​circle diagram of inversion J (in 4.1.12a). Drawn as more traditional
Figure 4.1.1b. When constructing this graph, we Cayley diagrams, these graphs might incorporate
assumed that 𝒯ℐ12 was acting on the space triad arrows of three different styles (or colors) for the
of major and minor triads; because that action three generators. The group 〈T1, R, I〉 of 4.1.12b is
is simply transitive, the action—​and therefore called the serial group 𝒮𝑒𝑟12, as it contains exactly
the diagram—​displays the entire structure of the row transformations inherent in the twelve-​
the group. (To read 4.1.1b as a complete Cayley tone system as codified by Schoenberg.20
diagram we must infer that all edges shown are
arrows, continuing the patterns shown in which Exercise 6.2.12 Review the definition of the
T1 arrows circumnavigate the two circles in contextual inversion operator J from Section 4.1.
opposite directions and I arrows connect the two Verify that J commutes with both T1 and R. The
circles radially.) The relation between 4.1.1b and group 〈T1, R, J〉 associated with Figure 4.1.12a
the Cayley diagram of 6.2.10b is similar to the is therefore a commutative counterpart to the

222 • E x plor i n g M u s i cal S paces

serial group, in much the same way that Exercise 6.2.14
G =​ 〈T1, P〉 is a commutative counterpart to (a) Construct a Cayley diagram for the mod-​12
𝒯ℐ 12 on the space triad.
   units group ℤ12× =​{1, 5, 7, 11}, using 5 and
7 as the two generators.
Figure 6.2.13 is a Cayley diagram for the group (b) The multiplication operator M5 does not
G =​ 〈T1, M〉 generated by the transposition oper- appear in the group G of Figure 6.2.13.
ator T1 and the multiplication operator M7 (here Construct a Cayley diagram for the group
shortened to M for convenience). Like 𝒯ℐ12, G is Gʹ =​ 〈T1, M5〉.
a noncommutative group of order 24. G consists (c) The affine group 𝒜𝑓𝑓12 is generated by T1, M
of the twelve transpositions Tk and twelve prod- (=​ M7), and I. This group has 48 elements:
ucts TkM. The operator M, like I, is an involution, twelve transpositions Tk, twelve inversions
but M interacts with transpositions differently Ik, twelve elements of the form TkM (as in
from I, and that difference is reflected in the Figure 6.2.13), and twelve of the form IkM.
appearance of the graph. The graph includes, for When the elements are named in this way,
example, a T1 arrow pointing from the element which one is equal to M5?21
M to the element T7M, illustrating the algebraic (d) Construct a Cayley diagram for 𝒜𝑓𝑓12, using
identity MT1 =​ T7M. This equation can be verified three different styles (or colors) of arrows
by the mod-​12 calculations   for the three generators T1, M, and I.22

T In general, the affine group 𝒜𝑓𝑓n consists of all

x M
→ 7 x 
1
→ 7 x + 1, transpositions Tk on ℤn, all multiplication opera-
T
x  7
→ x + 7 M → 7( x + 7 ) tors Ma with a coprime to n, and all other func-
= 7 x + 49 = 7 x + 1 (mod 12). tions generated from these by composition. In
other words, 𝒜𝑓𝑓n is the group generated by the
subgroups 𝒞n and ℳn; this definition accords
The twelve-​pointed star that appears in Figure with the description of 𝒜𝑓𝑓12 in the preceding
6.2.13 may be considered a visual representa- exercise because the group ℳ12 is generated by
tion of the relationship between the chromatic M7 and I. Generally I =​ M−1 ∈ ℳn, so 𝒜𝑓𝑓n contains
pitch-​class circle and the circle of fifths, as for- all inversion operators and therefore includes
malized by the operator M. The semitone and 𝒟n as a subgroup. Every affine transformation f:
the perfect fifth, the intervals that generate the ℤn → ℤn may be written in the form f(x) =​ ax +​ b for
two circles, are mapped to each other by M; recall some a in ℤn× and some b in ℤn. From Section 5.2
also Exercises 1.3.2b and 6.1.9b in this context. we know that there are φ(n) choices for the num-
ber a, where φ is Euler’s phi function, and there
are n choices for b, so 𝒜𝑓𝑓n is a group of order
n ∙ φ(n). For all n ≥ 7, this number is larger than
2n but smaller than n! ∕ 2, so 𝒜𝑓𝑓n takes its place
between 𝒟n and 𝒜n in the roster of permutation
groups acting on ℤn in Table 6.1.10.

Exercise 6.2.15
(a) In 𝒜𝑓𝑓12, let f(x) =​5x +​9 and g(x) =​7x +​6.
Calculate formulas for f −1, g−1, f ● g, and g ● f,
writing a formula of the form ax +​ b for each
of these functions.
(b) The permutations p and q from Exercise 6.1.6
both belong to 𝒜𝑓𝑓12. Write formulas of the
  form ax +​ b for both of these functions.

Exercise 6.2.16
(a) Consider P and L, the parallel and
leittonwechsel transformations acting on
triad. Both are involutions (PP =​ LL =​ E),
FIGURE 6.2.13 Cayley diagram for the group so they can generate other elements only
generated by T1 and M =​ M7 when combined in alternation: PL, PLP,

Permutations, Isomorphisms, and Other Topics • 223

 PLPL, … or LP, LPL, LPLP, … . There appear the labels identifying the generating arrows—​
to be infinitely many possible combinations we might convert this graph to Cayley diagrams
in these two lists, but in Exercise 4.2.12b for G2 and G3.
we observed that in fact PLP =​ LPL. Use this When we studied graph isomorphism in
fact to deduce other identities among the Section 3.2, an isomorphism was defined as
elements in the two lists, and then explain a structure-​preserving bijection. In the context
why P and L together generate a group of graphs, the essential structure that an iso-
of order 6, known as the PL group or the morphism must preserve is the adjacency rela-
hexatonic triad group.23 tion, which determines which pairs of vertices
(b) Draw a Cayley diagram for the hexatonic are joined by edges: a mapping from the ver-
triad group, on the generators P and L. tices of one graph to those of another graph
  Compare your graph to Figure 3.5.8b. is structure-​ preserving if adjacent vertices
always map to adjacent vertices. In the con-
Most of our musically motivated Cayley text of groups, the essential structure is pro-
diagrams are fairly small and straightforward. vided by the group operation, and a mapping
Cayley diagrams for larger groups can be stun- is structure-​preserving if corresponding ele-
ningly intricate and beautiful, revealing in strik- ments of the two groups combine in the same
ing ways the symmetry inherent in the structure way according to their respective operations.
of a group. See Carter 2009 for examples. Construction of such a function is one of the
most common ways to demonstrate relation-
6.3 ISOMORPHISM ships between two groups.
OF GROUPS To make this intuition precise, suppose (G, ⊕)
and (Gʹ, ⊗) are two groups. The group opera-
Figure 6.3.1 shows group tables for three groups tions in the two groups need not be the same, so
of order 4, called G1, G2, and G3. The group G1 we use two different symbols for them. A func-
is the modular group ℤ4, already familiar from tion f: G → Gʹ is called a homomorphism (more
Figure 6.2.1a; G2 is the cyclic group 𝒯4 =​{T0, specifically a group homomorphism) if, whenever
T1, T2, T3} that acts on ℤ4; and G3 is an abstract x, y, and z are elements of G satisfying x ⊕ y =​ z,
group with an identity element e and other ele- then the corresponding property f(x) ⊗ f(y) =​
ments called a, b, and c. f(z) holds among the elements f(x), f(y), and f(z)
Examination of the three tables shows that in the group Gʹ. To write this requirement a bit
all three exhibit the same basic pattern of more compactly, f is a homomorphism if the
entries: any one of the tables can be converted equation f(x) ⊗ f(y) =​ f(x ⊕ y) is true for all x and
to any other simply be renaming the elements y in G. This property is illustrated in Figure 6.3.2:
of the group. In a sense all three groups have the the dashed lines within each group show two
same essential structure; these are examples of elements combining to determine a third via the
isomorphic groups. All three are cyclic groups group operation, and the solid arrows show that
as defined in Section 5.3; they are generated the function f relates corresponding elements of
by the elements 1, T1, and a, respectively. The the two groups. An isomorphism from G to Gʹ is
isomorphism is visible also in the Cayley dia- a homomorphism that is also bijective (one-​to-​
grams for the groups: a Cayley diagram for G1 one and onto). If an isomorphism from G to Gʹ
appeared in Figure 6.2.5a, and it is easy to visu- exists, the groups G and Gʹ are said to be isomor-
alize how—​by changing the vertex labels and phic, written G ≅ Gʹ.

FIGURE 6.3.1 Group tables for three isomorphic groups of order 4

224 • E x plor i n g M u s i cal S paces

 may be ordered differently in the tables. The fol-
lowing exercise illustrates.

Exercise 6.3.4
(a) Construct a second group table for the
group G3 from Figure 6.3.1, but instead of
listing the elements in the order e, a, b, c
when constructing the table, list them
in the order e, b, a, c. Because the group
itself is unchanged, all the products will be
unchanged; for example, the table should
show that a ⊗ a =​ b, as the original table for
FIGURE 6.3.2 A group homomorphism G3 does. Does your new table have the same
pattern as the three tables in Figure 6.3.1?
(b) We already know that groups G1 and G3 are
In the case of the groups in Figure 6.3.1, an isomorphic: an isomorphism h: G1 → G3
example of an isomorphism from G1 to G2 is the was exhibited above. The existence of one
function f defined by f(i) =​ Ti for all i =​0, 1, 2, 3. isomorphism, however, does not imply that
This isomorphism makes the correspondence every bijection between the two groups is an
of the group tables explicit, as it maps each ele- isomorphism. Use your observation from
ment of G1 to the element that appears in all part (a) to construct a bijective function hʹ:
the corresponding positions in the table for G2   G1 → G3 that is not an isomorphism.
(row and column headings included). An iso-
morphism h: G1 → G3 is defined by h(0) =​ e, h(1) For every n, the modular group (ℤn, +​) and
=​ a, h(2) =​ b, and h(3) =​ c. As was the case with the transformation group (𝒯n, ●) acting on ℤn are
isomorphic graphs, isomorphic groups may isomorphic. The groups G1 and G2 in Figure 6.3.1
reasonably be considered “the same group” for are ℤ4 and 𝒯4, and in Chapter 5 we remarked
many purposes, as they have identical under- informally on the isomorphism between ℤ12 and
lying structure, the isomorphism amounting 𝒯12 (the interval group and transposition group
to nothing more than a renaming of the ele- of pc). One example of an isomorphism from ℤ12
ments. Nevertheless, as was also the case with to 𝒯12 is the function f defined by f(i) =​ Ti. Of the
graphs, isomorphic groups may appear super- many possible bijections between ℤ12 and 𝒯12,
ficially quite different, as the groups G1, G2, only a few are isomorphisms; in Section 6.5 we
and G3 illustrate: the three are isomorphic, but will determine exactly which functions these are.
G1 is a modular group of numbers, G2 a trans-
formation group, and G3 an abstract group of Exercise 6.3.5 Use the definition of a group
undefined symbols. homomorphism to explain why the following
two simple properties must always hold:
Exercise 6.3.3 Consider the group G4 =​{T0,
T3, T6, T9} consisting of four transposition (a) If f: G → Gʹ is a homomorphism, e is the
operators on pitch-​class space pc. Construct a identity element of G, and eʹ is the identity
group table for G4. Verify that this table has the element of Gʹ, then f(e) =​ eʹ.
same pattern as the tables in Figure 6.3.1, and (b) If f: G → Gʹ is a homomorphism, then f(x−1)
construct
   an isomorphism from G1 to G4.   =​ (f(x))−1 for every x in G.

It is true, as the above examples illustrate, Exercise 6.3.6 Confirm that isomorphism is an
that if one group table is converted to another equivalence relation on groups; that is, explain
by a simple renaming, the two groups are iso- why the relation ≅ is reflexive, symmetric, and
morphic. Conversely, if two groups are isomor- transitive.
   24

phic, then a table for one group may always be

converted to a table for the other by renaming. It Because isomorphism is an equivalence
is a mistake to conclude, however, that isomor- relation, it divides all groups into equivalence
phism can always be judged by noting whether classes, or isomorphism classes, isomorphic
two given group tables share the same pattern, groups always belonging to the same isomor-
because the elements of two isomorphic groups phism class. The concept of isomorphism classes

Permutations, Isomorphisms, and Other Topics • 225

is implicit in our use of abstract labels such as 𝒞n the row/​column permutation property to
and 𝒟n for cyclic and dihedral groups: 𝒞12 may deduce the remaining entries in the table.
be used as a name for any group in the entire Compare the result with your table from
isomorphism class of cyclic groups of order 12, part (a), and conclude that every group of
of which (ℤ12, +​) and (𝒯12, ●) are two specific   order 3 is isomorphic to ℤ3.
instances. The division into isomorphism classes
provides a useful way to classify groups. In the Speaking loosely, we may say based on the
case of the smallest groups, this classification is above observations that “there is only one group
quite simple. For example, it should be clear that of order 1, one group of order 2, and one group of
there is only one possible structure for a trivial order 3.” The smallest n for which there are two
group of order 1; that is, all groups of order 1 are different—​that is, non-​isomorphic—​groups of
isomorphic. There is, after all, only one cell in the order n is n =​4. One group of order 4 is the cyclic
group table for such a group, and whether this group ℤ4 or 𝒞4, represented by any of the tables
single element (which is automatically the iden- in Figure 6.3.1. But now consider the group G5 =​
tity element of the group) is called 0 or 1 or e or {T0, T6, I0, I6} acting on pc, whose table appears
something else does not matter much. in Figure 6.3.9.
It is almost as easy to see that all groups of
order 2 are isomorphic. Suppose that G =​({e, x}, ⊗) Exercise 6.3.10
is some group of order 2, where e is the identity (a) Use the formulas from Theorem 5.5.6 to
element, x is the other element, and ⊗ is the verify that the entries in the table for G5 in
group operation. The table for G has four cells, Figure 6.3.9 are correct.
but three of those are automatic simply because (b)  Verify that G5 is a group.25
e is the identity: the first row consists of e ⊗ e =​ e
and e ⊗ x =​ x, and the first element in the second The pattern of the table for G5 is different
row is x ⊗ e =​ x. The remaining entry in the table, from the pattern of the three tables in Figure
x ⊗ x, then can only be e, because of the row per- 6.3.1. For instance, the same element T0 appears
mutation property of group tables (each element in all four places down the diagonal of this table,
must appear once in row 2). We have shown that which was not true of any of the earlier tables.
the group table for G has (e, x) as its first row and Exercise 6.3.4 demonstrated that group tables
(x, e) as its second. This argument shows that may have different patterns even if the groups
the group tables for all groups of order 2 have are isomorphic; perhaps a simple reordering
the same pattern, so all such groups are isomor- could give this new table the same pattern as the
phic. This pattern is, of course, the pattern of the others. But that is not the case here: elements
group table for (ℤ2, +​), so we may conclude that on the diagonal of a group table are products
every group of order 2 is isomorphic to ℤ2 (or of the form x ⊗ x of an element with itself, and
equivalently to 𝒞2). therefore will remain on the diagonal following
any possible reordering. So the fact that only one
Exercise 6.3.7 Two other groups of order 2 element appears on the diagonal of the table for
have been mentioned previously in this chapter G5 while two different elements appear on the
or in Chapter 5: the multiplicative group {1, −1} diagonal of the table for G1 shows that G5 cannot
and the symmetric group 𝒮2 of permutations be isomorphic to G1 (or to G2 or G3).
of a two-​element set. (The elements of 𝒮2
might be labeled E, the identity permutation,
which fixes both elements of the set, and X, the
non-​identity permutation, which exchanges
them.) Verify that both of these groups are
isomorphic
   to ℤ2.

Exercise 6.3.8
(a) Construct the group table for (ℤ3, +​).
(b) Suppose G =​({e, x, y}, ⊗) is some other
group of order 3. Fill in the first row and
the first column of the table for G based FIGURE 6.3.9 Group tables for two Klein groups
on the behavior of the identity e. Then use of order 4

226 • E x plor i n g M u s i cal S paces

 The second table in Figure 6.3.9 shows an Exercise 6.3.11 By examining the group
abstract group with the same structure as G5. table you constructed in Exercise 6.2.3d for
This group structure is common and important the mod-​12 units group ℤ12×, show that ℤ12×
enough that it has a name: it is called the Klein is isomorphic to the Klein group 𝒱4. (You also
group (or the Klein four-​group), and commonly constructed a Cayley diagram for this group
denoted 𝒱4. The Klein group, represented by in Exercise 6.2.14a, so that diagram—​which
either of the tables in Figure 6.3.9, is not a cyclic required two generators, as the group is not
group like all of those in 6.3.1: because every cyclic—​may serve as a model Cayley diagram
non-​identity element of 𝒱4 combines with itself for  a Klein group.)
to produce the identity, no single element can
generate all of the others. The Klein group is, in Exercise 6.3.12 Let G be the transformation
fact, the smallest non-​cyclic group. In the case group G =​{T0, I0, R, RI0} acting on twelve-​tone
of the transformation group G5, the non-​cyclic rows, where I0 is inversion about C as usual,
structure means that repeated application of any and R is the retrograde operation. G acts on
single transformation (T6, I0, or I6) can never rec- the set of all possible twelve-​tone rows, but
reate the actions of all of the other transforma- you may find it helpful to consider the simply
tions in the group; this contrasts with the cyclic transitive action of G on a set consisting of four
group G4 =​{T0, T3, T6, T9} from Exercise 6.3.3, in forms of a particular row, S =​{P0, I0, R0, RI0}.
which repeated application of T3 will recapture Construct a group table for G. Is G a cyclic
all of the others. Like the cyclic groups, however, group
   or a Klein group?
the Klein group is commutative, as can be seen
from the symmetry of its table about the diag- Exercise 6.3.13 As noted above, every group
onal. In Section 6.4 we will see another way to of order 6 is isomorphic either to the cyclic
describe the structure of the Klein group, and we group 𝒞6 (or equivalently to the modular
will encounter other musical manifestations of group ℤ6) or to the symmetric group 𝒮3.
this group structure in exercises below and from One version of a group table for 𝒮3 was
time to time thereafter. presented in Figure 6.2.1b, and you
It turns out that every group of order 4 is iso- constructed a table for ℤ6 in Exercise 6.2.3a.
morphic to one of the two that we know—​either Listed below are three 6-​element subsets of
the cyclic group 𝒞4 or the Klein group 𝒱4—​so the dihedral group 𝒟12. For each of the listed
there are “only two groups of order 4.” The cyclic sets, verify that the set is a group; construct
group 𝒞5 is the only group of order 5, and in fact a group table or Cayley diagram if you find
if p is any prime number, the cyclic group 𝒞p it helpful; and determine whether it is
is the only group of order p. The smallest non- isomorphic to 𝒞6 or to 𝒮3.
commutative group is 𝒮3, the order-​6 symmet-
ric group on three elements discussed at some (a) {T0, T2, T4, T6, T8, T10}
length in Sections 6.1 and 6.2 above, where we (b) {T0, T4, T8, I0, I4, I8}
noted also that the symmetric group 𝒮3 is the (c)  {T0, T4, T8, I2, I6, I10}28
same group as the dihedral group 𝒟3. 𝒞6 and 𝒮3
are the only groups of order 6, and in fact if p is Exercise 6.3.14 Listed below are two more
any prime number ≥ 3, then the only groups of groups of order 6. One of them is isomorphic to
order 2p are 𝒞2p and 𝒟p. Group classification gets 𝒞6 and the other to 𝒮3. Which is which?
progressively more difficult as groups get larger,
particularly when the order of a group is a num- (a) The hexatonic triad group (PL group) from
ber with many factors (there is only one group of Exercise 6.2.1629
order 23, a prime number, but there are 15 dif- (b) The mod-​7 units group ℤ7× from
ferent groups of order 24).26 All finite groups can   Exercise 5.2.7b
be broken down into building blocks called sim-
ple groups of particular types; in what is widely Exercise 6.3.15 Recall the discussion of ski-​
considered one of the great achievements of hill lattices, metric states, and state classes
modern mathematics, exhaustive work by hun- from Section 3.4. Let S be the set of all
dreds of mathematicians over the last decades of possible state classes. Define transformations
the twentieth century, totaling many thousands f and g on S as follows: f reflects a state class
of pages dense with details, has completed the about a horizontal axis, and g reflects a state
classification of all the finite simple groups.27 class about a vertical axis.

Permutations, Isomorphisms, and Other Topics • 227

(a) Geometrically, what is accomplished by We have alluded several times, most recently
the composition f ● g? By g ● f? Do f and g in Sections 5.2 and 5.4, to a relationship between
commute? continuous pitch space cpitch and frequency
(b) Of the state classes represented by the space freq. These spaces have interval groups
metric states in Figure 3.4.12, which two (ℝ, +​) and (ℝ>0, ∙), respectively, and in fact these
are related by f? Which two are related by g? two groups are isomorphic. An example of
Which two are related by f ● g? an isomorphism f: (ℝ, +​) → (ℝ>0, ∙) is given by
(c) A state class may be described by a string f(x) =​2x. The function f is a bijection, its inverse
of 2s and 3s, reading the duration ratios is given by f −1(y) =​log2(y), and f is a homomor-
upward from the bottom of the graph. phism because 2x +​ y =​2x ∙ 2y. It then follows from
For example, the metric state common the interval-​ space homomorphism theorem
to parts (c) and (e) of Figure 3.4.12 is that cpitch and freq are isomorphic interval
(2, 2, 3). What is the effect of the action spaces, justifying our intuition that whether we
of f on a string of this kind? What is choose to measure distances between pitches by
the effect of g? Of f ● g? Interpret these subtraction in cpitch or as ratios in freq, we
relationships musically. are doing essentially the same thing. Moreover,
(d) By making a group table, show that f and g every subspace of cpitch is isomorphic to a
generate a Klein group of transformations corresponding subspace of freq. In particu-
  on S.30 lar, chromatic pitch space pitch, a subspace of
cpitch with the cyclic interval group ℤ, corre-
The concept of isomorphism is applicable to sponds under the above isomorphism to a sub-
infinite as well as finite groups. Certain exam- space of freq with interval group {…, 1 8 , 1 4 , 1 2 ,
ples are of particular importance. First, recall 1, 2, 4, 8, …}, an infinite cyclic multiplicative
from Section 5.3 the discussion of infinite cyclic group (generated by the element 2).31
groups of the form H =​ 〈a〉 =​{…, a−2, a−1, e, a, To conclude this section we revisit a bit of
a2, …}, consisting of all powers of an element a unfinished business from Section 5.6, where
along with their inverses. It is perhaps obvious one assertion in part (a) of Lewin’s theorem
that all infinite cyclic groups are isomorphic; an (Theorem 5.6.7) was left unproved. Given an
isomorphism maps a generator of one group interval space (S, G, int), for every i in G and
to a generator of the other. More explicitly, if every x in S, we defined Transi(x) as the unique
H is generated by a as above and Hʹ =​ 〈x〉 =​ y in S such that int(x, y) =​ i. We showed that Gʹ =​
{…, x−2, x−1, e, x, x2, …} is another infinite cyclic {Transi | i ∈ G} is a transformation group acting
group, generated by x, then the function f: H → on S, but left without proof the assertion that
Hʹ defined by f(ak) =​ xk for every integer k is auto- Gʹ is isomorphic to G. Now that we have formal-
matically an isomorphism. ized group isomorphism, we are in a position to
The integers form an infinite cyclic group verify that the function f: G → Gʹ defined by f(i) =​
(ℤ, +​) generated by the number 1, and the Transi is an isomorphism. This requires checking
even integers ℤeven also form an infinite cyclic that f is one-​to-​one, that f is onto, and that f is a
group, generated by 2. The preceding paragraph group homomorphism.
therefore implies that ℤeven ≅ ℤ, even though Saying that f is a homomorphism amounts
ℤeven is a proper subset of ℤ. We observed in to the equation Transi ● Transj =​Transi ⊕ j, which
Chapter 5 that ℤ is the interval group of chro- was already established in Section 5.6. That f is
matic pitch space pitch, and ℤeven is the inter- onto follows immediately from the definition of
val group of a subspace of pitch consisting of a Gʹ, which consists of the functions Transi and
whole-​tone scale. In Chapter 7 we will formally nothing else. Finally, to see that f is one-​to-​one,
define isomorphism of interval spaces, and we suppose i and j are intervals in G, and suppose
will establish a result called the interval-​space that f(i) =​ f(j), which means that Transi and
homomorphism theorem (Theorem 7.2.7), which Transj are the same function. For any x in S, if y
implies that if two interval spaces have isomor- =​ Transi(x), then y =​Transj(x) as well. From the
phic interval groups, then they are automatically definitions of Transi and Transj it follows that
isomorphic as interval spaces. The chromatic int(x, y) =​ i and also int(x, y) =​ j—​so in fact i =​ j.
and whole-​tone scales are therefore isomorphic This shows that f(i) and f(j) can be the same only
as interval spaces, even though one is a subset if i =​ j, which means that f is indeed one-​to-​one.
of the other. The proof of Lewin’s theorem is at last complete.

228 • E x plor i n g M u s i cal S paces

6.4 DIRECT-​P RODUCT of ℤ2 × ℤ2 are the ordered pairs (0, 0), (0, 1),
GROUPS (1, 0), and (1, 1). These elements are listed at the
top and at the left side of the table in the lexico-
Recall from Section 2.2 that if S and T are two graphic ordering, so that the elements whose first
sets, then S × T, the Cartesian product of S and component is 0 are listed before those whose
T, is the set of all ordered pairs (s, t) such that s first component is 1; other orderings are pos-
∈ S and t ∈ T. If G and H are groups, it is possible sible. The reader should be able to verify in the
to define a binary operation on the Cartesian table that ordered pairs are added component-
product G × H in such a way that G × H becomes wise, by the formula (x1, y1) +​(x2, y2) =​(x1 +​ x2,
a group. The group G × H, traditionally called the y1 +​ y2), where the addition in both components
direct product of G and H in this context, effec- is mod 2. (In direct products of additive groups,
tively combines the structure of G and H in a it is common to write +​for the group operation
single group. Many important groups in musical in the direct product as well as in the original
applications arise as direct products of smaller groups.)
groups. From Section 6.3 we know that as a group of
The group operation on the G × H is defined order 4, this group ℤ2 × ℤ2 must be isomorphic
componentwise. This means that if (g1, h1) and either to the cyclic group 𝒞4 or to the Klein
(g2, h2) are two elements of G × H (which means group 𝒱4. In fact, the pattern of entries in the
that g1 and g2 belong to G while h1 and h2 belong table of Figure 6.4.1a is the same as in the tables
to H), and if the group operations in G and H are for the Klein groups in Figure 6.3.9, not the cyclic
⊕ and ⊗ respectively, then the product of (g1, h1) groups of Figure 6.3.1: every non-​identity ele-
and (g2, h2) in G × H is defined to be (g1, h1) ∙ ment combines with itself to produce the iden-
(g2, h2) =​(g1 ⊕ g2, h1 ⊗ h2). Here the dot ∙ denotes tity. This is the alternative characterization of
the group operation in the direct-​product group. Klein groups promised in Section 6.3: the Klein
The first component g1 ⊕ g2 of this product cap- group 𝒱4 is isomorphic to the direct product
tures information about how elements combine ℤ2 × ℤ2, or equivalently 𝒞2 × 𝒞2.
in the group G, while the second component h1 The direct product ℤ2 × ℤ3, whose table
⊗ h2 captures the corresponding information appears in Figure 6.4.1b, may be slightly more
about H. The group axioms for G × H are direct confusing. In calculating the sums in this table
consequences of the same axioms for G and H; it is essential to remember that the sum in the
for example, if the identity elements in the two first component is always calculated by mod-​2
groups are eG and eH, then the identity element addition in ℤ2, while the sum in the second com-
in G × H is the ordered pair (eG, eH), and the ponent takes place in ℤ3 and so is calculated mod
inverse of an element (g, h) is (g, h)−1 =​(g−1, h−1), 3; this is why, for example, the sum (1, 1) +​(1, 1)
where g−1 and h−1 are calculated in the groups G is equal to (0, 2). It is interesting to notice, in
and H respectively. a direct-​product table like this one, how the
Figure 6.4.1 presents group tables for two structure of each of the two original groups is
simple direct products, starting with ℤ2 × ℤ2 in magnified in the product. If the table is divided
part (a). This is a group of order 4, so the table into four quadrants as shown by the dashed
is a 4-​by-​4 table of ordered pairs. The elements lines in the figure, the first component is 0 for all

FIGURE 6.4.1 Group tables for two direct-​product groups

Permutations, Isomorphisms, and Other Topics • 229

ordered pairs in the upper-​left and lower-​right the conformed neo-​Riemannian tonnetz a finite
quadrants and 1 for all ordered pairs in the other array suffices, suggesting the possibility of con-
two quadrants, so the 2-​by-​2 group table for ℤ2 is nections between tonnetzes and direct products
effectively reproduced at a larger scale in the first of small cyclic groups, to which we now turn our
components of these four large blocks. Within attention.
each block, meanwhile, the second components Figure 6.4.3 shows three Cayley diagrams for
form a 3-​by-​3 pattern identical to a group table the direct product ℤ4 × ℤ3. The first, drawn on
for ℤ3. a torus, arranges the twelve elements in recti-
The only groups of order 6 are the cyclic linear fashion reflecting the Cartesian-​product
group ℤ6 (or 𝒞6) and the symmetric group 𝒮3,
so ℤ2 × ℤ3 must be isomorphic to one of these.
We have seen that 𝒮3 is noncommutative, but
the symmetry of the table in Figure 6.4.1b
shows that the group ℤ2 × ℤ3 is commutative; in
fact, direct products of commutative groups are
always commutative. So ℤ2 × ℤ3 must be isomor-
phic to ℤ6—​that is, it must be a cyclic group. The
first part of the following exercise confirms that
this is the case.

Exercise 6.4.2
(a) Show that the element (1, 1) generates
ℤ2 × ℤ3.
(b) The pattern of entries in the table in Figure
6.4.1b probably differs from the pattern in
the table you constructed for ℤ6 in Exercise
6.2.3a, but we know that these patterns
depend on the order in which the elements
are listed. By listing the elements of ℤ2 × ℤ3
in an order corresponding to the order
in which they are generated by (1, 1),
construct an alternate group table for
ℤ2 × ℤ3 with the same pattern of entries as
in your earlier table for ℤ6.
(c) Let U be the subset {(0, 0), (1, 0)} of ℤ2 × ℤ3,
and let V =​{(0, 0), (0, 1), (0, 2)}. Verify
that U and V are subgroups of ℤ2 × ℤ3
with U ≅ ℤ2 and V ≅ ℤ3.
(d) Explain why ℤ2 × ℤ3 is isomorphic to
  ℤ3 × ℤ2.32

Having noted previously that ℤ2 × ℤ2 is not

isomorphic to ℤ4, one may be surprised to dis-
cover that ℤ2 × ℤ3 is isomorphic to ℤ6. It is the
two equal moduli in the product ℤ2 × ℤ2 that
make it impossible for one element to generate
this product. In general, it turns out that ℤm × ℤn
≅ ℤmn (or equivalently 𝒞m × 𝒞n ≅ 𝒞mn) if and only
if the integers m and n are coprime.
Recall from Figure 2.2.2 that Cartesian prod-
ucts may often be conveniently depicted in rect-
angular arrays. Tonnetzes are rectangular arrays,
and in fact Exercise 2.2.4a suggested labeling FIGURE 6.4.3 Cayley diagrams for ℤ4 × ℤ3: (a)
the vertices of an infinite tonnetz graph with on the generators (1, 0) and (0, 1); (b) and (c) on
ordered pairs of integers (elements of ℤ × ℤ). For the generator (1, 1)

230 • E x plor i n g M u s i cal S paces

structure, using the generators (1, 0) and (0, 1) (b) Because of the close correspondence
as the basis for the Cayley diagram. Addition of between intervals and transformations
(1, 0) to any element of the group corresponds described in Chapter 5, the same pictures
to a move one place to the right, and addition of may be considered transformation networks
(0, 1) is a move one place up. The now-​familiar on tonnetz space. The (1, 0) arrows in
convention by which opposite sides of a rectan- graph (a) correspond to the minor-​third
gular tile are conjoined enables a convenient rep- transposition T3 and the (0, 1) arrows to
resentation of the arrows that wrap around from the major-​third transposition T4. As a set,
the right side of the figure to the left and from tonnetz space is the same as ordinary
the top to the bottom. The rows of this figure are, pitch-​class space pc, and we know that
essentially, three copies of a Cayley diagram for the group 𝒯12 =​{T0, …, T11} acts in simply
ℤ4, while the columns are four copies of a Cayley transitive fashion on this set. What
diagram for ℤ3. transposition operator Tn is represented
The group ℤ4 × ℤ3, however, does not require by the (1, 1) arrows in graph (b)? What
two generators. The numbers 3 and 4 are coprime, is the musical interpretation of the cycle
so ℤ4 × ℤ3 must be a cyclic group isomorphic to comprising all the arrows in this graph?
ℤ12. As was the case above with ℤ2 × ℤ3, the ele- Sketch this cycle on a copy of the doughnut-​
ment (1, 1) is a generator of this group. Figure shaped tonnetz from Figure 1.4.4.
6.4.3b retains the toroidal organization of (a), (c) In tonnetz space, what is the ordered-​
but this graph uses the single generator (1, 1) as pair interval from E to F? From F to E?
the basis for the Cayley diagram. Addition of the What ordered pair corresponds to the
element (1, 1) to any element of the group corre- transposition operator T1? Give a general
sponds to a diagonal move up and to the right. The formula (in terms of n) for the ordered
wraparound connections are perhaps not as clear pair that corresponds to Tn. Can you find a
when the arrows are diagonal, but the arrows in general formula (in terms of a and b) for the
(b) trace a single cycle (0, 0) → (1, 1) → (2, 2) → transposition operator Tn that corresponds
(3, 0) → (0, 1) → ⋯ through all twelve vertices. to the ordered pair (a, b)?33
The graph in Figure 6.4.3c is the same as that (d) Four different elements of ℤ12 (1, 5, 7, and
in (b), but the vertices have been repositioned 11) are of order 12, which means that any
to make the cyclic structure more apparent. one of them generates the group. The same
Comparison with Figure 6.2.10a confirms that a must therefore be true of the isomorphic
group with this graph is isomorphic to 𝒞12. group ℤ4 × ℤ3. One of the four ordered pairs
Of course, the 3-​by-​4 module in Figure 6.4.3a capable of generating ℤ4 × ℤ3 is (1, 1). What
closely resembles the module of a conformed are the other three? To which transposition
tonnetz such as Figure 1.4.3, and in fact ℤ4 × operators Tn do they correspond?
ℤ3 may be regarded as the interval group for (e) The numbers 2 and 6 are not coprime, so
this familiar tonnetz space. Informally, the ton- the direct product ℤ6 × ℤ2 is not isomorphic
netz may be described as “the direct product of to the cyclic group ℤ12. Draw a Cayley
a diminished seventh chord and an augmented diagram for ℤ6 × ℤ2 similar to Figure 6.4.3a.
triad.” The following exercise explores several What happens if you try to construct a
aspects of this connection. tonnetz for mod-​12 pitch-​class space with
  this graph?
Exercise 6.4.4
(a) Redraw graphs (a) and (b) of Figure 6.4.3, Direct-​product groups underlie some other
replacing the ordered-​pair vertex labels with constructions we have studied in earlier chap-
pitch-​class names. Put C in place of (0, 0), ters as well. In Section 6.2 we remarked that
and follow the usual tonnetz arrangement, the double-​circle diagram from Figure 4.1.1a is
with minor thirds displayed horizontally effectively a Cayley diagram showing the action
and major thirds vertically. The arrows may on the space triad of a group G =​ 〈T1, P〉. The ele-
still be labeled with the ordered pairs (1, 0), ments of G include the transpositions T0, T1, …,
(0, 1), and (1, 1), now representing intervals T11, as well as the products of all of these with P,
in the modular two-​dimensional tonnetz namely T0P (=​ P), T1P, …, T11P. This is a complete
space. In this way tonnetz space may be list of the elements of G: because P commutes
viewed as an interval space tonn whose with T1 and therefore with every transposition
interval group is ℤ4 × ℤ3. (TnP =​ PTn) and is an involution (P2 =​ T0), any

Permutations, Isomorphisms, and Other Topics • 231

composition of any number of T1 and P opera- motion; in fact, the group of all possible move-
tions can be reduced to one of the 24 forms ments in the grid is the direct product of the
listed above. group of north-​south movements and the group
Of course, T1 by itself generates the cyclic of east-​west movements.
transposition group 𝒯12 =​{T0, …, T11}, and the
involution P generates a group of order 2, say Exercise 6.4.5
H =​{T0, P}. The group G therefore consists of all (a) Consider the function f: 𝒯12 × H → G above,
products (compositions) X ● Y of a transforma- where H =​{T0, P}, G =​ 〈T1, P〉, and all the
tion X in 𝒯12 with a transformation Y in H. In groups act on the space of triads. By the
this situation it is common to write G =​ 𝒯12H. definition of f, f(T5, P) =​ T5P and f(T9, T0) =​
This description makes G look very similar to the T9T0 =​ T9. According to the definition of
direct product 𝒯12 × H. The latter group consists the group operation in 𝒯12 × H, what is
not of compositions X ● Y but of ordered pairs (T5, P) ∙ (T9, T0)? In the group G, what is
(X, Y), specifically the ordered pairs (T0, T0), T5P ● T9? Verify that f(T5, P) ● f(T9, T0) =​
(T1, T0), …, (T11, T0) and (T0, P), (T1, P), …, (T11, P). f((T5, P) ∙ (T9, T0)), as is required in order for
The above enumeration of the elements of G f to be a homomorphism.34
shows that the function f: 𝒯12 × H → G defined (b) The group G, being isomorphic to a direct
by f(X, Y) =​ X ● Y is one-​to-​one and onto. In fact, product of two commutative groups (𝒯12
this function is an isomorphism. (The homomor- and H), is itself commutative. Is G a cyclic
phism property will be illustrated in Exercise group? What are the cyclic subgroups
6.4.5 below.) Therefore G ≅ 𝒯12 × H: the group   generated by the various elements of G?
G =​ 〈T1, P〉 is isomorphic to the direct product of
the group 𝒯12 =​ 〈T1〉 and the group H =​ 〈P〉. As a Exercise 6.4.6 Figure 4.1.1b showed the action
2-​element group, H is isomorphic to 𝒞2, so it fol- of the dihedral group 𝒯ℐ12 (or 𝒟12) on the space
lows that G ≅ 𝒞12 × 𝒞2. triad. In some ways 𝒯ℐ12 may appear similar
The fact that T1 and P commute is essential to the group G in the previous exercise: G is
to this conclusion, as Exercise 6.4.6 will demon- generated by T1 and the involution P, while
strate. A more general statement is true. Suppose 𝒯ℐ12 is generated by T1 and the involution I.
that H1 and H2 are two transformation groups Crucially, however, T1 commutes with P but not
acting on the same set S, suppose that H1 ∩ H2 with I. Suppose we try to reconstruct the above
=​ {E} (that is, H1 and H2 have no elements in reasoning in the case of 𝒯ℐ12. It is true that
common except the identity transformation), 𝒯ℐ12 =​{T0, …, T11, T0I, …, T11I}; it is true that
and finally suppose that every element of H1 com- I generates a subgroup Hʹ =​{T0, I} of order 2;
mutes with every element of H2. Let G be the set and it is true that the function f ʹ(X, Y) =​ X ● Y
H1H2, defined as {h1h2 | h1 ∈ H1, h2 ∈ H2}. (Here defines a one-​to-​one correspondence between
h1h2 means h1 ● h2. Because H1 and H2 are trans- the direct product 𝒯12 × Hʹ and 𝒯ℐ12. But this
formation groups on S, both h1 and h2 are func- function f ʹ is not a homomorphism. Demonstrate
tions defined on S, so they may be composed in this by showing that f ʹ(T5, I) ● f ʹ(T9, T0) ≠
the usual way.) Then G is a group isomorphic to f ʹ((T
  5, I) ∙ (T9, T0)).35
the direct product H1 × H2. It is not required that
H1 and H2 be commutative groups, only that In fact, the dihedral group 𝒟12 cannot be iso-
elements of each group always commute with morphic to the direct product of cyclic groups
elements of the other. Such groups are called 𝒞12 × 𝒞2, as the earlier group G was, because
pairwise commuting groups. 𝒟12 is noncommutative while direct products of
The pairwise commuting property effectively cyclic groups are always commutative. The differ-
ensures that the two groups act on S indepen- ent structures of these two groups are manifest
dently: transformations in H1 and transforma- in the directions of the arrows in Figure 4.1.1
tions in H2 may be mixed at will, without danger (which, as we have noted, are essentially Cayley
that the action of one group will somehow inter- diagrams for the groups).
fere with the action of the other. If this is the Another illustration of the way pairwise
case, then the group obtained by combining commuting groups generate direct products is
elements of both H1 and H2 is isomorphic to found in the group G5 =​{T0, T6, I0, I6}, whose
their direct product. For example, we observed table was presented previously in Figure 6.3.9.
in Section 2.3 that in a typical city street grid, This group is a subgroup of 𝒟12; its elements are
north-​south motion commutes with east-​west transformations on the space pc. The elements

232 • E x plor i n g M u s i cal S paces

T6 and I0 are involutions, so {T0, T6} and {T0, I0} Exercise 6.4.9 Figure 3.1.7 showed the triads
are subgroups of G5, both isomorphic to 𝒞2. in the hexatonic collection hex34 =​{0, 3, 4, 7, 8,
Also, T6 and I0 commute: transpositions do not 11} in the form of a cube depicting a Cartesian
generally commute with inversions, but it is product S =​ X × Y × Z, where X =​{11, 0}, Y =​{3,
true that T6I0 =​ I0T6 =​ I6. Therefore G5, the group 4}, and Z =​{7, 8} (these are subsets of pc). Let
generated by T6 and I0 together, is isomorphic ΔX be the non-​identity permutation on the set
to {T0, T6} × {T0, I0} and hence to 𝒞2 × 𝒞2, the X—​that is, the function that exchanges the
Klein group. two elements of X, so that ΔX(11) =​0 and
ΔX(0) =​11. Similarly, let ΔY and ΔZ denote
Exercise 6.4.7 Review the toroidal mode space the non-​identity permutations on Y and Z.
from Exercise 4.1.7. By joining horizontally Clearly ΔX, ΔY, and ΔZ are involutions, and
and vertically adjacent modes in this space each of them generates a group of order 2,
with arrows labeled (0, 1) and (1, 0), we may {E, ΔX}, {E, ΔY}, and {E, ΔZ}, all isomorphic to
view this mode space as an interval space 𝒞2. The functions ΔX, ΔY, and ΔZ each act on
whose interval group is ℤ12 × ℤ7. Because one coordinate of the ordered triples in S, and
12 and 7 are coprime, this direct-​product so they may be considered to act on S itself;
group is isomorphic to ℤ84. The cycle of for instance, ΔY(0, 4, 7) =​(0, 3, 7). Visualize
modes mentioned at the end of Exercise 4.1.7 the action of ΔX, ΔY, and ΔZ in Figure 3.1.7,
illustrates a generator of this cyclic group. and give a few examples to show that ΔX, ΔY,
What is the ordered-​pair representation of and ΔZ commute. Because they commute, they
this generator? What musical transformations generate a group G7 of order 8, isomorphic to
correspond
   to the intervals (0, 1) and (1, 0)? 𝒞2 × 𝒞2 × 𝒞2 and therefore to the group G6 of
row transformations discussed above. Consider
Direct products of more than two groups may how Figure 3.1.7 conveys the structure of a
also be considered. The group G5 =​{T0, T6, I0, I6} Cayley diagram for G7, and compare it with the
was defined as a transformation group on pitch-​ Cayley    diagram for G6 from Exercise 6.4.8.
class space, but these same transformations may
also act on twelve-​tone rows. Another involu- 6.5 GROUPS, EQUIVALENCE
tion defined on the space of twelve-​tone rows RELATIONS, AND
is the retrograde operation R, which generates
a group {T0, R}. R commutes with both T0 and
SYMMETRY
I6, and indeed with all transpositions and inver- Group structure lurks beneath the surface of
sions on twelve-​tone rows. It follows that T6, I0, many everyday musical phenomena, including
and R together generate a group of order 8, G6 =​ some of the equivalence relations that we have
〈T6, I0, R〉 =​{T0, T6, I0, I6, R, RT6, RI0, RI6}, isomor- already described in other ways, as well as famil-
phic to the triple direct product {T0, T6} × {T0, I0} × iar notions of symmetry. When two elements are
{T0, R} and therefore to 𝒞2 × 𝒞2 × 𝒞2. This group equivalent in some way, this often means that
is a subgroup of the serial group 𝒮𝑒𝑟12 =​ 〈T1, I0, R〉; they are related by the action of some transfor-
the serial group is a group of order 48 contain- mation group. To say that a thing is symmetri-
ing all the transpositions and inversions in 𝒟12 cal typically means that it is invariant under
together with R, and is isomorphic to 𝒯ℐ12 × the action of one or more transformations in a
{T0, R}, or to 𝒟12 × 𝒞2. group. Even when objects themselves are not
symmetrical, relationships between objects can
Exercise 6.4.8 Construct a Cayley diagram for often be described using transformation groups
G6 =​ 〈T6, I0, R〉. Which figure from Chapter 3 displaying symmetry in their structure. This sec-
does
   it closely resemble?36 tion explores the ramifications of these ideas.
We start with some general definitions.
Group-​ theoretic considerations involving Suppose G is a transformation group acting on
twelve-​ tone rows will be studied further in a set S. Two elements x and y of S are said to be
Sections 9.5 and 9.6, where the fact that the G-​equivalent, denoted x ~G y, if there is some trans-
retrograde operation R commutes with trans- formation f in G such that f(x) =​ y. It is not hard
positions and inversions will be seen as a special to verify that G equivalence is an equivalence
case of the more general principle that pitch-​ relation.37 The equivalence classes of the relation
class transformations always commute with order ~G are called the orbits of the action of G. More
transformations. specifically, if x ∈ S, the orbit of x is the set of all

Permutations, Isomorphisms, and Other Topics • 233

those elements of S to which x is mapped by some Exercise 6.5.1 Let K be the subgroup K =​ 〈T2〉 =​
transformation in G; this set could be written as {…, T−4, T−2, T0, T2, T4, …} of G. The action of
{y ∈ S | y =​ f(x) for some f ∈ G}, or more briefly as K partitions pitch into how many different
{f(x) | f ∈ G} or {y ∈ S | x ~G y}. This orbit will be orbits?
   What are these orbits?
denoted OrbG(x), or simply Orb(x) if there is no
likely confusion about which group G is involved. Enharmonic equivalence arises in a closely
Because of the reflexive property x ~G x, every x is analogous way from the space spc, the line of
automatically an element of its own orbit Orb(x). fifths from Figure 1.3.3. As in Section 2.6, let τn
Also, because orbits are equivalence classes, any denote upward transposition by n fifths, a move
two orbits of a group action are always either n places to the right in spc space, analogous to
identical or disjoint: if x ~G y, then OrbG(x) =​ Tn in chromatic pitch space. The spc transposi-
OrbG(y), but otherwise OrbG(x) ∩ OrbG(y) =​ ∅. tions τn form a group G* =​ 〈τ1〉 =​{…, τ−2, τ−1,
As an example, consider the transposition τ0, τ1, τ2, …}, isomorphic to the transposition
group G =​ 𝒯∞ =​{…, T−2, T−1, T0, T1, T2, …} acting group G =​ 𝒯∞ in pitch space considered above,
on chromatic pitch space pitch, as described in and the action of G* on spc, like the action of
Sections 5.1 and 5.2. The action of G on pitch G on pitch, is simply transitive. Moreover,
is simply transitive as defined in Section 5.6, the subgroup H* =​ 〈τ12〉 =​{…, τ−24, τ−12, τ0, τ12,
because for any two notes x and y in pitch, there τ24, …} plays a role in G* corresponding to H
is always exactly one Ti in G such that Ti(x) =​ y, above. Because every twelfth note on the line of
namely the Ti with i =​int(x, y). For any given x, fifths is enharmonically equivalent, H* equiva-
therefore, every y in pitch is some Ti(x), so every lence is nothing more than enharmonic equiva-
y is G-​equivalent to x. This group action, in other lence. The orbit of any note under the action of
words, has only one orbit: for every x, OrbG(x) is H* consists of all possible enharmonic respell-
the entire space pitch. The analogous statement ings of the given note; for example, OrbH*(C) =​
is true for every transitive group action: to say {…, E♭ ♭ ♭ ♭, D♭ ♭, C, B♯, A♯ ♯ ♯, …}.
that a transformation group G acts transitively
on a set S means that all elements of S belong to Exercise 6.5.2 Let J* be the subgroup J* =​ 〈τ7〉 =​
the same orbit. {…, τ−14, τ−7, τ0, τ7, τ14, …} of G*. The action of J*
More interesting orbits arise if instead of look- partitions spc into how many different orbits?
ing at the entire transposition group G =​ 𝒯∞ we What
   are these orbits?
consider subgroups of G. For example, let H be
the subgroup H =​ 〈T12〉 =​{…, T−24, T−12, T0, T12, For another familiar example, let G be the
T24, …} consisting of transposition operators dihedral group G =​ 𝒯ℐ12 =​{T0, T1, …, T11, I0, …,
whose indices are multiples of 12. Mathematically, I11} acting on the space pcset of all pitch-​class
𝒯∞ is isomorphic to the additive group of inte- sets. In this situation G equivalence is TI equiva-
gers ℤ, and H corresponds via this isomorphism lence. The action of G is not transitive on pcset,
to the subgroup 12ℤ =​{…, −24, −12, 0, 12, because many pairs of pc sets are not TI-​equiva-
24, …} (which, as a group, is again isomorphic to lent. The orbits of G are simply TI-​classes—​the
ℤ itself). Under the action of the subgroup H, the usual set classes. If X is a pitch-​class set, then
orbit of an element of x is not all of pitch, but OrbG(X) is the set class to which X belongs. For
the smaller set OrbH(x) =​{…, x − 24, x − 12, x, the cyclic subgroup 𝒯12 =​{T0, T1, …, T11} of trans-
x +​ 12, x +​24, …} consisting of all elements whose positions only, 𝒯12 equivalence is T equivalence.
distance from x is a multiple of 12. The action As defined in Section 2.4, T equivalence is a finer
of H partitions pitch into twelve distinct orbits, equivalence relation than TI equivalence, so
which are exactly the equivalence classes of the the action of 𝒯12 partitions pcset into a larger
mod-​ 12 congruence relation, enumerated in number of orbits than the action of 𝒯ℐ12. If X
Section 2.5. Musically, H consists of transposi- is a pc set with no symmetry, then the set class
tions by any whole number of octaves; the orbit Orb𝒯ℐ12(X) consists of 24 pc sets; under the
of a note x is the set of all notes octave-​equivalent action of 𝒯12 this 𝒯ℐ12 orbit splits into two 𝒯12
to x, so the orbits may be identified with pitch orbits of twelve sets each. If, on the other hand,
classes. The construction of the space pc as a X is inversionally symmetrical, then Orb𝒯ℐ12(X) =​
quotient space of pitch is familiar from Chapter Orb𝒯12(X). Altogether the 4,096 pitch-​ class
2, but now we see that the equivalence relation sets are partitioned into 352 orbits under
itself is determined by the action of a subgroup 𝒯12(T-​classes), and 224 orbits under 𝒯ℐ12 (TI-​
of the transposition group on pitch space. classes, or the usual set classes).38

234 • E x plor i n g M u s i cal S paces

Exercise 6.5.3 For each of the following automatically fixes every x in S (E(x) =​ x), stabi-
pitch-​class sets X, describe Orb𝒯ℐ12(X) and lizers (like orbits) are always nonempty sets: x is
Orb𝒯12(X). always an element of Orb(x), and E is always an
element of St(x). If the action of G on S is simply
(a) X =​the C-​major triad {0, 4, 7} transitive, then there can be only one f in G with
(b) X =​the C-​minor triad {0, 3, 7} the property that f(x) =​ x, and this must be f =​ E.
(c) X =​the C-​diminished triad {0, 3, 6} In the case of a simply transitive group action,
(d) X =​{0, 1, 4} then, orbits and stabilizers are not very interest-
(e)  X =​the Petrushka chord {0, 1, 4, 6, 7, 10} ing: there is only one orbit, and all stabilizers are
trivial, consisting of the identity transformation
Another familiar group action is that of the only. In other situations, however, St(x) may pro-
serial group 𝒮𝑒𝑟12 =​ 〈T1, I, R〉 on row, the space vide revealing information about the object x.
of all twelve-​tone rows. Two rows are 𝒮𝑒𝑟12-​ Consider again the action of the dihedral
equivalent if they are related by some combina- group 𝒯ℐ12 on pcset. If X =​{2, 4, 5, 7}, then
tion of transposition, inversion, and retrograde. Orb(X) is the set class 0235, consisting of X itself
In twelve-​tone theory such rows are considered and its transpositions by T1, …, T11. The inver-
“forms of the same row”; the orbits under the sions I0, I1, …, I11 are elements of 𝒯ℐ12 as well,
action of 𝒮𝑒𝑟12 are called row classes. Almost all but they do not contribute anything new to this
row classes contain exactly 48 rows, but a small orbit: because of the symmetry of X, its inver-
number of symmetrical rows give rise to row sions are the same as its transpositions. The
classes of cardinality 24 (recall Exercise 4.1.13, inversion I9 maps the specific set X to itself; the
and see Sections 9.5 and 9.6 for more details). other inversions map X to its transpositions. The
On a class of 48 rows, the action of 𝒮𝑒𝑟12 is sim- stabilizer St(X) is the set {T0, I9}; no transposition
ply transitive. Figure 4.1.12b, which shows this or inversion other than those two fixes the set X.
action on one row class, is essentially a Cayley Now let Y be the diminished seventh chord
diagram for 𝒮𝑒𝑟12, as we noted in Section 6.2. {1, 4, 7, 10}. In this case, Orb(Y), the set class
0369, is a small set, consisting only of Y and its
Exercise 6.5.4 two transpositions T1(Y) and T2(Y). The stabilizer
(a) The transposition group 𝒯12 =​{T0, T1, …, of Y, on the other hand, is large. In addition to the
T11} acts on row, as it is a subgroup of identity T0, Y is fixed by three other transposi-
𝒮𝑒𝑟12. Explain why every orbit of row tions and four inversions; in fact, St(Y) =​{T0, T3,
under the action of 𝒯12 contains exactly T6, T9, I2, I5, I8, I11}. The higher degree of symmetry
twelve rows, and use this fact to calculate of Y in comparison with X is reflected in the fact
how many orbits of this action (T-​classes of that its orbit is smaller and its stabilizer larger.
twelve-​tone rows) there are altogether. More can be said. Notice the following
(b) The dihedral group 𝒯ℐ12 =​{T0, T1, …, T11, I0, properties:
…, I11} is also a subgroup of the serial group
and therefore acts on row. Does every orbit • The inversion I9, like every inversion operator,
of this action contain exactly 24 rows? How is an involution, so the stabilizer St(X) =​{T0,
  many orbits are there altogether? I9} is a subgroup of 𝒯ℐ12. Using the formulas
from Theorem 5.5.6, one may check that the
For some of the pitch-​class sets X in Exercise stabilizer St(Y) =​{T0, T3, T6, T9, I2, I5, I8, I11} is
6.5.3, some transformation f in 𝒯ℐ12 has the also a subgroup of 𝒯ℐ12.
property that f(X) =​ X. In this situation we may • Orb(X), St(X), Orb(Y), and St(Y) are sets of
say that f fixes X, that f stabilizes X, or that X is cardinality 12, 2, 3, and 8 respectively. The
invariant under f. For example, the inversion products #Orb(X) ∙ #St(X) and #Orb(Y) ∙ #St(Y)
operator I6 fixes the diminished triad {0, 3, 6}. are both equal to 24, which is also the order of
Generally, if a transformation group G acts on the group 𝒯ℐ12.
a set S and if x ∈ S, the stabilizer of x (in G) is
defined to be the set {f ∈ G | f(x) =​ x}—​the set The above properties are very general.
of all those transformations in G that fix x. This Whenever any transformation group G acts on
stabilizer will be denoted StG(x) or simply St(x). any set S, the stabilizer St(x) of every element x
The stabilizer StG(x) is a subset of the group of S is always a subgroup of G, and the equation
G while the orbit OrbG(x) is a subset of the set #Orb(x) ∙ #St(x) =​#G always holds. This equation
S. Because the identity transformation E in G is called the orbit-​stabilizer theorem.39

Permutations, Isomorphisms, and Other Topics • 235

Exercise 6.5.5 List the elements in the orbit the group). Give examples of pc sets whose
and stabilizer of each of the following pitch-​ degrees of symmetry are 1, 2, 4, 6, 8, 12,
class sets under the action of 𝒯ℐ12, verify that and 24. (Although 3 is a divisor of 24, there
the stabilizer is a subgroup of 𝒯ℐ12, and verify are no pc sets whose degree of symmetry is
that the orbit-​stabilizer theorem holds: 3; see part (d) below.)
(b) Give an example of pc sets X and Y with
(a) the augmented triad {2, 6, 10} the property that X and Y have the same
(b) the French augmented-​sixth chord degree of symmetry relative to the action of
{0, 4, 6, 10} 𝒯ℐ12 but two different degrees of symmetry
(c) the C-​major triad {0, 4, 7} relative to the action of 𝒯12.
(d) the C-​major scale {0, 2, 4, 5, 7, 9, 11} (c) Give an example of pc sets X and Y with
(e) the whole-​tone scale {0, 2, 4, 6, 8, 10} the property that X and Y have the same
(f) the Petrushka chord {0, 1, 4, 6, 7, 10} degree of symmetry relative to the action of
(g)  the twelve-​tone aggregate40 𝒯12 but two different degrees of symmetry
relative to the action of 𝒯ℐ12.
Symmetry has been mentioned repeatedly in (d) Why are there no pc sets whose degree of
the above discussion; evidently, symmetry has symmetry relative to 𝒯ℐ12 is 3? One can
something to do with group actions. In pitch-​ construct such sets in pitch-​class spaces
class set theory it is widely recognized that sym- of cardinality other than 12, however. Can
metry is described by invariance. The pc set X =​ you find a subset of ℤ18 whose degree of
{2, 4, 5, 7}, for example, is inversionally symmet- symmetry is 3 relative to 𝒯ℐ18, the dihedral
rical because it is invariant under the inversion   group acting on ℤ18?42
operator I9; the transformation I9 may be called a
symmetry of X. A highly symmetrical set such as a Set-​class tables, as in Straus 2016 (378–​81)
diminished seventh chord has many symmetries; and other texts, sometimes list degrees of “T
that is, it is invariant under (fixed by) several dif- symmetry” and “I symmetry” for each set class.
ferent transposition and inversion operators, or, The T symmetry of a set is the number of trans-
to put it differently, it has a large stabilizer. Of position operators that fix the set, and the I
course, every pc set is fixed by the identity trans- symmetry is the number of inversion opera-
formation T0; a set that is fixed by no other Tn or tors. A set’s degree of symmetry relative to 𝒯12
In besides T0 is said to be asymmetrical. is therefore the same as its T symmetry, and the
This conception of symmetry may be extended sum of the T and I symmetries is the degree of
to actions of general transformation groups. If G symmetry relative to 𝒯ℐ12. (The I symmetry by
acts on S and x ∈ S, elements of StG(x) are called itself is not the degree of symmetry relative to
symmetries of x, and the cardinality of StG(x) is any group.) The possibility of defining the degree
called the degree of symmetry of x. A highly sym- of symmetry for a set class depends on the fact
metrical element is one with a large stabilizer—​ that all sets in the class have the same degree of
and therefore, by the orbit-​stabilizer theorem, a symmetry. That is, if pc sets X and Y are related
small orbit.41 If the degree of symmetry of x is 1, by some Tn or In, then X and Y are fixed by the
then x is asymmetrical, as it is fixed only by the same number of transposition operators and by
identity transformation. The group G itself may the same number of inversion operators (but not
be called a symmetry group (or group of symme- necessarily by precisely the same operators). The
tries) on S. A possible point of confusion is that following exercise explores a few examples of
the above definitions depend on the transforma- this property.
tion group, and more than one symmetry group
may act on the same set: an inversionally sym- Exercise 6.5.7 In parts (a)–​(d) below, fill
metrical pitch-​class set, for instance, will have a in each blank with the appropriate Tn or In
higher degree of symmetry relative to the action operator (disregarding T0). Also give examples
of the group 𝒯ℐ12 than it has relative to 𝒯12. of sets X and Y satisfying the given conditions
in each case, and use them to verify that your
Exercise 6.5.6 answers are correct.
(a) It is a consequence of the orbit-​stabilizer
theorem that the degree of symmetry of (a) If X is fixed by T4 (and therefore also by
every pitch-​class set relative to the action T8) and Y =​ T5(X), then Y is fixed by the
of 𝒯ℐ12 must be a divisor of 24 (the order of operators _​_​_​_​_​ and _​__​ ​_​_​ .

236 • E x plor i n g M u s i cal S paces

(b) If X is fixed by T6 and Y =​ I1(X), then Y is 9, or 11, while for other values of i neither Ti(1)
fixed by _​_​_​_​_​ . nor Ti(7) belongs to X.
(c) If X is fixed by I10 and Y =​ T3(X), then Y is
fixed by _​_​_​_​_​ . (a) Show in general that if x and y are
(d) If X is fixed by I5 and Y =​ I4(X), then Y is intervallically indistinguishable with
fixed by _​_​_​_​_​ . respect to X, then X is invariant under Tn,
(e) If a set is inversionally symmetrical (that is, where n =​ y − x. (In the example just given,
if its I symmetry is greater than 0), explain 7 − 1 =​6, and the Petrushka chord is
why its T symmetry and I symmetry must invariant under T6.)
be the same, so that its degree of symmetry (b) Show, conversely, that if a set X is invariant
relative to 𝒯ℐ12 must be exactly twice its under some Tn with n ≠ 0, and if x and y are
  degree of symmetry relative to 𝒯12.43 pitch classes such that Tn(x) =​ y, then x and
y are intervallically indistinguishable with
Every set is invariant under T0, so to be con- respect to X. In other words, intervallically
sidered transpositionally symmetrical, a set must indistinguishable elements exist only in the
have T symmetry greater than or equal to 2. It   case of transpositionally symmetrical sets.
should be obvious that there are no transposi-
tionally symmetrical sets of cardinality 1 or 11; Exercise 6.5.10 If X is a pitch-​class set, a mode
the next exercise shows that there are none of of X may be defined as the string of intervals
cardinality 5 or 7 either (the other mod-​12 inte- between successive notes of the set, starting
gers coprime to 12). This property has important with any desired note and wrapping around
consequences in the study of scales, and we will to the same note at the end. Intervallically
cite it several times in Part Four. The exercises indistinguishable starting notes, as defined in
that follow examine a few other consequences of the previous exercise, will give rise to the same
transpositional symmetry. mode. For example, an octatonic collection has
only two modes, given by the strings 12121212
Exercise 6.5.8 Suppose X is a pc set of and 21212121; notes of the collection related
cardinality c with 1 ≤ c ≤ 11, and suppose that X by interval class 3 or 6 are intervallically
is invariant under some Tn other than T0. Let d indistinguishable. A diatonic collection has
be the greatest common divisor of n and 12, and seven different modes, all of which may be
let k =​12 ∕ d. obtained by rotating the pattern 2212221
characteristic of a major scale. A mode of X is
(a) Show that under these assumptions n an ordered n-​tuple of positive integers, where
cannot be coprime to 12. Conclude that n =​#X and the integers in the string sum to 12.
2 ≤ d ≤ 6 and 2 ≤ k ≤ 6.
(b) Recall from Exercise 6.1.3 that every cycle (a) There is a simple equation, valid for every
in the cycle structure of Tn is of length nonempty pitch-​class set X, by which the
k. Conclude that c is divisible by k, and number of modes of X may be calculated
  therefore c cannot be coprime to 12. from the cardinality of X and the T
symmetry of X. What is this equation?44
Exercise 6.5.9 Suppose X is a pitch-​class set (b) Verify that your equation is valid for sets of
and x and y are pitch classes (which may or each of the following types: a whole-​tone
may not belong to X). We say that x and y are scale, a hexatonic collection, an enneatonic
intervallically distinguishable with respect to collection, a diminished seventh chord,
X if there is some interval i in ℤ12 such that a dominant seventh chord, a French
one of the two notes Ti(x) and Ti(y) belongs augmented-​sixth chord, and a major triad.
to X but the other does not. For example, if What word other than mode might more
X is the Petrushka chord {0, 1, 4, 6, 7, 10}, traditionally correspond to the modes of a
then the elements 1 and 6 are intervallically   triad or seventh chord?
distinguishable with respect to X, because
T1(6) ∈ X while T1(1) ∉ X. The elements 1 and Another important distinction is to be made
7, however, are intervallically indistinguishable between symmetry in pc space and symmetry in
with respect to X, because they are in identical pitch space. The former is the type of symmetry
intervallic relationships to the elements of X: most often studied in pitch-​class set theory; the
Ti(1) and Ti(7) both belong to X if i is 0, 3, 5, 6, latter is more restrictive. A finite set in pitch

Permutations, Isomorphisms, and Other Topics • 237

space cannot be invariant under any transposi- Exercise 6.5.12
tion other than T0. (An infinite set can be: for (a) Listed below are some significant pitch
instance, the infinite whole-​tone scale {…, −4, sets in the opening measures of “From
−2, 0, 2, 4, …} is invariant under Tn for every the Island of Bali,” No. 109 from volume
even n.) A finite set in pitch space can be invari- IV of Bartók’s Mikrokosmos. For each set,
ant under one inversion operator, but never determine the axis (or axes) of symmetry in
more than one; it can have at most one axis of pitch space and in pc space.
inversion, and at most one pitch can lie on that X =​{G♯3, A3, D4, E♭4} (left hand in mm. 1–​11)
axis. In pitch-​class space, it is possible for two Y =​ T3(X) =​{B3, C4, F4, G♭4} (right hand in
different notes to lie on an inversional axis. An mm. 1–​4)
inversion operator fixes a note that falls on its Z =​ T12(Y) =​{B4, C5, F5, G♭5} (right hand in
axis, and matches other notes in pairs. Either mm. 5–​11)
in pitch or pitch-​class space, therefore, if a set X ∪ Y (both hands together in mm. 1–​4)
of odd cardinality is inversionally symmetrical, X ∪ Z (both hands together in mm. 5–​11)
there must be exactly one note of the set on the Y ∪ Z (right hand in mm. 1–​11)
axis of inversion. (b) Study the score of “From the Island of Bali.”
If a set in pitch space is symmetrical, then Comment on the importance of the various
the corresponding set in pitch-​ class space is axes of symmetry you identified in (a) as
automatically symmetrical as well; in fact it is the piece progresses, and identify any other
invariant under the same In, reducing n mod 12   axes (in pitch or pc space) that appear.
if necessary. For example, the pitch set {E4, G4,
A4, C5} =​{4, 7, 9, 12} is invariant under the pitch Exercise 6.5.13
inversion I16, and in pitch space this set reduces (a) True or false: If pitch-​class sets X and Y
to {C, E, G, A} =​{0, 4, 7, 9}, which is invariant are related by In and pitch class x is an
under I4. The converse does not always hold: element of both X and Y, then In(x) also
for instance, the asymmetrical pitch set {C4, E4, belongs to both X and Y. (If the statement
G♯5} =​{0, 4, 20} reduces to {C, E, G♯} =​{0, 4, 8}, a is true, explain why; if it is false, give a
symmetrical augmented triad in pc space. The pc counterexample.)
set {0, 4, 8} does have symmetrical representa- (b) True or false: If two pitch-​class sets related
tions in pitch space, but whether such a repre- by In have exactly one note in common,
sentation exists for every symmetrical pc set is then that note must map to itself under In.
a subtler question than it might appear, as the (c) Is it possible for two pitch-​class sets related
following exercise demonstrates. by I5 to have exactly one note in common?
(d) Is it possible for two pitch-​class sets
Exercise 6.5.11 Listed below are eight related by I5 to have exactly three notes in
inversionally symmetrical sets in pc space.   common?
Determine the axis (or axes) of symmetry for
each of them. If possible, construct symmetrical Exercise 6.5.14 This exercise explores
representations for each set in pitch space. symmetry in spelled pitch-​class space—​the
What are the possible axes of symmetry in line of fifths.
pitch space? In one case you will find that even
though the pc set is symmetrical, there is no (a) Consider the following four spc sets,
possible symmetrical representation as a pitch where the numbers are spc numbers as in
set of the same cardinality. Why is this? Can Figure 1.3.3: {−2, 1, 4}, {−2, 1, 3, 6}, {−1,
you find a symmetrical pitch set in this case if 2, 3, 6}, {−5, −1, 2, 4}. Three of these sets
octave doublings are permitted? are symmetrical in spc space. Identify
these sets; for each of them, describe
(a) {C, D} (=​{0, 2}) a transformation that leaves the set
(b) {C, F♯} invariant, and locate the axis of symmetry
(c) {C, D, E} (which, as in pitch space, may be either an
(d) {C, C♯, D} integer or a half-​integer).
(e) {C, D, G} (b) For the four sets in (a), replace the spc
(f) {C, C♯, D, G} numbers with the equivalent note names
(g) {C, D, F♯, G♯} (for example, {−2, 1, 4} =​{C, A, F♯}), and
(h)  {C, D♭, F, A♭} reinterpret the spc sets as pc sets. Which

238 • E x plor i n g M u s i cal S paces

 of them are now symmetrical? What can Clearly M3 (or I or f) is a bijective mapping from
you conclude about a logical relationship ℤ4 to itself. Also, M3 satisfies the requirement
between symmetry in spc space and for a group homomorphism, because M3(x +​ y) =​
symmetry in pc space? How are axes of 3(x +​ y) =​3x +​3y =​ M3(x) +​ M3(y). Therefore M3
  symmetry in the two spaces related? is an isomorphism from ℤ4 to ℤ4 itself. An iso-
morphism from a group G to G itself is called
Exercise 6.5.15 Twelve-​tone rows can be an automorphism of G. Loosely speaking, auto-
symmetrical, but the possibilities are limited. morphisms of a group may be considered sym-
Within the confines of the serial group 𝒮𝑒𝑟12, no metries of the group itself. Our observation
twelve-​tone row can have a degree of symmetry about the interchangeability of the elements 1
greater than 2. Review the sample symmetrical and 3 in the group ℤ4 is an observation about
rows you constructed in Exercise 4.1.13, verify the symmetry of that group, made explicit by
that each has degree of symmetry 2, and the fact that multiplication by 3 is an automor-
determine which non-​identity serial operation phism of ℤ4.
fixes
   each row.45 For any group G, the identity function EG
on G is trivially an automorphism of G. If f:
The connections between groups and sym- G → G is an automorphism, then so is f −1; and if
metry are too varied and profound to explore f and g are automorphisms of G, then so is f ● g.
here in detail; see Carter 2009 for many excel- Consequently all automorphisms of G together
lent illustrations. One important aspect of form a new group, called the automorphism group
the symmetry of a group itself will be of some of G and denoted Aut(G).
interest to us, however. Consider the cyclic In the case of ℤ4, the only automorphisms
group ℤ4. The Cayley diagram in Figure 6.2.5a are the identity E and M3, so Aut(ℤ4) is a group
and the alternate diagram in Exercise 6.2.6 of order 2 isomorphic to ℤ2. The identity on ℤ4
perhaps suggest that the elements 1 and 3 is the same thing as the multiplication opera-
in this group are somehow interchangeable. tor M1, so Aut(ℤ4) =​{M1, M3}. This observation
Either 1 or 3 by itself generates the group; only is suggestive because 1 and 3 are the units of ℤ4
the direction in which the cycle is traversed is (the elements coprime to 4); thus Aut(ℤ4) may
affected by the choice. No other exchanges of also be identified with the units group ℤ4×. This
elements seem possible without violating the property holds for all cyclic groups: the automor-
group structure, however: the identity 0 is of phisms of ℤn are precisely the mod-​n multiplica-
course unique, and 2, as the only non-​iden- tion operators Ma for which a is coprime to n,
tity element that does not generate the entire and therefore the automorphism group Aut(ℤn)
group, plays a structural role different from is isomorphic to ℤn×. (Recall that ℤn is an addi-
that of any other element. tive group while ℤn× is multiplicative.)
This interchangeability of 1 and 3 can also be In particular, Aut(ℤ12) is the units group ℤ12×,
seen in the group table for ℤ4 from Figure 6.2.1a. which by Exercise 6.3.11 is isomorphic to the
If every 1 in this table is replaced by a 3 and vice Klein group. There are 12! bijective functions
versa (including those in the row and column from the set ℤ12 to itself, but only four of them
headings), the resulting table is still an accurate are automorphisms of the group: the multipli-
addition table for ℤ4. No other rearrangements cation operators M1 (=​ T0), M5, M7, and M11 (=​
of the table, involving either the number 0 or the I), all of which we have now encountered many
number 2, have this property. times. An automorphism of the additive group
The rearrangement we are considering here ℤ12 must map the generator 1 to one of the
is a bijection f: ℤ4 → ℤ4 given by the equations four generators 1, 5, 7, or 11; once that choice
f(0) =​0, f(1) =​3, f(2) =​2, f(3) =​1. There are some is made, the automorphism is completely deter-
other ways to describe this function. In the ter- mined by the homomorphism property. The
minology from Section 6.1, f is a permutation same reasoning gives the answer to a question
with the cycle representation (1, 3). Using sim- posed in Section 6.3: there are just four isomor-
ple modular arithmetic we may note that for phisms from ℤ12 to 𝒯12 =​{T0, T1, …, T11}, deter-
every x in ℤ4, f(x) =​−x (mod 4), or equivalently mined by mapping the generator 1 of ℤ12 to one
f(x) =​3x (mod 4). We may therefore identify f of the four generators T1, T5, T7, or T11 of 𝒯12.
with the inversion operator I: ℤ4 → ℤ4 or with Recall from Exercise 6.1.9b that the transforma-
the mod-​4 multiplication operator M3: ℤ4 → ℤ4. tion M7 relates the chromatic and circle-​of-​fifths

Permutations, Isomorphisms, and Other Topics • 239

orderings of pitch-​class space; the fact that this 6.6 QUOTIENT GROUPS;
transformation is an automorphism of ℤ12 CONSIDERATIONS WITH
underlies the structural similarity between the NONCOMMUTATIVE GROUPS
two pictures.
The group ℤ7, though smaller than ℤ12, has a We have made many observations about the
larger automorphism group, for ℤ7× is a group of relationship between chromatic pitch space
order 6 (see Exercise 5.2.7b). As measured by the pitch and chromatic pitch-​class space pc, and
count of automorphisms, ℤ7 is a more symmetri- also about the relationship between their inter-
cal group than ℤ12, even though as a transforma- val groups ℤ and ℤ12. In Chapter 1 we formed
tion group ℤ12 (or 𝒯12) acts on a larger space withpc by wrapping the line of pitch into a circle.
a greater variety of symmetries. In Section 2.4 we showed more precisely how to
obtain pc as a quotient set of pitch by apply-
Exercise 6.5.16 Consider generic pitch-​class ing an equivalence relation (octave equiva-
space gpc in the stepwise configuration of lence), and in Section 6.5 we observed that this
Figure 1.2.8, as well as generic fifths space equivalence relation is defined by the action
gfifth of 1.3.8 and generic thirds space of a transformation group H =​ 〈T12〉 =​{…, T−24,
gthird of 1.3.12. T−12, T0, T12, T24, …} on pitch. We also noted
that under the isomorphism of 𝒞∞ with ℤ, the
(a) Determine which automorphism of ℤ7 subgroup H of 𝒞∞ corresponds to the subgroup
(a mod-​7 multiplication operator Ma) maps 12ℤ =​ 〈12〉 =​{…, −24, −12, 0, 12, 24, …} of ℤ. In
the pitch-​class numbers in gpc to the Section 2.5, meanwhile, we showed how to con-
numbers in the corresponding locations struct ℤ12 as a quotient set of ℤ by the relation
in gfifth. Is this automorphism an of congruence mod 12. It will probably come as
involution, like M7 in ℤ12? no surprise that congruence mod 12 may also
(b) What automorphism of ℤ7 maps the be defined by a group action, specifically by an
pitch-​class numbers in gpc to the numbers action of the subgroup 12ℤ on the entire group
in the corresponding locations in gthird? ℤ. From this observation it follows that ℤ12 is
(c) What automorphism of ℤ7 maps the not merely a quotient set of ℤ but a quotient
pitch-​class numbers in gfifth to the group, a concept that we now make precise by
numbers in the corresponding locations in dissecting the construction of ℤ12 even a bit
  gthird? further than before.
By Cayley’s theorem (from Section 6.2), every
Another noteworthy automorphism group is group acts on itself by multiplication within the
that of the Klein group 𝒱4. In the group table for group: the transformation corresponding to an
𝒱4 in Figure 6.3.9, all three of the non-​identity element a in G is Multa(x) =​ xa, defined for all x
elements x, y, and z play symmetrical roles in the in G. Any subgroup H of G also acts on G in this
structure of this group: all are involutions, and way. In the present situation, the group G =​ ℤ
the product of any two of these elements is the and the subgroup H =​12ℤ are additive groups,
third. Consequently these three elements may so they act on ℤ by addition; changing the name
be rearranged in any way at all without changing of the transformation accordingly, the function
the content of the table. In other words, all pos- Adda is defined by Adda(x) =​ x +​ a for every x in
sible permutations of the set {x, y, z} are auto- ℤ. Elements x and y of ℤ are related by Adda for
morphisms of 𝒱4, and Aut(𝒱4) is the group of some a in 12ℤ if and only if x ≡ y (mod 12). The
permutations of a three-​element set—​the sym- relation of congruence mod 12 is therefore the
metric group 𝒮3. As we know, this is a group of same as the equivalence relation defined by the
order 6; the Klein group 𝒱4 has a higher degree group action of 12ℤ on G (in the notation from
of symmetry than the cyclic group 𝒞4, and the Section 6.5 this relation would be denoted ~12ℤ),
automorphism group of a Klein group is larger and the equivalence classes—​the elements of
than the Klein group itself. ℤ12—​are the orbits of this action. For example,
Automorphisms and automorphism groups Orb12ℤ(3) is the set {…, −21, −9, 3, 15, 27, …},
will come into play several times in the follow- the set denoted 3 in Section 2.5. Each such orbit
ing chapters, in our study of isomorphism of looks much like the subgroup 12ℤ, but shifted
interval spaces in Section 7.2, of transformation (“transposed”) by a constant value. That is, the
graphs and networks in Section 9.3, and of well-​ equivalence class 3 consists of all integers a
formed scales in Section 13.2. +​3 (or 3 +​a), where a ∈ 12ℤ. This set may be

240 • E x plor i n g M u s i cal S paces

denoted by the shorthand 12ℤ +​3 (or 3 +​12ℤ); of course this is precisely the set that we already
it is called a coset of the subgroup 12ℤ. know as ℤ12, the integers mod 12. The notation
In the set ℤ, the cosets of 12ℤ, orbits under ℤ ∕ 12ℤ is commonly pronounced “ℤ mod 12ℤ,”
the action of 12ℤ, and congruence classes mod showing its derivation from modular arithmetic;
12 are precisely the same sets. Every coset has the cosets of 12ℤ effectively define “modules”
multiple names: the coset 12ℤ +​3 could also that repeat over and over throughout the larger
be labeled 12ℤ +​15 or 12ℤ +​(−9), for example. group ℤ. Here ℤ12 =​ ℤ ∕ 12ℤ is not merely a quo-
There are twelve of these cosets, a situation tient set (the general term for the set of equiva-
summarized by saying that 12ℤ is a subgroup lence classes of an equivalence relation), but
of index 12 in the group ℤ. (Somewhat loosely, something more: it is a quotient group, specifically
this means that 12ℤ contains one-​twelfth of the the quotient of the group ℤ by the subgroup 12ℤ.
elements of ℤ—​but in this case of course both Further examination of the construction of this
ℤ and 12ℤ are infinite groups.) Figure 6.6.1 is quotient group, and especially the binary opera-
a Cayley diagram for ℤ in which the elements tion in this group, will enable us to understand
of each coset have been brought together. The the concept of quotient groups more generally.
arrows form a single spiral, not a series of con- In Section 2.5 we formalized addition in ℤ12 by
centric circles. This figure could be produced by the definition m + n = (m + n ). Translating this
bending the linear Cayley diagram from Figure definition into the language of cosets of 12ℤ, we
6.2.5c into a spiral; it may also be compared may define the sum of cosets (12ℤ +​ m) +​(12ℤ +​ n)
with Figure 1.2.1, a similar spiral representa- to be the coset 12ℤ +​(m +​ n). Defined in this
tion of pitch space. way, addition is a binary operation on ℤ ∕ 12ℤ.
The collection of the twelve cosets of the We observed in Chapter 2 that this addition is
subgroup 12ℤ within the group ℤ is frequently well-​defined; in terms of cosets, this observation
denoted ℤ ∕ 12ℤ. This notation ℤ ∕ 12ℤ (or more means that choosing different names for the
generally G ∕ H) is standard in group theory, but cosets 12ℤ +​ m and 12ℤ +​ n cannot result in a

FIGURE 6.6.1 A Cayley diagram for (ℤ, +​) on the generator 1, showing the cosets in ℤ/​12ℤ

Permutations, Isomorphisms, and Other Topics • 241

different sum. That is, if 12ℤ +​ m1 =​12ℤ +​ m2 quotient group G ∕ H, collapsing H to the iden-
and 12ℤ +​ n1 =​12ℤ +​ n2, then 12ℤ +​(m1 +​ n1) is tity? The general answer to this question is no,
the same coset as 12ℤ +​(m2 +​ n2). A special case for reasons explored below. As long as G is a
of this property is clearly visible in the graph of commutative group, however, a quotient group
Figure 6.6.1: if two numbers m1 and m2 lie in the G ∕ H can always be formed, exactly following the
same coset, then the arrows directed away from model of ℤ ∕ 12ℤ as described above.
m1 and m2 always point to the same coset. That
is, the coset containing m1 +​1 also contains m2 +​ Exercise 6.6.2 Review the above discussion
1. If two of these +​1 arrows with origins in the demonstrating that congruence mod 12 is a
same coset had their destinations in two differ- group congruence. Locate a specific step where
ent cosets, a coherent definition of addition in this explanation relied on the commutativity of
the quotient group would be impossible.   group ℤ.
the
Another general definition is useful here.
Suppose (G, ⊗) is a group and ~ is an equivalence To be precise, suppose (G, ⊕) is a commuta-
relation on G. The relation ~ is said to be a group tive group and H is a subgroup of G. If x ∈ G,
congruence if, whenever g1, g2, h1, and h2 are ele- write H ⊕ x for the set {h ⊕ x | h ∈ H}; this is
ments of G such that g1 ~ g2 and h1 ~ h2, then it a coset of the subgroup H in the group G. This
is also true that g1 ⊗ h1 ~ g2 ⊗ h2. This definition coset is also OrbH(x), where the subgroup H acts
means that the equivalence relation is compat- on the group G by “addition” (⊕). The number
ible with the group operation: when group ele- of different cosets is the index of H in G; if G is a
ments are replaced by other elements to which finite group, this index is simply #G ∕ #H, the ratio
they are equivalent (related by ~), the product of the cardinality of the group G to that of the
of a group operation may change, but the new subgroup H. The set of all the cosets is denoted
product will be equivalent to the original prod- G ∕ H (“G mod H”). Two elements x and y of G
uct in the same way. In these terms, the obser- are in the same coset if x ~H y, which means that
vation in the preceding paragraph says precisely y =​ h ⊕ x for some h in H; this relation ~H is a
that the relation ~12ℤ (congruence mod 12) on group congruence on G, which may be called
the group ℤ is a group congruence. congruence mod H. (As above, this conclusion
In fact, this is so precisely because 12ℤ is a depends on the commutativity of G.) We can
subgroup of ℤ. The hypotheses 12ℤ +​ m1 =​12ℤ +​ therefore define a binary operation on the cosets
m2 and 12ℤ +​ n1 =​12ℤ +​ n2 imply that m1 ≡ m2 in G ∕ H by the equation (H ⊕ x) ⊕ (H ⊕ y) =​ H ⊕
and n1 ≡ n2 (mod 12), so the differences m1 − m2 (x ⊕ y), and under this operation, G ∕ H becomes
and n1 − n2 belong to 12ℤ. Because the subgroup a group, the quotient group of G by the subgroup
12ℤ is closed under addition, it follows that the H. The identity element of G ∕ H is the coset H
sum of these two differences, (m1 − m2) +​(n1 − n2), itself (or H ⊕ 0, where 0 denotes the identity of
belongs to 12ℤ as well. But this sum is the same G), and the inverse of H ⊕ x in G ∕ H is H ⊕ (⊖x),
as (m1 +​ n1) − (m2 +​ n2), so it follows that m1 +​ n1 where ⊖x denotes the inverse of x.
≡ m2 +​ n2 (mod 12), and therefore it is indeed For a second example of a quotient group,
true that the cosets 12ℤ +​(m1 +​ n1) and 12ℤ +​ let G be the cyclic transposition group 𝒯12 =​{T0,
(m2 +​ n2) are the same. So it is the status of 12ℤ T1, …, T11} acting on pc, and let H be the sub-
as a subgroup of ℤ that allows us to define addi- group {T0, T3, T6, T9}. Of course G is commu-
tion on the quotient set, turning it into a quo- tative, and we observed in Section 6.1 that H
tient group. The identity element of this quotient is a cyclic subgroup of G. The index of H in G is
group is the coset 12ℤ +​0, the coset containing 12 ∕ 4 =​3. One of the three cosets is H itself; the
the identity element 0 of ℤ. As a subset of ℤ, this others may be denoted HT1 and HT2, obtained by
coset is equal to the group 12ℤ itself. Because “multiplying” H on the right by the elements T1
12ℤ is the only coset containing 0, it is therefore and T2 of G. “Multiplication” here is really com-
also the only coset that is itself a group; in the position of functions; the cosets could be writ-
quotient group ℤ ∕ 12ℤ, the subgroup 12ℤ effec- ten H ● T1 and H ● T2, but we omit the symbol ●
tively collapses to the identity. The inverse of for simplicity. Other representatives could have
the element 12ℤ +​ n in ℤ ∕ 12ℤ is 12ℤ +​(−n), the been chosen for these cosets, but T1 and T2 are
coset containing the integer −n; this coset could numerically the simplest.
also be written 12ℤ − n. Because the elements of H are T0, T3, T6, and
Is a similar construction possible for any sub- T9, the elements of the coset HT1 are T0T1 =​ T1,
group H of any group G? Can we always form a T3T1 =​ T4, T6T1 =​ T7, and T9T1 =​ T10, and similarly

242 • E x plor i n g M u s i cal S paces

HT2 =​{T2, T5, T8, T11}. The coset H is a subgroup the above generalized interval subspace
of G; HT1 and HT2 are not subgroups. Elements property describes.
such as T4 and T10, both in the coset HT1, are (b) In pitch-​class space pc, show how the
congruent mod H in the sense that T4 can be augmented triad {D, F♯, B♭} may be
multiplied by some element of H, namely T6, to generated from the note E♭ and a certain
give T10. Intuitively, two transpositions that are coset of a certain subgroup of the interval
congruent mod H might be considered “congru- group ℤ12.
ent modulo a diminished seventh chord”: they (c) Prove the generalized version of the interval
might map the note C, for example, to two dif- subspace property stated above, and
ferent places, but we can be certain that the two observe how commutativity of the interval
resulting notes will always have pitch-​class num-   group is essential in the proof.46
bers that differ by a multiple of 3. This relation
~H is a group congruence on G, so multiplication If G is a noncommutative group, things are
of cosets is well-​defined. For example, T4T11 =​ T3, more complicated. In general, some subgroups
and if we replace T4 and T11 with any other ele- will give rise to quotient groups while others
ments in the same cosets (HT1 and HT2 respec- will not. To illustrate, consider our most familiar
tively), such as T10 and T8, the resulting product, example of a noncommutative transformation
in this case T10T8 =​ T6, is guaranteed to be in the group, the dihedral group G =​ 𝒯ℐ12 =​{T0, …, T11,
same coset as T3, in this case the coset H itself. In I0, …, I11} acting on pc, and consider two sub-
the quotient group G ∕ H, therefore, the product groups of order 4: the cyclic subgroup H =​{T0,
(HT1)(HT2) is equal to H. We know from Section T3, T6, T9}, and K =​{T0, T6, I0, I6}, which as we
6.3 that every group of order 3 is cyclic; it is not observed in Section 6.3 is a Klein group.
difficult to see that G ∕ H is isomorphic to (ℤ3, +​) The index of each of these subgroups in G is
via the correspondence H ↔ 0, HT1 ↔ 1, HT2 ↔ 2. 24 ∕ 4 =​6. The cosets are obtained by multiplying
each element of each subgroup with other ele-
Exercise 6.6.3 In Section 5.4 we used the ments of G, which may be accomplished using
interval subspace property (Theorem 5.4.7) the now-​familiar formulas from Theorem 5.5.6.
to describe a “diminished-​seventh space” The six cosets of H are
{…, F♯3, A3, C4, E♭4, F♯4, …}, the subspace of pitch
generated by the element C4 and the subgroup H =​{T0, T3, T6, T9}, HI0 =​{I0, I3, I6, I9},
3ℤ of ℤ. A different diminished-​seventh space
such as {…, G3, B♭3, C♯4, E4, G4, …} could arise by HT1 =​{T1, T4, T7, T10}, HI1 =​{I1, I4, I7, I10},
choosing a reference point other than C4.
Now that we have studied cosets, a HT2 =​{T2, T5, T8, T11}, HI2 =​{I2, I5, I8, I11}.
generalized version of the interval subspace
property may be formulated in which the Calculation of the cosets containing inversion
reference point need not belong to the operators may be slightly less obvious than it
generated subspace. Suppose (G, ⊕) is a appears; in HI1, for instance, T3I1 =​ I10 (not I4)
commutative group; (S, G, int) is an interval while T9I1 =​ I4 (not I10). Nevertheless, the index
space; s0 is a fixed element of S; Gʹ is a subgroup numbers of the inversions within each coset
of G; and g0 is a fixed element of G. Define a remain congruent mod 3. The situation is some-
subset Sʹ of S consisting of all elements x in S what similar to the case considered above in
such that int(s0, x) is an element of the coset which G was not 𝒯ℐ12 but 𝒯12. The subgroup H is
Gʹ ⊕ g0. (The original interval subspace property the same while the group G is now twice as large,
required that int(s0, x) belong to Gʹ itself.) The so there are twice as many cosets, but three of
generalized interval subspace property states them are exactly the same as before, the trans-
that, under these conditions, (Sʹ, Gʹ, intʹ ) is formations in each coset are still “congruent
an interval space, where intʹ: Sʹ × Sʹ → Gʹ is the modulo a diminished-​seventh chord,” and the
same as the function int, but defined only on relation ~H is again a group congruence. The fol-
the elements of Sʹ. lowing exercise explores a few properties of this
congruence and the quotient group G ∕ H.
(a) Verify that the diminished-​seventh space
{…, G3, B♭3, C♯4, E4, G4, …} is the subspace Exercise 6.6.4
of pitch generated by the element C4 and (a) In the situation described above, observe
the coset 3ℤ +​1 of the subgroup 3ℤ, as that I4 ~H I7 (because I4 and I7 are in the

Permutations, Isomorphisms, and Other Topics • 243

 same coset HI1), and T2 ~H T8 (both are in T1 ~K I7 and I2 ~K T8. Is it true that T1I2 ~K
the coset HT2). Verify that I4T2 ~H I7T8. I7T8? Explain why your answer shows that
(b) In a similar manner, verify that T2I4 ~H T8I7. ~K is not a group congruence on G and
(c) Also verify that I4I3 ~H I7I0. that there is no coherent way to define the
(d) Use the results from parts (a)–​(c) to product (KT1)(KT2) of right cosets in G ∕ K.
determine the products (HI1)(HT2), (b) Also write K~ for “left-​coset equivalence.”
(HT2)(HI1), and (HI1)(HI0) in G ∕ H. Produce an example similar to part (a) to
(e) Is G ∕ H a commutative group? From Section show that K~ is not a group congruence on
6.3 we know that every group of order 6 is G and that there is no consistent way to
isomorphic either to the cyclic group 𝒞6 or   define a product of left cosets.
to the symmetric group 𝒮3. Which type of
  group is G ∕ H? Exercise 6.6.6 Return to the group H =​{T0, T3,
T6, T9} from Exercise 6.6.4. Verify that the left
Now consider the subgroup K =​{T0, T6, I0, I6} cosets
   of H are the same as its right cosets.
in place of H. This subgroup contains transposi-
tions and inversions in equal numbers, and a full The last two exercises capture the significant
set of cosets can be obtained by multiplying K by difference between the subgroups H and K. In
transpositions only: general, a subgroup of a group G is called a nor-
mal subgroup of G if its right cosets are the same
K =​{T0, T6, I0, I6}, KT3 =​{T3, T9, I3, I9}, sets as its left cosets. Of the above two examples
of subgroups of G =​ 𝒯ℐ12, the subgroup H =​{T0,
KT1 =​{T1, T7, I1, I7}, KT4 =​{T4, T10, I4, I10}, T3, T6, T9} is normal but the subgroup K =​{T0, T6,
I0, I6} is not. If H is a normal subgroup of G and
KT2 =​{T2, T8, I2, I8}, KT5 =​{T5, T11, I5, I11}. if x ∈ G, then the cosets Hx and xH are the same
set, and for any element h of H, the elements hx
Each transformation in 𝒯ℐ12 appears in one of and xh belong to the same coset as x itself. For
these cosets, and this list might appear to be a every normal subgroup H a relation ~H may be
logical partition of the elements of G; at least defined so that for x and y in G, x ~H y means that
it defines a sort of tritone equivalence of index x and y belong to the same coset of H (we need
numbers. But the presence of both transposi- not specify right or left cosets, because they are
tions and inversions in the same coset is a com- the same). As long as the subgroup H is normal,
plication. The first sign that something is amiss ~H is a group congruence; it is therefore pos-
is the realization that when multiplying K by the sible to define a binary operation on the cosets
transpositions T1, …, T5, it makes a difference in G ∕ H by the equation (Hx)(Hy) =​ H(xy), and
whether we choose to multiply on the right or on under this operation, G ∕ H becomes a group. No
the left. Recall, for example, that I0T1 =​ I1, while such quotient group construction is possible for
T1I0 =​ I11. Therefore I1 belongs to the right coset subgroups that are not normal.
KT1 in the above list, but does not belong to the If H is a normal subgroup of G, then every x
left coset T1K. The full list of left cosets of K is in G defines a coset Hx in G ∕ H, so we may define
a function QH: G → G ∕ H by the equation QH(x)
K =​{T0, T6, I0, I6}, T3K =​{T3, T9, I3, I9}, =​ Hx. This function is simply the quotient map
of the equivalence relation ~H, as defined in
T1K =​{T1, T7, I5, I11}, T4K =​{T4, T10, I2, I8}, Section 2.4, where such maps were denoted Q~.
The equation (Hx)(Hy) =​ H(xy), by which we have
T2K =​{T2, T8, I4, I10}, T5K =​{T5, T11, I1, I7}. defined the group operation in G ∕ H, then says
that QH(x)QH(y) =​ QH(xy), which means that the
Only K and T3K match the corresponding right quotient map QH is a homomorphism. We call
cosets. In attempting to form a quotient group such a quotient map a quotient homomorphism;
G ∕ K, we are faced with deciding which collection it is also often called the canonical (or natural)
of cosets to use. In fact, the following exercise homomorphism from G to G ∕ H. This mapping is
shows that neither one will work. automatically onto. It is not one-​to-​one unless
H is {e}, the trivial subgroup of G; this is because
Exercise 6.6.5 all elements of H map to the identity element H
(a) In the situation described above, write ~K of G ∕ H. In general, if f: G → Gʹ is a group homo-
for “right-​coset equivalence.” Therefore morphism, the set {x ∈ G | f(x) =​ e} is called the

244 • E x plor i n g M u s i cal S paces

kernel of f. In the present context, the kernel of is any element of H, then x−1hx ∈ H, thus
the quotient map QH: G → G ∕ H is the normal confirming that H is a normal subgroup
subgroup H itself. It happens that the kernel of of G.48
any group homomorphism f: G → Gʹ is always a (c) If G =​ 𝒯ℐ12 and K =​{T0, T6, I0, I6}, use one
normal subgroup of G. In fact, this is another of your formulas from part (a) to find
way to characterize normal subgroups: a sub- elements x in G and k in K such that x−1kx ∉
group of G is normal if and only if it is the kernel K, thus confirming that K is not a normal
of some homomorphism defined on G.   subgroup of G.
There are still other ways to define normal
subgroups. Rather than using the normal sub- Exercise 6.6.8 Consider the permutation
group H to define the congruence ~H, it is pos- group 𝒮3 acting on the set {A, B, C}, as in the
sible to work the other way around: if ~ is any table of Figure 6.2.1b and the Cayley diagram
group congruence on a group G, then H =​{x ∈ of 6.2.5b. Let H =​{E, (A, B, C), (A, C, B)} be the
G | x ~ e}, the set of elements of G congruent cyclic subgroup of order 3 generated by the
to the identity, is a normal subgroup of G, and 3-​cycle (A, B, C), and let K =​{E, (A, B)} be the
the cosets in G ∕ H are precisely the equivalence cyclic subgroup of order 2 generated by the 2-​
classes of the congruence ~. Another character- cycle (A, B). List the elements in all the right
ization involves the operation of conjugation. and left cosets of both H and K. One of the
A group element of the form x−1hx is called the two subgroups is normal and the other is not.
conjugate of h by x. A subgroup H of a group G is Which
   is which?49
normal if and only if, for every x in G and every
h in H, x−1hx belongs to H; that is, H is a normal The operation of conjugation is very impor-
subgroup of G if and only if H contains the con- tant in group theory. The abstract transformation
jugates of all of its elements by all elements of G. network in Figure 6.6.9a suggests one reason.
(If x−1hx =​ hʹ ∈ H, then hx =​ xhʹ. But hx belongs to Suppose that objects x and y in some space S
the right coset Hx, while xhʹ belongs to the left are related by a transformation g. Suppose also
coset xH; if this is true for every x, it follows that that f: S → Sʹ is a mapping from S into a space Sʹ
the cosets are the same.)47 (which may or may not be the same as S), and
suppose that we are interested in knowing how
Exercise 6.6.7 the elements f(x) and f(y) are related.50 There is,
(a) In 𝒯ℐ12, determine formulas for the conjugate in general, no guarantee that the same transfor-
of Tn by Tm, the conjugate of Tn by Im, the mation g that maps x to y also maps f(x) to f(y); it
conjugate of In by Tm, and the conjugate of is possible that the function g is not even defined
In by Im. That is, use the formulas from on the space Sʹ. But as long as the function f is a
Theorem 5.5.6 to simplify the expressions bijection, we can use its inverse function f −1 to
Tm−1TnTm, Im−1TnIm, Tm−1InTm, and Im−1InIm. trace a path
(b) If G =​ 𝒯ℐ12 and H =​{T0, T3, T6, T9}, use
appropriate formulas from part (a) to −1 g
f ( x ) f→ x  f
→ y →
 f ( y ),
explain why, if x is any element of G and h

FIGURE 6.6.9 Conjugation of functions

Permutations, Isomorphisms, and Other Topics • 245

from which we may conclude that while f(x) and A few other definitions relevant to noncom-
f(y) may not be related by g, they are at least mutative groups will be useful in later chapters. If
related by a conjugate of g, namely f −1gf. G is a group and A is a subset of G (not necessarily
This formula is valid whether or not f and g a subgroup), the centralizer of A in G is the set
belong to a transformation group, but is of spe-
cial importance if they do. Figure 6.6.9b shows a CentG ( A) = { x ∈G|xa = ax for every a ∈ A}.
simple case involving the action of the dihedral
group 𝒯ℐ12 on triads. Here the roles of the func-
tions f and g are played by the transposition T7 That is, the centralizer of A consists of those ele-
and the inversion I8, respectively. The C-​major ments of G that commute with all elements of A.
triad is mapped by I8 to the C♯-​minor triad. When It turns out that CentG(A) is always a subgroup of
both triads are transposed by T7, the resulting G. If A =​{a} consists of a single element, we write
G-​major and G♯-​minor triads are related not by CentG(a) (rather than the more cumbersome
I8 but by its conjugate T7−1I8T7, which (as one of CentG({a})) for the centralizer, which is simply
your formulas from Exercise 6.6.7a should con- the set of those elements that commute with a.
firm) is equal to I10. Conjugation relations of this The centralizer of the entire group, CentG(G) or
kind will play an important role in our discussion Cent(G) for short, is known as the center of the
of isomorphism of transformation networks in group G; Cent(G) consists of those elements of G
Section 9.3. that commute with every element of G. Cent(G)
is always a normal subgroup of G; centralizers of
Exercise 6.6.10 Make diagrams analogous to subsets of G may fail to be normal.
those in Figure 6.6.9b, always starting with the If x and y are elements of G, the commutator
C-​major triad, in which the following pairs of of x and y is the element x−1y−1xy. If x and y com-
transformations play the roles of f and g: mute, their commutator is equal to the identity
element e. The subgroup of G generated by all
(a) f =​ I8, g =​ T7 commutators of pairs of elements of G is called
(b) f =​ I3, g =​ I8 the commutator subgroup of G.
(c)  f =​ T7, g =​ T3 It is worth noting how much simpler all of
this is if the group G is commutative. In that
We are familiar with the fact that every group case, right-​multiplication and left-​multiplication
acts on itself by multiplication (or whatever the are the same; right and left cosets are the same;
group operation is); that is, for every element all subgroups of G are normal; every conjugate
a of a group G we can define a function Multa: x−1hx is equal to the element h being conjugated;
G → G by the equation Multa(x) =​ xa for all x in G. every element is in a conjugacy class by itself; the
We can also see now that conjugation defines an centralizer of every subset is the whole group;
action of a group on itself. That is, if G is a group the group is its own center; all commutators are
and a ∈ G, we may define a function Conja: G → equal to the identity; and the commutator sub-
G by the formula Conja(x) =​ a−1xa for all x in G. group is the trivial group {e}. For a noncommu-
In fact, for every a in G, the conjugation Conja tative group, some of these concepts measure in
is an automorphism of G, known as an inner some sense how close G is to being a commuta-
automorphism. tive group: conjugacy classes and commutator
In Section 6.5 we saw that any transforma- subgroups may be large, and centers and cen-
tion group G acting on a set S determines an tralizers may be small. The identity e, however,
equivalence relation ~G on S, defined so that x always commutes with every element, so e is
~G y if and only if y =​ f(x) for some f in G. When always in a conjugacy class by itself and belongs
G acts on itself by conjugation, the group G and to every centralizer.
the set S are the same thing. In this case, x ~G The following exercise explores how these
y if and only if x and y are conjugates—​that is, concepts apply in the case of the dihedral
if and only if there is some a in G such that y =​ group 𝒯ℐ12.
a−1xa. The equivalence classes of this relation
(the orbits of the conjugation action) are known Exercise 6.6.11
as conjugacy classes. The example in Figure 6.6.9b (a) Use the formulas from Exercise 6.6.7a to
shows that in the group 𝒯ℐ12, the inversions I8 make a list of all the conjugacy classes in the
and I10 belong to the same conjugacy class. group G =​ 𝒯ℐ12.

246 • E x plor i n g M u s i cal S paces

(b) Use the formulas from Theorem 5.5.6 to Consequently, whenever two groups are anti-​
determine CentG(T3), and verify that this isomorphic, they are automatically isomorphic
set forms a subgroup of G. also—​but if the groups are noncommutative, the
(c) Determine CentG(I3), and verify that it is a isomorphisms and anti-​isomorphisms are differ-
subgroup of G. ent functions.
(d) What is CentG({T3, I3})?
(e) What is Cent(G)?
(f) All possible commutators in G are of
NOTES
the forms Tm−1Tn−1TmTn, Tm−1In−1TmIn, 1. Hints for Exercise 6.1.3: In part (a), if 1 ≤ n ≤
Im−1Tn−1ImTn, and Im−1In−1ImIn. Calculate 11, then every cycle in the cycle structure of Tn is of
formulas for each of these. length 12 ∕ d, where d is the greatest common divisor
(g)  What is the commutator subgroup of G?51 of n and 12. For part (b), Figure 5.5.4 may be help-
ful. The sum of the answers to questions (a) and (b) is
Recall from Section 6.3 that the definition 8, but there are only seven different cycle types alto-
of a homomorphism from a group G to a group gether, because (as you should have discovered in the
Gʹ requires that f(x ⊕ y) =​ f(x) ⊗ f(y), where ⊕ previous exercise) one of the transposition cycle types
and ⊗ are the group operations in G and Gʹ duplicates one of the inversion cycle types.
respectively. When noncommutative groups are 2. Hints for Exercise 6.1.6d–​e: The second rows of
involved, the order of elements matters on both the matrix representations for p and q should be rec-
sides of this equation. Occasionally one encoun- ognizably similar. For (e), what do you get if you sub-
ters functions that reverse, rather than preserve, tract 3 from every number in the second row of the
the order of group products, satisfying the equa- matrix for p?
tion f(x ⊕ y) =​ f(y) ⊗ f(x) instead. Such a func- 3. In permutation theory 2-​cycles are commonly
tion is called an anti-​homomorphism. Extending called transpositions, as they simply “transpose” two
this definition in fairly obvious ways, an anti-​ elements. For obvious reasons, it seems prudent to
homomorphism that is one-​to-​one and onto is avoid this terminology in a musical context. For an
an anti-​isomorphism; an anti-​isomorphism from elementary discussion of permutations including a
a group to itself is an anti-​automorphism; and if proof of the theorem about the number of 2-​cycles
an anti-​isomorphism exists between two groups in a permutation representation, see Chapter 13 of
they are said to be anti-​isomorphic. In the case of Grossman and Magnus 1964.
commutative groups, homomorphisms and anti-​ 4. Hint for Exercise 6.1.8: In parts (b), (c), and (e),
homomorphisms are the same thing, but in the you should find even and odd permutations in equal
noncommutative case they may be different. numbers.
5. Other musical applications of permutation the-
Exercise 6.6.12 Define a function g: 𝒯ℐ12 → ory range from dice games for producing short com-
𝒯ℐ12 by the formulas g(Tn) =​ T−n and g(In) =​ In. positions, in vogue in the late 18th century (Ratner
Use the formulas from Theorem 5.5.6 to show 1970), to a formalization of “banjo pick permutations”
   g is an anti-​automorphism of 𝒯ℐ12.
that (Rockwell 2009). Permutation theory also figures
in techniques for enumerating set classes and other
The function g in the above exercise has an musical objects (Alegant 1993, Hook 2007d).
alternate definition: for every X in 𝒯ℐ12, whether 6. The reader who takes an interest in such details
a transposition or an inversion, g(X) is equal to may confirm that if n =​2, the inversions I0 and I1 are
X−1. An analogous formula g(x) =​ x−1 defines an the same as the transpositions T0 and T1—​inevitably,
anti-​automorphism on any group G whatsoever, as these are the only two permutations in the group
known as the inverse mapping on G (not to be 𝒮2—​so in this case 𝒟2 has only two elements, not
confused with an inverse function f −1). There are four, and 𝒞2 =​ 𝒟2 =​ 𝒮2. If n =​3, inversions are differ-
cases in which the most readily apparent map- ent from (and do not commute with) transpositions,
ping between two groups turns out to be an anti-​ but because 3! =​2 ∙ 3 =​6, the group 𝒟3 is the same
homomorphism or anti-​isomorphism; we shall as 𝒮3. Permutations in 𝒮3 will be studied in Example
see musically relevant examples in Chapters 7 6.1.13 below.
and 8. An anti-​isomorphism f: G → Gʹ may always 7. Hints for Exercise 6.1.9: If the answers to (a)
be composed with the inverse mapping g: Gʹ → Gʹ are not obvious, construct the matrix representa-
on the group Gʹ. In this case f ● g, the composi- tions for M1 and M11. For (c), use the formulas Tk(x) =​
tion of two order-​reversing anti-​isomorphisms, is x +​ k and M7(x) =​7x (mod 12) to write formulas for
an order-​preserving isomorphism from G to Gʹ. T1 ● M7 and M7 ● Tk. For (d), remember that T1

Permutations, Isomorphisms, and Other Topics • 247

generates all transposition operators, and T1 and I the entirety of “Farben,” Op. 16, No. 3. The analysis
generate all inversion operators, so it suffices to show of “Farben” in Burkhart 1973–​74 uncovers striking
that I and M11 can somehow be written as combina- patterns in the changing instrumentation, though
tions of M5, M7, and T1. those patterns cannot be described strictly in terms
8. Hints for Exercise 6.1.12a–​b: The case n =​12, k =​8 of permutations.
is instructive for part (a). In this case, what is d, what 15. Hint for Exercise 6.2.3b: The equations from
is j, and what is the cycle type of T8? For (b), note that Theorem 5.5.6 are all valid in 𝒯ℐ4, except that the sub-
if n (=​ d ∙ j) is odd, then d and j must both be odd, but if scripts are integers mod 4 rather than mod 12.
n is even, at least one of d and j must be even. 16. The row/​column permutation property is not
9. Hints for Exercise 6.1.12c–​f: The cycle types in sufficient to ensure that a group table is completely
parts (c)–​(e) may be expressed in terms of the num- correct; indeed, not every matrix satisfying this prop-
bers r and s. For (f), observe that if r is even, then n =​ erty is the group table for some group. Sudoku grids
0 (mod 4); if r is odd, then n =​2 (mod 4); if s is even, are not generally group tables, and although group
then n =​1 (mod 4); and if s is odd, then n =​3 (mod 4). theory underlies much of twelve-​tone theory in ways
10. The idea of applying permutation group theory to be described in Chapter 9, a twelve-​tone matrix is
to lines in a contrapuntal texture was suggested by not a group table either.
Daniel Harrison (1988) and pursued by David Lewin 17. The British mathematician Arthur Cayley
in notes later published in the Preface to the 2007 (1821–​95), who was employed as a lawyer during
reprinting of GMIT (Lewin [1987] 2007, xviii–​xx). For many of his most productive years of mathematical
an extended permutational analysis in four voices, work, was the first to write the modern definition of a
see Gollin 2009 (inspired by Hisama 1995), a study group and to recognize the equivalence of this defini-
of Ruth Crawford’s use of instrumental register in tion with permutation groups. The graphs mentioned
the third movement of her string quartet. Generally, in the title of Grossman and Magnus’s book Groups and
n-​voice counterpoint admits n! permutations. Writing Their Graphs (1964) are Cayley diagrams. Many more
out the 120 permutations of the five-​part counter- examples of Cayley diagrams and group tables appear
point that brings down the curtain on the “Jupiter” in Carter 2009. Gollin 2000 studies Cayley diagrams
Symphony has been suggested as apt punishment for in the context of musical transformation groups.
refractory music theory students. Joseph Riepel in 18. The function Multa is defined here by mul-
1755 did write out the 120 possible orders in which tiplication by a on the right (Multa(x) =​ xa). If the
five closely related keys may appear in a composition group G is not commutative, then multiplication by
(Ratner 1970, 353). a on the left (Multaʹ(x) =​ ax) may define a different
11. We originally defined an action of 𝒮3 on the set function. Either right-​or left-​multiplication may be
S =​{A, B, C}, not on the set of contrapuntal configu- used to define group actions and Cayley diagrams,
rations such as A/​B/​C. Extending this action to con- but we use right-​multiplication because it accords
figurations is a straightforward matter, however: if p ∈ better with the left-​to-​right functional orthography
𝒮3, we simply define p(A/​B/​C) to be the configuration that we generally favor; this definition ensures, for
p(A)/​p(B)/​p(C). It is this action of 𝒮3 on the space of example, that if a and b are two elements of G, then
six configurations that is simply transitive; the origi- the function Multab is the same as Multa ● Multb. The
nal action of 𝒮3 on the three-​element set {A, B, C} is distinction between right-​and left-​ multiplication
not simply transitive. will figure importantly in the discussion of quotient
12. Your analysis of the permutations in the C-​ groups in Section 6.6.
Minor Fugue may be compared with the table in 19. The dihedral group 𝒟n, rather than being
Schenker [1926] 1996, 33. defined as the group of rotations and reflections of an
13. The F-​Minor Sinfonia is among the works exam- n-​sided polygon, can instead be defined purely alge-
ined in Harrison 1988, as is the D-​Major Sinfonia. A braically as the group generated by two elements a and
persistent rumor holds that Bach never employs all six b such that an =​ e and ab =​ ba−1. Here T1 plays the role
of the configurations available in triple counterpoint, of a, I the role of b.
but the D-​Major Sinfonia is an instance in which he 20. A serial group 𝒮𝑒𝑟n acting on permutations of
does so. The rumor may have originated with Tovey n objects may be defined; of course 𝒮𝑒𝑟12 is the famil-
(1931, 70), but Tovey qualifies his observation, saying iar example. 𝒮𝑒𝑟12 is sometimes called the classical
only that “I know of no case” in which Bach uses all six serial group; in Chapter 9 we will extend 𝒮𝑒𝑟12 to larger
permutations. groups containing other operations such as serial
14. Changing instrumental colors were of great rotations.
importance to Schoenberg in many works, most 21. The elements of the group 𝒜𝑓𝑓12 are called
famously in the five-​voice structure that runs through twelve-​tone operators (TTOs) by Morris (1987, 66).

248 • E x plor i n g M u s i cal S paces

Morris and some other authors use the abbreviation tiny proportion of the permutations available in this
M for M5 rather than M7. group. His procedures in this work are described in
22. Hint for Exercise 6.2.14d: This graph, best visu- Berry [1976] 1987, 410–​13, and the implicit group
alized in three dimensions, is similar to Figure 4.1.12 structure is noted in Benson 2007, 358–​59.
but incorporates the twelve-​pointed star characteris- 28. Hint for Exercise 6.3.13: In this and the fol-
tic of M. The elements Tk and TkM occupy one layer, lowing exercise, you should be able to arrive at the
identical to Figure 6.2.13, and the elements Ik and IkM answers quickly by noting that if a group of order 6
form a similar figure in a second layer, connected to is commutative, it must be isomorphic to 𝒞6; if not,
the first layer by I. Recall that I and M are involutions it must be isomorphic to 𝒮3. Of course, it is still good
and that they commute with each other (see Exercise practice to construct the isomorphisms explicitly.
6.1.5c). To direct the T1 arrows properly in the second 29. Hint for Exercise 6.3.14a: The transformations
layer you will need to answer this question: if IkMT1 is P and L generate hexatonic triad cycles as in Figure
equal to IjM, what is the relationship between k and j? 3.5.8b. Cohn (1996, 18–​23) overlays a 𝒞6 group struc-
23. Hint for Exercise 6.2.16a: A “word” of more than ture on such a cycle to make the six triads into an
three letters, such as PLPL, may be simplified using interval space. In the context of the present exercise,
the identity LPL =​ PLP: that is, (PLP)L =​ (LPL)L =​ however, this apparent cyclic structure is misleading:
(LP)(LL) =​ LP. Words in musical transformation groups no single element of the PL group generates this cycle,
will be related to distance measures in Section 12.3. and the transformation that Cohn calls “T1” (rotation
24. Hint for Exercise 6.3.6: The corresponding by one position clockwise in Figure 3.5.8b) does not
facts about isomorphism of graphs were examined in correspond to a fixed element of the PL group. The
Section 3.2; those explanations carry over to groups Cayley diagram you constructed in Exercise 6.2.8b
with only minimal changes. may help you to see a relevant isomorphism for the PL
25. Hint for Exercise 6.3.10: G5 is a subset of the group. Clampitt 1998 compares the actions of several
group 𝒯ℐ12, so to verify that it is a group, it suffices different groups on a hexatonic cycle: Cohn’s cyclic
to check the requirements for subgroups from Section group, the PL group, and a group of transposition
5.2: closure, identity, and inverses. and inversion operators similar to those in Exercise
26. The classification of commutative groups is 6.3.13b–​c.
much easier: an elementary result called the funda- 30. Transformations equivalent to f and g are stud-
mental theorem of finite abelian groups states that every ied in Leong 2007 and Murphy 2009. Leong draws
finite commutative group is isomorphic to a direct parallels between these metric transformations and
product of cyclic groups. (Direct products are defined dramatic relationships in Humperdinck’s Hänsel und
in Section 6.4 below.) Of the 15 groups of order 24, Gretel and Wagner’s Parsifal, while Murphy postulates
only three are commutative; they are isomorphic to correspondences with relationships of key, harmony,
ℤ24, ℤ12 × ℤ2, and ℤ6 × ℤ2 × ℤ2. and form in Brahms.
27. For a “brief history” (running nearly 40 pages) 31. Perceptive readers will have noticed that
of work on the group classification problem, see the isomorphism f(x) =​2x does not map intervals
Solomon 2001. Humphreys 1996 includes detailed in cpitch, calculated from the pitch numbers of
study of many small groups, and an appendix describ- Figure 1.1.1, to the correct frequency ratios of
ing the structure of all groups up to order 31. The the corresponding musical notes. (For example,
simple groups include the cyclic groups 𝒞p of prime f(1) ∕ f(0) =​2 ∕ 1 =​2, but pitch numbers 0 and 1 are
order and the alternating groups 𝒜n. Some other only a semitone apart while 2 is the frequency ratio
“simple” groups are not so simple: a few of them are of an octave.) In fact, when two groups are isomor-
known as sporadic groups because they do not follow phic it is often possible to construct many different
any particular pattern, and one of these, famous as isomorphisms between them. In the present situ-
the monster group, has as its order a number of 54 ation, pitch numbers may be scaled up or down by
digits. Remarkably, one of the sporadic groups is at any constant factor before exponentiating, and the
least mathematically implicit in the serial transforma- exponentiation may be calculated using any base
tions used by Olivier Messiaen in “Île de feu 2” from b > 1 in place of 2. A more musically suitable iso-
Quatre études de rythme (1950). Messiaen derives the morphism might be the function given by f(x) =​ sx,
piece from two order-​number permutations whose where s =​21/12 ​ is the frequency ratio of an equal-​
cycle representations are (0, 6, 9, 1, 5, 3, 4, 8, 10, 11) tempered semitone.
(2, 7) and (0, 5, 8, 1, 6, 2, 4, 3, 7, 9, 10)(11); together 32. Hints for Exercise 6.4.2c–​d: There are several
these permutations generate a sporadic group of order ways to establish the isomorphisms in this exercise,
95,040 known to group theorists as the Mathieu the most explicit being to construct structure-​pre-
group M12. Needless to say, Messiaen uses only a serving bijections directly. Alternatively, you may

Permutations, Isomorphisms, and Other Topics • 249

use the transitive property of group isomorphism (if equivalence classes of this relation are not the orbits
two groups are both isomorphic to the same third of any standard group action.
group, they must be isomorphic to each other). These 39. See Humphreys 1996, 89–​97 for a thorough
properties are completely general: if G and H are any discussion of the orbit-​stabilizer theorem. If G is a
two groups at all, then U =​{(g, eH) | g ∈ G} and V =​ finite group, the theorem is an ordinary arithmetic
{(eG, h) | h ∈ H} are subgroups of G × H with U ≅ G equation. If G is infinite, then for every x in S, either
and V ≅ H (sometimes called the coordinate axis sub- Orb(x) or St(x) must be an infinite set (or both).
groups of G × H), and also G × H ≅ H × G. (The lat- 40. Hint for Exercise 6.5.5: In some order, the cardi-
ter isomorphism holds even if the groups G and H are nalities of the orbits of these seven sets are 1, 2, 4, 6,
noncommutative.) 12, 12, and 24.
33. Hint for Exercise 6.4.4c: If (a, b) corresponds to 41. Small orbits underlie Olivier Messiaen’s
T1, then (2a, 2b) corresponds to T2—​but remember descriptions of symmetry in terms of impossibil-
that a and 2a are integers mod 4, while b and 2b are ity, as when he refers to palindromic rhythms as
integers mod 3. “non-​retrogradable” and to transpositionally sym-
34. Hint for Exercise 6.4.5a: The statement to be metrical sets as “modes of limited transposition.”
verified may look complex, but this is exactly the situ- As he explains (Messiaen [1944] 1956, 8): “Modes
ation illustrated in Figure 6.3.2, the general picture of which cannot be transposed beyond a certain num-
a group homomorphism. The group called G in that ber of transpositions, because one always falls again
picture is the direct-​product group 𝒯12 × H, and the into the same notes; rhythms which cannot be used
group called Gʹ is the present transformation group G, in retrograde, because in such a case one finds the
generated by T1 and P. The operation ⊕ in Figure 6.3.2 same order of values again—​these are two striking
is the group operation on 𝒯12 × H, which is defined as impossibilities.”
usual for direct products by the equation (g1, h1) ∙ (g2, 42. Hints for Exercise 6.5.6: All the required exam-
h2) =​(g1g2, h1h2); the operation ⊗ is the group opera- ples for part (a) may be found in the preceding dis-
tion in G, namely composition of functions (which is cussion and exercises. The Petrushka chord, the only
also the group operation in both 𝒯12 and H, the com- set class that is transpositionally but not inversionally
ponent groups within the direct product). The ordered symmetrical, will be useful for (b), and you may be
pairs (T5, P) and (T9, T0) in 𝒯12 × H play the roles of able to find an example similar to the Petrushka chord
x and y in the figure. Of course any other elements that works for the last question in (d). For the first
of 𝒯12 × H could have served as well for illustrative question in (d), 𝒯ℐ12 has only one subgroup of order
purposes. 3, so that group is the only possible stabilizer for such
35. Hint for Exercise 6.4.6: Remember Theorem 5.5.6e. a set—​but you should be able to see that any pc set
36. Hint for Exercise 6.4.8: A Cartesian product of fixed by those transformations must be fixed by oth-
three sets is most conveniently depicted as a three-​ ers as well.
dimensional figure. The group 𝒞2 × 𝒞2 × 𝒞2, inciden- 43. Hint for Exercise 6.5.7: The formulas from
tally, is the smallest group that cannot be generated by Theorem 5.5.6 may be useful. In two of the first four
only one or two of its elements: any two non-​identity parts of this exercise, Y is fixed by exactly the same
elements of 𝒞2 × 𝒞2 × 𝒞2 generate a Klein subgroup operator(s) as X. The property in (e) is visible in set-​
of order 4. class tables that show degrees of symmetry; in the
37. The identity transformation in G maps every appendix to Straus 2016, for example, the numbers
x to itself, so x ~G x (the reflexive property). If f maps in the two symmetry columns are always of the form
x to y, then f −1 maps y to x, so x ~G y implies y ~G x “x, 0” or “y, y.”
(the symmetric property). Finally, if f maps x to y and 44. Hint for Exercise 6.5.10a: For the examples
g maps y to z, then f ● g maps x to z, so x ~G y and y ~G z mentioned in the introduction to the exercise, an
together imply x ~G z (the transitive property). octatonic collection has cardinality 8, T symmetry 4,
38. See Hook 2007d for combinatorial techniques and two modes, while a diatonic collection has cardi-
for enumerating orbits. Some sources give the num- nality 7, T symmetry 1, and seven modes.
ber of TI set classes as 222 or 220, disregarding sets 45. Symmetry of twelve-​tone rows will be dis-
of cardinalities 0 and 12, and possibly 1 and 11 as cussed in more detail in Sections 9.5 and 9.6, where we
well (there is just one set class of each of these car- will consider some groups larger than 𝒮𝑒𝑟12, admitting
dinalities). Not all equivalence relations are readily additional kinds of symmetry.
described by group actions. In Section 2.4 we noted 46. Hint for Exercise 6.6.3c: Review Exercise 5.4.8,
a third equivalence relation on pcset, coarser than TI in which the two important steps in proving the origi-
equivalence, namely interval-​content equivalence; the nal interval subspace property (corresponding to the

250 • E x plor i n g M u s i cal S paces

two properties required for a subset to be a subspace) parts (b)–​(e) and (g), in some order, are groups of
were outlined. order 2, 2, 4, 6, and 12.
47. Some authors write conjugates in the form
xhx−1 rather than x−1hx. Of course, both are conjugates SUGGESTED READING
of h; according to the terminology given here, x−1hx
is the conjugate of h by x, and xhx−1 is the conjugate Further details on the group-​theoretic topics in
of h by x−1. this chapter may be found in algebra texts such
48. Hint for Exercise 6.6.7b: Your formulas should as Dummit and Foote 2004, as well as in more
show that every subgroup of 𝒯12 (consisting of trans- advanced references such as Dixon and Mortimer
positions only) is a normal subgroup of 𝒯ℐ12. 1996 (permutation groups), Humphreys 1996 (the
49. Hint for Exercise 6.6.8: In any group, every sub- classification of finite groups), and Carter 2009
group of index 2 is normal. Your work for this exercise (group tables and Cayley diagrams, with many
will probably show you why. visually appealing presentations). Nathan Carter is
50. The situation described here is what Lewin also the developer of Group Explorer (nathancarter.
([1987] 2007, 148–​49) calls “system modulation.” github.io/​group-​explorer), a web application for
The point of the term is that the “system” contain- studying group structures and producing group
ing x and y is “modulated” by the function f to the tables and Cayley diagrams. An important early
system containing f(x) and f(y). A diagram similar article on group properties associated with pitch-​
to Figure 6.6.9a appears in Morris 2001, vol. 2, class space is Balzano 1980; in particular, Balzano
Example 4.7c. was the first to call attention to the importance
51. Hints for Exercise 6.6.11: You should find nine of the isomorphism between ℤ12 and ℤ4 × ℤ3 for
conjugacy classes in part (a). The five subgroups in tonnetz models of pitch-​class space.

Permutations, Isomorphisms, and Other Topics • 251

 7
Intervals

INTERVAL FUNCTIONS and interval spaces, difficult than it should. Even readers who have
introduced in Chapter 5, provide a valuable way not worked through all the theoretical devel-
of modeling musical relationships in many situ- opment may benefit from scanning the musi-
ations. This chapter further develops the theory cal examples in Sections 7.5 and 7.6. Detailed
of interval spaces and offers additional examples. proofs are not supplied for every assertion;
Among the concepts made precise here are label some proofs are suggested via exercises, and
functions (Section 7.1), homomorphisms and others are omitted entirely. Even though some
isomorphisms of interval spaces (mentioned sev- concepts in the following chapters formally
eral times in previous chapters and formalized depend on material in this chapter, readers
in Section 7.2), direct products and quotients who find the details here too knotty can prob-
of interval spaces (7.3 and 7.4), and general- ably skip ahead and read Chapters 8 and 9 more
ized kinds of transposition and inversion opera- comfortably.
tors (7.5 and 7.6). A few musical examples are
offered as illustrations; the chapter also includes 7.1 LABEL FUNCTIONS
an assortment of small mathematical theorems.
Some of these concepts have their origins in
FOR INTERVAL SPACES
Lewin’s GMIT, while others were introduced in The structure of an interval space is closely tied
Hook 2007a, but the presentation and examples to the structure of its interval group. This con-
here are new. nection is responsible for much of the theoreti-
This chapter is lengthy and sometimes cal appeal of interval spaces but also for some of
technical, especially in some parts of Sections their practical limitations. The label functions
7.2, 7.5, and 7.6, though complexity of nota- studied in this section make this correspondence
tion perhaps makes the material appear more between space and group more explicit.

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0007
 In the picture of pitch space from Figure 1.1.1, Then labu: S → G is the label function of S with
we may distinguish carefully between the pitch respect to the reference point u.1 The homogeneity
numbers and the pitches they stand for: pitches axiom for interval spaces guarantees that every
such as F3 and C5 are represented by the num- interval i in G occurs as the u-​label of exactly one
bers −7 and 12. From our understanding of pitch x in S (the unique x such that int(u, x) =​ i), so the
space as an interval space, we may recognize that function labu is both one-​to-​one and onto, put-
these pitch numbers are actually intervals, cho- ting the elements of the space S in one-​to-​one
sen from the interval group ℤ: a certain element correspondence with the elements of the group
of the space (middle C) is labeled with the iden- G. The unique point labeled with the identity
tity element of the group (0), and every other element of the group is the reference point u
pitch x is labeled with int(C4, x). Consequently itself: labu(u) =​int(u, u) =​ e.
the numerically labeled picture of the space Figure 7.1.1 shows two different label func-
pitch looks much like a picture of the group tions for pitch. The labeling in part (a) is the
ℤ: something resembling a Cayley diagram for a familiar one from Figure 1.1.1, namely labu,
group is used to depict a space with that inter- where u is the pitch C4. The labeling in (b) is labv,
val group. Identifying elements with intervals is where v is G4. All the labels are shifted by seven
perhaps even more familiar in the case of pitch-​ places from one picture to the other, 7 being the
class space pc. The usual pitch-​class numbers interval between the two reference points. The
are intervals reckoned from the referential pitch figure shows that we may regard a label func-
class C, but the correspondence is so ingrained tion as superimposing a picture of the interval
that we often imagine that the mod-​12 num- group (in this case ℤ) on a picture of the space
bers are pitch classes; again, a picture of the (pitch), effectively adjusting the relative posi-
space resembles a Cayley diagram for its interval tions of the two so that the identity element of
group ℤ12. the group aligns with the chosen reference point
The labeling of elements with intervals is in the space. The symmetry and homogeneity of
possible in any interval space. In Section 2.3 the group structure ensure that shifting of this
we noted that various label functions, mapping kind is always possible.
either from a space S to a set L of possible labels In pitch, the interval between two pitches
or vice versa, may be of interest. When (S, G, int) may be calculated by subtracting pitch num-
is an interval space, a convenient choice for a set bers: in traditional C4-​ based pitch numbers,
of labels for S is readily at hand in the group G, for example, int(E3, D5) =​14 − (−8) =​22. More
and a family of label functions labu: S → G may be generally, the equation int(x, y) =​labu(y) −
defined. Each such function labu depends on the labu(x) holds for any pitches x and y and any
choice of a reference point u in S. Any element of reference pitch u: in G4-​based labeling we could
S may be used as a reference point, though there make the above calculation in the form int(E3,
may be some musical or other reason for select- D5) =​7 − (−15) =​22, arriving at the same
ing a particular u. In pitch our reference point answer as before. The same equation works also
was C4; in pc it is traditionally the pitch class C. in pitch-​class space in mod-​12 arithmetic. In
Generally, for every x in S, labu(x), the u-​label of fact this equation holds in any interval space
x, is defined by the equation labu(x) =​int(u, x). whose interval group is additive (and therefore

FIGURE 7.1.1 Two label functions for pitch space: (a) labu, where u =​C4; (b) labv, where v =​G4

Intervals • 253
commutative). Some care must be taken in gen- it from the initial configuration at the top of the
eralizing to other cases, particularly the case of figure, the Cayley diagram in Figure 6.2.5b is the
a noncommutative interval group. If the inter- result. Because the symmetric group 𝒮3 acts on
val group is (G, ⊕), then the interval-​sum equa- the configurations in simply transitive fashion,
tion in the space S takes the form int(u, x) ⊕ Lewin’s theorem tells us that the configurations
int(x, y) =​int(u, y), or equivalently labu(x) ⊕ form an interval space with group 𝒮3, and the
int(x, y) =​labu(y). Techniques for solving this equa- label function just described identifies each ele-
tion for int(x, y) were derived in Section 5.3: add- ment of the space with an element of the group.
ing the inverse of labu(x) on the left to both sides
of the equation gives int(x, y) =​ ⊖labu(x) ⊕ labu(y). Exercise 7.1.4
This is the general equation for calculating an (a) In the interval space triad with interval
interval from the values of a label function: if group 𝒯ℐ12, using the C-​major triad as
labu(x) =​ i and labu(y) =​ j, then int(x, y) =​ ⊖i ⊕ j. the reference point, calculate labC(F)
If the operation ⊕ is the familiar commutative and labC(b♭), and verify that the interval
operation of ordinary addition (or mod-​12 addi- equation int(x, y) =​ ⊖labu(x) ⊕ labu(y) gives
tion), then ⊖i ⊕ j is the same as j − i. the correct result for int(F, b♭).3
(b) Verify that your equation from
Exercise 7.1.2 Verify that the equation Exercise 7.1.3 gives the correct result for
int(x, y) =​labu(y) − labu(x) holds in each of the int(F, b♭), where the two referential triads
following cases:   are u =​C and v =​f♯.
(a) S =​ pitch, u =​C4, x =​A4, y =​E3
(b) S =​ pitch, u =​D3, x =​A4, y =​E3 7.2 HOMOMORPHISMS
(c) S =​ pc, u =​C, x =​B, y =​E♭ AND ISOMORPHISMS
(d)  S =​ pc, u =​E♭, x =​B, y =​E♭
OF INTERVAL SPACES
Exercise 7.1.3 The interval equations derived The correspondence between space and group
above require that x and y be labeled with takes on additional significance when mappings
respect to the same reference point u. If u and from one space to another are considered. In
v are two different reference points, labu(x) =​ i, Section 6.3 we studied homomorphisms and iso-
labv(y) =​ j, and int(u, v) =​ k, write an equation morphisms of groups; these concepts have close
for int(x, y) in terms of i, j, and k. Give a simple counterparts in the theory of interval spaces.
equation valid in an additive interval group, As was the case with groups, construction of a
and also a more general equation valid in any homomorphism between two interval spaces is
(possibly noncommutative) group (G, ⊕). Check a valuable way of demonstrating relationships
that your equation works in pitch space when between the spaces. Because the interval group
u =​
  C4, v =​D3, x =​A4, and y =​E3.2 is an essential part of the structure of an interval
space, it should come as no surprise that homo-
We have noted other instances in which morphisms and isomorphisms of groups play
Cayley diagrams for groups closely resemble pic- central roles in the corresponding concepts for
tures of spaces on which the groups act. In many interval spaces. In general a homomorphism is
such cases a label function lies at the heart of a structure-​preserving mapping, the meaning
the correspondence. In Section 5.6, for exam- of the word “structure” depending on the kinds
ple, we saw that consonant triad space triad of objects being related by the homomorphism.
may be construed as an interval space whose When the objects are interval spaces, a homo-
interval group is the dihedral group 𝒯ℐ12. If we morphism must preserve something about the
start with the double-​ circle picture of triad space, something about the group, and some-
in Figure 4.1.1b but replace the name of each thing about the relationship between the two—​
triad X with labC(X) =​int(C, X), the interval a constraint that adds some complexity to the
from the C-​major triad to X in 𝒯ℐ12, the result- mathematical formalism required to express it.4
ing graph is exactly the Cayley diagram for Suppose that (S1, G1, int1) and (S2, G2, int2)
𝒯ℐ12 in Figure 6.2.10b. For another example, are two interval spaces, with interval groups
Figure 6.1.14b showed a space of contrapuntal (G1, ⊕) and (G2, ⊗) respectively. In this situation
configurations in Bach; if each configuration is two kinds of functions may be considered: map-
labeled with the permutation required to obtain pings f: S1 → S2 from one space to the other and

254 • E x p loring M usical S p aces

also mappings g: G1 → G2 from one group to the mapping associated with mod-​12 congruence,
other. We say that such functions f and g are Q12: ℤ → ℤ12, which reduces any integer to an
interval-​compatible (or simply compatible) if for integer mod 12. (All we need to know about Q12
all x and y in S1, g(int1(x, y)) =​int2(f(x), f(y)). As for the moment is that it is a group homomor-
depicted in Figure 7.2.1, interval compatibility phism, but in terminology from Section 6.6, Q12
shows something about how an interval in the is the quotient homomorphism QH from a group G
space S1 translates to an interval in S2. For any to its quotient group G/​H, where G is the addi-
pair of elements in S1, we can form their interval tive group ℤ of integers and H is the subgroup
(int1) in G1 and then map via g to an interval in 12ℤ. Such mappings are always well-​ defined
G2; alternatively we can first map the elements group homomorphisms as long as H is a normal
to S2 via f and then form the interval (int2) in G2. subgroup of G, which is always the case in com-
Compatibility of f and g means that the resulting mutative groups.) The interval functions are (in
interval in G2 will be the same in either case; in S1) int1 =​intpitch, which measures pitch intervals
short, compatibility says that f-​related elements in the space pitch, and (in S2) int2 =​intpc, which
form g-​related intervals. measures mod-​12 intervals in pc. The interval
An interval-​space homomorphism from S1 to S2 compatibility condition in this case says that
is a function f: S1 → S2 that is interval-​compat- octave reduction can take place either before or
ible with some group homomorphism g: G1 → G2. after the calculation of an interval. For example,
(Recall the definition illustrated in Figure 6.3.2: if given pitches x =​B4 and y =​E♭3, we may form
the group operations in G1 and G2 are ⊕ and ⊗ their interval in pitch space and then reduce it
respectively, then g: G1 → G2 is a homomorphism mod 12, or we may arrive at the same result by
if g(i) ⊗ g(j) =​ g(i ⊕ j) for all i and j in G1.) If g is calculating the pitch-​class interval between the
a group isomorphism (a homomorphism that pitch classes B and E♭. The first approach gives
is one-​to-​one and onto), then a function f that
is compatible with g is an interval-​space isomor- g(int1(x, y))
phism, and in that case S1 and S2 are isomorphic = Q12(intpitch(B4, E♭3)) =​ Q12(−20) =​4;
interval spaces. It is possible that S1 and S2 are
the second gives
the same space S; an interval-​space automorphism
of S is an isomorphism from S to S. int2(f(x), f(y))
For a first example of an interval-​space homo- =​ intpc(QO(B4), QO(E♭3)) =​intpc(B, E♭) =​4.
morphism, let S1 and S2 be discrete pitch space
pitch and pitch-​ class space pc, respectively, The reader should understand how the agree-
and consider the mapping from pitch to pc ment between these two calculations illustrates
that maps every pitch to its corresponding pitch the general situation depicted in Figure 7.2.1.
class. In terminology from Section 2.4, pc is a Interval compatibility ensures that the
quotient set and this mapping is the quotient map pitch-​class interval between two notes can
associated with octave equivalence, QO: pitch always be calculated from the pitch interval: if
→ pc. (In fact, pc is a quotient interval space, in we know that a pitch interval is −20, we can be
a way to be made precise in Section 7.4.) The sure that the corresponding pitch-​class interval
group mapping g with which QO is interval-​com- is 4, even without knowing the original notes.
patible is also a quotient map, namely the group The converse, of course, is not true: intervals in
pitch space cannot be deduced from intervals
in pc space. Generally, if we know an interval-​
space homomorphism f and if we know how
to calculate int1(x, y) in S1, then we can always
deduce the value of int2(f(x), f(y)) in S2. If the
homomorphism f maps S1 onto S2 (as is the case
in the above example), then all int2 intervals in
S2 can be deduced from corresponding intervals
in S1. But interval compatibility does not give
us a way to calculate intervals in S1 from inter-
vals in S2.
It is also easy to see that Q12 is the only map-
FIGURE 7.2.1 The interval compatibility condi- ping g: ℤ → ℤ12 with which f =​ QO could possi-
tion for an interval-​space homomorphism bly be compatible: the compatibility condition

Intervals • 255
g(intpitch(x, y)) =​intpc(QO(x), QO(y)) determines compatibility condition fails if the
what g(intpitch(x, y)) must be for all pitches x and equation f(u) =​ v is not true. (A more
y, and therefore what g(i) must be for every i in general equation, valid for interval-​space
the interval group G1. This uniqueness of g is a homomorphisms even if f(u) ≠ v, appears
general principle: a mapping of interval spaces in Theorem 7.2.13c below.)
f: S1 → S2 can be interval-​compatible with at most (c) Show that if g is a group homomorphism,
one function g: G1 → G2. (This principle depends then label compatibility implies interval
upon the homogeneity axiom for interval spaces, compatibility. That is, assuming that
which ensures that every element i of G1 occurs a mapping f: S1 → S2 and a group
as int1(x, y) for some x and y in S1.) In the case of homomorphism g: G1 → G2 satisfy
functions f that are not homomorphisms, there g(labu(x)) =​labv(f(x)) for every x in S1, show
is no such group mapping g at all; the following that g(int1(x, y)) =​int2(f(x), f(y)) for every x
exercise offers an example.   and y in S1.6

Exercise 7.2.2 Define f: pitch → pc so that Of course, pitch space and pitch-​class space
f maps C4 and all lower pitches to the pitch are not isomorphic as interval spaces. The func-
class C but maps all pitches higher than C4 to tion g =​ Q12: ℤ → ℤ12 is onto but not one-​to-​one,
D. Explain why no function g: ℤ → ℤ12 can be so g is not a group isomorphism and f =​ QO: pitch
interval-​compatible with f, so that f is not an → pc is not an interval-​space isomorphism. For
interval-​
   space homomorphism.5 an example of an interval-​space isomorphism,
let S1 be pitch as before, and let S2 be generic
Another important property illustrated by pitch space gpitch, first depicted in Figure 1.1.6.
the homomorphism QO is label compatibility. An isomorphism between pitch and gpitch was
If, as usual, pitches are labeled with integers mentioned informally in Exercise 1.1.9, and the
relative to the reference point u =​C4 and pitch graphs of the two spaces were observed to be
classes are labeled with mod-​12 integers relative isomorphic in Exercise 3.3.9; we are now able to
to v =​C, then the mod-​12 reduction of any pitch show that they are isomorphic as interval spaces.
label is the same as the corresponding pitch-​class Both spaces have the integers ℤ as their inter-
label: Q12(labu(x)) =​labv(QO(x)). Taking the pitch val group; the interval functions simply count
A5 as an example, Q12(labu(A5)) =​ Q12(21) =​9, steps within their respective scales. Because
while labv(QO(A5)) =​labv(A) =​9. Label compat- a mapping from pitch to gpitch maps from
ibility is a consequence of interval compatibility one scale to another, such a function is called
in general: if mappings f: S1 → S2 and g: G1 → G2 an interscalar mapping. The homomorphism
are interval-​compatible, and if u and v are ref- f: pitch → gpitch maps C4 to C4 and maps a
erence points in S1 and S2 such that f(u) =​ v, note k chromatic steps above or below C4 in pitch
then for every x in S1, the interval compatibil- to the note k generic steps above or below C4 in
ity condition applied to the elements u and x gpitch. More precisely, for any x in pitch, f(x) is
is g(int1(u, x)) =​int2(f(u), f(x)), which says that the unique y in gpitch such that intgpitch(C4, y) =​
g(labu(x)) =​labv(f(x))—​the general form of the intpitch(C4, x)—​or equivalently the unique y
label compatibility condition. If elements of the in gpitch whose v-​label is the same as the u-​
spaces S1 and S2 are labeled with their u-​ and label of x, where both reference points u and v
v-​labels, then label compatibility says that the are C4 in their respective spaces. For example,
mapping g on these labels has the same effect as f(G4) =​C5 because intpitch(C4, G4) =​intgpitch(C4,
the mapping f on the original elements. In the C5) =​7, and likewise f(E♭3) =​A2 because intpitch(C4,
above example, label compatibility is responsible E♭3) =​intgpitch(C4, A2) =​−9.
for the striking similarity between the quotient The interval compatibility condition in this
maps f =​ QO: pitch → pc and g =​ Q12: ℤ → ℤ12: if situation takes a special form: for this func-
pitches and pitch classes are identified with their tion f, int1(x, y) =​int2(f(x), f(y)) for every x
standard labels, f and g are the same function. and y in pitch. For example, the chromatic
interval between G4 and E♭3 is the same as the
Exercise 7.2.3 generic interval between f(G4) and f(E♭3) (that is,
(a) Draw a diagram similar to Figure 7.2.1 between C5 and A2); with f defined as above, both
illustrating label compatibility. intervals are −16. No function g appears in the
(b) Using the function f =​ QO above, give equation int1(x, y) =​int2(f(x), f(y)). The left side
an example to show that the label of the equation is, however, equal to g(int1(x, y)),

256 • E x p loring M usical S p aces

where g is the identity mapping Eℤ on ℤ (g(n) =​ n g(intpitch(x, y)) =​ Eℤ(intpitch(x, y))
for every integer n). That is, the group mapping =​intpitch(x, y)
with which f is interval-​compatible is Eℤ. The =​intpitch(x, C4) +​intpitch(C4, y)
identity mapping on a group is trivially an iso- =​−intpitch(C4, x) +​intpitch(C4, y)
morphism, so compatibility with Eℤ ensures that =​−intgpitch(C4, f(x)) +​intgpitch(C4, f(y))
f is an isomorphism between the two interval =​intgpitch(f(x), C4) +​intgpitch(C4, f(y))
spaces.   =​intgpitch(f(x), f(y)).
A function f such that int1(x, y) =​int2(f(x),
f(y)) for all x and y is said to be interval-​preserv- Interval-​space homomorphisms have many
ing. It is the fact that the two interval spaces distinctive properties, summarized in the
share the same interval group that makes the remainder of this section. The properties listed
interval-​preserving condition possible. One of in Theorem 7.2.5 are true of interval-​compat-
the intervals in the interval-​preserving condi- ible mappings in general; that is, they do not
tion is an int1 interval in the group G1 while require g to be a group homomorphism. The
the other is an int2 interval in G2, so the equa- later Theorems 7.2.7 and 7.2.13 depend on the
tion cannot be satisfied if the groups are dif- full homomorphism structure.
ferent. But it is common for different spaces
to share the same interval group, and in that Theorem 7.2.5: Properties of
situation interval-​ preserving mappings are interval-​compatible mappings
possible. As we saw in the previous example, Suppose (S1, G1, int1) and (S2, G2, int2) are
the interval-​ preserving condition says that interval spaces, and suppose the functions f:
the function is interval-​compatible with the S1 → S2 and g: G1 → G2 are interval-​compatible.
identity mapping on the group, and therefore
every interval-​ preserving mapping is auto- (a) Uniqueness of group mappings. The function
matically an interval-​ space homomorphism. g is uniquely determined by f. That is, if f is
Interval-​ preserving functions are closely interval-​compatible with both g: G1 → G2
akin to transposition operators, in ways to be and g ′: G1 → G2, then g =​ g ′.
detailed in Section 7.5. (b) Label compatibility. If u and v are reference
In particular, many spaces share the interval points in S1 and S2 such that f(u) =​ v, then
group ℤ, including not only pitch and gpitch for every x in S1, g(labu(x)) =​labv(f(x)).
but all the diatonic pitch spaces dpitch(n), (c) The function f: S1 → S2 is one-​to-​one if and
any of which could take the place of gpitch in only if g: G1 → G2 is one-​to-​one.
this discussion. Chromatic pitch space, generic (d) The function f: S1 → S2 is onto if and only if
pitch space, and all diatonic pitch spaces are   g: G1 → G2 is onto.
simply infinite scales; they may all be drawn
by relabeling the same picture, and they are all Parts (a) and (b) of Theorem 7.2.5 have been
isomorphic as interval spaces. There can be no established already. Parts (c) and (d) comprise
interval-​preserving function from chromatic four assertions—​the “if” and “only if” parts of
pitch-​class space pc to generic pitch-​class space each statement. Here we provide a proof that
gpc, because the interval groups are differ- if f is one-​to-​one, then g is one-​to-​one, leaving
ent: an interval in ℤ12 cannot also be an interval the other three parts for the following exercise.
in ℤ7. Interval-​preserving mappings are possible, Suppose f is one-​to-​one, and suppose i and j are
however, between gpc and a diatonic pitch-​class elements of G1 such that g(i) =​ g(j); we need to
space dpc(n). show that i =​ j. Choose any reference point u
in S1, and let x and y be the elements of S1 sat-
Exercise 7.2.4 In the above discussion of the isfying int1(u, x) =​ i and int1(u, y) =​ j (such ele-
isomorphism from pitch to gpitch, the interval ments exist by the homogeneity axiom for the
compatibility condition g(intpitch(x, y)) =​ interval space S1). The equation g(i) =​ g(j) there-
intgpitch(f(x), f(y)) was illustrated by example. fore says that g(int1(u, x)) =​ g(int1(u, y)), which
The chain of equalities below constitutes by interval compatibility translates to int2(f(u),
a more rigorous proof of this condition. f(x)) =​int2(f(u), f(y)). That is, in S2, both f(x) and
Explain every step in this chain, noting which f(y) lie at the same interval from f(u). But by
equalities illustrate general properties of the homogeneity axiom for S2, only a single ele-
interval spaces and which ones are specific to ment can lie at a particular interval from a given
this example. point; therefore f(x) =​ f(y). Since f is one-​to-​one

Intervals • 257
by hypothesis, it follows that x =​ y, and there- is easy to see why the f defined in the theorem is
fore i =​int1(u, x) =​int1(u, y) =​ j, which is what we the only possible function satisfying the stated
needed to prove. requirements; the interval compatibility of f and
g may be proved via a chain of equalities similar
Exercise 7.2.6: Complete the proofs of parts to that in Exercise 7.2.4 above.8
(c)  
and (d) of Theorem 7.2.5.7 It is convenient to introduce a notation for
the function f specified in the interval-​ space
Some parts of Theorem 7.2.5 reinforce our homomorphism theorem. Because f is completely
intuitions about the close parallels between the determined by the group homomorphism g and
structure of interval spaces and their interval the two reference points u and v, we shall write
groups. For example, because a group isomor- f = g uv , or occasionally in more explicit fashion
phism is both one-​to-​one and onto, it follows f = guv∈∈SS12 . The notation g uv may be pronounced “g-​
from (c) and (d) that the same is always true bar-​u-​v”; this function may be called the interval-​
of interval-​space isomorphisms, and moreover space homomorphism induced by g, mapping u to v.
that every interval-​space homomorphism that is Notice that u appears in this notation as the sub-
one-​to-​one and onto is an isomorphism, as we script and is pronounced before v, the superscript.
should expect. We will use similar notations in some other situ-
The functions f and g do not play entirely ations, always understanding that functions writ-
symmetrical roles. In particular, the uniqueness ten in this way map from subscript to superscript.9
property described in Theorem 7.2.5a works only The last sentence of the interval-​space homo-
in one direction. A given function f can be com- morphism theorem has a corollary already
patible with only one g, but a single function g noted in Chapter 6: if two interval spaces have
is typically compatible with more than one f. isomorphic interval groups, then the interval
Consider again the construction of an isomor- spaces themselves are isomorphic. If g: G1 → G2
phism from pitch to gpitch, and suppose we is a group isomorphism, then g uv is an interval-​
have decided that the group mapping g should be space isomorphism, interval-​compatible with g,
Eℤ—​that is, that f should be interval-​preserving. for any choice of reference points u and v at all.
If we know that f should map u =​C4 to v =​C4, we We commented above that chromatic, generic,
are left with no choice about how to define f(x) and diatonic pitch spaces are all isomorphic; in
for other elements x of pitch: the compatibility fact, any other type of scale—​a harmonic minor,
condition intgpitch(v, f(x)) =​intpitch(u, x) implies whole-​tone, or octatonic scale, for example—​also
that f must be the function described previously. defines an interval space whose interval group
We could obtain a different f, however, by declar- is ℤ and which is therefore isomorphic to pitch
ing that f should map u =​C4 to v =​C5, or by select- and gpitch. These spaces are all defined as infi-
ing some other pair of reference points entirely. nite scales in pitch space, not modular scales of
As the following theorem clarifies, the choice of pitch classes. Chromatic, generic, and whole-​tone
reference points in both spaces, together with the pitch-​class spaces have interval groups of differ-
homomorphism of interval groups, completely ent cardinalities and are not isomorphic. A har-
determines an interval-​space homomorphism. monic minor pitch-​class space is a seven-​note
space with interval group ℤ7 and so is isomor-
Theorem 7.2.7: Interval-​space phic to gpc. Scalar spaces will be studied in more
homomorphism theorem detail in Part Four.
Suppose that (S1, G1, int1) and (S2, G2, int2) Section 6.3 provided an example in which
are interval spaces, u ∈ S1, v ∈ S2, and g: G1 → G2 two interval groups are isomorphic but not iden-
is a group homomorphism. Then there exists a tical: continuous pitch space cpitch and fre-
unique interval-​space homomorphism f: S1 → S2, quency space freq, with interval groups (ℝ, +​)
interval-​compatible with g, such that f(u) =​ v. and (ℝ>0, ∙). As noted there, g(i) =​2i defines an
For every x in S1, f(x) is the unique y in S2 such isomorphism of the interval groups (this func-
that int2(v, y) =​ g(int1(u, x)) (or equivalently tion was called f in the discussion in Chapter 6),
labv(y) =​ g(labu(x))). The homomorphism f is an so the interval-​space homomorphism theorem
interval-​space isomorphism if and only if g is a confirms the isomorphism of the spaces.
group
   isomorphism.
Exercise 7.2.8
Mathematically proficient readers may try (a) The traditional reference points for the
writing out a rigorous proof of this theorem. It correspondence between pitch and freq

258 • E x p loring M usical S p aces

 are u =​A4 in pitch and v =​440 (Hz) in freq. The discussion is brief; readers are encouraged to
For this purpose the preferred interval supply missing details.
function in pitch space is intpitch′ =​ We start with a rather trivial example.
intpitch ∕ 12, which measures intervals in Discrete pitch space pitch is, of course, a sub-
octaves rather than semitones; the interval set of continuous pitch space cpitch. Define f:
function in frequency space, intfreq, pitch → cpitch by the equation f(x) =​ x—​that
calculates the ratio between two frequencies. is, for every x in pitch, f(x) is the same pitch,
The isomorphism f = g uv induced by the considered now as an element of cpitch. Such a
group mapping g(i) =​2i maps the pitch function is called an inclusion mapping. An inclu-
A4 to the frequency 440 Hz and assigns sion mapping is the same thing as an identity
to every other pitch its frequency in an mapping except that it is understood to map
equal-​tempered A-​440 tuning. To say that into a set larger than the domain; we write f =​
f is interval-​compatible with g means, for Epitch–​cpitch to indicate the inclusion mapping
example, that raising a pitch (x) by one octave
from pitch to cpitch. Then f is an interval-​space
raises the corresponding frequency (f(x)) by homomorphism. It is interval-​compatible with
a factor of 2 (=​g(1)). Calculate f(A5), f(E5),
the inclusion mapping g =​ Eℤ–​ℝ from the interval
and f(C4), making sure that you understand group ℤ into ℝ; this compatibility means simply
how the results follow from the way g uv is that the interval between two notes in discrete
defined.10 pitch space remains the same if they are consid-
(b) The reference points A4 and 440 are ered to inhabit continuous pitch space instead.
conventional but arbitrary. In principle, an The function f is therefore interval-​preserving,
isomorphism between pitch and freq may but this statement broadens our understand-
be constructed using any reference points ing of interval-​preserving mappings: for such a
at all. Suppose we choose u =​C4 in pitch mapping to exist, it is not strictly necessary that
and v =​1 in freq, while still using the same the groups G1 and G2 be equal, but it is neces-
interval functions intpitch′ and intfreq and sary that G1 at least be a subgroup of G2. In this
the same group mapping g(i) =​2i to define case the intervals of the form int2(f(x), f(y)) are
f = g uv . This choice of reference points always integers, but the group G2 =​ ℝ also con-
assigns to middle C one “frequency unit”—​a tains non-​integer intervals that do not arise as
different unit from the usual Hz. In this intervals in this way.
formulation, calculate f(C3), f(G4), The above function f is not an isomorphism,
and f(A4). as it does not map pitch onto cpitch, so it does
(c) How would the answers to part (b) be not have an inverse function. It is reasonable to
affected if we measured pitch intervals in ask whether a homomorphism can be defined
semitones rather than octaves—​that is, if in the other direction, from cpitch to pitch.
we used the usual interval function intpitch We might try, for example, to define f: cpitch
  rather than intpitch′? → pitch by rounding down: for every pitch x in
cpitch, f(x) is the highest note in pitch that
Exercise 7.2.9 Our construction of a is no higher than x (so, for example, all notes
homomorphism from pitch to gpitch started between C4 and C♯4 are mapped to C4, as is C4
with the group homomorphism g =​ Eℤ, the itself). This f is a valid function—​rounding func-
identity mapping on ℤ. Suppose we instead tions of this sort will find applications in our
choose the inverse mapping g(n) =​−n. Is g: ℤ → study of maximally even sets in Chapter 14—​
ℤ a group homomorphism? Use the reference but it is not an interval-​space homomorphism,
points u =​C4 in pitch and v =​C4 in gpitch to because it cannot be interval-​compatible with
define the induced homomorphism f = g uv from any mapping g: ℝ → ℤ. Observe that intcpitch(C4,
pitch to gpitch. Calculate f(G5), f(C5), and C4+​0.5) and intcpitch(C4+​0.5, C♯4) are both equal
♭3). Is f an isomorphism?
f(E   11 to 1 2 . Interval compatibility would therefore
require that g( 1 2 ) be equal to both intpitch(f(C4),
f(C4+​0.5)) and intpitch(f(C4+​0.5), f(C♯4)). But by
In the following discussion and exercises, we our definition of f, the first of these is equal to
present several additional examples of interval-​ intpitch(C4, C4), which is 0, while the second is
space homomorphisms, along with a few nega- intpitch(C4, C♯4), which is 1. Clearly g( 1 2 ) can-
tive examples—​mappings f: S1 → S2 that do not not be both 0 and 1; therefore no such g can
satisfy the requirements for a homomorphism. exist.12 This example suggests a general way

Intervals • 259
of showing that a function f: S1 → S2 is not which h is interval-​compatible is not the
an interval-​ space homomorphism (useful in identity mapping on ℤ12; what is it?
Exercise 7.2.12 below): if elements w, x, y, and (c) The functions f and h may be regarded as
z of S1 can be found so that int1(w, x) =​int1(y, z) mappings from the vertices of Figure 1.2.3
but int2(f(w), f(x)) ≠ int2(f(y), f(z)), then f cannot (chromatic pitch-​class space) to the vertices
possibly be interval-​compatible with any map- of Figure 1.3.1 (the circle of fifths). Only
ping g: G1 → G2. one of these two functions is a graph
In Section 5.4 we noted a likeness between isomorphism as defined in Chapter 3.
pc and fifth, enharmonically conformed fifths   Which is it?
space. Both spaces contain the same twelve pitch
classes, and both have ℤ12 as their interval group, In Section 1.3 we observed that the line of
but the interval functions are different: intfifth fifths (spelled pitch-​ class space spc) may be
measures intervals around the circle of fifths. wrapped into the circle of fifths in much the
Define f: pc → fifth so that f maps every pitch same way that pitch is wrapped into pc. We can
class in the chromatic circle (Figure 1.2.3) to the see now that isomorphic interval-​space struc-
pc that occupies the corresponding position in tures underlie both of these constructions. In
the circle of fifths (Figure 1.3.1); thus f(C) =​C, Section 5.4 we described an interval-​space struc-
f(C♯) =​G, f(D) =​D, and so on. As a mapping of ture on spc, with intervals measured along the
mod-​12 pitch-​class numbers, f is the multiplica- line of fifths. The interval group of spc is ℤ, so
tion operator M7, but the domain and range of as an interval space, spc is isomorphic to pitch.
the function are two different interval spaces, pc In Section 2.6 we defined a function π: spc → pc,
and fifth. Then f is an interval-​space isomor- which maps a spelled pitch class (for example, E♯)
phism; in fact, f is interval-​preserving. That may to its projection in pitch-​class space (pitch class
seem counterintuitive since intervals are obvi- 5, or F); π is the quotient map QE associated
ously measured differently in the two spaces, with the enharmonic equivalence relation on
but it reflects the way f is defined; for instance, spc. Although we described π as a mapping from
intfifth(f(C), f(C♯)) =​intfifth(C, G) =​1 =​intpc(C, C♯). spc to pc, we may also consider it as a mapping
As noted previously, the interval-​preserving pro­ from spc to fifth, the circle of fifths; this is the
perty means that f is interval-​compatible with wrapping of spc into fifth mentioned above.
the identity mapping on the interval group ℤ12. Moreover, π: spc → fifth is an interval-​space
homomorphism, compatible with the same
Exercise 7.2.10 Since the spaces pc and fifth group homomorphism Q12: ℤ → ℤ12 with which
are identical as sets, there is another obvious the interval-​space homomorphism QO: pitch →
pc is compatible. For example, the spc numbers
of B♭ and E♯ are −4 and 9 respectively, so the spc
mapping h: pc → fifth, namely the identity
mapping, which maps every pitch class in pc
to the same pitch class in fifth. Whereas the interval between them is 13. Their projections in
pc or fifth are the pitch classes π(B♭) =​10 and
π(E♯) =​5, and the interval from 10 to 5 in fifths
function f defined above satisfies f(C) =​C and
f(C♯) =​G, h satisfies h(C) =​C and h(C♯) =​C♯.
Sorting out the properties of the two functions f space is 1. The chain of equalities
and h is a good exercise in understanding various Q12(intspc(B♭, E♯)) =​ Q12(13) =​1
kinds of homomorphisms and isomorphisms. =​ intfifth(B♭, F) =​intfifth(π(B♭), π(E♯))
(a) The statements f(C) =​C and h(C) =​C are confirms the interval compatibility of π and Q12.
both true. According to the interval-​space
homomorphism theorem, only one function Exercise 7.2.11 Let f1: pitch → spc be the
from pc to fifth mapping C to C can be isomorphism that maps the pitch C4 to the
interval-​compatible with the identity mapping spelled pitch class C, induced by the identity
on ℤ12, and hence only one such function can mapping on the group ℤ, as described in the
be interval-​preserving. We noted above that interval-​space homomorphism theorem. This
f is interval-​preserving, so it follows that h means that f1 maps the pitch a certain number
cannot be. Give an example of the failure of of semitones above or below C4 to the spelled
the function h to preserve intervals.13 pitch class the same number of positions to
(b) Although h is not interval-​preserving, it is the right or left of C on the line of fifths; for
an interval-​space isomorphism from pc to example, f1(E♭4) =​A and f1(B4) =​E♯. Also let
fifth. The group automorphism of ℤ12 with f2: pc → fifth be the isomorphism called f in

260 • E x p loring M usical S p aces

Exercise 7.2.10 and the preceding discussion, interval-​compatible. Explain why the other two
mapping C to C and induced by the identity functions are not homomorphisms. Which of
mapping on ℤ12. the homomorphisms are isomorphisms?
Make a diagram showing the four spaces
pitch, spc, pc, and fifth, the isomorphisms (a) f: pc → pc is the transposition operator T3.
f1: pitch → spc and f2: pc → fifth, and the (b) f: cpitch → cpitch is the inversion
homomorphisms QO: pitch → pc and π: spc → operator I (inversion about C4).
fifth. Choose several elements of pitch and (c) f: pitch → cpitch is defined so that for
trace the elements to which they are mapped in every pitch x in discrete pitch space, f(x) is
the  other spaces. Is it true that QO ● f2 =​ f1 ● π? the pitch one quarter-​tone higher than x;
for example f(C4) =​C4+​0.5.
The noncommutative interval space triad (d) f: spc → pc maps any spelled pitch class
with the dihedral interval group 𝒯ℐ12 offers to the corresponding pitch class assuming
additional examples of interval-​ space homo- enharmonic equivalence. This f is same
morphisms. Let P: triad → triad denote the as the mapping π: spc → fifth described
neo-​Riemannian parallel transformation, which above and in Exercise 7.2.11, except
changes the mode of any triad while preserv- that for any spelled pitch class x, f(x) is
ing its root. For any triads X and Y, the interval considered to be an element of the space pc
inttriad(X, Y) (the Tn or In operation that maps X rather than an element of fifth.
to Y) is the same as inttriad(P(X), P(Y)). (Readers (e) f: pc → triad maps any pitch class x to the
should be able to convince themselves of the truth x-​m inor triad.
of this assertion by checking a few examples. It is (f) f: triad → pc maps any major or minor
a consequence of a principle to be established in triad to its root.
Chapter 8: Riemannian triadic transformations (g) f: triad → triad maps any triad, major
always commute with transpositions and inver- or minor, to the major triad with the
sions.) In other words, P is an interval-​preserving same root.
transformation on the space of triads, interval-​ (h) f is the neo-​Riemannian relative
compatible with the identity mapping on the   transformation R: triad → triad.
interval group 𝒯ℐ12, and therefore P is an auto-
morphism of the interval space triad. Theorem 7.2.13 below, the final theorem of
Now consider the function f: pc → triad this section, collects several properties of interval-​
defined so that for every pitch class x, f(x) is the space homomorphisms, especially induced homo-
v
major triad (the major triad whose root is morphisms g u as specified in the homomorphism
x-​
x). This f is an interval-​space homomorphism. theorem. The notation is sometimes cumber-
The compatible group mapping g: ℤ12 → 𝒯ℐ12 is some, but the concepts are straightforward. Some
defined so that, for any mod-​12 integer i, g(i) is motivation for these properties is given in the
the transposition Ti, an element of 𝒯ℐ12. Interval examples and exercises following and in Figure
compatibility of f and g amounts to a fairly obvi- 7.2.14. We will encounter further illustrations of
ous statement: if x and y are pitch classes with these properties in our study of generalized trans-
int(x, y) =​ i, then the transposition operator Ti position and inversion in Sections 7.5 and 7.6.
14

maps the x-​major triad to the y-​major triad (and

Ti is therefore the “interval” between those two Theorem 7.2.13: Properties of interval-​space
triads in the space triad). This homomorphism homomorphisms
is similar to an inclusion mapping: in this case the In the following statements, (S1, G1, int1),
interval group ℤ12 is isomorphic to a subgroup of (S 2, G2, int2), and (S3, G3, int3) are assumed to be
𝒯ℐ12 (the group 𝒯12 of transposition operators interval spaces; ⊕ is the group operation in G2;
only), and the space pc is likewise isomorphic, as and ⊖i denotes the inverse of the group element
interval spaces, to a subspace of triad (the sub- i in G2. The letters u, x, a, and b always denote
space containing major triads only). elements of S1; v and y are elements of S2; w is
an element of S3.
Exercise 7.2.12 Six of the eight mappings
described below are interval-​space (a) Composition of interval-​space homomorphisms.
homomorphisms; two are not. Identify the Suppose f1: S1 → S2 and f2: S2 → S3
homomorphisms and describe the group are interval-​space homomorphisms
homomorphism with which each of them is interval-​compatible with the group

Intervals • 261
FIGURE 7.2.14 Properties of interval-​space homomorphisms, from Theorem 7.2.13

homomorphisms g1: G1 → G2 and g2: (d) Intervals formed by induced interval-​space

G2 → G3 respectively. Then f1 ● f2: S1 → S3 homomorphisms. Suppose g: G1 → G2 and
is an interval-​space homomorphism h: G1 → G2 are group homomorphisms. Then
interval-​compatible with the group for every a and b in S1,
homomorphism g1 ● g2: G1 → G3. If f1 = g 1uv
and f2 = g 2 wv , then f1 • f2 = ( g1 ● g2)wu. int 2( g uv ( a), h xy(b ))
If both f1 and f2 are interval-​space = g (int 1( a, u)) ⊕ int 2( v, y ) ⊕ h( int 1( x , b )).
isomorphisms, then f1 ● f2 is an (e) Change of reference points. If g: G1 → G2 is a
interval-​space isomorphism. group homomorphism, then g xy = g uv if and
(b) Inverses of interval-​space isomorphisms.   only if g uv ( x ) = y .
Suppose f: S1 → S2 is an interval-​space
isomorphism interval-​compatible with Parts (a)–​(d) of the above theorem are illus-
the group isomorphism g: G1 → G2. trated in the corresponding parts of Figure
Then f −1: S2 → S1 is an interval-​space 7.2.14. Parts (a) and (b) show that compositions
isomorphism compatible with the group and inverses of interval-​space homomorphisms
isomorphism g−1: G2 → G1. If f = g uv, then behave as we should expect. As an example of (a),
f −1 = g–1wu. the mapping from pitch classes to minor triads in
(c) Label functions for induced interval-​space Exercise 7.2.12e may be obtained as the composi-
homomorphisms. Suppose labu: S1 → G1 tion of two other functions described previously,
and labv: S2 → G2 are the label functions f ● P, where f: pc → triad maps each pitch class to
associated with u and v, and suppose g: a major triad and P: triad → triad is the parallel
G1 → G2 is a group homomorphism. Then transformation. Because both f and P are interval-​
for every a in S1, space homomorphisms, the same is true of f ● P.
The group mapping interval-​compatible with f ● P
labv( g xy( a)) is g1 ● g2, the composition of the group mappings
= labv ( y ) ⊕ (⊖g ( labu(x ))) ⊕ g ( lab u(a)). associated with f and P, but g2 is the identity map
on 𝒯ℐ12, so the group mapping for f ● P is the same

262 • E x p loring M usical S p aces

one described previously for f, which maps i in ℤ12 two homomorphisms then allows j2 and j4 to be
to Ti in 𝒯ℐ12. replaced by g(i2) and h(i4), leading directly to the
In part (b) of the theorem it is necessary to formula in the theorem.
assume that f is an isomorphism because inverse Theorem 7.2.13e makes the point that the
functions are defined only for one-​to-​one, onto induced homomorphism notation g uv is not
mappings. The isomorphism f: pitch → gpitch unique: the reference points u and v may be
described above, for example, has an inverse replaced by any other pair of points related by
f −1: gpitch → pitch, which is also an isomor- the same mapping. For example, earlier we con-
phism. Both f and f −1 are interval-​compatible structed an isomorphism from pitch to gpitch
with the identity map Eℤ and both map C4 to C4, as g uv, where u and v were both C4. This function
so both could be denoted Eℤ CC44 ; at the risk of mapped G4 (in pitch) to C5 (in gpitch), so the
making the notation more unwieldy, the space same homomorphism could also be denoted g xy ,
names may be included in the subscript and where x =​G4 and y =​C5.
superscript if it is necessary to distinguish the
two. In the final statement of part (b), the posi- Exercise 7.2.15 Continuing your work
tions of u and v are exchanged in the inverse from Exercise 7.2.8, verify that the formula
function: f maps u to v, so f −1 maps v to u. in Theorem 7.2.13c holds in the following
Part (c) gives a very general formula for calcu- situation: S1 is the space cpitch; S2 is the space
lating the label of any point in S2 arising as the freq; g is the homomorphism from (ℝ, +​)
image of a point a in S1 under an induced interval-​ to (ℝ>0, ∙) given by g(x) =​2x; u, x, and a are
space homomorphism g xy, in terms of the labels respectively the pitches C4, A4, and E5; and v and
of a, x, and y, and reference points u and v in y are respectively 1 and 440. In pitch space use
the two spaces. If the interval group G2 is addi- the interval function intpitch′ =​intpitch ∕ 12, as
tive, the formula takes the slightly simpler form in  
the previous exercise.15
labv ( g xy( a)) = labv ( y ) − g ( labu( x )) + g ( labu( a)). In a
commutative group, the terms may be rear- Exercise 7.2.16 Verify that the formula
ranged if desired; the formula in the theorem in Theorem 7.2.13d holds in the following
gives the proper order of factors in the general, situation: S1 is the space pc; S2 is the space
possibly noncommutative, case. Figure 7.2.14c triad; g and h are both the homomorphism from
ℤ12 to 𝒯ℐ12 that maps the mod-​12 integer i to
provides some motivation for this formula,
using additive notation. The label in question, the transposition Ti; u, x, a, and b are respectively
labv ( g xy( a)), is equal to int 2( v, g xy( a)) , the inter- the pitch classes C, C, E, and G; and v and y are
val j1 in the figure. The interval-​sum equation respectively
   the C-​major and C-​minor triads.16
in S2 implies that j1 =​ j2 +​ j3. The interval com-
patibility of g xy with g means that j3 =​ g(i3). The Exercise 7.2.17 If x =​ u, y =​ v, and g =​ h, show
interval-​sum equation in S1 implies that i1 =​ i2 +​ that Theorem 7.2.13d reduces to the interval
i3, or equivalently i3 =​−i2 +​ i1; because g is a group compatibility
   condition.
homomorphism it follows that g(i3) =​−g(i2) +​
g(i1). Therefore j1 =​ j2 − g(i2) +​ g(i1); the assertion 7.3 DIRECT PRODUCTS OF
in the theorem then follows by expressing all of
these intervals as equivalent label functions. In
INTERVAL SPACES
the special case in which x =​ u and y =​ v, labu(x) The concepts of direct products and quotients
and labv(y) are the identity elements of their of groups, described in Chapter 6, have coun-
respective groups, so in this case the formula terparts in the theory of interval spaces. In
reduces to labv( g uv( a)) = g ( labu( a)), which simply fact, we have seen examples of both of these
restates the label compatibility property. constructions already. Product spaces are for-
Theorem 7.2.13d is another general formula, malized in this section, quotient spaces in
this one for the interval between two points of the next.
S2 arising as images of points a and b in S1 under Suppose (S1, G1, int1) and (S2, G2, int2) are two
two (possibly different) induced interval-​space interval spaces. The direct product of S1 and S2 is
homomorphisms g uv and h xy, in terms of various the interval space (S, G, int) defined as follows.
intervals formed by a, b, u, v, x, and y. In Figure The space S is S1 × S2, the Cartesian product of
7.2.14d, the interval-​sum equation in S2 implies the sets S1 and S2—​that is, the set of ordered
that j1 =​ j2 +​ j3 +​ j4; interval compatibility for the pairs (x, y) with x in S1 and y in S2. Similarly, the

Intervals • 263
group G is G1 × G2, the product of the two original ℤ × ℤ, or ℤ2, and ℝ × ℝ, or ℝ2, respectively. More
groups, with the direct-​product group structure generally, the n-​voice pitch-​tuple spaces pitchn
described in Section 6.4. The interval function and cpitchn have interval groups ℤn and ℝn.
int on S is defined componentwise. Specifically, Pitch-​class–​tuple spaces pcn and cpcn, with inter-
if (x1, x2) and (y1, y2) are elements of S, then x1 val groups (ℤ12)n and (ℝ12)n respectively, may
and y1 belong to S1 while x2 and y2 belong to S2, also be constructed.
so we can form the intervals i1 =​int1(x1, y1) in G1 Intervals in pitch-​tuple space, as elements of
and i2 =​int2(x2, y2) in G2; the interval int((x1, x2), ℤn or ℝn, are themselves n-​dimensional objects of
(y1, y2)) in the product space is by definition the the form (i1, i2, …, in). As an example of an inter-
ordered pair (i1, i2), an element of G1 × G2. val in pitch3, consider the chords α =​(G3, B3,
To be sure that S is an interval space, one F4) and β =​(C3, C4, E4), an incomplete dominant
must verify the homogeneity condition and the seventh chord and an incomplete tonic triad. (In
interval-​ sum equation from Section 5.4—​ but our study of voice-​leading spaces we will often
both are immediate consequences of the corre- use Greek letters to stand for pitch tuples.) Then
sponding properties of S1 and S2. For homogene- the interval int(α, β) =​(−7, 1, −1) tells us pre-
ity, if (x1, x2) ∈ S and (i1, i2) ∈ G, then the unique cisely how far, and in which direction, each of
element of S that lies at the interval (i1, i2) from the three voices moves as chord α progresses to
(x1, x2) is (y1, y2), where y1 is the unique ele- chord β. This interval function does not report
ment of S1 such that int1(x1, y1) =​ i1 and y2 is the a single number as the “distance” between the
unique element of S2 such that int2(x2, y2) =​ i2. two chords. In Chapter 12 we will introduce dis-
The interval-​sum equation in S takes the form tance functions that always take a single numeri-
cal value. In the context of voice-​leading spaces,
int((x1, x2), (y1, y2)) +​int((y1, y2), (z1, z2)) distance spaces, defined by distance functions,
  =​ int((x1, x2), (z1, z2)), will prove to be more broadly applicable than
interval spaces—​but the basic pitch-​tuple spaces
which follows directly from the direct-​product pitchn and cpitchn possess both an interval-​
structure of the group and the interval-​ sum space structure and a distance-​space structure.
equations in each component. The +​sign in the As the example of cpitchn shows, a direct-​
above equation represents the group operation product interval space may be used as a way of
in the product group G1 × G2, defined compo- combining information from multiple objects
nentwise (as in Chapter 6) using the group oper- in a single entity, such as several notes in one
ation of G1 in the first component and that of G2 chord or voices in a texture. In other situa-
in the second. tions, direct products may be used to capture
We noted in Section 2.2 that Cartesian prod- two or more kinds of information about a single
ucts of more than two sets are often useful; for object. Consider pitch × beat, the product of
example, chords in a four-​ voice texture may discrete pitch space with beat space. Beat space,
be modeled as ordered 4-​tuples of pitches, ele- introduced in Example 1.4.20 as an infinite
ments of the space pitch4. Our work with voice-​ chain of equally spaced beats, forms an interval
leading spaces in Part Three will make extensive space in which, for example, an equation like
use of product spaces of this form, but in the intbeat(x, y) =​3 means that beat y occurs 3 beats
continuous version: cpitchn, the direct product after x. For present purposes we assume a dis-
of n copies of continuous pitch space cpitch. crete beat space measured in quarter notes, with
The elements of cpitchn are ordered n-​tuples of interval group ℤ. This is the same interval group
pitches; each pitch may be represented by a real as that of discrete pitch space pitch, so from
number, so the elements of cpitchn correspond Section 7.2 it follows that the interval spaces
with points in n-​dimensional Euclidean space pitch and beat are isomorphic—​but the musi-
ℝn. Spaces of the form pitchn or cpitchn may cal information they convey is very different.
be called pitch-​tuple spaces. As we know, pitch An element of pitch × beat is a specific pitch
and cpitch are interval spaces with interval at a specific time, the two types of information
groups ℤ and ℝ respectively. The discrete and conveyed in a straightforward way by an ordered
continuous pitch-​pair spaces pitch2 and cpitch2 pair of integers.
(that is, the direct products pitch × pitch and In the canon from the closing-​theme group
cpitch × cpitch), which could be used to model in the first movement of Beethoven’s Fourth
sonorities in a two-​voice texture, therefore have Symphony (Figure 7.3.1), the bassoon imitates
interval-​space structure with the interval groups the clarinet one octave lower (pitch interval

264 • E x p loring M usical S p aces

−12) and one measure later (beat interval 4). of scale degrees presupposes a tonal center, and
Corresponding notes in the two voices are there- Rings comments extensively on “tonal qualia,”
fore consistently related by the interval (−12, 4) in perceptual attributes associated with events in
the product space. We noted in Section 5.6 that tonal contexts. The following example offers a
a kind of “transposition” Transi makes sense in few simple observations in this space, in a com-
any interval space, the statement “Transi(x) =​ y” parison of several occurrences of the tritone
being synonymous with “int(x, y) =​ i.” The bas- C♯–​G in Debussy’s “Afternoon of a Faun.”
soon part is a standard T−12 transposition of the
clarinet in pitch space, but it may also be con- Example 7.3.2: Debussy, Prélude à “L’après-​midi
sidered a T4 transposition of the clarinet in beat d’un faune”
space; both pieces of information may be cap- Measure 21 of Debussy’s score appears in
tured by saying that the bassoon transposes the reduction at the top left of Figure 7.3.3. The
clarinet at Trans(−12, 4) in the product space pitch flute’s C♯ sounds as an added sixth in a tonic
× beat. A consistent interval of transposition in triad in E major, so this note is represented by
this product space is characteristic of canonic the ordered pair (6,̂ 1) in the product space—​
and fugal imitation in general.17 pitch class 1 heard as scale degree 6̂. This
The reader can no doubt imagine many vari- ordered pair appears as the top-​left vertex in
ants on the preceding construction. Either pitch the network at the center of the figure. From C♯,
space or beat space could be replaced with its the flute melody descends almost chromatically
continuous counterpart. Chromatic pitch space to G♮, supported by a ♭VI chord over a tonic
could be replaced by a generic or diatonic space pedal. Scale-​degree space does not distinguish
(useful in the study of diatonic sequences) or between 3̂ and ♭3̂, but the ordered pair (3̂, 7)
by pitch-​class space. Lewin ([1987] 2007, 37–​ confirms that the note is G♮ rather than G♯. The
44) offers a detailed analysis of the opening of interval between the C♯ and G in the product
the third movement of Webern’s Variations for space is (4, 6): the second note is four steps
Piano, Op. 27, in the space pc × beat, modular- clockwise from the first in mod-​7 scale-​degree
izing the pitch dimension but retaining the infi- space (in this case a fourth down rather than
nite time dimension. One could also reverse the a fifth up) and six steps clockwise in mod-​12
order of the two components, forming a product pitch-​class space (in this case six semitones
space such as beat × pitch rather than pitch down). Rings sometimes conveys additional
× beat; the two interval spaces are isomorphic musical information via notations that
via the function that maps the ordered pair transcend the formal structure of the space;
(x, y) to (y, x). in this spirit this interval might be written in
Chromatic and generic (or diatonic) informa- the form (−3, −6), or even—​Rings’s preferred
tion may also be usefully combined in a product notation—​(4th−1, −6).
space.18 Steven Rings (2011b) has demonstrated The first harmony in the measure includes
many analytical virtues of the space sd × pc, the a G♯, and the flute’s descending line traverses
product of scale-​degree space (first introduced G♯ just before its arrival at G♮. In this context,
in Figure 1.2.10) with pitch-​ class space. The one might argue that a listener will hear G♮
interval group of this space is ℤ7 × ℤ12; the first not as ♭3̂ at all but as ♯2̂, a chromatic lower
component of an interval is a mod-​7 interval neighbor to 3̂. This hearing is modeled at
between scale degrees, while the second is a mod-​ the bottom left in the figure. The interval
12 interval between pitch classes. Identification spanned by the flute in this reading is not

FIGURE 7.3.1 The pitch × beat interval (−12, 4) in Beethoven, Symphony No. 4 in B♭ Major, Op. 60,
I, mm. 141–​45

Intervals • 265
(4, 6) but (3, 6) (or (5th−1, −6))—​a descending call this interval (1, 0) a pivot 2nd: there is no
fifth rather than a fourth. Intervals between change in pitch classes but the scale degrees
the two hearings appear as vertical arrows have shifted by a step, reflecting the change in
in the network. The two C♯s, (6̂, 1) in both tonal center. The bubble notation works here
hearings, are related by the identity interval because the interval group is commutative. The
(0, 0). The G♮ and F𝄪 are related by the interval bubbles imply two paths through the graph
(6, 0), indicating that while the pitch class is from (6̂, 1) to (4̂, 7): one can traverse the
unchanged (pc interval 0), the scale degree (4, 6) arrow followed by (1, 0), or the other
in the second hearing is one step lower (sd way around, and by the interval-​sum equation
interval 6 mod 7). Intervals between the first the composite transposition is (5, 6) in either
chord’s G♯ and the second chord’s G♮ (F𝄪), case—​the interval in sd × pc from C♯ in E major
though not shown in the graph, could also to G in D major.
be computed for both hearings. In the first In the central D♭-​major episode of the piece,
hearing this interval is (3̂, 8) ( 0, 11 )
→ (3̂, 7), shown at the lower right, the tritone D♭–​G♮
indicating a chromatic inflection of a fixed appears in the bass line as 1̂ and ♯4̂. The interval
scale degree, while in the second it is here is again (3, 6), the descending fifth;20 this
(3̂, 8) (
6, 11 )
→ (2̂, 7), a characteristic lower-​ D♭–​G is related to the descending fifth C♯–​F𝄪
neighbor relationship.19 in the second interpretation of m. 21 by the
Both the descending-​fourth and descending-​ interval (2, 0), a pivot 3rd, as the scale degrees
fifth hearings of the interval between have shifted by two steps. The vertical arrows
pitch classes 1 and 7 are instantiated less on the right side of the graph bear the interval
ambiguously elsewhere in the piece. At m. 11, labels (0, 0) and (1, 0): G♮ is (4̂, 7) in both the
as shown at the upper right, C♯ and G are 7̂ D-​major and the D♭-​major passages, but the scale
and 4̂ in an apparent D-​major tonal context. degree associated with C♯ (D♭) has changed. On
The interval between them is (4, 6), the the left side of the graph, the two C♯s are heard
descending fourth, matching the interval in in the same way while the Gs are interpreted
our first interpretation of m. 21. Though the differently, so the identity interval (0, 0) relates
interval between the two notes may be the only the C♯s; on the right side this situation is
same, the tonal context is different: these two reversed.
  
instances of C♯–​G are related in the product
space by the interval (1, 0). This information is Exercise 7.3.4
conveyed in the graph by bubble notation: in (a) By following arrows in the graph in Figure
the product space, C♯–​G in m. 11 is a Trans(1, 0) 7.3.3, six different paths may be traced from
transposition of C♯–​G in m. 21. Rings would (6̂, 1) at the upper left to (4̂, 7) at the lower

FIGURE 7.3.3 Interpretations of the interval C♯–​G in Debussy, Prélude à “L’Après-​midi d’un faune” in
the product space sd × pc

266 • E x p loring M usical S p aces

 right. Verify that the sums of the intervals the fact that S0 is an interval space follows from
along all six paths are the same. Why must the interval subspace property from Theorem
this be the case?21 5.4.7.) It should come as no surprise that this
(b) If the passage at m. 55 were notated in C♯ seven-​note interval space S0 is isomorphic to sd
major rather than D♭ major, would anything itself, and the embedding ε1 is an interval-​space
in the graph change? isomorphism from sd to S0. These constructions
(c) The last tritone heard in the piece is not are completely general: in any product interval
C♯–​G but E–​A♯, presented harmonically space, projections onto each component and
at m. 109 as 1̂ and ♯4̂ in E major. These embeddings of each of the original spaces into
are the same scale degrees as D♭–​G at isomorphic cross sections of the product space
m. 55, so the final E–​A♯ may be considered can always be defined in this way.
a transposition of m. 55’s D♭–​G. As an
interval in sd × pc, what is the interval of Exercise 7.3.5
  transposition from m. 55 to m. 109? (a) Rewrite the preceding paragraph to show
how to define the projection π2 of sd × pc
We may imagine combining many types of onto its second component and the
information in a multidimensional direct-​product embedding ε2 of pc into the cross section
interval space whose components might include, S1̂ =​{1̂} × pc of sd × pc.
for example, a chromatic pitch number, a generic (b) Carefully describe the group homomorphisms
pitch class, a tonal center, a scale degree, a time of interval groups with which the interval-​
point, and a beat-​class number. Such a space may space homomorphisms π1 and ε1 described
be thought of as a spreadsheet to be filled in with   above are interval-​compatible.
detailed information about various details of,
potentially, every note in a piece. This approach Exercise 6.4.4 described the conformed ton-
can enable a thorough accounting of musical netz of Figure 1.4.3 as an interval space whose
events (and relationships between events, in the interval group is ℤ4 × ℤ3. To move in this space
form of multidimensional intervals)—​but at the by an interval such as (2, 1) means to move two
cost of an overwhelming profusion of data. Even places rightward and one place up. It should
the comparatively simple product spaces of Lewin come as no surprise that this tonnetz is a direct-​
(pc × beat) and Rings (sd × pc) described above product interval space, formed as the product of
can seem complex. More important, while the spaces whose interval groups are ℤ4 and ℤ3—​or
spaces provide a framework for recording infor- at least it is isomorphic to such a product space.
mation, it remains up to the analyst to decide It does not much matter how the elements of
which of the myriad data points are of analytical the two smaller spaces are labeled; the ordered
significance in any given situation.22 pairs in the product space may in any case be
In Rings’s space sd × pc, if for every scale degree relabeled with pitch classes in the usual ton-
s and every pitch class x we define π1(s, x) =​ s, netz arrangement. (The relabeling is the iso-
then π1: sd × pc → sd (the projection mapping morphism.) We know from Section 6.4 that the
from the product onto its first component) is an group ℤ4 × ℤ3 is isomorphic to ℤ12, so it follows
interval-​space homomorphism from the product from the interval-​space homomorphism theo-
space sd × pc to scale-​degree space sd. By fixing rem (Theorem 7.2.7) that this tonnetz space
any pitch class such as 0, we can define a homo- is isomorphic as an interval space to ordinary
morphism in the opposite direction, ε1: sd → pitch-​class space pc.
sd × pc, by ε1(s) =​(s, 0); ε1 is an embedding of sd Other aspects of the relationship between
into the first component of the product space. tonnetzes and product spaces may be high-
The image of sd under this embedding is the set lighted by considering frequency spaces with
S0 =​ sd × {0} =​{(1̂, 0), (2̂, 0), …, (7̂, 0)}, a cross sec- multiplicative interval groups. In Exercise 2.3.25
tion of sd × pc consisting of only those elements we constructed Pythagorean pitch space in the
of the product space having 0 as their second form of a tonnetz generated by octaves and per-
component. This subset of the product space is fect fifths. If we orient this space so that octaves
an interval space in its own right: the interval run horizontally and fifths vertically, then the
function is that of the product space, but the sec- note at coordinates (x, y) has a frequency of
ond component of every interval formed within 2 x ⋅ ( 3 2 ) y ⋅ c Hz, where c is the frequency assigned
the subspace is 0. (Intervals of this type form to whichever note lies at the origin (0, 0). A hori-
a subgroup of the interval group ℤ7 × ℤ12, and zontal cross section of this space is a space of

Intervals • 267
octave-​related pitches such as {. . ., C2, C3, C4, C5, of all notes obtained from a given starting
C6, . . .}, which is itself a simple frequency space pitch via purely tuned octaves, perfect fifths,
whose interval group is { , 1 4 , 1 2 , 1, 2, 4, }, and major thirds. Using reasoning similar
the multiplicative group of powers of 2. A verti- to the above, show that just pitch space may
cal cross section, meanwhile, is a space of purely be constructed as a direct product of three
tuned fifth-​related pitches such as {. . ., B♭2, F3, C4, interval spaces, and that when regarded as
G4, D5, . . .}, an interval space with the multipli- a frequency space its interval group is the
cative interval group { , 4 9 , 2 3 , 1, 3 2 , 9 4 ,  } of multiplicative group Gjust =​{2a · 3b · 5c | a, b,
powers of 3 2 . Pythagorean frequency space pyth c ∈ ℤ}, which is isomorphic to ℤ × ℤ × ℤ, the
may be identified with the product of these two direct product of three copies of the additive
simpler spaces. In this conception, an interval in group
   of integers.24
pyth is technically an ordered pair of the form
(2 x, ( 3 2 ) y ) for some integers x and y, each of 7.4 QUOTIENTS OF
which may be positive, negative, or zero. It is use- INTERVAL SPACES
ful, however, to replace this ordered pair with the
product 2 x ⋅ ( 3 2 ) y, which is the exact frequency The frequency-​based tonnetz spaces described at
ratio between two notes in this relationship. the end of Section 7.3 are pitch rather than pitch-​
Because 2 x ⋅ ( 3 2 ) = 2 x − y ⋅ 3 y, intervals in this
y class spaces. To form pitch-​class spaces entails
group can always be written in the simple form removing the octave component (2a) from con-
2a · 3b for some integers a and b. The set Gpyth =​ sideration. The appropriate means for doing so is
{2a · 3b | a, b ∈ ℤ} is a multiplicative group—​the a quotient interval space, the topic of this section.
interval group of Pythagorean frequency space While product spaces combine information from
pyth. This group is a subgroup of the multiplica- two or more spaces in a single larger space, quo-
tive group (ℝ>0, ∙), the interval group of freq, so tient spaces reduce a given space to something
pyth’s status as an interval space is confirmed smaller by identifying some points with others,
by the interval subspace property. effectively discarding the information by which
we can tell those points apart. We have previ-
Exercise 7.3.6 After the previous discussion of ously described quotient sets (Chapter 2) and
tonnetzes as interval spaces, we might expect quotient groups (Chapter 6), formed by imposing
that intervals in the infinite two-​dimensional equivalence relations on sets or groups; we now
tonnetz of Pythagorean pitch space should be show how the quotient concept may be extended
depicted as ordered pairs of integers. In fact, to interval spaces under appropriate conditions.
ℤ × ℤ, the additive group of ordered pairs of Throughout this discussion it will be help-
integers, is isomorphic to the multiplicative ful to recall a simple and familiar example that
interval group Gpyth described above. One we have described from a number of different
possible isomorphism maps the ordered pair perspectives already: the construction of pitch-​
(a, b) to the number 2a · 3b. Verify the following class space from pitch space. In Section 1.2 we
properties of this mapping f, both of which obtained pc from pitch informally by wrapping
must be true for it to be an isomorphism: a line into a circle; in Section 2.5 we formalized
this construction via an equivalence relation;
(a) The mapping f must satisfy the homo­ and in Section 6.6 we formed the interval group
morphism property f(x) ⊗ f(y) =​ f(x ⊕ y). ℤ12 as a quotient group of ℤ. The discarded infor-
Replace x and y by ordered pairs (a, b) and mation that distinguishes pitches but not pitch
(c, d) and the operation symbols ⊕ and ⊗ classes is register information, present in the
by the appropriate operations in the two space pitch but not in pc.
groups, and verify that this equation holds. We know from Section 6.6 that a quotient
(b) The mapping f is evidently onto by group G ∕ H may be formed as long as H is a nor-
the definition of Gpyth, but to be an mal subgroup of G. The construction of a quo-
isomorphism it must also be one-​to-​one, tient interval space correspondingly requires a
which means that the same number cannot normal subgroup H of the interval group (G, ⊕).
be 2a · 3b for two different choices of (a, b). The subgroup H defines a group congruence ~H
  Why is this true?23 on G; two elements of G are related by ~H if they
belong to the same coset of H, which also means
Exercise 7.3.7 Review the construction of just that one element can be obtained from the other
pitch space in Exercise 2.3.26: this is the space by adding (using the group operation ⊕, on

268 • E x p loring M usical S p aces

either the left or the right) some element of H. ~H int(u, x), then int(u, y) =​int(u, x) ⊕ h for some
The elements of the quotient group G′ =​ G ∕ H are h in H. But the interval-​sum equation int(u, y) =​
the cosets of H, which are also the equivalence int(u, x) ⊕ int(x, y) holds as well, from which it
classes of the congruence ~H. In this situation H follows that int(x, y) =​ h ∈ H and therefore that
may be called the congruence subgroup, and the x ≈H y. Thus the equivalence relation ≈H on the
equivalence classes may also be called congruence space S corresponds exactly, via the label func-
classes. The requirement that H be a normal sub- tion labu, to the congruence ~H on the group G.
group is a limitation only in the noncommuta- Equivalence relations always enable the for-
tive case; many of the most important interval mation of quotient sets. Let S′ be the quotient
spaces have commutative interval groups, which set S ∕ ≈H, the set of equivalence classes of the
ensures that every subgroup is normal. normal equivalence relation ≈H. To complete the
The congruence ~H is defined on the group G, construction of the quotient interval space, we
but the subgroup H also provides a way to define must show how to define an interval function int′
an equivalence relation on the space S, a rela- on S′. If x and y are elements of S whose equiva-
tion we will denote ≈H. Two elements of S are H-​ lence classes in S′ are x and y , define int′( x , y )
equivalent if the interval between them belongs to be the equivalence class in the quotient group
to H: x ≈H y is true whenever int(x, y) ∈ H, or G′ of the interval int(x, y) in G. In the familiar
equivalently whenever int(x, y) ~H e. We say that example, this definition says that pitch-​ class
≈H is the equivalence relation on S induced by the intervals are obtained by reducing pitch inter-
normal subgroup H of G.25 We will use the term vals mod 12. To be sure that int′ is a well-​defined
normal equivalence relation for any equivalence interval function on S′, we must check that if a
relation on an interval space that is induced by a second pair of elements x1 and y1 were chosen as
normal subgroup of the interval group. representatives of the same equivalence classes
In the familiar example, S is pitch; the inter- x and y , then int(x1, y1) would belong to the
val group G is ℤ; H is the subgroup 12ℤ; ~H is same equivalence class as int(x, y), and so the
the relation of congruence mod 12; and the quo- definition of int′( x , y ) would be unaffected. To
tient group G′ is ℤ12, the integers mod 12. The see why this is true, observe that
equivalence relation induced by the subgroup
12ℤ is octave equivalence—​that is, the rela- int(x1, y1) =​int(x1, x) ⊕ int(x, y) ⊕ int(y, y1).
tion ≈12ℤ on pitch is just the familiar relation
~O—​because two pitches are octave-​equivalent Because x1 ≈H x and y1 ≈H y, the first and last
if and only if the interval between them belongs intervals in this sum both belong to H, so the
to 12ℤ. As an equivalence relation induced by entire sum is H-​congruent to int(x, y).
a normal subgroup, ~O is a normal equivalence Therefore (S′, G′, int′) is an interval space, the
relation on pitch. quotient interval space of S induced by the subgroup
As the familiar example demonstrates, the H. Moreover, there is always a quotient interval-​
congruence ~H on G (mod-​ 12 congruence of space homomorphism from the original space to
integers) and the equivalence ≈H on S (octave the quotient space, namely the quotient map
equivalence of pitches) are notably similar in Q≈H: S → S′. In the familiar example this is the
structure, because the structure of the space mapping QO: pitch → pc, discussed in Section
corresponds so closely to the structure of the 7.2. The definition of int′ ensures that Q≈H is
group. This space-​ group correspondence was interval-​
compatible with the quotient group
described in Section 7.1 via label functions. If u homomorphism Q~: G → G′; this compatibility
is a reference point and x and y are any elements with a group homomorphism ensures that Q≈H is
of S, we can now see that x and y are H-​equiv- an interval-​space homomorphism.
alent elements of the space (x ≈H y) if and only The familiar example immediately sug-
if their labels labu(x) =​int(u, x) and labu(y) =​ gests other examples, because we know sev-
int(u, y) are H-​congruent elements of the group eral other interval spaces sharing the interval
(labu(x) ~H labu(y)). To see this, note first that by group ℤ and therefore isomorphic to pitch. In
the interval-​sum equation, int(u, x) ⊕ int(x, y) =​ generic and diatonic pitch spaces gpitch and
int(u, y). If we know that x ≈H y, that means that dpitch(n), notes are octave-​ equivalent if the
int(x, y) ∈ H—​but then int(u, x) ⊕ h =​int(u, y) for interval between them is a multiple of 7, so
some h in H, which implies (because H is a nor- ~O is induced by the subgroup 7ℤ of ℤ (rather
mal subgroup) that int(u, x) and int(u, y) are H-​ than 12ℤ), giving rise to gpc and dpc(n) as quo-
congruent. The converse is true as well: if int(u, y) tient interval spaces. In spc, the subgroup 12ℤ

Intervals • 269
induces enharmonic equivalence; the quotient this requirement is transpositional equivalence
interval space is conformed fifths space fifth. ~T. In pitch3, two chords are transpositionally
Beat space beat also has ℤ as its interval group; equivalent if they are related by transposition Ti
any subgroup of the form nℤ gives rise to mod-​n for some integer i; the interval of transposition
beat-​class space bc(n) as a quotient space. must be the same in all three coordinates, which
Octave equivalence is a normal equivalence means that the interval between the chords is an
relation and determines a quotient interval space ordered triple of the form (i, i, i), a multiple of
not only in the discrete case (the formation of pc (1, 1, 1). The congruence subgroup for this rela-
from pitch) but also in the continuous case (cpc tion is the subgroup of ℤ3 consisting of all such
from cpitch), and not only in one dimension but triples (i, i, i). In the continuous space cpitch3,
in multiple dimensions (pcn from pitchn, and analogously, the congruence subgroup for ~T
cpcn from cpitchn). Taking pitch3 as an exam- consists of ordered triples (r, r, r), where r may
ple, the interval group is ℤ3, and the congruence be any real number. When we study the geom-
subgroup H =​(12ℤ)3 consists of those ordered etry of the OPTIC spaces in Chapter 11, we will
triples all of whose components are multiples of see that although cpitch3 is a three-​dimensional
12. The induced relation ≈H on pitch3 is precisely space, the quotient space cpitch3∕~T (three-​voice
octave equivalence, which we have previously T-​space) is only two-​ dimensional; it may be
denoted ~O: two ordered triples in pitch3 are obtained by flattening the three dimensions of
octave-​equivalent if the three-​dimensional inter- cpitch3 into a plane in a certain orientation.
val between them belongs to H, which is to say In discrete three-​voice T-​space pitch3∕~T, an
that the notes in corresponding components are interval intT(α, β) is an element of the quotient
octave-​equivalent. The quotient space pitch3∕~O group ℤ3 ∕ K, where K is the congruence subgroup
is the space pc3, three-​dimensional pitch-​class–​ {(i, i, i) | i ∈ ℤ}. As such, an interval is not an
triple space, consisting of ordered triples of pitch ordered triple of integers but an equivalence
classes. Because this space arises through octave class of ordered triples, a coset of K. If (x, y,
equivalence, in our study of general voice-​lead- z) belongs to a certain coset, then every triple
ing spaces in Part Three we will call it three-​voice of the form (x +​ i, y +​ i, z +​ i) belongs to that
O-​space. As an interval space, its interval group same coset. In Chapter 11 we will explore vari-
is the quotient group ℤ3 ∕ (12ℤ)3, which may be ous ways to describe these equivalence classes;
identified with (ℤ12)3, the direct product of three in one representation (which we will call the yz
copies of the integers mod 12. form), the three components are adjusted by
If α and β are chords written as ordered triples the same amount so that the first component
in pitch3, the interval between them in the quo- becomes 0. In the above example for which
tient space may be calculated by forming int(α, β) int(α, β) =​(−7, 1, −1) in pitch3, the interval
in pitch3 and reducing all three components mod intT(α, β) in T-​space is the equivalence class
12. In the general notation employed above, this of (−7, 1, −1); the yz form in this class is (0, 8,
–
interval would be denoted intʹ(α–, β ), but in this 6), obtained by adding 7 to each of the original
situation we might more conveniently write coordinates. This interval may be interpreted as
intO(α, β), understanding that the notation intO showing not precisely how the individual voices
refers to an interval between equivalence classes move from chord α to β but how they move rela-
in O-​space. For example, we noted in Section 7.3 tive to each other. If the three voices move from
that the incomplete dominant seventh chord α =​ α =​(G3, B3, F4) through intervals of 0, 8, and 6
(G3, B3, F4) and incomplete tonic triad β =​(C3, C4, semitones, the resulting chord (G3, G4, B4) is not
E4) are related in pitch3 by the interval int(α, β) =​ the same as β =​(C3, C4, E4) but is transposition-
(−7, 1, −1), whose components are integers; in ally equivalent to β.
O-​space they are related by intO(α, β) =​(5, 1, 11),
whose components are integers mod 12. Exercise 7.4.1 Two pitch triples in pitch3 are
Octave equivalence is one of the OPTIC rela- OT-​equivalent if they are related by O and T
tions introduced in Section 2.4, all of which will equivalence in any combination. For instance,
be discussed in more detail in Chapter 10 in the (G3, B3, F4) and (C4, E5, B♭3) are OT-​equivalent
context of the continuous spaces cpitchn. As because (G3, B3, F4) ~O (G4, B5, F4) ~T (C4, E5,
we will see in Chapter 12, not all of the OPTIC B♭3). Three-​voice OT-​space pitch3∕~OT may be
relations are normal equivalence relations and constructed as a quotient interval space from
give rise to quotient interval spaces; the only pitch3. Describe the congruence subgroup of ℤ3
relation besides octave equivalence that meets associated
   with OT-​equivalence.26

270 • E x p loring M usical S p aces

 We return now to the construction of quo- a congruence class always consists of a single
tients of continuous frequency space freq, power of 3 and its multiples by all possible pow-
mentioned at the beginning of this section. As ers of 2; for example, one congruence class is
we know from Section 7.2, the interval group { , 9 8 , 9 4 , 9 2 , 9, 18, 36, }, consisting of 32 =​9
of freq is the multiplicative group (ℝ>0, ∙), and all its power-​of-​2 multiples. We may take
which is isomorphic to the additive group 9 (the power of 3) to stand for the entire class;
(ℝ, +​); this group isomorphism confirms the taking the powers of 3 as representatives of the
isomorphism of freq and cpitch as inter- elements of Gpyth ⁄ H in this way, we see that
val spaces. In freq, the congruence subgroup the interval group of Pythagorean frequency-​
for octave equivalence is the powers-​ of-​
2 class space is simply the group of powers of
group H = {2a | a ∈ ℤ } = { , 1 4 , 1 2 , 1, 2, 4, }, a 3: GpYTH /H = { 3b | b ∈ ℤ } = { , 1 9 , 1 3 , 1, 3, 9, } .
cyclic subgroup of ℝ>0. Two notes are octave-​ We noted earlier that Gpyth is isomorphic to the
equivalent whenever the interval between additive direct-​product group ℤ × ℤ, and H is a
them, defined multiplicatively in the sense cyclic group isomorphic to the additive group ℤ.
of freq, belongs to H. The interval group of One of the ℤs in the product is effectively can-
frequency-​class space freq ⁄~O, therefore, must celed in the formation of the quotient group,
be the quotient group G′ =​ ℝ>0 ⁄ H. An element so that quotient {3b | b ∈ ℤ} is again isomor-
of this quotient group (a coset of H) is a con- phic to a single copy of ℤ. This should come as
gruence class of positive real numbers related no surprise, because the simplest depiction of
multiplicatively by powers of 2; for example, Pythagorean pitch-​class space is as a discrete
the number 5 belongs to the congruence class one-​dimensional space, the line of fifths. The
{ , 5 4 , 5 2 , 5, 10, 20, }, the coset 5H, which frequency ratio of a pure perfect fifth is 3 2 , not
is the element of G′ representing the interval 3, but because 3b and ( 3 2 )b always belong to the
from pitch class C to pitch class E in just tun- same congruence class in Gpyth ⁄ H, it does not
ing. Every such congruence class contains matter whether we choose powers of 3 or powers
exactly one number x in the range 1 ≤ x < 2 (for of 3 2 in the representation of this quotient group.
the class listed above, that element is 5 4 ), cor- The notion of interval sizes in the interval
responding to an ascending interval smaller group Gpyth ⁄ H may be somewhat unintuitive.
than an octave. The group G′ may be thought of Using the powers-​of-​3 representatives for ele-
as a multiplicative group of numbers between ments of the quotient group, the Pythagorean
1 and 2, with the understanding that larger or interval from C to G is 3, while the interval from
smaller numbers may be brought within that C to D is 9. It is important to remember that
range by multiplying or dividing by 2 as often these intervals are really “3 modulo the powers
as necessary. We have remarked before that a of 2” and “9 modulo the powers of 2.” For grasp-
tuning of pitch space may be specified by defin- ing the size of an interval, the number between
ing a mapping from pitch into freq; in a simi- 1 and 2 representing an equivalence class may be
lar way, a mapping from pc into the quotient more helpful than the power of 3; when 3 and 9
space freq ⁄~O may be regarded as a tuning of are reduced in this way, these intervals become
pitch-​class space. Such a mapping determines the more musically comprehensible 3 2 and 9 8 .
frequency ratios in the range 1 ≤ x < 2 between Expressing intervals as numbers between 1 and
any two pitch classes, thereby fixing the precise 2, on the other hand, disguises the cyclic struc-
acoustic interval between them, without the ture of the group, so it is useful to have both con-
need to specify the registers of the notes. ceptions available.
Pythagorean frequency space pyth, studied
in Exercise 7.3.6, has the multiplicative interval Exercise 7.4.2 Calculate each of the following
group Gpyth =​{2a · 3b | a, b ∈ ℤ}. Octave equiva-
intervals in Pythagorean frequency-​class space.
lence converts pyth into pyth ⁄~O, a Pythagorean
Represent each interval in two ways: as a power
frequency-​class space. The congruence subgroup
of 3 (for example, 9), and also as a number
for octave equivalence in Pythagorean space is
between 1 and 2 (for example, 9 8 ).
the same powers-​of-​2 group H described above,
which is a subgroup of Gpyth. An element of the (a) int(C, F)
quotient group Gpyth ⁄ H is a congruence class (b) int(G, E)
as before, but its elements are limited now to (c) int(E, G)
numbers of the form 2a · 3b. Because all power-​ (d) int(C, C♯) (the Pythagorean semitone)
of-​2 multiples of one number are H-​congruent, (e)  int(C, B♯) (the Pythagorean comma)

Intervals • 271
Exercise 7.4.3 A frequency ratio of y corresponds (a) int(C, D) (the C–​D whole step in the just
to an interval of log2(y) octaves, or 12 · log2(y) C-​major scale)
equal-​tempered semitones, or 1200 · log2(y) (b) int(D, E) (the D–​E whole step in the just
cents. (See Section 2.3 for some related formulas.) C-​major scale)
Use these formulas to calculate sizes in semitones (c) int(E, F) (the just diatonic semitone)
or cents for each of the intervals in Exercise 7.4.2. (d) int(E, E) (the syntonic comma)
The notes in the previous exercise were given as (e) int(C, D♭♭) (the enharmonic diesis, the
pitch classes; choose representatives from those difference between three pure major
classes to make the pitch-​space intervals as small   thirds and an octave)
as  
possible.
Exercise 7.4.5 Use the formulas given in
Just frequency space just, described in Exercise 7.4.3 to calculate sizes in semitones or
Exercise 7.3.7, has the interval group Gjust =​ cents
   for each of the intervals in Exercise 7.4.4.
{2a · 3b · 5c | a, b, c ∈ ℤ}. The powers-​of-​2 group
H, a subgroup of Gjust, is again the congru- As an interval space, the Riemannian ton-
ence subgroup for octave equivalence in this netz may be compared with two other two-​
space. Proceeding as above, we find that every dimensional figures from Chapter 1 that also
congruence class may be reduced to a number depict product interval spaces. Intervals in
of the form 3b · 5c, so the interval group of just signed registral letter space srl may be defined
frequency-​class space is Gjust ⁄ H =​{3b · 5c | b, as ordered pairs derived from the positions of
c ∈ ℤ}. One congruence class in this quotient notes in Figure 1.1.11; for example, intsrl(G4,
group, for example, is { , 3 10 , 3 5 , 6 5 , 12 5 , 24 5 , }, E♯4) =​(−2, 1), because from G4, the note E♯4
represented by the number 3 5 = 31 ⋅ 5−1; this class lies two places to the left (two generic steps
corresponds to the pitch-​class interval from E down from G4 to E4) and one place up (from ♮
to G in the just C-​major scale (and differs from to ♯). The interval group of srl is ℤ × ℤ, so as an
the Pythagorean int(E, G) in Exercise 7.4.2c interval space it is isomorphic to discrete pitch-​
above). We noted previously that Gjust is iso- pair space pitch2, to Pythagorean frequency
morphic to the additive direct-​product group space, or to just frequency-​class space. Signed
ℤ × ℤ × ℤ; forming the quotient by the subgroup
letter space sl, illustrated in Figure 1.2.11, is
H again cancels one ℤ, so Gjust ⁄ H is isomorphic to similar but lacks octave designations; there are
ℤ × ℤ. A picture of just frequency-​class space as a
only seven distinct columns in 1.2.11a, and the
two-​dimensional grid is familiar to us: the uncon- interval group of sl is ℤ7 × ℤ rather than ℤ × ℤ.
formed Riemannian tonnetz of Figure 1.4.14. Octave equivalence ~O, generic equivalence
The powers of 3 correspond to the axis of per- ~G, and enharmonic equivalence ~E are all well-​
fect fifths in the tonnetz (a horizontal row of defined equivalence relations on srl, giving rise
Figure 1.4.14), while the powers of 5 correspond to quotient maps QO: srl → sl, QG: srl → gpitch,
to an axis of major thirds (a diagonal). The just and QE: srl → pitch. The exercise below shows
C-​major scale consists of pitch classes F, C, G, that only two of these three equivalence relations
and D from one row of the tonnetz and A, E, and on srl are induced by subgroups of ℤ × ℤ, and
B from the row above, so the interval from E to G correspondingly only two of the three quotient
mentioned above is really the interval from E to mappings are interval-​space homomorphisms.
G; to say that this interval is 31 · 5−1 means that
one can get from E to G by moving one place in Exercise 7.4.6
the positive direction along the “3” (fifths) axis (a) Identify the subgroup of ℤ × ℤ that induces
and one place in the negative direction along the octave equivalence ~O on srl, and show
“5” (thirds) axis. that QO: srl → sl is an interval-​space
homomorphism.
Exercise 7.4.4 Calculate each of the following (b) Identify the subgroup of ℤ × ℤ that induces
intervals in just frequency-​class space and generic equivalence ~G on srl, and show
visualize them in the Riemannian tonnetz. that QG: srl → gpitch is an interval-​space
Represent each interval in two ways: as a homomorphism.
number of the form 3b · 5c (for example, 3 5 ), (c) In Exercise 1.1.12 you drew a picture
and also as a number between 1 and 2 (for of enharmonic equivalence in srl; the
example, 6 5 ). apparent complexity of this relation was

272 • E x p loring M usical S p aces

 discussed further in Section 2.4. Explain interval i in G, Transi(x) is defined to be the unique
why no subgroup of ℤ × ℤ induces the y in S such that int(x, y) =​ i. Exactly one such y
relation ~E on srl, and why the quotient exists by the homogeneity condition for interval
map QE: srl → pitch is not an interval-​ spaces. For every i, Transi: S → S is a well-​defined
space homomorphism. transformation, and it is an immediate conse-
(d) As a set, sl is identical to spelled pitch-​ quence of the basic properties of interval spaces
class space spc (the line of fifths). The two that Transi is one-​to-​one and onto. We know from
cannot be isomorphic as interval spaces, Chapter 5 that the transposition operators form
however, because the interval group of spc a group isomorphic to G and that the equations
is ℤ while the interval group of sl is ℤ7 × Transi ● Transj =​Transi⊕j and Transi−1 =​Trans⊖i
ℤ, and the two groups are not isomorphic. generally hold, where ⊕ is the group operation
Let Esl–​spc denote the function that maps a in G and ⊖i denotes the inverse of the interval i.
spelled pitch class in sl to the same spelled The reader may suspect that every transposition
pitch class as an element of spc. Is Esl–​spc operator is an interval-​space automorphism of S;
  an interval-​space homomorphism?27 that is true, but we defer discussion of this prop-
erty for a few pages, turning our attention first to
In light of part (c) of the above exercise, while other properties of transpositions.
the space pitch may accurately be described as a We start by using generalized transposi-
quotient set of srl modulo enharmonic equiva- tions to formulate a general statement of the
lence, it is not a quotient interval space, since no so-​called Babbitt hexachord theorem, which
subgroup of its interval group induces ~E. The in its basic form states that a six-​note pitch-​
same is true of the relationship between signed class set always has the same interval content
letter space sl and pitch-​class space pc. As sets, as its complement.28 Here is the more general
however, sl is identical to spelled pitch-​class statement:
space spc while pc is identical to fifths space
fifth, and we noted above that fifth is a quo- Theorem 7.5.1: Generalized hexachord
tient interval space of spc: the interval structure theorem
on spc ensures that ~E is induced by the sub- Suppose (S, G, int) is a finite interval space. If
group 12ℤ of the interval group ℤ. X is a subset of S and i is an interval in G, let αi(X)
denote the number of elements z in X such that
Transi(z) is also in X. If X ′ is the complement of X
7.5 TRANSPOSITION S, then αi(X ′) − αi(X) =​#X ′ − #X.
in  
OPERATORS AND INTERVAL-​
PRESERVING MAPPINGS If S is pc and 1 ≤ i ≤ 5, then αi(X) is simply
In this and the following section we discuss the number of occurrences of interval class i in
extensions of the concepts of transposition and the pitch-​class set X. For i =​6, α6(X) is twice the
inversion operators in the setting of general number of occurrences of interval class 6 in X,
interval spaces, and the closely related ideas because every tritone {z, z +​6} in X is counted
of interval-​preserving and interval-​ reversing once for z and once for z +​6. If #X (the cardi-
mappings. With the exception of inversions nality of X) is 6, then #X ′ is also 6, so the theo-
in non-​commutative interval spaces, these are rem asserts that αi(X ′) =​ αi(X) for every interval
special cases of the general concept of interval-​ i, showing that the interval content of X and of
space homomorphisms from Section 7.2. X ′ are identical. More generally, the theorem
Transposition and inversion operators act in shows, for a set X of any cardinality in any finite
reasonably familiar ways as long as the interval interval space, exactly how the interval content
spaces in question are commutative, but their of X and that of X ′ are related.
behavior (especially that of inversions) may be To prove the theorem, consider all ordered
counterintuitive and confusing when noncom- pairs of the form (z, Transi(z)) for z in S. Because
mutative interval groups are involved. We will X and X ′ are complements, these pairs may be
explore some of those complications at the end partitioned into four sets:
of Section 7.6, but for some purposes we will dis- ZXX =​{(z, Transi(z)) | z ∈ X, Transi(z) ∈ X}
regard them and assume commutativity. ZXX ′ =​{(z, Transi(z)) | z ∈ X, Transi(z) ∈ X ′}
Recall the generalized definition of transposi-
tion from Section 5.6: if (S, G, int) is an interval ZX′X =​{(z, Transi(z)) | z ∈ X ′, Transi(z) ∈ X}
space, then for every element x of S and every ZX ′X ′ =​{(z, Transi(z)) | z ∈ X ′, Transi(z) ∈ X ′}

Intervals • 273
 Because αi(X) counts the z’s for which (z, function from pc to pc, or T4(10) =​2, a function
Transi(z)) ∈ ZXX, it is clear that #ZXX =​ αi(X); sim-
from ℤ12 to ℤ12.
ilarly #ZX ′X ′ =​ αi(X ′). Every element of X occurs In some common spaces the operators Transi
as the first element of exactly one ordered pair are known to us by other names. In pitch or pc,
in either ZXX or ZXX ′, so #X =​#ZXX +​#ZXX ′. Every Transi is nothing more or less than the familiar
element of X ′ occurs as the second element of Ti, where i is either an integer (for transposi-
exactly one ordered pair in either ZXX ′ or ZX ′X ′, so
tions in pitch) or an integer mod 12 (in pc). In
#X ′ =​#ZXX ′ +​#ZX ′X ′. When the first of these twocontinuous space cpitch, analogously, a trans-
equations is subtracted from the second, the position takes the form Tr where the interval
#ZXX ′ terms cancel, leaving #X ′ − #X =​#ZX ′X ′ − r may be any real number. Among other trans-
#ZXX =​ αi(X ′) − αi(X), as the theorem states. positions we have encountered are the generic
To illustrate the theorem in a case in which operators ti. We saw in Figure 4.3.4, for exam-
#X and #X ′ are not the same, consider a tetra- ple, that diatonic sequences are characterized by
chord X of prime form 0257 and its comple- patterns of generic transpositions. The opera-
ment X ′, an eight-​note segment of the circle of tors ti may be defined either on generic or on
fifths. The traditional interval-​class vector of X diatonic spaces, and like their chromatic coun-
is 021030.29 By the theorem, for every i, αi(X ′) terparts, they may be applied either to pitches
exceeds αi(X) by 4, the difference in cardinality or to pitch classes. The pitch spaces gpitch
between X and X ′. It follows that the first five and dpitch(n) are infinite spaces with interval
entries in the interval-​class vector of X ′ must be group ℤ, so infinitely many different opera-
46547, obtained by adding 4 to the correspond- tors ti are available in these spaces, whereas in
ing entries in the vector of X. For the i =​6 posi- gpc or dpc(n)—​modular spaces with interval
tion we must add only 2, not 4, because as noted group ℤ7—​there are only seven operators ti. Of
above, α6 counts tritones twice; the vector of X ′ course, corresponding operators may be defined
is 465472. in other scalar spaces as well, for which we may
adapt the ti notation as needed; for example, we
Exercise 7.5.2 may introduce a notation such as tiO for transpo-
(a) For each of the following small subsets X sition by i steps within an octatonic scale. These
of pc, determine the interval-​class vectors scalar (or intrascalar) transpositions measure
of X and X ′, and verify the relationship their intervals by counting steps within their
described in Theorem 7.5.1: X =​ ∅, X =​{C}, respective scales, and therefore they differ from
X =​{C, C♯}, X =​{C, F♯}. transpositions of the same music in chromatic
(b) Use Theorem 7.5.1 to calculate the interval-​ space.31 Another kind of transposition in our
class vector of an enneatonic collection repertoire is spc transposition along the line of
by first determining the vector of its fifths, introduced in Section 2.6; the operators τi
complement, an augmented triad. are the transposition operators Transi in spelled
(c) Messiaen’s sixth mode of limited pitch-​class space spc.
transposition is a scale of prime form One more notation will prove useful: if u and
0124678t. Use Theorem 7.5.1 to find the v are specific elements of S, we will sometimes
  interval-​class vector of this collection.30 write Transvu for the transposition operator
Transi, where i =​int(u, v). This notation is similar
to the notation g uv introduced for interval-​space
The behavior of transpositions may be homomorphisms in Section 7.2. Like g v , Trans v
u u
expressed simply using label functions: Transi is a function that maps u to v (pronounced
adds i to the label of any element to which it is “Trans-​u-​v,” not “Trans-​v-​u”), and a given trans-
applied. More precisely, labu(Transi(x)) =​labu(x) ⊕ position Trans may be written as Trans uv for
i
i; this equation is nothing more than the interval-​ many different pairs of reference points u and
sum equation int(u, Transi(x)) =​int(u, x) ⊕ int(x, v—​all pairs for which int(u, v) =​ i.
Transi(x)). Transposition operators are technically
functions Transi: S → S, but in many familiar Exercise 7.5.3
settings, it is common to write them as func- (a) The interval spaces pitch, gpitch, and
tions on labels instead, so that the operators are spc share the interval group ℤ. When the
regarded as functions Transi: G → G given by the transposition operators Ti (on pitch), ti
simple formula Transi(j) =​ j +​ i. In pc, for exam- (on gpitch), and τi (on spc) are considered
ple, we may write either T4(B♭) =​D, implying a as functions on ℤ (that is, on labels, as

274 • E x p loring M usical S p aces

 discussed above), all three are the same intervals (i, j) defines a transposition operator
function: for example, the equations Trans(i,j) in the product space. Transposition
T5(3) =​8, t5(3) =​8, and τ5(3) =​8 are all by an ordered-​pair interval works component-
correct. Explain what these three equations wise: the first component is a mod-​7 interval
mean in the spaces pitch, gpitch, and spc. in sd, while the second is a mod-​12 interval
As reference points for the label functions, in pc. For example, the relationship between
use C4 in pitch and gpitch; use D in spc. C♯ and G at m. 21, modeled by the interval (4,
(b) The transposition operator acting on an 6) in Figure 7.3.3, may be described equiva-
interval space may be distinguished from lently by the equation Trans(4,6)(6̂, 1) =​(3̂,
the one on its interval group by writing 7): the scale degree is transposed up by four
them as Transi: S → S and Transi ′: G → steps (t4(6̂) =​3̂) and the pitch class by six semi-
G. The group mapping Transi ′ is then the tones (T6(1) =​7).
conjugate of the space mapping Transi by the In sd × pc, let X be the set {(1̂, 0), (3̂, 4), (5̂,
label function labu: 7)}, or {(1̂, C), (3̂, E), (5̂, G)}, a tonic triad in C
major. Here are a few transpositions of X:
Transi ′ =​labu−1 ● Transi ● labu.
X = {(1̂, C), (3̂, E), (5,̂ G)} — ​a
Explain why this equation is correct and tonic triad in C major
show how it applies in the case of the Trans(0,7)(X) =​ {(1,̂ G), (3,̂ B), (5,̂ D)} — ​a
  examples in (a).32 tonic triad in G major
Trans(4,7)(X) =​ {(5̂, G), (7̂, B), (2,̂ D)} — ​a
Having defined transposition operators on
dominant triad in C major
the elements of a space (such as pitch classes)
we may then apply them also to sets of elements Trans(3,0)(X) =​ {(4,̂ C), (6,̂ E), (1,̂ G)} — ​a
(pitch-​class sets). This is a straightforward appli- subdominant triad in G major
cation of a general convention about functions Trans(2,4)(X) =​ {(3̂, E), (5̂, G♯), (7,̂ B)} — ​an
noted in Section 2.3: if f: S1 → S2 is a function E-​major triad (V/​vi) in C major
and A is a subset of S1, then f(A) denotes the set
{f(x) | x ∈ A}. Thus we may transpose pitch sets
A transposition such as Trans(0,7), which leaves the
or pitch-​class sets by Ti, sets of generic or dia-
scale-​degree component unchanged, is termed
tonic pitch classes by ti, and sets of spelled pitch
a real transposition by Rings (2011b, 81–​83). An
classes by τi. Ordered sets may be similarly trans-
interval such as (3, 0), as noted in Section 7.3, is
posed, by transposing each component through
a pivot interval, and the corresponding transposi-
the same interval: in pitch3, for example, we
tion Trans(3,0), which relates two different tonal
may write T7(G3, E4, C4) =​(D4, B4, G4). (In this
interpretations of the same set of pitch classes,
case “T7” is equivalent to transposition by the
may be called a pivot transposition. The last exam-
interval (7, 7, 7) in the product space pitch3.)
ple in the list above, Trans(2,4), is one instance of
Depending on the nature of the interval space,
what Rings generally calls chromatic transposition.
the relationship between a set A and a transpo-
Trans(2,4)(X) is an E-​major triad, not the E-​minor
sition Transi(A) may or may not resemble the
triad that would be a diatonic iii chord in C major.
usual understanding of the word transposition—​
No transposition by an interval in sd × pc relates
but because transposition operators are always
a major triad to a minor triad. We know one way
one-​to-​one, we can say at least that if A is a finite
to relate a I and a iii chord using the diatonic
set, Transi(A) is always a set of the same cardi-
transposition operator t2; Rings accomplishes
nality as A.
something comparable by defining diatonic
scales within his system and defining diatonic
Exercise 7.5.4 Show that it is possible for a transposition within those scales.
major triad and a minor triad to be related
by t1O, transposition by one step within
an octatonic scale. What do you get if you Exercise 7.5.5 Let Y be the set {(2,̂ B♭), (4,̂ D),
transpose
   a major triad by t2O? By t3O? (6̂, F)} in sd × pc. Calculate the result of
applying each of the transposition operators
The Debussy analysis in Example 7.3.2 Trans(2,0), Trans(4,0), Trans(4,7), and Trans(3,6) to Y,
made reference to several ordered-​pair inter- and describe musical situations in which each of
vals in Rings’s space sd × pc. Each of these these
   transpositions might be of interest.

Intervals • 275
 Transpositions are closely related to interval-​ label-​preserving, then f is interval-​preserving,
preserving mappings. As defined in Section 7.2, and therefore that presuv is the only u–​v label-​
a transformation f: S → S on an interval space (S, preserving
   transformation on S.33
G, int) is interval-​preserving if int(f(x1), f(x2)) =​
int(x1, x2) for all x1 and x2 in S. Equivalently, f is In Section 7.1 we remarked that a label func-
interval-​preserving if it is interval-​ compatible tion essentially superimposes a picture of the
with the identity mapping EG on the interval interval group on a picture of the space. An
group G. Because EG is an automorphism of G, our interval-​preserving mapping presuv shifts the
work in Section 7.2 (particularly Theorem 7.2.5) alignment of the two pictures to a different refer-
shows that every interval-​preserving transfor- ence point, moving the label 0 (the identity ele-
mation on S is automatically an interval-​space ment) from u to v and exchanging u-​based labels
automorphism of S. Also, for every u and v in S for v-​based labels. To see how this works, the
there exists one and only one interval-​preserving reader may consult Figure 7.1.1 again: if u =​C4
function f: S → S such that f(u) =​ v, namely f = g uv , and v =​G4, then presuv maps any pitch in 7.1.1a to
where g =​ EG. For any x in S, f(x) is the unique y in the pitch with the same number in 7.1.1b. In this
S such that int(v, y) =​int(u, x). Consequently an case, presuv is simply the transposition operator
interval-​preserving function is completely deter- T7. This shift is also analogous to the relabeling
mined once its value at any point is known: if f: S of Cayley diagrams noted in Exercise 6.2.9.
→ S and f ′: S → S are both interval-​preserving It is not by accident that the interval-​preserv-
and if f(u) =​ f ′(u) for some u in S, then f and f ′ are ing mapping in this example is a transposition.
the same function. We write presuv for the unique Figure 7.5.7 compares the characteristic behav-
interval-​preserving function that maps u to v. iors of transpositions in part (a) with those of
As was the case with g uv and Transuv, the function interval-​ preserving functions in (b). The ele-
presuv is well-​defined, but the same function may ments u and v, appearing at the bottom of each
be written as presuv for many different choices of figure, may be thought of as reference points
the reference points u and v. relative to which other elements are situated;
f is a function mapping u to v, and f also maps
Exercise 7.5.6 The equation int(v, y) =​int(u, x) some other element x to y. If f is a transposition
may be rewritten using label functions in the Transi, as in 7.5.7a, then the intervals int(u, v)
form labv(y) =​labu(x), so presuv( x ) is the element and int(x, y) must be the same, because both are
of S whose v-​label is the same as the u-​label equal to the interval of transposition i. Hence
of x. We describe this situation by saying the horizontal dashed arrows in (a) depict equal
that the transformation presuv is u–​v label-​ intervals. The condition required of an interval-​
preserving. The u–​v label-​preserving property preserving function, as in part (b), says not that
is a special case of the interval-​preserving int(u, v) =​int(x, y) but instead that int(u, x) =​
property, requiring only that a function int(v, y): in this case it is the other two intervals,
preserve intervals of the form int(u, x) with those drawn (almost) vertically in the figure,
the starting point u. But this property implies that are required to match.
the interval-​preserving property in its full These two conditions, though different, are
generality. Suppose u and v are elements of an closely related. The geometry of Figure 7.5.7 per-
interval space S. Show that if f: S → S is u–​v haps suggests that if the two arrows in one pair are

FIGURE 7.5.7 (a) Matching intervals under a transposition Transi; (b) matching intervals under an
interval-​preserving mapping Pres uv

276 • E x p loring M usical S p aces

equal in length and direction, the same must be Commutativity is essential to most of the
true of the other pair—​but of course we must be conclusions of the preceding paragraph; indeed,
careful about extrapolating from plane geometry in noncommutative interval spaces, transpo-
to a general interval space. Suppose that f is a trans- sitions and interval-​ preserving functions are
position Transi, so that the horizontal intervals different things. Even in the noncommutative
match as in Figure 7.5.7a, and consider the verti- case, however, transpositions are interval-​space
cal interval int(v, y) on the right side of the figure. automorphisms. To see this, consider a transpo-
By the interval-​sum equation, int(v, y) =​int(v, u) ⊕ sition Transi =​ Transuv, for fixed elements u and v
int(u, x) ⊕ int(x, y), where ⊕ is the group operation and interval i =​int(u, v), with no assumption of
in G (not necessarily ordinary addition). This inter- commutativity. The discussion above established
val sum can be visualized in the figure by tracing a the equation int(v, Transuv(x)) =​ ⊖i ⊕ int(u, x) ⊕
path from v to u to x to y. Because f is the transpo- i. In the commutative case, the right side of this
sition Transi, int(u, v) =​int(x, y) =​ i, and therefore equation reduces to int(u, x), but in general it is
int(v, u) =​ ⊖i, the inverse of i in the group G. Hence equal to Conji(int(u, x)), the conjugate of int(u, x)
the equation becomes int(v, y) =​ ⊖i ⊕ int(u, x) ⊕ i. by i =​int(u, v). (The general formula for the con-
If the group G is commutative, then we may rear- jugate of a group element, from Section 6.6, is
range the terms on the right side of the equation Conji(a) =​ i−1ai—​or, in the notation we are using
and cancel ⊖i with ⊕i, concluding that int(v, y) =​ for the interval group, Conji(a) =​ ⊖i ⊕ a ⊕ i.)
int(u, x)—​the condition required of an interval-​ Therefore int(v, Transuv(x)) =​Conji(int(u, x)), or
preserving function. If G is commutative, therefore, equivalently labv(Transuv(x)) =​Conji(labu(x)). This
every transposition is also interval-​preserving. A equation, valid for every u, v, and x in any inter-
similar argument shows that if int(u, x) =​int(v, y) val space S, means that for every u and v, the
and if G is commutative, then int(u, v) =​int(x, y), so transposition Transuv: S → S is label-​compatible
every interval-​preserving mapping is a transposi- with the mapping Conji: G → G, where i =​int(u, v).
tion. In a commutative interval space, therefore, Because (from Section 6.6) every conjugation
transpositions and interval-​preserving mappings is an automorphism of G, Exercise 7.2.3c then
are the same thing, and the functions Transuv and implies that the mappings Transuv and Conji are
presuv are always the same. We already established interval-​compatible, from which it follows that
in general that presuv is an interval-​space automor- Transi =​ Transuv is an interval-​space automor-
phism of S, so the same is true of Transuv in the phism of S.
commutative case. These properties offer a first Often the functions of greatest interest
illustration of the general principle that transposi- are not those of the form f: S → S defined on
tion and inversion are better behaved in commu- a single interval space but rather mappings
tative spaces, and they also explain why our only f: S1 → S2 from one interval space (S1, G1, int1)
examples of interval-​preserving transformations to another space (S2, G2, int2). Such mappings
so far have been transpositions: the most famil- may relate two entirely different types of objects
iar interval spaces are commutative, so no other and accordingly are sometimes called cross-​type
examples are possible. transformations. In Figure 7.5.8, the elements u
Returning once more to the Debussy analy- and x belong to S1, while v and y belong to S2.
sis in Figure 7.3.3, we can see several illustra-
tions of the interval-​ preserving property of
transpositions in the commutative interval
space sd × pc. For example, the arrow labeled
(1, 0) at the top of the graph may be inter-
preted as a transposition Trans(1,0), which maps
the ordered pairs (6̂, 1) and (3̂, 7) to (7̂, 1) and
(4̂, 7), respectively. The fact that the intervals
int((6̂, 1), (3̂, 7)) and int((7̂, 1), (4̂, 7)) in the
two bubbles are both equal to (4, 6) shows the
interval-​preserving behavior of this transpo-
sition. We remarked earlier that the bubble
notation works because of the commutativity
of the interval group; we could just as well say FIGURE 7.5.8 A cross-​type interval-​preserving
that it works because of the interval-​preserving mapping between two spaces with the same inter-
nature of transpositions. val group

Intervals • 277
The conditions described in the preceding para- int1(u, x), or equivalently labv(y) =​labu(x). If the
graphs do not necessarily apply in this more gen- groups are the same, then this interval equa-
eral setting. The intervals int(u, v) and int(x, y) tion means that presuv is label-​compatible and
are not even defined, as the elements involved therefore interval-​compatible with the identity
belong to two different spaces. The intervals mapping on the group, and hence is an interval-​
int(u, x) and int(v, y) exist, but they cannot in space isomorphism. If G1 is a proper subgroup
general be compared because they potentially of G2, then presuv is interval-​compatible with the
belong to two different interval groups. inclusion mapping EG1 −G2 from G1 to G2, which is
We noted in Section 7.2, however, that the a group homomorphism but not an automor-
interval-​preserving condition can be satisfied if phism, so in this case presuv is an interval-​space
the groups G1 and G2 are the same, or even if G1 homomorphism but not an isomorphism.
is a subgroup of G2. As we know, chromatic pitch Technically a cross-​type mapping presvu∈∈SS21 is
space pitch, generic pitch space gpitch, and an interval-​preserving function, not a transposi-
spelled pitch-​class space spc (the line of fifths) tion. There can be no “interval of transposition”
all share the interval group ℤ. If f is a function from a space S1 to another space S2 unless inter-
from one of these spaces to another, the interval-​ vals from points of S1 to points of S2 are well-​
preserving property int1(x1, x2) =​int2(f(x1), f(x2)) defined, which is not generally the case unless S1
makes sense: the spaces may be different and is the same as (or a subspace of) S2—​a consider-
the interval functions may be different, but ably more restrictive condition than saying that
as long as the interval functions take values in G1 is the same as (or a subgroup of) G2. Interval-​
the same group the equation may be satisfied. preserving cross-​type mappings have sometimes
For an example in which G2 is a larger group been called transpositions, however;34 the term
than G1, consider the inclusion mapping Epitch–​ is convenient and there is probably no harm in
cpitch from chromatic pitch space into continu- it as long as the interval groups are commuta-
ous pitch space, with interval groups ℤ and ℝ, tive and there is no other kind of “transposition”
respectively; this function, which maps every with which presuv could be confused. To main-
note in pitch to the same note in cpitch, is cer- tain precision in the following examples we will
tainly interval-​preserving, but the intervals in adhere to the presuv notation, but in the narrative
pitch, which are always integers, do not exhaust we will occasionally refer to such transforma-
the real-​number intervals in cpitch. tions informally as “transpositions,” in quota-
As long as G1 ⊆ G2, then for any choice of u in S1 tion marks.
and v in S2, there is a unique interval-​preserving
function presuv : S1 → S2 such that presuv (u) =​ v.
The name of the function may be written more Example 7.5.9: Wagner, Tristan und Isolde
explicitly as presvu∈∈SS21 , indicating the spaces Cross-​type mappings between scales with
involved. This is the function shown in Figure interval group ℤ are particularly useful.
7.5.8; presuv (x) is defined, for every x in S1, to be Melodic motives are transformed from one
the unique element y of S2 such that int2(v, y) =​ scale basis to another in many works, and

FIGURE 7.5.10 Scalar and interscalar “transpositions” in Wagner, Tristan und Isolde, Prelude to Act I,
mm. 2–​7, and Prelude to Act III, mm. 1–​4

278 • E x p loring M usical S p aces

interscalar cross-​type mappings often model in pitch. The interval-​preserving mapping
such situations effectively. Figure 7.5.10 shows presGB43 ∈∈pITCH
DpITCH( −4)
between the two second
a well-​known example, a comparison of the statements
   completes the network.
oboe line at the beginning of the Prelude to Act
I of Tristan und Isolde with the similar melodic Example 7.5.11: Grieg, Sonata in C Minor
fragment that opens the Prelude to Act III. for Violin and Piano, Op. 45, I
The Act I excerpt appears at the bottom of the The second theme in the first movement of
figure; several analytical figures in this chapter Grieg’s C-​Minor Violin Sonata undergoes the
are organized chronologically from bottom to unusual series of melodic transformations
top, so that the subscript-​to-​superscript reading shown in Figure 7.5.12. The first phrase (mm.
of the transformational labels matches the 59–​66), at the bottom of the figure, is diatonic
orientation of the arrows. The Act I melody is in E♭ major, emphasizing the notes of the tonic
chromatic, while the Act III melody inhabits the triad. The second phrase transposes the first
F-​minor diatonic collection dpitch(−4). From to G♭ major. This minor-​third transposition,
the starting notes G♯4 and G3, both melodies which we just encountered in Wagner as well,
ascend through four-​note segments of their was a favorite of Grieg’s, appearing in many
respective scales, so these initial fragments are works. It could of course be modeled as T3 in
related by an interscalar “transposition” chromatic pitch-​class space (or in this case T15
presGG3♯ ∈∈DpITCH(
4 pITCH
−4)
. Other notations for the same in chromatic pitch space), but as shown in the
mapping are possible; this one takes the initial figure it may also be described as an interval-​
notes in both voices as reference points. Even preserving mapping between the two diatonic
though the diatonic motive spans a larger scales, DPC(−3) and DPC(−6). The last pitch
absolute interval than the chromatic one, class of the first phrase, B♭, reappears as the
each motive spans three steps of its own scale, first pitch class of the second, forming a pivot
allowing the interval-​preserving transformation third (5, 0) from (5̂, 10) to (3̂, 10) in the product
to relate them exactly. space sd × pc.
In both Preludes the opening fragment Measure 79 marks the beginning of a 23-​
is followed by a restatement. In the Act III measure passage in which almost every note
Prelude the second fragment is identical to in both the violin and the piano belongs to the
the first, shown here as a t0 transposition in octatonic collection oct23. Like other scales, an
dpitch(−4), but in Act I, the second statement is octatonic scale may be regarded as an interval
transposed up a minor third, a T3 transposition space in either of two ways: as a pitch space

FIGURE 7.5.12 Scalar and interscalar “transpositions” in Grieg, Violin Sonata in C Minor, Op. 45, I

Intervals • 279
with interval group ℤ, or as a pitch-​class space Exercise 7.5.13
with a modular interval group, in this case (a) Measures 75–​78, omitted from Figure
ℤ8. At a small risk of confusion, we use labels 7.5.12, transpose mm. 71–​74 down a
such as oct23 for both kinds of spaces. In the whole step to E major. Give as many
present situation an octatonic pitch space must transformational descriptions as you can of
be used for the cross-​type mappings, because relationships between this phrase and the
only the pitch space shares its interval group surrounding music.
with the diatonic spaces. The tonal focus of (b) Although the melody notes in mm. 79–​82
Grieg’s octatonic passage is indistinct, but belong to dpitch(+​3), the relationship
much of it unfolds over a B pedal, perhaps between the third and fourth phrases in
suggesting a dominant prolongation in Figure 7.5.12 (even ignoring the asterisked
E. Melodically, the four measures of mm. 79–​ notes) cannot be described as an interval-​
82 form a diminution of the previous eight-​ preserving mapping between diatonic
measure phrases, reconfigured to conform to spaces. Why not?
the octatonic scale; the pitches are derived (c) The melody notes in mm. 83–​86 belong
almost exactly from the melody in mm. 67–​74 to the E harmonic minor scale. This scale
by the interval-​preserving cross-​type mapping forms an interval space isomorphic to the
shown in the figure. The only deviations other scales; in Chapter 14 we will call it
from this “transposition,” marked by Xs in hmin(+​2). Can the relationship between the
the example, serve to highlight the melody’s third and fourth phrases be described using
triadic frame even more strongly than before, an interval-​preserving mapping between
with striking effect because the triad in   dpitch(+​3) and hmin(+​2)?35
question is now diminished in quality.
These four measures are then restated at Example 7.5.14: Ginastera, Danzas
a higher pitch level in mm. 83–​86. At first argentinas No. 1
glance this may appear to be a straightforward In the first of Ginastera’s Danzas argentinas,
chromatic T1 transposition, but the the right hand plays on white keys and the left
A in m. 81—​the only note of mm. 79–​82 hand on black almost exclusively; Ginastera
not belonging to the G♯-​diminished triad—​is writes a five-​flat key signature on the lower
transposed by T2, not T1. This deviation allows staff only. The white-​key diatonic pitch space is
mm. 83–​86 to conform to the prevailing dpitch(0). For pentatonic collections we adopt
octatonic scale—​and in fact the transposition a naming convention based on the observation
of the four-​measure phrase is exact as a scalar that a pentatonic collection, like a diatonic one,
transposition within oct23 (t1O rather than is a consecutive segment of the line of fifths; we
chromatic T1). index the collection by the central one of its five
Other transformational descriptions of notes. The pentatonic collection G♭–​D♭–​A♭–​E♭–​B♭
this passage are possible. If one considers the is thus indexed by the number −6, the spc
middle two lines of Figure 7.5.12 and disregards number assigned to A♭, so our label for this
(1) register; (2) the key signature; (3) accidentals; space is pent(−6).36 As with the octatonic space
and (4) the notes marked with Xs, the two lines in the previous example, this label could refer
become completely identical: both consist of the either to a pitch space or to a modular pitch-​
same notes B –​D –​B –​G –​⋯. In other words, class space (in this case with interval group ℤ5);
the two phrases are identical in generic pitch-​ in the present example it refers to a pitch space.
class space. The second phrase inhabits the Two short excerpts from the piece appear
diatonic collection dpitch(−6). The third phrase in parts (a) and (c) of Figure 7.5.15. At
was described above as octatonic, but all the m. 26 the right hand plays a series of diatonic
melody notes in this phrase (whatever its tonal transpositions of the initial trichord B–​C–​E,
orientation may be) also belong to dpitch(+​3). ascending through the diatonic scale as shown
The transformation between the two could in (a). “Planing” chords of this sort, either in
therefore be accomplished through alteration diatonic or in chromatic space, are common
of an implied key signature, leaving the generic in some musical styles and are easily modeled
notes unchanged—​an example of a signature using transposition operators of the appropriate
transformation (in this case s9, from six flats kind. The top portion of the analytical graph
to three sharps), a concept to be formalized in in (b) is typical of such analyses: not only are
Chapter
   13. successive chords related by diatonic t1, but

280 • E x p loring M usical S p aces

FIGURE 7.5.15 Scalar and interscalar “transpositions” in Ginastera, Danzas argentinas No. 1

also the lines formed by the three voices are scales result in motion that is not closely
related to each other by t1 and t2, reflecting the parallel at all when viewed chromatically. The six
intervals in the planing chord. chords exhibit five different chromatic interval
What makes this passage distinctive is structures; only the second and third chords are
the left hand, which is doing the same thing related exactly by chromatic transposition, as
in its pentatonic world. Here t1P indicates major seconds happen to occur simultaneously
transposition by one step in the space pent(−6), in all four voices at that moment. The pentatonic
relating successive notes of the pentatonic voice, having fewer notes to play, traverses an
scale. The pentatonic line is related to any of the octave more quickly than the diatonic voices do,
diatonic lines by an interscalar “transposition” starting a perfect fourth below the next lowest
such as pres GB3♭3∈∈DpITCH(0)
pENT( −6), shown between the
voice but nearly catching up to it by the end
lowest two voices. Although all the motion is of the measure. The example shows why the
in some sense parallel, the mixtures of interval interscalar mapping must be defined on pitch
sizes within both the diatonic and pentatonic spaces rather than on pc spaces: the two G♭s in

Intervals • 281
the pentatonic line correspond to two different appropriate pitch-​class spaces (although the
pitch classes in any of the right-​hand lines. pitch-​space mapping probably remains more
A similar passage a few measures letter, useful). Why is a pc-​space mapping possible in
shown in (c) and analyzed in (d), moves a this
  situation?38
second voice into pentatonic space. The two
right-​hand voices move diatonically in parallel These examples will suffice for the time being;
thirds. The left hand plays a pentatonic version we will later see examples of cross-​type mappings
of parallel thirds; that is, the two left-​hand in other settings. The remainder of this section
voices are consistently related by pentatonic assembles many properties of transpositions and
transposition t2P, but some of these pentatonic interval-​preserving functions, many of which
“thirds” are perfect fourths in the chromatic have been observed already, in a series of theo-
world. Consequently the entire two-​voice rems (Theorems 7.5.19, 7.5.20, and 7.5.24). The
units in the two hands are related exactly by formulations given here are sometimes more gen-
an interscalar “transposition,” as shown by the eral than in the discussion above. We will not have
large dashed bubbles in (d); this mapping is the occasion to apply all of these properties, so com-
same one that related the lowest two voices plete proofs are not presented, but a few remarks
in (b). Except for the literal chord repetitions, are given below by way of elaboration, and some
no two chords in this example have the same readers may enjoy working out the details.
chromatic
   structure.
Theorem 7.5.19: Properties of transposition
Exercise 7.5.16 Examine the fugue subject
operators
in mm. 1–​5 of the first movement of Bartók’s
In the statements below, (S1, G1, int1), (S2,
Music for String Instruments, Percussion, and
G2, int2), and (S3, G3, int3) are interval spaces
Celesta and the transformed version in mm.
satisfying S1 ⊆ S2 ⊆ S3 and G1 ⊆ G2 ⊆ G3. (In
204–​09 of the fourth movement. Show how
particular, the spaces may all be the same.) It is
the relationship between the two may be
assumed that the interval functions of the larger
expressed as an interval-​preserving interscalar
spaces extend those of the smaller (for example,
transformation.
   37
for all x1 and x2 in S1, int1(x1, x2) =​int2(x1, x2)).
The symbol ⊕ denotes the group operation in
Exercise 7.5.17 Debussy’s “Mouvement” (No.
G2, and ⊖i is the inverse of i. The letters u, x, a,
3 of Images I), mm. 67–​73, offers an unusual
and b always denote elements of S1; v and y are
hybrid of diatonic and chromatic planing. The
elements of S2; w is an element of S3.
top-​voice melody is diatonic in DPC(0), but the
harmonies are of the nondiatonic prime form
(a) Definitions and basic properties. For every
01369, and successive chords are transposed
exactly in chromatic space. Examine the score, i in G2, Transi: S1 → S2 is the interval-​
reduce it to a five-​voice texture, and construct space homomorphism induced by the
an analysis in the style of Figure 7.5.15, conjugation mapping Conji: G1 → G2.
assigning
   each voice its own scalar space. For every x in S1, Transi(x) is the unique
element y of S2 such that int2(x, y) =​ i. For
Exercise 7.5.18 Write the subject from the every u in S1 and v in S2, Transuv is the
C-​Minor Fugue from Bach’s WTC I (to the same function as Transi, where i =​int2(u,
downbeat of m. 3). Then write a transformed v). If S1 =​ S2, then Transi is an interval-​
version of the same subject, starting on the space automorphism of S1.
same pitch C5, in which every note n semitones (b) Identity mapping. For every u in S1, Transuu
above or below C5 is replaced by a note n whole is the inclusion mapping ES1 −S2 from S1 to
steps above or below C5. (The first four notes, S2. If S1 =​ S2, then Transuu is the identity
originally C5–​B4–​C5–​G4, thus become C5–​B♭4–​ mapping ES1 on S1.
C5–​D4.) Show how the relationship between (c) Composition of transpositions. For every u in
the two subjects may be expressed as an S1, v in S2, and w in S3, Transuv ● Transwv
interval-​preserving interscalar transformation. =​ Transuw .
This transformation, unlike those considered (d) Inverses of transpositions. If S1 =​ S2, then for
above between scales of different cardinalities, every u and v in S1, ( Transuv )−1 = Transuv . If
defines a coherent mapping even on the Transuv = Transi , then ( Transvu )−1 = Trans⊖i.

282 • E x p loring M usical S p aces

(e) Label functions for transpositions. Suppose S2 → G2 are the label functions associated
labv: S2 → G2 is the label function with u and v. Then for every a in S1,
associated with v. Then for every a in
S1, labv (Trans xy(a)) =​labv(a) ⊕ i, where labv(presxy( a))
i =​int2(x, y). = labv ( y ) ⊕ ( ⊖ labu( x )) ⊕ labu( a).
(f) Intervals formed by transpositions. For every (f) Intervals formed by interval-​preserving
u, x, a and b in S1 and for every v and y in S2, mappings. For every u, x, a, and b in S1 and
int2(Transuv (a), Trans xy (b)) for every v and y in S2,
=​ int2(v, u) ⊕ int1(a, b) ⊕ int2(x, y).
int 2(pres vu( a), pres xy ( b ))
(g) Change of reference points. For every u and = int 1( a, u) ⊕ int 2(v, y ) ⊕ int 1( x , b ).
x in S1 and for every v and y in S2, Transuv =​
y
Trans x if and only if Transuv(x) =​ y—​that is, (g) Change of reference points. For every u
  if and only if int2(u, v) =​int2(x, y). and x in S1 and for every v and y in
S2, presuv = pres xy if and only if presuv (x)
Theorem 7.5.20: Properties of =​ y—​that is, if and only if
interval-​preserving mappings int1(u, x) =​int2(v, y).
In the statements below, (S1, G1, int1), (S2,
G2, int2), and (S3, G3, int3) are interval spaces (h) Intervals formed (commutative case). If G2 is
whose interval groups satisfy G1 ⊆ G2 ⊆ G3. (The commutative, then for every u, x, and a in S1
three spaces and the interval functions may be and for every v and y in S2,
entirely different as long as the groups are in this int 2(presvu ( a), pres xy( a))
relationship.) The symbol ⊕ denotes the group = int 1( x , u) ⊕ int 2( v, y ).
operation in G2, and ⊖i is the inverse of i. The
letters u, x, a, and b always denote elements of S1; (i) Change of reference points (generalized
v and y are elements of S2; w is an element of S3. formula in commutative case). If G2 is
commutative, then for every u and x in S1
(a) Definitions and basic properties. For every and for every v and y in S2,
u in S1 and v in S2, presvu∈∈SS21 : S1 → S2 is y
presx = ( Transi )−1 • presuv • Transj ,
the interval-​space homomorphism
mapping u to v, induced by the inclusion where i =​int1(u, x) and j =​int2(v, y).
mapping EG1 −G2 from G1 to G2. For every
x in S1, presuv( x ) is the unique element (j) Relationship with transposition. If S1 =​ S2
y of S2 such that int2(v, y) =​int1(u, x), or and G1 is commutative, then presuv is the
equivalently such that labv(y) =​labu(x). same function as Transuv, which is the
presuv is the unique interval-​preserving transposition operator Transi: S1 → S1,
mapping from S1 to S2 that maps u to v. where i =​int1(u, v). If S1 is a subspace of S2
If G1 =​ G2, then presuv : S1 → S2 is an and G2 is commutative, then presuv(x) =​
interval-​space isomorphism, induced by Transi(x) for all x in S1, where i =​int2(u, v);
the identity mapping EG1 on G1. that is, presuv is the same as the trans­
(b) Identity mapping. If S1 is a subspace of S2, position operator Transi, defined only on
then for every u in S1, presuu is the inclusion   the elements of S1.
mapping ES1 −S2 from S1 to S2. If S1 =​ S2, then
presuu is the identity mapping ES1 on S1. In both Theorems 7.5.19 and 7.5.20, parts
(c) Composition of interval-​preserving mappings. (c)–​(g) follow from the five parts of Theorem
For every u in S1, v in S2, and w in S3, 7.2.13, though in some cases more direct proofs
presuv • preswv = presuw. are possible. The earlier theorem, about interval-​
(d) Inverses of interval-​preserving isomorphisms. space homomorphisms in general, is here
If G1 =​ G2, then for every u in S1 and v in S2, applied in the special cases of transpositions
( presuv )−1 =​ presvu ; this function is an and interval-​preserving mappings, for which the
interval-​preserving interval-​space interval-​compatible group homomorphism is
isomorphism from S2 to S1. either a conjugation or an identity mapping. It
(e) Label functions for interval-​preserving is also instructive to compare the corresponding
mappings. Suppose labu: S1 → G1 and labv: parts of the two theorems in detail.

Intervals • 283
 in the next higher voice. Verify the formula
for labv(pres xy (a)) in Theorem 7.5.20e in
this case.
(b) Let x and y be as in part (a) of this exercise,
but now let u =​G♭4 and v =​C5, so that pres uv
is a cross-​type mapping from the lowest
voice to the highest in Figure 7.5.15b,
defined using the final notes in
the two voices as reference pitches.
Let a and b be A♭3 and E♭4 respectively, the
FIGURE 7.5.21 Change of reference points in second and fifth notes in the pentatonic
interval-​preserving mappings (Theorem 7.5.20i) voice. Let int1 and int2 be the interval
functions in pent(−6) and dpitch(0)
respectively. Verify that the formulas for
Theorem 7.5.20g applies in the situation int2(presuv (a), presxy(b)) and int2(presuv(a),
shown previously in Figure 7.5.8, where int(u, x) presxy (a)) in parts (f) and (h) of Theorem
and int(v, y) are equal intervals and presuv and 7.5.20 are correct in this case, and verify
pres xy are the same function. Theorem 7.5.20i that the relationship between presuv and
then provides, in the commutative case, a more   presxy is as described in part (i).
general formula relating presuv and pres xy , valid
even if int(u, x) and int(v, y) are different. Here Exercise 7.5.23 In the formula in part (h) of
Transi and Transj are transpositions within the Theorem 7.5.20, the element a does not appear
spaces S1 and S2 respectively. This situation is to the right of the =​sign. This means that in the
depicted in Figure 7.5.21, where it is clear that commutative case, the interval from presuv(a)
(Transi)−1 ● presuv ● Transj maps x to y, as does
to pres xy (a) is the same for all elements a in S1.
pres xy . Assuming commutativity, the transposi-
If S1 =​ S2, presuv = Transi , and presxy = Transj ,
tions Transi and Transj are interval-​preserving,
what is this interval? Express the answer in
and therefore so is the above composition of
terms
   of i and j.
functions, which then must be the same func-
tion as pres xy by the uniqueness condition in (a).
The final theorem of this section summarizes
Even if the interval spaces are commutative, the
functions in a statement such as (i) must still be the symmetrical relationship between transpo-
combined in the order shown: these functions sitions and interval-​ preserving mappings—​ a
are not elements of the interval group and do duality that manifests itself in several ways.
not commute with each other. Though some of the properties enumerated in
The Ginastera analysis from Figure 7.5.15 the theorem are by now familiar, we have not yet
provides simple illustrations of some of the prop- seen any examples demonstrating the full extent
erties in Theorem 7.5.20. One can get from any of this duality, so the theorem may seem rather
note to any other note in its network by some abstract at this point. Readers eager for exam-
combination of the interval-​preserving mappings ples will find their patience rewarded in Section
shown; the mappings combine as specified in part 8.3, where the mappings involved will be triadic
(c) of the theorem, and arrows may be traversed transformations; one of the two dual groups is
in reverse using (d). The following exercise offers a the familiar transposition-​inversion group 𝒯ℐ12,
few illustrations of the later parts of the theorem. while the other comprises Riemannian transfor-
mations such as P, L, and R. Another example,
Exercise 7.5.22 involving descriptions of twelve-​tone rows as
(a) Label pitches in pent(−6) using labu, permutations, will be encountered in Section 9.6.
where u is the reference pitch D♭4, and For any interval space (S, G, int), let 𝒯S denote
label pitches in dpitch(0) as usual with the set {Transi | i ∈ G} of all transposition opera-
respect to v =​C4. Let x and y be G♭3 and B3 tors on S, and let 𝒫S be the set {presuv | u, v ∈ S}
respectively, so that pres xy is the cross-​type of all interval-​preserving mappings on S. Both 𝒯S
mapping shown in Figure 7.5.15b. Let a be and 𝒫S are subsets of 𝒮S, the group of all bijective
G♭4, the last note in the lowest voice in the transformations from S to S; in fact, the theorem
figure, so that pres xy (a) is G4, the last note shows that they are both subgroups of 𝒮S.39

284 • E x p loring M usical S p aces

Theorem 7.5.24: Duality of transpositions Part (c) illustrates another aspect of the
and interval-​preserving mappings symmetry: when either the transformation
Suppose (S, G, int) is an interval space, with f(i) =​Transi or gu(i) =​ presuv is applied to an ele-
𝒯S, 𝒫S, and 𝒮S defined as above. ment x of S, the u-​label of x is adjusted by the
addition of the interval i, but Transi adds i to the
(a) 𝒯S is a transformation group on S right of labu(x), while presuv adds i on the left.
isomorphic to G; the mapping f: G → This property of Transi was noted earlier in this
𝒯S defined by f(i) =​Transi is a group section; for the corresponding property of presuv
isomorphism. The action of 𝒯S on S is see Exercise 7.5.26.
simply transitive. The commutativity of Transi and presuv in
(b) 𝒫S is a transformation group on S (d) follows from the definitions of transposi-
isomorphic to G. For any reference point tions and interval-​ preserving functions, as
u in S, define a mapping gu: G → 𝒫S by shown in Exercise 7.5.27a. The theorem does
gu(i) =​ presuv , where v =​Transi(u). Then gu is not say that transpositions always commute
an anti-​isomorphism. The action of 𝒫S on S with other transpositions or that interval-​pre-
is simply transitive. serving mappings always commute with other
(c) Suppose u ∈ S, i ∈ G, f(i) =​Transi and interval-​preserving mappings; indeed, if the
gu(i) =​ presuv as in parts (a) and (b) above, interval group is noncommutative there are
and suppose ⊕ is the group operation in G. always some non-​commuting pairs of both of
Then for every x in S, labu(Transi(x)) =​labu(x) those types. The fact that every Transi com-
⊕ i, while labu(presuv(x)) =​ i ⊕ labu(x). mutes with every presuv implies, in terminology
(d) For every i in G and every u and v in S, from Chapter 6, that the groups 𝒫S and 𝒯S are
Transi ● presuv =​ presuv ● Transi. Within pairwise commuting and that each of them is at
the group 𝒮S, each of 𝒯S and 𝒫 S is the least a subset of the centralizer of the other.
centralizer of the other: 𝒫 S =​Cent𝒮S(𝒯S) The theorem says something stronger: each of
and 𝒯S =​Cent𝒮S(𝒫S). 𝒫S and 𝒯S is exactly the centralizer of the other.
(e) Define a new interval function int~: This means that a transformation on S is inter-
S × S → 𝒫 S so that for all u and v in S, val-​preserving if and only if it commutes with
int~(u, v) =​ presuv . Then (S, 𝒫S, int~) is all transpositions, and is a transposition if and
an interval space, called the dual of the only if it commutes with all interval-​preserving
original interval space (S, G, int). The mappings. Parts (b) and (c) of Exercise 7.5.27
original space (S, G, int) and its dual outline these proofs.
(S, 𝒫 S, int~) are isomorphic. In (S, 𝒫 S, Part (e) of the theorem completes the
int~) the transpositions are the original description of the duality between the groups
functions presuv , and the interval-​ 𝒯S and 𝒫S by showing that it is possible to rede-
preserving mappings are the original fine the interval function on S so that the roles
  functions Transi. of the two groups are exchanged: the functions
in 𝒫 S, originally interval-​ preserving, become
Part (a) of the above theorem, the group the new “transpositions,” while those in 𝒯S, the
structure of transposition operators, is noth- original transpositions, become “interval-​ pre-
ing new, as it restates a part of Lewin’s theorem serving.” Part (e) follows from the earlier parts
(Theorem 5.6.7). It is repeated here for compari- of the theorem together with Lewin’s theorem.
son with the corresponding property of interval-​ The construction of the new interval function is
preserving mappings in (b). The mapping f in given to us directly by part (b) of Lewin’s theo-
(a) is an isomorphism while gu in (b) is an anti-​ rem, which shows how any set acted upon by a
isomorphism. From Section 6.6 we know that if simply transitive transformation group may be
two groups are anti-​isomorphic, they are auto- made into an interval space whose transposi-
matically isomorphic. In the case of the groups tion operators are the transformations in the
G and 𝒫S it happens to be easier to construct an given group. In the present situation we use the
anti-​isomorphism than an isomorphism; there transformation group 𝒫S (whose action on S is
is actually a whole family of anti-​isomorphisms, simply transitive by (b)) for this purpose. The
one function gu for each reference point u of the elements of 𝒫S then automatically become the
space.40 Exercise 7.5.25 below outlines a proof “transpositions” in the new space. By part (d) of
that gu is an anti-​isomorphism. the present theorem applied to the new space,

Intervals • 285
the new “interval-​ preserving” mappings are Exercise 7.5.27 This exercise outlines the proof
the transformations that commute with these of Theorem 7.5.24d.41
“transpositions” (elements of 𝒫 S)—​but from
part (d) applied to the original space we know (a) Given functions Transi and presuv , take
any element a in S. Let b = presuv( a), let
that these 𝒫 S-​commuting transformations are
x =​Transi(a), and let y = presuv ( x ). Use the
just the elements of 𝒯S. Lewin’s theorem tells
interval-​preserving property of presuv to
us that the functions in any simply transitive
show that Transi(b) =​ y, and explain why it
transformation group can become the trans- follows that Transi and presuv commute.
positions in an interval-​space structure on the (b) To show that interval-​preserving mappings
group; the present theorem tells us that those are the only transformations in the
same functions may, if we define the interval-​ centralizer of 𝒯S, suppose g: S → S
space structure in the dualized way, become the commutes with every transposition
interval-​preserving mappings instead. on S; we wish to show that g is interval-​
As with several other results in this section, preserving. To do this, suppose int(u, v) =​ i;
Theorem 7.5.24 is of interest primarily in the use the fact that g commutes with Transi to
noncommutative case. If the interval group G is show that int(g(u), g(v)) =​ i.
commutative, then transpositions and interval-​ (c) Finally, to show that transpositions are the
preserving mappings are the same thing; they all only transformations in the centralizer of
commute with each other; no other transforma- 𝒫S, suppose f: S → S commutes with every
tions commute with all of them; adding i on the interval-​preserving mapping on S; we wish
right is the same as adding i on the left; and the dual to show that f is a transposition. Choose
space is essentially the same as the original space. any u in S and let i =​int(u, f(u)); the goal is
to show that f =​Transi. For any v in S, use
Exercise 7.5.25 To show that the function the interval-​preserving property of presuv
gu: G → 𝒫S in Theorem 7.5.24b is an anti-​ and the fact that pres uv commutes with f
homomorphism, we must show that if i and j to show that int(v, f(v)) =​ i and therefore
are elements of G, then gu(i ⊕ j) =​ gu(j) ● gu(i). As   f(v) =​Transi(v).
described in the definition of gu, gu(i) =​ pres uv ,
where v =​Transi(u) or equivalently int(u, v) =​ i; 7.6 INVERSION OPERATORS
similarly, gu(j) =​ pres uw, where int(u, w) =​ j; and AND INTERVAL-​R EVERSING
gu(i ⊕ j) =​ pres uz , where int(u, z) =​ i ⊕ j. MAPPINGS
(a) Use basic interval properties to show that In extending the concept of transposition opera-
int(v, z) =​ j. tors to general interval spaces in the preceding
(b) Use the interval-​preserving property of section, we formulated two kinds of interval-​
pres uv to show that int( v, presuv( w )) = j. space homomorphisms: Transi (or Transuv ), a
(c) Using the results of (a) and (b) together straightforward generalization of the familiar
with Theorem 7.5.20g, show that transposition operators Tn on pitch and pitch-​
presuv = preswz . class spaces, and pres uv , based on the interval-​
(d) Use Theorem 7.5.20c to complete the proof preserving property. When the interval spaces
involved are commutative, the two formulations
that gu(i ⊕ j) =​ gu(j) ● gu(i).
are equivalent. The situation involving inversion
(e) Complete the proof of Theorem 7.5.24b by
operators is similar up to a point: below we pres-
explaining why the group mapping gu:
ent a generalized definition of inversion opera-
G → 𝒫S is one-​to-​one and onto, and why tors Invi (or Inv uv ) analogous to the familiar In,
  the action of 𝒫S on S is simply transitive. and also a definition of interval-​reversing map-
pings. If the interval spaces are commutative all
Exercise 7.5.26 In the situation described in is well: inversions and interval-​reversing map-
Theorem 7.5.24c, where gu(i) =​ presuv , use the pings turn out to be the same thing, and they are
interval-​preserving property of presuv to show interval-​space homomorphisms.
that if z =​ presuv( x ), then int(v, z) =​labu(x). Then In the noncommutative case, however, the
use the interval-​sum equation to show that potential for confusion turns out to be much
u( presu ( x )) =​ i ⊕ labu(x).
lab   v
greater for inversions than for transpositions.

286 • E x p loring M usical S p aces

In a noncommutative space there are two equally reference, we refer to these two properties as
reasonable but inequivalent ways to define Inv uv, inversion condition 1 and inversion condition 2:
neither of which is assured of being an
interval-​ space homomorphism, and neither Inversion condition 1 for Inv[u]‌i(x) =​ y:
of which is interval-​reversing. In fact there int(u, x) ⊕ int(u, y) =​ i
are no interval-​reversing mappings at all in Inversion condition 2 for Inv[u]i(x) =​ y:
a noncommutative space. For these reasons, int(u, y) ⊕ int(u, x) =​ i
and because the large majority of the interval
spaces studied in this book are commutative, For a given u, i, and x, each of the inversion
we will assume commutativity in much of this conditions is satisfied for a unique y. Assuming
section, deferring most of our discussion of the commutativity, the same y satisfies both; hence
peculiarities of the noncommutative case until Inv[u]i: S → S is a single well-​defined function in
the end of the section. Some results are pre- the commutative case. Moreover, for a given u, i,
sented without complete proofs. Because we and y, the conditions are satisfied for a unique x,
will often need to consider the same intervals which means that the function Inv[u]i is one-​to-​
forward and backward, we remind the reader one and onto.
of the inverse interval property from Section Let v be the element Transi(u), so int(u, v) =​ i.
5.4: for any elements x1 and x2 in an interval According to the interval-​sum equation, int(u, x)
space, int(x2, x1) =​ ⊖int(x1, x2). ⊕ int(x, v) =​int(u, v) =​ i. Comparing this equa-
As usual, it is helpful to keep pitch-​class space tion with inversion condition 1 above, we see
in mind as a familiar example. In pc, one dif- that both equations take the form int(u, x) ⊕ j =​ i;
ference between transpositions and inversions in one case j is int(u, y) while in the other case j
is that the numbering of inversions In depends is int(x, v). As discussed in Section 5.3, such an
on the arbitrary choice of C as a referential pitch equation can be solved uniquely for j (the solution
class. The transposition T10 transposes pitch is j =​ ⊖int(u, x) ⊕ i), so it follows that int(u, y) =​
classes by ten semitones, but the inversion I10 int(x, v). That is, in order to satisfy condition 1,
has no inherent “10-​ness.” Instead, as detailed in Inv[u]i(x) must be the unique element y of S for
Section 5.5, the index number 10 locates the axis which int(u, y) =​int(x, v), where v =​Transi(u).
of inversion relative to C: two pitch classes are Similar reasoning applies to inversion con-
related by I10 if their pc numbers—​their labels dition 2, the equation int(u, y) ⊕ int(u, x) =​ i.
with respect to C—​sum to 10 in ℤ12. If we chose By the interval-​sum equation again, int(u, y) ⊕
D rather than C as the basis for pc numbering, int(y, v) =​int(u, v) =​ i. Comparing these two
the indexing of the inversion operators would equations shows, as above, that int(y, v) must
change, and the name I10 would refer to a differ- be equal to int(u, x), or equivalently int(v, y) =​
ent transformation. ⊖int(u, x) =​int(x, u). So condition 2 implies that
In the general setting of an interval space Inv[u]i(x) is the unique y in S for which int(v, y) =​
(S, G, int), analogously, the definition of Invi int(x, u), where v =​Transi(u).
depends on the choice of a reference point u of Both of these properties are illustrated in
the space. To make the dependence on u clear Figure 7.6.1a. The heavy arrow depicts the
when necessary, we write the function as Inv[u]i. function f =​Inv[u]i. The arrangement is simi-
If u ∈ S, i ∈ G, and x ∈ S, then Inv[u]i(x) =​ y should lar to that of Figure 7.5.7a illustrating Transi,
be true whenever labu(x) ⊕ labu(y) =​ i, or equiva- but the elements v (=​Transi(u)) and y (=​ f(x))
lently int(u, x) ⊕ int(u, y) =​ i. By this definition, have exchanged places. The equation int(u,
the usual inversion operator In in pc is Inv[C]‌n for y) =​int(x, v), derived above from condition 1,
each n in ℤ12. says that the two horizontal intervals in the
If commutativity is assumed, the equation figure are equal; the second property, int(y,
labu(x) ⊕ labu(y) =​int(u, x) ⊕ int(u, y) =​ i is equiv­ v) =​int(u, x), derived from condition 2, says
alent to labu(y) ⊕ labu(x) =​int(u, y) ⊕ int(u, x) =​ i. that the vertical intervals are equal. In a com-
This is the first sign of a complication in the mutative interval space, the two properties
noncommutative case. In the absence of com- are equivalent and the inversional relation-
mutativity, the two orderings of the sum are not ship y =​Inv[u]i(x) ensures that both of them
necessarily equal, and it is not clear which one are true.
should be used in the definition of inversion. What happens if we apply the function Inv[u]i
Even in the commutative case, each ordering to the reference point u itself? If we substitute
has its own distinctive consequences. For future x =​ u in either condition 1 or condition 2, the

Intervals • 287
FIGURE 7.6.1 Interval relationships in a commutative interval space: (a) under the inversion Inv[u]i;
(b) under an interval-​reversing mapping f

term int(u, u) appears, which of course is 0 to which we will work our way in the course of
(the identity element of the interval group), and the discussion below.
the condition reduces to int(u, y) =​ i. Because
int(u, v) =​ i, it follows that in this case y =​ v; in Exercise 7.6.2
other words, Inv[u]i(u) =​ v =​Transi(u). We therefore (a) Suppose u and w are two different reference
introduce the notation Inv uv for the function Inv[u]i, points in a commutative interval space
where i =​int(u, v). We have just established that S. Let j =​int(w, u), and let k =​ j ⊕ j. Show
Inv uv : S → S maps u to v, consistent with the simi- that for every interval i and every x in S,
lar notations for the functions Transuv and pres uv . Inv[u]i(x) =​Inv[w]i⊕k(x), and therefore Inv[u]i
In pitch-​class space, pitch classes x and y are and Inv[w]i⊕k are the same function.42
related by I10 (the inversion we are now calling (b) Suppose u, v, w, and z are reference points
Inv[C]‌10) if their labels relative to the reference in a commutative interval space S. Let
point u =​C sum to 10 in ℤ12. The property Inv[u]i(u) i =​int(u, v), j =​int(w, u), and k =​ j ⊕ j. Show
=​ v derived above says that Inv[C]10(C) =​B♭, the pc that the functions Inv vu and Inv wz are the
whose interval from C is 10. Therefore Inv[C]10 same if and only if int(w, z) =​ i ⊕ k.
could also be called Inv CB♭ . But these are not the (c) Use the results of (a) and (b) to verify that
only names for this inversion. The inversion that Inv[C]‌10, Inv[D]6, Inv[B♭]2, Inv[F]0, Inv CB♭, Inv DA♭,
maps C to B♭ also maps D to A♭, B♭ to C, and F to Inv CB ♭ , and Inv FF are all the same function
F (among other things), so we might expect that   on pitch-​class space.
♭
Inv BC is the same function as Inv DA♭ , Inv BC♭ , and
Inv F . Because int(D, A ♭) =​6, int(B♭, C) =​2, and
F

int(F, F) =​0, these are by definition the same as The notation Inv FF shows explicitly that F is
Inv[D]6, Inv[B♭]2, and Inv[F]0, respectively. Indeed, all a fixed point of this inversion (the usual I10).
of these things are the same function as Inv[C]10 =​ Naming a function by a fixed point in this way
♭ can be convenient; such a function generally
Inv CB , but this is perhaps not obvious from
the definitions above, in which functions such takes the form Inv uu or equivalently Inv[u]0, and
as Inv[D]6 and Inv DA♭ are framed in terms of the may be called inversion about u. But not every
comparatively unfamiliar labeling of elements inversion operator arises in this way: recall from
relative to the pc D rather than C. For example, Figure 5.5.4 that in pitch or pc only the even-​
Inv[D]6(C) =​B♭ is a true statement because labD(C) numbered inversion operators have fixed points.
+​ labD(B♭) =​10 +​8 =​6 (mod 12). Also, the notation Inv FF disguises the fact that in
Two questions arise: Under what conditions pc, this inversion has a second fixed point (B).
is Inv[u]i, the inversion with reference point u and
interval i, the same function as Inv[w]j, the inver- Exercise 7.6.3 Refer back to the pictures
sion with reference point w and interval j? And of pitch space in Figure 7.1.1. Let u =​C4 and
under what conditions is Inv vu , the inversion with v =​G4. We remarked previously that the
reference points u and v, the same function as interval-​preserving transformation presvu maps
Inv wz, the inversion with reference points w and z? every pitch in 7.1.1a to the correspondingly
The following exercise answers the first question numbered pitch in 7.1.1b. Construct another
and provides a preliminary answer to the second. labeling of pitch space so that the inversion
The second question has a more elegant answer, Inv vu maps every pitch in 7.1.1a to the

288 • E x p loring M usical S p aces

correspondingly numbered pitch in your figure. The label-​ reversing property labv(f(x)) =​
This inversion is the same as Inv[u]i for some i, ⊖labu(x) is a special case of the interval-​revers-
and it is also Inv[v]j for some j. What are i and j? ing property, requiring only that f reverse inter-
Does
   this inversion have a fixed point? vals int(u, x) with the starting point u, but saying
nothing about any other intervals. Exercise 7.5.6
Exercise 7.6.4 In a commutative interval space, examined the analogous case of u–​v label-​pre-
   that if Inv[u]i(x) =​ y, then Inv[u]i(y) =​ x.43
show serving functions, concluding in that situation
that the special implied the general: a u–​v label-​
Exercise 7.6.4 implies that Inv[u]i ● Inv[u]i =​ ES, preserving function is completely interval-​pre-
the identity mapping on the space S: in a com- serving. That conclusion required no assumption
mutative interval space, every inversion is an of commutativity. The situation with label-​and
involution. Of course, this property is familiar in interval-​reversing functions is not so agree-
pitch-​class space; recall Theorem 5.5.6g. able: in a noncommutative space, it turns out,
In pitch-​class space, transpositions preserve u–​v label-​reversing functions exist, but they are
intervals while inversions reverse them. For not fully interval-​reversing. In a commutative
example, consider the pitch classes C♯ and E, for interval space, however, a u–​v label-​reversing
which int(C♯, E) =​3. Applying T4 to both pcs gives function is indeed interval-​reversing. This can
T4(C♯) =​F and T4(E) =​G♯, and the interval between be proved directly (see Exercise 7.6.5 below),
the transformed pcs is still the same: int(F, G♯) =​3. but it follows also from other principles we
But if we apply I4 instead, we get I4(C♯) =​E♭ and have established. In terminology from Section
I4(E) =​C, and the new interval is int(E♭, C) =​9, the 7.2, the interval-​ reversing property int(f(x1),
inverse (in ℤ12) of the original interval. f(x2)) =​ ⊖int(x1, x2) says that an interval-​revers-
Generally, a function f: S → S on an interval ing mapping on an interval space (S, G, int) is
space (S, G, int) is interval-​reversing if int(f(x1), interval-​compatible with the inverse mapping
f(x2)) =​ ⊖int(x1, x2) for all x1 and x2 in S, or equiva- g(i) =​ ⊖i on the interval group G. As noted at the
lently (by the inverse interval property) int(f(x1), end of Section 6.6, an inverse mapping g(x) =​ x−1
f(x2)) =​int(x2, x1). This property is illustrated in is generally an anti-​automorphism of a group,
Figure 7.6.1b, where the elements x1, x2, f(x1), and not an automorphism; this is because the inverse
f(x2) are renamed u, x, v, and y, respectively, to clar- of xy is not x−1y−1 but rather y−1x−1, reversing the
ify the similarity with the inversion Inv[u]i =​ Inv vu in order of the product (that is, g(xy) =​ g(y)g(x)).
part (a). The equality int(y, v) =​int(u, x) between In a commutative group, however, x−1y−1 and
the two vertical intervals in (a), derived from y−1x−1 are the same thing, homomorphisms
inversion condition 2, appears in (b) in the form and anti-​homomorphisms are the same thing,
int(v, y) =​ ⊖int(u, x), or in other words int(f(u), and the inverse mapping is an automorphism. In
f(x)) =​ ⊖int(u, x): the function has reversed the a commutative interval space, therefore, every
interval, and the arrow has reversed direction. The interval-​reversing mapping is an interval-​space
equation may be written using label functions as automorphism.
labv(y) =​ ⊖labu(x): the v-​label of y is the inverse of Moreover, it follows from the interval-​space
the u-​label of x. In these last two formulations, homomorphism theorem (Theorem 7.2.7) that
the inverse symbol ⊖ is the only thing that distin- for any u and v in S, there is a unique interval-​
guishes the interval-​reversing property from the space automorphism f: S → S, mapping u to v,
interval-​preserving property studied previously. compatible with the inverse mapping g: G → G—​
For any given u and v, the discussion above which is to say that f is interval-​reversing. That
shows that the function f =​ Inv vu satisfies the equa- automorphism, called g uv in Section 7.2, must be
tion labv(f(x)) =​ ⊖labu(x) for every x. A function u–​v label-​reversing. But we observed above that
satisfying this property is called u–​v label-​revers- the only u–​v label-​reversing transformation on S
ing. Because labu(u) =​0, a u–​v label-​reversing is Inv vu. Hence it follows, in a commutative inter-
function must satisfy labv(f(u)) =​0, and there- val space S, that g uv =​ Inv vu; that Inv vu is both the
fore f(u) =​ v. The label-​reversing property tells us unique u–​v label-​reversing transformation on S
exactly what f(x) must be for every x in S: it must and the unique interval-​reversing transforma-
be the unique y such that labv(y) =​ ⊖labu(x). For tion on S that maps u to v; and, more generally,
a given u and v, therefore, there is one and only that inversions and interval-​reversing transfor-
one u–​v label-​reversing function, and that func- mations are precisely the same thing. In noncom-
tion must be Inv vu. mutative spaces these statements are false.

Intervals • 289
Exercise 7.6.5 Suppose f: S → S is a u–​v to cross-​type mappings from one interval space
label-​reversing function in a commutative (S1, G1, int1) to another space (S2, G2, int2). The
interval space. Prove directly that f is interval-​reversing property int2(f(x1), f(x2)) =​
interval-​reversing by writing a chain of ⊖int1(x1, x2) relates an int1 interval to an int2
equalities, beginning with interval; even if these intervals are in differ-
ent spaces, they can potentially be equal if the
int(f(x1), f(x2)) interval groups G1 and G2 are the same, or if G1
=​ int(f(x1), f(u)) ⊕ int(f(u), f(x2)) is a subgroup of G2. Interval-​reversing mappings
are applicable in some cross-​type situations in
(which follows from the interval-​sum
which inversions are not: the definition of Inv vu,
equation)
   and ending with ⊖int(x1, x2).44
like that of Transvu, requires u and v to inhabit
the same space, and therefore applies only in the
Our general work with interval-​space homo-
more restrictive case in which the spaces S1 and
morphisms in Section 7.2 now provides the
S2 are the same, or S1 is a subspace of S2.
elegant answer to the question raised above: in
Generally, we write Rev uv , or more explic-
a commutative interval space, under what condi-
itly Rev vu∈∈SS21, for a cross-​type interval-​reversing
tions are Inv vu and Inv wz the same function? By
mapping. This notation presupposes commuta-
Theorem 7.2.13e, if g: G1 → G2 is a group homo-
tive interval spaces, for reasons just outlined.
morphism, then g wz = g uv if and only if g uv( w ) = z.
If u ∈ S1, v ∈ S2, and the commutative interval
In the present situation, g is the inverse map-
groups satisfy G1 ⊆ G2, then Rev uv exists and
ping, an automorphism of G, and g uv is Inv vu .
is an interval-​space homomorphism; it is also
So whenever Inv vu ( w ) = z, we may conclude that
the unique interval-​reversing function from S1
Inv vu = Inv wz. Only one inversion can map a par-
to S2 that maps u to v. If the spaces S1 and S2
ticular w to a particular z; if two differently
are the same, then Rev uv is the same thing as
named inversions do so, they are merely two dif- Inv vu . If the spaces are different, then the cross-​
ferent names for the same function. Moreover,
type Rev uv is not strictly speaking an inversion,
because we have established that Inv vu ( v ) = u, it
but we may occasionally call it an “inversion,”
follows that in a commutative interval space,
in quotation marks, just as cross-​type inter-
Inv vu and Inv uv are the same function. (This prop-
val-​preserving mappings are sometimes called
erty may seem obvious, but in a noncommutative
“transpositions.”45
space it is not true.) In pitch-​class space, there-
fore, simply from knowing that the inversion
Inv BC♭ maps D to A♭, we can conclude that Inv BC♭ Example 7.6.6: Bach, Fugue in G Major
and Inv DA♭ are the same function. (WTC I)
The definition of interval-​ reversing map- In tonal music, diatonic inversions in some
pings, like that of interval-​ preserving map- space dpitch(n) or dpc(n) arise more commonly
pings, may in some circumstances be extended than chromatic inversions in pitch or pc.

FIGURE 7.6.7 Scalar inversions and interscalar “transpositions” in Bach, WTC I, Fugue in G Major

290 • E x p loring M usical S p aces

When more than one diatonic scale is involved, What are the fixed points of the other
cross-​type mappings can be useful. The network two inversions? Explain why an inversion
in Figure 7.6.7 relates six appearances of the operator in a mod-​12 pitch-​class space can
subject in the G-​Major Fugue from WTC I. The have either no fixed points or two fixed
analysis is carried out in seven-​note pitch-​ points, while every inversion operator in
class spaces sharing the interval group ℤ7, a mod-​7 pitch-​class space has exactly one
beginning with dpc(+​1) at the lower left. (A fixed point.47
pitch-​space analysis, starting in dpitch(+​1) (b) If the analysis in Figure 7.6.7 were carried
instead, would be very similar.) The real answer out in pitch space rather than pitch-​class
in mm. 5–​7, shown at center left, transposes space (dpitch(+​1) rather than dpc(+​1), and
the subject from mm. 1–​3 by the usual T7 in pc, so on), would every inversion operator have
but is described here as an interval-​preserving   one fixed point?
mapping, or “transposition,” from dpc(+​1) to
dpc(+​2)—​that is, from G major to D major. As Example 7.6.10: Brahms, Symphony
reference points in the presvu notation, we use G No. 1, III
and D, the initial notes of the excerpts, but of The inversions in Figure 7.6.7 operate
course there are other options. strictly within their respective mod-​7 spaces.
A counterexposition in inversion begins Some interval qualities change when the fugue
at m. 20, whereupon the inverted subject subject is inverted, so the transformations are
appears in the same two diatonic scales heard not inversions in chromatic space. The opening
previously. Each inverted subject is related to clarinet melody from the Allegretto of Brahms’s
the corresponding recto subject from the first First Symphony, shown in Figure 7.6.11, offers
exposition by a diatonic (intrascalar) inversion an example of an interscalar diatonic “inver-
operator, indicated by a horizontal Inv vu arrow, sion” that is also a chromatic inversion. The
and the two inverted subjects are related to each measure phrase, starting on E♭ in
first five-​
other by the interscalar “transposition” presDA as dpc(−4) (A♭ major), is immediately inverted in
shown. Finally, the top of the figure shows two dpc(−3), starting on C as shown. The interscalar
subject entries in mm. 38–​40 and 43–​45, where interval-​reversing mapping shown in the fig-
a recto statement in E minor is answered by an ure accurately describes the relationship—​but
inverted statement in the same key. Both of in this case every interval quality is preserved
these entries conform to the E harmonic minor in the inversion. That is, the transformation
scale, labeled hmin(+​2) (a name to be explained reverses not only diatonic intervals but also
in Chapter 14). These subjects are related to chromatic intervals, so despite the diatonic
the earlier diatonic subjects by cross-​type nature of the material the relationship is cor-
“transpositions,” and to each other by a scalar rectly modeled by the inversion operator I3 in
inversion
   within hmin(+​2). chromatic space.
Diatonic inversions that are also chromatic
Exercise 7.6.8 In Figure 7.6.7, two arrows inversions are rare in tonal music, at least among
(from mm. 1–​3 to mm. 5–​7 and from mm. passages of more than a few notes. Recall that
20–​22 to mm. 24–​26) represent cross-​type a diatonic collection, as a subset of chromatic
“transpositions” from dpc(+​1) to dpc(+​2). space, is symmetric about its dorian tonic. Of
The reference points shown for these two
cross-​type mappings presvu are different. Are
they actually different functions, or are they
different names for the same function? Are the
two interval-​preserving mappings shown from
dpc(+
   ​2) to hmin(+​2) the same function?46

Exercise 7.6.9
(a) Inspection of Figure 7.6.7 shows that
each of the three inversions in the figure
has exactly one fixed point. For example,
the inversion Inv GD in dpc(+​1) has B as a FIGURE 7.6.11 An interscalar diatonic “inver-
fixed point, so Bs occur in corresponding sion” or chromatic inversion in Brahms, Symphony
places in mm. 1–​3 and in mm. 20–​22. No. 1 in C Minor, Op. 68, III

Intervals • 291
the seven possible intrascalar inversions of a holds for the interscalar “inversion” in
diatonic pitch-​class space, the only one that pre- Figure 7.6.11, and verify that it does not
serves all interval qualities is the inversion about hold for either of the two scalar diatonic
that note. An interscalar “inversion” from one inversions in Figure 7.6.7.
diatonic scale to another will preserve all inter- (b) The third inversion in Figure 7.6.7 takes
val qualities only if the dorian tonic of the first place within a harmonic minor scale. No
scale maps onto the dorian tonic of the second. scalar inversion of this space can be a
That is precisely what happens in Figure 7.6.11:   chromatic inversion. Why not?
the mapping is identified here as RevEC♭ based on
the initial notes of the phrases, but it also maps Example 7.6.13: Szymanowski, Etude,
B♭, the dorian tonic for dpc(−4), to F, the dorian Op. 33, No. 3
tonic for dpc(−3). The third of Karol Szymanowski’s
Another way to model the mapping in this Twelve Etudes, Op. 33 (1916), like the
example is by using an inversion operator in Ginastera example considered previously
Rings’s product space sd × pc. Doing so requires in Example 7.5.14, opposes the white-​key
identifying a tonic for the second phrase. C and diatonic collection in one hand with the black-​
E♭ are both plausible choices; if we choose C, key pentatonic in the other, but in this work
the applicable inversion is Inv vu, indexed by u =​ gestures in the two hands often mirror each
(5̂, 3) and v =​(1̂, 0), the starting notes of the other. Figure 7.6.14a shows a representative
two phrases (E♭ as scale degree 5̂ and C as 1̂). two-​measure passage.48 The analysis in
The first interval in m. 1 is (6, 10): a descending (b) shows relationships between the top voice
step in mod-​7 scale-​degree space, and a descend- in the right hand and the middle voice in the
ing whole step in mod-​12 pitch-​class space. The left. (In the first measure the bass moves more
second interval is (6, 11), only a semitone. An slowly than the other voices; otherwise any
interval in this space is inverted by inverting of the three voices in the left hand could be
both its mod-​7 and mod-​12 components; the used in this analysis.) As with the Ginastera
inverses of (6, 10) and (6, 11) are (1, 2) and (1, 1) example, the analysis must be carried out
respectively, the first two intervals in the sec- in the pitch spaces S1 =​ dpitch(0) and
ond phrase. Had we chosen E♭ rather than C as S2 =​ pent(−6), not in the corresponding pitch-​
the tonic for the second phrase, the scale-​degree class spaces, because the scales are of different
numbers and the notation for the inversion cardinalities.
operator would change, but the intervals within The analysis takes four-​note cells as the units
each of the two phrases would not. Because an of transformational action. Each such cell in the
inversion in sd × pc must operate in both com- right hand mirrors the simultaneous cell in the
ponents, such an analysis is possible only with left via a cross-​type “inversion” Rev uv from S1 to
an inversion that works both diatonically and S2. In three of the four cells, this transformation
chromatically. amounts to nothing more than reversing the
contour and flatting the notes (transforming
B–​A to A♭–​B♭ in the first cell, for example). The
Exercise 7.6.12
third cell is of particular interest, however,
(a) Inversions that function both diatonically because the diatonic step G–​F is inverted to
and chromatically may be conveniently the pentatonic “step” E♭–​G♭, a minor third.
described using spelled pitch-​class numbers. Horizontally within each voice, successive cells
The dorian tonic is the central note of a are shifted stepwise within their respective
diatonic collection along the line of fifths, scales (the contour of the figure being modified
and its spc number is the index number between the second cell and the third); the
of the collection. If the referential notes u two right-​hand cells in the second measure are
and v of a cross-​type interval-​reversing related by the pentatonic transposition t1P but
mapping Rev vu∈∈DpC ( n)
DpC ( m ) from dpc(m) to dpc(n) not by chromatic transposition. These stepwise
are identified by their spc numbers, the shifts move in the same direction in both
mapping will be a true chromatic inversion hands, so that for example diatonic t−1 occurs
if and only if u and v are symmetrically simultaneously with pentatonic t−1P, not with
situated relative to the dorian tonics of t1P. The “inversions” Rev uv operate only at a local
their respective scales, which is true if and level, on no more than four notes at a time; at
only if u +​ v =​ m +​ n. Verify that this formula the  larger scale the hands move in parallel.

292 • E x p loring M usical S p aces

FIGURE 7.6.14 Scalar transpositions and interscalar “inversions” in Szymanowski, Etude, Op. 33, No.
3, mm. 7–8

The following theorem summarizes many unique u–​v label-​reversing mapping from
properties of interval-​reversing mappings and S1 to S2, and it is also the unique interval-​
inversions. Section 7.5 presented similar prop- reversing mapping from S1 to S2 that maps
erties of transpositions and interval-​preserving u to v. If G1 =​ G2, then Rev vu: S1 → S2 is an
mappings in two separate theorems (Theorems interval-​space isomorphism, induced by the
7.5.19 and 7.5.20), but here we assume com- inverse mapping on G1.
mutativity, so that interval-​ preserving map- (b) Relationship with inversion. If S1 =​ S2 and
pings are the same things as inversions, and we int1 =​int2, then Rev vu is the same function as
combine the results in a single theorem. For the the inversion operator Inv vu ,
greatest generality in the cross-​type setting the which is the same as Inv[u]i: S1 → S1, where
theorem primarily uses the notation Rev vu, but i =​int1(u, v). If S1 is a subspace of S2, then
in the case of mappings within a single space, Rev vu( x ) = Inv vu( x ) =​Inv[u]i(x) for all x in
as clarified in part (b), Rev uv is the same func- S1, where i =​int2(u, v); that is, Rev vu is
tion as Inv vu . the same as the inversion operator Inv vu,
defined only on the elements of S1.
Theorem 7.6.15: Interval-​reversing (c) Compositions of interval-​preserving and
mappings and inversions in commutative interval-​reversing mappings. For every u in
spaces S1, v in S2, and w in S3,
In the following statements, (S1, G1, int1), presvu ● presvw =​ presuw ,
(S2, G2, int2), and (S3, G3, int3) are interval
Rev vu ● presvw =​ Rev uw ,
spaces whose interval groups are commutative
and satisfy G1 ⊆ G2 ⊆ G3; the symbol ⊕ denotes presvu ● Rev vw =​ Rev uw ,
the group operation in G2, and ⊖i is the inverse and Rev vu ● Rev vw =​ presuw .
of i. The letters u, x, a, and b always denote (d) Compositions of transpositions and inversions.
elements of S1; v and y are elements of S2; w is If S1 =​ S2 and int1 =​int2, then for every u in
an element of S3. S1 and every i and j in G1,
(a) Definition and basic properties. For every Transi ● Transj =​Transi⊕j,
u in S1 and v in S2, Rev vu∈∈SS21 : S1 → S2 is the Inv[u]i   ● Transj =​Inv[u]i⊕j,
interval-​space homomorphism mapping Transi ● Inv[u]j   =​ Inv[u](⊖i)⊕j,
u to v, induced by the inverse mapping
and Inv[u]i   ● Inv[u]j    =​ Trans(⊖i)⊕j.
g(i) =​ ⊖i from G1 to G2. For every x in S1,
Rev vu∈∈SS21( x ) is the unique element y of S2 such (e) Inverses of interval-​reversing isomorphisms.
that int2(v, y) =​ ⊖int1(u, x), or equivalently If G1 =​ G2, then for every u in S1 and v
such that labv(y) =​ ⊖labu(x). Rev uv is the in S2, (Revuv )−1 =​ Rev uv ; this function is an

Intervals • 293
 interval-reversing isomorphism from S2 to Exercise 7.6.17
S1. If S1 =​ S2 and int1 =​int2, then Rev vu =​ Draw a picture similar to Figure 7.5.21 to
Rev uv =​ Inv vu =​ Inv uv ; this function is an illustrate
   Theorem 7.6.15i.
involution.
(f) Label functions for interval-​reversing Exercise 7.6.18 Return to the Brahms analysis
mappings. Suppose labu: S1 → G1 and labv: in Example 7.6.10. Taking the tonics of the two
S2 → G2 are the label functions associated scales, A♭ and C, as the reference points u and
with u and v. Then for every x and a in S1 v, and the initial notes of the two phrases, E♭
and every y in S2, and C, as x and y, show how the relationship
between the final notes of the two phrases, B♭
labv(Revxy( a)) and  F, illustrates Theorem 7.6.15f.
=​ labv(y) ⊕ labu(x) ⊕ (⊖labu(a)).
Exercise 7.6.19
(g) Intervals formed by interval-​reversing
(a) Return to the Szymanowski analysis in
mappings. For every u, x, a, and b in S1 and
Example 7.6.13. Take the interscalar
for every v and y in S2,
mappings of the first and third four-​note
int 2(Revuv( a),Revxy( b )) cells as Rev vu and Revxy . That is, take u =​B3
=​ int1(u, a) ⊕ int2(v, y) ⊕ int1(b, x) and x =​G3 in dpitch(0), and take v =​A♭4
and y =​E♭4 in pent(−6). Verify that the first
and int 2(Revuv( a),Revxy( a )) interval formula in Theorem 7.6.15g holds
  =​ int1(u, x) ⊕ int2(v, y). when a is the note A3 in the first left-​hand
cell and b is the note F3 in the third left-​
(h) Change of reference points. For every u and x hand cell.
in S1 and every v and y in S2, Rev vu =​ Rev xy if (b) Verify that Theorem 7.6.15i holds for the
and only if Revuv ( x ) = y—​that is, if and only   same two mappings Rev vu and Revxy .
if int(v, y) =​ ⊖int(u, x).
(i) Change of reference points (generalized formula). Many of the properties described above go
For every u and x in S1 and every v and y in S2, badly awry if the interval spaces involved are
Rev xy =​(Transi)−1 ● Rev vu ● Transj, not commutative. We conclude this section with
a few observations about the noncommutative
  where i =​int1(u, x) and j =​int2(v, y). case. This discussion is perhaps of more theo-
retical than practical interest, as noncommuta-
Parts (c) and (d) of the above theorem warrant tive interval spaces are not a significant focus of
special comment. If the commutative spaces S1 and attention in this book, and we will not employ
S2 are the same, we know that interval-​preserving inversion operators in any such spaces.
and interval-​reversing mappings are the same thing Let (S, G, int) be an interval space, no longer
as transpositions and inversions, respectively. The assumed commutative. A complication we have
statements in (c) (which follow immediately from already noted is that in the absence of commu-
Theorem 7.2.13a) could be converted into state- tativity, conditions 1 and 2 in the definition of
ments about transpositions and inversions simply inversion are no longer equivalent; a function may
by replacing presvu by Transvu and Rev vu by Inv vu . satisfy one condition but not the other, and either
When these statements are translated into the of the two could reasonably be taken as the basis
notation of Transi and Inv[u]i as in part (d), the for defining inversion. It is convenient to recall
resulting formulas take forms recognizable from the two conditions in the forms below, which
Theorem 5.5.6 as properties of transpositions and were shown above to be equivalent to our origi-
inversions in pitch or pc. It also follows from these nal formulations (proposed as interpretations of
properties (and previously established facts about the statement Inv[u]i(x) =​ y, or Inv vu( x ) =​ y, where
the inverses of Transi and Inv vu ) that the transpo- i =​int(u, v)):
sition and inversion operators on a commutative
interval space always form a group. Inversion condition 1: int(u, y) =​int(x, v)
Inversion condition 2: int(v, y) =​int(x, u)
Exercise 7.6.16 Show how the statements in
part (d) of Theorem 7.6.15 may be derived from Lewin ([1987] 2007, 51) defines his inversion
v
the
  statements in (c).49 operator Iu in a way equivalent to condition

294 • E x p loring M usical S p aces

2, so we will use this as our definition of Inv vu Recall from Section 6.6 that the center of a
in the general noncommutative case. That group G, Cent(G), is the subgroup consisting
is, Inv vu( x ) is the unique y such that int(v, y) =​ of those elements that commute with all ele-
int(x, u), or equivalently int(v, y) =​ ⊖int(u, x). In ments in G.
our terminology this statement says that Inv vu
is u–​v label-​reversing. Theorem 7.6.21: Identity of dual inversions
Condition 2 does not imply condition 1 in For every u and v in an interval space (S, G,
general, so if Inv vu( x ) =​ y, we cannot conclude int), Inv vu =​ Inv uv if and only if the interval
that int(u, y) =​int(x, v). There is a kind of dual- int(u,
   v) belongs to Cent(G).
ism between the two conditions, however, which
makes it easy to see exactly what the function is To prove this theorem, recall that the state-
that satisfies condition 1. The following exercise ment Inv vu( x )=​ y is equivalent to condition 2,
explores this dualism. int(v, y) =​int(x, u), or, reversing both intervals,
int(y, v) =​int(u, x). Use the interval-​sum equa-
Exercise 7.6.20 tion to split the right side of this equation into
(a) Show that Inv vu (u) =​ v and also two intervals:
that Inv vu(v )=​ u.50
int(y, v) =​int(u, v) ⊕ int(v, x)
(b) Let f be the function analogous to Inv vu
(equivalent to Inv vu( x ) =​ y).
defined using condition 1 instead of
condition 2. That is, for fixed reference Meanwhile, Inv uv( x ) =​ y is equivalent to con-
points u and v in S, define f: S → S so dition 1, which says that int(u, y) =​int(x, v), or,
that for every x in S, f(x) is the unique y reversing both intervals, int(y, u) =​int(v, x).
satisfying int(u, y) =​int(x, v). Show that Adding int(u, v) to the right side of both sides
f(x) =​ y if and only if Inv vu( y ) =​ x. Conclude of the last equation gives int(y, u) ⊕ int(u, v) =​
that f is the function (Inv vu)−1, the inverse int(v, x) ⊕ int(u, v), which by the interval-​sum
function of Inv vu.51 equation is equivalent to
(c) For the same function f as in (b), show that
f(x) =​ y if and only if Inv uv( x )=​ y. (Note the int(y, v) =​int(v, x) ⊕ int(u, v)
reversal of the reference points: Inv uv, (equivalent to Inv uv( x ) =​ y).
not Inv vu .) Conclude that f is the Of the last two displayed equations, the first
function Invvu .52 is equivalent to Inv vu( x ) =​ y and the second to
   Inv uv( x ) =​ y, and the two are identical except
To summarize: for elements u and v in a for the order in which the two intervals on the
noncommutative interval space, the inversions right side are added. If int(u, v) and int(v, x)
Inv vu and Invvu may be different. Inv vu satisfies commute, therefore, Inv vu( x ) =​ Inv uv( x ). If the
condition 2, while Invvu satisfies condition 1. interval int(u, v) belongs to Cent(G), then it
Both functions map u to v and v to u, and the commutes with int(v, x) for every x, and in that
two functions are inverses. Inv vu is u–​v label-​ case the functions Inv vu and Invvu are the same.
reversing, which means that it reverses every If, on the other hand, int(u, v) ∉ Cent(G), then
interval involving the element u. Inv vu , analo- there is some interval j such that j ⊕ int(u, v) ≠
gously, is v–​u label-​reversing, which means that int(u, v) ⊕ j. In that case we may choose x so that
it reverses every interval involving the element int(v, x) =​ j. For that x, Inv vu( x ) and Inv uv( x ) will be
v. Neither Inv vu nor Invvu is interval-​reversing different, so the functions are different.
in general.
The two short theorems below restate results Theorem 7.6.22: Impossibility of interval-​
of Lewin.53 It is reasonable to wonder whether reversing mappings in noncommutative
Inv vu and Invvu can ever be the same function even spaces
if the interval group is noncommutative. In fact Suppose (S1, G1, int1) and (S2, G2, int2) are
this is possible, and Theorem 7.6.21 describes interval spaces, with G1 ⊆ G2. If an interval-​
exactly when it happens. Theorem 7.6.22 shows, reversing mapping f: S1 → S2 exists, then G1 is a
as promised previously, that interval-​reversing commutative group. In particular, if an interval-​
mappings cannot exist in a noncommutative reversing mapping f: S → S on a single interval
interval space. space
   S exists, the interval space is commutative.

Intervals • 295
 interval group 𝒯12 (isomorphic to the interval
space pc with interval group ℤ12). The mapping
C → C, D♭ → B, D → B♭, … defines an interval-​
reversing transformation on this commutative
space. This mapping may be considered to map S1
into the larger noncommutative interval space S2
comprising the 24 major and minor triads with
interval group 𝒯ℐ12, but there is no possible way
of extending this mapping to an interval-​revers-
FIGURE 7.6.23 Proof that an interval-​reversing ing transformation on the entire larger space.
mapping implies commutativity
NOTES
The proof of Theorem 7.6.22 is illustrated in 1. Label functions of this type were introduced
Figure 7.6.23. Assume that f: S1 → S2 is interval-​ by Lewin in his original development of Generalized
reversing, and let i and j be any elements of G1; Interval Systems (Lewin [1987] 2007, 31–​32). Lewin
we must show that i and j commute. Let x be writes “ref” for the reference point u and LABEL for
any element of S1, and choose y and z so that the function labu; thus the name of the function does
int1(x, y) =​ i and int1(y, z) =​ j. The figure shows not explicitly indicate the reference point.
x, y, and z in S1 and their images f(x), f(y), and 2. Hint for Exercise 7.1.3: You may find it helpful to
f(z) in S2. The interval-​reversing property implies draw a diagram showing the points u, v, x, and y, with
that int2(f(y), f(x)) =​int1(x, y) =​ i; that is, the two arrows depicting four relevant intervals.
intervals labeled i in the figure are the same. 3. Hint for Exercise 7.1.4a: The group operation in
Likewise int2(f(z), f(y)) =​int1(y, z) (the two j’s are 𝒯ℐ12 is composition of functions, which is noncom-
the same), and int2(f(z), f(x)) =​int1(x, z) (the two mutative, so care must be taken with the order of
k’s). But by the interval-​sum equation, the inter- terms in the equation.
val k on the left side of the figure is int1(x, z) =​ 4. The presentation of interval-​space homomor-
i ⊕ j, while the same interval k on the right side is phisms here is considerably reworked from Hook
int2(f(z), f(x)) =​ j ⊕ i. Therefore i ⊕ j =​ j ⊕ i. 2007a, where a homomorphism is defined as an
In Section 7.5 we demonstrated that trans- ordered pair of functions (f, g). Equivalent formula-
positions Transvu and interval-​preserving map- tions of some of these concepts are developed inde-
pings presvu may be defined in every interval pendently in Kolman 2004, where the structural
space and that if the space is commutative, correspondences between groups and Generalized
then Transvu =​ presvu . One might expect some- Interval Systems are called “transfer principles.”
thing similar to be true of inversions Inv vu and 5. Hint for Exercise 7.2.2: The interval functions
interval-​reversing mappings Rev vu, but we now intpitch and intpc here are still the usual ones.
see that the situation is more complicated. In Consider the implications of the equations g(intpitch
the noncommutative case there are two inver- (C4, C♯4)) =​intpc(f(C4), f(C♯4)) and g(intpitch(C♯4, D4)) =​
sions to consider, Inv vu and Inv vu ; both of these intpc(f(C♯4), f(D4)).
inversions reverse some intervals, but neither 6. Hint for Exercise 7.2.3c: Start by writing g(int1(x,
of them can reverse all intervals, as no interval-​ y)) =​ g(int1(x, u) ⊕ int1(u, y)), using the interval-​sum
reversing function Rev vu exists at all. If int(u, v) ∈ equation to write int1(x, y) in terms of intervals that
Cent(G), then Inv vu and Inv vu are the same func- can be described by label functions. Extend this equa-
tion, but that function is not interval-​reversing tion to a chain of equalities ending with int2(f(x), f(y)).
unless the entire group is commutative. You will need to use the inverse interval property (to
In the general setting in which G1 is a subgroup replace int1(x, u) with ⊖int1(u, x)), the fact that g is a
of a possibly larger group G2, the interval-​revers- group homomorphism (the property mentioned in
ing property implies that all G1 intervals map Exercise 6.3.5b may be useful), and the assumption of
into G1 intervals, so this property says nothing label compatibility.
about elements of G2 lying outside G1. It is pos- 7. Hints for Exercise 7.2.6: If g is one-​to-​one and
sible, therefore, for an interval-​reversing map- f(x) =​ f(y), and if u is a reference point in S1, what can be
ping to map from a commutative interval space said about int2(f(u), f(x)) and int2(f(u), f(y)), and there-
into a commutative subspace of a larger noncom- fore about int1(u, x) and int1(u, y)? If f is onto and j ∈ G2,
mutative space. For example, the twelve major choose v and w in S2 such that int2(v, w) =​ j; can you then
triads form a commutative interval space S1 with use interval compatibility and the assumption that f is

296 • E x p loring M usical S p aces

onto to find i in G1 such that g(i) =​ j? If g is onto and y ∈ 18. For early instances of pitch structures repre-
S2, and if reference points u in S1 and v in S2 are chosen sented by ordered pairs of mod-​7 and mod-​12 inte-
so that f(u) =​ v, choose i in G1 such that g(i) =​int2(v, y); gers, see Brinkman 1986 and Agmon 1989. Some
can you use this i to find x in S1 such that f(x) =​ y? related constructions will be explored in the study of
8. A proof of a differently formulated but equiva- signature transformations in Chapter 13.
lent statement of the interval-​space homomorphism 19. Rings (2011b, 76–​78) studies a similar ♭3/̂ ♯​2̂
theorem is found in the appendix to Hook 2007a. ambiguity in Tatyana’s letter scene in Tchaikovsky’s
9. Some readers will recognize this reading-​from-​ Eugene Onegin.
the-​bottom-​up convention from standard mathematical 20. Figure 7.3.3 shows the bass line descending
from D♭ to G♮ in m. 55, as Debussy’s basses do. The
b
notations; in calculus, for example, ∫a is the integral
from a to b, not from b to a. In Sections 7.5 and 7.6 widely anthologized piano transcription by Leonard
we will introduce other notations that follow the same Borwick misleadingly shows D♭ ascending to G here
convention. instead.
10. Hint for Exercise 7.2.8a: The frequency a pure 21. Hint for Exercise 7.3.4a: If you are having trou-
perfect fifth above 440 would be 660; f(E5) is not pre- ble finding six paths, remember that an arrow join-
cisely 660, but something very close to that. ing two bubbles theoretically stands for two different
11. Hint for Exercise 7.2.9: Inverse mappings of the arrows between vertices.
form g(x) =​ x−1 were discussed at the end of Section 22. For an example of an ambitious undertaking
6.6, where it was noted that in general an inverse map- somewhat resembling the multidimensional approach
ping is an anti-​automorphism of a group. Remember, described here, see Pople 2004.
however, that the interval group under consideration 23. Hints for Exercise 7.3.6: The operation ⊕ is com-
in the present situation is commutative. ponentwise addition in ℤ × ℤ; ⊗ is ordinary multipli-
12. The only group homomorphism from ℝ to cation; and f(a, b) is 2a · 3b. If the same f(a, b) results
ℤ is the trivial one, which maps every real number from different choices of a and b, you should be able
to 0. It follows that the only possible interval-​space to deduce an equation in four variables showing that
homomorphisms from cpitch to pitch are trivial as a power of 2 is equal to a power of 3. Why must the
well: every note in cpitch must be mapped to the exponents on the 2 and 3 both be 0?
same element of pitch. 24. Just frequency space appears as Example 2.1.5
13. Hint for Exercise 7.2.10a: This conclusion may in Lewin [1987] 2007, 17–​20. The discussion is some-
sound paradoxical: if h is the identity mapping, how what informal, as it occurs before Lewin has defined
can it not be interval-​preserving? In fact, h is an iden- generalized interval systems and direct products
tity mapping of sets, but not an identity mapping of thereof, but it includes commentary on Lewin’s
interval spaces. In this situation the interval-​preserving way of hearing intervals in this space as “harmonic
condition requires comparison of intervals measured sensations.”
by two different interval functions, intpc and intfifth. 25. In Lewin’s formulation of quotient GISes
14. Some parts of Theorem 7.2.13 are proved in the (Lewin [1987] 2007, 33–​35), the equivalence relation
appendix to Hook 2007a. on S is induced by a group congruence rather than by
15. Hints for Exercise 7.2.15: The function g xy in a normal subgroup. The two formulations are equiva-
this example was called g uv in Exercise 7.2.8a. The lent, because normal subgroups give rise to group con-
interval group of frequency space is multiplicative, gruences and vice versa. A group congruence ~ on G
so in the formula from the theorem, ⊕ must be determines a normal subgroup H =​{x ∈ G | x ~ e}; the
interpreted as multiplication, and ⊖i means i−1, the relation on S induced by ~, as defined by Lewin, is the
reciprocal of i. Intervals in S1, including u-​labels, are same as the relation ≈H defined here.
fractions of an octave (multiples of 1 12 for equal-​tem- 26. Hint for Exercise 7.4.1: The example in the exer-
pered pitches). cise shows that the interval (5, 17, −7) is an element
16. Hint for Exercise 7.2.16: The interval group of S2 of this congruence subgroup. The subgroup may be
is the noncommutative group 𝒯ℐ12 in which the group described as H +​ K, the set of all ordered triples that
operation ⊕ is composition of functions ●, so the order can be obtained as the sum of a triple in H (the congru-
of factors in the equation is important. The formulas ence subgroup for ~O) and a triple in K (the congruence
from Theorem 5.5.6 may be useful. subgroup for ~T). (Rather than constructing OT-​space
17. Nowhere in this discussion have we needed to directly as a quotient of pitch-​triple space via a single
specify which beat is labeled 0—​or, for that matter, congruence subgroup, as described in this exercise, we
which pitch is labeled 0. Intervals do not depend on could instead construct it in two stages, first forming
label functions. O-​space and then forming a quotient of that quotient

Intervals • 297
to account for T-​equivalence or vice versa. Interested additional examples of interscalar mappings similar
readers are encouraged to work out the details.) to those presented below.
27. Hints for Exercise 7.4.6c–​d: For generic equiva- 35. Hints for Exercise 7.5.13: In part (a), chromatic
lence recall Exercise 2.4.5. The answer to the question transposition, interscalar mappings of diatonic scales,
in (d) must be no: Esl–​spc is a bijection, so if it were pivot intervals on common tones, and signature
a homomorphism, according to Theorem 7.2.5c–​d it transformations are all possible. You should be able to
would be an isomorphism, and as noted in the present answer (c) by examining only the first interval in each
exercise, sl and spc cannot be isomorphic. For both of the two phrases.
(c) and (d), remember a technique from Section 7.2 for 36. A pentatonic collection is a subset of three dif-
showing that a particular function f: S1 → S2 is not an ferent diatonic collections—​in this case dpitch(−7),
interval-​space homomorphism: find two pairs of ele- dpitch(−6), and dpitch(−5)—​and shares its index
ments of S1, separated by the same interval, whose number with the middle one. The pentatonic index
images under f are separated by two different inter- number does not reflect the number of sharps or
vals in S2. flats in the scale, as a diatonic index number does.
28. The hexachord theorem, attributed to Babbitt, Pentatonic index numbers with that property could
apparently first appeared in print in Lewin 1960 and not be unambiguously defined, because for example
has since been rediscovered frequently and proved three different pentatonic collections contain white
and generalized in many different ways. The state- notes only.
ment presented here could be generalized further by 37. Hint for Exercise 7.5.16: The mapping is from
replacing Transi with any bijective transformation on chromatic pitch space to the pitch space of the acous-
a finite set S—​a formulation essentially equivalent tic scale C–​D–​E–​F♯–​G–​A–​B♭. In notation to be intro-
to the one given in Lewin [1987] 2007, 144–​45. For duced in Chapter 14, the latter space could be called
other versions see Rahn 1980, 105–​07, and Morris mmin(0). See also Hook 2007a, 17–​19.
1987, 74–​77. 38. Hint for Exercise 7.5.18: The mapping is from
29. For interval-​ class vectors see, for example, chromatic space to the whole-​tone space wt0. This
Straus 2016, 16–​17. example was given by Joseph Schillinger (1946,
30. Hint for Exercise 7.5.2: As a check on your vol. 1, 216), who advocated this kind of transfor-
answers, the sum of the entries in the interval-​class mation, which he called “geometric expansion,” as
vector for a set of cardinality n is always n(n − 1) ∕ 2, the a compositional technique. Subsequently Nicolas
total number of intervals (2-​element subsets) present Slonimsky wrote a short fugue on the whole-​tone
in the set. A curious corollary of this observation, rele- subject in No. 48 of his 51 Minitudes for piano
vant to the first two questions in (a), is that the unique (1972–​76), subtitled “Bach × 2 =​Debussy.” See
set classes of cardinalities 0 and 1 are the only pair of Hook 2007a, 19–​21.
set classes of different cardinalities sharing the same 39. Underlying the 𝒯S-​𝒫S duality is the following
interval-​class vector. theorem of pure mathematics: Suppose a transfor-
31. Tymoczko (2011, 119–​23) uses the term sca- mation group G acts on a set S in simply transitive
lar transposition. Messiaen illustrates “parallel suc- fashion, and let H be the centralizer of G, the group
cessions of chords,” related by scalar rather than of all bijective transformations on S that commute
chromatic transposition, in several of his modes of with all transformations in H. Then H is isomorphic
limited transposition (Messiaen [1944] 1956, 87–​95). to G, the action of H on S is simply transitive, and
Neidhöfer 2005 studies Messiaen’s use of such trans- G is the centralizer of H. Among mathematicians
positions in composition. this theorem seems not to be well known, but it is a
32. Hint for Exercise 7.5.3b: Recall the discus- consequence, for example, of Theorem 4.2A in Dixon
sion of conjugation from Section 6.6; in particular, and Mortimer 1996, 109. Musically, the duality is
Figure 6.6.9a may be helpful. Interpret the set S′ in described telegraphically in the final paragraph of
the figure as the group G. Lewin’s GMIT (Lewin [1987] 2007, 253), where the
33. Hint for Exercise 7.5.6: A short way to prove this groups 𝒯S and 𝒫S are called STRANS and STRANS′,
result is to apply the result of Exercise 7.2.3c, using respectively, and more completely in Lewin 1995,
the identity mapping on the interval group for the 100–​01, where they are called SIMP and COMM.
function g. For more thorough accounts with proofs and many
34. In Hook 2007a, 16–​ 21, I used the term generalizations, see Fiore and Satyendra 2005 and
“cross-​type transposition” and the notation T uv for Peck 2010.
interval-​preserving mappings in a cross-​ type set- 40. The interval-​ preserving mapping gu(i) cor-
ting; Tymoczko 2011 (140–​53) uses the term “inter- responds with the function labeled P ref i in Lewin
scalar transposition.” See both of those sources for [1987] 2007, 47. Lewin’s definition of this function

298 • E x p loring M usical S p aces

is motivated by the property described in Theorem 52. Hint for Exercise 7.6.20c: Inv uv ( x ) = y is equiva-
7.5.24c. lent to a statement obtained by exchanging u and v in
41. Hint for Exercise 7.5.27: In all parts of this exer- condition 2. You must show that such a statement is
cise you may find it helpful to draw a diagram showing equivalent to condition 1.
the elements and mappings involved. 53. Corollary 3.5.4 on page 53 of Lewin [1987] 2007
42. Hint for Exercise 7.6.2a: Compare the sums matches Theorem 7.6.21, stated in terms of Lewin’s
int(u, x) ⊕ int(u, y) and int(w, x) ⊕ int(w, y) (from inversions I uv and I vu . Theorem 3.6.4 on page 59
inversion condition 1) relative to the two reference matches Theorem 7.6.22, but only in the case of map-
points u and w. Use the interval-​sum equation and pings on a single interval space. Lewin’s proofs of
commutativity to show that these two sums always both properties are longer and more complex than
differ by the interval k, independent of x and y. If the proofs presented here.
y =​Inv[u]i(x), you should then be able to conclude that
y =​Inv[w]i⊕k(x). To visualize this relationship, you
may find it useful to add points w and z to a copy of SUGGESTED READING
Figure 7.6.1a, positioned so that an arrow from w to u
has the same length and direction as an arrow from v In Chapter 3 of GMIT (Lewin [1987] 2007), Lewin
to z, implying that int(w, z) =​ i ⊕ k and Inv[w]i⊕k =​ Inv wz . develops some aspects of the theory of transposi-
43. Hint for Exercise 7.6.4: If you exchange x and y tion and inversion operators, including the non-
in condition 1 for the inversion, what do you get? commutative case, beyond what is considered here;
44. Hint for Exercise 7.6.5: In addition to the products and quotients of generalized interval
interval-​sum equation, you will need to use the u–​v systems are defined in the same chapter. Many
label-​reversing property, the inverse interval prop- readers find this part of GMIT especially dense and
erty, and commutativity. forbidding, and comprehension is further hindered
45. Hook 2007a refers to “cross-​type inversions” by a few mathematical errors (compounded, in
and extends Lewin’s notation I uv to such mappings, some printings, by typographical errors; see Hook
which in the present terminology are Rev vu. 2007b, 181–​82). Lewin does not define GIS homo-
46. Hint for Exercise 7.6.8: See Theorem 7.5.20g. morphisms or isomorphisms, and his formaliza-
The two questions in this exercise have different tion of transposition and inversion allows only for
answers. transformations from a space to itself, never from
47. Hint for Exercise 7.6.9a: Recall Figure 5.5.4. one space to another. The treatment of some of
48. A longer excerpt from this work is analyzed in these topics here is reworked (substantially) from
Hook 2007a, 22–​25. Hook 2007a.
49. Hint for Exercise 7.6.16: Remember that
presvu =​Transi and Rev vu =​Inv[u]i, where in both cases A decade before GMIT, Lewin 1977 works with
i =​int(u, v). In some cases you may need to change pitch classes in a more abstract way than is usual in
the reference point u in the definition of Inv[u]i; such a set theory, avoiding the assignment of integer labels
change may be made with the aid of the formula from to the twelve pcs. This article effectively presents
Exercise 7.6.2a. pitch-​class space as an interval space, developing
50. Hint for Exercise 7.6.20a: Recall that Inv vu( x ) = y the interval function, the interval-​space axioms,
if and only if condition 2 holds. To show that Inv vu( u) =​ and definitions of transposition and inversion
v, therefore, you must show that if you replace x with operators in this special case.
u and y with v in condition 2, the resulting statement
is always true. The analytical chapters in Rings 2011b employ
51. Hint for Exercise 7.6.20b: In the commutative the product space sd × pc with a richness and
case, by Theorem 7.6.15e, (Inv vu)−1 =​ Inv vu. In the non- subtlety that the short summary and quasi-​
commutative case this is not necessarily true. The Ringsian Debussy analysis in Section 7.3 can only
assertion Inv vu( y ) = x is equivalent to a statement suggest. The book’s earlier chapters offer valu-
obtained by exchanging x and y in condition 2. You able commentary on transformational techniques
must show that such a statement is equivalent to con- in general, with emphasis on applications to
dition 1, which defines f(x) =​ y. tonal music.

Intervals • 299
 8
Transformations I
Triadic Transformations

IN THIS chapter and the next, we expand our elements of the space triad. By a triadic transfor-
repertoire of transformations and related ideas. mation we mean a mapping f: triad → triad. For
The study of triadic transformations in this the time being we will be concerned only with
chapter encompasses and extends our previous transformations that are bijective (one-​to-​one
observations about the neo-​Riemannian P, L, and and onto); this eliminates from consideration,
R, clarifying aspects of algebraic structure that for example, a function that maps both the
have so far been left unexplored. Many impor- C-​major and C-​minor triads to G major.
tant triadic transformations may be formalized An example of a bijective triadic transforma-
as uniform triadic transformations, studied in tion is the relative transformation R. When R
Section 8.1. In Section 8.2 the focus narrows maps the C-​major triad to the A-​minor triad, the
to the 24 transformations in the Riemann group triad’s root descends by three semitones and its
(the so-​called schritts and wechsels, including P, mode (or quality) changes from major to minor.
L, and R). The scope broadens again in Section The same observations describe the action of R
8.3, which examines other triadic transforma- on any other major triad as well. When R acts
tions and related topics. Analytical examples on minor triads, the rule is different—​the chord
illustrate the many and varied applications of root now ascends by three semitones—​but again
these techniques. the action of the transformation on one minor
triad is representative of its action on all minor
8.1 UNIFORM TRIADIC triads. The behavior of the transformation is
TRANSFORMATIONS completely characterized by three pieces of
information: first, the mode of the triad always
In this section the word triad always refers to a changes; second, if the input triad is major, the
consonant (major or minor) triad, one of the 24 root (as a pitch class) transposes by T9; and

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0008
third, if the input triad is minor, the root trans- is F minor. In this way we may associate the
poses by T3. space triad with the Cartesian product ℤ12 ×
We summarize the above observations by 𝒮𝑖𝑔𝑛; here 𝒮𝑖𝑔𝑛 is the sign group {+​, −}, which
saying that the action of R on triad is uniform, as noted in Section 5.2 may be identified with
which means loosely that R transforms all major {1, −1}, a multiplicative group of order 2. The
triads “the same way” and likewise transforms elements of this product are ordered pairs (r,
all minor triads “the same way.” If a uniform σ), but we will sometimes use a more com-
transformation maps C major to A minor, then pact notation in which the sign is written as a
it must also map D♭ major to B♭ minor, D major to superscript on the pitch-​class number; by this
B minor, and so on. More generally, if a uniform convention A♭ major is 8+​ and F minor is 5−.
transformation maps a triad X to a triad Y, then We will on occasion use a variable such as X
it must also map T1(X) to T1(Y), T2(X) to T2(Y), to stand for a triad; the components r and σ
and indeed Tn(X) to Tn(Y) for every transposi- of the triad X may be written as rX and σX if
tion Tn. The equation U(Tn(X)) =​ Tn(U(X)) there- confusion with other r and σ is possible.
fore holds for every uniform transformation U, Both ℤ12 and 𝒮𝑖𝑔𝑛 are groups. Signs multiply
every triad X, and every mod-​12 integer n. But by the rules +​+​=​−− =​+​and +​− =​−+​=​−; algebra-
to say that U(Tn(X)) =​ Tn(U(X)) for all X implies ically, the group 𝒮𝑖𝑔𝑛, like every group of order 2,
that the composite functions Tn ● U and U ● Tn is isomorphic to ℤ2 or 𝒞2. According to Section
are the same function. A triadic transformation 6.4, therefore, the product G =​ ℤ12 × 𝒮𝑖𝑔𝑛 has a
is uniform, therefore, if and only if it commutes direct-​product group structure. The space triad
with all transpositions.1 We refer to this property may be regarded as an interval space whose inter-
as the uniformity condition. Uniformity ensures val group is this product group G. This is a differ-
a kind of musical coherence of the transforma- ent interval-​space structure on triad from the
tion; a bijective triadic transformation satisfying one discussed in Sections 5.6 and 7.1–​7.2 whose
the uniformity condition is called a uniform tri- interval group is 𝒯ℐ12. In the space considered
adic transformation, or UTT. here, the interval from A♭ major to F minor is
Other examples of UTTs include the neo-​ int(8+​, 5−) =​(9, −). This last ordered pair (9, −)
Riemannian P and L as well as the transpositions represents not a triad but an interval between
Tn themselves. Transpositions are defined on triads. The first component (9) is not a pitch
many things besides triads, but they map triads class but rather int(8, 5), an interval between
bijectively to triads and so may be considered tri- roots in pc space, and the second component (−)
adic transformations in their own right. A trans- indicates not the minor mode but the “interval”
position does not change the mode of the triad to between the signs +​and − of the original triads,
which it is applied, and in the case of a transpo- the sign by which +​must be multiplied to yield −.
sition the associated root motion is the same for When a +​occurs in the second component of
triads of both modes. The inversion operators In an interval it indicates that the two triads being
are familiar examples of transformations whose compared are of like mode, while a − means they
action on triads is not uniform, since we know differ in mode. The ordered-​pair notation for
from Section 5.5 that inversions do not com- triads, in this interpretation, is a label function
mute with transpositions. When the inversion I as described in Section 7.1, using C major as the
=​ I0 (inversion about the pitch class C) is applied reference point of the space: the C-​major triad
to a C-​major triad, the result is an F-​minor triad, is labeled with (0, +​), the identity element of G,
but when I is applied to D♭ major (one semitone and every other triad is labeled with its interval
higher), the result is not F♯ minor as uniformity from C major.
would require, but rather E minor, a semitone A UTT may now be defined as an ordered
lower—​an explicit example of non-​uniform action. triple, conventionally written in the form
A few remarks about the structure of the U =​ 〈σ, t+​, t−〉 using angle brackets.2 Here σ is a
space triad are in order before we proceed sign (called the sign of U and denoted σU when
with our formalization of UTTs. A major or necessary), and t+​ and t− are mod-​12 integers,
minor triad may be represented by an ordered called the transposition levels for major and minor
pair (r, σ), where r is a mod-​12 integer denot- triads respectively. If U =​ 〈σU, t+​, t−〉 is a UTT and
ing the triad’s root and σ (sigma) is a sign, X =​(r, σX) is a triad, then U acts on X by trans-
either +​or −, denoting its mode (+​for major, − posing its root upward by either t+​ or t− semi-
for minor). For example, the ordered pair (8, +​) tones, depending on the mode of the triad X
represents the A♭-​ major triad, while (5, −) (depending, that is, on whether σX is +​or −),

Triadic Transformations • 301

and by adjusting that mode depending on the
sign of the UTT U (that is, changing the triad’s
sign if and only if σU is −). More precisely, if σX
is +​, then U(X) =​(r +​ t+​, σU), while if σX is −, then
U(X) =​(r +​ t−, −σU). A UTT whose sign is +​always
maps major triads to major triads and minor to
minor, and is accordingly called mode-​preserv-
ing (or quality-​preserving); a UTT whose sign is
− maps major to minor and minor to major, and
is mode-​reversing (or quality-​reversing).
The UTT notation for the relative trans-
formation is R =​ 〈−, 9, 3〉. To apply this trans-
formation to the A♭-​major triad X =​(8, +​), we FIGURE 8.1.1 Calculating the action of the UTT
observe that the starting triad is major, so the R =​ 〈−, 9, 3〉: two examples
applicable transposition level is t+​ =​9, which we
add to the triad’s root to obtain 8 +​9 =​5 (mod
12); we then change the triad’s mode because
the sign of the UTT is −. Therefore R(X) =​(5, −),
the F-​minor triad. If instead we apply the same
R to the C♯-​minor triad (1, −), we add t− =​3
to the root 1 and change the mode, obtaining
(4, +​), the E-​major triad. Figure 8.1.1 summa-
rizes these calculations.
Notably, R is the only UTT that maps A♭ → f
and c♯ → E: these actions, together with uni-
formity, imply exactly how the transformation
must behave when applied to any other triad.
Generally, two sample actions—​one application
to a major triad and one to a minor triad—​are
sufficient to determine a UTT completely. FIGURE 8.1.2 Calculating the product of two
The UTT notations for the other uniform UTTs: two examples
transformations mentioned previously are sum-
marized here, along with R, and with representa-
tive actions of each:

Tn =​〈 +​, n, n〉 for n =​0, 1, …, 11:

T5 T5
C  → F, c  →f
P =​〈 −, 0, 0〉 (parallel):
C P
→ c, c P
→C
L =​〈 −, 4, 8〉 (leittonwechsel):
C L → e, → A♭
c L
R =​〈 −, 9, 3〉 (relative):
FIGURE 8.1.3 Calculating the inverse of a UTT:
C R
→ a, → E♭
c R two examples

Transpositions, by definition, transpose every-

thing the same way: when Tn is written as a UTT, property that the two transposition levels are
its two transposition levels t+​ and t− are both mod-​12 inverses: that is, 0 +​0 =​4 +​8 =​9 +​3 =​
equal to the number n. The case n =​0 deserves 0 (mod 12). The action of one of these UTTs on
special notice: T0 =​ 〈+​, 0, 0〉 leaves every triad minor triads, that is, is equal and opposite to its
unchanged, and is therefore nothing more than action on major triads. This property shared by
the identity mapping Etriad on the space triad. P, L, and R comports with a conception of har-
The transformations P, L, and R all have the monic dualism espoused by Hugo Riemann and

302 • E x p l oring M usica l S p aces

others, whereby minor triads, scales, and keys the usual way: in both calculations in the figure,
are seen as upside-​down images of their major a mode-​preserving (+​) and a mode-​reversing (−)
counterparts.3 Accordingly, UTTs for which UTT combine to produce a mode-​reversing (−)
t+​ +​ t− =​0 (mod 12), or equivalently t− =​−t+​, are UTT. The transposition levels combine differ-
called Riemannian UTTs. ently in the two calculations, because the rule
Riemannian as defined here is a technical for composing UTTs depends on the sign of the
term identifying certain UTTs. We have previ- first UTT. If the first UTT is mode-​preserving, as
ously referred to P, L, and R as neo-​Riemannian, in the product 〈+​, 4, 5〉〈−, 9, 3〉, then the second
a looser term encompassing a constellation UTT is applied to a triad of the same mode as
of related concepts, useful especially in situa- the first, so corresponding transposition levels
tions in which it is important to distinguish the are added (4 +​9 and 5 +​3). But if the first UTT is
modern perspective from that of Riemann (for mode-​reversing, as in the product 〈−, 9, 3〉〈+​, 4,
example, the Riemannian and neo-​Riemannian 5〉, then the second UTT acts on a triad opposite
tonnetzes of Figures 1.4.14 and 1.4.5, respec- in mode from the first, so opposite transposi-
tively). In the theoretical development in this tion levels must be added (9 +​5 and 3 +​4).
chapter the distinction between Riemannian As bijective mappings, UTTs always have
and neo-​Riemannian is of less importance than inverse functions. Figure 8.1.3 illustrates the
the distinction between Riemannian and non-​ calculation of inverses. The inverse of a mode-​
Riemannian—​between the few UTTs that sat- preserving UTT 〈+​, t+​, t−〉 is simply 〈+​, −t+​, −t−〉,
isfy t+​ +​ t− =​0 and the larger number that do obtained by replacing each transposition level
not. Many principles of recent neo-​Riemannian with its mod-​12 inverse, but the inverse of a
theory can be derived from the distinctive prop- mode-​reversing UTT 〈−, t+​, t−〉 is 〈−, −t−, −t+​〉,
erties of Riemannian UTTs, and we will have which also requires exchanging the positions of
more to say about them in the next section. In the two transposition levels.
particular, some Riemannian transformations
are recognized for their smooth voice leading: Exercise 8.1.4
when we write C L → e, we often imagine two (a) Use the method illustrated in Figure 8.1.1
stationary voices and one voice moving by semi- to calculate the result when each of the
tone. It is important to realize, however, that UTTs P, L, and R is applied to a D-​major
a UTT or any triadic transformation defined triad, and also to a G♯-​minor triad. Verify
on triad operates on a triad as an indivisible that the results are consistent with the
entity and does not directly describe action at familiar behavior of these neo-​Riemannian
the level of individual pitch classes: the action transformations.
of L applied to pitch class 0, for instance, is not (b) Let U and V be the UTTs U =​ 〈+​, 3, 2〉 and
defined. A C-​major triad played by cellos, fol- V =​ 〈−, 2, 3〉. Calculate the result when each
lowed by an E-​minor triad two octaves higher in of the transformations U and V is applied to
the woodwinds, may be modeled as an L trans-   a G-​major triad, and also to a G-​minor triad.5
formation despite the lack of coherent voice
leading between the chords. The smooth voice Exercise 8.1.5
leading commonly associated with L is inher- (a) Consider the UTTs 〈+​, 4, 5〉 and 〈−, 9, 3〉
ent in the transformation only in the sense of appearing in Figure 8.1.2. Starting with an
potential: an L relation between two elements F-​minor triad, first apply 〈+​, 4, 5〉, then apply
of triad admits a possible realization in pitch 〈−, 9, 3〉 to the result. Verify that the answer
or pitch-​class space with smoother voice lead- is the same as if you had applied the single
ing than is possible for most other triadic transformation 〈−, 1, 8〉 to the F-​minor triad
transformations.4 (as it should be, since the figure shows that
When two or more UTTs are applied in suc- 〈+​, 4, 5〉〈−, 9, 3〉 =​ 〈−, 1, 8〉). Perform a similar
cession, the composite transformation may be check to confirm the product of the same
calculated by simple arithmetic. Figure 8.1.2 transpositions in the opposite order.
shows two such calculations, composing the (b) Use the method illustrated in Figure 8.1.2
UTTs 〈+​, 4, 5〉 and 〈−, 9, 3〉 in both possible to calculate each of the following UTT
orders. The operation by which the two UTTs products:6
are combined is composition of functions, 〈+​, 4, 5〉〈+​, 9, 3〉 and 〈+​, 9, 3〉〈+​, 4, 5〉
denoted in the figure by the symbol ● but often 〈+​, 8, 0〉〈−, 6, 5〉 and 〈−, 6, 5〉〈+​, 8, 0〉
omitted in practice. The two signs multiply in 〈−, 3, 5〉〈−, 8, 10〉 and 〈−, 8, 10〉〈−, 3, 5〉

Triadic Transformations • 303

(c) Let U and V be the UTTs U =​ 〈+​, 3, 2〉 Exercise 8.1.8 Use the above formulas to
and V =​ 〈−, 2, 3〉. Show that the equations reproduce the results of Exercises 8.1.4b,
PU =​ V and PV =​ U both hold, where P is the 8.1.5b,
   and 8.1.6b.
parallel transformation 〈−, 0, 0〉. How do
your answers to Exercise 8.1.4b illustrate Exercise 8.1.9 Consider the action of a UTT
these equations? Are the equations UP =​ V U =​ 〈σU, t+​, t−〉 on a triad X =​(r, σX). In the
  and VP =​ U true as well? interval-​space structure on triad with interval
group ℤ12 × 𝒮𝑖𝑔𝑛, the interval from X to U(X)
Exercise 8.1.6 is either (t+​, σU) or (t−, σU) depending on the
(a) Verify the inverses shown in Figure 8.1.3 mode of X. In the spirit of the general formulas
by calculating the UTT products 〈+​, 4, 5〉 presented above, write a single formula for this
〈+​, 8, 7〉 and 〈−, 4, 5〉〈−, 7, 8〉 and showing interval,
   valid in all cases.
that both are equal to T0 =​ 〈+​, 0, 0〉.
(b) Use the method illustrated in Figure 8.1.3 Under the operation of composition as
to calculate the inverse of each of the described above, the UTTs form a group 𝒰 with
following UTTs: 〈+​, 2, 6〉; 〈−, 6, 5〉; 〈+​, 7, 7〉 a rich and complex structure. The order of the
(=​ T7); and 〈−, 9, 3〉 (=​ R). group is 2 ∙ 12 ∙ 12 =​288, as there are two pos-
(c) The last calculation in (b) confirms that R sibilities for the sign of a UTT and twelve pos-
is an involution—​that is, R−1 =​ R. Verify sibilities for each of its two transposition levels.
that the neo-​Riemannian P and L are also Elements of the group 𝒰 are in one-​to-​one cor-
involutions, and identify a few other UTTs respondence with elements of the direct-​product
that are involutions. group ℤ2 × ℤ12 × ℤ12 (or 𝒞2 × 𝒞12 × 𝒞12), but as a
(d) Suppose U and V are any two UTTs such group, 𝒰 is not isomorphic to that direct prod-
that PU =​ V. Use the fact that P is an uct. A direct product of commutative groups
involution to show that the equation PV =​ U is commutative, but examples such as those in
must be true as well. (Exercise 8.1.5c Figure 8.1.2 and Exercise 8.1.5 show that the
  illustrated this property.) UTT group is not commutative.
Algebraically, the group 𝒰 is called a wreath
Exercise 8.1.7 Three different UTTs U have product of the groups 𝒞12 and 𝒞2, symbolized
the property that U3 =​ P. (Here U3 indicates the 𝒞12 ≀ 𝒞2. Figure 8.1.10, which is similar in con-
product UUU.) One of them is P itself. What are struction to double-​circle diagrams in Chapters
the
  other two? 4 and 6, may be helpful in understanding the
structure of this group. Imagine two copies of
the cyclic group 𝒞12 at work, one transposing
The formulas for UTT actions, products, and
inverses given above presented two separate rules
to follow depending on whether a certain sign was
+​or −. It is possible, in each instance, to write a
single formula applicable in all cases, though the
formulas may be rather difficult to read:

Action of a UTT: If U =​ 〈σU , t+​, t−〉 is a UTT

and X =​(r, σX) is a triad, then
U(X) =​(r +​ t(σ ), σX ∙ σU).
X

Product formula: If U =​ 〈σU , t+​U , t−U〉 and

V =​ 〈σV, t+​V, t−V〉 are UTTs, then
UV =​ 〈σU ∙ σV, t+​U +​ t(σ )V, t−U +​ t(−σ )V〉.
U U

Inverse formula: If U =​ 〈σ, t+​, t−〉 is a

UTT, then U−1 =​ 〈σ, −tσ, −t(−σ)〉.

In these formulas, the signs σX and σU

sometimes occur in superscripts. The nota-
tion t(σ )V, for example, means t+​V if σU is +​, or
U
FIGURE 8.1.10 The wreath-​ product structure
t−V if σU is −. 𝒞12 ≀ 𝒞2 of the UTT group 𝒰

304 • E x p l oring M usica l S p aces

the +​circle of major triads by means of the add a third configuration to the right of
transposition level t+​, the other transposing the Figure 8.1.11, showing the result of applying
− circle of minor triads via t−. These two copies the UTT 〈−, 8, 7〉 to the configuration on the
of 𝒞12 are called the fiber groups of the wreath right side of the original figure. Calculate the
product. If these transpositions, acting indepen- UTT product 〈−, 2, 5〉〈−, 8, 7〉 and verify that
dently on major and minor triads, were the only if this product is applied to the configuration
possible transformations, the group would be on the left side of the figure, the resulting
𝒞12 × 𝒞12. But the sign σ of a UTT comes into configuration
   matches the one you drew.
play by allowing us to exchange triad modes,
effectively swapping the roles of the two circles. As described in Chapter 6, a group action
The sign group 𝒮𝑖𝑔𝑛, a group of order 2 isomor- on a set may be conceived in terms of permuta-
phic to 𝒞2, is the control group of the wreath prod- tions of the set. There are theoretically 24! per-
uct. Despite the simplicity of this control group, mutations of the set triad—​ a vast number,
its action entangles with the actions of the fiber 24 digits long—​but most of them are of little
groups to create a group structure more complex apparent musical interest, and most of them are
than an ordinary direct product. Metaphorically not UTTs. Double-​circle diagrams clarify the sig-
we may imagine a two-​drawer dresser in which nificant way in which the uniformity condition
major triads occupy one drawer and minor triads constrains permutations: the only admissible
the other. Within each drawer we are allowed to configurations are those obtained by rotating
permute the contents cyclically, but additionally the two circles and exchanging them. We may
we are allowed to pull out the two drawers and not mix major and minor triads on one circle,
exchange them.7 nor may we rearrange the triads on either cir-
Double-​circle diagrams are also useful aids cle in anything other than clockwise-​ascending
for visualizing actions of specific UTTs, as Figure order. These constraints limit the number of
8.1.11 shows. When the UTT U =​ 〈−, 2, 5〉 is configurations, and the number of UTTs, to the
applied to any triad in the configuration on the manageable total of 288.
left, we may look to the same position in the con- Still, the UTT group is much larger than the
figuration on the right to see the result; thus for set of 24 triads on which it acts. We have noted
example the actions U(7+​) =​9− and U(11−) =​4+​ that R is the only UTT that maps A♭ → f and
may be read from the picture. Geometrically, c♯ → E. If we know that a UTT maps A♭ → f but do
the new configuration is derived from the old not know how it acts on any minor triad, then
by rotating the two circles counterclockwise as we do not have enough information to identify
determined by the transposition levels t+​ =​2 and the UTT. Such a UTT must be mode-​reversing
t− =​5, and (because the sign of the given UTT is −) and its +​transposition level must be 9, but the
exchanging the +​and − signs. − transposition level cannot be determined from
the given information. Such a UTT may be R =​
Exercise 8.1.12 To see how double-​circle 〈−, 9, 3〉, but it may equally well be 〈−, 9, 0〉, or
configurations can illustrate UTT products, T9P =​ 〈−, 9, 9〉, or any of nine other possibilities

FIGURE 8.1.11 Double-​circle illustration of the action of a UTT

Triadic Transformations • 305

of the form 〈−, 9, x〉. The action of the group 𝒰 The transformations P =​ 〈−, 0, 0〉 =​ W0, L =​
on the space triad is clearly not simply tran- 〈−, 4, 8〉 =​ W4, and R =​ 〈−, 9, 3〉 =​ W9 are all exam-
sitive, which as defined in Section 5.6 would ples of wechsels. We are familiar already with
require that for any two triads X and Y there be many characteristics of these transformations.
one and only one UTT U such that U(X) =​ Y. As In Section 4.1 we described them as contextual
we observed there, simple transitivity is possible inversions, in contrast to the fixed inversion oper-
only for a group of the same size as the set on ators In; every wechsel maps a triad to an inver-
which it acts. (In the terminology of Exercise sionally related triad, but not via a fixed axis in
5.6.5c, the action of 𝒰 on triad is 12-​tuply tran- pitch-​class space. Figures 4.2.5 and 4.2.8 showed
sitive: for every X and Y there are exactly twelve neo-​Riemannian triad space nrtriad, or Riemann
possible UTTs.) space for short, a toroidal tonnetz-​style arrange-
The multiplicity of UTTs sharing the same ment of the 24 triads foregrounding P, L, and R
action contributes to the algebraic richness and relationships; we may now regard the Riemann
analytical potential of the group 𝒰, but it can group ℛ as a transformation group for navigat-
lead to indecision in practical situations: if we ing Riemann space. While some properties of P,
want to model A♭ → f transformationally, how L, and R are specific to those transformations,
do we decide which of the twelve available UTTs we will see that others are characteristic of wech-
to choose? For some purposes, a smaller and sels or Riemannian UTTs more generally. For
simpler group may be preferred, especially one example, Exercise 8.1.6c verified that P, L, and R
whose action on triad is simply transitive. One are all involutions—​a property easily seen to be
such group of particular interest in analysis com- true of all wechsels. The following exercise pro-
prises the Riemannian UTTs, to which we next vides examples of several properties involving
turn our attention. P, L, R, and a few other Riemannian transforma-
tions, to be described more generally in Theorem
8.2 RIEMANNIAN UTTs 8.2.2 below.
AND NEO-​R IEMANNIAN Exercise 8.2.1
ANALYSIS (a) Verify that the wechsel W5 =​ 〈−, 5, 7〉 is an
Riemannian UTTs were defined in the previous involution.
section as UTTs for which t+​ +​ t− =​0. In gen- (b) Verify that the inverse of the schritt
eral, t+​ +​ t−, the sum of the two transposition S5 =​ 〈+​, 5, 7〉 is S7 =​ 〈+​, 7, 5〉.
levels of a UTT U =​ 〈σ, t+​, t−〉, is a mod-​12 inte- (c) Verify that RL =​ W9W4 =​ S5 (=​ S9 –  4) while LR
ger called the total transposition of U, denoted =​ W4W9 =​ S7 (=​ S4 –​9). Therefore RL ≠ LR, but
tt(U). Riemannian UTTs are therefore those by (b), (RL)−1 =​ LR.
UTTs U for which tt(U) =​0. There are only 24 (d) Verify that S5L =​ S5W4 =​ W9 (=​ W5 + ​4) =​ R
Riemannian UTTs, twelve of each of two kinds: while LS5 =​ W4S5 =​ W11 (=​ W4 – ​5).
mode-​ preserving Riemannian UTTs of the (e) Exercise 4.2.12 demonstrated in the
form Sn =​ 〈+​, n, −n〉, known as schritts, and tonnetz that PLR =​ RLP. Use UTTs to
mode-​reversing Riemannian UTTs of the form show that both of these composite
Wn =​ 〈−, n, −n〉, known as wechsels.8 In both cases transformations are equal to W5.
n may range over the integers mod 12, but if (f) Use UTTs to confirm that PLP =​ LPL
either transposition level of a Riemannian UTT   (another identity from Exercise 4.2.12).
is known, the other is uniquely determined.
The set of all 24 Riemannian UTTs is denoted Theorem 8.2.2: Properties of Riemannian
ℛ; we will show below that ℛ is a group, which transformations
we call the Riemann group. (It is also sometimes (a) Products of schritts and wechsels may be
known as the schritt-​wechsel group or the PLR calculated by the following formulas, in
group.) We will also use the notation ℛ+​ for which the subscripts are integers mod 12:
the set of schritts, which is a subgroup of ℛ SmSn =​ Sm+​n,
called the schritt group. The schritts S0 =​ 〈+​, 0, 0〉 SmWn =​ Wm+​n,
and S6 =​ 〈+​, 6, 6〉 are the same as the transpo- WmSn =​ Wm−n,
sitions T0 (the identity) and T6; these are the WmWn =​ Sm−n.
only two transformations that belong to both (b) Inverses of schritts and wechsels may be
ℛ and 𝒯ℐ12. calculated by the formulas Sn−1 =​ S−n and

306 • E x p l oring M usica l S p aces

 Wn−1 =​ Wn. In particular, every wechsel is an (U1U2 ⋯ Uk)−1 =​ Uk−1 ⋯ U2−1U1−1 holds for any
involution. UTTs U1, U2, …, Uk in the group 𝒰. If all of these
(c) If U1, U2, …, Uk are wechsels, then UTTs are wechsels—​that is, if each Ui is equal to
(U1U2 ⋯ Uk)−1 =​ Uk ⋯ U2U1. If k is odd, some Wn—​then every Ui is an involution by part
then U1U2 ⋯ Uk =​ Uk ⋯ U2U1. (b) of the theorem, so we may replace each Ui−1
(d) ℛ is a normal subgroup of 𝒰, the kernel on the right side of this equation with Ui itself,
of the group homomorphism tt: 𝒰 → ℤ12. giving the first statement in (c).
ℛ is isomorphic to 𝒟12, the dihedral group Generally, the product of an even number of
of order 24, and acts in simply transitive mode-​ reversing (−) UTTs is mode-​ preserving
fashion on the set triad. (+​
), while the product of an odd number of
(e) The two UTTs S1 =​ 〈+​, 1, 11〉 and P =​ W0 mode-​reversing UTTs is mode-​reversing. Thus
=​ 〈−, 0, 0〉 together generate ℛ. The two the product of an odd number of wechsels is
wechsels L =​ W4 =​ 〈−, 4, 8〉 and R =​ W9 =​ again a wechsel. If k is odd, therefore, U1U2 ⋯ Uk
〈−, 9, 3〉 also generate ℛ. in the first part of (c) is itself a wechsel
(f) ℛ+​, the set of schritts Sn, is a normal subgroup and therefore an involution, so we may replace
of 𝒰 isomorphic to 𝒞12, the cyclic group of (U1U2 ⋯ Uk)−1 with U1U2 ⋯ Uk, yielding the second
order 12, generated by S1 =​ 〈+​, 1, 11〉. ℛ+​ statement in (c).
acts in simply transitive fashion on the Parts (a) and (b) show that ℛ is closed under
set of major triads, and also on the set of the formation of UTT products and inverses,
  minor triads. which is sufficient to ensure that ℛ is a sub-
group of 𝒰 as (d) asserts, but we will show this
We will outline proofs of all parts of this theo- in another way as well. According to the product
rem; readers are encouraged to fill in the missing rule, if U =​ 〈σU, t+​U, t−U〉 and V =​ 〈σV, t+​V, t−V〉 are
details. The properties in (a) and (b), which were UTTs, the two transposition levels of the prod-
illustrated in parts (a)–​(d) of Exercise 8.2.1, fol- uct UV are either t+​U +​ t+​V and t−U +​ t−V (if σU is +​)
low directly from the UTT product and inverse or t+​U +​ t−V and t−U +​ t+​V (if σU is −). Either way,
formulas and the definitions of the schritts and the sum of the two transposition levels of the
wechsels involved. For example, product is equal to the sum t+​U +​ t−U +​ t+​V +​ t−V ;
in other words, tt(UV) =​tt(U) +​tt(V). This equa-
WmSn = 〈−, m, − m〉〈+, n, − n〉 tion says that the total transposition function tt:
𝒰 → ℤ12 is a group homomorphism as defined in
= 〈−, m + ( −n), ( −m) + n〉
Section 6.3. The set ℛ is the kernel of this homo-
= 〈−, (m − n), − (m − n)〉 morphism as defined in Section 6.6: the set of
= Wm− n . elements U of 𝒰 for which tt(U) =​0, the identity
element in ℤ12. The kernel of a group homomor-
Part (c) of the theorem presents two use- phism is always a normal subgroup (also defined
ful properties about products of wechsels. in Section 6.6), so it follows that ℛ is a normal
The inverse of any product of wechsels may be subgroup of 𝒰. (Normality can be established in
obtained by reversing the order of the product, other ways; see Exercise 8.2.10 below.) The last
and any product of an odd number of wechsels assertion in (d), the simple transitivity of the
remains unchanged if the order is reversed. In action of ℛ on triad, follows from the observa-
particular, these properties hold for products of tion that if triads X and Y are given, there are
P, R, and L in any combination. Parts (c) and (e) twelve different UTTs U for which U(X) =​ Y, but
of Exercise 8.2.1 offered two illustrations in the only one of the twelve is Riemannian—​the one
equations (RL)−1 =​ LR and PLR =​ RLP. The first for which t− =​−t+​.
of these equations involves even-​length prod- The dihedral structure of ℛ is illustrated
ucts, so the inverse is required; the second state- by either of the two Cayley diagrams in Figure
ment, involving odd-​length products, includes 8.2.3, each of which shares the structure of the
no inverses, but the equation would remain true Cayley diagram for the dihedral group 𝒯ℐ12 from
if written in the form (PLR)−1 =​ RLP. Figure 6.2.10b. The graphs in Figure 8.2.3 are
To see why these properties hold, recall constructed using two different pairs of genera-
from Exercise 5.3.3 that group inverses gener- tors for ℛ. In (a) the generators are the schritt
ally reverse the order of products: the equation S1 and the parallel transformation P (=​ W0).
(a1a2 ⋯ ak)−1 =​ ak−1 ⋯ a2−1a1−1 holds for any ele- Following the usual practice for Cayley dia-
ments a1, a2, …, ak in any group. In particular, grams, the vertices are labeled with elements of

Triadic Transformations • 307

FIGURE 8.2.3 Two Cayley diagram for the Riemann group ℛ: (a) on the generators S1 and P; (b) on the
generators LR and R, acting on the space triad

the group. The solid arrows representing S1 point is generated by L and R alone. (It follows that
from Sn to Sn+​1 and from Wn to Wn−1, reflecting P must be expressible in terms of L and R; see
the identities SnS1 =​ Sn+​1 and WnS1 =​ Wn−1. The Exercise 8.2.5 below.)
dashed edges representing P are bidirectional The cyclic structure of the schritt group ℛ+​,
arrows showing that SnP =​ Wn and WnP =​ Sn. generated by S1 as Theorem 8.2.2f asserts, is
All of these equations are special cases of the probably clear; it follows from the product for-
formulas in Theorem 8.2.2a. Figure 8.2.3a also mula SmSn =​ Sm+​n for schritts. Of course (S1)n =​
demonstrates that S1 and P generate the entire Sn, and (S1)12 is S0, the identity. For a proof that
group ℛ, the first statement in part (e) of the the subgroup ℛ+​ is normal, see Exercise 8.2.10
theorem. below. If X and Y are any two triads, we already
The arrow pattern in 8.2.3b matches that know there is exactly one Riemannian U such
in (a), but here the solid arrows represent the that U(X) =​ Y; if X and Y are of the same mode,
transformation LR (=​ S7) while the dashed that U must be mode-​preserving and therefore
lines represent R. In this diagram the vertices a schritt. Therefore ℛ+​ acts in simply transitive
are labeled not with group elements but with fashion both on the set of major triads and on
the elements of a set on which the group acts the set of minor triads. While the twelve schritts
in simply transitive fashion, namely the space form a group, the twelve wechsels do not, since
triad. (For this use of Cayley diagrams, recall the set of wechsels is not closed under the for-
the discussion in Section 6.2 comparing Figures mation of products—​the product of two wech-
6.1.14b and 6.2.5b.) The arrangement of triads sels is a schritt, not a wechsel—​and does not
here is the circle-​of-​fifths arrangement of con- contain the identity S0.
sonant triad space with relative keys aligned,
encountered previously in Figure 4.1.4; the pres- Exercise 8.2.4 Redraw Figure 8.2.3b as a
ent figure may be seen as a Riemannian inter- traditional Cayley diagram for ℛ whose vertices
pretation of this familiar diagram, contrasted are labeled with elements of the group. Use the
with the reading implicit in the earlier figure labels Sn and Wn, putting the identity S0 in place
involving T7. Figure 8.2.3b shows that LR and of  
the C-​major triad in the figure.
R together generate the group ℛ, from which it
follows immediately that L and R generate ℛ, Exercise 8.2.5 Show how the Riemannian
the second statement in (e). The Riemann group UTTs S3, S6, W3, and W6 can be written as
is often called the PLR group, understood to be products of S1 and P. Also show how the same
the group generated by P, L, and R. That is true, UTTs, as well as P (=​ W0), can be written as
but P is redundant in this list of generators: ℛ products
   of L and R.9

308 • E x p l oring M usica l S p aces

Exercise 8.2.6 (b) For which pairs of mod-​12 integers m and n
(a) Compare the formulas for products and   do Wm and Wn generate ℛ?
inverses of schritts and wechsels from
parts (a) and (b) of Theorem 8.2.2 with the Exercise 8.2.10
corresponding formulas for transpositions (a) We can show directly that the group ℛ is a
and inversions from Theorem 5.5.6. The normal subgroup of 𝒰 via a method from
formulas are similar but not identical; if Section 6.6, by showing that ℛ contains
each S in Theorem 8.2.2 is replaced with T the conjugates of all of its elements by all
and each W with I, the resulting formulas elements of 𝒰. If U ∈ ℛ and V ∈ 𝒰, explain
are not correct. In other words, the mapping why tt(V−1UV) must equal 0, and therefore
f: ℛ → 𝒯ℐ12 defined by f(Sn) =​ Tn and V−1UV ∈ ℛ.12
f(Wn) =​ In is not an isomorphism between (b) In the same situation, if U ∈ ℛ+​ (that is, if
the Riemann group ℛ and the transposition-​ U is a schritt rather than a wechsel), explain
inversion group 𝒯ℐ12. Use the formulas from why V−1UV must also be in ℛ+​; conclude
both theorems to show that f is an anti-​   that the subgroup ℛ+​ of 𝒰 is also normal.
isomorphism between the two groups.10
(b) Compare the Cayley diagram for ℛ in Figure The simple transitivity of ℛ simplifies the
8.2.3a with the Cayley diagram for 𝒯ℐ12 in problem of determining whether two different
Figure 6.2.10b. Schritts and transpositions composites of P, L, and R determine the same
are arranged identically in the two pictures, function. We know from Section 2.3 that if f and
but the arrangement of the inversions is g are functions such that f(x) =​ g(x) for some par-
reflected relative to that of the wechsels. The ticular x, we cannot generally conclude that f =​ g,
vertex labels in the two diagrams are therefore which requires knowing that f(x) =​ g(x) for every
related by the mapping g: ℛ → 𝒯ℐ12 defined x. But if f and g belong to a group and x is an ele-
by g(Sn) =​ Tn and g(Wn) =​ I−n. The identical ment of a set on which the group acts in simply
arrow structure of the two diagrams suggests transitive fashion, then f(x) =​ g(x) implies f =​ g.
that the mapping g is an isomorphism. Use The following exercise illustrates the application
the formulas from Theorems 5.5.6 and 8.2.2 of this principle to the group ℛ.
to verify that this is the case.
(c) Define h: ℛ → 𝒯ℐ12 by the equations h(Sn) =​ Exercise 8.2.11
T−n and h(Wn) =​ In. Is h an isomorphism or (a) Show that the composite transformations
  an anti-​isomorphism? PLRL and LPRP, applied to C major, yield
the same triad. Conclude that PLRL =​ LPRP.
Exercise 8.2.7 Show that for any mod-​12 (b) Show that LRLR and RPRL, applied to F
integer k, the mapping f: ℛ → ℛ defined by minor, yield the same triad. Conclude that
f(Sn) =​ Sn, f(Wn) =​ Wn+​k is an automorphism LRLR =​ RPRL.
  isomorphism from the group ℛ to itself).
(an (c) Is it true that RLRP =​ LRPR?
(d) Is it true that PLRLR =​ RPRPL?
Exercise 8.2.8 Many other pairs of (e) One of the two statements in (c) and (d)
Riemannian transformations generate the is false. Can you make it true by changing
group ℛ besides the two pairs mentioned in the last transformation in one of the two
Theorem 8.2.2e. Some of the five pairs listed   products?
below generate ℛ while others do not. Which
are the generating pairs?11 Exercise 8.2.12
(a) Exercise 4.2.15 analyzed the chord
(a) S4 and S9 succession D–​d–​B♭–​b♭–​D♭–​d♭–​A–​f♯–​D from
(b) S4 and W9 (=​ R) Schubert’s “Trout” Quintet as a series
(c) S7 and W3 of moves in a tonnetz. Write the string
(d) W0 (=​ P) and W4 (=​ L) of Riemannian transformations that
(e)  W2 and W7 produces this series of triads. Because
the progression begins and ends on the
Exercise 8.2.9 Generalize the previous exercise same triad, the simple transitivity of
by answering the following questions: ℛ implies that the composite of all the
transformations in the string must equal
(a) For which pairs of mod-​12 integers m and n T0, the identity. Use properties of these
do Sm and Wn generate ℛ? transformations to reduce the string to the

Triadic Transformations • 309

 identity. (For example, you may replace no common tones and voice leading that is awk-
PLP with LPL, or PLR with RLP, or you may ward at best. Furthermore, we will see below that
eliminate two consecutive Ps.) some of the same transformation groups that
(b) Rewrite the same string of transformations act on triads may act also on spaces of seventh
using the Wn notation and use the product chords, twelve-​ tone rows, and other musical
formulas from Theorem 8.2.2a to show that entities, and voice leading in many of these con-
  the composite transformation is S0. texts bears little resemblance to the triadic case.
Because group structure and voice leading have
To derive the structure of the Riemann group been commingled in the neo-​Riemannian litera-
using UTTs defined by root-​interval relation- ture, this point is worth reiterating: the algebraic
ships, as above, is far from the only possible structure of a musical transformation group does
way to describe this group. In fact, the dihedral not determine the voice-​leading properties of the
structure of ℛ was established before UTTs were transformations involved.14
introduced into the literature.13 One can, for
example, define P, L, and R as contextual inver- —​—​—​
sions (for instance, R inverts a triad to preserve
its major third) and show that they generate a We turn now to an examination of several
group isomorphic to 𝒟12. The latter approach Riemannian transformations besides P, L, and
highlights distinctive features of P, L, and R; for R that have proven to be of much theoretical
example, two triads share two common tones if and analytical interest. The reader will note in
and only if they are in one of these three relation- this discussion that wechsels generally receive
ships. Such properties have played central roles more attention than schritts. Several justifi-
in the development of neo-​Riemannian theory, cations may be cited for this imbalance, chief
are essential to the status of neo-​Riemannian among them the involution property of wech-
triad space as the dual graph of the triangulated sels (Theorem 8.2.2b). A wechsel pairs each
tonnetz (Figure 4.2.8), and contribute to the triad uniquely with another in a reciprocal rela-
musical coherence of PL-​cycles and many other tionship.15 While wechsels are transformations
triadic progressions of which we saw examples (functions), their status as involutions allows us
in Chapter 4 (and more that we will see in this to think of them also as a special kind of binary
section). The root-​interval approach, in contrast, relation between triads; indeed, we have already
situates P, L, and R democratically among nine described P, L, and R at times as relations.
other wechsels in a way that clarifies algebraic Twelve distinct root relationships are possible
relationships but may leave the reader guess- between two triads of opposite mode, and the
ing about the reasons for the disproportionate twelve wechsels Wn become apt descriptors for
attention paid to these three transformations. these relationships. Wechsels, the prototypical
We noted in Section 8.1 that the definition of contextual inversions, are more convenient for
a triadic transformation implies nothing about this purpose than the fixed inversion operators
voice leading at the level of individual pitch In, because (owing to the non-​uniformity of In) a
classes. Algebraic properties of transformations root relationship cannot be determined from an
are likewise distinct from voice-​leading behavior, In label.16 Schritts, not being involutions, display
and algebra confers no special privileges upon P, no reciprocity of this kind: for example, S2(D) =​
L, or R. Theorem 8.2.2e tells us that L and R gen- E but S2(E) ≠ D. In the case of two triads of the
erate the group ℛ, but by Exercise 8.2.9, many same mode such as D major and E major, there
other pairs of schritts and wechsels generate ℛ is seldom a compelling reason to identify the
as well. Moreover, given any two wechsels Wm relationship as a schritt S2 rather than a trans-
and Wn, by Exercise 8.2.7 we can always con- position T2—​a choice to be considered further
struct an automorphism of ℛ that maps Wm to in conjunction with Figure 8.2.21 and Example
Wn. The possibility of constructing such auto- 8.2.27 below.
morphisms means in effect that all twelve wech- A wechsel that has attracted abundant atten-
sels play indistinguishable roles in the algebraic tion is the hexatonic pole relation, illustrated in
structure of the group ℛ—​but these wechsels Figure 8.2.13. Disregarding the arrows labeled
include not only P, L, and R, celebrated for their H for the moment, the PL-​cycle in part (a) was
common-​tone retention and smooth voice lead- presented previously in Figure 3.5.8b; (b) shows
ing, but also other wechsels such as W10, which as the same cycle in a tonnetz. The cycle is maxi-
a triadic transformation (C → b♭, c → D) exhibits mally smooth in the sense that adjacent triads

310 • E x p l oring M usica l S p aces

FIGURE 8.2.13 The hexatonic pole relation H: (a) in a PL-​cycle; (b) in the neo-​Riemannian tonnetz;
(c) characteristic voice leading

are connected by single-​semitone voice leading relation engage all six notes of the governing
in one voice. (Among Riemannian transforma- hexatonic collection.
tions only P and L have this property; R pre- A pair of hexatonic poles cannot belong to
serves two common tones but moves the third a single diatonic collection. They cannot even
voice by whole step.)17 The reader may also recall come close: if a triad belongs to a diatonic col-
Figure 3.3.6, a common-​tone graph of the tri- lection dpc(m) and its hexatonic pole belongs
ads in this cycle. The division of the octave into to dpc(n), then m and n must differ by at least
three major thirds is characteristic of the PL-​ 3. For example, the C-​major triad belongs to
cycle, and this cycle engages only the notes of dpc(−1), dpc(0), and dpc(+​1) (the diatonic col-
the hexatonic collection hex34 =​{0, 3, 4, 7, 8, lections from one flat to one sharp), while its
11}, the pitch classes in two adjacent columns hexatonic pole G♯ minor belongs to the distant
of the triangulated tonnetz in the background collections dpc(+​4), dpc(+​5), and dpc(+​6).
of Figure 8.2.13b. Presumably because of this tonal remoteness,
Two vertices in a 6-​cycle can be no more than direct juxtaposition of hexatonic poles is uncom-
three positions apart. Triads at that maximum mon in music before the late nineteenth century.
distance in a PL-​cycle, such as C major and G♯ Figure 8.2.14 shows two often-​cited passages in
minor, are hexatonic poles. Such triad pairs are Wagner that exemplify the exotic, “uncanny”
opposite in mode and related by the Wechsel W8, sound often associated with this juxtaposition.19
denoted H in Figure 8.2.13.18 From either part The hexatonic pole in (a) acts as a chromatic
(a) or (b) of the figure it is clear that one may neighbor chord to a tonic triad, while in (b) the
get from C major to G♯ minor via either PLP or combination of H and L leads to other distant
LPL; as a transformation, therefore, H =​W8 =​ tonal regions. More examples of hexatonic poles
PLP =​ LPL, consistent with the calculation in will be encountered in exercises later in this sec-
Exercise 8.2.1f. Hexatonic poles share no com- tion. So foreign to diatonic tonality are these
mon tones but are related by single-​semitone progressions that merely spelling them is prob-
displacements in all three voices, as shown in lematic. In Figure 8.2.14a, Wagner writes each
Figure 8.2.13c. Taken together, two triads in this linear interval as a minor second but is forced

Triadic Transformations • 311

FIGURE 8.2.14 Hexatonic poles in two excerpts by Wagner: (a) Das Rheingold, Scene 4, m. 3836; (b)
Parsifal, Act III, mm. 1098–​1100

into a nonsensical spelling of the E-​minor chord; use the tonnetz geometry from Figure 4.2.5b, in
in 8.2.14b, as in 8.2.13c, the triads are spelled which the P, L, and R edges are of equal length
coherently but chromatic semitones appear in and the hexagons are regular. Within one of
the voice leading. these hexagonal cells, the three pairs of oppo-
Algebraically, P and L generate a group of sitely positioned triads exhibit three different
order 6, the hexatonic triad group, studied previ- relationships. These three wechsels have been
ously in Exercises 6.2.16 and 6.3.14. In schritt-​ dubbed Pʹ, Lʹ, and Rʹ by Robert Morris; the Pʹ
wechsel terminology this group consists of all arrow runs parallel to the P edges in the ton-
transformations Sn and Wn for which n ≡ 0 (mod netz, Lʹ parallel to L, and Rʹ parallel to R.20 Parts
4): the schritts S0 (=​ T0, the identity), S4, and (b) and (c) of Figure 8.2.15 show the same three
S8, and the wechsels W0 (=​ P), W4 (=​ L), and W8 transformations in relation to a starting triad of
(=​ H). The PL-​cycle graph (Figure 8.2.13a with- C major.
out the H arrows) is a Cayley diagram for this We list here several properties of Pʹ, which
group on the generators P and L. Although the apply with appropriate adjustments to Lʹ and Rʹ
graph is a cycle, the group is not cyclic: two differ- as well:
ent generators are involved, and the group is not
even commutative. Exercise 6.3.14 showed that • Pʹ relates triads at diametrically opposite
the hexatonic triad group is isomorphic to the vertices of one PLR hexagon, in the direction
symmetric group 𝒮3, or equivalently the dihedral parallel to the P edges. From any triad in
group 𝒟3. As the hexatonic triad group acts on the tonnetz (such as from C major in Figure
triad, its orbits are the four hexatonic cycles 8.2.15b), P and Pʹ point in opposite directions.
(the triads in hex01, …, hex34), and the group • Pʹ =​ LPR =​ RPL. The two products are equal by
acts in simply transitive fashion on each orbit, so Theorem 8.2.2c, and P occupies the central
we may regard this group as the subgroup of ℛ position in both of them.
appropriate for navigating among the six triads • Because Pʹ relates C major and C♯ minor, or by
of a particular cycle. direct calculation of the product LPR or RPL,
PL-​cycles are not the only 6-​cycles formed Pʹ is the wechsel W1.
by Riemannian transformations. Another hex- • P preserves two common tones of any triad to
agonal assemblage of triads familiar to us from which it is applied; Pʹ preserves one common
Chapter 4 is the PLR-​cycle, an example of which tone. When P is applied to a triad, its root
is given in Figure 8.2.15a. The figure shows a and fifth are unchanged while its third moves
portion of Riemannian triad space, the dual by semitone. When Pʹ is applied to the same
graph of the triangulated tonnetz. To emphasize triad, its third is unchanged while its root and
the symmetries inherent in the relationships we fifth move by semitone in the other direction.

312 • E x p l oring M usica l S p aces

FIGURE 8.2.15 The wechsels Pʹ, Lʹ, and Rʹ: (a) in a PLR-​cycle; (b) in relation to C major; (c) character-
istic voice leadings

• P may be defined as the contextual inversion of term nebenverwandt.22 Lʹ, like Pʹ, features paral-
a triad about its perfect-​fifth interval. Pʹ may lel semitonal motion in two voices. While paired
be defined as the contextual inversion about Lʹ-​related triads may be aurally unremarkable,
the triadic third, the note that does not belong examples below will show that Lʹ often partici-
to the perfect-​fifth interval. pates in sequential motion and in character-
istic ways of navigating certain regions in the
Exercise 8.2.16 Summarize properties of Lʹ and tonnetz.
Rʹ   
corresponding to the above properties of Pʹ. Analytical invocations of Rʹ, the W7 relation
between C major and G minor, are conspicuously
Both Pʹ (W1) and Lʹ (W5) have been widely rarer than those of Pʹ or Lʹ. The voice leading asso-
studied, though often under other names. It ciated with Rʹ is less smooth, two voices moving
was Lewin ([1987] 2007, 178) who first defined by whole step. In tonal contexts, the relation is
the slide transformation as the operation that common enough: the ii–​V progression in a major
“preserves the third of a triad while changing its key is an Rʹ relation. If either of two Rʹ-​related tri-
mode,” which is precisely Pʹ; the name slide has ads is the tonic, the presence of the other imparts
been abbreviated to S by Cohn (2012) and oth- a modal coloration, mixolydian I–​v or dorian i–​
ers.21 The startling chromatic effect of Pʹ can rival IV—​impressions that are indeed pronounced in
that of H, as at m. 29 in the second movement the few examples of Rʹ that we will encounter (see
of Schubert’s String Quintet in C Major, D. 956, Figure 8.2.21 and Exercise 8.2.24).
when the opening section’s serene E major gives The transformations Pʹ, Lʹ, and Rʹ offer new
way without warning to a tempestuous F minor; and expedient ways to describe certain motions
or at the moment of transfiguration (m. 229) in in a tonnetz. For example, Exercise 4.2.13f con-
Schoenberg’s Verklärte Nacht, when the chilly sidered the progression
loneliness of E♭ minor is dispelled by a warm
D-​major triad. f–​A♭–​c–​E♭–​g–​B♭–​e♭–​G♭–​b♭–​D♭–​f
The relation Lʹ (W5), as between C major
and F minor, lacks the exotic character of Pʹ, from Chopin’s F-​minor Fantasy, in which an RL-​
being ubiquitous in tonal music as the relation chain, ascending by thirds, is disrupted by one
between a minor tonic and its major dominant irregularity, the B♭ –​e♭ move. We may now ana-
or between a major tonic and its minor subdomi- lyze that irregularity as a substitution of Lʹ for
nant. Cohn’s label for Lʹ is N, from Weitzmann’s L. While a strict RL-​chain would cycle through

Triadic Transformations • 313

FIGURE 8.2.17 Hexatonic triad families and hexatonic bridge families in the tonnetz

all 24 triads before repeating, the substitution the smooth voice leading associated with P and
enables Chopin to return to the starting key of F L makes a PL-​chain the most common way to
minor far sooner.23 move through a hexatonic triad family (as in
Like the hexatonic pole transformation, Pʹ, Lʹ, some examples from Section 4.2), other paths
and Rʹ participate in subgroups of ℛ appropriate are possible: the Parsifal example from Figure
for navigating through certain tonnetz regions, 8.2.14b explores hex23 by alternating H and L.
several of which are illustrated in Figures 8.2.17 Pairings of two different wechsels provide the
and 8.2.18. The vertical axis is emphasized in most systematic routes through all the regions
8.2.17, which groups together triads related by we will study, and many of our most interesting
major thirds, while 8.2.18 similarly highlights examples are of this kind.24
the horizontal axis and minor-​third relations. The shaded regions labeled hb0, …, hb3 in
The four hexatonic triad families encircled in Figure 8.2.17 group the triads differently, pair-
8.2.17 are familiar to us as PL-​cycles, each con- ing the three major triads from one hexatonic
taining three major and three minor triads. Each triad family with the three minor triads from
hexatonic triad family is labeled in the figure with an adjacent family. Accordingly we call these
the name of the hexatonic collection (hex01, …, regions hexatonic bridge families. The number-
hex34) comprising the pitch classes of its tri- ing is such that, for example, family hb2 bridges
ads. The toroidal topology of the tonnetz loops hexatonic collections hex12 and hex23. Family
the six triads of each family into a cycle in the hbn is centered on an axis of pitch classes consti-
vertical direction and brings family hex34 on tuting the augmented triad that contains pitch
the left side of the figure adjacent to hex01 on the class n; for example, the central axis of hb2 is
right. These families are the orbits of the hexa- the D-​augmented triad, labeled at the top of the
tonic triad group {S0, S4, S8, W0, W4, W8}, whose axis. The full pitch-​class content of the triads in
three wechsels are P, L, and H. Within each fam- hb2 is the enneatonic collection enn123 =​{1, 2, 3,
ily, triads opposite in mode are related by these 5, 6, 7, 9, 10, 11}. The same-​mode triads within a
three wechsels, as shown for the C-​major triad in hexatonic bridge family are related by the major-​
the hex34 region. The same-​mode triads within third schritts S4 and S8; opposite-​mode triads are
each family, aligned vertically in the figure, are related by W1, W5, and W9, also known as Pʹ, Lʹ,
related by S4 and S8, not shown explicitly. While and R. Dashed arrows in the figure show the Pʹ-​,

314 • E x p l oring M usica l S p aces

FIGURE 8.2.18 Octatonic triad families and octatonic bridge families in the tonnetz

Lʹ-​, and R-​partners of C major within hb0. This Octatonic bridge families have not been
set of transformations {S0, S4, S8, W1, W5, W9} is widely studied, but hexatonic bridge families
another group (see Exercise 8.2.22 below), the correspond closely with what Cohn has called
hexatonic bridge group, whose orbits are the four Weitzmann regions. The six triads in a hexatonic
hexatonic bridge families. Just as some musical bridge family are minimal displacements of the
passages use P, L, and H to navigate through a augmented triad that forms the family’s central
hexatonic triad family, others explore a hexa- axis in the tonnetz, each differing from that aug-
tonic bridge family via some combination of Pʹ, mented triad by a single semitone in one voice.
Lʹ, and R. For Cohn, a Weitzmann region includes not only
The triadic groupings in Figure 8.2.18 are the six consonant triads but also the augmented
oriented horizontally in the tonnetz, along triad itself, regarded as an essential member
the minor-​third dimension, but are otherwise of the region.25 An alternative picture facilitating
closely analogous to the major-​third–​oriented this conception is shown in Figure 8.2.19. The
families from 8.2.17. An octatonic triad family, toroidal topology of the tonnetz is abandoned
encircled in the figure, contains the eight triads here in favor of a symmetrical graph that dis-
of a PR-​cycle (see Exercise 4.2.13c for an exam- plays the four augmented triads as links between
ple), whose notes form an octatonic collection, hexatonic triad families. Edges in this graph,
oct01, oct12, or oct23. Each of these families including the dashed edges forming connections
is an orbit of the octatonic triad group {S0, S3, with the augmented triads, exclusively indicate
S6, S9, W0, W3, W6, W9}, whose wechsels include single-​semitone voice-​leading relationships,
P (=​ W0) and R (=​ W9). Shaded in Figure 8.2.18 so the less smooth Pʹ, Lʹ, and R relationships
are the octatonic bridge families ob0, ob1, and ob2, are not shown explicitly as they were in Figure
each of which combines the four major triads of 8.2.17. Figure 8.2.19 is inspired by the graph
one octatonic triad family with the four minor called “Cube Dance,” devised by Jack Douthett.26
triads of another. Each octatonic bridge family The reader should be able to visualize the four
is an orbit of the octatonic bridge group {S0, S3, S6, segments of this graph stretching from one aug-
S9, W1, W4, W7, W10}. Three of the four wechsels mented triad to the next as distended cubes; the
in this group are Pʹ (=​ W1), L (=​ W4), and Rʹ (=​ W7). cube on the left side replicates the hexatonic
The fourth wechsel is W10, the tonnetz pole, which triad graph from Figure 3.1.5, which was deliber-
as noted in Exercise 4.2.17b is the remotest rela- ately structured in the form of a cube.
tion possible as defined by distance in the ton- The solid edges within the hexatonic triad
netz, though it is familiar from the progression families of Figure 8.2.19 represent P and L, but
iv–​V in a minor key. no transformations adequately model the dashed

Triadic Transformations • 315

FIGURE 8.2.19 Another picture of hexatonic triad families and hexatonic bridge families (after
Douthett, “Cube Dance”)

lines, inasmuch as UTTs and other conventional Figure 8.2.21 also occurs twice in succession,
triadic transformations are not defined on aug- this one in the coda of the finale of the Sonatine;
mented triads. Exercise 3.5.11 addressed the here the harmonies are simple triads but the
difficulty of defining such mappings; the func- pattern is less regular. The major triads come
tions defined there operate in one direction only, from the minor-​third cycle F♯ –​A –​C –​E♭ –​F♯,
from consonant triads to augmented triads but but the minor triads include two from the ton-
not vice versa, and because they are not one-​to-​ netz row above and one from the row below, so
one they cannot participate in transformation that the progression as a whole veers asymmet-
groups. The important voice-​ leading relation- rically between ob1 and oct01.
ship between consonant and augmented triads A strictly chronological Riemannian analy-
is more satisfactorily addressed through a geo- sis of the tonnetz path in Figure 8.2.21b would
metric approach than a transformational one. read
We will return to figures much like Figure 8.2.19
R ′ ♯ S4 S3 ♯
F♯ →  E♭ →
as we begin our geometric study of voice-​lead- S5 L S3
 c → a → e → C → F .
ing spaces in Chapter 10 (see in particular Figure
10.2.11); for the time being we set this figure
aside and return to the tonnetz-​based depictions Logical organization is hard to discern in this
of the regions in Figures 8.2.17 and 8.2.18.27 reading; in particular, the schritts S4 and S5 seem
The two short but dramatic excerpts from to contravene the minor-​third cycle, where S3 or
piano works of Ravel in Figures 8.2.20 and S9 might be expected. The analysis in 8.2.21c is
8.2.21 explore some of these regions. At the more satisfactory: here the minor-​third cycle is
astonishing climax of “Ondine” (Figure 8.2.20), shown as a T3-​chain even though these chords
the extravagant texture, extended tertian are not consecutive in the score. One of these tri-
harmonies, and whole-​ tone melody disguise ads, A major, does not occur at all, but the analy-
the simplicity of the underlying progression, sis shows A major as an important nexus binding
which cycles through hexatonic bridge fam- together its T3 relatives and the minor triads.
ily hb2 via a regular alternation of R and Lʹ. A The A-​minor triad may be heard as a parallel
second complete traversal of the same cycle substitution for the absent A major, and the C♯-
follows in the next measure.28 The passage in minor and E-minor triads hint at the mixolydian

316 • E x p l oring M usica l S p aces

FIGURE 8.2.20 (a) Ravel, Gaspard de la nuit, I, “Ondine,” m. 66; (b) path through hb2 in the tonnetz

flavor associated with the Rʹ relation. That modal division of the octave is traversed over the three
coloration is one of several aspects of this coda segments of a two-​bar hemiola in ; here the sec-
that recall earlier elements in the piece: ♭7̂ plays ond half of the cycle is compressed into the final
important melodic and harmonic roles in all hemiola segment by omission of the Rʹ relatives
three movements of the Sonatine, and octatoni- of the C and E♭ chords.
cism and the key of A major are especially sig- In Figure 8.2.21c, the use of the transposition
nificant in the last movement. All of these traits T3 rather than the schritt S3 warrants commen-
are on prominent display in an extended passage tary. The convenience of wechsel notation for rela-
in mm. 60–​94, which prolongs an A-​major har- tions between triads of opposite mode was noted
mony, repeats the motive A–​G♮–​E unceasingly, above, and in the company of wechsels, one might
and inhabits oct01 exclusively.29 assume that algebraic consistency should dictate
The two Ravel examples share a metrical the use of schritts for same-​mode relationships.
peculiarity. In each of them, the octave divi- Indeed, in a strictly Riemannian analysis of the
sion is at odds with the metric structure, neces- Ravel passage, the horizontal arrows could only
sitating adjustments in harmonic rhythm. In be labeled S3. When non-​Riemannian transposi-
“Ondine,” a three-​part division of the octave is tions mingle with the Riemannian P, L, and Rʹ,
traversed over a 44 measure; Ravel halves the the transformations do not all belong to one sim-
harmonic rhythm in the second half of the mea- ply transitive group. Moreover, in terminology
sure so that B minor will return on schedule at to be introduced in Section 9.2, the network in
the next downbeat. In the Sonatine, a four-​part Figure 8.2.21c is not path-​consistent, because the

Triadic Transformations • 317

FIGURE 8.2.21 (a) Ravel, Sonatine, III, mm. 162–​64; (b) path through oct01 and ob1 in the tonnetz;
(c) analysis

product of the wechsels Rʹ and L is not equal to Exercise 8.2.22

T3 (it is W7W4 =​ S3). For reasons like these, some (a) Construct a multiplication table for the
authors, unwilling to abide a wechsel-​transposi- hexatonic triad group {S0, S4, S8, W0, W4, W8}
tion hybrid, might advocate the use of schritts in and construct an isomorphism between
an analysis of this kind.30 these six transformations and a group of
Many analysts, however, have not considered mod-​3 transpositions and inversions to
themselves bound by such constraints. Cohn show that the group is isomorphic to the
(2012, 90–​95) displays a “T11 sequence” gener- dihedral group 𝒟3.
ated by P and Pʹ, whereas in fact PPʹ =​ S11; he also (b) Do the same for the hexatonic bridge group
presents a multiplication table for wechsels in {S0, S4, S8, W1, W5, W9}.
the hexatonic triad and hexatonic bridge groups (c) To what familiar group are the octatonic triad
in which, for example, the product of L and R is   group and octatonic bridge group isomorphic?
given as “T±5”—​a calculation as musically intui-
tive as it is algebraically uncomfortable. Presented Exercise 8.2.23 This exercise generalizes the
with two major triads, most of us will probably construction of the hexatonic and octatonic
describe the relationship more readily as a trans- triad and bridge groups. All subgroups of ℛ
position than as a schritt; the latter choice seem- that contain both schritts and wechsels belong
ingly requires us to imagine an alternate universe to the family of groups defined below; the only
in which the triads are minor and the interval other subgroups are cyclic groups of schritts
between them is reversed. To say that if the Ravel only, subgroups of ℛ+​.
example were in a minor key it would descend by
thirds is at best unverifiable, at worst laughable. (a) Suppose d is one of the numbers 2, 3, 4,
These considerations motivate the wechsel-​trans- or 6—​a divisor of 12—​and k is an integer
position analysis in Figure 8.2.21 and in Example with 0 ≤ k < d. Let ℛd,k denote the subset
8.2.27 below. Chapter 9 will offer further justifi- of ℛ containing all schritts Sm for which
cations for the viability of non–​path-​consistent m ≡ 0 (mod d) and all wechsels Wn for
networks. which n ≡ k mod (d). For example, ℛ4,1

318 • E x p l oring M usica l S p aces

 is {S0, S4, S8, W1, W5, W9}, the hexatonic (d) Glazunov, Symphony No. 3 in D Major, Op.
bridge group. Verify that every ℛd,k is 33, I, mm. 276–​92:
closed under products and inverses and is
therefore a subgroup of ℛ. e♭–​A♭–​c–​F–​a–​D–​f♯–​B–​e♭
(b) The hexatonic triad group is ℛd,k for which
values of d and k? What about the octatonic (e) Rachmaninoff, Sonata in G Minor for Cello
triad group and the octatonic bridge group? and Piano, Op. 19, II, mm. 116–​20:
(c) List the schritts and wechsels in the group
ℛ4,2. List the triads in one orbit of this    E–​f–​A♭–​a
group. Write a chord progression that
cycles through this orbit using the wechsels Exercise 8.2.25 The chord succession below
W2 and W10. models the essential harmonic structure of
(d) The chord progression C –​g –​D –​a –​E –​⋯ Schubert’s Fantasy in F Minor for piano four
is a cycle through an orbit of which group hands, D. 940, mm. 57–​92:
ℛd,k? (The progression uses Rʹ and Lʹ in
alternation, and the group in question is f–​C–​d♭–​E–​a–​E–​f
generated by Rʹ and Lʹ.)
(e) Any two wechsels Wm and Wn generate Though the progression is not sequential,
either a group ℛd,k or the entire group ℛ it remains within one triadic family
(which may be considered to be ℛ1,0). Given from Figure 8.2.17. Identify the family,
m and n, how can you determine the values analyze the progression using Riemannian
of d and k? What is the order of the group? transformations, and trace the path of the
In other words, how long are the cycles progression in the tonnetz and in Figure
generated by Wm and Wn? 8.2.19. Noting that the progression begins
(f) Does the definition of ℛd,k make sense if d =​ and ends on the same triad, verify that the
  12? What is the group ℛ12,9? product of all the transformations involved is
the  identity.32
Exercise 8.2.24 Each of the triadic
progressions below traces a path through Exercise 8.2.26 Progressions (a)–​(e) below
one of the triadic families in Figure 8.2.17 or do not remain within any of the named
8.2.18. Most of the progressions are sequences; triadic families we have identified, but they
one is slightly irregular but remains within a are strictly sequential, generated by regular
single family. Analyze each progression using alternation of wechsels, and they trace
Riemannian transformations, identify or regular paths through the tonnetz. For each
describe the subgroup of ℛ generated by these progression, identify the transformations
transformations, identify the triadic family involved and draw or describe the tonnetz
traversed by the progression, and draw or path. Answer the additional questions in
describe the path in the tonnetz.31 (f)–​( h).33

(a) Schubert, Octet in F Major, D. 803, VI, mm. (a) Bach, Fantasia and Fugue in G Minor, BWV
172–​75: 542, mm. 31–​34:

c–​E♭–​a♭–​C♭–​e–​G–​c–​E♭ D–​g–​G–​c–​C–​f–​F–​b♭–​B♭–​e♭–​E♭–​a♭–​A♭–​d♭–​D♭

(b) Schubert, Mass in E♭ Major, D. 950, Sanctus, (b) Mozart, Requiem, K. 626, Confutatis,
mm. 1–​7: mm. 10–​12:

E♭–​b–​g–​e♭–​C♭ C–​c–​G–​g–​D–​d–​A–​a–​E–​e

(c) Liszt, Grande fantaisie symphonique über (c) Schubert, Sonata in C Minor, D. 958, II,
Themen aus Berlioz’ “Lélio,” mm. 195–​98: mm. 105–​07:

b♭–​A–​c♯–​C–​e–​E♭–​g–​F♯–​b♭–​ ⋯ C–​f–​D♭–​f♯–​D–​g

Triadic Transformations • 319

(d) Liszt, Études d’exécution transcendante, No. when the right hand moves in sixteenth notes,
8, “Wilde Jagt,” mm. 180–​84: one of the two notes of each pair consistently
completes a major or minor triad with the
E♭–​g–​D–​f♯–​D♭–​f–​C concurrent left-​hand interval. The first sequence
starts in the middle of m. 2 with the triads
(e) Prokofiev, Scherzo in A Minor, Op. 12, No. c–​E–​d–​F♯–​e–​G♯, generated by alternate
10, mm. 161–​64: application of the wechsels H (=​ W8) and W10,
shown in the analysis by solid black arrows.
C–​c♯–​C♯–​d–​D–​d♯–​D♯–​e Triads are labeled in abbreviated ordered-​
pair notation (0− for C minor, 4+​ for E major)
(f) How are sequences (a) and (b) related? How to facilitate recognition of W10 and other
are (c) and (d) related? somewhat unfamiliar transformations that
(g) Which of the progressions in this exercise occur in this example. Two-​chord segments
illustrates the same basic sequence type as are related by T2, indicated by dashed arrows
Figure 2.3.4b? Which is the same type as linking the bubbles. Each pair of hexatonic
Figure 4.5.24? poles completes a hexatonic collection, which
(h) What can you say about the group combines with its T2 transposition to form a
generated by the transformations in any of twelve-​tone aggregate in a span of four chords.
  these sequences?34 The choice of the transposition T2 in
preference to a schritt in this analysis is not
Example 8.2.27: Martin, Prelude No. 2 only intuitively appealing but technically
The second of Frank Martin’s Eight Preludes required: no schritt could take the role of T2
for piano (1948) projects a distinctive character in the figure’s bubble notation. In Riemannian
through its pervasive use of sequences that terms, successive minor triads are related
pair hexatonic poles, then transpose them by W8W10 =​ S10 while successive major triads
successively by whole steps. Figure 8.2.28a are related by W10W8 =​ S2—​but in both
shows the first phrase. The harmony is triadic; cases the transposition is up by whole step.

FIGURE 8.2.28 Triadic sequences in Martin, Prelude No. 2

320 • E x p l oring M usica l S p aces

Bubble notation, as we have remarked when does not admit satisfactory representations
using similar notations previously, requires of these sequences. Algebraically, however,
that the transformation linking the bubbles these patterns are closely akin to others whose
(here T2) commute with all transformations component transformations are more typical.
within bubbles (H). The fact that all UTTs Either the H–​W10 pair in sequence (a) or the
commute with transpositions—​the uniformity H–​W6 pair in (b) or (c) generates a subgroup of
condition—​uniquely qualifies transpositions for ℛ containing all the even-​numbered schritts
analyses of this kind. and wechsels; in the notation from Exercise
The phrase deviates from the pattern at 8.2.23 this group is ℛ2,0. An orbit of this group
both ends. The first bubble in the analysis contains six major and six minor triads, all
shows a hypothetical D-​major triad on the with roots belonging to one whole-​tone scale.
downbeat of m. 2, which fits the pattern but In Figure 8.2.28, the sequences in (a) and (c)
is supplanted in the score by its L-​partner, F♯ trace the even-​numbered (wt0) orbit, while
minor. In m. 4 C major occurs where B♭ major is in (b) the chord roots are odd; both orbits
expected, another T2 transposition. The phrase occur several times in the piece, with frequent
is bookended by the opening B-​minor triad and phrase-​level modulations between the two.
the bass B in m. 4, which stand out as the only Although these orbits do not form compact
odd numbers among the chord roots in the tonnetz regions, they are triad families
analysis, each related by semitone to the nearest somewhat analogous to those in Figures 8.2.17
chord of the sequence. and 8.2.18, and any of Martin’s sequences,
The second phrase introduces a related sufficiently prolonged, will cycle through all
pattern, shown in reduction in Figure 8.2.28b. twelve triads of one of the two families.
The structure is similar, but here the hexatonic-​ The networks in Figure 8.2.28 will be revisited
pole pairs transpose not by T2 but by T10, in Section 9.3 as illustrations of the concept of
descending by whole step. This arrangement network
   isomorphism.
entails a tritone root relation between each
major triad and the following minor triad; Exercise 8.2.29 Progressions (a)–​(e) below
chord-​to-​chord Riemannian analysis now are strictly sequential, but some of them
reveals an alternation of the wechsels H (=​ W8) include some less familiar transformations,
and W6. and none can be conveniently graphed in
The sequence patterns from (a) and (b) recur the tonnetz. Analyze each sequence using
several times throughout the short piece. Riemannian transformations, identify or
The coda introduces a new variant, shown in describe the subgroup of ℛ generated by
(c). Here, for the only time, the major triad these transformations, and describe the
comes first within each hexatonic-​pole pair. set of triads traversed by the progression.
The result is a sequence that combines the You may identify wechsels when possible
T2 transpositions from (a) with the W8W6 using familiar names such as P, H, or Lʹ, but
wechsel pattern from (b); W8 is now applied some will need to be identified by number
to major triads and W6 to minor, the opposite (for example, W10). Answer the additional
arrangement from (b). A sequence of type (c) question in (f).36
may be obtained from (a) by reversing the order
of the two chords in each pair (compare the first (a) Beethoven, String Quartet in B♭ Major,
four chords in (c) with m. 3 of (a)), or from (b) Op. 18, No. 6, IV, mm. 20–​28:
by playing the entire progression backward (in
the figure, (c) exactly retrogrades (b) a semitone B–​e–​F♯–​b–​C♯–​f♯–​G♯–​c♯–​E♭–​a♭–​ ⋯
lower).35
The H relations in Martin’s sequences are (b) Liszt, Dante Symphony, I, “Inferno,”
presented with the characteristic semitonal mm. 232–​51:
voice leading in all three voices, but the
other wechsels that appear, W6 and W10, B–​f–​D♭–​g–​E♭–​a
do not allow for comparably smooth voice
leading. Moreover, to express H or W6 as a (c) Saint-​Saëns, Symphony No. 3 in C Minor,
composite of P, L, and R requires at least Op. 78, I, mm. 63–​70:
three transformations, while W10 requires
five. For these reasons the usual tonnetz c–​G–​e♭–​B♭–​f♯–​C♯–​a–​E–​c

Triadic Transformations • 321

(d) Rimsky-​Korsakov, Legend for orchestra are of theoretical and analytical interest, as are
(Skazka), Op. 29, mm. 529–​33: many subgroups of 𝒰 besides ℛ and 𝒯12; this sec-
tion provides an overview of some of these topics.
c♯–​F–​b♭–​D–​g–​B–​e–​G♯–​c♯ Some transformations, introduced by Lewin
alongside P, L, and R, are designed to model
(e) Britten, Peter Grimes, Passacaglia, familiar tonal or functional relations between tri-
mm. 63–​69: ads, without the constraints of the Riemannian
dualism condition.37 Among these are the domi-
f♯–​F–​e–​E♭–​d nant and mediant transformations, D and M. As a
transformation, D is nothing new, as it is simply
(f) Which two of the sequences in this T5 =​ 〈+​, 5, 5〉. Lewin conceives of D as mapping
exercise are retrograde-​related in the same from a dominant to a tonic—​what happens at a
way as the sequences in parts (b) and (c) of V–​I cadence. Importantly, D is mode-​preserving:
Figure 8.2.28? it maps V–​I or v–​i, never V–​i. In a tonal context,
   D can also map I–​IV or vi–​ii, for example. The
inverse of D is D−1 =​ T7 =​ 〈+​, 7, 7〉, the subdominant
Exercise 8.2.30 Many sequential patterns transformation, which maps IV–​I (among other
of Riemannian transformations are possible possibilities).38 As transpositions, D and D−1 are
besides those represented in the above already familiar elements of our transformational
examples and exercises. Seven combinations of repertoire. Because 1, 5, 7, and 11 are genera-
wechsels are listed below. For each one, write tors of ℤ12, either D or D−1 by itself generates the
a chord progression that repeats the given transposition group 𝒯12, as do T1 and T11.
transformations cyclically at least twice. In The mediant transformation M, on the other
(a), for example, you must write five chords hand, is no mere transposition. While D maps
to cycle through P and H twice. Begin with any triad to the triad of which it is the same-​
either a major or a minor triad. Try to keep all mode dominant, M is defined so that it maps
voices strictly sequential, but also strive for any triad to the triad of which it is the opposite-​
appropriate voice leading. (These two guidelines mode diatonic mediant: thus M maps C major
may conflict; you may find more satisfactory to A minor and C minor to A♭ major. As a UTT,
solutions by writing in four voices rather than M =​ 〈−, 9, 8〉; its inverse, the submediant trans-
three, and you may find it helpful to repeat one formation M−1 =​ 〈−, 4, 3〉, maps C → e and c → E♭.
chord in a pattern twice in succession, with Two successive applications of M, as in C → a → F
different voicing.) Trace the pattern in the or a → F → d, always yield the same-​mode triad a
tonnetz, and identify the triad region traversed fifth lower, illustrating the equation M2 =​ D. In
if appropriate. the transposition group 𝒯12, D has no “square
root”—​there is no transposition Tn such that
(a) P, H (Tn)2 =​ T5—​but in the larger UTT group, M is just
(b) R, Rʹ such a transformation.39
(c) Pʹ, Lʹ
(d) Pʹ, Rʹ Exercise 8.3.1
(e) Pʹ, Lʹ, Rʹ (a) Use the product formula for UTTs to
(f) L, P, R, P confirm that M2 =​ D.
(g) Pʹ, R, Pʹ, L (Start (f) and (g) on the same major (b) M is not the only UTT whose square is D;
  triad, and compare the two sequences.) give an example of another.
(c) More generally, explain why if U is any
8.3 OTHER TOPICS IN mode-​reversing UTT, then U2 is always
TRIADIC TRANSFORMATION equal to the transposition Tn, where
  n =​tt(U).
THEORY
Only one-​twelfth of the 288 UTTs are Riemannian, The six consonant triads in any diatonic col-
but to this point, the only non-​Riemannian UTTs lection are adjacent along an M-​chain:
about which we have had much to say are the
transpositions Tn =​ 〈+​, n, n〉. Many other triadic
G M
→ e M
→ C M
→ a M
→ F M
→d
transformations, both uniform and non-​uniform,

322 • E x p l oring M usica l S p aces

The same triadic succession may also be described as part of an M-​chain. Our discussion of the
as an RL-​chain: earlier figure alluded to the importance of
descending-​third relationships in functional
harmony, an observation that gives M-​based
G R L
 C R
→ e → L
 F R
→ a → →d analysis a certain appeal for common diatonic
progressions:
As a transformation, M =​ 〈−, 9, 8〉 is not the
same as either R =​ 〈−, 9, 3〉 or L =​ 〈−, 4, 8〉, but it 2 −5 3
shares one transposition level with each of them, I M
→ vi M→ ii M
→ V M→ vi M
→
−4
so that M behaves like R when it is applied to a IV M
→ V M→ I
2

major triad and like L when applied to a minor

triad. The submediant transformation M−1 =​
A Riemannian (schritt-​wechsel) analysis of this
〈−, 4, 3〉 flips the pairings of transposition lev-
progression would be unrevealing, but because
els with R and L and reverses the arrows in the
of the proximity of the diatonic triads in the M-​
M-​chain:
chain, the relationship between any two such
triads may always be modeled by a small positive
M −1 M −1 M −1 M −1 M −1 or negative power of M. Functional progressions
d  → F  → a  → C  → e  →G
typically mix small positive powers of M with
R L R L R a few larger negative powers, as in the above
d  → F →  a  → C → e  →G example, which intersperses five small moves to
the right in the M-​chain with two larger jumps
For either of these chains of thirds, the M-​based to the left (M−5 and M−4). The common root
analysis could be said to have the advantage motions by descending third, descending fifth,
of efficiency, as it models the entire chain by and ascending step are described in most cases
repeated application of a single transformation. by the transformations M, M2 (=​ D), and M3,
The RL analysis makes it clear that two differ- respectively, all corresponding to counterclock-
ent triadic relationships appear in alternation, wise motion in the circle and to small rightward
something the M analysis disguises. An RL-​ moves in the M-​chain. The exceptions are the
chain, however, may either ascend or descend by moves from pre-​dominant to dominant, ii–​V and
thirds, whereas M-​based analysis distinguishes IV–​V in the above example. These are also coun-
the two: an M-​chain always descends by thirds, terclockwise in the circle, but they occur at the
while its ascending-​ thirds counterpart is an point where continued rightward motion in the
M−1-​chain. M-​chain would escape the family of diatonic tri-
Figure 8.3.2 redraws Figure 4.3.8, a circle-​ ads (and where the diminished triad intervenes
of-​
thirds arrangement of diatonic triads, in the circle), so the leftward leap compensates.
along with a diagram showing the same triads If M2 is applied to the ii chord, the result is not

FIGURE 8.3.2 Diatonic triads in a diatonic circle of thirds and in an M-​chain

Triadic Transformations • 323

V but the non-​diatonic minor v; the seven posi- group ℳ will likely align better with common
tions separating V and v in the M-​chain account intuitions about functional behavior.41
for the difference between M2 (ii–​v) and M−5 The groups ℛ and ℳ are not the only sim-
(ii–​V ).40 ply transitive subgroups of 𝒰, and M is not the
only UTT capable of generating chains of all 24
Exercise 8.3.3 Major V and minor v are related triads. As it turns out, there are exactly 24 sub-
by M7, but they are also related by P. Is it true groups of 𝒰 that act in simply transitive fashion
that
   M7 =​ P? on triad.42 All of these groups are of order 24;
recall from Section 5.6 that the order of a simply
Exercise 8.3.4 Recall from Section 6.6 that a transitive group always matches the cardinality
group element of the form a−1xa is called the of the set on which it acts. By Lewin’s theorem
conjugate of x by a, denoted Conja(x). If U is (Theorem 5.6.7), each of the simply transitive
a UTT, then ConjM(U) =​ M−1UM. Determine groups defines its own interval-​space structure
general formulas for ConjM(Sn) and ConjM(Wn). on the space triad. In the interval space with
What
   are ConjM(R) and ConjM(L)? the group ℳ, the interval from the G-​major triad
to the A-​minor triad is int(G, a) =​M3—​effec-
Because M2 =​ D =​ T5, it follows that M24 =​ D12 =​ tively measured on the M-​cycle, corresponding
T0, the identity. An M-​chain, continued beyond loosely to intuitions about triadic distance in
the bounds of the six triads in one diatonic diatonic progressions as discussed above. In the
collection, follows the course of the RL-​chain, interval space with the group ℛ, int(G, a) is the
cycling through all 24 major and minor triads wechsel W2, which is less easily interpreted as a
before returning to the initial triad. All powers “distance.” (Chapter 12 will introduce distance
of M up through M23 are different; as an element functions, which are distinct from interval func-
of a transformation group, M generates a cyclic tions.) Both of these interval structures on triad
group ℳ =​{T0, M, M2, M3, …, M23} of order 24, are different from the one with the group 𝒯ℐ12
the mediant group. As UTTs, the first few powers introduced in Section 5.6, in which int(G, a) =​I11.
of M are M =​ 〈−, 9, 8〉, M2 =​ D =​ 〈+​, 5, 5〉, M3 =​ 〈−, Six of the simply transitive groups, including
2, 1〉, M4 =​ D2 =​ 〈+​, 10, 10〉, M5 =​ 〈−, 7, 6〉, … . In ℳ (but not ℛ), are cyclic; each of them contains
looking at the odd and even powers separately, it eight different elements that generate the group
is not hard to see the pattern: the mode-​preserv- (this is true of any cyclic group of order 24).
ing UTTs in the group ℳ are the transpositions These 48 elements are all the UTTs of order 24,
Tn =​ 〈+​, n, n〉, while the mode-​reversing UTTs are and any of them will generate a chain through
the twelve UTTs of the form 〈−, n, n − 1〉. Because the 24 triads. The order-​24 UTTs are easy to
all 24 triads lie in the same M-​cycle, the action identify: they are the mode-​reversing UTTs U
of the group ℳ on triad, like the action of ℛ, is whose total transposition tt(U) is equal to 1, 5,
simply transitive. 7, or 11 (mod 12)—​a requirement satisfied by
The transformations appearing in an analy- M =​ 〈−, 9, 8〉, since tt(M) =​9 +​8 =​5 (mod 12). All
sis need not always remain within one simply UTTs not satisfying this condition are of order
transitive group, as we saw when mixing wech- 12 or less.
sels with transpositions in Figure 8.2.21 and
Example 8.2.27. An advantage of a simply tran- Exercise 8.3.5 Identify the eight generators of
sitive group is that it admits only one possible   cyclic group (ℤ24, +​).43
the
analysis for any transformation. If we are work-
ing in the group ℳ and we encounter the chord Exercise 8.3.6 By Theorem 8.2.2d, ℛ is a
succession G–​a, we have no choice but to identify normal subgroup of 𝒰. The subgroup ℳ, on the
it as M3 =​ 〈−, 2, 1〉, but if we are working in ℛ, other hand, is not normal. Calculate the UTT
we must choose W2 =​ 〈−, 2, 10〉 instead. The two LML, show that it does not belong to the group
analyses of the descending-​thirds chain above ℳ, and explain why this shows that ℳ is not a
(RL-​chain or M-​chain?) may therefore be seen as    subgroup of 𝒰.44
normal
byproducts of selecting a simply transitive group
in which to carry out the analysis (ℛ or ℳ?). A Just as M offers an alternative description
variety of considerations may come into play in of an RL-​chain, some other UTTs provide new
making such a choice, but the above discussion ways to conceive of some of the sequential pat-
suggests that if a transformational analysis of a terns analyzed using wechsels in Section 8.2. For
diatonic progression is desired, analysis in the example, the sequence B–​e–​F♯–​b–​C♯–​f♯–​G♯–​c♯–​ ⋯

324 • E x p l oring M usica l S p aces

was analyzed in Exercise 8.2.29a as an alterna- that it is isomorphic to 𝒞12 × 𝒞2 or equivalently
tion of two wechsels, but it could instead be ℤ12 × ℤ2 (but not to the cyclic group ℤ24, because
generated entirely by the single transformation 12 and 2 are not coprime). The double-​circle
U =​ 〈−, 5, 2〉. This is one of the order-​24 UTTs; its diagram from Figure 4.1.1a is a Cayley diagram
total transposition (7) appears in the sequence for this group. A triadic analysis in this simply
as the interval between successive same-​mode transitive group identifies every transforma-
triads, because U2 =​ T7. Writing out U3 =​ 〈−, 0, tion as some Tn or TnP, which amounts to little
9〉, U4 =​ 〈+​, 2, 2〉, U5 =​ 〈−, 7, 4〉, …, we see that more than an accounting of root movement and
this chain generates a group consisting of all mode changes. In the triadic interval space with
the transpositions Tn together with the mode-​ this group, int(G, a) =​T2P; this interval space is
reversing UTTs of the form 〈−, n, n +​9〉. This is essentially the same as the space with the group
another of the cyclic simply transitive groups; ℤ12 × 𝒮𝑖𝑔𝑛, introduced in Section 8.1, where this
its structure is similar to that of ℳ as described interval takes the form (2, −).
above. This structure is characteristic of the six
cyclic simply transitive groups, each of which Exercise 8.3.8
contains the transpositions Tn together with the (a) Use UTT notation to verify that the group
mode-​reversing UTTs of the form 〈−, n, n +​ b〉 generated by T1 and P is commutative.46
for some odd number b. (b) Analyze the chord succession G–​e–​C–​a–​F–​d
  using transformations from this group.
Exercise 8.3.7 Each of the triadic successions
below is generated by a UTT of order 24. In The UTT group has a rich and complex sub-
each case, identify the UTT U, identify the group structure.47 Some maps and transforma-
transposition Tn =​ U2 between successive same-​ tional systems for triadic space are implicitly
mode triads, and describe the elements of the based on groups containing more than 24 trans-
cyclic group generated by U.45 formations (and therefore not simply transitive)
but smaller than the full UTT group. For exam-
(a) Bach, Fantasia and Fugue in G Minor, BWV ple, Weber space, first presented in Figure 4.2.3,
542, mm. 31–​34 (from Exercise 8.2.26a): combines a horizontally oriented PR-​chain with
D relations arranged vertically. A group gener-
D–​g–​G–​c–​C–​f–​F–​b♭–​ ⋯ ated by P, R, and D therefore enables navigation
through Weber space freely in both the horizon-
(b) Verdi, La traviata, Act I, Scene 4, “Si ridesta,” tal and vertical dimensions; we call this group
mm. 18–​32 (pattern from Figure 2.3.4b): 𝒲, the Weber group. The Riemannian P and R
generate the octatonic triad group G =​{S0, S3, S6,
f–​D♭–​g♭–​D–​g–​E♭–​a♭–​E–​ ⋯ S9, W0, W3, W6, W9} from Figure 8.2.18, while D
generates the transposition group 𝒯12. Because
(c) Sequence of keys from Bach, Well-​Tempered transpositions commute with all UTTs, the
Clavier: group 𝒲 consists of all products of the form SmTn
and WmTn, where m is restricted to the values 0,
C–​c–​C♯–​c♯–​D–​d–​E♭–​e♭–​ ⋯ 3, 6, and 9. It may appear that there should be
4 ∙ 12 =​48 transformations of the form SmTn and
(d) Sequence of keys from Chopin, Preludes, another 48 of the form WmTn, but actually there
Op. 28: are only half that many, because the groups G
and 𝒯12 intersect nontrivially: the schritt S6 in G
   C–​a–​G–​e–​D–​b–​A–​f♯–​ ⋯ is the same as the transposition T6 =​ 〈+​, 6, 6〉 in
𝒯12, so the products SmTn and WmTn with 6 ≤ n ≤
Another of the simply transitive groups, dis- 11 duplicate those with 0 ≤ n ≤ 5. 𝒲 is therefore
cussed briefly in Sections 6.2 and 6.4, is the group a group of order 48, six times the size of the octa-
generated by T1 and P, which contains all trans- tonic triad group. The following exercise offers
positions Tn together with the mode-​reversing another description of the Weber group.
transformations TnP. In UTT notation, Tn =​ 〈+​,
n, n〉 while TnP =​ 〈−, n, n〉, so this group consists Exercise 8.3.9 Give a simple arithmetical
of all UTTs with two equal transposition levels. description of the UTTs in 𝒲. Your description
This group is commutative (see Exercise 8.3.8a should make it possible to look at any UTT
below) but not cyclic; we showed in Chapter 6 〈σ, t+​, t−〉 and see immediately whether it

Triadic Transformations • 325

belongs to the group or not. Is the simply • The transposition T3, as a permutation of
transitive group generated by T1 and P, triad, is of cycle type 46, consisting of six
described
   above, a subset of 𝒲?48 4-​cycles such as (C, E♭, F♯, A): three cycles of
major triads and three more of minor tri-
An important study by Brian Hyer (1995) ads. A 4-​cycle, or any cycle of even length,
develops a group of triadic transformations gen- is an odd permutation, but the product of
erated by P, L, R, and D. The Weber group does not an even number of such cycles is even, so T3
contain L, so the Hyer group is larger. Without is even.50 In fact, all transpositions of triads
recourse to UTTs, Hyer gives standardized rep- are even in the sense of permutation, even
resentations for the transformations in his group though several different cycle structures
as products of the four generators, and a table for occur among them, because the cycle struc-
multiplying them.49 The Hyer group still does not ture of a transposition always divides into
contain all UTTs, however; in particular, M can- matching halves for major and minor triads.
not be expressed as a product of P, L, R, and D. • The cycle structure of the mediant trans-
To understand the structure of the Hyer formation M is a single 24-​cycle, M =​(C, a,
group, a new definition will be helpful. A UTT F, d, …, e). Because an even cycle is an odd
U =​ 〈σ, t+​, t−〉 is even (more specifically even in the permutation, M is odd in the sense of per-
sense of total transposition) if its total transposi- mutation. Every order-​24 UTT shares this
tion tt(U) =​ t+​ +​ t− is an even number; U is odd cycle type and is therefore odd.
if tt(U) is odd. Equivalently, U is even if its two
transposition levels are of the same parity (both The reader may have noted that in the above
even or both odd), and odd if they differ in par- examples, the two definitions of even and odd
ity. All of the UTTs of order 24 are odd, because UTTs are in agreement: wechsels and transposi-
as noted above the total transposition of such a tions are even both in the sense of total trans-
UTT must be 1, 5, 7, or 11. The even UTTs include position and in the sense of permutation, while
all transpositions, whose two transposition lev- M and the other order-​24 UTTs are odd in both
els are the same, as well as all Riemannian UTTs, senses. This equivalence turns out to be com-
for which tt(U) is 0. By Theorem 8.2.2d, the func- pletely general: a UTT is even in the sense of
tion tt: 𝒰 → ℤ12 is a group homomorphism; that total transposition if and only if it is even in the
is, tt(UV) =​tt(U) +​tt(V) for all U and V. It follows sense of permutation.51 We may therefore speak
that the product of two UTTs of the same parity simply of even or odd UTTs with no ambiguity.
is always even, while the product of two UTTs of
opposite parity is always odd. Exercise 8.3.10 Determine the parity of each
There is a second sense in which UTTs may be of the following UTTs in the sense of total
called even or odd. Every UTT is a permutation transposition and in the sense of permutation,
of the elements of the set triad. From our study and verify that the two are the same.
of permutations in Section 6.1 we know that
every permutation can be written as a product of (a) The schritt S1 =​ 〈+​, 1, 11〉
cycles, which can be broken down further into a (b) The UTT 〈+​, 0, 1〉
product of 2-​cycles; a permutation is even or odd (c) The UTT M3 =​ 〈−, 2, 1〉
depending on whether the number of 2-​cycles is (d)  The UTT T3P =​ 〈−, 3, 3〉
even or odd. While different 2-​cycle representa-
tions of the same permutation are possible, the Exactly half of the 288 UTTs are even. Because
parity is the same for all of them. Accordingly, products and inverses of even UTTs are even, the
we may classify each UTT as even or odd in the 144 even UTTs form a subgroup 𝒰even of 𝒰; this
sense of permutation. A few examples illustrate group 𝒰even, which contains P, L, R, and D but not
this definition: M, is precisely the Hyer group. The same group
arises in other contexts as well. Our discussion of
• The Riemannian R, as a permutation of Figure 8.2.21 and Example 8.2.27 noted the ben-
triad, is the product of twelve 2-​cycles: R efits of analyzing triadic progressions via combi-
=​(C, a)(D♭, b♭)(D, b) ⋯ (B, g♯). In notation nations of transpositions and wechsels; as UTTs
from Chapter 6, the cycle type of R is 212. these are all even, and together they generate
It follows that R is even in the sense of per- 𝒰even. (Transpositions and wechsels do not form
mutation. All twelve wechsels share this a group by themselves, and no simply transitive
cycle type and are therefore even. group contains them all.) David Kopp (2002)

326 • E x p l oring M usica l S p aces

defines a larger set of basic transformations than Uniformity requires a transformation to com-
Hyer, but as UTTs, all of Kopp’s transformations mute with every Tm, but by Theorem 5.5.6e,
are even and expressible within Hyer’s system, so inversions instead satisfy the rule TmIn =​ InT−m.
the group structure implicit in Kopp’s theory and In the circles on the right, consequently, triads
his many analyses is also 𝒰even.52 ascend by semitone in the counterclockwise direc-
tion; the circles have been not merely rotated
—​—​—​ as uniform action would accomplish, but also
reflected. We refer to such action as anti-​uniform.
Some groups of triadic transformations even The theory of UTTs may be extended to allow
larger than 𝒰 are of interest. In particular, the anti-​uniform action; we summarize the appro-
UTT formalism may be extended to accom- priate formalism here only briefly.53 Reflections
modate the inversion operators In, which may are accommodated by the introduction of two
be regarded as mode-​reversing triadic trans- additional signs into each transformation,
formations but are not UTTs. Though inver- called the reflection factors for major and minor
sions violate the uniformity condition, they triads and labeled ρ+​ and ρ− respectively (using
do so in a way that is easy to describe. Figure the letter rho for reflection). A reflection factor
8.3.11a uses double-​ circle diagrams in the indicates whether the action of a transforma-
manner of Figure 8.1.11, here illustrating the tion is uniform (+​), in which case the circle is
action of I0, inversion about the pitch class C, as rotated only, or anti-​uniform (−), in which case
a triadic transformation. The figure shows, for it is rotated and reflected. There are two such
instance, that I0(0+​) =​5− while I0(1+​) =​4−; that factors, allowing the possibility that a transfor-
is, the C-​major triad inverts about C to F minor mation may act uniformly on major triads but
while D♭ major (a semitone higher) inverts to E anti-​uniformly on minor triads or vice versa.
minor (a semitone lower). These actions illus- These generalized transformations, which may
trate the transformational equation T1I0 =​ I0T11. be uniform, anti-​uniform, or a hybrid of the

FIGURE 8.3.11 (a) Double-​circle illustration of the anti-​uniform action of the inversion I0 as a triadic
transformation; (b) double-​circle illustration of the action of a QTT

Triadic Transformations • 327

two, are called quasi-​uniform triadic transforma- In QTT notation, inversion operators may
tions, or QTTs. A QTT is written as an ordered be expressed in the form Jn =​ 〈−, n, n, −, −〉.
5-​tuple U =​ 〈σ, t+​, t−, ρ+​, ρ−〉. For example, J0 exchanges triads in the pat-
If the reflection factors ρ+​ and ρ− are both +​, tern 0+​ ↔ 0−, 1+​ ↔ 11−, 2+​ ↔ 10−, …—​that is,
then the action of the QTT 〈σ, t+​, t−, +​, +​〉 matches C ↔ c, D♭ ↔ b, D ↔ b♭, …. At the level of pitch
that of the UTT 〈σ, t+​, t−〉. In general, the action classes, triadic inversion maps the root of one
of a QTT U =​ 〈σ U, t+​, t−, ρ+​, ρ−〉 on a triad X =​ triad to the fifth of another, so J0 corresponds
(r, σ X) may be calculated in the following stages. to the inversion we know as I7. In general, In, as
First, as usual, the root r is incremented by the a triadic transformation, is the same as J7 − n.
appropriate transposition level, either to r +​ t+​ The transformations Jn are just the inversions
(if σ X is +​) or to r +​ t− (if σ X is −). Next, if the In, renumbered; the group 𝒯ℐ12, which we have
reflection factor ρ+​ or ρ− corresponding to the generally defined as the collection of all trans-
mode σ X is −, the transposed root is replaced by formations Tn and In, could equally well be
its mod-​12 inverse; this step reflects the circle, said to comprise all transformations Tn and Jn.
changing uniform behavior to anti-​ uniform While the mapping from ℛ to 𝒯ℐ12 defined by
when appropriate. Finally, if the sign σ U of the Sn → Tn and Wn → In is an anti-​isomorphism of
QTT is −, the mode σX is changed to −σ X. the two dihedral groups by Exercise 8.2.6a, the
For example, to apply the QTT U =​ 〈−, 2, 5, mapping defined by Sn → Tn and Wn → Jn is an
+​, −〉 to the B-​minor triad 11−, we first transpose isomorphism.
the root 11 by the transposition level 5, yielding Schritts, wechsels, and transpositions are
4; then, because the initial triad is minor and the UTTs, while inversions are QTTs, so both the
reflection factor ρ− is −, we replace 4 with its mod-​ Riemann group ℛ and the transposition-​inver-
12 inverse 8; finally, because the sign σU of the sion group 𝒯ℐ12 are subgroups of 𝒬, and both
QTT is also −, we change the triad’s mode from act in simply transitive fashion on triad. The
− to +​. The result is 8+​, the A♭-​major triad. The transpositions in 𝒯ℐ12 commute with every-
double-​circle picture in Figure 8.3.11b illustrates thing in ℛ, because transpositions commute
the action of this QTT; consistent with the calcu- with all UTTs. It is perhaps less obvious that the
lation above, the triad 8+​ appears on the right in inversions in 𝒯ℐ12 also commute with all trans-
the same position as 11− on the left. Comparison formations in ℛ. The following examples illus-
of this figure with Figure 8.1.11, the correspond- trate the commutativity of the inversion J5 with
ing diagram for the UTT 〈−, 2, 5〉, shows that the the schritt S3:
two pictures are identical except that the inner
circle on the right, showing the application of U S J J S
to minor triads, has been reflected about the axis 0 + 
3
→ 3 + 
5
→ 4 −; 0 + 
5
→ 7 − 
3
→ 4−
through 0+​ and 6+​.
S J J S
0 − 
3
→ 9 − 
5
→ 10 +; 0 − 
5
→ 7 + 
3
→ 10 +
Exercise 8.3.12
(a) Calculate the result of applying the QTT
〈+​, 6, 4, −, +​〉 to an F-​major triad and to a In general, if either SmJn or JnSm is applied to the
C♯-​minor triad. major triad (r, +​), the result is the minor triad
(b) What is the inverse of the QTT 〈−, 2, 5, +​, −〉? ((−r − m − n), −), and if either of those compos-
Of 〈+​, 6, 4, −, +​〉? ite transformations is applied to the minor triad
(c) Calculate the QTT products 〈−, 2, 5, +​, −〉 (r, −), the result is the major triad ((m − r − n), +​).
  〈+​, 6, 4, −, +​〉 and 〈+​, 6, 4, −, +​〉〈−, 2, 5, +​, −〉.54 Exercise 8.3.13 below works through similar calcu-
lations for the combination of an inversion and a
Because of the two reflection factors, QTTs wechsel. Elements of 𝒯ℐ12, therefore, always com-
are four times as numerous as UTTs, numbering mute with elements of ℛ, even though both 𝒯ℐ12
288 ∙ 4 =​1152 in all. They form a group 𝒬, of and ℛ are noncommutative groups. Inversions
which 𝒰 is a subgroup. Algebraically, 𝒬 is a wreath generally do not commute with each other or with
product 𝒟12 ≀ 𝒞2. The two dihedral fiber groups transpositions, and wechsels do not commute with
𝒟12 (in place of the cyclic fiber groups 𝒞12 for the each other or with schritts—​but all the transfor-
group 𝒰, from Figure 8.1.10) allow for either uni- mations in one group commute with all the trans-
form or anti-​uniform action on each of the two formations in the other group. As described in
circles of triads, while the control group 𝒞2, as Section 6.4, 𝒯ℐ12 and ℛ are pairwise commuting
before, allows the two circles to be exchanged. groups.

328 • E x p l oring M usica l S p aces

 More can be said: not only does every transfor- This description of two simply transitive trans-
mation in 𝒯ℐ12 commute with every transforma- formation groups that are both isomorphic and
tion in ℛ, but the transformations in ℛ are the pairwise commuting will carry a familiar ring to
only triadic transformations that commute with readers who worked through the discussion of
every transformation in 𝒯ℐ12, and vice versa. In duality in Section 7.5. Theorem 7.5.24 described
terminology from Section 6.6, ℛ is the centralizer a duality between the group of transpositions 𝒯S
of 𝒯ℐ12 and vice versa. To see this, suppose we and the group of interval-​preserving mappings
know that a triadic transformation f commutes 𝒫S in any interval space S: both groups act on S
with every Tn and Jn; we claim that f must be in simply transitive fashion, the two groups are
some schritt or wechsel. The specific Sn or Wn may isomorphic (though it is easier to construct an
be determined from knowing f(0+​). Suppose, for anti-​isomorphism between them than an isomor-
instance, we know that f(0+​) =​3−—​that is f(C) =​ phism), and each group is the centralizer of the
e♭. Using the assumed commutativity and the for- other. The triadic transformation groups 𝒯ℐ12
mulas for the actions of Tn and Jn, the action of f and ℛ provide an example of this duality. In this
on any major triad n+​ is determined by setting, S is the set triad; depending on how we
define “intervals” between triads, either 𝒯ℐ12 or ℛ
can play the role of 𝒯S, and the other then auto-
f ( n+ ) = f (Tn (0+ )) = Tn ( f (0+ )) = Tn (3− ) matically becomes 𝒫S.55
= (n + 3)− , We noted the possibility of regarding triad
as an interval space with 𝒯ℐ12 as its interval
group first in Section 5.6, again in Section 7.2
while the action on any minor triad n− must be in illustrating interval-​space homomorphisms,
and earlier in this section in discussing simply
transitive groups on triad. As Lewin’s theo-
f ( n − ) = f ( J − n (0 + )) = J − n ( f (0 + )) = J − n (3− )
rem describes, 𝒯ℐ12 serves both as the group
= ( n − 3 )+ . of “intervals” and as the associated group 𝒯S of
“transposition” operators on this space. The cau-
The only transformation that maps all triads tionary quotation marks remind us of the non-
as these equations specify is the wechsel W3. traditional meanings of the words: “intervals”
Exercise 8.3.14 below illustrates the similar rea- in this sense are defined by Tn and In relations
soning showing that the only transformations between triads (for example, int(G, a) =​I11),
that commute with all the transformations in ℛ and those same mappings Tn and In are also the
are the transpositions and inversions. “transpositions” Transi in this interval space (G
major “transposed” by I11 produces A minor).
But from Section 7.5 we know that every
Exercise 8.3.13 interval space S also has a group 𝒫S of “inter-
(a) Calculate the result of applying the val”-​preserving mappings, and that 𝒫S is the
composite transformations W3J5 and J5W3 centralizer of 𝒯S. Because the centralizer of
to the C-​major and C-​minor triads 0+​ and the triadic transformation group 𝒯ℐ12 is the
0−, and verify that the two composites give Riemann group ℛ, it follows that ℛ plays the
the same result in each case. role of 𝒫S in the space triad. That is, when
(b) Generally, what is the result of applying we use the transformations Tn and In to define
WmJn or JnWm to a major triad (r, +​)? To a “intervals,” the “interval”-​preserving mappings
minor triad (r, −)? are the Riemannian transformations Sn and Wn.
(c) Only one of the following two equations is The duality tells us, moreover, that we could
true in general: Jn =​ J0Tn or Jn =​ TnJ0. Which have approached the construction the other
  equation is the correct one? way around, starting with the simply transitive
group ℛ, taking schritts and wechsels to define
Exercise 8.3.14 Suppose a triadic transformation “intervals” (for example, int(G, a) =​W2). In this
f commutes with every Sn and Wn and satisfies alternate universe ℛ plays the role of 𝒯S (Sn and
f(0+​) =​7−. Determine the action of f on every Wn are now “transpositions”), while 𝒫S is 𝒯ℐ12
major and minor triad, and show that f can only (Tn and In are no longer “transpositions,” but
be  the inversion J5 (or equivalently I2). instead are “interval”-​preserving).

Triadic Transformations • 329

 The duality between the groups ℛ and 𝒯ℐ12 the transposition group 𝒯12 is the UTT group
may be described in other ways. We noted above 𝒰. The duals of the transpositions Tn are the
that the mapping defined by schritts Sn, elements of the group ℛ+​, so by dual-
ity we may conclude that the transformations
Sn =​ 〈+​, n, −n, +​, +​〉 → Tn =​ 〈+​, n, n, +​, +​〉, that commute with all schritts are exactly the
Wn =​ 〈−, n, −n, +​, +​〉 → Jn =​ 〈−, n, n, −, −〉 dual-​UTTs: 𝒰~ is the centralizer of ℛ+​.

is an isomorphism between the two groups. This Exercise 8.3.15

mapping may be extended to a function defined (a) Verify that the equation (U~)~ =​ U holds if
on the entire group 𝒬, a mapping that inverts U =​ 〈−, 2, 5, +​, −〉.
the second transposition level of any QTT and (b) Verify that the equation (UV)~ =​ U~V~ holds
reverses the two reflection factors if and only if   if U =​ 〈−, 2, 5, +​, −〉 and V =​ 〈+​, 6, 4, −, +​〉.
the QTT is mode-​reversing. That is, the function
f: 𝒬 → 𝒬 defined by Exercise 8.3.16
(a) What is the dual of the M-​chain G–​e–​C–​a–​
F–​d? First determine the QTT M~, where
f ( 〈σ, t +, t −, ρ+, ρ− 〉) = 〈σ, t +, −t −, σρ+, σρ− 〉 M is the mediant transformation. Then,
starting with the G-​major triad 7+​, apply
M~ repeatedly. Is the M~-​chain a sequence
satisfies f(Sn) =​ Tn and f(Wn) =​ Jn. The statements formed by transpositions of two-​chord
f(Tn) =​ Sn and f(Jn) =​ Wn are also true; f is an blocks? Why or why not? Alternate chords
involution. in the M-​chain, such as G–​C–​F, are related
Let us write U~ in place of f(U) for the result by M2 =​ D; are alternate chords in the M~-​
of applying this mapping to the QTT U. The chain related by D~?
mapping ~ is an involution, since (U~)~ =​ U for (b) Now consider G–​e–​C–​a–​F–​d as an RL-​chain
every U. Moreover, ~ is a group homomorphism, rather than an M-​chain, and determine its
which means that the equation (UV)~ =​ U~V~ dual. (That is, start with 7+​ and apply R~
always holds. Clearly ~ is one-​to-​one and onto, and L~ in alternation.) Do you get the same
so ~ is an automorphism of 𝒬. We call U~ the dual   triads as in the M~-​chain? Why or why not?
of the QTT U. The mapping ~ extends the dual
relationship between ℛ and 𝒯ℐ12 to the entire There are several additional ways in which
group 𝒬. Dual QTTs may be said to play identi- the UTT theory may be generalized, a few of
cal roles in the algebraic structure of the group which we sketch briefly in the remainder of
𝒬—​ even though they obviously may differ in this section. We have described UTTs and QTTs
significant ways. as transformations on the set triad, but the
Because ~ is an automorphism of 𝒬, any true ordered pairs (r, σ) in ℤ12 × 𝒮𝑖𝑔𝑛 on which these
algebraic statement about QTTs remains true if transformations act could stand for other kinds
the QTTs are replaced by their duals. For exam- of musical objects. In particular, the 24 sets in
ple, the UTTs form a subgroup 𝒰 of 𝒬, and their any set class of asymmetrical pitch-​class sets
duals also form a subgroup, the group of dual-​ may be labeled with (r, σ) pairs.56 Any such set
UTTs 𝒰~ =​{U~ | U ∈ 𝒰}. From the definition of class, therefore, admits its own groups of UTTs
~, it is clear that the dual of a mode-​preserving and QTTs, algebraically isomorphic to the tri-
UTT 〈+​, t+​, t−, +​, +​〉 is another mode-​preserving adic case, complete with transpositions and
UTT, 〈+​, t+​, −t−, +​, +​〉, but the dual of a mode-​ inversions, schritts and wechsels, M-​chains,
reversing UTT 〈−, t+​, t−, +​, +​〉 is 〈−, t+​, −t−, −, −〉, the ℛ-​𝒯ℐ12 duality, the other simply transitive
which is anti-​uniform and therefore not a UTT at subgroups, and more.57 A number of studies
all. So the dual-​UTT group 𝒰~ consists of QTTs have developed transformational systems on
of these two types: mode-​preserving uniform set classes other than triads, systems that are
transformations and mode-​reversing anti-​uni- easily reinterpreted within the UTT formalism.
form transformations. Equivalently, 𝒰~ consists For example, Childs 1998 and Gollin 1998 inde-
of all QTTs whose three signs σ, ρ+​, and ρ− are pendently proposed dualistic transformational
all the same. theories for the set class of dominant and half-​
Recall that the UTTs are exactly the transfor- diminished seventh chords (prime form 0258),
mations that commute with all transpositions effectively reconstructing on this set class the
Tn; thus in the large group 𝒬, the centralizer of Riemann group ℛ; the systems of Childs and

330 • E x p l oring M usica l S p aces

Gollin could be subsumed in larger groups, r, so the +​/​− distinction cannot indicate chord
counterparts of 𝒰 and 𝒬.58 qualities; in this analysis r+​ denotes a root-​
There are other possibilities. Pairs of the form position triad whose root is generic pitch class
(r, σ) may be associated with ordered sets of pitch r (0 =​C, 1 =​D, …), while r− denotes the corre-
classes, the 24 ordered sets in one equivalence sponding first-​inversion triad. Even sequences
class being related by transposition or inver- with all chords in root position may be modeled
sion.59 Such an ordered set may be a twelve-​tone in this way, using the +​/−​ distinction simply to
row or some other serial structure. So far we indicate a chord’s position as the first or second
have assumed that the +​and − “modes” of the element of a sequential block. Mod-​7 structure
objects (r, σ) are related by inversion, but the will be explored further in Part Four.
UTT formalism does not depend on this assump-
tion. In principle the two “modes” of objects on Exercise 8.3.18 The diatonic sequences below
which generalized UTT-​ type transformations were graphed in a generic tonnetz in Figure
act can be defined by any sort of binary opposi- 4.3.4. Determine a mod-​7 UTT that generates
tion—​they could be major triads and dominant each sequence.
seventh chords, or triads in root position and
in first inversion—​as long as the two modes are (a) G–​C–​F–​bo–​e–​a
distinguishable somehow and objects of both (b)  C–​G–​a–​e–​F–​C
kinds can be transposed to twelve different pitch
levels. If the objects are twelve-​tone rows, we Example 8.3.19: Rimsky-​K orsakov,
may use (r, −) to denote the retrograde, rather Christmas Eve We conclude this section with
than the inversion, of (r, +​). In Section 9.5 we an analysis that engages four “modes” of
will explore serial applications of UTTs further, chords, major and minor triads and dominant
refining our observations from Example 4.5.1 and half-​diminished seventh chords, in a
about Webern’s String Quartet, Op. 28, by apply- comparison of two excerpts from Rimsky-​
ing UTTs to ordered pairs (r, +​) and (r, −) repre- Korsakov’s 1894–​95 opera Christmas Eve.61
senting BACH tetrachords. The opera, a charming blend of comedy and
Yet another constraint subject to loosening fantasy, is an adaptation of a story by Gogol
is the assumption of an underlying mod-​12 in which the village blacksmith Vakula wins
pitch-​class structure. Mod-​n UTTs are mathe- his bride by flying to Petersburg on the
matically just as viable as mod-​12 UTTs, though Devil’s back in order to procure for her the
with some differences in details of algebraic Empress Catherine’s slippers. The Prelude
structure for different values of n.60 In particu- to the opera opens with the series of root-​
lar, mod-​7 UTTs may be used to model some position major triads shown at the top of
diatonic or generic structures. In the same way Figure 8.3.20. What the figure does not
that some chromatic sequences are generated show is the remarkable orchestration: every
by repeated application of a single mod-​ 12 chord is scored for a different combination
UTT, the diatonic sequence in Figure 8.3.17 is of winds. Only rarely does one player play a
generated entirely by the UTT U =​ 〈−, 5, 3〉7 (the note in two consecutive chords; the concept
subscript signaling a mod-​7 UTT). The trans- of voice leading is foreign to this music. The
position levels in this UTT indicate generic chords are projected as a series of isolated,
transposition operators t5 and t3, as in analyses frozen jewels, an evocation of the moonlit,
of similar sequences in Section 4.3. In generic snow-​covered landscape on which the curtain
space there is only a single triad on any root is about to rise.

FIGURE 8.3.17 A diatonic sequence generated by the mod-​7 UTT U =​ 〈−, 5, 3〉7

Triadic Transformations • 331

FIGURE 8.3.20 Transformational relationships in Rimsky-​Korsakov, Christmas Eve: Act I, Prelude
(top); Act III, Procession of Comets (bottom)

Our analysis regards this succession of the third relations are now chromatic, with
triads as a chromatic alteration of a now-​ prominent cross relations. The major triads
familiar pattern, the RL-​chain E–​c♯–​A–​f♯–​ ⋯, on C♯, F♯, and B are analyzed as parallel (P)
which appears across the top row of the substitutions for the minor triads in the RL-​
transformation network in the upper of the chain, or simply as T9 transpositions of the
two bubbles in the figure. Rimsky-​Korsakov’s preceding chords.
music follows not this row of transformations The scene of Vakula’s flight is divided into
but the zigzag path of solid arrows, in which several fantastic episodes, among them the
all chords are major. The chord roots descend Procession of Comets, which opens as shown
alternately by minor and major thirds as in the at the bottom of the figure. Apart from an
RL-​chain, and alternate chords are dominant-​ identical tempo indication, this excerpt bears
related as shown by the curved arrows, but little outward resemblance to the opening of the
the exclusive use of major triads means that Prelude. A quartet of horns, doubled by other

332 • E x p l oring M usica l S p aces

winds, plays dominant seventh chords, cycling seventh chords—​cross-​type mappings, as we
systematically through all four inversions used the term in Chapter 7. A particularly
with efficient voice leading. The instability of straightforward mapping of this kind is the
these sonorities is intensified by violin trills, transformation X (for exchange) that maps
doubled in woodwinds, with highlights in every triad to the seventh chord represented
harp and celesta. The pattern continues nine by the same (r, σ) pair and vice versa; thus X
measures beyond the seven in the figure, the maps the B-​major triad to the B7 chord and
trills ascending inexorably into the realm of the vice versa, and the G♯-​minor triad to the g♯ø7
comets. The musical depiction of flight is as vivid chord and vice versa. The bubble notation in
as was the crystalline stillness of the Prelude. the figure shows that the upper bubble is DX
It is remarkable, then, to find exactly the same of the lower: starting from any seventh chord
root-​interval relationships in both excerpts. in the lower bubble, if we apply the dominant
While the transformations in the upper bubble transformation D (that is, transpose by T5)
are the usual triadic transformations, those in and exchange the seventh chord for a triad,
the lower bubble are transformations of the we get the chord at the same location in the
seventh chords in the Procession of Comets. upper bubble. Indeed, every dominant seventh
But within each bubble all transformations chord in the Procession of Comets is V7 of the
belong to the group 𝒰, whose action on the set corresponding triad in the Prelude.
sev of dominant and half-​diminished seventh Evidently X is an involution; it generates a
chords, as described above, corresponds exactly group 𝒳 =​{E, X} of order 2, isomorphic to 𝒞2.
to its action on triads, at least in terms of root We could write XD in place of DX, because X
relationships. For example, R (or W9) relates the and D commute; in fact X commutes with all
ordered pairs (4, +​) and (1, −); in triad this is UTTs. By principles outlined in Section 6.4, the
a relationship between E-​major and C♯-​minor UTTs in 𝒰, together with X, generate a group of
triads, while in sev it is a relationship between order 288 ∙ 2 =​576 isomorphic to 𝒰 × 𝒳, which
the E7 and c♯ø7 chords. now acts not just on triad but on the larger set
As we have noted in other contexts, an triad ∪ sev, a space of cardinality 48 containing
algebraic isomorphism between two musical triads and seventh chords, two qualities of each.
passages does not imply that they share other, Half of the transformations in this large group
non-​algebraic properties. Voice-​leading behavior are the UTTs U; the others are of the form UX
is not preserved by this isomorphism between (=​ XU). All of these transformations are defined
triad and sev: R-​related triads are related by a on all 48 chords: UTTs map triads to triads and
two-​semitone displacement in one voice, while seventh chords to seventh chords, while the
R-​related seventh chords require two voices to transformations UX exchange the two. (When
move by semitone. Triads related by R belong the set triad ∪ sev is regarded in this way as
to a common diatonic scale; seventh chords the space of transformational action, X need no
related by R do not. These differences are even longer be considered a cross-​type mapping at
more pronounced for the relationships among all; like the UTTs, it maps to and from the same
the same-​mode chords that actually appear in space, triad ∪ sev.)62
the excerpts: Exercise 8.3.21 below points out a However one chooses to describe the
difference between the voice-​leading behaviors transformational patterns exhibited by these
of the transpositions T8 and T9 on triads and two excerpts, Rimsky-​Korsakov seems to
the behaviors of those same transformations have been enamored with them. The Prelude’s
on seventh chords, a difference for which we opening chords recur several times during
will see a geometric explanation in Section 11.5. the opera. Once the seventh-​chord version is
For reasons like these, readers may justifiably introduced, it becomes a motto for Vakula’s
be less inclined to hear the seventh-​chord flight, and forms the basis for the khorovod
progression as an altered RL-​chain than to hear that follows the Procession of Comets. The
the triadic progression that way. csárdás that comes next opens with a newly
We could regard the two bubbles in Figure animated statement of the triadic progression
8.3.20 as separate transformation networks; in which the subtle rhythmic acceleration of
the two networks are then isographic in the Prelude chords, all but imperceptible at
terminology to be made precise in Section first, gains new urgency. The seventh-​chord
9.3. It is possible, however, to define variant immediately follows, then a hybrid
transformations directly relating triads and version in which the top voice follows the

Triadic Transformations • 333

ascending pattern of the Prelude chords but 6. Hint for Exercise 8.1.5b: In at least one of these
the harmonies are seventh chords in third cases, the product is unchanged when the order of the
inversion.
   63 two UTTs is reversed, demonstrating that some pairs
of UTTs may commute even if no transpositions are
Exercise 8.3.21 involved.
(a) Determine the smallest possible total 7. Generally, if G and H are groups of orders m and
number of semitones of voice-​leading n, acting on sets of cardinalities j and k, respectively,
displacement needed to connect two major we may imagine k drawers with a copy of a set of car-
triads related by T9. Assume that the chords dinality j in each drawer. For the UTT group, m =​ j =​
are written in three voices and may be in 12 and n =​ k =​2. The wreath product G ≀ H is a group
any inversion. Do the same for T8. Which of order mk ∙ n acting on a set of cardinality jk (all ele-
of the two transpositions is potentially ments in all drawers), involving k copies of the fiber
smoother? group G (one acting on each drawer) interrelated by
(b) Answer the same questions as in (a) for the control group H (which controls how the drawers
the transpositions T9 and T8 applied to may be exchanged). A wreath product is a special case
  dominant seventh chords. of a more general construction known as a semidirect
product; see Humphreys 1996, 163–​73. Wreath prod-
ucts play a central role in a sweepingly general theory
NOTES of abstract forms and the storage of information in
Leyton 2001, including a description of sequential
1. In order to show that a transformation com-
structure in music (477–​93).
mutes with all transpositions, it suffices to show that
8. Riemann’s terms schritt and wechsel have been
it commutes with T1 (or with T5, T7, or T11—​any one
widely adopted, but the terminological variants are
of the generators of the transformation group 𝒯12). If,
many. See Table 1 in Hook 2002, 80–​81, for a com-
for example, U commutes with T1, we can write UT2 =​
parison of names by which schritts and wechsels have
UT1T1 and exchange U with the T1s one by one to show
been known. The notations Sn and Wn appear in sev-
that U commutes with T2.
eral sources, including Hook 2002 and Clough 2008.
2. There is no particular theoretical reason for the
This convention conflicts with an earlier use of the
convention of writing UTTs and some other related
same notations in Gollin 1998, based not on intervals
transformations in angle brackets, but the visual con-
between the traditional roots of triads but on inter-
trast is helpful in distinguishing UTTs from other
vals between Riemannian dual roots (the dual root
tuples, written in parentheses, that may appear nearby
of a minor triad being its fifth). Gollin’s schritt labels
in various contexts. In Chapters 5 and 6 we used angle
match the notation here, but the wechsel we call Wn is
brackets to denote groups generated by certain ele-
Gollin’s Wn+7.
ments, an unrelated notation that is not employed in
9. Hints for Exercise 8.2.5: You can use Figure
this chapter.
8.2.3b, or your redrawn figure from Exercise 8.2.4,
3. Riemann introduced terms and notations for
to express any element of ℛ in terms of LR and R,
triadic relationships akin to our P, R, and L, though
and thereby in terms of L and R. Because L and R are
his conceptions of these relations were not explic-
involutions, the only way to generate other transfor-
itly “transformational” in the modern sense. For
mations from them is to use the two in alternation,
Riemann and harmonic dualism more generally, see
LRLRL⋯ or RLRLR⋯. Both chains produce all 24
Harrison 1994.
Riemannian transformations. If you start with L and
4. A transposition operator such as T3 may seem
find that more than 12 transformations are required,
to be an exception to the generalization that triadic
you can obtain a shorter expression by starting with R.
transformations do not directly imply voice lead-
See Table 1 in Hook 2002.
ing, since the action of T3 on a triad is determined
10. Hint for Exercise 8.2.6a: As defined in Section
by the action of T3 on the pitch classes of that triad:
6.6, f is an anti-​isomorphism if it reverses, rather than
“T3 of pitch class 0” makes sense in a way that “L
preserves, the order of products—​that is, if the equa-
of pitch class 0” does not. It nevertheless remains
tion f(U)f(V) =​ f(VU) holds for all U and V in ℛ. The fol-
true that once we admit T3 as a triadic transforma-
lowing chain of equalities demonstrates this property
tion, that object called T3—​a function from triad to
when U is a schritt and V is a wechsel:
triad—​implies nothing about the mapping of indi-
vidual notes. f(Sm)f(Wn) =​ TmIn =​ In−m =​ f(Wn−m) =​ f(WnSm).
5. Hint for Exercise 8.1.4b: The four calculations do (To see that Wn−m is the same as WnSm requires
not have four different answers. exchanging the roles of m and n in the equation

334 • E x p l oring M usica l S p aces

WmSn =​ Wm − n from Theorem 8.2.2a.) The other cases of cardinality 1) form a trivial example, the cycle C –​
can be verified similarly. Anti-​isomorphic groups are C♯ –​D –​⋯. A more interesting example is obtained
always isomorphic, so this anti-​isomorphism is con- by arranging the twelve diatonic collections (sets of
sistent with our observation that ℛ and 𝒯ℐ12 are cardinality 7) in circle-​of-​fifths order. The only other
both isomorphic to the dihedral group 𝒟12. examples are formed by the complements of the sets
11. Hints for Exercise 8.2.8: Two of the five pairs just mentioned: sets of cardinality 11, pentatonic col-
generate ℛ. As with transpositions, any one of the lections, and the nine-​note complements of conso-
four schritts S1, S5, S7, or S11 by itself generates the nant triads.
cyclic group ℛ+​, so if you can produce any one of these 18. The use of the label H for hexatonic poles is
four, you can generate all twelve schritts. If you can attributed to Robert Cook. Some authors have used
produce all the schritts and any one wechsel, then you the abbreviated name hexpole.
can generate ℛ. 19. For analyses of the Grail theme from Parsifal
12. Hint for Exercise 8.2.10a: By Theorem 8.2.2d, tt: (8.2.14b) see Lewin [1984] 2006; Cohn 1996;
𝒰 → ℤ12 is a group homomorphism. What is tt(V−1UV) Clampitt 1998; and particularly Rings 2011b, 83–​88,
in terms of tt(U) and tt(V)? wherein Rings’s interval space sd × pc is invoked to
13. See Engebretsen 2008 for a study of implicit reinterpret the chromatic Grail theme as a distortion
group structure in the work of Riemann and his pre- of the diatonic form heard previously. Cohn has stud-
decessors. The root-​ interval approach, predating ied hexatonic cycles and hexatonic poles extensively;
Riemann’s more celebrated formalization of the con- see Cohn 1996, Cohn 2004, and Chapter 2 of Cohn
cept of harmonic function, may be traced through 2012. Cohn’s hexatonic pole examples date back to
Riemann 1880, Oettingen 1866, and Hauptmann Gesualdo, Monteverdi, and Bach, but most are by the
1853. Following and expanding upon Oettingen, likes of Wagner, Franck, Grieg, Mahler, Strauss, and
Riemann employs such terms as Quintschritt and Puccini. For examples of a major hexatonic pole used
Kleinterzwechsel (“fifth-​ schritt” and “minor-​ third–​ as a chromatic neighbor chord to a minor tonic rather
wechsel”), named to convey (dualistic) root rela- than the other way around, see the opening sections of
tions. Klumpenhouwer 1994 identifies Riemann’s both the first and third movements of the Piano Trio
system as a group; the dihedral structure, implicit in in G Minor, Op. 3, by Ernest Chausson. The opening of
Klumpenhouwer’s formulation, is noted specifically in Vaughan Williams’s Sea Symphony owes its jaw-​drop-
Gollin 1998, 203. ping effect to hexatonic poles; more recent examples
14. In this context the reader may recall the ton- may be found in Shostakovich and Adams.
netz of 014 trichords studied in Exercise 4.4.3. The 20. See Morris 1998, 184–​88. Morris calls Pʹ, Lʹ,
transformations P*, L*, and R* defined there gener- and Rʹ the obverses of P, L, and R. Importantly, they
ate a group isomorphic to the Riemann group, even are not the inverses of P, L, and R: all six of these trans-
though they differ markedly from P, L, and R in their formations are involutions, so each is its own inverse.
voice-​leading behavior. 21. We avoid the S notation here as it risks confu-
15. A wechsel does not intrinsically privilege one sion with Sn (Pʹ is a wechsel, not a schritt); the letter
of the two triads over the other; depending on the S has also been used elsewhere for the subdominant
context, either of the two may be the more stable or transformation discussed in the next section. In this
the more structurally salient. For commentary on this book we have also used S as a label for a general set or
notion of reciprocity see Chapter 3 of Cohn 2012. interval space, a notation that we avoid in any context
16. Wechsels enjoy other advantages over inver- involving transformations labeled S.
sions as well. As noted in Section 7.6, the numbering 22. See Cohn 2012, 47 and 61–​62, and Cohn 2000,
of inversions is arbitrary in the sense that it depends 92. For Weitzmann, as for Hauptmann before him,
on the choice of C as pitch class 0; the numbering C-​major and F-​minor triads were neighbors along a
of wechsels does not depend on this choice (though line of thirds, F–​A♭–​C–​E–​G. Weitzmann conceived the
the numbering of both schritts and wechsels would nebenverwandt relation as the inversion of either triad
be reversed if we identified minor triads as the about the central note of that series—​the root of the
+​mode). Also, a well-​trained listener without abso- major triad or the fifth of the minor triad.
lute pitch can identify a wechsel Wn by ear, but not 23. Cohn (2012, 96–​97) analyzes an RL-​chain in
an inversion In. Brahms’s Deutsches Requiem in which Lʹ substitutes
17. Cohn (1996, 16) observes that the triadic PL-​ not for an L but for an R. Chopin’s sequence ascends
cycle is one of only a few possible examples of maxi- by thirds while Brahms’s descends, but in both cases Lʹ
mally smooth cycles consisting of more than two sets is applied to a major triad, effectively inserting a V–​i
from the same set class. The twelve pitch classes (sets resolution into the middle of a sequence.

Triadic Transformations • 335

 24. Patterns formed by a pair of wechsels in alterna- While the three sequence types that Martin uses
tion are called uniform flip-​flop circles in Clough 2008. are seemingly foreign to common-​ practice tonal-
25. See Chapter 4 of Cohn 2012. The name is in ity, examples of type (d) occur in a number of well-​
homage to Carl Friedrich Weitzmann’s 1853 trea- known works, from Bach (B–​g–​A–​f–​G in m. 8 of the
tise on the augmented triad, in which voice leading A-​Minor Prelude from WTC II) to Schubert (g–​A–​f–​G
between augmented and consonant triads is a primary in mm. 12–​15 of “Morgengruss,” No. 8 of Die schöne
concern. Müllerin) to Debussy (E–​c–​D–​b♭–​C in the distant,
26. The Cube Dance graph was devised by Douthett evocative horns at m. 107 in the coda of L’après-​midi
in 1992 and first published in Douthett and Steinbach d’un faune).
1998, 254. A version appears also in Cohn 2012, 86. 36. The Liszt example appears in Bass 1996, 271–​72,
As noted in Cohn 2000, many features of this graph the Rimsky-​Korsakov in Taruskin 1985, 97.
are alluded to in Weitzmann 1853, though Weitzmann 37. See Lewin [1987] 2007, 175–​80. At first Lewin
does not present the complete figure. used the transformational labels PAR, LT, REL, DOM,
27. A chord progression containing augmented tri- and MED, but within a few years these were largely
ads can be traced through the graph of Figure 8.2.19, supplanted by the single-​ letter labels used here
but such a path is not a transformational analysis. (except M). The notation D brings with it a need to
Cohn 2012 includes some graphs of this kind. More exercise care in distinguishing italic D, the dominant
often, Cohn graphs progressions in the triangulated transformation, from roman D, the D-​major triad.
tonnetz, in which consonant triads are triangles and 38. Lewin ([1987] 2007, 177) devotes considerable
the occasional augmented or diminished triad may be effort to explaining why the dominant transforma-
depicted as a straight line segment. Cohn’s analysis of tion is defined so that dominants point to their tonics
Schubert’s “Auf dem Flusse” (2012, 122–​25), which rather than the other way around, going so far as to
floats among hexatonic triad and bridge families, is criticize Riemann’s function theory for not conceiv-
germane to the discussion here; readers may find it ing of dominant-​tonic relations in this way. Formally,
instructive to map the “voice-​leading zones” in Cohn’s either way is as good as the other—​in a group, if we
Figure 6.8 onto our Figure 8.2.17. have D, we also have D−1—​but the additional layer of
28. Cohn 2012 gives many examples of RLʹ-​chains interpretation is evidently of considerable importance
(“N/​R chains” in his terminology), by composers to Lewin.
including Haydn, Beethoven, Schubert, and Liszt. 39. Some authors have used the label M for triadic
29. Baur (1999, 568–​71) studies mm. 60–​94 in the transformations other than 〈−, 9, 8〉. In Kopp 2002,
context of a wide-​ranging study of octatonicism in “M” is the transposition T8; in Segall 2017, it is the
Ravel’s music. schritt S3.
30. Such a preference is expressed, for example, in 40. If the M-​chain is tuned in pure major and minor
Klumpenhouwer 1994, and as we will note in Chapter thirds, the roots of triads seven places apart differ by a
9, Lewin requires path consistency of transformation syntonic comma; recall Exercise 1.4.16.
networks. 41. It would be an overstatement to claim that the
31. The two Schubert examples and the Glazunov choice of a simply transitive group makes all analyti-
appear in Taruskin 1985, 84–​86 and 130–​31. For the cal decisions inevitable. As we know, transformations
Schubert Mass see also Cohn 2012, 30–​32 and the need not always model chord-​to-​chord relationships
rather different discussion in Salzer and Schachter in strictly chronological fashion; the determination
1969, 215–​18. The Liszt example is cited in Douthett that we wish to model the chord succession G–​a with a
2008, 102, and in Cohn 2012, 91–​92. transformation is an analytical decision that no group
32. This Schubert example is studied in Cohn can make for us.
1998b, 290–​91. 42. See Hook 2002, 84–​93, for details of the simply
33. The Bach sequence in (a) is studied in Harrison transitive subgroups of 𝒰, and see Hook and Douthett
1994, 33–​34. 2008 for additional structural properties and applica-
34. Hint for Exercise 8.2.26h: How many different tions of these groups. The 24 simply transitive groups
triads would appear in any of these sequences before are labeled 𝒦a,b, where a and b are mod-​12 integers;
the pattern repeats? a must be one of the mod-​12 units (1, 5, 7, or 11),
35. This family of sequence patterns would theo- and b must be such that ab ≡ b (mod 12). Each group
retically be completed by a fourth type, in which the 𝒦a,b consists of twelve mode-​preserving UTTs of the
major chord of each pair comes first and successive form 〈+​, n, an〉 and twelve mode-​reversing UTTs of
pairs descend by T10. This hypothetical sequence—​ the form 〈−, n, an +​ b〉. The Riemann group ℛ is 𝒦11,0,
call it (d)—​ would be a W8W10 sequence like (a), while ℳ is 𝒦1,11. The group generated by T1 and P, dis-
but with W8 applied to major triads, W10 to minor. cussed below, is 𝒦1,0. Only the twelve groups 𝒦1,b are

336 • E x p l oring M usica l S p aces

commutative, and of those, only the six groups with I0-​inversion of (0, +​) with the ordered pair (0, −), and
odd values of b are cyclic. the transpositions of that set with (n, −). While this
43. Hint for Exercise 8.3.5: The generators are the strategy can be applied in general, it is not the only
units mod 24 (the numbers coprime to 24). possibility, and in fact it does not match the stan-
44. Hint for Exercise 8.3.6: Remember this charac- dard assignment of (r, σ) labels to triads. If the sets
terization of normal subgroups from Section 6.6: H is are labeled differently, some transformational labels
a normal subgroup of G if and only if H contains the change, but the group structure is unaffected.
conjugates of all of its elements by all elements of G. 57. The first T in the abbreviation UTT is obsolete
45. Hint for Exercise 8.3.7: The four patterns are if the transformed objects are not triads, but we retain
generated by four different UTTs, but two of them the name, as triads remain the prototypical example
generate the same group. for these transformations. Similarly, the r in a pair
46. Hint for Exercise 8.3.8a: Because transpositions (r, σ) may no longer be the “root” of the object in ques-
commute with all UTTs, all that must be shown is that tion in any traditional sense.
two different transformations of the form TnP (say 58. Gollin points out the isomorphism between
TmP and TnP) commute with each other. his group and the triadic schritt-​wechsel group, and
47. Hook 2002, 100–​04, describes some subgroups notes that the same group structure is applicable to
of 𝒰 not considered here. other set classes. For another example, Lewin [1993]
48. Hint for Exercise 8.3.9: Write the UTT notations 2007, 16–​67, develops a transformational system for
for the eight transformations in the octatonic triad pentachords of prime form 01236 in Stockhausen’s
group. Note the difference between the two transposi- Klavierstück III, paying particular notice to the contex-
tion levels of each of them, and consider whether this tual inversion that preserves the chromatic tetrachord
difference is preserved when the UTT is composed within a pentachord of this type. By combining this
with a transposition. wechsel with transpositions, Lewin essentially works
49. See Hyer 1995, 120, for the general form of not in the group ℛ but in one of the other simply
his multiplication table. Table 1 in Hook 2002, 80–​81, transitive subgroups of 𝒰. See Hook 2002, 111–​13,
may be used to convert Hyer’s canonical forms to UTT for more details about the UTT reformulations of the
notation. systems of Childs, Gollin, and Lewin described here.
50. This conclusion about T3 may seem to con- 59. In some cases the theory may be applicable
tradict Exercise 6.1.8a, which determined that T3 is to a class of ordered sets even if, because of sym-
an odd permutation—​but in that context “T3” was a metry, it is not applicable to the associated class of
permutation of the 12-​element set pc rather than of unordered sets; for instance, there are 24 distinct
the 24-​element set triad, so there were only three 4-​ TI-​relatives of the ordered pitch-​class set (0, 3, 6, 9)
cycles rather than six. but not of the unordered set {0, 3, 6, 9}. Any ordered
51. See Hook 2002, 97. The equivalence can be set containing two distinct pitch classes forming an
proved by exhibiting a generating set of transfor- interval class other than ic6 will have 24 different
mations for the group 𝒰 and verifying that the two forms. Fiore and Satyendra 2005 note that under
definitions agree for each of the generators and there- this assumption, a transformational system isomor-
fore for any product thereof. One such generating phic to triadic transformations can be constructed;
set consists of the three transformations 〈+​, 0, 1〉, while the action of transpositions and inversions on
〈+​, 1, 0〉, and P. ordered sets is straightforward, Fiore and Satyendra
52. For a summary of Kopp’s basic transformations show that schritts and wechsels are also well-​defined
see Kopp 2002, 146. Some of Kopp’s notations conflict in this setting and the ℛ-𝒯ℐ12 duality applies in
with traditional transformational terminology. full. Lewin 2008 examines a class of pentachords in
53. The idea of extending UTTs in this way was Schoenberg’s Op. 23, No. 3 as both unordered and
suggested by John Clough. See Hook 2002, 106–​11, ordered sets, defining contextual inversions appli-
for a more thorough formalization of QTTs. cable in both senses.
54. Hint for Exercise 8.3.12: You may find it 60. For a summary of some of these differences see
helpful to make double-​circle diagrams similar to Hook 2002, 114–​15.
Figure 8.3.11. 61. The Rimsky-​Korsakov analysis is adapted from
55. This dual relationship between 𝒯ℐ12 and ℛ is Hook 2007a, 32–​35.
noted in Clough 1998. 62. The direct product 𝒰 × 𝒳 is the simplest group
56. To be explicit, we may associate the set whose that accommodates UTTs on both triads and seventh
pitch-​class numbers match the prime form of the set chords and provides a way to relate the two, but other
class with the ordered pair (0, +​) and the transposi- strategies, involving higher-​order wreath products, are
tions of that set with (n, +​); then we may associate the also possible. We could take two copies of the group 𝒰,

Triadic Transformations • 337

one acting on triads and the other on seventh chords, among the examples in Rimsky-​ Korsakov’s post-
as the fiber groups in a wreath product 𝒰 ≀ 𝒞2 (or humously published treatise on orchestration, but
(𝒞12 ≀ 𝒞2) ≀ 𝒞2). A transformation in this group is of they are widely separated in the book and there is no
the form 〈σ, Utriad, Usev〉, consisting of two indepen- discussion of a connection between them or of the
dent UTTs for triads and seventh chords along with a harmonic progressions involved (Rimsky-​ Korsakov
sign that allows the two to interact. Alternatively, we [1913] 1964, Examples 106 and 143).
could take four copies of the transposition group 𝒯12,
one for each of the four qualities of chords in triad
∪ sev, and combine them in a more complex wreath
SUGGESTED READING
product 𝒯12 ≀ 𝒮4 (algebraically, 𝒞12 ≀ 𝒮4), whose control Lewin turns his attention to triadic transforma-
group is not 𝒞2 but the symmetric group 𝒮4. A UTT-​ tions for a few pages at the beginning of Chapter
style transformation in this group incorporates four 8 of GMIT (Lewin [1987] 2007, 175–​80)—​a pas-
transposition levels, and, in place of a sign, a permuta- sage that may be seen in hindsight as the de facto
tion in 𝒮4, enabling the four modes to be permuted in birth of neo-​Riemannian theory, among the most
any desired way: a transformation might, for example, influential pages in Lewin’s writings. Many scholars
employ the permutation (in cycle notation) (M, m, contributed to the subsequent development of neo-​
ø7)(Mm7), which permutes three of the chord quali- Riemannian theory, none more influentially than
ties cyclically while leaving dominant seventh chords Richard Cohn. Many examples and insights in Cohn
unchanged. Both of these proposed groups are much 1996 and Cohn 1997 remain valuable today; Cohn’s
larger than 𝒰 × 𝒳—​the order of 𝒰 ≀ 𝒞2 is 2882 ∙ 2 =​ work culminates in his book Audacious Euphony
165,888, while that of 𝒯12 ≀ 𝒮4 is 124 ∙ 4! =​ (2012), whose compelling tonnetz-​based analyses
497,664—​and musically unwieldy; in many musical engage ideas in Section 8.2 of this book. Analytical
situations it is far from clear how one should deter- applications in the early years of neo-​Riemannian
mine which of the many possible transformations in theory concentrated on late nineteenth-​century
one of these two groups is most suitable. examples, but triadic transformations have also
63. In Rimsky-​Korsakov’s harmony text, written been applied to a diverse assortment of more recent
almost a decade before Christmas Eve, direct juxtapo- music, including pop-​rock examples (Capuzzo
sition of third-​related triads of the same mode falls 2004) and works by Gavin Bryars (Cook 2009),
into the category called “false progressions” (Rimsky-​ John Adams (Johnson 2011), and Alfred Schnittke
Korsakov [1886] 1930, 108–​11). Writing late in his (Segall 2017).
life about Christmas Eve, he regretted that his free
incorporation of mythological elements into Gogol’s The discussion of UTTs in this chapter incorporates
tale had been a mistake—​“but a mistake which offered material first published in Hook 2002, but many
the opportunity of writing a wealth of interesting technical details from that article are omitted here.
music” (Rimsky-​ Korsakov [1909] 1942, 343–​ 44). A concise introduction to UTTs may also be found
Both the Prelude and the Procession of Comets appear in Hook and Douthett 2008, 92–​98.

338 • E x p l oring M usica l S p aces

 9
Transformations II
Transformation Graphs and Networks; Serial Transformations

TRANSFORMATION GRAPHS and networks musical insights that might escape notice in a
were introduced in Section 3.5 and have made conventional twelve-​ tone analysis. UTTs may
appearances in every subsequent chapter of this also play the role of coordinate transformations
book. This chapter examines some formal and relating different labeling schemes for the forms
theoretical considerations involving graphs and of a twelve-​tone row. Section 9.6 describes the
networks that we have not yet considered. Some duality between pitch-​class numbers and order
of this material originates in the work of David numbers.
Lewin, but the formalism presented here departs
from Lewin in several ways, notably by allowing 9.1 TRANSFORMATION
cross-​type networks and relaxing Lewin’s path GRAPHS AND NETWORKS:
consistency condition (discussed in detail in
Section 9.2). Section 9.3 describes several kinds
BASIC PROPERTIES
of isomorphism and isography of graphs and We begin by using a series of small transforma-
networks, and Section 9.4 is a brief introduction tion graphs and networks to review basic defini-
to Klumpenhouwer networks. tions and to define some new properties. These
The final two sections of the chapter exam- graphs are designed less to model specific musi-
ine transformational relationships among cal passages than to illustrate the aspects of net-
serial structures, particularly twelve-​tone rows, work structure that are now our primary focus.
in greater depth than we have to this point. As defined in Section 3.5, a transformation
In Section 9.5 we show how UTTs and simi- network is a directed graph endowed with two
lar transformations may be applied to rows; sets of labels: the vertices are labeled with the
examples from the serial music of Webern names of elements in some space S while the
demonstrate that such an analysis may yield arrows are labeled with transformations on S.

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0009
The analyst constructing a transformation net- space is not possible for this network, because S
work is bound by few constraints. There are no is not closed under some of the transformations
structural constraints on the space: S is often appearing here (R may be considered a function
an interval space, for example, but this is not a from S to S, but D and L cannot). In fact, R and
requirement. No rules specify which elements of L together generate a cycle through all 24 triads,
the space must appear in the network, how often so the smallest space possible for this network is
any element may appear, which pairs of verti- the complete space triad. This figure, like many
ces must be joined by arrows, or the direction in this chapter, includes a legend identifying the
in which a particular arrow should point. The space associated with the network. (While the
transformations appearing in a network must network label N1 in Figure 9.1.1a is merely for
be functions defined on S, which implies that ease of reference, the space label triad may tech-
S must be closed under each of them—​but the nically be considered part of the network.) The
transformations are not required to belong to a precise extent of the space on which a network is
group, so they need not be one-​to-​one or onto.1 defined is not always a significant consideration,
Formally, a transformation network may be but careful attention to spaces is needed in the
defined by enumerating four things: a directed case of cross-​type networks, discussed below. The
graph, a space S, and two label functions, of transformations in the figure are all functions f:
which one assigns an element of S to every ver- triad → triad, as required; in fact all are UTTs,
tex in the graph and the other assigns a function elements of the group 𝒰 from Chapter 8. Label
f: S → S to every arrow.2 The only other general consistency is easily checked; for example, the
constraint on transformation networks is label network shows that the F-​minor triad is simul-
consistency: arrow labels must be consistent with taneously D of C minor and R of A♭ major, both
vertex labels in the sense that an arrow labeled true statements.
f from element x to element y is permitted in a Part (b) of the figure shows the underlying
network only if the implied equation f(x) =​ y is transformation graph of the network N1, labeled
true. (The path consistency condition, imposed by gr(N1). The configuration of vertices and arrows
Lewin as an additional requirement, will be dis- in the graph gr(N1) matches that in the network
cussed in Section 9.2.) N1, and the arrows are labeled with the same
Figure 9.1.1a presents a simple transforma- transformations, but the vertices in a transfor-
tion network called N1. The elements are major mation graph are unlabeled. Even though no ele-
and minor triads; the transformations R, L, and D ments of the space S appear in a transformation
are familiar from Chapter 8. While in principle it graph, to make sense of the arrow labels we must
is possible that the space of some transformation assume that a space has been specified as a part of
network might consist only of S =​{E♭, c, A♭, f}, the the definition of the graph; Figure 9.1.1b makes
four triads appearing in the figure, that minimal the space triad explicit via the same legend as
in (a).3 A transformation graph thus consists of
all the same components as a transformation
network except for the vertex labeling: a directed
graph, a space S, and a label function that assigns
to each arrow of the graph a function f: S → S.
Because there are no vertex labels, label consis-
tency is not a consideration for transformation
graphs.

FIGURE 9.1.1 (a) A triadic transformation

network N1; (b) the underlying transformation FIGURE 9.1.2 A triadic transformation net-
graph gr(N1) work N2

340 • E x ploring M usi c al S pa c es

 Every transformation network N has an
under­lying transformation graph gr(N), obtained
by removing its vertex labels. In comparison with
transformation networks, transformation graphs
are less explicit and more abstract. Their use-
fulness lies in the fact that one graph can con-
vey information common to several networks.
Figure 9.1.2 shows a network N2 that is identical
to N1 except for its vertex labels. The two net-
works, that is, share the same underlying trans-
formation graph: gr(N1) =​gr(N2). This property,
called strong isography, is one of several kinds of
isomorphism and isography of transformation
graphs and networks to be studied in Section
9.3. A network is said to realize its underlying
graph; thus the networks N1 and N2 both realize
the graph in Figure 9.1.1b.

Exercise 9.1.3 How many different realizations

are  possible for the graph in Figure 9.1.1b?4 FIGURE 9.1.5 (a) An event network; (b) a spa-
tial network
The network in Figure 9.1.4a captures a
relationship between the first two triads from
networks N1 and N2, employing two shorthand R. The bubble notation in the figure indicates
notations we have used previously. When an that the transposition T9 relates pairs of corre-
involution such as P, L, R, or T6 appears among sponding vertices in both bubbles; because there
the transformations in a graph or network, it may are two vertices in each bubble, the T9 arrow
be indicated by means of a bidirectional arrow—​ stands for two separate vertex-​to-​vertex arrows,
or by a more efficient shorthand that omits E♭ → C and c → a. By means of these abbrevia-
both arrowheads, leaving a simple line segment tions Figure 9.1.4a, though it displays only a
that we understand to represent two oppositely single arrowhead, represents a network with six
directed arrows bearing the same label. The two arrows, shown completely in 9.1.4b. The appear-
segments labeled R in 9.1.4a therefore stand for ance of the transformation T9 in this figure con-
four separate applications of the transformation veys something about the similarity between the
networks N1 and N2 and the role of transposition
in that similarity, considerations to be examined
in the discussion of isomorphism and isography
in Section 9.3.
As with other graphs, the elements of a trans-
formation network may be arranged in many
different ways on the page, though there may
be musical or aesthetic reasons for preferring a
certain layout. Figure 9.1.5, based on a Schubert
progression from Exercise 8.2.25, revisits the
distinction between spatial networks and event
networks first introduced in Section 3.5. Figure
9.1.5a is an event network of an unusually
straightforward type, simply accounting for the
triads in the order in which they appear and
identifying transformations between succes-
sive chords. (For the transformations Pʹ and Lʹ
see Section 8.2.) Part (b) of the figure displays
FIGURE 9.1.4 (a) A transformation network the same triads and transformations in a spa-
incorporating shorthand notations; (b) the full tial network, organized like a partial map of
network implied in (a) neo-​Riemannian triad space. (The space label

Graphs and Networks; Serial Transformations • 341

nrtriad implies this spatial organization, as in
Figure 4.2.5b; as a set, nrtriad is identical to
triad.) The spatial network reveals an organiza-
tional logic not apparent in the event network,
possibly at the cost of obscuring the chronol-
ogy of the passage. F-​minor and E-​major triads
each occur twice in the progression and in Figure
9.1.5a; in (b), F minor appears twice but E major
only once. Duplicate vertex labels in a network
indicate that two or more vertices represent the
same element of the space, something encoun-
tered commonly in event networks but not nec-
essarily in spatial networks. The duplication of
F minor in (b) arises because the progression FIGURE 9.1.6 (a) A cross-​type transformation
completes a circuit around the toroidal topology network; (b) the underlying cross-​type transfor-
of the space between the first appearance of F mation graph
minor and the last. Nothing of the sort happens
between the two E-​major triads, so both are rep-
resented by the same vertex. a single space but a mapping from one space to
Figure 9.1.5b provides spatial context for the another—​and the network is a cross-​type trans-
chord progression by showing some nearby tri- formation network. Figure 9.1.6b, the underlying
ads in the space. Formally, a vertex or an arrow transformation graph of the network in (a), is a
either belongs to a network or it does not; the cross-​type transformation graph.
nearby triads such as E minor and A major may Though not accounted for in Lewin’s formal-
be considered external to the network, or alter- ization of transformation networks, cross-​type
natively we may regard the figure as a larger net- graphs and networks are often useful. Chapter
work that incorporates them, along with a large 7 offered examples of mappings relating two dif-
number of bidirectional P, L, and R arrows corre- ferent scalar spaces and networks featuring such
sponding to the unlabeled edges. Similar consid- mappings; Figure 8.3.20 presented a network
erations apply to other graphic conventions we combining triads and seventh chords. In a gen-
have observed from time to time, such as enclos- eral cross-​type graph or network there may be
ing some vertex labels in parentheses or display- not just two spaces but an entire family of spaces
ing some transformations as dashed arrows. The S1, S2, …. Each vertex is assigned to one space (as
difference between spatial and event networks indicated in Figure 9.1.6 by the dual legends),
is itself only an informal distinction, and many and in a network must be labeled with an ele-
transformation networks do not fall cleanly into ment of that space.5 Even in a cross-​type trans-
either category. Network N1 from Figure 9.1.1a, formation graph with unlabeled vertices, the
for example, resembles an event network for the space associated with each vertex must be iden-
progression E♭ –​c –​A♭ –​f: the chords proceed tified. This is necessary in order to know what
through the network from left to right, though labels are available for each arrow: an arrow from
the network also includes two indirect transfor- a vertex in space S1 to a vertex in space S2 may be
mations linking nonconsecutive chords. On the labeled only with a function f: S1 → S2. Thus each
other hand, the zigzag path between major and QO arrow in Figure 9.1.6 points from a vertex in
minor triads hints at a spatial arrangement, and pitch to a vertex in pc; the label T12 would not be
it would be a simple matter to adjust the posi- possible on those arrows, and a QO arrow cannot
tions of the four vertices to match the configura- map from pitch to pitch. Functions f: S1 → S2
tion of the triads in a tonnetz. and g: S3 → S4 may be composed in the order f ● g
The network in Figure 9.1.6a illustrates octave only if S2 ⊆ S3; the transformations appearing in a
equivalence by displaying elements from two dif- cross-​type graph or network therefore cannot be
ferent spaces, pitch space and pitch-​class space. composed arbitrarily but only in certain orders.
The pitch transposition T12 relates two pitches All the transformations in a cross-​type graph
an octave apart, both of which are mapped to the generally do not belong to a single group, semi-
same pitch class by the quotient map QO: pitch → group, or any other familiar algebraic structure.
pc from Section 2.4. The quotient map is a In conjunction with Figure 8.3.20 we noted
cross-​type transformation—​ not a function on that a cross-​type network may sometimes be

342 • E x ploring M usi c al S pa c es

regarded as, or replaced by, an ordinary single-​ transpositions relating the four seventh
type network on a single space S1 ∪ S2 ∪ ⋯. chords, Riemannian transformations relating
Because a function must be defined on every the triads, and an appropriate cross-​type
element of its domain, this interpretation transformation relating each seventh chord
requires that every transformation in a cross-​ to the following triad. Compare your network
type network be defined on all elements of that with
   Figure 9.1.1a.7
large space—​and there is not always a musically
coherent way to define all the functions in this 9.2 CONSISTENCY
way. Even in the simple case of Figure 9.1.6 it is PROPERTIES
not clear how QO could be sensibly defined on
the set pitch ∪ pc. Similarly, many interscalar Lewin’s formalization of transformation graphs
mappings in Chapter 7 were designed explicitly and networks includes a provision that we have
to map from one space to another, with no clear so far disregarded, a requirement that in fact
way to apply them to larger domains. For such many graphs and networks in this book and else-
reasons many transformation networks are most where have violated: the so-​called path consis-
effectively presented in cross-​type fashion.6 tency condition. Path consistency applies when
In addition to being the first cross-​type trans- two different directed paths lead from one vertex
formation to appear in this chapter, the quotient to another, a situation we will refer to as paired
map QO in Figure 9.1.6 is also the first transfor- paths. Figure 9.2.1 shows paired paths from
mation that lacks an inverse. To have an inverse, vertex v to vertex w, one path with transforma-
a function must be a bijection, and while QO tions f1, f2, …, fm and the other with g1, g2, …, gn.
certainly maps pitch onto pc, a mapping from Paired paths need not be of the same length, so
an infinite space to a finite space can never be m and n may differ. Path consistency requires
one-​to-​one. When all transformations appearing that the products of the transformations along
in a graph or network have inverses, the graph is paired paths be equal: f1 ● f2 ● ⋯ ● fm =​ g1 ● g2 ● ⋯ ● gn.
reversible as defined in Section 3.5: every arrow If this equality holds for every instance of paired
may be replaced by an oppositely directed arrow paths in a graph or network, the graph or net-
labeled with the inverse of the original transfor- work is path-​consistent. Path consistency applies
mation. (Recall the T-​net in Figure 3.5.1b and both to transformation graphs and to networks,
its reverse in 3.5.3d.) A network and its reverse but vertex labels play no role in the property, so
convey essentially identical information about a transformation network is path-​consistent if
structural relationships among the elements and only if its underlying graph is path-​consis-
of the space or spaces, though arrow directions tent. Lewin presents the condition as part of his
may convey information of other kinds such definition of transformation graphs, a require-
as the chronology of a musical passage. All the ment that must be satisfied by every proper
networks and graphs in Figures 9.1.1–​9.1.5 are transformation graph or network. Here we
reversible, but those in 9.1.6 are not. Cross-​type extricate path consistency from the definition,
graphs can be reversible only if all of the spaces regarding it as an independent property that a
are of the same cardinality, a necessary condition graph or network may or may not satisfy.8
for the existence of bijections between them.
Even a single-​type graph may fail to be reversible
if some of its transformations are not bijective.
Group elements always have inverses, however,
so a single-​type graph whose transformations
belong to a group is always reversible.

Exercise 9.1.7 Construct a cross-​type

transformation network to model this
sequence from Mozart’s Piano Quartet in
E♭ Major, K. 493, I, mm. 172–​75:

B♭7 –​E♭ –​G7 –​c –​E♭7 –​A♭ –​C7 –​f

Arrange seventh chords in one row of the FIGURE 9.2.1 Paired paths and the path consis-
network and triads in a second row. Include tency condition

Graphs and Networks; Serial Transformations • 343

 If Figure 9.2.1 occurs as part of a transfor- transformations; in fact all are UTTs. The R–​P–​
mation network in which element x occupies R chain resembles neo-​Riemannian analyses of
vertex v and element y occupies vertex w, and if similar passages in Chapters 4 and 8. The sec-
we write f for the product f1 ● f2 ● ⋯ ● fm and g ond half of the passage is an exact transposition
for g1 ● g2 ● ⋯ ● gn, then label consistency (which of the first half, a relationship captured by no
we have adopted as a requirement in all net- transformation other than T9. Path consistency
works) ensures that f(x) =​ g(x) =​ y. We know from effectively forbids the analyst from combining
Section 2.3, however, that an equation of the these observations in one network, not because
form f(x) =​ g(x) for a particular x does not guar- of a contradiction in this network but because of
antee that the functions f and g are the same, a hypothetical contradiction in some other net-
as path consistency demands. Path consistency work, a mode-​reversed one that does not model
thus implies a stronger conclusion than label the musical passage in question. With R and P
consistency, and path consistency may fail even arrows placed as shown, the only transformation
for a network that displays transformational that could appear on the E♭ → C arrow in a path-​
relationships accurately. Figure 9.2.2, which consistent network would be RP (though it could
slightly modifies Figure 9.1.4a, offers a simple be called 〈+​, 9, 3〉 or S9 or some other name for
example. This network is based on an analysis by the same function). This label RP would not fit
Robert Cook of a passage from the introduction the c → a arrow, which could only be labeled PR
to the first movement of Franck’s Piano Quintet (or S3), so bubble notation would be unavailable
(also analyzed in Exercise 4.2.13d): a modula- and the second half of the passage could not be
tion from E♭ major to C minor is followed by a described as the simple transformation of the
shift to the parallel major and a repetition of first half that it manifestly is. A path-​consistent
the same modulation, transposed down a minor labeling is essentially redundant; no path-​con-
third.9 As in Figure 9.1.4, the bubble nota­tion sistent network can convey the same fusion of
implies two arrows,E♭ T 9
→ c and c T9
→a . musical insights that the non–​path-​consistent
The network thus includes paired paths from one can.
E♭ to C, a path E♭ T 9
→ c of length 1 and a path Wittingly or unwittingly, many analysts have
♭
E  R
→ c → P
 c of length 2. As UTTs, T9 is flouted Lewin’s requirement, producing net-
〈+​, 9, 9〉, while RP is 〈+​, 9, 3〉, the Riemannian works for which path consistency fails. First to
schritt S9. These two transformations will do so was Lewin himself: at least three networks
always produce the same result when applied in the pages of GMIT fail the path consistency
to a major triad as they are here, but never test.11 Rings (2011b, 115–​ 19) explicitly sets
when applied to a minor triad; they are not aside Lewin’s path consistency requirement in
the same function, the transformational prod- his work, noting the usefulness of non–​path-​
ucts along the paired paths do not match, consistent graphs in analysis and offering sev-
and the network is not path-​ consistent.10 eral examples.12
Lewin provides little justification for requir- Path consistency interacts with other con-
ing path consistency in every case, saying only cepts we have explored. The failure of path
that it is needed “to ensure that no contradiction consistency in Figure 9.2.2 involves the simulta-
can possibly arise.” But Figure 9.2.2 contains no neous presence of Riemannian transformations
contradictions. The network is label-​consistent, P and R along one path and the non-​Riemannian
and the transformations are legitimate triadic transposition T9 along the other. In the path-​
consistent alternative proposed above, with
RP and PR in place of T9, all transformations
are Riemannian. As detailed in Section 8.2,
the Riemannian transformations form a group
whose action on triad is simply transitive. Every
network all of whose transformations belong to
one simply transitive group, it turns out, is path-​
consistent. To see this, recall that any network
containing paired paths as in Figure 9.2.1 must
satisfy f(x) =​ g(x) =​ y, where x is the element at
vertex v, y is the element at w, and f and g are the
FIGURE 9.2.2 A triadic transformation network composite functions along the paired paths. If
that is not path-​consistent in this situation all the transformations belong

344 • E x ploring M usi c al S pa c es

to a simply transitive group, f and g must be the Exercise 9.2.3 Seven transformation networks
same function. (This reasoning relies on vertex from this book are listed below. Determine
labels and label consistency, so the conclusion is whether each of them is path-​consistent.13
valid for transformation networks only, not for
transformation graphs. The conclusion holds, (a) The network of row forms in Figure 3.5.8d
however, in the case of realizable transformation (b) The Debussy network in Figure 7.3.3
graphs, discussed below.) (c) The PL-​cycle in Figure 8.2.13a (consider the
Path consistency of networks with sim- network to contain bidirectional arrows for
ply transitive groups applies, for example, to each P, L, and H)
either of the two networks in Figure 9.1.5. The (d) The Ravel network in Figure 8.2.21c
network in 9.1.5a is trivially path-​consistent: (e) The Rimsky-​Korsakov network in
a graph with no paired paths cannot violate Figure 8.3.20
path consistency. But 9.1.5b is path-​consistent (f) Network N1 in Figure 9.1.1a
as well, even if all the unlabeled edges in the (g)  The cross-​type network in Figure 9.1.6a
figure are included in the network with appro-
priate PLR labels, because all of the transforma- Exercise 9.2.4 Is the network you constructed
tions are Riemannian. This complete version of in  
Exercise 9.1.7 path-​consistent?
Figure 9.1.5b includes many examples of paired
paths, such as E R → d♭ →  D♭ →
P L
 f and In addition to path consistency, several other
E → L ♯
 g →  A♭ 
P R
→ f, but path consistency consistency properties of graphs and networks
holds in every case (recall from Section 8.2 that may be formulated. The four graphs in Figure
RPL =​ LPR =​ Pʹ ). 9.2.5 illustrate some of these properties. All of
Similarly, any triadic transformation network these figures are transformation graphs, not
whose transformations are all transpositions networks. Spaces are not shown; all four graphs
or inversions is automatically path-​consistent, are understood to be defined on the space triad.
because the action of 𝒯ℐ12 on triad is simply The argument for requiring path consistency
transitive. By Lewin’s theorem (Theorem 5.6.7), is perhaps motivated by graphs like the non-
interval spaces always give rise to simply tran- sensical one in Figure 9.2.5a. The graph is not
sitive groups, so any transformation network path-​consistent: the transformational products
whose transformations are transpositions in along the paired paths from the leftmost to the
an interval space—​ any network depicting an rightmost vertex are T1 and T0. More obviously,
interval configuration, in the generalized termi- though, our intuition tells us immediately that
nology from Section 5.4—​is path-​consistent. In this graph is impossible. Formally, we say that a
Example 8.2.27, however, we noted analytical transformation graph is realizable if some net-
benefits of combining transpositions with wech- work realizes it—​that is, if its vertices can be
sels. No simply transitive group mixes trans- labeled with elements of the appropriate space
positions and wechsels, and the paired paths (or spaces, in the case of a cross-​type graph) to
involving H, W10, and T2 prevent Figure 8.2.28a produce a valid transformation network. The
from being path-​consistent. underlying graph of any transformation network

FIGURE 9.2.5 Consistency properties of transformation graphs on triad

Graphs and Networks; Serial Transformations • 345

is therefore automatically realizable, but a graph unrealizable for essentially the same reason,
like 9.2.5a has no possible triadic realization: and they differ only in the directions of arrows
in any proper network realization, the three labeled T0. If the objective is to prohibit self-​con-
T0 arrows imply (by label consistency) that the tradictory graphs like these, path consistency
same triad must occupy all four vertices, ren- would appear to fall short of what is needed.
dering the T1 arrow impossible. If, however, the It is a simple matter to strengthen the defini-
same figure is considered to be a transformation tion of path consistency to disallow a graph like
graph on the space of all pitch-​class sets, rather 9.2.5c. This graph contains no paired directed
than just triads, it becomes realizable, as we may paths of the sort we have been considering, but
place the complete twelve-​tone aggregate or the there are paired undirected paths from the left-
empty set in all four vertices. This observation most vertex to the rightmost—​ by which we
reminds us why the space on which a graph is mean they are paths if arrow directions are disre-
defined is an important part of the definition of garded. In order to travel from one vertex to the
the graph: changing the space can change some other along either of these paths, an edge must be
properties of the graph. traversed backward, in the direction opposite the
The graph in (b) presents a subtler case. This arrow. We may form the transformational prod-
graph is unrealizable also, but it may appear to uct along such a path, replacing a transforma-
contain no paired paths and therefore no poten- tion by its inverse whenever an edge is traversed
tial violations of path consistency. In fact, how- backward. We call such products directionally
ever, there are paired paths in this graph. One adjusted. The directionally adjusted products
can follow two different paths from a vertex to along the paired undirected paths in (c) are T0−1
itself by going around the cycle either once or ● T =​ T along the upper path, T −1 ● T =​ T along
1 1 0 0 0
twice: one path is of length 4, with transfor- the lower. A graph is strongly path-​consistent if
mational product T1, and the other is a path all paired undirected paths have equal direction-
of length 8 traversing every edge twice, with ally adjusted transformational products. By this
product T2. Graph (b), like (a), is therefore not definition, graph (c), though path-​consistent, is
path-​consistent. In general, if the product of not strongly path-​consistent. Replacing trans-
transformations around a directed cycle is f, path formations with their inverses is possible only if
consistency demands that f ● f =​ f; if f is bijective all transformations have inverses, so strong path
(in particular, if f belongs to a group), it follows consistency is defined only for reversible graphs
that f must be the identity mapping. and networks; in particular, it is defined for all
Path consistency similarly constrains oppo- single-​type transformation graphs or networks
sitely directed paired arrows like those shown whose transformations belong to a group.
previously in Figure 9.1.4b. Two such arrows The realizability condition, too, may be
form a cycle of length 2, so by the argument in strength­ened. Figure 9.2.5d is a subgraph of
the preceding paragraph, for such a graph to be the underlying transformation graph of Figure
path-​consistent the product f of the transforma- 9.2.2. As we have noted, this graph is not path-​
tions on the two arrows must satisfy f ● f =​ f. If consistent because T9 ≠ RP. The graph is never-
the transformations are bijections, then f is the theless realizable; its realization in Figure 9.2.2
identity, the two transformations are inverses, motivated our argument that non–​ path-​con-
and they can be equal only if the transformation sistent graphs should not be rejected outright.
is an involution. Further, if any loops (arrows The possible realizations of this graph, however,
starting and ending at the same vertex) appear are limited: if the leftmost vertex is occupied by
in a transformation graph, path consistency a major triad, the graph can be realized, but if
requires that each loop be labeled with a trans- a minor triad is placed there it cannot. (Applied
formation f satisfying f ● f =​ f—​which, if bijec- to C minor, for example, T9 yields A minor while
tive, can only be the identity.14 RP yields E♭ minor.) We say that a transforma-
With Figure 9.2.5c, we encounter an unrealiz- tion graph is universally realizable if it remains
able arrangement that truly contains no paired realizable after any single vertex of the graph is
paths, and which is therefore trivially path-​ labeled arbitrarily—​that is, if for any labeling of
consistent. Of the four graphs in the figure, one vertex with any element of the appropriate
only (c) is path-​consistent and therefore a valid space, some labeling of the remaining vertices
transformation graph according to Lewin’s defi- is label-​consistent with the arrows. Universal
nition. It may seem odd that path consistency realizability fails for Figure 9.2.5d for the rea-
does not apply equally to (a), (b), and (c): all are sons we have noted: the leftmost vertex may be

346 • E x ploring M usi c al S pa c es

labeled not arbitrarily but only in certain ways. different ways. A distinction between syntax
(The same is true for each of the other vertices: and semantics is recognized in linguistics, formal
the rightmost vertex must be major, the lower logic, and computer programming. Path consis-
vertex minor.) tency and strong path consistency are syntactic
properties, constraints on the formal structure of
Exercise 9.2.6 Consider the underlying a graph. Realizability and universal realizability,
transformation graphs of the seven networks in contrast, are semantic properties, concerned
from Exercise 9.2.3 and the one from Exercise less with the structure of a graph than with its
9.2.4. Because all these graphs were originally meaning: its ability to model a real musical situ-
presented as networks, all are realizable. Which ation. The following theorem shows, remarkably,
of  
them are universally realizable?15 that the stronger of the two syntactic properties
and the stronger of the two semantic properties
Exercise 9.2.7 are logically equivalent.
(a) Using two vertices, two opposing arrows,
and the same bijective transformation Theorem 9.2.9: Completeness theorem
on both arrows, construct a triadic for transformation graphs A reversible
transformation graph that is realizable but transformation graph is strongly path-​consistent
not universally realizable. if and
   only if it is universally realizable.
(b) Construct a realizable triadic
transformation graph with a single vertex For a proof of the theorem see Hook 2007a,
and a loop bearing a transformational label 36–​38. The theorem is called the complete-
  f such that f ● f ≠ f.16 ness theorem by analogy with certain results
in mathematical logic known by that name,
Exercise 9.2.8 If a reversible graph includes asserting equivalence of syntactic and semantic
an undirected cycle (as in part (a) or (c) of concepts.18
Figure 9.2.5), and if the graph is strongly
path-​consistent (which those figures are not), Exercise 9.2.10
then by reasoning similar to the discussion (a) It is clear that strong path consistency
above, the directionally adjusted product of implies path consistency and universal
transformations around the cycle must be realizability implies realizability.
the identity. Explain why the converse is also These observations, together with the
true: if a reversible transformation graph has completeness theorem, limit the possible
the property that the directionally adjusted combinations of consistency properties
transformational product around every cycle is that a graph may satisfy to five: a graph
the identity transformation, then the graph is may be (1) neither path-​consistent
strongly
   path-​consistent.17 nor realizable; (2) path-​consistent but
not strongly path-​consistent, and not
If a subgraph of a transformation graph realizable; (3) realizable but not universally
violates path consistency, the same violation realizable, and not path-​consistent;
is present in the full graph. A realization of a (4) path-​consistent and realizable but
graph, meanwhile, is also a realization of every not strongly path-​consistent and not
subgraph of that graph. It follows from these universally realizable; or (5) strongly
observations that all four of the consistency path-​consistent and universally realizable.
properties defined in this section—​path consis- Give examples of triadic transformation
tency, strong path consistency, realizability, and graphs of all five types.19
universal realizability—​are hereditary, mean- (b) For a graph all of whose transformations
ing that if a graph has a certain property, all its belong to a simply transitive group, only
subgraphs have the property also. It is possible, three of the five types are possible. Which
though, for a subgraph to have stronger consis-   types are these?
tency properties than the larger graph of which
it is a part. For example, omission of the P arrow 9.3 ISOMORPHISM AND
would make Figure 9.2.2 into a path-​consistent ISOGRAPHY
network.
The four consistency properties describe In this book we have discussed the concept of
transformation graphs in two fundamentally isomorphism as it applies to several different

Graphs and Networks; Serial Transformations • 347

kinds of structures: graphs and directed graphs of N, then the object at the corresponding
in Chapter 3, groups in Chapter 6, and interval vertex of Nʹ is h(x), and if a transformation
spaces in Chapter 7. In this section we explore f: S → S appears on an arrow in N, then the
isomorphism of transformation graphs and transformation on the corresponding arrow in
networks. The topic is complex: structural like- Nʹ is Conjh(f ) =​ h−1 ● f ● h: Sʹ → Sʹ.
nesses between graphs and especially networks • Strong isomorphism of transformation networks.
take many different forms, so we will define rela- Two transformation networks are strongly
tionships of several kinds. Varied as they are, isomorphic if they are both isomorphic
these definitions do not exhaust the possible and strongly isographic—​that is, if their
approaches to the subject.20 For ease of reference underlying transformation graphs are
we set out the relevant definitions first, with dis- identical and the objects at corresponding
cussion following. vertices are related by an isomorphism h as
described above. We will see that a network
• Isomorphism of directed graphs. As defined in isomorphism h is strong if and only if h
Section 3.4, an isomorphism from a directed commutes with all of the transformations
graph A to a directed graph Aʹ is a bijective that appear on the arrows in the network.
function θ from the vertices of A to the • Isomorphism of interval spaces. Interval-​space
vertices of Aʹ with the property that there is isomorphisms, defined in Section 7.2, are not
an arrow from θ(v1) to θ(v2) in Aʹ if and only mappings of graphs or networks, but they are
if there is an arrow from v1 to v2 in A. An potentially applicable if the spaces S and Sʹ
isomorphism of directed graphs is required on which networks N and Nʹ are defined are
for all the other types of graph and network interval spaces. In many musically significant
isomorphism and isography listed below. cases a network isomorphism h is also an
• Isomorphism of transformation graphs. interval-​space isomorphism, ensuring a kind
Suppose A and Aʹ are transformation graphs of intervallic coherence to the correspondence
on spaces S and Sʹ with transformation between networks.
groups G (transformations on S) and Gʹ
(on Sʹ ), respectively, and suppose A and Isomorphism of directed graphs, discussed
Aʹ are isomorphic as directed graphs. An previously, describes a correspondence between
isomorphism of transformation graphs is arrangements of vertices and arrows but says
then a group isomorphism φ: G → Gʹ such nothing about transformational structure. Of
that if a transformation f appears on an the conditions that involve transformations,
arrow in A, then the transformation on the isomorphism of transformation graphs is the
corresponding arrow in Aʹ is φ(f).21 weakest and the most abstract; detailed discus-
• Isography of transformation networks. An sion of this property will be deferred until later
isography of transformation networks in this section. The remaining properties in the
is an isomorphism of the underlying above list, which we discuss first, apply to trans-
transformation graphs. Two networks are formation networks. Isomorphism of networks
isographic if their underlying graphs are is stronger than isography, in that isography
isomorphic. depends only on the underlying transformation
• Strong isography of transformation networks. graphs without reference to the objects (vertex
Two transformation networks are strongly labels) in the networks; network isomorphism
isographic if their underlying transformation also involves a mapping of the objects. The con-
graphs are identical (not merely isomorphic). straints on network isomorphism are consider-
This requires that all corresponding able, making isography a more useful relation in
transformations be equal as mathematical many cases.
functions, which implies that the spaces on We start with an example of strong network
which they act must also be the same. isomorphism. Figure 9.3.1 reproduces the tri-
• Isomorphism of transformation networks. Suppose adic networks N1 and N2 from Figures 9.1.1 and
N is a network on a space S and Nʹ is a network 9.1.2. Assume for the moment that the figure
on a space Sʹ, and suppose the underlying shows two separate networks, disregarding the
graphs gr(N) and gr(Nʹ ) are isomorphic two bubbles and the arrow joining them. There
as directed graphs. An isomorphism of is an inconsequential change in notation from
transformation networks is a bijection h: S → Sʹ the earlier figures: the dominant transformation
such that if an element x of S appears at a vertex D has been renamed T5 to clarify relationships

348 • E x ploring M usi c al S pa c es

FIGURE 9.3.1 The transposition T9 as isomorphism between networks N1 and N2

with other transformations in the discussion transformations within the bubbles. All the
below. N1 and N2 are two different networks, but transformations in the network N1 are UTTs,
as noted in Section 9.1, the underlying transfor- which always commute with transpositions.
mation graphs gr(N1) and gr(N2) are identical in It is precisely because T9 commutes with R, for
every respect: they have identical configurations example, that R may occur in the same locations
of vertices and arrows as well as identical trans- in both bubbles: the A-​minor triad in N2 may be
formational labels on corresponding arrows. The obtained from E♭ major in N1 via either R ● T9 or
identity of underlying transformation graphs T9 ● R. Commutativity thus ensures the strong
means that the networks N1 and N2 are strongly isography of the networks: an isomorphism
isographic. Any realizable transformation graph is generally not required to commute with all
may be regarded as an equivalence class of trans- transformations in the network to which it is
formation networks under the relation of strong applied, but if it does, the underlying transfor-
isography. Strong isography, as its name sug- mation graph is unchanged in the transformed
gests, is a special case of the more general phe- network, so the original and transformed net-
nomenon of isography, to be considered below. works are strongly isographic, and the isomor-
More can be said about the relationship phism is strong.
between N1 and N2. The triads appearing in the Network isomorphism implies isography, but
two networks are related in a systematic way: it does not imply strong isography. Networks
compared with the triads in N1, those in N2 can be isomorphic without being strongly iso-
are transposed by T9. The four-​chord progres- graphic or vice versa. Bubble notation may be
sions modeled in the two networks are, in fact, generalized to accommodate isomorphisms that
parts of the same harmonic sequence—​the long do not commute with all transformations in a
Beethoven sequence from Exercise 4.2.13e. network, giving rise to forms of network iso-
(Network analysis of sequential passages often morphism lacking the identical transformation
reveals strongly isographic blocks.) The bubble graphs characteristic of strong isography. The
notation in Figure 9.3.1 shows the T9 relationship network N3 appearing in the right half of Figure
explicitly, extending the mapping first shown in 9.3.2, for example, attempts to recreate the con-
Figure 9.1.4 to encompass the entirety of the figuration of N1 and N2 but with a minor triad
networks. Figure 9.3.1 may now be regarded as in the leftmost position. In comparison with N1
a single large network, but it remains useful to and N2, the vertices, arrows, and R and L labels
think of it as a simultaneous depiction of the two appear just the same in N3, but the former T5
networks N1 and N2 (indicated by the separate labels have had to be replaced by T7. The under-
labels) and the transposition that relates them. lying transformation graph gr(N3), therefore, is
We say that N2 is the T9-​transform of N1, and we not the same as gr(N1) and gr(N2); the relation-
write N2 =​ T9(N1).22 The transposition T9 is an iso- ship between N3 and either N1 or N2 is not one of
morphism of transformation networks, and since strong isography.
this isomorphism is also a strong isography, we It is, nevertheless, a kind of isomorphism.
call it a strong isomorphism. Networks N1 and N2, Figure 9.3.2 depicts the relationship between
that is, are both strongly isographic and isomor- N1 and N3 using the bubble transformation I0
phic, and therefore strongly isomorphic. in place of the earlier T9: each triad in N3 is the
We have noted previously that bubble nota- I0-​inversion of the corresponding triad from N1.
tion works as long as the bubble transforma- We know from Section 8.3 that inversions In
tion (T9 in Figure 9.3.1) commutes with all commute with Riemannian transformations, so

Graphs and Networks; Serial Transformations • 349

FIGURE 9.3.2 The inversion I0 as isomorphism between networks N1 and N3

the R and L arrows in N3 match those in N1. But As detailed in Chapter 6, symmetry manifests as
inversions do not commute with transpositions; invariance under some operation; the network
the arrows in Figure 9.3.2 confirm the formula N2 is invariant under the operation of reverse-​I0.
T5 ● I0 =​ I0 ● T7 from Theorem 5.5.6e. The mapping The isomorphisms T9 (from N1 to N2 in Figure
I0 in this example is a network isomorphism but 9.3.1) and I0 (from N1 to N3 in Figure 9.3.2) are
not a strong isomorphism. both bijective transformations on triad. Any
bijective mapping of spaces may give rise to a
Exercise 9.3.3 Using bubble notation, draw network isomorphism in the same way.23 Figure
a network showing an isomorphism between 9.3.4 shows the general situation: a network N is
the networks N2 and N3. Is this isomorphism a defined on a space S, and h: S → Sʹ is a bijective
strong
   isomorphism? mapping from S into a possibly different space
Sʹ. A new network Nʹ is defined on the space Sʹ
The reader may have noticed another kind of as follows. The vertices and arrows of Nʹ are in
resemblance between N2 and N3 besides the one one-​to-​one correspondence with those of N; that
described in the above exercise: the four triads is, the directed graphs are isomorphic. If an ele-
of N2—​C, a, F, and d—​appear also in N3, rear- ment x of S appears at a certain vertex in N, then
ranged. If we form the reverse of N2—​leaving its the element at the corresponding vertex in Nʹ is
triads unchanged, reversing the direction of each xʹ =​ h(x), an element of Sʹ.
arrow, and replacing each transformation with There remains only to specify how the trans-
its inverse (leaving R and L in place but replacing formational label f on an arrow x → y in N is
T5 with T7)—​and if we also perform the cosmetic converted to a label f ʹ on the corresponding
alteration of rotating the picture by 180 degrees, arrow xʹ → yʹ in Nʹ, where xʹ =​ h(x) and yʹ =​ h(y).
the resulting network is exactly N3. The relation- In fact the transformations are related by con-
ship between N2 and N3 can thus be described jugation: f ʹ is the function Conjh(f) =​ h−1 ● f ● h,
in two different ways: as an isomorphism by which satisfies f ʹ(xʹ ) =​ yʹ in Nʹ. (Label consis-
pitch-​class inversion or as a network reversal. tency in N requires f(x) =​ y, from which we see
−1
Neither of these relationships implies the other; that x ′ h→ x f → y h → y ′. For conjugation
both are applicable here because the networks recall Figure 6.6.9a; the role of the function g in
under consideration exhibit a certain symmetry. that figure is now played by f, and the original f
is now h.) Because h is bijective, h−1 exists and
a conjugate of this kind can formed for every f;
it is not necessary that f and f ʹ be bijective. If
h commutes with f, then f and f ʹ are the same
function. If h commutes with all transforma-
tions appearing in N, as was the case with h =​
T9 in Figure 9.3.1, then the underlying graphs
gr(N) and gr(Nʹ ) are identical, so N and Nʹ are
strongly isographic and (because h is an isomor-
phism) strongly isomorphic. The isomorphism
h =​ I0 in Figure 9.3.2 illustrates the case in which
full commutativity does not hold, so that some
FIGURE 9.3.4 Isomorphism of transformation transformations in the two networks differ even
networks though the networks are isomorphic.

350 • E x ploring M usi c al S pa c es

Exercise 9.3.5 In Figure 9.3.2, verify that every transformation M from Section 8.3. The two
transformation appearing in N3 is the conjugate networks are strongly isographic, as their under-
Conjh(f) of the corresponding transformation f lying transformation graphs are identical. For
in  
N1, where h =​ I0. them to be isomorphic, some bijective mapping
on triad would have to map the triads in N4 to
In Section 6.6 we noted that if a is an element the correspondingly positioned triads in N5, but
of a group G, then Conja: G → G is an automor- there is no such mapping: no bijective function
phism of G. If S =​ Sʹ in Figure 9.3.4, and if the maps both C → d and F → d, and no function at
transformations in the network N belong to a all maps both a → B♭ and a → F. The presence of
group G of transformations on S to which the the same element at more than one vertex of a
isomorphism h also belongs, then Conjh is an network can be a complicating factor in the con-
automorphism of G. Even if S and Sʹ are differ- struction of network isomorphisms; this is one
ent, as long as the transformations in N belong reason why isomorphism is not always the most
to a group G, Conjh is an isomorphism from G useful way to describe relationships between
to a group Gʹ =​Conjh(G) of transformations on similar networks and why we must explore other
Sʹ. Whenever the transformations in a network ways to characterize those relationships.
belong to a group, therefore, an isomorphism
of transformation networks implies an isomor- Exercise 9.3.7 Are networks N4 and N5 path-​
phism of groups.24 consistent?
   Are they strongly path-​consistent?25
The “prime” notation we have been using (N,
Nʹ; S, Sʹ; x, xʹ; f, f ʹ; G, Gʹ ) disguises the fact that For an example of isomorphic networks
an isomorphism of transformation networks defined on different spaces, the reader may again
involves several different mappings between the consult Figure 8.3.20, modeling two passages
two networks. First, there must be an isomor- in Rimsky-​Korsakov’s Christmas Eve. If the two
phism of directed graphs, a mapping of the ver- bubbles in the figure are regarded as separate net-
tices of N to the vertices of Nʹ which preserves works, one on the space triad and the other on
arrow relationships. Second, there is the func- sev (the space of dominant and half-​diminished
tion h (informally, “the isomorphism”), mapping seventh chords), the bubble mapping DX is an
the elements of S bijectively to the elements of isomorphism between the two. Though the trans-
Sʹ. Finally, Conjh (under appropriate conditions, formations within both bubbles are notated iden-
a group isomorphism) maps the transformations tically, they are defined on different spaces and are
in N (elements of G) to the transformations in Nʹ therefore not really the same functions; the two
(elements of Gʹ ). If Conjh is the identity mapping networks, while isomorphic, are not strongly iso-
on G, then the same transformations appear graphic. The identically notated transformations
in both networks, which are therefore strongly are related to each other by ConjDX; for example,
isomorphic. Rtriad =​(DX)−1 ● Rsev ● DX.
Figure 9.3.2 demonstrated that isomorphic In Section 8.1 we noted that triad forms an
transformation networks are not always strongly interval space with the commutative interval
isographic; Figure 9.3.6 shows that strongly iso- group G =​ ℤ12 × 𝒮𝑖𝑔𝑛. In this interval space, for
graphic networks need not be isomorphic at all. example, int(E, c♯) =​(9, −). Likewise, sev may
Networks N4 and N5 combine the Riemannian be regarded as an interval space with the same
transformations R and L with the mediant group G; in this space int(B7, g♯ø7) =​(9, −). The
isomorphism DX between the two networks is
also an interval-​space isomorphism; in fact, DX
is interval-​ preserving, which in terminology
from Section 7.2 means that DX is interval-​com-
patible with the identity mapping on the group
G. The fact that the network isomorphism is also
an interval-​space isomorphism ensures that the
similarity of network structure carries over to
encompass intervallic structure as well.
The definition of network isomorphisms car-
ries no guarantees about intervallic structure,
FIGURE 9.3.6 Networks N4 and N5: strongly iso- and it is possible to construct many network
graphic but not isomorphic isomorphisms that do not have this property.

Graphs and Networks; Serial Transformations • 351

In the general setting of Figure 9.3.4, any generally in Section 9.4. Network N6 shows
bijection h: S → Sʹ theoretically gives rise to a the notes of an 014 pc set related by transpo-
network isomorphism as long as the transfor- sitions T1, T3, and T4. N7 is the I0-​transform of
mations in Nʹ are h-​conjugates of those in N. N6: its pitch classes have been inverted about
Of the vast number of possible bijective map- C, forming an inverted 014 set, and its trans-
pings between two sets (such as the 24! bijec- positions invert those in N6. The mapping I0: pc
tions between triad and sev), most have no → pc is a bijection and therefore defines a net-
apparent musical relevance, and such mappings work isomorphism as described above. But pc
may be used to produce isomorphic networks is an interval space, and I0 is an interval-​space
that, while theoretically valid, appear musically isomorphism. The definition of interval-​space
nonsensical. For example, one could produce a isomorphism requires that there be a group iso-
network isomorphic to N1 in which the posi- morphism g: ℤ12 → ℤ12 with which I0 is interval-​
tions originally occupied by the triads E♭, c, A♭, compatible in the sense that int(I0(x), I0(y)) =​
and f are filled by the seventh chords fø7, f♯ø7, g(int(x, y)) for all pitch classes x and y (recall
bø7, and A7, respectively. There exist bijections Figure 7.2.1). In fact, int(I0(x), I0(y)) =​−int(x, y),
h: triad → sev that map chords in this way, but so g is the mapping g(n) =​−n (which as a func-
any such mapping will result in an incoherent tion on ℤ12 is the same as I0). It is useful for
jumbling of intervallic relationships between future reference to recall that I0 could also be
chords, which is to say that such functions h are called M11, mod-​12 multiplication by 11 (or −1),
not interval-​space isomorphisms. The transfor- and also that I0 is an involution, so the arrow
mations in the h-​transformed network, being joining the bubbles for N6 and N7 points both
conjugations-​by-​jumbling of the original trans- ways. Because N6 and N7 are two different net-
formations, will likewise be musically incoher- works sharing the same space pc and the same
ent (though, inasmuch as the transformations interval group ℤ12, it is technically correct to say
in N1 are UTTs, the new transformations will that Figure 9.3.8 shows an automorphism of both
all belong to a group of incoherent transforma- the interval space and the group, but an isomor-
tions isomorphic to the UTT group). For these phism (not an automorphism) of networks.
reasons, the most revealing network isomor- In a similar way, N8 is obtained from N6 via
phisms are often interval-​space isomorphisms the multiplication operator M7. In comparison
as well. Exercise 9.3.11 below confirms that the with N6, both the pitch-​class numbers and the
isomorphisms T9 and I0 in Figures 9.3.1 and index numbers of the transpositions in N8 have
9.3.2 have this property if interval spaces are been multiplied by 7. This mapping too is both
defined appropriately. a network isomorphism and an interval-​space
Figure 9.3.8 simplifies intervallic consider- isomorphism (because g =​ M7 is a group auto-
ations by using pitch classes rather than triads morphism of ℤ12), as well as an involution. The
and limiting the transformations in the net- pc-​set in N8 is a C-​major triad, the M7-​transform
works to ordinary transpositions: the four net- of N6’s 014 set.
works in this figure are T-​nets. T-​nets may be The last network in Figure 9.3.8, N9, may be
considered to be especially simple examples of described as the M2-​transform of N6: in compari-
Klumpenhouwer networks, to be studied more son with N6, the interval sizes in N9 have been

FIGURE 9.3.8 Intervallic relations in four T-​nets

352 • E x ploring M usi c al S pa c es

doubled. While the process by which this new T9 in Figure 9.3.1 is an interval-​space
network is formed resembles the processes that isomorphism in this space. What is the
produced N7 and N8, there is a fundamental dif- group automorphism?
ference in that the mapping M2: ℤ12 → ℤ12 is not (d) With the same interval space as in (c),
a bijection, a group isomorphism, or an involu- show that I0 in Figure 9.3.2 is an interval-​
tion. Odd-​numbered pitch classes or transposi- space isomorphism. What is the group
tions can never appear in an M2-​transformed   automorphism?28
network; had any tritone-​related pitch classes
appeared in the original network N6, the M2 We have discussed isomorphism and strong
transformation would have introduced a pitch-​ isography of transformation networks, but we
class duplication. M2 has no inverse: a second have yet to say much about the more general
application of M2 would not map N9 back to N6, kind of isography. Network isography depends
nor would any other multiplication operator on only on the transformation graphs underlying
ℤ12 do so. For these reasons M2 fails to be a net- the networks, not on the objects appearing as the
work isomorphism. Notwithstanding its similar- vertex labels. Two networks are isographic if and
ity to the other three networks in the figure, N9 only if their underlying transformation graphs
is not isomorphic to them.26 are isomorphic—​so we now turn our attention
to isomorphism of transformation graphs.
Exercise 9.3.9 Our understanding of network isomorphism
(a) Recall from Section 6.5 that the only reveals something about graph isomorphism,
automorphisms of ℤ12 are M1 (=​ T0), because every network isomorphism is also an
M5, M7, and M11 (=​ I). The M7-​ and M11-​ isomorphism of the underlying transformation
transforms of the network N6 are N8 and N7 graphs. Suppose N is a transformation network
respectively. Draw the M5-​transform of N6. defined on a space S, whose transformations
(b)  What isomorphism relates N7 and N8? belong to a group G of transformations on S,
and let A =​gr(N) be the underlying transforma-
Exercise 9.3.10 Let N8ʹ be the T2-​transform tion graph.29 We have seen that a network iso-
of N8, a network whose notes form a D-​major morphism from N to another network Nʹ entails
triad. This network is isomorphic to N6, N7, and a bijection h from S to the space Sʹ associated
N8. Write a formula for the bijection h: pc → pc with Nʹ, and also a group isomorphism Conjh
that maps the pitch classes of N6 to those of N8ʹ. from G to a group Gʹ of transformations on Sʹ.
In this case the group automorphism g: ℤ12 → Because we have stipulated that the space asso-
ℤ12 with which h is interval-​compatible is not ciated with a transformation graph must be
the same function as h; in fact h is not a group specified as part of the definition of the graph,
isomorphism.
   What is g?27 the mappings h and Conjh can be described for
the graph A without referring to the specific
Exercise 9.3.11 elements of S that appear in N. That is, we can
(a) Consider triad to be an interval space define such mappings on transformation graphs,
with the commutative interval group G referring to the spaces but without reference to
=​ ℤ12 × 𝒮𝑖𝑔𝑛 as described in Section 8.1; vertex labels; we can do so even for unrealizable
for instance, int(E♭, c) =​(9, −). Verify graphs, for which no corresponding network N
that the network isomorphism T9 from exists at all.
N1 to N2 in Figure 9.3.1 is an interval-​ The description in the preceding paragraph,
space isomorphism. What is the group which manufactures a graph isomorphism from a
automorphism g: G → G with which T9 is bijection of spaces, accounts for many instances
interval-​compatible? of graph isomorphism, but it does not account for
(b) With the same interval space as in (a), all of them. We can see as much by returning to
verify that the network isomorphism Figure 9.3.6, the two networks that are strongly
I0 from N1 to N3 in Figure 9.3.2 is an isographic but not isomorphic. The underlying
interval-​space isomorphism. What is the graphs of these two networks, being identical,
corresponding group automorphism? are surely isomorphic—​but as we have noted, no
(c) A different interval structure on triad, bijection h correctly relates the objects in the two
discussed in Sections 5.6 and 7.2, has the networks. Some examples of graph isomorphism,
dihedral group 𝒯ℐ12 as its interval group; therefore, arise from bijective mappings on the
for instance, int(E♭, c) =​I10. Verify that associated spaces, but some others do not.

Graphs and Networks; Serial Transformations • 353

FIGURE 9.3.12 (a) Two isomorphic transformation graphs; (b) two isographic transformation
networks

What a graph isomorphism requires in gen- LʹR-​cycle previously as well: it maps the trajectory
eral is a group isomorphism φ: G → Gʹ between traced by the excerpt from Ravel’s “Ondine” in
the transformation groups. To be precise, if A Figure 8.2.20 through the hexatonic bridge family
and Aʹ are transformation graphs on spaces S hb2. Some other sequential paths through hexa-
and Sʹ with transformation groups G and Gʹ, tonic triad and bridge families generate 6-​cycles
respectively, then an isomorphism from A to that are isomorphic, as transformation graphs, to
Aʹ requires an isomorphism of the underlying those in Figure 9.3.12a. See Exercise 9.3.15 below
directed graphs and a group isomorphism φ: G for some related examples.
→ Gʹ. If the transformational label on an arrow Isomorphic transformation graphs are iso-
in the graph A is f, then the label on the corre- morphic as directed graphs, but the converse
sponding arrow in Aʹ must be φ(f). In some cases does not hold: if two transformation graphs are
φ is Conjh for some bijection h: S → Sʹ, but in gen- isomorphic as directed graphs, we cannot con-
eral there need not be such an h, and φ need not clude that they are isomorphic as transforma-
take the form of a conjugation. tion graphs. Two other transformation graphs
In Figure 9.3.12 the triadic transformations taking the form of 6-​cycles are shown in Figure
belong to the Riemann group ℛ. Exercise 8.2.7 9.3.13, and neither of these is isomorphic to
showed that for any mod-​12 integer k, the map- the graphs in 9.3.12. The transformations in
ping φ: ℛ → ℛ defined by φ(Sn) =​ Sn, φ(Wn) =​ Wn+​k the PLR-​cycle in 9.3.13a are not in one-​to-​one
is an automorphism of ℛ. For k =​5, φ maps the correspondence with those in the graphs from
transformation P (=​ W0) to Lʹ (=​ W5) and likewise 9.3.12—​ three transformations occur twice
maps L (=​ W4) to R (=​ W9). The two transforma- each, rather than two transformations three
tion graphs in 9.3.12a, a PL-​cycle and an LʹR-​cycle, times each—​so no group isomorphism φ could
are therefore isomorphic, as φ maps the transfor- match the transformations properly. The T1T3-​
mations in the first graph to the corresponding cycle in 9.3.13b is obviously not isomorphic to
transformations in the second graph. Part (b) of the graphs in 9.3.12 as it stands, since its arrows
the figure supplies triads to turn these isomor- point only one way—​but even if the graphs
phic graphs into a pair of isographic networks. from 9.3.12 were redrawn with one-​way arrows,
The PL-​cycle, familiar from the Brahms Double they would not be isomorphic to the T1T3-​cycle.
Concerto example first introduced in Section 3.1, The earlier figure’s transformations are involu-
circumnavigates the hexatonic triad family hex34 tions; the image of an involution under a group
from Figure 8.2.17. We have encountered the isomorphism must be another involution, and

354 • E x ploring M usi c al S pa c es

 centers of both hexagons for the hexatonic
pole relation H (=​ W8) in N10 and the slide
relation Pʹ (=​ W1) in N11. Networks N12 and N13
depict relationships among the same triads
(rearranged) using inversions In rather than
wechsels Wn. This exercise shows that the
underlying transformation graphs gr(N10)–​
gr(N13) are all isomorphic, and therefore the
networks N10–​N13 are all isographic.
FIGURE 9.3.13 Two transformation graphs not
isomorphic to those in Figure 9.3.12 (a) Verify that the isomorphism φ: ℛ →
ℛ defined by φ(Sn) =​ Sn, φ(Wn) =​ Wn+​5,
neither T1 nor T3 is an involution. The graphs in described above in conjunction with
Figure 9.3.13 are isomorphic to those in 9.3.12 Figure 9.3.12, determines an isomorphism
only as directed graphs; as transformation graphs of the expanded transformation graphs
they are not isomorphic. gr(N10) and gr(N11). Call this mapping φ10–​11.
(b) Define φ10–​12: ℛ → 𝒯ℐ12 by φ10–​12(Sn) =​
Exercise 9.3.14 The two isographic networks T−n, φ10–​12(Wn) =​ In+​7. Verify that φ10–​12
in Figure 9.3.12b are isomorphic as is a group isomorphism and that φ10–​12
transformation networks. Describe the maps each transformation in N10 to the
isomorphism h: triad → triad and verify that corresponding transformation in N12. Thus
   h(L) =​ R and Conjh(P) =​ Lʹ.30
Conj φ10–​12 determines an isomorphism from
gr(N10) to gr(N12).31
Exercise 9.3.15 The triadic networks N10–​N13 (c) Define an appropriate mapping φ12–​13: 𝒯ℐ12
in Figure 9.3.16 illustrate isomorphisms and → 𝒯ℐ12 and verify that φ12–​13 determines an
automorphisms involving the dihedral groups isomorphism from gr(N12) to gr(N13).32
ℛ and 𝒯ℐ12. Arrowheads are omitted in this (d) Finally, define an appropriate mapping
figure; all the transformations are wechsels or φ11–​13: ℛ → 𝒯ℐ12 and verify that φ11–​13
inversions, and all arrows are bidirectional. determines an isomorphism from gr(N11)
Networks N10 and N11 expand the networks to gr(N13). Is it true that φ10–​12 ● φ12–​13 =​
from Figure 9.3.12b, adding arrows across the   φ10–​11 ● φ11–​13?

FIGURE 9.3.16 Four triadic transformation networks with groups ℛ and 𝒯ℐ12, for Exercise 9.3.15

Graphs and Networks; Serial Transformations • 355

 In Figure 9.3.17, portions of the three net- networks are related by the inversion operator I8
works from Figure 8.2.28, depicting sequences (g–​F♯, B–​d, and so on). I8 commutes with wech-
in Martin’s Prelude No. 2, have been rear- sels but not with transpositions, so the same
ranged in order to highlight isomorphisms. wechsels appear in N15 and N16, while the trans-
Each of the three large networks N14–​N16 position T2 in N16 is the I8-​conjugate of T10 in
consists of two two-​chord blocks encircled by N15. This isomorphism I8, unlike U, is an involu-
dashed bubbles. Networks N14 and N16 depict tion, so its arrow is bidirectional.
sequences ascending by T2, while the sequence Networks N14 and N15 contain neither the
in N15 descends by T10. If the two bubbles in same wechsels nor the same transpositions,
one network are regarded as small networks and they are related neither by a UTT nor by
in their own right, the six small networks are an inversion. From the relationships we have
all strongly isographic, and the two small net- already established between these networks and
works within one large network are strongly N16, however, we can deduce that the composite
isomorphic, befitting the sequential pattern- mapping UI8 must be an isomorphism from N14
ing. The isomorphisms are the bubble transfor- to N15.
mations T2 and T10, which commute with the
transformation W8 (=​ H) inside the bubbles. Exercise 9.3.18
Our primary interest now is the relationships (a) Verify these assertions from the above
among the three large networks, each of which discussion of Figure 9.3.17: U−1W8U =​ W8;
contains a slightly different assortment of U−1W10U =​ W6; I8−1T10I8 =​ T2.
transformations. (b) Verify that UI8 maps the triads in N14 to the
Consider first the ascending sequences in N14 corresponding triads in N15. What is UI8
and N16. The UTT U =​ 〈−, 10, 6〉 maps each triad   as a QTT?
in N14 to the corresponding triad in N16: each
major triad in N14 corresponds in N16 to a minor We close this section with a small theorem.
triad a whole step lower (E–​d, F♯–​e), and each We have observed that it is possible for transfor-
minor triad maps to a tritone-​related major triad mation networks to be isographic without being
(c–​F♯, d–​A♭). This U: triad → triad is a network isomorphic. The networks in Figure 9.3.6 that
isomorphism, playing the role of the mapping illustrated this possibility combined Riemannian
called h in previous examples. Transpositions and non-​Riemannian transformations, and in
commute with all UTTs, including U, so T2 the next section Klumpenhouwer networks will
appears in both N14 and N16. U does not com- provide more examples. The theorem shows that
mute with all wechsels, however; the wechsels in no such example is possible when the transfor-
N16 are obtained from those in N14 by ConjU (see mation groups are simply transitive: in that case
Exercise 9.3.18 below). every isography is a network isomorphism.
Next consider N15 and N16. One sequence
ascends and the other descends, so the tri- Theorem 9.3.19: Isography of networks
adic mapping is not uniform. But it is straight- with simply transitive groups Suppose N is
forward: corresponding triads in these two a transformation network on a space S whose

FIGURE 9.3.17 Network isomorphisms in Martin, Prelude No. 2

356 • E x ploring M usi c al S pa c es

transformations belong to a simply transitive
group G of transformations on S, and Nʹ is a
transformation network on a space Sʹ whose
transformations belong to a simply transitive
group Gʹ of transformations on Sʹ. If the
networks N and Nʹ are isographic via a group
isomorphism φ: G → Gʹ, then N and Nʹ are
isomorphic via a bijection h: S → Sʹ for which
φ   
=​Conjh.

Exercise 9.3.20 This rather abstract exercise

outlines a proof of Theorem 9.3.19; fill in the
gaps.
Choose any element a of S and any element
aʹ of Sʹ. For any x in S, let gx denote the unique
function in G such that gx(a) =​ x, and let gxʹ
denote φ(gx), an element of Gʹ and therefore
a transformation on Sʹ. Define h(x) =​ gxʹ(aʹ ).
Explain why this definition implies that h is a
bijection and h(a) =​ aʹ.
To prove that φ =​Conjh, we must show that FIGURE 9.4.1 (a) Tetrachords α and β related
for every transformation f in G, if f ʹ =​ φ(f), by dyad transposition; (b) network showing
then f ʹ =​ h−1 ● f ● h. To establish this, we must dyad transpositions; (c) strongly isographic
show that for every xʹ in Sʹ, f ʹ(xʹ ) =​ h(f(h−1(xʹ ))). Klumpenhouwer networks K1 and K2
If x =​ h−1(xʹ ) and y =​ f(x), then both x and y are
elements of S, and the equation to be proved
reduces to f ʹ(xʹ ) =​ h(y). You may find it helpful them into networks called K1 and K2, one net-
to draw a diagram showing the elements a, x, work for each chord. T3 and T6 arrows join notes
and y in S and aʹ, xʹ, and f ʹ(xʹ) in Sʹ, along with within the ic3 and ic6 dyads as in (b). The T5 and
the functions relating them. T7 arrows from α to β have been abandoned, but
Use simple transitivity to show that f =​ gx−1 ● gy, I arrows have been added, connecting notes of
and therefore f ʹ =​ φ(f) =​(φ(gx))−1 ● φ(gy) =​ the two dyads within each chord. The arrow con-
(gxʹ )−1 ● gyʹ, and so f ʹ(xʹ ) =​ gyʹ((gxʹ )−1(xʹ )). From figurations and labels in the two networks match
h(x) =​ gxʹ(aʹ ), deduce that (gxʹ )−1(xʹ ) =​ aʹ, and exactly: as transformation networks, K1 and K2
conclude
   that f ʹ(xʹ ) =​ gyʹ(aʹ ) =​ h(y). are strongly isographic.
K1 and K2 are Klumpenhouwer networks, or
9.4 KLUMPENHOUWER K-​nets: networks of pitch classes whose trans-
NETWORKS formations are transpositions and inversions
from the group 𝒯ℐ12. First studied by Henry
Chords α and β in Figure 9.4.1a are all-​interval Klumpenhouwer, K-​nets were introduced into
tetrachords, both with interval-​ class vector the music theory literature in Lewin 1990 and
111111. The chords are Z-​related, not related by enjoyed a modest explosion of analytical atten-
any Tn or In operation. The figure shows, how- tion for some years thereafter. A large part of the
ever, that each chord may be partitioned into appeal of K-​nets lies in their ability to model (via
two dyads, one forming interval class 3 and the network isography) relationships between pc
other forming interval class 6. The network in sets belonging to different set classes, as in Figure
part (b) of the figure, a transformation network 9.4.1; K-​nets have most often been employed in
on the space pc, shows transpositions T3 and T6 analysis of atonal passages lacking set-​class con-
defining these dyads as well as transpositions T5 sistency. This introductory example also perhaps
and T7 relating the same-​size dyads in the two raises a question about the appropriateness of
chords. the K-​net approach: which is a more intuitive
The network in (b) is disconnected: no arrows description of the relationship between chords
connect notes of the ic3 dyads to notes of the α and β, the K-​net isography of Figure 9.4.1c or
ic6 dyads. Part (c) shows a different set of con- the paired transpositions of 9.4.1b? In the years
nections among the same pitch classes, splitting after K-​nets were introduced, such skepticism

Graphs and Networks; Serial Transformations • 357

was occasionally voiced, most pointedly in are preserved when the inversionally related
Buchler 2007, and K-​net applications have more elements are transposed symmetrically in this
recently receded in prominence. Neither all the way. If chord β is replaced with its transposition
uses of K-​nets nor all of the attendant contro- T1(β), the paired transpositions relating the T-​
versy can be described here, nor will we present sets in α and T1(β) become T6 and T8, no longer
any substantive K-​net analyses, but this section equal and opposite; the K-​net for T1(β), shown as
will summarize some of the most important con- K3 at the right side of Figure 9.4.2, consequently
siderations, including a few theoretical points requires different inversions, I9 and I6 in place of
that have generally been overlooked. the original I7 and I4.
The partitioning of the pitch classes into dyads Though K3 is not strongly isographic to K1
in K1 and K2 illustrates a feature that deserves and K2, it is nevertheless isographic to them.
closer scrutiny. We say that network K1 implies As defined in Section 9.3, network isography
a T-​partition of the tetrachord {D♯, E, G, A} into requires an isomorphism of transformation
dyads {D♯, A} and {E, G}. A T-​partition, when one groups, in this case an automorphism of 𝒯ℐ12.
exists, divides the vertices of a network into two The transformations appearing in K1 (or K2) are
parts that we will call T-​sets. Within each T-​set, related to those in K3 by the mapping φ2: 𝒯ℐ12
the vertices are joined only by T arrows—​in K1 → 𝒯ℐ12 defined by φ2(Tn) =​ Tn, φ2(In) =​ In+​2, and
each T-​set is a dyad with a single T arrow—​so this function φ2 is an automorphism of 𝒯ℐ12. In
each T-​set by itself forms a smaller network of fact, the mapping φk defined by φk(Tn) =​ Tn, φk(In)
the kind we have called a T-​net. Notes in oppos- =​ In+​k defines an automorphism of 𝒯ℐ12 for every
ing T-​sets in the same network are joined only mod-​12 integer k (analogous automorphisms for
by I arrows. Any T arrow may theoretically the isomorphic group ℛ were noted in Exercise
be reversed, the transposition replaced by its 8.2.7). K-​nets whose transformations are related
inverse; I arrows, representing involutions, are by an automorphism φk—​K-​nets, that is, whose
bidirectional. The network K1 does not include corresponding transpositions always match,
I arrows joining all possible pairs of notes from and whose corresponding inversions differ by a
the two T-​sets, but the other I relationships may constant number k—​are said to be positively iso-
be deduced from the ones that are present; for graphic. Strong isography is a special case of posi-
example, D♯ and G are related by I7T3 =​ T6I4 =​ I10. tive isography, the case in which k =​0.
(Both products give the same result because the The pitch classes in K3 were obtained as T1
network is path-​consistent; consistency proper- transpositions of those in K2, and this T1 map-
ties of K-​nets will be discussed further below.) ping illustrates a K-​net isomorphism between K2
The paired transpositions T5 and T7 in Figure and K3. Generally if all pitch classes of a K-​net
9.4.1b are now seen to be transpositions of the are transposed by Tk, each inversion In appear-
T-​sets of the networks K1 and K2. The two net- ing in the original K-​net must become In+2k in
works, as we have seen, are strongly isographic. the transposed K-​ net; the transpositions are
Equivalence classes of K-​ nets under strong unchanged in the new K-​net, so the transfor-
isography may be identified with K-​graphs, the mations are related by φ2k and the old and new
underlying transformation graphs of K-​ nets. K-​nets are positively isographic. But Tk is also a
The fact that that the same inversions I7 and I4 network isomorphism in this case. Exercise 9.4.3
appear in both K1 and K2 may be traced to the below confirms that the group automorphism
fact that the paired transpositions T5 and T7 are φ2k arises via conjugation by the pitch-​ class
equal and opposite: inversional relationships mapping Tk, as we know from Section 9.3 that
it must. Of the three networks K1–​K3, only K2
and K3 are isomorphic. The pitch classes of K1 are
not related to those of either K2 or K3 by trans-
position, and the relationship of K1 to either of
the other two networks is one of isography, not
isomorphism.

Exercise 9.4.3 If h =​ Tk, verify that Conjh =​ φ2k;

that
   is, Conjh(Tn) =​ Tn and Conjh(In) =​ In+​2k.

FIGURE 9.4.2 K-​net K3 (right), positively iso- Positive isography is not the only kind of K-​
graphic to K1 and K2 and isomorphic to K2 net isography. Because the T-​sets in the pc sets

358 • E x ploring M usi c al S pa c es

 Positive and negative isography together do
not exhaust the possible varieties of K-​net isog-
raphy. The automorphisms of 𝒯ℐ12 are 48 in
number; they may be enumerated in the form
ωj,k, where j is one of the mod-​12 units (1, 5, 7, or
11) and k is any mod-​12 integer.33 The automor-
phism ωj,k acts on 𝒯ℐ12 by the rules ωj,k(Tn) =​ Tjn,
ωj,k(In) =​ Ijn+​k. Indices of transposition, that is,
FIGURE 9.4.4 K-​net K4 (right), negatively iso- are transformed by the multiplication operator
graphic to K1 and K2 Mj (that is, n → jn), while indices of inversion are
transformed by the affine mapping n → jn +​ k. The
twelve automorphisms of the form ω1,k are the
we have been considering are simply dyads, the same as the ones we have called φk, giving rise to
T-​sets in α can be mapped to those in T1(β) not positive isography of K-​nets; strong isography is
only by transposition but also by inversion. The the special case ω1,0. The twelve automorphisms
K-​net K4, shown in Figure 9.4.4 along with K1 and ω11,k are the same as ψk, giving rise to negative
K2, illustrates. We may imagine that the ic3 dyad isography. Of the examples discussed above, K1
E–​G from K1 has been transformed by I5 to C♯–​B♭, and K3 are isographic via ω1,2, while K1 and K4
and the ic6 dyad D♯–​A by I8 to F–​B. This opera- are isographic via ω11,1. The automorphisms ωj,k
tion affects not only the K-​net’s inversion opera- with j =​5 or j =​7 have been invoked in analytical
tors but its transpositions as well: each Tn in K1 practice much less frequently than positive and
has been replaced in K4 by T−n, so the original T6 negative isography; they correspond to trans-
is unchanged but the original T3 has become T9. forming T-​sets by the multiplication operators
Each inversion operator In, meanwhile, has been M5 and M7 rather than by transposition or inver-
replaced by I1−n. In fact, the mapping ψk defined sion. In some cases, networks produced by these
by ψk(Tn) =​ T−n, ψk(In) =​ Ik−n defines an automor- isographies may duplicate networks produced by
phism of 𝒯ℐ12 for every k. Two K-​nets whose simpler positive or negative isographies, as the
transformations are related by one of these following exercise shows.
automorphisms ψk—​K-​nets whose correspond-
ing transposition indices always sum to 0, and Exercise 9.4.6 Construct a K-​net isographic to
whose corresponding inversion indices sum to K1 via the automorphism ω5,10. Put the note B♭
a constant number k—​are called negatively iso- in your network in the vertex occupied by E in
graphic. Networks K1 and K4 are negatively iso- . This K-​net should look familiar. What is it?
K1  
graphic via ψ1.
The following exercise examines the signifi- Exercise 9.4.7
cance of the index numbers of the mappings φk (a) Show that K3 and K4 are isographic via ω11,3.
and ψk in positive and negative isography. (b) The chain of isographies linking K1, K3, and
K4 suggests that ω1,2 ● ω11,3 =​ ω11,1. Verify
Exercise 9.4.5 that this equation is true.
(a) The isography between K1 and K3 is a positive (c)  Derive a general formula for ωj,k ● ωu,v.
isography φk with k =​2. The number 2 is also
the mod-​12 sum of the index numbers of the As is evident from the above examples, the
paired transpositions T6 and T8 relating the possibilities for K-​net isography are diverse, and
T-​sets in these networks. Explain why this is two pc sets may often be modeled by isographic
the case. That is, show that if pitch classes x K-​nets in more than one way. This “promiscuity”
and y are related by In, then Ti(x) and Tj(y) are of K-​net relations is another of the grounds on
related by In+​k, where k =​ i +​ j. which K-​net analysis has been criticized (Buchler
(b) The isography between K1 and K4 is a negative 2007). The possibilities expand further if differ-
isography ψk with k =​1. The number 1 is also ent basic configurations of T and I arrows are
the mod-​12 sum of the index numbers of the considered. K-​nets K5 and K6 in Figure 9.4.8
paired inversions I5 and I8 relating the T-​sets partition tetrachords into T-​sets of cardinali-
in these networks. Explain why this is the ties 3 +​1 rather than 2 +​2; that is, each of these
case. That is, show that if pitch classes x and y networks shows three pitch classes related by
are related by In, then Ii(x) and Ij(y) are related transposition, with the remaining pitch class
  by Ik−n, where k =​ i +​ j. related to each of those three by inversion. These

Graphs and Networks; Serial Transformations • 359

 K-​nets in analytical applications are connected,
with every vertex linked to every other by some
chain of T and/​or I arrows. (Some T arrows may
need to be traversed backward in such a chain;
I arrows are almost always understood to be
bidirectional.) In principle a T-​net with only T
arrows, like networks N6–​N9 from Figure 9.3.8,
is a special kind of K-​net, but sometimes it is
assumed that a proper K-​net must contain both
T and I arrows. Duplication of one pitch class
FIGURE 9.4.8 Negatively isographic K-​nets K5 in two different vertices, though theoretically
and K6 possible in K-​nets, is rarely encountered.35
Consistency properties of K-​ nets, though
rarely discussed, deserve consideration.36 Most
K-​nets model the same pc sets α and T1(β) as in writers on K-​nets have implicitly adopted Lewin’s
previous examples, and K5 and K6 are negatively definition of transformation networks and
isographic via ω11,10. This isography is possible thereby assumed path consistency as a require-
because both tetrachords contain trichordal ment—​but many useful features of K-​nets, and
subsets of prime form 026; the isography must properties often assumed to be true of them
be negative because the two trichords are inver- generally, depend not only on path consistency
sionally related. but on strong path consistency. (Realizability,
of course, is not a concern for networks, which
Exercise 9.4.9 always realize their underlying graphs.) Non–​
(a) Construct a series of K-​nets like K5 but path-​consistent K-​nets like Figure 9.4.10a have
with the pitch class E replaced by another not appeared in the literature—​but neither have
pitch class chosen from the eight pcs not examples like 9.4.10b, which is trivially path-​
appearing in this network. Leave the other consistent (for the same reasons as Figure 9.2.5c
three pcs and the transpositions unchanged, discussed previously: it has no paired directed
adjusting the inversion operators as needed. paths) but not strongly path-​consistent. Neither
Explain why all K-​nets obtained in this way network in Figure 9.4.10 has a proper T-​parti-
are positively isographic to K5. How many tion, a point to which we shall return shortly.
different tetrachordal set classes can be We know from Section 9.2 that every network
modeled by these isographic K-​nets? all of whose transformations belong to a simply
(b) Construct three different K-​nets of a 2 +​1 transitive group is strongly path-​consistent. But
type for a trichord of prime form 026 such the action of 𝒯ℐ12 on pc is not simply transitive;
as CDF♯. Each network should be a triangle as defined in Exercise 5.6.5, it is doubly transitive.
formed by one T arrow and two I arrows. That is, for any two pitch classes x and y, exactly
The T arrows in the three networks should two transformations in 𝒯ℐ12 map x to y: the
be labeled T2, T4, and T6. transposition Ty−x and the inversion Ix+​y. It is this
(c) Explain why every three-​note pitch-​class set duplication of transformational action that gives
can be modeled by a K-​net that is positively K-​nets the flexibility to model one pitch-​class set
isographic to one of the three networks you in so many different ways—​but it is also what
  drew in (b).34 allows path consistency to fail as in the K-​nets of
Figure 9.4.10.
A number of tacit conventions have perme- The following exercise shows that strong path
ated the K-​net literature: K-​nets appearing in consistency for K-​nets is easily recognized.
analytical applications have overwhelmingly
shared certain characteristics not formally Exercise 9.4.11 Explain why a K-​net is strongly
required by any published definition of K-​nets. path-​consistent if and only if every cycle in
The K-​ net configurations we have encoun- the network (disregarding arrow directions)
tered—​the 2 +​2 (“box”) and 3 +​1 (“umbrella”) contains
   an even number of I arrows.37
arrangements for tetrachords, and the 2 +​1
trichordal type in Exercise 9.4.9b—​ are the The two K-​nets of Figure 9.4.10, each con-
ones most frequently used. Significantly larger sisting of a cycle with a single inversion, fail the
K-​nets have appeared only rarely. Almost all criterion provided by the above exercise. These

360 • E x ploring M usi c al S pa c es

 fact, strong path consistency makes such a con-
tradiction impossible: taken together the two
paths would form a cycle with an odd number
of I arrows, which according to Exercise 9.4.11
cannot happen as long as the K-​net is strongly
path-​consistent.

Exercise 9.4.13
FIGURE 9.4.10 (a) A K-​net that is not path-​con- (a) Show how the above algorithm can be used
sistent; (b) a K-​net that is path-​consistent but not to identify the T-​partitions of the K-​nets K1
strongly path-​consistent and K5.
(b) Show that the coloring algorithm fails for
both networks in Figure 9.4.10, which are
K-​nets also lack the characteristic feature that   not strongly path-​consistent.
motivated our introduction of K-​nets: neither of
these networks admits a T-​partition. In each of In Figure 9.4.12, the T-​partition consists
these networks, all the vertices are linked by T of a black T-​set containing five vertices and a
arrows; when the single I arrow is included, each white T-​set with four. The white T-​set is T-​con-
network implies both transpositional and inver- nected, meaning that it is connected as a T-​net:
sional relationships between any two vertices. all the white vertices are linked by T arrows. The
Every K-​ net that is both connected and black T-​set is not T-​connected; we cannot navi-
strongly path-​consistent, however, implies a T-​ gate from the upper three black vertices to the
partition. Suppose we are given such a K-​net; lower two without traversing some I arrows. The
we will describe how a T-​partition may be con- network implies transpositional relationships
structed, identifying the two T-​sets by coloring among all the white vertices, but the transpo-
vertices black and white, as shown schematically sition that relates an upper black vertex to a
in Figure 9.4.12. All we need to know are the lower black vertex cannot be deduced from the
locations of T and I arrows; the numerical indi- T arrows alone.
ces are irrelevant, as are the directions of the T We call a K-​net T-​complete if both of its T-​
arrows, so the figure omits these details. Start sets are T-​ connected. Of the strongly path-​
by choosing any vertex v0—​perhaps the leftmost consistent K-​ nets presented so far, most are
vertex in the figure—​and coloring it black. Now T-​complete, but a K-​net with a configuration like
color every other vertex v by the following rule: Figure 9.4.12 is not. A T-​complete K-​net’s poten-
if a path from v0 to v contains an even number tial for isography, it turns out, can be deduced
of I arrows, color v black, but if a path from v0 from its T arrows alone. In K1 from Figure 9.4.1c,
to v contains an odd number of I arrows, color for example, the T arrows are T3 and T6. Suppose
v white. Because the network is assumed to be Kʹ is any K-​net of the 2 +​2 type with the same
connected, this process will assign a color to configuration of T and I arrows as K1, and sup-
every vertex. It is reasonable to ask if a contra- pose the T arrows of Kʹ are also T3 and T6. Then
diction in coloring might arise: what if there are without even knowing the index numbers on the
two paths from v0 to v, one with an even num- I arrows of either network we can assert that Kʹ
ber of I arrows and one with an odd number? In is positively isographic to K1. If instead the T
arrows of Kʹ are T9 and T6, then we know Kʹ is
negatively isographic to K1.
To see why this property holds, suppose K is a
connected, strongly path-​consistent, and T-​com-
plete K-​net with T-​sets S1 and S2 (imagine black
and white vertices). Suppose Kʹ is another K-​net
with the same configuration of T and I arrows as
K, and with T-​sets S1ʹ and S2ʹ corresponding to
S1 and S2, and suppose the transpositions in Kʹ
match those of K exactly. Let x be the pitch class
at any vertex in S1, let xʹ be the pitch class at the
FIGURE 9.4.12 Coloring the vertices of a con- corresponding vertex in S1ʹ, and let i =​int(x, xʹ ),
nected, strongly path-​consistent K-​net so that x ʹ =​ Ti(x). Because S1 is T-​connected, every

Graphs and Networks; Serial Transformations • 361

other vertex in S1 is linked to x by some combi- recursion is that positive and negative isography
nation of T arrows, and all the vertices of S1ʹ are of K-​nets resemble transposition and inversion,
linked to xʹ by corresponding T arrows, identi- so networks of isographic K-​nets can resemble
cally labeled. Consequently the (black) pitch K-​nets themselves. This intuition is supported
classes in S1ʹ must all be Ti-​transpositions of the by the fact that the 24 automorphisms φk and
corresponding (black) pitch classes in S1. By the ψk associated with positive and negative isogra-
same reasoning, the (white) pitch classes of S2ʹ phy form a dihedral group isomorphic to 𝒯ℐ12.
must be some transposition Tj of those of S2. This isomorphism is traditionally highlighted by
(This deduction is not possible in the absence of the suggestive notation 〈Tk〉 (“hyper-​Tk”) for the
T-​completeness; see Exercise 9.4.14 below.) Ti relationship of positive isography via φk, and 〈Ik〉
and Tj are precisely the paired transpositions of (“hyper-​Ik”) for negative isography via ψk.38 In
T-​sets that motivated our first example of K-​net this notation, our discussion earlier in this sec-
isography. tion shows that K1 and K3 are related by 〈T2〉, K3
The index numbers of the inversions in Kʹ and K4 by 〈I3〉, and K1 and K4 by 〈I1〉.
may not match those in K exactly, but if an inver- The hyper-​transformations 〈Tk〉 and 〈Ik〉 are
sion In occurs between pitch classes x (in S1) and associated with the mappings φk and ψk, but they
y (in S2) in the network K, the corresponding are not the same objects. While φk and ψk are
inversion in Kʹ must map Ti(x) to Tj(y), and must automorphisms of the group 𝒯ℐ12, 〈Tk〉 and 〈Ik〉
therefore be Ti−1InTj, which is In+i+​j. All the indices are commonly described as relations between
of inversion in Kʹ therefore differ from those in K-​nets. Technically 〈Tk〉 and 〈Ik〉 are not well-​
K by the constant value k =​ i +​ j, and the two net- defined transformations on K-​nets, because a
works are positively isographic via φk. If instead given K-​net can be related by a particular 〈Tk〉 or
the transpositions in Kʹ are the inverses of those 〈Ik〉 to more than one other K-​net. For example,
in K, similar reasoning shows that the T-​sets in the two strongly isographic K-​nets K1 and K2 are
Kʹ are related by inversion, say Ii and Ij, to those related by 〈T0〉 not only to themselves but also to
in K, and the two networks are negatively iso- each other, and K3 is related to both of them by
graphic via ψk, where k =​ i +​ j. 〈T10〉. Formally, 〈Tk〉 and 〈Ik〉 may be understood
Although the possibility of deducing isogra- as transformations defined on K-​graphs (strong
phy from T arrows alone has sometimes been isography classes of K-​nets); we may write cor-
asserted or assumed for K-​nets generally, the rectly that 〈T10〉(gr(K3)) =​gr(K1) =​gr(K2).
following exercise shows that this property does Translated into the notation of hyper-​trans-
not hold for networks that are not T-​complete. formations, the formula from Exercise 9.4.7b
describing the chain of isographies from K1 to K3
Exercise 9.4.14 Form two K-​nets from the to K4 becomes 〈T2〉 ● 〈I3〉 =​ 〈I1〉, apparently obey-
graph in Figure 9.4.12, filling in pitch classes ing the same rule of composition Tm ● In =​ In−m
and specific transpositions and inversions in satisfied by ordinary transpositions and inver-
two different ways, so that the T labels in the sions according to Theorem 5.5.6c. The following
two networks match exactly but the index exercise confirms this observation, establishing
numbers of the inversions in the two K-​nets that the mapping Tk → 〈Tk〉, Ik → 〈Ik〉 defines
do not differ by a constant value. The K-​nets an isomorphism from 𝒯ℐ12 to the analogously
therefore
   cannot be isographic. notated group of 〈T〉 and 〈I〉 operations—​a group
we will call 〈𝒯ℐ12〉.
Any strongly path-​consistent K-​net that is
not T-​complete can be made T-​complete by add- Exercise 9.4.15 Use your formula from
ing one or more T arrows. In Figure 9.4.12 a sin- Exercise 9.4.7c to verify that all possible
gle additional T arrow, joining some vertex in the compositions of hyper-​T and hyper-​I
upper part of the black T-​set to some vertex in transformations satisfy equations analogous
the lower part, suffices to make the network T-​ to formulas from Theorem 5.5.6 (〈Tm〉 ● 〈In〉 =​
complete. The new T arrow effectively provides 〈In−m
   〉 and three others).
the missing information about the network’s
potential isographies. Figure 9.4.16 illustrates K-​ net recursion.
We conclude this section with a short dis- Networks K7–​K10 in (a) model four different all-​
cussion of the use of K-​nets to build networks interval tetrachords.39 All of these K-​nets are iso-
of networks, a technique commonly described graphic to each other and to the networks K1–​K4
as recursion. The intuition underlying K-​ net presented previously; in fact K8 is identical to K1.

362 • E x ploring M usi c al S pa c es

Part (b) of the figure is a network of networks, might question whether the use of the word
showing 〈T〉 and 〈I〉 relations among the K-​nets recursion is strictly appropriate.) Isography
K7–​K10 (or more properly among their graphs). across levels imposes artificial limitations on
Strikingly, the pattern of 〈T〉 and 〈I〉 relations the number of objects that can be analyzed:
in (b) exactly duplicates the pattern of T and I since the networks K7–​K10 in Figure 9.4.16a
relations in the network K8. It is tempting to say depict tetrachords, the isography between K8
that the hypernetwork is strongly isographic to and the hypernetwork in (b) is possible only
K8, but that is not strictly accurate, as the trans- because we chose to construct a hypernetwork
formations in the two networks belong to two of four chords. A given chord may be modeled
different (though isomorphic) groups: the group by a K-​net in numerous ways; an analyst can
𝒯ℐ12 acting on pitch classes and the group 〈𝒯ℐ12〉 forage for isography by rearranging the pitch
acting on K-​graphs. The relation is certainly an classes within each of the several K-​nets of a
isography, however, and since K8 is already iso- hypernetwork in various ways to see what
graphic to the other three K-​nets in (a), it follows works—​and the arrangement that works fre-
that the hypernetwork is isographic to all four quently bears no relationship to the registral
of them. positions of notes within chords or to the voice
K-​net recursion has garnered considerable leading between chords.
attention, proving both tantalizing and con- A more technical objection is that in order
troversial. Lewin 1994 extends recursion to a to identify isomorphic mappings between 𝒯ℐ12
third level of structure—​networks of networks and 〈𝒯ℐ12〉, we must disregard half of all pos-
of networks. Higher-​level networks are said sible K-​net isographies, those obtained via the
to “prolong” lower-​level ones, and recursion is mappings ωj,k with j =​5 and j =​7.40 Moreover,
seen as evidence of large-​scale structural unity as noted above, the correspondence Tk → 〈Tk〉, Ik
comparable with what we may find in the lev- → 〈Ik〉 is not truly a strong isography, so it is per-
els of a Schenkerian analysis. But a variety of haps misleading to suppose that the hypernet-
reservations have been expressed about K-​net work of Figure 9.4.16b is somehow more like K8
recursion, starting with the nature of the levels than like any of the other K-​nets in 9.4.16a with
themselves: in a Schenkerian analysis each level which it is also isographic. It is also reasonable to
is a subset of the previous one, with objects of question what an assertion of structural likeness
the same kinds (pitches) appearing at all levels, between networks at different levels is supposed
whereas in K-​net recursion the objects at the to mean, musically or phenomenologically, or
lowest level are pitch classes, those at the sec- whether it is something one can hear. These
ond level are K-​nets (or K-​graphs), and those points are reinforced by the following exercise,
at the third level are something still more which shows that a correspondence of index
abstract. (For this reason a mathematician numbers between networks and hypernetworks

FIGURE 9.4.16 (a) K-​nets K7–​K10; (b) a hypernetwork illustrating K-​net recursion

Graphs and Networks; Serial Transformations • 363

may vanish if the networks are transposed by a this class simply Orb(A). Thus Orb(A) =​Orb(B) if
constant interval.41 and only if A and B are “forms of the same row,”
related by some serial transformation. One orbit
Exercise 9.4.17 Construct four networks is only a small part of the full space row, but
T1(K7)–​T1(K10) by transposing all pitch classes Orb(A) may nevertheless be regarded as a space
in K7–​K10 by T1 and making the necessary in its own right. We encountered a few depic-
adjustments to the inversions. Construct a tions of this space in Section 4.1; Figure 4.1.12b,
hypernetwork of the four transposed networks. in particular, is a Cayley diagram for 𝒮𝑒𝑟12 acting
Do the index numbers of the hyper-​inversions on such a space.
match the index numbers of the inversions in Almost all orbits are sets of 48 rows on
any
  of the transposed networks? which the action of 𝒮𝑒𝑟12 is simply transitive.
Exceptions to this generalization occur in the
9.5 SERIAL cases of symmetrical rows, to be considered in a
TRANSFORMATIONS moment. The 48 forms of an asymmetrical row
comprise four T-​classes of twelve row forms each,
AND UTTs related by transposition. The T-​classes are related
Some of the first musical applications of group by transformations we may call E (the identity),
theory were in the realm of twelve-​tone struc- I, R, and RI. For example, if two row forms are
ture, notably in the writings of Milton Babbitt related by some Tn they are in the same T-​class,
([1960] 2003, [1961] 2003). This work also which is to say their T-​classes are E-​related; if
played an important role in shaping the devel- two row forms are related by some RTn their T-​
opment of Lewin’s transformational theory. classes are related by R, and so on. The group {E,
The final two sections of this chapter investi- I, R, RI} acting on the four T-​classes, which is the
gate some aspects of serial structure in relation quotient group 𝒮𝑒𝑟12 ∕ 𝒯12 of the serial group by
to other concepts we have explored, including the subgroup of transpositions, is isomorphic
UTTs, transformation networks, and symmetry. to the direct product {E, I} × {E, R} or 𝒞2 × 𝒞2, a
Some preliminary remarks on notation are in Klein group.
order. We will write twelve-​tone rows in the form Symmetrical rows are invariant under some
A =​(a0, a1, …, a11). Letters denoting rows will non-​identity transformation in 𝒮𝑒𝑟12. The reader
generally be boldfaced. The twelve pitch classes may recall the construction of some symmetrical
a0, …, a11 are indexed by order positions 0–​11; rows in Exercise 4.1.13 as well as some related
thus the first note of a row occupies order posi- discussion in Section 6.5. No Tn or In operation
tion 0. Using order numbers 0–​11 rather than other than T0 fixes all twelve pitch classes, so the
1–​12 ensures that the same mod-​12 integers only possible nontrivial symmetries of a twelve-​
that denote pitch classes also denote order posi- tone row in the group 𝒮𝑒𝑟12 are operations of the
tions, a correspondence that will be essential in form RTn and RIn. No row can be invariant under
Section 9.6. In the remainder of this chapter R more than one such operation, so relative to the
always denotes the retrograde operation acting group 𝒮𝑒𝑟12, the degree of symmetry of a row (as
on ordered sets, not the neo-​Riemannian R on defined in Section 6.5) can be no larger than 2.
triads. By the orbit-​ stabilizer theorem, symmetrical
Chapter 6 introduced the serial group 𝒮𝑒𝑟12, a rows always have exactly 24 distinct forms, half
group of order 48 generated by T1, I, and R, act- of the usual 48, and two distinct T-​classes rather
ing on the space row of all 12! twelve-​tone rows. than four.
Every transformation in 𝒮𝑒𝑟12 takes one of the A symmetry of either the R or the RI type
forms Tn, In, RTn, or RIn for some mod-​12 integer imposes constraints on the relationship between
n. The last two of these expressions could equally notes in symmetrically situated order positions. If
well be written TnR and InR, as R commutes with RIn(A) =​ A =​(a0, a1, …, a11), then ak +​ a11−k =​ n for
all transpositions and inversions—​ a property every order position k, whereas if RTn(A) =​ A, then
that also ensures that 𝒮𝑒𝑟12 is isomorphic to a a11−k =​ ak +​ n for every k; the latter condition is
direct-​product group 𝒯ℐ12 × {T0, R}, or 𝒟12 × 𝒞2. possible only for n =​6 (see Exercise 9.5.2 below).
If A is a row, then Orb𝒮𝑒𝑟12(A), the orbit of A The severity of these restrictions makes sym-
under the action of 𝒮𝑒𝑟12, is a row class contain- metry statistically rare among twelve-​tone rows:
ing all the T-​, I-​, and R-​related rows traditionally fewer than one-​tenth of one percent of all rows
considered “forms of A.” When no confusion possess either type of symmetry. When all twelve-​
with orbits of other groups is likely, we denote tone rows are considered, the average degree of

364 • E x ploring M usi c al S pa c es

symmetry is about 1.0007, barely larger than 1, necessarily taken to be an inverted form that
the degree of symmetry for an asymmetrical row. starts on the same note as P0; then Pn and In are
The average number of rows per row class is about Tn(P0) and Tn(I0) respectively. In recent decades
47.97—​barely smaller than 48, the number for fixed-​0 row labeling, in which Pn and In always
an asymmetrical row class—​and the number of denote row forms starting with pitch class n,
row classes, 9,985,920, is only slightly larger than has gained favor. The equations Rn =​ R(Pn) and
1/​48 of 12!, the total number of twelve-​tone rows. RIn =​ R(In) hold in all traditional labelings.
If composers used symmetrical rows in propor- (Observe the orthographic distinction between
tion to their frequency in the universe of all possi- boldface R and I and italic R and I: the first two
ble rows, we would rarely if ever see a symmetrical occur as part of row labels, while the latter two
row. In fact, while the proportion of twelve-​tone are transformations.)
compositions based upon symmetrical rows is While either a fixed-​0 or a movable-​0 system
fairly small, it is much larger than random selec- provides workable labels for all 48 row forms,
tion would dictate; nearly 20 percent of Webern’s the fixed-​0 system enjoys two modest advan-
twelve-​tone rows are symmetric. The exercise tages. First, it reduces the number of arbitrary
below includes a few examples.42 decisions to be made when labeling rows—​
though one must still decide somehow which of
Exercise 9.5.1 Six twelve-​tone rows are given the four T-​classes is to be called P. Perhaps more
below as strings of pitch-​class numbers. Some important, fixed-​0 labeling ensures that the con-
but not all of these rows are symmetrical. venient equation In =​ In(P0) always holds. In a
Identify the symmetrical rows, and for each of fixed-​0 system, one can calculate the row label
them determine the operation RT6 or RIn under that results when any transformation in 𝒮𝑒𝑟12
which the row is invariant.43 is applied to any row form, using nothing but
the row labels and transformations involved.
(a) Schoenberg, Serenade, Op. 24: In every fixed-​0 system, for example, RI7(I2) =​
(9, 10, 0, 3, 4, 6, 5, 7, 8, 11, 1, 2) R(I7(I2(P0))) =​ R(T5(P0)) =​ R(P5) =​ R5. (In left-​to-​
(b) Webern, Symphony, Op. 21: right orthography, the composition of the inver-
(9, 6, 7, 8, 4, 5, 11, 10, 2, 1, 0, 3) sions in this equation is I2 ● I7, which is T5 by
(c) Webern, Concerto for Nine Instruments, Theorem 5.5.6d.) In a movable-​0 system, RI7(I2)
Op. 24: (11, 10, 2, 3, 7, 6, 8, 4, 5, 0, 1, 9) is always some R form, but not necessarily R5.
(d) Berg, Lyric Suite: Relationships between fixed-​and movable-​ 0
(5, 4, 0, 9, 7, 2, 8, 1, 3, 6, 10, 11) coordinate systems, and among various row
(e) Lutosławski, Musique funèbre: labelings more generally, will be studied further
(5, 11, 10, 4, 3, 9, 8, 2, 1, 7, 6, 0) later in this section.
(f) Nono, Il canto sospeso:
  (9, 10, 8, 11, 7, 0, 6, 1, 5, 2, 4, 3) Exercise 9.5.3
(a) Does the equation RIn =​ In(R0) always hold
Exercise 9.5.2 If RTn(A) =​ A, explain why n in a fixed-​0 system? In a movable-​0 system?
must be 6, and also explain why the central (b) In a fixed-​0 system, is it true that
interval of A (between a5 and a6) must be a RIn(RIn) =​ P0?
tritone.
   (c) In a movable-​0 system in which both P0 and
I0 are row forms starting with F♯, does the
The traditional assignment of labels Pn, In,   equation In =​ In(P0) hold?44
Rn, and RIn to the forms of a row depends on
choosing two inversionally related forms to serve Exercise 9.5.4 In a movable-​0 system in which
as P0 and I0. There are many ways to do this; we the first pitch class of P0 is i and the first pitch
shall refer to a particular labeling of the rows in class of I0 is j, explain why In(P0) =​ Ik, where
a class Orb(A) as a coordinate system for Orb(A). k =​ n − i − j. In the same situation, what is
Conventionally the first row form in a piece, or In(P
  m)? What is In(Im)?
some form that appears prominently near the
beginning, is chosen as a P form, but there is Even though a row class Orb(A) is defined via
not always a single obvious choice for such a row. the transformations in 𝒮𝑒𝑟12, those transforma-
Historically, many twelve-​tone analysts labeled tions are not the only ways to model relations
this first row P0 regardless of its starting note. between rows in Orb(A). While Figure 4.1.12b
In such a movable-​0 system, I0 is usually but not was a Cayley diagram for 𝒮𝑒𝑟12, the alternative

Graphs and Networks; Serial Transformations • 365

picture in 4.1.12a was a Cayley diagram for a dif-
ferent group 𝒮𝑒𝑟12ʹ acting on the same row forms,
a group generated by T1, R, and the contextual
inversion operator J that inverts any row about
its first note. The group 𝒮𝑒𝑟12ʹ offers an alternate
way of conceiving of row relationships; as noted
in Exercise 6.2.12, 𝒮𝑒𝑟12ʹ may be considered a
commutative counterpart to 𝒮𝑒𝑟12, and its action
on any asymmetrical row class is also simply
transitive.
The central observation of this section is that
serial transformations can also be modeled by
UTTs and related transformations.45 As a first
illustration we examine the celebrated mirror
canon that constitutes the second movement of
Webern’s Variations for Piano, Op. 27. The move-
ment opens with two row forms we call D and C,
for dux and comes:

First row in dux:

D =​(10, 9, 1, 11, 2, 0, 6, 5, 4, 8, 7, 3) FIGURE 9.5.5 Row transformation networks for
First row in comes: Webern, Variations for Piano, Op. 27, II: (a) with
C =​(8, 9, 5, 7, 4, 6, 0, 1, 2, 10, 11, 3) transpositions and inversions; (b) with schritts
and wechsels; (c) with schritts and wechsels, in
Rows D and C are related by I6: the canon’s two bubble notation
voices famously mirror each other about A4, and
I6 is the inversion that relates pitch class A to about A4 persists throughout, so the I6 rela-
itself. tionship established between the first pair of
Let us provisionally choose a movable-​0 coor- rows holds for every pair of simultaneous rows.
dinate system in which row D plays the role of The other six transformations in the network,
P0 and the inverted form beginning with the however, are all different. Because I6 does not
same pitch class (B♭) is I0. Row C is then T10(I0), commute with the other transformations, the
or I10. In this system, I10 is I6(P0), not I10(P0) as transformations in the comes, T7–​I11–​T9, differ
would be the case in every fixed-​0 system. While from those in the dux, T5–​I1–​T3. A close relation-
the names D and C, on the one hand, and P0 and ship between these two chains of transforma-
I10, on the other, refer to the same two rows, tions is nevertheless evident: corresponding
we interpret these names in conceptually differ- index numbers are mod-​12 inverses, with one
ent ways: D and C denote fixed strings of pitch transformation always acting on a P form as the
classes, while P0 and I10 are labels for those rows, other acts on an I form.
their coordinates in a particular coordinate sys- UTTs offer a way to generate both voices of
tem. Observing this distinction will be crucial as the canon using exactly the same transforma-
we compare coordinate systems below. tions—​ not transpositions and inversions but
The transformation network in Figure 9.5.5a schritts and wechsels. Let us write (n, +​), or
shows relationships among successive and more briefly n+​, for the row form Pn, and simi-
simultaneous row forms throughout the move- larly (n, −) or n− for In. We can then apply UTTs
ment, labeled in the system just described. The to the objects n+​ and n− exactly as in Chapter 8,
comes is placed above the dux in the network for where the ordered pairs (n, σ) represented triads.
easiest comparison with the score: though not Figure 9.5.5b redrafts the Op. 27 analysis in this
consistently higher in register, the comes is usu- fashion. Here a single schritt-​wechsel chain S5–​
ally (but not always) written on the upper staff W7–​S9 appears in both voices of the canon; the
and played by the right hand. The transforma- schritt S5 behaves like the transposition T5 when
tions in this network are dictated by the simply applied to a P form but like T7 when applied
transitive group 𝒮𝑒𝑟12, but because only P and I to an I form. This serial manifestation of the
row forms occur in the movement, all the trans- Riemannian dualism condition, more familiar
formations here belong to 𝒯ℐ12. The mirroring in its triadic form, is exactly what is needed to

366 • E x ploring M usi c al S pa c es

model the similarity between the canon’s two under consideration—​our row D, the first row
voices. in the dux. The table shows the relationship
The analysis in Figure 9.5.5b requires only between rows and labels in two different ways.
four different transformations altogether, com­ The top portion of each column identifies the
pared with seven in (a). The two voices are rows upon which the appropriate coordinate
always related by the fixed inversion I6, not by system bestows the labels P0 and I0, describing
any contextual inversion, so the vertical arrows them in terms of D and C: in system ①, P0 is D
in (b) retain the I6 label. Inversions are not UTTs, while I0 is T2(C). In general, a coordinate system
though they are QTTs as described in Section is completely determined by the choice of two
8.3; all transformations in the network belong inversionally related row forms to serve as P0
to the QTT group 𝒬. The network shows directly and I0, as all other row labels may be determined
that I6 commutes with each transformation in from these.47 The bottom part of the table shows
the schritt-​wechsel chain; for instance, the first the coordinates (row labels) that each system
square shows that S5I6 =​ I6S5. Part (c) of the fig- assigns to D and C themselves: in system ①, D is
ure recasts (b) in bubble notation, with dux and P0, or 0+​, while C is I10, or 10−.
comes bubbled separately and related by I6. If the Coordinate system ②, adopted by Catherine
bubbles are regarded as separate networks, the Nolan (1995, 50), is a fixed-​0 system adopting
two are strongly isomorphic: I6 is the isomor- a more global perspective. Nolan’s P0 is T4(C),
phism (the mapping called h in Section 9.3), and chosen because it starts with pitch class 0—​a
the isography is strong because the transforma- row that occurs nowhere in the movement.
tions in both bubbles are identical. Peter Westergaard (1963, 108) uses system ③, a
local, movable-​0 system different from system
Exercise 9.5.6 The equation I6(0+​) =​10−, ①. Westergaard takes the first pair of rows in
implied by Figure 9.5.5b, would not be correct the movement as his P0 and I0 notwithstanding
in the usual triadic interpretation of the their different starting notes: C, the first row of
symbols 0+​ and 10−, but it is correct in the the comes, is P0, while D is I0.48 Finally, system ④
present row interpretation. Verify that this is a movable-​0 system, but not one that is local-
equation is consistent with an appropriate ized for the second movement: its row labels are
formula
   from Exercise 9.5.4. derived from the form P0 stated at the beginning
of the third movement. This row, which happens
Exercise 9.5.7 As discussed in Section 9.3, to be the retrograde of C, is identified in Webern’s
bubble notation can be feasible even when the sketches as the original form of the row, and
bubble transformation does not commute with Bailey uses this coordinate system in her primary
all transformations within the bubbles. Redraw analyses of all three movements of Op. 27.
Figure 9.5.5a in this way, and verify that the The retrograde row forms in system ④ will
transformations in the two bubbles are related require some refinements to the n+​/​n− notation,
by Conjh, where h =​ I6. Are the two bubbles to be detailed below. A relationship between
isomorphic?
   Are they strongly isographic? systems ③ and ④, however, is easy to describe:
system ④’s P0 and I0 are the retrogrades of the
Exercise 9.5.8 Draw another network for the correspondingly labeled rows in system ③, and
second movement of Op. 27, using exclusively the same is therefore true for all row forms in
transformations from the group 𝒮𝑒𝑟12ʹ these two systems. Converting coordinates
generated
   by T1, R, and J.46 between system ③ and system ④, therefore, is
straightforward. We now extend this idea to
The many published analyses of Op. 27 are show more generally how UTTs and related
notable for the variety of row-​labeling conven- transformations may be used to convert from
tions employed by different authors. Table 9.5.9 one coordinate system to another.
summarizes four different coordinate systems For the moment we continue to assume that
that have appeared in the literature for this ordered pairs n+​ and n− represent row labels Pn
work. System ① in the table’s first column is and In respectively; the actual rows represented
the movable-​0 system described above, which by these labels depend on the choice of a coordi-
appears in what Kathryn Bailey (1991, 350) calls nate system. Comparing the row labels for coor-
her “transposed” analysis of this movement. It dinate systems ① and ② in the bottom portion of
may also be called a local analysis, in that it takes Table 9.5.9, we see that system ①’s 0+​ (row D) is
as its P0 a particular row form in the excerpt system ②’s 10−, and system ①’s 10− (C) is system

Graphs and Networks; Serial Transformations • 367

Table 9.5.9. Row coordinate systems for Webern, Op. 27, II

CO OR DI NAT E SY S T E M

① ② ③ ④
M O VA BL E -​0 F I X E D -​0 MOVA BL E-​0 MOVA BL E-​0
Bailey Nolan Westergaard Bailey
(transposed analysis) (primary analysis)

P0 P0 =​ D (10, 9, …, 7, 3) P0 =​ T4(C) (0, 1, …, 3, 7) P0 =​ C (8, 9, …, 11, 3) P0 =​ R(C) (3, 11, …, 9, 8)

I0 I0 =​ T2(C) (10, 11, …, 1, 5) I0 =​ T2(D) (0, 11, …, 9, 5) I0 =​ D   (10, 9, …, 7, 3) I0 =​ R(D) (3, 7, …, 9, 10)

D D =​ P0 =​0+​ D =​ I10 =​10− D =​ I0 =​0− D =​ RI0 =​0−−

C C =​ I10 =​10− C =​ P8 =​8+​ C =​ P0 =​0+​ C =​ R0 =​0+​−

D =​(10, 9, 1, 11, 2, 0, 6, 5, 4, 8, 7, 3)
   (first row in dux)
    C =​(8, 9, 5, 7, 4, 6, 0, 1, 2, 10, 11, 3) =​I6(D) (first row in comes)

FIGURE 9.5.10 Row coordinate transformations: (a) in Op. 27, II, first two rows of dux in coordinate
systems ① and ②; (b) general case, rows A and B in coordinate systems Φ and Ψ

②’s 8+​. More generally, system ①’s n+​ is always sys- transformation networks in Section 9.3, espe-
tem ②’s (n − 2)−, and system ①’s n− is system ②’s cially Figure 9.3.4; indeed, the coordinate trans-
(n − 2)+​. It follows that the UTT 〈−, 10, 10〉, formation h =​ hΦ–​Ψ is an isomorphism of the
applied to any row label in system ①, yields the two small networks within the bubbles. In the
label for the same row in system ②. We call this general setting of Figure 9.3.4 the two bubbles
UTT the coordinate transformation from system ① contain different objects; in 9.5.10b they con-
to system ② and label it h①–​②. tain different labels for the same objects. In
Figure 9.5.10a shows the labels for the first either case, objects (or labels) in one bubble are
two rows in the dux of the Op. 27 canon in sys- mapped to objects (or labels) of the other by h,
tems ① and ② and the transformation h①–​② =​ and a transformation f in one bubble becomes
〈−, 10, 10〉 that relates them. The rows labeled f =​Conjh(f ) =​ h−1 ● f ● h in the other.
0+​ and 5+​ here match the first two rows of the By similar reasoning, the coordinate transfor-
dux in Figure 9.5.5b, which was based on system mation from system ① to system ③ in Table 9.5.9
①; 10− and 3− are the labels for the same rows in is h①–​③ =​ 〈−, 0, 2〉, and from system ② to system ③
system ②. Figure 9.5.10b, more abstractly, shows is h②–​③ =​ 〈+​, 4, 2〉.
transformations between two row forms A and
B in coordinate systems called Φ and Ψ. Here AΦ Exercise 9.5.11
and BΦ are the labels assigned to the two rows (a) In Figure 9.5.10a, verify that S7 =​Conjh(S5),
in the system Φ, and AΨ and BΨ are the labels where h =​ h①–​② =​ 〈−, 10, 10〉.
for the same rows in Ψ. The situation is famil- (b) Verify that the coordinate transformations
iar from our general study of isomorphisms of   identified above satisfy h①–​② ● h②–​③ =​ h①–​③.

368 • E x ploring M usi c al S pa c es

 The ordered-​pair notation (n, σ) (either n+​ possible to incorporate a retrograde sign into the
or n−) accommodates only the 24 P and I row UTT notation to accommodate the transforma-
forms. To handle the R and RI forms in sys- tions that include R, but writing such transfor-
tem ④ we may append a second sign, forming mations in the form UR, where U is a standard
an ordered triple (n, σ, ρ), or in abbreviated UTT, is sufficient for present purposes. (In the
form nσρ. Here σ, the inversion sign, is the same next section we will devise a more comprehen-
as before, while ρ is the retrograde sign. If ρ is sive row-​ labeling system accounting for both
+​, then nσρ =​ nσ+​ represents the same P or I row pitch-​class and order transformations.)
label that we have previously denoted nσ; if ρ is As we have observed, the coordinate transfor-
−, then nσρ =​ nσ− is the retrograde of nσ. Table mation h③–​④ from system ③ to system ④ is R itself.
9.5.9, in the column for coordinate system ④, Consequently h①–​④ =​ h①–​③ ● h③–​④ =​ 〈−, 0, 2〉R and
shows the row forms R0 and RI0 in this nota- h②–​④ =​ h②–​③ ● h③–​④ =​ 〈+​, 4, 2〉R. Figure 9.5.12 repro-
tion as 0+​− and 0−− respectively. duces the Op. 27 analysis from Figure 9.5.5b in all
All 48 row forms may be notated in this way four coordinate systems, displaying the coordi-
and related by transformations of the form U or nate transformations as bubble transformations.
UR (=​ RU) for various UTTs U (or more generally The four bubbles, regarded as separate networks,
QTTs). The formalism is simplified by the fact are all isomorphic, the coordinate transforma-
that the retrograde operation R commutes with tions acting as network isomorphisms. Within
all UTTs. Whenever a transformation involving each bubble, the horizontal arrows are schritts
R is applied to a row nσρ, the retrograde sign ρ and wechsels, always matching in both voices of
changes. All such transformations form a group the canon (as discussed above for system ①); in
isomorphic to 𝒰 × 𝒞2 (or 𝒬 × 𝒞2). It would be different bubbles, these transformations are not

FIGURE 9.5.12 Analysis of Op. 27, II in coordinate systems ①–​④, with coordinate transformations

Graphs and Networks; Serial Transformations • 369

necessarily the same but are related by conju- space of BACH forms and a spatial network
gation. We noted in Section 9.3 that a network showing the path traced by repeated
isomorphism is strong if and only if it commutes applications of U in this space. We may now
with all the transformations in the network. Of identify U with the UTT 〈−, 4, 4〉, which
the coordinate transformations in Figure 9.5.12, generates the group G =​{T0, T4, T8, R, T4R, T8R}.
the only one that commutes with all transforma- This group has a direct-​product structure {T0, T4,
tions in the bubbles is h③–​④ =​ R. Because of this T8} × {T0, R} isomorphic to 𝒞3 × 𝒞2, but because
commutativity, conjugation by R has no effect, so 2 and 3 are coprime we know from Section 6.4
the transformations appearing in bubbles ③ and that such a group is cyclic, generated in this case
④ are identical; these two networks are strongly either by U =​ T4R or by U−1 =​ T8R.
isographic and hence strongly isomorphic. The BACH forms in the chain above are those
with index numbers 3, 7, and 11. These six
Example 9.5.13: Webern, String Quartet, forms constitute one of the four orbits of G, an
Op. 28, I (revisited) Our understanding of orbit we shall call 3 . The other orbits are 0, 1,
serial transformations enables us to refine and 2; each orbit m contains all forms n+​ and
the analysis of Webern’s String Quartet, Op. n− for which n ≡ m (mod 4). If the symbols n+​
28, from Example 4.5.1.49 As noted there, the and n− are interpreted as triads, these orbits are
quartet is composed entirely of “BACH forms,” the hexatonic triad families from Section 8.2.
serial transformations of the ordered tetrachord In the triadic interpretation the transformation
B♭–​A–​C–​B♮, and the work’s structure is described U =​ 〈−, 4, 4〉 cycles through the triads of one
in terms of BACH forms more readily than orbit via an LH-​chain (and not, for example, an
complete twelve-​tone rows. Following the LP-​chain, which would be more a more common
notation introduced in Chapter 4, we index P way of navigating the triadic family).
forms of BACH by their first notes and R forms The U-​chain breaks at the second appearance
by their last, writing 10+​ for the tetrachord (10, of the BACH form 3+​ =​(3, 2, 5, 4), whose last
9, 0, 11) (B♭–​A–​C–​B♮) and 10− for (11, 0, 9, 10), two notes overlap with the first two notes of the
the retrograde of 10+​. The sign in this notation is next form 5+​ =​(5, 4, 7, 6). With this T2 move
a retrograde sign, but because of the tetrachord’s Webern departs from the orbit 3 for the first
RI symmetry, R forms are also I forms, and the time, moving to 1 , where the theme of this
24 forms n+​ and n− exhaust all BACH forms. variations movement soon concludes after two
The quartet’s first movement, a theme with more U moves, from 5+​ to 9− and then 1+​.50
variations, opens with the chain of BACH forms The event network in Figure 9.5.14 tells the
transformational story of BACH forms in the
7+​, 11−, 3+​, 7−, 11+​, 3−, 7+​, 11−, 3+​, theme and the first variation. Variation 1 is
a canon in which two chains of BACH forms
generated entirely by the transformation proceed in counterpoint. BACH forms 6+​ and 3+​
U =​ T4R. Figure 4.5.3 presented a double-​circle initiate the dux and comes respectively. Of these,

FIGURE 9.5.14 Event network for Webern, String Quartet, Op. 28, I, theme and variation 1

370 • E x ploring M usi c al S pa c es

3+​ is another T2 transformation of the last-​ row in Schoenberg’s String Quartet No. 4—​a
heard tetrachord 1+​, while 6+​ is obtained from row we will call Q, to be used for illustrations
1+​ via T5. The tetrachords 6+​ and 3+​ represent throughout this section—​is identified with the
the orbits 2 and 3 , and both of the canonic ordered set
voices in variation 1 are generated entirely by U,
so the voices remain in these orbits throughout Q =​(2, 1, 9, 10, 5, 3, 4, 0, 8, 7, 6, 11).
the variation. Each voice traces an exact pitch-​
class transposition of the first nine BACH forms This simple definition suffices for many pur-
of the theme, and the canon always maintains a poses, and we will continue to display rows in
T9 relationship between the voices, established this way when convenient, but some aspects of
by the initial BACH forms 6+​ and 3+​ and shown serial structure are illuminated more effectively
in the figure as a bubble transformation. by a different formalization.
The techniques introduced in these The notes of a row may be indexed by order
opening measures foreshadow much of the numbers 0, 1, …, 11. In the notation (a0, a1,
transformational activity in the movement, and …, a11) from the previous section, these order
indeed in the entire quartet.51 All six variations numbers are the subscripts. We write order =​
are two-​voice canons at T9, with motion via {0, 1, …, 11} for the set of order numbers. Each
U predominating in each canonic voice. A few note of a row is also a pitch class, an element of
short U−1-​chains reverse the order of BACH pc =​{0, 1, …, 11}. Numerically, the sets order
forms from the prevailing U-​chains; more than and pc are identical—​both are simply ℤ12—​but
80 percent of the BACH forms in the movement in the context of a row they are interpreted dif-
are obtained from an immediately preceding ferently. A number in order specifies a posi-
BACH form via either U or U−1. Consequently tion within a row without naming the note that
the music consistently inhabits a pair of adjacent appears there, while the same number in pc speci-
orbits such as the first variation’s 2 /​ 3 pair. The fies a pitch class without indicating where in a row
most common interorbital moves, and the most it occurs. A row is specified by pairing each order
frequent transformations other than U and U−1, number with a specific pitch class; that is, a row is
are T2 and T5, whose first occurrences were noted a bijection between the sets order and pc. A row
above. In particular, T5 occurs several times at may be identified with a set of twelve ordered
variation boundaries, modulating to a new orbit pairs (i, x), a subset of order × pc, in which the
pair for the next variation. The 5+​ –​9− –​1+​ codetta first component i of each pair is an order number
that introduces a new orbit at the end of the from the set order, while the second component
theme does not reappear consistently but has x is a pitch-​class number from pc—​and with the
later ramifications: variation 2 contains a similar crucial property that each mod-​12 integer occurs
codetta, and variations 3 and 5 each switch to a exactly once in each component. In this notation
different orbit pair for three tetrachords before Schoenberg’s row is the set
switching back. Variations 4 and 6, like variation
1, remain entirely in orbits 2 /​ 3 . The similarities Q =​{(0, 2), (1, 1), (2, 9), (3, 10), (4, 5),
between orbit pairs and key areas in tonal music (5, 3), (6, 4), (7, 0), (8, 8), (9, 7), (10, 6),
are suggestive, with the pair 2 /​ 3 playing the (11, 11)}.
role of the movement’s tonic—​an idea recalling
Lewin’s concept of twelve-​tone areas, illustrated This notation tells us that pitch class 2 (D) occurs
previously in Figure 4.1.18. U-​directed motion is at order position 0 (the first note), pitch class 1
even more prevalent in the two later movements at order position 1, and so on. The same informa-
of the quartet: the second movement is generated tion may be conveyed in the form of a 2-​by-​12
without exception by U and U−1, never straying matrix similar to the permutation matrices stud-
from
   the orbit 0 . ied in Section 6.1,

9.6 TRANSFORMATIONS  0 1 2 3 4 5 6 7 8 9 10 11

Q= ,
OF PITCH CLASSES AND  2 1 9 10 5 3 4 0 8 7 6 11
ORDER NUMBERS
To this point we have assumed that a twelve-​ each order-​pc pair now appearing as one col-
tone row is defined as an ordered 12-​tuple of umn of the matrix.52 For theoretical purposes, in
the twelve pitch classes. For example, the first this section we regard rows as sets of order-​pc

Graphs and Networks; Serial Transformations • 371

pairs, but we will sometimes display these pairs gp transforms a row’s ordered pairs by the rule
in matrix form. gp(i, x) =​(i, g(x)). Such a mapping gp is called
The set of order-​pc pairs is itself an unor- a pitch-​class transformation on twelve-​ tone
dered set (as the curly braces in the ordered-​ rows. Familiar operations such as Tk or Ik can
pair representation of Q imply); the row’s order play the role of g: Tk and Ik are fundamentally
information is determined by the ordered pairs mappings on pc, but we may apply them to
in the set, not by the order in which they are a row by applying them to all the row’s pitch
listed. Thus the appearance of the order-​pc classes. Applying T3p and I6p to the row from
pair (3, 10) tells us that a row’s fourth note Schoenberg’s quartet, for example,
(order position 3) is B♭ (pitch class 10) regard-
less of where (3, 10) may appear in the list of
ordered pairs. We may, if we wish, rearrange the  0 1 2 3 4 5 6 7 8 9 10 11
Q= ,
 2 1 9 10 5 3 4 0 8 7 6 11
ordered pairs within the braces, or equivalently
rearrange the columns of the matrix, listing
them in numerical order by the pc component and therefore
rather than by the order component. The row
Q then takes the appearance
 0 1 2 3 4 5 6 7 8 9 10 11
T3P(Q ) = 
 5 4 0 1 8 6 7 3 11 10 9 2 
 7 1 0 5 6 4 10 9 8 2 3 11
Q= ,
 0 1 2 3 4 5 6 7 8 9 10 11
while

which tells us successively that pitch class 0 (C)

is in order position 7, pitch class 1 is in order  0 1 2 3 4 5 6 7 8 9 10 11
I6P (Q ) = 
 4 5 9 8 1 3 2 6 10 11 0 7 
.
position 1, and so on. This is the same informa-
tion conveyed in the previous matrix; the set of
order-​pc pairs is the same either way. The last two rows may be called simply T3(Q) and
Order numbers may be regarded as defining I6(Q), but the notations T3p and I6p remind us
the horizontal dimension of a row and pitch-​ that it is the pitch-​class components that have
class numbers as the vertical dimension—​ a been transposed and inverted.
conception facilitated by the ordered-​pair nota- The order components—​ the first compo-
tion. This perspective reveals complete symme- nents of the ordered pairs—​ may be trans-
try between the two domains. The symmetry is formed in all the same ways. Any bijection
manifest in particular in the relations between f : ℤ12 → ℤ12 may be regarded as a transforma-
pitch-​class transformations and order transforma- tion on order, giving rise to an order trans-
tions, two distinct kinds of mappings that never- formation f o on twelve-​tone rows, defined by
theless exhibit completely isomorphic behavior. f o(i, x) =​(f(i), x), the superscript o signaling the
These topics have been studied extensively; the order transformation. Operations such as Tk or
discussion in this section will be limited to a Ik can play the role of this f just as they played
few observations about the algebraic structures the role of g above. Applying T3o to Schoenberg’s
involved.53 row Q entails transforming the first component
Of the classical serial operations Tk, Ik, and of each ordered pair by T3:
R, the first two are pitch-​ class transforma-
tions, while the last is an order transforma-
tion. We consider the more familiar pitch-​class  0 1 2 3 4 5 6 7 8 9 10 11
Q= ,
transformations first. Any bijective function  2 1 9 10 5 3 4 0 8 7 6 11
g: ℤ12 → ℤ12 may be regarded as a transforma-
tion on pc, a permutation of the twelve pitch
and therefore
classes. As such, g determines a transforma-
tion gp: row → row that acts on twelve-​tone
rows by applying g to the second component  3 4 5 6 7 8 9 10 11 0 1 2 
T3O (Q ) = 
 2 1 9 10 5 3 4 0 8 7 6 11
of each ordered pair in the row, the pitch-​class .
component (hence the superscript p). That is,

372 • E x ploring M usi c al S pa c es

Rearranging the matrix by order number, composers have made use of them.54 In any
case, the algebraic structure of the rotation
and retrograde transformations in the order
 0 1 2 3 4 5 6 7 8 9 10 11 domain is completely isomorphic to that of
T3O (Q ) = 
 7 6 11 2 1 9 10 5 3 4 0 8  transpositions and inversions in the pitch-​class
.

domain. We know that the transformations

p p
In ordered-​set notation, this row is T3 (Q) =​ Tk and Ik form a dihedral
o
p
group of order 24,
(7, 6, 11, 2, 1, 9, 10, 5, 3, 4, 0, 8). The origi- which we now call 𝒯ℐ12 to distinguish it fromo
nal row Q has been rotated three places to the the analogous (and isomorphic) group 𝒯ℐ12
o o
right, a transformation sometimes denoted r3. of order transformations Tk and Ik . All alge-
In general, Tko =​ rk. Row rotations are “order braic properties of 𝒯ℐ12 that we have studiedo
transpositions,” counterparts in the order in previous chapters remain valid in both 𝒯ℐ12
p o
domain of transpositions in the pitch-​ class and 𝒯ℐ12 ; for ­example, the equation Ik =​ R ●
domain. r k +​1 in the previous paragraph may be written
o o o
Among the order-​number inversions Iko, the in the form Ik =​ I11 ● Tk+​1 , which is a special
case k =​11 is notable: case of the equation Im Tn =​ Im+​n from Theorem
●

5.5.6b. Transformations in 𝒯ℐ12o may be writ-

ten using the common notations rk and R, but
 0 1 2 3 4 5 6 7 8 9 10 11 the notation with Tko and Iko clarifies the cor-
Q= ,
 2 1 9 10 5 3 4 0 8 7 6 11 respondence with the pitch-​class transforma-
tions in 𝒯ℐ12p. The transformations in 𝒯ℐ12o and
so 𝒯 ℐ12p together generate the extended serial group
𝒮𝑒𝑟12+​, which contains order transformations,
pitch-​class transformations, and composites of
 11 10 9 8 7 6 5 4 3 2 1 0  the two. This is a larger group than 𝒮𝑒𝑟12, the
I 11O (Q ) = 
 2 1 9 10 5 3 4 0 8 7 6 11 “classical” serial group discussed in the previous
section, which contains combinations of Tk, Ik,
 0 1 2 3 4 5 6 7 8 9 10 11 and R but not rk.
= .
 11 6 7 8 0 4 3 5 10 9 1 2  Transpositions (as pitch-​ class transforma-
tions) do not generally commute with inversions
This row is (11, 6, 7, 8, 0, 4, 3, 5, 10, 9, 1, 2), the (also pitch-​ class transformations), and rota-
retrograde of the original row Q: I11o is another tions (order transformations) do not commute
name for the transformation we have denoted with retrogrades (also order transformations).
R. The other order-​number inversions Iko are Conveniently, however, order transforma-
rotated retrogrades. For example, for k =​0, tions always commute with pitch-​class trans-
formations: statements of the form f o ● gp =​
gp ● f o hold in general, a principle that makes
 0 11 10 9 8 7 6 5 4 3 2 1  the structure of groups such as 𝒮𝑒𝑟12+​ easy to
I 0 (Q ) = 
O

 2 1 9 10 5 3 4 0 8 7 6 11 grasp.55 Consider the result of applying both an

order transformation f o and a pitch-​class trans-
 0 1 2 3 4 5 6 7 8 9 10 11  formation gp to a row, where f: ℤ12 → ℤ12 and
=  . g: ℤ12 → ℤ12 are any bijections. Because f o affects
 2 11 6 7 8 0 4 3 5 10 9 1
only the first component of an ordered pair and
gp affects only the second, the order in which
This row is (2, 11, 6, 7, 8, 0, 4, 3, 5, 10, 9, 1), the two transformations are applied makes no
obtained by rotating the retrograde form by r1 difference: an ordered pair (i, x) will be trans-
(or T1o). That is, I0o =​ R ● r1. The general equa- formed to (f(i), g(x)) regardless of whether f or g
tion is Iko =​ R ● rk+​1; retrogrades are “order is applied first.
inversions.” The group 𝒮𝑒𝑟12+​, large as it is, contains only a
Retrogrades have of course been a part of the small fraction of all possible order and pitch-​class
serial system since its invention by Schoenberg. transformations. The multiplication operators
Row rotations have not been employed com- M5 and M7 do not belong to this group, but they
positionally on an equal footing with the other may be applied to any row, either as order trans-
transformations, though a number of serial formations or as pitch-​ class transformations.

Graphs and Networks; Serial Transformations • 373

For that matter, any of the 12! bijections from terms of rk and/​or R. For example, the transfor-
ℤ12 to ℤ12 may be applied to any row as a trans- mation I5oT4p (or T4pI5o) combines the retrograde
formation of either kind. Given any two rows rotation I5o, or equivalently Rr6, with the ordi-
A and B at all, we can always obtain B from A nary pitch-​class transposition T4. This transfor-
by transforming the pitch classes; that is, there mation may be applied to Schoenberg’s row
always exists a unique bijection g: ℤ12 → ℤ12 such
that gp(A) =​ B. (The function g maps every pc x to
0 1 2 3 4 5 6 7 8 9 10 11
the pc y that occupies the same order position in Q=
2 1 9 10 5 3 4 0 8 7 6 11
B that x occupies in A. That is, if the order-​pc
pair (i, x) occurs in A and (i, y) occurs in B, then
g(x) =​ y.) But it is equally true that we can always by transforming the order numbers in the top
obtain B from A by transforming order numbers: line of the matrix by I5 and the pc numbers
there always exists a unique bijection f: ℤ12 → in the bottom line by T4, then reordering the
ℤ12 such that f o(A) =​ B. (If (i, x) occurs in A and columns:
(j, x) occurs in B, then f(i) =​ j.) These observations
describe two different simply transitive actions
5
I5O T4P(Q ) = 
4 3 2 1 0 11 10 9 8 7 6
of the symmetric group 𝒮12, the group of the 12! 5 1 2 9 7 8 4 0 11 10 3
6
permutations of ℤ12, on row, the space of all 12!
twelve-​tone rows: one action permutes pitch-​ 0 1 2 3 4 5 6 7 8 9 10 11
=
class numbers while the other permutes order  7 9 2 1 5 6 3 10 11 0 4 8 
numbers. We use the notations 𝒮12p and 𝒮12o for
these two isomorphic groups acting on row.
This situation offers another example of the —the row (7, 9, 2, 1, 5, 6, 3, 10, 11, 0, 4, 8).
𝒯S-​𝒫S duality described in Theorem 7.5.24, at
The traditional system of row labels such as
the grand scale of the complete space row.56 We P0 and RI5 accommodates only the 48 trans-
may use the simply transitive action of 𝒮12p to formations in the group 𝒮𝑒𝑟12, but this system
turn row into an interval space whose interval may be expanded to accommodate the 576 row
group is 𝒮12 and whose “transpositions” are all forms that arise from the larger group 𝒮𝑒𝑟12+​.
the pitch-​class transformations gp in 𝒮12p. The One way to do this is to use labels of the form
“interval”-​ preserving transformations in this FoGp, where the labels F and G are boldfaced ver-
space are precisely the order transformations f o sions of the transformations f and g. For better
in 𝒮12o. Conversely, the transformations f o could correspondence with the traditional notation we
be the “transpositions” and gp the “inversions.” may substitute the label P for T. As usual, some
Because the cardinality of row is 12!, the total referential row form must first be selected as
number of bijections from row to row is the P0 (or more completely P0oP0p); there are advan-
unimaginably large number (12!)! (12 factorial tages to a fixed-​0 coordinate system in which
factorial); within that vast group the two iso- this referential row starts with pitch class 0 and
morphic subgroups 𝒮12o and 𝒮12p, both of order thus contains the order-​pc pair (0, 0). If we des-
12!, are centralizers of each other.57 ignate Schoenberg’s row Q, which starts with D,
The group 𝒮𝑒𝑟12+​ is much more manageable as P0oP2p (=​ T2p(P0oP0p)), then I5oT4p(Q), the row
in size than 𝒮12o and 𝒮12p. Because 𝒮𝑒𝑟12+​ is gen- form calculated above, will bear the label I5oP6p.
erated by the elements of the two groups 𝒯ℐ12o A row form traditionally labeled Ik becomes
and 𝒯ℐ12p, and because elements of 𝒯ℐ12o always P0oIkp in this notation, while Rk becomes I11oPkp
commute with elements of 𝒯ℐ12p, it follows and RIk becomes I11oIkp (recall that the retro-
that 𝒮𝑒𝑟12+​ is a group of order 24 ∙ 24 =​576 iso- grade transformation R is I11o).
morphic to the direct product 𝒯ℐ12o × 𝒯ℐ12p, or
𝒟12 × 𝒟12. The serial group 𝒮𝑒𝑟12, as noted in
Exercise 9.6.1 Calculate each of the following
Section 6.4, is a group of order 48 isomorphic to transformations of Schoenberg’s row Q:
𝒟12 × 𝒞2; 𝒮𝑒𝑟12 incorporates the full structure of
𝒯ℐ12p but only the subgroup {T0, R} of 𝒯ℐ12o. (a) T4oI5p(Q)
Every element of 𝒮𝑒𝑟12+​ may be written in (b) I11oI11p(Q)
the form f ogp (or equivalently gpf o), where both (c) T6oT6p(Q)
f and g are some Tk or Ik. The order component (d) Row form P6oI5p (where Q =​ P0oP2p)
of the notation may alternatively be written in (e)  Row form I2oI10p

374 • E x ploring M usi c al S pa c es

Exercise 9.6.2 Each of the rows given below from the row (1, 2, 10, 9, 5, 6, 0, 11, 3, 4,
is obtained from Q by some transformation in 8, 7). What is the degree of symmetry of
𝒮𝑒𝑟12+​. For each one, identify the transformation this row relative to 𝒮𝑒𝑟12? What is its degree
in the form f ogp, and give the row label in the of symmetry relative to 𝒮𝑒𝑟12+​? List all
form FoGp.58 symmetries of this row in 𝒮𝑒𝑟12+​ (there are
more than two).
(a) (5, 10, 9, 1, 2, 11, 6, 7, 8, 0, 4, 3) (c) If a chromatic scale (0, 1, …, 11) is considered
(b) (9, 10, 6, 2, 1, 0, 5, 8, 7, 3, 4, 11) as a twelve-​tone row, what is its degree
(c)  (10, 11, 9, 4, 5, 1, 0, 3, 8, 7, 6, 2) of symmetry relative to 𝒮𝑒𝑟12? What is its
  degree of symmetry relative to 𝒮𝑒𝑟12+​?60
Exercise 9.6.3 Each of the six transformational
identities below is true. Illustrate each of them
by applying the given transformations to the The symmetry between order numbers and
row Q, and verify that the transformations pitch classes suggests another kind of serial
on both sides of each equation give the same operation. The order/​pitch-​class exchange trans-
result. In (b) and (c), I2 and M7 refer to the formation X acts on any ordered pair (i, j) by the
pitch-​class transformations I2p and M7p; these formula X(i, j) =​(j, i), reinterpreting the original
two formulas illustrate the commutativity order number as a pitch-​class number and vice
of order transformations with pitch-​class versa.61 By exchanging what we have called the
transformations. The formulas in parts (d)–​(f) horizontal and vertical dimensions of a row,
are order-​number manifestations of the rules X effectively turns a twelve-​ tone row “side-
of composition for pitch-​class operations from ways”—​an intuition for which we will develop
Theorem 5.5.6. a more exact visualization shortly. When a
row is represented in its 2-​by-​12 matrix form,
(a) R ● r8 =​ r4 ● R X exchanges the top and bottom lines of the
(b) I2 ● r3 =​ r3 ● I2 matrix. Using Schoenberg’s row Q again as an
(c) M7 ● r4 =​ r4 ● M7 example,
(d) I7o ● T3o =​ I10o
(e) T3o ● I7o =​ I4o
(f)  I4o ● I1o =​ T9o 0 1 2 3 4 5 6 7 8 9 10 11
Q=
1 9 10 5 3 4 0 8 7 6 11
,
2
Exercise 9.6.4 Recall from Figure 5.5.4 that
pitch-​class inversion operators Ik always have
either two fixed points (if k is even) or no fixed 2 1 9 10 5 3 4 0 8 7 6 11
so X (Q ) = 
0 1 2 3 4 5 6 7 8 9 10 11
points (if k is odd). What analogous property
is true of the order transformations Iko on 0 1 2 3 4 5 6 7 8 9 10 11
=
 7 1 0 5 6 4 10 9 8 2 3 11
twelve-
   ​tone rows? .

Exercise 9.6.5 As we know from Section

6.5, the degree of symmetry of a musical That is, X(Q) is the row (7, 1, 0, 5, 6, 4, 10, 9, 8,
entity depends on the symmetry group under 2, 3, 11). In terms of interval or subset structure,
consideration. In Section 9.5 we observed that X(Q) may appear to bear scant resemblance to Q;
the possibilities for symmetrical rows relative for example, tritones occur three times between
to the group 𝒮𝑒𝑟12 are very limited. The larger successive notes of X(Q) while Q has no tri-
group 𝒮𝑒𝑟12+​, however, makes additional kinds of tones at all. On the other hand, the row’s many
symmetry possible. semitones are preserved under X; the chromatic
segment 8–​7–​6 in order positions 8–​9–​10 of Q
(a) The row of Berg’s Violin Concerto, (7, 10, 2, 6, converts directly to pitch classes 10–​9–​8 in order
9, 0, 4, 8, 11, 1, 3, 5), is not symmetrical in the positions 6–​7–​8 of X(Q).
traditional sense of 𝒮𝑒𝑟12, but is symmetrical The exchange transformation seems not to
relative to 𝒮𝑒𝑟12+​. What non-​identity have been employed directly by composers,
transformation in 𝒮𝑒𝑟12+​ is a symmetry of this but it enables further study of the relationship
row (maps the row to itself)?59 between pitch classes and order numbers, may
(b) Babbitt’s Composition for Four Instruments be used to model additional kinds of symme-
makes use of a trichordal array derived try exhibited by some rows, and gives rise to

Graphs and Networks; Serial Transformations • 375

algebraic structures related to others we have involution, X−1 could be replaced by X in this
seen. When X is combined with the transforma- equation, but the fact as stated is a useful
tions in 𝒮𝑒𝑟12+​, a still larger group 𝒮𝑒𝑟12X is gener- one: order transformations are X-​conjugates
ated. Clearly X is an involution, and the group of pitch-​class transformations and vice versa.
𝒮𝑒𝑟12X is twice as large as 𝒮𝑒𝑟12+​. It is not isomor- In particular, any order transformation may
phic to 𝒮𝑒𝑟12+​ × 𝒞2, however, because X does not be rewritten as a combination of a pitch-​class
commute with the transformations in 𝒮𝑒𝑟12+​. The transformation and two exchanges, and simi-
interaction of X with those transformations is larly any pitch-​class transformation may be
nevertheless easy to describe. For the row Q we rewritten using an order transformation and
calculated above that two exchanges.
The availability of X doubles the number of
rows in a typical row class, from 576 in a 𝒮𝑒𝑟12+​-​
T3 ( Q ) = r3( Q ) =
O  0 1 2 3 4 5 6 7 8 9 10 11 . class to 1152 in a 𝒮𝑒𝑟12X-​class (compared with 48
 7 6 11 2 1 9 10 5 3 4 0 8 
in a classical 𝒮𝑒𝑟12-​class). The FoGp row-​labeling
system described above for rows in a 𝒮𝑒𝑟12+​-​class
Therefore may be extended to 𝒮𝑒𝑟12X-​classes by including
the letter X in the label. If Schoenberg’s row
Q is P0oP2p, then X(Q) could be called either
(
X (T3 ( Q )) =
O 7 6 11 2 1 9 10 5 3 4 0 8
0 1 2 3 4 5 6 7 8 9 10 11 ) P0oP2pX or XP2oP0p (note the reversal of the
o and p components). We arbitrarily choose to

( )
0 1 2 3 4 5 6 7 8 9 10 11
place X first when it occurs, so our standard label
= 10 4 3 8 9 7 1 0 11 5 6 2 . for X(Q) is XP2oP0p.
Readers who worked through Section 8.3 have
seen the number 1152 before: it is the order of
We can obtain the same result by applying X the group 𝒬 of quasi-​uniform triadic transforma-
first (as calculated above), followed not by T3o tions (UTTs extended to accommodate inversion
but by T3p: operators). In fact, the groups 𝒬 and 𝒮𝑒𝑟12X are
isomorphic, sharing the wreath-​product struc-

( )
ture 𝒟12 ≀ 𝒞2. We will illustrate this isomorphism
0 1 2 3 4 5 6 7 8 9 10 11
X (Q ) = 7 1 0 5 6 4 10 9 8 2 3 11 , with double-​circle configurations like those we
introduced for UTTs and QTTs in Chapter 8—​
using them now to visualize the transformations
and therefore in 𝒮𝑒𝑟12X.
Figure 9.6.7 shows six such configurations
P
(
T3 ( X( Q )) =
0 1 2 3 4 5 6 7 8 9 10 11
10 4 3 8 9 7 1 0 11 5 6 2 ) ,
and transformations relating them. The configu-
rations are like those in Chapter 8 (for example
in Figure 8.3.11) except that the superscripts
+​and − have been replaced with o and p (arbi-
matching X(T3o(Q)) above. In left-​ to-​
right trarily associating the order domain with the +​
o ●
orthography, that is, T3 X gives the same mode). The symbols no and np appearing in the
result as X ● T3p. More generally, the equation configurations do not represent specific objects;
f o ● X =​ X ● f p holds for every order transforma- rather, a configuration as a whole represents one
tion f o, and similarly gp ● X =​ X ● go for every of the 1152 rows in a 𝒮𝑒𝑟12X-​class. The orienta-
pitch-​class transformation gp. The most general tions of the two circles show how each configu-
form of the equation, showing the interaction ration is related to the referential configuration
of X with any transformation f ogp in 𝒮𝑒𝑟12+​, is at the upper left, which depicts the row form
f ogp ● X =​ X ● gof p. P0oP0p. In the next configuration to the right,
depicting P2oP0p, the o circle is rotated two posi-
Exercise 9.6.6 Use the equations f o ● X =​ tions counterclockwise. This row transforma-
X ● f p and gp ● X =​ X ● go to derive the general tion is T2o, and the symbols no in the o circle are
o p o p
form
   of the equation, f g ● X =​ X ● g f . transposed by T2. In Chapter 8 we modeled this
transformation of a double-​circle configuration
The equation gp ● X =​ X ● go implies that with the UTT 〈+​, 2, 0〉, or equivalently the QTT
g =​  X−1 ● gp ● X =​ConjX(gp). Because X is an 〈+​, 2, 0, +​, +​〉.
o

376 • E x ploring M usi c al S pa c es

 The third configuration in the top row, for The identities T2o ● X =​ X ● T2p and I7p ● X =​
P2oI7p, is obtained from P2oP0p by I7p; here the X ● I7o are visible in the figure, along with cor-
symbols np are transformed by I7. The corre- responding identities involving QTT prod-
sponding QTT is 〈+​, 0, 5, +​, −〉, which according ucts. The network, that is, is path-​consistent,
to the instructions from Section 8.3 rotates the whether we choose to regard it as a network of
− circle (now the p circle) five places counter- transformations in 𝒮𝑒𝑟12X acting on row forms
clockwise and replaces each np with its mod-​12 or as a network of QTTs acting on double-​circle
inverse −np. configurations. Both of these group actions are
The configurations in the bottom row simply transitive, so we know from Section 9.2
of Figure 9.6.7 are the X-​transforms of that the network must be path-​consistent. The
those above. Graphically, X leaves numbers composite mapping from P0oP0p at the upper
unchanged but swaps the superscripts o and p; left to XI7oP2p at the lower right is XI7oT2p (or
in these configurations, therefore, the o circle equivalently T2oI7pX) as a serial transforma-
is on the inside. The QTT corresponding to X tion, or 〈−, 2, 5, +​, −〉 as a QTT.
is 〈−, 0, 0, +​, +​〉, which in its triadic guise as The groups 𝒮𝑒𝑟12X and 𝒬 are isomorphic
the parallel transformation P swaps the +​and − via an isomorphism that maps the row trans-
modes. The transformations relating the three formation Tko to the QTT 〈+​, k, 0, +​, +​〉, Iko to
configurations in the bottom row are the same 〈+​, −k, 0, −, +​〉, Tkp to 〈+​, 0, k, +​, +​〉, Ikp to 〈+​, 0, −k,
as those above but with o and p exchanged: +​, −〉, and X to 〈−, 0, 0, +​, +​〉. Transformations
T2p and I7o in place of T2o and I7p. On these in the group 𝒯ℐ12o map to QTTs of the form
configurations, T2p rotates the p (outer) cir- 〈+​, t+​, 0, ρ+​, +​〉, while those in 𝒯ℐ12p map to QTTs
cle while I7o reflects and rotates the o circle. of the form 〈+​, 0, t−, +​, ρ−〉. The QTTs of these

FIGURE 9.6.7 Double-​circle configurations for row forms and transformations in 𝒮er12X

Graphs and Networks; Serial Transformations • 377

two kinds form two pairwise commuting sub- process; numbers on the axis wrap around from
groups, both isomorphic to 𝒯ℐ12. top to bottom in the manner of a torus. For a
pitch-​class inversion Ikp the pitch-​class numbers
Exercise 9.6.8 are transformed by Ik, which also reverses the
(a) Identify the QTT corresponding to each of orientation of the axis. The corresponding order
the following transformations in 𝒮𝑒𝑟12X: transformations Tko and Iko are achieved via the
I11oT0p (=​ R); XT8oT4p; and I9oI6pX. same relabelings of the order axis. Finally, to
(b) Identify the transformations in 𝒮𝑒𝑟12X accomplish the exchange transformation X, we
corresponding to each of the following simply switch the labels order and pc, toggling
QTTs: 〈+​, 3, 9, +​, +​〉 (the schritt S3); the interpretations of the two axes. Compared
  〈−, 0, 5, −, +​〉; and 〈−, 6, 6, −, −〉. with the array in 9.6.9a, the array in (b) trans-
forms order numbers by T2 and pitch-​ class
Double-​circle configurations like those in numbers by I7; the resulting order-​pc pairs are
Figure 9.6.7 are helpful in visualizing trans- correct for the row T2oI7p(Q) (=​ P2oI5p). Part (c)
formational algebra in 𝒮𝑒𝑟12X, but they reveal then applies X to (b), showing the row XI5oP2p.
nothing about pitch-​class relationships within
a particular row. A standard twelve-​tone matrix Exercise 9.6.10
displays the 48 row forms in one classical row (a) Relabel the axes of the array in Figure 9.6.9
class in a compact 12-​by-​12 format.62 Perhaps to illustrate the following rows in the 𝒮𝑒𝑟12X-​
surprisingly, it is not difficult to construct an class of Q: I5oI6p; XP0oP0p; and
array of the same size that makes many row XP2oI5p. Use the array to write the string
characteristics visually apparent, and from of pitch classes for each of these rows.
which all 1152 row forms in a 𝒮𝑒𝑟12X-​class can Remember that the original row Q is
be read easily. Figure 9.6.9 illustrates, return- P0oP2p, not P0oP0p.
ing to Schoenberg’s row Q for the last time. The (b) Rather than relabeling the axes to perform
array for Q (=​ P0oP2p ) in part (a) of the figure these transformations, it would be possible
maps the row’s order-​pc pairs, its horizontal to leave the labels unchanged but to
and vertical dimensions labeled order and transform the array itself, translating it
pc in the manner of the axes of a graph. The horizontally or vertically (wrapping around
entries in the array could be 1s and 0s; we have as needed) for Tko or Tkp, reflecting about
opted to show black dots and blank squares a vertical or horizontal axis for Iko or Ikp,
instead.63 or reflecting about a diagonal axis for X.
Any row in the 𝒮𝑒𝑟12X-​class of Q may be read Redraw arrays for the rows in part (a) of
from this same array by an appropriate relabel-   this exercise in this alternative way.
ing of the axes. A pitch-​class transposition Tkp is
obtained by transposing every number on the pc The group 𝒮𝑒𝑟12X further expands the possi-
axis by Tk, shifting the 0 point of the axis in the bilities for row symmetry, and arrays like those

FIGURE 9.6.9 Array representations of twelve-​tone rows and serial transformations

378 • E x ploring M usi c al S pa c es

just considered can aid in visualizing this sym- Its degree of symmetry is 2 relative to the classi-
metry. A simple example is offered by the BACH cal serial group 𝒮𝑒𝑟12, but increases to 4 relative
tetrachord, viewed as a four-​note row (1, 0, 3, 2). to 𝒮𝑒𝑟12X; the symmetries of W in 𝒮𝑒𝑟12X are T0,
As a set of order-​pc pairs, this row is {(0, 1), RI11 (=​ I11oI11p), X, and XRI11 (=​ XI11oI11p).
(1, 0), (2, 3), (3, 2)}; as such, it is invariant under
the exchange transformation X. Figure 9.6.11a Exercise 9.6.12
shows a 4-​by-​4 array for this short row. The (a) Relabel the axes of the array in Figure
invariance of this array under 180-​degree rota- 9.6.11b as appropriate for the last three
tion signals the tetrachord’s RI symmetry, while transformations named above, and verify
the symmetry about the diagonal signals X-​ that these relabelings all determine the
invariance. The array is also symmetric about original row W.
the diagonal; this is a consequence of the two (b) The array in 9.6.11b was constructed using
symmetries already noted, because the combina- a row form transposed to start with pitch
tion of a 180-​degree rotation with a reflection class 1—​a row form that might be called
over one diagonal is equivalent to a reflection P1 in a fixed-​0 coordinate system or P0oP1p
over the other diagonal. in the more general row labeling described
Exact X-​invariance is rare in longer rows, but above—​because that form is invariant
the twelve-​tone row of Webern’s String Quartet, under X. Other forms of the row are no
Op. 28, offers an example, at least in one care- less symmetrical, but their symmetries
fully selected transposition. Example 9.5.13 are more complex transformations. Draw
analyzed the piece via transformations of BACH the array for P0 (that is, W transposed
forms, but we now consider the complete row, down by semitone) and visualize its
comprising three BACH forms in succession, in symmetry. Calculate X(P0) and overlay
the transposed form W =​(1, 0, 3, 2, 6, 7, 4, 5, 9, it in the same array, using a different
8, 11, 10). The array for W in Figure 9.6.11b has symbol to distinguish it from P0. What
the same symmetries as the tetrachord in (a), other transformations applied to P0
and W is invariant under X. Like the BACH tetra- yield the same row as X(P0)? Under what
chord, the complete row W is also RI-​symmetric. transformations is P0 invariant?
(c) Of the transpositions of W, only P1 is
X-​invariant. But one inversion of W
(some row P0oIkp) is X-​invariant, as is one
rotated transposition of W (some row
PjoPkp). Identify these row forms and draw
  their arrays.

Exercise 9.6.13 The first row in Babbitt’s song

cycle Du is B =​(3, 0, 5, 1, 2, 4, 10, 8, 7, 11, 6, 9).
Divide the row into hexachords B1 and B2.
Write each hexachord as a set of order-​pc
pairs, using order numbers 6–​11 for B2. Apply X
to each hexachord; what other transformations
give the same results? What transformations
in 𝒮𝑒𝑟12X fix each hexachord? Make an array for
the entire row and note the appearance of the
two hexachords. What transformations in 𝒮𝑒𝑟12X
   B1 to B2?
map

Exercise 9.6.14
(a) Construct a twelve-​tone row that
has no nontrivial symmetry in 𝒮𝑒𝑟12
but is invariant under the exchange
transformation X, and draw its array.
FIGURE 9.6.11 Arrays for Webern, String (b) We informally described the action of X as
Quartet, Op. 28: (a) BACH tetrachord; (b) com- turning a row “sideways.” That is not quite
plete twelve-​tone row W accurate; as we have seen, X transforms the

Graphs and Networks; Serial Transformations • 379

 shape of a row (as depicted in its array) by 7. Hint for Exercise 9.1.7: The mapping X from Figure
reflection across a diagonal, not by rotation. 8.3.20 may be useful in labeling the cross-​type trans-
What happens if you take the array for formations. For an analysis of this Mozart sequence in
Q from Figure 9.6.9a and rotate it 90 generic space see Hook [2014] 2020, 83–​85.
degrees clockwise, leaving the order and 8. Path consistency is condition D in GMIT’s defi-
pc labels in their original locations? What nition of transformation graphs (Lewin [1987] 2007,
transformation in 𝒮𝑒𝑟12X is this? 195). The commentary and additional definitions in
(c) Construct a row that is invariant under the this section are revised from Hook 2007a, where the
transformation described in (b), and draw term path consistency was introduced.
  its array. 9. See Cook 2005, 126. Acknowledging the lack
of path consistency, Cook describes the network as
“malformed.”
10. The failure of path consistency is not attrib-
NOTES utable to the bubble notation. Figure 9.2.2 remains
1. Lewin ([1987] 2007, 195) requires the trans- non–​path-​consistent if the bubbles and the single T9
formations in a transformation graph or network to arrow are replaced by two arrows as described above—​
belong to a semigroup, but this stipulation does not or even by just the single arrow from E♭ to C. Figure
constrain the functions that can appear: the set of all 9.1.4a, which incorporates the bubble notation but
functions f: S → S is always a semigroup, from which omits the P arrow, is path-​consistent.
the transformations appearing in any transformation 11. Shortly after the publication of GMIT, Lewin
network on S may be selected. realized that the “Tarnhelm network” of his Figure
2. For Lewin ([1987] 2007, 196) a transforma- 8.2a (Lewin [1987] 2007, 179) ran afoul of his rules;
tion network is an ordered 6-​tuple: the sets of verti- he offered a path-​consistent alternative in the notes
ces and arrows in the directed graph are enumerated that became the Preface to the 2007 reprinting, and
separately, and the transformation semigroup is also further revisions in Lewin [1992] 2006. Though Lewin
included. considered the revised analysis “better,” Hook 2007a
3. The space S is not explicitly included in Lewin’s argues that it sacrifices musical insight in the name
definition of transformation graphs ([1987] 2007, of path consistency. Both of the networks in Figure
195); Lewin requires only that arrows be labeled with 9.14, GMIT p. 213, modeling the slow movement
elements of some semigroup, not necessarily a semi- of Beethoven’s “Appassionata,” are also non–​ path-​
group of transformations on any particular space. consistent, as observed by Rings (2011b, 115) but
(For Lewin, a transformation graph is an ordered 4-​ apparently unnoticed by Lewin. When I presented
tuple comprising the sets of vertices and edges, the preliminary observations about path consistency at
semigroup, and the label function for the arrows.) the 2002 meeting of the Society for Music Theory, I
In musical applications, the semigroup elements are noted that non–​path-​consistent networks appeared in
always transformations, and we assume here that we two other papers in the same session.
always know what kind of elements the transforma- 12. Further support for relaxing the path consis-
tions act on, even if we do not know the specific ele- tency requirement may be found in mathematical
ments. This assumption is essential in the formulation practice. The definitions of transformation graphs
of cross-​type transformation graphs and has some and networks, like some other definitions in this
ramifications for the discussion of strong isography in book, are fundamentally mathematical, not musical.
Section 9.3. Mathematicians construct such graphs frequently,
4. Hints for Exercise 9.1.3: In a proper realization, typically referring to them as diagrams. A diagram
all vertices must be labeled with triads and label con- satisfying the condition we call path consistency is
sistency must be observed. Is it possible to construct called a commutative diagram; in common parlance,
a realization in which the leftmost vertex is occupied “the diagram commutes” (Dummit and Foote 2004,
by a minor triad? 100). The notion of a commutative diagram may be
5. Formally, the definition of a cross-​type transfor- said to generalize the concept of a commutative binary
mation network includes a set 𝒮 of spaces and a space operation: if functions f and g are defined on a set S,
assignment function, mapping from the vertices of the a simple diagram—​that is, a transformation graph on
graph into 𝒮. In the case of a single-​type network, 𝒮 S—​can be constructed with the property that the dia-
consists of a single space, so the space assignment gram commutes if and only if f ● g =​ g ● f. Many impor-
function is trivial and may be disregarded. tant mathematical concepts are described in terms
6. See Hook 2007a for additional examples of of commutative diagrams, which play a central role
cross-​type transformations and networks. in the field of category theory. Importantly, however,

380 • E x ploring M usi c al S pa c es

commutativity is not a requirement for all diagrams, 20. The definitions of isomorphism of transfor-
and mathematicians construct noncommutative dia- mation graphs and isography of transformation net-
grams often and for many purposes. works presented here are equivalent to those of Lewin
13. Hints for Exercise 9.2.3: Four of the networks [1987] 2007, 199–​ 200, though formulated some-
are path-​consistent. In (a), it is helpful to realize that what differently. Strong isography and isomorphism
T6, I, and R all commute. The concepts of paired paths of networks were first defined by Klumpenhouwer
and path consistency are fully applicable to cross-​type in the special case of Klumpenhouwer networks, as
networks as in (e) and (g). In (e), the answer is unaf- described in Lewin 1990.
fected if only the triadic portion of the network is 21. The reference to “corresponding arrows” in the
considered. In (g), the question is whether T12 ● QO definition of isomorphism of transformation graphs,
is the same function as QO (a mapping from pitch to to be precise, is a statement of compatibility between
pc). An equation of the form ab =​ b is not possible for the group isomorphism φ and the bijection θ between
elements of a group unless a is the identity, but recall the vertices of the two graphs, mentioned in the pre-
that cross-​ type transformations generally do not vious definition of isomorphism of directed graphs.
belong to a group. Specifically, if the transformational label on an arrow
14. An idiosyncrasy in Lewin’s formalization of from vertex v1 to v2 in graph A is f, then the label on
directed graphs (node-​arrow systems in his terminol- the arrow from θ(v1) to θ(v2) in Aʹ must be φ(f). Similar
ogy) is that he postulates a loop at every vertex: “we compatibility requirements are implicit elsewhere
shall stipulate that every node is in the arrow relation in this section when corresponding arrows or corre-
with itself” (Lewin [1987] 2007, 193). The reasons sponding vertices are mentioned.
for this unusual assumption are not explained, and 22. This is a minor abuse of notation. The vertices,
Lewin’s figures never show the loops. His intention arrows, and transformations that are integral parts
may have been to rule a graph like Figure 9.2.5b inad- of network N1 are not “transposed” by T9—​only the
missible on the grounds that path consistency would triads are transposed. It is nevertheless convenient to
require the transformational product around the cycle think of T9 as a transformation applied to networks,
to equal the transformation on the loop, which (if the making Figure 9.3.1 a “network of networks” as
transformations belong to a group) must be the iden- described by Lewin ([1987] 2007, 204–​06).
tity. In fact, as we have seen, no loop is needed to con- 23. The following discussion assumes single-​type
clude that this graph is not path-​consistent. networks: each network is defined on a single space,
15. Hint for Exercise 9.2.6: Four of the seven graphs but different networks may have different associated
in Exercise 9.2.3 are universally realizable. In consid- spaces. The theory of network isomorphism can be
ering the bottom vertex in (g), note that universal extended to cross-​type networks, each of the several
realizability does not require that a labeling of one spaces of one network being in bijective correspon-
vertex determine a unique realization of the graph. dence with one space in the other. This more general
16. Hints for Exercise 9.2.7: Neither of these graphs definition presents no theoretical obstacles apart
can be path-​consistent. Consider writing transforma- from notational complexity, but we will not pursue
tions in UTT notation. Which non-​identity UTTs can it here.
map a triad to itself? 24. If the transformations in N belong not to a
17. Hint for Exercise 9.2.8: To prove strong path group but to a semigroup K of transformations on
consistency, assume the graph contains a set of paired S, Conjh is an isomorphism from K to a semigroup of
undirected paths, and consider the cycle formed by transformations on Sʹ.
the two paths together. 25. Hint for Exercise 9.3.7: See Exercise 8.3.4.
18. For example, Gödel’s completeness theorem for 26. Though M2 is not an automorphism of the
first-​order languages asserts that a set of statements additive group ℤ12, it is a group homomorphism. It
(axioms) within a certain kind of formal system is con- is possible to define homomorphisms of transforma-
sistent, in the sense that it cannot be used to derive a tion graphs and networks, relaxing the bijection con-
logical contradiction, if and only if it has a model—​a straints for both the objects and the transformations.
mathematical structure in which interpretations are Analytical benefits of network homomorphisms are
assigned to the symbols of the system and in which less apparent than of isomorphisms, in part because
all the axioms are true. See Mendelson 2015, 82–​92. homomorphism does not produce an equivalence rela-
19. Hint for Exercise 9.2.10a: Graphs of all types tion on networks.
except (4) have appeared among examples in this 27. Hints for Exercise 9.3.10: The isomorphism h is
chapter. For type (4), try combining Riemannian and an affine transformation (discussed in Section 6.2); g
non-​Riemannian transformations in a graph with a must be the function that maps int(x, y) to int(h(x),
configuration of arrows like Figure 9.2.5c. h(y)) for all pitch classes x and y.

Graphs and Networks; Serial Transformations • 381

 28. Hints for Exercise 9.3.11: For each part of the 37–​42 of Elliott Carter’s Scrivo in vento. Childs 2006,
exercise, draw a diagram showing the interval int(X, inspired by Capuzzo 1999, analyzes the Carter pas-
Y) between two triads X and Y, the mapping T9 or I0 sage, describing relationships among the tetrachords
as appropriate, and the interval int(T9(X), T9(Y)) or via dyadic transposition but without using K-​nets.
int(I0(X), I0(Y)). (You may need two diagrams, depend- 40. Appendix B to Lewin 1990 considers the pos-
ing on whether X and Y are of the same mode or oppo- sibility of admitting M5 and M7 as pitch-​class trans-
site mode.) In each case determine the mapping g formations in K-​nets, and Peck 2004–​05 examines
that converts the original interval to the transformed recursion in networks with M operators in more detail.
one, and determine whether g is an automorphism of If the group of basic transformations is expanded in
the interval group. In particular, the homomorphism this way, however, the automorphism group expands
requirement g(i)g(j) =​ g(ij) must be checked. as well, so the possible isographies still are not in one-​
29. We assume a group structure in this discus- to-​one correspondence with the basic transforma-
sion, but similar remarks apply if the transformations tions. There exist groups that are isomorphic to their
belong only to a semigroup. own automorphism groups (the dihedral group 𝒟4 is
30. Hint for Exercise 9.3.14: The isomorphism h, an example), but none of the most commonly used
which must map every triad in the PL-​cycle in the musical transformation groups have this property.
figure to the correspondingly located triad in the LʹR-​ 41. Losada 2007, par. 12, presents this observa-
cycle, can be expressed as a QTT. tion in an equivalent form: a correspondence between
31. Hint for Exercise 9.3.15b: To verify that φ10–​12 is inversional subscripts in a network and hypernetwork
a group homomorphism you must show φ(X) ● φ(Y) =​ may disappear if some pitch class other than C is cho-
φ(X ● Y) for all transformations X and Y in ℛ. Consider sen as the reference point for pc labeling. Losada con-
schritts and wechsels in all possible combinations, cludes that the original correspondence was “merely
using formulas from Theorem 8.2.2 to calculate X ● coincidental.”
Y and the analogous formulas from Theorem 5.5.6 to 42. For principles of enumeration involving
calculate φ(X) ● φ(Y). twelve-​tone rows (and generalizations to n-​tone rows)
32. Hint for Exercise 9.3.15c: The way to define see Hook 2007d, pars. 77–​80. For statistical confirma-
φ12–​13 on inversions should be clear from the graphs. tion that composers choose symmetrical rows dispro-
To determine the behavior of φ12–​13 on transpositions, portionately often, see Hunter and von Hippel 2003.
remember that Tn =​ I0In and that φ12–​13 must be a 43. Hint for Exercise 9.5.1: Five of the rows are
homomorphism. symmetrical.
33. Unsurprisingly, automorphisms of dihedral 44. Hint for Exercise 9.5.3: Of the four yes-​no ques-
groups are mathematically well known. Appendix A to tions in this exercise, three have the same answer.
Lewin 1990 provides a proof that all automorphisms 45. The possibility of applying UTTs to twelve-​tone
of 𝒯ℐ12 are of the types described here. rows was suggested by Jack Douthett in a 2001 con-
34. Hint for Exercise 9.4.9c: Every pc set of cardinal- ference paper. The discussion in this section is con-
ity ≥3 contains at least one even interval class. densed from Hook and Douthett 2008, which explores
35. Lewin 2002 offers rare counterexamples to sev- the idea in more detail, with an extended analysis of
eral of these generalizations: his Example 1.3 is a large trichordal structure in Webern’s Concerto for Nine
K-​net with nine vertices but only eight pitch classes Instruments, Op. 24, in addition to the examples con-
because of a duplication, and Example 6.1 shows a dis- sidered here.
connected K-​net. Lewin 1990 opens with several T-​net 46. Hints for Exercise 9.5.8: The group 𝒮𝑒𝑟12ʹ is com-
examples. Figure 4.10 in Morris 2001 also includes mutative, so JTn =​ TnJ; you may call this transforma-
some large K-​nets of uncommon types. tion Jn (it is not the same transformation called Jn in
36. Brief discussions of path consistency of K-​nets, Section 8.3). Caution: most of the transformations Jn
in the name of “well-​formedness,” may be found in are not involutions.
O’Donnell 1998, 57, and Peck 2004–​05, 26–​28. 47. There are similarities between coordinate sys-
37. Hints for Exercise 9.4.11: Show both implica- tems for row labeling and label functions for interval
tions: if the K-​net is strongly path-​consistent, then spaces, discussed in Section 7.1. While Orb(A) can
every cycle contains an even number of I arrows, and be an interval space with interval group 𝒮𝑒𝑟12, row
vice versa. The result of Exercise 9.2.8 may be useful. labeling is more complex than interval-​space label-
38. The 〈Tn〉-​〈In〉 notation, introduced in Klum- ing. A label function labu in an interval space is deter-
penhouwer 1991, is employed in Lewin 1994 and mined by the selection of a single reference point u to
many subsequent publications. be labeled with the identity element of the group. A
39. The four tetrachords in Figure 9.4.16a occur row labeling, in contrast, depends on two reference
together with other all-​interval tetrachords in mm. points: we must select two inversionally related rows

382 • E x ploring M usi c al S pa c es

to be labeled P0 and I0. The number of possible row 54. The rotational arrays used by Stravinsky in sev-
labelings is therefore larger than the cardinality of the eral compositions of the 1960s employ rotation not to
space. If we require that P0 and I0 begin with the same entire twelve-​tone rows but to hexachords. Babbitt’s
pitch class, then it becomes possible to formalize row trichordal arrays deploy multiple rows simultane-
labeling in terms of interval-​space labeling, but at the ously, some of which are often related by rotation (see
price of excluding viable labelings such as coordinate Exercise 9.6.5b). For discussion of these practices see
system ③ in Table 9.5.9. Straus 2016, 328–​38. Examples of rotation can be
48. In Westergaard’s unusual row numbering, cus- found in the early serial music of the Second Viennese
tomized for the mirror canon, I forms bear negative School as well. Mead 1985 discusses rotation in
indices, so paired rows are labeled P0 and I0, P7 and I−7 Schoenberg’s Wind Quintet, Op. 26; see also Exercise
(rather than I5), and so on. 9.6.5a. In non-​serial contexts, rotation in the horizon-
49. For a more complete analysis of BACH forms tal (time) dimension can be related to beat-​class trans-
and related structures in the first movement of the positions, invoked in analytical study of the phase
quartet see Hook and Douthett 2008, 110–​19, and techniques of Steve Reich (Cohn 1992b, Roeder 2003).
Moseley 2018, 186–​93. Moseley (165–​69) also stud- 55. The general commutativity of pitch-​ class
ies similar cycles of BACH forms in the second move- operations with order operations is noted in Morris
ment of the same work. A study of the work by David 1987, 108.
Clampitt (2009, 207–​15) describes a transformation 56. There is nothing special about the number 12
functionally equivalent to the UTT U =​ 〈−, 4, 4〉 dis- here. Permutations of a finite set of any cardinality
cussed here, as well as an alternative analysis making may be described as permutations of elements or as
use of two contextual inversions in alternation, corre- permutations of order positions, and the two concep-
sponding in UTT notation to the wechsels W4 and W8. tions give rise to transformation groups that are dual
50. The move from 3+​ to 5+​ is an example of RI-​ to each other in this way.
chaining, dubbed RICH by Lewin ([1987] 2007, 180–​ 57. The 𝒯S-​𝒫S duality does not apply in full to the
84). The last two notes of any pitch or pitch-​class series smaller groups 𝒯ℐ12o and 𝒯ℐ12p, even though they are
always coincide with the first two notes of some RI-​ isomorphic and pairwise commuting, because these
related form of the same series; here the pitch classes two groups do not act in simply transitive fashion on
5 and 4, though they belong to two BACH forms, are the same space. An orbit of rows under the action of
presented only once, not twice. For the symmetri- 𝒯ℐ12p (rows related by transposition and inversion) is
cal BACH forms, RI-​related forms are also related by not the same as an orbit under 𝒯ℐ12O (rows related by
transposition, and RICH(n+​) =​ T2(n+​) while RICH(n−) rotation and retrograde). When the full groups 𝒮12p
=​ T10(n−). On BACH forms in general, therefore, RICH and 𝒮12o are considered, however, the same 12! rows
is the same as the schritt S2 =​ 〈+​, 2, 10〉—​but the label are produced by either group.
T2 suffices here because Webern applies the technique 58. Hint for Exercise 9.6.2: Look for distinctive
only to P forms. See Hook and Douthett 2008, 108–​ interval patterns within the row, such as a trichord
10, for further UTT analyses of RI-​chaining (and the that could be a transformation of the chromatic seg-
related T-​chaining), including commentary on some of ment 8–​7–​6 from Q.
Lewin’s analyses. 59. Hint for Exercise 9.6.5a: Consider rotating the
51. A network in Hook and Douthett 2008, 117, row by r3. What other row operation yields this same
expands Figure 9.5.14 to encompass the entirety of form of the row?
the first movement. 60. Hint for Exercise 9.6.5c: The chromatic row has
52. To make the correspondence with permuta- multiple symmetries obtained by combining rota-
tions explicit, if the matrix for Q is regarded as the tions with transpositions. Consider not only these
permutation matrix for a permutation q defined on symmetries but also combinations of them with the
the space pc, then the row Q is the result of apply- one nontrivial symmetry of this row in 𝒮𝑒𝑟12.
ing the permutation q to the pitch classes of the chro- 61. This X differs from the transformation called X in
matic row (0, 1, …, 11). These order-​pc matrices are of Example 8.3.19, though that mapping also exchanged
course entirely different from the usual row matrices elements between two isomorphic structures.
of twelve-​tone theory, which are 12-​by-​12 matrices, 62. For twelve-​ tone matrices see, for example,
not 2-​by-​12. Straus 2016, 301–​02.
53. The symmetry between pitch-​class numbers 63. O’Connell 1968 displays figures similar to the
and order numbers was noted in Babbitt [1960] 2003 arrays presented here. A matrix of 0s and 1s with a
and subsequently developed in O’Connell 1968 and single 1 in each row and column is an alternative form
other sources. For the most thorough study see Mead of permutation matrix, describing the permutation of
1988 and 1989. Morris 2001 offers a brief overview. pitch classes required to transform the chromatic row

Graphs and Networks; Serial Transformations • 383

(0, 1, …, 11) to the row under consideration. Mathe- Important mileposts in the K-​net literature include
maticians would probably prefer a default configuration Lewin 1990 and 1994; the Klumpenhouwer
in which the pitch-​class numbers 0–​11 are ordered from Network Symposium in Intégral (vol. 12, 1998), par-
top to bottom, in the orientation of a matrix rather than ticularly O’Donnell 1998; a dedicated issue of Music
a graph; this is one reason we refer to our figures as arrays Theory Spectrum (vol. 24 (2), 2002); the skeptical
rather than matrices (another reason is to avoid confu- reconsideration in Buchler 2007; and the series of
sion with traditional twelve-​tone matrices and with the responses to Buchler in the following issue of Music
2-​by-​12 permutation matrices employed earlier). The Theory Online (vol. 13 (3), 2007).
bottom-​to-​top ordering adopted here is consistent with
musical intuitions of direction in pitch-​class space, but Morris 2007 provides a valuable overview of math-
the distinction is of little importance once transforma- ematical literature on serial theory. O’Connell 1968
tions that reverse that direction are allowed. is a rich and underappreciated study, among the
first to examine multiplication operators, the order-​
pc duality (studied more thoroughly in Mead 1988
SUGGESTED READING and 1989), and arrays similar to those in Section
Transformation graphs and networks are formal- 9.6. Twelve-​tone topics of mathematical inter-
ized in Chapter 9 of GMIT (Lewin [1987] 2007, est not explored here include invariance matrices
193–​219), though some examples appear earlier (Morris 1991, Chapter 14, and in greater generality
in the book. The discussion here of cross-​type net- Morris 2001, Section 4.6); partitions of the aggre-
works (not considered by Lewin) and the path con- gate (Morris and Alegant 1988); and complex multi-
sistency condition is revised from Hook 2007a. plication (Losada 2014).

384 • E x ploring M usi c al S pa c es

 PART THREE
Geometric Music Theory
The OPTIC Voice-​Leading Spaces
 10
Spaces III
Introduction to Voice-​Leading Spaces

If two notes are close together in pitch space, behavior. Recall, for example, that pitch-​class
one note may be transformed to the other by space pc and fifths space fifth are isomorphic
motion through a small interval—​that is, by a as interval spaces—​but while pc may be consid-
small voice leading. (Small voice leadings are com- ered a voice-​leading space, fifth surely is not.
monly called “smooth,” “efficient,” or “parsimo- Moreover, we shall see that the group-​theoretic
nious.”) Pitch space may therefore be taken as structure required of interval spaces creates
a prototypical example of a voice-​leading space: constraints that many important voice-​leading
a space with the property that distances in the spaces do not, and cannot, satisfy.
space reflect the possibility of connecting ele- A related point is that voice leadings are gener-
ments by efficient voice leadings. ally not transformations, and for a broad assort-
The discussion of musical spaces in the fore- ment of reasons, transformations (mathematical
going chapters, particularly the triadic spaces functions) are of limited value as models of voice
introduced in Chapter 4 and studied further in leading. To be sure, transformations have been
Chapter 8, has included occasional commentary employed productively in the study of voice
about voice-​leading relationships among the ele- leading,1 and some characteristic voice-​ leading
ments of those spaces. For various reasons, how- motions may be elegantly modeled using trans-
ever, many of the spaces studied to this point do formations—​the neo-​Riemannian P, L, and R
not meet the above criterion for a voice-​leading are obvious examples—​but there is no reason to
space, and many of the techniques employed expect this to be true in general. In Section 2.4 we
so far are unsatisfactory as tools for studying considered some of the difficulties in defining a
voice leading. In particular, the structure of an function that adequately depicts the voice leading
interval space, developed at length in Chapters in a characteristic viio7–​i resolution of a fully dimin-
5–​7, guarantees nothing about voice-​ leading ished seventh chord. Numerous complications

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0010
arise, including the enharmonic equivalence of from pitch-​class set theory, may be extended to
differently spelled diminished seventh chords, the elements in any of the OPTIC spaces. Normal
unequal cardinalities of the two chords as pitch-​ forms become a convenient way of labeling
class sets, the fact that there are only three dimin- points in the spaces and will prove useful in our
ished seventh chords but twelve different minor study of specific examples in Chapter 11, where
triads to which they may resolve, the merging of the focus will turn increasingly from the algebraic
two voices in the first chord into a single note of and transformational to the topological and geo-
the second, the need to identify not just the initial metric as we examine some spaces in detail, iden-
and final chords but the movement of individual tifying musical objects within them and tracing
voices in order to differentiate between two differ- trajectories through them. Chapter 12 then con-
ent voice leadings joining the same pair of chords, siders the meaning of the word distance in a more
and the desire to define a function not just on a rigorous way, introducing distance functions as an
single input chord or a few chords of a particular alternative to interval functions, better suited to
type but on a domain containing enough chords to application in voice-​leading spaces.
make the function broadly applicable.
To avoid such difficulties, we will now work 10.1 THE HEXATONIC TRIAD
primarily with chords not just as pitch-​class sets GRAPH AS A CONTINUOUS
but as points in more general kinds of spaces—​
voice-​leading spaces. There is not just one such
VOICE-​L EADING SPACE
space but a large family of them. Voice lead- A few figures in previous chapters may be legiti-
ings are modeled as paths through appropriate mately regarded as representations of voice-​
spaces—​a kind of relation more general than leading spaces. Although these examples are
a function. A particular bo7–​ c resolution, for somewhat limited in scope, they embody some
example, determines a path joining two points essential aspects common to voice-​leading spaces
in a space of chords in four voices, and the path in general, and it is worthwhile to review them
shows how each voice moves. here with attention to those aspects. One such
The voice-​leading spaces considered here were example, developed at some length in this sec-
introduced to music theory in remarkable work by tion, is based on the hexatonic triad graph, sev-
Clifton Callender, Ian Quinn, and Dmitri Tymoczko eral versions of which were explored in Section
(2008).2 This area of research is sometimes called 3.1. This example will lead to a broader explora-
geometric music theory, and the spaces are referred tion of the complexities of larger spaces of three-​
to variously as CQT spaces, orbifold spaces, or OPTIC note chords in Section 10.2, where we will also
spaces (the first of these names derives from the revisit a few other familiar spaces in relation to
authors’ initials; the other two will be explained in the requisite conditions for voice-​leading spaces.
due course). One way to describe this approach is Figure 10.1.1 redraws the hexatonic triad
to say that it adds a geometric component to pitch-​ graph, combining information from Figures 3.1.5
class set theory; indeed, OPTIC spaces offer many and 3.1.7 along with some new features. The
ways to generalize and extend the traditional study reader should be able to visualize this graph as a
of pc sets. The OPTIC spaces are continuous spaces, three-​dimensional figure. The x, y, and z coordi-
not discrete. They are multidimensional; chords of nate axes from Figure 3.1.7 have been reoriented;
larger cardinalities generally require more dimen- these axes remain at right angles to each other
sions, and the geometry sometimes includes sur- in three dimensions, but the entire coordinate
prising complexities. For these reasons, OPTIC system is now tilted so that the cube balances on
spaces contain many unfamiliar chords and can be the corner representing the B-​augmented triad,
difficult to visualize. and the graph as a whole more closely resembles
The first two sections of this chapter explore, the orientation of Figure 3.1.5.3 The dimensions
tentatively and informally, a voice-​leading space of the cube are the same in all three coordinates,
of three-​note chords—​a three-​voice space, as we but distances in the z coordinate appear fore-
shall call it. In Section 10.3, recognizing the need shortened in the perspective of the figure. A ver-
for a more rigorous theoretical development, we tical axis, distinct from the x, y, and z axes, passes
revisit the OPTIC relations, or OPTIC symmetries, through the B-​ augmented and C-​ augmented
first introduced in Section 2.4, and describe them triads and through the center of the cube. If the
more systematically, partly in group-​ theoretic cube were spun like a top about this central axis,
terms. Each subset of the five OPTIC relations the two augmented triads would remain in place
gives rise to a voice-​leading space. Section 10.4 while the other six triads would revolve about the
shows how the notion of normal form, familiar axis. The planes slicing horizontally across the

388 • E x p l o rin g M u sica l S paces

FIGURE 10.1.1 The hexatonic triad graph as a voice-​leading space

cube will be discussed below and may be disre- the cube from the C-​minor to the C-​major triad,
garded for the time being. passing through intermediate points such as the
As in Figure 3.1.7, each of the x, y, and z coor- “triads” (0, 3.01, 7), (0, π, 7), (0, 31 2 , 7), and (0,
dinates defines a single voice in the triadic sonor- 3.99, 7) of possibly indeterminate quality. It is
ities. Each voice is free to move within its own useful to think of a glissando of this sort as an
dimension independent of the other two voices, idealized model of smooth voice leading, even if
but for the purposes of this figure, the move- such gestures are rarely executed in practice.
ment of each voice is restricted to the gamut of We may even imagine all three voices playing
a single semitone. Major, minor, and augmented intermediate pitches simultaneously, defining
triads are customarily regarded as discrete enti- points such as (11.9, 3.2, 7.3) in the interior of the
ties, which is tantamount to restricting the coor- cube, not on any edge or face at all. This point is
dinates x, y, and z to the integer values shown: x closer to the C-​minor triad (0, 3, 7) than to any
=​11 or 0, y =​3 or 4, and z =​7 or 8. Coordinates other lattice point, but the voices are displaced
(x, y, z) in three-​dimensional space for which from C, E♭, and G by −0.1, +​0.2, and +​0.3 semitones,
x, y, and z are all integers are commonly called respectively; adopting a shorthand introduced pre-
lattice points.4 Figure 10.1.1, however, may be viously, the chord might be labeled C−0.1–​E+​0.2–​G+​0.3.
regarded as a depiction of a continuous space, in Microtonal chords of this sort, though unfamiliar,
which each voice is free to execute a continuous are acoustically just as viable as the chords we play,
glissando through the intermediate non-​integer and to understand voice-​leading spaces it is impor-
values. We may imagine, for example, a violist tant to regard the familiar equal-​tempered objects
sliding up slowly from E♭ to E♮ while a cellist and as a distinguished set of lattice points in a continu-
violinist sustain the notes C and G. In this situ- ous space also populated by the more numerous
ation, y would increase continuously from 3 to 4 microtonal sonorities. In this way the entire solid
while x and z remain fixed at 0 and 7; this motion cube becomes a voice-​leading space. Every point of
traces a straight-​line path along a back edge of the cube is given by coordinates (x, y, z) with 11 ≤ x

Introduction to Voice-Leading Spaces • 389

≤ 0 (mod 12), 3 ≤ y ≤ 4, and 7 ≤ z ≤ 8, and every such consistently to pitch classes rather than pitches
point represents some three-​note chord. If a string (E rather than E4, for example), and we have also
trio plays a B-​augmented triad and then sharpens assumed that the values 11 and 0 in the x coordi-
it to a C-​augmented triad by glissandos in each nate differ by 1, as they do only in mod-​12 arithme-
instrument, the sonority will trace a path through tic (ℤ12, or in the continuous case ℝ12, rather than
the cube from the bottom vertex to the top: a ℤ or ℝ). We know from Section 1.4 that modular-
straight path up the central axis if the three instru- izing two dimensions independently, effectively
ments move in perfectly parallel motion, or a more wrapping lines into circles, creates the topology of a
circuitous path if they move at different rates. Any torus 𝕋 2, which, it should be recalled, is fundamen-
continuous path through the space may be taken to tally a two-​dimensional surface even though its
represent a particular voice leading from one three-​ well-​known representation as a doughnut requires
note chord to another. three. In the present situation, we are modular-
izing three dimensions independently, suggest-
Exercise 10.1.2 In Figure 10.1.1, locate, as ing that the appropriate topology, rather than ℝ3,
accurately as possible, each of the following may be the 3-​torus 𝕋 3, the solid analog of the usual
points, given by coordinates (x, y, z). For each torus, mentioned previously in conjunction with
point, indicate whether it lies in the interior of the three-​dimensional tonnetz of Figure 4.4.6.
the cube, on a face of the cube, or on an edge of In fact, even 𝕋 3 fails to capture the complex-
the cube, and also indicate which of the eight ity of three-​voice chord space, because of another
lattice points it lies closest to. equivalence relation that we have yet to consider.
Triads and other pitch-​class sets are customarily
(a) (0, 3, 7 2 3 ) regarded as unordered sets, but the coordinate sys-
(b) (0, 33 8, 71 4) tem of Figure 10.1.1 requires them to be plotted as
(c) (11, 3.8, 7.8) ordered triples. The C-​major triad, the set {0, 4, 7},
(d)   (111 2 , 31 2 , 7 1 2 ) appears in this figure at the coordinates (0, 4, 7).
The point (0, 7, 4), though it represents the same
Exercise 10.1.3 For each pair of points given chord, would fall at a different place in this coordi-
below, visualize a straight-​line path from one nate system. (This point would lie in the same hori-
point to the other in Figure 10.1.1. Describe the zontal plane with the three major triads in Figure
paths musically in terms of the motion of each 10.1.1, but it would lie far outside the cube, some
of the three voices in the chords. Each of these distance back and to the left within that plane.)
paths passes through exactly one of the points The triples (0, 4, 7) and (0, 7, 4) are permutationally
given in Exercise 10.1.2 above; determine which equivalent; permutational equivalence was intro-
point belongs to each path. Also determine, for duced in Section 2.4, and permutations were stud-
each path, whether the path passes through the ied algebraically in Chapter 6. A proper three-​voice
interior of the cube or whether it lies entirely in chord space should somehow recognize the equiva-
one face or one edge. lence of the six possible permutations of a chord,
but neither the coordinate system of Figure 10.1.1
(a) the path from the B-​augmented triad to the nor the toroidal topology of 𝕋 3 does so. This com-
E-​major triad plication and others will be addressed in the more
(b) from the E-​major triad to the C-​minor triad general constructions later in this chapter. The rel-
(c) from the C-​minor triad to the A♭-​major triad ative simplicity of the limited space encompassed
(d)  from the A♭-​major triad to the point (0, 31 2, 7) by the cube of Figure 10.1.1 allows us to disregard
these considerations for the time being.
When Figure 10.1.1 is considered in isolation, Each face of the cube may be described by
it is not too much of a simplification to identify holding one of the three voices fixed on a specific
this space with a portion of ordinary three-​dimen- pitch while allowing the other two to move freely.
sional Euclidean space ℝ3, as the xyz coordinates For example, the face on the lower right side of
imply, and it is tempting to assume that ℝ3 might Figure 10.1.1, bounded by the triads B+​, g♯, A♭, and
serve as a model for a more comprehensive space c, is defined by fixing y =​3—​that is, by holding
of three-​voice chords. There are complications, one voice to the note E♭. The value y =​3 specifies
however, which gain in significance as larger tracts a single point on the y axis, but in three-​dimen-
of this space are explored. First, within each of the sional space there are two other dimensions in
three dimensions, we have effectively presupposed which points may move, so this equation defines
a modular pitch-​class space, not pitch space. We a plane orthogonal (perpendicular) to the y axis.
have tacitly assumed octave equivalence, referring (The oblique orientation of the text label “y =​3”

390 • E x p l o rin g M u sica l S paces

in the figure is intended to suggest that the label integer sum classes 9, 10, 11, and 0 (mod 12) are
itself lies in this plane.) Fixing y =​4 instead defines shown, but in the continuous space, intermediate
the opposite face, bounded by the triads e, E, C+​, planes corresponding to non-​integer sum classes
and C. Fixing an intermediate value of y between exist as well. Every sum-​class plane intersects the
3 and 4 will describe another plane, parallel to central axis in a single point, and in fact sum classes
and intermediate between these two faces of the may be used to label points on that axis, from 9 at
cube. For instance, the equation y = 31 5 describes the bottom to 0 at the top. For this reason, we may
a plane parallel to the B+​–​g♯–​A♭–​c face, but off- think of the central axis as the s axis (“sum axis”);
set by one-​fifth of the distance to the opposite every point in the three-​dimensional space has an s
face. Points such as (0, 31 5, 7 ), (11, 31 5, 8), and coordinate satisfying the equation x +​ y +​ z =​ s.
( 111 2 , 31 5 , 71 4 ), “triads” one of whose notes is The intersections of the sum-​class planes with
20 cents higher than E♭, all lie somewhere on this the tilted cube are of some interest. The sum-​
intermediate plane. class-​9 and sum-​class-​0 planes each intersect the
The sum classes 9, 10, 11, and 0 from Figure cube in only a single point, the augmented triads
3.1.5, indicating the mod-​12 sums of a triad’s B+​ and C+​ respectively. The planes for sum classes
three pitch classes, reappear in Figure 10.1.1, 10 and 11 each cut through the cube in a triangu-
where these sum classes also determine planes. lar cross section; the sum-​class-​11 plane is shown
Each sum-​class plane is, by definition, specified in Figure 10.1.4 as viewed from above. The tri-
by a simple equation: for example, the equation angle is equilateral; its only lattice points are the
for the plane defining sum class 11 is x +​ y +​ z =​11, three major triads at its corners, but every other
where the arithmetic is mod 12. This equation, point in the triangle is represented by some non-​
like the equation y =​3 for one face of the cube, integer coordinates (x, y, z) that sum to 11. Some
is a linear equation—​that is, an equation of the of these points, those whose coordinates are all
form ax +​ by +​ cz =​ d, involving sums of x, y, and z, multiples of 1 6 , are shown in the figure, arranged
multiplied by constant coefficients a, b, and c. (All in a triangular lattice of their own.
three of the coefficients in the equation x +​ y +​ z =​ A few of these intermediate points are of par-
11 are equal to 1, while in the equation y =​3 the ticular interest. At the midpoints of the triangle’s
coefficients of the variables x and z are 0.) In gen- three sides lie quarter-​tone chords, whose coordi-
eral, any linear equation in three variables defines nates are multiples of 1 2 . For instance, the chord
a plane in ℝ3; more generally, a linear equation (0, 31 2 , 71 2 ), midway between the C-​major and
involving the variables in an n-​dimensional space A♭-​major triads, shares the note 0 (C) with both
defines an (n − 1)-​dimensional subspace.5 of those triads, but its other two notes bisect the
The sum-​class planes are not faces of the cube semitones E♭–​E and G–​A♭; this chord could be writ-
but cut across the tilted cube horizontally, orthogo- ten C–​E♭+​0.5–​G+​0.5. The lower of the two “thirds” in
nal to the central axis. The planes corresponding to this “triad,” the interval C–​E♭+​0.5, is 31 2 semitones

FIGURE 10.1.4 The sum-​class-​11 plane of Figure 10.1.1

Introduction to Voice-Leading Spaces • 391

in size, exactly halfway between equal-​tempered z axes lie not in this plane but at angles to it; the
major and minor thirds; the upper third E♭+​0.5–​G+​0.5 axes shown in Figure 10.1.4 are the projections of
is exactly the size of an equal-​tempered major third, the true axes onto the sum-​class-​11 plane. In the
though both of its notes are displaced from the plane of this figure, the three coordinates are not
usual equal-​tempered pitches by a quarter-​tone. independent—​a consequence of labeling points
At the center of the triangle, equidistant in a two-​dimensional plane using three-​dimen-
from the three major triads at the vertices, lies sional coordinates. In moving down away from
the point (112 3 , 32 3 , 72 3 ). This chord lies on the the C-​major triad as described above, not only
central axis shown in Figure 10.1.1, at the point does the value of z increase, but x and y simulta-
where that axis intersects the sum-​ class-​
11 neously decrease as required to preserve the sum
plane. The intervals between the notes of this class 11. (To understand why all three coordi-
chord are precisely 4, which means that the chord nates must change, the reader should visualize
is an augmented triad, all of whose notes are dis- this movement not only in the plane of Figure
placed from equal-​tempered pitches by the same 10.1.4 but also in the sum-​class-​11 plane within
amount: the chord is two-​thirds of a semitone the three-​dimensional view of Figure 10.1.1.) Any
higher than a B-​augmented triad and one-​third two of the three coordinates theoretically suffice
of a semitone lower than C-​augmented. to determine any point within this plane; the
In fact, every point on Figure 10.1.1’s central axis equation x +​ y +​ z =​11 can then be used to deduce
is an exact augmented triad, ranging continuously the third.8
over the semitone span from B+​ to C+​. Every such
point satisfies the equations y =​ x +​4, z =​ y +​4, and Exercise 10.1.6
x =​ z +​4 (mod 12). Any one of these three equations (a) Draw a picture corresponding to Figure
may be deduced from the other two: for example, if 10.1.4 showing the sum-​class-​10 plane as it
y =​ x +​4, then z =​ y +​4 implies z =​ x +​8 =​x − 4 (mod slices through the cube.
12), or equivalently x =​ z +​4. So the central axis is (b) Draw a picture showing the sum-​class-​91 2
defined by two independent equations, such as y =​ plane as it slices through the cube.
x +​4 and z =​ y +​4, which may be combined in the (c) Draw a picture showing the sum-​class-​101 2
single formula z =​ y +​4 =​x +​8 that appears at the   plane as it slices through the cube.9
top of Figure 10.1.1. In three-​dimensional space, a
single linear equation defines a plane (reducing the Two particularly important kinds of voice-​
dimensionality of the space by 1); adding a second leading motion are implicit in the above dis-
independent equation then defines a line within cussion. Vertical motion in the space of Figure
this plane (reducing the dimensionality again). 10.1.1, along the central axis or any other line
Independence of the equations is essential: if one parallel to that axis, corresponds musically to par-
equation can be deduced from the other (such allel voice leading in which all three voices ascend
as x +​ y =​1 and 2x +​2y =​2), then the two equa- or descend by the same amount. For example, the
tions together simply define the same plane that is B-​augmented and C-​augmented triads are related
already defined by either equation alone.6 by parallel voice leading. Parallel voice leading
may be thought of as a process of continuous
Exercise 10.1.5 The x axis in Figure 10.1.1 is transposition, and B+​ may be transformed into C+​
a line in three-​dimensional space. What pair of by such a process: the motion is straight up the
independent
   equations defines this line?7 central axis, from the bottom vertex of the tilted
cube to the top, passing through all of the other
Restricting our attention to the plane of Figure non-​equal-​tempered augmented triads that pop-
10.1.4, we may note that points aligned horizon- ulate this axis, and stopping when the interval of
tally in this view share the same z coordinate. As transposition is precisely one semitone. Earlier we
a point moves downward in this figure, away from described the vertical axis as the sum axis, but it
the C-​major triad—​for instance, from (0, 4, 7) to may also be thought of as the axis of transposition;
(115 6 , 35 6 , 71 3 ) to (112 3 , 32 3 , 72 3 ) to (111 2 , 31 2 , 8)—​ this simple depiction of the ubiquitous transposi-
the z coordinate steadily increases. Similarly, the x tion operation is one reason why we have oriented
and y coordinates of points in this plane increase the figure in this way. The T1 transposition from
with movement away from the E-​major and A♭-​ B+​ to C+​ might be notated (11, 3, 7) → (0, 4, 8);
major triads, respectively, in the directions shown numerically, this transposition operates by add-
in the small system of coordinate axes supplied at ing 1 to every coordinate (11 → 0, 3 → 4, 7 → 8), or
the upper right. In three dimensions, the x, y, and by adding the vector (1, 1, 1) componentwise.10 In

392 • E x p l o rin g M u sica l S paces

the geometry of Figure 10.1.1, parallel voice lead- two voices (4 → 3, 7 → 8, or in musical terms E →
ing always corresponds to addition of some (pos- E♭, G → A♭) while the other voice remains station-
sibly non-​integer) multiple of (1, 1, 1). Apart from ary. In the other example, (0, 4, 7) → (112 3 , 32 3 ,
the two augmented triads, no other pairs of lattice 72 3 ), all three voices move, but the sum class
points in Figure 10.1.1 are related by parallel voice is unchanged because the one ascending voice
leading, but many pairs of microtonal chords inte- (7 → 72 3 ) moves twice as far as each of the two
rior to the cube are so related. For example, the descending voices (0 → 112 3 , 4 → 32 3 ).
chords (B+​0.25, E♭, G+​0.25) =​(111 4 , 3, 71 4 ) and (C−0.25, A single moving voice in a three-​note chord
E−0.5, A♭−0.25) =​(113 4 , 31 2 , 73 4 ) are related by parallel cannot execute either parallel or zero-​sum voice
voice leading, because the second chord is a T0.5 leading by itself; both concepts depend on the
transposition of the first. interaction of multiple voices. For parallel voice
leading, all three voices must move together. If
Exercise 10.1.7 Locate the points (111 4 , 3, voice leading is constrained to be parallel, then
71 4 ) and (113 4 , 31 2 , 73 4 ) in Figure 10.1.1, and motion of any one voice uniquely determines
visualize
   the relationship between them.11 how the others must move, so parallel voice
leading is essentially a one-​ dimensional phe-
The second important type of voice leading nomenon, along vertical lines in Figure 10.1.1.
is represented by horizontal motion in Figure For zero-​sum voice leading, ascending motion
10.1.1. This kind of motion takes place within in one voice must be balanced by descending
one sum-​class plane, such as the sum-​class-​11 motion in some other voice or voices. This con-
plane of Figure 10.1.4. As a point moves within dition allows freedom of movement throughout
this plane—​for instance, from the C-​major triad one sum-​class plane, so zero-​sum voice leading is
(0, 4, 7) to the A♭-​major triad (0, 3, 8), or from C essentially two-​dimensional. More generally, in a
major to the point (112 3 , 32 3 , 72 3 ) on the central space of n-​voice chords, zero-​sum voice leading
axis—​some voices ascend while others descend, is (n − 1)-​dimensional, while parallel voice lead-
with the result that the sum of the coordinates ing always remains one-​dimensional.
remains fixed. For this reason we refer to this The sum-​class planes in which zero-​sum voice
type of voice leading as zero-​ sum voice lead- leading takes place are all orthogonal to the lines
ing.12 The first of the examples just mentioned, in which parallel voice leading takes place. It
(0, 4, 7) → (0, 3, 8), involves contrary motion in should be clear geometrically that it is possible to

FIGURE 10.1.8 The parallel and zero-​sum components of a voice leading

Introduction to Voice-Leading Spaces • 393

get from any point in Figure 10.1.1 to any other 10.2 A LARGER SPACE OF
by some combination of motion along a vertical THREE-​VOICE CHORDS
line and motion in a horizontal plane. Musically,
this means that given any two three-​note chords, The cube discussed in the preceding section
say α and β, even though α and β may not be related illustrates the association between voices of a
to each other by either parallel or zero-​sum voice chord and dimensions of a space, but its musical
leading, we can always find some other chord γ for limitations are apparent, starting with the fact
which it is possible to move from α to γ by paral- that the cube allows only a single semitone of
lel voice leading, and then from γ to β by zero-​sum motion in each voice. In this section we consider
voice leading. That is, γ is a transposition of α and how this cube may be expanded to a space that
shares a sum class with β. A caveat, however, is that includes more chords. This is easy enough up to
even if α and β are equal-​tempered chords, γ may a point, but complications (some of which have
not be. Figure 10.1.8 illustrates this principle when already been suggested) soon begin to arise with
α is the B-​augmented triad and β is A♭ major. The which any attempt to construct a comprehensive
most direct path from B+​ =​(11, 3, 7) to A♭ =​(0, 3, space of three-​note chords will need to contend.
8) traverses a diagonal of the y =​3 face of the cube, Figure 10.2.1, which adjoins several new hexa-
exhibiting neither parallel nor zero-​sum voice lead- tonic triad cubes to the original one, illustrates
ing. But starting at B+​, it is possible instead to move both the straightforwardness of the basic idea
up the central axis via parallel voice leading to the and some of the complications.
point where that axis intersects the sum-​class-​ The cube from Figure 10.1.1 appears at the
11 plane in which the A♭-​major triad resides. This bottom of Figure 10.2.1; let us work our way
intersection point, the chord γ, is the augmented up. The lattice points forming the corners of the
triad (112 3 , 32 3 , 72 3 ) at the center of Figure 10.1.4. original cube, depicted here as small spheres,
From there, the motion to A♭ major can take place have coordinates limited to the hexatonic collec-
in the sum-​class-​11 plane, via zero-​sum voice lead- tion hex34 =​{11, 0, 3, 4, 7, 8}, and within this
ing in which y descends 32 3 → 3 while the other cube in continuous space the three coordinates
two voices ascend 112 3 → 0 and 72 3 → 8. In three-​ are limited to the ranges 11 ≤ x ≤ 0, 3 ≤ y ≤ 4, and
dimensional space, the arrows corresponding to 7 ≤ z ≤ 8. If we transpose each of these ranges
the parallel and zero-​sum components of the voice up a semitone to 0 ≤ x ≤ 1, 4 ≤ y ≤ 5, and 8 ≤ z
leading form a right angle, and the straight-​line ≤ 9, the second cube in the stack is the result.
path from B+​ to A♭ is the hypotenuse of a right tri- The vertices of this second cube inhabit the
angle. This principle, that any voice leading may be hexatonic collection hex01 =​{0, 1, 4, 5, 8, 9}, a T1
broken into parallel and zero-​sum components at transposition of hex34. This cube introduces six
right angles to each other, holds in suitably gener- new major and minor triads, T1 transpositions
alized forms in many of the spaces we will consider of the triads in the original cube, along with the
in this chapter and the next.13 D♭-​augmented triad. The two cubes are joined at
the C-​augmented triad, which is shared by the
Exercise 10.1.9 In each part of this exercise, two hexatonic collections.
points α and β are given, representing chords With appropriate numerical modifications,
in the voice-​leading space of Figure 10.1.1. everything we have said about the hex34 cube
Determine in each case a third chord γ, related is true also about the hex01 cube. It may be
to α by parallel voice leading and to β by zero-​ regarded as a continuous space, bounded by
sum voice leading. Visualize the path from α to planes defined by fixed values in each coordi-
γ to β in the figure.14 nate, including the new boundary values x =​1,
y =​5, z =​9. It is sliced horizontally by planes cor-
(a) α =​C+​ =​(0, 4, 8), β =​e =​(11, 4, 7) responding to sum classes 0, 1, 2, and 3. Its cen-
(b) α =​c =​(0, 3, 7), β =​C =​(0, 4, 7) tral axis is a continuation of the original central
(c) α =​(0, 31 2 , 8), β =​(111 2 , 31 2 , 7) axis, given by the same formula z =​ y +​4 =​x +​ 8
(d)  α =​A♭ =​(0, 3, 8), β =​B+​ =​(11, 3, 7) as before. Parallel voice leading is still parallel
to the central axis, while zero-​sum voice leading
Exercise 10.1.10 Suppose points α and β have takes place within one sum-​class plane.
coordinates α =​(xα, yα, zα) and β =​(xβ, yβ, zβ)
respectively. Derive a general formula for the Exercise 10.2.2
coordinates of the point γ =​(xγ, yγ, zγ) that is (a) If instead of transposing the ranges spanned
related to α by parallel voice leading and to β by by each coordinate of the original cube we
zero-​
   sum voice leading.15 instead extended the ranges, allowing the

394 • E x p l o rin g M u sica l S paces

FIGURE 10.2.1 A chain of hexatonic triad cubes, joined at the augmented triads

full ranges of values 11 ≤ x ≤ 1, 3 ≤ y ≤ 5, observing how the two smaller hexatonic

and 7 ≤ z ≤ 9, we would obtain a larger cube, triad cubes fit within it. The chords at the
two units on each side, bounded at the six outer vertices of the large cube are not
bottom by the B+​ triad and at the top by D♭+​. major and minor triads. What are they?
Visualize this large cube in Figure 10.2.1, Like the major and minor triads of the

Introduction to Voice-Leading Spaces • 395

 smaller cubes, do they consist of three central axis, such as (1.01, 4.99, 9.02), cor-
representatives each of two inversionally respond to chords that divide the octave into
related chord types? three nearly equal parts (in this case intervals
(b) Altogether, how many lattice points (equal-​ of size 3.98, 4.03, and 3.99). In the equal-​tem-
tempered chords) lie within or on the pered world, apart from the augmented triads,
boundary of this 2 × 2 × 2 cube? Besides the the three-​ note chords that divide the octave
major, minor, and augmented triads and most nearly evenly are the major and minor tri-
the chords you identified in part (a), what ads shown in the figure, sets of prime form 037
are the others? that divide the octave into intervals of sizes 3,
(c) You have now identified several set classes 4, and 5. Trichords of prime forms other than
to which the equal-​tempered chords within 048 and 037 divide the octave less evenly, and
the 2 × 2 × 2 cube belong. Of course, the theoretically lie at larger distances from the cen-
cube does not contain every chord in any of tral axis of the figure; loosely speaking, a chord’s
these set classes. Perhaps more surprisingly, proximity to the central axis may be consid-
a chord in one of these set classes may fail ered a measure of the evenness with which the
to lie within this cube even if all its notes chord’s notes are spaced through the octave. The
belong to hex34 ∪ hex01, the total set of idea that chords that divide the octave evenly or
equal-​tempered pitch classes available in nearly so lie on or near the central axis of a space
the cube. Give an example of such a set is an important one and will reappear in several
  lying outside the cube.16 other contexts in this and subsequent chapters.

The remainder of Figure 10.2.1 is constructed Exercise 10.2.3

by successive T1 transpositions of the original (a) What augmented triad lies closest to the
hexatonic triad cube. Triads in the third and C♯-​minor triad (1, 4, 8)? If only equal-​
fourth cubes in the stack belong to the hexatonic tempered chords are considered, the answer
collections hex12 =​{1, 2, 5, 6, 9, 10} and hex23 =​ is the C-​augmented triad (0, 4, 8), but some
{2, 3, 6, 7, 10, 11}, respectively. The fifth cube other points on the central axis in Figure
depicts the same collection hex34 as the first 10.4 lie closer to c♯ than C+​ does. Which one
(there are, after all, only four different hexatonic is closest?
collections in equal-​tempered pc space) and con- (b) The distance from c♯ =​(1, 4, 8) to C+​ =​
tains the same triads. (The augmented triads (0, 4, 8) is, of course, 1. Calculate the
labeled B+​ and C+​ in the first cube are the same distance from the C♯-​minor triad to the
pitch-​class sets as E♭+​ and E+​ in the fifth cube.) point you identified in part (a), and verify
The relationship between the first and fifth   that this distance is smaller than 1.17
cubes will be examined further below.
The first four cubes correspond to the four Example 10.2.4: Fauré, Barcarolle No. 4 in A♭
hexatonic triad families studied in transforma- Major, Op. 44 The central episode of Fauré’s
tional terms in Section 8.2, a connection that fourth Barcarolle illustrates smooth voice
will be described further below. All twelve major leading through the chain of hexatonic triad
triads, all twelve minor triads, and all four aug- cubes. Following a cadence in A♭ major at m. 39,
mented triads are accounted for in these four the episode begins at m. 41 with a shift to the
cubes. Each major or minor triad appears in only parallel minor, the A♭-​minor triad then promptly
one hexatonic system, each augmented triad in giving way to E♭ augmented and G major as
two. All twelve sum classes are also accounted shown in Figure 10.2.5a. In the space of only
for in the four cubes, as indicated at right. Dark-​ two measures Fauré effectively modulates down
colored vertices represent chords of even sum a semitone by the simple means of dropping
class, while light vertices represent chords of each voice of the triad in turn: the third (A♭–​a♭),
odd sum class. Every edge of a cube joins a dark then the root (a♭–​G+​, spelled as E♭+​), then
vertex to a light vertex, demonstrating that the the fifth (G+​–​G).18 Figure 10.2.5b shows this
graph of the entire cube chain, like the hexatonic motion as a three-​step descent from A♭ major
triad graph of Chapter 3, is a bipartite graph. in the hex34 cube (the uppermost cube from
Every point on the central axis of Figure Figure 10.2.1) to G major at the corresponding
10.2.1 represents an augmented triad such as location in the hex23 cube.
(1, 5, 9), dividing the octave exactly into three That modulation is the essential harmonic
equal intervals of size 4. Points very near the motion of the first phrase of the episode

396 • E x p l o rin g M u sica l S paces

FIGURE 10.2.5 Fauré, Barcarolle No. 4 in A♭ Major, Op. 44: chord progressions and paths in the
cube chain

(mm. 41–​44). The second phrase (mm. 45–​48) 39–​58 is shown in parts (d) and (e) of Figure
begins with a subdominant triad in G major 10.2.5, omitting the second phrases of the
and ultimately cadences in G again. The path G-​major and G♭-​major sections. The path
from IV to I, however, is not a traditional engages all four hexatonic families en route
diatonic progression but rather the functionally from A♭ major to F major, its regular geometry
confounding C–​C+​–​E–​E♭+​–​G. As shown in Figure reflecting the systematic if somewhat
10.2.5c, this second phrase recapitulates the mysterious descent of the chords, one note
hex34–​ hex23 motion of the first phrase, but at a time dropping by semitone. The pattern
with different triads representing hex34; the breaks with an E chord at m. 60, which could be
emphasis is again on smooth voice leading seen as an early arrival of hex34, the hexatonic
through hexatonic systems with the augmented system that might have been expected to arrive
triad as a crucial link. The slip from E to E♭+​ is a few measures later. Of course, hex34 is also
the only point in either phrase where two voices the system in which the pattern began, and the
move simultaneously, expressed geometrically home system of the piece’s A♭-​major tonic; the
by a diagonal move across one face of the hex34 E chord harbors a seventh and eventually
cube rather than a move along a single edge. resolves
   as an augmented sixth to A♭.19
In mm. 49–​56, the two phrases just heard
repeat a semitone lower, cadencing now in G♭ Exercise 10.2.6 The chord progression at the
major. At m. 57 a third cycle begins like the first Barcarolle’s final cadence (mm. 92–​94) is F–​f–​
two, dropping briefly to F major in m. 58 before A♭+​–​C–​c–​A♭. Trace this path in a diagram similar
veering off course. The entire trajectory in mm. to those in Figure 10.2.5. Observe that all of

Introduction to Voice-Leading Spaces • 397

these triads lie in a plane. What does this mean third, not an octave. Indeed, chords in corre-
musically?
   sponding locations in the two cubes are not the
same chord, but are related by T4: A♭ major in the
Exercise 10.2.7 A chord progression in first cube appears in the same position as C major
Bruckner’s Psalm 150 (mm. 207–​11) is in the fifth.20 Of course, C major is already pres-
ent in the first cube, but at a different location.
D–​B♭+​–​g–​E♭+​–​a♭–​A♭+​–​a–​F+​–​B♭ A hexatonic triad cube is closed under T4: the T4
transposition of a chord in such a cube always
(Bruckner’s spelling). Write this progression to lies elsewhere in the same cube. The two shaded
illustrate smooth voice leading (the bass note of faces in Figure 10.2.1 call attention to the dif-
the first chord is A), and trace the progression fering orientations of the two hex34 cubes: both
in  a hexatonic triad cube chain. shaded faces contain the same chords, but they
are angled in different directions. Compared with
We now consider more closely the relation- the orientation of the first cube, the fifth has
ship between the two hex34 cubes appearing at been rotated one-​third of a turn (120 degrees)
the top and bottom of Figure 10.2.1. It is tempt- about the central axis of the space.
ing to suppose that these cubes coincide simply It is instructive to consider the coordinates
because octave equivalence wraps the cube chain of the chords in these two cubes in the xyz coor-
around on itself like the circle of pitch-​class space. dinate system. The C-​major triad in the bottom
To be sure, octave equivalence comes into play: it is cube has coordinates (0, 4, 7); we can see in
octave equivalence that allows us to label the front-​ Figure 10.2.1 that this point lies at the intersec-
facing plane at the top of the top cube z =​0 rather tion of the planes x =​0, y =​4, and z =​7. The C-​
than z =​12, and to consider the sum classes 9, 10, major triad in the top cube, in contrast, lies in
11, and 0 in the top cube to be the same as those planes labeled x =​4, y =​7, and z =​0, so its coor-
in the bottom. Because of octave equivalence, the dinates are (4, 7, 0). The coordinates for these
mod-​12 coordinates shown in the figure are effec- two C-​major triads, (0, 4, 7) and (4, 7, 0), are not
tively coordinates in the three-​dimensional torus identical but permutationally equivalent. Both
𝕋 3 rather than in unbounded Euclidean space ℝ3. octave and permutational equivalence play into
(We have equivocated so far about whether Figure the geometry of our three-​dimensional voice-​
10.2.1 is supposed to depict ℝ3, 𝕋 3, or something leading space. A 120-​degree rotation of any point
else; such questions will be addressed more rigor- in the space about the central axis effects a cyclic
ously later.) permutation of its coordinates, transforming
But octave equivalence is not the whole story. (a, b, c) into (b, c, a). We can therefore describe
Moving from one cube to the next in the chain the top cube as derived from the bottom cube
corresponds to a T1 transposition; by this rea- in any of several different ways: by T4 transposi-
soning, the relationship between the first (bot- tion, by 120-​degree rotation, or by cyclic permu-
tom) and fifth (top) cubes should be T4—​a major tation of coordinates.

FIGURE 10.2.8 A voice-​exchange progression joining two C-​major triads

398 • E x p l o rin g M u sica l S paces

 The chord progression shown in Figure of the new hex34 cube match that of either the
10.2.8a, in which a root-​position C-​major triad top or the bottom cube in Figure 10.2.1, or
moves to a first-​inversion C-​major triad through does it differ from both of them? The fifth cube
a chromatic voice exchange, illustrates further in the new cube chain contains yet another
implications of permutational equivalence as point representing a C-​major triad; what are its
well as additional complexities that arise as we coordinates?
  
move away from the central axis of the cube
chain. Let us suppose that the point with coor- The various C-​major triads considered above,
dinates (0, 4, 7) at the back corner of the origi- in the first and fifth cubes of Figure 10.2.1 and
nal hexatonic triad cube represents the initial at the right side of Figure 10.2.8b, are fairly
C chord in this progression, so that the x coor- distant from each other in our chordal space.
dinate corresponds to the bass voice, y to the Permutationally equivalent points are not
soprano, and z to the middle voice. In each suc- always so widely separated, however. The path in
ceeding chord, x increases by 1 as the bass rises 10.2.8b contains two occurrences of the pitch-​
chromatically, while y simultaneously decreases class set {1, 3, 7}, at coordinates (1, 3, 7) and (3,
by 1 as the top voice descends. The z coordinate 1, 7), and those two points are closer together
remains stationary at 7. The resulting chain than any two of the C-​major triads. This is pos-
sible because the set {1, 3, 7} contains two notes
(0, 4, 7) →
 (1, 3, 7) → (2, 2, 7) → (3, 1, 7) that are closer together than any pair of notes in
→ (4, 0, 7) a triad; points differing only in the permutation
of those two notes are therefore separated by
is traced by the arrow in Figure 10.2.8b. only a small voice-​leading distance. In continu-
The original hexatonic triad cube appears in ous space, we could also consider points such as
outline at the left side of this figure. To allow for (1.99, 2.01, 7) and (2.01, 1.99, 7), located at a
the changing x and y coordinates in the progres- small distance to the left and right of (2, 2, 7)
sion under consideration, the plane z =​7, which on the path: these points are permutationally
includes the triads C, e, c, and B+​ in that cube, equivalent and represent the same pitch-​class
is extended to show a grid of squares spanning set, and they are spatially very near to each other.
other x and y coordinates beyond the limited The central chord (2, 2, 7) is itself noteworthy
range of the original cube. The progression traces because of the doubled D (2). As an ordinary
a rightward path, cutting diagonally across a pitch-​class set, this chord is not a trichord at all
series of grid squares in the z =​7 plane (7 being but a dyad {2, 7}; somewhat more informatively,
the pitch class common to all the chords in the it could be associated with the (unordered) mul-
chain). The path also illustrates a zero-​sum voice tiset {2, 2, 7}, which shows the doubling without
leading, as all the chords in the chain belong to specifying which voices are involved.
sum class 11. In fact, as the picture shows, the We might hope that an idealized space of three-​
straight line of this path is the intersection of note chords should respect both octave and permu-
the sloping z =​7 plane with the horizontal sum-​ tational equivalence: that is, a pitch-​class set such as
class plane x +​ y +​ z =​11. The final chord in the {0, 4, 7} should be represented only by a single point
progression, (4, 0, 7), is another permutation of in the space, not by various points at various dis-
the coordinates of the C-​major triad, different tances from each other, as seems to be the case in the
from both of the C-​major triads (0, 4, 7) and (4, spaces we have considered so far. In such a space there
7, 0) seen previously in Figure 10.2.1. should be only a single hex34 hexatonic triad cube,
not two or more. Our ideal space must also contain
Exercise 10.2.9 Although (4, 0, 7) is a multisets such as {2, 2, 7} to accommodate the com-
permutation of (0, 4, 7), it is not a cyclic mon phenomenon of doublings; in effect our three-​
permutation, and this point lies far from the voice space must contain two-​note pitch-​class sets as
central axis of the original hexatonic triad cube well. In Chapter 11 we will describe three-​voice OP-​
chain. This new C-​major triad is nevertheless space, which satisfies these conditions.
part of its own hexatonic triad cube, all of As a preliminary attempt at grasping such a
whose chords are permutationally equivalent to space, we may try wrapping the cube chain of
the chords in the original hex34 cube—​and this Figure 10.2.1 into a large cycle of four cubes,
cube is part of its own chain of hexatonic triad so that the fifth cube coincides with the first.
cubes. Visualize the position of this new cube Because of the different orientations of those
chain in Figure 10.2.8b. Does the orientation two cubes, however, the wrapping must be

Introduction to Voice-Leading Spaces • 399

accomplished simultaneously with some sort of such a twisted cube chain. The successive rota-
twisting. A 30-​degree rotation of each cube rela- tions of the cubes are evident from the shifting
tive to the preceding cube will accomplish the orientations of the shaded faces, each of which is
120-​degree twist of the fifth cube required to a T1 transposition of the shaded face in the next
align it with the first. Figure 10.2.10 illustrates lower cube but is angled in a slightly different
direction. The twist enables the two hex34 cubes
at the top and bottom of the chain to appear in
the same orientation: in this figure, unlike the
simpler chain in Figure 10.2.1, triads in the
top and bottom cubes appear in corresponding
locations.
Figure 10.2.11 then wraps the cube chain into
a cycle, effectively gluing together the first and
fifth cubes. Part (a) of this figure attempts to per-
form this wrapping without a twisting; (b) takes
the twisting into account, rather like a Möbius
strip. In both pictures the hex34 cube appears at
the left side in approximately its original orien-
tation; from there, one may imagine that (a) has
been produced by bending the untwisted cube
chain from Figure 10.2.1 into a circle, while in
(b) it is the twisted arrangement from 10.2.10
that has been bent. Part (a) is more regular in
appearance than (b), but in the absence of the
twisting, intervallic relationships cannot remain
consistent from cube to cube. The dashed line in
each figure traces successive T1 transpositions of
the E-​major triad. In (a), one sees that the triads
E, F, F♯, and G all occupy corresponding locations
in their respective cubes, but then A♭ jumps to a
new location—​the twist happens all at once. In
(b) these triads shift progressively from a loca-
tion in one cube to a nearby location in the next,
as the cubes gradually rotate.
The four cubes in these figures comprise the
same major and minor triads as the four hexatonic
triad families studied in Section 8.2. In particular,
Figure 10.2.11a resembles a more explicitly three-​
dimensional version of Figure 8.2.19, the graph
inspired by Douthett’s Cube Dance graph, which
included augmented triads as links between hexa-
tonic triad families. Chapter 8’s transformational
apparatus explored the P and L relations among
the major and minor triads within each family at
length, but did not accommodate the augmented
triads. The edges of the graph in Figure 8.2.19
depicted single-​ semitone voice-​ leading displace-
ments, now represented similarly by the edges
of the cubes, but the earlier graph (like previous
published versions of the Cube Dance) made no
attempt to account for the twisting of the cube
chain as in Figure 10.2.11b, and was not conceived
as part of a larger space also containing other kinds
FIGURE 10.2.10 A chain of hexatonic triad cubes, of three-​note chords, whose construction is our
with a twist present goal.

400 • E x p l o rin g M u sica l S paces

FIGURE 10.2.11 (a) A cycle of hexatonic triad cubes (after Douthett, “Cube Dance”); (b) a cycle of
hexatonic triad cubes, with a twist

Introduction to Voice-Leading Spaces • 401

Exercise 10.2.12 Trace each of the following clear how the path from Figure 10.2.8b could be
chord progressions in the two graphs of Figure traced in this space, as it moves away from the
10.2.11: hexatonic triad cubes but nevertheless some-
how ends up back at the C-​major triad where it
(a) Bruckner, Psalm 150 (from Exercise 10.2.7): started, after apparently passing through many
points twice along the way.22
D–​B♭+​–​g–​E♭+​–​a♭–​A♭+​–​a–​F+​–​B♭ Even for some simpler graphs like the famil-
iar circle of pitch-​class space (Figure 1.2.3), we
(b) Smetana, Overture to The Bartered Bride have observed that the two-​dimensional plane
(from Exercise 4.2.13c): in which the figure is drawn is a musical fiction:
points lying off the circle cannot be associated
C–​c–​E♭–​e♭–​G♭–​f♯–​A–​a–​C with pitch classes. Something similar is true of
a picture like Figure 10.2.11. The discussion of
(c) Wolf, Mörike-​Lieder, No. 22, “Seufzer,” finite and infinite tonnetz graphs in Chapter 3
mm. 5–​8: is also worth recalling in this context. In Figure
10.2.1, as in an infinite tonnetz, every point of
D♭–​g♭–​G♭–​G♭+​–​D–​g–​G–​E♭+​–​E♭–​a♭–​A♭–​A♭+​–​E–​a–​A the musical space—​in this case every three-​note
   chord—​ appears at multiple locations. Figure
10.2.11, like the finite tonnetz, eliminates (or
Exercise 10.2.13 Construct a hexatonic triad at least reduces) this redundancy at the cost
cube chain analogous to Figure 10.2.1 but using of introducing new geometric and topological
a T9 relationship (rather than T1) between complications—​and still leaves many questions
successive cubes. Why is this possible (whereas unanswered.
a T2 relationship, for example, would not be When we study three-​voice OP-​space more
possible)? Is a twist required in order for the rigorously in Chapter 11, we shall see that it is
first and fifth cubes to appear in the same indeed a three-​dimensional space, but for rea-
orientation? What shortcomings, if any, does sons that should by now be apparent, its geome-
this
  arrangement have?21 try and topology cannot be as straightforward as
Euclidean space ℝ3 or even a 3-​torus 𝕋3, and even
The cycle of cubes in Figure 10.2.11 has intui- though an xyz coordinate system underlies the
tive appeal but also some unintended byprod- construction of the space, locations in the space
ucts. Some triadic relationships are misleadingly cannot be modeled by those coordinates alone.
depicted even in the twisted version in (b): for In this space, a chain of hexatonic triad cubes
example, A-​major and B♭-​minor triads appear must somehow loop back on itself in the man-
to be close together, while E-​major and F-​minor ner of Figure 10.2.11—​but at the same time, it
triads, which are in the same musical relation- must be possible to move from a C-​major triad
ship, appear more distant. A more fundamen- along the path of Figure 10.2.8b and also arrive
tal problem is that the coordinate system that back at the starting point. The space must appar-
determined the geometry of Figure 10.2.1 has ently contain multisets such as {2, 2, 7}, and the
been distorted beyond recognition. In Figure above discussion suggests that there must be
10.2.1, every point in three-​dimensional space, something unusual about the geometry in the
even a point lying far outside the cube chain, vicinity of such points, in that a chord such as
may be labeled unambiguously with xyz coordi- {1.99, 2.01, 7}, though itself a single point of the
nates and therefore associated with a musical space, is somehow simultaneously located on
chord. Already in Figure 10.2.10, the direction two opposite sides of {2, 2, 7}.
associated with the x, y, or z coordinate varies Before we develop the theory required to
from cube to cube, and the circular wrapping grasp the structure of such spaces, it is appro-
of the cube chain in 10.2.11 further confounds priate to reconsider a few other figures from
the association. At best it seems that a coordi- previous chapters in relation to voice-​leading
nate system might be applicable at a local level in spaces. We have noted some points of contact
such a figure, but not globally. It is not clear what between our discussion of voice-​ leading rela-
musical chord a point lying outside the cubes tionships among three-​note chords and triadic
shown should represent, even if its location transformations. Let us consider the tonnetz, in
relative to the points appearing in the figure can the form of the neo-​Riemannian triad space of
somehow be specified exactly. It is also far from Figure 4.2.5b or 4.2.8. Certainly the P, R, and L

402 • E x p l o rin g M u sica l S paces

relations, elegantly modeled by these graphs, are inspection of Figure 4.2.5b, the neo-​Riemannian
characterized by efficient voice leading, and voice distance between C major and F minor is 3.
leading often figures prominently in tonnetz-​ Nevertheless, there is a smooth voice leading
based analysis. Nevertheless, comparison of from a C-​major triad to an F-​minor triad, requir-
the tonnetz with the space we have been tenta- ing only that two voices (E and G) be raised by
tively exploring in this section suggests that as a semitone. We say that the voice-​leading distance
potential voice-​leading space, the tonnetz leaves from C major to F minor is 2. This is the same
much to be desired.23 Among its shortcomings as the voice-​leading distance between R-​related
may be noted the following: triads such as C major and A minor, whose neo-​
Riemannian distance is only 1. So two triads,
• The tonnetz is fundamentally a discrete graph; both apparently at distance 2 from C major in
the above discussion suggests that a continu- terms of voice leading, lie at two very different
ous space is required to model voice leading distances (1 and 3) from C major in the tonnetz.
adequately.
• The logic of the tonnetz is based on interval- Exercise 10.2.14
lic relationships, which also enable certain (a) Calculate the neo-​Riemannian distance and
voice-​leading behavior. From time to time (for the voice-​leading distance between C-​major
instance, in Exercises 1.4.9 and 1.4.17 and and F-​major triads. Compare the results
Figure 1.4.18) we have encountered alternate with the distances between C-​major and
tonnetzes based on intervallic relationships F-​minor triads discussed above. You should
different from the usual ones. The fact that find that as measured by neo-​Riemannian
the resulting graphs may resemble the usual distance, C is nearer to F than to f, but as
tonnetz even while the corresponding voice-​ measured by voice-​leading distance, C is
leading properties are dramatically different nearer to f than to F.
suggests that the association between the ton- (b) Make a table analogous to Table 4.2.16
netz and voice leading is only incidental. showing the voice-​leading distance from the
• The only chords in the standard tonnetz are C-​major triad to every other major or minor
major and minor triads. Consequently, the triad. What is the largest voice-​leading
only kind of voice leading that can be depicted distance from C major to any other major or
in the tonnetz is from one major or minor minor triad?24
triad to another. Many possible smooth voice (c) Unlike the distance measurements in Table
leadings, such as the progressions in Figures 4.2.16, voice-​leading distance can be defined
10.2.5 (with augmented triads) and 10.2.8 for other kinds of chords besides major and
(with chords of other types, including dou- minor triads. Calculate the voice-​leading
blings) cannot be adequately modeled in a distance from the C-​major triad to each of
tonnetz. the following chords: the C-​augmented triad;
• The tonnetz treats the P, R, and L relations the C♯-​augmented triad; the C-​diminished
symmetrically, but connections between triad; the C♯-​diminished triad; the chords
P-​ or L-​related triads, which entail only a {D♭, E♭, G} and {D, D, G} from Figure 10.2.8;
one-​semitone displacement in one voice, are and the multiset {B♭, B♭, B♭}.
intrinsically more efficient than those between (d) What is the largest voice-​leading distance
R-​related triads, which require a two-​semitone between any two three-​note chords
displacement. (allowing doublings)? Give an example of
• Adjacent vertices in the tonnetz represent   two chords separated by that distance.25
chords connected by smooth voice leading
(with the qualification just noted), but dis- Having seen some inadequacies of the ton-
tances between non-​adjacent vertices do not netz as a voice-​leading space, the reader may
accurately reflect voice-​leading distances. now be surprised by Figure 10.2.15, which
reveals a tonnetz lurking inside the hexatonic
The last point deserves elaboration. Distances triad cube chain. As a finite graph with 24 verti-
between vertices in the tonnetz are measured by ces (six of which are repeated at the top and bot-
what we called the neo-​Riemannian distance in tom of the figure), this picture is isomorphic to
Section 4.2. C-​major and F-​minor triads are not the neo-​Riemannian triad space of Figure 4.2.5.
adjacent in the tonnetz, as they are not related The abstract graph, however, is oblivious to the
by P, R, or L. According to Table 4.2.16, or by deeper aspects of the organization of the cube

Introduction to Voice-Leading Spaces • 403

 connected by additional edges of the graph
representing R. Construct another graph of
neo-​Riemannian triad space by drawing the
three PR-​cycles and supplying additional
edges for L. (The PR-​cycles will not look like
  cubes.)

Another graph worth reconsidering in rela-

tion to voice-​ leading spaces is Figure 3.3.1,
depicting trichordal set-​ class space. Like the
hexatonic triad cube chain, this graph shows
three-​note chords, and here as well, edges depict
single-​semitone displacements connecting one
chord to another. Nevertheless, the two pictures
convey different information and differ mark-
edly in appearance. The distinguishing feature
of Figure 3.3.1 is that its vertices stand for set
classes, not sets. C-​major and E-​major triads, for
example, are two different points in the voice-​
leading spaces we have considered in this chap-
ter, but both are represented by the same vertex
in Figure 3.3.1: the vertex labeled 037, which
also represents the other 22 major and minor
triads.
Figure 3.3.1 can, in fact, be regarded as a voice-​
leading space, and in Chapter 11 we will gain
some understanding of why this figure takes the
form that it does. It is, however, a voice-​leading
space of a different sort from those discussed so
far. By replacing sets with set classes, this figure
FIGURE 10.2.15 A tonnetz (neo-Riemannian effectively adds transpositional and inversional
triad space) in a hexatonic triad cube chain equivalence into the mix along with octave
and permutational equivalence. The OPTIC
equivalence relations introduced in Section
chain. In particular, the augmented triads are 2.4—​ octave, permutational, transpositional,
not part of the tonnetz graph. Figure 10.2.15 inversional, and cardinality equivalence—​ may
shows three edges representing the R transfor- appear in different combinations and give rise
mation apparently intersecting at a point cor- to different kinds of voice-​leading spaces. While
responding to each augmented triad in the cube the hexatonic triad cube chain is a part of three-​
chain—​but that point is not a vertex of the ton- voice OP-​space, Figure 3.3.1 shows three-​voice
netz graph, and in more familiar representations OPTI-​space. An understanding of voice-​leading
of the tonnetz those edges do not intersect at all. spaces in general depends on a detailed under-
The tonnetz may be conceived as a part of voice-​ standing of the OPTIC relations and the ways in
leading space, in other words, but the full space which they interact with each other, as well as
contains a great deal of additional information on knowing which equivalence relations apply in
that the tonnetz lacks. any given situation. It is to these matters that we
turn our attention in the next section.
Exercise 10.2.16
(a) Trace the path of a PL-​cycle, a PR-​cycle, 10.3 THE OPTIC RELATIONS
an RL-​cycle, and a PLR-​cycle in the graph
of Figure 10.2.15. (See Exercises 4.2.13 Several figures in Sections 10.1 and 10.2 were
and 4.2.14 for musical examples of these constructed using an xyz coordinate system as
cycles.) a way of labeling points in three-​dimensional
(b) In Figure 10.2.15, each cube contains space with ordered triples of numbers. At times,
a PL-​cycle; the four PL-​cycles are then without considering all the implications, we have

404 • E x p l o rin g M u sica l S paces

assumed the coordinates to be mod-​12 numbers ℝn and cpitchn are well-​defined spaces for
representing pitch classes, but in the simplest sit- every value of n =​1, 2, 3, …. We will also have
uation they are real numbers—​positive or nega- occasion to refer to the sets cpitch≤n or ℝ≤n,
tive, with no upper or lower bounds—​standing defined as ℝ≤n =​ ℝ1 ∪ ℝ2 ∪ ℝ3 ∪ ⋯ ∪ ℝn, the union
for pitches. The association of continuous pitch of all the pitch-​tuple spaces of n or fewer voices.
space cpitch with the real numbers ℝ has been The most general space we will encounter is
familiar since Chapter 1, and in Section 7.3 we cpitch* or ℝ*, defined as ℝ* =​ ℝ1 ∪ ℝ2 ∪ ℝ3 ∪ ℝ4 ∪
considered pitch-​tuple space cpitchn as an inter- ⋯, the union of all the sets ℝn, theoretically infi-
val space. Pitch-​tuple space will form the basis nite in number, comprising ordered tuples of any
for the construction of voice-​leading spaces in number of real numbers or musical pitches.
this chapter and the next, but its interval-​space As preliminary illustrations of the OPTIC
structure will now figure less prominently than relations, consider the following pitch tuples,
its geometry and topology. labeled with Greek letters for reference:
A pitch tuple in cpitchn is an ordered n-​
tuple of pitches, which may be identified with α =​(0, 4, 7) =​(C4, E4, G4)
an ordered n-​tuple of real numbers, a point in β =​(24, 4, −5) =​(C6, E4, G3)    α  ~O β
n-​dimensional Euclidean space ℝn. As long as we γ =​(4, 7, 0) =​(E4, G4, C4)     α  ~P γ
consider ordered n-​tuples only, with no equiva- δ =​(2, 6, 9) =​(D4, F♯4, A4)           α  ~T δ
lence relations imposed, this identification is ε =​(0, −4, −7) =​(C4, A♭3, F3)           α  ~I ε
uncomplicated—​but it is important to recognize ζ =​(0, 0, 4, 7) =​(C4, C4, E4, G4)  α  ~C ζ
that the ordered tuples in cpitchn are not the
same thing as unordered sets of n numbers (or As pitch tuples, α, β, γ, δ, ε, and ζ are all dif-
notes). Associating the pitch C4 with the num- ferent; that is, they are six different points in the
ber 0 as in previous applications, the numerical space cpitch*. The first five belong to cpitch3,
ordered triple (0, 4, 7) in ℝ3 indicates the pitch while ζ is an element of cpitch4. As indicated
triple (C4, E4, G4). This is a more specific con- at right in the above list, however, each of the
struct than the pitch-​class triple (C, E, G) consid- last five tuples is equivalent to α via one of the
ered in the previous section, and it is still more OPTIC relations:
specific than the concept “C-​major triad,” which
corresponds not to an ordered triple at all but • Tuples α and β are octave-​equivalent (~O)
to the unordered set {C, E, G}. Unordered sets because one can be obtained from the other
arise through the application of permutational by transposing some notes up or down by
equivalence, and set classes arise through addi- whole octaves. If notes are considered as pitch
tional equivalence relations. As we have noted classes instead of pitches, α and β become
in Sections 2.4 and 7.4, application of an equiva- indistinguishable. “Octave-​equivalent” may be
lence relation effectively discards a certain kind understood as “identical except for octaves”; a
of information, such as order or register, which mathematician might say that octave-​equiva-
differentiates some elements that we may wish lent pitches are “equal modulo the octave.”
to consider “equivalent” for some purposes. As • Tuples α and γ are permutationally equiva-
suggested in the preceding section, however, lent (~P) because they consist of the same
applying an equivalence relation can complicate pitches in a different order. They are “identical
the geometry, so we begin our construction of except for permutation,” or “equal modulo
voice-​leading spaces with the more general and permutation.”
geometrically simpler Euclidean space of ordered • Tuples α and δ are transpositionally equivalent
tuples. The pitches in an ordered pitch tuple may (~T, “equal modulo transposition”) because
sometimes be regarded as the notes in a melodic one can be obtained from the other by trans-
line, but for present purposes it is usually more position; in this case δ =​ T2(α).
suggestive to regard them as the voices in an n-​ • Tuples α and ε are inversionally equivalent (~I)
voice chord, with the voices listed in a particular because they are related by inversion about C4
order.26 We will refer to spaces of pitch n-​tuples, (0); that is, ε =​ I(α).
n-​note chords, or n-​note set classes as examples • Finally, tuples α and ζ are cardinality-​equiva-
of n-​voice spaces. Thus cpitch3 could be called lent (~C) because they become identical if the
three-​voice pitch-​tuple space, or simply pitch-​triple doubled C4 in ζ is reduced to a single C4. These
space; informally we may also call it three-​dimen- tuples are “identical except for cardinality”;
sional pitch space. according to the definition of the cardinality

Introduction to Voice-Leading Spaces • 405

 of an ordered set from Section 2.2, the cardi- note must be transposed up two octaves while
nality of α is 3 while the cardinality of ζ is 4. the third is transposed down an octave and the
middle note is left unchanged. Uniform trans-
Recall that equivalence relations are reflexive, position of a pitch triple by octaves, as in the
symmetric, and transitive. By symmetry, any of relation (0, 4, 7) ~O (24, 28, 31), is a special case
the above relations could just as well have been of octave equivalence (and also a special case of
written the other way around (β ~O α rather than transpositional equivalence); the general case is
α ~O β, for example). much more flexible.
Traditional music theory pedagogy includes
ample discussion of the O, T, and I relations but Exercise 10.3.1 Determine whether the
is much less explicit about our frequent reliance following statements about octave equivalence
on P and C. All five relations are conceptually are true or false.27
straightforward and musically suggestive, but
there are enough subtleties in their application (a) (0, 4, 7) ~O (−24, 28, 31).
that it is important to have a firm mathematical (b) (0, 4, 7) ~O (24, −28, 31).
understanding of exactly how they are defined, (c) (F3, G4, A5) ~O (F5, G4, A3).
so even though we already have some familiar- (d) (F3, G4, A5) ~O (A3, G4, F5).
ity with most of these relations, in this section (e)  (F3, F3, G4, A5) ~O (F5, G3, G4, A3).
we will work through each of them in turn in
some depth. We will see that four of the five rela- A group action underlies this equivalence. We
tions may be described via group actions of the know from Chapter 7 that cpitch3 is a direct-​
sort described in Section 6.5. Combinations of product interval space with interval group ℝ3.
these equivalence relations correspond to vari- Addition in ℝ3 is defined componentwise; that is,
ous symmetry groups; indeed, the OPTIC rela- (x1, x2, x3) +​(y1, y2, y3) =​(x1 +​ y1, x2 +​ y2, x3 +​ y3). The
tions may be called the OPTIC symmetries. In the interval int(σ, τ) from σ =​(s1, s2, s3) to τ =​(t1, t2, t3)
discussion below, we work mainly with abstract is the ordered triple τ − σ =​(x1, x2, x3), where x1 =​
ordered triples σ =​(s1, s2, s3) and τ =​(t1, t2, t3) in t1 − s1, x2 =​ t2 − s2, and x3 =​ t3 − s3. The pitch triples
ℝ3; the general situation in ℝn works essentially σ and τ are octave-​equivalent if and only if all three
the same way. Cardinality equivalence, which of the componentwise differences x1, x2 and x3
relates points in spaces of different dimensions, belong to the set 12ℤ =​{…, −24, −12, 0, 12, 24, …},
will require some special considerations. the subgroup of the additive group ℝ generated by
the number 12. In other words, σ ~O τ is true if and
—​—​—​ only if τ − σ ∈ 12ℤ × 12ℤ × 12ℤ =​(12ℤ)3.
The set (12ℤ)3, consisting of ordered triples
Octave equivalence has been studied from sev- whose components are all multiples of 12, is a
eral perspectives already. As we know, it is based subgroup of ℝ3. In Section 7.4 we described
on mod-​12 congruence of pitch numbers, and (12ℤ)3 as the congruence subgroup associated
pitch space (pitch or cpitch) is converted into with the relation ~O; it may also be regarded as
pitch-​class space (pc or cpc) by the formation of the symmetry group for octave equivalence in
equivalence classes of this relation. In Section three dimensions. Strictly speaking, the congru-
7.4 we extended the idea of octave equivalence ence subgroup is a group of intervals in cpitch3
from the single dimension of cpitch to the n-​ while the symmetry group is a group of transfor-
dimensional space cpitchn. Two pitch triples mations on cpitch3—​but these are effectively
in cpitch3 are octave-​equivalent whenever all the same because of the general equivalence of
pairs of corresponding notes in both triples are intervals and transpositions in interval spaces,
octave-​equivalent. That is, if σ =​(s1, s2, s3) and outlined in Section 5.6. That is, the triple (x1, x2,
τ =​(t1, t2, t3), then σ ~O τ means that the congru- x3) may be regarded as the interval between pitch
ences s1 ≡ t1, s2 ≡ t2, and s3 ≡ t3 (mod 12) are all triples σ and τ, or it may be identified with the
true. For α =​(0, 4, 7) and β =​(24, 4, −5) as above, transposition operation T(x1, x2, x3) which, applied
the statements 0 ≡ 24, 4 ≡ 4, and 7 ≡ −5 (mod to σ, will produce τ.
12) are all true, so it is indeed the case that α ~O The equivalence between cosets of subgroups
β. Octave equivalence does not require that β be and orbits of group actions noted in Section 6.6
obtained by transposing the entire ordered triple is also relevant here. The statement σ ~O τ is true
α uniformly by a fixed number of octaves: in the if and only if σ and τ belong to the same coset of
present example, to produce β from α, the first the subgroup (12ℤ)3 of ℝ3. Equivalently, σ ~O τ if

406 • E x p l o rin g M u sica l S paces

and only if σ is mapped to τ by transposition by associated with ordered triples of pitch classes
some element of the subgroup (12ℤ)3—​that is, if (rather than pitches), so this quotient space may
σ and τ lie in the same orbit of the transforma- be identified with continuous pitch-​class–​triple
tion group (12ℤ)3. The equivalence classes of the space cpc3. This quotient space cpc3 remains
relation ~O, or O-​classes, are the orbits; we write three-​dimensional, but it is now bounded; that
σO for the class containing σ and all its transposi- is, the whole space fits inside a box of finite
tions by elements of (12ℤ)3. In terminology from size. Octave equivalence generally preserves
Section 6.5, octave equivalence is the same thing the dimensionality of a space but reduces an
as H equivalence, where H is the subgroup (12ℤ)3. unbounded space to a bounded one.
As an additive group, 12ℤ is isomorphic to ℤ Topologically the space cpc3 is a three-​dimen-
itself, so (12ℤ)3 is isomorphic to ℤ3—​but ℤ3 and sional torus 𝕋 3, with the property (characteris-
(12ℤ)3 are two different subgroups of ℝ3, and the tic of tori in general) that if we exit the cube on
symmetry group in question is (12ℤ)3, not ℤ3. In one side we will re-​emerge at the corresponding
n dimensions, the symmetry group for octave point on the opposite side. Algebraically, octave
equivalence is (12ℤ)n. equivalence reduces ℝ3 (the interval group of
If we imagine three-​ dimensional space ℝ3 pitch-​triple space cpitch3) to the quotient
filled with 12 × 12 × 12 cubes (“cubic octaves”), group ℝ3/​(12ℤ)3. In one dimension, the quotient
the points of (12ℤ)3 form a discrete grid of lattice group ℝ/​12ℤ is ℝ12, the real numbers mod 12;
points at the corners of the cubes. These cubes the three-​dimensional quotient group ℝ3/​(12ℤ)3
form a repeating pattern of modules in three may be identified with (ℝ12)3, the direct product
dimensions, filling up the entire space in a man- of three copies of ℝ12. Indeed, (ℝ12)3 is the inter-
ner similar to the two-​dimensional tiles we have val group of the quotient interval space cpc3. (In
encountered in many plane figures. If we focus Section 7.4 we considered the analogous situa-
on the single cube in which all three coordinates tion involving discrete spaces, identifying the
lie between 0 and 12, then every pitch triple in quotient space pitch3/​~O with discrete pitch-​
the entire space cpitch3 is octave-​equivalent to class–​triple space pc3, whose interval group is
some point in this cube; the cube may be called not (ℝ12)3 but (ℤ12)3.)
a fundamental region for octave equivalence. The
fundamental region represents the quotient space —​—​—​
cpitch3/​~O, also known as three-​voice O-​space.28
Fundamental regions (or fundamental Much of what we need to know about permuta-
domains) are an important concept in topology. tional equivalence was developed in Chapter 6. In
Generally a fundamental region for an equiva- the present situation the three components s1, s2,
lence relation ~ defined on a space S is a region s3 of the ordered triple σ are being permuted. Two
R within S having the property that every point elements of ℝ3 are permutationally equivalent if
in S is equivalent via ~ to a unique point in R.29 they have the same components but possibly in a
In a minor abuse of terminology, we sometimes different order. This is the case, for instance, with
refer to a fundamental region for an equiva- the ordered triples α =​(0, 4, 7) =​(C4, E4, G4) and
lence relation (such as octave equivalence), while γ =​(4, 7, 0) =​(E4, G4, C4) in the list at the begin-
at other times we may refer instead to a fun- ning of this section. We know from Section 6.1
damental region for a space (such as O-​space that the number of permutations of three ele-
cpitch3/​~O, the quotient space arising from ments is 3! =​6, and that these permutations,
that relation). That is, if ~ is an equivalence regarded as transformations, form the symmet-
relation on a space S, and if S′ is the quotient ric group 𝒮3. In ℝ3, then, permutational equiva-
space S/​~, then a region R as described above (a lence is the same as 𝒮3 equivalence; the P-​class αP
subset of S) may be called either a fundamental consists of the six permutations of the triple α.
region for ~ or a fundamental region for S′—​and The cycle notation for permutations from
because the points of R are in one-​to-​one cor- Section 6.1 can be adapted to fit the present sit-
respondence with the points of S′, the funda- uation. We write permutations of the numbers
mental region may often be identified with the 1, 2, and 3, indicating the three coordinates in
quotient space S′ itself. an ordered triple. For example, the 2-​cycle (1, 2)
A fundamental region is not unique: any 12 × interchanges the first two components, while
12 × 12 cube whose edges lie parallel to the coor- the 3-​cycle (1, 2, 3) moves the first component
dinate axes could serve as a fundamental region to the second position, the second component
for octave equivalence. Points in O-​space may be to the third position, and the third component to

Introduction to Voice-Leading Spaces • 407

the first position. In the above example, the tri- (also called P-​space), therefore, should intuitively
ple α =​(0, 4, 7) is mapped to γ =​(4, 7, 0) by the be only one-​sixth the size of the original space
3-​cycle (1, 3, 2), because the first component (0) cpitch3. As it turns out, permutational equiva-
moves to the third position, the third component lence divides cpitch3 into six regions, any one
(7) to the second position, and the second com- of which may serve as a fundamental region
ponent (4) to the first position. Permutational representing the quotient space, and the planes
equivalence is symmetric, so some permutation mentioned above form the boundaries between
must map γ to α as well; in fact, γ is mapped to these regions. All six regions are three-​dimen-
α by the 3-​cycle (1, 2, 3), which is the inverse of sional and unbounded.
(1, 3, 2).30 Unlike the symmetry group for octave equiv-
Unlike the symmetry group ℝ3/​(12ℤ)3 for alence, the group 𝒮3 for permutational equiva-
octave equivalence, 𝒮3 is a noncommutative lence has no clear relation to the interval group
group. Because there are six permutations in 𝒮3, of cpitch3; we shall see in Chapter 12 that while
a typical ordered triple in ℝ3 is permutationally P-​space is a valid quotient space, it cannot be
equivalent to five other triples. Orbits of the constructed as a quotient interval space. Points
action of 𝒮3, that is, generally contain six ele- in P-​space may be regarded as unordered pitch
ments. Pitch triples with doublings, however, sets, or, allowing for doublings, pitch multisets—​
are exceptions: for example, nothing new results three-​note chords in pitch space. The complexity
if the first two components of the ordered triple increases with the number of dimensions; recall
(0, 0, 7) are interchanged. That triple is therefore that the order of the symmetric group 𝒮n is n!.
fixed by the 2-​cycle (1, 2), so its orbit under the In four-​dimensional pitch space cpitch4, most
action of 𝒮3 contains fewer than six elements. orbits of 𝒮4 have 4! =​24 elements, so the quo-
tient space cpitch4/​~P , four-​voice P-​space, is
Exercise 10.3.2 Determine whether the only 1 24 the size of cpitch4.
following statements about permutational Constructing a space of unordered chords in
equivalence are true or false. For each true pitch-​class space entails using the relations ~O and
statement, give the cycle representation of ~P in combination. OP-​space is among the most
a permutation in 𝒮3 or 𝒮4 that maps the first important voice-​ leading spaces; the discussion
point to the second.31 in Section 10.2 began to probe some of its com-
plexity, which will be further explored in Chapter
(a) (0, 4, 7) ~P (0, 7, 4). 11. New considerations arise when two or more
(b) (0, 0, 7) ~P (7, 0, 0). equivalence relations interact. Of the pitch triples
(c) (0, 0, 7) ~P (7, 7, 0). in the above examples, we know that β ~O α and
(d) (F4, G4, A4, B4) ~P (B4, A4, G4, F4). α ~P γ. Equivalence relations are transitive (if x ~ y
(e)  (F4, G4, G4, A4) ~P (A4, G4, F4, G4). and y ~ z, then x ~ z), but ~O and ~P are two differ-
ent relations, so we cannot conclude that either
Exercise 10.3.3 List the elements in the orbit of β ~O γ or β ~P γ is true—​and indeed, β =​(C6, E4,
(0, 0, 7) under the action of 𝒮3. What subgroup G3) and γ =​(E4, G4, C4) are neither octave-​equiva-
of 𝒮3 is the stabilizer of (0, 0, 7)? Verify that lent nor permutationally equivalent. We may say,
the orbit-​stabilizer theorem from Section 6.5, however, that β and γ are OP-​equivalent, written
#Orb
   G(x) ∙ #StG(x) =​#G, holds in this case. β ~OP γ. In general, two tuples are OP-​equivalent
whenever their equivalence follows from O and P
The preliminary discussion of three-​ voice equivalence in any combination; that is, σ ~OP τ
chord space in the previous section suggests if there exist some tuples μ1, μ2, …, μk such that
that permutational equivalence introduces some σ ~ μ1 ~ μ2 ~ ⋯ ~ μk ~ τ, where each ~ in the chain
geometric complications. These will be explored is either ~O or ~P.32 In the above example the OP
more thoroughly in Chapter 11, but a few general equivalence of β and γ follows from the fact that α
observations may be made now. We have noted =​ (C4, E4, G4) is octave-​equivalent to β and permu-
the exceptional nature of pitch triples with dou- tationally equivalent to γ.
blings. An ordered triple (x, y, z) contains a dou- O equivalence or P equivalence by itself
bling only if it lies on one of the planes x =​ y, x =​ implies OP equivalence, so the three tuples
z, or y =​ z in ℝ3. If those three planes are excluded α, β, and γ are all OP-​equivalent and therefore
from consideration, points contain no doublings belong to the same OP-​class, which could be
and orbits under the action of 𝒮3 always contain called αOP (or βOP or γOP). OP-​classes in general
six elements. The quotient space cpitch3/​~P are larger than either O-​classes or P-​classes; in

408 • E x p l o rin g M u sica l S paces

terminology from Section 2.4, OP equivalence of whose components are the same, or, regarded
is a coarser equivalence relation than either O as a vector, it must be a multiple of the vector (1,
or P equivalence. Regarded as pitch-​class sets, 1, 1). In the case of the triples α =​(0, 4, 7) =​(C4,
triples α, β, and γ all reduce to {0, 4, 7}. In fact, E4, G4) and δ =​(2, 6, 9) =​(D4, F♯4, A4) in the list at
OP-​classes correspond closely with the familiar the beginning of this section, the difference δ − α
concept of pitch-​class set, in which octave dis- is the triple (2, 2, 2). From this we may conclude
placements and note order are disregarded. The that δ =​ T2(α) and therefore that α ~T δ.34
only qualification to this correspondence is that The discrete transposition group 𝒯∞ =​{Tn | n
doubled notes are also routinely disregarded in ∈ ℤ} introduced in Chapter 5 is an infinite cyclic
pc set analysis (for example, the pitch triple (C4, group, isomorphic to the additive group (ℤ, +​).
G4, C5) is associated with the two-​note pc set {C, The continuous transposition group {Tr | r ∈ ℝ},
G} rather than with the three-​note multiset {C, or {Tr} for short, is not cyclic. In fact, this group is
C, G}), so pc set theory implicitly assumes car- isomorphic to the additive group (ℝ, +​). The iso-
dinality equivalence ~C as well. We may say that morphism is given by Tr → r; the equation Tr ● Ts =​
OP-​classes correspond to pitch-​class multisets, Tr+​s confirms that this mapping is a homomor-
or to pitch-​class sets as long as no pitch-​class phism. (In the transposition group {Tr}, as usual
doublings are involved. with group actions, the group operation is composi-
tion of functions.) This continuous group {Tr} is the
Exercise 10.3.4 Determine whether the following symmetry group for transpositional equivalence.35
statements about OP equivalence are true or false. The same transpositions Tr that act on cpitch act
For each true statement, identify a tuple that is also on cpitch3 (or cpitchn), so the transposi-
octave-​equivalent to one of the two given tuples tion group {Tr} is independent of the number of
and permutationally equivalent to the other.33 dimensions of the space, in contrast to the symme-
try groups underlying octave and permutational
(a) (0, 4, 7) ~OP (28, 31, −24). equivalence, which grow larger and more complex
(b) (0, 0, 7) ~OP (7, −24, 31). as the number of dimensions increases.
(c) (F4, G4, A4, B4) ~OP (B2, A3, G4, F5). We described the vertical axis in Figures
(d) (F4, G4, F4, G4) ~OP (F3, F4, G4, G5). 10.1.1 and 10.2.1 as the axis of transposition,
(e)  (F4, G4, G4, A4) ~OP (A4, G4, F4, G4). because all triples lying along that axis are
related by transposition. Any other line parallel
—​—​—​ to the central axis of those figures is also a line
of transposition: the orbits of the action of the
Transpositional equivalence is, of course, defined transposition group {Tr} on the space cpitch3
in terms of transposition operators. Many prop- are straight lines. These lines are T-​classes; the T-​
erties of the transpositions Tn on discrete pitch class αT is the vertical line through α. The lines of
space pitch were studied in Chapter 5. In con- transposition are vertical because of the tilt we
tinuous pitch space cpitch, transpositions take imposed on the coordinate system when draw-
the form Tr, where r, the interval of transposi- ing the figures; in a traditionally oriented coor-
tion, may be any real number, not necessarily an dinate system with horizontal and vertical axes,
integer. The characteristic behavior of transposi- transpositional orbits would be diagonal lines,
tions remains the same: for any pitch x and any though they would still all be parallel to the vec-
interval of transposition r, the equation Tr(x) =​ tor (1, 1, 1) and to each other.
x +​ r holds. In the quotient space cpitch3/​~T, or T-​space,
In multiple dimensions, a transposition opera- each line of transposition reduces to a single point.
tor Tr acts on pitch tuples by transposing every We may locate a set of such points representing
component of the tuple by the same interval T-​space by projecting onto any horizontal plane
r: that is, Tr(s1, s2, s3) =​(Tr(s1), Tr(s2), Tr(s3)) =​ in Figure 10.1.1 or 10.2.1; these are the planes
(s1 +​ r, s2 +​ r, s3 +​ r). In other words, Tr in three identified previously as the sum-​class planes. In
dimensions is the same thing as transposition by particular, we may take the sum-​class-​0 plane to
the interval (r, r, r) in the product space cpitch3. represent T-​space: every pitch triple in cpitch3 is
Two triples σ =​(s1, s2, s3) and τ =​(t1, t2, t3) are transpositionally equivalent to a unique triple in
transpositionally equivalent (~T) if τ =​ Tr(σ) for the sum-​class-​0 plane.36 Importantly, three-​voice
some r. For this to be true, the interval int(σ, τ) =​ T-​space reduces to a two-​dimensional plane. In
(t1 − s1, t2 − s2, t3 − s3) in pitch-​triple space must be effect, the one-​dimensional orbits of the transpo-
of the form (r, r, r): it must be an ordered triple all sition group {Tr} are collapsed by the equivalence

Introduction to Voice-Leading Spaces • 409

relation ~T, leaving one fewer dimension in the the more precise concept of T-​classes of pitch-​
quotient space. In four-​dimensional pitch space, class sets. (As pc sets, the above triads are {0, 4,
the tuples of sum class 0 (or any other fixed 7} and {2, 6, 9}: not the same pc set, but trans-
sum class) form a three-​dimensional subspace. positionally related sets.) This sense of chord
In general, n-​voice T-​space cpitchn/​~T is not quality does not correspond to the set classes of
dimensional but (n − 1)-​dimensional. In one-​
n-​ pitch-​class set theory, which incorporate inver-
dimensional pitch space cpitch1, all points are sional equivalence also. Inversion preserves set
transpositionally equivalent, so one-​voice T-​space class, but it does not preserve chord quality; a
reduces to a single point—​ a zero-​dimensional major triad cannot be OPT-​equivalent to a minor
space. This reduction in dimensionality results triad. As in the discussion above regarding OP-​
from the straight-​line orbits of the transposition classes, a qualification about doublings applies:
group {Tr}, and is related also to the continuous OPT-​classes correspond to T-​classes of pitch-​
nature of the group—​very different from the dis- class multisets, or to T-​classes of pitch-​class sets
crete symmetry groups for ~O and ~P, discussed as long as no pitch-​class doublings are involved.
above, and ~I, to be considered shortly.
Exercise 10.3.7 Four of the following five pairs
Exercise 10.3.5 Determine whether the of pitch tuples are related by ~OPT —​that is, by
following statements about transpositional some combination of octave, permutational,
equivalence are true or false. For each and transpositional equivalence. Connect
true statement, determine the interval of each such pair by a chain of ~O, ~P, and ~T
transposition relating the two tuples.37 relations similar to the chain above, introducing
intermediate tuples as needed. The tuples in one
(a) (0, 4, 7) ~T (−7, −4, 0). OPT-​equivalent pair are actually PT-​equivalent,
(b) (0, 4, 7) ~T (−32 3 , 1 3 , 31 3 ). so ~O is not needed in the chain; which pair is
(c) (0, 0, 7) ~T (−7, 0, 0). this? Which pair is not related by ~OPT?39
(d) (C4, E4, G4) ~T (A5, C♯5, E5).
(e)  (F4, G4, A4, B4) ~T (D6, C6, B♭5, A♭5). (a) (0, 4, 7) and (−2, 3, 7)
(b) (0, 0, 7) and (0, 5, 17)
Exercise 10.3.6 (c) (C4, E4, G4) and (F3, A♭3, C4)
(a) What pitch triple in the sum-​class-​0 plane (d) (F4, G4, A4, B4) and (D6, C6, B♭5, A♭5)
is transpositionally equivalent to the (e)  (G3, B3, C♯4, F4) and (D4, E4, A♭4, B♭4)
augmented triad (0, 4, 8)?
(b) What pitch triple of sum class 0 is —​—​—​
transpositionally equivalent to the
C-​major triad (0, 4, 7)? Inversional equivalence is almost always consid-
(c) In cpitch3, write a general formula for ered in conjunction with transpositional equiva-
the pitch triple of sum class 0 that is lence, as was the case in Chapters 2 and 5, where
transpositionally equivalent to (s1, s2, s3). we described the relation ~TI but not a relation
(d) In cpitch4, what pitch tuple of sum class ~I. Inversional equivalence by itself is simple in
0 is transpositionally equivalent to the conception but narrow in application, because
  dominant seventh chord (0, 4, 7, 10)?38 it must be defined in reference to a single (and
arbitrary) axis of inversion in pitch space. A
The triples (0, 4, 7) =​(C4, E4, G4) and (9, 14, pitch triple σ =​(s1, s2, s3) is therefore inversion-
−6) =​(A4, D5, F♯3), though they differ in pitch-​ ally equivalent to only two things. Equivalence
class content, in register, and in ordering, are relations must be reflexive, so σ must be equiva-
both major triads: they are chords of the same lent to σ itself. The other is the triple −σ =​ I(σ) =​
quality. They are also OPT-​equivalent, as the fol- (I(s1), I(s2), I(s3)) =​(−s1, −s2, −s3), in which every
lowing chain of equivalences demonstrates: component has been inverted via the inversion
operator I. As usual in pitch space, I denotes I0,
(0, 4, 7) ~O (12, −8, 7) ~P (7, 12, −8) the fixed inversion about pitch 0 (C4), described
~T (9, 14, −6). by the equation I(x) =​−x. As with transposition,
I acts on pitch triples in cpitch3 by acting the
OPT-​classes of pitch tuples correspond closely same way on each component pitch class.40
to the usual notion of chord quality, to the more Of the pitch triples listed at the beginning of
informal understanding of “chord type,” or to this section, α =​(0, 4, 7) =​(C4, E4, G4) and ε =​(0,

410 • E x p l o rin g M u sica l S paces

−4, −7) =​(C4, A♭3, F3) are I-​related, and therefore those types. The same properties hold in mul-
inversionally equivalent, because I(0) =​0, I(4) =​ tiple dimensions. Pitch triples σ =​(s1, s2, s3) and
−4, and I(7) =​−7. No other triple can be related τ =​(t1, t2, t3) are TI-​equivalent if and only if either
to either α or ε by ~I; the I-​class αI is the two-​ τ =​ Tr(σ) or τ =​ Ir(σ) for some real number r. If
element set {α, ε}. Every I-​class consists of such τ =​ Tr(σ), then σ and τ are T-​equivalent (which
a pair of I-​related pitch triples, with one excep- implies TI equivalence). If τ =​ Ir(σ), then σ and τ
tion: the triple (0, 0, 0) is its own I-​inversion and are TI-​equivalent but not T-​equivalent (and not
therefore occupies an I-​class by itself. The sym- I-​equivalent unless r =​0). The symmetry group
metry group for inversional equivalence is the for TI equivalence is {Tr, Ir | r ∈ ℝ}, or {Tr, Ir} for
group {E, I} of order 2, consisting of the identity short. This is a continuous analog of the infi-
E and the inversion I. We know from Chapter 6 nite dihedral group 𝒯ℐ∞ (or 𝒟∞) introduced in
that all groups of order 2 are isomorphic to the Section 5.5; it is sometimes known as the gener-
cyclic group 𝒞2, or equivalently ℤ2, the additive alized dihedral group of the real numbers, 𝒟ℝ.
group {0, 1} of integers mod 2. The symmetry
group for inversional equivalence, like that for Exercise 10.3.8 Determine whether the
transpositional equivalence, is independent of following statements about TI equivalence are
the dimensionality of the space on which it acts. true or false. For each true statement, identify a
One might suppose that inversional equiva- transposition Tr or inversion Ir that relates the
lence could be defined more inclusively so that two tuples. Do any of the statements remain
triples σ =​(s1, s2, s3) and τ =​(t1, t2, t3) are inver- true if ~TI is replaced with ~I?42
sionally equivalent if they are related by inver-
sion about any pitch; that is, we might suppose (a) (0, 4, 7) ~TI (0, 3, 7).
σ ~I τ to be true whenever τ =​ Ir(σ) =​(r − s1, r − s2, (b) (0, 4, 7) ~TI (7, 3, 0).
r − s3) for any real number r. But Exercise 2.4.11e (c) (0, 0, 7) ~TI (0, 0, −7).
showed that this definition fails to produce an (d) (F4, G4, A4, B4) ~TI (B4, A4, G4, F4).
equivalence relation. This definition implies logi- (e)  (F4, G4, F4, G4) ~TI (B4, A4, B3, A3).
cally enough that C-​major and C-​minor triads are
I-​equivalent, and that C-​minor and F-​major tri- OPTI equivalence, the combination of the
ads are I-​equivalent. But transitivity would then four equivalence relations considered so far, cap-
imply that the C-​major and F-​major triads must tures the concept of set class from pitch-​class set
be I-​equivalent as well, whereas of course these theory as long as no doublings are involved. For
two major triads are related by transposition, example, the triples (0, 4, 7) =​(C4, E4, G4) and (9,
not inversion. As we know from Chapter 5, two 14, −7) =​(A4, D5, F3), their various differences
inversions combine to produce a transposition, notwithstanding, are a major and a minor triad,
so if we are to allow inversion about more than so as pitch-​class sets they both belong to set
one axis we must allow transposition also. The class 037. The chain below confirms that these
equivalence relation obtained by extending the triples are OPTI-​equivalent:
proposed definition above to ensure transitivity
is not I equivalence but rather TI equivalence.41 (0, 4, 7) ~I (0, −4, −7) ~T (9, 5, 2) ~P (9, 2, 5)
Compared with strict inversional equivalence, ~O (9, 14, −7).
TI equivalence is more familiar, more useful, and
more complex. In continuous pitch space, trans-
position operators Tr (described by the equation Exercise 10.3.9 Four of the following five pairs
Tr(x) =​ x +​ r) and inversion operators Ir (equal to of pitch tuples are related by ~OPTI. Connect
I ● Tr, and described by the equation Ir(x) =​ r − x) each such pair by a chain of ~O, ~P, ~T, and ~I
continue to obey the properties described in the relations, or as many of these as are necessary.
discrete case in Section 5.5, except that now the Which pair is not related by ~OPTI? Are any pairs
subscripts may be any real numbers. In particu- related by ~OPT or other combinations of three
lar, the formulas given in Theorem 5.5.6 hold in or fewer relations?43
general; for example, inversions Ir and Is about
two different axes combine to produce a transpo- (a) (0, 4, 7) and (0, 4, −3)
sition Ir ● Is =​ Ts−r (5.5.6d), and every inversion Ir (b) (0, 4, 7) and (−18, 1, 22)
is an involution (5.5.6g). The formulas also show (c) (0, 0, 7) and (7, 0, 7)
that any combination of operators Tr and Ir will (d) (C4, E4, G4, C5) and (C4, E♭4, G4, C5)
always produce some other operator of one of (e)  (G3, C4, E4, E4) and (G3, C4, E♭4, E♭4)

Introduction to Voice-Leading Spaces • 411

 As the number of equivalence relations fuse(0, 0, 0, 0, 4, 7, 7, 7) deleting three 0s and
increases, the associated symmetry group =​(0, 4, 7) two 7s
expands, and the resulting quotient space shrinks. fuse(G3, G3, B3, B3, F4) deleting one G3 and
The symmetry group for ~OPTI is an infinite non- =​ (G3, B3, F4) one B3
commutative group generated by all the relations fuse(G3, G3, B3, F4, F4, B3) deleting one G3 and
in all of the groups described above for ~O, ~P, =​ (G3, B3, F4, B3) one F4
~T, and ~I: for pitch triples this group contains fuse(G2, G3, B3, F4) deleting nothing
transposition of individual coordinates by whole =​ (G2, G3, B3, F4)
octaves (elements of the group (12ℤ)3), permu-
tations (elements of 𝒮3), transpositions Tr, the Although the definition says that the repetitions
inversion operator I, and all possible composi- are deleted, it is useful to imagine, as the terminol-
tions of these. In a way that we will describe more ogy suggests, that the repeated notes are fused, or
exactly in Section 11.4, three-​voice OPTI-​space merged. The non-​immediate repetition of B3 in the
corresponds roughly to a continuous version of fourth example is not eliminated; we will see below
the trichordal set-​class space from Figure 3.3.1, that it may be eliminated using ~C and ~P in com-
a compact triangular region in two dimensions bination. Also, the repeated pitch class G in the last
and a vastly smaller space than the unbounded example is not eliminated, which would require ~C
three-​dimensional pitch-​triple space with which together with ~O. The last example also shows that
our study began. (See Figure 11.4.12.) Inversional if there are no immediate pitch repetitions, the
equivalence contributes only a group of order 2, {E, fuse function leaves the original tuple unchanged.
I}, to the large symmetry group—​a smaller contri- Two tuples σ and τ are cardinality-​equivalent if
bution than that of any of the other equivalence fuse(σ) =​fuse(τ). From the above examples, we see
relations. In ways that we will grasp more fully in that the tuples (0, 0, 4, 7) and (0, 0, 0, 0, 4, 7, 7, 7)
Chapter 11, OPTI-​space is therefore about half the are cardinality-​equivalent, because fusing yields (0,
size of OPT-​space, and TI-​space is roughly half the 4, 7) in both cases. Of course, (0, 4, 7) itself is also
size of T-​space. cardinality-​equivalent to both of the original tuples;
every pitch tuple σ is automatically cardinality-​
—​—​—​ equivalent to fuse(σ). In C-​space, the quotient space
Uniquely among the OPTIC relations, cardinality cpitch*/​~C, the three tuples (and many others) col-
equivalence can relate tuples of different sizes, a lapse to one equivalence class, and the triple (0, 4, 7)
property that engenders numerous complications. may be taken to stand for that class.
Musically, however, cardinality equivalence is use- The tuples in the last three of the above
ful and familiar. We rely on it whenever we say, for examples of the fuse function, θ1 =​(G3, G3,
example, that a three-​voice triad and a four-​voice B3, B3, F4), θ2 =​(G3, G3, B3, F4, F4, B3), and θ3
triad with a doubled root can both be “C-​major =​ (G2, G3, B3, F4), yield three different tuples
triads”; in effect ~C allows us to equate multi- upon fusing, so no two of them are cardinality-​
sets with sets, which are more easily described. equivalent. The tuples θ1 and θ2, however, are
Cardinality equivalence is usually invoked in com- PC-​equivalent, because they can be related by
bination with permutational equivalence, and a combination of permutational and cardinality
often octave equivalence as well.44 equivalence:
Taken by itself, cardinality equivalence allows
for adding or dropping immediately repeated θ1 =​(G3, G3, B3, B3, F4) ~C (G3, G3, B3, B3, F4, F4)
pitches in a tuple; that is, one note may split into ~P (G3, G3, B3, F4, F4, B3) =​ θ2.
two identical adjacent notes, or two identical
adjacent notes may fuse into one.45 The formal Moreover, θ1 and θ3 are OC-​equivalent:
definition of cardinality equivalence makes use
of a fuse function defined on all of cpitch*, the θ 1 =​ (G 3 , G 3 , B 3 , B 3 , F 4 ) ~ C (G 3 , G 3 , B 3 , F 4 )
set of pitch tuples of all cardinalities. If σ =​(s1, ~O (G2, G3, B3, F4) =​ θ3.
s2, …, sn) is a pitch tuple, then fuse(σ) is the tuple
obtained from σ by deleting all immediate rep- All three tuples θ1, θ2, and θ3 are OPC-​equivalent. Of
etitions—​that is, by deleting every component course, all three of these tuples reduce to the same
pitch si for which si =​ si−1. For example: pitch-​class set, the set {F, G, B} or {5, 7, 11}. We noted
above that OP-​classes correspond to pitch-​class sets
fuse(0, 0, 4, 7) deleting the second 0 in the absence of doubled pitch classes; OPC-​classes
=​(0, 4, 7) correspond exactly to pitch-​class sets in the general

412 • E x p l o rin g M u sica l S paces

case in which doublings are allowed. Any other tuple (c) (D3, A3, A3, F4, D5, D5) ~C (D3, D3, A3, F4, F4, D5).
(of any cardinality) whose notes include Fs, Gs, and (d) (D2, A2, A3, F4, D5, D6) ~C (D2, D3, A3, F4, F5, D6).
Bs (in any registers) but nothing else will always be (e) (D4, F4, D4, F4, A4, F4, A4, F4)
OPC-​equivalent to the tuples above.   ~C (F4, D4, F4, D4, F4, A4, F4, A4).
Similarly, we noted earlier that OPT-​classes
and OPTI-​classes correspond to chord qualities Exercise 10.3.12 Each of the following five
and set classes, respectively, in the absence of pairs of pitch tuples is related by ~OPC. Connect
doublings. To account for doublings, we may use each pair by a chain of ~O, ~P, and ~C relations
OPTC-​classes and OPTIC-​classes for these pur- (or as many of these as are necessary). Are any
poses. For example, the 9-​tuple pairs related by ~OP, ~OC, or ~PC?47

β =​(−12, −5, 0, 4, 7, 10, 16, 22, 28) (a) (0, 4, 7) and (−17, −5, 0, 4, 7, 12, 16)
=​ (C3, G3, C4, E4, G4, B♭4, E5, B♭5, E6) (b) (0, 0, 7) and (0, −5, 7, −5, 0)
(c) (E♭4, G4, B♭4) and (E♭3, G3, E♭3, B♭2, E♭3, G3, B♭3, E♭3)
represents the first chord in Beethoven’s Symphony (d) (G4, G4, G4, E♭4, F4, F4, F4, D4)
No. 1, a dominant seventh chord with many octave and (D4, E♭4, F4, G4)
doublings, with the pitches listed in registral order. (e) (F♭2, A♭2, C♭3, F♭3, G3, B♭3, D♭4, E♭4)
The 4-​tuple   and (A♭3, B♭3, C♭4, D♭4, E♭4, F♭4, G4, A♭4)

ω =​(−7, −1, 3, 8) =​(F3, B3, D♯4, G♯4) Exercise 10.3.13 Four of the following five
pairs of pitch tuples are related by ~OPTIC.
is Wagner’s Tristan chord, a half-​diminished sev- Connect each such pair by a chain of ~O, ~P, ~T,
enth chord with no doublings. Dominant and ~I, and ~C relations (or as many of these as are
half-​diminished seventh chords belong to the necessary). Which pair is not related by ~OPTIC?
same set class 0258, and indeed these two tuples Are any pairs related by some combination of
are OPTIC-​equivalent, as the following extended four or fewer relations?48
chain confirms:
(a) (0, 4, 7) and (3, 6, 10, 15, 18, 22)
β=​(−12, −5, 0, 4, 7, 10, 16, 22, 28) (b) (0, 4, 7, 10) and (−16, −9, −7, −4, 0, 3, 12)
~P (−12, 0, 4, 16, 28, −5, 7, 10, 22) (c) (G4, G4, G4, E♭4, F4, F4, F4, D4)
~O (0, 0, 4, 4, 4, 7, 7, 10, 10) and (C4, G3, A3, B3, C4, G3, A3, B3, C4)
~C (0, 4, 7, 10) ~I (0, −4, −7, −10) (d) (A♯3, C♯4, E4, G♯4) and (B♭3, D4, F4, A♭4)
~T (3, −1, −4, −7) ~O (3, −1, 8, −7) (e)  (B♭1, D5, A♭5, F♭6) and (E♭2, G4, D♭5, A5, E♭6)
~P (−7, −1, 3, 8) =​ω.
Mathematically, cardinality equivalence is the
All five of the OPTIC relations appear in this chain, problem child in the OPTIC family. Even if one
and all are essential: the tuples β and ω are not equiv- desires to study tuples with no more than three
alent via any four or fewer of the OPTIC relations. notes, a proper study of cardinality equivalence
requires considering a space such as cpitch≤3 =​
Exercise 10.3.10 Apply the fuse function to cpitch1 ∪ cpitch2 ∪ cpitch3, because some pitch
each pitch tuple as indicated. triples with doublings are cardinality-​equivalent
to pitch pairs or single pitches. To be applied in
(a) fuse(0, 0, 4, 4, 7, 7, 4, 5, 5, 2, 2, −1, −1, −5) a useful way, therefore, ~C must be defined on
(b) fuse(0, −5, −3, −1, 0, −5, −3, −1, 0) cpitch* or at least some space cpitch≤n rather
(c) fuse(G4, G4, G4, E♭4, F4, F4, F4, D4) than on a fixed cpitchn. A quotient space involv-
(d) fuse(F♯4, F♯4, G4, A4, A4, G4, F♯4, E4, D4, D4, E4, ing cardinality equivalence must somehow fuse
F♯4, F♯4, E4, E4) different dimensions, something that is difficult
(e)  fuse(D4, F♯4, A4, A4, A5, A5, F♯5, F♯5) to fathom geometrically. For example, the pitch
triples (0, 0, 7) and (0, 0.01, 7) lie near each
Exercise 10.3.11 Determine whether the other in cpitch3, but in C-​space cpitch*/​~C, the
following statements about cardinality former somehow occupies the same point as the
equivalence are true or false.46 ordered pair (0, 7) in cpitch2, while the latter
can exist only in three or more dimensions. We
(a) (0, 4, 7) ~C (7, 7, 4, 4, 0, 0). will see in Chapter 12 that while distance mea-
(b) (0, 4, 7) ~C (0, 0, 0, 0, 7, 7, 7, 7). sures in spaces involving cardinality equivalence

Introduction to Voice-Leading Spaces • 413

can be defined, they are less intuitive than dis- are less complicated and easier to describe than the
tances in the other OPTIC spaces. There are alge- foregoing description may suggest.
braic complications as well. Uniquely among the The five OPTIC relations may be applied in any
OPTIC symmetries, cardinality equivalence is combination. The set of five relations has 25 =​32
not defined by a group action: no transforma- subsets, ranging from the empty set to the full
tion group maps tuples to all the other tuples to set of five relations. In principle, therefore, 32
which they are cardinality-​equivalent. different composite equivalence relations may be
imposed on pitch-​tuple space, and 32 different
Exercise 10.3.14 The fuse function maps quotient spaces are formed as a result, ranging
a pitch tuple to another tuple to which it is from the full pitch-​tuple space (if no equivalence
cardinality-​equivalent—​but this function cannot relations at all are applied) to the much smaller
belong
   to any transformation group. Why not?49 OPTIC-​space (in which all five are applied). The
foregoing discussion suggests that some of these
The good news is that as long as the tuples or combinations are of greater musical relevance
chords under consideration contain no doublings, than others: inversional equivalence is unlikely
C equivalence is immaterial. We will see in Chapter to be of much interest without transpositional
11 that multisets with doublings often lie on some equivalence, and cardinality equivalence is usu-
sort of boundary of a space, so for example OPC-​ ally applied only in conjunction with permu-
space looks just like OP-​space everywhere except on tational equivalence. These restrictions—​ no
that boundary. Moreover, in some low-​dimensional I without T, and no C without P—​reduce the
spaces of interest, especially those involving chords number of possibilities to 18. Table 10.3.15 enu-
with no more than three voices, even the boundaries merates these 18 quotient spaces and describes

Table 10.3.15. Eighteen quotient spaces of the OPTIC relations (after Callender,
Quinn, and Tymoczko)

no T or I T TI
T-​classes TI-​classes

no P or C pitch-​tuple space: T-​space: TI-​space:

ordered tuples pitch tuples T-​classes of TI-​classes of
pitch tuples pitch tuples

no O P P-​space: PT-​space: PTI-​space:

pitches multisets pitch multisets T-​classes of TI-​classes of
pitch multisets pitch multisets
PC PC-​space: PTC-​space: PTIC-​space:
sets pitch sets T-​classes of TI-​classes of
(chords) pitch sets pitch sets
O O-​space: OT-​space: OTI-​space:
ordered tuples pitch-​class tuples T-​classes of TI-​classes of
pitch-​class tuples pitch-​class tuples
(line species)
O OP OP-​space: OPT-​space: OPTI-​space:
pitch multisets pitch-​class multisets T-​classes of TI-​classes of
classes pitch-​class multisets pitch-​class multisets
(multiset classes)
OPC OPC-​space: OPTC-​space: OPTIC-​space:
sets pitch-​class sets T-​classes of TI-​classes of
pitch-​class sets pitch-​class sets
(chord qualities; (set classes)
chord species)

414 • E x p l o rin g M u sica l S paces

the elements of each.50 Most of the descriptions immediately above capture the same concepts in
simply restate the definitions of the equivalence a less complete way, without allowing for cardinal-
relations or recast them in other familiar terms. ity change—​but these “non-​C ” spaces exhibit geo-
The names “line species” and “chord species” will metric structure of great elegance and beauty, and
be used in our study of scales in Chapter 13. it is these spaces that will be the primary focus of
In analytical work we often switch, tacitly and our study in Chapter 11. The effect of this restric-
with little effort, from one type of space to another. tion is that we can consider only a single cardi-
When we speak of major triads and half-​dimin- nality at a time and that we must allow for the
ished seventh chords we are talking about objects possibility of multisets containing doubled notes.
in OPTC-​space. When we describe the relationship
between the pitch-​class sets {G, G♯, B} and {D♭, E, 10.4 NORMAL FORMS IN
F} in the opening measures of Schoenberg’s Piano OPTIC SPACES
Piece, Op. 11, No. 1, we are working in OP-​ or OPC-​
space, but when we describe properties of the set Students of pitch-​class set theory are taught to
class 014 we are talking about an element of OPTIC-​ represent set classes using prime forms. En route to
space. If by the “Tristan chord” we mean specifically calculating a prime form, one generally determines
the chord {F3, B3, D♯4, G♯4}, then the Tristan chord is a normal order (or normal form) for the set. The
a pitch set, an element of PC-​space.51 normal-​order calculation involves reducing pitches
to pitch classes, eliminating repetitions, and reor-
Exercise 10.3.16 dering the pitch classes following an algorithm to
(a) If we speak of a category of chords such as optimize the spans between notes, ensuring that
“major triads with a doubled root” or “half-​ they are “packed to the left” as much as possible.
diminished seventh chords with a doubled These steps invoke O, P, and C equivalence. The nor-
third,” we are implicitly referring to an mal order, in fact, stands for an OPC equivalence
object in which quotient space? What if we class: two pitch tuples will have the same normal
speak more specifically about “a C-​major order if and only if they are OPC-​equivalent. To cal-
triad with a doubled root” or “an eø7 chord culate the prime form one takes the normal order
with a doubled third”? and performs additional steps invoking T​ and I
(b) The opening attack in the last movement of equivalence: transposing the normal order to start
Schubert’s String Quintet in C Major, D. 956, with 0, repeating the calculation for an inverted
consists of a doubled G2, a doubled G3, a single form of the set, and comparing the two candidates
G4, and a single G5. The preceding sentence by the span-​optimization criteria. The prime form
describes an object in which quotient space? thus represents an OPTIC equivalence class.
(c) The opening melody of Chopin’s Sonata The concepts of normal order and prime form
in B Minor, Op. 58, begins with the notes can be extended to define what we will call a
G–​F♯–​D–​B–​F♯. The preceding sentence normal form for a pitch tuple under any possible
describes an object in which space? What if combination of the OPTIC relations. These gen-
we instead describe the melody as G5–​F♯5–​ eralized normal forms become convenient names
D5–​B4–​F♯4? As 6–̂ ​5̂–​3̂–​1̂–​5̂ in a minor key? for OPTIC equivalence classes of all types: just as
(d) The chords (pitch sets) {F3, B3, D♯4, G♯4} and an OPC normal form may be used as the name for
{F3, B♭3, D3, A♭4} are related by inversion an OPC-​class, we may use OT normal forms, PTI
in pitch space. Both of these chords are normal forms, OPTC normal forms, and so on, as
therefore represented by the same element names for the corresponding equivalence classes.
  of which quotient space? These less familiar normal forms are potentially
useful in a variety of analytical and theoretical sit-
OPC-​, OPTC-​, and OPTIC-​space, which appear uations. In Chapter 11 we will use normal forms
in the last row of Table 10.3.15, have intuitive as an aid in identifying fundamental regions, and
appeal as spaces of pitch-​class sets, chord quali- also to label the points of various OPTIC spaces.
ties (T-​classes of pitch-​class sets), and set classes, An elementary application arises, for example,
respectively. The anomalous behavior of cardinal- when examining the trichords of prime form
ity equivalence unfortunately makes the struc- 014 in “Nacht” from Schoenberg’s Pierrot lunaire
ture of these spaces complex and mathematically (some of which were studied in Exercise 1.4.25).
intractable, and the Callender-​Quinn-​Tymoczko These trichords are of two types: the “014 type,”
model does not describe them completely. The transpositionally related to the prime form, and
simpler OP-​, OPT-​, and OPTI-​space in the row the “034 type,” inversionally related to the prime

Introduction to Voice-Leading Spaces • 415

form. The informal language of the preceding parts of the algorithm are familiar from the calcula-
sentence is potentially confusing: there are 24 tion of the usual normal order and prime form; the
trichords of prime form 014, but only twelve of subtleties arise in the decisions about what needs
them are of the “014 type”—​and if 034 is not a to be done, and in what order, when some of the
prime form, what is it? Generalized normal forms OPTIC relations are in force and others are not.
allow us to clarify what we mean: [0, 1, 4]OPTIC, as Calculations involving octave, permutational, and
we shall write it, is the OPTIC normal form for a cardinality equivalence, as in the usual normal order,
class of 24 trichords, but those trichords have two are performed first, in steps 1–​5. Transpositional
different OPTC normal forms, [0, 1, 4]OPTC and [0, and inversional equivalence are saved for steps
3, 4]OPTC. In mm. 1–​2, for example, the trichords 6–​7, but step 7 for ~I requires revisiting steps 1–​5
E1–​G1–​E♭1 and G1–​E♭1–​G♭1 both have the OPTIC for an inversionally related tuple in order to choose
normal form [0, 1, 4]OPTIC, but while E1–​G1–​E♭1 between two competing candidates. Because the
has the OPTC normal form [0, 1, 4]OPTC, G1–​ procedure is designed for application in continuous
E♭1–​G♭1 has the OPTC normal form [0, 3, 4]OPTC, spaces, it is applicable to all tuples in ℝn, interpreted
because without appealing to inversion it can be as pitch tuples in cpitchn. Steps not applicable for
reduced to [0, 3, 4] but nothing simpler. (Neither the normal form under consideration are to be dis-
trichord includes any pitch-​class repetitions, so regarded. For example, step 1, for O+​normal forms,
C equivalence is irrelevant in these calculations; should be skipped in a non-​O context.52
the OPTIC normal forms are also OPTI normal Steps 4 and 5 require choosing among two
forms, and OPTC normal forms match OPT nor- or more candidates in such a way that a cer-
mal forms.) tain quantity (either a number representing a
We use square brackets to denote normal note or an interval between two notes) is “opti-
forms. For example, while αOPT denotes the OPT mized.” This word is interpreted as follows. If
equivalence class of a pitch tuple α, the OPT nor- the quantities being compared are real num-
mal form of α, a particular representative of the bers, “optimized” means “minimized in absolute
class αOPT, is denoted [α]OPT. When tuples are value”—​that is, of the available candidates, we
written numerically, a notation such as [0, 3, should choose the number that is as close to 0
4]OPT indicates that the tuple inside the brackets as possible. If facing a choice between a posi-
is the normal form. We might write, for exam- tive and a negative number equally far from 0,
ple, (7, 3, 6)OPT to stand for the entire OPT-​class choose the positive number. (For example, −3
of which the triple (7, 3, 6) is a member; many is preferred over −5; −5 is preferred over 6; 3 is
other notations for that class are possible, but preferred over −3.) If the quantities being com-
only [0, 3, 4]OPT is its normal form. A bracketed pared are numbers mod 12, “optimized” means
tuple such as [0, 3, 4] without a subscript indi- “as close to 0 as possible in the ascending direc-
cates that the tuple is some sort of normal form, tion”—​that is, measured clockwise in the usual
presumably clear from the context. clock diagram. (For example, 4 is preferred over
For convenience we will use terms such as 6; 6 is preferred over 7; 9 is preferred over 11.)
O+​ space and non-​O space. By an O+​ space we
Steps involving ~O, ~P, and ~C:
mean any OPTIC space whose defining equiva-
lence relations include ~O and possibly others; 1. 
If calculating an O+​normal form, reduce all
a non-​O space is any OPTIC space whose defin- components to numbers mod 12, writing them
ing equivalence relations do not include ~O. Thus in the range 0 ≤ ai < 12. For an O+​ normal
OPC-​space is an example of an O+​space and also form, this step immediately reduces the real
a non-​T space, while PTI-​space is an example of numbers to real numbers mod 12, and all
a T+​space and also a non-​O space. We may com- subsequent calculations are then performed
bine these descriptors, referring for instance to mod 12, with tuples in (ℝ12)n. (In some cases
a PC+​, non-​O space, meaning any OPTIC space the initial tuple may be presented as a tuple
whose defining relations include ~P and ~C but of pitch classes rather than pitches; then an
not ~O . Analogously we may refer to O+​ normal O+​normal form is assumed, and the notes
forms, non-​T normal forms, and so on. should be written as mod-​ 12 numbers
The various normal forms of a pitch tuple α =​ throughout the calculation.)
(a1, a2, …, an) may be calculated by applying steps 2. 
If calculating a P+​normal form, arrange the
1–​7 in the order given below. These calculations are components in ascending order: a1 ≤ a2 ≤ ⋯
illustrated in Table 10.4.1 with the 32 normal forms ≤ an. In a P+​, non-​O normal form, these
of the 7-​tuple α =​(−5, −11, 14, 9, 14, 14, 2). Most components are real numbers representing

416 • E x p l o rin g M u sica l S paces

Table 10.4.1. The 32 normal forms of the pitch tuple α =​(−5, −11, 14, 9, 14, 14, 2)

N OR M A L F OR M PRO C E D UR E

Normal forms involving O, P, and C only

α       =​ (−5, −11, 14, 9, 14, 14, 2) initial pitch tuple α =​(a1, a2, …, an)
[α]O     =​[7, 1, 2, 9, 2, 2, 2] apply step 1 to α: reduce all components mod 12
[α]P      =​[−11, −5, 2, 9, 14, 14, 14] apply step 2 to α: arrange components in ascending order
[α]C  =​[−5, −11, 14, 9, 14, 2] apply step 3 to α: fuse immediate repetitions
[α]OP    =​ [7, 9, 1, 2, 2, 2, 2] apply step 2 to [α]O: arrange ascending; rotate/​optimize by step 4
     =​GAC♯DDDD (normal order of α as a pitch-​class multiset)
[α]OC  =​[7, 1, 2, 9, 2] apply step 3 to fuse [α]O
[α]PC       =​[−11, −5, 2, 9, 14] apply step 3 to fuse [α]P
[α]OPC  = [7, 9, 1, 2] apply step 3 to fuse [α]OP; rotate/​optimize by step 4
     = GAC♯D (normal order of α as a pitch-​class set)
Normal forms involving T but not I
[α]T      =​[0, −6, 19, 14, 19, 19, 7] apply step 6 to α: transpose in pitch space so a1 =​0
[α]OT   =​[0, 6, 7, 2, 7, 7, 7] apply step 6 to [α]O: transpose in pc space so a1 =​0
[α]PT     =​[0, 6, 13, 20, 25, 25, 25] apply step 6 to [α]P: transpose in pitch space so a1 =​0
[α]TC    =​[0, −6, 19, 14, 19, 7] apply step 6 to [α]C: transpose in pitch space so a1 =​0
[α]OPT   =​[0, 2, 6, 7, 7, 7, 7] apply step 6 to [α]OP: transpose in pc space so a1 =​0
(T type of α as a pitch-​class multiset)
[α]OTC   =​[0, 6, 7, 2, 7] apply step 6 to [α]OC: transpose in pc space so a1 =​0
[α]PTC     =​[0, 6, 13, 20, 25] apply step 6 to [α]PC: transpose in pitch space so a1 =​0
[α]OPTC = ​[0, 2, 6, 7] apply step 6 to [α]OPC: transpose in pc space so a1 =​0
(T type of α as a pitch-​class set)
Normal forms involving I but not T
[α]I     =​[5, 11, −14, −9, −14, −14, −2] step 7: compare α and I(α); choose a1 > 0 by step 5
[α]OI    =​[5, 11, 10, 3, 10, 10, 10] step 7: compare [α]O and [I(α)]O; optimize a1 by step 5
[α]PI   =​ [−14, −14, −14, −9, −2, 5, 11] step 7: compare [α]P and [I(α)]P; optimize spans by step 4
[α]IC =​[5, 11, −14, −9, −14, −2] step 7: compare [α]C and [I(α)]C; choose a1 > 0 by step 5
[α]OPI =​[10, 10, 10, 10, 11, 3, 5] step 7: compare [α]OP and [I(α)]OP; optimize spans by step 4
[α]OIC =​[5, 11, 10, 3, 10] step 7: compare [α]OC and [I(α)]OC; optimize a1 by step 5
[α]PIC =​[−14, −9, −2, 5, 11] step 7: compare [α]PC and [I(α)]PC; optimize spans by step 4
[α]OPIC =​[10, 11, 3, 5] step 7: compare [α]OPC and [I(α)]OPC; optimize spans by step 4
Normal forms involving T and I
[α]TI   =​ [0, 6, −19, −14, −19, −19, −7] step 7: compare [α]T and [I(α)]T ; choose a2 > 0 by step 5
[α]OTI =​[0, 6, 5, 10, 5, 5, 5] step 7: compare [α]OT and [I(α)]OT ; optimize a3 by step 5
[α]PTI  =​[0, 0, 0, 5, 12, 19, 25] step 7: compare [α]PT and [I(α)]PT ; optimize spans by step 4
[α]TIC =​[0, 6, −19, −14, −19, −7] step 7: compare [α]TC and [I(α)]TC ; choose a2 > 0 by step 5
[α]OPTI =​[0, 0, 0, 0, 1, 5, 7] step 7: compare [α]OPT and [I(α)]OPT ; optimize spans by step 4
(TI type of α as a pitch-​class multiset)
[α]OTIC =​[0, 6, 5, 10, 5] step 7: compare [α]OTC and [I(α)]OTC ; optimize a3 by step 5
[α]PTIC  =​[0, 5, 12, 19, 25] step 7: compare [α]PTC and [I(α)]PTC; optimize spans by step 4
[α]OPTIC  = [0, 1, 5, 7] step 7: compare [α]OPTC and [I(α)]OPTC; optimize spans by step 4
(TI type of α as a pitch-​class set; traditional prime form of α)

Introduction to Voice-Leading Spaces • 417

 pitches. In an OP+​ normal form, the com- Steps involving ~T and ~I:
ponents will already have been reduced in
step 1 to mod-​12 numbers representing 6. 
If calculating a T+​normal form, first use steps
pitch classes. In step 2 they should simply 1–​5 to calculate the normal form on the
be arranged in ascending order between 0 basis of all applicable relations among ~O ,
and 12; rotations of this arrangement will ~P , and ~C ; then transpose the tuple so that
be considered in step 4. (In some cases the its first component is a1 =​0. In a T+​, non-​O
initial collection may be presented as an normal form, this transposition takes place
unordered set or multiset rather than as in pitch space, so the components are real
an ordered tuple; then a P+​normal form numbers; in an OT+​normal form the trans-
is assumed, and the components should be position takes place in pitch-​ class space
arranged in ascending order in step 2.) and all components remain numbers mod
3. 
If calculating a C+​normal form, apply the 12. Every T+​normal form begins with 0. (A
fuse function, reducing the cardinality of non-​T normal form can begin with 0 only
the tuple as appropriate, to ensure that no incidentally, if the first component deter-
two consecutive components are identi- mined by other criteria happens to be 0.)
cal. In a PC+​normal form, identical com- 7. 
If calculating an I+​normal form, first use
ponents will already have been grouped steps 1–​6 to calculate the two normal forms
together by step 2, so step 3 will eliminate [α] and [I(α)] on the basis of all applicable
all duplications. In a C+​, non-​P normal relations except ~I; then choose between the
form, nonconsecutive duplications may two using the tiebreakers in step 4 (for PI+​
remain. For tuples with no duplicated normal forms) and, if necessary, step 5.
pitch classes, step 3 has no effect, and
every C+​normal form will be the same as The 7-​tuple in Table 10.4.1 has been chosen
the corresponding non-​C normal form. deliberately, if artificially, so that all 32 of its nor-
4. 
Span-​ optimization tiebreaker (P+​normal mal forms are distinct (a property satisfied by only
forms only): When necessary, choose a small percentage of pitch tuples). Notes summa-
among the candidates for a P+​normal form rizing each calculation are included in the table.
by the following procedure: first choose Several of the normal forms, as indicated, may be
the candidate(s) for which the difference described in somewhat familiar ways: [α]OP, [α]OPT,
an − a1 is optimized (as defined above); if and [α]OPTI may be called, respectively, the normal
ties remain, choose the candidate(s) for order, T type, and TI type of α as a pitch-​class mul-
which an−1 − a1 is optimized; and so on.53 tiset, while [α]OPC, [α]OPTC, and [α]OPTIC are the
This step, which implements the “packed-​ corresponding forms of α as a pitch-​class set.54
to-​left” criterion, is applicable only in a P+​ For the non-​T, non-​I normal forms listed in the
context, so the components will already top section of the table, the numbers appearing
have been arranged in ascending order in in the normal forms are pitch numbers (in non-​O
step 2. For an OP+​normal form, where the normal forms) or pitch-​class numbers (in O+​ nor-
components are mod-​12 numbers, all rota- mal forms). As soon as either T or I equivalence
tions (cyclic permutations) of the ascend- is introduced, actual pitches or pitch classes are
ing order must be compared in this way. no longer discernible from the normal form. For
For a non-​O normal form whose compo- O+​normal forms whose coordinates are integers,
nents are real numbers, there are no com- familiar conventions may be observed such as
peting candidates at the first pass through omitting commas, spaces, and brackets, and writ-
step 4, but in the case of PI+​normal forms, ing t and e for 10 and 11. In OP and OPC normal
even in a non-​O context, step 7 will require forms in particular, it is often convenient to replace
a second pass through the span-​optimiza- the numbers with pitch-​class names. For example,
tion tiebreaker in order to compare two the 5-​tuple (2, 6, 8, 10, 14) has the OP normal
inversionally related candidates. form (2, 2, 6, 8, 10), which could be written 2268t
5. 
First-​note tiebreaker (if necessary): If the or DDF♯A♭B♭, and the OPC normal form (6, 8, 10,
previous steps fail to produce a unique nor- 2), which could be written 68t2 or F♯A♭B♭D. In this
mal form, choose the candidate whose first case the span-​optimization tiebreaker produces
component a1 is optimized. If necessary, a different ordering when the doubling is elimi-
consider subsequent components a2, a3, … nated: (2, 2, 6, 8, 10) is preferred over (6, 8, 10, 2,
by the same criteria until the tie is resolved. 2), but (6, 8, 10, 2) is preferred over (2, 6, 8, 10).

418 • E x p l o rin g M u sica l S paces

FIGURE 10.4.2 Lattice of normal forms of the pitch tuple α =​(−5, −11, 14, 9, 14, 14, 2)
 The 32 normal forms in Table 10.4.1 are dis- to calculate α2 =​[α]PT, it should suffice to
played another way in Figure 10.4.2. This graph calculate the P normal form of α1. Show that
takes the form of a subset lattice for the set {O, this supposition is false by showing that [α1]P
P, T, I, C}, similar in concept to the lattice from ≠ α2, and moreover that [α1]P is not the same
Figure 3.4.6a. The complete set of relations is dis- as any of the 32 normal forms of α. In general,
played in the OPTIC normal form (prime form) when calculating [[α]X]Y, it is important that
at the bottom of the lattice, while the empty set all the relations in Y be considered, including
(no equivalences at all—​that is, the original pitch those that were considered previously in the
tuple) appears at the top. The lattice shows the calculation of [α]X.
hierarchical structure of the composite relations (e) Show that [α4]PT =​ α4 and [α5]PTC =​ α5.
more clearly than the table. One consequence of Generally, if X ⊂ Y, then [[α]Y]X =​[α]Y : every Y
this hierarchy is that normal forms may be calcu-   normal form is also its own X normal form.
lated progressively, adding one new relation at a
time, as detailed in Exercise 10.4.5 below. Exercise 10.4.6 Consider the eight ordered
trichords from Schoenberg’s “Nacht” listed
Exercise 10.4.3 In order for all of the 32 below. Make a table showing the O, P, OP, T,
normal forms of α in Table 10.4.1 to be OT, PT, OPT, and OPTI normal form of each.
different, it is necessary that the sixteen I+​ All the OPTI normal forms are the same, and
normal forms in the lower half of the table be because the trichords contain no repeated
the same as the corresponding non-​I normal notes, there is no need to consider C+​ normal
forms of the tuple I(α) =​(5, 11, −14, −9, −14, forms. Whenever two or more trichords share
−14, −2) rather than the corresponding non-​I a normal form, verify that they are indeed
normal forms of α itself (which appear in the related by the appropriate relations, and also
top half of the table). Confirm the entries in verify that they share all other normal forms
the lower half of the table by carrying out the that include those relations (for example, if two
calculation of all the non-​I normal forms of trichords share an OT normal form, then they
I(α) and verifying that all sixteen of them win should be OT-​equivalent, and they should also
the tiebreakers (from steps 4 and 5) over the share an OPT normal form).
corresponding
   normal forms of α.
(a) (E1, G1, E♭1) (m. 1, piano)
Exercise 10.4.4 Calculate the 32 normal forms (b) (B♭1, G♭1, A1) (mm. 1–​2, piano)
of  
the 6-​tuple (−5, −5, 4, 1, 7, −5).55 (c) (E3, G3, E♭3) (m. 4 or m. 8, bass clarinet, or
m. 10, voice)
Exercise 10.4.5 For the α of Table 10.4.1 and (d) (B♭2, A2, G♭3) (mm. 5–​6, bass clarinet, or mm.
Figure 10.4.2, suppose we wish to calculate 6–​7, cello)
normal forms by adding relations in the order T, (e) (G3, B♭3, G♭3) (m. 8, bass clarinet)
P, C, I, O. Let α1 =​[α]T, α2 =​[α]PT, α3 =​[α]PTC, α4 (f) (E♭4, D4, B3) (m. 16, voice)
=​ [α]PTIC, and α5 =​[α]OPTIC. (g) (E2, G2, E♭3) (m. 17, cello)
(h)  (D5, B♭4, D♭5) (m. 19, piano)
(a) Show that [α1]PT =​ α2. That is, if we have
already calculated the T normal form for α, Exercise 10.4.7 An example in Section 10.3
in order to calculate the PT normal form we compared the chords β =​(−12, −5, 0, 4, 7, 10, 16,
may start with the T normal form instead of 22, 28) and ω =​(−7, −1, 3, 8) from Beethoven’s
starting with α itself. Symphony No. 1 and Wagner’s Tristan
(b) Continuing in this way, verify that [α2]PTC =​ respectively. These tuples share the prime form
α3, [α3]PTIC =​ α4, and [α4]OPTIC =​ α5. (OPTIC normal form) [β]OPTIC =​[ω]OPTIC =​0258,
(c) Show also that [α1]PTIC =​ α4 and [α3]OPTIC =​ but they differ in all of the other 31 normal
α5. The general principle illustrated here is forms. Calculate the OPC, OPT, OPTC, OPIC, and
that if X and Y are subsets of {O, P, T, I, C}    normal forms of both tuples.56
OPTI
and if X ⊂ Y, then [[α]X]Y =​[α]Y : calculation
of the normal form based on the larger set Exercise 10.4.8 The pulsing dominant ninth
of relations (Y) may begin with calculation of chord that introduces the E-​minor theme in the
the normal form based on the smaller set (X). development section of the first movement of
(d) Since ~T was considered in the calculation of Beethoven’s Symphony No. 3 (mm. 280–​82) is ε =​
α1 =​[α]T, one might suppose that in order (B1, B2, F♯3, A3, C4, D♯4), or (−25, −13, −6, −3, 0, 3).

420 • E x p l o rin g M u sica l S paces

Each of the following tuples is the normal form of the coordinate system, (0, 0, 0). In general, a line
of ε under some subset of the OPTIC relations. parallel to the central axis of Figure 10.1.1 and passing
Identify an appropriate combination of through a point (a, b, c) is given by the formula z =​ y +​
relations in each case. In some cases more than (c − b) =​ x +​(c − a).
one combination may be possible. 7. Hint for Exercise 10.1.5: Both equations are very
simple, and neither of them involves the variable x.
(a) [0, 12, 19, 22, 25, 28] 8. This redundant system, in which n-​dimensional
(b) [0, 3, 6, 9, 16, 28] coordinates are used to plot points in an (n − 1)-​
(c) [0, 1, 3, 6, 9] dimensional space, is a form of what are sometimes
(d) [0, 2, 3, 6, 9] called simplicial coordinates.
(e) [0, 1, 1, 3, 6, 9] 9. Hints for Exercise 10.1.6: The picture for (a) should
(f) [0, 2, 2, 3, 6, 9] be a triangle the same size as Figure 10.1.4 whose cor-
(g) ABCD♯F♯ ners are the three minor triads from Figure 10.1.1. The
(h)  ABBCD♯F♯ picture for (b) is a smaller triangle; one of its corners
is the point (11, 31 2 , 7). The picture for (c) is not a tri-
NOTES angle but something more complex. Observe that all
of the following points have coordinates that sum to
1. See, for example, Roeder 1994, Straus 2003, and 101 2 : (11, 31 2 , 8), (111 2 , 3, 8), (0, 3, 71 2 ), (0, 31 2 , 7),
Straus 2005b. While Straus’s techniques of “transfor- (111 2 , 4, 7), (11, 4, 71 2 ), and (111 2 , 31 2 , 71 2 ).
mational voice leading” have yielded valuable analyti- 10. A detailed understanding of vectors is not
cal insights, it is fair to note that in the most literal essential here, but loosely speaking, the vector (1, 1,
senses of the terms they often involve neither trans- 1) may be thought of as a movable arrow pointing
formations nor voice leading. “Fuzzy” transposition straight up in Figure 10.1.1, parallel to the central
and inversion operators are relations, not transfor- axis. If the tail of the vector is placed at (11, 3, 7), its
mations, because a given operator, applied to a given tip will be at (0, 4, 8). The vector’s coordinates (all 1s
input, does not produce a uniquely determined out- in this case) are the same as the coefficients in the lin-
put. “Transformational voice leading” relates a note ear equations for the sum-​class planes (x +​ y +​ z =​ d
in one chord to a note in another chord to which it for various values of d). In general, a vector with coor-
is mapped (possibly fuzzily) by the appropriate opera- dinates (a, b, c) is orthogonal to a plane given by the
tion, which may or may not be the note to which it equation ax +​ by +​ cz =​ d.
moves by “voice leading” in the traditional sense. 11. Hints for Exercise 10.1.7: The coordinates of the
2. Relevant publications prior to 2008 include Cohn second point (113 4 , 31 2 , 73 4 ) sum to 11 (mod 12), so
2003, Callender 2004, and Tymoczko 2006 (which intro- you should be able to locate this point somewhere in
duces the OP-​spaces, a subfamily of the OPTIC spaces). the plane of Figure 10.1.4. It is not one of the points
Preliminary aspects of the CQT work were presented at labeled in that figure, but lies midway between two
the John Clough Memorial Conference at the University labeled points. Having located this point in Figure
of Chicago in 2005 and in a special session at the Society 10.1.4, find the corresponding location in the sum-​
for Music Theory annual meeting in 2007. class-​ 11 plane of Figure 10.1.1. The other point,
3. As a matter of graphical convenience, Figure (111 4 , 3, 71 4 ), must lie on a vertical line (parallel to the
10.1.1 shows x, y, and z axes intersecting not at the central axis) somewhere directly below the first point
usual origin (0, 0, 0) of a three-​dimensional coordinate you located. This point also lies on the face of the cube
system, but at the point (11, 3, 7) corresponding to defined by the equation y =​3, and on a boundary of
the B-​augmented triad. The axes correctly show the the sum-​class-​91 2 plane that you drew in Exercise
directions of motion possible in the three mutually 10.1.6b.
perpendicular dimensions. 12. Geometric interpretations of parallel and zero-​
4. This use of the word lattice is distinct from sum voice leading are discussed in Tymoczko 2011,
the directed graphs known as lattices, discussed in 81–​85. Tymoczko refers to zero-​sum voice leading as
Section 3.4. “pure contrary motion.” This term is apt when only
5. Other sorts of expressions such as x2, 2x, xy, or two voices are moving but less so when there are more
log x are not permissible in linear equations. than two, which is why we prefer the “zero-​sum” name.
6. The simpler formula z =​ y =​ x (which also incor- 13. In Figure 10.1.8, the straight-​line path and
porates two independent linear equations) corre- the two components may all be regarded as vectors.
sponds to a line parallel to the central axis. While the The straight-​line path is the vector (1, 0, 1), which is
line z =​ y +​4 =​x +​8 passes through the point (0, 4, the sum of the parallel component ( 2 3 , 2 3 , 2 3 ) and the
8), the line z =​ y =​ x passes through the true origin zero-​sum component ( 1 3 , –​ 2 3 , 1 3 ).

Introduction to Voice-Leading Spaces • 421

 14. Hints for Exercise 10.1.9: Although α and β 23. Dmitri Tymoczko has critiqued the tonnetz at
always lie in (or on the boundary of) the cube, in some some length on these grounds. See the brief discussion in
cases γ may lie outside the cube. The chords given in Appendix C of Tymoczko 2011; see also Tymoczko 2009b
part (d) are the same as those in Figure 10.1.8, but the for further exploration of related topics, and Tymoczko
roles of α and β are reversed, and γ will be different. 2012 for a detailed study of the relationship between gen-
15. Hint for Exercise 10.1.10: To be related to α by eralized tonnetz graphs and voice-​leading spaces.
parallel voice leading, γ must be of the form (xα +​ r, yα +​ 24. Hint for Exercise 10.2.14b: The voice-​leading
r, zα +​ r) for some number r. You must determine what distance between two chords should be defined as
value of r will put γ in the same sum-​class plane with β. the smallest possible total number of semitones of
16. Hints for Exercise 10.2.2: For part (b), there motion required to get from one chord to the other.
are three integer possibilities for the x coordinate of In moving from a C-​major triad to a D-​minor triad, for
a chord (x =​11, 0, or 1), and likewise three possibili- example, one could move the voices either up or down
ties each for y and z, so the total number of lattice to the nearest available note; in this case moving up
points must be 3 ∙ 3 ∙ 3. For (c), note that the x, y, and requires five semitones of motion while moving down
z ranges for the cube do not overlap, so for example a requires seven, so the voice-​leading distance is 5. In
chord in the cube cannot contain two notes from the some cases, such as C major and E major, the most effi-
set {11, 0, 1}. cient voice leading may require one voice to move up
17. Hints for Exercise 10.2.3: To get from c♯ to while another moves down; both must be counted, so
the nearest point on the central axis, the shortest the voice-​leading distance in that case is 2. The voice-​
path is horizontal (perpendicular to the axis), and leading distance and other related distance measures
therefore a zero-​ sum voice leading—​ so the near- will be studied in more detail in Chapter 12.
est point must share the same sum class with the 25. Hint for Exercise 10.2.14d: The greatest distance
triad c♯. In three-​ dimensional space ℝ3, the dis- that any one voice may need to move is 6 semitones.
tance between two points (x1, y1, z1) and (x2, y2, z2) Can you find a pair of multisets requiring that all three
is ( x2 − x1 )2 + ( y2 − y1 )2 + ( z2 − z1 )2. This exercise voices move 6 semitones?
assumes a traditional Euclidean understanding of 26. Tymoczko (2011, 36) refers to pitch tuples as
distance in ℝ3; in Chapter 12 we will consider some “basic musical objects.”
alternative ways of measuring distance in this space. 27. Hint for Exercise 10.3.1: Only two of the state-
18. The spelling in Figure 10.2.5 is Fauré’s. The geo- ments are true.
metric structure under consideration here does not 28. Some area and volume calculations in OPTIC
depend on spelling; the analysis treats the chords as spaces will be considered in Chapter 12. A 12 × 12
pitch-​class sets, effectively disregarding both spelling × 12 cube—​a cubic octave—​is much larger than the
and chord inversion. cubes in Figure 10.2.1 and the other figures presented
19. Hook 2013 includes the Barcarolle example previously in this chapter, which are cubic semitones
as a supplement to a study of a passage from Fauré’s measuring only a single unit on each side. All of
String Quartet, Op. 121, the latter illustrating motion Figure 10.2.1 would fit inside a 5 × 5 × 5 cube, tilted
through a four-​dimensional chord space. along with the coordinate system employed there.
20. A chord in the top cube is a T4 transposition Remember that many of the relationships discussed
of the corresponding chord in the bottom cube, but in conjunction with that figure involved something
the sum classes traverse an entire octave in between. other than octave equivalence.
When a three-​note chord is transposed by T1, its sum 29. To account for the boundary points precisely
class increases by 3; thus the T4 and T8 transpositions in the present example, the fundamental region for
belong to the same sum class as the original chord. octave equivalence may be defined to consist of triples
21. Hint for Exercise 10.2.13: Consider whether the (s1, s2, s3) such that s1, s2, and s3 are greater than or equal
xyz coordinate system relates in a consistent way to to 0 but strictly less than 12. Points on region boundar-
the triads in the cube chain, and whether the graph ies are often exceptional in some way, and for most
depicts a voice-​leading space according to the criteria purposes we may simply call them “boundary points”
described earlier in this chapter. without worrying about whether they are elements of
22. In both parts of Figure 10.2.11, the cubes them- the region or not.
selves appear somewhat distended; in some published 30. This is a slightly different application of the
versions of the Cube Dance graph they are stretched to cycle notation in comparison with Example 6.1.13,
the point of being unrecognizable as cubes. Assuming where the elements appearing in the cycles were the
perfect cubes identical in size, the construction shown objects being permuted rather than numbers speci-
in the figure is not physically possible in three-​dimen- fying the positions of those objects in a tuple. When
sional space without the cubes intersecting each other. a fixed set of n objects is permuted, the same n!

422 • E x p l o rin g M u sica l S paces

permutations may be obtained either by permuting 38. Hint for Exercise 10.3.6: Remember that in
the objects themselves or by permuting their order cpitch3, the sum class of (0, 4, 8) is 12, not 0. Even
positions. We observed this principle in the case of if a pitch triple has integer coordinates, its transpo-
twelve-​tone rows in Section 9.6, where the two groups sitional equivalent in the sum-​class-​0 plane may not.
were called 𝒮12p and 𝒮12o; the two groups are dual to 39. Hint for Exercise 10.3.7: Many different chains
each other in the sense described in Theorem 7.5.24. of ~O , ~P , and ~T relations are possible; the three
The order-​position permutations employed here allow may be combined in any order. You may find it easi-
for the greater generality required in the present situ- est to start with ~T : find a transposition of one of
ation, in which the components of an ordered triple the given tuples that contains the pitch classes of the
will not always be the same three objects but can be other tuple.
any real numbers at all. 40. In the generalized inversion terminology from
31. Hint for Exercise 10.3.2: Four of the statements Section 7.6, the inversion I on cpitch3 is the same as
are true. If two points with doublings are permuta- the inversion Inv[u]i, where u is the referential pitch
tionally equivalent, more than one permutation will triple (0, 0, 0) in cpitch3 and i is the interval (0, 0, 0)
map one to the other. In part (b) there are two such in the interval group ℝ3.
permutations, a 2-​cycle and a 3-​cycle. 41. The choice of C4 as the axis of inversion in the
32. In the case of OP-​equivalent pitch tuples, a sin- definition of ~I is arbitrary. One could define an inver-
gle intermediate tuple can always be found; that is, σ sion operator on cpitch3 that inverts about any other
~OP τ if and only if there exists some tuple μ such that fixed pitch, giving rise to an alternate definition of ~I.
σ ~O μ and μ ~P τ. The general definition of composite Each pitch triple would still have a unique inversion,
equivalence relations of this sort, however, requires and the resulting form of inversional equivalence would
that chains of arbitrary length be considered. be just as limited in application as the relation that we
33. Hint for Exercise 10.3.4: Four of the statements have defined. The relation of TI equivalence would be
are true. unaffected by the change; the definition of ~I depends
34. We know from Chapter 7 that transposition on the arbitrary choice of axis, but the definition of ~TI
may be defined in any interval space, including prod- does not.
uct spaces such as cpitch3. This generalized defini- 42. Hint for Exercise 10.3.8: Three of the statements
tion allows for transposition through any interval (r, are true as written; one of those remains true if ~TI is
s, t) whose components are not necessarily the same, replaced with ~I. Remember that two tuples are not
and in this sense (as in any interval space) any two TI-​equivalent if octave or permutational equivalence
elements are related by some transposition. To be must be invoked to relate them. The two tuples in one
transpositionally equivalent in the sense defined in part of this exercise, however, are both permutation-
this chapter, elements of cpitch3 must be related by ally equivalent and TI-​equivalent. What special prop-
a transposition of the form (r, r, r). erty of this tuple makes this possible?
35. The continuous symmetry groups {Tr} and 43. Hint for Exercise 10.3.9: One pair is related by
{Tr, Ir} described in this section are both examples of ~OPT and two others by ~PTI; only one pair requires all
Lie groups (pronounced “Lee”), a class of groups having four relations. In constructing chains of equivalences
both algebraic and topological structure. you may find it easiest to take the OPTI relations in
36. Figure 10.2.1 appears to show two different reverse order, starting with I, as in the example pre-
sum-​class-​0 planes, but in cpitch3 there is only one. ceding the exercise.
(If the discussion in Section 10.2 suggested otherwise, 44. Twelve-​tone rows are often treated as elements
it is because we sidestepped questions of octave equiv- of OC-​space, in that octave displacements and imme-
alence at that time.) If the C-​augmented triad near the diate note repetitions are disregarded—​an otherwise
bottom of Figure 10.2.1 has coordinates (0, 4, 8), then rather uncommon situation in which cardinality
the E-​augmented triad at the top will have coordinates equivalence is invoked without permutational equiva-
(4, 8, 12). These two triads lie in planes corresponding lence. The OPTIC relations do not account for all the
to sum classes 12 and 24, respectively; the true sum-​ symmetries of twelve-​tone theory studied in Sections
class-​0 plane lies below the bottom of the picture. 9.5 and 9.6.
37. Hint for Exercise 10.3.5: Only one statement is 45. The terms split and fuse were introduced in
true. Remember that this exercise involves transpo- Callender 1998, 224. Callender describes fusing as a
sitional equivalence only; two tuples are not trans- voice-​leading event, in which a D and an E may fuse
positionally equivalent if other operations such as directly into an E♭. Our usage here is stricter: we con-
permutation or octave displacements of individual sider the voice-​leading move by which D and E move
notes are also required in order to express the rela- to a doubled E♭ separately from the fusion of the two
tionship between them. E♭s into one.

Introduction to Voice-Leading Spaces • 423

 46. Hint for Exercise 10.3.11: Only one statement there are some sets for which Forte’s prime form dif-
is true. fers from the others. For example, Straus and Rahn’s
47. Hint for Exercise 10.3.12: One pair is related prime form [0, 1, 5, 6, 8] corresponds to Forte’s [0, 1,
by ~PC. The observation implicit in part (e) is that the 3, 7, 8]). The algorithm given here is most suitable for
pitch-​class content of the “Augurs of Spring” chord our generalization here, in which the left-​packing step
from Stravinsky’s Rite of Spring is exactly an A♭ har- may or may not be combined with other calculations
monic minor scale; see Tymoczko 2002, 78. involving transposition and inversion.
48. Hint for Exercise 10.3.13: One pair of tuples is 54. Rahn 1980, 75–​77, uses the names “Tn-​type”
related by ~OPTC, one by ~PTIC, and one by ~PTI. and “Tn/​TnI-​type” for our “T type” and “TI type.”
49. Hint for Exercise 10.3.14: You will find the 55. Hint for Exercise 10.4.4: There are some duplica-
answer in Section 5.5. tions among the 32; your list should include 26 differ-
50. A briefer version of this table appears in ent normal forms.
Tymoczko 2011, 40. For more extensive commentary 56. Hint for Exercise 10.4.7: Of the ten normal
see Callender, Quinn, and Tymoczko 2008, especially forms specified in the question (not counting the two
the supporting online material. OPTIC normal forms), four are equal to 0258.
51. The names for a few of the OPTIC spaces create
some potential for confusion. PC-​space, accommodat-
ing permutational and cardinality equivalence, is not
SUGGESTED READING
the same thing as “pc space” (pitch-​class space pc), The most comprehensive published exploration of
familiar since Chapter 1; continuous pc space is actu- the OPTIC spaces, in the journal Science (Callender,
ally one-​voice O-​space, cpitch/​~O . Also, we use the Quinn, and Tymoczko 2008), is recommended
general term OPTIC spaces (in which OPTIC may be to readers who can negotiate its many technical
pronounced like the word optic) to refer to the entire aspects. A previous publication in Science, Tymoczko
family of quotient spaces formed from pitch-​tuple 2006, introduces only the OP-​spaces. For additional
space by application of any subset of the O, P, T, I, and commentary on the Callender-​Quinn-​Tymoczko
C symmetries; in contrast, the specific term OPTIC-​ work see Hook 2006 and Hall 2008. Tymoczko 2011
space, with a hyphen (and with the five letters pro- takes a gentler approach than the Science articles,
nounced individually), refers to the particular space in largely eschewing group theory and the origins
which all five symmetries are applied. of the OPTIC symmetries therein and describing
52. The normal-​form algorithm introduces equiva- only a few of the lower-​dimensional OPTIC spaces
lence relations in the order OPCTI, and Table 10.4.1 most relevant to the musical applications discussed
is organized accordingly. In the subscripts identifying elsewhere in the book. Hook 2013 contrasts the
normal forms, we continue to use the mnemonically geometric approach with other methods in com-
convenient OPTIC ordering. parative analyses of a passage from Fauré’s String
53. The left-​packing algorithm given here matches Quartet, Op. 121. The CQT work owes a debt to
that in Rahn 1980, 33–​38, and the first three editions early work on voice leading by John Roeder (1987
of Straus’s Introduction to Post-​Tonal Theory (for exam- and 1994) and engages in complex and profound
ple, Straus 2005a, 35–​38), among other texts. Spans ways with the Fourier balances studied by Ian Quinn
are considered in the order an − a1, an−1 − a1, …, a2 − a1. (2006 and 2007, inspired by Lewin 1959); the lat-
Forte 1973, 3–​5, gives a different algorithm, taking the ter work is not considered here. An important link
spans in the order an − a1, a2 − a1, a3 − a1, …, an−1 − a1. between neo-​Riemannian theory and geometric
Straus 2016, 45–​46, introduces a new wrinkle, choos- theory is Douthett and Steinbach 1998. Douthett
ing as the normal form the ordering that is most packed and Steinbach’s discrete lattices were developed as
to the right in cases where that ordering corresponds generalizations of the tonnetz but depict voice-​lead-
to the inversion of the prime form. For example, for ing distances more faithfully, and several of them
the “ ♯9 chord” C–​E–​G–​B♭–​D♯, our algorithm gives [10, can be located within the central regions of OPTIC
0, 3, 4, 7] as the OP or OPC normal form, matching the spaces.
“normal form” in Rahn and in Straus 2005a; for Forte,
the “normal order” of this set is [3, 4, 7, 10, 0], and Linear equations and vectors, mentioned in Section
for Straus 2016, it is [7, 10, 0, 3, 4]. All are reorder- 10.1, are part of the province of mathematics
ings of the same set, and representatives of the same known as linear algebra. Many texts on linear alge-
OPC-​class. In this case all the algorithms agree that the bra are available; Lay, Lay, and McDonald 2016
prime form (OPTIC normal form) is [0, 1, 4, 6, 9], but offers a good introduction.

424 • E x p l o rin g M u sica l S paces

 11
Spaces IV
The Geometry of OPTIC Spaces

HAVING LAID the groundwork for the study to the reader who wishes to demystify these
of OPTIC spaces in the previous chapter, we are spaces and understand some of the many ways
now in a position to examine these spaces more in which music can move through them.
systematically. Section 11.1 introduces the con- Some cautions are in order. First, most of
cept of orbifold, which comes into play in topo- our attention will be devoted to a few spaces of
logical descriptions of OPTIC spaces. This section particular importance, the OP-​, T-​, PT-​, OPT-​,
also reviews the various spaces whose elements and OPTI-​spaces in two, three, and four voices.
are single notes, straightforward spaces mostly Because the complexities of cardinality equiva-
familiar to us already. The power of OPTIC spaces lence make C+​spaces particularly intractable,
lies in the higher-​dimensional examples, which such spaces will receive only minimal consid-
can model collections of notes of any cardinality eration here; although we continue to refer to
under assumptions corresponding to any desired OPTIC spaces, this chapter deals almost exclu-
subset of the OPTIC relations. Subsequent sec- sively with non-​C spaces, arising from combina-
tions of this chapter take up two-​, three-​, and tions of the other four relations. The disregard
four-​voice OPTIC spaces in turn. As the number for cardinality equivalence is of little conse-
of notes increases, the spaces become progres- quence in two-​voice spaces and some three-​voice
sively more complex, and some aspects of their spaces, but it becomes increasingly significant
geometry can seem abstruse even after inten- as the number of voices increases. A second cau-
sive study. The discussion here, though far from tion is that this chapter, like the preceding one,
encyclopedic, is in many cases more detailed contains occasional references to “distances”
than other published descriptions of the OPTIC between points in a space. Distances in OPTIC
spaces, and it is hoped that the examples and spaces can behave in complex and sometimes
exercises here will provide hands-​on assistance counterintuitive ways, and in fact there is not

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0011
just one way to conceive of distances in these at them in a more global way. The same thing is
spaces. This chapter will not attempt a thorough true of tori in higher dimensions. For example,
study of distance measurements; the topic will two-​voice O-​space, consisting of ordered pairs of
be explored more fully in Chapter 12. Finally, in pitch classes, is a 2-​torus 𝕋 2, a 2-​manifold locally
light of the abstract nature of some of the spaces indistinguishable from a plane ℝ2, and three-​
considered here, readers should not be discour- voice O-​space is 𝕋 3, a 3-​manifold. Spheres 𝕊n of
aged if they find some of the exercises in this various dimensions constitute another familiar
chapter to be quite challenging. class of manifolds. The plane-​like appearance of
a small region on the sphere 𝕊2 (the two-​dimen-
11.1 MANIFOLDS AND sional surface of a three-​dimensional ball) led
ORBIFOLDS; ONE-​VOICE many generations of human beings to assume,
reasonably enough, that the earth was flat.2
SPACES Some combinations of OPTIC relations,
The quotient spaces arising from the O, P, T, and however, yield quotient spaces that are not
I symmetries belong to a class of mathematical quite manifolds. As an example, consider the
spaces known as orbifolds. Orbifolds are a gen- simple if somewhat unfamiliar construction of
eralization of manifolds, mentioned in Section one-​voice I-​space cpitch/​~I shown in Figure
1.2. The precise definitions of manifolds and 11.1.1. As described in Section 10.3, inversional
(especially) orbifolds are technical and will not equivalence, in the absence of transpositional
be given here, but the basic underlying ideas are equivalence, relates pairs of notes symmetrically
not hard to grasp.1 An n-​dimensional manifold positioned about C4 (0). The picture of pitch space
(or n-​manifold) is a space that locally resembles in 11.1.1a shows pitch numbers and the associ-
ℝn, ordinary n-​ dimensional Euclidean space. ated I normal forms, representing I-​classes. The
More precisely, an n-​manifold has the property first-​note tiebreaker in the normal-​form algo-
that about any point, there is some small region, rithm from Section 10.4 gives preference to pos-
called a neighborhood in topological parlance, itive numbers, so pitches 1 and −1 (C♯4 and B3)
that is topologically indistinguishable from a both have the I normal form [1]‌I, here written
neighborhood about a point in ℝn. simply as 1. The I-​class containing any pitch s is
Pitch-​tuple space cpitchn, described in the set sI =​{s, −s} of points with the same normal
Section 10.3, is just ℝn itself, so it is an n-​mani- form; thus sI is the same class as (−s)I. That is, 1I =​
fold by default. O-​space (cpitchn/​~O, or cpcn) is −1I, 2I =​−2I, and so on—​or in musical terms,
an n-​dimensional torus 𝕋 n, which is also an n-​ (C♯4)I =​(B3)I, (D4)I =​(B♭3)I, and so on.
manifold. In one dimension, one-​voice O-​space is While cpitch (Figure 11.1.1a) is the space
the 1-​torus 𝕋  1, which is nothing more than the of all pitches, represented by pitch numbers,
continuous pitch-​class circle cpc. Every point in cpitch/​~I (11.1.1b) is the space of all I-​classes,
this circle has a neighborhood that is topologi- which correspond to I normal forms. In the
cally indistinguishable from a neighborhood of a conversion to I-​space, pitch space is effectively
point in continuous pitch space cpitch; cpc and folded over on itself at the point 0, reducing the
cpitch cannot be distinguished except by looking line to a half-​line. Most points on this half-​line

FIGURE 11.1.1 (a) A path in continuous pitch space; (b) the corresponding path in one-​voice I-​space

426 • E x p l o rin g M u sica l S paces

have neighborhoods indistinguishable from includes ~T. An ordered 1-​tuple does not allow
neighborhoods of points in the line ℝ, but there for permutation and, as there is no possibility
is one exception: the point 0I (or (C4)I) in I-​space of doublings, cannot be reduced to a smaller
is now a boundary point, and a neighborhood of tuple, so permutational and cardinality equiva-
this point does not resemble that of any point lence yield nothing new. Inversional equivalence
in ℝ. Motion is possible in two directions from produces one-​voice I-​space cpitch/​~I, the half-​
any point in ℝ, but there is only one direction line of Figure 11.1.1b. Apart from these spaces,
in which to move from 0I: this point has only only one other space arises from applying OPTIC
a one-​sided neighborhood instead of the usual symmetries to one-​voice pitch space; this space
two-​sided neighborhood. Such a point is called is studied in the following exercise.
a singularity of the space: a point where the local
topology is different from the topology typical Exercise 11.1.2 One-​voice OI-​space, resulting
of most points of the space. The fact that I(0) =​ from the combination of octave and inversional
0 is significant: the singularity of I-​space occurs equivalence, is not the same as any of the spaces
at the point of the quotient space corresponding just mentioned. Describe this space cpitch/​~OI
to the unique fixed point in ℝ of the inversion and draw a picture of it, using OI normal forms
operator I defining the equivalence relation.3 as  
labels.5
The presence of even this one singularity pre-
vents I-​space from being a manifold, but it is a 11.2 TWO-​VOICE SPACES
one-​dimensional orbifold, which may be thought
of as a space like a manifold but with singulari- Spaces whose elements are two-​ note sonori-
ties allowed. While the singularity in one-​voice ties exhibit a significant increase in complexity
I-​space is an isolated point, we shall see that a in comparison with the single-​voice spaces just
space of higher dimension may contain an entire considered. Two-​voice pitch-​tuple space cpitch2,
line or plane of singularities—​but in these cases or pitch-​pair space, is graphed in Figure 11.2.1 in
also, the singularities generally correspond to an ordinary xy coordinate system whose axes are
fixed points of some symmetry operation.4 tilted 45 degrees. In this and many subsequent
The bold arrows in Figure 11.1.1 show a path figures, a small open circle marks the origin of
through pitch space and the image of that path the coordinate system. The diagonal orientation
in I-​space. Imagine that a violinist plays the note of the axes, analogous to the tilt of the x, y, and
F4 on the G string and then begins to slide her z axes in some three-​dimensional figures from
finger away from the bridge. In pitch space, the Chapter 10, ensures that lines of transposition
moving pitch traces a simple downward path are vertical. Three of these lines of transposi-
(leftward in 11.1.1a). Nothing of particular tion are shown, along with four sample points
interest happens when this path passes through on each line; the reader may verify that the pitch
C4: the pitch keeps descending, to B3 and beyond. pairs shown on the same line are always related
If we trace the corresponding path in I-​space, by some transposition Tr. Particular examples of
however, the points reached after C4 are the the transpositions T6 and T21 2 are shown on one
same as those traversed previously, in reverse: line. Four horizontal lines corresponding to sum
remember that (B3)I is the same as (C♯4)I and classes are also shown; these are analogous to
so on. In I-​space, the path hits the singularity the horizontal sum-​class planes encountered in
at C4 and bounces, heading back in the direction Chapter 10.
from which it came. As we shall see, this mirror
boundary property also characterizes many sin- Exercise 11.2.2 The three lines of
gularities in higher-​dimensional spaces. transposition shown in Figure 11.2.1 happen
One-​voice spaces are simple, and there is little to be equally spaced. Explain in musical terms
else to say about them here beyond a brief enu- what
   this means.6
meration. Starting with the line of pitch space
cpitch, octave equivalence produces one-​voice Figure 11.2.1 also identifies a fundamental
O-​space cpitch/​~O , familiar to us as circular region for ~O . We saw in Section 10.3 that a
pitch-​class space cpc. Because all single pitches fundamental region for octave equivalence in
are transpositionally equivalent, one-​ voice three dimensions is a 12 × 12 × 12 cube ori-
T-​space cpitch/​~T reduces to a single point, ented with its edges parallel to the x, y, and
as does every one-​voice space obtained from z axes. Analogously, a fundamental region
any combination of equivalence relations that for octave equivalence in two dimensions is a

The Geometry of OPTIC Spaces • 427

FIGURE 11.2.1 Pitch-​pair space cpitch2, with a fundamental region for ~O

12 × 12 square whose sides are parallel to the x feature. This square consists of ordered pairs
and y axes, and therefore tilted in the view (x, y) for which x and y lie between 0 and 12.
of Figure 11.2.1—​the shaded square in the fig- (Let us not worry about whether the points on
ure. The plane of cpitch2 is tiled by a diagonal the boundary of the square, for which one coor-
grid of such squares; the figure includes one dinate is equal to 0 or 12, belong to the region
complete square and portions of several neigh- or not.) According to step 1 from Section 10.4’s
boring squares. In two dimensions the symme- normal-​form algorithm, ordered pairs in this
try group for octave equivalence is 12ℤ × 12ℤ, range are precisely the ones that can arise as
or (12ℤ)2; every other square in the grid is the O normal forms of pitch pairs. Points in
related by some transposition in this group to any fundamental region for ~O will always be
the shaded square. Consequently every pitch in one-​to-​one correspondence with elements
pair in the plane is O-​equivalent to some point of the quotient space cpitch2/​~O, and there-
in the shaded square; this is the defining char- fore in one-​ to-​
one correspondence with O
acteristic of a fundamental region. The shaded normal forms—​but the shaded square is the
region may be taken as a representation of only fundamental region whose elements’
two-​voice O-​space cpitch2/​~O , which is topo- coordinates are O normal forms. We therefore
logically a two-​dimensional torus 𝕋 2, a 2-​mani- call this particular fundamental region the nor-
fold. (The boxed legend identifies Figure 11.2.1 mal region for octave equivalence. In a normal
as cpitch2, the space of the plane as a whole. region, coordinates match normal forms; there
O-​space cpitch2/​~O corresponds to the funda- are many possible fundamental regions, but
mental region within that larger space.) only one normal region. Later in this chapter
Fundamental regions are not unique: any we will use normal forms to identify normal
12 × 12 square with sides parallel to the coordi- regions for various combinations of OPTIC
nate axes could serve as a fundamental region relations, thereby providing some insight into
for octave equivalence, and it would even be the appearance of the corresponding quotient
possible to construct a fundamental region that spaces.
is not square at all. The shaded square in Figure In some cases, however, the normal region
11.2.1, however, has a particularly noteworthy may be less convenient to work with than some

428 • E x p l o rin g M u sica l S paces

FIGURE 11.2.3 (a) A fundamental region for two-​voice OP-​space; (b) cpitch2 tiled with regions for
OP-​space, showing the dyads BD♯ and D♯F in each region

other fundamental region. Such a situation arises square with horizontal and vertical edges, not
when we consider octave equivalence in combi- a tilted square like the fundamental region for
nation with permutational equivalence. Of all ~O (which appears here in dashed outline). This
two-​voice spaces, OP-​space has the most subtle is a compact and convenient region, but we will
and interesting structure. Figure 11.2.3a iden- see that it is only a fundamental region for ~OP,
tifies a fundamental region for ~OP: an upright not a normal region.

The Geometry of OPTIC Spaces • 429

 It may at first be far from obvious that this sense that OP-​space labels can be deduced from
smaller square is a fundamental region for cpitch2 labels: the two labels for any point are
~OP —​that is, that every point in the plane is OP-​ related by the quotient map from cpitch2 to
equivalent to some point in this square. Before cpitch2/​~OP . Nevertheless, inclusion of both
considering this property, let us take note of the labels is useful for illustrating the relationship
size of this square. The symmetry group for per- between the two spaces.
mutational equivalence in two dimensions is 𝒮2, OP-​classes are multisets rather than sets
a group of order 2 (there are after all only two because in the absence of cardinality equivalence,
ways to permute two notes). We should there- doublings must be accounted for. The four corners
fore expect that two-​voice OP-​space should be of the shaded square are the ordered pairs (0, 0),
half the size of two-​voice O-​space—​and indeed, (6, 6), (12, 0), and (6, −6), two of which belong to
the area of the shaded region in Figure 11.2.3a is the OP-​class {0, 0} (OP normal form CC) and two
exactly half that of the normal region for ~O . The of which belong to {6, 6} (F♯F♯), though only two of
smaller square is split by its diagonal into two the four are doublings in pitch space. The plane of
triangles of equal size; four such triangles could Figure 11.2.3b is tiled with squares the same size as
be assembled to fill the larger square. the shaded square. The corners of all these squares
Figure 11.2.3b shows this ~OP fundamen- represent the same doublings, CC and F♯F♯. The
tal region surrounded by eight other similar figure shows that the OP-​classes BD♯ and D♯F each
squares. Some points in this figure are labeled have exactly one representative in each square—​
in two notations: xy coordinates in cpitch2 as and the same is true of every other two-​note pitch-​
before, and also OP normal forms, written with class multiset (with the caveat, already noted, that
pitch-​class names. Recall from Table 10.3.15 some OP-​classes may appear twice on the bound-
that OP-​ classes are pitch-​class multisets. For ary of a square). This is, again, the defining feature
example, the ordered pair (3, −1) in the shaded of a fundamental region. Within the interior of
square is an element of cpitch2 that could also the shaded square or any one of the other squares
be labeled (D♯4, B3). Its OP-​class is the pitch-​ (which are also fundamental regions), the OP-​
class multiset {3, 11}, whose OP normal form is space label function is one-​to-​one.
[11, 3]OP (the traditional normal order); here 3
and 11 are numbers mod 12 representing pitch Exercise 11.2.4
classes, so this normal form may also be writ- (a) Locate points representing the OP-​class A♯B
ten in the form BD♯. The musical note names are in the fundamental region and in each of
suggestive but disguise the continuous nature of the other square regions in Figure 11.2.3b.
OP-​space, which also contains nearby multisets (b) Also locate points representing the OP-​class
such as {3.1, 10.75}. The OP-​class BD♯ is repre-   C♯G in each square.
sented by many different points in the plane of
cpitch2. Figure 11.2.3b shows nine such points As suggested above, the fundamental region
(small black circles), one in each square, among highlighted in Figure 11.2.3a is not a normal
them the ordered pairs (−1, 3) =​(B3, D♯4), (3, 11) region for ~OP . It cannot be, because it is not a
=​ (D♯4, B4), and (11, 3) =​(B4, D♯4). Points repre- subset of the normal region for ~O (the tilted
senting the multiset D♯F =​{3, 5} are also shown square identified previously). Points in the lower-​
(open circles). right half of the OP fundamental region, such as
In terminology from Section 2.3, the two (3, −1), do not lie within the O normal region, are
labelings of Figure 11.2.3b illustrate two differ- not O normal forms, and therefore cannot be OP
ent label functions defined on the plane of the normal forms either. We have already remarked
figure. The ordered-​pair labeling is the standard that the OP normal form of (3, −1) is not [3, −1]
labeling of a plane in xy coordinates; the val- but [11, 3]. The point (5, 3), in the other half of
ues of this function are ordered pairs in ℝ2. The the OP fundamental region, does correspond
range of the other label function is two-​voice to an O normal form, but not to an OP normal
OP-​space, cpitch2/​~OP, whose points are identi- form: its OP normal form is [3, 5]. The OP normal
fied by OP normal forms. The plane as a whole region in this case takes a surprisingly complex
depicts cpitch2, and the label function from form, revealed in the following exercise.
cpitch2 to ℝ2 is a bijection. The OP-​space label
function, in contrast, is not one-​to-​one, because Exercise 11.2.5
of the multiple appearances of the same normal (a) Plot the ordered pairs (1, 2), (1, 6), (1, 8),
forms. The two labelings are redundant in the (1, 11), (4, 11), (6, 11), (8, 4), and (10, 11)

430 • E x p l o rin g M u sica l S paces

 in cpitch2. Those ordered pairs are in O images. This is why the points representing the
normal form, so the points all lie within the OP-​classes BD♯ and D♯F appear in the left half
O normal region of Figure 11.2.3. of some squares and in the right half of others.
(b) Which of the eight ordered pairs from part There is no mirror reversal in the vertical direc-
(a) are in OP normal form? For the ordered tion: BD♯ lies in the bottom half of every square
pairs that are not already in OP normal and D♯F in the top half. In Exercise 11.2.4 you
form, calculate the OP normal forms, and should have found that the OP-​class A♯B lies on
plot those points also, in the same figure. the opposite side of each square from BD♯ and
(c) Determine the shape of the OP normal D♯F, while the tritone C♯G lies exactly on the cen-
region, keeping in mind that it must include tral axis of every square, midway between the
exactly half the area of the O normal region, left and right edges.
that all OP normal forms must lie within We have encountered a tiling of the plane
it, and that ordered pairs not in OP normal with rectangles in two opposite orientations
  form must lie outside of it.7 previously, in Figure 1.4.18, a tonnetz gener-
ated by semitones and tritones. We observed
Any of the squares in Figure 11.2.3b could on that occasion that the implied topology was
theoretically serve as a fundamental region for that of a Möbius strip—​and two-​voice OP-​space
~OP , but there is an important way in which is topologically a Möbius strip as well. Figure
these squares are not all identical. The squares 11.2.6, based on a figure by Tymoczko, is an
in this figure appear in two different orienta- enlarged picture of the fundamental region for
tions, related to each other as left-​right mirror OP-​space.8 Lattice points representing all 78

FIGURE 11.2.6 Two-​voice OP-​space (after Tymoczko)

The Geometry of OPTIC Spaces • 431

equal-​tempered two-​note pitch-​class multisets Vertical lines in Figure 11.2.6 are lines of
are labeled with their OP normal forms; the grid transposition. The dyads on any such line are
of lattice points, like the underlying coordinate characterized by a fixed interval class, from ic0
system, is tilted on a diagonal.9 on the edges of the strip to ic6 at the center. As
In Figure 11.2.6, as in the preceding figures, with the ic0 boundary edges discussed above,
horizontal lines correspond to sum classes. every interval class other than ic6 appears to be
Because of octave equivalence, these sum classes represented by two different lines of transposi-
are now mod-​12 numbers; both the top and bot- tion, but in the Möbius strip the two segments
tom edges of the figure correspond to sum class are merely two parts of the same line. Moving
0. All the sum-​class-​0 dyads, symmetric about C, up the ic3 line from D♯F♯, for example, we pass
appear on both the top and bottom edges, but in through EG, then exit the figure at the top and
the reverse order, as indicated by the oppositely re-​enter at the bottom of the other ic3 segment
directed arrowheads and on those edges. to continue with FG♯, F♯A, and so on. After two
If we were to wrap the figure around to glue the trips around the strip we will return to the start-
top and bottom together, a twist would have to ing point. The central ic6 line is unique in that
be introduced, characteristic of the Möbius strip. it returns to its starting point after a single
As we remarked in Chapter 1, the twist is not an circumnavigation. This property is musically
intrinsic feature of the musical space but merely familiar: the tritone, which divides the octave in
something required in order to realize the topol- half, is the only interval that may be transposed
ogy of that space within our three-​dimensional to itself by a transposition less than an octave.
Euclidean surroundings. This is the two-​voice counterpart of the property
We encountered the term false boundary in of augmented triads observed in Section 10.2,
Chapter 1 as well. In Figure 11.2.6, the top and whereby sets that divide the octave equally lie on
bottom edges of the square are false boundar- the central axis of a space.
ies of OP-​space: they look like boundaries of the Only the equal-​tempered dyads are labeled
figure, but they are not true boundaries of the in Figure 11.2.6, but OP-​space is continuous,
space. One can move uninterrupted off the top not discrete. In principle both non-​integer sum
of the figure and re-​emerge at the bottom, but classes and non-​integer interval classes are pres-
on the opposite side of the central axis. Points ent here. One could locate, for example, the dyad
on these boundaries are not singularities of consisting of pitch-​class numbers 7.1 and 9.8, at
OP-​space, and there is nothing unusual about the intersection of the horizontal line for sum
the topology in the vicinity of such a point. It class 4.9 and the vertical line for interval class
would be possible to draw OP-​space with the 2.7, near the dyad labeled GA♯ in the figure.
false boundary positioned elsewhere, perhaps The continuous nature of OP-​space is use-
coinciding with the sum-​ class-​
6 line rather ful to keep in mind when studying voice
than the sum-​class-​0 line; in such a picture leading—​ the ways in which chords move in
the points of sum class 0 would not look like space. We define a voice leading in n voices to
boundary points at all. be a path through, or a succession of points
The left and right edges of Figure 11.2.6 con- in, n-​voice pitch-​tuple space cpitchn or any of
tain unison dyads, but those on the left side the OPTIC spaces arising as quotient spaces of
are different from those on the right, so these cpitchn (such as OP-​space). A voice leading is
two edges are not conjoined as the top and bot- not completely determined by its initial and
tom edges are. These sides are true boundaries of final sonorities; rather, a voice leading describes
the space—​the edges of the Möbius strip. The a particular way of getting from one sonority
unison dyads are singularities, with one-​sided to another, through a particular succession of
neighborhoods unlike the neighborhoods of all intermediate states—​sonorities that occur en
other points in the space; these singularities route.11 A path that can be traced as a continu-
make OP-​space an orbifold, not a manifold.10 ous line (not necessarily a straight line) defines
The properties of these boundaries will be a continuous voice leading, while a discrete succes-
explored below, but we note now that the plu- sion of points, not joined by continuous paths,
ral noun boundaries is not quite appropriate: is a discrete voice leading. The voice leadings in
because of the Möbius-​strip topology, the two most familiar musical situations are discrete,
sides of the figure are two segments of the same but as long as the voices move relatively short
line. A Möbius strip has only a single boundary distances from chord to chord, it is conceptually
edge, not two. a straightforward matter to regard discrete voice

432 • E x p l o rin g M u sica l S paces

leadings as subsets of implicit continuous voice upward trajectory through another tritone
leadings, an understanding of which can clarify while the lower voice reverses course and
the underlying geometry. Importantly, while descends; this contrary motion continues
some voice leadings may be described in trans- until γ =​(−1, 15) (B3–​D♯5) is reached. Finally,
formational terms (for example, a parallel voice the upper voice holds D♯5 while the lower voice
leading may be described as a transposition), a ascends through an octave, yielding δ =​(11,
voice leading is not itself a transformation, and 15) (B4–​D♯5).
in many cases there may be no transformation Figure 11.2.8b graphs this voice leading
(that is, no mathematical function defined on in pitch-​pair space cpitch2, where it forms
the space) that captures the essential features three straight line segments linking the four
of a voice leading. mileposts, changing directions at β and γ.
A voice leading in cpitchn automati- Several intermediate points, discussed below,
cally defines a unique voice leading in each are also labeled. We will use the term straight-​
of the OPTIC spaces that are quotient spaces line voice leading for any voice leading that
of cpitchn. Because the quotient maps from traces a straight line through pitch-​tuple space
cpitchn to its quotient spaces are onto but not cpitchn, and the term segmental straight-​
one-​to-​
one, many different pitch tuples can line voice leading for a voice leading like the
map to the same point of a quotient space, and complete one in the figure, formed by joining
therefore a voice leading defined in OP-​space or together several segments of straight-​line voice
some other quotient space does not determine a leading. The three segments will be denoted
unique voice leading in cpitchn. α → β, β → γ, and γ → δ; this arrow notation
The following example explores a voice lead- will be used consistently to denote straight-​
ing in two voices in some detail, first in cpitch2 line voice leadings between the indicated
and then in two-​voice OP-​space. The example is sonorities. Although α, γ, and δ are all BD♯
an artificial one, designed to illustrate some of dyads, they are distinct elements of cpitch2
the complexities that can arise even in appar- and lie in three different square fundamental
ently unpretentious two-​voice spaces. regions from Figure 11.2.3b, while β, which
is not a BD♯ dyad, lies in still another region.
Example 11.2.7: A voice leading in two voices The continuous nature of the voice leading is
Consider the succession of two-​note reflected in the fact that the entire trajectory
sonorities shown in Figure 11.2.8a, consisting from α to δ is a continuous path through
of milepost sonorities α, β, γ, and δ joined by cpitch2, which can be drawn without lifting
continuous voice leading. The first sonority, one’s pen from the paper.
α, is a major third B3–​D♯4. In cpitch2, α may The definition of straight-​line voice leading
be represented by an ordered pair, either (−1, is subtler than it may appear. For one thing, it
3) or (3, −1), depending on which voice we turns out that a straight-​line voice leading may
assign to the first coordinate; we arbitrarily not always look like a single straight line when
choose (−1, 3).12 The corresponding points in graphed in OP-​space or some other quotient
O-​space and OP-​space are given by the normal space; we will nevertheless continue to call it a
forms [11, 3]O and BD♯, respectively; each of straight-​line voice leading as long as the original
these notations is shown in the figure. From path in cpitchn is straight. Also, to be sure
this initial sonority α, both voices ascend that the segment from α to β is a straight line,
in parallel motion through the interval of a we need to know that the two voices ascend in
tritone in continuous pitch space, arriving at perfectly parallel fashion as they move from
another major-​third dyad β =​(5, 9) (F4–​A4). one dyad to the other. If the upper voice moves
The lines joining the dyads indicate that the faster in the first part of the motion and the
voice leading is understood to be continuous, lower voice catches up later on, so that intervals
each voice sliding smoothly through pitch larger than a major third are present along the
space joining one notated sonority to the way, the voice leading will trace some sort of
next. One could, of course, reduce this curve or zigzag path from α to β rather than
continuous voice leading to a discrete voice a straight line. The simplest way to ensure
leading consisting only of the equal-​tempered straight-​line motion is to assume that all voices
dyads playable on the piano, ascending move at a uniform speed between milepost
chromatically. In the next segment of the sonorities, which we will assume to be the case
voice leading, the upper voice continues its unless otherwise specified.13

The Geometry of OPTIC Spaces • 433

FIGURE 11.2.8 (a) A voice leading in two voices; (b) the corresponding path in cpitch2; (c) the cor-
responding path in two-​voice OP-​space

434 • E x p l o rin g M u sica l S paces

 The three segments of straight-​line voice jump from one α ′ to the other is akin to what
leading in Figure 11.2.8b illustrate voice we described in Section 1.4 as “teleporting,”
leadings of three different types. Segment or passing through a “wormhole,” from one
α → β is a parallel voice leading, depicted by side of a tonnetz tile to the other—​but as we
motion along a vertical line of transposition remarked in the context of the tonnetz, the
in cpitch2. The contrary-​motion segment β event is no more mysterious than passing from
→ γ, a horizontal move from (5, 9) to (−1, 15) one side of a map to the other when traveling
in cpitch2, is a zero-​sum voice leading within around the world. No musical features
the sum-​class-​14 line, as ascending motion in distinguish this moment from any other in
one voice is balanced by descending motion in the α → β segment, and the two points labeled
the other. In cpitch2, parallel and zero-​sum α′ are the same point of OP-​space, so the
voice leading always correspond to vertical and teleportation does not indicate any motion
horizontal motion, respectively, as was also at all in that space. Although α → α′ and
the case in the hexatonic triad cube studied in α′ → β appear in Figure 11.2.8c as separate
Section 10.1. Finally, segment γ → δ is oblique line segments, the two together form a single
in two senses: as the word is traditionally used straight line in the Möbius strip, and α → β
with respect to voice leading (one stationary as a whole remains a single straight-​line voice
voice and one moving voice), and also in the leading by definition because of its appearance
sense of a diagonal line in the plane of the in cpitch2.
figure. Voice leading along a diagonal line At β the contrary-​motion portion of the voice
parallel to either the x or y axis in pitch-​pair leading begins. In OP-​space as in cpitch2, the
space is always oblique in both senses; in this path turns left here, now following segment
example γ → δ lies parallel to the x axis because ③, a portion of the sum-​class-​2 line. Another
all its points share the y coordinate 15, signaling region boundary is reached at β′, the dyad
the held note D♯5. C♯C♯. This sonority is an octave (1, 13) in the
The three segments of this path correspond original voice leading, but in OP-​space it is
in a straightforward way to the pitch-​space indistinguishable from a unison. The vertical
motions of the original voice leading. Four edges are true boundaries of OP-​space, and
times, however—​at the points labeled α′, β′, γ′, the behavior of the path at β′ is different from
and γ ′′—​the path crosses boundaries between what we observed at the false boundary point
two of the square fundamental regions for α′. In cpitch2 several points on the path in the
OP equivalence. Things therefore become vicinity of β′ are (3, 11) (labeled β ʺ in Figure
considerably more complicated when the path 11.2.8b), (2, 12), (1, 13) (=​β′), (0, 14), and
is graphed in OP-​space, as in Figure 11.2.8c. (−1, 15) (=​γ). The corresponding OP normal
The square in 11.2.8c is theoretically the forms are BD♯ (βʺ ), CD, C♯C♯ (β′), CD, and BD♯
same size as each of the square regions in (b), (γ), from which we can see that in OP-​space
but has been enlarged for clarity. This figure (11.2.8c) the path bounces at β′ (C♯C♯), heading
superimposes all the path segments in a single rightward again along segment ④, retracing part
fundamental region corresponding to Figure of segment ③ in reverse. The left and right edges
11.2.6. Some of the path segments in 11.2.8c of the figure together form the single edge of
look just like the corresponding segments in (b), the Möbius strip; all the points on this edge are
but others appear in mirror reflection because singularities of OP-​space, and the edge acts as
of the differing orientations of the fundamental a mirror boundary, a higher-​dimensional analog
regions. The resulting path appears to consist of of the unique singularity in the one-​voice I-​
no fewer than seven separate segments, labeled space of Figure 11.1.1b. Like the singularity in
①–​⑦, which we now examine in turn. I-​space, these singularities in OP-​space arise
The path begins at α, the dyad BD♯.14 The from fixed points of a symmetry operation, in
initial segment, segment ① in Figure 11.2.8c, this case the permutation that interchanges the
is vertical as before. When the path hits α′, the two components in cpitch2. The bounce at β′
dyad EG♯, it has reached the false boundary represents a sort of voice crossing in OP-​space.
at the top of the figure. In the Möbius strip The voices do not literally cross in pitch space,
the same point α′ appears also on the bottom as they are an octave apart; the motion from βʺ
edge; from α′ the voice leading continues with (dyad D♯4–​B4 in cpitch2) to γ (B3–​D♯5) describes
segment ②, which captures the final semitone what music theorists would more likely call a
of the parallel ascent to β, the dyad FA. The voice exchange. But if the two moving voices

The Geometry of OPTIC Spaces • 435

were traced in a pitch-​class circle, they would ⑦ are continuations of segments ① and ⑥,
meet here, moving in opposite directions. respectively, at the false boundary, and the
At γ the BD♯ dyad is reached again, and the two specular reflections at the side boundary
oblique voice leading begins, during which one illustrate what straight lines do at mirror
voice ascends through an octave while the other boundaries of orbifolds. Only at β and γ,
holds D♯. The path in 11.2.8c turns sharply at where at least one voice changes its pattern
γ and proceeds along segment ⑤. In 11.2.8b of motion, is there a true change of musical
this segment angled instead; the difference direction.
stems from the two different orientations of The path in OP-​space calls our attention to
the square fundamental regions in (b), and aspects of the voice leading that we might have
specifically the fact that this portion of the path overlooked in pitch space. For instance, while
traverses a region whose orientation is opposite it was apparent all along that the milepost
to the one we chose as the basis for our map sonorities α, γ, and δ are all BD♯ dyads and
of OP-​space. The oblique motion soon leads therefore all lie at the same point in OP-​space,
to another octave at γ′ (D♯D♯), where the path the path actually visits this point four times:
again bounces off the mirror boundary, this another BD♯ dyad, βʺ, occurs partway through
time obliquely, proceeding along segment the contrary-​motion segment ③, as a minor
⑥. The false boundary at the top of the figure sixth D♯4–​B4. Also, path segments ① and ⑥ cross
is reached again at γ′′, the tritone D♯A. After each other at D♯G, because this dyad occurs on
another top-​to-​bottom wormhole, the path two separate occasions: in segment ① as αʺ, the
continues with segment ⑦, again, until the major third D♯4–​G4, and in segment ⑥ as γ‴, the
final BD♯ dyad δ is reached, coinciding in OP-​    sixth G4–​D♯5.
minor
space with α and γ.
Ricochets like those at β′ and γ′ may be Exercise 11.2.9
likened to billiard balls bouncing off the (a) In Figure 11.2.8b, points α, β, and δ all lie
cushions of a pool table. A physicist might on the same vertical line. Interpret this
refer to these bounces as specular reflections, statement musically.
meaning that they satisfy the property “angle (b) In Figure 11.2.8b, points αʺ, βʺ, and γ′ all
of incidence equals angle of reflection.” At lie on a diagonal line parallel to the y axis.
β′ both angles are 90 degrees, so the path Interpret this statement musically.
exactly reverses course. At γ′ both angles are 45 (c) The complete voice leading visits interval-​
degrees. Specular reflections are characteristic class-​3 dyads four times and also visits
of the behavior of straight-​line voice leadings at interval-​class-​5 dyads four times. Use
mirror boundaries in OPTIC spaces. Musically, Figure 11.2.8c to identify all these dyads.
in each case the voice leading after the ricochet (d) Suppose that instead of listing the lower
merely continues whatever was happening voice first in our ordered pairs, we had
before; in cpitch2 there is no change of adopted the opposite convention, listing
direction at all at β′ or γ′. the upper voice first. How, if at all, would
The path in Figure 11.2.8c appears to consist the paths in Figures 11.2.8b and 11.2.8c be
of seven different line segments, but we still   affected by this change?15
describe it as a single continuous voice leading,
divided into three segments of straight-​line Exercise 11.2.10
voice leading demarcated by the original (a) Consider the voice leading given in Figure
milepost dyads α, β, γ, and δ. Segments ② and 11.2.11a. Draw corresponding paths in

FIGURE 11.2.11 Voice leadings for Exercise 11.2.10

436 • E x p l o rin g M u sica l S paces

 cpitch2 and in two-​voice OP-​space similar the lines of transposition shown in Figure 11.2.1
to those in Figure 11.2.8.16 contains the points labeled (−31 2 , −11 2 ), (−1, 1),
(b) Consider the two-​voice counterpoint in (31 2 , 51 2 ), and (91 2 , 111 2 ). The point (−1, 1) lies on
Figure 11.2.11b. Draw corresponding paths the sum-​class-​0 line, so we may take this point
  in cpitch2 and in two-​voice OP-​space.17 as a representative of the entire T-​class—​but
we could equally well choose the point (−2, 0),
Our discussion in this section has addressed which lies just below (−1, 1) on the x axis, or (0,
pitch-​pair space cpitch2 and two-​ voice OP-​ 2), which lies above (−1, 1) on the y axis. We call
space cpitch2/​~OP; we have yet to consider T (−1, 1) the balanced form of the T-​class, (−2, 0)
or TI equivalence. Recall from Section 10.3 that the x form, and (0, 2) the y form. The y component
T equivalence reduces the dimensionality of a of an x form and the x component of a y form are
space by one. T-​classes appear in cpitch2 (Figure always 0. The y form of a two-​note T-​class is the
11.2.1) as vertical lines of transposition, and to same as the T normal form defined in Section
impose T equivalence is to collapse every such 10.4, so technically the normal region for ~T in
line to a single point, essentially eliminating the Figure 11.2.1 is the y axis—​a one-​dimensional
vertical dimension from the picture. Two-​voice “region” in the two-​dimensional space.
T-​ and TI-​spaces, cpitch2/​~T and cpitch2/​~TI, Figure 11.2.12a shows the line of two-​voice
are therefore one-​dimensional spaces, and the T-​space and several points on that line, with the
same is true of any two-​voice T+​space, possibly balanced forms of many of these points indi-
incorporating ~O and/​or ~P as well. cated and the x and y forms of somewhat fewer.
One way to achieve this collapsing is to proj- The y forms are bracketed in recognition of their
ect each vertical line in cpitch2 onto a point of status as normal forms. The multiple labels
a single referential line, which could be a hori- serve as a reminder that a point on this line is
zontal line like the sum-​class-​0 line or a diago- not a specific pitch pair but a T-​class consist-
nal line like the x or y axis. For example, one of ing of many different pairs—​which is also the

FIGURE 11.2.12 (a) Two-​voice T-​space; (b) two-​voice OT-​space; (c) two-​voice OPTI-​space
(interval-​class space)

The Geometry of OPTIC Spaces • 437

reason why points in T-​space cannot be labeled Figure 11.2.12a, the point moves rightward
with note names. Like all the spaces studied in from (0, 0), passing through the points with x
this chapter, T-​space is a continuous space. All forms (1, 0), (2, 0), and so on. When x reaches
the discrete points shown on the line, including 12, the pitch pair is (12, 0); other labels for this
those with half-​integers in their balanced forms, point in T-​space include the y form [0, −12] and
can arise as T-​classes of equal-​tempered dyads; the balanced form (6, −6). But (12, 0) ~O (0, 0),
for instance, the pitch pair (B3, D4) has balanced so in O-​ or OT-​space (0, 0) and (12, 0) are the
form (−11 2 , 11 2 ), x form (−3, 0), and y form [0, same point. Octave equivalence wraps the line
3]. It should be evident from the picture that of two-​voice T-​space around on itself to form a
two-​voice T-​space cpitch2/​~T is isomorphic to circle in the same way that the line of one-​voice
cpitch—​ordinary one-​voice pitch space.18 pitch space cpitch is converted to the circle of
pitch-​class space cpc. Figure 11.2.12b shows
Exercise 11.2.13 This exercise illustrates a the circle of cpitch2/​~OT. The portion of T-​space
simple property of inversion and permutation shown in 11.2.12a corresponds to two full loops
in two-​voice spaces: an inversion of a pitch around this circle. Points in OT-​space can still be
pair may always be accomplished by combining identified by x forms or y forms, (x, 0)OT or [0,
a permutation with a transposition, and a y]OT respectively, but the coordinates are now
permutation of a pitch pair may always be numbers mod 12; the y forms proceed around
accomplished by combining an inversion with the circle in the opposite direction from the x
a transposition. It follows that in two-​voice T+​ forms. The y form [0, y]OT matches the OT nor-
spaces, I equivalence means the same thing as mal form.19
P equivalence—​or, to put it differently, in two In light of the observations in Exercises 11.2.13
voices, PT, TI, and PTI equivalence all amount and 11.2.14 above, two-​voice OPT-​space, OTI-​
to the same thing. space, and OPTI-​space are all the same thing. This
space is obtained by imposing inversional equiva-
(a) Pitch pairs (5, 3) and (−5, −3) are I-​equivalent. lence onto the circle of OT-​space. Every point on
Show that they are also PT-​equivalent. the right side of the circle in Figure 11.2.12b is both
(b) Pitch pairs (2, 7) and (7, 2) are P-​equivalent. inversionally and permutationally equivalent to a
  Show that they are also TI-​equivalent. point on the left side. For example, the pitch pair
(5, 3) is represented by the point labeled with x
Exercise 11.2.14 Because two-​voice T-​space form (2, 0) and y form [0, 10]OT, which appears on
cpitch2/​~T (Figure 11.2.12a) is isomorphic to the right side of the circle; (5, 3) is T-​equivalent to
one-​voice pitch space cpitch, it should come as (2, 0) and OT-​equivalent to (0, 10). Its inversion
no surprise that two-​voice TI-​space cpitch2/​~TI (−5, −3) and its permutation (3, 5) both appear
is isomorphic to one-​voice I-​space cpitch/​~I, on the left side at the point (10, 0) or [0, 2]OT. In
shown previously in Figure 11.1.1b. Moreover, OPTI-​space the original point and its inversion/​
as noted in the previous exercise, two-​voice permutation are the same: the ordered pairs
TI-​space, PT-​space, and PTI-​space are all the (5, 3), (3, 5), (2, 0), (0, 2), (10, 0), and (0, 10) are all
same space. Draw a picture of cpitch2/​~TI. OPTI-​equivalent. The OPTI normal form for this
Label several points in this space with several OPTI-​class is [0, 2]OPTI. Any other interval-​class-​
representative ordered pairs, including some 2 dyad will be represented by this same point in
that illustrate inversional or permutational OPTI-​space. OPTI-​space effectively conflates the
equivalence.
   two halves of the circle, leaving only a single line
segment of length 6, as shown in Figure 11.2.12c.
It is a straightforward matter to impose According to Table 10.3.15, elements of
octave equivalence on the two-​ voice T-​space OPTI-​space are multiset classes. The numbers
just considered. An intuition similar to that in 0–​6 in Figure 11.2.12c identify those classes: 0
the above exercise might lead us to expect that, stands for the multiset class of unison doublings
because two-​voice T-​space is isomorphic to one-​ {x, x}, while numbers greater than 0 correspond
voice pitch space, two-​ voice OT-​space should to two-​note set classes. Thus the number 2 rep-
be isomorphic to one-​voice O-​space, the pitch-​ resents [0, 2]OPTI, the set class 02, commonly
class circle cpc—​and this is indeed the case. known as interval class 2. In fact, two-​voice
Consider a voice leading starting with the pitch OPTI-​space may be called interval-​class space. It
pair (0, 0) in which the x voice slowly ascends is a continuous space with two singularities, at
while y remains fixed at 0. In the T-​space of the unison 0 and the tritone 6. The voice leading

438 • E x p l o rin g M u sica l S paces

described above, in which one voice holds the and PT-​space described in Exercises 11.2.13 and
pitch 0 while the other rises through an octave, 11.2.14 is a feature unique to two-​voice spaces,
describes a path in interval-​class space starting not valid in higher dimensions.
at 0, moving rightward as far as 6, then bouncing We remarked in the introduction to this chap-
back leftward and ending at 0. If the ascending ter that disregarding cardinality equivalence has
voice continues beyond an octave, the path will no serious consequences in the context of two-​
bounce rightward again upon hitting 0. voice spaces. The reason is simple: two different
pitch pairs can never be cardinality-​equivalent.
Exercise 11.2.15 Consider the five pitch pairs The only effect of cardinality equivalence in
(B3, D4), (D4, B3), (D4, B4), (B3, D5), and (B4, D5). cpitch2 is that a unison doubling (x, x), which
Identify each pair with a point in two-​voice lies on a vertical line through the origin in Figure
T-​space (Figure 11.2.12a), OT-​space (11.2.12b), 11.2.1, is cardinality-​ equivalent to the single
and
  OPTI-​space (11.2.12c).20 pitch x in cpitch. The C normal form of (x, x) is
[x]C, but no other pitch pair has that C normal
Exercise 11.2.16 Consider the three pitch pairs form. Elements of C-​space cpitch2/​~C are there-
(5, 3), (5, −5), and (−4, 18). Identify each pair fore in one-​to-​one correspondence with the ele-
with a point in two-​voice T-​space, OT-​space, and ments of cpitch2 itself; for sonorities with no
OPTI-​space. Use the procedure from Section more than two notes, cardinality equivalence
10.4 to verify that the OPT, OTI, and OPTI does not give rise to any new spaces.
normal
   forms of all three pairs are [0, 2].
11.3 THREE-​VOICE
Exercise 11.2.17 Reconsider the voice leading
examined in Example 11.2.7 and Figure 11.2.8.
OP-​S PACE
Trace the image of this voice leading in two-​voice Three-​voice spaces are more complex than the
space, OT-​space, and OPTI-​space.21
T-​   two-​voice spaces just described, but conceptually
many aspects of their structure are similar. We
Exercise 11.2.18 start our discussion in this section with three-​
(a) Review your picture of dyadic set-​class voice OP-​space, then turn our attention in the
space from Exercise 3.3.4, and compare it next section to spaces incorporating T equiva-
with the spaces considered in this section. lence. It will be helpful here to recall our pre-
(b) Review the picture of one-​voice OI-​space you liminary work in Sections 10.1 and 10.2, which
drew in Exercise 11.1.2. Because two-​voice developed an informal picture of a portion of
T-​space is isomorphic to one-​voice pitch three-​voice OP-​space.22
space, we might expect that two-​voice OTI-​ The description of two-​voice OP-​space in the
space (which is the same as the OPTI-​space previous section remains relevant as we consider
of Figure 11.2.12c) should be isomorphic to its three-​voice counterpart. Two-​voice OP-​space
  one-​voice OI-​space. Is it? may be constructed by starting with a cross sec-
tion depicting sum class 0—​the horizontal line
One principle evident in the above examples segment that marks the bottom edge of Figure
and exercises remains true in higher dimen- 11.2.6—​ and projecting it through a second
sions. Because of the reduction in dimensional- (vertical) dimension to account for other trans-
ity associated with transpositional equivalence, positions of the same dyads. The transpositions
n-​voice T-​space cpitchn/​~T is always isomorphic range not through an entire octave but only
to (n − 1)-​voice pitch space cpitchn−1. Our pri- from T0 through T6: when both notes of a dyad
mary interest here is topological equivalence, but are transposed by Tr, the sum class increases by
we observed in Section 7.4 that T-​space forms a 2r, so under the assumption of octave equiva-
quotient interval space; in fact cpitchn/​~T and lence a T6 transposition of a sum-​class-​0 dyad
cpitchn−1 are isomorphic as interval spaces. The is another sum-​class-​0 dyad. Our study of the
topological isomorphism remains valid if addi- analogous three-​voice space will likewise begin
tional equivalence relations are imposed: for with a cross section containing the chords of
example, n-​voice OPT-​ space is isomorphic to sum class 0. This slice will now be two-​dimen-
(n − 1)-​voice OP-​space, and n-​voice OPTI-​space sional, and the full figure will then be obtained
is isomorphic to (n − 1)-​voice OPI-​space. (We will by projecting it through a third dimension to
see in Chapter 12 that some of these spaces are account for other transpositions. The transposi-
not interval spaces.) The equivalence of TI-​space tions will now range only from T0 through T4: the

The Geometry of OPTIC Spaces • 439

sum class of a three-​note chord increases by 3r relative to it; the projections of the axes onto the
under a Tr transposition, so a T4 transposition plane are shown at the upper right.
leaves the sum class unchanged. Motion along any of the light dashed grid-
The sum-​class-​0 cross section of three-​voice lines in the figure is not motion in a single coor-
OP-​space is the triangular region shown in dinate: motion in one voice cannot take place
Figure 11.3.1. The equal-​tempered chords within within one sum-​class plane. Rather, along any
this region are labeled in the figure, where they gridline, ascending motion in one coordinate is
form a grid of equilateral triangles. Elements balanced by descending motion in another, so
of this space are OP-​classes (pitch-​class multi- that the sum class is unchanged. For example, in
sets), labeled by their OP normal forms, written the motion in O-​space from (0, 4, 8) (CEA♭) at the
with note names. The figure also shows ordered center of the figure to the nearby point (1, 3, 8)
triples in xyz coordinates. These are O normal (C♯E♭A♭), the x coordinate increases from 0 to 1 (C
forms, elements of O-​ space rather than OP-​ to C♯) while the y coordinate decreases from 4 to
space. For example, B♭BE♭, near the bottom of 3 (E to E♭). The z coordinate is unchanged, as the
the figure, is the OP normal form for the pitch-​ line of motion is perpendicular to the z axis. If
class multiset {3, 10, 11}, while the accompany- this path through O-​space is extended further in
ing notation [11, 3, 10] denotes the O normal the same direction, it will lead to (2, 2, 8) (DDA♭).
form of one of the six O-​space representatives This chord, a three-​voice chord with a doubled
of the OP-​class B♭BE♭. The figure with ordered-​ note, lies on a boundary of the figure. Extended
triple labels may be regarded as a portion of the still further, the path will lead to (3, 1, 8); this
sum-​class-​0 plane of O-​space—​a fundamental point of O-​space lies outside the region of the fig-
region for ~OP within that plane. The dual nota- ure, but it is permutationally equivalent to (1, 3,
tion—​multisets in OP-​ space and ordered tri- 8), and in OP-​space it represents the same C♯E♭A♭
ples in O-​space—​is similar to that employed in chord visited previously. The complete path from
Figure 11.2.3b, where ordered pairs in cpitch2 (0, 4, 8) to (3, 1, 8) is a straight-​line voice leading
and two-​note multisets in OP-​space were shown. as defined in Section 11.2 and describes a single
As in Figure 10.1.4, another triangular cross sec- line segment in O-​space, but in OP-​space it hits
tion of a three-​dimensional space, the x, y, and the boundary at DDA♭ and bounces back, in the
z axes corresponding to the O-​space coordinates manner of some of the paths considered previ-
do not lie in the plane of the figure but are tilted ously in two-​voice OP-​space. All three edges of

FIGURE 11.3.1 The sum-​class-​0 cross section of three-​voice OP-​space

440 • E x p l o rin g M u sica l S paces

the figure contain chords with doublings and are Figure 11.3.1 shows only one cross section of
mirror boundaries of OP-​space. three-​voice OP-​space. To obtain the rest of the
space, we must extrude this triangular region
Exercise 11.3.2 Consider moving from (0, 2, along an axis of transposition through a third
10) to (1, 2, 9) in O-​space (B♭CD to AC♯D in OP-​ dimension, forming the prism shown in Figure
space). Locate this path in Figure 11.3.1. If this 11.3.4.24 The perspective of the figure disguises
straight-​line voice leading is extended further the fact that the triangular cross sections are
in the same direction, what are the O-​space equilateral. In the orientation shown, the sum-​
coordinates of the next two lattice points that class-​0 plane from Figure 11.3.1 appears at the
will be reached? What are the corresponding bottom—​and also at the top, as a T4 transposi-
points in OP-​space? Does the path bounce tion of the bottom layer. The top layer is rotated
off one of the boundaries of the space? Write relative to the bottom, evident in the differing
the corresponding voice leading in musical positions of the unison dyads CCC, EEE, and
notation.
   A♭A♭A♭, consistent with the T4 transposition. This
rotation is a higher-​dimensional analog of some-
Figure 11.3.1, the sum-​ class-​0 cross sec- thing we observed in the two-​voice OP-​space
tion of three-​voice OP-​space, resembles Figure of Figure 11.2.6, where the sum-​ class-​
0 line
10.1.4, a region in the sum-​class-​11 plane of segment at the bottom of the figure reappears
the same space. The earlier figure covers a much at the top in the opposite orientation. In two
smaller portion of the space, as it allows for only dimensions, gluing the top and bottom of the
one semitone of motion in each coordinate. The figure together requires a half-​twist and yields a
points at the corners of Figure 10.1.4 are major Möbius strip. In three dimensions, the top and
triads, but there are no major triads among the bottom layers are likewise conjoined, but the
chords identified in Figure 11.3.1, because no gluing now requires only a one-​third twist (120
equal-​tempered major triad belongs to sum class degrees), resulting in a three-​dimensional orbi-
0. The following exercise investigates questions fold that we will call a Möbius prism—​a higher-​
suggested by these observations. dimensional analog of a Möbius strip. As in the
two-​dimensional case, the top and bottom of the
Exercise 11.3.3 figure are false boundaries of the space; one can
(a) Three points near the center of Figure pass continuously from one to the other, allow-
11.3.1, not lattice points, are marked by ing for the twist just described.
open circles and labeled ρ, σ, and τ. Each The walls forming the three lateral faces of
of these points represents a major triad, the prism are mirror boundaries, as the above
some transposition Tr of CEG, but the exploration of the sum-​class-​0 slice suggests. All
interval of transposition r is not an integer. the points on these boundary walls are trichords
Determine O-​space coordinates for ρ, σ, and with doublings, such as DDA♭ in the sum-​class-​
τ, consistent with the O-​space coordinates 0 layer. The vertical line segments at the three
given for the surrounding points in the corners of the prism consist entirely of triple
figure. (There will be fractions in the unisons such as CCC. When the gluing is taken
coordinates.) into account, these three line segments are seen
(b) If you draw the triangular region bounded to be segments of the same line: if the triple uni-
by ρ, σ, and τ at a larger scale and rotate son CCC is transposed continuously upward, it
it 30 degrees clockwise, you should find first traces a path up one corner segment to EEE,
that it matches Figure 10.1.4, except that where it crosses the sum-​class-​0 seam and con-
the coordinates of all the points in the tinues along the next corner segment as far as
figure have been uniformly transposed, A♭A♭A♭, then finally traces the third corner seg-
accounting for the difference in sum class. ment back to CCC. Moreover, the three appar-
What is the interval of transposition that ent boundary walls are three portions of one
relates these two triangles?23 continuous surface; for example, the top of the
(c) There are also three minor triads somewhere front face, the edge labeled CCC–​EEE, contin-
in the region of Figure 11.3.1. Where are ues across the seam to become the bottom of
they, and what are their coordinates? the right rear wall. While the Möbius strip of
Compare the triangle bounded by the three two-​voice OP-​space has only one edge (a mir-
minor triads to the sum-​class-​10 region you ror boundary consisting of doubled notes), the
  drew in Exercise 10.1.6a. Möbius prism of three-​voice OP-​space has only

The Geometry of OPTIC Spaces • 441

FIGURE 11.3.4 Three-​voice OP-​space (after Tymoczko)

a single face (a mirror boundary consisting of augmented triads, is also the central axis of the
trichords with doublings) and a single edge (con- entire space. The major and minor triads in the
sisting of triple unisons). cube chain divide the octave slightly less evenly
This space at last provides a proper setting than the augmented triads, and correspondingly
in which to investigate the questions raised in lie near the central axis but not on it. The labels
Section 10.2. Visible in the middle of the prism in on the triads are OP normal forms, which for
Figure 11.3.4 is a chain of hexatonic triad cubes. major and minor triads always begin with the
The four cubes shown here match the top four of root. If space permitted, it would be useful to
the five cubes in Figure 10.2.1, in approximately display O normal forms here also, ordered triples
the same perspective.25 In the earlier figure the corresponding to the coordinate system used in
bottom cube duplicated the top (hex34) cube constructing Figure 10.2.1; the O normal forms
except for a 120-​degree rotation; we now under- would in some cases show the voice-​leading rela-
stand that OP-​space contains only a single hex34 tionships between nearby triads more clearly
cube, and we see how both the cycle of linked than the OP forms.26 For example, the single-​
cubes (Figure 10.2.11) and the rotation are semitone displacement from a B-​minor to a G-​
manifested in OP-​space. The central axis of the major triad is more apparent from the notation
cube chain, the line of transposition through the (2, 6, 11) → (2, 7, 11) than from BDF♯ → GBD.

442 • E x p l o rin g M u sica l S paces

 The three segments of the line of transpo- related by 120-​degree rotations. These proper-
sition for major triads appear as dashed lines ties refine some of our observations about the
in Figure 11.3.4; this line passes through the cube chain from Chapter 10 and show how they
points ρ, σ, and τ in the sum-​class-​0 plane (from extend to the rest of OP-​space.
Exercise 11.3.3 above), as well as through all Allowing for the duplication of the top and
twelve equal-​tempered major triads in the cube bottom layers, every three-​note pitch-​class mul-
chain. The minor triads in the cube chain lie tiset (OP-​class) appears exactly once somewhere
on a different line of transposition, not shown in the Möbius prism. As an example, let us locate
explicitly. Almost all the vertical lines of trans- the multiset CE♭F ({0, 3, 5}) in this space. Figure
position in the Möbius prism are divided into 11.3.5 illustrates. Because the pitch-​class num-
three segments in this way. The unique excep- bers sum to 8, this point lies somewhere in the
tion is the central axis of augmented triads, sum-​class-​8 plane. To pinpoint its location in
which returns to its starting point after a single that plane, we first calculate a point in the sum-​
T4 transposition. Layers corresponding to all class-​0 plane of which CE♭F is a transposition
possible sum classes are present within the T4 between T0 and T4. To convert the multiset {0,
range of transposition of the prism, as shown 3, 5} of sum class 8 to something transposition-
at the right side of the figure. Again with the ally equivalent but with sum class 0, we must
exception of augmented triads, every transpo- subtract 8 3 , or 22 3 , from each of the three coor-
sition class (that is, every trichordal OPT-​class) dinates, yielding {91 3 , 1 3 , 21 3 }; that is, {0, 3, 5} =​
occurs three times in every layer, the three Tr({91 3 , 1 3 , 21 3 }), where r =​22 3 . (The transposition
chords related by T4/​T8 and the three points will be an integer only if the difference in sum

FIGURE 11.3.5 Locating the trichord CE♭F in three-​voice OP-​space

The Geometry of OPTIC Spaces • 443

classes is a multiple of 3, the cardinality of the there any other sum-​class planes containing
chords.) equal-​tempered trichords at still different
The next step involves locating this trichord   locations?
{91 3 , 1 3 , 21 3 } in the sum-​class-​0 plane of Figure
11.3.1—​a potentially tricky task for two rea- Exercise 11.3.8 Using reasoning similar to
sons. First, the coordinates are not integers, that in the above discussion, locate each of the
so it will not be among the points labeled in following OP-​classes in the Möbius prism of
the figure; and second, while the multiset in Figure 11.3.4.28
question has OP normal form [91 3 , 1 3 , 21 3 ],
the coordinates of the O normal form in the (a) CC♯D
figure may appear in a different order. One of (b) BDF
the small triangles formed by the gridlines in (c) CCD
the right side of Figure 11.3.1 is bounded by (d) CCE
points with O-​space coordinates (0, 3, 9), (1, (e) CDF
2, 9), and (0, 2, 10). The point (1 3 , 21 3 , 91 3 ), a (f) CCF♯
permutation of the desired multiset, must lie (g) C♯E♭G
somewhere within that triangle, because its x, (h) DDG
y, and z coordinates all lie within the ranges of (i)  {1.99, 2.01, 7}
the corresponding coordinates at the corner
points. In fact, this point lies precisely at the Exercise 11.3.9 In a copy of Figure 11.3.4,
center of that small triangle. (The center of an locate the seven triads that can be formed
equilateral triangle lies two-​thirds of the way using notes of the C-​major diatonic collection.
from any vertex to the midpoint of the oppo- Six of them are already labeled in the figure,
site side.) The original multiset CE♭F therefore and the other is the B-​diminished triad from
lies on the line of transposition directly above Exercise 11.3.8b. Join the seven chords with
the point (1 3 , 21 3 , 91 3 ), at the intersection of line segments to show the smooth voice leading
that line of transposition with the sum-​class-​8 characteristic of diatonic triads moving by
plane of OP-​space. Figure 11.3.5 shows these    (C –​a –​F –​⋯).29
thirds
points, as well as the other two points in the
sum-​class-​0 plane transpositionally equivalent Comparison of the locations of the major tri-
to CE♭F (trichords of OPT normal form 035), ads in Figure 11.3.4 and the 035OPT chords in
the three segments of the line of transposition Figure 11.3.5 illustrates a relationship between
for this OPT class, and all the other equal-​tem- the voice-​leading potential of chords of a given
pered chords in this class. type and the locations of those chords in OP-​
space. The other equal-​ tempered 035 chords
Exercise 11.3.6 Determine the coordinates of closest to CE♭F in OP-​space are those directly
the other two points in the sum-​class-​0 plane above and below it on its line of transposition,
that are transpositions of CE♭F, and the note C♯EF♯ and BDE. Moving from CE♭F to one of
names corresponding to each of the other these chords requires all three voices to move
nine equal-​tempered chords shown on the line by semitone, for a total of three semitones of
of transposition in Figure 11.3.5. Which sum motion. Of course, a T1 or T11 transposition of
classes are represented by the equal-​tempered this kind is possible for any trichord. Starting
chords?
   from a C-​major triad, however, two other major
triads can be reached via only two semitones of
Exercise 11.3.7 total motion: an E-​major triad is available by the
(a) Carefully draw a picture similar to Figure semitone moves C–​B, G–​G♯ (corresponding to
11.3.1 that shows the sum-​class-​8 cross the neo-​Riemannian LP) and an A♭-​major triad
section of three-​voice OP-​space. Your figure by the moves E–​E♭, G–​A♭ (neo-​Riemannian PL).
should show the locations of all the equal-​ The major triads in each of the hexatonic triad
tempered trichords of sum class 8.27 cubes in Figure 11.3.4 are related in this way.
(b) In which other sum-​class planes do the Because major triads lie near the central axis
equal-​tempered trichords fall in the same of OP-​space, they automatically lie close (in the
locations as in the sum-​class-​0 plane? In voice leading sense) to their T4 and T8 transposi-
which other planes do they fall in the same tions, to which they are related by rotation about
locations as in the sum-​class-​8 plane? Are that axis. Consequently, major triads may be

444 • E x p l o rin g M u sica l S paces

joined to other major triads by smoother voice Part (b) of the figure shows this voice leading
leading than is possible for 035 chords or most in OP-​space. The path begins at α, the D♭-​aug-
other chord types. (An 035 chord does, however, mented triad at the center of the sum-​class-​3
lie close to one of its inversions in OP-​space, as plane, and ends at η, the E♭-​augmented triad at
can be seen by comparing the locations of CE♭F the corresponding location in the sum-​class-​9
in Figure 11.3.5 and CDF in Exercise 11.3.8e.) plane. Chords γ and ε contain pitch-​class dou-
Meanwhile, the picture you constructed in blings and therefore lie on boundary walls of
Exercise 11.3.9 should confirm that within a dia- the prism, in the sum-​class-​5 and sum-​class-​7
tonic scale, smoother voice-​leading is possible planes respectively. The OP-​space path forms
between third-​related triads than between triads three straight segments α → γ, γ → ε, and ε →
related in any other way. η, reflecting specularly off the walls at γ and ε.
We are now equipped for a better under- Chords β, δ, and ζ lie at the midpoints of the
standing of the voice-​ exchange progression three segments. The three segments are of equal
from Figure 10.2.8 as a voice leading in OP-​ length and slope; the projection of the path onto
space. The progression may be identified with a horizontal plane such as the sum-​class-​0 plane
the straight-​ line zero-​sum voice leading (0, is an equilateral triangle, visible at the bottom
16, 7) → (4, 12, 7) in pitch-​triple space (this of the figure.
notation matches the order of voices in Figure
10.2.8, but the coordinates in the original figure Exercise 11.3.11
assumed octave equivalence). The voice leading (a) The projection of the path in the sum-​class-​
begins at the C-​major triad in Figure 11.3.4’s 0 plane passes through points α′, β′, …,
uppermost cube and passes through the points η′, located directly below (and therefore
C♯E♭G and DDG, identified in parts (g) and (h) of transpositionally equivalent to) α, β, …, η.
Exercise 11.3.8. This much of the path forms a The starting and ending points α′ and η′
straight line segment in the sum-​class-​11 plane coincide at the augmented triad {0, 4, 8} at
of OP-​space, from the C-​major triad to the point the center of the plane. Calculate numerical
DDG on the front face of the prism. At DDG representations (as pitch-​class multisets)
the voice leading hits the mirror boundary at a for the other five points β′ through ζ′ in this
right angle and bounces back, retracing its path plane.
through C♯E♭G and back to the initial C-​major (b) Of the points α′ through η′ in the sum-​
triad. Immediately before and after the bounce class-​0 plane, only α′, δ′, and η′ (the same
at DDG, the path passes through the multiset as α′) are lattice points representing equal-​
{1.99, 2.01, 7} from Exercise 11.3.8i. We noted tempered chords. The path in this plane
in Section 10.2 that “there must be something does, however, pass through two other
unusual about the geometry” in the vicinity of lattice points, one between β′ and γ′ and
a point such as DDG in order for a straight-​line the other between ε′ and ζ′. Determine the
path through that point to pass through the pitch-​class multisets represented by these
same point on both sides. This something is the two lattice points, and determine numerical
mirror-​boundary property—​the singularity of representations for the corresponding
the orbifold. points in the three-​dimensional path above
The voice leading just described takes place (which are not lattice points).
within a single horizontal slice of OP-​space. (c) What will the path in Figure 11.3.10b
Figure 11.3.10 illustrates a voice leading that look like if the voice leading in (a) is
exploits the vertical dimension of the space as transposed by T4?
well. The chords in part (a) are labeled with the (d) What if the voice leading is transposed
first seven Greek letters. Because each of the   by T2?31
three voices moves by semitone in a consistent
direction (the bass descending as the upper two Exercise 11.3.12 In Euclidean space ℝn, only
voices ascend), the motion is a straight-​ line one straight line segment is possible from one
voice leading α → η, or (9, 13, 17) → (3, 19, 23) in given point to another. In orbifolds, things are
pitch-​triple space.30 Ordered-​triple coordinates not so simple. For example, the points α (D♭FA)
are given in cpitch3 (ordering the voices from and η (E♭GB) in Figure 11.3.10 may be joined
lowest to highest) and in O-​space, as well as OP by an infinite number of different straight-​line
normal forms with note names. voice leadings in OP-​space.

The Geometry of OPTIC Spaces • 445

FIGURE 11.3.10 (a) A voice leading in three voices; (b) the corresponding path in three-​voice OP-​space

(a) There is a straight-​line voice leading above suggest several others. Describe a
ascending from α to η along the central family of voice leadings of this type. How
axis of the prism, and another straight-​ many are there, and how are they related to
line voice leading descending from α to η each other musically?
(departing the bottom of the figure upon (c) Construct a straight-​line voice leading from
reaching the sum-​class-​0 plane and re-​ α to η that reflects off a wall only once, and
emerging at the top). Describe each of describe it musically.
those voice leadings musically. (d) Construct a straight-​line voice leading from
(b) The path shown in Figure 11.3.10b, which α to η that reflects off walls more than twice.
reflects twice off the walls of the prism, is (e) Construct a straight-​line voice leading from
another straight-​line voice leading from α to α to η that crosses the sum-​class-​0 plane
η, and parts (c) and (d) of Exercise 11.3.11   more than once before reaching η.32

446 • E x p l o rin g M u sica l S paces

FIGURE 11.3.14 Three-​voice harmonic progression from Saariaho, Vers le blanc

Exercise 11.3.13 Figure 11.3.14 shows a (c) As you consider a path from α through γ
continuous voice leading that takes place to β, you will realize that this path must
over the course of fifteen minutes in Kaija at some point, say δ, cross the sum-​class-​
Saariaho’s 1982 electronic composition Vers le 0 plane. Locate δ by calculating the value
blanc.33 Over this time span, one voice ascends of t for which f1(t) +​ f2(t) +​ f3(t) =​0 (mod
continuously from C3 to E3 while a second voice 12). Does δ occur before or after γ? In your
descends from A3 to D3 and a third descends picture of the path, the point δ will appear
from B3 to F3. In cpitch3, this motion describes at both the top and bottom of the Möbius
the straight-​line voice leading (−12, −3, −1) prism, as usual in two different places in
→ (−8, −10, −7). The two pitch triples are those two cross sections.
ordered to respect the voice leading; the lowest (d) Complete your picture of the path by
notes of the two sonorities appear in different putting the points α, β, γ, and δ in the
components because they are in different proper order and joining them with straight
voices. Follow the steps below to graph this   line segments.
motion in OP-​space.
11.4 THREE-​VOICE T-​, PT-​,
(a) The passage describes a path joining the PTI-​, OPT-​, AND OPTI-​S PACE
multisets α (ABC) and β (DEF). Locate
these two points in the Möbius prism of Three-​voice OP-​space, described in the previous
OP-​space. section, does not account for transpositional
(b) The path from α to β in OP-​space is not the or inversional equivalence. Because T equiva-
most obvious one. It must bounce off a wall lence reduces the dimensionality of a space,
of the prism, because the lower two voices quotient spaces such as three-​ voice T-​space
cross near the end of the time span, at γ in (cpitch3/​~T) and OPTI-​space (cpitch3/​~OPTI)
Figure 11.3.14. To calculate where γ is, it is are two-​dimensional. The reduced dimension-
helpful to describe each of the three voices ality makes these spaces easier to depict on
as mathematical functions of time, as in paper than three-​ dimensional OP-​space, but
Section 2.3. For simplicity, consider the they nevertheless take forms that may be far
duration of the passage to be one time unit, from obvious. In this section we shall explore
so t ranges from 0 to 1 over the course of three-​voice T-​space and a series of other two-​
the voice leading. During this time, the top dimensional spaces obtained by imposing other
voice (the third component in the ordered-​ OPTIC relations on T-​space.34
triple notation in the figure) descends six Our conception of three-​ voice T-​space
semitones from −1 to −7 in pitch space, so depends on viewing the three dimensions of
that voice may be described by the function pitch-​triple space cpitch3 from below. Consider
f3 given by the equation f3(t) =​−1 − 6t. a cube, one semitone unit on each side, with one
Write similar equations for functions f1 corner at the origin (0, 0, 0). If we recall the ori-
and f2 describing the other two voices, and entation of the space cpitch3 as we have been
calculate the value of t for which f1(t) =​ f2(t). considering it, with the coordinate axes tilted,
The coordinates at that time point will and if we view this cube from directly below, it
determine γ, which you should then locate will appear as in Figure 11.4.1a. The origin, in
in OP-​space. (It is not an equal-​tempered the center of the picture, is the nearest point of
trichord.) the cube; the three coordinate axes angle away

The Geometry of OPTIC Spaces • 447

FIGURE 11.4.1 (a) A unit cube in cpitch3, viewed from below; (b) three-​voice T-​space, conceived by
viewing cpitch3 from below; (c) locating the point (4, 2, 5)T =​(−3, −5, −2)T in T-​space

from us at equal angles as we look up. The figure In other words, we are looking at T-​space.
could be rotated, depending on the orientation As detailed in Section 10.3, T equivalence col-
of the viewer’s head. The negative portions of lapses each line of transposition in pitch-​triple
the coordinate axes, for which x < 0, y < 0, and space (each orbit of the transposition group) to a
z < 0, extend in our direction, as shown partially single point, effectively flattening the space into
in dashed lines. Three of the cube’s six faces are a plane—​the plane of the figure. If we imagine
visible, as are seven of its eight vertices. The all of cpitch3 divided into wire-​frame cubes a
eighth vertex, at (1, 1, 1), is not visible; from this semitone on each side, and if we imagine view-
angle it lines up precisely behind the vertex at ing this space from far below, perhaps from a
the origin. It is significant that the two vertices point such as (−100, −100, −100) far down the
that align in this view, (0, 0, 0) and (1, 1, 1), are central axis through the origin, Figure 11.4.1b is
transpositionally equivalent: the line of transpo- what we will see. Both the positive and negative
sition through these points is parallel to our line portions of the x, y, and z axes are now shown in
of sight, so from this perspective we cannot dis- their entirety; these axes do not lie in a plane,
tinguish any point on this line from any other. but angle away from the viewer at equal angles.

448 • E x p l o rin g M u sica l S paces

The space is filled by a grid of 1 × 1 × 1 cubes T-​class; there is no x for which (4, 2, 5) ~T (x, 0,
formed by straight lines parallel to the three 0) (and we should not expect that all the points
coordinate axes; from our vantage point we see in a two-​dimensional space could be described
all these cubes in corner view like the one in part by a single variable). Two coordinates are suf-
(a) of the figure, their edges forming a grid of ficient, however, because by appropriate sub-
equilateral triangles in this projection. Because traction we can always find a transpositionally
our line of sight is parallel to the lines of trans- equivalent point for which any one coordinate is
position, we are not able from this angle to dis- 0. For example, (4, 2, 5) is T-​equivalent to (0, −2,
tinguish transpositionally equivalent points. The 1) (subtracting 4 from all three coordinates); it
projection in (b) will be our standard model of is also equivalent to (2, 0, 3) (subtracting 2), and
three-​voice T-​space, though we note that, as was to (−1, −3, 0) (subtracting 5). For this T-​class, we
the case with (a), the orientation of the figure call (0, −2, 1) the yz form, (2, 0, 3) the xz form,
is somewhat arbitrary; the entire picture could and (−1, −3, 0) the xy form.
be rotated, and if viewed from above instead of
from below it would appear in mirror image. Exercise 11.4.2 In T-​space, trace the paths from
Plotting chords in T-​space by their coordi- the origin to (0, −2, 1), (2, 0, 3), (−1, −3, 0), and
nates is straightforward, as Figure 11.4.1c illus- (1 3 , −12 3 , 11 3 ), and verify that these points all fall
trates using the pitch triple (E4, D4, F4) =​(4, 2, at   the same location as (4, 2, 5).
5)—​a triple we shall use for many examples in
this section. Starting at the origin and following The yz form for a T-​class is the same as its T
the solid black arrows, we move four units in the normal form; we will often use yz forms in the
positive x direction (that is, in the direction discussion below. Such points are easily plot-
of the arrowhead on the x axis), then two units ted in Figure 11.4.1, using the y and z coordi-
in the positive y direction ( ), and finally five nates only and ignoring the x axis. Figure 11.4.3
units in the positive z direction ( ). The result- enlarges the area around the origin in T-​space,
ing point, which lies to the lower right of the ori- giving the balanced form and the yz form for
gin in the orientation of the figure, represents each equal-​tempered trichordal T-​class in this
the T-​class (4, 2, 5)T. This T-​class also stands for small region. The relationship between yz forms
any other pitch triple transpositionally equiva- and the xyz coordinates deserves attention.
lent to (4, 2, 5). One such triple is (A3, G3, B♭3) =​ Movement in the positive y or z direction coin-
(−3, −5, −2), which is T−7(4, 2, 5); the dashed cides with an increase in the corresponding coor-
arrows in the figure confirm that (−3, −5, −2), dinate of the yz form, for example from [0, 1, 1]
plotted in T-​space, falls at the same point as to [0, 2, 1] (increasing y) or to [0, 1, 2] (increas-
(4, 2, 5). Because the coordinates of this triple ing z). The x coordinate of a yz form is always 0,
are negative numbers, the dashed arrows point however, and cannot change even with move-
in the negative x, y, and z directions, opposite ment along the x axis; movement in the positive
to the corresponding solid arrows. In the three x direction brings about a decrease in both the y
dimensions of cpitch3, the two chains of arrows and z coordinates of a yz form, for example from
shown in the figure would terminate in two dif- [0, 1, 2] to [0, 0, 1].
ferent points, but from our viewpoint far below, The T-​classes that appear in Figure 11.4.3
the two points line up because the two chords lie close to the origin because they consist of
are transpositionally equivalent. three closely spaced notes. The triple (E7, D7,
Every line of transposition in cpitch3 inter- F7) shares the same yz form [0, −2, 1] with (E4,
sects the sum-​class-​0 plane in a single point. As D4, F4) and lies at the same point of T-​space, but
in the two-​voice space studied in Section 11.2, the triple (E4, D4, F7) lies far from the origin in
therefore, every T-​class contains exactly one bal- T-​space (its yz form is [0, −2, 37]). The triples (E4,
anced form whose three pitch-​space coordinates D4, F4) and (E4, D4, F7) are octave-​equivalent but
sum to 0. For the T-​class (4, 2, 5)T, the balanced not transpositionally equivalent. Our construc-
form is (1 3 , −12 3 , 11 3 ), obtained by subtracting tion of T-​space so far takes no account of octave,
11 =​32 from each coordinate in (4, 2, 5). The permutational, or inversional equivalence, so
3 3
subtracted quantity is the original sum class (11) triples that are equivalent in any of those ways
divided by the number of dimensions (3). may be distant from one another in T-​space. We
In two-​voice T-​space we also noted an x form now proceed to impose each of these additional
and y form for each T-​class. In three-​voice space equivalence relations one by one, starting with
a single coordinate is not enough to identify a permutational equivalence.

The Geometry of OPTIC Spaces • 449

FIGURE 11.4.3 Balanced forms and yz forms of points in three-​voice T-​space

FIGURE 11.4.4 Permutations of (4, 2, 5) in three-​voice T-​space, with regions for PT-​space

Figure 11.4.4 plots the six permutations of been obtained by interchanging x and y. Every
(4, 2, 5) in T-​space, also showing the yz form other permutation of (4, 2, 5) can be obtained
(T normal form) of each. A notation such as by a series of such exchanges, illustrating a gen-
(4, 2, 5)T =​[0, −2, 1]T shows two different labels eral principle noted in Section 6.1: every permu-
for the same T-​class, of which the second, in tation in a permutation group 𝒮n can be written
brackets, is the T normal form. The six yz forms as a product of 2-​cycles. The three coordinate
are not all permutationally related, though they axes divide the plane into six sectors, each con-
consist of three pairs whose y and z coordinates taining one permutation of (4, 2, 5). The sectors
are interchanged. The six P-​equivalent points may be defined by order relations among the
form a figure that is symmetric about all three three coordinates, as shown in the figure: for
axes. If (4, 2, 5)T is reflected about the z axis, example, the sector containing (4, 2, 5)T, at the
the resulting point is (2, 4, 5)T, which could have lower right, consists of all T-​classes (x, y, z)T for

450 • E x p l o rin g M u sica l S paces

which y < x < z. In other words, the sectors are Now consider the same voice leading in PT-​
contour classes: all triples (x, y, z) satisfying y < x space, represented by the normal region as shown
< z share a “middle-​low-​high” contour.35 Points in Figure 11.4.5c. In PT-​space, α and β are the same
with two equal coordinates lie on the coordi- point (PT normal form [0, 2, 3]PT), while γ ([0, 0,
nate axes, the boundaries between sectors: for 2]PT) lies on a boundary of the space. The path starts
example, the z axis comprises points of the form at α, bounces off the boundary at γ, and retraces
[0, 0, z]T for which x =​ y. Such points are permu- its path to α, illustrating the fact—​not apparent in
tationally symmetric, as they are left invariant by T-​space—​that the sonorities heard after the voice
some permutation of the coordinates (such as crossing are the same as those before, in reverse
interchanging x and y). order. The coordinate axes in T-​space become mir-
The six permutations of (4, 2, 5) represent ror boundaries of PT-​space: whenever a continuous
six different T-​classes, but they all represent the path in T-​space crosses an axis, the corresponding
same PT-​class. The plane of T-​space contains path in PT-​space reflects off a boundary. Because
every T-​ class exactly once, but (disregarding of this boundary behavior, PT-​space cannot be a
points on the axes) contains every PT-​class six manifold: pitch-​triple space and T-​space are mani-
times, once in each sector. Any one of the six folds, but PT-​space is an orbifold.
sectors could therefore serve as a fundamental
region for PT equivalence and a representation Exercise 11.4.6 Reconsider the Saariaho voice
of three-​voice PT-​space cpitch3/​~PT. The sector leading (−12, −3, −1) → (−8, −10, −7) from
containing (2, 4, 5)T, shaded in Figure 11.4.4, is Exercise 11.3.13 and Figure 11.3.14.
the normal region. In this region the coordinates
appear in ascending order x < y < z, as is always (a) Graph this motion in T-​space and PT-​space.
the case in P or PT normal forms (except on the What happens at the point γ where the
boundary of the sector, where at least x ≤ y ≤ z lower two voices cross?
is always true). Every pitch triple α is P-​equiva- (b) Suppose the initial pitch triple α =​(C3, A3,
lent to some point in the normal region, namely B3) is transposed up an octave to α′ =​(C4,
its P normal form [α]P, obtained by arrang- A4, B4) but β =​(E3, D3, F3) remains in its
ing the three coordinates in ascending order. original register. In T-​space, is the path
For α =​(4, 2, 5), the P normal form is [2, 4, 5]P, from α′ to β the same as the path from α to
and the PT normal form is [0, 2, 3]PT, obtained by β? What about the path in PT-​space? Does
transposing [α]P to start with 0. Because (2, 4, 5) the crossing point γ represent the same
and (0, 2, 3) are transpositionally equivalent, pitch-​class set as before? What if α and β
they lie at the same point in the plane of T-​space.   are both transposed up an octave?36
PT normal forms such as [0, 2, 3]PT will be our
standard labels for points in PT-​space. PT-​space, Next we introduce inversional equivalence.
like T-​space, is an unbounded two-​dimensional The image of (4, 2, 5) under the inversion
space, but it is only one-​sixth of the plane—​as operator I is I(4, 2, 5) =​(−4, −2, −5), a point
we should expect, because PT-​space is the quo- that does not coincide with (4, 2, 5) in T-​space
tient space obtained from T-​space through the or PT-​space. Like every pitch triple, however,
action of a group 𝒮3 of order 6. it is permutationally equivalent to a point in
Consider the straight-​ line voice leading the fundamental region for PT equivalence, in
(2, 4, 5) → (4, 2, 5) shown in Figure 11.4.5a, mov- this case (−5, −4, −2), with yz form [0, 1, 3]. As
ing from α =​(D4, E4, F4) to β =​(E4, D4, F4). The shown in Figure 11.4.7, the two inversionally
first voice ascends continuously through a whole related points (4, 2, 5)PT =​[0, 2, 3]PT and (−4, −2,
tone from 2 to 4 (D4 to E4) while the second voice −5)PT =​[0, 1, 3]PT are symmetrically positioned
executes the opposite motion from E4 to D4, the in opposite halves of the sector of PT-​space. The
third voice all the while fixed at F4. In the plane of dashed line that bisects PT-​space in the figure is
T-​space, shown in part (b) of the figure, this voice an axis of inversional symmetry. Points on this
leading describes a straight line from (2, 4, 5)T line include yz forms such as [0, 1, 2], [0, 2, 4],
(yz form [0, 2, 3]) to (4, 2, 5)T (yz form [0, −2, 1]). and [0, 3, 6]—​inversionally symmetric chords,
The two moving voices cross at E♭4, momentarily invariant under some inversion in pitch space.37
forming the pitch triple γ =​(3, 3, 5) (yz form [0, For triples (x, y, z) on this line, y is equidistant
0, 2]), a permutationally symmetric point where in pitch space between x and z; the line is char-
the path in T-​space crosses the z axis. acterized by the equation y − x =​ z − y, or more

The Geometry of OPTIC Spaces • 451

 compactly 2y =​ x +​ z. Inversion I maps any point
in PT-​space to its reflection across this axis of
symmetry.
With inversional equivalence added to the
arsenal of available equivalence relations, every
point is now equivalent to some point in each
half of PT-​space; we may therefore take either
one of these half-​ sectors as a fundamental
region for PTI equivalence. The lower half-​sector
is the normal region—​the PTI normal form of
(4, 2, 5) is [0, 1, 3], not [0, 2, 3]—​so we will use
this normal region as a representation of three-​
voice PTI-​space. This region of T-​space consists
of points (x, y, z) with x < y < z such that y is
closer to x than to z. PTI-​space is half the size
of PT-​space—​plausibly, inasmuch as inversion
generates a group of order 2—​and one-​twelfth
the size of T-​space. Excepting boundary singu-
larities, the full plane of T-​space includes every
T-​class once, every PT-​class six times, and every
PTI-​class twelve times.

Exercise 11.4.8
(a) Each of the six sectors of T-​space shown in
Figure 11.4.4 (any of which could represent
PT-​space) has its own axis of inversional
symmetry, comparable to the dashed line in
Figure 11.4.7. Locate the six permutations
of (0, 2, 4) in T-​space and verify that they
all lie on one of these axes. Determine
equations analogous to x +​ z =​2y for each of
the other axes.
(b) The six permutations of the pitch triple
(−4, −2, −5) may be obtained by applying the
FIGURE 11.4.5 (a) The voice leading (2, 4, 5) → inversion operator I to the six permutations
(4, 2, 5); (b) the corresponding path in three-​voice of (4, 2, 5) shown in Figure 11.4.4. Plot
T-​space; (c) the corresponding path in three-​voice these six new points in T-​space, noting
PT-​space how they are related to each other (by
reflection about the coordinate axes) and
to the original permutations of (4, 2, 5)
(by reflection about the axes of inversional
  symmetry in the six sectors).

Consider the simple voice leading (2, 4, 5) →

(2, 3, 5) shown in Figure 11.4.9a, in which the
middle voice descends continuously through
the semitone from E4 to E♭4 while the other two
voices hold the pitches D4 and F4. In PT-​space, as
shown in (b), this voice leading traces a straight-​
line path from α =​(2, 4, 5) (PT normal form [0, 2,
3]PT) to β =​(2, 3, 5) ([0, 1, 3]PT). This path crosses
the boundary separating the two half-​sectors of
PT-​space; the crossing occurs at γ =​(2, 31 2 , 5)
FIGURE 11.4.7 Inversion in three-​voice PT-​space, (PT normal form [0, 11 2 , 3]PT). Although γ is not
with the normal region for PTI-​space an equal-​tempered trichord, it is inversionally

452 • E x p l o rin g M u sica l S paces

 FIGURE 11.4.10 Two octave-​equivalent points
in three-​voice PTI-​space

direction parallel to any of the coordinate axes

will always yield an O-​equivalent point, pro-
ducing a closed path of some sort in any space
that accounts for O equivalence. In the setting
that already includes P, T, and I equivalence,
however, the modular structure created by O
equivalence turns out to be considerably more
complex than the mod-​12 cubes that arise in
the case of 𝕋 n.
Consider once more the pitch triple (E4, D4,
F4) =​(4, 2, 5). We know that this triple appears
in PTI-​space at the point whose yz form is [0, 1,
FIGURE 11.4.9 (a) The voice leading (2, 4, 5) → 3]. Suppose now that the triple’s second note
(2, 3, 5); (b) the corresponding path in three-​voice is octave-​transposed from D4 to D5, yielding
PT-​space; (c) the corresponding path in three-​ the O-​equivalent triple (4, 14, 5). This triple is
voice PTI-​space P-​equivalent to (4, 5, 14), whose yz form is [0,
1, 10]. This is another point of PTI-​space, lying
seven units to the right of the original point at
symmetric. In PTI-​space, as shown in (c), α and coordinates [0, 1, 3]. The idea that transposing
β are the same point, and γ lies on a boundary of a pitch by twelve semitones causes the position
the space. It is another mirror boundary: at γ the of a point to shift by seven units may seem unex-
path in PTI-​space bounces back in the direction pected—​but as Figure 11.4.10 shows, these two
from which it came. points are symmetrically positioned about a ver-
Coordinates in the PTI normal region, such tical line through the point [0, 0, 6]. This line is
as [0, 1, 3], often resemble prime forms of a line of symmetry, but what kind may not be
pitch-​class sets—​but because we have not yet immediately clear. Another point on this line
introduced octave equivalence, the triples here has the yz form [0, 2, 7]. In pitch space, {0, 2,
still stand for sets of pitches, not pitch classes, 7} is not a symmetrical set, but pitch-​class sets
and chords with widely separated notes, lying of prime form 027 are inversionally symmetric.
far from the origin, may have coordinates The symmetry of [0, 2, 7] and other points on
larger than 12. Like PT-​space, PTI-​space is an the vertical line depends on both octave and
unbounded two-​ dimensional orbifold; only inversional equivalence; the line is a line of OI
octave equivalence will produce a bounded symmetry for PT-​classes.38 All points on this line
space. Octave equivalence converts unbounded satisfy the equation x +​ y +​12 =​2z.
pitch space to bounded pitch-​class space—​or Figure 11.4.11 shows a broader picture of
in multiple dimensions, converts unbounded lines of O, P, and I symmetry in the plane of
pitch-​tuple space ℝn to bounded pitch-​class–​ T-​space.39 The original x, y, and z axes (solid
tuple space 𝕋 n. Moving twelve semitones in a black lines) are lines of permutational symmetry,

The Geometry of OPTIC Spaces • 453

FIGURE 11.4.11 Axes of symmetry in the plane of three-​voice T-​space, with the normal region for
OPTI-​space (after Callender)

intersecting at the origin and dividing the direction, and to combine these translations in
plane into six sectors. The central axes of the any desired way. The translated copies of the
six sectors (dashed black lines) are the lines axes are in gray.
of inversional symmetry described in Exercise The various axes and their translated copies
11.4.8 above. The full picture in Figure 11.4.11 tile the plane with right triangles, all identical
emerges through octave equivalence, which in size but occurring in twelve different orien-
effectively allows us to translate this entire tations. Any one of these triangles could serve
system of axes by twelve units along the x, y, as a fundamental region for OPTI equivalence.
or z axis, in either the positive or the negative The normal region is the shaded triangle, the

FIGURE 11.4.12 Three-​voice OPTI-​space

454 • E x p l o rin g M u sica l S paces

triangle closest to the origin within the funda- (b) Figure 3.3.1 showed only a single loop
mental region for PTI equivalence defined pre- at the trichord 037, but Figure 11.4.12
viously. This triangle, representing three-​voice reveals that two different loop motions
OPTI-​space, is reproduced at a larger scale in in OPTI-​space are possible here: a single-​
Figure 11.4.12. All the lattice points within and semitone voice leading may return to 037
on the boundary of the triangle are identified by after bouncing off either the sloping axis
their yz forms, which in this region coincide with of inversional symmetry or the vertical
OPTI normal forms—​traditional prime forms of axis of OI symmetry. One of these loops
pitch-​class multisets. The normal forms are writ- corresponds with the transformation P and
ten in compressed fashion without brackets or the other with L. Which is which?
commas. Multisets with doublings, with normal (c) The loops just discussed are single-​semitone
forms 00z, occur along the lower boundary. The voice leadings. If the neo-​Riemannian R is
arrangement is consistent with that of other fig- regarded as a voice leading in which one
ures in this section, but more interesting is the voice moves continuously through a whole
resemblance to the graph of trichordal set-​class tone, then R creates yet another kind of
space from Figure 3.3.1—​or, better, with the loop at 037. How does the voice leading
graph constructed in Exercise 3.3.5, with mul- associated with R appear in OPTI-​space?
tisets added. The appearance of this graph was (d) Graph the voice leadings associated with
described as “oddly asymmetrical” in Chapter 3; the Riemannian transformations P′, L′, R′,
our work with continuous voice-​leading spaces and H (from Section 8.2) in OPTI-​space. All
in this chapter effectively validates the earlier of these paths in OPTI-​space should match
graph and shows how it fits within a larger struc- others you have considered previously. Why
ture (Figure 11.4.11) that exhibits symmetry of   is this?40
many kinds.
The gridlines in Figure 11.4.12 indicate Exercise 11.4.14 Write a straight-​line voice
continuous motion in the x, y, and z coordi- leading in three voices (a straight line in
nates; as in the graph in Chapter 3, chords at cpitch3) which, in OPTI-​space (Figure 11.4.12),
adjacent vertices along any gridline differ by starts at 016, bounces off the corner of the
a single semitone in one voice. The loops in space
   at 006, and returns to 016.
the earlier graph, the edges joining a vertex to
itself, appear here not as loops but as bounces: Exercise 11.4.15 Locate all points in Figure
places where a single-​semitone motion reflects 11.4.11 that are OPTI-​equivalent to (E4, D4, F4) =​
off a mirror boundary and returns to its start- (4,  
2, 5).
ing point. The bounce from 013, off the axis of
inversional symmetry, was illustrated in Figure Exercise 11.4.16 Figure 11.4.11 enables us
11.4.9c above; similar bounces are possible at to deduce something about the symmetry
001, 025, and 037. Meanwhile, 016 and 037 of pitch triples from their position relative
admit a different kind of bounce, off the ver- to the various axes of symmetry in T-​space.
tical axis of OI symmetry. For example, the Specifically, a pitch triple such as (0, 3, 3) lying
T-​space voice leading (0, 1, 6) → (0, 1, 7) on a solid black axis is P-​symmetric, meaning
bounces in OPTI-​space at (0, 1, 61 2 ), an OPTI-​ that it has a doubled pitch; a point such as (0,
symmetric trichord, only to return to its start- −3, 9) on a solid gray axis is OP-​symmetric,
ing point 016 in OPTI-​space. meaning that it has a doubled pitch class; a
point such as (0, 3, 6) on a dashed black axis is
Exercise 11.4.13 PTI-​symmetric, meaning that it is inversionally
(a) Consider the neo-​Riemannian PL-​cycle symmetric as an unordered multiset in pitch
c –​C –​e –​E –​g♯ –​A♭ –​c. Start with the pitch space; and a point such as (0, −3, 6) on a dashed
triple (0, 3, 7), and realize each P and L gray axis is OPTI-​symmetric, meaning that it
transformation through a single-​semitone is inversionally symmetric as an unordered
voice leading. Write the cycle as a series of multiset in pitch-​class space.
pitch triples; then, conceiving of the voice First, verify the examples in the preceding
leading as continuous motion, graph the paragraph, locating each point in T-​space,
cycle in three-​voice T-​space, PT-​space, PTI-​ identifying the axis on which it lies, and
space, and OPTI-​space. verifying the symmetry numerically. Then

The Geometry of OPTIC Spaces • 455

determine the type(s) of symmetry, if any, for equivalence. Figure 11.4.18d shows the OPTI-​
each of the following pitch triples.41 space path. One way to conceive of this path
is to take the original straight-​line path from
(a) (−4, 1, 3) (b), visualize each of the seven segments in its
(b) (5, 1, 3) own triangular tile, and map those seven sepa-
(c) (5, 1, 5) rate tiles into the normal region by a series of
(d) (5, 1, 7) reflections across boundaries. Because α and β
(e) (5, 1, 13) are octave-​equivalent, they are the same point
(f)  (7, 1, 13) in OPTI-​ space; the path bounces six times
between the two, including the first bounce
Exercise 11.4.17 Write equations in the at γ, already encountered in PTI-​space. At δ =​
variables x, y, and z for all of the axes (black or (2, 4, 9) =​(D4, E4, A4) (OPTI normal form 027)
gray,
   solid or dashed) visible in Figure 11.4.11.42 it bounces at a right angle off the vertical axis
of OI symmetry and heads leftward again for
Consider the voice leading (2, 4, 5) → (2, 4, segment ③; here we may envision the triangle
17) shown in Figure 11.4.18a, in which two containing segment ③ in (b) mapped onto the
voices hold the pitches D4 and E4 while the third normal region by reflection about the vertical
ascends continuously through an octave from line through δ. Point ε =​(2, 4, 12) =​(D4, E4, C5)
F4 to F5. In pitch-​triple space this voice leading is OPT-​equivalent to γ, and therefore the same
describes a straight line parallel to the z axis from point of OPTI-​space (OPTI normal form 024).
α =​(2, 4, 5) to β =​(2, 4, 17). It is also a straight Here the path bounces back in the direction of
line in T-​space, extending 12 units rightward α, and in fact the path passes through αOPTI =​
because the only motion is in the z coordinate. 013 in the middle of segment ④.
The three voices remain in ascending order x < y The final three bounces are closely spaced,
< z throughout the voice leading, so the T-​space separated by single semitones of voice leading.
path lies entirely within the normal region for The first of these occurs at ζ =​(2, 4, 14) =​(D4,
~PT, and therefore the graph in PT-​space looks E4, D5), an OP-​symmetric point on a solid gray
identical to the graph in T-​space. This graph in axis in (b) and (c), with OPTI normal form 002.
T-​space or PT-​space is shown in (b). A right-​angle bounce follows at the OPTI-​sym-
The straight-​line path in T-​space does not metric point η =​(2, 4, 15) =​(D4, E4, E♭5) (OPTI
cross any axes of permutational symmetry normal form 012), after which there is a final
(Figure 11.4.11’s solid black lines), but it crosses bounce at θ =​(2, 4, 16) =​(D4, E4, E5), another
no fewer than six other axes of symmetry en route OPT-​symmetric point coincident with ζ in OPTI-​
from α to β. The six crossing points are labeled in space, leading back to the starting point at β =​(2,
(b) with the next six Greek letters, γ through θ, 4, 17) =​(D4, E4, F5).
and they divide the entire path into seven seg-
ments, labeled ①–​⑦. The first axis encountered, Exercise 11.4.19 To show that the pitch triple
at the symmetrical pitch triple γ =​(2, 4, 6) =​(D4, γ =​(2, 4, 6) is PTI-​symmetric, we could write,
E4, F♯4), is a dashed black line of PTI symmetry, for example,
so in PTI-​space the path bounces off a boundary
at γ. The path enters the lower half of PT-​space (2, 4, 6) ~I (−2, −4, −6) ~T (6, 4, 2) ~P (2, 4, 6).
at γ—​the normal region for ~PTI—​and remains
in that region through segments ②–​⑦. Segment In a similar way, verify numerically the
① from α to γ, however, lies in the upper half of assertions below, all of which figured in the
PT-​space and must therefore be reflected into above description of the voice leading in
the lower half when the voice leading is graphed OPTI-​space:
in PTI-​space. As we have seen, α maps onto the
point (0, 1, 3) in PTI-​ space; Figure 11.4.18c (a) Pitch triple δ =​(2, 4, 9) is OPTI-​symmetric.
shows that the graph of the voice leading in PTI-​ (b) Pitch triples γ =​(2, 4, 6) and ε =​(2, 4, 12)
space differs from that in T-​ or PT-​space only in are OPT-​equivalent.
the appearance of segment ①. (c) Pitch triple ζ =​(2, 4, 14) is OP-​symmetric.
In OPTI-​space, things get much more com- (d) Pitch triple η =​(2, 4, 15) is OPTI-​symmetric.
plicated, because the path encounters five (e) Pitch triple θ =​(2, 4, 16) is OP-​symmetric.
more mirror boundaries, all gray lines from (f) Pitch triples ζ =​(2, 4, 14) and θ =​(2, 4, 16)
Figure 11.20 depending in some way on octave   are OPTI-​equivalent.

456 • E x p l o rin g M u sica l S paces

FIGURE 11.4.18 (a) The voice leading (2, 4, 5) → (2, 4, 17); (b) the corresponding path in three-​voice
T-​space; (c) the corresponding path in three-​voice PTI-​space; (d) the corresponding path in three-​voice
OPTI-​space

The Geometry of OPTIC Spaces • 457

Exercise 11.4.20 What are the coordinates obtained by combining the triangular normal
of the pitch triple that lies between ε and ζ on region for ~OPTI (Figure 11.4.12) with its reflec-
segment ④ of the path described above and tion across the dashed axis of inversional sym-
which coincides in OPTI-​space with α and β? metry. This region contains two representatives
Verify numerically that this triple is OPTI-​ of every asymmetrical three-​note multiset class,
equivalent
   both to α and to β. one for each of its two T-​classes, symmetrically
positioned on opposite sides of the inversional
We will not attempt to cover every pos- axis. For example, the figure shows points at
sible combination of the OPTIC symmetries coordinates [0, 1, 3] and [0, 2, 3], both of which
for three-​voice spaces, but OPT-​space deserves are in OPT normal form, though they share the
attention. Because ~OPT resembles ~OPTI but single OPTI normal form [0, 1, 3].
without inversion, we should expect that OPT-​ This normal region differs in important
space should be twice as large as the OPTI-​space respects from others we have studied. The two
described above. The normal region for ~OPT is shorter boundary segments of this figure are
shown in T-​space in Figure 11.4.21a, where it false boundaries, not mirror boundaries. This
takes the form of the shaded kite-​shaped figure is because the pitch triples on one of these two

FIGURE 11.4.21 (a) The normal region for three-​voice OPT-​space; (b) the voice leading (2, 4, 5) →
(2, 4, 17) in OPT-​space

458 • E x p l o rin g M u sica l S paces

boundary segments are OPT-​equivalent to those Exercise 11.4.24 The OPT normal region
on the other: for example, of Figure 11.4.21, formed by combining the
triangular OPTI normal region with the triangle
(0, 6, 6) ~T (6, 12, 12) ~O (6, 0, 0) ~P (0, 0, 6) above it in Figure 11.4.11, is not the only
possible fundamental region for OPT-​space.
and
Consider these two possibilities:
(0, 5, 7) ~T (7, 12, 14) ~O (7, 0, 2) ~P (0, 2, 7).43
(a) Combine the OPTI normal region with the
These two edges of the region therefore repre- triangle below it (its reflection across the z
sent the same line segment in OPT-​space; the axis), forming an equilateral triangle.
two are glued together in the orientation implied (b) Combine the OPTI normal region with the
by the arrowheads on these edges.44 triangle to its right (its reflection across
Figure 11.4.21b traces the voice leading (2, the dashed vertical axis of OI symmetry),
4, 5) → (2, 4, 17) from Figure 11.4.18 in OPT-​   forming a broad, flat isosceles triangle.
space. Unlike the path in OPTI-​space, the path
in OPT-​space does not retrace its own course in One of these two possibilities yields a valid
reverse; instead, corresponding segments mirror construction of a fundamental region for OPT-​
each other about the axis of inversional symme- space, but the other does not. Which is which,
try. The voice leading starts by moving rightward and why? Plot the (2, 4, 5) → (2, 4, 17) voice
from αOPT =​023. At γOPT =​024, the path crosses leading from Figures 11.4.18 and 11.4.21b
the inversional axis, but this is not a boundary using the alternate fundamental region you
of OPT-​space, so there is no change of direc- have identified, and describe advantages and
tion here; the path simply continues rightward disadvantages of using this region as a model of
into the other half of OPT-​space. At δOPT =​027, OPT-​space instead of the normal region.46
one of the false boundaries of the fundamental
region is reached. Rather than bouncing back as We conclude this section with some brief
in OPTI-​space, the path re-​emerges at the other remarks about cardinality equivalence in three-​
boundary point representing δ, still moving at voice spaces. In the OPTI-​space of Figure 11.4.12,
right angles to the boundary, and continues its C equivalence turns out to be inconsequential,
journey from there via segments ③ and ④, which as was the case in all two-​voice spaces. The only
mirror segments ② and ① in reverse. The path trichords potentially affected are those with dou-
crosses its previous course at ε, which coincides blings, on the lower boundary of the triangle
with γ on the line of inversional symmetry. In in the figure. The OPTI normal form of such a
quick succession, the path then reflects off the z trichord is of the form [0, 0, z]OPTI for some z with
axis at ζ, crosses the inversional axis again at η, 0 ≤ z ≤ 6, which reduces via C equivalence to the
and bounces off the x axis at θ, finally arriving at OPTIC normal form [0, z]OPTIC (or [0]‌OPTIC in the
β, which coincides with α. case of the triple unison [0, 0, 0]OPTI). The line seg-
ment of trichords with doublings looks just like
Exercise 11.4.22 the line segment of two-​voice OPTI-​space (Figure
(a) Verify that all labeled points are correctly 11.2.12c), and in all cases there is a one-​to-​one
plotted in Figure 11.4.21b. correspondence between OPTI normal forms
(b) In Exercise 11.4.20 you identified another and OPTIC normal forms. As a result, three-​voice
pitch triple in this voice leading that OPTIC-​space is essentially the same thing as
coincided with α and β in OPTI-​space. three-​voice OPTI-​space.
  Where is this pitch triple in OPT-​space? It would be a mistake to infer, however, that
cardinality equivalence has no bearing on any
Exercise 11.4.23 Return once more to the three-​voice spaces. While two different pitch
Saariaho voice leading from Figure 11.3.14 pairs cannot have the same C normal form, two
and Exercises 11.3.13 and 11.4.6. Graph this different pitch triples can: for example, (0, 0, 5)
voice leading in PTI-​space, in OPTI-​space, and and (0, 5, 5) share the C normal form [0, 5]C.
in OPT-​space. Your picture should show that In cpitch3 or any quotient space of cpitch3 in
in addition to the initial and final sonorities, which pairs of points like (0, 0, 5) and (0, 5, 5)
one other OPTI-​class occurs twice in this remain distinct, cardinality equivalence will
voice leading; identify the coordinates of this affect the form of the space. In OPT-​space, for
point.
   45 example, [0, 0, 5]OPT and [0, 5, 5]OPT are two

The Geometry of OPTIC Spaces • 459

different points (both are in OPT normal form), are difficult to conceptualize in a three-​dimen-
so the fact that they are cardinality-​equivalent sional world and even harder to draw in a two-​
means that three-​voice OPTC-​space cannot be dimensional picture, but Figure 11.5.1a shows
the same as the OPT-​space of Figure 11.4.21.47 one way to visualize this construction. Vertices
are labeled with four-​dimensional coordinates (w,
Exercise 11.4.25 Locate two different points x, y, z), each coordinate taking the values 0 and
in the Möbius prism of three-​voice OP-​space 1. The original cube, in xyz space (all the vertices
(Figure 11.3.4) that must be conjoined in have w =​0), is shown in black, in an orientation
   space.48
OPC-​ similar to Figure 10.1.1 and other figures we have
examined in three-​voice spaces. The translated
Three-​voice OPTI-​space turns out to be the copy, on the vertices with w =​1, appears in gray;
last space for which we may say that cardinal- the projection through the fourth (w) dimen-
ity equivalence does not make a difference: all sion, connecting the two cubes, is shown by
spaces of four or more voices are substantively dashed lines.
altered by the inclusion of ~C, usually in ways
too complicated to describe here. The “worm- Exercise 11.5.2 A 4-​cube has 16 vertices
holes” joining distant points become even more (twice as many as an ordinary 3-​cube), 32
difficult to visualize when spaces of different one-​dimensional edges (line segments), 24
dimensions are involved. Even in the case of two-​dimensional square faces, and eight three-​
three-​voice OP-​space, as in the above exercise, dimensional facets or “hyperfaces,” which are
the identifications required to account for ~C themselves ordinary cubes. Identify all of these
are complex. Cardinality equivalence matters objects
   in Figure 11.5.1a.50
only for chords with doublings, which charac-
teristically occupy only a very small portion of a Figure 11.5.1b shows an octatonic seventh-​
space: a single line in a plane, a boundary edge, chord hypercube, which may look familiar if you
or a two-​dimensional slice within a three-​dimen- solved Exercises 3.1.17 and 3.1.18. The verti-
sional space, for example. While these small ces of this 4-​cube are identified with four-​note
regions may seem statistically insignificant, they chords, labeled with O normal forms in wxyz
are of considerable musical importance because coordinates and also with chord names repre-
of the prevalence of sonorities with doublings—​ senting OP-​classes. The hypercube in (b) is like
and because these regions often occur on space the one in (a), but its coordinates have been
boundaries, they are often atypical in other ways translated to a different location in four-​dimen-
as well. See Exercise 11.6.17 at the end of this sional space: the coordinates in (b), which are O
chapter for another illustration. normal forms, may be obtained by adding (0, 3,
6, 9) componentwise to the coordinates in (a). As
11.5 FOUR-​VOICE OP-​S PACE with the three dimensions of the hexatonic triad
cube, each dimension of this hypercube encom-
The complexity of OPTIC spaces increases dra- passes a single semitone of motion in one voice:
matically as the number of notes increases, and a the w voice (the first coordinate of the ordered
thorough study of higher-​dimensional examples 4-​tuple) moves between 0 and 1, x between 3
is beyond the scope of this book. In the final two and 4, y between 6 and 7, and z between 9 and
sections of this chapter we confine ourselves to a 10. All notes of the chords at the vertices of the
brief overview of some four-​voice spaces. graph therefore belong to the octatonic collec-
We began our study of three-​voice spaces with tion oct01.
the hexatonic triad cube. The analogous construc- The OP-​classes in (b) could have been labeled
tion in four notes is an octatonic seventh-​chord with OP normal forms, rotations of the O nor-
hypercube. A hypercube is a higher-​dimensional mal forms as determined by the span-​optimiza-
analog of a cube, in this case a four-​dimensional tion criterion from Section 10.4. For example,
cube or 4-​cube.49 Just as a three-​dimensional cube the OP normal form derived from O normal
may be constructed from a (two-​ dimensional) form [0, 3, 7, 10]O would be [7, 10, 0, 3]OP, which
square by projecting it through a third dimension could be written in note names as GB♭CE♭. Rather
perpendicular to the first two, a 4-​cube may be than displaying OP normal forms in this way, the
constructed by starting with a three-​dimensional figure instead employs familiar chord names,
cube and projecting it through a fourth dimension so that the chord just discussed is labeled Cm7.
perpendicular to the first three. Four dimensions Fully diminished seventh chords co7 and c♯o7

460 • E x p l o rin g M u sica l S paces

FIGURE 11.5.1 (a) Construction of a 4-​cube; (b) an octatonic seventh-​chord hypercube

appear at the bottom and top vertices of the just two trichordal set classes in the hexatonic
figure. The other fourteen vertices include the triad cube (037 and 048). The mod-​ 12 sum
half-​diminished, minor, and dominant seventh classes of the chords in Figure 11.5.1b range
chords whose roots are pitch classes 0, 3, 6, from 6 for co7 at the bottom of the figure to 10
and 9, as well as two French augmented-​sixth for c♯o7 at the top. Dark vertices represent chords
chords, labeled CFr and E♭Fr. For example, CFr of even sum class (6, 8, and 10), while light ver-
is the chord C–​E–​F♯–​A♯, a chord that might also tices are chords of odd sum class (7 and 9). Each
be called F♯Fr, C♭⁷₅, or F♯ ♭⁷₅. The octatonic seventh-​ edge, depicting a single-​semitone voice leading,
chord hypercube thus includes chords in four joins a dark vertex to a light vertex; the graph of
different tetrachordal set classes (prime forms a cube in any number of dimensions is always a
0258, 0268, 0358, and 0369), compared with bipartite graph. The above discussion describes

The Geometry of OPTIC Spaces • 461

the chords at the vertices of the hypercube as seventh-​chord hypercubes in four-​ voice OP-​
discrete objects, but the octatonic 4-​cube, like space. There are, after all, only three different
the hexatonic 3-​cube, is a continuous structure, octatonic collections, and the d♯o7 chord at the
points in its four-​dimensional interior corre- top of Figure 11.5.3 is the same OP-​class as co7
sponding to non-​equal-​tempered chords. at the bottom. The chain must therefore loop
The hypercube in Figure 11.5.1b is part of a back on itself somehow in the manner of Figure
chain of similar hypercubes, shown in Figure 10.2.11, presumably with some sort of twist.
11.5.3. Diminished seventh chords form the Apart from the co7–​d♯o7 duplication, every other
links between successive hypercubes, playing equal-​
tempered chord of prime form 0258,
a role analogous to augmented triads in the 0268, 0358, and 0369 appears exactly once in
hexatonic triad cube chain from Figure 10.2.1. Figure 11.5.3.51
This chain is a short one: the three hyper- Like the hexatonic cube, a single octatonic
cubes in the figure are the only three octatonic hypercube may be conceived as a subset of

FIGURE 11.5.3 A chain of octatonic seventh-​chord hypercubes

462 • E x p l o rin g M u sica l S paces

FIGURE 11.5.4 The sum-​class-​0 tetrahedron in four-​voice OP-​space

pitch-​tuple space or O-​space, but when the full doubled pitch classes, such as DDEE. Multisets
chain is considered, complexities arise requir- with a single doubling fall on the faces of the tet-
ing a fuller conception of OP-​ space. Here the rahedron: for example, C♯D♯EE falls near the top
increased dimensionality of four-​ voice space of the rightward-​facing face, midway between
presents a greater conceptual challenge. Recall the labeled points CEEE and DDEE. The inte-
that three-​voice OP-​space takes the form of the rior of the tetrahedron is filled with tetrachords
Möbius prism in Figure 11.3.4, whose triangular without doublings; for example, C♯DEF is also
top and bottom layers, consisting of trichords of relatively near the top of the tetrahedron, but
sum class 0, are glued together with a 120-​degree within the interior, not on a face.
rotation. In four-​ voice pitch-​tuple space, tetra-
chords of sum class 0 lie not in a plane but in a Exercise 11.5.5
three-​dimensional subspace, consisting of solu- (a) Some tetrachords of sum class 0, such as the
tions to a single linear equation w +​ x +​ y +​ z =​ 0 whole-​tone tetrachord E♭FGA, are invariant
in four variables. In four-​voice OP-​space, the sum-​ under the inversion I0—​that is, they are
class-​0 tetrachords form the tetrahedron shown in symmetric about the C–​F♯ axis in pitch-​class
Figure 11.5.4; zero-​sum voice leading takes place space. Where do such tetrachords lie in the
not in a plane but in a tetrahedron of this sort. tetrahedron of Figure 11.5.4?
As a picture of the sum-​class-​0 cross section (b) Some tetrachords of sum class 0, such as
of OP-​space, Figure 11.5.4 is far from complete. the Em7 chord BDEG, are invariant under
The analogous figure for three-​ voice spaces, I6, symmetric about the E♭–​A axis. Where do
Figure 11.3.1, showed all the equal-​tempered such tetrachords lie in the tetrahedron?
trichords of sum-​class 0 and included O normal (c) A few tetrachords of sum class 0, such as
forms in xyz coordinates as well as OP normal the D♭Fr chord FGBD♭, are symmetric about
forms. In Figure 11.5.4, no attempt is made to both the C–​F♯ and E♭–​A axes. Where do such
show wxyz coordinates, and only 22 of the 116   tetrachords lie in the tetrahedron?52
equal-​tempered four-​note multisets present in
the tetrahedron are labeled in the figure—​those Exercise 11.5.6 The four-​voice straight-​line
on its edges. These 22 tetrachords, all of which voice leading shown in Figure 11.5.7a is similar
include multiple pitch-​class doublings, are iden- to the three-​voice example studied previously
tified by OP normal forms. The multisets CCCC, in Figure 11.3.10a.53 Now, however, two voices
E♭E♭E♭E♭, F♯F♯F♯F♯, and AAAA at the corners of the descend while two others ascend, so the voice
tetrahedron are the quadruple unisons of sum leading is zero-​sum, and it lies entirely within
class 0, analogous to the multisets CCC, EEE, the sum-​class-​0 tetrahedron of Figure 11.5.4.
and A♭A♭A♭ at the corners of Figure 11.3.1. Four
of the six edges of the tetrahedron contain mul- (a) Graph this voice leading in the tetrahedron.
tisets with a tripled pitch class, such as DDDF♯; First locate α and η, which are among the
the other two edges contain multisets with two chords labeled on the edges of the figure.

The Geometry of OPTIC Spaces • 463

FIGURE 11.5.7 (a) A zero-​sum voice leading in four voices, for Exercise 11.5.6; (b) omnibus progres-
sion for Exercise 11.5.8d

Then locate γ and ε, each of which contains also be notated C+​0.25C+​0.25C+​0.25C+​0.25. Draw
one doubled pitch class and therefore lies this tetrahedron in a manner resembling
on a face of the tetrahedron. The complete Figure 11.5.4, labeling chords in the latter
path consists of straight line segments notation.
α → γ, γ → ε, and ε → η; chords β, δ, and (b) The vertices of the sum-​class-​1 tetrahedron
ζ, interior to the tetrahedron, lie at the are not equal-​tempered tetrachords, but
midpoints of these three segments.54 many equal-​tempered chords belong to sum
(b) If the voice leading continues in the same class 1 and therefore may be found here.
manner beyond chord η (all four voices Locate the triple unisons C♯EEE, FFFB♭,
continuing to move in their established and AAAB♭ on the edges of the tetrahedron.
directions), what path will be traced Also locate the (non-​equal-​tempered) chord
through the tetrahedron? E♭+​0.5E♭+​0.5AA on one edge. Do any equal-​
(c) Chord δ is the chord labeled D♭Fr in the tempered chords lie on that edge?
oct12 hypercube of Figure 11.5.3, also (c) Consider the straight-​line voice leading
mentioned in Exercise 11.5.5c. The chord (−3, −2, 9, 9) → (−3, 31 2 , 31 2 , 9) in four-​
EFr also belongs to sum class 0 and is voice pitch-​tuple space. During the course
symmetric about the same axes; visualize of this voice leading, voice x ascends
its location within the tetrahedron. through 51 2 semitones while y descends
(d) The four minor seventh chords in the through the same interval; the entire
oct12 hypercube also belong to sum class voice leading remains within the sum-​
0. Can you visualize their locations in the class-​1 tetrahedron. In OP-​space, this
tetrahedron? If you draw a line segment voice leading describes a straight line
through the tetrahedron from Em7 to B♭m7, from AAAB♭ to E♭+​0.5E♭+​0.5AA, two of the
what (non-​equal-​tempered) chord lies at the points identified in part (b). The w and z
  midpoint of that segment?55 coordinates hold two pitches (−3 and 9)
representing the same pitch class, and the
Exercise 11.5.8 entire voice leading in OP-​space remains
(a) Tetrachords of sum class 1 form a within one face of the tetrahedron.
tetrahedron like Figure 11.5.4, but with After x and y have moved through four
every pitch class 1 4 semitone higher. One semitones, the chord (−3, 2, 5, 9)OP (DFAA)
vertex of this tetrahedron is the quadruple is encountered, a D-​minor triad with a
unison OP-​class [ 1 4 , 1 4 , 1 4 , 1 4 ]OP , which may doubled fifth. Locate this chord on one face

464 • E x p l o rin g M u sica l S paces

 of the tetrahedron. In a similar fashion, sum classes as transpositions of sum class 0. The
locate the chords A♭C♭E♭E♭ and FA♭CC. axis of transposition extends only from T0 to T3
(d) The omnibus progression shown in Figure before the original tetrahedron reappears in a
11.5.7b is a zero-​sum voice leading entirely different orientation. Any attempt to show this
within the sum-​class-​1 tetrahedron. As four-​dimensional construction in a two-​dimen-
a whole it is not a straight-​line voice-​ sional image, as in Figure 11.5.9, is inevitably
leading (why not?), but it divides into three less than completely satisfactory. The tetrahedra
segments of straight-​line voice leading at at the top and bottom of this figure both cor-
chords γ and ζ. Chords α, δ, and η were respond to the three-​dimensional sum-​class-​0
identified in part (c) above; the other four tetrahedron from Figure 11.5.4; they are identi-
chords are dominant seventh chords. Graph cal except for their orientation. The vertical lines
this omnibus in the tetrahedron.56 connecting the two appearances of this tetrahe-
(e) The omnibus could continue through six dron depict the fourth dimension in this figure.
more chords, arriving back at A♭C♭E♭E♭. The sum-​class indications at the right side of the
  Graph this extended voice leading. figure run the risk of being somewhat mislead-
ing, as every sum class occupies a three-​dimen-
The sum-​ 0 tetrahedron is a three-​ sional tetrahedral region in four-​
class-​ dimensional
dimensional cross section of four-​ voice OP-​ space. The two appearances of the sum-​class-​0
space. The full space, which we will call a Möbius tetrahedron cap the hyperprism of four-​voice
hyperprism, is obtained by extruding the tetrahe- OP-​space at its ends; conjoining the two tetra-
dron through a fourth dimension corresponding hedra in the appropriate orientations imposes a
to transposition, thereby producing the other twist through four-​dimensional space analogous

FIGURE 11.5.9 Four-​voice OP-​space

The Geometry of OPTIC Spaces • 465

to the twist described previously for the Möbius root motion in alternating minor and major
prism of three-​voice OP-​space.57 thirds; in the triadic excerpt the implied voice
The octatonic seventh-​ chord hypercube leading is smoother in the major-​third steps,
chain surrounds the central axis of four-​voice while in the seventh-​chord excerpt the minor-​
OP-​ space, much as the hexatonic triad cube third steps are smoother.
chain does in the three-​voice space. The hyper-
cubes, unlike the tetrahedra, are four-​dimen- Exercise 11.5.10 In three voices, many chords
sional objects in themselves. The hypercubes are related by single-​semitone voice leading to
are joined at the diminished seventh chords, inversionally related chords (P-​ and L-​related
but unlike augmented triads, no equal-​ tem- triads being among the many examples). Do any
pered diminished seventh chord belongs to sum equal-​tempered tetrachords have this property?
class 0. Consequently the oct12 hypercube is Give an example or explain why no such
split between the top and bottom of the figure; example
   can exist.59
the gluing of the two tetrahedra unites the two
halves in a complete hypercube. The intersection Example 11.5.11: Chopin, Prelude in E
of the oct12 hypercube with the sum-​class-​0 Minor, Op. 28, No. 4 Figure 11.5.12a shows
tetrahedron, appearing at both the top and the in slightly simplified form the chords in mm.
bottom of Figure 11.5.9, contains the four minor 2–​8 of Chopin’s E-​Minor Prelude, a widely
seventh and two French sixth chords examined studied passage celebrated for its remarkable
in Exercise 11.5.6c–​d. The center of the oct12 voice leading in four voices.60 Part (b) traces
hypercube coincides with the center of the sum-​ this voice leading in OP-​space. The vertical
class-​0 tetrahedron; this is the midpoint men- sonorities include all five qualities appearing
tioned in the last question in Exercise 11.5.6d. in the octatonic seventh-​chord hypercube but
We remarked in Section 11.3 that a major no others, and the passage consists entirely
or minor triad divides the octave nearly evenly, of smoothly descending motion within the
lies near the central axis of three-​ voice OP-​ hypercube chain. The path begins at B7 (V7 in
space, shares a hexatonic triad cube with its T4 E minor) in the oct23 hypercube and descends
and T8 transpositions, and therefore is related through oct12 and oct01 to reach oct23 a
to those two chords by smooth voice leading. second time, tracing different routes through
The analogous property for seventh chords near each hypercube but always passing through the
the central axis of four-​voice OP-​space is that a diminished seventh chords at cube junctions.
seventh chord falls in the same octatonic hyper- When g♯o7 is reached for the second time at the
cube with its T3, T6, and T9 transpositions. One end of the passage, it leads to iv6 in preparation
can see in the hypercube chain of Figure 11.5.3 for a half cadence in E minor (not shown). In
that an E7 chord can move to G7, B♭7, or D♭7 by general, voices move individually by semitone,
moving two voices by semitone (such a move traversing edges of the hypercubes. The only
occurs in chords β–​γ of the omnibus progres- exceptions are segments ⑥ and ⑦, in which two
sion of Figure 11.5.7b), while at least four semi- voices move together twice in succession. With
tones of motion are needed to reach any other the aid of these double moves the path traverses
dominant seventh chord. Tymoczko describes the oct12 and oct01 hypercubes more rapidly
this property as a fundamental contrast in the and
  spends most of its time in oct23.
voice-​leading potential of triads and seventh
chords: for triads the smoothest voice leadings Exercise 11.5.13 Figure 11.5.14 shows three
connect major-​ third–​related chords, but for chromatic passages from Chopin’s Etudes, Op.
seventh chords the smoothest voice leadings 10. Chopin’s complex textures are reduced to
connect minor-​third–​related chords.58 In both block chords, but the voice leading in the figure
cases, however, the location of chords near the faithfully follows the score, and the spellings are
central axis allows for a richer and more varied Chopin’s. Graph each of these voice leadings in
network of voice-​ leading connections than if four-​
   voice OP-​space.61
we restricted our attention to the symmetrical
chords on the axis itself. The reader may find it Exercise 11.5.15 Figure 11.5.16 shows a
instructive to revisit the pair of examples from sequence from Dvořák’s Seventh Symphony.
Rimsky-​Korsakov’s Christmas Eve in Example (The root movement in this passage was
8.3.19 and the associated Exercise 8.3.21 with examined in Exercise 3.5.7c.) The voice leading
this contrast in mind. Both examples feature shown is simplified, the instrumental parts

466 • E x p l o rin g M u sica l S paces

FIGURE 11.5.12 Chopin, Prelude in E Minor, Op. 28, No. 4: (a) voice leading in mm. 2–​8; (b) path in
four-​voice OP-​space (after Tymoczko)

being highly disjunct. Graph this passage in Exercise 11.5.19 The hypercube chain of
OP-​space, noting the systematic alternation of Figure 11.5.3, situated within the Möbius
zero-​
   sum and parallel voice leading. hyperprism of Figure 11.5.9, is the four-​voice
analog of the three-​voice cube chain first shown
Exercise 11.5.17 Figure 11.5.18 shows two in Figure 10.2.1, which lies within the Möbius
excerpts from works by Liszt, each featuring prism of Figure 11.3.4. What is the analog
smooth voice leading in the vicinity of a single of this cube chain within the Möbius strip of
diminished seventh chord. Graph both passages two-​voice OP-​space (Figure 11.2.6)? Observe
in  
OP-​space. the types of sonorities appearing in this chain,

The Geometry of OPTIC Spaces • 467

FIGURE 11.5.14 Voice leading from Chopin Etudes for Exercise 11.5.13: (a) Etude in A♭ Major, Op.
10, No. 10, mm. 49–​53; (b) Etude in F Major, Op. 10, No. 8, mm. 37–​40; (c) Etude in E♭ Minor, Op. 10,
No. 6, mm. 33–​38

FIGURE 11.5.16 Chordal reduction of Dvořák, Symphony No. 7 in D Minor, Op. 70, IV, mm. 233–​40,
for Exercise 11.5.15

FIGURE 11.5.18 Two excerpts by Liszt for Exercise 11.5.17: (a) Années de pèlerinage, troisième année,
No. 2, “Aux cyprès de la Villa d’Este: Thrénodie I,” mm. 107–​11; (b) Valse oubliée No. 1, mm. 1–​7

and the type of twist required to glue its ends and depict these spaces. The abbreviated discus-
together.
   62
sion below suggests one way in which they may
be conceived.63
11.6 FOUR-​VOICE T-​, OPT-​, Because T equivalence reduces the dimen-
AND OPTI-​S PACE sionality of a space, four-​voice T-​space and its
compounds such as OPT-​ and OPTI-​space are
The construction of OPT-​ and OPTI-​space, three-​dimensional. We introduced T equivalence
already complex in the case of three-​note chords, for three-​voice spaces by projecting an xyz coor-
becomes even more so in the case of four-​note dinate system into a plane (Figure 11.4.1b); all
chords; moreover, there are many ways to describe three coordinate axes (or more accurately their

468 • E x p l o rin g M u sica l S paces

FIGURE 11.6.1 Four-​voice T-​space in three dimensions

two-​dimensional projections) are visible in this for the T-​class (3, 7, 10, 12)T (an added-​sixth
plane, radiating away from each other at equal or inverted minor seventh chord), the balanced
angles. In an analogous construction of four-​ form is (−5, −1, 2, 4), the wxy form is (−9, −5,
voice T-​space, the four dimensions of a wxyz −2, 0), and the xyz form is [0, 4, 7, 9]; xyz forms
coordinate system must be projected into a three-​ are the same as T normal forms. Points may be
dimensional space, as shown in Figure 11.6.1. In plotted in T-​space, using any form of their coor-
this three-​dimensional space, the four axes (or dinates, by moving appropriate distances in the
more accurately their three-​dimensional projec- direction of each axis in turn, as was the case in
tions) point away from each other, forming equal the three-​voice T-​space of Figure 11.4.1b. Figure
angles at the origin. To understand the orienta- 11.6.1 shows the xyz forms of the coordinates
tion of these four axes in three dimensions, the of the cube’s eight corners. In the three dimen-
reader may find it helpful to envision the cube sions of four-​voice T-​space, distances along all
shown in the picture (this is an ordinary three-​ four axes are measured in the same units, and
dimensional cube, not a 4-​cube). The center of all eight corners lie at a distance of 12 from
the cube lies at the origin of the coordinate sys- the origin. The two-​dimensional projection in
tem, and each of the four coordinate axes passes the figure, however, foreshortens distances in
through two opposite corners of the cube. The some directions. In particular, distances in the w
scale of the figure, in which the diagonals of the dimension appear compressed because that axis
cube are of length 24, and its orientation with points roughly toward the viewer.
the z axis pointing to the right and the w and z
axes lying in a horizontal plane, have been cho- Exercise 11.6.2 Calculate the wyz and wxz
sen to facilitate the upcoming discussion.64 The forms of the T-​class (3, 7, 10, 12)T. Visualize the
view of T-​space in this figure is from a vantage location of this T-​class in Figure 11.6.1, using
point slightly above the wz plane. at least two different forms of its coordinates
Points in the three-​ dimensional space of (both should, of course, lead you to the same
Figure 11.6.1 stand for T-​ classes of pitch 4-​ location).
  
tuples. As was the case in three-​voice T-​space, any
T-​class may be represented by a balanced form Exercise 11.6.3 This exercise requires some
whose coordinates sum to 0; alternatively, any facility in trigonometry. Show that the angle
one of the four coordinates may be set equal to 0, formed at the center of a cube by any two
resulting in an xyz form, a wyz form, a wxz form, axes oriented like those in Figure 11.6.1 (for
and a wxy form for each T-​class. For example, example, the angle between the positive

The Geometry of OPTIC Spaces • 469

directions of the w and z axes shown there) is Layer 0 consists of multisets with doublings.
approximately
   109.5 degrees.65 Generally, for k =​0, 1, 2, or 3, the smallest inter-
val between any two notes of any tetrachord in
In the wz plane of Figure 11.6.1 lies a triangle layer k is exactly k. Some inversionally related
bounded by the positive z axis, the negative w pairs of tetrachords appear in the tetrahedron,
axis, and one edge of the cube. The three vertices such as 0239T and 0679T in layer 1; such pairs
of this triangle are the origin [0, 0, 0, 0]T, the T-​ are symmetrically positioned about the dotted
class [0, 0, 0, 12]T on the z axis, and [0, 12, 12, diagonal line in each layer. Apart from some
12]T, which is the xyz form of the T-​class on the duplications on the boundaries (discussed
negative w axis that may also be denoted (−12, below), every tetrachordal OPT normal form
0, 0, 0)T. Figure 11.6.4a shows the construction appears exactly once in the tetrahedron. This is
of a shallow tetrahedron whose base is this tri- why our tetrahedron is of interest: it is a fun-
angle. The vertex at the top of this tetrahedron damental region for four-​voice OPT-​space, the
is the T-​class [0, 3, 6, 9]T. A tetrahedron has space of T-​classes of pitch-​class multisets. (It is
four triangular faces; we will refer to the faces of not the normal region, because for many points
this one as the base (the triangle in the wz plane the T and OPT normal forms do not match. The
described above), the front face (with corners at relations 0 =​w ≤ x ≤ y ≤ z, however, imply that all
the origin, [0, 0, 0, 12]T, and [0, 3, 6, 9]T), the T normal forms within the tetrahedron are also
left face (barely visible in the perspective of the PT normal forms, so the tetrahedron is a subset
figure), and the right rear face (with corners at [0, of the normal region for PT-​space.)
0, 0, 12]T, [0, 12, 12, 12]T, and [0, 3, 6, 9]T). As in the other spaces considered in this
chapter, proximity in this space indicates prox-
Exercise 11.6.5 imity in the sense of voice leading. For example,
(a) The right rear face of the tetrahedron is a the tetrachord in layer 1 of the tetrahedron
portion of one face of the cube from Figure with OPT normal form 0236OPT is related by
11.6.1. The plane of this face is defined by single-​semitone voice leading to the tetrachords
the equation w +​ x +​12 =​y +​ z. Verify that 0125OPT, 0235OPT, 0237OPT, and 0347OPT in layer
the four corners of one face of the cube 1 as well as to 0226OPT and 0336OPT in layer 0
satisfy this equation, as does (0, 3, 6, 9). and 0246OPT in layer 2, and it lies near each of
(b) Write similar equations for the other faces those points in the tetrahedron. As an aid to
  of the cube. visualizing relationships among the various
layers in Figure 11.6.4b, light outlines in each
Within the tetrahedron, lattice points whose layer show the extent of the higher layers. The
xyz forms have integer coordinates are arranged tetrachord 0236OPT in layer 1 lies directly over a
in four horizontal layers, shaded and numbered point midway between 0226OPT and 0336OPT in
0 through 3 in the figure. Layer 0, with the layer 0. The triangles formed by the lattice points
lightest shading, is the base of the tetrahedron; within each layer are not equilateral (nor are the
layer 3 is the single vertex [0, 3, 6, 9]T at the top. entire layers); according to Exercise 11.6.3, the
Top-​down views of the four layers are shown in angle formed at the origin between the positive
Figure 11.6.4b, where all 152 lattice points are z and negative w axes is approximately 180° −
labeled in two ways. The labels omit commas 109.5° =​70.5°, not 60° as in an equilateral tri-
and brackets for compactness, and the letters t, angle. In layer 1, 0236OPT lies slightly closer to
e, and v stand for 10, 11, and 12. The upper label the tetrachords named above than to 0126OPT
on each point is its xyz form (T normal form). or 0346OPT, two tetrachords to which 0236OPT is
The point labeled 000v at the lower right corner not related by single-​semitone voice leading.
of layer 0 is the same point that appears in both Careful readers may have noted an omis-
Figures 11.6.1 and 11.6.4a with the label [0, sion in the above list of tetrachords adjacent
0, 0, 12]T. The coordinates of xyz forms within to 0236OPT. By moving any of the four voices
the tetrahedron satisfy the relations 0 =​w ≤ x ≤ up or down, 0236OPT is related by single-​semi-
y ≤ z ≤ 12. The lower label on each point is its tone voice leading to eight other tetrachords,
OPT normal form, which may be obtained only seven of which have been accounted for.
from the T normal form by the algorithm from Evidently 0136OPT should appear in this list—​
Section 10.4. For instance, the OPT normal but the tetrachord with OPT normal form 0136
form corresponding to the T-​class 0239T in appears at a considerable distance from 0236OPT
layer 1 is 0356OPT. in layer 1 of Figure 11.6.4b, seemingly too far

470 • E x p l o rin g M u sica l S paces

FIGURE 11.6.4 (a) A fundamental region for four-​voice OPT-​space in four-​voice T-​space; (b) layers of
lattice points in four-​voice OPT-​space

The Geometry of OPTIC Spaces • 471

FIGURE 11.6.6 The voice leading (0, 2, 3, 6) → (0, 1, 3, 6) through the boundary of four-​voice
OPT-​space

for a single-​semitone connection. To understand We can now explain the enigmatically

this apparent anomaly, we must examine the large apparent distance between 0236OPT and
complex nature of the boundaries of the tetrahe- 0136OPT. In T-​ space, the straight-​ line voice
dron—​complexities we have learned to expect in leading from α =​(0, 2, 3, 6) to β =​(0, 1, 3, 6)
fundamental regions. is a short line segment parallel to the x axis.
In each layer, some OPT-​classes appear twice, To understand the path of this voice leading
along the edges parallel to the w and z axes. For in OPT space we must recognize that although
example, OPT normal form 0125OPT appears tetrachords αOPT =​0236OPT and βOPT =​0136OPT
twice in layer 1: on the bottom edge (that is, the both lie in layer 1, the continuous voice leading
front face of the tetrahedron) it appears as the α → β includes intermediate points that are not
OPT normal form of the T-​class 0125T, while on equal-​tempered and lie not in any of the four
the left edge (the left face of the tetrahedron) it layers shown in Figure 11.6.4 but between those
appears as the OPT normal form of a different T-​ layers. At its midpoint this voice leading passes
class, 0789T. In fact, in OPT-​space the front face through (0, 11 2 , 3, 6), which lies in layer 11 2 ,
of the tetrahedron is conjoined with the left face, midway between layers 1 and 2. As Figure
as signaled by arrowheads along the bottom 11.6.6 shows, [0, 11 2 , 3, 6]OPT lies on the bound-
and left edges of the layers in Figure 11.6.4b. ary of layer 11 2 of the tetrahedron (it contains
The right rear face, meanwhile, divides into two occurrences of pitch-​ class interval 11 2 ),
two halves at the diagonal dotted line, and in and therefore appears twice in that layer. One
every layer the same OPT normal forms (such as of the two appearances, labeled γ, corresponds
0156OPT in layer 1) appear on that face on both to T normal form [0, 11 2 , 3, 6]T and lies nearly
sides of that line. The two halves of the right rear above 0236OPT in layer 1, while the other,
face are thus conjoined in reversed orientation, γ′, corresponds to T normal form [0, 6, 71 2 ,
indicated by double arrowheads . Generally, 9]T and lies closer to 0136OPT. In OPT-​space,
points on the boundary of layer k are tetra- the voice leading α → β begins with segment
chords in which the smallest pitch-​class interval ① in Figure 11.6.6, but it never gets to β =​[0,
k occurs twice; thus 0125OPT and 0156OPT each 1, 3, 6]T, which lies outside the tetrahedral fun-
contain two occurrences of pitch-​class interval 1. damental region for OPT-​ space. Instead, the
Because of these duplications, only 116 different path hits the boundary of the tetrahedron at
OPT normal forms are represented by the 152 γ, teleports to the other boundary point γ′ rep-
lattice points in the tetrahedron. Of these 116 resenting the same OPT-​class, and from there
OPT normal forms, 73 lie in layer 0 and contain follows segment ②, arriving at its destination
pitch-​class doublings, leaving 43 distinct OPT β′, the representative of the same OPT-​class as
normal forms of four-​note pitch-​class sets in lay- β within the fundamental region, after a total
ers 1, 2, and 3. motion through only a single semitone.

472 • E x p l o rin g M u sica l S paces

Exercise 11.6.7 Consider the two straight-​ tetrahedron. As with other false boundaries we
line voice leadings (0, 2, 5, 8) → (0, 3, 5, 8) and have encountered, points on these faces are not
(0, 2, 5, 8) → (0, 2, 5, 9). In OPT-​space, these singularities; it would be possible to construct
voice leadings define two different paths from different fundamental regions for OPT-​space
0258OPT (a half-​diminished seventh chord) to in which these points do not lie on boundaries
0358OPT (a minor seventh chord). Describe both at all. The base of the tetrahedron (layer 0), in
paths
   geometrically.66 contrast, is a true mirror boundary. For an illus-
tration of a reflection at this boundary, consider
Exercise 11.6.8 Draw geometrically accurate the three central chords β → γ → δ from the five-​
pictures of the front, left, and right rear faces chord omnibus progression in Figure 11.6.9a.
of the tetrahedron, labeling all lattice points in These chords execute a voice exchange in which
each
   face with OPT normal forms.67 a dominant seventh sonority (β, OPT normal
form 0368OPT) passes to an inverted form of
The front, left, and right rear faces of the the same chord (δ) via a second-​inversion minor
tetrahedron are false boundaries of OPT-​space: triad (γ). In OPT-​space, as shown in segments ③
a straight-​line voice leading similar to the one and ④ in Figure 11.6.9b, the voice leading moves
in Figure 11.6.6 can pass through any of these from β =​0368OPT to γ =​0377OPT and then back
boundaries and re-​ emerge elsewhere in the to δ =​ β =​0368OPT. Chord β lies in layer 2, and

FIGURE 11.6.9 (a) A five-​chord omnibus progression; (b) the omnibus progression in four-​voice OPT-​
space; (c) a long omnibus progression

The Geometry of OPTIC Spaces • 473

γ lies directly beneath β in layer 0; at γ the path Part (c) of the figure extends the progres-
bounces back vertically in the direction from sion γ → δ → ε into a so-​called long omnibus,
which it came. The path crosses through layer 1 modulating repeatedly by minor third. The long
in both directions, but no equal-​tempered tetra- omnibus, studied previously in OP-​space in
chord occupies that location in layer 1. Exercise 11.5.8d–​e, is a segmental straight-​line
The five-​chord omnibus is a straight-​line voice voice leading; the bass descends steadily, but
leading, the outer voices both moving at the rate in cpitch4 or in OP-​space the motion changes
of one semitone per chord. Chords α, β, δ, and direction at ε (and at ε1, ε2, …), whenever one of
ε, however, are all the same point 0368OPT in the three upper voices stops moving as another
layer 2 of OPT-​space. This may seem counterin- starts. In OPT-​ space, the long omnibus sim-
tuitive: how can a straight line pass through the ply retraces segments ③–​⑥ from Figure 11.6.9b
same point four times in such close succession? again and again. This path reflects off the base
The bounce off the mirror boundary at γ is only of the tetrahedron at every appearance of γ (γ1,
part of the answer. A complete answer must take γ2, …). The paired segments ⑤–​⑥ occur in every
into account the other characteristic parts of the cycle, the path turning a corner each time ε is
omnibus: the distinctive moves α → β and δ → ε, reached, heading down to γ via segment ③ to
pairs of consecutive 0368OPT chords related by begin the next cycle.
contrary single-​semitone voice leading in two The α → β and δ → ε moves in the omnibus
voices. The two moves are essentially the same, progression illustrate smooth voice leading
as δ → ε retrogrades α → β in OPT-​space; we shall between T3-​related dominant seventh chords.
consider the δ → ε pair. As noted previously, such a connection is pos-
In cpitch4, enumerating voices from the sible because the chords divide the octave nearly
bass upwards, δ and ε are the 4-​tuples (2, 8, 11, evenly. In OP-​space, this nearly equal division
16) and (1, 8, 11, 17) respectively. Both lie in results in a location near the central axis of the
layer 2 of OPT-​space because in both cases the space. In OPT-​space, that central axis of dimin-
smallest pitch-​class interval between two notes ished seventh chords collapses to the single
of the chord is 2: between pitch classes 2 and 4 point 0369; accordingly, the most intricate parts
in chord δ and between 11 and 1 in chord ε. The of the voice leading in the omnibus progression
straight-​line path δ → ε, however, rises above unfold between layers 2 and 3 of the tetrahedron
layer 2. It is tempting to suppose that this path of OPT-​space, near the 0369 vertex.
reaches its apex in OPT-​space at its midpoint,
the point (11 2 , 8, 11, 161 2 ) in layer 21 2 (21 2 being Exercise 11.6.10 Between chords α and ε,
the pitch-​class interval between 11 and 11 2 ). the five-​chord omnibus in OPT-​space passes,
That is not correct: the high point occurs ear- twice each, through one point in layer 1 and
lier, only one-​third of the way from δ to ε, at the two different points in layer 21 2 . Determine the
point η =​(12 3 , 8, 11, 161 3 ) in layer 22 3 . Pitch-​ coordinates of these three points in cpitch4
class interval 22 3 occurs twice in η, between and their OPT normal forms, and locate them
pitch classes 11 and 12 3 and also between 12 3 on  
the path in Figure 11.6.9b.
and 41 3 . Therefore η, with OPT normal form [0,
3, 52 3 , 81 3 ]OPT, lies on a boundary of layer 22 3 Exercise 11.6.11 As an alternative to the
and appears on two different faces of the tetra- long omnibus of Figure 11.6.9c (though less
hedron. Figure 11.6.9b shows these two points, viable musically), suppose that the five-​chord
η and η′, both representing the same OPT-​class. progression from 11.6.9a is extended as a
The point η, on the tetrahedron’s front face, single straight-​line voice leading, the top voice
lies directly above δ. The omnibus voice lead- continuing to ascend as the bass descends. After
ing, after completing the vertical segment ④ (γ those two voices have each traversed an octave,
→ δ), continues via segment ⑤ up to η, where it chord α will reappear, its bass an octave lower
hits the false boundary; it re-​emerges on the left and soprano an octave higher than before. Trace
face at η′ and returns via segment ⑥ to ε (the the graph of this extended straight-​line voice
same point as δ in OPT-​space). Though δ and ε leading
   in OPT-​space.
are consecutive chords in the omnibus, the path
joining them encompasses both segments ⑤ Light dotted lines bisect the base and the
(δ → η) and ⑥ (η′ → ε). The α → β portion of the right rear face of Figure 11.6.4’s tetrahedron and
omnibus (segments ①–​②) traces segments ⑤–​⑥ each of its layers, dividing the tetrahedron in
in reverse, α → η′ and η → β. half. These dotted lines, along with the edge from

474 • E x p l o rin g M u sica l S paces

0000 to 0369, lie in a vertical plane; this plane Exercise 11.6.12 Five tetrachords are listed
separates the two halves of the tetrahedron and below in OPT normal form. Determine the
consists of inversionally symmetric tetrachords OPT normal form of the inversion of each
such as 0336OPT in layer 0, 0347OPT in layer 1, tetrachord, and locate the corresponding pairs
and 0358OPT in layer 2. The vertical plane acts as of points in Figure 11.6.4b.
a plane of inversional symmetry for OPT-​space.
If a tetrachord lying on one side of this plane is (a) 0237
inverted, the inverted tetrachord will occupy the (b) 0368
corresponding point in the opposite half of the (c) 0344
tetrahedron. For example, OPT-​class 0136OPT (d) 0158
lies near the upper left corner of layer 1 in Figure (e)  0027
11.6.4b, and its inversion 0356OPT lies at the cor-
responding point near the lower right corner of Each half of the tetrahedron is a smaller tet-
the same layer. rahedron in its own right, and either half could

FIGURE 11.6.13 (a) A fundamental region for four-​voice OPTI-​space in four-​voice T-​space; (b) layers
of lattice points in four-​voice OPTI-​space

The Geometry of OPTIC Spaces • 475

FIGURE 11.6.15 Four voice leadings for Exercise 11.6.14

serve as a fundamental region for four-​voice represent only 72 different OPTI normal forms.
OPTI-​space. Thus one OPT fundamental region Of these, 43 lie in layer 0 and contain pitch-​
comprises two OPTI fundamental regions, as class doublings, leaving 29 distinct OPTI normal
was also the case with the corresponding three-​ forms of four-​note pitch-​class sets in layers 1, 2,
voice spaces (Figures 11.4.12 and 11.4.21; also and 3—​the 29 tetrachordal set classes enumer-
recall Exercise 11.4.24). We choose the half-​tet- ated in standard set-​class tables.
rahedron adjacent to the z axis (the front half in
the view of Figure 11.6.4a) as our OPTI funda- Exercise 11.6.14 Four straight-​line voice
mental region. Figure 11.6.13a shows a three-​ leadings are shown in Figure 11.6.15. Each
dimensional view of this smaller tetrahedron of them crosses or reflects off of one of the
in T-​space; the adjacent rear half of the origi- boundaries of the OPTI fundamental region.
nal larger tetrahedron appears in light outline. Determine the OPT and OPTI normal forms of
Figure 11.6.13b shows the 84 lattice points in each chord, and trace the voice leadings in OPT-​
the four layers of the OPTI fundamental region, space
   and in OPTI-​space.68
labeled with their OPTI normal forms. Every
OPTI normal form in 11.6.13b matches one of Exercise 11.6.16 A tetrachord of OPTI normal
the OPT normal forms in 11.6.4b, but some of form 0247 may be related by single-​semitone
the OPTI normal forms come from one half of the voice leading to tetrachords belonging to eight
original tetrahedron, some from the other half. other OPTI set classes. Graph these eight single-​
Though the geometry is difficult to visualize, the semitone voice leadings in OPT-​space and in
three dimensions of four-​voice T-​space are filled    space.69
OPTI-​
with copies of the tetrahedral OPTI fundamental
region in various orientations, analogous to the Exercise 11.6.17 Locate three different points
tiling of the plane of three-​voice T-​space by the in four-​voice OPTI-​space (Figure 11.6.13) that
OPTI fundamental region in Figure 11.4.11. are all OPTIC-​equivalent and must therefore be
Reducing the OPT fundamental region to the conjoined in OPTIC-​space. (Nothing of the sort
smaller OPTI region complicates the boundary is possible in three-​voice OPTI-​space. The fact
behavior further. The base of the tetrahedron, that three different points must be identified
layer 0, has been reduced to half its former begins to hint at the complexity of cardinality
size but remains a mirror boundary. The origi- equivalence
   in higher-​dimensional spaces.)70
nal plane of inversional symmetry is now a left
rear face of the smaller tetrahedron and has
become a mirror boundary also. The right rear NOTES
face, formerly a false boundary, is also halved in
size and has become yet another mirror bound- 1. The concept of orbifolds is a relatively recent one,
ary. Finally, the front face, though geometrically dating from the 1950s; the most important work on orbi-
unchanged, now consists of two halves conjoined folds is that of William Thurston from the 1970s. While
with each other in reverse orientation (indicated Weeks 2020 offers an excellent nontechnical introduc-
with double arrowheads in the layers of tion to manifolds, published descriptions of orbifolds
Figure 11.6.13b); these are false boundaries tend to be highly technical (e.g., Thurston 1980, 300).
like the halves of the right rear face of the OPT 2. The circle, which is both a one-​dimensional torus
fundamental region in Figure 11.6.4. Exercise 𝕋 1 and a one-​dimensional sphere 𝕊1, is the simplest
11.6.14 below provides examples of voice-​lead- example of a sphere among the OPTIC spaces. Spheres
ing behavior at all four of these boundary faces. can also result from the introduction of additional
Because of the boundary identifications, the 84 equivalence relations. For example, Tymoczko (2009a,
lattice points in the OPTI fundamental region 251) observes that 𝕊2 can arise musically as a space

476 • E x p l o rin g M u sica l S paces

we might denote cpitch4/​~TM, the space of ordered Similar notations appear in Tymoczko 2006 and in
four-​note pitch sets modulo transposition and multi- Callender, Quinn, and Tymoczko 2008.
plication by positive real numbers. 11. The word state suggests an analogy with the
3. The curious properties of this example pro- mathematical theory of computation: a space may
vide further justification for a fact noted in Chapter be regarded as a sort of abstract machine that at any
10: I equivalence is rarely invoked in the absence of moment may be in any of various possible states;
T equivalence. The pairings of notes in I-​classes and a voice leading describes a series of transitions from
the location of the singularity in I-​space are arbitrary, state to state.
determined by the choice of C4 as the zero point of 12. In this example, the lower voice is listed first in
pitch space. The arbitrariness vanishes when T equiva- all ordered pairs. It is not always possible to observe
lence is introduced (although spaces of two or more such a rule consistently while respecting voice leading,
voices must be considered in order for TI equivalence because voice leadings may include voice crossings.
to be a nontrivial relation). 13. Extending an idea from Section 2.3, it would be
4. Every manifold (without singularities) is an possible to formalize the temporal nature of voice lead-
orbifold by definition. The formal definition of orbi- ings by modeling them not merely as paths in cpitchn
folds requires that every point have a neighborhood but as functions from time into cpitchn. For example,
topologically indistinguishable from some point in a segment α → β in Figure 11.2.8 could move from α =​
quotient space of ℝn under some group action. The (−1, 3) at time t =​0 to β =​(5, 9) at t =​1 via the function
spaces derived from the O, P, T, and I symmetries are f(t) =​(−1 +​6t, 3 +​6t). A straight-​line voice leading is
quotient spaces of exactly this kind and so are orbi- one modeled by a linear function, all of whose compo-
folds automatically. nents are of the form a +​ bt (as they are in this exam-
5. Hint for Exercise 11.1.2: The pitches F3, G3, F4, ple). We will not rely on such formalism often, but it is
and G4 are all OI-​equivalent (they share the OI normal useful occasionally; see Exercise 11.3.13.
form 5). This space has not one but two singularities. 14. In a convenient abuse of notation, we use the
6. Hint for Exercise 11.2.2: All the points (x, y) on a letter α both for a point in cpitch2 (the ordered pair
given line of transposition share a common value of (−1, 3)) and for a point in OP-​space (the OP-​class BD♯).
int(x, y), an interval in pitch space. What intervals cor- At times we may use the same letter for corresponding
respond to the three lines shown? equivalence classes in other quotient spaces as well.
7. Hint for Exercise 11.2.5c: If you draw vertical In a more precise but cumbersome notation, we could
lines through (0, 6), (0, 0), and (6, 0), dividing the O write αOP for the OP-​class.
normal region into four parts (two triangles and two 15. Hints for Exercise 11.2.9: In part (c), although
trapezoids), the OP normal region will be the union of the dyads must be exactly the same size as equal-​
two of these subregions; your answers to (b) should tempered ic3 and ic5, the pitches involved need not
help you figure out which two. The OP normal region be equal-​tempered, and so the dyads need not corre-
is not a connected region in the plane, but if you con- spond to lattice points in Figure 11.2.6. (For example,
ceive of the O normal region as a torus, the OP normal if both notes in an equal-​tempered DF dyad are raised
region is a connected region within that torus. by a quarter-​tone, the resulting dyad is still ic3.) In
8. Though oriented differently, Figure 11.2.6 is part (d), you should find that one of the two pictures
otherwise similar to Figure 2 in Tymoczko 2006, 73, is unaffected by the change.
and to Figure 3.3.1 in Tymoczko 2011, 71. 16. Hints for Exercise 11.2.10a: This voice leading
9. The number of equal-​tempered two-​note pitch-​ consists of four segments of straight-​line voice lead-
class sets is the number of “combinations of 12 things ing, but three of the four must be broken into two
2 at a time,” which is 12! ∕ (2!10!) =​66. (See Hook apparently separate segments in OP-​space. A straight
2007d for the combinatorial principles underlying line from the point labeled DD to the point labeled AD
such calculations.) In addition to these 66 dyads, there in Figure 11.2.6 is not the correct way to model the
are twelve unison multisets CC, C♯C♯, …, for a total of first segment. The last segment involves straight-​line
78 multisets. There are 85 labeled points in Figure voice leading that is not parallel, zero-​sum, or oblique,
11.2.6, because the seven points along the top edge so this line segment should appear at an angle differ-
duplicate those at the bottom. ent from any of those in Figure 11.2.8.
10. The orbifold of two-​voice OP-​space cpitch2/​~OP 17. Hints for Exercise 11.2.10b: Although this coun-
might be called 𝕋 2/​𝒮2 by mathematicians, indicat- terpoint would presumably be performed in discrete
ing that it is a quotient space of O-​space (the torus pitches, you may model it as a continuous voice lead-
𝕋 2) produced by the action of the symmetry group ing as in previous analyses. The regularity of the chain
(𝒮2) associated with permutational equivalence. of 7–​6 suspensions may not be apparent in your pic-
Analogously, n-​voice OP-​space is the orbifold 𝕋 n/​𝒮n. tures. Why not?

The Geometry of OPTIC Spaces • 477

 18. In this and a few subsequent examples we use as the G-​major, B-​major, and E♭-​major triads visible in
the word isomorphic in a topological sense that we Figure 11.3.4. These trichords do not lie in the same
have not defined precisely. In these cases the corre- places in the triangular cross section as the equal-​tem-
spondence between points of the spaces should be pered trichords of sum class 0 in Figure 11.3.1; they lie
clear even if the conditions that such an isomorphism at the centers of some (not all) of the small equilateral
is required to satisfy have not been spelled out, and triangles formed by the gridlines in that figure. The
the similarity of form should be clear as well. The major triads lie at points corresponding to ρ, σ, and τ.
intuition underlying topological isomorphism was 28. Hints for Exercise 11.3.8: Start by locating the
described in Section 1.1: topological properties are projections of the listed points in the sum-​class-​0
preserved if a space is stretched, compressed, bent, plane in a copy of Figure 11.3.1, indicating the layer
twisted, or continuously deformed in other ways, of the prism to which each point belongs by noting
but without tearing, punching holes, or pasting its sum class. In the perspective of Figure 11.3.4, the
together parts that were not originally connected. four points containing pitch-​class doublings all lie on
Topological isomorphisms are called homeomorphisms the front face of the prism; the other points are inte-
by mathematicians. rior to the prism, but all lie fairly close to the front
19. Figure 11.2.12b does not show anything analo- face. Techniques other than those described above can
gous to the balanced forms of 11.2.12a. In fact, every often be devised to locate points in OP-​space. Locating
two-​voice OT-​class contains not one but two ordered the B-​diminished triad for part (b), for example, may
pairs, related by T6, whose coordinates sum to 0 mod be facilitated by the observation that the straight-​line
12. For example, (11, 1) and (5, 7) are both “balanced” voice leading DFA → B♭DF, easily located in Figure
in this way, but they are OT-​equivalent, both corre- 11.3.4, could be extended through another semitone
sponding to the normal form [0, 2]OT. to arrive at BDF.
20. Hint for Exercise 11.2.15: The five pairs are rep- 29. Hint for Exercise 11.3.9: To depict the con-
resented by four different points in T-​space but only nection from the bottom of the figure to the top
two different points in OT-​space and a single point in accurately, notice what chord the straight-​line voice
OPTI-​space. leading between the C-​ major and A-​ minor triads
21. Hint for Exercise 11.2.17: One of the straight-​ passes through at its midpoint. See Figure 3.11.6b in
line voice leading segments in Figure 11.2.8b consists Tymoczko 2011, 108.
entirely of points that map to a single point of T-​space; 30. Passages similar to Figure 11.3.10a, featuring
this segment therefore corresponds to no movement major thirds, augmented triads, and other whole-​
at all in any of the spaces in Figure 11.2.12. tone sonorities moving chromatically in contrary
22. Three-​voice OP-​space is described in Tymoczko motion, occur in Grieg’s “Illusion,” Op. 57, No. 3,
2006 (especially the supporting online material) and at mm. 27–​31, and in several late works of Sibelius,
in Chapter 3 of Tymoczko 2011. among them the storm depiction in the Prelude to
23. Hint for Exercise 11.3.3b: The major triad CEG The Tempest, Op. 109, and mm. 523–​576 of Tapiola,
belongs to sum class 11. What fractional transposition Op. 112.
Tr of this triad belongs to sum class 0? 31. Hints for Exercise 11.3.11c–​d: For the T4 trans-
24. Figure 11.3.4 resembles Figure S5 in the position, the starting and ending points α and η
supporting online material to Tymoczko 2006, as remain unchanged, but the path in between is differ-
well as Figure 3.8.2 in Tymoczko 2011, 86, one of ent. For the T2 transposition, the path starts at η and
the most characteristic and widely cited images in ends at α—​but it still ascends!
Tymoczko’s work. 32. Hints for Exercise 11.3.12: For (c), construct
25. Figure 11.3.4 slightly exaggerates the vertical a path that reflects off the front wall of the prism.
dimension for clarity, giving the cubes a somewhat Because α and η lie at the centers of the sum-​class-​3
elongated appearance. and sum-​class-​9 planes, respectively, a bounce point
26. Adding O normal forms to Figure 11.3.4 θ at the center of the front edge of the sum-​class-​6
requires choosing one of the six possible orderings of plane will work. Determine the chord represented by
the three voices. Six different regions in O-​space result θ (it is one of the points located in Exercise 11.3.8),
from these choices, but as depictions of OP-​space they and describe the voice leading from α to θ to η. For (d),
all look the same. It is difficult to visualize, but six consider a path similar to the one in Figure 11.3.10,
copies of the Möbius prism slice at angles across the but compressed in the vertical dimension so that each
12 × 12 × 12 cube of O-​space, completely filling the cube. segment traverses a smaller vertical distance, and
27. Hint for Exercise 11.3.7a: The equal-​tempered therefore more segments, and more reflections, are
trichords of sum class 8 include CE♭F and its transpo- required to reach η. For (e), imagine several copies of
sitions by T4 and T8 (shown in Figure 11.3.5), as well the prism stacked on top of each other, and try joining

478 • E x p l o rin g M u sica l S paces

widely separated appearances of α and η via a path 41. Hints for Exercise 11.4.16: First put each triple
with a single bounce. in yz form. You should find one point illustrating each
33. Saariaho describes some of the procedures type of symmetry mentioned in the first paragraph of
at work in Vers le blanc in Saariaho 1987, 104–​05. the exercise, one point that illustrates two different
Callender 2004 analyzes this voice leading, displaying types because it lies at the intersection of two lines
graphs in T-​space, PTI-​space, and OPTI-​space, which of symmetry, and one point with no symmetry at all.
will be explored in Exercises 11.4.6 and 11.4.23 below. 42. Hints for Exercise 11.4.17: All the solid lines
Callender does not consider OP-​space, whose struc- have equations of the form x =​ y +​ c, x =​ z +​ c, or y =​ z +​
ture had yet to be described at the time of the article’s c, and all the dashed lines have equations of the form
publication. x +​ y =​2z +​ c, x +​ z =​2y +​ c, or y +​ z =​2x +​ c, where c is
34. The organization of Section 11.4 is indebted to some integer multiple of 12. Parallel lines have equa-
Callender 2004. tions that differ only in the value of c.
35. In contour theory (Morris 1993), the “middle-​ 43. Two different OPT-​equivalent pitch triples can-
low-​ high” contour is notated 〈1 0 2〉. The online not both be in OPT normal form. In this case [0, 2, 7]
supplemental material to Callender, Quinn, and and [0, 0, 6] are in OPT normal form, while (0, 5, 7)
Tymoczko 2008 includes commentary on contour and (0, 6, 6) are not. Only the lower of the two false
theory in relation to the OPTIC spaces; see especially boundary segments contains normal forms.
their Figure S14, which proposes that contours may 44. Three-​voice OPT-​space may be described as a
be distinguished more finely by associating a contour cone. In Figure 11.4.21a, the reader may imagine pull-
not with a large region in T-​space but with a direction ing the vertex at [0, 4, 8] up out of the plane of the
from the origin. paper as the two false boundary segments are glued
36. Hints for Exercise 11.4.6: The path in T-​space is a together, producing a cone-​shaped figure with a sharp
straight line from αT to βT. See Figure 12 in Callender point at [0, 4, 8]. See Callender, Quinn, and Tymoczko
2004. (Callender writes <9, 11>T for the T-​class whose 2008, 347, for a three-​dimensional depiction of this
yz form is [0, 9, 11].) The answer to the first question space. Topologically, the cone could be flattened to a
in (b) may seem counterintuitive; remember that two plane region such as a disk, but the cone may be dif-
pitch triples coincide in T-​space if they are transpo- ferentiated from a disk using distance measurements
sitionally equivalent, so it is possible for two voice to be described in Chapter 12.
leadings to describe the same path through this space 45. Hints for Exercise 11.4.23: In your PT-​space
even if they do not comprise the same pitch-​class sets. graph from Exercise 11.4.6, you should see that both
Describing the moving triple precisely as a function of segments, from α to γ and from γ to β, cross the axis
time, as in Exercise 11.3.13, is not necessary for this of inversional symmetry, so two new bounces will
exercise, though you may use the result of your cal- appear in the PTI-​space graph. Using your formu-
culations in the earlier exercise to locate γ precisely in las from Exercise 11.3.13 and the equation for the
PT-​space. axis of symmetry, you should be able to locate these
37. Strictly speaking, the dashed line is a line of I points exactly. In PTI-​space, α lies to the right of the
symmetry for PT-​classes, not for pitch triples. A pitch line of OI symmetry, so in OPTI-​space the first part of
triple such as (0, 2, 4) is not I-​symmetric but PTI-​ this graph will need to be reflected to the left side of
symmetric, as illustrated by the chain of equivalences that line. As pitch-​class sets, α and β have the same
(0, 2, 4) ~I (0, −2, −4) ~T (4, 2, 0) ~P (0, 2, 4). prime form, so the entire path forms a closed loop in
38. The pitch triple (0, 2, 7) is not OI-​symmetric OPTI-​space, perhaps bearing a loose resemblance to
but OPTI-​symmetric, as illustrated by the chain of the graph in Figure 11.4.18d. See Figures 14 and 15
equivalences (0, 2, 7) ~O (0, 2, −5) ~P (2, 0, −5) ~T in Callender 2004 for graphs in PTI-​ and OPTI-​space.
(0, −2, −7) ~I (0, 2, 7). The OPT-​space graph (not studied by Callender) may
39. Figure 11.4.11 is similar to Figure 9 in be produced by reflecting appropriate portions of the
Callender 2004. OPTI-​space graph to the opposite side of the inver-
40. Hints for Exercise 11.4.13: Two of the four sional axis, keeping in mind the special nature of the
graphs in part (a) are identical. For (c), what sonority false boundary segments described above. Sonorities
occurs midway through the R voice leading? Recall the α and β are OPT-​equivalent, so this path too should
discussion of the apparent intersection points formed end where it began.
by the edges representing R in Figure 10.2.15. For (d), 46. Hint for Exercise 11.4.24: In a valid fundamen-
a pure parallel voice leading represents no motion at tal region for ~OPT, every OPT-​class (with the possible
all in OPTI-​space, so if two voice leadings are the same exception of some on the boundary) must be repre-
except for a transposition of one chord, their paths in sented exactly once. In one of the regions proposed
OPTI-​space will look the same. in the exercise, you will find, for example, that the

The Geometry of OPTIC Spaces • 479

OPT-​class (0, 1, 3)OPT appears twice and (0, 2, 3)OPT in four-​voice pitch-​tuple space, in which voices w
not at all. When one fundamental region is exchanged and z ascend through four semitones while x and y
for another, a point originally on a false boundary may descend through an equal interval. In OP-​space, this
no longer be on a boundary at all, but a point on a mir- voice leading forms a straight segment from CEEE to
ror boundary must remain on a boundary. A♭A♭A♭C lying entirely on one face of the sum-​class-​0
47. The points [0, 0, 5]OPT and [0, 5, 5]OPT lie at tetrahedron, passing through ε at its midpoint.
corresponding locations on the two mirror boundary 55. Hints for Exercise 11.5.6d: The voice leading (7,
segments of OPT-​space. To form OPTC-​space these 11, 1, 5) → (11, 11, 1, 1) describes a straight line seg-
two mirror boundaries must be conjoined, forming ment through the interior of the tetrahedron from
a false boundary that essentially continues the false D♭Fr to BBC♯C♯, passing (though not at its midpoint)
boundary already defined by the other two edges of through the C♯m7 chord. The midpoint of the segment
the kite-​shaped region in Figure 11.4.21. When all mentioned in the last question is the center of the
appropriate identifications are made, three-​ voice tetrahedron. The six chords mentioned in (c) and (d)
OPTC-​space has the topology of a sphere. form the vertices of an octahedron.
48. Hint for Exercise 11.4.25: One of the OP normal 56. Hint for Exercise 11.5.8d: The voice leading in
forms labeled in Figure 11.3.1 is OPC-​equivalent to cpitch4 consists of three straight-​line segments, but
one of the trichords in Exercise 11.3.8. in OP-​space there are more segments, as the path
49. A 4-​cube is sometimes called a tesseract. In light must bounce off a face of the tetrahedron at any chord
of the several ways to draw a cube that we encountered with a doubled pitch class.
in Section 3.1, it should come as no surprise that many 57. What we are calling a hyperprism is more
different graphic depictions of a 4-​cube are possible. technically known as a uniform prismatic polytope.
50. Hint for Exercise 11.5.2: In a 3-​cube, each vertex The Möbius hyperprism in Figure 11.5.9 (without the
belongs to three edges, each edge belongs to two faces, chain of hypercubes) resembles Figure S6c in the sup-
and each face belongs (of course) to one 3-​cube. In a porting online material to Tymoczko 2006, as well as
4-​cube, each vertex belongs to four edges, each edge Figure 3.9.1c in Tymoczko 2011, 94.
belongs to three faces, each face belongs to two facets, 58. See Tymoczko 2011, 97–​103, a section titled
and each facet belongs to one 4-​cube. “Triads are from Mars; seventh chords are from Venus.”
51. In Section 10.2 we noted the relation- Tymoczko presents statistics showing that chord pro-
ship between the hexatonic triad cube chain and gressions in the works of Schubert and Chopin reflect
Douthett’s Cube Dance graph. The octatonic seventh-​ this difference, with juxtapositions of major-​ third–​
chord hypercube chain is closely related to another of related triads and minor-​third–​related seventh chords
Douthett’s graphs, dubbed “Power Towers” (Douthett occurring more commonly than the reverse.
and Steinbach 1998, 256). The three “towers” in the 59. Hint for Exercise 11.5.10: Think about sum
published Power Towers graph are not complete classes.
hypercubes (the French augmented-​sixth chords are 60. Especially relevant to the present analysis
missing), though an earlier version, presented by of Chopin’s Prelude are discussions of the piece by
Douthett in 1993, included complete hypercubes. Tymoczko and Cohn, both of whom study it in rela-
Another version of the graph appears in Cohn 2012, tion to the similar voice leading in Chopin’s Mazurka
158, with the name “4-​Cube Trio.” in F Minor, Op. 68, No. 4. Tymoczko (2011, 287–​93)
52. Hints for Exercise 11.5.5: Tetrachords of a given does not present the graph in Figure 11.5.12b explic-
sum class symmetric about a single axis in pitch-​class itly, but this path is implicit in his discussion. Cohn
space lie in a plane in OP-​space. The chords labeled in 2012, 160–​66 maps the path in his 4-​Cube Trio graph,
Figure 11.5.4 include enough chords symmetric about equivalent to the hypercube chain. The reduction in
the C–​F♯ and E♭–​A axes that you should be able to Figure 11.5.12a respells Chopin’s E♭ in mm. 2–​3 as D♯
locate these two planes. Tetrachords simultaneously (which does not affect the geometry) and disregards
symmetric about both axes lie in the intersection of the upper neighbor notes in the right-​ hand mel-
the two planes—​a straight line. ody. In generic space the labeled chord roots form a
53. Bulger and Cohn 2016 study subspaces of descending-​thirds pattern; see Exercise 14.1.32.
OP-​space spanned by voice leadings similar to those 61. Hints for Exercise 11.5.13: The patterns in (a)
in Figures 11.3.10 and 11.5.7a, in which some voices and (b) are sequential with the exception of the first
are constrained to move in strict parallel or contrary chord in (a), and remain entirely within the octatonic
motion. seventh-​chord hypercube chain. A pattern similar to
54. Hint for Exercise 11.5.6a: The most difficult task (a) in the cadenza of Chopin’s Prelude in C♯ Minor,
is locating γ and ε. One way to locate ε is to consider Op. 45, is studied in Childs 1998, 182–​84 and 189–​
the straight-​line voice leading (4, 12, 4, 4) → (8, 8, 0, 8) 90. The voice leading in (c) is somewhat irregular and

480 • E x p l o rin g M u sica l S paces

includes one chord lying outside the hypercube chain. are identical; the two halves of the right rear face are
This chord may plausibly be omitted, or it may be mirror images of each other. Each face is an isosceles
located approximately in the picture by studying the triangle with (0369) at its apex.
correspondence between the four voices and the four 68. Hints for Exercise 11.6.14: The voice leading in
dimensions of the nearest hypercube. part (a) repeats segments ③–​④ from Figure 11.6.9a,
62. Hint for Exercise 11.5.19: A two-​dimensional whose path in OPT-​space has already been described.
cube is a square. The cube chain in two-​voice OP-​space The voice leading in (b) lies within one layer of OPT-​
is depicted in Tymoczko 2011, 98, in an orientation space and passes through the plane of inversional
that differs from Figure 11.2.6. symmetry; in OPTI-​space this voice leading remains
63. For somewhat different presentations of four-​ in the same layer but bounces off the left rear bound-
voice OPT-​ and OPTI-​spaces, see Callender, Quinn, ary. The paths in (c) and (d) each begin at a point in
and Tymoczko 2008, particularly Figure S5 in the one layer and soon reach a boundary point in a differ-
Supporting Online Material. See also their Table ent layer. The path in (c) reflects at a right angle to its
S2, which describes ways to construct fundamental starting point in OPTI-​space; the path in (d) does not,
regions for n-​voice spaces involving all possible com- though it does return to its original layer.
binations of the O, P, T, and I symmetries. For ear- 69. If multisets with doublings are not consid-
lier depictions of tetrachordal set classes in four-​voice ered, 0247 is the only equal-​tempered four-​note set
OPTI-​space, see the two-​dimensional graph in Straus class related to eight other set classes in this way
2003, 339, and the tetrahedron in Cohn 2003, Figure (an observation attributed to Ian Quinn in Cohn
10. Cohn’s tetrahedron is not a complete fundamen- 2003, par. 11). With doublings allowed, several other
tal region for four-​voice OPTI-​space, as (like Straus’s OPTI-​classes have this property, including 0236 as
graph) it does not include multisets with doublings, noted above.
but it could be extended to a fundamental region dif- 70. Hint for Exercise 11.6.17: Start with an asym-
ferent from the one described here. metrical trichord and double one note at a time.
64. The units of distance in this discussion will be
called v-​units in Section 12.5. See Exercise 12.5.28 for
additional distance calculations in four-​voice T-​space.
SUGGESTED READING
65. Hint for Exercise 11.6.3: Do not use the four-​ Callender, Quinn, and Tymoczko 2008 remains the
dimensional coordinates shown in Figure 11.6.1; use reference most relevant to much of this chapter.
the geometry of an ordinary three-​dimensional cube. Section 11.4 is importantly influenced by Callender
Consider a triangle formed by the center of the cube 2004, which describes three-​voice T+​spaces and
and two appropriately chosen corners, and use the analyzes the voice leading in Saariaho’s Vers le blanc,
law of cosines: if a, b, and c are the three sides of a the subject of several exercises in this chapter. The
triangle and θ is the angle opposite side c, then c2 =​ a2 OP-​spaces described here are developed in some detail
+​ b2 − 2ab ∙ cosθ. in Tymoczko 2006. Cohn 2003 presents a not-​quite-​
66. Hint for Exercise 11.6.7: The OPT-​classes complete tetrahedron for four-​voice OPTI-​space (lack-
0258OPT and 0358OPT both lie in layer 2 of the tetra- ing the multiset classes with pitch-​class doublings);
hedron. One of the two paths lies entirely within layer the description of the four-​voice spaces here differs
2; the other hits a boundary at layer 21 2 . significantly from those of both Cohn and CQT. For
67. Hints for Exercise 11.6.8: There are only two other related items see the suggested reading for
different pictures to draw, as the front and left faces Chapter 10.

The Geometry of OPTIC Spaces • 481

 12
Distances

CONCEPTS OF distance capture important geometric structure as well as a topological one,

musical intuitions in many spaces, and the word but we have said little so far about exactly how
distance has appeared many times in the preced- distances are measured in multidimensional
ing chapters. Already in Chapter 1 we noted that spaces or how the distance measures we have
musical spaces can sometimes be distinguished formulated in various spaces are related.
from each other by different ways of measuring In this chapter we present a general formu-
distances. A graph-​ theoretic kind of distance lation of distance functions and distance spaces
was introduced in Chapter 3, offering a way to (known to mathematicians as metrics and metric
measure the distance between two objects if spaces). Distance spaces are in some ways sim-
those objects can be modeled by the vertices in a pler than interval spaces, and readers may find
graph. Specific applications appeared as we stud- them easier to grasp intuitively. For one thing,
ied distances in graphs of consonant triad space distances are always real numbers, not ordered
in Chapter 4. Interval functions, introduced in tuples or elements of some other abstract group,
Chapter 5, convey information something like and so it is clear when one distance is larger or
distances, but no sooner had we presented the smaller than another. We shall see that some
definition of interval functions than we began distance functions bear simple relationships to
to notice ways in which those functions do not interval functions while others do not.
correspond with common intuitions of distance. Section 12.1 reviews some of the shortcom-
Distances between points in OPTIC spaces were ings of interval functions as a way of describing
mentioned occasionally in Chapters 10 and 11, musical distances; many OPTIC spaces, as we
and in particular the concept of voice-​leading dis- shall see, cannot be interval spaces at all. This
tance was introduced in Section 10.2. The possi- first section is not essential for what follows,
bility of measuring distances gives these spaces a but it ties together several concepts from recent

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0012
chapters and provides some justification of the as well.1 The undirected pitch interval between
need for a different kind of formalism for mea- F5 and C4 is 17 regardless of the order in which
suring distance. Section 12.2 presents the gen- the two notes are listed, and the undirected
eral theory of distance functions and distance pitch-​ class interval (interval class) between
spaces, describes some categories of these spaces F and C is 5, the shortest distance separating
that relate most directly to interval spaces, and the two notes in the circle. These undirected
introduces the notion of isometry. Section 12.3 interval measurements cannot be modeled by
examines distance spaces derived from graphs or interval functions: there is no interval-​space
groups; such spaces arise in many musical set- structure on pitch in which int(F5, C4) and
tings but do not generally correspond to inter- int(C4, F5) are both 17, and no interval-​space
val spaces. A study of products and quotients structure on pc in which int(F, C) and int(C,
of distance spaces in Sections 12.4 and 12.5 F) are both 5. It is true that the undirected
shows how distance functions may be defined on intervals can be deduced from the directed
OPTIC spaces; we will see that a variety of differ- intervals—​an undirected interval may be con-
ent distance functions may be defined in some sidered an equivalence class of two directed
of these spaces. intervals—​so in that sense the interval spaces
pitch and pc contain the essential information
12.1 INTERVAL FUNCTIONS about intervals of both types. But to make even
AND MEASURES OF elementary observations about undirected
intervals, strictly speaking, requires stepping
DISTANCE out of the formalism of interval spaces.
A single theoretical construction cannot be These simple observations suggest that the
expected to solve every problem. The attention group-​ theory–​ based conception of interval
devoted to interval spaces in Chapters 5–​ 7, spaces is perhaps not sufficiently flexible to
along with work by Lewin, Rings, and others, account for everything that we might want to
should suffice to convince the reader that inter- call an “interval.” Other questions arise if we
val spaces possess a powerful and appealing wish to think of an interval as a measure of the
structure, valuable for an assortment of theo- distance separating two points of a space.2 To
retical and analytical study. But the restrictions regard an interval as a measure of distance would
imposed by that structure make interval spaces seem to require that intervals measure size in
less than completely satisfactory even as a way some concrete way, enabling us to deduce that
of modeling some things that are traditionally some intervals are smaller or larger than oth-
called “intervals,” and even more problematic as ers and therefore that some pairs of elements of
a way of modeling some other musically relevant the space are closer or more distant than others.
notions of distance. This section explores some Perception of interval size generally depends on
of these limitations in preparation for the intro- the structure of the interval group, which may
duction of the concept of distance functions in take a wide variety of forms; indeed, any group
Section 12.2. at all can theoretically serve as an interval group.
In an interval space, as we know from The group axioms make no mention of the “size”
Chapter 5, the interval from x to y is generally of group elements, so if the only thing we know
not the same as the interval from y to x but about an interval is that it is an element of a
rather its inverse in the interval group. In pitch group, that may or may not give us any sense of
space pitch, for example, int(F5, C4) is −17, the how “large” it is.
descending perfect eleventh, while int(C4, F5) is In the case of a real interval space, in which
+17, the additive inverse of −17. In pitch-​class the interval group is the additive group ℝ of real
space pc, int(F, C) is 7, measured clockwise in the numbers or some subgroup of ℝ, the notion of
pitch-​class circle, while int(C, F) is 5, the inverse interval size is relatively unproblematic: intervals
of 7 in the modular group ℤ12. The interval mea- far from 0 may be considered larger than intervals
sured by an interval function, in other words, is close to 0. In pitch, therefore, we may say that C4
always a directed (or ordered, or oriented) interval; is closer to A♭4 (interval 8) than to E♭3 (interval 9).
the particular kinds of intervals just mentioned (Notice that this argument tacitly makes use of
may be called directed pitch intervals and directed undirected rather than directed pitch intervals:
pitch-​class intervals. the size of the interval from x to y is taken to be
In pitch-​class set theory, undirected (unor- not int(x, y), which could be a negative number,
dered, unoriented) intervals are routinely invoked but rather the absolute value |int(x, y)|.) But we

Distances • 483
FIGURE 12.1.1 An impossible interval in three-​voice OPTI-​space

have seen many examples of interval spaces with =​8 in some interval space, the homogeneity
other groups. In Section 6.3 we noted that con- axiom from Section 5.4 then requires that every
tinuous pitch space is isomorphic (as an interval other trichord lie at interval 8 from something.
space) to frequency space freq, whose interval For example, there must be some trichord β such
group is multiplicative. In freq, an ascending that int(014, β) =​8—​but if intervals correspond
(equal-​tempered) minor sixth corresponds to an to voice-​leading distances, there can be no such
interval of approximately 1.59, while a descend- β, for every other trichord lies at a voice-​leading
ing major sixth is about 0.59; it is far from obvi- distance of 4 or less from 014. In pitch-​triple
ous, looking at those numbers, which interval is space cpitch3, there exist points at a voice-​lead-
larger. We have several times considered conso- ing distance of 8 from (0, 1, 4); for instance,
nant triad space triad as an interval space whose (−4, 1, 8) is such a point. But in OPTI-​space,
interval group is the dihedral group 𝒯ℐ12. The (−4, 1, 8) is represented by the point labeled
intervals from C major to two other triads in this with its OPTI normal form, which is 005—​and
space are int(C, A♭) =​ T8 and int(C, a) =​I4; there this point lies quite close to 014 in OPTI-​space
are no clear grounds for saying that one of these (at voice-​leading distance 2).
intervals is larger than the other. Homogeneity requires that all points of an
Various problems arise if we attempt to rec- interval space look alike as far as the interval
oncile the requirements of interval spaces with structure is concerned, but some OPTIC spaces
the ideas of voice-​leading distance that we began by their very nature are not homogeneous. As
to develop in the OPTIC spaces of Chapters 10 we know from Chapter 11, many of these spaces
and 11. Three-​ voice OPTI-​ space (the space of are orbifolds with boundary points that do not
three-​note multiset classes), shown originally in resemble other points in the spaces. Many of
Figure 11.4.12 and reproduced in Figure 12.1.1, them are also not as symmetrical as an interval
offers a plausible model of voice-​leading distances space would require. Interval spaces effectively
between trichords. (Exercise 3.3.3 explored display the symmetry of their interval groups:
related questions in the simpler setting of the the Cayley diagram for an interval group can be
trichordal set-​class space of Figure 3.3.1, of which relabeled with the elements of an interval space
the later figure is a refinement.) The trichords with that group. But three-​voice OPTI-​space does
most widely separated in this figure are 000 and not “look like a group”: it is hard to imagine what
048. The voice-​leading distance from 000 to 048 sort of group could have a Cayley diagram that
is eight semitones: in the coordinates shown in resembles the asymmetrical triangle in Figure
Figure 12.1.1, one path from 000 to 048 con- 12.1.1. In Section 12.2 we will present a defini-
sists of four semitones of motion in the positive tion of homogeneity for distance spaces; while
z direction and four in the negative x direction, some distance spaces are homogeneous, many
indicating that a triple unison can be transformed others are not.
into an augmented triad if two voices move by Figure 12.1.2 illustrates another problem that
four semitones in opposite directions.3 arises if intervals are conflated with distances.
No interval function can capture this concep- Part (a) shows, at ①, a tritone CG♭ resolving to a
tion of distance in OPTI-​space. If int(000, 048) major third D♭F, followed by a series of semitone

484 • E x p l o rin g M u sica l S paces

FIGURE 12.1.2 Tritone resolutions in two-​voice OP-​space

transpositions of this resolution. Part (b) then OP-​space in part (b) of the figure shows a way
plots these resolutions in two-​voice OP-​space, of understanding what is happening topologi-
the Möbius strip from Figure 11.2.6. Suppose for cally: as the dyads are transposed, the arrow
the moment that OP-​space is an interval space; shifts through the space, seemingly maintain-
then each of these resolutions defines a single ing its original orientation—​but by the time it
interval between a pair of dyads. Let i denote has circumnavigated the strip, that orienta-
int(CG♭, D♭F), the supposed interval between the tion has been reversed. Arrows ① and ⑦ begin at
dyads in voice leading ①. It stands to reason that the same point but aim in opposite directions,
voice leading ②, C♯G–​DF♯, should define the same a phenomenon not possible if both arrows are
interval i: after all, some dyad must lie at interval supposed to represent the same interval in an
i from C♯G, and surely the intervallic relationship interval space.
between the dyads C♯G and DF♯ is the same as Lest the reader conclude that none of the
that between CG♭ and D♭F. By this reasoning, all voice-​leading spaces studied in Chapters 10
the successive transpositions should define the and 11 are interval spaces, we should point out
same interval i between dyads. When the voice that in fact a few of them are. Pitch-​tuple space
leading has been transposed by T6 (voice leading cpitchn, the basis for the construction of all of
⑦), it becomes F♯C–​GB, so presumably int(F♯C, the OPTIC spaces, is itself an interval space with
GB) is i like the others. But int(F♯C, GB) cannot the additive interval group ℝn. In Section 7.4 we
be the same as int(CG♭, D♭F), because the dyads observed that octave and transpositional equiva-
F♯C and CG♭ are the same point in OP-​space, and lence give rise to quotient interval spaces. In one
homogeneity does not permit one point of an dimension, O-​space is simply the circle of contin-
interval space to lie at the same interval from uous pitch-​class space cpc, an interval space with
two other points. This contradiction shows that interval group ℝ12 (the real numbers mod 12).
no interval function on OP-​space can behave More generally, n-​voice O-​space cpitchn/​~O,
as we supposed. The Möbius-strip structure of also known as cpcn in Section 10.3, is an n-​torus:

Distances • 485
an interval space whose group is (ℝ12)n, the share the crucial feature that they are both groups
direct product of n copies of ℝ12. The interval of transpositions in the interval space cpitch3.
structure of T-​space is less familiar but was also By the usual equivalence of intervals and
described briefly in Chapter 7: cpitchn/​~T is an transpositions, GO and GT may be identified
(n − 1)-​dimensional space whose interval group with subgroups of the interval group ℝ3, GO′ =​
is the quotient group ℝn/​K, where K is the con- (12ℤ)3 and GT′ =​{(r, r, r) | r ∈ ℝ}. Pitch triples α
gruence subgroup consisting of all multiples of and β are octave-​equivalent if and only if int(α,
the n-​dimensional vector (1, 1, …, 1). As was β) ∈ GO′, and transpositionally equivalent if and
suggested in Exercise 7.4.1, OT-​space, obtained only if int(α, β) ∈ GT′. A practical consequence
by combining O and T equivalence, is an interval of this observation, explored in Exercise 12.1.3
space as well. below, is that one can tell whether two pitch tri-
But that is as far as we can go: permutational ples are O-​ or T-​equivalent from knowing noth-
equivalence, inversional equivalence, and cardinal- ing more than the interval between them. Also,
ity equivalence do not produce quotient interval because ℝ3 is a commutative group, GO′ and GT′
spaces, and any space incorporating one or more are automatically normal subgroups; in termi-
of these relations has no interval-​space structure nology from Section 7.4, this means that ~O and
that is compatible with voice-​leading distances. In ~T are normal equivalence relations on cpitch3
n voices for n ≥ 3, of the 32 spaces that may theo- and therefore give rise to quotient interval
retically be constructed from different combina- spaces. The interval-​space structure of O-​ and
tions of the OPTIC relations, only four are interval T-​space is thus seen to be a consequence of the
spaces: pitch-​tuple space itself (with no equiva- transpositional-​intervallic structure of the sym-
lences applied), O-​space, T-​space, and OT-​space. metry groups for these two relations.
The lack of compatibility with interval struc- The situation with permutational and inver-
ture should come as no surprise in the case of car- sional equivalence is very different. The sym-
dinality equivalence, which has already proven metry group for permutational equivalence GP
troublesome in many other ways. We know is the symmetric group 𝒮3 of permutations of
from Section 10.3, however, that P and I equiva- three objects; this group acts on pitch tuples
lence, like O and T equivalence, are defined by in cpitch3 by permuting the coordinates. The
symmetry groups, so one might expect that the symmetry group for inversional equivalence is
formation of quotient interval spaces should be GI =​{E, I}, whose only non-​identity transforma-
possible for these relations. Our work in Chapter tion is the inversion operator I. Neither GP nor GI
11 provides a glimmer of a reason why this is not is a group of transpositions in cpitch3; there can
the case: permutational and inversional equiva- be no groups GP′ and GI′ analogous to GO′ and
lence are precisely the relations that introduce GT′ above. (Because GP and GI are finite groups,
singularities when quotient spaces are formed, neither one is isomorphic to any subgroup of ℝ3,
in violation of the homogeneity required of which has no nontrivial finite subgroups.) We
interval spaces. But the symmetry groups come therefore cannot conclude that ~P and ~I give
into play in other ways, and it is worth taking a rise to quotient interval spaces. In fact they do
moment to consider them. not: Exercise 12.1.4 below shows that the equiv-
Using three-​voice spaces to illustrate, let us alence relations ~P and ~I, as well as cardinality
refer to the symmetry groups for ~O, ~T, ~P, and equivalence ~C, are not normal, and correspond-
~I in cpitch3 as GO, GT, GP, and GI respectively. GO ing quotient interval spaces therefore cannot be
consists of transpositions through three-​dimen- formed.
sional intervals all of whose components are mul-
tiples of 12: writing Trans(x, y, z) for transposition Exercise 12.1.3
through the three-​dimensional interval (x, y, z) (a) Suppose α and β are pitch triples for
in the interval space cpitch3, GO =​{Trans(x, y, z) | which int(α, β) =​(−7, −7, −7). Are α
x, y, z ∈ 12ℤ}. A pitch-​tuple transposition Tr cor- and β octave-​equivalent? Are they
responds in cpitch3 to a transposition through transpositionally equivalent? First answer
an interval of the form (r, r, r); the symmetry the questions in the abstract; then
group for transpositional equivalence is therefore produce examples of specific triples α and
GT =​{Trans(r, r, r) | r ∈ ℝ}. These groups GO and β in the given relationship and verify that
GT are different in structure—​the former is a dis- your answers are correct.
crete three-​dimensional group while the latter is (b) Answer the same questions if int(α, β) =​
a continuous one-​dimensional group—​but they (24, 0, −12).

486 • E x p l o rin g M u sica l S paces

(c) Answer the same questions if int(α, β) =​ 12.2 DISTANCE FUNCTIONS;
(−24, −24, −24). REAL AND MODULAR
(d) Answer the same questions if int(α, β) =​ INTERVAL SPACES AS
  (6, −6, 0).
DISTANCE SPACES
Exercise 12.1.4 Some of the musical spaces with which we are
(a) To show that permutational equivalence acquainted are distance spaces of particularly
cannot be deduced from intervals in simple kinds. In this section we define dis-
cpitch3, give examples of pitch triples tance functions and distance spaces, and show
α1, β1, α2, and β2 for which int(α1, β1) =​ that simple distance functions can always be
int(α2, β2) =​(6, −6, 0) but for which α1 and defined in two of the most important kinds of
β1 are permutationally equivalent while α2 interval spaces: real interval spaces and modu-
and β2 are not. lar interval spaces. At the end of this section we
(b) To show that inversional equivalence cannot discuss distance-​preserving mappings, known
be deduced from intervals in cpitch3, give as isometries. In the following sections we will
examples of pitch triples α1, β1, α2, and β2 describe other kinds of distance spaces, includ-
for which int(α1, β1) =​int(α2, β2) =​(−24, −24, ing many that do not correspond to interval
−24) but for which α1 and β1 are inversionally spaces at all.
equivalent while α2 and β2 are not. A word on terminology is needed. In the
(c) To show that cardinality equivalence cannot mathematical community, the function that we
be deduced from intervals in cpitch3, give are calling a distance function is known as a met-
examples of pitch triples α1, β1, α2, and β2 for ric, and what we are calling a distance space is
which int(α1, β1) =​int(α2, β2) =​(0, 2, 0) but a metric space. These terms are well established,
for which α1 and β1 are cardinality-​equivalent dating to the early years of the twentieth cen-
while α2 and β2 are not. tury, but employing them here seems inadvis-
(d) Explain why it follows from (a), (b), and (c) able because of possible associations of the
that the equivalence relations ~P, ~I, and ~C word metric with musical meter rather than dis-
on cpitch3 are not normal and therefore do tance—​and particularly because the literature
  not give rise to quotient interval spaces.4 on rhythm and meter has often employed the
term metric space in reference to meter.6 In an
One way of generalizing the notion of inter- effort to avoid this confusion, we will substitute
val is suggested by the observation that an the customized terms distance function and dis-
interval defines an equivalence class of ordered tance space in our definitions and in most other
pairs.5 If (S, G, int) is an interval space, an equiv- circumstances (though we may occasionally refer
alence relation ~ may be defined on S × S so that to a metric when no confusion seems likely).
(x1, y1) ~ (x2, y2) whenever int(x1, y1) =​int(x2, y2). The definition of distance spaces, like the
The structural requirements of interval spaces definition of interval spaces in Chapter 5, is
limit the sorts of equivalence relations that arise axiomatic. Formally, a distance space (S, dist)
in this way; it seems plausible that other equiva- consists of a set S (often referred to as the
lence relations on S × S, not defined by interval space itself) and a function dist: S × S → ℝ≥0,
spaces, may offer more flexible conceptions of called the distance function of the space. The
interval (or distance). We may imagine such con- notation dist: S × S → ℝ≥0 means that the dis-
structions as a way of addressing the problems tance function maps ordered pairs of elements
brought to light by Figures 12.1.1 and 12.1.2: for of the space to nonnegative real numbers; the
an appropriate equivalence relation ~ on three-​ number dist(x, y) may be called “the distance
voice OPTI-​space, perhaps there is no trichord from x to y.” The following three distance-​space
β such that (000, 048) ~ (014, β), while in two-​ axioms must be satisfied:
voice OP-​space, perhaps the relation (CG♭, D♭F) ~
(F♯C, GB) could be possible after all. Equivalence • Positivity. For every element x in the space S,
classes of ordered pairs may of course be defined dist(x, x) =​0. For all x and y in S with x ≠ y,
in many ways; the distance functions introduced dist(x, y) > 0.
in the next section give rise to their own concep- • Symmetry. For all x and y in S, dist(y, x) =​
tion of equivalence classes—​one that behaves in dist(x, y).
the ways just described and is compatible with • Triangle inequality. For all x, y, and z in S,
our understanding of voice-​leading distance. dist(x, z) ≤ dist(x, y) +​dist(y, z).

Distances • 487
 While the notation for distance functions is we may speak of the distance “between x and y”
similar to that for interval functions, some differ- rather than specifying “from x to y”; the specific-
ences are apparent. The first difference involves ity of the latter is always necessary with interval
the values taken by the function: interval func- functions.
tions can map into any group, but distances are Parts (c) and (d) of Figure 12.2.1 contrast
always nonnegative real numbers. The relation the triangle inequality with the interval-​ sum
symbols > and ≤ appearing in the axioms make equation, one of the interval-​space axioms from
sense as long as the things surrounding them are Section 5.4. The symbol +​appearing in the trian-
real numbers; these relations are not defined for gle inequality always indicates ordinary addition
general group elements. In practical situations of real numbers, in contrast with the operation ⊕
distances are measured in units of some sort: in the interval-​sum equation, which may denote
perhaps miles, feet, or millimeters for distances some other group operation. Intervals add like
in physical space, or octaves, semitones, or cents vectors, but distances do not: the distance from
for distances in pitch space. In Section 12.3 it will x to z cannot be greater than the sum of the dis-
be convenient to consider the distances between tances from x to y and from y to z, but it may
some pairs of points to be infinite; in such situa- well be less than that sum. The inequality takes
tions we may take the distance function to map its name from a familiar property of distances in
into the set ℝ≥0 ∪ {∞}. the everyday world, where the length of one side
The positivity axiom says that while the dis- of a triangle is always less than the sum of the
tance from any point to itself is 0, the distance lengths of the other two sides: you will put fewer
between two different points is always a positive miles on your car driving directly from Chicago
number. The symmetry axiom says that the dis- to Pittsburgh than if you drive from Chicago to
tance between two elements does not depend on Nashville and from Nashville to Pittsburgh.
the order in which the two are listed. Distances Pitch space, in either its discrete form pitch
are therefore fundamentally undirected, in con- or its continuous form cpitch, is a distance space
trast with the directed nature of interval func- with the distance function defined by dist(x, y) =​
tions noted in Section 12.1. Parts (a) and (b) of |y − x| for all pitches x and y. The notation |x|,
Figure 12.2.1 contrast the symmetry axiom with as usual, denotes the absolute value of the num-
the inverse interval property of interval func- ber x. If pitches are numbered in the familiar way
tions from Section 5.4. Because of symmetry, from Figure 1.1.1, then this distance function
measures distances in semitones. This distance
function may be called distpitch or distcpitch in
situations where the space may be in question or
confusion with some other distance function is
a concern. Some examples of this distance func-
tion are given here:

dist(F4, F4) =​|5 − 5| =​|0| =​0;

dist(F4, A♭5) =​|20 − 5| =​|15| =​15;
dist(A♭5, F4) =​|5 − 20| =​|−15| =​15;
dist(A♭5, C5) =​|12 − 20| =​|−8| =​8;
dist(F4, C5) =​|12 − 5| =​|7| =​7.

The positivity axiom is illustrated by the fact

that the first distance in the list, between the
note F4 and itself, is zero, while all the distances
between two different notes are positive. The
second and third examples, the distances from
F4 to A♭5 and vice versa, illustrate the symmetry
FIGURE 12.2.1 Comparison of interval func- axiom; the numbers y − x and x − y are always
tions and distance functions: (a) the inverse negatives of each other, so they have the same
interval property for interval functions; (b) the absolute value. The triangle inequality is illus-
symmetry axiom for distance functions; (c) the trated by the second, fourth, and fifth exam-
interval-​sum equation for interval functions; (d) ples: dist(F4, C5) is 7, much less than the sum
the triangle inequality for distance functions dist(F4, A♭5) +​dist(A♭5, C5) =​15 +​8 =​23.

488 • E x p l o rin g M u sica l S paces

 For this distance function, the triangle pitch interval. Distance functions thus allow us
inequality takes the form |z − x| ≤ |y − x| +​|z − y|. to formalize some such types of “intervals” that
To see why this inequality must always be true, are not accommodated by the theory of interval
write a for y − x and b for z − y; then a +​ b =​ z − x, functions. (Undirected pitch-​class intervals will
so the inequality asserts that |a +​ b| ≤ |a| +​|b|. be considered shortly.) With the exception of the
In fact, the latter statement (which is itself also value 0, every possible value of the (undirected)
sometimes called the triangle inequality) is true distance function on cpitch corresponds to two
for all real numbers a and b. If a and b are of the possible values of the (directed) interval func-
same sign (both positive or both negative), or if tion: for example, a distance of 8 may indicate an
either of them is zero, then |a +​ b| =​|a| +​|b|, interval of either 8 or −8.
while if the two numbers are of opposite sign, In fact, every real interval space is also a dis-
then |a +​ b| is always strictly less than |a| +​ tance space, with distances corresponding to
|b|. What this means in pitch space is that if the undirected intervals in exactly this way. Loosely
intervals from x to y and y to z are both ascending speaking, if a space looks like a number line,
or both descending—​that is, if y lies between x we can measure distances along that line. To be
and z—​then the distances dist(x, y) and dist(y, z) precise, suppose (S, G, int) is an interval space
sum to dist(x, z), but if one interval is ascend- whose interval group G is the additive group
ing and the other descending (as is the case in (ℝ, +​) or some subgroup of (ℝ, +​) (such as the
the above example involving F4, A♭5, and C5), then integers (ℤ, +​)). If we define dist(x, y) =​|int(x,
dist(x, z) is smaller than that sum. y)| for all x and y in S, then dist is a distance
function, and (S, dist) is a distance space.
Exercise 12.2.2 Below are five pitch triples x, y,
z. The first two triples repeat the example above Exercise 12.2.4 Use the properties of the
but with the notes in different orders. For each interval function int in the real interval
triple, calculate dist(x, y), dist(y, z), and dist(x, z) space (S, G, int) to explain why the distance
in the space pitch, and verify that these function dist defined by dist(x, y) =​|int(x, y)|
distances satisfy the triangle inequality. satisfies the three axioms required of distance
functions.
   8

(a) F4, C5, A♭5

(b) A♭5, F4, C5 The equation dist(x, y) =​|int(x, y)| ensures
(c) F4, A♭5, E6 that an interval function and distance function
(d) F4, E♭4, C4 related in this way are metrically compatible in the
(e)  E♭4, C4, F4 sense that equal intervals imply equal distances:
whenever x, y, x′, and y′ are elements of the space
Exercise 12.2.3 such that int(x, y) =​int(x′, y′), then dist(x, y) =​
(a) If you know that x, y, and z are three notes dist(x′, y′) is true as well.9 The converse does not
in cpitch such that dist(x, y) =​15 and hold: in pitch space, as we have seen, a single dis-
dist(y, z) =​8, what are the possibilities for tance may correspond to two different intervals.
dist(x, z)? The musical spaces we have encountered by
(b) If x, y, and z are three points in any distance now include several different real interval spaces,
space (not necessarily in cpitch) with so we therefore have an assortment of distance
distances as in (a), what can you conclude spaces at our disposal as well. Generic pitch space,
  about dist(x, z)?7 for example, is an interval space with interval
group ℤ, in which int(C4, C5) =​7 and int(C5, C4)
We introduced distance functions as an alter- =​−7; it is also a distance space in which dist(C4,
native to interval functions, but the word inter- C5) =​dist(C5, C4) =​7. Other examples of real
val has already crept into our discussion. Pitch interval spaces with a corresponding distance-​
space is both an interval space and a distance space structure include diatonic pitch spaces
space, and the interval function int and the dis- dpitch(n), other scalar spaces such as whole-​
tance function dist are closely related: dist(x, y) tone and octatonic pitch spaces, the (continuous)
is the same as |int(x, y)|, the absolute value of time domain time, (discrete) beat space beat,
int(x, y). In terminology discussed in Section and spelled pitch-​class space spc (the line of
12.1, int(x, y) is the directed (or ordered or ori- fifths). In spc, dist(G♭, G♯) =​14, because fourteen
ented) pitch interval from x to y, while dist(x, y) fifths separate G♭ and G♯. In discrete spaces with
is the undirected (or unordered or unoriented) the interval group ℤ, all distances are integers,

Distances • 489
but in continuous spaces such as cpitch with We turn our attention next to pitch-​ class
group ℝ, non-​integer distances are possible. space, either discrete (pc) or continuous (cpc),
where we wish to model the undirected pitch-​
Exercise 12.2.5 Calculate each of the distances class interval between two pcs (the traditional
below in the space indicated. “interval class”) by means of a distance func-
tion. Intuitively it seems plausible to say that,
(a) dist(A3, E♭5) in pitch for example, the distance between the pitch
(b) dist(A3, E♭5) in dpitch(−2) (two-​flat diatonic classes E♭ and G is 4 because four semitones sep-
pitch space) arate them in the pitch-​class circle. Of course,
(c) dist(A3, D♯5) in dpitch(+​4) as shown in Figure 12.2.6, we could travel eight
(d) dist(A3, E♭5) in wt1 (the whole-​tone pitch semitones from E♭ to G (or vice versa) by taking
space containing the notes in question) the long way around the circle, or even farther
(e) dist(A3, E♭5) in oct01 by circumnavigating the space more than once—​
(f) dist(A3, E♭5) in oct23 but if we want to define a single number as the
(g) dist(A, E♭) in spc distance, we should choose 4, the length of the
(h)  dist(A, D♯) in spc shortest possible route. This idea of distance as
the length of the shortest possible path joining
In Section 5.4 we noted that a subset of an two points will be revisited in several other con-
interval space is not generally an interval space texts later in this chapter, notably including the
in its own right, for instance because the inter- context of voice-​leading distances.
vals formed by elements of the subset may fail The mathematical description of the distance
to form a group. For distance spaces there are no function in pitch-​class space is not quite as simple
such restrictions: every subset of a distance space as the formula distpitch(x, y) =​|y − x| in pitch
is itself a distance space. The distance function on space.10 A special notation will be useful. If x is
the subspace is inherited from the larger space; either a real number or a real number mod 12, we
that is, any distance in the subspace is the same write |x|12 for the smallest real number r ≥ 0 such
as the distance between the same elements in the that either r ≡ x (mod 12) or r ≡ −x (mod 12). For
larger space. The triad {C4, E4, G4} in pitch space example, |4|12, |8|12, |−4|12, |16|12, and |20|12
does not form an interval space—​in Chapter 5 we are all equal to 4. We may think of |x|12 as the
introduced the term interval configuration for this “mod-​12 absolute value” of x, although |x|12 is
reason—​but this triad does form a distance space. technically a nonnegative real number, not a num-
The only distances occurring in this small space are ber mod 12. Intuitively, |x|12 is the length of the
0, 3, 4, and 7; these numbers do not form a group shortest path from 0 to x in the mod-​12 circle, just
or any other sort of algebraic structure, nor are as the usual absolute value |x| is the length of the
they required to. The same triad is also a subset of shortest path from 0 to x in the number line.
the distance space oct01, from which it inherits a A similar definition of |x|n is possible for any
different distance function, one for which the only modulus n; for example, |3|4 =​1 and |11|7 =​3.
distances occurring in the triad are 0, 2, 3, and 5. The following properties of |x|n are direct conse-
quences of the definition:

• The inequalities 0 ≤ |x|n ≤ n/​2 always hold; in

particular, 0 ≤ |x|12 ≤ 6.
• |x|n =​0 if and only if x ≡ 0 (mod n).
• |x|n =​|y|n if and only if either x ≡ y (mod n)
or x ≡ −y (mod n); in particular, |−x|n =​|x|n.
• If x is a real number with |x| ≤ n/​2, then
|x|n =​|x|, whereas if |x| > n/​2, then |x|n
< |x|. The inequality |x|n ≤ |x| therefore
always holds: the mod-​n absolute value of
a number can never be larger than its usual
absolute value.
• If x is an integer, then |x|n is also an integer.
The definition of |x|n, however, makes sense
FIGURE 12.2.6 Three paths from E♭ to G in also for non-​integers (numbers in ℝ or ℝn);
pitch-​class space for example, |8.3|12 =​3.7.

490 • E x p l o rin g M u sica l S paces

 With this definition in hand, the distance general, let |a|12 =​ r and |b|12 =​ s. Then r and s are
function in pitch-​class space (distpc or distcpc) is nonnegative numbers such that a ≡ ±r (mod 12)
given by the formula dist(x, y) =​|y − x|12, where (that is, either a ≡ r or a ≡ −r) and b ≡ ±s (mod 12).
x and y are mod-​12 pitch-​class numbers. Here It follows that a +​ b is congruent mod 12 to
are a few examples of this distance function, one of the four numbers of the form (±r) +​(±s),
which the reader should visualize in the pitch-​ and therefore that |a +​ b|12 =​|(±r) +​(±s)|12.
class circle: Using the property |x|12 ≤ |x| and the triangle
inequality in the form |a +​ b| ≤ |a| +​|b| for the
dist(F, F) =​|5 − 5|12 =​|0|12 =​0; usual absolute value (both of which were noted
dist(F, A♭) =​|8 − 5|12 =​|3|12 =​3; previously), it follows that
dist(A♭, F) =​|5 − 8|12 =​|9|12 =​3;
dist(A♭, C) =​|0 − 8|12 =​|4|12 =​4;
dist(F, C) =​|0 − 5|12 =​|7|12 =​5. |a + b|12 =|( ± r ) + ( ± s )|12 ≤ |( ± r ) + ( ± s )|
≤ |± r|+|± s| = r + s =|a|12 +|b|12 .
In the above examples x and y are numbers
mod 12, and the calculation of y − x is performed
mod 12. Treating x and y as real numbers, how- Exercise 12.2.7 Below are five pitch-​class
ever, would yield the same results: for instance, triples x, y, z. For each triple, calculate
we could calculate dist(A♭, F) =​|5 − 8|12 =​|−3|12 dist(x, y), dist(y, z), and dist(x, z) in the space
=​3, the same answer as above. As in pitch space, pc, and verify that these distances satisfy the
the equation for the distance function may be triangle inequality.
expressed in terms of the interval function:
dist(x, y) =​|int(x, y)|12. Here int denotes the (a) F, C, A♭
usual directed interval function on pc or cpc, (b) A♭, F, C
and dist gives the undirected pitch-​class inter- (c) F, A♭, E
val—​an interval class in the usual definition of (d) F, E♭, C
the term. (An interval class is a distance, not (e)  E♭, C, F
an interval.) The number 0, though not usu-
ally recognized as an interval class, is a possible Exercise 12.2.8 If you know that x, y, and z
value of this distance function, and non-​integer are three notes in cpc such that dist(x, y) =​3
values are also possible in cpc. As with the dis- and dist(y, z) =​4, what are the possibilities for
tance function in pitch space, one value of this dist(x,
   z)?
distance function typically corresponds to two
values of the interval function—​a pitch-​class The distance function on pitch-​ class space
distance of 4 may correspond to a directed pitch-​ suggests a way to define an undirected distance
class interval of either 4 or 8. But in pitch-​class function on any modular interval space. Suppose
space there are two exceptions to this generaliza- (S, G, int) is an interval space whose interval
tion, not just one as in pitch space: distances 0 group G is a modular group (ℝn, +​) or any sub-
and 6 each correspond to a single interval. group of (ℝn, +​) such as (ℤn, +​). Because int(x, y)
The first three examples in the above list illus- is a number mod n for all x and y in S, the
trate the positivity and symmetry properties notation |int(x, y)|n makes sense, so we may
of the distance function, which are immediate define a distance function on S by the equation
consequences of the definition of the mod-​12 dist(x, y) =​|int(x, y)|n. The positivity and sym-
absolute value. The general validity of the tri- metry axioms follow easily from the properties
angle inequality for this function is perhaps not of the interval function and of the mod-​n abso-
so obvious. Before proving it, we note that the lute value function. The triangle inequality may
second, fourth, and fifth items in the above list be proved exactly as in pitch-​class space above,
provide an example: dist(F, C) is 5, less than the writing a and b for int(x, y) and int(y, z). As was
sum dist(F, A♭) +​dist(A♭, C) =​3 +​4 =​7. the case with real interval spaces, this definition
In pitch space, by writing a =​ y − x and of the distance function ensures that the interval
b =​ z − y, we showed that the triangle inequality function and the distance function are metrically
was equivalent to the statement |a +​ b| ≤ |a| +​ compatible: equal intervals imply equal distances.
|b|. In similar fashion, the triangle inequality in Besides pc and cpc, familiar modular interval
pitch-​class space reduces to |a +​ b|12 ≤ |a|12 +​ spaces in our repertoire include generic and dia-
|b|12. To see why this inequality must be true in tonic pitch-​class spaces gpc and dpc(n). These

Distances • 491
are discrete mod-​7 spaces with interval group ℤ7, and inversions are isometries in every real or
so the distance function dist(x, y) =​|int(x, y)|7 modular interval space. Because these spaces are
takes only the values 0 (for unisons/​octaves), commutative, we know from Sections 7.5 and
1 (generic or diatonic seconds/​ sevenths), 2 7.6 that every transposition is interval-​preserv-
(thirds/​sixths), and 3 (fourths/​fifths). In gpc, ing and every inversion is interval-​reversing. The
for instance, dist(F, B) =​dist(F, C) =​3. Other distance-​preserving property for transpositions
modular spaces of various cardinalities include then follows from the metric compatibility that
scalar spaces such as whole-​tone and octatonic we have noted (equal intervals imply equal dis-
pitch-​class spaces. tances). For inversions, the property |−x| =​|x|
We have seen that the elements of some (or |−x|12 =​|x|12) ensures that opposite inter-
modular spaces may be rearranged to produce vals imply equal distances as well. The following
different interval spaces with similar modu- exercise provides illustrations.
lar structure, so those spaces also are distance
spaces. The most familiar example is mod-​12 Exercise 12.2.10
fifths space fifth (the circle of fifths), a rear-
rangement of pc. In fifth, intfifth(C, C♯) =​7 and (a) In pitch, verify that dist(E4, C♯5) =​dist(T−5(E4),
therefore distfifth(C, C♯) =​|7|12 =​5: the distance T−5(C♯5)).
from C to C♯ around the circle of fifths is 7 in the (b) In pitch, verify that dist(E4, C♯5) =​dist(I(E4),
clockwise direction, but 5 if we are allowed to I(C♯5)), where I (=​ I0) is inversion about C4.
choose the shortest path. The spaces of generic (c) In pc, verify that dist(A, G) =​dist(T6(A),
fifths and thirds, gfifth and gthird (Figures T6(G)).
1.3.8 and 1.3.12), along with their diatonic coun- (d) In pc, verify that dist(A, G) =​dist(I9(A),
terparts, are examples of mod-​7 interval spaces I9(G)).
on which distance structures may be defined (e) In gpc, verify that dist(C, F) =​dist(t3(C),
similarly. t3(F)).
(f) In spc, verify that dist(B♭ ♭, G♭) =​
Exercise 12.2.9 Calculate both the interval and dist(τ15(B♭ ♭), τ15(G♭)), where (as in Sections
the distance from C to B in each of the indicated 2.6 and 7.5) τk transposes a spelled pitch
spaces.11   class by k positions along the line of fifths.

(a) int(C, B) and dist(C, B) in pc Exercise 12.2.11 Use the distance-​space

(b) int(C, B) and dist(C, B) in fifth axioms and the definition of isometry to
(c) int(C, B) and dist(C, B) in gpc explain why every isometry of distance spaces is
(d) int(C, B) and dist(C, B) in gfifth a one-​
   to-​one mapping.
(e)  int(C, B) and dist(C, B) in gthird
Exercise 12.2.12
As interval spaces, the mod-​ 12 spaces pc (a) Is it true that distfifth(C, B) =​
and fifth are isomorphic, while all three of distfifth(T3(C), T3(B))? Are the usual
the mod-​7 spaces gpc, gfifth, and gthird are transposition operators Ti isometries in
isomorphic. Mathematicians usually do not use the distance space fifth?
the term isomorphic for metric spaces (distance (b) Let QE: spc → fifth be the quotient map
spaces), preferring the term isometric. An isom- of the enharmonic equivalence relation;
etry is a distance-​preserving mapping of distance that is, QE maps every spelled pitch class
spaces, defined analogously to the interval-​pre- (for example, E♯) to the corresponding pitch
serving mappings of interval spaces studied in class (5) in the circle of fifths. Is QE
Chapter 7. That is, if (S1, dist1) and (S2, dist2) are   an isometry?12
two distance spaces, a function f: S1 → S2 is an
isometry if dist2(f(x), f(y)) =​dist1(x, y) for all x We observed in Section 7.2 that the interval
and y in S1. Every isometry is a one-​to-​one func- spaces pc and fifth are isomorphic via the map-
tion (see Exercise 12.2.11 below); if some isom- ping f(C) =​C, f(C♯) =​G, …, which associates every
etry maps S1 onto S2, then the distance spaces S1 pitch class in the chromatic pitch-​class circle
and S2 are said to be isometric. with the pitch class in the corresponding loca-
Familiar examples of isometries include the tion in the circle of fifths. This mapping is also
usual transposition and inversion operators on an isometry of distance spaces. The analogous
pitch and pitch-​class space. In fact, transpositions mappings among the mod-​7 spaces gpc, gfifth,

492 • E x p l o rin g M u sica l S paces

and gthird are likewise interval-​space isomor- some group isomorphism g: ℤ → 2ℤ with
phisms as well as isometries. which f is interval-​compatible.)
In Section 12.1 we noted that interval spaces (b) Explain why the spaces wt0 and wt0′ are
are required to be homogeneous in ways that dis- not isometric, and in particular why the
tance spaces are not. The concept of isometry   function f in part (a) is not an isometry.
enables us to give a precise definition of homo-
geneity for distance spaces. A distance space (S, A mapping such as f in the above exercise,
dist) is said to be homogeneous if any point of which multiplies all distances by a constant factor,
the space can be mapped isometrically to any is called a dilation. Spaces related by dilation, such
other—​that is, if for every x and y in S, there as wt0 and wt0′, have distance structures that
exists an isometry f: S → S such that f(x) =​ y. are completely proportional and therefore may
The corresponding property for interval spaces be said to convey equivalent information about
always holds, but while some distance spaces distances, even though they are not isometric.13
are homogeneous, others are not. All the dis-
tance spaces considered so far in this section are 12.3 DISTANCE FUNCTIONS
homogeneous. If S =​(S, G, int) is a real or modu- DEFINED BY GRAPHS
lar interval space such as pitch, cpitch, pc, or
cpc, then S is also a distance space as described
OR GROUPS
in this section, and for any x and y in S, the trans- In our study of graphs in Section 3.1, we defined
position Ti, where i =​int(x, y), is an isometry that the distance between two vertices in a graph to be
maps x to y. The discussion from Section 12.1 the length of the shortest path joining them, the
suggests that many OPTIC spaces are not homo- length of a path being the number of edges of the
geneous, as we will verify in Section 12.5. graph in that path. This graph-​theoretic notion of
It is reasonable to regard an isometry as an distance is a valid distance function in the sense
“isomorphism of distance spaces,” but in some defined in this chapter, and therefore every graph
cases things may not be as straightforward as may be regarded as a distance space. More pre-
they appear. Recall from Section 7.2 that an cisely, if A is any undirected graph, a distance func-
interval-​space isomorphism is not required to be tion distA may be defined on the vertices of A—​or
interval-​preserving; it is therefore possible that on any set used to label those vertices as long as
two interval spaces may be isomorphic while the the label function is a bijection. (That is, we may
associated distance spaces are not isometric. The use a graph A to define distA(x, y) as long as there
following exercise offers an illustration. is exactly one vertex in A labeled x and exactly one
vertex labeled y.) In a disconnected graph, pairs of
Exercise 12.2.13 Consider the whole-​tone vertices in different components are joined by no
pitch space wt0 =​{. . ., A♭3, B♭3, C4, D4, E4, …}. paths at all; if the graph A provides no route from
In Section 7.2 we described wt0 as an interval x to y, distA(x, y) is defined to be ∞.
space with interval group ℤ, isomorphic to It should be clear that distA satisfies the
pitch, in which intervals are measured by positivity and symmetry axioms for distance
counting scale-​steps—​in this case whole steps, functions. Symmetry depends on the ability to
so that, for example, int(C4, D4) =​1. But wt0 is traverse a path in either direction, as in an undi-
also an interval space with the interval function rected graph. In a directed graph, requiring that
it inherits from pitch, in which intervals are paths be traversed in the direction of the arrows
measured in semitones, so that int(C4, D4) =​2. may lead to a notion of “distance” that violates
In this space only even intervals are present, and the symmetry axiom and therefore is not a
the interval group is 2ℤ. Let us write wt0 for the proper distance function as defined here. (Such
space with group ℤ and wt0′ for the space with functions are sometimes useful—​in a city with
group 2ℤ. Because they are real interval spaces, one-​way streets, distances can be asymmetric—​
wt0 and wt0′ are therefore distance spaces as but we will not consider them here.)
well, with different distance functions: dist(C4, To see why the triangle inequality always
D4) =​1 in wt0 while dist′(C4, D4) =​2 in wt0′. holds for distA, suppose distA(x, y) =​ r and
distA(y, z) =​ s. This means that r is the length of
(a) Consider the mapping f: wt0 → wt0′ that the shortest path from x to y and s is the length
maps every pitch in wt0 to itself. Explain of the shortest path from y to z. By combining
why f is an interval-​space isomorphism. (As these two paths we may obtain a path from x to
in Section 7.2, you must show that there is z whose total length is r +​ s. This combined path

Distances • 493
may or may not be the shortest path from x to z, The idea that there may be different ways of
but the fact that such a path exists at all shows measuring distance in the same space should be
that distA(x, z) ≤ r +​ s =​distA(x, y) +​distA(y, z). If unsurprising. The distance traveled by a vehicle
either distA(x, y) or distA(y, z) is ∞, the inequality constrained to follow roads is longer than the
distA(x, z) ≤ distA(x, y) +​distA(y, z) is true regard- distance flown by the idiomatic crow. We rarely
less of what distA(x, z) is.14 consider that in three-​ dimensional Euclidean
In Sections 4.1 and 4.2 we defined three space, the shortest distance between widely sep-
graph-​theoretic distance functions on consonant arated cities is the length of a straight line that
triad space triad: the circle-​of-​fifths distance, burrows beneath the earth’s surface, a shorter
Weber distance, and neo-​Riemannian distance distance than the crow’s if not a realistic travel
(PLR distance). Table 4.2.16 and Exercise 4.2.17 route. Triad space has many possible pictures,
compared these distance functions and explored not limited to the graphs in Figure 12.3.1. In
some of their properties. The graphs from which particular, double-​circle spaces come in many
these distance functions are derived are repro- possible configurations. The arrangement in
duced in Figure 12.3.1, with some modifications 12.3.1a, unchanged from Figure 4.1.4b, is based
as noted below. on T7 (which generates the circles of major and
minor triads) and the relative transformation R
(dictating the radial juxtaposition of R-​related
triads). Diagonal edges such as the one joining C
major to E minor ensure that all keys tradition-
ally considered to be closely related are separated
by a distance of 1.
For other double-​ circle alternatives, the
reader may refer back to the two graphs of
Figure 4.1.1, each of which defines a differ-
ent distance function and therefore a different
distance space. Both of these graphs arrange
the triads by semitone rather than by fifth and
eliminate the diagonal edges. To understand the
notion of distance determined by each graph,
one may think of the graph as a kind of game
board on which the edges define the admis-
sible moves. The arrowheads appearing on a
few edges in Figure 4.1.1 may be disregarded, as
each edge may be traversed in either direction.
In 4.1.1a the admissible moves are semitone
transpositions (up or down) and mode changes
(P); thus the distance from E♭ major to C minor
is 4, because a minimum of three semitones of
transposition and a mode change are required
to get from one to the other. In 4.1.1b, P is
replaced with the fixed inversion operator I as
an admissible move, and the distance between
the same two triads is only 3. If this distance
function seems less intuitive than the others we
have considered, the following exercise suggests
one possible reason.

Exercise 12.3.2
(a) In the distance space defined by the graph
in Figure 4.1.1b, give an example to show
FIGURE 12.3.1 Graphs defining triadic distance that transposition is not an isometry. That
functions: (a) circle-​of-​fifths distance (from Figure is, show that the distance between two
4.1.4b); (b) Weber distance (Figure 4.2.3); (c) neo-​ triads can change when both are transposed
Riemannian distance (Figures 4.2.5b, 4.2.8) by the same interval.

494 • E x p l o rin g M u sica l S paces

(b) Is the inversion operator I an isometry in More generally, rather than assuming that all
  this space? edges of a graph have length 1, we may assign
each edge its own length independent of all the
In Weber space or neo-​ Riemannian triad others. A distance function may be defined as
space, the relative transformation R is an admis- above, using the length of the shortest path. The
sible move and the distance between the relative only requirement, needed to ensure the positiv-
triads E♭ major and C minor is therefore only 1. ity axiom, is that all edge lengths be positive real
As described in Chapter 4, both of these spaces numbers.15
are toroidal; parts (b) and (c) of Figure 12.3.1
show only a single tile of each. In comparison Exercise 12.3.3
with the original picture of weber in Figure (a) Calculate distNRVL(C, f♯) and distNRVL(C, g♯),
4.2.3, Figure 12.3.1b incorporates a shearing and compare them with distNR(C, f♯) and
transformation to convert the parallelogram tile distNR(C, g♯).
into a rectangle. The graph’s formerly vertical (b) In Exercise 4.2.17b you identified B♭
edges are now slanted, but the continuations of minor as the triad at the greatest possible
these edges across the boundary are traced more neo-​Riemannian distance (5) from C
easily when the boundary is rectangular. Three major. What is distNRVL(C, b♭)? Is this the
kinds of moves are admissible in either weber largest neo-​Riemannian voice-​leading
or nrtriad. P and R are available in both spaces distance from C major to any other triad?
(these two transformations alternate in each (c) The smoothest voice leading from a
row of Weber space); weber also allows transpo- C-​major triad to an F-​minor triad requires
sition up or down by perfect fifth, while nrtriad only that two voices move by semitone.
allows moves by the leittonwechsel L. What is distNRVL(C, f)? Considering that
The graph of nrtriad and the associated neo-​ the definition of distNRVL was motivated
Riemannian distance function suggest still other by voice-​leading distances, why is this
metric structures that may be defined on the set   distance not equal to 2?
triad. One of these involves a simple refine-
ment of the graph-​theoretic definition of dis- The last part of the exercise above, which
tance. In the standard interpretation of Figure recalls a discussion from Section 10.2 about
12.3.1c, a move by P, L, or R traverses a single shortcomings of the tonnetz as a voice-​lead-
edge of the graph, a path of length 1. Musically, ing space, suggests that distNRVL is not entirely
however, it could be argued that R traverses a adequate as a measure of voice-​leading distance.
longer distance than P or L, because R entails a More satisfactory voice-​leading distances will be
two-​semitone displacement in one of a triad’s examined in the next section. In the meantime,
voices rather than a single semitone. The figure the following exercise suggests something closer
calls attention to this difference by depicting to the mark.
the R edges as dotted lines. If we consider the
dotted edges to be of length 2 rather than 1, a Exercise 12.3.4 Let S be the set consisting of
new distance function emerges, which we call the 24 major and minor triads and the four
the neo-​Riemannian voice-​leading distance. (Each augmented triads. Let distCube be the distance
R edge in Figure 12.3.1c could be converted to a function on S defined by the Cube Dance
path of length 2 by adding a new vertex within it; graph from Figure 10.2.11. Either 10.2.11a
the rather different picture of neo-​Riemannian or 10.2.11b will work, as will Figure 8.2.19
triad space from Figure 10.2.15 may be helpful (including the dashed edges); all share the same
as another way of visualizing this distance func- graph. Since triad is a subset of S, distCube also
tion.) The distance between two triads remains defines a distance function on triad.
the length of the shortest path joining them; the
length of a path, however, is no longer simply the (a) Determine distCube(C, c), distCube(C, a), and
number of P, L, or R edges, but instead the sum distCube(C, f), and compare these distances
of the lengths of those edges—​now some com- with distNR and distNRVL. As a function on
bination of 1s and 2s. Let us write distNR for the triad, is distCube the same as either distNR or
original neo-​Riemannian distance and distNRVL distNRVL?
for the neo-​Riemannian voice-​leading distance. (b) What is distCube(C, b♭)? Is this the greatest
Then distNR(C, e), distNR(C, a), and distNRVL(C, e) possible Cube Dance distance from C major
are all equal to 1, but distNRVL(C, a) =​2.   to any other triad?

Distances • 495
 The only (major or minor) triads sharing two R-​, and L-​partners), but this relationship breaks
common tones are those related by P, L, or R—​ down if chords other than triads are considered.
exactly the triads at neo-​Riemannian distance 1. For example, the C-​major triad {C, E, G} shares
The neo-​Riemannian distance function does not two common tones with the trichord {C, D♭, E},
always correlate so perfectly with the number so the common-​ tone distance between these
of common tones (see Exercise 12.3.5 below), two chords is only 1, but at least six semitones
but a common-​tone distance function distCT can of total voice-​ leading displacement (in some
be defined directly—​and not only for triads. combination of voices) are required to connect
Suppose S is a collection of pitch-​class sets of the two.
some fixed cardinality n (with no doublings),
and suppose X and Y are sets in S. As in Exercise Exercise 12.3.7 This book includes many
2.1.5, write X\Y for the set X ∩ Y ′ consisting of graphs in which the concept of distance is
those elements that belong to X but not to Y. of some musical interest. Consult graphs
Then distCT(X, Y) is defined to be #(X\Y), the car- from earlier chapters as noted to answer the
dinality of X\Y. This is not the number of com- questions below, making sure you understand
mon tones but the number of other notes—​the the musical implications of the distances being
number of notes of the set X that must change calculated. Use the standard graph-​theoretic
in moving from X to Y. The number of common distance functions, assuming that all edges are
tones is n − distCT(X, Y), so as the number of com- of length 1.
mon tones decreases, the distance increases, as
we should expect. Among triads, for example, (a) In the graph of twelve-​tone row space in
distCT(B♭, g) =​1 while distCT(B♭, G) =​2. Exercise Figure 4.1.12a or Figure 4.1.16 (the graphs
12.3.6c confirms that distCT is a proper distance are the same), determine dist(P10, RI9).
function. Also determine dist(P10, RI9) in Figure
4.1.12b (a different graph).
Exercise 12.3.5 True or false: If two triads (b) In signed letter space, depicted by any
share one common tone, the neo-​Riemannian of the three graphs in Figure 1.2.11,
distance
   between them is always 2. determine dist(G♭, G♯), dist(F, A♯), dist(F, B♭),
and dist(F♯ ♯, B♭ ♭). Compare each of these
Exercise 12.3.6 distances with the distance between the
(a) When S is the space triad, determine same notes in spelled pitch-​class space
distCT(C, F♯), distCT(C, f), and distCT(C, g♯), (Figure 1.3.3).
and compare these distances with the (c) In the triangulated neo-​Riemannian
values of distNR, distNRVL, and distCube. tonnetz of Figure 3.3.10 (the dual graph of
(b) What is the largest possible common-​tone neo-​Riemannian triad space), determine
distance from C major to any other triad, dist(C, D), dist(C, E), dist(C, A♭), and
and how many triads lie at that distance dist(C, B♭). What is the greatest distance
from C major? between any two pitch classes in
(c) Explain why distCT satisfies the three this graph?
axioms (positivity, symmetry, and the (d) Assume that the Riemannian tonnetz of
triangle inequality) required of a distance Figure 1.4.14 is triangulated in the obvious
  function.16 way, with edges joining notes related
by major and minor thirds and perfect
Voice-​leading distances, clearly, are not gov- fifths. The distances dist(C, D), dist(C, E),
erned by common tones alone. The function dist(C, A♭), dist(C, C♯), and dist(C, B♯) along
distCT is defined entirely in terms of shared pitch-​ the central row are the same as in spc.
class content, without regard to the voice leading Calculate these distances and compare them
involved in moving from one set to another. The with dist(C, D), dist(C, E ), dist(C, Ab),
focus on triads and the tonnetz makes common-​ dist(C, C♯), and dist(C, B♯ ). Also determine
tone distances appear more similar to voice-​ dist(C, C), dist(C, C♯), and dist(C, B♯).
leading distances than they are for many other (e) A graph of Pythagorean pitch space was
kinds of chords. In a voice-​leading space such described in Exercise 2.3.25 and revisited
as three-​voice OP-​space (Figure 11.3.4), every in Section 7.3. Assume that edges in this
triad lies in close proximity to the other triads graph connect each note to its neighbors
with which it shares two common tones (its P-​, in both the fifth and octave dimensions, so

496 • E x p l o rin g M u sica l S paces

 that for example B4 is adjacent to E4, F♯5, B3, lengths can be determined in fairly straightfor-
and B5. Determine dist(C4, D4), dist(C4, D5), ward ways and offer intuitive ways of conceiving
dist(C4, D♭4), dist(C4, C♯4), dist(C4, F♯4), and of distances.
dist(C4, G♭4) in this graph. The second connection is group-​ theoretic,

determine dist(h.,  ) and dist(h., w). Does

(f) In the ski-​hill lattice of Figure 3.4.10, and we will explore it in more detail. In our
discussion of edges in a graph as “admissible
it make musical sense to you that these moves,” the reader may have noticed that many
distances should be the same? of these moves are well-​defined transformations,
(g) In Exercise 4.1.7 you constructed a space of perhaps elements of a transformation group. In
diatonic modes, a toroidal graph with a 12-​ the cyclic graph of pc, the admissible moves are
by-​7 modular tile. Assume that edges in this T1 and its inverse T11; both must be allowed in
graph connect each mode to its circle-​of-​ order that edges of the graph may be traversed
fifths neighbors in both the chromatic and in either direction. Either T1 or T11 generates the
diatonic dimensions, so that, for example, cyclic group 𝒯12, this group is isomorphic to the
G lydian is adjacent to C lydian, D lydian, interval group of the interval space pc, and we
C♯ locrian, and D ionian. In this graph, know from Chapter 6 that the Cayley diagram
determine dist(G ionian, G aeolian), dist(G of the group is essentially the same as the graph
ionian, E aeolian), dist(G ionian, D♭ ionian), of the space with which we started. A statement
  and dist(G ionian, C♯ locrian). such as dist(C, E♭) =​3 may be interpreted as a
statement about either the graph or the group.
The graphs in parts (b), (d), (e), and (f) of the In the graph, this statement means that the ver-
above exercise are infinite graphs. Each of these tices representing C and E♭ may be joined by a
graphs is connected, so the distance between path of length 3 but not by any shorter path.
any two vertices is a finite number, but there is As a statement about the group, dist(C, E♭) =​3
no upper bound to how large distances in these means that C is mapped to E♭ by the transforma-
graphs may be. tion T1T1T1, a word formed by combining three
admissible moves (in this case three identical T1
—​—​—​ moves)—​but not by any transformation that can
be written as a shorter word.
Graph-​ theoretic distance functions are con- In neo-​Riemannian triad space nrtriad,
nected with two other concepts introduced pre- each of the admissible moves P, L, and R is an
viously. First, the idea that a distance may be involution, so in this case there is no need to
defined as the length of the shortest path from consider inverses separately. Figure 12.3.1c may
one point to another emerged also in Section be seen as a Cayley diagram for the group ℛ of
12.2 in the context of pitch-​class space; recall Riemannian transformations from Section 8.2,
the illustration in Figure 12.2.6. If the discrete or as a transformation network showing the
space pc is drawn as a cyclic graph with twelve action of the generators P, L, and R on the space
vertices, the graph-​ theoretic distance is the triad. The neo-​Riemannian distance between
same as the distance defined in Section 12.2—​ two triads is the length of the shortest path join-
but the notion of the shortest path makes sense ing them, which is also the length of the shortest
also in the continuous space cpc, which is not word composed of the transformations P, L, and
a graph at all. This point illustrates a way in R that maps one triad to the other.
which distance functions can sometimes be con- Starting from any triad in the graph, paths
structed. In some spaces (including graphs, but through the graph correspond exactly to words
other spaces as well), prior to the definition of composed of P, L, and R. For example, distNR(A,
a distance function, it may be possible to define B♭) =​4, a large distance by the standards of neo-​
paths—​ways of getting from a point x to a point Riemannian triad space. It is characteristic of
y—​and there may be a simple way of measur- this space that triads separated by large dis-
ing the lengths of those paths. Then the distance tances may often be linked by several different
between x and y may be defined as the length of paths of minimal length.18 There is not just one
the shortest such path. This idea can be formal- path of length 4 from A major to B♭ major; Figure
ized in a general mathematical setting, though 12.3.8 shows that there are four such paths.
the general theory of path length is rather tech- These paths correspond to four different words
nical and we will not need the details here.17 In in the symbols P, L, and R: the paths labeled ①,
many cases of musical interest, however, path ②, ③, and ④ correspond respectively to the words

Distances • 497
 set obtained by adjoining to Z additional sym-
bols denoting the inverses of the elements in
Z. For some elements of Z it is possible that the
inverses are already present in Z (this is auto-
matically true in the case of elements that are
involutions); because those inverses are already
accounted for, it is not necessary to introduce
new symbols for them in Z*.
The elements of Z* correspond to the admis-
sible moves in the above discussion. Inverses
must be included to ensure the symmetry of the
distance function to be defined below. A word on
Z is a string of symbols selected from Z*. Even
FIGURE 12.3.8 Four paths of length 4 from A if Z is an infinite set, words are always finite;
major to B♭ major in neo-​Riemannian triad space, the length of a word is the number of symbols
corresponding to four PLR words of length 4 appearing in it. For example, if a, b, and c are
elements of Z, then abcccb−1b is a word on Z of
length 7. All repetitions of the same symbol are
LPRP, RPLP, RLPL, and PLRL, and the graph counted; b−1 is considered a single symbol, the
shows that each of these words, interpreted as symbol denoting the inverse of b. The substring
a series of transformations in the group ℛ, leads ccc may be abbreviated c3, but c3 still contributes
from A major to B♭ major. 3 to the word length. The substring b−1b contrib-
It is important to note something the pre- utes 2 to the word length, even though multiply-
ceding paragraph did not say: we did not assert ing by b−1b leaves the group element unchanged.
that four different transformations in the group There is no upper limit to the length of a word, so
ℛ map A major to B♭ major. We have identified infinitely many different words are possible even
four different words, but they all denote the on a small finite set Z.
same transformation. Whatever the input triad, Because every element of Z* denotes an ele-
the four words always yield identical outputs, so ment of the group G, every word on Z denotes
they are simply four different ways of realizing an element of G as well, obtained by combin-
the same mapping. Recall from Section 2.3 the ing the elements in the proper order using the
general principle that if f(x) =​ g(x) for all x, then group operation. If G is a group of transforma-
f and g are the same function; also recall from tions acting on some set S, then every word on
Theorem 8.2.2d that the action of ℛ on the set Z also defines a transformation on S, formed by
triad is simply transitive, which implies that if composition of functions. For example, if P, L,
two words produce the same output for even a and R denote the neo-​Riemannian transforma-
single input triad, those words must represent tions on triad, then the word LPLPRL denotes
the same transformation. The four words listed L ● P ● L ● P ● R ● L, which is also a transformation
above are not the only possible notations for the on triad; transformations are composed using
unique Riemannian transformation that maps A left-​to-​right orthography as usual. Two words on
to B♭: while no other words of length 4 define the Z are equivalent if they denote the same element
same mapping, many longer words do, such as of G; the above discussion of Figure 12.3.8 shows
LPLPRL or LRPRPR. (Exercises 4.2.12 and 6.2.16 that LPRP, RPLP, RLPL, and PLRL are equivalent
included some other relevant examples.) words in the group ℛ. This notion of equivalence
The concept of words is related to the idea of is indeed an equivalence relation on the set of all
group generators, introduced in Section 5.3.19 possible words.
Suppose G is a group and Z is a set of symbols To say that Z generates the group G means
standing for some elements of G. We may iden- that every element of G is denoted by some word
tify Z with a subset of G, but it is useful to dis- on Z. The transposition group 𝒯12 on pc is gener-
tinguish the group elements from the symbols ated by the single element T1 (that is, by the set
that stand for them. The set Z may theoretically Z =​{T1}), while the dihedral group 𝒯ℐ12 is gener-
be finite or infinite, but in most applications ated by the two elements T1 and I. As we know
Z consists of only a small number of elements from Section 8.2, the Riemann group ℛ is gen-
of a much larger group. Because G is a group, erated by P, L, and R. If G is a finite group gen-
every element has an inverse; let Z* denote the erated by Z, then we can draw a Cayley diagram

498 • E x p l o rin g M u sica l S paces

for G using the generators in Z. As described in inverses, T11 or T5). The choice of T1 (or T11) gives
Section 6.2, the vertices of a standard Cayley dia- rise to the chromatic pitch-​class circle and the
gram are labeled with the elements of the group, usual distance function on pc, while T7 (or T5)
but if G is a transformation group acting on a leads instead to the circle of fifths and its very
space S in simply transitive fashion, then the different conception of distance. In the Riemann
elements of G are in one-​to-​one correspondence group, it may seem natural to regard the trans-
with the elements of S, and the vertices may formations P, L, and R as somehow “primitive”
instead be labeled with elements of S in such a and all others as “composite,” but many other
way that the Cayley diagram becomes a transfor- generating sets could be chosen for this group—​
mation network on S. This is why the graph in including a smaller generating set, as Exercise
Figure 12.3.1c may be regarded both as a Cayley 12.3.11 below demonstrates.
diagram for ℛ and as a transformation network
on the space triad. (Strictly speaking, the edges Exercise 12.3.9 In the general setting
in this graph should be labeled in some way to described above, explain why the word-​length
identify them with P, L, and R—​the elements of distance distZ satisfies the three axioms
the generating set Z—​and each edge should be (positivity, symmetry, and the triangle
interpreted as a bidirectional arrow, each of the inequality)
   required of a distance function.21
generators being an involution.)
As Figure 12.3.8 illustrates, we may start at Exercise 12.3.10 As noted in Exercise 12.3.3b,
any vertex in such a graph and associate any word distNR(C, b♭) =​5. There are no fewer than nine
on Z with a path through the graph. The length different paths of length 5 from C to b♭ in the
of the path matches the length of the word (4 graph of Figure 12.3.1c, and correspondingly
in the case of the words in Figure 12.3.8). The nine different words of length 5 on the
distance between two points in the space may be symbols P, L, and R that denote the unique
defined in either of two equivalent ways: in the transformation in ℛ that maps C to b♭. Can you
graph, by minimal path length, or in the group, find
   all nine?
by minimal word length. For any transformation
group G with generating set Z acting on a space Exercise 12.3.11
S, a word-​length distance function may be defined (a) The inclusion of P in the generating set {P,
on S without reference to graphs at all: for all ele- L, R} for the group ℛ is redundant, because
ments x and y in S, distZ(x, y) is defined to be the by Theorem 8.2.2e the smaller set Z =​{L, R}
length of the shortest word on Z that denotes a generates ℛ. Construct a transformation
transformation mapping x to y.20 network on triad in the form of a Cayley
Many of the distance functions described pre- diagram for ℛ on the generators L and R.
viously in this chapter may be defined as distZ for (b) With the distance function distLR on triad
an appropriately chosen group G and generating defined by this smaller generating set Z,
set Z. The neo-​ Riemannian distance function what is the LR distance between parallel
distNR is the same as distZ, where Z is the generat-   triads such as C major and C minor?22
ing set {P, L, R} for the group ℛ. For the Weber
distance function, Z could be either {P, R, T7} or Exercise 12.3.12 Each of the distance
{P, R, T5}; because of the inclusion of inverses in functions listed below, considered earlier in
Z*, either T7 or T5 guarantees the availability of this chapter, may be defined as a word-​length
the other. The group generated by this set Z is the distance function distZ for some generating
Weber group 𝒲, introduced in Section 8.3; the set Z in an appropriate group. Describe the
action of 𝒲 on triad is not simply transitive. transformations in Z in each case. (The distance
The role of the generating set Z is an impor- functions in parts (c)–​(g) are the same as some
tant one. Word length is not an intrinsic prop- of those in Exercise 12.3.7.)23
erty of a group element. Groups usually have
more than one possible set of generators, lead- (a) The usual distance function in discrete pitch
ing to different sets of possible words and dif- space pitch
ferent distance functions; there may be no clear (b) The distance function in spelled pitch-​class
reason why one generating set should be privi- space spc (Figure 1.3.3)
leged over another. For example, the transposi- (c) Two distance functions in twelve-​tone row
tion group 𝒯12 may be generated by the element space, corresponding to the two graphs in
T1 or equally well by T7 (or by either of their Figure 4.1.12

Distances • 499
(d) The distance function in signed letter space may be defined—​ there are three particularly
sl, as in any of the graphs in Figure 1.2.11 important ones—​giving rise to several different
(e) The distance function in the (triangulated) (but strongly correlated) notions of voice-​lead-
conformed tonnetz (Figure 3.3.10) ing efficiency. This means that each OPTIC space
(f) The distance function in the (triangulated) is technically not just a single distance space but
Riemannian tonnetz (Figure 1.4.14) several different ones. Informally, we will con-
(g) The distance function in the ski-​hill lattice tinue to regard each OPTIC space as a single space,
  (Figure 3.4.10) but whenever we refer to distances we must spec-
ify which metric we are using.25
Exercise 12.3.13 By the methods described in We constructed the OPTIC spaces as quo-
this section, every Cayley diagram for a group tients of pitch-​tuple space cpitchn. Continuous
defines a distance function on the group itself. pitch space cpitch is a real interval space and
This observation relates to Cayley’s theorem, therefore a distance space as discussed in Section
discussed in Section 6.2, whereby every Cayley 12.2; our concern now is with products and quo-
diagram may be regarded as a transformation tients of distance spaces. We investigate products
network on the group. in Section 12.4, quotients in Section 12.5. As it
turns out, it is always possible to define a dis-
(a) Figure 6.4.3 showed Cayley diagrams for tance function on a product of distance spaces,
the group ℤ4 × ℤ3 on two different sets of but there is not just one way to do so. A different
generators: Z =​{(1, 0), (0, 1)} in part (a), complication arises in the case of quotients: it is
and Z =​{(1, 1)} in parts (b) and (c). These not always possible to define distance functions
diagrams define two different distance on quotient spaces formed from distance spaces.
functions on ℤ4 × ℤ3. Determine dist((0, 0), In the case of the OPTIC spaces, distance func-
(3, 1)) and dist((0, 0), (3, 2)) according to tions can always be defined, but with some cave-
each of these distance functions. What is ats to be noted involving cardinality equivalence.
the largest possible distance between two We begin with the two-​ dimensional space
elements of the group according to each cpitch2, a Cartesian product of two copies of
function? cpitch. Figure 12.4.1 shows two pitch pairs α
(b) The distance space defined by one of these =​ (x1, y1) and β =​(x2, y2) in cpitch2, along with
two distance functions on ℤ4 × ℤ3 is isometric three proposed ways of measuring the distance
to another distance space that we have between them. The x and y axes in this figure
  described previously. Which space is this? are oriented horizontally and vertically, unlike
the tilted coordinate systems used in pictures of
cpitch2 in Chapter 11. We introduce the nota-
12.4 DISTANCE FUNCTIONS tions Δx and Δy for the displacements in the
ON PRODUCT SPACES x and y coordinates between the two points:
The primary aim of the remainder of this chap- that is, Δx =​dist(x1, x2) =​|x2 − x1| and Δy =​
ter is to enhance our understanding of the geom-
etry of the OPTIC spaces by studying distances in
these spaces. Paths through these spaces depict
voice leadings as detailed in Chapters 10 and
11, and distance measures provide ways to make
precise many intuitions about voice leading. The
length of a path is a measure of voice-​leading effi-
ciency; longer paths require the voices to do more
musical “work” by moving through larger inter-
vals.24 In general, the distance between two points
in an OPTIC space is the length of the shortest
path joining them—​a measure of the size of the
most efficient possible voice leading. The distance
between two chords can be highly dependent on
the choice of space in which the chords are consid-
ered (pitch-​tuple space, OP-​space, or OPTI-​space,
for example). Additionally, we will see that even in FIGURE 12.4.1 The max, Euclidean, and voice-​
a single space several different distance functions leading distance functions in cpitch2

500 • E x p l o rin g M u sica l S paces

dist(y1, y2) =​|y2 − y1|, where dist denotes the 5.83, and distVL(α, β) =​5 +​3 =​8. Musical inter-
usual one-​ dimensional distance function in pretations of the voice-​leading distance and the
cpitch. The displacements Δx and Δy are thus max distance are straightforward: the former
the lengths of the horizontal and vertical legs accounts for the eight semitones of displace-
of the triangle in the figure. The three proposed ment in the two voices combined, while the lat-
measures of the distance between α and β are as ter measures the largest distance moved by one
follows: voice (five semitones). The Euclidean distance,
meanwhile, is geometrically accurate as a mea-
• The max distance is defined as distMax(α, β) =​ sure of distance in the plane of cpitch2. As this
max(Δx, Δy), the larger of the two displace- example illustrates and as Figure 12.4.1 probably
ments Δx and Δy. In Figure 12.4.1, Δx is larger suggests in general, the inequalities distMax(α, β)
than Δy, so distMax(α, β) =​ Δx. ≤ distEuc(α, β) ≤ distVL(α, β) are true for all α and
• The Euclidean distance is distEuc(α, β) =​ β, with equality holding only if either Δx or Δy is
( ∆x )2 + ( ∆y )2 =​[(Δx)2 +​(Δy)2]1 ⁄ 2, the length 0. (The “max” in the max distance refers to the
of the hypotenuse of the triangle. maximum displacement in any voice; for fixed
• The voice-​leading distance is distVL(α, β) =​ α and β, the max distance is the smallest of the
Δx +​ Δy, the sum of the two displacements. three distance functions.)
This definition matches the way we have The values of all three distance functions are
described voice-​leading distances previously, completely determined by the displacement multi-
starting in Section 10.2. set {Δx, Δy}, which is {3, 5} in the above example.
The distances computed by any of the functions
The Euclidean distance formula corresponds would be the same for any other pair of points
to the usual way of calculating the diagonal dis- for which Δx and Δy are 3 and 5—​in either order,
tance between two points in the xy coordinate which is why we write {Δx, Δy} as an unordered
plane. The formula is nothing more than the multiset rather than an ordered pair. Because Δx
familiar Pythagorean theorem in geometry: (Δx)2 and Δy may be the same in some cases, {Δx, Δy}
+​ (Δy)2 =​ c2, where c is the length of the hypot- must be regarded as a multiset rather than as a
enuse of the triangle. Euclidean distances can be set. All the standard distance functions in the
measured in the plane with an ordinary ruler, OPTIC spaces are calculated from displacement
assuming that the space is drawn accurately and multisets. The distance between two chords, or
the ruler is marked in the appropriate units. the efficiency of a voice leading, is essentially a
The other two distance formulas are simpler: measure of the “size” of the associated displace-
the voice-​leading distance merely adds the two ment multiset, in a sense that involves not only
displacements—​no square root needed—​while the cardinality of the multiset but also the size of
the max distance uses only one of the two, its elements.26 A complication is the fact that in
whichever is larger. The voice-​leading distance some OPTIC spaces there may be more than one
may also be called the sum distance; mathema- straight-​line path from one point to another,
ticians sometimes refer to it as the city-​block with different displacement multisets and there-
metric or the taxicab metric because it records fore different path lengths; in such a case the
the distance you must travel from one point to distance is defined using whichever path pro-
another if constrained to follow a rectilinear duces the smallest value for the chosen distance
grid rather than cutting across the grid diago- function. The possibility of multiple paths will
nally. Although we reserve the name voice-​lead- come into play in the quotient spaces studied in
ing distance for this sum, one of the points of the Section 12.5, but this is not a concern at present,
present discussion is that the Euclidean and max because in cpitchn the straight-​line path joining
distances are relevant measures of some aspects two points is unique.
of voice leading also. Exercise 12.4.2 For each of the following
For example, suppose α is the pitch pair α =​ pairs (α, β) of points in cpitch2, determine
(A3, F♯4) =​(−3, 6) while β =​(D4, E♭4) =​(2, 3). Then the displacement multiset and the distances
Δx =​|2 − (−3)| =​5 and Δy =​|3 − 6| =​3: in mov- distMax(α, β), distEuc(α, β), and distVL(α, β).
ing from α to β, one voice (the x coordinate in
the ordered pairs) moves by five semitones, the (a) α =​(G4, E♭4), β =​(F4, D4)
other (y) by three. It follows that distMax(α, β) =​ (b) α =​(C4, G5), β =​(C♯4, A5)
max(5, 3) =​5, distEuc(α, β) =​ 52 + 32 = 34 ≈ (c) α =​(C4, C4), β =​(C♯4, C♯4)

Distances • 501
(d) α =​(C4, E♭4), β =​(E♭4, F♯4) First, the same construction works in any prod-
(e) α =​(C4, E♭4), β =​(F♯4, E♭4) uct of distance spaces at all. That is, if (S1, dist1)
(f) α =​(A3, F4), β =​(C4, A4) and (S2, dist2) are distance spaces, we can define
(g)  α =​(C4, G4), β =​(C5, C5) max, Euclidean, and voice-​ leading (sum) dis-
tances on the Cartesian product S1 × S2 exactly
All three of the functions distMax, distEuc, and as above, substituting dist1(x1, x2) for Δx and
distVL are proper distance functions. The positiv- dist2(y1, y2) for Δy in our original formulas. (For
ity and symmetry axioms should be evident in all distances in the product space to be meaningful,
cases. The triangle inequality for the Euclidean the distances in S1 and S2 should be measured
distance is intuitively clear, as it simply asserts, in in the same units—​typically semitones in the
two-​dimensional Euclidean space, the property for cases of interest to us.) Second, and of particu-
which the inequality is named: the length of one lar importance for our study of OPTIC spaces,
side of a triangle can never exceed the sum of the the formulas generalize readily to more than
lengths of the other two sides.27 To understand two dimensions. For pitch triples α =​(x1, y1, z1)
the triangle inequality for the other two distance and β =​(x2, y2, z2) in cpitch3, the three distance
functions requires consideration of three points, α functions take the following forms (of course, Δz
=​ (x1, y1), β =​(x2, y2), and γ =​(x3, y3). For the voice-​ denotes dist(z1, z2) =​|z2 − z1|):
leading distance the inequality takes the form
• distMax(α, β) =​max(Δx, Δy, Δz);
distVL(α, γ) ≤ distVL(α, β) +​distVL(β, γ). • distEuc(α, β) =​​ [(Δx)2 +​(Δy)2 +​(Δz)2]1 ⁄ 2;
• distVL(α, β) =​ Δx +​ Δy +​ Δz.
Because the voice-​leading distance is the sum of
the componentwise displacements Δx and Δy, The extension to spaces of four or more dimen-
which are simply distances in one dimension, sions should be clear.30
this inequality is equivalent to
Exercise 12.4.5 For each of the following pairs
[dist(x1, x3) +​dist(y1, y3)] ≤ [dist(x1, x2) +​ of points in cpitch3 or cpitch4, calculate the
dist(y1, y2)] +​[dist(x2, x3) +​dist(y2, y3)]. displacement multiset and the distances distMax,
distEuc, and distVL.
We know from Section 12.2 that the triangle
inequality holds in the one-​dimensional space (a) (C4, E4, G4), (D♭4, F4, A♭4)
cpitch. This applies in both the x and y com- (b) (C4, E4, G4), (B♭3, D♭4, F4)
ponents; that is, the inequalities dist(x1, x3) ≤ (c) (C4, E4, G4), (B♭3, F4, D♭5)
dist(x1, x2) +​dist(x2, x3) and dist(y1, y3) ≤ dist(y1, (d) (C4, E4, G4), (B♭3, G♭4, C5)
y2) +​dist(y2, y3) both hold. The inequality in the (e) (C4, D4, E4), (D4, E4, F♯4)
two-​dimensional case follows by adding together (f) (C4, D4, E4), (F♯4, D4, E4)
these two simpler inequalities. (g) (B3, D♯4, A4), (C4, E4, G4)
(h) (C3, E4, G4, B♭4), (F3, C4, F4, A4)
Exercise 12.4.3 Using the one-​dimensional (i) (F3, B3, D♯4, G♯4), (E3, B3, D4, G♯4)
inequalities in the x and y components as above, (j)  (F3, B3, D♯4, G♯4), (E3, G♯3, D4, B4)
explain why the triangle inequality holds for the
max distance: Parts (c) and (d) of the above exercise illus-
trate an important property. The displacement
max[dist(x1, x3), dist(y1, y3)] multisets are {1, 2, 6} in (c) and {2, 2, 5} in (d).
≤ max[dist(x1, x2), dist(y1, y2)] The first multiset can be converted to the sec-
+​ max[dist(x2, x3), dist(y2, y3)].28 ond by decreasing a large displacement in the
   first set by one semitone (changing 6 to 5) while
Exercise 12.4.4 Suppose we try to define increasing a smaller displacement by the same
the “min distance” in cpitch2 by the formula amount (changing 1 to 2). This change, which
distMin(α, β) =​min(Δx, Δy), the smaller of the has the effect of leveling the distribution of
two componentwise displacements. Is distMin a displacements, does not affect the voice-​lead-
proper
   distance function?29 ing (sum) distance, but it reduces both the max
and Euclidean distances. This is a general prin-
The three distance functions just described ciple: if a displacement multiset is made more
in cpitch2 may be generalized in several ways. even while preserving its sum, the Euclidean

502 • E x p l o rin g M u sica l S paces

distance will decrease, as will the max distance (b) Suppose S1 and S2 are homogeneous
if the largest displacement is decreased by the distance spaces. Show that with any of
change.31 the distance functions distMax, distEuc, or
Pitch-​tuple space cpitchn, as we know, is distVL, the product space S1 × S2 is also
both an interval space and a distance space. The   homogeneous.32
interval-​space structure on cpitchn is deter-
mined by the general principles of product Figure 12.4.7 compares the three distance
interval spaces described in Section 7.3, with functions on cpitch3 as they apply to two pairs
intervals calculated componentwise; in cpitch3, of triads: the R-​related F major and D minor, and
for example, the interval between the two pitch the hexatonic poles F major and C♯ minor. The
triples in Exercise 12.4.5g is int((B3, D♯4, A4), (C4, displacement multisets are {0, 0, 2} for the F–​d
E4, G4)) =​(1, 1, −2). The distance-​space struc- pair, {1, 1, 1} for F–​c♯. All three functions agree
ture, meanwhile, arises from the choice of any that the F–​d distance is 2 (they will always agree
of the distance functions distMax, distEuc, or dis- when the motion takes place in a single voice),
tVL introduced above. By any of these metrics, but the three metrics return different distances
the distance is calculated from the displace- for F–​c♯: the voice-​leading distance between the
ment multiset, and this multiset ({1, 1, 2} in hexatonic poles is 3, while the other two dis-
the present example) can be deduced from the tances are smaller than 2. The distance functions
multidimensional interval. It follows that equal are therefore in disagreement about which voice
intervals imply equal distances, which is to say leading covers a larger distance. Many musicians
that the interval and distance functions are met- may share an intuition that F major seems closer
rically compatible as defined in Section 12.2. to D minor than to C♯ minor; this judgment is
Moreover, the product space cpitchn is also a likely in part an effect of the common tones, but
homogeneous space as defined in Section 12.2 musical intuitions about “closeness” may also be
(any point can be mapped isometrically to any influenced by many factors besides voice-​leading
other): the transposition T(1,1,−2) is an isometry displacements, in this case perhaps including the
of cpitch3 that maps (B3, D♯4, A4) to (C4, E4, G4). familiarity of the relative relation between triads,
The following exercise provides more general the fact that triads in this relation (unlike hexa-
statements of these properties. tonic poles) belong to a common diatonic scale,
or associations between (for example) C♯ minor
Exercise 12.4.6 as the name of a triad and as the name of a scale
(a) Suppose S1 and S2 are spaces on which both or key.
interval-​space structures and distance-​space The above example, in any case, is relatively
structures are defined, and suppose that exceptional. In a broader sense the three dis-
on both spaces the interval function and tance functions are not very different. The three
distance function are metrically compatible. functions produce different numbers, to be sure,
On the product space S1 × S2, show that but the three are strongly correlated; more often
the product interval function is metrically than not, they are in agreement about which of
compatible with all of the distance two distances is larger. Figure 12.4.8 returns to
functions distMax, distEuc, and distVL. a previous example in cpitch2, involving dyads

FIGURE 12.4.7 Comparison of max, Euclidean, and voice-​leading distances between triads

Distances • 503
FIGURE 12.4.8 Zone of disagreement for distance functions in cpitch2 (after Hall and Tymoczko)

α and β for which Δx =​5 and Δy =​3. Originally distVL(α, χ) =​8. Threading its way between the
α was (−3, 6) while β was (2, 3), but this figure two squares is a circle of radius 34 compris-
assumes that α =​(0, 0) while β =​(5, −3). This ing the points χ for which distEuc(α, χ) =​ 34.
adjustment does not affect the values of the The three regions may be regarded as general-
distances calculated previously, which depend ized spheres, or at least as something analogous
only on Δx and Δy: distMax(α, β) =​5, distEuc(α, to spheres for each of the three distance func-
β) =​ 34 , and distVL(α, β) =​8. We are interested tions, in that they consist of all points at a fixed
in comparing these distances from α to β with distance from the fixed point α: in two dimen-
distances from α to other points. In particular, sions a max-​distance sphere is a square, a voice-​
we are interested in the question of whether leading–​distance sphere is a tilted square, and a
dist(α, β) is larger or smaller than dist(α, χ) for Euclidean sphere is a circle. The “radii” of these
various points χ, and whether the three distance spheres, 5, 8, and 34, are chosen to match the
functions are in agreement about which of the respective distances from α to β, and therefore
two distances is larger. The preceding paragraph β lies on all three spheres, at a point where the
provided an example in cpitch3 in which the three region boundaries intersect.
functions disagreed. The area shaded in the figure consists of points
The three large regions of cpitch2 outlined lying inside at least one of the three regions but
in the figure—​two squares and one circle, all outside at least one other. For a point χ within
centered on α—​each show, for one of the three the shaded area, therefore, dist(α, χ) may be
distance functions, the locations of all dyads either larger or smaller than dist(α, β) depend-
χ for which dist(α, χ) takes the same value as ing on which distance function is chosen. These
dist(α, β). The points χ for which distMax(α, χ) =​5 areas of disputed territory form only a small part
lie on the boundary of the 10 × 10 square with of the plane, however: the zone of disagreement
sides parallel to the coordinate axes. The larger fills less than half as much area as the unshaded
tilted square consists of those χ for which octagonal central region, in which dist(α, χ)

504 • E x p l o rin g M u sica l S paces

< dist(α, β) for all three distance functions, and distance has been used, explicitly or implicitly,
is of course vastly smaller than the unbounded in many studies of voice leading.35
exterior region where dist(α, χ) > dist(α, β) for But the voice-​ leading distance too has its
all three functions. Similar observations apply in drawbacks. A large value of distVL may disguise
higher dimensions.33 the smoothness of a voice leading in many
voices. Moreover, it is a mistake to assume that
Exercise 12.4.9 Ten points in cpitch2 are listed voice leadings through the smallest distances
below. Taking each point in turn as χ, locate χ in (by any measure) are always musically desir-
Figure 12.4.8. Calculate distMax(α, χ), distEuc(α, able. The two chords in Exercise 12.4.5h above
χ), and distVL(α, χ) for each χ, compare these imply a V7–​I resolution illustrating voice leading
values with the corresponding distances from α traditionally considered proper, the large voice-​
to β, and verify that the measures disagree about leading distance notwithstanding (a smaller
the relative sizes of dist(α, β) and dist(α, χ) only distance would result from “wrongly” resolving
for those points χ lying within the indicated the top three voices upward). Meanwhile we are
zone of disagreement. Some of the points lie taught to avoid parallel motion as in part (a) of
exactly on a boundary of this zone. Do the the same exercise, for which the voice-​leading
distance functions disagree at these points? distance is small. In resolving a bo7 chord to a C-​
minor triad (with a doubling) in four voices—​a
(a) (4, 6) traditional leading-​tone resolution of the dimin-
(b) (41 2 , 41 2 ) ished seventh—​ the smallest possible voice-​
(c) ( 17 , 17 ) leading distance is 5 semitones. In contrast, a
(d) (5, 4) common-​tone resolution of bo7 to B minor can
(e) (6, 3) be accomplished with a voice-​leading distance of
(f) (6, 2) only 3 semitones, and the nonstandard resolu-
(g) (6, 1) tion of bo7 to C♯ minor with 4. The max distance is
(h) (51 2 , 0) 2 for all three resolutions, perhaps reflecting an
(i) (5, −1) intuition that all three are comparably smooth;
(j)  (4, −2) the preference for the resolution to C minor evi-
dently depends on considerations other than
The choice of an appropriate distance function voice leading. For some purposes a voice lead-
should depend on the purposes for which it is to ing in which all voices move stepwise (by no
be used. For studying cpitchn and the OPTIC more than two semitones) may be considered
spaces geometrically, the Euclidean distance appropriately “smooth”; only the max distance
function is likely to be the most useful choice. recognizes this property. In other cases we may
Several references to distances in these spaces choose to give priority to a voice leading that
in Chapters 10 and 11 implicitly assumed the minimizes the number of moving voices, a quan-
Euclidean distance, and awareness of Euclidean tity not measured by any of the three distance
distances is required of anyone who sets out to functions in this section (but, for chords of fixed
draw accurate pictures of voice leadings through cardinality, by the common-​tone distance func-
OPTIC spaces. Calculation of areas (in cpitch2) tion introduced in Section 12.3).36
and volumes (in cpitchn for n ≥ 3), such as the If the chords are triads there are still more
areas mentioned above in Figure 12.4.8, also options, in the form of the circle-​ of-​fifths,
typically presupposes a Euclidean understand- Weber, and neo-​Riemannian distances described
ing of distance.34 previously. These distance functions are not
On the other hand, Euclidean distances may directly comparable to distMax, distEuc, and dis-
sometimes seem computationally awkward or tVL, partly because they are defined only for tri-
musically unintuitive. To assert a microtonal ads, but also because they regard those triads as
value such as 34 semitones for the distance pitch-​class sets—​OP-​ or OPC-​classes—​rather
between two equal-​tempered dyads may strike than pitch triples. In order to see how the dis-
musicians as peculiar. The voice-​ leading dis- tance functions defined on cpitch3 may apply
tance probably matches musical intuitions best to OP-​space and other OPTIC spaces, we must
in many cases, as reflected in the name we have consider the problem of distance functions on
chosen to use for it. Indeed, the voice-​leading quotient spaces—​the topic of the next section.

Distances • 505
12.5 DISTANCE FUNCTIONS grasp of this example will aid our understanding
ON QUOTIENT SPACES; of the more general problem.
OPTIC SPACES AS Figure 12.5.1a shows the torus of cpitch2/​~O.
This space corresponds to the fundamental
DISTANCE SPACES region in cpitch2 from Figure 11.2.1. The square
If ~ is an equivalence relation on a set S, then as in the earlier figure was tilted along with the
described in Section 2.4, a quotient set S/​~ may coordinate axes, but in Figure 12.5.1a the axes
be formed as a set of equivalence classes. If S is are horizontal and vertical. Points in this space
endowed with some additional algebraic, topo- correspond to O normal forms [x, y], ordered
logical, or geometric structure, and if we would pairs of mod-​12 real numbers representing pitch
like to define the same sort of structure on classes. The edges of the figure are false boundar-
S/​~, we may or may not be able to do so: the ies, as usual for a torus, and the point (C, C) =​[0,
relation ~ must be compatible with the struc- 0] appears at all four corners of the square.
ture in some way in order for the structure to Points labeled α, β, γ, and δ are shown in the
carry over to the quotient space. In Section interior of the square. Path ①, near the center,
6.6 we studied the conditions under which an depicts a straight-​line voice leading from α =​(D♯,
equivalence relation ~ on a group G allows for F) =​[3, 5] to β =​(G, E) =​[7, 4]. The component-
the formation of a quotient group G/​~, and in wise displacements are Δx =​4 and Δy =​1, from
Section 7.4 we considered the analogous ques- which the values of the distance functions are
tion for an interval space (S, G, int); in both readily calculated as distMax(α, β) =​4, distEuc(α, β)
cases the required condition involved a normal =​ 17 ≈ 4.12, and distVL(α, β) =​5.
subgroup of G. It should not come as a surprise, Now consider the same calculation for the
then, to learn that if ~ is an equivalence relation points γ =​(A, B) =​[9, 11] and δ =​(C♯, A♯) =​[1,
on a distance space (S, dist), we may or may not 10]. Because γ and δ are tritone transposi-
be able to define a distance function on the quo- tions of α and β, dist(γ, δ) should presum-
tient space S/​~, depending on the properties of ably be equal to dist(α, β) according to any of
the relation ~. the standard distance functions. Indeed, cal-
A general characterization of the require- culations of Δx and Δy for γ and δ, using the
ments for a quotient distance space turns out to distance function in cpc from Section 12.2,
be a thorny problem, which we will not consider give Δx =​|1 − 9|12 =​4 and Δy =​|10 − 11|12 =​ 1,
in its full generality. Fortunately, the relations matching the displacements between α and β,
of octave, permutational, transpositional, and so all of the distance calculations will give the
inversional equivalence are well-​behaved in this same results as well. But a need to distinguish
regard, and distance functions in quotient spaces between distance and path length becomes more
involving these relations are easily described apparent in the case of γ and δ, which lie near
conceptually (though not always so easy to cal- corners of the square.
culate in practice). Cardinality equivalence, as is The shortest path from γ to δ in the torus, path
so often the case, poses extra challenges, to be ②, crosses a false boundary, wrapping around to
considered at the end of this section. the other side of the square. For this path, the
We begin with two-​voice O-​space cpitch2/​~O, displacements in the x and y coordinates are 4
a straightforward case. We know from Chapters and 1, matching the calculation above. Path ③ is
10 and 11 that two-​voice O-​space may be iden- another straight-​line voice leading from γ to δ,
tified with cpc2, a product of two copies of the one that does not wrap across the boundary—​
continuous pitch-​class circle cpc, represented by but path ③ is longer than path ②. Path ③ gives
a torus with a 12-​by-​12 modular tile. As shown displacements Δx =​8 and Δy =​1, so the length
in Section 12.2, cpc is itself a distance space, so of path ③ is 8 by the max distance, 65 by the
distance functions distMax, distEuc, and distVL can Euclidean distance, or 9 by the voice-​leading dis-
all be defined on cpc2 via the general construc- tance. These are all larger than the corresponding
tion of distance functions on product spaces values calculated from paths ① and ②. Because
from Section 12.4. This product construction path ② is the shortest path from γ to δ, the dis-
works only for O-​space (P-​, T-​, and I-​spaces in tance between γ and δ is equal to the length of
two or more voices cannot be constructed as path ②; path ③ is irrelevant to the distance cal-
products of one-​dimensional spaces), but some culation. In cpitchn there is only one straight-​
aspects of the distance functions on two-​voice line voice leading from one point to another, and
O-​space carry over to the other relations, so a so we did not need to distinguish distance from

506 • E x p l o rin g M u sica l S paces

FIGURE 12.5.1 (a) Paths in two-​voice O-​space; (b) corresponding voice leadings and distances; (c)
corresponding points in cpitch2

Distances • 507
path length. In quotient spaces, however, there distance between any representatives of those
can be many such paths with different displace- equivalence classes in cpitch2.
ment multisets and therefore different lengths; This reasoning generalizes to more than two
the distance between two points is the length of dimensions and applies to P, T, and I equiva-
the shortest path. lence as well. To be precise, suppose ~ denotes
Voice leadings corresponding to paths ①, ②, ~O, ~P, ~T, ~I, or a composite equivalence rela-
and ③ are shown in musical notation in Figure tion obtained by combining some of these (such
12.5.1b along with a summary of the path-​length as ~OP or ~OPTI), and suppose dist is any of the
calculations. We are working in O-​space, not distance functions distMax, distEuc, or distVL on
OP-​space, so the order of elements within each cpitchn. Then a corresponding distance func-
ordered pair is fixed: the musical notation shows tion dist ~ may be defined on the quotient space
the first pitch class in each ordered pair with a cpitchn/​~ by the quotient distance formula
downward stem, the second with an upward stem.
In moving from α =​(D♯, F) to β =​(G, E) along path
①, D♯ must move to G and F to E. (As shown in dist~( α, β) = min {dist(α 0 , β0 )|α 0 ∈α, β0 ∈β} .
Exercise 12.5.2a below, there is a shorter path
from α =​(D♯, F) to β′ =​(E, G), but β and β′ are
The elements α and β of cpitchn/​~ are equiva-
two different points in O-​space.) Moreover, even
lence classes of points in cpitchn modulo the
though we are working under the assumption of
octave equivalence, D♯ must move up four semi-
relation ~; the quotient distance formula says
that the distance dist~(α, β) in the quotient
tones to G—​not down eight or up sixteen, for
space is the smallest possible value of dist(α0, β0)
example—​in order to match path ①, which shows
among all possible pairs of points (α0, β0) in
the x coordinate increasing by 4 over the course
cpitchn such that α0 belongs to the equivalence
of the voice leading. We may place the notes of α
in whatever register we wish—​D♯ could be writ-
class α and β0 belongs to the equivalence class β.
This definition implies that dist~ is measured in
ten in a higher register than F, for example—​but
the same units as the original distance function
in order to obtain the displacements implied by
path ①, D♯ must move up four semitones to G,
dist, typically semitones. We generally omit the
subscript ~ from the notation dist~, using the
and F must descend by semitone to E.
same notation for the distance function on the
Figure 12.5.1c shows another way to think
quotient space as for the one on cpitchn.
about these distances. This figure shows a por-
While the above form of the quotient dis-
tion of the infinite plane of cpitch2, includ-
tance formula applies to O, P, T, and I equiva-
ing nine copies of the fundamental region for
lence, it fails to work for many other equivalence
octave equivalence. Each region contains one
relations. We will see later that this formula does
point representing each pitch-​ class pair; the
not produce a valid distance function on quo-
points representing α, β, γ, and δ are shown with
tient spaces involving cardinality equivalence,
their coordinates in cpitch2. It is evident in this
and Exercise 12.5.3 below shows another exam-
figure that many pairs of points representing
ple of a quotient space, unrelated to the OPTIC
α and β, and many other pairs representing γ
relations, on which no suitable distance function
and δ, are in the relationship corresponding
can be defined at all.37
to paths ① and ② in part (a), with Δx =​4 and
Δy =​1; for instance, the points α0 =​(3, 5) and Exercise 12.5.2
β0 =​(7, 4) form such a pair, as do γ0 =​(9, 11) (a) Let β′ be the pitch-​class pair (E, G) obtained
and δ0 =​(13, 10). Many α–​β pairs are separated by exchanging the components of β in the
by greater distances than that, but no pairs above discussion. For each of the three
are closer together. The path from any α to distance functions, calculate the distance
the nearest β lies entirely in one fundamental in O-​space from α =​(D♯, F) to β′. Draw
region, but the point from any γ to the nearest the shortest path from α to β′, and write
δ crosses a region boundary. By whichever dis- a corresponding voice leading in musical
tance function we choose, the distance between notation. Observe that the distances from α
any of these minimal-​distance pairs of points to β′ are shorter than those from α to β.
in cpitch2 matches the value we calculated for (b) For each of the three distance functions,
the distance in O-​space (4, 17, or 5). That is, calculate the distance from β =​(G, E) to
the distance between two points in the quotient γ =​(A, B). Draw the shortest path and write
space cpitch2/​~O equals the smallest possible a corresponding voice leading.

508 • E x p l o rin g M u sica l S paces

(c) For each of the three distance functions, distance formula implies dist~(3, 4) =​0. Since
calculate the distance from β =​(G, E) to δ =​ 3 and 4 are two different equivalence classes
(C♯, A♯). There is not a unique shortest path (elements of the quotient space ℝ/​~), the
in this case (why not?), but the distance is positivity axiom from Section 2.2 requires that
uniquely determined. Write voice leadings a proper distance function record a positive
corresponding to each of the possible distance between them, not 0.
shortest paths. (c) Calculate dist~(1, 3), dist~(3, 5), and
(d) In Figure 12.5.1c, draw a diagonal line
dist~(1, 5) using the modified formula, and
graphing the equation x =​ y. Of the pairs of
show that these values violate the triangle
points representing α and β and separated
  inequality.
38
by the smallest possible distance, some
pairs (such as (3, 5) and (7, 4)) have the
property that the path joining them crosses Before leaving O-​ space, we recall from
this diagonal line, while other pairs (such Section 12.1 that O-​ space is one of the few
as (15, 5) and (19, 4)) do not have this OPTIC spaces to possess an interval-​space struc-
property. What does this property say about ture. We should therefore expect O-​space to
  a voice leading? be a homogeneous distance space. Indeed it is:
every pitch-​class pair is mapped to every other
Exercise 12.5.3 For any real number x, let pair by some transposition, and transpositions
⎣x⎦ denote the greatest integer in x, the largest are isometries of O-​space, so any point can be
integer n such that n ≤ x. We may interpret this mapped isometrically to any other. (This homo-
function musically as a mapping from cpitch geneity also follows from Exercise 12.4.6b, using
to pitch that rounds every note down to the the product-​space construction of O-​space.) We
next lower equal-​tempered semitone. Define a will see below that many other OPTIC spaces fail
relation ~ on ℝ so that x ~ y is true whenever to be homogeneous.
⎣x⎦ =​ ⎣ y⎦—​that is, x and y round off to the The quotient distance formula shows con-
same number (or note). All numbers x such that ceptually how distances can be measured in any
3 ≤ x < 4 are equivalent by this relation, as they of the quotient spaces described in Chapter 11
all round down to 3. Write n for the equivalence (excluding cardinality equivalence). On the other
class containing the integer n; that is, n =​ hand, the formula can be impractical to apply, as
{x ∈ ℝ | n ≤ x < n +​1}. The point of this exercise it requires that we compare distances dist(α0, β0)
is to show that, if we start with the usual for all possible representatives α0 and β0 of the
distance function dist on ℝ (or on cpitch), equivalence classes α and β. There are often a
then the quotient distance formula dist~(m, n) great many representatives to choose from—​in
=​ min{dist(x, y) | x ∈ m, y ∈ n } fails in several some cases (as in Figure 12.5.1c) infinitely many.
different ways to produce a proper distance When calculating a distance in a quotient space,
function on the quotient space ℝ/​~. it is usually more feasible to consider a funda-
mental region for the space, as in 12.5.1a, con-
(a) Consider how we would calculate dist~(3, 5) taining only a single representative of each class.
using the quotient distance formula. Show Because the shortest path joining two points in
that for any number r slightly larger than 1, the fundamental region may cross the boundar-
such as r =​1.001, one can find numbers x ∈ ies of the region, recognizing the shortest path
3 and y ∈ 5 such that dist(x, y) =​ r, but that requires an understanding of the geometry of
there are no such x and y for which dist(x, the region and the topological properties of its
y) =​1 exactly. Therefore the minimum value boundaries. This is a fairly simple matter in the
specified in the formula does not exist. torus of O-​space, but potentially more confus-
(b) The problem described in part (a) may be ing in some other spaces. Writing voice leadings
circumvented by defining dist~(m, n) to be in musical notation is often helpful as well, as a
the greatest lower bound of all suitable values way of visualizing displacement multisets. The
of dist(x, y) rather than the minimum. The examples and exercises in the remainder of this
greatest lower bound of the real numbers section offer many illustrations.
greater than 1 is 1, so dist~(3, 5) =​1 by Figure 12.5.4a shows several voice-​leadings
this modified formula. But there are other in two-​voice OP-​space, the Möbius strip from
problems not solved by this modification. Figure 11.2.6. The associated musical notation
Show that this redefinition of the quotient and distance calculations are presented in Figure

Distances • 509
FIGURE 12.5.4 (a) Paths in two-​voice OP-​space; (b) corresponding voice leadings and distances

12.5.4b. The orientation of the space here is as joining these dyads is the straight line shown,
in Chapter 11, so transposition is vertical in the corresponding to the voice leading D♯ → E, F → G
figure, zero-​sum voice leading is horizontal, and with both voices ascending. If it is not clear how
motion in a single voice is diagonal. Dyads are the path translates to this voice leading, it may
denoted by OP normal forms; because permuta- be helpful to divide the motion into single-​semi-
tional equivalence is assumed, there is no fixed tone moves, D♯F → EF → EF♯ → EG. Each step in
association of displacements in the x and y com- this chain is easily traced in the space as a move
ponents with directions in the figure. to a nearby lattice point, and the three steps in
Path ① connects dyads D♯F and EG, the OP-​ combination produce the given voice leading.
classes corresponding to the pitch-​class pairs The displacement multiset for this voice lead-
α =​(D♯, F) and β =​(G, E) studied in O-​space in ing is {1, 2}. In O-​space the displacement mul-
Figure 12.5.1. Because the dyads are now unor- tiset from α to β was {1, 4}, so the distance, by
dered, a note in the first dyad is now free to move whichever metric we choose, is reduced in OP-​
to either note in the second. The shortest path space: the max distance is 2 (instead of 4), the

510 • E x p l o rin g M u sica l S paces

Euclidean distance is 5 ≈ 2.24 (instead of 17 ≈ 3–​4–​5 right triangle. In OP-​space, therefore, the
4.12), and the voice-​leading distance is 3 (instead calculation distMax(DF♯, F♯A) is governed by the
of 5). These distances should agree with the dis- minimal length (4) of path ⑤, while distVL(DF♯,
tances computed in Exercise 12.5.2a between α F♯A) =​5 because of path ④. For distEuc(DF♯, F♯A)
=​ (D♯, F) and β′ =​(E, G). The pitch-​class pairs β =​ the minimal length is attained by both paths.
(G, E) and β′ =​(E, G) are two different points of
O-​space corresponding to the same dyad in OP-​ Exercise 12.5.5 For each of the following pairs
space. The distance between D♯F and EG in OP-​ of dyads, determine the shortest path between
space is the shortest possible distance between the two in OP-​space, write a corresponding
representatives of those classes in O-​space—​and voice leading in musical notation, and calculate
in this case the shortest distance is dist(α, β′) the distance by each of the three distance
rather than dist(α, β). functions.
Path ② in Figure 12.5.4 transposes path ① by
tritone to the dyads AB and A♯C♯ corresponding
(a) CD and FA
(b) CD♯ and G♯C♯
(c)  F♯G♯ and AC♯
to γ and δ in Figure 12.5.1. Path ② mirrors path
① across the central axis of the Möbius strip; the
distance remains the same. Path ③ transposes
path ② up an additional semitone to A♯C and BD.
Exercise 12.5.6 In OP-​space, draw “spheres”
centered on the point DF♯, similar to the regions
Our standard picture of OP-​space places these in Figure 12.4.8, as follows: a sphere of radius
two dyads near opposite corners of the funda- 4 using the max distance, a sphere of radius
mental region, but they are joined by a short 5 using the Euclidean distance, and a sphere
path across the false boundary. In tracing voice of radius 5 using the voice-​leading distance.
leadings and calculating distances across a false Because these radii match the distances from
boundary, it is sometimes helpful to imagine the DF♯ to F♯A calculated above, the dyad F♯A
continuation of the figure beyond the boundary, should lie at a point of intersection of all three
as suggested by the extra copy of the dyad BD spheres.
   39
included at the top of Figure 12.5.4a. The por-
tion of OP-​space around BD falls in the lower-​ The examples in Figure 12.5.4 include situa-
left portion of the fundamental region, but it is tions in which the shortest path from one point
directly adjacent to the upper-​right portion, in to another crosses the false boundaries at the top
the relationship implied by the arrow to BD. The and bottom of the figure, but the mirror bound-
displacements and distances in path ③ are the aries at the sides of the Möbius strip do not come
same as those in paths ① and ②. into play in any of these examples. Straight-​line
Paths ④ and ⑤ involve a different pair of voice leadings may bounce off of these boundar-
dyads, DF♯ and F♯A. Both are relatively near the ies (recall Figure 11.2.8c, in which this happens
center of the fundamental region, and in the twice), but paths of this sort never need to be
figure the most obvious path joining them is considered in distance calculations with any of
the straight-​line path ④, which corresponds to the standard distance functions, because they
a voice leading in which D descends five semi- can always be replaced by shorter paths, or at
tones to A. But the alternative path ⑤ across a worst equally short paths, that do not include
false boundary must also be considered; in this such reflections. A bounce off a boundary indi-
voice leading both voices ascend. From a quick cates a voice crossing in pitch-​class space—​a
glance at paths ④ and ⑤ in (a) and the voice temporary doubling of pitch classes—​and the
leadings in (b), it is perhaps not obvious which most efficient path between two non-​ unison
of the two is shorter or more efficient. In fact, dyads never involves such a voice crossing.
this example illustrates a situation in which This point may be illustrated by comparing
the shortest path depends on which distance two voice leadings from D♯F to EG: voice lead-
function is chosen. The displacement multisets ing ① from Figure 12.5.1b, which includes a
are {0, 5} for path ④ and {3, 4} for path ⑤, from voice crossing, and voice leading ① from Figure
which it follows that path ⑤ is shorter as mea- 12.5.4b, which does not. Assuming that both
sured by the max distance but path ④ is shorter voices move continuously at constant speed,
by the voice-​ leading distance. The Euclidean the crossing in the first example is a momentary
distance gives exactly the same length (5) for unison at a pitch class between E and F, a point
both paths: path ⑤ traces the hypotenuse of a on the mirror boundary of the Möbius strip.

Distances • 511
The corresponding path in OP-​space, shown as a be considered, as must the 120-​degree twist
dashed arrow near the top left of Figure 12.5.4a, imparted to the prism when those two faces are
starts at D♯F, bounces off the boundary at the glued together.
point corresponding to the doubling, and con- Figure 12.5.9a shows two paths joining
tinues to EG. It is apparent geometrically, and the chromatic trichords G♯AB♭ and CC♯D. Both
the distance calculations confirm, that the path chords lie in the sum-​class-​3 plane, so one pos-
without the crossing (path ① in Figure 12.5.4a) is sible path, path ①, is a straight line within that
shorter, just as it is clear musically that the voice plane—​a zero-​sum voice leading. In this voice
leading without the crossing is more efficient. As leading, as shown in (b), G♯ descends by tri-
the following exercise shows, there are some sit- tone to D while the other two voices ascend by
uations in which the voice-​leading distance may three semitones. (In part (b) the second chord
be the same with or without a crossing, but it is is spelled B♯C♯D for ease of notation. If the voice
never shorter for the path with the crossing, and leading corresponding to path ① is not clear, you
for the other two distance functions the path may wish to consult Figure 11.3.1, the arrange-
with the crossing is always longer. ment of trichords within the sum-​class-​0 plane,
of which the sum-​class-​3 plane is a T1 transposi-
Exercise 12.5.7 For each of the following pairs tion.) Because CC♯D is a T4 transposition of G♯AB♭,
of dyads, write two voice leadings, one joining another possible path, path ②, follows a vertical
the two dyads with a voice crossing (which will line of transposition, crossing the false bound-
always take place at the dyad AA) and the other ary at the trichord BCC♯ in the sum-​class-​0 plane.
without a crossing. Draw the corresponding Path ② corresponds to the parallel voice leading
paths in OP-​space and calculate the lengths in which all three voices ascend four semitones.
of both paths according to each of the three The displacement multiset {4, 4, 4} for path ② is
distance functions. a more even distribution with the same sum as
path ①’s multiset {3, 3, 6}, so the two paths have
(a) GA♯ and G♯B the same length by the voice-​leading distance but
(b) GA and AB path ② is shorter by either the max or Euclidean
   ♯ and A♯B
(c) GG distance. The points in this example lie near the
corners of their sum-​class layer, so the distance
Exercise 12.5.8 Because permutational between them within that plane is large enough
equivalence introduces singularities into to make the zero-​sum voice leading inefficient in
quotient spaces in the form of mirror comparison with the parallel voice leading.
boundaries, we should expect that OP-​space, as
a distance space, fails to be homogeneous. This Exercise 12.5.10 Listed below are three
exercise confirms this expectation. pairs of T4-​related trichords. For each pair,
locate the chords in three-​voice OP-​space
(a) In OP-​space, what is the largest distance (the two always lie in the same sum-​class
from the unison dyad AA to any other dyad? plane). Draw a straight-​line zero-​sum voice
Answer the question for each of the three leading joining the two chords, and also
distance functions. a parallel voice leading through the false
(b) For each distance function, what is the boundary; write the corresponding voice
largest distance from CF♯ to any other dyad? leadings in musical notation. Using each
(c) Use the answers to parts (a) and (b) to of the three distance functions, determine
show that, using any of the three distance which path is shorter.41
functions, there can be no isometry of OP-​
space that maps AA to CF♯, and therefore OP-​ (a) DFA♭ and F♯AC
  space is not a homogeneous distance space.40 (b) FGA♭ and ABC
(c)  DDD and F♯F♯F♯
In three-​voice OP-​space, the Möbius prism
from Figure 11.3.4, considerations are similar to Major and minor triads inhabit three-​voice
those in the two-​voice OP-​space just discussed. OP-​space, so every distance function in this
The walls of the prism are mirror boundaries, space provides a measure of distance in con-
and the shortest path between two points never sonant triad space triad, supplementing the
reflects off a wall. The false boundaries at the several other triadic distance measures we have
top and bottom of the prism, however, must already studied. It would be possible to construct

512 • E x p l o rin g M u sica l S paces

FIGURE 12.5.9 (a) Two paths from G♯AB♭ to CC♯D in three-​voice OP-​space; (b) corresponding voice
leadings and distances

tables similar to Table 4.2.16 showing the max, F-​major, D-​minor, and C♯-​minor triads from
Euclidean, and voice-​leading distances between Figure 12.4.7, though conceived in cpitch3,
all possible pairs of triads, but we limit the dis- remain valid in OP-​ space because the voice
cussion here to a few general remarks and spe- leading shown there was already the most effi-
cific exercises. cient possible. Figure 12.5.11 shows the relative
The major and minor triads, together with positions of these triads in the chain of cubes as
the augmented triads, form the chain of hexa- a way of visualizing the distance calculations in
tonic triad cubes surrounding the central axis OP-​space.
of OP-​ space (Figure 11.3.4). Every edge in
each of these cubes signifies a single-​semitone Exercise 12.5.12
voice leading, a distance of length 1 by all three (a) In Exercise 10.2.3 you identified a non-​
measures, but the three functions give diver- equal-​tempered augmented triad lying at
gent values for distances between more widely the smallest possible Euclidean distance
separated triads. The calculations involving the from the C♯-​minor triad. What augmented

Distances • 513
FIGURE 12.5.11 Max, Euclidean, and voice-​leading distances between triads in OP-​space

triads (equal-​tempered or not) lie at the descriptions without complete proofs, generally
smallest possible max and voice-​leading focusing on the voice-​leading distance (usually
distances from C♯ minor? the most musically intuitive) and the Euclidean
(b) For the neo-​Riemannian distance, the distance (the most geometrically revealing). The
unique triad at the maximum distance from three-​voice spaces involving transpositional
C major is B♭ minor. What equal-​tempered equivalence were derived from the triangular
major or minor triad or triads lie at the grid of T-​space first exhibited in Figure 11.4.1b.
largest possible max, Euclidean, and voice-​ Voice-​leading distances are measured in this grid
  leading distances from C major?42 in a straightforward fashion, by counting units
parallel to the x, y, and z axes in whatever way
Exercise 12.5.13 We have now identified many
yields the smallest total. The T-​class (4, 2, 5)T,
different distance functions on consonant triad
first used as an illustration in Figure 11.4.1c, is
space triad, including the following:
shown again in Figure 12.5.14a. The voice-​lead-
• The circle-​of-​fifths, Weber, and neo-​ ing distance from the origin (0, 0, 0)T to this
Riemannian distances (Figure 12.3.1); point is 3 semitones, because the shortest path
• The neo-​Riemannian voice-​leading distance from (0, 0, 0)T to (4, 2, 5)T combines two semi-
(Exercise 12.3.3); tones of motion in the negative y direction with
• The cube distance (Exercise 12.3.4); one semitone in the positive z direction. The
• The common-​tone distance (Exercise 12.3.6); components of this motion match the compo-
• The LR distance (Exercise 12.3.11); nents of [0, −2, 1], the T normal form of (4, 2,
• The max, Euclidean, and voice-​leading 5), which was also called the yz form in Chapter
distances inherited from OP-​space.   11. The voice-​leading distance from the origin to
any point can always be read directly from the
The above list (which excludes some of the xy form, the xz form, or the yz form, whichever
variant double-​circle space distances considered gives the smallest sum of displacements.
in Section 12.3) appears to contain ten distance Euclidean distances may be measured with
functions, but only nine of them are different. a ruler in the plane of T-​space, but with two
Which two are the same function on triad? caveats. First, the triangular grid does not lend
itself to distance calculations as easily as a tra-
For the remaining OPTIC spaces consid- ditional rectangular grid. The Euclidean distance
ered here, we limit our discussion to brief formula [(Δx)2 +​(Δy)2]1 ⁄ 2 is applicable only if the

514 • E x p l o rin g M u sica l S paces

 to one side of the triangle, which is 3 4 ≈ 0.866.
Distances between grid points can be figured by
combining v-​units and u-​units in two perpen-
dicular directions, to which the usual Euclidean
distance formula is then applicable. Figure
12.5.14c shows that one can move from the ori-
gin to (4, 2, 5)T by moving two v-​units along the
positive z axis in combination with two u-​units
in a perpendicular direction, yielding a distance
of [(2v)2 +​(2u)2]1 ⁄ 2. If we substitute v =​1 and u =​
0.866, this formula yields a Euclidean distance of
approximately 2.65 v-​units (the exact value is 7
v-​units) from (0, 0, 0)T to (4, 2, 5)T.
There is another complication. It may come
as a surprise to realize that the v-​unit, the dis-
tance in T-​space from (0, 0, 0)T to (0, 0, 1)T, is not
equal to one semitone according to the quotient
distance formula for the Euclidean metric on the
quotient space cpitch3/​~T. Certainly distEuc((0,
0, 0), (0, 0, 1)) =​1 in cpitch3, but in T-​space,
distEuc((0, 0, 0)T, (0, 0, 1)T) is the smallest pos-
sible Euclidean distance in cpitch3 between (0,
0, 0) and α, where α is any pitch triple of the
form (x, x, x +​1), transpositionally equivalent to
(0, 0, 1). This minimal distance turns out to be
v =​ 2 3 ≈ 0.816, attained when α is (−1 3 , −1 3 , 2 3 ).
A semitone of motion in one voice brings about
motion through a distance of only v ≈ 0.816
semitones in T-​space, not one whole semitone.
(Recall from Section 11.4 that three-​voice T-​
space is obtained by projecting cpitch3 onto a
sloping plane perpendicular to the vector (1, 1,
1). The projections of the points (0, 0, 0) and (0,
0, 1) onto this plane are slightly closer together
than the original points are.) The shorter u-​unit
is therefore u = 3 4 ⋅ 2 3 = 1 2 ≈ 0.707. A proper
ruler for Euclidean measurements in the plane of
T-​space is one that shows 0.816 units between
adjacent grid points. By this measure, distEuc((0,
0, 0)T, (4, 2, 5)T) is not 2.65, as suggested above,
FIGURE 12.5.14 (a) Calculation of a voice-​lead- but that value multiplied by v ≈ 0.816, or approx-
ing distance in three-​voice T-​space; (b) the units imately 2.16.
v and u for measuring Euclidean distance in T-​ As a practical matter, the length of the v-​unit
space; (c) calculation of a Euclidean distance in is simply a scale factor for the Euclidean distance
three-​voice T-​space function. We may regard the v-​unit as a “voice-​
leading unit” for Euclidean T-​space, a convenient
measure because it corresponds to a motion
displacements Δx and Δy are perpendicular, so we through a single semitone in a single voice.
cannot use the displacements 2 and 1 in Figure When studying Euclidean relationships inter-
12.5.14a to calculate distEuc((0, 0, 0)T, (4, 2, 5)T). nal to T-​space (or any space assuming T equiva-
Suppose the Euclidean distance between adja- lence), there is often little harm in measuring
cent grid points such as (0, 0, 0)T and (0, 0, 1)T distances in v-​units, effectively assuming that
is labeled v, as in Figure 12.5.14b. The spacing u v =​1 rather than the theoretically correct
between parallel gridlines is shorter than v, by smaller value. When comparing distances in T-​
the ratio of the altitude of an equilateral triangle space to distances in cpitch3, however, or when

Distances • 515
comparing the numerical size of Euclidean dis- voice-​leading distance. A sphere of this kind is
tances to voice-​ leading distances, the scaled neither a circle nor a square; what is it? Because
values reflecting the correct size of the v-​units the radius matches the voice-​leading distance
should be used: 1 v-​unit ≈ 0.816 semitones in the from the origin to (2, 4, 5)T calculated above,
Euclidean metric on the quotient space. the point (2, 4, 5)T should lie on the boundary
of  
the region.
Exercise 12.5.15
(a) An alternative path from (0, 0, 0)T to (4, Exercise 12.5.19 The max distance was not
2, 5)T in Figure 12.5.14c would move 21 2 included in the above discussion of distances
v-​units along the negative y axis combined in T-​space. Determine distMax((0, 0, 0)T, (0, 0,
with one u-​unit in a perpendicular 1)T) and distMax((0, 0, 0)T, (4, 2, 5)T). In T-​space
direction. Verify that this calculation gives there is a simple relationship between the max
the same result as above for the Euclidean distance and the voice-​leading distance. Can you
distance (about 2.65 v-​units, or 2.16 see  what this relationship is, and why?44
semitones).
(b) Because distEuc((0, 0, 0)T, (4, 2, 5)T) ≈ 2.16, The features of distances in T-​space carry
there must be a pitch triple α in cpitch3, over to quotient spaces formed from T-​space.
transpositionally equivalent to (4, 2, 5), False boundaries must be considered when
such that distEuc((0, 0, 0), α) ≈ 2.16. Can you they are present, but of the several quotients
  find this α?43 of three-​voice T-​ space examined in Section
11.4, only OPT-​space (Figure 11.4.21) has false
Exercise 12.5.16 Determine the voice-​leading boundaries. PT-​space (Figure 11.4.4), PTI-​space
and Euclidean distances between each of the (11.4.7), and OPTI-​space (11.4.12) have mirror
following pairs of points in T-​space. Give the boundaries only, which do not come into play
Euclidean distances both in v-​units and in in distance calculations. Measuring distances
semitones. in these spaces is therefore no more difficult
than in T-​ space—​ perhaps easier, because of
(a) (0, 0, 0)T and (0, 4, 7)T (a triple unison and a the smaller sizes of the spaces. Figure 12.5.20
major triad) locates the multiset classes 002 and 025 in
(b) (4, 2, 5)T and (0, 4, 7)T OPTI-​space. The displacements along the grid-
(c) (4, 2, 5)T and (2, 4, 5)T (recall Figure lines, 2 and 1, are the same as between (0, 0, 0)T
  11.4.5b) and (4, 2, 5)T in Figure 12.5.14a, so the distance
calculations will give the same results: in OPTI-​
Exercise 12.5.17 Determine the size of the space, distVL(002, 025) =​3 and distEuc(002, 025)
v-​unit in two-​voice T-​space. That is, calculate =​ [(2v)2 +​(2u)2]1 ⁄ 2, about 2.65 v-​units or 2.16
dist
  Euc((0, 0)T, (0, 1)T) in cpitch2/​~T. semitones.
Exercise 12.5.18 In three-​voice T-​space, draw Exercise 12.5.21 Construct minimal voice
a “sphere” of radius 3 centered on the origin, leadings to confirm the above distance
similar to the regions in Figure 12.4.8, using the calculations. That is:

FIGURE 12.5.20 Voice-​leading and Euclidean distances between multiset classes in three-​voice
OPTI-​space

516 • E x p l o rin g M u sica l S paces

(a) Identify two pitch triples representing multiset 000 is 8 semitones by the voice-​leading
OPTI-​classes 002 and 025, separated in distance or 5.66 semitones by the Euclidean
cpitch3 by a voice-​leading distance of 3 distance. (The Euclidean eccentricity is the
semitones. radius of the smallest circle about the given
(b) Identify two pitch triples representing point that encompasses the entire space.) The
OPTI-​classes 002 and 025, separated point 048 has the same eccentricity as 000
in cpitch3 by a Euclidean distance of by either measure, but these are points of
  approximately 2.16 semitones.45 maximum eccentricity.

Exercise 12.5.22 Locate the inversionally (a) Determine the voice-​leading eccentricity of
related OPT-​classes 016 and 056 in three-​ the point 014. Some other points share this
voice OPT-​space (Figure 11.4.21a). Although same value of the voice-​leading eccentricity;
these points appear widely separated in the what are they? Do any points have a smaller
fundamental region, they are separated only by voice-​leading eccentricity than this?
a voice-​leading distance of one semitone and (b) Determine the Euclidean eccentricity of 014.
a Euclidean distance of one v-​unit. Construct One equal-​tempered trichord has a smaller
minimal voice leadings to confirm these   Euclidean eccentricity than 014; what is it?
distances.
  
Some sense of how distances vary from one
OPTI-​space is bounded, and distances in this space to another may be gleaned from Table
space are limited in size. The most distant points 12.5.24. The bass clarinet line in mm. 4–​ 6
in Figure 12.5.20 are 000 and 048. The voice-​ of “Nacht” from Schoenberg’s Pierrot lunaire
leading distance between them is eight semi- begins with the ordered trichord α =​(E3, G3,
tones; the Euclidean distance (the length of the E♭3) and ends with β =​(B♭2, A2, G♭3). (These are
diagonal line joining them) is eight u-​units, the two of the eight trichords whose normal forms
equivalent of 48 ≈ 6.93 v-​units or 32 ≈ 5.66 were studied in Exercise 10.4.6.) The table
semitones. The straight-​line voice leading (0, 0, shows the distance between α and β in cpitch3
0) → (−4, 0, 4) in cpitch3 has both a voice-​lead- and in seven of the most important OPTIC
ing length of 8 and a Euclidean length of 32 . spaces, using all three of the standard distance
These observations provide a more rigorous con- functions. The displacement multiset that
firmation of an intuition we first formulated in produces each distance is also shown. Several
conjunction with Figure 12.1.1. For a multiset things are apparent in this table:
class such as 014 near the middle of OPTI-​space,
there are no other points in the space at dis- • As more equivalence relations are added and
tances as large as a voice-​leading distance of 8 or spaces shrink, distances generally decrease.
a Euclidean distance of 5.66. There can therefore • Trichords α and β share the prime form 014,
be no isometry from OPTI-​space to itself that so they are the same point in OPTI-​space,
maps 000 to 014; it follows that OPTI-​space is separated by zero distance. They are different
not a homogeneous distance space (which is as points in every other space listed, however,
we should expect of any P+​ or I+​ space). including OPT-​space, because as pitch-​class
More can be said. Because the triangle of sets they are related by inversion.
OPTI-​space is not symmetrical, there can be no • The distances in PT-​ and PTI-​space are the same,
isometry from OPTI-​space to itself at all except- because α and β lie in the same half-​region of
ing the trivial case of the identity mapping. PT-​space.
Points in OPTI-​space are not interchangeable • In the first three rows of the table, the
as far as their distances from other points are same voice leading is minimal for all three
concerned (in the way that points in an interval distance functions, so the same displacement
space must be): every multiset class is unique in multiset appears for all three. This is always
the distance relations that it forms with other true in cpitch3 and in O-​space, and often in
points in the space. OP-​space (the dyads DF♯ and F♯A in Figure
12.5.4 were an exception). In T+​ spaces,
Exercise 12.5.23 Define the eccentricity of a different minimal voice leadings for the
point in OPTI-​space to be the greatest distance three distance functions are common.
from that point to any other point.46 The above • The last five rows should confirm your
discussion shows that the eccentricity of the observation from Exercise 12.5.19 about a

Distances • 517
Table 12.5.24. Max, Euclidean, and voice-​leading distances between two trichords
from Schoenberg’s “Nacht” in selected OPTIC spaces

S PAC E N OR M A L F OR M S DI S TA N C E S A ND A S S O C I AT E D
DI S PL AC E ME N T MU LT I S E T S

α β distMax( α , β ) distEuc( α , β ) distVL( α , β )

cpitch3 (−8, −5, −9) (−14, −15, −6) 10 145 ≈ 12.04 19

=​ (E3, G3, E♭3) =​ (B♭2, A2, G♭3) {3, 6, 10} {3, 6, 10} {3, 6, 10}
cpitch3/​~O [4, 7, 3]O [10, 9, 6]O 6 49 =​ 7 11
{2, 3, 6} {2, 3, 6} {2, 3, 6}
cpitch3/​~OP [3, 4, 7]OP [6, 9, 10]OP 5 33 ≈ 5.74 9
=​E♭EG =​G♭AB♭ {2, 2, 5} {2, 2, 5} {2, 2, 5}
cpitch3/​~T [0, 3, −1]T [0, −1, 8]T 61 2 798 3 ≈ 9.42 13
{21 2, 61 2 , 61 2} {12 3 , 52 3 , 71 3 } {0, 4, 9}

cpitch3/​~PT [0, 1, 4]PT [0, 1, 9]PT 21 2 150 3 ≈ 4.08 5

{21 2 , 21 2, 21 2 } {12 3 , 12 3 , 31 3} {0, 0, 5}
cpitch3/​~PTI [0, 1, 4]PTI [0, 1, 9]PTI 21 2 150 3 ≈ 4.08 5
{21 2 , 21 2, 21 2 } {12 3 , 12 3 , 31 3} {0, 0, 5}
cpitch3/​~OPT [0, 1, 4]OPT [0, 3, 4]OPT 1 24 3 ≈ 1.63 2
{1, 1, 1} {2 3, 2 3, 11 3 } {0, 0, 2}
cpitch3/​~OPTI [0, 1, 4]OPTI [0, 1, 4]OPTI 0 0 0
{0, 0, 0} {0, 0, 0} {0, 0, 0}

consistent relationship between the max of one cubic semitone in OP-​space. Three-​voice
and voice-​leading distances, which holds in OPTI-​space (Figure 11.4.12 or 12.5.20) is a right
all T+​ spaces. triangle with legs 6v and 4u; its area is therefore
• The three distance functions give different (6v)(4u)/​2 =​ 12uv =​ 12 ⋅ 1 2 ⋅ 2 3 = 48 ≈ 6.93
values, but the three remain loosely square semitones. The exercises below show some
proportional from one row of the table to ways in which areas and volumes can be used to
another, confirming our observation in Section compare the sizes of different portions of a space.
12.4 that the three are “not very different.” Awareness of areas and volumes can also
give us a fresh perspective on symmetry. In any
Exercise 12.5.25 OPTIC space in two or more voices, symmetrical
(a) Construct voice leadings to confirm the chords form lines or planes of a smaller dimen-
distance calculations in Table 12.5.24. sionality than the space, and therefore a set
(b) Construct a similar table for the trichords α whose area or volume, as measured in the larger
and β in Saariaho’s Vers le blanc from Figure space, is zero. Recall, for example, the axes of
  11.3.14 and several exercises in Chapter 11. symmetry in T-​space shown in Figure 11.4.11:
many such axes crisscross the plane, but they
Euclidean distances can be used as the basis for are all lines with no area. Symmetrical chords fill
calculating areas and volumes in OPTIC spaces, zero percent of the area of any OPTIC space; if
measured in the possibly odd-​sounding units of we choose a point in an OPTIC space at random,
square semitones and cubic semitones. The musi- the probability that we will choose a symmetrical
cal significance of such a calculation may seem set is zero.47 In this regard our experience with
obscure, but some areas and volumes are fairly equal-​tempered chords in the mod-​12 world is
intuitive; for example, a hexatonic triad cube highly misleading: five of the twelve trichordal
(Figure 10.1.1, or any of the cubes surrounding set classes, or 42%, are symmetrical in some way,
the central axis in Figure 11.3.4) has a volume as are twelve of the nineteen trichordal multiset

518 • E x p l o rin g M u sica l S paces

classes, or 63% (remember that multisets with is represented by six different points in this
doublings are P-​symmetric). In the wider uni- cross section (three related by transposition
verse of continuous spaces, symmetry is statisti- and three others by inversion). We should
cally a rare phenomenon. therefore expect the area of the cross section
Near-​symmetry, on the other hand, is very to be exactly six times the area of OPTI-​
common. In three-​ voice OPTI-​space (Figure space. Verify that this proportion is correct.
12.5.20), the symmetrical multiset classes are (c) The height of the prism of three-​voice OP-​
those on the boundaries of the triangle. Among space (the vertical dimension of Figure
the equal-​ tempered multiset classes not on 11.3.4) is the same as the Euclidean length
the boundary, 026 lies at a Euclidean distance of path ② in Figure 12.5.9. By multiplying
of one v-​unit (0.816 semitones) from two dif- this height by the area you calculated in (a),
ferent points on the boundary (027 and 036), determine the volume of OP-​space.
and every other multiset class lies closer than (d) What percentage of OP-​space consists of
that to some boundary point. Even among the trichords lying at a Euclidean distance of 1
non-​equal-​tempered multiset classes, the large or less from some augmented triad? What if
majority lie within one v-​unit of some boundary the distance is 2 or less?50
point—​that is, of some symmetrical multiset (e) Three-​voice O-​space is a 3-​torus whose
class (see Exercise 12.5.26c below). fundamental region is a cube of volume
123 =​1,728 cubic semitones (one cubic
Exercise 12.5.26 octave). Except for symmetrical sets, every
(a) The trichords containing two pitch classes OP-​class is represented by six different
separated by a pitch-​class interval of one permutations in this cube, so we should
semitone or less form a strip along the bottom expect the volume of OP-​space to be exactly
edge of three-​voice OPTI-​space (Figure one-​sixth of the volume of O-​space. Is this
12.5.20), bounded at the top by the horizontal   expectation correct?
line through 012 and 016. Calculate the
area of this strip. What percentage of the Volumes can be calculated in higher dimen-
total area of OPTI-​space is contained in this sions also. The following exercises develop
strip? This percentage may be regarded as the analogs in four-​voice spaces of some of the prop-
probability that a randomly chosen trichord erties described above in lower dimensions. It
in OPTI-​space will contain two notes within a is instructive to note the wide variation in the
semitone of each other. magnitudes of the distances, areas, and volumes
(b) What percentage of the twelve equal-​ calculated here: some of these spaces are much
tempered trichordal set classes contain an larger than others.
interval of a semitone? What percentage
of the nineteen multiset classes contain an
interval of size 0 or 1? Exercise 12.5.28
(c) The points of OPTI-​space lying at a (a) Show that the length of the v-​unit in four-​
Euclidean distance greater than one v-​unit voice T-​space (the Euclidean distance from
from the nearest point on the boundary of (0, 0, 0, 0)T to (0, 0, 0, 1)T) is v =​ 3 4 ≈ 0.866
the space form a small triangle in the center semitones. (When distances in four-​voice
of the space. This triangle is similar (that is, T-​space were mentioned informally in
proportional) to the triangle formed by the Section 11.6, the implicit units of measure
whole space; its sides lie parallel to the sides were v-​units, not semitones. All calculations
of the space, at a distance of one v-​unit, below are in semitones.)
with 026 at its upper-​right corner. Show (b) Recall that the v-​unit in three voices is 2 3 ;
that this triangle contains less than 5% of you calculated the v-​unit in two voices in
  the area of OPTI-​space.48 Exercise 12.5.17. Can you deduce a general
formula for the v-​unit in n voices?
Exercise 12.5.27 (c) Show that distEuc((0, 0, 0, 1)T, (0, 1, 1, 1)T) is
(a) Calculate the area of the sum-​class-​0 cross exactly one semitone.
section of three-​voice OP-​space shown in (d) Consider layer 0 of the fundamental region
Figure 11.3.1.49 for four-​voice OPT-​space from Figure
(b) With the exception of symmetrical multisets 11.6.4, specifically the small triangle at the
(which contribute no area), every OPTI-​class lower-​left corner with vertices 0000, 0001,

Distances • 519
 and 0111. Use trigonometry together with (d) With the exception of symmetrical
the results of parts (a) and (c) to show that multisets, every four-​voice OPTI-​class
the angle at 0000 is approximately 70.5 is represented by how many different
degrees, and calculate the other two angles points in this sum-​class-​0 cross section?
in that triangle.51 Verify that the ratio of the volume of
(e) Show that the distance between horizontal OPTI-​space to the volume of the sum-​
rows of lattice points in layer 0 (or any class-​0 tetrahedron is consistent with this
other layer) of Figure 11.6.4b is d =​ 2 3 ≈ observation.
0.816 semitones. (e) As was shown in Figure 11.5.9, the Möbius
(f) Use the above values of v and d to show that hyperprism of four-​voice OP-​space is
the area of layer 0 is 36 ⋅ 2 ≈ 50.9 square obtained by extruding the sum-​class-​0
semitones. tetrahedron through a fourth dimension.
(g) The tetrachord 0246 in layer 2 of the Show that the length of the hyperprism
tetrahedron lies directly above 0336 in in this fourth dimension is exactly
layer 0. Calculate the Euclidean distance six semitones, and therefore that the
between these two points, and use it to four-​dimensional volume of four-​voice
show that the spacing between layers is OP-​space is exactly 864 (measured in
1 ≈ 0.707 semitones. semitones-​to-​the-​fourth-​power).53
2
(h) If a tetrahedron has a triangle of area (f) Four-​voice O-​space is a 4-​torus whose
A as its base and if its height measured fundamental region is a 4-​cube of volume
perpendicular to the base is h, then its 124 =​20,736. Except for symmetrical
volume is Ah/​3. Show that the volume of multisets, every OP-​class is represented
four-​voice OPT-​space is exactly 36 cubic by how many permutations in this cube?
semitones, and therefore that the volume Verify that the ratio of the volume of OP-​
of four-​voice OPTI-​space is exactly 18 cubic space to the volume of O-​space is consistent
semitones.   with this observation.
(i) Show that the probability that a randomly
chosen tetrachord in four-​voice OPTI-​space In calculating distances as in many other
contains two notes separated by a semitone respects, cardinality equivalence presents chal-
  or less is slightly more than 70%. lenges unlike any of the other OPTIC relations.
A consideration of the voice-​leading distance
between the A♭7 and c♯ø7 chords in OP-​space
Exercise 12.5.29 and OPC-​space illustrates some of the compli-
(a) Figure 11.5.4 showed the sum-​class-​0 cations involved. Figure 12.5.30a locates these
tetrahedron in four-​voice OP-​space. Show two chords in the octatonic hypercube chain
that four edges of this tetrahedron (such that surrounds the central axis of OP-​space
as the edge CCCC–​E♭E♭E♭E♭) have Euclidean (from Figures 11.5.3 and 11.5.9). A path join-
length 6 ⋅ 3 ≈ 10.4 semitones while the ing the two chords in a series of six single-​semi-
other two (such as CCCC–​F♯F♯F♯F♯) are tone voice-​leading moves is highlighted. Many
of length 12 semitones. If line segments other paths of length 6 can be found in the fig-
AAAA–​E♭E♭AA and E♭E♭AA–​E♭E♭E♭E♭ are ure, but they all correspond to the voice leading
drawn to bisect two of the tetrahedron’s shown in Figure 12.5.30b, in which two voices
faces, show that each of these two descend by one semitone and two others by two
segments is of length 6 ⋅ 2 ≈ 8.49 semitones; the various paths differ only in the
semitones.52 order in which the voices are assumed to move.
(b) Use the lengths calculated in (a) to show If the voices moved simultaneously, a straight-​
that the bisecting line segments form a line path would be traced; that path would be
right angle at E♭E♭AA. If the base of the shorter as measured by Euclidean distance but
tetrahedron lies in a horizontal plane, its length by the voice-​leading distance would
therefore, E♭E♭E♭E♭ lies directly above E♭E♭AA. remain 6. In fact, by the voice-​leading distance
Where else in the tetrahedron do two faces there can be no shorter path: a single-​semi-
meet at a right angle? tone move always changes the sum class of a
(c) Show that the volume of the sum-​ chord by 1, and the A♭7 chord is of sum class
class-​0 tetrahedron is exactly 144 cubic 5 while c♯ø7 is of sum class 11, so a minimum
semitones. of six moves are required to move from one to

520 • E x p l o rin g M u sica l S paces

FIGURE 12.5.30 The distance from A♭7 to c♯ø7 in OP-​space and OPC-​space: (a) a path of voice-​leading
length 6 in four-​voice OP-​space; (b) a corresponding voice leading; (c) a voice leading of apparent length
5; (d) with split/​fuse relations in OPC-​space; (e) in five-​voice OP-​space

the other. In four-​voice OP-​space, therefore, Apparently, then, while the voice-​leading dis-
distVL(A♭7, c♯ø7) =​6. tance from A♭7 to c♯ø7 in OP-​space is 6, in OPC-​
space it is only 5.
Exercise 12.5.31 How many different paths But which OPC-​space? A voice leading such as
of length 6 join A♭7 and c♯ø7 in the octatonic that in (c) or (d) cannot be modeled adequately
hypercube
   chain?54 in cpitch4/​~OPC, a quotient space formed from a
space cpitch4/​~OP of four-​voice chords, because
If the voice-​leading distance between A♭7 and the complete voice leading cannot be realized in
c♯ø7 is 6, then what are we to make of the voice four voices: the voice created by fusion is not the
leading in Figure 12.5.30c, in which the two same one that subsequently splits. To realize the
chords are linked by only five single-​semitone voice leading in a space of a fixed dimension, five
voice-​leading moves? The trick is that this is not voices are required, as shown in (e). Indeed, in
a voice leading in four-​voice OP-​space. As (d) five-​voice OP-​space, the voice-​leading distance
shows in more detail, this voice leading shuttles between the multisets CCE♭G♭A♭ and BC♯EGG is
between four-​voice and three-​voice spaces, two 5, not 6. In the curious world of C-​space, it seems
Gs fusing into a single voice at the second chord that the shortest path between two chords in
and a single C splitting in two later on. The fuse four-​dimensional space does not lie in that space:
and split moves relate chords in spaces of differ- it must pass through either a three-​dimensional
ent dimensions, but the chords so related are C-​ or a five-​dimensional space along the way.
equivalent (that is, CE♭GG ~C CE♭G and CEG ~C
CCEG). The fuse and split do not correspond to Exercise 12.5.32 Explain this apparent
any motion at all in OPC-​space and contribute paradox in Figure 12.5.30c: three of the five
nothing to the length of a path in that space. single-​semitone voice-​leading moves ascend

Distances • 521
while only two descend, yet the last chord in through the judicious deployment of one fuse
the progression is lower in register than the and one split, the need for a large displacement
first. Does the voice leading in (e) resolve this like six semitones is avoided altogether.
paradox?
   The tuples in this example are labeled γ1, δ1,
γ2, δ2, γ3, and δ3; they are of cardinalities 3, 3, 2,
Exercise 12.5.33 Determine the voice-​leading 2, 3, and 3, respectively. For each i (i =​1, 2, or
distance between the trichords CDF♯ and D♭FG 3), γi and δi are tuples of the same cardinality,
in three-​voice OP-​space. Show that under the so the voice-​leading distances between them—​
assumption of C equivalence, a shorter path 1, 2, and 1—​may be computed normally. The
may be obtained, passing through either a two-​ links from δ1 to γ2 and from δ2 to γ3 are the fuse
or  
a four-​voice space. and split; these links join C-​equivalent tuples of
different cardinalities, so there is no distance
An adequate understanding of cardinality between them in C-​space. The first tuple γ1 is the
equivalence requires that we work not in a space same as α—​that is, γ1 is a pitch triple belonging
of a fixed dimensionality but in a space such as to the class αC —​while the last tuple δ3 is the
cpitch*, the union of the spaces cpitchn for all same as β (belongs to βC). The total length of the
n. For any given example, a space of the form voice leading in C-​space is thus only four semi-
cpitch≤n consisting only of spaces up through tones, which is indeed the voice-​leading distance
a certain finite dimension n will suffice, but between αC and βC in C-​space: distVL(αC, βC) =​4.
we have just seen that the limiting dimension This example shows that the quotient distance
n may need to be larger than the number of formula presented earlier in this section does
apparent voices in the example, so for simplic- not work for cardinality equivalence. That for-
ity we refer to the all-​embracing space cpitch*. mula would require us to find tuples α0 and β0 in
We will work directly with pitch tuples for the cpitch*, in the C-​classes αC and βC respectively,
time being, thereby disregarding complications for which distVL(α0, β0), computed normally
due to octave and permutational equivalence. within some cpitchn, is as small as possible. But
In the following discussion, the name C-​space there are no such α0 and β0, of any cardinality, for
refers to cpitch*/​~C, the quotient space of which distVL(α0, β0) is as small as 4. The best we
cpitch* under the assumption of cardinality can do is distVL(α0, β0) =​6, for the tuples α0 =​(0,
equivalence. 1, 6, 6) and β0 =​(2, 2, 7, 8) in cpitch4.
To examine the behavior of distances in C-​ The calculation in Figure 12.5.34 suggests a
space more closely, consider the pitch triples α =​ chain technique by which distances in C-​space may
(C4, D♭4, G♭4) =​(0, 1, 6) and β =​(D4, G4, A♭4) =​(2, 7, be defined in general. To determine distVL(αC, βC)
8). Both α and β belong to cpitch3, but we con- for any C-​classes αC and βC, we must examine
sider them now as equivalence classes αC and βC chains of pitch tuples γ1, δ1, γ2, δ2, …, γk, δk with
in C-​space; for example, αC includes not only the the following properties:
original triple (0, 1, 6) but also larger tuples such
as (0, 1, 6, 6) or (0, 0, 0, 1, 6, 6, 6, 6). In cpitch3, • γ1 is a representative of the C-​class αC;
the displacement multiset between α and β is {2, • δk is a representative of the C-​class βC;
2, 6}, and the voice-​leading distance is therefore • for each i =​1, 2, …, k, the pitch tuples γi and δi
10 semitones. Figure 12.5.34 shows, in a fash- are of the same cardinality;
ion similar to Figure 12.5.30d, that the distance • for each i =​1, 2, …, k − 1, the pitch tuples δi
between αC and βC in C-​space is much smaller: and γi +​1 are C-​equivalent.

FIGURE 12.5.34 A voice-​leading distance in C-​space

522 • E x p l o rin g M u sica l S paces

We then form the sum distVL(γ1, δ1) +​distVL(γ2, (a) Using techniques similar to those illustrated
δ2) +​ ⋯ +​distVL(γk, δk), where each segmental dis- in Figure 11.3.5, locate αOP and βOP in the
tance distVL(γi, δi) is calculated in the appropriate Möbius prism of three-​voice OP-​space.
pitch-​tuple space cpitchn. The distance distVL(αC, (b) The voice leading C → D, D♭ → G, G♭ → A♭
βC) in C-​space is the smallest possible sum of this is not the shortest voice leading joining
form, for any such chain of intermediate points. αOP and βOP in OP-​space. Identify the
A few comments on this definition are in order. shortest voice leading (it does not cross
Each segmental distance distVL(γi, δi) is a straight- the false boundary) and use it to show that
forward voice-​leading distance in some n-​dimen- distVL(αOP, βOP) =​6, distEuc(αOP, βOP) =​ 18,
sional space, but successive segments may lie in and distMax(αOP, βOP) =​4.
spaces of different dimensions. Because δi and (c) In OPC-​space, the voice leading in
γi +​ 1 are the same point in C-​space, the distance Figure 12.5.34 is possible. Using a
between them is 0, so distances distVL(δi, γi +​ 1) do straightforward shorthand, this voice
not affect the sum: these links in the chain—​the leading may be written
0s in Figure 12.5.34, the splits and fuses, worm-
holes from one dimension to another—​come for (0, 1, 6) → (1, 1, 6) ⇢ (1, 6) → (2, 7)
free. The definition does not require that α and ⇢ (2, 7, 7) → (2, 7, 8),
β be tuples of the same cardinality; tuples of all
sizes determine C-​classes which can always be where the dashed arrows indicate the split
linked by appropriately devised chains, so the and fuse, which link C-​equivalent points and
chain technique enables a distance to be calcu- therefore cover no distance in C-​space. Trace
lated in C-​space between any two tuples at all. the segments of this path in three-​voice and
A voice-​leading distance calculated in this way two-​voice OP-​space as appropriate. The total
measures the length of a voice leading of the kind length of this path in C-​space by the voice-​
illustrated in Figure 12.5.30d or 12.5.34, but it leading distance is 4. Show that the length
does not necessarily correspond to a voice lead- is 3 by the max distance and approximately
ing in a fixed number of voices. 3.41 by the Euclidean distance.
The above definition applies to C-​space as the
quotient of cpitch* under cardinality equiva- (d) The voice-​leading distance between αOPC and
lence, but the same principles apply when other βOPC is indeed 4, but the path in (c) is not
OPTIC relations are present. To calculate a dis- the only way to achieve it. Moreover, that
tance in OPC-​space cpitch*/​~OPC, for instance, path is not minimal for either of the other
we consider chains of points in OP-​spaces, pos- distance functions. Repeat the calculations
sibly of various cardinalities, measuring each for the path
segmental distance distVL(γi, δi) within the
appropriate OP-​space cpitchn/​~OP. The above (0, 1, 6) → (1, 1, 7) ⇢ (1, 7) ⇢ (1, 7, 7)
discussion assumes that we are working with → (2, 7, 8),
the voice-​leading distance—​probably the most
intuitive way to understand distances in C-​ in which the split and fuse occur at the same
space—​but the chain technique may be applied point [1, 7]OPC of OPC-​space. This path turns
with either the Euclidean or the max distance, out to be minimal for all three distance
the segmental distances dist(γi, δi) being calcu- functions; your calculations should show
lated in the manner appropriate for the chosen that distVL(αOPC, βOPC) =​4, distEuc(αOPC, βOPC)
distance function. For the Euclidean distance,   ≈ 2.83, and distMax(αOPC, βOPC) =​2.
square roots may be involved in the calculation
of the segmental distances distEuc(γi, δi), but not Exercise 12.5.36 Figure 12.5.9 showed paths ①
in the final sum—​the linking distances are sim- and ② joining the trichords [8, 9, 10]OP =​G♯AB♭
ply added. The exercises below illustrate several and [0, 1, 2]OP =​CC♯D in OP-​space. Because
of these points. both points lie near the boundaries of the prism
where trichords with pitch-​class doublings may
Exercise 12.5.35 The example above involved be found, shorter paths are possible in OPC-​
the pitch triples α =​(0, 1, 6) and β =​(2, 7, 8). space. Consider the following paths:
Consider the corresponding trichords in three-​
voice OP-​space, αOP =​[0, 1, 6]OP =​CD♭G♭ and Path ③: (8, 9, 10) → (81 2 , 81 2 , 10) ⇢ (81 2 , 10)
βOP =​[2, 7, 8]OP =​DGA♭. → (0, 11 2 ) ⇢ (0, 11 2 , 11 2 ) → (0, 1, 2)

Distances • 523
 Path ④: (8, 9, 10) → (81 2 , 81 2 , 101 2 ) other pitch triples separated by the same interval
⇢ (81 2 , 101 2 ) → (111 2 , 11 2 ) are not.
5. This general conception of intervals as equiva-
⇢ (111 2 , 11 2 , 11 2 ) → (0, 1, 2)
lence classes was suggested by Tymoczko (2009a, 241).
Path ⑤: (8, 9, 10) → (9, 9, 9) ⇢ (9)
6. See, for example, Cohn 2001, Leong 2007, and
→ (1) ⇢ (1, 1, 1) → (0, 1, 2)
Murphy 2009.
7. Hint for Exercise 12.2.3: In cpitch, the possibili-
One of these three paths is minimal for the
ties depend only on whether the intervals from x to y
voice-​leading distance, another for the max
and from y to z are ascending or descending. In other
distance. Trace each of these paths in the
distance spaces the number of possibilities may be
appropriate OP-​spaces. Calculate the length of
larger, but the triangle inequality implies a numerical
each path by both the voice-​leading and max
range, valid in all cases (consider permuting the vari-
distance functions, and compare the results
ables in the statement of the inequality).
   the values for paths ① and ②.55
with
8. Hint for Exercise 12.2.4: The proof of the triangle
inequality in the general case relies on the property
|a +​ b| ≤ |a| +​|b| noted above.
9. Tymoczko (2009a, 245–​46) defines a Lewinian
NOTES interval system to be a space on which both an interval
1. Forte (1973, 14) discusses directed and undi- function and a distance function are defined, metri-
rected pitch-​ class intervals (though not by those cally compatible in this way. Real interval spaces and
names), introducing the term interval classes for the (as discussed later in this section) modular interval
latter. Several more recent sources offer complete pre- spaces always satisfy this condition.
sentations of directed and undirected pitch and pitch-​ 10. Texts in pitch-​class set theory typically define
class intervals, among them Rahn 1980, 20–​29, and undirected pitch-​class intervals in a simple and intui-
Straus 2016, 9–​12. tive way, saying for instance that the interval between
2. The reservations raised here about the suitability pcs x and y is whichever of the mod-​12 numbers y − x
of interval spaces for measuring distances are devel- and x − y is “smaller.” The intent of such a definition is
oped at greater length in Tymoczko 2009a. In par- that 4 is “smaller” than 8; this is easy to understand
ticular, see pp. 243–​48 for the question of “whether but does not stand up to mathematical scrutiny. For
Lewinian intervals are supposed to have size.” numbers mod 12, the concept of “smaller” is not
3. This formulation of the total distance, in which defined: ℤ12 (or ℝ12) is not an ordered set. The defini-
the distances traversed by different voices are added, tion of |x|12 in this section is one way of making this
corresponds to a path in Figure 12.1.1 from 000 to idea more precise.
004 to 048; this description is also consistent with the 11. Hint for Exercise 12.2.9: Four different numbers
discussion of voice-​leading distance given in Section appear among the five distances in the answers to this
10.2. We will see in Sections 12.4 and 12.5 that there exercise.
is not just one way to measure distances in OPTI-​ 12. Hint for Exercise 12.2.12b: Distances in spc can
space; whether a straight-​line path from 000 to 048 be greater than 6.
may represent a distance smaller than 8 depends on 13. Metric spaces automatically acquire a topologi-
which distance function we choose. Regardless of that cal structure; spaces related by dilation have the same
choice, however, the distance between 000 and 048 topology.
exceeds the distance between any other pair of points 14. For readers who may be wondering about arith-
in the space. metic with the ∞ symbol, the statements ∞ +​∞ =​∞,
4. Hints for Exercise 12.1.4: In parts (a)–​(c), most ∞ +​ x =​∞, ∞ ≤ ∞, and x ≤ ∞ are true for all real num-
triples separated by the given interval are not equiva- bers x. Elements of the set ℝ ∪ {∞, −∞} are sometimes
lent in the desired way; the tricky part is to find one called extended real numbers; only the nonnegative half
pair of triples, α1 and β1, that are equivalent. In part of the extended real number line is being used here.
(a), the first two components of α1 must be six semi- Many arithmetic operations can be carried out with
tones apart. In (b), there is only one possibility for extended real numbers, although they do not form a
α1. In (c), try making the first two components of α1 group or any other standard kind of algebraic struc-
the same, and consider what other pitch triple could ture. They do form an ordered set with a well-​defined
be cardinality-​equivalent to this α1. For (d), review topological structure.
the definition of normal equivalence relations from 15. Theoretically, edge lengths do not even need to
Section 7.4 to see why it is not possible for two pitch be integers. A graph with edge lengths defined inde-
triples to be equivalent by such a relation while two pendently may contain configurations that appear

524 • E x p l o rin g M u sica l S paces

to violate the triangle inequality—​for example, a tri- be able to find two generators, one producing horizon-
angle with sides of length 1, 2, and 5. The violation is tal motion in Figure 1.2.11a and the other producing
illusory, however: the edge of length 5 will not occur vertical motion. The generators for the two tonnetzes
in the shortest path joining any two vertices, and is (parts (e) and (f)) are musically very similar, but the
therefore irrelevant to the distance function. generators in the Riemannian tonnetz are more dif-
16. Hints for Exercise 12.3.6c: Symmetry depends ficult to describe mathematically. If you wish to do
on the assumption that all sets in the space S are of this, you may find it helpful to review the discussion
the same cardinality. For the triangle inequality, use of just frequency-​class space in Section 7.4. For the
the result of Exercise 2.1.5d. conformed tonnetz, it should be clear that the set
17. See Chapter 2 of Burago, Burago, and Ivanov of three generators is redundant, because any two
2001. The authors define a class of admissible paths of them generate the group. Is this true also for the
whose length can be measured; the underlying space Riemannian tonnetz? For the ski-​hill graph (part (g)),
is initially assumed to have a topological structure recall Figure 3.5.8c.
but no metric structure. A length function defined on 24. Cohn (2012, 6) defines voice-​leading work in
the class of admissible paths, satisfying certain speci- a way equivalent to the voice-​leading distance as we
fied properties, is then used to define a metric on the define it here.
space. Metric spaces constructed in this way are called 25. In some circumstances distances between
length spaces; not every metric space is a length space. chords may also be regarded as a measure of the
18. Figure 4.2.18 made a somewhat similar point similarity of the chords, shorter distances represent-
about Weber space, but the distance function implicit ing greater similarity. This idea, not explored here,
in the discussion of that figure, in which diagonal engages the sizable literature on similarity measures
moves were available, was not the same as the Weber that determine, for any given pair of chords, a numeri-
distance. cal measure of their similarity. Ian Quinn (2006 and
19. Chapter 6 of Grossman and Magnus 1964 pro- 2007) comments extensively on similarity measures
vides a straightforward introduction to words in group en route to his construction of a spatial representation
theory and their relation to graphs. Determining of the concept of chord quality based on Fourier prop-
whether two different words represent the same ele- erties introduced in telegraphic fashion in Lewin 1959.
ment is straightforward in the Riemann group and in 26. The distance functions described here are
many other musical applications, but in general this related to the concept of norm, important in linear
problem (known as the word problem for groups) is algebra and related areas of mathematics. Norms are
difficult. Gollin 2000 uses words and word lengths to defined axiomatically, but the relevant examples are
study the structure of many Lewinian transformation as follows. If A =​(a, b) or A =​{a, b} is either an ordered
groups. or an unordered set or multiset of real numbers of car-
20. The definition of distZ does not require that the dinality 2, the max norm of A is ||A||Max =​max(|a|,
action of G on S be simply transitive. If it happens that |b|), the Euclidean norm of A is ||A||Euc =​(a2 +​ b2)1 ⁄ 2,
no word on Z maps x to y, then distZ(x, y) =​∞. If two and the sum norm of A is ||A||Sum =​|a| +​|b| (with
different transformations in G map x to y, then words analogous formulas in higher dimensions). The norm
representing those transformations are not equiva- of A, loosely speaking, is a measure of the “size” of A,
lent, even though they both correspond to paths from or of the distance of A from the origin in some appro-
x to y in the graph. priate space. Our distance functions are therefore cal-
21. Hints for Exercise 12.3.9: The empty word of culated as the norms of the associated displacement
length 0 represents the identity element of G, which multisets.
maps every element of S to itself. Symmetry depends 27. Though the triangle inequality for the
on the availability of inverses of elements of Z; using Euclidean distance seems obvious, a rigorous proof is
the result of Exercise 5.3.3c, you should be able to see, more difficult than might be expected. Exercise Q1.4
for example, that if the word ab−1c represents a transfor- in Ó Searcóid 2007, 279, provides a general proof
mation that maps x to y, then c−1ba−1 represents a trans- applicable in multiple dimensions.
formation that maps y to x. For the triangle inequality, 28. Hint for Exercise 12.4.3: The quantity on the left
if you have a word that maps x to y and one that maps y side of this inequality is always equal to either dist(x1,
to z, how can you obtain a word that maps x to z? x3) or dist(y1, y3), so the inequality may be established
22. Hint for Exercise 12.3.11: Exercises 4.2.2 and by showing that both of those are less than or equal to
8.2.5 are relevant. the quantity on the right.
23. Hints for Exercise 12.3.12: All of the generat- 29. Hint for Exercise 12.4.4: Do not overlook
ing transformations are reasonably familiar with the the obvious. If α =​(C4, E4) and β =​(C4, G4), what is
exception of those in sl (part (d)). In sl you should distMin(α, β)?

Distances • 525
 30. Mathematically, all three of the distance func- the authors note applications not only to the study of
tions considered here belong to a family of so-​called voice leading but also to economics.
p-​norm metrics—​or, more precisely, the Euclidean and 34. Callender 2004 explicitly uses Euclidean dis-
voice-​leading distances are p-​norm metrics, and the tance calculations in several spaces.
max distance is a limiting case of that family. Let p be 35. For example, Straus (2003, 320–​22) defines
any real number ≥ 1. In three dimensions, the p-​norm smoothness in terms of the voice-​leading distance;
distance is given by [(Δx)p +​(Δy)p +​(Δz)p]1 ⁄ p, the pth much of the neo-​ Riemannian literature of the
root of the sum of the pth powers of the component- 1990s adopts the same understanding implicitly.
wise displacements. This formula defines a distance Though usage has been inconsistent, some authors
function distp for every p ≥ 1. In particular, distp for have drawn subtle and often informal distinctions
p =​1 is the same as distVL, while distp for p =​2 is the between “smooth” and “parsimonious” voice lead-
same as distEuc. As p grows larger, the p-​norm dis- ings, perhaps using the voice-​leading distance as a
tance between a fixed pair of points of the space gets measure of “smoothness” and either the max dis-
smaller. As p approaches infinity, the p-​norm distance tance or the number of common tones as a measure
approaches max(Δx, Δy, Δz), the max distance, as a of “parsimony.” For related discussion see Cohn
limiting value, so the max distance may be considered 1997, Douthett and Steinbach 1998, Cook 2005,
to be distp for p =​∞. and Tymoczko 2006 (supplemental online material).
31. See the online supplementary material to Roeder (1987, 381–​84) may have been the first to
Tymoczko 2006 for a more precisely formulated ver- contrast the Euclidean and voice-​leading distances
sion of this property in a form called the distribution in a musical context, considering the latter more
constraint; see also Hall and Tymoczko 2012. Straus appropriate as a measure of distance in his “ordered
2003 discusses uniformity as a property of a voice interval space” (not exactly the same situation con-
leading; a transposition, for example, is a completely sidered here).
uniform voice leading. While a highly uniform voice 36. Cohn (2012, 5) documents references to har-
leading will feature a level displacement multiset, the monic proximity based on common tones in treatises
converse is not necessarily true, because a multiset in throughout the nineteenth century, beginning with
which all displacements are the same may include a Francesco Galeazzi in 1796.
mix of displacements in the ascending and descending 37. Suppose the following three conditions are
directions. true of an equivalence relation ~ defined on a met-
32. Hints for Exercise 12.4.6: The proofs involve ric space (S, dist): (1) the relation ~ is defined by the
little more than a careful understanding of the defi- action of a symmetry group G on S (that is, the equiv-
nitions, but some notational complexity arises in the alence classes are orbits of G); (2) all elements of G
details. Use notations such as (S1, G1, int1), (S2, G2, are isometries of the space S; and (a rather technical
int2), (S1, dist1), and (S2, dist2) for the spaces involved. requirement) (3) all orbits of G are closed subsets in
For part (a), we are assuming that equal intervals the topology of S. Under these conditions, a distance
imply equal distances in both S1 and S2. That is, if function on the quotient space S/​~ may be defined
x1, y1, x1′, and y1′ are elements of S1 with int1(x1, y1) by the quotient distance formula presented here. The
=​ int1(x1′, y1′), then dist1(x1, y1) =​dist1(x1′, y1′), and relations ~O, ~P, ~T, and ~I satisfy these three condi-
a corresponding statement holds for S2. You must tions. Cardinality equivalence does not, nor does the
show that the corresponding property holds in S1 × relation in Exercise 12.5.3.
S2: if int((x1, x2), (y1, y2)) =​int((x1′, x2′), (y1′, y2′)), then 38. Many texts on the subject of metric spaces do
dist((x1, x2), (y1, y2)) =​dist((x1′, x2′), (y1′, y2′)), where not even consider quotient spaces. It is always pos-
int is the product interval function and dist is any of sible to define a pseudometric on a quotient space,
distMax, distEuc, or distVL. (Think about what the inter- a pseudometric being a function that satisfies the
val equation implies about the one-​dimensional inter- properties required of a metric except for the sec-
vals and the displacement multisets.) In part (b) there ond part of the positivity axiom (that is, distances
are no interval spaces to consider. By homogeneity, we between distinct points are allowed to be 0). The
know that if x1 and y1 are elements of S1, then there is simple quotient distance formula given in this sec-
an isometry f1: S1 → S1 such that f1(x1) =​ y1, along with tion, even with the refinement suggested in Exercise
a corresponding property of S2. Show how to combine 12.5.3, is not sufficient for defining the quotient
these two isometries to produce an isometry f: S1 × S2 pseudometric in general: the general case requires a
→ S1 × S2 that maps (x1, x2) to (y1, y2). chain technique similar to that discussed below for
33. The comparison of metrics in ℝn is of interest cardinality equivalence. (See, for example, Burago,
in several disparate fields. Figure 12.4.8 is inspired Burago, and Ivanov 2001, 62, where pseudomet-
by Figure 5 in Hall and Tymoczko 2012, 270, where rics are called semi-​metrics.) A pseudometric can be

526 • E x p l o rin g M u sica l S paces

reduced to a metric by gluing together all points and displacing one of the two pitch triples by mul-
separated by zero distance, but the metric space that tiples of 1 3 .
results may be very different from the quotient set 46. Eccentricity is defined this way for vertices in
S/​~. For the space in the exercise, the quotient pseu- graphs; see Harary 1969, 35.
dometric turns out to be 0 in all cases, so when it 47. To say that the probability of an event is zero
is reduced to a metric the space becomes trivial—​a is not the same as saying that it is impossible. Many
single point. The general question of when the quo- nonempty sets in a plane have zero area. If S is a set of
tient pseudometric is actually a metric does not have zero area and a point is selected from the plane at ran-
a simple answer. dom, the probability of selecting a point of S is zero—​
39. Hints for Exercise 12.5.6: The spheres are large but the only set from which it is impossible to select an
in comparison with the size of the space: most dyads element is the empty set. (This discussion disregards
in OP-​space lie within one or more of them. It may some other technical considerations, such as the ques-
be helpful to plot the shapes first by imagining the tion of what it means to choose a point “at random”
figure continuing beyond the boundaries (both the from an unbounded set.)
false boundaries and the mirror boundaries), then 48. Hints for Exercise 12.5.26: The area of the strip
determining which parts of the fundamental region in (a) may be calculated as the difference between the
correspond to the areas beyond the boundaries. areas of two triangles. In (c), the height of the small
Because of the orientation of the figure, the sphere triangle is 2u − v; its base may be calculated from the
for the voice-​leading distance is a square in the ori- proportion with the large triangle.
entation , while the sphere for the max distance is 49. Hint for Exercise 12.5.27a: The small triangles
a tilted square (the opposite was true in Figure forming a grid in Figure 11.3.1 are not the same size
12.4.8). as the triangles in T-​space, so distances are not mea-
40. Hint for Exercise 12.5.8c: A very simple proof is sured in v-​units here. The triangles are equilateral,
possible for the voice-​leading and Euclidean distances. however, so the ratio of altitude to base is 3 4 , the
For the max distance this simple reasoning does not same as the ratio between u and v—​which is also the
work, but see if you can find a way to establish the same as the ratio for the entire cross section.
same result by considering the number of different 50. Hint for Exercise 12.5.27d: These trichords form
points at the greatest possible distance from each of a cylinder of radius 1 (or 2) about the central axis of
the two dyads. the prism.
41. Hint for Exercise 12.5.10: In one case the parallel 51. Hint for Exercise 12.5.28d: This result is con-
voice leading is shorter according to all three distance sistent with Exercise 11.6.3, which showed that the
functions; in one case the zero-​sum voice leading is obtuse angle at the origin is approximately 109.5
shorter; and in one case the distance functions are in degrees. As in that exercise, the law of cosines will be
disagreement about which path is shorter. useful. Like the tetrahedron as a whole, the triangle is
42. Hints for Exercise 12.5.12: In part (a), you symmetrical about the dashed diagonal line.
should find three different augmented triads for the 52. Hint for Exercise 12.5.29a: Some of these line
three distance functions. For part (b), if you tabulate segments are not the shortest paths joining their
the displacement multisets from C major to every endpoints in OP-​space. To calculate the length of a
other triad, you should be able to see which ones maxi- segment you must determine the displacement mul-
mize the distance functions. tiset for the particular voice leading the segment
43. Hint for Exercise 12.5.15b: Try multiples of 1 3 represents.
for the coordinates. Readers versed in calculus may 53. Hint for Exercise 12.5.29e: The same sum-​class-​
be able to write a formula for distEuc((0, 0, 0), (x, x 0 tetrahedron appears at both the top and bottom of
− 2, x +​1)) (or, more simply, the square of that dis- the hyperprism. The length of the line segment CCCC–​
tance) in terms of x, then apply a standard technique E♭E♭E♭E♭ in the fourth dimension is not the same as the
involving derivatives to find the minimum value of length of the line segment CCCC–​E♭E♭E♭E♭ within the
this function. tetrahedron at either end.
44. Hint for Exercise 12.5.19: To calculate a max 54. Hint for Exercise 12.5.31: A path of length 6 may
distance in T-​space you must find a transposition of either ascend or descend from A♭7 to c♯ø7; the two kinds
one of the triples that minimizes the max distance in of path are not equal in number, and only the descend-
cpitch3. Try transposing by multiples of 1 2 . ing paths are visible in Figure 12.5.30a. Requiring that
45. Hint for Exercise 12.5.21b: As in some pre- the path stay within the octatonic hypercube chain
vious examples, a minimal voice leading for the eliminates some paths that would otherwise be possi-
Euclidean distance may be found by starting with a ble in OP-​space. The number of possible paths is large
minimal voice leading for the voice-​leading distance enough that you will probably arrive at the correct

Distances • 527
number more easily through some simple calculations SUGGESTED READING
than by trying to locate them all.
55. Hints for Exercise 12.5.36: The first segment of Several general introductions to the theory of met-
path ③ joins G♯AB♭ to one of the two nearest possible ric spaces are available. Ó Searcóid 2007 is quite
points on the boundary of the prism, in the same accessible; some special topics mentioned in this
sum-​class layer. Path ④ leaves that sum-​class layer. chapter, including path lengths, group actions,
The most efficient voice leading for the Euclidean dis- and quotients of metric spaces, are treated more
tance is not any of these, passing instead through the extensively in Burago, Burago, and Ivanov 2001.
point (81 2 +​ r, 81 2 +​ r, 10 +​r), a transposition of the Tymoczko 2009a discusses the unsuitability of
point (81 2 , 81 2 , 10) from path ③ by Tr where r =​ 1 3 , Lewinian intervals in some geometric settings and
giving a Euclidean distance of approximately 5.77 proposes some approaches different from the dis-
between G♯AB♭ and CC♯D in OPC-​space. tance functions considered here.

528 • E x p l o rin g M u sica l S paces

PART FOUR
Theory of Scales
Diatonic and Beyond
 13
Scales I
Diatonic Spaces

SCALES HAVE made many appearances in this broaden in the following chapter as we increas-
book. This chapter and the next delve into scalar ingly turn our attention to scalar spaces other
spaces in more detail. In Section 13.1 we inves- than the diatonic, including some with some-
tigate the structure of diatonic or generic spaces thing other than seven or twelve notes.
with seven notes per octave, consolidating some The placement of these chapters late in the
of our observations about these spaces from book is motivated by their many points of con-
previous chapters and noting that while some tact with ideas introduced previously; in par-
properties of chromatic spaces carry over to the ticular, there are occasional references to OPTIC
mod-​7 case essentially without change, there are spaces and normal forms. Many of the concepts
other ways in which the mod-​7 universe is very here, however, are fairly elementary, and much
different. Sections 13.2–​ 13.4 examine mod-​ 7 of the material depends only incidentally on ear-
and mod-​12 spaces not in isolation but in tan- lier chapters.
dem, as subset and superset, with an exploration
of the subtle and complex relationship between 13.1 DIATONIC AND
a diatonic scale and the chromatic space in which GENERIC SCALES AS
it lies. These sections introduce concepts from
the fields sometimes known as diatonic theory
MUSICAL SPACES
(or diatonic set theory) and scale theory, pioneered Diatonic and generic spaces have been familiar
by John Clough in several publications around to us since Chapter 1. The reader may wish to
1980 and developed by many scholars in the review the construction of generic pitch space
ensuing decades. Among these concepts are well-​ gpitch (Figure 1.1.6), diatonic pitch space
formed scales, signature transformations, and the dpitch(n) (1.1.7), generic pitch-​class space gpc
principle cardinality equals variety. The scope will (1.2.8), diatonic pitch-​class space dpc(n) (1.2.9),

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0013
scale-​degree space sd (1.2.10), generic fifths using the mod-​7 absolute value from Section
space gfifth (1.3.8), and generic thirds space 12.2. Exercises 12.2.5 and 12.2.9 explored inter-
gthird (1.3.12). val and distance calculations in these spaces; the
Recall also the difference between diatonic largest possible distance between two points in a
and generic. A diatonic space corresponds to a discrete mod-​7 space is 3.
particular diatonic collection; thus dpc(+​4), the As interval spaces, the spaces with interval
diatonic pitch-​class space with four sharps in its group ℤ7 are all isomorphic (this is a consequence
key signature, contains the note D♯ but not D♮ or of Theorem 7.2.7, the interval-​space homomor-
D♭. Generic pc space gpc, in contrast, contains phism theorem), and the corresponding distance
an element D that is understood to stand for spaces are isometric. Several intervallic relation-
an entire class of generically equivalent notes, ships among the circles of gpc, gfifth, and
including D♭ ♭, D♭, D♮, D♯, and D𝄪; this single space gthird are easily derived.1 In generic pitch-​class
may therefore represent any diatonic scale, as space, a fifth is equal to four steps. It follows that
well as other collections that contain the seven if x and y are pitch classes such that intgfifth(x, y)
letter names once each, such as a harmonic =​1 (for example, x =​D and y =​A), then intgpc(x,
minor scale. (Generic equivalence was intro- y) =​4. More generally, the equation
duced in Section 2.4 as an equivalence relation
on signed letter space.) In the next chapter we intgpc(x, y) =​4 ∙ intgfifth(x, y) (mod 7)
will broaden our use of the term generic to accom-
modate some other situations, but at present it holds for all generic pitch classes x and y. This
continues to refer to these abstract spaces with equation expresses intgpc in terms of intgfifth.
seven notes per octave, identified by the usual We can derive an equation for the inverse rela-
seven letter names. tionship—​intgfifth as a function of intgpc—​by
In generic pitch-​class space, or in white-​note observing that in generic pc space, a step such
diatonic space dpc(0), the trichord CDF is related as D–​E is equal to two fifths (D–​A–​E); therefore
by generic or diatonic transposition to EFA and
by inversion to FAB, even though in chromatic intgfifth(x, y) =​2 ∙ intgpc(x, y) (mod 7)
space these three sets represent three different
set classes. (See Sections 7.5 and 7.6 for musical for all generic pitch classes x and y. The examples
examples involving diatonic transposition and intgpc(A, F) =​5 and intgfifth(A, F) =​3 illustrate
inversion.) The normal-​form algorithm, intro- both of these equations, because 5 =​4 ∙ 3 and
duced in Section 10.4 for mod-​12 space, is easily 3 =​2 ∙ 5 (mod 7).
adapted to a mod-​7 context. The trichords just The relationship between the two equations
mentioned share the generic OPTI normal form displayed above deserves closer scrutiny. The
013; the first two also have 013 as their OPT nor- second equation can be derived from the first
mal form, while FAB has the OPT normal form by solving for intgpc. A few observations from
023. The generic set classes, to be enumerated earlier chapters are relevant here. Because 7 is a
shortly, are few in number; 013 and its comple- prime number, we know from Sections 2.5 and
ment 0124 turn out to be the only asymmetric 5.2 that every nonzero integer mod 7 has a mul-
OPTI-​classes. tiplicative inverse, and if a and b are mod-​7 inte-
As detailed in Chapter 7, diatonic and generic gers with a ≠ 0, then the equation an =​ b (mod 7)
pitch spaces are infinite discrete interval spaces always has a unique solution n =​ a−1b. (Group
with interval group ℤ. The interval function structure is lurking here: by Exercise 5.2.7b the
counts scale steps, considered positive in the nonzero integers mod 7 form a multiplicative
ascending direction. From Section 12.2 we know group ℤ7×, and by Exercise 5.3.7, every non-
that these spaces are also distance spaces with zero integer mod 7 generates ℤ7 as an additive
the distance function defined by the equation group.) Write b =​intgpc(x, y) and n =​intgfifth(x,
dist(x, y) =​|int(x, y)|. The spaces gpc, dpc(n), y); then the first of the two equations displayed
sd, gfifth, and gthird, meanwhile, are modular above says 4n =​ b (mod 7). The mod-​7 multipli-
interval spaces with the interval group ℤ7; in any cative inverse of 4 is 2, so solving this equation
of these spaces the interval function measures for n yields n =​4−1b =​2b (mod 7), which is the
intervals as mod-​7 numbers in the clockwise second equation above.
direction in the appropriate circular figure. Each
of these spaces is also a distance space with the dis- Exercise 13.1.1 Write formulas, valid in ℤ7,
tance function defined by dist(x, y) =​|int(x, y)|7, relating the interval functions as specified

532 • E x p l o r i n g M u sical S p aces

below, and verify that your formulas are correct of this section when we discuss analogs of the
for the various intervals from A to F. OPTIC spaces in a mod-​7 setting.

(a) Write a formula for intgthird as a function Theorem 13.1.4: Sum classes of generic
of intgpc. transpositions Let X be a generic pitch-​class
(b) Write a formula for intgpc as a function of set whose cardinality #X satisfies 1 ≤ #X ≤ 6, or
intgthird. more generally a generic pitch-​class multiset
(c) Write a formula for intgthird as a function X such that #X ≠ 0 (mod 7). Then the generic
of intgfifth. transpositions X, t1(X), …, t6(X), in some
(d) Write a formula for intgfifth as a function of order, are sets whose mod-​7 sum classes are
  intgthird. 0, 1, …, 6. That is, for every mod-​7 integer m,
there is exactly one mod-​7 integer n such that
Exercise 13.1.2 tn(X)
   is a set of sum class m.
(a) If x and y are generic pitch classes such that
intgpc(x, y) =​6, what can you conclude about An analogous property does not hold in
distgfifth(x, y)? chromatic space, where, for example, all twelve
(b) If distgfifth(x, y) =​3, what can you conclude major triads are of sum class 2, 5, 8, or 11. To
about intgthird(x, y)? see why the theorem holds in mod-​7 space, sup-
(c) If distgthird(x, y) =​2, what can you conclude pose the given multiset X is of cardinality #X =​
  about distgpc(x, y)?2 k and of sum class s. Then t1(X) is of sum class s
+​ k, because each of the k notes of t1(X) is raised
Exercise 13.1.3 by one generic step relative to the corresponding
(a) Refer to the depictions of gpc and gfifth note of X. More generally, tn(X) is of sum class
in Figures 1.2.8 and 1.3.8. Let mi denote s +​ kn. We are therefore interested in finding a
mod-​7 multiplication by i. Show that the number n such that s +​ kn =​ m (mod 7), where s
function m4 maps each gpc number in (the sum class of X), k (the cardinality of X), and
gpc to the number at the corresponding m (the desired sum class) are known. Because 7
location in gfifth, while m2 maps similarly is prime and k ≠ 0 (mod 7) by assumption, every
in reverse, from gfifth to gpc. (Recall such equation has a unique solution.
from Exercise 6.1.9b that the mod-​12
multiplication operator M7 maps in both Exercise 13.1.5 In the situation described in
directions between pc and fifth.) Theorem 13.1.4, write a general formula for n
(b) Which operator mi maps from gpc to in terms of k, s, and m. Then verify that your
gthird (Figure 1.3.12)? From gthird formula is correct in each of cases (a)–​(e) below:
to gpc? From gfifth to gthird? From
  gthird to gfifth? (a) X is the dyad CD; m =​0 (that is, what generic
transposition of CD is of sum class 0?)
Musically, the solvability of equations of the (b) X is the triad CEG; m =​0
form an =​ b (mod 7) implies that every nonzero (c) X is the triad CEG; m =​1
interval in generic pitch-​ class space divides (d) X is the asymmetric trichord CDF; m =​0
evenly into every other interval, and also that (e) X is the asymmetric trichord FAB; m =​0
every interval can be divided evenly into n parts
for every nonzero number n. The possibility of Finally, show that the theorem’s conclusion
equal division depends on octave equivalence: fails for the multiset CCCCEGG (for which k =​0
a step in gpitch cannot be divided in half, (mod 7)), and notice why your formula does not
but in gpc a step is equal to two fifths. The apply
   in that case.
analogous properties in chromatic (mod-​ 12)
pitch-​class space do not always hold: even Generic space gpc has many characteristics
allowing for octave equivalence, a minor sev- in common with chromatic space pc, but when
enth cannot be divided evenly into minor there are differences in their behavior, they
thirds, and a perfect fourth cannot be divided generally have to do either with the fact that 7,
into two equal parts. unlike 12, is a prime number, or with the even
Another consequence of the solvability of simpler fact that 7 is a smaller number. We
mod-​ 7 equations is presented in the follow- noted in Section 4.3 that the number 7 is just
ing theorem, which will prove useful at the end the right size to guarantee an appealing feature

Diatonic Spaces • 533

of the triangulated generic tonnetz, reproduced The fact that every nonzero integer mod 7 is a
in Figure 13.1.6: each of the six nonzero generic generator of the additive group ℤ7 has important
transpositions t1–​t6 corresponds to motion in a ramifications for sequential behavior in generic
particular direction along one of the three axes space. If a musical fragment is repeated sequen-
of this tonnetz. The axes are the seconds axis, the tially, transposed by the same generic trans-
thirds axis, and the fifths axis; we may imagine position operator tn at each repetition, seven
that the circular spaces gpc, gthird, and gfifth iterations will always cycle through every pos-
have been unrolled in these three directions to sible generic transposition level. Here is another
form the tonnetz. We noted in Chapter 4 that difference between generic and chromatic space:
as a finite graph, this tonnetz is K7, a complete successive chromatic transpositions by T2 or T3,
graph on seven vertices. for example, will never cycle through the twelve
Two common sequence patterns were possible transposition levels, because 2 and 3
graphed in the generic tonnetz in Figure 4.3.4. are not coprime to 12. Generic (or diatonic)
sequences, in contrast, automatically complete
a generic aggregate without repetitions; Clough
(1979–​80, 469–​73) refers to this property as a
“serial” feature of the diatonic system.
The structure of ℤ7 also ensures that generic
sequences form predictable kinds of hierarchi-
cal patterns. Figure 13.1.7 illustrates with a
short passage from Beethoven’s Violin Sonata,
Op. 30, No. 3.3 The passage is not completely
diatonic, but it is sequential in generic space,
the chord roots descending by step (generic
transposition t6). As is commonly the case in
sequences, motivic and metric organization
groups the chords in pairs, in this case form-
ing a descent by thirds (t5) at the one-​measure
level. Changes of register and instrumenta-
tion group the two-​chord blocks themselves in
pairs, creating larger two-​measure groupings,
descending by fifth (t3). The C♯ (generic C) chord
at the end of the example hints at the possibil-
ity of an even higher four-​measure level of struc-
ture, descending again by step, recapturing in
eight-​chord blocks the t6 transpositions of the
FIGURE 13.1.6 Transpositions t1–​t6 in the generic original pattern. If the sequence were to con-
tonnetz tinue indefinitely, every level in this hierarchy

FIGURE 13.1.7 Hierarchy of generic sequential motion in Beethoven, Sonata in G Major for Violin
and Piano, Op. 30, No. 3, I, mm. 57–​61

534 • E x p l o r i n g M u sical S p aces

would illustrate the “serial” property, cycling (b) If a t2 sequence is a k-​fold extraction of a t3
through all seven generic pitch classes before sequence, what is k?
repeating; the fact that 7n =​0 (mod 7) for every (c) If a t4 sequence is a k-​fold subdivision of a t6
mod-​7 integer n ensures that the eighth generic sequence, what is k?
pitch class at any level will always be the same (d) What sort of sequence can be a 5-​fold
as the first. The three transpositions involved,   subdivision of a t2 sequence?5
t6, t5, and t3, circumnavigate the three axes of
generic space—​the seconds, thirds, and fifths Exercise 13.1.9 Construct a hierarchy
axes respectively. Adjacent levels in the hierar- analogous to that in Figure 13.1.7, starting with
chy illustrate the transformational identities a sequence that ascends by step (t1). Compare
t6 ● t6 =​ t5, t5 ● t5 =​ t3, and t3 ● t3 =​ t6. Any sequence the result with the original figure, and also
generated by t6, t5, or t3, sufficiently prolonged, consider connections between this hierarchy
will reproduce this hierarchy.4 and the formulas relating intgpc, intgfifth, and
This equation t6t6 =​ t5 (we henceforth sup- int  
gthird from Exercise 13.1.1.
6

press the ● symbol) holds because 6 +​6 =​5

(mod 7), or equivalently 6 ∙ 2 =​5 (mod 7), and Example 13.1.10: Brahms, Symphony
it implies that if every second note is selected No. 4, I Schoenberg and many subsequent
from a t6 sequence, the result is a t5 sequence. commentators have observed that the opening
Similarly, from the equation 6 ∙ 3 =​4 (mod 7), if theme of Brahms’s Fourth Symphony outlines
every third note is selected from a t6 sequence, a chain of descending thirds.7 A close analysis
the result is a t4 sequence. The notes D, A, and E of generic structure, summarized in Figures
in Figure 13.1.7 illustrate this property, though 13.1.11 and 13.1.12, shows that this opening
this grouping is not articulated musically in the spawns a rich network of related sequential
example. Clough (1979–​80, 476) uses the word patterns featuring extraction, subdivision,
extraction for this process: the t5 sequence is a and shrewdly altered inversion, which figure
2-​fold extraction of the original t6 sequence, and significantly in developmental processes
the t4 sequence a 3-​fold extraction. Generally, a k-​ throughout the movement.8
fold extraction of a tn sequence is a tm sequence, The opening theme is shown at level 3 of
where m =​ kn (mod 7). Figure 13.1.11a. Only the melody is shown; the
A hierarchy such as that in Figure 13.1.7 motivic structure is sequential even though
may be read from the bottom up, so that new the harmony is not. The pattern is generic but
sequences are formed by interpolating a fixed not strictly diatonic; the pitch content is a
number of new notes between each pair of adja- harmonic minor scale. The numbers appearing
cent notes in a given sequence: the figure shows in Figures 13.1.11 and 13.1.12 indicate generic
that a t3 sequence is a 2-​fold subdivision of a t6 transpositions directed from left to right, so the
sequence. Extraction and subdivision are inverse 5s at level 3 describe the descending thirds as
processes: if a k-​fold extraction of a tn sequence generic t5 transpositions.
is a tm sequence, then a k-​fold subdivision of a tm The upper portion of Figure 13.1.11a shows
sequence is a tn sequence. If any two of k, m, and n successive extractions from this t5 sequence.
are known nonzero integers mod 7, the equation The short-​long rhythm implies a grouping in
m =​ kn (mod 7) uniquely determines the third. two-​note segments, each segment related to
If m and n are known, we can be sure that some the next by t3 (=​ t5t5). This 2-​fold extraction
k can be found for which a tm sequence occurs appears at level 2, where the long and metrically
as a k-​fold extraction from a given tn sequence. accented note of each pair is extracted. Because
If k and m are known, then we know that for a the segments appear alternately as descending
given tm sequence, a k-​fold subdivision exists as thirds and ascending sixths, the two-​note
a tn sequence for some n. groups form a sequence in gpc but not in
gpitch, and traditional terminology fails to
Exercise 13.1.8 recognize a melodic sequence at this level. A
(a) Exercise 5.3.2g asked that you solve clear melodic sequence emerges at the four-​note
the equation 4 ∙ k =​3 (mod 7). Having level, though there are only two such segments,
determined the solution k, verify that a related by t6 (=​ t3t3), before the pattern changes.
k-​fold extraction of a t4 sequence is a t3 The figure shows two musically feasible
sequence, and a k-​fold subdivision of a t3 alternatives for notating this extraction: metric
sequence is a t4 sequence. accent favors the notes at level 1a, while those

Diatonic Spaces • 535

FIGURE 13.1.11 Sequential hierarchies in Brahms, Symphony No. 4 in E Minor, Op. 98, I: (a) mm.
1–​4 and 19–​22; (b) mm. 1–​9

at 1b are registrally prominent. Both options the melodic thirds at level 3 reverse direction,
are t6 sequences, 2-​fold extractions from level now ascending (t2) rather than descending (t5).
2’s t3 sequence, and 4-​fold extractions from Crucially, however, the pattern is not precisely
level 3’s t5 sequence. inverted: the occasional repeated pitch classes,
The second statement of the theme at m. 19 in the form of octave leaps, are new. A strict
is elaborated in violas and woodwinds as shown inversion of mm. 1–​4 would produce a t4
in simplified form at level 5. The pattern is three sequence at level 2, as a 2-​fold extraction of a t2
descending steps followed by an ascending step, sequence and the inversion of the t3 sequence
or t6t6t6t1 =​ t5: the subscripts sum to 5 mod 7, that occurs at this level in mm. 1–​4. The repeated
as they must, since the original t5 sequence is notes, however, result in a t4t2 pattern at level 2
reproduced in every fourth note of this pattern. and a t6 sequence at level 1a above, allowing the
Of course, each ascending step effectively t6 established at this level in mm. 1–​4 to continue
negates an adjacent descending step. These without interruption to m. 9. Three ascending
countervailing moves (two 6s and two 1s in each thirds in this second part of the melody occur
measure) are eliminated in the reduction at level in the time of four descending thirds in the first
4, which is a straight t6 sequence, a descending part, and both groupings produce t6 at the two-​
scale—​and a 2-​fold subdivision of the original measure level (t2t2t2 =​ t5t5t5t5 =​ t6). One might
t3 sequence from level 3. Levels 1 (a or b) and expect that if a sequence is inverted, all its
4 display t6 sequences at two different levels, extractions should also be inverted—​as indeed
and the t6–​t3–​t5–​t6 hierarchy between these two they must be in cases of strict inversion. Here, a
levels essentially matches that of Figure 13.1.7. subtle deviation from strict inversion allows the
Figure 13.1.11b extends levels 1a, 2, and 3 extraction at level 1a to maintain, rather than
of 13.1.11a to encompass mm. 1–​9. At m. 5, invert, its original pattern.

536 • E x p l o r i n g M u sical S p aces

FIGURE 13.1.12 Sequential elaborations in Symphony No. 4, I: (a) elaboration of t6 sequences in mm.
37–​40; (b) elaboration of t5 sequences in mm. 157–​61; (c) elaboration of t3 sequences in mm. 169–​75

The developmental techniques to which the woodwinds and passes to the first violins,
Brahms subjects his materials in this movement the up-​stemmed notes spell out B–​G–​E–​C,
feature a great deal of small-​scale sequential the first notes of the opening theme (level 3
patterning, prominently featuring the t5, from 13.1.11a). The other parts, meanwhile,
t3, and t6 relationships introduced in the elaborate t6 in other ways. A t3 pattern, four
opening measures. Figure 13.1.12 shows three notes per measure, starts in the first violins
such passages, structured as elaborations and passes to the cellos and basses. The second
of sequences generated by t6 (13.1.12a), t5 violins present two forms of a t5t1 subdivision:
(b), and t3 (c). In (a), the essential motion of in the third measure the metric relationship
all the parts shown is parallel, and they all between the two notes is reversed, reshaping
exhibit t6 motion at the half-​measure level, the figure to resemble a retrograde inversion of
indicated with downward stems and 6s on all the original.
staves. Throughout this figure, stems indicate Figure 13.1.12b, from the beginning of
segments of sequential structure, not parts in the development section proper, features t5
the instrumental texture. The t6 motion at the sequences and elaborations thereof. The down-​
half-​measure level implies t5 at the one-​measure stemmed notes in the flute spell out a G-​minor
level; in the simple scalar t6 line that begins in statement of the opening theme, its original

Diatonic Spaces • 537

octave displacements intact. One of every two In addition to the transposition operators
descending thirds in this theme is filled in t0–​t6, generic space admits inversion operators
stepwise, creating a t6t6t5 =​ t3 pattern in one-​ i0–​i6, calculated by the equation in(x) =​ n − x
measure blocks. Because the t5 leaps coincide with (mod 7). In the mod-​12 case, structural differ-
rests in the line, this pattern is easily conceived ences between inversion operators with even
as a gapped t6 sequence whose every fourth index and those with odd index may be recalled.
note is missing (shown with small noteheads in For example, Figure 5.5.4 showed that even
parentheses). The violins, meanwhile, elaborate t5 inversions have two fixed points in pc while
not with t6 but with a t4t1 pattern that resembles odd inversions have none; parts (c) and (d) of
the t5t1 elaboration of t6 in 13.1.12a. This generic Exercise 6.1.12 showed that the two kinds of
pattern persists for several beats even after the inversions have different cycle structures when
rhythm changes at the beginning of the fourth regarded as permutations. In mod-​ 7 space,
measure of the example. these differences vanish: all inversions have the
Figure 13.1.12c reintroduces inversion, and same cycle structure, determined in Exercise
specifically the idea that a not-​quite-​literal 6.1.12e, and every in has a single fixed point at
inversion may align with the original pattern at a generic pitch class 4n (mod 7), a property noted
larger level. The violin line begins with the same for inversions in diatonic pitch-​class space in
t6t6t5 =​ t3 pattern just encountered in (b). The Exercise 7.6.9a. Cayley diagrams for the cyclic
bass line, however, moves in three-​note segments group 𝒯7 (or 𝒞7) and 𝒯ℐ7 (or 𝒟7) were con-
of an ascending (t1) rather than a descending structed in Exercise 6.2.11.
scale. At a glance the two parts appear to form a In some situations in which we might employ
canon in inversion, at the time interval of a half inversions in chromatic space, the correspond-
measure. Closer inspection reveals, however, that ing inversions in generic space turn out to be
the inversion is not exact: no notes are missing superfluous. We have noted, for example, the
from the bass line’s scale, which forms t1t1t1 =​ t3 simply transitive action of the dihedral group
in every measure rather than t1t1t2 =​ t4, the literal 𝒯ℐ12 on triad, the set of 24 major and minor
inversion of the upper voice. Three ascending triads. The distinction between major and
steps in the bass line therefore appear in the minor triads cannot be made in generic space,
time of four descending steps in the violins. Both where there is only a single generic triad on a
parts form a descending-​fifths pattern (t3) at the given root (CEG, DFA, …). Generic triads are
one-​measure level (t1t1t1 =​ t6t6t6t6 =​ t3), allowing symmetrical sets of mod-​ 7 prime form 024;
the relationship between the parts to remain the set {0, 2, 4} (CEG) is invariant under i4. The
consistent from measure to measure so that inversions in therefore do not produce any sets
the sequence can continue. The relationship is that cannot already be produced by transposi-
notably similar to the relationship between levels tions tn. Generic triad space gtriad is a set of
1a and 3 in the two halves of Figure 13.1.11a; in cardinality 7, not 14; for many purposes we may
fact, if all transpositions in the equation t1t1t1 identify a generic triad with its root and con-
=​ t6t6t6t6 =​ t3 are doubled, the result is t2t2t2 =​ sider gtriad to be the same space as gpc. Such
t5t5t5t5 =​ t6, precisely the equation that described an identification was implicit in Figure 13.1.7,
that earlier situation. After four measures of where the note names stood for triads, as well as
Figure 13.1.12c, the counterpoint is inverted: the in the sequence diagrams in Figure 4.3.4.
gapped descending scales move to the bass, the Close affinities between third-​related triads
ascending
   scales to the violins.9 were remarked upon previously in Sections 1.3
and 4.3. In generic space, third-​related triads
Exercise 13.1.13 Construct an analysis similar share two common tones and are related by
to Figure 13.1.11a for the opening theme of single-​step voice leading in the remaining voice
Mozart’s Symphony No. 40 in G Minor, K. (for example, CEG and EGB are joined by C →
550. Show a hierarchical structure of three or B), a smoother connection than between triads
more levels, two of them outlining t6 patterns related by fifth (which requires motion in two
proceeding at different rates. How does the voices) or by step (which requires motion in
pattern change at m. 21? When the theme all three). For these reasons, the generic circle
recurs to open the development section, it is of thirds gthird offers an appealing organiza-
altered chromatically; is the generic pattern the tion for generic triad space. Implications of the
same as before? Study other related passages in generic circle of thirds for functional progres-
the
  development as well. sions were observed in Figure 4.3.8.

538 • E x p l o r i n g M u sical S p aces

 by pairing triads with seventh chords as in Figure
13.1.14a. In this arrangement the circles follow the
organization of gthird, generated by t2, and each
generic triad is radially aligned with the generic
seventh chord sharing the same root. The relation-
ship between these two chords may be described
by the Rameau-​inspired operation of superposi-
tion (“sup” in the figure), which extends a stack
of thirds upward to include a fourth note (CEG →
CEGB).10 Superposition, thus defined, is a function
from the set gtriad of generic triads to the set
gsev of generic seventh chords; it is not a trans-
formation defined on the entire space gtriad ∪
gsev. We can obtain a function sup* on this larger
space by combining sup with its inverse: that is,
if X ∈ gtriad, then sup*(X) =​sup(X), while if Y ∈
gsev, then sup*(Y) =​sup−1(Y). Thus sup*(CEG) =​
CEGB and sup*(CEGB) =​CEG. Defined in this way,
sup* is an involution on gtriad ∪ gsev, so that
replacing sup with sup* allows the radial arrows in
the double-​circle space to become bidirectional. In
the graph in 13.1.14b these arrowheads have been
omitted entirely, as has been our practice in other
graphs with involutions.
Figure 13.1.14b is a Cayley diagram for a
transformation group acting on gtriad ∪ gsev,
with the generators t2 and sup*. The transposi-
tion group 𝒯7, generated by t2, effects circum-
ferential motion around either circle, while
movement from one circle to the other is accom-
plished by sup*, which by itself generates a group
{E, sup*} of order 2. Because sup* commutes with
t2 (see Exercise 13.1.15 below), we know from
Section 6.4 that the full transformation group
is the direct product of these two groups, 𝒯7 ×
{E, sup*}, which is isomorphic to 𝒞7 × 𝒞2. This is
not the dihedral group 𝒟7 but rather a commuta-
tive group of order 14, acting in simply transitive
fashion on gtriad ∪ gsev. (This group is in fact
cyclic; see Exercise 13.1.16.)
A generic triad can give rise to a seventh
chord not only by superposition but also by sub-
position (sub), which extends its stack of thirds
downward (CEG → ACEG). The four functions
FIGURE 13.1.14 (a) A double-​circle diagram of sup, sub, sup−1, and sub−1 are all different: the
generic triads and seventh chords; (b) a Cayley first two are functions from gtriad to gsev
diagram of a transformation group on gtriad ∪ while their inverses are functions from gsev to
gsev generated by t2 and sup*; (c) an expanded gtriad. As with superposition, subposition may
diagram showing sup* and sub* be combined with its inverse to create a trans-
formation sub* defined on all of gtriad ∪ gsev.
The symmetric relationship between superpo-
We introduced double-​circle spaces in Section sition and subposition, and between sup* and
4.1 in part as a way to display major and minor sub*, is highlighted by offsetting the two circles
triads in the same space. Though generic space as in Figure 13.1.14c. This figure is the graph of
lacks the major-​minor distinction, double-​circle a voice-​leading distance function on the space
diagrams may be put to other uses, for instance gtriad ∪ gsev; see Exercise 13.1.20.

Diatonic Spaces • 539

Exercise 13.1.15 gtriad ∪ gsev) isomorphic to the group 𝒯7
(a) Show that if X ∈ gtriad, then sup(t1(X)) =​ × {E, sup*} discussed above?
t1(sup(X)), and also that if Y ∈ gsev, then (c) Complements are defined for all subsets
sup−1(t1(Y)) =​ t1(sup−1(Y)). of gpc, but for some purposes it is
(b) Explain how Figure 13.1.14a illustrates convenient to consider functions defined
relationships like those in part (a) but with on triads and seventh chords separately.
t2 in place of t1. Write comp: gtriad → gsev for the
(c) Explain why the result of either (a) or (b) function that maps each generic triad to
implies that sup* commutes with every its seventh-​chord complement. Show that
  generic transposition tn. comp =​sub ● t1 =​sup ● t6, and write similar
equations for comp−1 (a function from
Exercise 13.1.16 Because 7 and 2 are coprime,   gsev to gtriad).
we know from Section 6.4 that the direct product
𝒞7 × 𝒞2, the group whose Cayley diagram appears Exercise 13.1.20
in Figure 13.1.14b, must be isomorphic to 𝒞14, (a) Show that third-​related seventh chords
a cyclic group. Identify a single transformation may be connected by single-​step voice
that
   generates the group in the figure.11 leading—​that is, by a voice leading in
which only one of the four voices moves by
Exercise 13.1.17 a single step.
(a) Draw a double-​circle diagram similar to (b) Show that a triad and seventh chord related
Figure 13.1.14b in which triads and seventh by sup* or sub* may be connected by single-​
chords related by sub* (rather than sup*) are step voice leading in four voices, with an
aligned radially. appropriate doubling in the triad.
(b) Do the equations from Exercise 13.1.15a (c) Parts (a) and (b) above, together with our
hold if sup is replaced with sub? previous observations about triadic voice
(c) Your picture in (a) is the Cayley diagram of leading, show that adjacent chords in Figure
a transformation group on gtriad ∪ gsev, 13.1.14c are always related by single-​step
generated by t2 and sub*. Is this group voice leading. No other pairs of chords in
isomorphic to the group 𝒯7 × {E, sup*} gtriad ∪ gsev are so related. Using the
  discussed above? techniques developed in Section 12.4, we
may therefore regard the distance function
Exercise 13.1.18 defined by this graph as the generic voice-​
(a) Show that sup =​t2 ● sub =​sub ● t2. leading distance on gtriad ∪ gsev. What is
(b) Show that sup−1 =​ t5 ● sub−1 =​sub−1 ● t5. the greatest generic voice-​leading distance
(c) Is the equation sup* =​t2 ● sub* =​sub* ● t2 between any two chords in this space?
correct? The functions in this equation are Which pairs of chords are separated by this
defined on gtriad ∪ gsev.   maximum distance?13
(d) Write equations analogous to those in
(a) and (b), reversing the roles of sub and Exercise 13.1.21 Expand Figure 13.1.14c to
sup—​that is, expressing sub and sub−1 a larger diagram with five concentric circles
as compositions of sup and sup −1 with showing generic tertian chords of cardinalities
appropriate transpositions. 1 through 5. The innermost circle contains the
(e) Is sub* an element of the group generated seven generic pitch classes; the second circle
by t2 and sup*? Is sup* an element of the contains the generic thirds CE, EG, …; the next
group generated by t2 and sub*? Are the two two circles are the triads and seventh chords;
  groups the same?12 and the outermost circle consists of generic
ninth chords such as CEGBD. Edges connect
Exercise 13.1.19 chords in adjacent circles whenever they are
(a) In generic space, the complement of a triad related by superposition or subposition; for
(such as CEG) is a seventh chord (BDFA). example, EG is adjacent to E, G, CEG, and
Construct a double-​circle space in which EGB, as well as to CE and GB in its own circle.
triads are aligned with their seventh-​chord Arrowheads may be omitted. Adjacent circles
complements. should be offset like the two circles in 13.1.14c,
(b) Is the group generated by t2 and the but EG in the second circle should align radially
complement transformation (acting on with CEGB in the fourth. Drawn symmetrically,

540 • E x p l o r i n g M u sical S p aces

the complete picture should resemble a large with mod-​12 pitch-​class sets, as the smaller size
blossom with seven petals. Does the distance of generic space limits the possibilities consider-
function defined by the graph as a whole ably. In mod-​12 space, as noted in Section 6.5,
describe
   a voice-​leading distance?14 there are 4,096 pitch-​class sets, which are parti-
tioned into 352 orbits (T-​classes) by the action
There are, of course, other kinds of generic of the transposition group 𝒯12, or into 224 orbits
pitch-​class sets—​subsets of gpc—​besides triads, (the usual set classes, or TI-​classes) by the action
seventh chords, and the other tertian chords of the dihedral group 𝒯ℐ12. In mod-​ 7 space,
mentioned in the last exercise above. The vari- the corresponding numbers are dramatically
ety of such sets, however, is much smaller than smaller: there are only 128 pitch-​class sets, 20
we might expect on the basis of our experience T-​classes, and 18 TI-​classes.15

FIGURE 13.1.22 The generic set classes (after Clough)

Diatonic Spaces • 541

 The 224 mod-​12 TI-​classes are tabulated in to the tetrachordal complements of these two
Straus 2016 and other texts on post-​tonal theory. trichords, which are inversionally related and
The complete roster of mod-​7 T-​ and TI-​classes, share an OPTIC normal form while their OPTC
first enumerated by Clough, appears in Figure normal forms differ. These four T-​classes and
13.1.22.16 Each of the clock diagrams in the first two TI-​classes are the only exceptions to the
column of this figure shows two complementary generalizations in the preceding paragraph, and
T-​classes of sets in gpc: a set of cardinality ≤ 3 in they explain why the ten circles in Figure 13.1.22
the pitch classes shown as small black circles, and account for twenty T-​classes but only eighteen
its complement, a larger set, in white circles. The TI-​classes.
last clock diagram shown, for example, depicts
generic triads and seventh chords: the black Exercise 13.1.23
pcs form a triad such as CEG or any of its rota- (a) Verify that the generic T-​classes [0, 1, 3,
tions, while the white pcs form a seventh chord 4]OPTC, [0, 1, 3, 5]OPTC, [0, 1, 2, 3, 5]OPTC,
such as BDFA. The ten circles thus depict twenty and [0, 1, 2, 4, 5]OPTC are generated by the
T-​classes, each labeled with OPTC and OPTIC intervals shown for them in the last column
normal forms. Every generic pitch-​class set is a of Figure 13.1.22.
rotation of one of these twenty sets.17 (b) Explain why every generated set in
If the sets in the shaded area of Figure 13.1.22 generic space must be inversionally
are disregarded for the time being, the remain- symmetric. Does this property hold also in
ing sets share a number of appealing properties. chromatic space?
First, as shown by the axes drawn on the circles, (c) Explain why the complement of every
they are inversionally symmetric; consequently, generated set in generic space is also a
T-​classes are the same as TI-​classes for all these generated set. Does this property hold in
sets, and OPTIC normal forms match OPTC nor- chromatic space?
mal forms. Moreover, each of these sets is a gen- (d) Apart from the trivial cases of the empty
erated set—​a partial interval cycle, constructed set and its complement (the generic
by starting from a particular note k and mov- aggregate), no generic set class possesses
ing repeatedly through a single generic interval transpositional symmetry. That is, no
g to obtain the other notes (k +​ g), (k +​2g), … nonempty proper subset of gpc is invariant
(the addition performed mod 7). For example, under any mod-​7 transposition operator
the set {0, 1, 4} is generated from generic pitch tn other than t0. Use the fact that 7 is a
class k =​1 by generic interval g =​3: the gener- prime number to explain why nontrivial
ated order for the set is 1–​4–​0, each note after transpositional symmetry is impossible in
the first being obtained by adding 3 (mod 7) to generic space.
its predecessor. (Musically, this means that a (e) Figure 13.1.22 also shows that no
chord such as CDG can be written as a quartal nonempty proper subset of gpc is
chord—​a stack of fourths—​in generic space.) symmetric about more than one axis.
The description columns show that all of these Explain why this must be the case, again
sets are easily described: besides the trivial sets   using the fact that 7 is prime.18
of cardinalities 0 and 1 and their 6-​and 7-​note
complements, the other sets include the generic The prevalence of the properties noted
interval classes identified as gic 1, 2, and 3 (each above in generic space differs markedly from
a set of cardinality 2); generic clusters (scale seg- the prevalence of the same properties in chro-
ments, generated by gic 1); tertian chords (tri- matic space. Sixteen of the eighteen TI-​classes
ads, seventh chords, and others from Exercise in gpc (89 percent) are inversionally sym-
13.1.21, generated by gic 2); and quartal/​quintal metric and generated; in pc, only 96 of 224
chords (generated by gic 3). TI-​classes (43 percent) are inversionally sym-
The sets in the shaded portion of the figure metric, and only 33 TI-​classes (15 percent) are
differ from the others in several ways. The two generated. In chromatic space 63 TI-​classes
trichords here, with OPTC normal forms [0, 1, are inversionally symmetric but not generated
3]OPTC and [0, 2, 3]OPTC, are not inversionally (for example, prime form 0134), a combination
symmetric but are related to each other by inver- that does not occur at all in generic space. Also,
sion; they therefore share the OPTIC normal sixteen chromatic TI-​ classes exhibit trans-
form [0, 1, 3]OPTIC. No generic interval gener- positional symmetry, and fifteen of those are
ates either of these sets. Similar remarks apply inversionally symmetric about more than one

542 • E x p l o r i n g M u sical S p aces

axis—​properties that, as noted in the above indicates the smooth voice leading between
exercise, are impossible in mod-​7 space except third-​
   related triads.20
in trivial cases.
The OPTIC spaces were presented in Chapters
Exercise 13.1.24 10 and 11 on the assumption of an underlying
(a) Write a generic interval-​class vector for mod-​12 space, but OPTIC spaces may be con-
each of the TI-​classes in Figure 13.1.22. structed for other moduli as well.21 The choice
This will be a three-​place vector indicating of modulus has no bearing on the topological
the number of occurrences of gic 1, 2, and 3 structure of the continuous spaces. In com-
within the set. parison with the mod-​12 spaces, mod-​7 OPTIC
(b) What is the relationship between the spaces are scaled down by a factor of 7 12 in every
generic interval-​class vector of a set and dimension but otherwise take the same forms,
that of its complement?19 determined only by the number of voices in
(c) Are there any Z-​related sets in generic the chords and the choice of OPTIC relations,
space? That is, do two different TI-​classes not by the cardinality of the underlying scale.
  ever share the same interval-​class vector? Figure 13.1.26 illustrates with the mod-​7 ver-
sion of two-​voice OP-​space, which may be com-
Exercise 13.1.25 Construct a voice-​leading pared with the mod-​12 space from Figure 11.2.6.
graph analogous to Figure 3.3.1 for generic The Möbius-​strip structure of the mod-​12
trichordal set-​class space. Include the four space remains intact in its mod-​7 counterpart.
three-​note OPTIC-​classes shown in Figure As in the earlier figure, horizontal lines indicate
13.1.22 as well as the four multiset classes fixed sum classes, vertical lines indicate inter-
[0, 0, 0] through [0, 0, 3]. Vertices in the graph val classes, and motion in the vertical direction
should be adjacent whenever the set classes indicates transposition. The numerical values,
they represent are joined by single-​generic-​step however, have changed. In the mod-​7 space the
voice leading. Include loops on any set classes sum classes range continuously from 0 to 7 (not
that can lead to themselves by single-​step 0 to 12); interval classes range only from 0 to 31 2
voice leading; for example, a loop on [0, 2, 4] (not 0 to 6); and the transposition intervals

FIGURE 13.1.26 Two-​voice mod-​7 OP-​space

Diatonic Spaces • 543

shown at the left side of the figure likewise range pitch space, producing the smoothest
only from t0 to t3.5 (rather than from T0 to T6). possible voice leading.
In the mod-​12 space, the dyad at the lower-​right
corner was [6, 6]OP, or F♯F♯; the correspondingly GC –​CF –​FB –​BE –​EA –​AD –​DG –​GC
positioned dyad in mod-​7 space is [31 2 , 31 2 ]OP,
which cannot be expressed in note names as it is (b) Trace the following similar progression in
not a lattice point in generic space. Lattice points chromatic OP-​space (Figure 11.2.6):
(that is, points whose mod-​7 OP normal forms
have integer coordinates) are depicted as black GC –​CF –​FB♭ –​B♭E♭ –​ ⋯ –​AD –​DG –​GC
circles, labeled with normal forms written in
generic note names; [31 2 , 31 2 ]OP and a few other (c) Comment on the differences between the
non-​lattice points are shown as open circles, progressions in (a) and (b).24
their OP normal forms indicated numerically. (d) Also trace the following progression in
We have emphasized several times that mod-​7 OP-​space, assuming that thirds
generic and diatonic spaces, at least in our usual alternate with sixths:
conception of them, are discrete by their very
nature: there are no generic notes between C GB
   –​CE –​FA –​BD –​EG –​AC –​DF –​GB
and D. A continuous space such as that of Figure
13.1.26 is therefore in some sense a musical fic- As in the mod-​12 case, each horizontal line
tion. Points other than lattice points have no across the continuous space of Figure 13.1.26
obvious musical interpretation, and the reader represents a sum class and contains every inter-
may justifiably wonder what to make of a num- val class twice, except that the central interval
ber such as 31 2 in a mod-​7 normal form or index class occurs only once. Each row of lattice points,
of transposition.22 Nevertheless, the fiction is a however, contains each integer-​valued generic
useful one. As in the mod-​12 case, a picture of interval class (gic 0, 1, 2, and 3) exactly once. This
the continuous space is automatically a picture property is ensured by the feature of sum classes
of the discrete space as well, and displaying an of generic multisets demonstrated in Theorem
underlying continuous space in this manner 13.1.4: the seven generic transpositions of a two-​
enables us to see relationships among the dis- note set range through the representatives of an
crete points that are hard to visualize in any interval class, and the theorem states that those
other way. (Because we are considering continu- seven transpositions exhaust the seven sum
ous spaces in the remainder of this section, when classes, once each. The corresponding property
restricting our attention to lattice points we will does not hold in mod-​12 space (Figure 11.2.6),
sometimes refer to “integer-​ valued” generic where some rows of lattice points contain only
pitch classes, pc sets, and so on.) even interval classes and other rows only odd
Even as the essential structure of the contin- interval classes.
uous mod-​7 OP-​space mimics that of the mod-​ More generally, Theorem 13.1.4 says that
12 space, the discrete spaces display notable when only integer-​valued multisets, transposi-
differences, reflected in the different distribu- tions, and sum classes are considered, each OPT-​
tions of lattice points in the two pictures.23 The class occurs exactly once in every sum class.
central axis of mod-​12 space consists of tritones This principle is useful to keep in mind in the
(ic 6), but because 7 is an odd number, no generic two challenging exercises below, which lead the
interval divides an octave in half and no lattice reader through the construction of a few addi-
points appear on the central axis of mod-​7 space, tional examples of mod-​7 OPTIC spaces.
an axis that corresponds to a non-​integer inter-
val class (gic 31 2 ). The following exercise explores Exercise 13.1.28
dyadic chord progressions in the vicinity of the (a) Determine the twelve integer-​valued
central axis in both spaces. generic trichordal multisets of sum class 0,
expressing them using note names.25
Exercise 13.1.27 (b) Draw the sum-​class-​0 cross section of three-​
(a) Trace the following dyadic circle-​of-​fifths voice mod-​7 OP-​space, analogous to the
progression in the mod-​7 OP-​space of mod-​12 picture in Figure 11.3.1. Show the
Figure 13.1.26. Assume that the dyads locations of the twelve lattice points from
alternate between fifths and fourths in (a) within this picture.26

544 • E x p l o r i n g M u sical S p aces

(c) Draw a picture of all of three-​voice mod-​7 Relationships among generic, diatonic, and
OP-​space, a Möbius prism analogous to chromatic scales may be conceived in several
Figure 11.3.4, making sure you understand different ways, some of which are illustrated in
how the sum-​class layers relate to each Figure 13.2.1 using the three-​flat diatonic col-
other. Label all lattice points appearing on lection dpc(−3) as an example. The three circu-
the various edges of the prism (the three lar pictures in the top row of this figure depict
vertical edges and the boundaries of the top generic, diatonic, and chromatic pitch-​ class
and bottom triangles).27 space—​gpc, dpc(−3), and pc—​ in the scalar
(d) One of the mirror boundary walls of ordering. Those in the bottom row depict the
this prism is a rectangle with CCC at two circle-​
of-​
fifths ordering of the same spaces,
opposite corners. Draw this rectangle gfifth, dfifth(−3), and fifth. In the center is
carefully, showing the locations of all lattice
spelled pitch-​class space spc, the line of fifths.
points that lie on this face.28 Generic (mod-​ 7) pitch-​
class numbers are
(e) Locate the seven generic triads in this space,given in the generic spaces gpc and gfifth. As
and graph a triad cycle descending by thirds, noted in Exercise 13.1.3, the gpc numbers in the
CEG –​ACE –​FAC –​DFA –​BDF –​GBD –​EGB two generic spaces are related by the mod-​7 mul-
–​CEG. Compare this picture of the generic tiplication operators m4 and m2, as shown by the
cycle with your picture of the diatonic triad pair of large curved arrows at the left side of the
cycle descending by thirds, C –​a –​F –​d –​bofigure. At right, the usual mod-​12 pc numbers
  –​G –​e –​C, from Exercise 11.3.9.29 are shown in pc and fifth, related by M7 in both
directions. In the middle, it would be possible to
Exercise 13.1.29 use either gpc or pc numbers to label the notes
(a) Construct a picture of mod-​7 three-​voice of the diatonic spaces dpc(−3) and dfifth(−3),
OPTI-​space, analogous to Figure 11.4.12. but in the figure these spaces are labeled instead
Label the eight lattice points in the figure with spelled pitch-​class numbers from spc. This
with their OPTI normal forms.30 numbering distinguishes, for example, E♭ (spc
(b) Compare your picture of OPTI-​space with −5) from E♮ (2), a distinction not recognized by
the voice-​leading graph you constructed gpc numbers (both are gpc 2), and it also distin-
in Exercise 13.1.25. Each of the loops in guishes E♭ (−5) from D♯ (7), not distinguished by
the earlier graph should determine a path pc numbers (both are pc 3).
through the space, reflecting off a mirror The spaces dpc(n) and spc do not recognize
  boundary; identify these paths.31 enharmonic equivalence; the note names shown
in these spaces are spelled in the only correct way
13.2 DIATONIC SCALES IN for notes in dpc(−3). Every diatonic collection
CHROMATIC SPACE dpc(n) is a consecutive seven-​note segment of
the line of fifths, comprising spc numbers from
The generic properties discussed in the previ- n − 3 to n +​3. Thus the spcs in dpc(−3), the exam-
ous section are premised on nothing more than ple in the figure, range from −6 to 0—​that is,
mod-​7 structure; they do not require that the from A♭, the leftmost, or flattest, note of dpc(−3),
seven-​note scale be embedded within a larger to D, the rightmost, or sharpest. In the chromatic
scale or musical space. But of course we are spaces pc and fifth, note names are shown for
accustomed to thinking of a diatonic scale as convenience only; pitch class 3 could be labeled
a subset of a chromatic scale. That is, we often D♯ as well as E♭. The designation edpc(9) and the
associate generic space gpc with a diatonic space solfège syllables appearing in some of the spaces
dpc(n), which in turn we identify with a seven-​ will be explained below.
note subset of pc. It is the subtle and complex A diatonic collection may be interpreted
relationship between a diatonic scale and an in several ways in relation to the other spaces
underlying chromatic space to which we next shown in Figure 13.2.1. We may regard dpc(n) as
turn our attention. This section presents some a contiguous subset of spc, or we may interpret
preliminary observations and definitions, to be it in either a generic (mod 7) or a specific (mod 12)
followed by discussions of signature transfor- sense. The projection functions γ: spc → gpc
mations in Section 13.3 and the principle car- and π: spc → pc were introduced in Section 2.6
dinality equals variety and related properties in as the quotient maps QG and QE of the relations
Section 13.4. of generic and enharmonic equivalence. That

Diatonic Spaces • 545

FIGURE 13.2.1 The diatonic collection dpc(−3) in relation to various spaces

is, γ maps any spelled pitch class to the generic spcs: spcs −9 through −3 in the former case, +​3
pitch class with the same letter name (stripping through +​9 in the latter. These collections are, of
the spc of its accidentals, if any), while π maps course, enharmonically equivalent in the usual
any spc to the mod-​12 pitch class corresponding sense of the term, because spcs differing by 12
to that spc and all others enharmonically equiva- (or multiples of 12) are always enharmonically
lent to it. For example, γ(E♭) is the generic pitch equivalent; the projection π maps both dpc(−6)
class E, and π(E♭) is pitch class 3. We will at times and dpc(+​ 6) to the same seven-​ note subset
identify dpc(n) with gpc, via γ, and at other of pc, theoretically along with infinitely many
times with a subset of pc, via π; in a minor abuse other spaces whose indices differ by multiples
of notation we will not always make explicit ref- of 12, such as dpc(−18) and dpc(+​30). It is use-
erence to the functions γ and π. ful to introduce a different name for these dia-
Allowing for multiple sharps and flats, the tonic subsets of pc, predicated upon enharmonic
number of distinct collections dpc(n) is theo- equivalence; we refer to them as the enharmonic
retically infinite. The six-​flat and six-​sharp col- diatonic pitch-​class spaces edpc(n). There are only
lections dpc(−6) and dpc(+​6) contain different twelve of these enharmonic spaces, and their

546 • E x p l o r i n g M u sical S p aces

index numbers are integers mod 12. Thus the A diatonic scale may be interpreted generi-
space edpc(6) is associated with both the six-​ cally via γ, and conversely a generic space may
flat and six-​sharp diatonic collections. In Figure be interpreted as a diatonic scale. We introduce
13.2.1, the enharmonic space associated with the diatonic interpretation functions δn: gpc →
dpc(−3) is labeled edpc(9); theoretically this dpc(n) for this purpose: for any integer n, δn
subset of pc could represent a nine-​sharp dia- maps every generic pitch class to the note in
tonic collection (the D♯-​major scale) as well as the dpc(n) with the same letter name. The function
three-​flat collection. The notes of an enharmonic δn may be regarded as the application of a key
space edpc(n) are not spelled pitch classes like signature with the appropriate number of sharps
the notes of dpc(n), but ordinary pitch classes, (for positive n) or flats (for negative n). Thus δ−3,
which are enharmonic equivalence classes of shown in the figure, maps the gpcs B, E, and A to
spcs. For convenience we may sometimes refer the corresponding flatted notes in dpc(−3) while
to these notes by letter names, but in an enhar- mapping the other four gpcs to notes with no
monic space such as edpc(9), as in the space pc sharps or flats. As the figure suggests, for a fixed
of which it is a subset, it makes no difference value of n the functions δn: gpc → dpc(n) and γ:
whether we call a note E♭ or D♯. Similarly, we will dpc(n) → gpc are inverses.
sometimes use familiar scale names such as “E♭
major,” but this too is for convenience only: a Exercise 13.2.2 If k is a mod-​7 gpc number
collection edpc(n) implies nothing either about and n is any integer, the spc number of δn(k) is
the spelling of its notes or about a tonal center the number congruent to 2k +​5 mod 7 that lies
for the collection. closest to n. Verify that this formula matches
Intervals between notes of dpc(n) may be   above description of δ−3 for all seven gpcs.32
the
measured in several different ways. The generic
interval function intgpc may be used to count As sets, the circular spaces on the bottom row
steps of the scale as if the notes were elements of Figure 13.2.1 are identical to those in the top
of gpc; alternatively, intervals may be measured row (being merely rearrangements of the same
specifically (chromatically), by applying the mod-​ objects), so γ and π may also be regarded as func-
12 interval function intpc to the corresponding tions from spc to gfifth and fifth, respectively,
notes in pc. We will sometimes write intg and and δn as a function from gfifth to the diatonic
ints for the generic and specific interval func- fifths space dfifth(n).33 Relationships between
tions on a diatonic space. Thus in dpc(−3) we the diatonic scale in spc and the various circles
may write intg(A♭, F) =​5 and ints(A♭, F) =​9. The of fifths are particularly suggestive. The map-
equation intg(A♭, F) =​5 really means intgpc(γ(A♭), pings γ: spc → gfifth and π: spc → fifth simply
γ(F)) =​intgpc(A, F) =​intgpc(5, 3) =​5, correspond- wrap the line of fifths into a circle: conjoining
ing to the generic transposition t5(A) =​F; the every seventh note produces gfifth, while con-
equation ints(A♭, F) =​9 really means intpc(π(A♭), joining every twelfth note produces fifth. The
π(F)) =​intpc(8, 5) =​9, corresponding to the chro- wrapping preserves the adjacency of fifth-​related
matic transposition T9(A♭) =​F. The line of fifths notes, so the seven notes of a diatonic collection
offers yet another way to measure the interval remain clustered in the mod-​12 circle of fifths.
between A♭ and F: intspc(A♭, F) =​3, corresponding The same cannot be said of the chromatically
to the spc transposition τ3(A♭) =​F. (As defined in organized circle of mod-​12 space pc at the top
Section 2.6, τk shifts every spelled pitch class by right: rather than clustering together, the pitch
k places in the line of fifths, a rightward shift if classes of edpc(n) form in pc the familiar pat-
k > 0 or leftward if k < 0.) The space dpc(n) is a tern of whole and half steps characteristic of the
modular interval space with the generic interval diatonic scale.
function intg. Via either ints or intspc, dpc(n) is
not by itself an interval space because the homo- Exercise 13.2.3 A diatonic scale in fifth has
geneity condition is not satisfied: there is no the same appearance as a cluster of prime
note x in dpc(−3) for which ints(A♭, x) =​1, nor form 0123456—​a chromatic heptachord—​in
is there a y for which intspc(D, y) =​1. But dpc(n) chromatically arranged pitch-​class space pc.
is a subset of spc and edpc(n) is a subset of pc, Verify that M7, the multiplication operator
and both of these larger spaces are proper inter- that transforms the space pc to fifth and
val spaces in which the missing notes (A in both vice versa, also transforms a seven-​note
cases) may be found. chromatic cluster to a diatonic collection and

Diatonic Spaces • 547

vice versa. What does a chromatic heptachord skill called position-​finding in Browne 1981. For
look
   like if drawn in fifth? example, if we hear a note x with the property
that x, T1(x), and T7(x) all belong to some edpc(n),
Several familiar properties of diatonic col- we can deduce that x is the note corresponding
lections emerge in these constructions. First, to scale degree 3̂ of a major scale, or the tonic of
as a subset of pc, edpc(n) is a generated set. We a phrygian scale. Position-​finding of this kind is
observed in Section 13.1 that generated sets are not possible in transpositionally symmetric scales
proportionally much rarer in mod-​12 space than such as whole-​tone or octatonic. Intervallic dis-
in mod-​7, but the diatonic collection is such a tinguishability ensures that the seven modes of a
set: the generated order matches the order diatonic scale all have different interval arrange-
of spcs in the line of fifths. In Figure 13.2.1, ments; the relationship between the number of
edpc(9) is generated from pitch class 8 (A♭, the modes of a collection and its transpositional sym-
leftmost note of dpc(−3) in the line of fifths) metry was explored in Exercise 6.5.10.
by the generating interval 7 (the perfect fifth); Descriptions such as “the note correspond-
the pitch-​class numbers in generated order are ing to scale degree 3̂ of a major scale, or the
8, (8 +​7), (8 +​2 ∙ 7), …, (8 +​6 ∙ 7) (mod 12). tonic of a phrygian scale” are awkward. It is use-
Alternatively, edpc(9) may be generated from ful to have a way of referring to the position of a
pitch class 2 (D, the rightmost note) by the gen- note within a diatonic collection whose tonic we
erator 5, the inverse of 7 in the interval group do not know. Solfège syllables may be used for
ℤ12, corresponding to a right-​ to-​left traversal this purpose: whatever the tonic and mode, the
of the line of fifths. The generation process also note just described is mi.35 If the scale is major,
ensures that edpc(n) is inversionally symmetric: then mi is 3̂; if it is minor (aeolian), then mi is 5̂;
its symmetry in the line of fifths or the circle of if it is phrygian, then mi is 1̂—b ​ ut the syllable
fifths is obvious, and its symmetry in pc will be unambiguously describes the position of a note
discussed further below. (Recall also Exercise within a diatonic collection even if the mode
13.1.23b.) and tonic are unspecified. In dpc(−3), do is E♭;
Though inversionally symmetric, edpc(n) syllables are provided for dpc(−3) and edpc(9)
is not transpositionally symmetric. Transposi­ in Figure 13.2.1.
tional symmetry is an uncommon property, and Each of the seven diatonic positions has its
according to Exercise 6.5.8b a set of cardinality own characteristics. Along the line of fifths, the
7 in pc can never be T-​symmetric. A few scales notes of a diatonic collection always proceed
of other cardinalities, notably the whole-​tone in the order fa–​do–​sol–​re–​la–​mi–​ti from left to
and octatonic, are T-​ symmetric. The lack of right—​the generated order. The framing notes
such symmetry for diatonic scales carries the fa and ti form the collection’s unique tritone;
important consequence of intervallic distinguish- fa is the flattest note of the collection and ti
ability: any element of a diatonic collection can the sharpest. The centrally positioned re is the
be distinguished from any other by its mod-​12 note about which the collection is symmetric,
intervallic relationships to the other notes in and the note whose spelled pitch-​class number
the collection. More precisely, if x and y are two matches the index number n of the collection
different elements of edpc(n), then there must dpc(n). Re is the dorian tonic, and dorian is the
be some interval i such that Ti(x) ∈ edpc(n) but only mode whose interval arrangement is sym-
Ti(y) ∉ edpc(n). Intervallic distinguishability fol- metric when written as an eight-​note scale from
lows from the lack of transpositional symmetry 1̂ to 1̂. (We have remarked on aspects of this
according to Exercise 6.5.9, but it may also be symmetry previously, for instance in Example
seen as a consequence of the fifth-​generation 7.6.10, an instance of inversion in the Allegretto
procedure. Suppose that x lies to the left of y in of Brahms’s First Symphony.) In pc and fifth, if
the line of fifths, both within the span of dpc(n). an axis of symmetry passes through one note, it
Then clearly there is some number k such that passes also through the tritone relative of that
τk(x) lies within dpc(n) but τk(y) does not. As note (recall Figure 5.5.4c); in either pc or fifth,
shown in Section 2.6, the pitch-​class transposi- edpc(n) is symmetric not only about re but also
tion corresponding to τk is T7k, so it follows that about the pitch class re +​6, a note not present
T7k(x) ∈ edpc(n) but T7k(y) ∉ edpc(n).34 in edpc(n) (it is the note that falls chromatically
Musically, intervallic distinguishability is what between sol and la). The same axis of symmetry
enables us to infer a note’s location within a dia- for edpc(n) appears in both pc and fifth; when
tonic scale by hearing intervallic relationships—​a n is odd, as in the example of dpc(−3), the axis

548 • E x p l o r i n g M u sical S p aces

notes re and re +​6 trade places between the two set X, then as long as 1 ≤ i ≤ 5, the sets X and Ti(X)
pictures. have exactly k common tones.39 If the set is a dia-
tonic collection, then a transposition by fifth—​
Exercise 13.2.4 As an inversionally symmetric τ1 or τ−1 in the line of fifths, T7 or T5 in pc—​is
set, edpc(n) must be invariant under some a transposition to a closely related key, such as
inversion operator Ik in pc. Write a formula for from edpc(9) (E♭ major) to edpc(10) (B♭ major).
k in
  terms of n.36 If two collections in this relationship are viewed
in spc or in fifth, it is clear that they share six
Exercise 13.2.5 Suppose a chord is written on common tones, in agreement with the 6 in the
a five-​line staff with no accidentals. Then reflect ic5 position of the interval-​class vector. Diatonic
the notation about a horizontal axis, so that a collections separated by two fifths—​τ2, τ−2, T2, or
note originally on the bottom line is now on the T10, for example edpc(9) and edpc(11)—​share
top line and so on. Whatever clef one chooses five common tones, corresponding to the 5 in
to place on the staff, the new chord is always an the ic2 position; collections separated by three
inversion of the original chord in generic space. fifths (ic3) share four common tones; and so on.
Give an example to show that if the staff is read With each successive shift around the circle of
in treble clef, one or more interval qualities may fifths, each new diatonic collection has one fewer
change in the inversion, so that the new chord common tone with the original collection, pro-
fails to be an exact inversion of the original ducing a clean hierarchy of common-​tone rela-
chord in chromatic space. If the staff is read in tionships among keys.
bass clef, however, the inversion will always be The pattern breaks with the tritone transpo-
exact
   in chromatic space as well. Why is this?37 sition. If the same relationship were to persist
through transposition by six fifths (τ6 or T6), then
Another intervallic property of the diatonic a diatonic collection and its transposition by T6
collection, much rarer than the symmetry just would share only a single common tone. It is
discussed, involves the collection’s interval-​class true that a diatonic collection contains only one
vector, the enumeration of the occurrences of instance of ic6, and it is true that τ6-​related dia-
each of the six interval classes ic1–​ic6 within a tonic collections dpc(n) in spc share only a single
pc set. (Recall from Section 12.2 that interval common tone, but this is not true of T6-​related
classes are actually distances, not intervals, in diatonic collections edpc(n) in pc. For example,
the space pc.) The vector of the diatonic collec- in the line of fifths, properly spelled A-​major and
tion—​that is, the vector of any set edpc(n) as a E♭-​major collections share only the note D, but
subset of pc—​is 254361, which has the remark- edpc(3) and edpc(9) share pitch classes 2 (D)
able property that each interval class occurs a and 8 (G♯/​A♭). In fact, it is not possible for two
different number of times. A set with this prop- seven-​note pitch-​class sets to have only a single
erty is known as a deep set.38 note in common (see Exercise 2.1.4b). Because
The deep property is readily explained the tritone divides the octave in half, each occur-
through the generation of diatonic scales by rence of ic6 in any set brings about not one but
fifths. In the space fifth (Figure 13.2.1, lower two common tones between the set and its T6
right), where distfifth(C, G) =​1, distfifth(C, D) transposition in pitch-​class space. In fifth, if
=​2, and so on, it is apparent that a diatonic col- the seven-​note cluster of a diatonic collection is
lection contains six pairs of notes separated by rotated halfway around the circle, the new clus-
distance 1, five pairs at distance 2, and so on, up ter overlaps the original at both ends, resulting
through a single pair at distance 6; we may say in two shared pitch classes: tritone-​related dia-
that the fifth-​distance vector of the diatonic col- tonic collections share a common tritone, with
lection is 654321. These are the same six digits the roles of fa and ti interchanged.
as in the interval-​class vector and they enumer-
ate pairwise distances among the same notes, Exercise 13.2.6
but the distances are measured differently and (a) In the notation from Section 12.2, show
listed in a different order. Exercise 13.2.6 below that if x and y are pitch classes such that
explores the details. distfifth(x, y) =​ k, then distpc(x, y), the interval
The deep property of the diatonic collection class formed by x and y, is equal to |7k|12.
has important ramifications involving key rela- (b) Use the result of (a) to explain why a set
tionships. Generally, if the number k appears in with fifth-​distance vector 654321 has
the ith position of the interval-​class vector for a   interval-​class vector 254361.

Diatonic Spaces • 549

Exercise 13.2.7 Well-​formedness may appear straightforward
(a) Verify that a chromatic heptachord of in the circle of gpc, but the property is a subtle
prime form 0123456 (as in Exercise 13.2.3) one. For one thing, well-​formedness does not
has interval-​class vector 654321 and preclude other intervals from spanning the same
fifth-​distance vector 254361—​the same number of scale-​steps as the generating interval:
two vectors as the diatonic collection, but in dpc(−3) the generic interval from D to A♭ is
interchanged—​and is therefore another also 4, even though D–​A♭ is not an instance of
example of a deep set. the generating interval (that is, D and A♭ are
(b) Besides diatonic collections and chromatic not adjacent in the line of fifths). Moreover,
heptachords, there are only two other many generated sets, even some fifth-​generated
TI-​classes of deep sets in pc. Like the two sets, are not well-​formed, as the following exer-
collections just mentioned, they are related cise shows.
to each other by M7. Their interval-​class
vectors are 543210 and 143250. What Exercise 13.2.8
  are they? (a) A diatonic (Guidonian) hexachord such as
CDEFGA is a six-​note fifth-​generated set.
Several important properties of diatonic Draw a circle with these six notes spaced
scales involve relationships between chromatic equally around the circle in scalar order.
and generic interval and subset structure. Such Draw line segments across the circle to
properties include Carey and Clampitt’s formula- connect notes in generated order (F–​C–​G–​
tion of well-​formed scales, Clough and Myerson’s D–​A–​E), and observe that the segments
property cardinality equals variety, and Clough are not all equal in length, because not all
and Douthett’s maximal evenness, each of which instances of the generating interval span
we will examine in this chapter or the next. The the same number of steps within this
well-​formed property involves fifths—​the gen- scale. Also write out the mapping of mod-​6
erating interval—​and in particular the relative pitch-​class numbers that relates the two
positions of fifth-​ related notes in a diatonic orderings, and show that this mapping
scale, interpreted generically. In the circle of is not an automorphism of ℤ6. Each of
gpc (or in dpc(0), identified with gpc via γ) it is these observations shows that the diatonic
evident that the intervals intgpc(F, C), intgpc(C, hexachord is not a well-​formed scale.
G), …, intgpc(E, B) are all equal to 4. If you draw (b) Consider fifth-​generated sets with five
straight line segments cutting across the circle notes (pentatonic collections such as
of gpc to connect notes in generated order (F –​C CDFGA), four notes (CDFG), and three
–​ G –​ ⋯), all such segments will be equally long. notes (CFG). Verify that two of these are
This is Carey and Clampitt’s definition: a well-​   well-​formed scales. Which one is not?
formed scale is a generated set in which every
instance of the generating interval spans the 13.3 SIGNATURE
same number of steps within the scale.40 TRANSFORMATIONS
The well-​formed property of a diatonic col-
lection captures something we have already Figure 13.3.1 shows a prototypical example of a
observed about the generic circles gpc and signature transformation in the first movement
gfifth in Figure 13.2.1. Because of the sys- of Schubert’s B♭-​Major Sonata. After eighteen
tematic intervallic relationship between the measures of B♭ major the music veers into G♭,
two circles, the fact that fifth-​related notes are shifting from dpc(−2) to dpc(−6). The phrase
always adjacent in one circle guarantees that in G♭ imitates the melody in B♭ from the outset
they are always separated by the same distance of the movement, but it has not been trans-
in the other. Because 4 and 7 are coprime, the posed. Instead, it repeats at its original regis-
multiplication operator m4, which maps pitch-​ tral position, with adjustments as needed to
class numbers in gpc to those in gfifth, is fit the new key: the notes A, C, and D are now
an automorphism of the group ℤ7, illustrat- flatted. Schubert does not write the six-​flat key
ing another characterization of well-​ formed signature, instead notating the inflections with
scales also formulated by Carey and Clampitt: a accidentals, but the diatonic context is clear and
scale is well-​formed if and only if the mapping the figure employs the signature as an analytical
from scalar order to generated order is a group shorthand. The shift of diatonic scale is modeled
automorphism.41 with the signature transformation s−4, indicating

550 • E x p l o r i n g M u sical S p aces

FIGURE 13.3.1 The signature transformation s−4 in Schubert, Sonata in B♭ Major, D. 960, I

FIGURE 13.3.2 Comparison of the signature transformation s−4 and the spc transposition τ−4 as map-
pings from dpc(−2) to dpc(−6)

that the underlying diatonic collection has When s−4 is applied to music in dpc(−2), as in
shifted four places in the flatwise direction—​the Figures 13.3.1 and 13.3.2, C changes to C♭, but if
four new flats in the key signature—​but with the s−4 is applied instead to music in dpc(0), the note
notes otherwise unchanged. C will be unaffected, since C♭ is not one of the
Signature transformations are a powerful first four flats added to the key signature. Before
tool in analyzing music that moves among dia- we can apply a signature transformation, there-
tonic collections; we encountered them briefly fore, we must know the key signature. Signature
in a discussion of changing scalar contexts transformations may be applied to single notes
in Adams’s Phrygian Gates in Example 4.5.5. within diatonic spaces, to lines (ordered sets of
Musical notation with key signatures, as in notes), chords (unordered sets), melodies as in
Figure 13.3.1, often provides the easiest way Figure 13.3.1, or combinations of these—​essen-
to grasp the behavior of these transformations, tially to anything that can be notated on a musi-
but the subtlety of their behavior benefits from cal staff with a traditional key signature and no
additional formalization. This section presents accidentals—​but the key signature must always
such a formalization, highlighting the intricate be specified.
interactions between signature transformations It is convenient to introduce a term that can
and the more familiar diatonic and chromatic be used loosely to describe musical fragments
transposition operators.42 of this sort. We will use the general term design,
Figure 13.3.2 shows dpc(−2) and dpc(−6) modified by various adjectives, to refer to any
as subsets of spc, and compares the action musical structure that can be notated on a staff
of the signature transformation s−4 with that of subject to certain restrictions. In particular:
the spc transposition τ−4. Both s−4 and τ−4 map
the seven notes of dpc(−2) to the seven notes • A diatonic design is a design all of whose notes
of dpc(−6), but they are not the same map- belong to one diatonic scale dpc(n).
ping: τ−4 shifts every note four places to the • An enharmonic design is a design all of whose
left, whereas s−4 shifts four notes seven places notes belong to one enharmonic space
to the left while leaving the other three notes of edpc(n).
dpc(−2) unchanged. The notes remaining fixed • A generic design is a design whose notes are
under the action of s−4 are E♭, B♭, and F, the notes generic pitch classes in gpc.
in the intersection of dpc(−2) and dpc(−6); the • A chromatic design is a design whose notes are
other notes of dpc(−2) must leapfrog leftward mod-​12 pitch classes in pc.
across the fixed notes to find their new homes • An spc design is a design whose notes are
in dpc(−6). spelled pitch classes in spc.

Diatonic Spaces • 551

FIGURE 13.3.3 Examples of diatonic, enharmonic, generic, and chromatic designs

The first four types are illustrated in Figure equivalence. The figure shows only a single
13.3.3. As the annotations show, several of the enharmonic design, but it appears in two differ-
designs here are nothing more than lines or ent notations corresponding to the two common
chords: ordered or unordered sets of notes from notations for the enharmonic space edpc(6),
a space of the appropriate kind. In general, how- the six-​flat and six-​sharp diatonic collections.
ever, a design may include more complex musical These notations could represent two different
textures with rhythmic notation (as in the first diatonic designs, one in dpc(−6) and the other
diatonic design shown, reproduced from Figure in dpc(+​6), but as enharmonic designs they are
13.3.1) and possibly other elements. In all of the equal. A diatonic design in dpc(n) may always
cases considered here, the notes of a design are be interpreted as an enharmonic design in the
understood to be pitch classes, not pitches; that corresponding enharmonic space edpc(n). The
is, octave equivalence is assumed, so if a design last n is a mod-​12 integer: to be more precise we
contains the note C it does not matter whether could write dpc(n) and edpc(n), where n denotes
the note is positioned on the staff as C4 or C5. the mod-​12 congruence class of n. Enharmonic
In the case of a diatonic design, the underly- equivalence of diatonic designs will be examined
ing diatonic collection dpc(n), typically signaled in greater detail below.
by a key signature, is understood to be a part The notes of a generic design are generic pitch
of the design. Of the four diatonic designs in classes (letter names) only, with no key signa-
Figure 13.3.3, three belong to dpc(−2) and one ture and no accidentals. Visually a generic design
to dpc(−4). The two three-​note diatonic lines looks like a diatonic design in dpc(0), but the two
both contain the notes B♭–​G–​F, but they are dif- are conceptually distinct in the same way that
ferent designs because they belong to two differ- the spaces gpc and dpc(0) are distinct. In Figure
ent diatonic collections. A G-​minor triad by itself 13.3.3 the space name gpc signals the generic
is not a diatonic design unless we have speci- designs. A diatonic design may be interpreted as
fied whether it belongs to dpc(−1), dpc(−2), or a generic design via the mapping γ from Section
dpc(−3), any of which is theoretically possible. 13.2, which essentially discards the original key
It is diatonic designs to which signature trans- signature. Conversely, a generic design may be
formations may be applied most directly, though interpreted as a diatonic design in any space
we will show below that their application may be dpc(n) by the interpretation mapping δn—​that
extended to enharmonic designs as well. is, by applying the appropriate signature. The
An enharmonic design is an equivalence two diatonic lines in the top row of Figure 13.3.3
class of diatonic designs under enharmonic both have as their generic interpretation the

552 • E x p l o r i n g M u sical S p aces

generic line B–​G–​F in the third row. If any of the shown previously in Figure 13.3.1, are related
three generic designs in the figure is interpreted by the signature transformation s−4. These two
via δ−2, the result is the diatonic design appear- designs have different key signatures but their
ing above it. notes are generically the same (B–​B–​A–​B–​C–​D):
Every diatonic or enharmonic design may signature transformations preserve generic pitch
be interpreted as a chromatic design by regard- classes. The two designs in dpc(−2) are related
ing its notes as pitch classes in pc rather than by diatonic transposition t2; they have differ-
in dpc(n) or edpc(n). The first chromatic design ent notes, but the diatonic transposition pre-
in Figure 13.3.3 is the chromatic interpretation serves the key signature. The two designs in the
of the enharmonic design in the figure. Spelling second row of the figure (mm. 19–​21 and mm.
is of no consequence in chromatic designs: this 347–​49) have neither the same key signature
design could equally well be spelled with flats, nor the same notes, but the notes of these two
or even with a combination of sharps and flats, designs are positioned in the same way relative
as long as its pitch classes are unchanged. The to their scales, reflected in matching solfège syl-
first of the two three-​note chromatic lines in the lables. Both begin on mi, and the first interval
figure is the chromatic interpretation of both of in both is a whole step, in contrast to mm. 0–​2,
the diatonic lines in the top row of the figure. where the first note is do and the first interval is
The second chromatic line, B♭–​G–​F♯, is not the a semitone. (Solfège syllables indicate diatonic
interpretation of any diatonic design at all, as no position as described in Section 13.2. Mi need
diatonic collection contains all its pitch classes. not always imply scale degree 3,̂ though in this
Spc designs, the last item in the above list, will figure it does because all of the excerpts are in
not concern us in this chapter, but we will find the major mode.) The figure shows these two
uses for them in Chapter 14. An spc design may designs related by T4, a chromatic transposition
be thought of as a chromatic design in which that relates not only the notes of the designs but
each note has a specified spelling, not merely also the diatonic collections in which they reside.
an identity as a pitch class. A diatonic design The label T4 is in quotation marks in the figure
uniquely determines an spc design because its because the application of a chromatic transpo-
notes have well-​defined spellings, but this is not sition to the notes of a diatonic space dpc(n) is
true of enharmonic or chromatic designs, which theoretically problematic for reasons related to
may be interpreted as spc designs in more than enharmonic equivalence—​a deficiency to be rec-
one way. tified shortly.
Figure 13.3.4a exhibits three interrelated dia- The three designs in Figure 13.3.4a share the
tonic designs from Schubert’s sonata. Three dif- basic form given in 13.3.4b. The lack of clef and
ferent kinds of transformational relationships key signature in this figure indicates that both
are shown, each preserving a different feature the diatonic scale and the exact notes are vari-
of the design. The two designs that start on B♭, able. We may imagine an idealized C clef free

FIGURE 13.3.4 (a) Transformational relationships among three diatonic designs in Schubert; (b) the
genus common to the three designs

Diatonic Spaces • 553

to slide up and down the staff to any position, order of notes does not matter, and cardinality
capable of fixing any line or space as C, and equivalence eliminates doublings (a line genus
applying any traditional key signature to define may have repeated notes).
the governing scale as a particular diatonic col- A few line and chord genera are shown in
lection dpc(n).43 We refer to the form in (b) as Figure 13.3.5. OT-​ and OPTC-​classes are identi-
a genus, and we say that the three designs in fied by their mod-​7 normal forms. In the next
(a) realize that genus. Via the addition of a clef section we will work with both mod-​ 7 and
and key signature, any genus may be realized in mod-​12 normal forms, which we distinguish by
many ways, the various realizations related to means of superscripts (7) and (12). The mod-​12
each other by combinations of signature trans- counterpart of a genus is called a species; a line
formations and diatonic transposition. If the species is a mod-​12 OT-​class while a chord spe-
notes of the genus in (b) were shifted uniformly cies is a mod-​12 OPTC-​class. The reader may
higher or lower on the uncleffed staff, the genus find it useful to review the description of vari-
would not change. A genus is not the same thing ous OPTIC equivalence classes in Table 10.3.15
as a generic design, since in a generic design (where mod-​12 structure was assumed).44
the generic pitch classes are specified; thus The first two line genera in Figure 13.3.5 are
the generic designs in Figure 13.3.3 have clefs different because the differing order of the notes
but no key signature. A genus may be formally leads to different OT normal forms. The second
defined in either of two ways: as an equivalence and third, however, are the same because a genus
class of generic designs related by generic trans- is a transpositional equivalence class. Likewise
position tj, or as an equivalence class of diatonic the first two chord genera in the bottom row are
designs related by all possible sk and tj. the same. The third chord genus differs from the
If a design is a line or chord, the OPTIC rela- first two because inversion is not among the rela-
tions provide a precise means of identifying its tions defining a genus. The fourth and fifth chord
genus: a line genus is a mod-​7 OT-​class, while genera are the same because O and C equivalence
a chord genus is a mod-​7 OPTC-​class. A mod-​7 are among those relations for a chord genus: uni-
O-​class is a line (ordered set) of generic pitch son or octave doublings do not matter. All of the
classes (letter names), and a line genus is an chord genera shown here can be found among
equivalence class of those lines under generic the OPTC normal forms in Figure 13.1.22;
transposition. A mod-​7 OPC-​class is a chord in indeed, the twenty OPTC-​classes in that figure
generic space—​an unordered subset of gpc—​ exhaust the possible chord genera. The number
and a chord genus is an equivalence class of of possible line genera, in contrast, is unlim-
those chords under generic transposition. Chord ited. The Schubert genus from Figure 13.3.4b,
genus and line genus differ in the presence or reduced to a line genus (and therefore stripped
absence of the relations P and C: permutational of its rhythmic information), is the fourth
equivalence for a chord genus means that the line genus in Figure 13.3.5, [0, 0, 6, 0, 1, 2](OT7 ) .

FIGURE 13.3.5 Examples of line and chord genera

554 • E x p l o r i n g M u sical S p aces

 The last chord genus in Figure 13.3.5 has an with Figure 13.3.3, the enharmonic equivalence
unusual property. In general, as we have noted, of these two diatonic designs ensures that they
the exact generic pitch classes present in a both represent the same enharmonic design in
genus are unspecified. In this chord, however, it edpc(6). The transformation s12t6 always relates
is clear which gpcs are present: all of them. No enharmonically equivalent diatonic designs; we
matter how we position a clef on this staff, this will see below that when applied to enharmonic
chord is a generic aggregate CDEFGAB. Such a designs, s12t6 is the identity mapping.
chord is transpositionally symmetric in mod-​7 Enharmonic equivalence of diatonic designs
space, unchanged by any tj. A diatonic or generic is a strong condition requiring not only that
design that is invariant under mod-​7 transposi- corresponding notes of the two designs be
tion is called a trivial design, and the associated enharmonically equivalent but also that the
genus, such as this seven-​note chord genus, is underlying diatonic collections be enharmoni-
a trivial genus. (The only transpositional sym- cally equivalent. The last design in Figure 13.3.6,
metry possible in mod-​7 space is a trivial kind: even though its notes match those of the six-​
every nonzero transposition generates the mod-​ sharp design, is not enharmonically equiva-
7 transposition group 𝒯7, so if a set or design lent to the six-​flat design in this sense, because
in mod-​7 space is invariant under one nonzero dpc(+​ 5) is not enharmonically equivalent to
transposition it is invariant under all of them.) dpc(−6). The solfège syllables in this design
Triviality implies that whenever any gpc is pres- also differ from the others. The diatonic design
ent, all seven must be present simultaneously. in dpc(+​ 5) represents an enharmonic design
There are only two trivial chord genera, the in edpc(5), which differs from the enharmonic
empty set and the generic aggregate in the fig- design in edpc(6) represented by the other two
ure. Every nonempty line genus is nontrivial, diatonic designs.
even a seven-​note line such as a scale, whose As we have described them so far, signature
seven generic transpositions are distinguishable transformations are defined on diatonic spaces
because they arrange the seven gpcs in differ- dpc(n). More precisely, the signature transfor-
ent orders. Trivial genera may seem unworthy mation sk is a mapping from dpc(n) to dpc(n +​ k),
of much attention, but we define them carefully which maps any note in dpc(n) to the note in
here because they will be the only exceptions to dpc(n +​ k) sharing the same letter name (or
some of the general principles described in the equivalently the same generic pitch-​class num-
remainder of this chapter. ber). Some of the most interesting features of
Of the three diatonic designs in Figure 13.3.6, signature transformations, however, involve
the two in the top row are enharmonically equiv- enharmonic equivalence and therefore require
alent. One is a design in dpc(−6) and the other applying the transformations to enharmonic
in dpc(+​6). They are related by the composite designs in edpc(n). Because the elements of
transformation s12t6: the signature transforma- edpc(n) are pitch classes that have neither fixed
tion s12 converts from dpc(−6) to dpc(+​6) and spellings nor generic pitch-​class numbers asso-
the diatonic transposition t6 (or t−1) shifts all the ciated with them, it may be unclear how a sig-
generic pitch classes one step down (from B to A nature transformation can be defined on such a
for the first note). As we observed in conjunction design. Before we give this definition formally,

FIGURE 13.3.6 Enharmonic equivalence of diatonic designs

Diatonic Spaces • 555

further consideration of the spaces edpc(n) will the (mod-​12) scale index and the (mod-​7) solfège
be helpful, with attention to ways of describing index. The enharmonic coordinates for the note
relationships between notes in different enhar- B♭ in the G♭-​major scale, or equivalently for A♯ in
monic spaces. the F♯-​major scale, are (6, 2).
There are twelve different spaces edpc(n),
each containing seven notes. We define the uni- Exercise 13.3.7
fied enharmonic space edpc* to be a space of car- (a) Write the enharmonic coordinates for the
dinality 84 containing the elements of all twelve note D in the E♭-​major scale.
spaces edpc(n). This definition itself requires (b) Identify the scale and note specified by
elaboration: if we consider the elements of the enharmonic coordinates (5, 5). Write
edpc(n) to be ordinary pitch classes, then it is your answer in two enharmonically
not sufficient to define edpc* as the union of the equivalent ways.
twelve spaces edpc(n), as that union is nothing (c) For what value of q do the enharmonic
more than mod-​12 pitch-​class space pc. Let us coordinates (4, q) specify the note F♯?
write each note of edpc(n) as an ordered pair (d) For what value of n do the enharmonic
(n, q), where n is the index number of the space   coordinates (n, 3) specify pitch class 10?
edpc(n) and q is a solfège syllable (or, as described
below, a mod-​7 integer associated with a solfège Exercise 13.3.8
syllable) identifying one of the seven notes of (a) If q is a solfège index (an integer mod 7),
that space. For example, the ordered pair (6, mi) then as noted above, q is also the gpc
indicates the note mi of edpc(6), which could be number of the note in dpc(0) whose solfège
written either as B♭ in the G♭-​major scale or as A♯ index is q. The spc number of that note of
in the F♯-​major scale. dpc(0) is then δ0(q), which according to the
A similar ordered-​ pair notation could be formula from Exercise 13.2.2 is the number
introduced to identify notes in diatonic spaces congruent to 2q +​5 mod 7 that lies closest
dpc(n), but in that case there would be other to 0—​that is, in the range −3 ≤ δ0(q) ≤ 3.
options for the second component of the ordered For example, if q =​5, this formula gives
pair. For example, (n, g) could designate the note δ0(q) =​1, and indeed the note in dpc(0)
in dpc(n) whose generic pitch-​class number is g. whose solfège is la (solfège index 5) is A
Generic pitch-​class numbers cannot be used in (spc 1). Write the values of δ0(q) for all
this way in edpc(n), because gpcs are not defined seven values of q.
for notes in enharmonic spaces: gpc numbers (b) The pitch-​class number x of the note with
correspond to letter names, while notes in enharmonic coordinates (n, q) may be
edpc(n) may be spelled in more than one way. calculated by the formula x =​7(n +​ δ0(q)) +​
Notes in edpc(n) may theoretically be identi- 2 (mod 12). For example, if (n, q) =​(6, 2),
fied by their mod-​12 pitch-​class numbers, but then the formula from part (a) gives δ0(q)
doing so is awkward since for each n only seven =​2, and x =​7(6 +​2) +​2 =​10 (mod 12),
of the twelve pitch classes are used. Solfège syl- confirming that the coordinates (6, 2)
lables circumvent these difficulties, as a note of represent the note B♭ or A♯ as described
a space edpc(n) is uniquely determined by its above. Verify that this formula gives correct
diatonic position. It is helpful to have a numeric results for all four of the sets of enharmonic
representation for the syllables, so we introduce   coordinates in Exercise 13.3.7.
the solfège indices 0 =​do, 1 =​re, 2 =​mi, 3 =​fa, 4
=​ sol, 5 =​la, 6 =​ti. These are mod-​7 integers; in Enharmonic coordinates identify edpc* with
any major scale, the note with solfège index 0 is the Cartesian product ℤ12 × ℤ7. Indeed, edpc* is
scale degree 1̂, solfège index 1 is 2̂, and so on. For an interval space with that direct product as its
scales such as C major or C♯ major in which do interval group.45 Because the numbers 12 and
corresponds with the letter name C, solfège indi- 7 are coprime, we know from Section 6.4 that
ces match gpc numbers. With this convention, ℤ12 × ℤ7 is a cyclic group isomorphic to ℤ84—​a
every note of every enharmonic space edpc(n), structure that the perspective provided by sig-
and therefore every note of the unified space nature transformations will clarify. We now
edpc*, is represented by a unique ordered pair proceed to show that mappings of the three
(n, q), where n is an integer mod 12 and q is an types that appeared in Figure 13.3.4a—​chro-
integer mod 7. We refer to this notation as enhar- matic transpositions Ti, diatonic transposi-
monic coordinates, and to the two components as tions tj, and signature transformations sk—​are

556 • E x p l o r i n g M u sical S p aces

all well-​defined functions on the enharmonic (c) What equation in enharmonic coordinates
spaces edpc(n) and therefore on edpc*. All describes the diatonic transposition from C♯
these mappings are elements of the simply   to G in the D-​major scale?
transitive transformation group associated
with the interval group of edpc*, and all are Chromatic transposition Ti technically can-
important ways of navigating this space. These not be defined on the diatonic spaces dpc(n)
transformations act on individual notes within because the mod-​ 12 operator Ti presumes
the space, and thereby on enharmonic designs enharmonic equivalence: if we wish to trans-
comprising such notes. pose dpc(0) (C major) by T6, we have no crite-
Of the three types of transformations, rion for deciding whether it should transpose
the action of tj is easiest to describe. Diatonic to dpc(−6) (G♭ major) or to dpc(+​6) (F♯ major).
transposition operates on the second compo- The appearance of T4 in Figure 13.3.4a was
nent—​the mod-​7 component—​of enharmonic problematic for this reason. Notes in enhar-
coordinates, as given by the formula tj(n, q) =​(n, monic spaces edpc(n), however, have identi-
q +​ j): that is, tj transposes a note’s solfège index ties as mod-​12 pitch classes but are not tied to
up j steps, leaving the diatonic scale edpc(n) specific spellings, and therefore may be trans-
unchanged. For example, the equation t3(6, 4) posed by Ti.
=​(6, 0) in enharmonic coordinates indicates Generally Ti is not a mapping from edpc(n)
a transposition from sol (solfège index 4) to do to the same edpc(n), as tj is; Ti alters the scale,
(solfège index 0) in edpc(6), a transposition that mapping from edpc(n) to some other enhar-
could be interpreted as t3(D♭) =​G♭ in G♭ major or monic space edpc(m). To see how the new
t3(C♯) =​F♯ in F♯ major. The space edpc(n), like scale index m is related to the original index n
dpc(n), is a mod-​7 interval space, and a transpo- and the interval of transposition i, recall from
sition operator tj on edpc(n) works just as it does Section 2.6 that chromatic transposition Ti cor-
on dpc(n). Because tj is defined on every space responds to spc transposition by τk for any k
edpc(n), it is defined on the unified space edpc*. with k =​7i (mod 12). When a diatonic collection
The seven diatonic transpositions tj on edpc* dpc(n) is transposed by τk, it shifts by k posi-
form a cyclic group of order 7. The action of this tions in the line of fifths, to dpc(n +​ k). The new
group is simply transitive on each space edpc(n) scale index m is therefore n +​7i (mod 12); Ti is
individually but not on the unified space edpc*, a well-​defined mapping Ti: edpc(n) → edpc(n +​
inasmuch as diatonic transposition cannot map 7i). In Figure 13.3.4a, therefore, while it is not
from one space edpc(n) to another. technically correct to say that T4 maps from
dpc(−6) to dpc(−2), we may say instead that T4
Exercise 13.3.9 maps from edpc(6) to edpc(10), which is con-
(a) Verify that the equation t4(8, 5) =​(8, 2) is sistent with the formula above because the new
consistent with the formula given above for index number is n +​7i =​6 +​7 ∙ 4 =​10 (mod 12).
tj in enharmonic coordinates, and interpret Chromatic transposition may be applied to any
the statement musically. enharmonic design in edpc(n), always transpos-
(b) The equation t3(D♭) =​G♭ in edpc(6) could ing the underlying scale along with the notes.
be written using pitch-​class numbers in the Ti is defined on every enharmonic space
form t3(1) =​6. This notation is awkward in edpc(n) and is thus a well-​defined mapping on
several ways, among them the fact that the the unified space edpc*. In enharmonic coor-
action of tj depends on the space edpc(n) dinates (n, q), Ti operates on the first (mod-​12)
relative to which tj is defined. While t3(1) =​ coordinate only, adjusting the scale index by 7i
6 is a true statement in edpc(6), there are as described above; the solfège syllable associ-
some mod-​12 integers n for which t3(1) is ated with a note remains the same in the new
not defined in edpc(n) because pitch class 1 scale. That is, Ti(n, q) =​(n +​7i, q). For example,
does not belong to edpc(n) in the first place. the equation T4(6, 2) =​(10, 2) describes the T4
For at least one value of n, t3(1) is defined transposition of B♭ in G♭ major to D in B♭ major,
in edpc(n) but is something other than 6. as in the first notes of the two T4-​related designs
Give an example of such an n. Is there more in Figure 13.3.4a. All twelve transpositions Ti
than one such n? What familiar property of may be obtained by iterating T1; because T1
diatonic scales explains the answer to the adjusts the mod-​12 coordinate by 7, and 7 is a
last question? generator of the cyclic group ℤ12, it follows that

Diatonic Spaces • 557

the chromatic transpositions on edpc* form Figure 13.3.12 illustrates the theorem, show-
a cyclic group of order 12, consistent with our ing the action of s1 on the two enharmonically
expectations. equivalent designs from Figure 13.3.6. (This
transformation does not occur in Schubert’s
Exercise 13.3.10 sonata; it is shown for illustrative purposes
(a) Verify that the equation T9(4, 1) =​(7, 1) is only.) While s1 adds one sharp to any key signa-
consistent with the formula given above for ture that already contains sharps, it removes one
Ti in enharmonic coordinates, and interpret flat from a signature with flats, with the same
the statement musically. effect. The original diatonic collections dpc(−6)
(b) What transposition Ti maps the C♯-​major and dpc(+​6) are enharmonically equivalent, and
scale to the C♭-​major scale? First answer the same is true of the transformed collections
the question by figuring an interval of dpc(−5) and dpc(+​7). The only notes affected
transposition in the usual way; then by s1 are the enharmonically equivalent notes
express the two scales as edpc(n) and C♭ and B, which occur at the same places in the
edpc(m) and verify that m =​ n +​7i (mod two designs and are altered in the same way by
12). Write an equation in enharmonic the transformation (raised chromatically), so
coordinates describing what happens to the the transformed designs remain enharmoni-
note B♯ in this transposition. cally equivalent. The original note, whether C♭ in
(c) Write a general formula for determining dpc(−6) or B in dpc(+​6), is (6, 3) in enharmonic
the transposition Ti that maps from coordinates, while the transformed note, C in
edpc(n) to edpc(m), expressing the dpc(−5) or B♯ in dpc(+​7), is (7, 6) in either case
mod-​12 integer i in terms of n and m. (originally fa, it is now ti).
Verify that your formula is correct in the The line-​of-​fifths perspective should make
  examples from (a) and (b).46 the general validity of the theorem clear.
Enharmonically equivalent notes and spaces
We turn now to the action of the signature look just alike in the line of fifths, shifted left or
transformation sk on the enharmonic spaces right by twelve places or some multiple thereof,
edpc(n) and edpc*. The possibility of defining sk and the action of a signature transformation on
on edpc(n) depends on a simple theorem: the two spaces takes exactly the same form.
To define the action of a signature transfor-
Theorem 13.3.11: Enharmonic compatibility mation such as s1 on a note such as (6, 3)—​B or
of signature transformations If a note x in a C♭ in the enharmonic space edpc(6)—​we may
space dpc(m) and a note y in a space dpc(n) are assume any convenient spelling for the space;
enharmonically equivalent (in the strong sense that is, we may choose dpc(+​6) or any other
defined above for diatonic designs), and if the space dpc(n) enharmonically equivalent to
same signature transformation sk is applied to dpc(+​6), such as dpc(−6). If we choose dpc(+​6),
both of them, then the notes sk(x) in dpc(m +​ k) then the note is B; applying s1 to B in dpc(+​6)
and sk(y) in dpc(n +​ k) are also enharmonically gives B♯ in dpc(+​7). Finally, we reinterpret this
equivalent.
   result as a pitch class in the enharmonic space

FIGURE 13.3.12 The signature transformation s1 applied to two enharmonically equivalent designs

558 • E x p l o r i n g M u sical S p aces

FIGURE 13.3.13 Transformations of a diatonic scale by diatonic transposition, chromatic transposi-
tion, and signature transformation

edpc(7), confident that if we had performed the solfège indices unchanged. Had we chosen to
calculation in a different spelling we would have notate this scale in sharps, the enharmonic coor-
reached the same result. In this way we obtain dinates would be unaffected.
a well-​defined mapping s1: edpc(6) → edpc(7), The signature transformation s1, though it
or more generally sk: edpc(n) → edpc(n +​ k). leaves the written notes unchanged, alters both
Because sk is defined on every enharmonic space the scale index (from 0 to 1) and the solfège indi-
edpc(n), it is also a well-​defined mapping on the ces. The resulting scale still starts on C but is now
unified space edpc*. lydian; the C that was originally do (0) is fa (3) in
The transformations of the white-​note scale the new scale, and the indices of the other notes
in Figure 13.3.13 may be helpful in understand- shift likewise by 3 (mod 7). The number 3 may be
ing how the transformations Ti, tj, and sk act on explained by the observation that do in dpc(n)
enharmonic coordinates. Each scale is a simple and do in dpc(n +​1) lie three diatonic steps
diatonic design—​a line through the seven notes apart. Accordingly, s1 adds 1 to the mod-​12 coor-
of one enharmonic space edpc(n). (We write dinate and 3 to the mod-​7 coordinate of every
seven-​note scales because of octave equivalence; note: s1(n, q) =​(n +​1, q +​3). Other signature
the eighth note could be written, but it would transformations sk may be produced by iterating
duplicate the first.) In edpc(n) the spelling of s1; the general formula is sk(n, q) =​(n +​ k, q +​3k).
notes is arbitrary; three of the four scales in the
figure have only one likely spelling, but the D♭-​ Exercise 13.3.14
major scale could be written as C♯ major instead, (a) Verify that the equation s5(4, 3) =​(9, 4) is
with a seven-​sharp key signature. The names consistent with the formula given above for
shown for the scales are thus somewhat arbi- sk in enharmonic coordinates, and interpret
trary, as signaled by the cautionary quotation the statement musically.47
marks. A space edpc(n) does not imply a tonal (b) What equation in enharmonic coordinates
center, but the scale names take the first note of expresses the application of the signature
each scale to be the tonic. transformation s−4 to the note G♯ in an A-​
The transposition operators t1 and T1 trans- major scale? Confirm that your equation is
form the scale in accordance with the descrip-   consistent with the formula for sk.
tions above. Diatonic t1 shifts each note one
step higher within edpc(0) and adds 1 to the The next exercise shows, empirically, that
solfège index of each note in enharmonic coordi- some composite functions formed of Ti, tj, and
nates, converting the original C-​major scale to D sk give results that seem to match other simpler
dorian. Chromatic T1 maps C major to D♭ major, transformations. Explanations for these obser-
adding 7 (mod 12) to the scale index but leaving vations will emerge in the discussion below.

Diatonic Spaces • 559

Exercise 13.3.15 pair reduces to (n, q +​1), which is nothing more
(a) Starting with enharmonic coordinates (0, 6) than t1(n, q), a one-​step diatonic transposition.
(the note B in C major), apply the composite Whereas s7 is the same as chromatic T1, s12 is the
transformation T1s5 (that is, apply T1, then same as diatonic t1. Again consider the chain of
s5). Give the enharmonic coordinates of the s1-​related scales starting on C major, and con-
resulting note, and interpret the answer sider extending the chain beyond C♯ major (s7
musically. What simpler transformation, of C major). For convenience we may respell C♯
applied to the same note, would yield the major as D♭ major with no change in enharmonic
same result? coordinates. The next s1 produces D♭ lydian, after
(b) Also calculate the result of applying T1s5 to which D♭ is raised to D♮ and the cycle of modes
enharmonic coordinates (2, 0) and (10, 5). starts again on D: locrian, phrygian, aeolian.
Does the simpler transformation you The fifth application of s1 after D♭ major, or the
found in (a) give the same results in these twelfth after C major, produces D dorian, a t1
cases also? transposition of the original C-​major scale.48
(c) Identify a simpler transformation The equivalence of s12 and t1 is intimately
equivalent to t1s2 by applying this connected with the meaning of enharmonic
composite transformation to several equivalence. Return once more to the two enhar-
different notes in enharmonic coordinates. monically equivalent diatonic designs from
(d) What simpler transformation is equivalent Figure 13.3.6. These two designs are related by
  to T7t3? s12t6, but we now know that s12 =​ t1, so s12t6 =​
t1t6 =​ t0, the identity mapping on edpc*. From a
Because there are only twelve enharmonic slightly different perspective, Figure 13.3.6 itself
spaces edpc(n), one might expect that the sig- may be taken as a demonstration of the equa-
nature transformation s1 generates a 12-​cycle tion s12 =​ t1. When s12 is applied to the design in
of enharmonic designs. The truth is much more dpc(−6), the resulting design in dpc(+​6) does
interesting. Because s1 alters both the mod-​12 not match the dpc(+​6) design shown in the fig-
and mod-​7 components of the enharmonic coor- ure but is one diatonic step higher in the F♯-​major
dinates (n, q), iteration of s1 produces not a 12-​ scale, starting on B rather than on A♯ (because
cycle or a 7-​cycle but an 84-​cycle. As odd as the the signature transformation leaves generic pitch
idea of adding dozens of sharps or flats to a key classes unchanged). This observation shows that
signature may sound, signature transformations the action of s12 matches that of t1.
sk with indices much larger than 12 or smaller As the s1-​chain of scales starting on C major
than −12 are of considerable theoretical and ana- is extended beyond s12, every seventh scale is a
lytical interest. major scale, some Ti of C major: D♭ major at s7, D
The behavior of the transformations s7 and major at s14, E♭ major at s21, and so on. Meanwhile,
s12 merits particular attention. Adjusting a key every twelfth scale is a scale in dpc(0), some tj of
signature seven places in the sharpwise direction C major: D dorian at s12, E phrygian at s24, F lydian
always has the effect of raising the scale and all at s36, and so on. Because 12 and 7 are coprime,
its notes by one semitone; hence s7 is enharmon- the two interleaved cycles will realign only after
ically the same mapping as the chromatic trans- 12 ∙ 7 =​84 iterations of s1, when C major finally
position T1. We may confirm this by plugging k reappears. By that time, all 84 modal scales will
=​7 into the formula for sk, which gives s7(n, q) =​ have been traversed: seven modes starting on
(n +​7, q +​21). Because the second component is each of the twelve pitch classes.49
mod 7, this reduces to (n +​7, q), which is exactly The formula for sk gives s84(n, q) =​(n +​84, q +​
T1(n, q). If we apply s1 repeatedly to the C-​major 3 ∙ 84), which reduces to (n, q) because 84 is divis-
scale in Figure 13.3.13, after C lydian, the next ible by both 12 and 7. Indeed, s84, which is equal
application of s1 will raise C to C♯, producing a to both T112 and t17, is the identity transforma-
C♯ locrian scale; repeated applications of s1 will tion on edpc*, and could be written equivalently
then cycle through the modes on C♯ in the pro- as T0, t0, or s0. The signature transformation s1
gressively brightening order locrian–​phrygian–​ generates a cyclic group of order 84, the signa-
aeolian–​dorian–​mixolydian, arriving at C♯ major ture group. Every Ti, tj, and sk belongs to this
with the seventh application of s1. group, and every element of the group may be
If k =​12, the formula for sk gives s12(n, q) =​ written as sk for some k between 0 and 83; for
(n +​12, q +​36). Because the first component is instance, T5 may be written as s35, t5 as s60, and s−5
mod 12 and the second is mod 7, this ordered as s79. We will not always write transformations

560 • E x p l o r i n g M u sical S p aces

in these ways, however, because chromatic and from mm. 0–​2 to mm. 121–​23 is therefore s−4s−3
diatonic transpositions and small negative sig- =​ s−7, which is the same as T−1 or T11, and indeed,
nature transformations remain the most intui- both the key signature and the notes are a semi-
tive descriptions of some musical relationships. tone lower at mm. 121–​23 than at the outset.
As a cyclic group, the signature group is auto- (Measures 121–​ 23 tonicize F♯ minor, not A
matically commutative; all transformations Ti, major, but recall that a diatonic or enharmonic
tj, and sk commute. The space edpc* has 84 ele- space dpc(n) or edpc(n) does not imply a tonal
ments given by enharmonic coordinates (n, q), center.)
and action of the signature group on this space At the bottom of the figure, relations are
is simply transitive. shown also between mm. 121–​23 and mm. 347–​
The formulas for Ti and tj as signature trans- 49. These two designs are not related by any Ti
formations have many implications. They may or tj or by any simple sk, but we can describe the
be used to derive the observations in Exercise relationship between them in at least two differ-
13.3.15 above; for example, the calculation T7t3 ent ways and show that they are equivalent. The
=​ T17t13 =​ s77s123 =​ s49s36 =​ s85 =​ s84s1 =​ s1 confirms first possible description is T1t2: that is, we imag-
the result of 13.3.15d. They also show why some ine transforming the three-​sharp design by T1 to
of the transformations we have observed in put it back in B♭ major, then raising the notes
Schubert’s sonata combine as they do. Figure diatonically by two additional steps to match
13.3.16 revisits the sonata for a final time, show- mm. 347–​49. Alternatively, we could accomplish
ing several related designs that appear in the the same thing by s−5t3, obtaining the desired
movement. The three fragments arranged along key signature via a five-​step flatwise signature
the left side of this figure are those from Figure transformation from three sharps to two flats,
13.3.4a, related by s−4, T4, and t2. The designs after which adjustment by three diatonic steps
shown here are enharmonic designs, so the use is required. The algebra of signature transforma-
of the chromatic transposition T4 is now justi- tions reassures us that both of these composite
fied. Because T4 =​ T14 =​ s74 =​ s28, the composite transformations are the same: T1t2 =​ s7s24 =​ s31
transformation from mm. 0–​2 to mm. 347–​49 is while s−5t3 =​ s−5s36 =​ s31. While there may be a
s−4T4 =​ s−4s28 =​ s24, which is the same as t2 since variety of ways to think of this relationship, s31 is
t1 =​ s12. The enharmonically spelled form in six the only transformation in the signature group
sharps, from Figure 13.3.6, does not occur in that relates these two designs as shown.
the piece but is included here to clarify the s−3 Any two enharmonic designs in the same
relationship with the three-​sharp design below, genus are always related by some signa-
which appears at mm. 121–​23. The mapping ture transformation sk. In group-​ theoretic

FIGURE 13.3.16 Network of signature transformations and transpositions in Schubert, D. 960, I

Diatonic Spaces • 561

terminology from Section 6.5, a genus is an orbit realizations on each staff share a key signature
of the signature group. We have seen that the and all belong to the same space edpc(n); these
chromatic and diatonic transpositions Ti and seven designs are therefore related by diatonic
tj may be expressed in terms of sk. Our discus- transposition tj. (The line for edpc(6) appears in
sion of the last two designs in Figure 13.3.16 the figure twice, notated in sharps and equiva-
suggests, conversely, that in lieu of a signature lent flats.) Vertically aligned realizations, mean-
transformation, we can always find some com- while, are related by chromatic transposition Ti.
posite function of the form Titj to describe the The array could have been organized stepwise,
relationship between two designs of the same by t1 horizontally and T1 vertically, but instead
genus, choosing Ti to obtain the correct relation we have chosen a fifth-​generated arrangement
between key signatures and tj to shift the notes in both dimensions, so that horizontally adja-
to the proper position. In fact, we can say exactly cent designs are related by t4, vertically adjacent
which transpositions Ti and tj will combine to designs by T7, as the arrows show. The figure
produce a given signature transformation sk; the wraps around on itself in both the horizontal and
following exercise provides the details.50 vertical dimensions in the manner of a torus. An
advantage of arranging the realizations by fifths
Exercise 13.3.17 For any integer k, let i be a is that the signature transformation s1 becomes
number congruent to 7k mod 12, and let j be a a diagonal move , illustrated by the diagonal
number congruent to 3k mod 7. Show that the chain of arrows starting at the top right. This
composite transformation Titj is equal to the diagonal move combines a vertical move by
signature
   transformation sk. T7 with a leftward move by t4−1 =​ t3, so the fig-
ure shows explicitly that T7t3 =​ s1, as calculated
The two enharmonically equivalent designs above. The reader may trace this s1-​chain beyond
in Figure 13.3.16 are the same enharmonic the twelve arrows shown and verify that it traces
design in edpc(6), but as diatonic designs in a path through all 84 realizations. The structure
dpc(−6) and dpc(+​6) they are different. The for- of the array illustrates the group isomorphism
mulas t1 =​ s12 and s84 =​ s0 are valid in enharmonic ℤ12 × ℤ7 ≅ ℤ84 in much the same way that Figure
space edpc* but not in the diatonic spaces 6.4.3 illustrated the simpler isomorphism ℤ4 ×
dpc(n). Notes in dpc(n) have fixed letter names, ℤ3 ≅ ℤ12. The figure suggests two different Cayley
which signature transformations are powerless diagrams for the signature group, one (compris-
to change, so no sk can duplicate the effect of t1 ing all the horizontal and vertical arrows) on the
on these spaces; moreover, dpc(n) differs from pair of generators t4 and T7 and highlighting the
dpc(n +​12), to say nothing of dpc(n +​84). On direct-​product structure, the other (the single
the family of spaces dpc(n), s1 generates an infi- 84-​cycle of diagonal arrows) on the single gen-
nite cyclic group. If two diatonic designs belong erator s1 and showing the cyclic structure.
to the same genus but contain different generic Some pairs of realizations share the same
pitch classes, they will not be related by any notes. For example, the first four realizations
signature transformation, but they will always in the s1-​chain in Figure 13.3.18 all have the
be related by some combination of transforma- notes A♮–​D♮–​B♮; not until the fourth applica-
tions tj and sk. tion of s1, when D♮ becomes D♯, do any of these
In enharmonic space edpc*, an s1-​cycle like notes require adjustment. Interpreted as chro-
the one described above for scales can be recre- matic designs in pc, these four realizations
ated for any genus: iteration of s1 always pro- are identical, but they differ as enharmonic
duces a cycle through 84 designs, which exhaust designs because of the differing key signatures;
the realizations of the genus. The transforma- in enharmonic coordinates the note A is (0, 5)
tion s1 is not the only generator of the group; in the first design in the chain (la in edpc(0))
any sk for which k is coprime to 84 will produce a and (1, 1) in the second (re in edpc(1)). Because
complete cycle. Other ways to arrange the cycle s1 always changes the key signature, no enhar-
may be revealing. Figure 13.3.18 shows the 84 monic design can be invariant under s1, even if
realizations of the line genus [0, 3, 1](OT7 ) , one of its notes are unchanged.
the line genera shown earlier in Figure 13.3.5. The commonality among the four A♮–​D♮–​B♮
Each three-​note group is one realization of this realizations has further implications. Vertically
genus; the 84 realizations are arranged in a aligned realizations are related by Ti, so as chro-
12-​by-​7 array. Reading horizontally, the seven matic designs, all 48 designs in the rightmost

562 • E x p l o r i n g M u sical S p aces

FIGURE 13.3.18 The 84 realizations of the line genus [0, 3, 1](OT
7)
, related by t4, T7, and s1

four columns of the array are related by chro- these 48 realizations of the given genus. There are
matic transposition. These realizations all con- a total of three species for this genus, as indicated
sist of an ascending perfect fourth followed by at the bottom of the figure: [0, 5, 1](OT12 ) is the spe-
a descending minor third; that is, they all share cies for all the realizations in the first two col-
the mod-​12 OT normal form [0, 5, 2](OT12 ). This umns, and [0, 6, 2](OT12 ) is the species for the third
mod-​12 OT-​class is the species associated with column alone. The genus-​ species relationship

Diatonic Spaces • 563

will be examined in detail in the next section. A common phenomenon easily modeled by
That this three-​note genus comes in exactly three a signature transformation is the inflection of a
species is no accident; it illustrates the impor- triad from major to minor or vice versa, which
tant principle cardinality equals variety. we have described previously using the parallel
An array like Figure 13.3.18 may be con- transformation P. Signature transformations
structed for any genus. For most genera the can sometimes reproduce the behavior of other
84 realizations are all different as enharmonic triadic transformations as well. An analysis of
designs, as is the case for the line genus in the a triadic progression via signature transforma-
figure. The only exceptions to this generalization tions, however, requires assigning a diatonic
are the genera that we have called trivial—​those context to each triad in the form of a key signa-
that are transpositionally symmetric in mod-​ ture or a space edpc(n), an additional layer of
7 space, invariant under any tj, such as a chord interpretation that a standard transformational
containing all seven generic pitch classes. A triv- analysis does not require, and which may be con-
ial genus has only twelve different realizations, sidered desirable in some situations but not nec-
distinguished by key signatures. Every nontriv- essarily in others.
ial genus has 84 different realizations, and the Consider, for example, the PR-​cycle C–​c–​
action of the signature group on them is simply E♭–​e♭–​G♭–​f♯–​A–​a–​C from Smetana’s Bartered
transitive. Bride overture, examined previously in Exercises
The following theorem consolidates the main 4.2.13c and 10.2.12b. For each triad three dia-
observations we have made about signature tonic collections are available. For consistency,
transformations and transpositions. assume that each triadic root is a temporary
tonic of a major or aeolian scale; that is, the root
Theorem 13.3.19: Properties of signature of every major triad is do and the root of every
transformations and transpositions The minor triad is la. With this understanding, the
chromatic transpositions Ti, diatonic C-​major triad belongs to edpc(0) while C minor
transpositions tj, and signature transformations and E♭ major belong to edpc(9). The C–​c pro-
sk, all defined on unified enharmonic space gression and the other parallel moves become
edpc*, satisfy the following properties: s−3, while c–​E♭ and the other relative moves are
t2. The composite transformation s−3t2 =​ s−3s24 =​
(a) The chromatic transpositions Ti form a s21 =​ T3 captures the relationship between the C-​
cyclic group of order 12, generated by major triad in edpc(0) and the E♭-​major triad in
T1; T12 is the identity T0. In enharmonic edpc(9), and repeating s−3 and t2 in alternation
coordinates, Ti(n, q) =​(n +​7i, q). produces the entire cycle. Another feature of the
(b) The diatonic transpositions tj form a cyclic signature-​ transformation approach is that s−3
group of order 7, generated by t1; t7 is the and t2, unlike P and R, are defined on individual
identity t0. In enharmonic coordinates, notes within triads; the transformational action
tj(n, q) =​(n, q +​ j). maps root to root, so in the case of R it does not
(c) The signature transformations sk form reflect the smooth voice leading characteristic of
a cyclic group of order 84, the signature the triadic transformation.
group, generated by s1; s84 is the identity s0. Signature transformations and transposi-
In enharmonic coordinates, sk(n, q) =​ tions are not involutions as are the wechsels P
(n +​ k, q +​3k). and R. While s−3 captures the behavior of P on
(d) All transpositions Ti and tj belong to the a major triad and t2 captures the behavior of R
signature group: Ti =​ s7i, while tj =​ s12j. All on a minor triad, these transformations must
transformations Ti, tj, and sk commute. be replaced by their inverses if the modes
(e) Every signature transformation sk may be are reversed. Thus an oppositely directed
written as the product of some Ti and tj. PR-​ chain, descending by thirds, alternates
Specifically, sk =​ Titj, where i =​7k (mod 12) s3 and t5.
and j =​3k (mod 7).
(f) Every nontrivial diatonic genus has exactly Exercise 13.3.20 Use signature transformations
84 different realizations in enharmonic to analyze the sequence C–​c♯–​C♯–​d–​D–​d♯–​D♯–​e
spaces edpc(n). The signature group acts in from Prokofiev’s Scherzo in A Minor,
simply transitive fashion on the set of these Op. 12, No. 10 (analyzed as a P′P-​chain in
  84 realizations. Exercise
   8.2.26e).

564 • E x p l o r i n g M u sical S p aces

Exercise 13.3.21 The relationship between the piano’s E-​major
(a) A signature-​transformation alternative to and C-​major chords is evidently not s−3. The
the leittonwechsel transformation L is less pitch classes in the chords are transposed by T8,
intuitive than that for P or R. In edpc(0), but as we have defined chromatic transposition
C-​major and E-​minor triads are related by on enharmonic designs, T8 transposes music
t2 or t5, but the note E is mi, not la, in this in edpc(4) to music in edpc(0), not edpc(1).
diatonic context. Above we considered s3 The figure therefore shows this transposition as
the most appropriate representation for T8s1, the unique element of the signature group
minor-​to-​major P. Show that, starting on a that maps an E-​major triad in edpc(4) to a C-​
C-​major triad in edpc(0), an alternation of major triad in edpc(1). We may note that T8s1 =​
t2 and s3 does not generate an LP-​c ycle. s56s1 =​ s57, and the index numbers of s−3 and s57
(b) The deficiency of the t2s3-​chain in (a) as a are congruent mod 12, as must always be true
model of an LP-​chain may be rectified in of signature transformations whose initial and
two ways: by choosing a representation final diatonic spaces agree.
for L different from t2, or by choosing This melody is the first of several features
a representation for P different from s3 of Debussy’s development section for which
(and therefore a different “P” from the a signature-​transformation analysis proves
one used above in the PR-​chain). Solve revealing. In mm. 133–​39, as shown in part
the problem both ways, and show that the (b) of the figure, the violin hovers around G♭
solutions both produce the same composite while the piano arpeggiates the perfect fifth
transformation at the two-​chord level (that E♭–​B♭. The violin’s G♭ is neighbored by A♭ and the
is, the same transformation equivalent piano’s B♭ by C♮, so the mode is E♭ dorian; the
  to “LP”).51 diatonic space is dpc(−5) or enharmonically
edpc(7). The echo at m. 140 introduces a new
Example 13.3.22: Debussy, Sonata for Violin modal context: the piano’s fifths drop to D–​A ,
and Piano, I The first movement of Debussy’s the violin’s G♭ and A♭ are respelled as F♯ and G♯,
Violin Sonata is a loose sonata form whose and the mode is D lydian—​dpc(+​3) or edpc(3).
development section opens with arpeggiations This too is a simple signature transformation,
of a first-​inversion triad in the piano, over now s−4, clarified in the figure by first respelling
which a violin melody enters at m. 88, as E♭ dorian as D♯ dorian. In contrast to the s−3
shown in greatly simplified form at the left side transformation of the earlier melody, this s−4
of Figure 13.3.23a.52 The diatonic context is acts on the entire violin-​and-​piano texture as
unambiguous: the arpeggiated triad is E major, a single enharmonic design, though it has the
and the entire ten-​measure melody remains effect of changing only the notes in the piano.
within dpc(+​4), the E-​major scale, engaging In all four of the segments in parts (a)
every note of the scale except B, which appears and (b) of Figure 13.3.23, the violin enters
in the piano arpeggiations. The arpeggiations on the same F♯, neighbored each time by a
return in m. 106 and the melody in m. 110, but prominent G or G♯, but the modal context
there are differences. The arpeggiated chord is is different each time. Part (c) assembles
now C major, but the melody is not in dpc(0): it this information into a network that spans
is in dpc(+​1), apparently C lydian. The melody the entire development section. The first
is not transposed: it starts on F♯ both times, two “measures” in (c) are a shorthand form
and the remaining notes are identical except of (a), while the next two repeat (b). Three
for chromatic alterations to the notes C, D, different transformations, s6, s−9, and s6t1,
and G to fit dpc(+​1)—​a textbook example are shown spanning from the measures of (a)
of the signature transformation s−3. The to the measures of (b); see Exercise 13.3.25
transformation is applied scrupulously through below for interpretations of these. The note-​
all 37 notes of the melody, beginning as shown grouping bubbles in (c) vary in size to show
in the figure. Because enharmonic equivalence the extent of the designs on which the various
will soon come into play in this analysis, the transformations act.
figure gives the enharmonic labels for the At m. 146 the continuing piano figuration is
spaces, edpc(4) and edpc(1) rather than transformed by s−5 to dpc(−2). The movement’s
dpc(+​4) and dpc(+​1). primary tonality is G minor, and the recapitulation

Diatonic Spaces • 565

FIGURE 13.3.23 Signature transformations in Debussy, Sonata for Violin and Piano, I, mm. 88–​150

is signaled at m. 150 by G in the bass and the parallel motion. Although these segments do
return of the triadic melody from the opening of not linger for long in one diatonic collection,
the movement. (This melody is first heard in the at any moment a listener readily assimilates
piano, as the instruments trade material briefly; it the music into an imagined diatonic context,
is then restated in the violin and is shown in the and signature transformations are useful
violin line of Figure 13.3.23c for convenience.) The for modeling these changing perceptions, as
piano’s A and D are the only notes of the D lydian illustrated in Figure 13.3.24. The top half of
scale that are unaffected by this s−5. As B♮ gives way the figure shows mm. 98–​106; the bottom half
to B♭, however, an ingenious role reversal takes shows mm. 120–​28.
place: in the G-​minor context, B♭ is heard not as The first of these passages immediately
an upper neighbor to A, as B♮ has been, but as the follows the E-​major section described above,
stable note to which the now dissonant A resolves. so the E–​B dyad at m. 98 is initially heard
Figure 13.3.23 skips over two eight-​measure in a four-​sharp context, edpc(4). Already in
passages featuring perfect-​fifth dyads in the next measure (the perfect-​fifth dyads

566 • E x p l o r i n g M u sical S p aces

FIGURE 13.3.24 Signature transformations in Debussy, mm. 98–​106 and 120–​28

move regularly at one chord per measure), 13.3.24 are pivot transformations of this kind,
the D–​A dyad challenges that assumption, relating dyads with the same notes in different
so our hypothetical listener fits it into the key signatures. A short horizontal t6 arrow
nearest available collection with D♮, edpc(3). shows the relationship between the E–​B and
Retrospectively, she realizes that the E–​B D–​A dyads in edpc(3).
dyad belongs to edpc(3) as well, so it may be The F–​C dyad that follows poses a greater
reinterpreted that way, as a sort of pivot chord. challenge to diatonic hearing as it requires a
E–​B in edpc(4) and E–​B in edpc(3) are related larger scale shift, an s−3 all the way to edpc(0).
by s−1. All double-​line vertical arrows in Figure Again, though, retrospective hearing allows the

Diatonic Spaces • 567

entire segment (so far) to be accommodated are in edpc(7), the diatonic context for the E♭
in this white-​note scale; in the figure, a pivot dorian passage from Figure 13.3.23b, which
s−3 maps a bubble containing both the E–​B and follows immediately. The plagal t4 into m. 128,
D–​A dyads in edpc(3) to a similar bubble in however, is new: until this point the dyads have
edpc(0). This new context then holds sway for moved exclusively by step or by third (t1, t2,
three more measures. The fifth and sixth dyads t5, or t6). The longest arrow in Figure 13.3.24,
repeat the first two, but are no longer likely to looping around the left side of the figure, is s−9
be heard in reference to edpc(4) or edpc(3) from m. 98 to m. 128. This s−9 appears also in
because of what has intervened. With the B♭–​F Figure
   13.3.23c; see the exercise below.
dyad in m. 104, the possibility of hearing the
entire passage in one diatonic collection is Exercise 13.3.25
finally shattered, as this dyad forces a shift to a (a) Explain the transformations s−9 and
one-​flat collection, edpc(11), incompatible with s6t1 appearing in the piano in Figure
the previously heard B♮s. Here the preceding D–​ 13.3.23c, and verify that the composite
A is the only possible pivot. transformation (T8s1)(s6t1) in the bass is
The final dyad in this first segment, C–​G, equal to s−9.
becomes absorbed in the arpeggiations of the (b) In the violin part in the same figure, the
C-​lydian section discussed previously. From transformation from m. 88 to m. 133
edpc(11), an s2 pivot to edpc(1) is required is s−3s6, which is s3, not s−9. Explain the
when the lydian context emerges. The arrow discrepancy.
at the top right in Figure 13.3.24 describes the (c) Explain the transformations T3s−3 and t2s−6
relationship between E major at m. 98 and C shown from m. 120 to m. 128 in Figure
lydian at m. 110 in two ways. The first, T8s1, 13.3.24, and verify that both are equal to s18.
was noted earlier in 13.3.23a. Alternatively, (d) Verify that the transformations s57 and s18
an E–​B dyad is mapped to a C–​G dyad by t5 combine to produce s−9, as implied by the
(and the diatonic transpositions to this point,   three longest arrows in Figure 13.3.24.
t6t2t1t5t6t5t1, combine to produce t5), so this t5
may be combined with the signature change s−3
from edpc(4) to edpc(1). Calculation confirms 13.4 GENUS AND SPECIES
that T8s1 and t5s−3 are both equal to s57. Indeed
they must be equal: the simply transitive action Figure 13.3.18 noted the three species arising
of the signature group ensures that one and as realizations of the line genus [0, 3, 1](OT7 ) .
only one group element maps E–​B in edpc(4) Figure 13.4.1 shows a more familiar example,
to C–​G in edpc(1). the chord genus of triads. The seven generic tri-
The T8 transposition has a further ads CEG, DFA, …, diagrammed in gpc in the top
ramification: the second passage of perfect-​ row of this figure, share the mod-​7 OPTC nor-
fifth dyads, in mm. 120–​27, is an exact mal form [0, 2, 4]OPTC
(7 )
. The bottom row shows
T8 transposition of the first, as shown by the corresponding white-​note chords in mod-​
the vertical arrow joining the two largest 12 space pc and demonstrates a familiar fact:
bubbles in Figure 13.3.24. The two passages the triads within a diatonic scale include three
are, however, in different relationships with major triads, three minor triads, and one dimin-
their surroundings. The music preceding the ished triad.
C–​G dyad of m. 120 is not in C major but in Let us carefully examine what this observa-
C lydian, so the initial pivot transformation tion means. We may interpret the generic pitch
here is s−2 rather than s−1. This s−2 reverses the classes in the triads in gpc as white notes, dia-
s2 with which the previous segment ended, tonic pitch classes in dpc(0) (technically using
leading back to edpc(11), and the following the interpretation function δ0 to do this). We
measures push further flatward, to edpc(8) and may then regard those white notes as mod-​12
eventually edpc(7). In edpc(11), the first two pitch classes in edpc(0), a subset of pc (techni-
dyads at mm. 120–​21 retrograde the last two cally using the projection function π), obtain-
dyads of the earlier passage; in fact, allowing ing triads in pc. While the original chords in
for signature adjustments as needed to fit gpc all had the same mod-​7 interval structure,
edpc(8), the first four dyads of this passage the new subsets of pc exhibit three different
retrograde the last four of the previous passage. mod-​ 12 interval structures, commonly called
The final three dyads of the second passage qualities but known in this context as species.

568 • E x p l o r i n g M u sical S p aces

FIGURE 13.4.1 One chord genus and three chord species of triads

The three major triads have mod-​12 OPTC every diatonic, enharmonic, or generic design
normal form [0, 4, 7](OPTC 12 )
, the three minor tri- realizes a unique genus. Among the categories of
ads [0, 3, 7]OPTC , and the one diminished triad
( 12 )
designs defined in Section 13.3, only a chromatic
[0, 3, 6](OPTC
12 )
. There are no augmented triads in a design does not determine a genus. A chord
diatonic scale, nor other kinds of “triads” such as genus is a mod-​7 OPTC-​class, which may also
CEG♭. Normal forms may be calculated using the be described as a t-​class of generic pitch-​class
algorithm from Section 10.4. Although major sets; a line genus is a mod-​7 OT-​class, a t-​class of
and minor triads belong to the same traditional generic pitch-​class lines. If any diatonic design in
set class (OPTIC-​class), they belong to different a genus is trivial (transpositionally symmetric in
species (OPTC-​classes) because I equivalence is generic space), then all are trivial; in this case we
not among the relations defining a species. In call the genus itself trivial as well.
summary, the diatonic transpositions of one A species, generally, is an equivalence class of
triad through generic space yield three different chromatic designs related by chromatic trans-
kinds of sets in chromatic space; the single genus position Ti. Because diatonic and enharmonic
of triads is realized in three different species. designs may be interpreted chromatically, a
Clough and Myerson introduced the terms unique species may be determined for any dia-
genus and species, and the associated adjec- tonic, enharmonic, or chromatic design, but not
tives generic and specific, in exactly this context, for a generic design. (A generic design may be
when studying the various possible forms that a interpreted as a diatonic design in multiple ways,
generic pitch-​class set may take when it is situ- and as we saw in the case of triads, the different
ated as a subset of a diatonic scale within chro- interpretations do not generally share the same
matic space.53 Clough and Myerson considered species.) A chord species is a mod-​12 OPTC-​class,
only lines and chords, special cases subsumed or a T-​ class of ordinary pitch-​ class sets—​ the
by the broader notion of designs introduced in same thing that in other contexts may be called a
Section 13.3. We noted there that a genus is an chord quality. A line species is a mod-​12 OT-​class,
equivalence class of diatonic designs related by a T-​class of lines in pitch-​class space.
any combination of diatonic transpositions tj A diatonic or enharmonic design realizes both
and signature transformations sk, or an equiva- a genus and a species. In such a case we may say
lence class of enharmonic designs related by also that the species realizes the genus. Every
sk. Genera may also be determined for generic genus, clearly, is realized by one or more species,
designs: if X is a generic design, the various dia- because the notes of any diatonic design in the
tonic interpretations δn(X) are identical except genus may always be interpreted as pitch classes
for their key signatures, so they are related by in chromatic space. We have seen that the line
signature transformations and therefore belong genus [0, 3, 1](OT7 ) and the chord genus [0, 2, 4](OPTC7)

to a common genus. Two generic designs belong are realized in three species each. Some spe-
to the same genus if and only if they are related cies do not realize any genus at all because they
by some generic transposition tj, leading to yet imply pitch-​class patterns not occurring in any
another definition of genus: a genus is an equiv- diatonic scale—​an obvious example being the
alence class of generic designs related by tj. Thus chord species [0, 1, 2](OPTC
12 ) . If a species does realize

Diatonic Spaces • 569

a genus, the genus is almost always unique; the seventh chords come in more species: every dia-
rare exceptions will be discussed below. tonic scale includes two major seventh chords,
Every design, of any type, can be reduced to three minor sevenths, one dominant seventh,
a chord: the unordered set of all pitch classes and one half-​diminished seventh. Whereas tri-
in the design. Consequently, any genus can be ads—​ three-​
note sets—​ were realized in three
reduced to a chord genus and any species to a species, seventh chords—​four-​note sets—​are
chord species. For example, the generic lines realized in four. Clough and Myerson’s observa-
C–​E–​G, G–​E–​C, E–​C–​G–​G, and C–​E–​C–​G–​C–​ tion, stated formally in the following theorem, is
E–​G–​ C all reduce to the generic triad CEG, that the number of species always matches the
and the corre-sponding line genera [ 0, 2, 4](OT7 ) , cardinality of the genus.
[0, 5, 3](OT7 ) , [0, 5, 2, 2](OT7 ) , and [0, 2, 0, 4, 0, 2, 4, 0](OT7 )
all reduce to the chord genus [0, 2, 4](OPTC 7)
. (OT Theorem 13.4.2: Cardinality equals variety
normal forms for line genera may be reduced Every nontrivial genus of cardinality k is
to OPTC normal forms for chord genera by the realized
   in exactly k species.
algorithm from Section 10.4.) When the line G–​
E–​C is transposed diatonically through dpc(0), Cardinality equals variety is often abbrevi-
it produces three line species, which, reduced to ated “C =​V.” Recall that the only trivial chord
chord species, match the chord species identified genera are the genera of the empty set and the
previously for the genus of triads. The generic 7-​
note generic aggregate, and that all non-
transpositions of E–​C–​G–​G also produce three empty line genera, including those of cardi-
different line species, each containing a repeated nality 7, are nontrivial. For chords and lines,
note—​but these line species reduce to the same therefore, the theorem may be recast in the fol-
three three-​note chord species as before. While lowing forms:
a line genus or any other genus may always
be reduced to a unique chord genus, a chord • C =​V for chords: If 1 ≤ k ≤ 6, then every chord
genus is potentially associated with many differ- genus of cardinality k is realized in exactly k
ent line genera—​an unlimited number, allowing species.
for both permutations and repeated notes. • C =​V for lines: If 1 ≤ k ≤ 7, then every line
In the context of designs, genera, and spe- genus of cardinality k is realized in exactly k
cies, we will use the word cardinality to mean species.
pitch-​class cardinality. That is, the cardinality of
a design is the number of distinct pitch classes It is these two forms in which Clough and
(generic, diatonic, or chromatic) in the design; Myerson presented the theorem and in which
equivalently, it is the cardinality of the design’s it has usually been stated, but both are encom-
chordal reduction. Thus the generic line E–​C–​ passed by the more general formulation given in
G–​G is of cardinality 3. All designs belonging to Theorem 13.4.2.
one genus or species are of the same cardinality, We shall prove the theorem after considering
so we may speak of the cardinality of a genus a few more examples, starting with the assort-
or species without ambiguity. In particular, the ment in the exercises below.
cardinality of a genus is the cardinality of any
design in that genus. (It is not the number of
Exercise 13.4.3 For each of the following chord
designs in the genus.) The largest possible car-
genera, determine the associated species and
dinality of a diatonic, generic, or enharmonic
verify that C =​V holds.
design, or of a genus, is 7; the largest possible
cardinality of a chromatic design or of a spe- (a) [0, 1](OPTC
7)
(generic interval class 1)
cies is 12. (b) [0, 2](OPTC
7)
(gic2)
The fact that there are three qualities of tri-
ads in a diatonic scale is elementary, but Clough (c) [0, 3](OPTC
7)
(gic3)
and Myerson recognized that it illustrates a (d) [0, 1, 2](OPTC
7)
(a three-​note generic cluster)
broadly applicable principle, one of the most (e) [0, 1, 3](OPTC
7)
(one of the two asymmetric
elegant theorems in mathematical music theory.
trichord genera from Figure 13.1.22)
Consider the generic seventh chord, the genus
[0, 1, 3, 5](OPTC7)
. In generic pitch-​class space there (f) [0, 2, 3](OPTC
7)
(the other asymmetric
are just as many seventh chords as triads, but trichord genus)

570 • E x p l o r i n g M u sical S p aces

(g) the genus of the chord CDFG The genera of cardinality 2 merit special
(h) [0, 1, 2, 3, 4](OPTC
7)
(a five-​note generic attention. A chord genus of cardinality 2 is a
generic interval class. The first three parts of
  cluster)
Exercise 13.4.3 formalize the familiar fact that
each generic interval class is realized in exactly
Exercise 13.4.4 For each of the following
two species—​for example, a third (gic2) can be
line genera, determine the associated species
either minor (ic3) or major (ic4). A line genus of
and verify that C =​V holds; also identify the
cardinality 2 is a directed mod-​7 interval: the
associated chord genus and species.
line genus [0, 5](OT7 ) in Exercise 13.4.4b also comes
(a) [0, 2](OT7 ) in two species, of which [0, 8](OT12 ) could refer to
(b) [0, 5](OT7 ) either an ascending minor sixth or a descend-
ing major third while [0, 9](OT12 ) is an ascending
(c) [0, 3](OT7 ) major sixth or descending minor third. The par-
(d) [0, 4](OT7 ) ticular case of C =​V for directed intervals (lines
of cardinality 2) is known as Myhill’s property.56
(e) [0, 3, 1, 4](OT7 ) (compare with Exercise The line species [0, 6](OT12 ) appears among the
13.4.3g) realizations for both parts (c) and (d) of 13.4.4
(f) the genus of the line E–​C–​D–​G because ic6 can occur diatonically as either an
(g)  [0, 1, 2, 3, 4, 5, 6](OT7 ) augmented fourth or a diminished fifth. Any line
species of cardinality 2 that reduces to the chord
Exercise 13.4.5 species [0, 6](OPTC
12 )
realizes two different genera,
(a) Consider a diatonic design consisting of but this is a rare property: no other line species
the two chords GBDF–​CEG in dpc(0). This realizes two genera, nor does any chord species
is a design of cardinality 6; verify that is at all. (The chord genus for both an augmented
realized in six different species. fourth and a diminished fifth is [0, 3](OPTC 7)
, which
(b) Consider a diatonic design consisting of is the only genus realized by the species [0, 6](OPTC 12 )
.)
the two chords BDFA–​CEG in dpc(0). As a chord, the generic aggregate, the unique
This is a nontrivial design of cardinality 7; generic pitch-​class set of cardinality 7, is a trivial
verify that it is realized in seven different design, realized not by seven species but by only
  species.54 one: [0, 1, 3, 5, 6, 8, 10](OPTC
12 )
, the OPTC-​class of the
diatonic collection as a subset of pc. Thus C =​V
fails for chords of cardinality 7. An ordered set
Exercise 13.4.6 The C-​m ajor fugue from
containing all seven generic pitch classes, how-
Shostakovich’s Preludes and Fugues, Op.
ever, such as the scale in Exercise 13.4.4g, is
87, is played entirely on white keys. The
realized by seven different line species, as is the
subject, shown in Figure 13.4.7, is treated in
two-​chord genus of cardinality 7 from Exercise
the course of the fugue to each of the seven
13.4.5b.
possible diatonic transpositions within the
The examples above offer few clues to suggest
white-​note scale. As a diatonic design, this
why C =​V should be true. In Figure 13.4.1, there
subject is of cardinality 5. Write the seven
is no obvious reason why the species should
diatonic transpositions of the subject, locate
be equal in number to the notes of the chord.
them in the score if possible, and verify
Because the chord genera in Figure 13.1.22 are
that they belong to exactly five different
limited in number, one could verify C =​V for
species.
   55
chords by enumerating the associated species for

FIGURE 13.4.7 Shostakovich, Twenty-​Four Preludes and Fugues, Op. 87: Fugue No. 1 in C Major,
subject

Diatonic Spaces • 571

each of them. In lieu of such a tedious exercise, The fifth-​ string describes the appearance
we will exhibit a species-​enumeration process of the set in the line of fifths by enumerating
that may be applied to the designs of any genus; the gaps between notes in order from left to
we will then establish the validity of the theo- right. We also say that the span of this spc set
rem by showing that for nontrivial designs this is 3: the span of a set is the fifth-​interval from
process produces species equal in number to the the set’s leftmost (flattest) spc to its rightmost
cardinality.57 (sharpest), and is equal to the sum of the num-
The process is a familiar one: it is the cycle bers in the fifth-​string. A fifth-​string and span
generated by the signature transformation are defined for every finite nonempty spc set.
s1. We have observed that a 12-​by-​7 array like Two spc sets have the same fifth-​string if and
Figure 13.3.18 may be constructed for any only if they are related by some spc transposi-
genus, and that if the genus is nontrivial, the tion τk; a fifth-​string may be regarded as a sort
array contains 84 distinct enharmonic designs. of normal form for spc sets.58 A set is completely
The twelve designs in any one column of the determined by its flattest spc and its fifth-​string.
array are related by chromatic transpositions Because a diatonic collection dpc(n) comprises
Ti and therefore automatically belong to the seven consecutive notes in spc, the span of a dia-
same species—​but as Figure 13.3.18 showed, tonic spc set (a subset of some dpc(n)) can never
more than one column can share a species. Our exceed 6. Conversely, every spc set of span ≤ 6
objective, therefore, is to determine how spe- is a diatonic set. Importantly for the discussion
cies are distributed among the seven columns below, one set may be a subset of more than one
of the array. diatonic collection.
The discussion below emphasizes chords, Figure 13.4.8 situates each chord of the
but at the end we will comment on the applica- s1-​
chain in the appropriate diatonic collec-
tion to other genera. The process may be dem- tion dpc(n) within spc and shows the chord’s
onstrated for any nontrivial genus; we illustrate fifth-​string and species. The species of ABD is
12 ) ,
with the chord genus [0, 1, 3](OPTC7) ,
the chordal [0, 2, 5](OPTC the OPTC normal form of ABD as
reduction of the line genus from Figure 13.3.18. an ordinary pitch-​class set (technically, the set
The beginning of an s1-​cycle for this chord genus obtained from the diatonic set ABD by the map-
appears in Figure 13.4.8. Any [0, 1, 3] chord ping π).
in any space edpc(n) could serve as the start- The first three applications of s1 do not change
ing point for the cycle; we choose to start with any of the chord’s notes. The figure shows the
ABD in edpc(0), the chord corresponding to the boundaries of the diatonic collections from
three-​note line A–​D–​B at the upper-​right corner dpc(0) to dpc(+​3) shifting rightward around the
of Figure 13.3.18, which was also the starting fixed notes D, A, and B. These four ABD chords,
point of the s1-​chain illustrated in that figure. situated in four different diatonic collections,
The chords shown in musical notation at constitute what we will call a stable segment of
the left side of Figure 13.4.8 are obtained by the s1-​chain. These are the only four diatonic
repeated application of s1 to this initial chord. collections containing the chord A♮B♮D♮; see the
These chords are labeled as enharmonic designs exercise below. Since the notes of the chord are
in the spaces edpc(n) as noted. In principle, any unchanged, the fifth-​string and species are also
of these designs could be spelled in more than unchanged by these first three applications of s1.
one way. To get started we may choose a spelling
for the first chord; for convenience we of course Exercise 13.4.9 Write a simple formula,
choose dpc(0) (rather than, say, dpc(+​12), involving the span of the set, for the number
which is also theoretically possible). This three-​ of different diatonic collections of which a
note spelled pitch-​class set A♮B♮D♮ is pictured in given
   diatonic spc set is a subset.
the top row of the figure as a subset of dpc(0), a
contiguous seven-​note subset of spc. ABD is the In dpc(+​3), D♮ is the flattest note (fa); it
mod-​7 OP normal form for the generic set, but is not a note of dpc(+​4) at all. As such, D is
the line-​of-​fifths order for this set is D–​A–​B. A altered by the next application of s1, giving
single fifth separates D and A, while two fifths way to D♯, the sharpest note (ti) of dpc(+​4).
separate A and B; that is, intspc(D, A) =​1 and The next set in the chain, consequently, is not
intspc(A, B) =​2, using the interval function intspc ABD but ABD♯. As the curved arrow in the fig-
from Section 7.2. We say that the fifth-​string of ure shows, when D moves to D♯ it jumps seven
the spc set ABD is 1–​2. places to the right, leapfrogging the other two

572 • E x p l o r i n g M u sical S p aces

 7)
FIGURE 13.4.8 Demonstration of C =​V for the chord genus [0, 1, 3] (OPTC

notes of the chord and dramatically changing At this point, in dpc(+​6), B is fa, so the next
the fifth-​string of the set, from 1–​2 to 2–​4. s1 replaces B with B♯. But A♯B♯D♯ in dpc(+​7) is
The chord’s species changes accordingly, to identical to ABD in dpc(0), except that the dia-
[0, 2, 6](OPTC
12 )
. A horizontal line separates the spe- tonic collection and all the notes of the chord
12 ) .
cies [0, 2, 5](OPTC
12 )
and [0, 2, 6](OPTC have shifted seven places to the right; the
The note A has become fa in dpc(+​4), so the initial fifth-​string 1–​2 and species [0, 1, 5](OPTC
12 )

chord changes again at the next step, to A♯BD♯ recur here. Recall that in enharmonic space
in dpc(+​5), bringing with it another new fifth-​ s7 =​ T1 by Theorem 13.3.19d; in the array of
string, 4–​1, and another new species, [0, 1, 5](OPTC
12 )
. Figure 13.3.18, the first seven applications
This chord A♯BD♯ is unaffected by the following s1. of s1 cycle through all seven columns of the

Diatonic Spaces • 573

array, back to the initial column, where the The converse—​sets of the same species have
T1 transposition of the initial design appears. the same fifth-​ string—​ holds only for diatonic
As far as species are concerned, therefore, sets, spc sets of span ≤ 6. We can see as much
the first seven designs in the s1-​chain tell us by considering the implications of enharmonic
everything we need to know; the same pattern equivalence. We could respell ABD as AC♭D, replac-
of fifth-​strings and species will simply repeat ing B with C♭, twelve places to the left. The new
over and over thereafter. In the remainder of set is enharmonically equivalent to the old—​it
this section we use the term s1-​c ycle for a seg- defines the same subset of pc—​so it certainly
ment of an s1-​chain consisting of seven designs belongs to the same species, but as an spc set it
and seven s1 transformations. An s1-​c ycle takes a very different form (its fifth-​string is 9–​1).
is therefore not a real cycle of chords, as the With a span of 10, this is of course not a diatonic
eighth chord differs from the first by T1 trans- set. If some but not all of the notes of a diatonic
position, but it is a cycle of fifth-​strings and set are respelled enharmonically, the result will
species. The first seven lines of Figure 13.4.8 almost always be a set whose span is greater than
constitute an s1-​c ycle; the three species in this 6, hence a non-​diatonic set whose fifth-​string dif-
cycle, in groupings of 4, 1, and 2 from top to fers from that of the original set. (There is a single
bottom, correspond to the species in the col- exception: if the set is a tritone, whose fifth-​string
umns of Figure 13.3.18, read from right to left. is the single number 6, transposing its leftmost
Here, then, is an intuitive explanation of note by τ12 or its rightmost note by τ−12 yields a
C =​V: in the course of an s1-​cycle, the chord respelled tritone, also with fifth-​string 6.) The
changes every time a note jumps to the right, only way that two diatonic sets can be enharmon-
and because every note in the chord jumps once ically equivalent is if the entire sets are related by
per cycle, the number of jumps is equal to the τ12i for some i (or τ6i if the set is a tritone), as is the
number of notes in the chord. case, for example, with ABD and B♭♭C♭E♭♭. In par-
This explanation is plausible and ultimately ticular, enharmonically equivalent diatonic sets
correct, but it is not quite a rigorous proof. The (even tritones) always share the same fifth-​string.
consecutive chords between jumps, in the stable To complete the proof of Theorem 13.4.10,
segments of the cycle, are all the same, so clearly suppose X and Y are diatonic spc sets of the same
they all belong to one species. Chords from dif- species. Then as pc sets, X and Y—​or technically
ferent stable segments always contain some dif- π(X) and π(Y)—​are related by some chromatic
ferent notes—​but to conclude C =​V we need to transposition in pc, say Ti(π(X)) =​ π(Y). An spc
know that such chords always belong to different transposition corresponding to Ti is any τk with
species. The three stable segments of the cycle k =​7i (mod 12). The equality of the pc sets Ti(π(X))
in Figure 13.4.8 all give rise to different species, and π(Y) does not imply that the spc sets τk(X) and
but how can we be sure that is always the case? Y are the same, but it does imply that τk(X) and Y
The question is not a simple one. The answer are enharmonically equivalent. But we have just
depends on two observations, one (implicit in noted that enharmonically equivalent diatonic
the discussion above) involving the equivalence sets always share the same fifth-​string, so in the
of species and fifth-​strings, the other about the present situation Y has the same fifth-​string as
numbers appearing in the fifth-​string for a given τk(X), which is the same as the fifth-​string of X.
chord. We present the first observation in the The second observation involves the num-
form of the following theorem. ber of times each chord species occurs in the
( 12 )
s1-​cycle of Figure 13.4.8. The species [0, 2, 5]OPTC
( 12 )
Theorem 13.4.10: Equivalence of species and occurs four times in this cycle, [0, 2, 6]OPTC once,
fifth-​strings For diatonic spc sets, species are in and [0, 1, 5](OPTC
12 )
twice. These numbers 4, 1, and 2
one-​to-​one correspondence with fifth-​strings. That are the same numbers that appear in the fifth-​
is, two diatonic spc sets have the same fifth-​string strings 1–​2, 2–​4, and 4–​1 for the three species.
if and
   only if they belong to the same species. It is not difficult to see why this is the case. For
example, the number 2 in two of these fifth-​
Half of this theorem is clear. Any two spc sets strings reflects the fact that the notes A and B
with the same fifth-​string are related by some are separated by two fifths, which ensures that
spc transposition τk; the corresponding pc sets two steps of the cycle separate the moves A♮
are therefore related by T7k, so the sets belong to → A♯ and B♮ → B♯; the stable segment between
the same species. these moves is therefore a segment of length 2.

574 • E x p l o r i n g M u sical S p aces

 Figure 13.4.11 shows the chord ABD in the complex type, a note of the design changes at
generic circle of fifths gfifth. The three numbers the same step where the corresponding note of
1, 2, and 4 are clearly visible here as the gfifth the chordal reduction changes, so the species for
intervals between successive notes around the such a genus are in one-​to-​one correspondence
circle: intgfifth(D, A) =​1, intgfifth(A, B) =​2, and with the species for the chord. It follows that C =​
intgfifth(B, D) =​4. The chord in gfifth has a V holds for every such genus.
cyclical fifth-​string that could be read in any of the Generic designs of cardinality 7, which we
rotations 1–​2–​4, 2–​4–​1, or 4–​1–​2, depending on will call complete designs, require special consid-
where one chooses to start. The fifth-​strings in eration. Some such designs are trivial and so are
Figure 13.4.8 are the same, but with the final explicitly excluded in the statement of Theorem
interval—​the wraparound interval—​omitted 13.4.2. We have noted that the trivial chord of
from each. The various fifth-​strings associated cardinality 7 does not satisfy C =​V; the last step
with the stable segments of an s1-​cycle for any in the above proof is invalid in this case because
chord are always obtained by rotating a single all rotations of the cyclical fifth-​string 1–​1–​1–​
cyclical fifth-​string in this way. 1–​1–​1–​1 are the same. But such cases are rare:
In light of this observation, the question that the only trivial designs are those in which all
lingers from our explanation of C =​V may be seven generic pitch classes are always present
rephrased and answered at last. If it is somehow simultaneously or not at all, such as a pattern of
possible for two different stable segments of an cardinality-​7 chords alternating with rests. Other
s1-​cycle to belong to the same species, then by complete designs, such as a seven-​note line or
Theorem 13.4.10 they must also have the same the two-​chord genus from Exercise 13.4.5b, are
fifth-​string. Because all the fifth-​strings in an s1-​ nontrivial, and C =​V holds for them. For non-
cycle arise through rotation of a single cyclical trivial complete genera, C =​V may be regarded
fifth-​string, it follows that two different rota- as a generalization of the intervallic distinguish-
tions of one cyclical fifth-​string must match. This ability property of the diatonic scale discussed in
means that the picture of the set in the generic Section 13.2: loosely speaking, every such genus
circle of fifths is rotationally symmetric, and has seven distinguishable “modes.”
therefore the set is invariant under some generic To see why this is true, consider a complete
transposition tj (a one-​place clockwise rotation genus, interpreted as a diatonic design in some
in the generic circle of fifths corresponds to diatonic space dpc(n). Suppose two different
generic t4). In other words, this situation is pos- diatonic transpositions belong to the same spe-
sible only for transpositionally symmetric sets cies, say X and tj(X) for some j with 1 ≤ j ≤ 6. The
in generic space, the kind we have called trivial. corresponding designs in chromatic space, π(X)
This completes the proof of C =​V for nontrivial and π(tj(X)), are then related by some chromatic
chord genera—​that is, for nonempty chords of transposition Ti. But the total pitch-​class con-
cardinality ≤ 6. tent of X is dpc(n), which as a subset of chro-
Any other kind of genus of cardinality matic space is not invariant under any Ti other
≤ 6 reduces to a chord for which C =​V holds. than T0. It follows that the chromatic designs
In an s1-​cycle for a line or a genus of a more π(X) and π(tj(X)) must be identical. Two different
complete diatonic designs in one space dpc(n)
can never be the same in chromatic space, so
this means the diatonic designs X and tj(X) are
identical as well. Once again, we have shown
that a duplication of species implies that the
original design must be trivial (transpositionally
symmetric).
This discussion of complete designs shows
that the transpositions tj within one diatonic
space dpc(n) produce all the species, as illus-
trated previously for triads in Figure 3.4.1. The
s1-​cycle in Figure 13.4.8 produces the species
FIGURE 13.4.11 The set ABD in generic fifths in a different way, via signature transforma-
space gfifth, with the cyclical fifth-​string 1–​2–​4 tions. The array in Figure 13.3.18 shows that

Diatonic Spaces • 575

either of these processes will always produce the your example summing to 12 could describe
same species. Diatonic transpositions occur on the interval structure of a transpositionally
one row of the array, while the s1-​chain traces a symmetric
   pitch-​class set.
diagonal path; either process will cycle through
all seven columns, and the chords obtained by Exercise 13.4.13
diatonic transposition are chromatic transposi- (a) In the s1-​cycle in Figure 13.4.8, verify that
tions of those in the s1-​cycle. both fa and ti occur in exactly one set per
A fifth-​string such as 4–​1 for a diatonic set species, while no other solfège syllable has
may be considered an abbreviated form of a that property.
cyclical fifth-​string 4–​1–​(2), obtained by adjoin- (b) Write the seven diatonic transpositions
ing the wraparound interval for the set in the of the chord ABD in dpc(0), transposing
generic circle of fifths. The wraparound interval, the set repeatedly by t3. Verify that the
now given in parentheses, is redundant in the transpositions produce the same species
sense that it is uniquely determined by the other as in Figure 13.4.8, in the same order
intervals: it is 7 − s, where s is the span of the and each with the same multiplicity. Why
set in spc. The above discussion shows that the does transposing by t3 (rather than t1 or
number of occurrences of a given species within something else) guarantee that the order of
one s1-​cycle is always equal to the wraparound species will be the same?60
interval of the fifth-​string associated with that (c) Of the seven sets in dpc(0) from (b), verify
species. Thus in Figure 13.4.8, the species with that the notes F and B each occur in exactly
the fifth-​string 1–​2–​(4) occurs four times, 4–​1–​ one set per species, while no other note of
(2) twice, and 2–​4–​(1) only once. This property,   dpc(0) has this property.
dubbed structure implies multiplicity by Clough
and Myerson (1985, 254), holds not only for Exercise 13.4.14 For each chord genus listed
chords but for other genera as well, using the below, enumerate the species by producing
fifth-​string of the design’s chordal reduction.59 an s1-​cycle similar to Figure 13.4.8. For each
As a consequence of this property, a species such species, also verify the “structure implies
as the one with the fifth-​string 1–​2, which takes multiplicity” property and verify that fa and ti
a compact shape in the line of fifths, will occur occur only once per species.61
more frequently than a species whose notes are (a) [0, 3](OPTC
7)
(generic interval class 3)
more widely dispersed, such as 4–​1. Exercise
13.4.15 below presents another interpretation (b) [0, 1, 4](OPTC
7)

of the wraparound interval. (c)  the genus of seventh chords

Figure 13.4.8 shows that the notes fa and ti
play special roles in the s1-​cycle. If a set in dpc(n) Exercise 13.4.15
contains the note fa of that scale, the next appli- (a) Wraparound intervals were defined above
cation of s1 will raise that note, producing a new by situating a diatonic set within a generic
species in which the raised note is ti in dpc(n +​ circle of fifths. Instead, such a set may be
1). In the seven designs in one s1-​cycle, there- drawn in the appropriate diatonic circle
fore, fa and ti each occur in exactly one set per of fifths; a picture of A♯BD♯ in dpc(+​5),
species. A species containing a tritone, such as for example, shows the correctly spelled
2–​4, is always of span 6 and can occur only once notes of dpc(+​5), including A♯ and D♯. Draw
in the cycle. pictures of the eight chords from Figure
13.4.8 in this way. The pictures will show
Exercise 13.4.12 The fact that 7 is a prime eight different diatonic collections; arrange
number is crucial to the conclusion that a all of the circles with C or C♯ at the top.
cyclical fifth-​string other than 1–​1–​1–​1–​1–​ A diatonic circle of fifths always includes
1–​1 can never be rotationally symmetric, and one diminished fifth, between ti and fa.
therefore crucial to the C =​V property. Give an Observe that the diminished fifth always
example of a string of positive integers (not all occurs within the wraparound interval
equal) summing to 8 that matches one of its for each chord. Trace the movement of
rotations. Also give a similar example summing the diminished fifth around the circle as
to 9 and one summing to 12, noting that the s1-​cycle unfolds, and observe how this

576 • E x p l o r i n g M u sical S p aces

 process illustrates the “structure implies briefer analysis of the opening theme of the Scherzo
multiplicity” property. from Beethoven’s Symphony No. 7 (Clough 1979–​80,
(b) Draw another series of diatonic circles of 480–​81).
fifths to illustrate the species of this same 9. A variety of other observations may be made
genus, this time rotating the set within the about relationships between the two parts in Figure
circle of fifths for a fixed diatonic collection. 13.1.12c. For example, one can define two contextual
  Compare these pictures with those in (a). inversion operators on three-​note generic scale seg-
ments: operator J inverts a segment in generic space
about its first note, operator K about its last note.
The first segment in the violins (generic F–​E–​D) is
NOTES K-​inverted to produce the first segment in the bass
1. Exercise 6.5.16 is relevant to the discussion here; (B–​C–​D), which is then J-​inverted to produce the next
in that exercise, similar relationships were described segment in the violins (B–​A–​G), and so on.
in terms of automorphisms of ℤ7. 10. The mod-​7 OP normal form of a triad proceeds
2. Hints for Exercise 13.1.2: Some of the functions upward from the root (CEG), but the OP normal form
in this exercise are interval functions while others of a seventh chord begins with the chordal seventh
are distance functions, and three different spaces are (BCEG, not CEGB). In the present discussion, for ease
involved. In two of the three parts, the desired inter- of recognition, we list the notes of both chords begin-
val or distance is uniquely determined by the informa- ning with the root.
tion given. In the other part, the answer is not unique, 11. Hint for Exercise 13.1.16: The group has several
but the possibilities are limited. possible generators. A generator must be a transfor-
3. Figure 13.1.7 and Example 13.1.10 are revised mation f such that repeated application of f on a chord
from Hook [2014] 2020. such as CEG will cycle through all fourteen chords in
4. The pattern in Figure 13.1.7 may be compared gtriad ∪ gsev. All transformations in the group are
with a somewhat similar sequential hierarchy in chro- of the form tn or tn ● sup* for some n. Because an appli-
matic space from Figure 4.5.25c. Chromatic sequences cation of tn can never lead from one circle to the other,
do not obey the “serial” property, nor does the chro- no tn by itself can generate the entire group.
matic hierarchy exhibit the recursive structure of the 12. Hints for Exercise 13.1.18: The equation in (a)
generic one, with the same transpositions appearing expresses a relationship between sup and sub, as func-
on different levels. tions from gtriad to gsev. To prove it, you must show
5. Hint for Exercise 13.1.8: These problems may be that if X ∈ gtriad, then the seventh chord sup(X) is
solved algebraically through the careful use of mod-​ the same as sub(t2(X)) or t2(sub(X)). How does Figure
7 multiplicative inverses. There are, however, only six 13.1.14c illustrate this relationship? For (b), you must
possible answers in each case, so the desired number show that if Y ∈ gsev, then sup−1(Y) =​sub−1(t5(Y)) =​
can probably be found quickly by trial and error. t5(sub−1(Y)). The two groups in (e) were considered in
6. The hierarchy constructed in Exercise 13.1.9 Exercise 13.1.17c, where the question was whether
arises less frequently in tonal music than the t6–​t5–​t3 they are isomorphic. The present exercise asks if
hierarchy from Figure 13.1.7, as sequences moving by they consist of precisely the same transformations on
descending step, descending third, and descending gtriad ∪ gsev—​a stronger condition.
fifth are all more common than their opposites. 13. The single-​step voice leadings CEG → ACEG
7. See Schoenberg [1947] 1975, 405–​ 06. The and CEG → CEGB require two different doublings
descending-​thirds description has not gone unchal- in the triad. A path such as ACEG → CEG → CEGB
lenged: Rothgeb (1987, 210) writes that “the idea in Figure 13.1.14c therefore cannot be realized with
that the theme of the Fourth Symphony consists of single-​step voice leading in a four-​voice setting. The
(seven) descending thirds … contains not the small- generic voice-​leading distance described in Exercise
est grain of truth, and is in fact inimical to a correct 13.1.20 is a distance function defined on triads and
understanding of the most fundamental properties seventh chords as generic pitch-​class sets, not multi-
of Brahms’s tonal language.” Rothgeb then presents a sets—​elements of mod-​7 OPC-​space, not OP-​space.
voice-​leading framework on which the theme is said to See Section 12.5 for related considerations in chro-
be truly based. matic space.
8. See also Clough 2008, 41–​44, for an analysis 14. The graph constructed in Exercise 13.1.21
using techniques related to those in this section. The is not a Cayley diagram. The functions sup, sub,
analysis here may also be compared with Clough’s sup−1, and sub−1 are defined on most, but not all, of the

Diatonic Spaces • 577

chords depicted here, and transformations sup* and 22. One could construct a musically real example
sub* are not well-​defined on this space. of a continuous mod-​7 space by dividing the octave of
15. Hook 2007d describes combinatorial tech- continuous pitch-​class space into seven equal steps,
niques for enumerating sets and set classes in a space or by imposing mod-​7 rather than mod-​12 congru-
of cardinality n. The number of pitch-​class sets is sim- ence on continuous pitch space (using the perfect
ply 2n ; there is not a simple formula for counting T-​ or fifth rather than the octave as the modulus for scale
TI-​classes. structure)—​but neither of these is a musically famil-
16. See Clough 1979, 48. Clough uses the term iar construction, and neither one corresponds to the
“diatonic chords” for generic T-​ and TI-​classes, con- intended interpretations of the generic spaces in this
ceiving them as subsets of a diatonic scale. Rather section.
than identifying them by the normal forms given 23. As we will discuss in more detail in Section
here, he employs an interval-​string notation inspired 14.2, Tymoczko distinguishes several families of lat-
by Chrisman 1971 and Regener 1974. For example, tices depending on the relationship between the car-
OPTC normal form [0, 2, 3] is (214) in Clough’s nota- dinality of the chords and that of the underlying scale.
tion, the three digits signifying intervals between Among two-​voice OP-​spaces, the mod-​12 and mod-​7
successive notes in the circular diagram. Clough also cases are of different types because 2 is a divisor of 12
calculates an interval-​class vector for each set; see but is coprime to 7.
Exercise 13.1.24 below. 24. Hints for Exercise 13.1.27c: Among other dif-
17. Recall from Table 10.3.15 that the concept of ferences, the two progressions circumnavigate their
pitch-​class set presupposes octave, permutational, respective Möbius strips a different number of times.
and cardinality equivalence. T-​ and TI-​classes of pitch-​ Also, if the progressions are realized in pitch space
class sets are therefore the same as OPTC-​ and OPTIC-​ with smooth voice leading, both starting on a perfect
classes of pitch tuples, respectively. Figure 13.1.22 fifth, one progression will arrive back at GC on a per-
is a complete enumeration of mod-​ 7 OPTC-​ and fect fifth, the other on a perfect fourth. The progres-
OPTIC-​classes. These normal forms are also OPT and sion in (a) is shown in Tymoczko 2011, 109, Figure
OPTI normal forms, respectively, but the list is not a 3.11.7; the one in (b) is implied in Tymoczko 2011,
complete enumeration of OPT-​ or OPTI-​classes when 98, Figure 3.10.2a.
multisets with doublings are considered; for example, 25. Hints for Exercise 13.1.28a: By Theorem 13.1.4,
[0, 0, 1, 3] is an OPTI normal form but not an OPTIC sum class 0 contains one representative of each OPT-​
normal form. class. Five three-​note OPT-​classes appear among the
18. Hints for Exercise 13.1.23: For (b), let n be the OPTC-​classes in Figure 13.1.22. Multisets with dou-
sum of the first and last generic pitch-​class numbers blings account for the other seven; note that [0, 0, 1],
in the generated order of the set, and consider the [0, 0, 2], and [0, 0, 3] exhibit permutational but not
inversion operator in. For (c), remember that every inversional symmetry.
interval cycle in generic space exhausts all seven pitch 26. Hints for Exercise 13.1.28b: Like Figure 11.3.1,
classes; what can you conclude about the complement this cross section is an equilateral triangle and may
of a partial cycle? The yes-​no questions in (b) and (c) be oriented with CCC at the lower right. The other
have different answers. For (d), reconsider Exercise two corners of the triangle are OP normal forms [21 3 ,
6.5.8a, modified as needed for the mod-​7 context. For 21 3 , 21 3 ] and [42 3 , 42 3 , 42 3 ], which also have OPT
(e), use the result of (d) together with the result of normal form 000 but are not lattice points. The five
Exercise 6.5.7e. lattice points without doublings lie in the interior of
19. Hint for Exercise 13.1.24b: The relationship can the triangle; the symmetrical ones fall on an axis from
perhaps be seen by inspecting a few examples, but it CCC to the midpoint of the opposite side, which is
may also be deduced from Theorem 7.5.1. the only axis about which the arrangement of lattice
20. Hint for Exercise 13.1.25: The graph should points is symmetrical. Multisets with doublings lie on
include eight vertices and sixteen edges, includ- the boundary of the triangle. Along the bottom edge,
ing loops. proceeding leftward from CCC, one of the other lattice
21. Mod-​7 spaces make occasional appearances points lies 3 7 of the way to the lower-​left corner, and
in Tymoczko 2011. See in particular pp. 103–​ 18; another lies 6 7 of the way.
Tymoczko’s Figure 4.1.4b resembles our Figure 27. Hints for Exercise 13.1.28c: The sum-​class-​0
13.1.26. Frederick 2019 examines this and several layer from (b) appears as both the bottom layer and
other mod-​7 OPTIC spaces in detail, noting the inde- the top layer of the prism, in two different rotations.
pendence of the structure of the space from the cardi- The other six sum-​class layers are arranged in between;
nality of the underlying scale. each layer is a t1/​3 transposition of the layer below it.

578 • E x p l o r i n g M u sical S p aces

Lattice points do not fall in the same positions in all starting with those in Figure 13.2.1. The formula may
the layers; the configuration of lattice points occurs in also be derived as follows. The index n of the collection
three different rotations. See also Exercise 14.2.15 for dpc(n) is the spc number of re. Use the formula for
another approach to the construction of this space. π(n) from Section 2.6 to convert this spc number to a
28. Hints for Exercise 13.1.28d: All multisets with pc number, and recall from Section 5.5 that the axis of
the note C doubled fall on the diagonal of the face symmetry through a note j corresponds to invariance
from CCC to CCC. Some doublings of D, E, A, and B under the inversion I2j.
also appear on this face (but no doublings of F or G). 37. Arthur von Oettingen exploited this property
29. Hints for Exercise 13.1.28e: When a generic of the bass-​clef staff in illustrating aspects of his dual-
trichord descends by thirds (t5), its sum class increases ist harmonic theory. See Harrison 1994, 249–​50.
by 1. It is difficult to graph such progressions clearly in 38. The deep property of the diatonic collection
these spaces, but the mod-​7 picture should look more is noted in Babbitt [1965] 2003, 195. An early study
regular than the mod-​12; the difference is analogous of deep sets is Gamer 1967; Gamer borrows the
to the difference between the sequence graphs in parts name deep from an unpublished manuscript by Terry
(a) and (c) of Figure 2.3.4. Winograd.
30. Hints for Exercise 13.1.29a: This space is a tri- 39. For the relationship between interval-​ class
angle in the same proportions as Figure 11.4.12; in vectors and common tones under transposition, see
fact, it may be obtained by slicing off that figure at an Straus 2016, 96–​100.
appropriate vertical line. As in the mod-​12 case, the 40. Well-​formed scales were introduced in Carey
three edges of the triangle contain trichords exhibit- and Clampitt 1989, with further developments in
ing P, I, and OI symmetry. The only lattice point in the Carey and Clampitt 1996.
interior of the mod-​7 triangle is the unique trichordal 41. The mapping between the scalar and generated
OPTI-​class that exhibits none of those symmetries. orders is described here in terms of generic pitch-​class
31. Hint for Exercise 13.1.29b: All these reflections numbers, but it may also be described in terms of the
are at right angles to the boundary. The path corre- two interval functions, as in the formula intgpc(x, y) =​
sponding to the loop at [0, 2, 4] does not pass through 4 ∙ intgfifth(x, y) (mod 7) from Section 13.1. As noted
a corner of the triangle. in Section 6.5, the automorphisms of a cyclic group ℤn
32. Hint for Exercise 13.2.2: The relationship between are precisely the mod-​n multiplication operators Ma
gpc and spc numbers in Table 2.6.1 may be helpful in for which a is coprime to n. In the present situation n =​
understanding why δn(k) is always congruent to 2k +​5 7 and a =​4. Because 7 and 12 are coprime and the sca-
mod 7. Exercises 2.6.2b and 2.6.7b are also relevant. lar and fifth-​generated arrangements of the chromatic
33. This interpretation of π was noted in Section pitch-​class circle are related by the mod-​12 operator
7.2, where we established that π: spc → fifth is an M7, it follows that the 12-​note chromatic scale is also
interval-​ space homomorphism. Some relationships well-​formed.
shown in Figure 13.2.1 are potentially misleading for 42. Signature transformations were introduced in
reasons connected with the dual depictions of γ and Hook 2008, but the presentation here is significantly
π. For example, as mappings of sets, the functions γ: different. The illustrations from Schubert’s B♭-​Major
spc → gpc and γ: spc → gfifth are the same, so it Sonata may be compared with more extensive exam-
should not be inferred from the arrows in the figure ples from the same piece and other works by Schubert
that the composition of γ: spc → gpc with m4: gpc in Lam 2020.
→ gfifth is equal to the mapping γ: spc → gfifth. 43. Clough 1979–​80 includes several clefless exam-
Mathematicians would say that the diagram “does not ples of this sort, with the stated aim of identifying fea-
commute.” tures of each example that are shared by all possible
34. Again we are abusing notation slightly, using realizations.
the names x and y both for spelled pitch classes and for 44. This use of the terms genus and species general-
the corresponding pitch classes: technically if spelled izes, but is consistent with, the usage in Clough and
pitch classes x and y are related by τk, then T7k relates Myerson 1985, to be discussed further in Section 13.4.
the pitch classes π(x) and π(y). Of course, Clough and Myerson did not use the OPTIC
35. The use of solfège syllables to indicate diatonic terminology, which was not available in the 1980s.
position without implying tonic or mode was sug- 45. As an interval space, edpc* is isomorphic to
gested by Nathan Lam. This convention is consistent Rings’s space sd × pc, described in Section 7.3, though
with the practice in some pedagogical traditions of the two spaces were conceived in different ways and
singing melodies in do major and la minor. for different purposes.
36. Hint for Exercise 13.2.4: You may be able to 46. Hint for Exercise 13.3.10c: Solve the congruence
induce a formula by considering a few examples, m =​ n +​7i (mod 12) for i.

Diatonic Spaces • 579

 47. Hint for Exercise 13.3.14a: In interpreting the 55. In effect the subject is presented in all seven
statement musically, you may find it helpful to break modes, but because two generic pitch classes are omit-
s5 into a composite transformation s2s3. The initial ted from the subject, two pairs of modes are inter-
s2 leads to edpc(6), where for convenience you may vallically indistinguishable. The fugal exposition, as
change spelling before continuing with s3. is traditional, alternates subjects in the tonic with
48. An s1-​chain much like the one described here answers in the dominant, here a t4 transposition.
occurs in the opening section of Michael Torke’s 1985 Because the subject lacks scale degree 7̂, neither F♮ nor
composition The Yellow Pages. A two-​ measure seg- F♯ occurs in the t4 transposition, so the theoretically G
ment of diatonic music—​a diatonic design—​is played mixolydian answer cannot be distinguished from the
thirteen times, with an additional sharp in the scale expected G major. The next transposition introduced,
at each iteration. The effect is that the music initially at m. 40, is t2 of the subject—​but this E phrygian pre-
heard in G major is in A♭ major after seven transforma- sentation is indistinguishable from aeolian because
tions and in A dorian after twelve. At this point the the subject lacks 2̂ as well. Only when the phrygian
cycle breaks with a t6 transposition back to G major, statement is answered at m. 48 with a B locrian (t6)
illustrating the equation t6 =​ t−1 =​ s−12. See Hook 2008, entry in the bass do we hear the subject in a strik-
147–​49. ingly unexpected intervallic setting. (Although only
49. For another description of this cycle of scales five modes are distinguishable in the subject itself, the
see Exercise 14.2.5d. contrapuntal texture projects all seven; for example,
50. Arithmetically, this property says that any the countersubject presented concurrently with the
integer may be written as the sum of a multiple of 12 G mixolydian answer contains F♮.) The opening of
and a multiple of 7, which is true because 12 and 7 Henryk Górecki’s Symphony No. 3, Op. 36 (Symphony
are coprime. This property is related to the Chinese of Sorrowful Songs), features a long diatonic melody
remainder theorem, mentioned in Exercise 2.5.5, as of cardinality 7, immediately subjected to all seven
well as to the calculations in Exercise 2.6.6. diatonic transpositions, all of them intervallically
51. Hint for Exercise 13.3.21b: For the “L” alterna- distinguishable.
tive, consider the signature transformation that maps 56. The name recognizes John Myhill, a mathema-
a C-​major triad in edpc(0) to an E-​minor triad in tician who worked with Clough and Myerson.
edpc(1). For the “P” alternative, having settled on t2 57. An enumeration of all the chord genera and
for “L,” choose “P” to ensure that “LP” maps a C-​major species is found in Clough and Myerson 1985, 251–​
triad in edpc(0) to an E-​major triad in edpc(4). 53. The use of signature transformations differenti-
52. This analysis is abridged from Hook 2008, ates our approach here from the proofs of the theorem
151–​58. in Clough and Myerson 1985 and 1986, but there
53. The genus-​ species terminology originates are points of contact; our representations of sets in
in Clough and Myerson 1985 and 1986, though the the line of fifths may be compared with Clough and
essential idea is found in some of Clough’s earlier writ- Myerson’s rotations of sets in the diatonic circle of
ings on diatonic theory. For example, Clough (1979, fifths. Clough and Myerson’s proofs turned out to
46) observes that A–​C–​E, A–​C♯–​E, and A–​C♭–​E “are contain a logical gap; Silverman and Wiseman 2006
all equivalent collections in a diatonic world,” just as offer a fix.
A–​C–​E, A–​B♯–​E, and A–​D♭ ♭–​E are equivalent in the 58. The interpretation of fifth-​strings as normal
chromatic world. The terminology has been adopted forms can be formalized. In principle, one could define
by many other writers, but even in the literature on an entire family of normal forms for tuples of spelled
diatonic theory and scale theory the usage has not pitch classes in spcn , closely analogous to the OPTIC
been entirely consistent, especially when the terms family of normal forms for tuples in cpitchn . P and C
are extended to scales of other cardinalities. In other equivalence mean the same things in both spaces, and
contexts the same words have been used in mark- T and I equivalence are closely analogous (referring in
edly different ways; notable examples include Forte’s spc to equivalence under τ-​transposition and inver-
(1988) system of pitch-​class set genera, Parks’s (1989) sion along the line of fifths). Mod-​12 congruence in
categorization of genera in the music of Debussy, and spc, however, refers to enharmonic rather than octave
Quinn’s (2006, 2007) far-​ reaching study of chord equivalence: spc normal forms are thus based not on
quality. OPTIC relations but on EPTIC relations. In this termi-
54. Hint for Exercise 13.4.5: Normal forms are not nology the fifth-​string 1–​2 for an spc set corresponds
defined for designs like those in this exercise; the spe- to the PTC normal form [0, 1, 3]PTC. (It is merely coin-
cies of BDFA–​CEG in dpc(0) could be recorded, for cidental that [0, 1, 3] is also the mod-​7 OPTC normal
example, in a form such as [0, 3, 6, 10]–​[1, 5, 8]. form for the chord ABD.)

580 • E x p l o r i n g M u sical S p aces

 59. This property holds even for nontrivial designs The discussion of well-​formed scales in Carey and
of cardinality 7: the cyclical fifth-​string of the chordal Clampitt 1989 delves much deeper than the brief
reduction of such a design is 1–​1–​1–​1–​1–​1–​(1), and discussion here, and Carey and Clampitt 1996
each of the seven species occurs once in the s1-​cycle. explore the concept further still, in relation to scale
60. Hint for Exercise 13.4.13b: Figure 13.3.18 may descriptions from early medieval treatises. The
suggest an answer to the last question. genus-​species terminology and the principle cardi-
61. Hint for Exercise 13.4.14: If the generic set you nality equals variety originated in the collaboration
start with does not contain ti, it will not be the “first” of music theorist John Clough and mathematician
set in its species as described above. The s1-​cycle will Gerald Myerson, who published this work twice,
then start with an incomplete enumeration of a spe- in the Journal of Music Theory (1985) and in the
cies; the remaining sets in this species will occur at the American Mathematical Monthly (1986). For a more
end of the s1-​cycle. extensive study of the mod-​7 OPTIC spaces see
Frederick 2019.
SUGGESTED READING The discussion of signature transformations here
Among important early landmarks in diatonic is a substantially new formalization of a concept
theory, Clough 1979 and Clough 1979–​80 exam- first published in Hook 2008. This formalism and
ine the structure of generic space, engaging with the enharmonic coordinates introduced here are
much of the material in Section 13.1. Browne related to a transformational system developed
1981 studies the diatonic as a subset of the chro- in Lam 2020, which also tracks a scale’s tonic and
matic, as in Section 13.2, with special attention to mode, aspects not accounted for by signature
intervallic distinguishability and related matters. transformations.

Diatonic Spaces • 581

 14
Scales II
Beyond the Diatonic

IN THIS chapter the focus broadens progres- that its surface can barely be scratched here,
sively from the diatonic to scales of other kinds. and in any case the common scales are probably
In Section 14.1 we examine non-​diatonic seven-​ the examples of greatest interest to most read-
note subsets of the chromatic scale as spelled ers. Nevertheless, a brief examination of some
pitch-​class sets, or spelled heptachords, an unorthodox scale systems may enable us to see
approach with applications in the study of chro- the distinctive features of the familiar ones in
matic harmony. Section 14.2 discusses maximal a new and revealing light, and to understand
evenness, a property often noted of the diatonic more completely what makes diatonic and chro-
scale but of considerable interest in other con- matic scales so remarkable.
texts as well; the geometric perspective pro- In some applications in this chapter it will be
vided by the OPTIC spaces proves fruitful in this useful to broaden our definitions of the words
connection. generic and specific. Until now, we have used the
There are of course many kinds of scales word generic exclusively in reference to scalar
besides those with seven or twelve notes. spaces with seven notes per octave, and specific
Section 14.3 generalizes observations from in relation to the familiar twelve-​note chromatic
this and the previous chapter to scales of other scale. These terms may be generalized to other
cardinalities, with special attention to ways in situations in which one scale (the generic) is a
which a subset of a larger scale may be consid- subset of another (the specific). In Section 14.2
ered analogous to the 7-​in-​12 diatonic scale. we will consider maximally even sets of various
This final section is, of necessity, highly selec- generic sizes within the usual chromatic scale of
tive and somewhat cursory in nature. The litera- specific cardinality 12. Section 14.3 will explore
ture on scales and microtonal tunings is so vast other specific cardinalities as well.

Exploring Musical Spaces. Julian Hook, Oxford University Press. © Oxford University Press 2023. DOI: 10.1093/oso/9780190246013.003.0014
14.1 SEVEN-​N OTE are identical as mod-​12 pitch-​class sets but very
SCALES AND SPELLED different when viewed through a diatonic lens.
HEPTACHORDS The first, a dominant seventh chord, is a subset
Figure 14.1.1 shows the dissonant opening of a unique diatonic collection, dpc(−3). The sec-
chords of Ravel’s Menuet antique alongside two ond, an augmented-​sixth chord, does not belong
modified restatements of these chords later in to any diatonic collection but may be considered
the piece. A generic aggregate—​a complete col- an inflected form of a chord from a different
lection of the seven generic pitch classes—​is diatonic collection, perhaps B♭DFG in dpc(−1).
present in each of the three passages, and in fact Attentiveness to spelling allows distinctions like
all three are identical in generic space, but one or these, crucial in tonal contexts, to be recognized.
more accidentals are altered at each subsequent We begin with some general observations
appearance. In terminology from Chapter 13, about spc sets before narrowing our focus to
the three may be considered inflected versions of heptachords. Table 14.1.2 displays the two
a single generic design. The relationships among sets just mentioned along with several others
them are similar to relationships described using as subsets of spc, the line of fifths. The dots in
signature transformations in Section 13.3, but the fifth-​orderings stand in for notes in the line
signature transformations are not applicable of fifths absent from the sets in question; we
here because only the last of the three passages, will sometimes identify spc sets in forms such
in dpc(+​6), is diatonic. as A♭∙B♭F∙∙D for the B♭7 chord or B♭F∙∙D∙∙∙∙∙G♯ for
The pitch collection in the first excerpt in is the augmented sixth. The fifth-​string and span
an F♯ harmonic minor scale. That in the second are shown for each set; these are as defined in
excerpt may also be regarded as an inflected Section 13.4, but for compactness we now omit
diatonic scale, though one of a less familiar the dashes previously included in fifth-​strings. A
type. The labels hmin(+​4) and nmin(+​3) will fifth-​string is a string of c − 1 numbers, where c
be explained in the course of this section; both is the cardinality of the set; these numbers sum
labels identify spelled heptachords—​seven-​note to the span of the set. The column headed α will
sets in which each letter name is represented be explained below.
once. A spelled heptachord may be thought of Every spc set determines an ordinary pitch-​
as a generalized key signature; the techniques class set via the projection mapping π from
described here offer one way of extending the Section 2.6. For example, the B♭7 and augmented-​
application of signature transformations to con- sixth chords in the table share the π-​projection
texts that are not entirely diatonic. As we will {2, 5, 8, 10}; the spelling information present
see, spelled heptachords offer a perspective ele- in each spc set is lost in the pc set. An spc set
gant in its theoretical aspects and one with wide-​ also determines a generic (mod-​7) pitch-​class set
ranging implications for the study of chromatic by the generic projection mapping γ. The γ-​pro-
harmony.1 jection of the B♭7 chord is {D, F, A, B}, or in gpc
Spelled pitch-​class space spc is an essential numbers {1, 3, 5, 6}, while that of the augmented
part of this perspective. In order to interpret a sixth is {D, F, G, B} or {1, 3, 4, 6}.
collection as an inflected form of something dia- The first three chords in the table are domi-
tonic, we must know how its notes are spelled. For nant seventh chords. The π-​projections of dom-
example, in a way that was tentatively explored inant sevenths are always pc sets of species [0,
in Exercise 2.6.8, the chords B♭DFA♭ and B♭DFG♯ 12 )
3,6, 8](OPTC and are all related by T-​transposition

FIGURE 14.1.1 Spelled heptachords in Ravel, Menuet antique

Beyond the Diatonic • 583

Table 14.1.2. Eight spelled pitch-​class sets

DE S C R IP T ION F I F T H -​OR DE R I N G FIF T H -​S T R IN G S PA N α

B♭7 chord B♭DFA♭ A♭ ∙ B♭ F ∙ ∙ D 213 6 −2

F7 chord FACE♭ E♭ ∙ F C ∙ ∙ A 213 6 −1
7
C chord CEGB ♭ B♭ ∙ C G ∙ ∙ E 213 6 −1
♯ø7 ♯
f chord F ACE C ∙ ∙ A E ∙ F♯ 312 6 1
Ger+​6 chord B♭DFG♯ B♭ F ∙ ∙ D ∙ ∙ ∙ ∙ ∙ G♯ 136 10 0
f♯o7 chord F♯ACE♭ E♭ ∙ ∙ C ∙ ∙ A ∙ ∙ F♯ 333 9 0
Whole-​tone scale E♭FGABC ♯ E♭ ∙ F ∙ G ∙ A ∙ B ∙ C♯ 22222 10 0
♭ ♯
Octatonic scale ABCDE FF G ♯ E♭ ∙ F C ∙ D A ∙ B F♯ ∙ G♯ 2121212 11 1

in pc; their γ-​projections (along with those of τ-​related to the standard spellings and may
other species of seventh chords) are gpc sets be asymmetrical. Generally, inversional sym-
of genus [0, 1, 3, 5](OPTC
7)
, related by t-​transposi- metry of a set in spc implies inversional
tion in gpc. In spc, dominant sevenths share symmetry of its π-​projection in pc, but not
the fifth-​string 213 and span 6, and all are vice versa.
related by spc transposition (τk)—​that is, they An spc set may contain a letter doubling such
appear identical in spc space except for left-​ as F–​F♯, an enharmonic doubling such as E♯–​F, or
right translation along the line of fifths. The f♯ø7 even a crossing such as E♯–​F♭. The π-​projection of
chord, the fourth chord in the table, is inver- an spc set with an enharmonic doubling may be
sionally related to the dominant sevenths in regarded as a multiset, the doubled notes becom-
spc, just as their π-​projections are inversionally ing an element of multiplicity 2. An spc set with
related in pc. If we write ɩ (iota) for inversion no letter doublings is called letter-​distinct; an spc
about D (spc 0) in the line of fifths, then f♯ø7 is set with no enharmonic doublings or crossings is
related to C7 by ɩ, and to any other dominant called proper. The largest possible cardinality for
seventh by some ɩτk. Inversionally related spc a letter-​distinct spc set is 7 (the octatonic scale
sets have the same span, and their fifth-​strings in Table 14.1.2 is not letter-​distinct as it con-
are retrogrades. As noted in Section 13.4, a tains the letter doubling F–​F♯), while the largest
fifth-​string may be used as a label for a τ-​trans- possible cardinality for a proper spc set is 12. The
position class of spc sets, hereafter called a τ-​ π-​projection of a proper spc set is a pc set of the
class. A pair of retrograde-​related fifth-​strings same cardinality.
together define a τɩ-​class. Section 2.6 introduced the accidental index
While a τk relation between spc sets always α(n) of the spc n, an integer that counts the
implies a T7k relation between their π-​projec- sharps or flats in a note name. We define the
tions as pc sets, a T relation between pc sets does accidental index α(X) of an spc set X to be the
not imply a τ relation between spc sets. The B♭7 sum of the accidental indices of its notes—​that
and augmented-​sixth chords in the table are not is, a count of all accidentals in the set, counting
τ-​ or τɩ-​related; they have different fifth-​strings, sharps positive and flats negative. The acciden-
and the large span of the augmented sixth sig- tal index generalizes the index numbers we have
nals its non-​diatonic status (only sets of span ≤ 6 used for diatonic collections throughout this
are subsets of diatonic collections). By Theorem book: dpc(n) is an spc set of accidental index
13.4.10, however, a T relation does imply a τ n. The final column of Table 14.1.2 displays the
relation for diatonic spc sets. accidental index of each listed set. For example,
Fully diminished seventh chords, whole-​ the accidental index of the whole-​tone scale in
tone scales, and octatonic scales, in the spell- the table is 0 (its one sharp balanced by one flat),
ings shown in Table 14.1.2, are symmetrical in and that of the octatonic scale is 1 (two sharps
spc space as they are in pc space. Nonstandard minus one flat). Other enharmonic spellings of
spellings of the same pc sets, such as {C, D ♯ , the same scales may have different accidental
F♯ , B♭ ♭ } for the diminished seventh, are not indices.

584 • E x p l o r i n g M u sical S p aces

Exercise 14.1.3 Several different notations for spelled hep-
(a) Calculate the fifth-​string, span, and tachords are useful depending on the circum-
accidental index of the spc set X =​{C, D♯, F♯, stances. For example, a G harmonic minor
B♭ ♭}. Your fifth-​string should confirm that X scale defines a spelled heptachord that may be
is asymmetrical in spc. written in the form GAB♭CDE♭F♯. This notation
(b) Give an example of two spc sets X and Y highlights the familiarity of the scale, but in
such that π(X) and π(Y) are related by some principle a spelled heptachord is an unordered
Ik in pc even though X and Y are not τɩ-​ set and does not imply a tonic. Writing hepta-
related in spc. chords in C-​scale order, in this case CDE♭F♯GAB♭,
(c) If all the sets in Table 14.1.2 are transposed can facilitate comparison of spelled heptachords
by τ1, the accidental index changes in with the white-​key diatonic scale CDEFGAB and
some cases but not in others. How can with each other. A general spelled heptachord
you determine easily what will happen takes the form Cλ0Dλ1Eλ2Fλ3Gλ4Aλ5Bλ6, where the
to the accidental index of a set under τ1 seven letter names are inflected by accidentals
transposition? λ0–​λ6, which may be identified with integers
(d) To describe how the accidental index of a (♯ =​1, ♭ ♭ =​−2, and so on). A spelled heptachord is
set changes when the set is transposed by completely determined by these seven acciden-
τ7, one needs to know only one simple thing tals, which may be presented in a seven-​place
about the set. What is it? Write a formula vector such as (0, 0, −1, 1, 0, 0, −1) for G har-
for α(τ7(X)) in terms of α(X). monic minor. Finally, the line-​of-​fifths repre-
(e) In pc there are two whole-​tone collections. sentation E♭B♭∙CGDA∙∙F♯ may be useful, because
Is it possible to write standard spellings the structure of a spelled heptachord in spc is
with fifth-​string 22222 for both of them so often of interest.
that both have the same accidental index? Spelled heptachords have some distinctive
(f) The octatonic scale in Table 14.1.2 could be properties not shared by spc sets generally,
respelled with A♭ in place of G♯. The respelled and proper spelled heptachords have some even
set has fifth-​string 1212121 and span 10, more notable properties. We summarize these
making it more compact than the spelling properties in two theorems, Theorem 14.1.5 for
in the table. There is no single correct way general spelled heptachords, and, later, Theorem
to spell an octatonic scale, but spellings of 14.1.11 for proper spelled heptachords.
the 2121212 type are generally preferred to
those of the 1212121 type. Why? What is Theorem 14.1.5: Properties of spelled
  awkward about the spelling with A♭?2 heptachords
(a) If H is a spelled heptachord and k is
Exercise 14.1.4 an integer, then τk(H) is also a spelled
(a) If n is an spc number, then ɩ(n), the heptachord, and α(τk(H)) =​ α(H) +​ k.
inversion of n about D, is the spc −n. (b) Every τ-​class of spelled heptachords
Explain why π(−n) =​ I4(π(n)) (mod 12), contains exactly one heptachord of every
where I4 is the usual inversion operator of accidental index.
that name in pc. (c) If H is a spelled heptachord with α(H) =​0,
(b) Write a similar formula relating γ(−n) and then the sum of the spc numbers of the
  γ(n) in gpc.   notes of H is equal to 0.

A spelled heptachord is defined to be a letter-​ Part (a) of this theorem involves the behavior
distinct spc set of cardinality 7. Letter-​distinct- of accidentals as a spelled heptachord is τ-​trans-
ness ensures that each generic pitch class (letter posed in spc space. The example of a diatonic
name) appears exactly once in every spelled scale is illuminating. Transposition of a diatonic
heptachord. The generic projection γ(H) of every scale dpc(n) by τ1 always alters exactly one acci-
spelled heptachord H is therefore {C, D, E, F, dental, yielding dpc(n +​1):
G, A, B}, the generic aggregate. Because letter
names recur with periodicity 7 in spc, the spc τ1(CDEFGAB) =​GABCDEF♯ =​CDEF♯GAB.
numbers of the notes of a spelled heptachord
include exactly one representative of every mod-​ In the case of a diatonic scale the signature
7 congruence class. transformation s1 gives the same result as τ1,

Beyond the Diatonic • 585

but signature transformations are available only heptachords, as we have done all along with
in diatonic spaces. Part (a) implies, more gener- diatonic scales dpc(n).
ally, that an analogous property holds for the τ1 In particular, every τ-​class of spelled hepta-
transposition of every spelled heptachord: τ1(H) chords contains exactly one balanced form H0 for
is another spelled heptachord, and its accidental which α(H0) =​0. The only spelled heptachord with
index is α(τ1(H)) =​ α(H) +​1. Whatever accidentals no sharps or flats is of course the balanced diatonic
may inflect the six notes CDEFGA of a spelled scale dpc(0). The spc numbers of dpc(0) sum to 0,
heptachord H, the notes GABCDE will appear in and any other balanced spelled heptachord may
the transposed heptachord τ1(H) with those same be obtained from dpc(0) by adding sharps and
accidentals. But because τ1(B) =​F♯, the acciden- flats in equal numbers. The spc sum increases by
tal originally on B must be adjusted one place in 7 with each added sharp and decreases by 7 with
the sharpwise direction when it appears on F in each added flat, so the sharp-​flat balance ensures
τ1(H). For example, if H is the C harmonic minor that the sum remains 0, as stated in part (c) of the
scale, then theorem.
A spelled heptachord may be regarded as a
τ1(H) =​ τ1(CDE♭FGA♭B) =​GAB♭CDE♭F♯, seven-​note space similar to a diatonic scale, and
we will use labels analogous to dpc(n) for some
while if H is the G melodic minor scale, then of the most important spelled heptachords: a
descriptive name identifying the τ-​class, written in
τ1(H) =​ τ1(GAB♭CDEF♯) =​DEFGABC♯. small capitals, followed by the accidental index of
the heptachord in parentheses.3 We may imagine
In both cases, of the accidentals in the trans- the accidentals of a spelled heptachord attached
posed heptachord only the one on F differs from to the lines and spaces of a musical staff, acting as
the original accidental. The general property may a kind of generalized key signature. For another
be written in the form metaphor, we may associate a spelled heptachord
with a positioning of the pedals of an idealized
τ1(Cλ0Dλ1Eλ2Fλ3Gλ4Aλ5Bλ6 ) harp, one that allows for unlimited numbers of
=​ Gλ0Aλ1Bλ2Cλ3Dλ4Eλ5Fλ6+ ​1: sharps or flats.
Table 14.1.6 illustrates the properties from
the first six accidentals are unchanged but λ6 has Theorem 14.1.5 as they apply to the τ-​class
been incremented to λ6 +​1, so the accidental hmin, with fifth-​string 121113, comprising the
index of the set as a whole has increased by 1. various transpositions of the harmonic minor
Because τ1 increases the accidental index scale. The balanced form of this τ-​class, hmin(0),
of a spelled heptachord by 1, it follows that is the D harmonic minor scale, which has one flat
τk increases it by k; this is part (a) of the and one sharp. Its spc transpositions hmin(n) =​
theorem. A τ-​class consists of all spc trans- τn(hmin(0)) are also shown for values of n from
positions of one heptachord, so (b) is an −2 to +​2. With each successive τ1 transposition
immediate consequence of (a). In light of (b), one note crosses either the B♭–​F or B–​F♯ bound-
we may use the accidental index as a means of ary with a resulting increment in its accidental,
indexing the elements of any τ-​class of spelled ensuring that α(hmin(n)) =​ n for every n. In

Table 14.1.6. Five representatives of the τ-​class 121113 =​ hmin (harmonic

minor scales)

HE P TAC HOR D S C A L A R OR DE R I N G F IF T H -​OR DE R IN G COMMON NA ME

hmin(−2) CDE♭FGA♭B A♭ E♭ ∙ F C G D ∙ ∙ B C harmonic minor

hmin(−1) GAB♭CDE♭F♯ E♭ B♭ ∙ C G D A ∙ ∙ F♯ G harmonic minor
hmin(0) DEFGAB♭C♯ B♭ F ∙ G D A E ∙ ∙ C♯ D harmonic minor
hmin(+​1) ABCDEFG♯ F C ∙ D A E B ∙ ∙ G♯ A harmonic minor
hmin(+​2) EF♯GABCD♯ C G ∙ A E B F♯ ∙ ∙ D♯ E harmonic minor

586 • E x p l o r i n g M u sical S p aces

accordance with Theorem 14.1.5c, the sum of the heptachords may be regarded as pitch-​class sets,
spc numbers of hmin(0) is whose chord species (mod-​12 OPTC-​classes) are
identified by OPTC normal forms in the last col-
(−4) +​(−3) +​(−1) +​0 +​1 +​2 +​5 =​0. umn; a correspondence between spelled hepta-
chords and chord species will be made explicit in
The properties described in Theorem 14.1.5 Theorem 14.1.11 below.
are specific to heptachords. The following exer- Several of the heptachord labels in the first
cise shows that they do not generally hold for spc column of Table 14.1.8 refer to familiar scale
sets of cardinalities other than 7. types, but it should be reiterated that a spelled
heptachord is an unordered set and does not
Exercise 14.1.7 imply a tonic. Thus the familiar dpc(0) could be
(a) Give an example of an spc set X and a positive C major or any of six other modes. The melodic
integer k for which α(τk(X)) < α(X) +​ k, and minor heptachord mmin(0) could be a G melodic
another example for which α(τk(X)) > α(X) +​ k. minor scale, GAB♭CDEF♯, but it could also be a C
(b) Give an example of a τ-​class of spc sets acoustic scale (a major scale inflected by ♯4̂ and
containing no sets of accidental index 0, ♭7̂), CDEF♯GAB♭.5 The balanced harmonic minor
and another example containing at least heptachord hmin(0), as seen previously in
two different sets of accidental index 0. Table 14.1.6, is D harmonic minor; hmaj(0) is
(c) Give an example of an spc set of accidental a G harmonic major scale (a major scale with ♭6̂),
  index 0 whose spc numbers do not sum to 0.4 GABCDE♭F♯.
The remaining heptachords in Table 14.1.8
Twelve spelled heptachords of types arising are as follows. A Neapolitan minor scale is a
in many analytical examples are listed in Table harmonic minor scale inflected by ♭2̂, while
14.1.8. An identifying label for each appears in Neapolitan major is a major scale inflected with
the first column, the index number 0 signify- ♭2̂; the balanced forms are nmin(0), A Neapolitan
ing that all are given in their balanced forms. minor (AB♭CDEFG♯), and nmaj(0), G Neapolitan
The notes of each heptachord are given in C-​ major (GA♭BCDEF♯).6 Double harmonic is an ethni-
scale order and in fifth-​ordering, along with the cally neutral name for the scale sometimes called
fifth-​string and span. Via the projection π the the Gypsy or Hungarian minor scale, distinctive

Table 14.1.8. Balanced forms of twelve proper spelled heptachords

NA M E C -​S C A L E F I F T H -​OR DE R IN G FIF T H -​ S PA N C HOR D

OR DE R S T R IN G S PE C I E S

dpc(0) CDEFGAB F C G
DA E B 111111 6 013568t
mmin(0) CDEF♯GAB♭ B♭ ∙ C G D A E ∙ F♯ 211112 8 013468t
hmin(0) C♯DEFGAB♭ B♭ F ∙ G D A E ∙  ∙ C♯ 121113 9 0134689
hmaj(0) CDE♭F♯GAB E♭ ∙ ∙ C G D A ∙ B F♯ 311121 9 0135689
nmin(0) CDEFG♯AB♭ B♭ F C ∙ D A E ∙ ∙ ∙ G♯ 112114 10 0124689
nmaj(0) CDEF♯GA♭B A♭ ∙ ∙  ∙ C G D ∙ E B F♯ 411211 10 0135789
dharm(0) C♯DE♭F♯GAB♭ E♭ B♭ ∙  ∙ G D A ∙ ∙ F♯ C♯ 131131 10 0125689
swt(0) C♯DE♭FGAB E♭ ∙ F ∙ G D A ∙ B ∙ C♯ 221122 10 012468t
socta(0) CDE♭FG♯AB E♭ ∙ F C ∙ D A ∙ B ∙ ∙ G♯ 212123 11 0134679
soctb(0) C♯DEFGA♭B A♭ ∙  ∙ F ∙ G D ∙ E B ∙ C♯ 321212 11 0235689
shexa(0) CD♯EFGA♭B A♭ ∙ ∙ F C G ∙ ∙ E B ∙ ∙ ∙ D♯ 311314 13 0124589
shexb(0) CD♭EFG♯AB D♭ ∙ ∙ ∙ F C ∙ ∙ A E B ∙ ∙ G♯ 413113 13 0134589

Beyond the Diatonic • 587

for its two augmented seconds and two consecu- (c) Can a finite spc set be symmetrical about
tive semitones; the balanced form dharm(0) is more than one note in spc?
GAB♭C♯DE♭F♯. The super–​whole-​tone (or whole-​ (d) Can a seven-​note pc set be symmetrical
tone-​plus-​one) collection is a seven-​note superset about more than one note of the set
of a whole-​tone scale; the balanced form swt(0)   in pc?9
adds the central D to the whole-​tone collection
from Table 14.1.2. There are two different suboc- Spelled heptachords, like other spc sets,
tatonic heptachords socta and soctb, seven-​ may have enharmonic doublings or cross-
note subsets of an octatonic collection; socta(0) ings; for instance, the whole-​tone heptachord
matches the octatonic scale from Table 14.1.2 CDEF♯G♯A♯B♯ contains the enharmonic doubling
but for the missing F♯. Finally, shexa and shexb B♯–​C. All the heptachords in Table 14.1.8, how-
are the two superhexatonic heptachords, seven-​ ever, are proper—​free of doublings or cross-
note supersets of a hexatonic collection.7 ings—​which ensures that as the seven letters
Other heptachords of the same types are proceed around the circle of generic pitch-​class
readily obtained by τ-​transposing the balanced space, the corresponding π-​projected notes
forms in the table. To write the B♭ harmonic proceed similarly around the circle of pc, with-
major scale as a spelled heptachord, for example, out pausing (for an enharmonic doubling) or
one may observe that hmaj(0) is G harmonic reversing direction (for a crossing). Proper
major and that B♭ is τ−3(G), from which it follows spelled heptachords satisfy several attractive
that B♭ harmonic major is hmaj(−3). properties beyond those satisfied by all spelled
Horizontal lines in Table 14.1.8 partition the heptachords.
heptachords into singletons and pairs. The hep- One such property involves enharmonic
tachords dpc, mmin, dharm, and swt, appear- equivalence. We have seen examples of enhar-
ing singly, are inversionally symmetric in spc monically equivalent spc sets with very different
space, as their palindromic fifth-​strings confirm. structures in spc space (for example, dominant
As a consequence of this ɩ symmetry in spc, the seventh and augmented-​sixth chords), but for
π-​projections of these sets are also symmetric proper spelled heptachords, respellings are
in pc. The other eight heptachords in the table limited to those with the same fifth-​ string.
come in four pairs, hmin/​hmaj, nmin/​nmaj, Consider, for instance, the E♭ harmonic minor
socta/​soctb, and shexa/​shexb. The two hep- heptachord hmin(−5) =​E♭FG♭A♭B♭C♭D, and sup-
tachords in each pair are ɩ-​related, and their pose we wish to respell this set as a proper
π-​projections are I-​related in pc. spelled heptachord containing B♮ rather than C♭.
Letter-​distinctness requires that the original B♭
Exercise 14.1.9 be respelled, and to avoid crossings it must be
(a) Explain why the sum of the spc respelled as A♯ rather than C♭ ♭ or G♯ ♯ ♯ or some
numbers of the notes in every spelled more remote possibility. Continuing in this way,
heptachord (balanced or not) must be we find that the only possible repelling of the
divisible by 7. Can you give a simple set is D♯E♯F♯G♯A♯BC𝄪, which is the D♯ harmonic
formula for this sum in terms of the minor heptachord hmin(+​7), the τ12 transpo-
accidental index? sition of the original set. (This process closely
(b) If H is a spelled heptachord, explain why the resembles the application of the signature trans-
sum class of π(H) (the sum of all the pitch-​ formation s12, which respells a diatonic scale in
class numbers of the notes of H) is equal to the sharpwise direction, but signature transfor-
  α(H) +​2 (mod 12).8 mations are not applicable for this non-​diatonic
collection.) In short, the only possible ways to
Exercise 14.1.10 respell a proper spelled heptachord as another
(a) The balanced form of every inversionally proper spelled heptachord are the obvious ones,
symmetric spelled heptachord is symmetric obtained by respelling the entire set in the same
about D. Explain why this balanced form way: enharmonically equivalent proper spelled
must always contain the note D. heptachords are always related by τ12k for some
(b) What can you say about the note on the axis integer k.
of symmetry of an inversionally symmetric The following theorem restates this observa-
spelled heptachord of accidental index +​1? tion in (a) and presents several other properties
Of index −1? Of index n? as consequences.

588 • E x p l o r i n g M u sical S p aces

Theorem 14.1.11: Properties of proper Part (d) follows from (c) by letting H1 and H2
spelled heptachords be the same heptachord H. Symmetry of π(H) in
pc means that π(H) is I-​related to π(H), so by (c),
(a) If H1 and H2 are enharmonically equivalent H is ɩ-​related to H—​that is, H is symmetric in spc.
proper spelled heptachords, then H2 =​ Part (b) implies that τ-​classes of proper spelled
τ12k(H1) for some integer k. heptachords are in one-​ to-​
one correspondence
(b) For proper spelled heptachords, τ with T-​classes of seven-​note pitch-​class sets, and
equivalence is the same as T equivalence. (c) likewise implies that τɩ-​classes are in one-​to-​
That is, if H1 and H2 are proper spelled one correspondence with TI-​classes. T-​classes of
heptachords, then H2 is a τ-​transposition of pc sets are OPTC-​classes (recall Table 10.3.15), also
H1 in spc if and only if π(H2) is a called chord species in Chapter 13, while TI-​classes
T-​transposition of π(H1) in pc. are OPTIC-​classes, the traditional set classes. Set
(c) For proper spelled heptachords, τɩ tables show 38 seven-​note set classes (TI-​classes),
equivalence is the same as TI equivalence. of which ten are symmetrical and each of the
That is, if H1 and H2 are proper spelled other 28 divides into two T-​classes. The number of
heptachords, then H1 and H2 are inversionally T-​classes is therefore 2 ∙ 28 +​10 =​66, as noted in (e).
related in spc if and only if π(H1) and π(H2) For example, the τ-​class mmin from Table
are inversionally related in pc. 14.1.8 contains all proper spelled heptachords
(d) For proper spelled heptachords, ɩ symmetry whose π-​projections have the symmetrical prime
is the same as I symmetry. That is, if H form 013468t, which is both the OPTIC and
is a proper spelled heptachord, then H is OPTC normal form for the twelve pc sets in that
inversionally symmetric in spc if and only if class. The τ-​classes hmin and hmaj both contain
π(H) is inversionally symmetric in pc. heptachords of prime form (OPTIC normal form)
(e) There are 66 τ-​classes of proper spelled 0134689; the heptachords in hmin have OPTC
heptachords, corresponding to the 66 seven-​ normal form 0134689 while those in hmaj have
note chord species (mod-​12 OPTC-​classes, the inversionally related OPTC normal form
or T-​classes of seven-​note pitch-​class sets). 0135689. It is the τ-​classes, not the τɩ-​classes,
There are 38 τɩ-​classes of proper spelled to which we have assigned labels such as hmin
heptachords, corresponding to the 38 seven-​ and hmaj and to which we will devote the bulk
  note set classes (mod-​12 OPTIC-​classes). of our attention. For a table of the 66 τ-​classes of
proper spelled heptachords, including the span,
Parts (b)–​(d) of this theorem are “if and only fifth-​string, balanced form, OPTC normal form,
if” statements. In each case one implication Forte number, and other information for each,
(τ equivalence implies T equivalence, τɩ equiva- see Hook 2011, 96–​97.10
lence implies TI equivalence, and ɩ symmetry
implies I symmetry) holds for spc sets in general. Exercise 14.1.12 Write the spcs in each of the
We must demonstrate the converses, which are spelled heptachords listed below.11
specific to proper spelled heptachords; all follow
from (a), which was established above. (a) mmin(+​3)
To prove (b), suppose π(H2) =​ Ti(π(H1)). (b) hmaj(−4)
Because Ti is the same as τ7i up to enharmonic (c) nmin(+​5)
equivalence, it follows that π(H2) =​ π(τ7i(H1)), (d) swt(−6)
which means that τ7i(H1) and H2 are enharmoni- (e)  swt(+​6)
cally equivalent. Both are proper spelled hep-
tachords, so (a) implies that τ7i(H1) and H2 are Exercise 14.1.13 Each of the following
related by some τ12k, from which it follows that heptachords, written in C-​scale order, is some
H1 and H2 are τ-​related. spc transposition of one of the heptachords in
For (c), suppose π(H2) =​ Ik(π(H1)). The set Table 14.1.8. Identify each of them by name and
π(H1) is also I-​related to π(ɩ(H1)), from which it accidental index.
follows that the pc sets π(ɩ(H1)) and π(H2), both
I-​related to the same set π(H1), are T-​related to (a) C♯DE♯F♯GA♯B
each other. By (b), therefore, ɩ(H1) and H2 are (b) C♭D♭E♭F♭GA♭B♭
τ-​related, from which it follows that H1 and H2 (c) C♭DE♭F♯GAB♭
are inversionally related in spc (that is, they are (d) C♯DEF♯GAB♭
related by ɩτj for some j). (e)  CD♭E♭F♭G♭A♭B♭

Beyond the Diatonic • 589

Exercise 14.1.14 The heptachords in Table interpretations δ0(X) and δ−1(X). If H is hmin(0)
14.1.8 feature relatively uniform step sizes, a (C♯DEFGAB♭, from Table 14.1.8), then X ∕ H is the
characteristic of most common scales. The notes spc set C♯EGB♭, a diminished seventh chord; we
of some other heptachords are more tightly may write CEGB ∕ hmin(0) =​C♯EGB♭. Similarly,
packed, the most extreme example being the CEGB ∕ dharm(0) is C♯E♭GB♭, an augmented-​sixth
chromatic heptachord of prime form 0123456. chord. In these examples the design X is an unor-
Write the balanced form of this set as a spelled dered set (chord), so the order in which the notes
heptachord; also write its fifth-​string and span. are written does not matter either in X or in X ∕ H.
What quality of “second” or “seventh” is the The same kinds of inflections are possible, how-
interval formed by the notes at opposite ends of ever, for lines and other designs in which order
the
  chromatic cluster?12 is significant.
A variety of intuitions may shape our under-
Exercise 14.1.15 The super–​whole-​tone standing of the role of the heptachord H in this con-
heptachord swt(0) from Table 14.1.8 consists of struction. We will sometimes refer to H as a field of
a whole-​tone scale plus the note D. Suppose we spcs defining the notes available in a particular spc
replace the D in this set with E, a different whole-​ design. This conception of a field is somewhat akin
tone outlier. The old and new sets are different, to Tymoczko’s idea of macroharmony (Tymoczko
but they are T-​related as pitch-​class sets and they 2011, especially Chapters 1 and 5), described as
both have accidental index 0. If T equivalence a collection of notes controlling the pitch content
implies τ equivalence by Theorem 14.1.11b, why of music over a certain span of time. We may also
does this example not violate Theorem 14.1.5b, regard H as a filter through which the notes of X are
which states that one τ-​class cannot contain two passed, determining their chromatic inflections;
heptachords
   of the same accidental index? this perspective relates to the filtered point-​sym-
metry models to be discussed in Section 14.2.
In Section 13.2 we defined the diatonic inter- Figure 13.3.13 showed transformations of a
pretation functions δn: gpc → dpc(n). For any diatonic scale by diatonic transposition (t1 of C
integer n, δn maps every generic pitch class to the major =​D dorian), chromatic transposition (T1 of
note with the same letter name in the diatonic C major =​D♭ major), and signature transformation
collection dpc(n). Later in Chapter 13 we applied (s1 of C major =​C lydian). Analogous transforma-
δn more broadly to gpc sets, lines, and generic tions may be performed on scales defined by other
designs of various kinds. A similar construction spelled heptachords—​indeed, not merely on scales
is possible for spelled heptachords of other τ-​ but on any design X ∕ H. Figure 14.1.16 illustrates
classes besides the diatonic. In principle we could in the case of the D harmonic minor scale. Though
define a large family of functions δ MMIn
n , δ n , and
HMIn
we have informally been describing some spelled
so on; for example, δ 2 applied to the gpc D
HMIn
heptachords as “scales,” a spelled heptachord is
yields the spc D♯, because the note of hmin(+​2) an unordered set of spcs and does not capture the
with the letter name D is D♯. Such an abundance note order of a scale. The D harmonic minor scale
of functions is unwieldy in analytical practice, in the figure is therefore formalized as X ∕ H, where
however, so we will adopt a simpler notation. X is a generic D scale (an ordered set of seven gpcs,
Suppose X is a generic design as defined in D–​E–​F–​G–​A–​B–​C) and H is the spelled heptachord
Section 13.3—​ any musical fragment whose hmin(0). The boxes at the top of the figure display
notes are generic pitch classes—​and H is a spelled the components from which the spc designs below
heptachord. Imagining H as a generalized key are formed: the generic designs X and t1(X) in the
signature, we may read the gpcs of X as spcs by outlined box and the spelled heptachords H and
attaching the accidentals of H to the correspond- τ1(H) in the shaded box. Note order matters in the
ing gpcs wherever they appear; this process, tan- generic designs but not in the spelled heptachords,
tamount to applying a particular function δ as which are shown in black noteheads in C-​scale
described above, reinterprets X as an spc design order. Several subsequent figures will similarly
rather than a gpc design. This spc design will be display generic designs in open boxes and spelled
called X inflected by H, denoted X ∕ H. heptachords in shaded boxes.
For example, suppose X is the generic sev- Three transformations of the scale X ∕ H are
enth chord CEGB. If H is dpc(0), then X ∕ H is shown. First, we may transpose the notes of X
the major seventh chord C♮E♮G♮B♮, whereas if generically by t1, leaving the heptachordal field
H is dpc(−1), X ∕ H is the dominant seventh H unchanged; the resulting spc design t1(X) ∕ H
chord CEGB♭; these are the same as the diatonic is a different mode of the same scale, starting

590 • E x p l o r i n g M u sical S p aces

FIGURE 14.1.16 Three transformations of a D harmonic minor scale

on E. Alternatively, we may transpose the entire function from hmin(+​4) to nmin(+​3) that pairs
spc design X ∕ H by τ1, shifting each of its notes notes with the same letter name, this strategy
one position to the right in the line of fifths. As would result in an unmanageable profusion of
always, transposing spcs by τ1 has the effect of notations; simply recording the change of field is
transposing the corresponding pcs by T7; this sufficient for our purposes.
transposition changes the original D harmonic The use of the word mode in the above discus-
minor scale to A harmonic minor, τ1(X ∕ H). sion merits elaboration. Any proper spelled hep-
Finally, we may change the heptachordal field H tachord H may, like hmin(0) in Figure 14.1.16, be
while leaving the generic scale X unaltered. This realized as a scale via a construction of the form
process is a kind of inflection, changing only the X ∕ H, where X is a generic scale. Transposing X
accidentals of the spc design. When H is trans- generically enables the scale to start on any of
posed by τ1 it becomes hmin(+​1), the A harmonic its seven notes, yielding seven different modes
minor collection; the D scale X inflected by this of the heptachord. The seven modes of a proper
new heptachord τ1(H) is X ∕ τ1(H), the D mode of spelled heptachord always have distinct interval
the A harmonic minor scale. The figure shows a patterns: if two modes were the same, the pc set
special arrow style that will signal inflections in π(H) would be invariant under some transposi-
subsequent analytical illustrations. tion Tn, which is impossible because there are no
Although we have described the scale altera- transpositionally symmetric pc sets of cardinal-
tions in Figure 14.1.16 as transformations, a ity 7 (a principle noted first in Exercise 6.5.8 and
mathematical function is not necessarily the several times in Chapter 13).
most convenient description for a general inflec- In Figure 14.1.17, the seven modes of the D
tion—​a change of heptachordal field. The three harmonic minor scale are arranged in successive t1
Ravel excerpts from Figure 14.1.1, for example, transpositions from left to right across the center
are related to each other by inflection, as the of the figure. The modes are numbered with mod-​
field changes first from hmin(+​4) to nmin(+​3) 7 integers so that each mode of the balanced form
and then to dpc(+​6). While we could define a of the heptachord H =​ hmin(0) is assigned the gpc

Beyond the Diatonic • 591

FIGURE 14.1.17 The seven modes of harmonic minor scales obtained by t-​transposition or
τ-​transposition

number of its starting note; mode 0, for example, fact X ∕ τ1(H) is equal to τ1(t3(X) ∕ H)—​and their π-​
is the balanced heptachord’s C scale, which in this projections in pc space are related by T-​transposi-
example happens to start on C♯. With diatonic tion. Designs of different modes cannot be τ-​ or
scales we are familiar with the idea that the seven T-​related.
modes can be obtained in two different ways, by The same pattern arises for every proper
starting the same scale on different notes or by spelled heptachord H. The generic design X need
changing the key signature. Something analogous not be a scale. As long as X is complete and non-
is true of proper spelled heptachords in general, trivial as defined in Chapter 13—​that is, as long
as illustrated by the inflections of the C scale run- as X contains all seven generic pitch classes and
ning up and down the left side of the figure. The is not invariant under any nonzero generic trans-
inflections are obtained by τ-​transposing the hep- position—​the spc design X ∕ H gives rise to seven
tachord, producing the other harmonic minor col- distinguishable modes, which may be obtained
lections hmin(n) =​ τn(H). Transposition of H by τ1 either by t-​transposing X or by τ-​transposing H. In
adjusts the numbered mode by 3 mod 7, the same the general (non-​scalar) case the modes may more
mode shift as generic transposition of X by t3 (as appropriately be called species, using the sense of
when a major scale is altered to become lydian). the word introduced in Chapter 13. Designs of the
Vertically, the modes repeat in a mod-​7 pattern: same species are τ-​related in spc and T-​related in
for any seven consecutive values of n, the C-​scales pc, while designs of different species are not. If H
of hmin(n) exhaust the seven modes. The figure is a diatonic heptachord, the seven species illus-
could be filled out as a seven-​by-​seven array by trate the “cardinality equals variety” principle for
supplying D-​, E-​, …, and B-​modes for each scale nontrivial designs of cardinality 7, as detailed in
X ∕ τn(H) in the first column; a pattern of the same Section 13.4—​but as described here the analo-
kind would appear in every column. Two designs gous seven-​mode principle holds for any complete
of the same mode, such as t3(X) ∕ H and X ∕ τ1(H) nontrivial generic design and any proper spelled
(both mode 3), are related by τ-​transposition—​in heptachord, diatonic or otherwise.

592 • E x p l o r i n g M u sical S p aces

Exercise 14.1.18 C =​V does not hold in full (c) With X and H as in Figure 14.1.17, write the
generality for non-​diatonic heptachords. Give scale τ1(X ∕ H). By comparing τ1(X ∕ H) with
an example of a generic design X of cardinality X ∕ τ1(H) in the figure, write an equation of the
k < 7 for which the seven spc designs form τ1(X ∕ H) =​ tb(X) ∕ τ1(H) for some integer b,
tj(X)
   ∕ hmin(0) do not form exactly k species. implying the more general equation τk(X ∕ H)
=​ tbk(X) ∕ τk(H). Check that your formula holds
Exercise 14.1.19 The arrangement of Figure when X is the generic line E–​F–​G–​A, H is
14.1.17 is similar to that of Figure 13.3.18,   swt(−3) =​CDEFG♭A♭B♭, and k =​4.
which showed the 84 diatonic realizations
of a line genus, but there are differences. The spelled-​heptachord perspective enables
Horizontally adjacent realizations in the earlier us to formalize and broaden our understand-
figure were related by t4 rather than t1 as in the ing of chromatic inflections. In principle any
present figure, and in the vertical dimension generic design may be inflected by any spelled
the ordering is reversed. Also, the earlier figure heptachord; if the design is complete and non-
assumed enharmonic equivalence via the use trivial, seven different species will be available
of the spaces edpc(n) rather than dpc(n). No for each of the 66 τ-​classes of heptachords, mak-
such assumption underlies Figure 14.1.17, ing a total of 462 different musical realizations
which theoretically has infinitely many distinct of the same design, all of which are inflectionally
rows (though the modes repeat in a seven-​row related but no two of which are τ-​related in spc
pattern). or T-​related in pc. While many of these realiza-
It is possible, however, to define enharmonic tions may be considered musically unsuitable for
versions of spelled heptachords. The a variety of reasons, many others—​especially
enharmonically equivalent E♭ and D♯ harmonic those constructed from the common heptachord
minor heptachords hmin(−5) and hmin(+​7) types from Table 14.1.8—​may be viable, and
may be identified with the enharmonic space composers may exploit the similarities and dif-
ehmin(7), whose index number is the mod-​12 ferences among them.
integer 7. If X is the generic C scale as in Figure Figure 14.1.21 illustrates with 21 different
14.1.17, are the designs X ∕ hmin(−5) and inflections of the generic design BDFA–​CEG—​a
X ∕ hmin(+​7) enharmonically equivalent? If not, generic B7 chord resolving to a generic C triad,
how   are they related? a design we will call “the VII7–​I design.” All are
shown in the transpositions at which the root of
Exercise 14.1.20 the C chord is C♮; the two chords in the design
(a) The equation X ∕ τ1(H) =​ τ1(t3(X) ∕ H) relating are unordered sets, which in musical examples
two different designs of the same species may appear in inversions different from those
was noted above. Reiterating this τ1 shown. Analytical notations show that several
transposition k times produces the more of the realizations are common in the tonal rep-
general formula X ∕ τk(H) =​ τk(t3k(X) ∕ H), ertoire, but the others occur in the literature as
valid for any generic design X, spelled well, some with some frequency. Across the 21
heptachord H, and integer k. Show that if realizations, the first chord of the pair in particu-
X is a generic C scale, H is hmin(−5), and lar exhibits a wide variety of qualities beyond the
k =​12, this formula confirms the familiar types of seventh chords, and a variety
observation at the end of Exercise 14.1.19.13 of voice-​leading relationships to the triad that
(b) Several other formulas describing follows.
relationships among generic transposition,
spc transposition, and field transposition Exercise 14.1.22 Each of the examples below
may be derived. By comparing the two is an inflection of the VII7–​I design belonging
designs of mode 1 appearing in Figure to the same species as one of the examples in
14.1.17, determine an integer a for which Figure 14.1.21. Identify each spelled heptachord
t1(X) ∕ H =​ τa(X ∕ τ−a(H)). Repeating this t1 by name and accidental index. Only two of
transposition j times, conclude that tj(X) ∕ H these examples are the same species; which two
=​ τaj(X ∕ τ−aj(H)). This formula is again valid are these?
for every X and H. Check that it holds when
X is the generic seventh chord ACEG, H is (a) AC♭E♭G♭–​B♭D♭F (Chopin, Nocturne in B♭
dharm(+​2) =​ CD♯EFG♯AB, and j =​3. Minor, Op. 9, No. 1, mm. 80–​81)

Beyond the Diatonic • 593

FIGURE 14.1.21 Twenty-​one inflections of the VII7–​I design BDFA–​CEG

(b) C♯E♭GB♭–​DFA (Clara Schumann, Ballade, case one or more notes may be respelled to
Op. 6, No. 4, mm. 109–​10) produce an inflection of the VII7–​I design.
(c) CE♭G♭B♭–​D♭FA (Tchaikovsky, Romeo and Make the necessary respellings and identify the
Juliet, m. 193) heptachords; if possible, consult the scores and
(d) BD♭FA–​CEG (Franck, Symphony in D consider possible reasons for the composers’
Minor, I, mm. 135–​36) spelling choices. One of the examples, when
(e) C♯EG♯B♭–​DF♯A (Bruckner, Symphony No. 9, respelled, matches a species shown in Figure
II, mm. 1–​22) 14.1.21; the other does not, though the
(f) C♯EG♯B♭–​DFA (same movement, heptachord type matches one in the figure.
mm. 47–​50)
(g) G♯BD♭F–​ACE (Brahms, Symphony No. 4, (a) D♭EA♭B–​FAC (Rachmaninoff, “Daisies,” Op.
III, mm. 78–​79) 38, No. 3, mm. 2–​3)
(h) D♯F𝄪A♯C♯–​EG♯B (same movement, m. 149) (b) CE♭G♭B♭–​EGB (Szymanowki, Etude in B♭
(i) GBDF–​AC♯E (Brahms, Trio in A Minor, Op.   Minor, Op. 4, No. 3, mm. 22–​23)
114, I, mm. 221–​22)
(j) EG♯B♭D–​FAC (Strauss, Sonata for Cello and Exercise 14.1.24 Consult scores of the two
Piano, Op. 6, III, m. 2) Brahms works below and identify the spelled
(k) EG♯B♭D♭–​FAC (Strauss, Till Eulenspiegels heptachord that governs each passage.
lustige Streiche, mm. 47–​49)
(l) DF♭A♭ ♭C♭–​E♭G♭B♭ (Dukas, Piano Sonata in E♭ (a) Violin Concerto, Op. 77, I, mm. 90–​97 (first
Minor, I, mm. 161–​62) solo entrance)
(m) ACEG–​B♭DF (Ravel, String Quartet, I, m. 1, (b) Trio in A Minor, Op. 114, I, mm. 217–​24
beats 3–​4 (end of movement). (The final cadence
(n) CE♭GB♭–​D♭F♭A♭ (Holst, The Planets, “Venus,” was analyzed in Exercise 14.1.22i above.
   mm. 4–​5) Do the scale passages present the same
  heptachord?)
Exercise 14.1.23 The spelling of notes is
generally crucial in working with spelled Spelled heptachords offer possible interpre-
heptachords, but composers do not always tations of a wide assortment of chromatic phe-
spell things as we might expect. Two examples nomena at a more granular level than is often
are given below in the original spellings, possible with other approaches, in that notes
which involve letter doublings, but in each and chords that might otherwise be dismissed

594 • E x p l o r i n g M u sical S p aces

FIGURE 14.1.25 Enharmonic equivalence and inflection of spelled heptachords in Liszt, Valse
oubliée No. 1

as mere embellishments or byproducts of voice enharmonic equivalence is strong in a sense sim-

leading may be assigned to specific scalar con- ilar to that described in Section 13.3: not only
texts.14 Of countless analytical examples that are all the notes enharmonically equivalent, but
could be shown at this point, we will present the underlying heptachords are as well.
only a few, suggesting others as exercises. The F♯-​major and G-​ minor tonic triads in
Among unambiguous uses of non-​ diatonic the Liszt example are potentially related by the
heptachords, the classes hmin and mmin are triadic transformation we know as P′ or slide
particularly common. Minor-​key music typically from Section 8.2, but the heptachordal inflec-
mixes dpc, hmin, and mmin heptachords freely. tion models a more comprehensive relationship
Occasionally, however, one of the non-​diatonic encompassing tonic, dominant, and decorative
heptachords is used exclusively over an extended harmonies and a wide-​ ranging melodic line:
span. Figure 14.1.25 shows how the first main a sixteen-​measure passage starting at m. 124
section of Liszt’s first Valse oubliée is trans- inflects the earlier music from m. 17 note for
formed from the original F♯ major to G minor at note, the only exceptions being a very few
its later reappearance.15 The figure postulates a extrascalar notes such as the G𝄪 visible in the
two-​step transformation, first via enharmonic score at m. 22.16 The inflection from dpc(−6)
respelling from dpc(+​6) (F♯ major) to a hypo- to hmin(−1) alters five of the heptachord’s
thetical dpc(−6) (G♭ major), then by inflection seven notes; one of the two unaffected notes,
from dpc(−6) to hmin(−1) (G harmonic minor). E♭ (originally D♯), plays an important pivot role,
The enharmonic change combines a t1 appearing prominently in the melody on every
transposition of the generic notes with a τ−12 downbeat and functioning as 6̂ in both keys. The
respelling of the diatonic heptachord. The flat- music is governed by hmin(−1) until m. 144, at
wise enharmonic respelling ε−1 shown here acts which point the transformation is undone, E♭
on any design X ∕ H by the formula ε−1(X ∕ H) =​ again becoming D♯ in F♯ major.
t1(X) ∕ τ−12(H); the counterbalancing sharpwise Near the end of the A section of the third
enharmonic respelling ε1 is given by ε1(X ∕ H) =​ movement of Dvořák’s second piano quartet,
t−1(X) ∕ τ12(H). As a consequence of Theorem an mmin heptachord governs a twelve-​measure
14.1.11a, respellings of these types account span of music, the beginning of which appears
for all possible instances of enharmonic equiv- in Figure 14.1.26. The alternation of tonic triads
alence for realizations of complete generic with modally mixed plagal embellishing chords
designs by proper spelled heptachords. This is characteristic of the harmonic major scale,

Beyond the Diatonic • 595

FIGURE 14.1.26 The spelled heptachord mmin(−5) in Dvořák, Piano Quartet in E♭ Major, Op. 87, III,
mm. 94–​96

FIGURE 14.1.27 Inflections in Schubert, Impromptu in G♭ Major, D. 899 (Op. 90), No. 3, mm. 35–​38

here hmaj(−4) =​C♭DE♭FGA♭B♭. But the violin’s two-​bar segments are generically identical
echo of the piano’s melodic motive persists in but inflected differently. Perhaps surprisingly,
stating D♭, not D♮, situating the whole passage Schubert presents the darker inflection first,
in mmin(−5) (C♭D♭E♭FGA♭B♭). The E♭ mode of this drifting flatward as far as mmin(−9) and, for the
scale is not either one of the two modes most augmented-​sixth chord in the second measure,
commonly associated with mmin, the melodic dharm(−8). The second two-​bar presentation is
minor and acoustic scales. a nearly diatonic restatement of the same music.
The same rare mmin mode—​a major scale with In the opening measures of Gounod’s Faust,
♭6̂ and ♭7—
̂ ​appears within the C♭-​major episode shown in Figure 14.1.28, generic seventh chords
in the middle section of Schubert’s Impromptu DFAC resolve to generic triads CEG four times,
in G♭ Major, shown in Figure 14.1.27. The two but the species is different each time, and none

596 • E x p l o r i n g M u sical S p aces

FIGURE 14.1.28 Inflections of the generic design DFAC–​CEG in Gounod, Faust, Act I, Introduction,
mm. 2–​8

of the four are diatonic. The DFAC–​CEG design is the applicable heptachords and considering
of cardinality 6; the heptachords identified in the the relationships between the scales and the
figure assume that the missing note in every case underlying harmonies.
is B♮, the leading tone of the chord of resolution.
Figure 14.1.29 presents a spelled-​heptachord (a) Mozart, Sonata in F Major, K. 533/​494, I,
analysis of a more extended passage, the cen- mm. 82–​87
tral section of Chopin’s first Etude. Between the (b) Mozart, Don Giovanni, Overture, mm. 23–​26
cadence chord for the opening period at m. 16 (c) Schubert, Sonata in A Major, D. 664
and the thematic return at m. 49, an extended (Op. 120), I, mm. 57–​64
t3 sequence makes five nearly complete circuits (d) Chopin, Nocturne in B Major, Op. 62, No. 1,
of the generic circle of fifths, shortcut only by mm. 21–​24 and 81–​88
a descending-​ third move at mm. 20–​ 21 and (e) Brahms, Sextet in B♭ Major, Op. 18, II,
an ascending third to the dominant at m. 48.   mm. 65–​80 (Variation 3)
Almost every chord succession is analyzed as
t3, τ−1, or an inflection with no change of root. Exercise 14.1.31 The passage in Figure 14.1.27
Only mm. 38–​45, lodged in C major, strikes the is one of several instances in Schubert’s G♭-​
ear as obviously sequential; elsewhere the under- Major Impromptu of inflected repetition with
lying regularity is obscured by the variety of little or no change in generic structure. Analyze
heptachordal fields. An initial tonicization of A each of the pairs of passages listed below using
minor (hmin(+​1)) is followed by two short chro- spelled heptachords.
matic descending-​fifths sequences (mm. 23–​29
and 35–​38) illustrating the identity τ−1(X ∕ H) (a) mm. 13–​14 and 21–​22
=​ t3(X) ∕ τ−1(H), a special case of the equation (b) m. 30 and m. 31
from Exercise 14.1.20c. The analysis interprets (c) mm. 42–​43 and 46–​47
the A♭-​minor triad at m. 32 as an fø7 chord miss- (d)  mm. 74–​77 and 78–​8118
ing its root and postulates two enharmonically
equivalent spellings for the French augmented Exercise 14.1.32 Analyze the first phrase (mm.
sixth in m. 34, one rooted on B♭ and the other 1–​12) of Chopin’s Prelude in E Minor, Op. 28,
(theoretically the next link in the descending-​ No.
  4, using spelled heptachords.19
fifths chain) on E. This enharmonic equivalence
is weak in the sense that while the chords are 14.2 MAXIMAL EVENNESS
enharmonically equivalent, the tritone-​related AND THE GEOMETRY OF
fields they inhabit (dharm(−5) =​C♭DE♭F♭GA♭B♭ SCALES
and dharm(+​1) =​ C♯DEFG♯AB♭) are not.17
Consider a circular dinner table with twelve
Exercise 14.1.30 Each of the following chairs, and suppose seven people wish to sit
excerpts features a series of scale passages in seven of the chairs, distributing themselves
outlining several different spelled heptachords around the table as evenly as possible. How
in succession. Analyze each of them, identifying should they do it? The answer is that they should

Beyond the Diatonic • 597

FIGURE 14.1.29 Spelled-​heptachord analysis of Chopin, Etude in C Major, Op. 10, No. 1, mm. 16–​49 (chordal reduction)
occupy a diatonic subset of the twelve chairs, augmented triad, diminished seventh chord,
with gaps between occupied chairs in the pattern and whole-​tone scale.
2–​2–​1–​2–​2–​2–​1. Maximal evenness is one of the Cardinalities 8, 9, and 10 are almost as
most frequently noted properties of the diatonic straightforward. These numbers are not divisors
collection, explored at length in Clough and of 12, and the ME sets—​the octatonic, ennea-
Douthett 1991 and in much subsequent work tonic, and tritone-​complement collections—​are
in scale theory. Though the property is readily not perfectly even, nor are they generated by a
grasped intuitively, it is perhaps not clear what single interval. But these sets are the comple-
precisely is meant by “as evenly as possible.” We ments of the ME sets of cardinalities 4, 3, and
examine this question in this section and pres- 2 respectively, and the sets related by comple-
ent some justification for the musical relevance mentation are similar in other ways as well. The
of the concept, acknowledging at the outset that enneatonic collection (d =​9), for example, may
the technical nature of some of this work pre- be regarded as the union of three augmented
cludes anything like a complete discussion; some triads (ic4 cycles of cardinality 3 =​12 − d); alter-
properties will be stated without proof.20 natively, it may be constructed as a cluster cycle,
Figure 14.2.1 illustrates maximally even sets cyclically generated not from a single start-
(commonly abbreviated “ME sets”) of each car- ing note but from a three-​note chromatic clus-
dinality d from d =​2 to d =​10 within pitch-​class ter (three adjacent white notes in the picture)
space pc. The pitch classes shown in black and combined with copies of itself in a cycle of T4
white form complementary sets; if a set is ME, transpositions.
so is its complement. For some cardinalities, As in several other situations we have encoun-
the maximally even set is obvious. If d is 2, 3, tered, something is distinctive about the numbers
4, or 6, as shown in black notes in four of the 5 and 7, which are coprime to 12. The penta-
pitch-​class circles in the figure, then d is a divi- tonic and diatonic collections are the only sets
sor of 12, and the maximally even distribution in Figure 14.2.1 that are neither perfectly even
is perfectly even: it is a division of the octave interval cycles nor cluster cycles, and maximal
into d equal parts, a complete interval cycle evenness is probably less obvious for these sets
generated by an interval of 12/​d semitones. than for the others. Figure 14.2.2 shows four sets
These sets are of course the familiar tritone, of cardinality 7, labeled with spelled-​heptachord

FIGURE 14.2.1 Maximally even subsets of pc

FIGURE 14.2.2 A maximally even seven-​note set, and three impostors

Beyond the Diatonic • 599

 Quantization processes like that in Figure
14.2.3 are characteristic of the filtered point-​
symmetry models developed by Jack Douthett
and Richard Plotkin. Douthett conceives of the
inner circle as a beacon sending seven beams of
light outward; the beams are then redirected
through a filter when they reach the outer circle.
Details and many examples may be found in
Douthett 2008, Plotkin and Douthett 2013, and
Plotkin 2019; the following exercises explore a
few possibilities.

Exercise 14.2.4
(a) Use a quantization process similar to that of
Figure 14.2.3 to construct a maximally even
five-​note set, and verify that the resulting
FIGURE 14.2.3 Construction of a maximally even set is a pentatonic collection.
set from a perfectly even set via quantization (b) Use a similar process to construct a
maximally even eight-​note set, and verify
names from Section 14.1 and with OPTC normal that the resulting set is an octatonic
forms: a diatonic scale, a melodic minor or acous-   collection.
tic scale, a harmonic minor scale, and a super–​
whole-​ tone collection. Intuition may suggest Exercise 14.2.5 Many variants of the
that the diatonic is the most even of the four, but quantization process are possible, yielding
it can be difficult to explain why, at least in a way essentially the same results. This exercise
that could apply generally to sets and scales of explores a few such variants.
other cardinalities.
As it turns out, the maximal evenness con- (a) Repeat the above construction of a
dition may be formalized in several equivalent maximally even seven-​note set, but round
ways. Perhaps the most intuitive formulation all fractions up instead of down, so that
is based on the observation that if we consider for example the number 15 7 will produce
subsets not just of pc but of continuous pitch-​ pitch class 2 rather than 1. Show that the
class space cpc, then a perfectly even division of resulting set is another diatonic collection.
the octave into seven equal parts (or any desired (b) Repeat the construction again, this time
number) is possible. Such a set is an interval rounding each fraction to the nearest
cycle generated by a fractional interval, in this integer, so that 15 7 will produce pitch class
case 12 7 , consisting of the numbers 0, 15 7 , 33 7 , 2 while 33 7 will produce 3. Again show that
51 7 , 66 7 , 84 7, and 102 7 , shown in the inner circle of the resulting set is a diatonic collection.21
Figure 14.2.3. Truncating each of these numbers (c) The relative alignment of the two circles in
to an integer, discarding the fractional remain- Figure 14.2.3 may be adjusted; that is, the
ders, yields the set {0, 1, 3, 5, 6, 8, 10} shown in perfectly even set in cpc need not contain
the outer circle. This set is a diatonic collection, 0. If the pitch classes in the inner circle are
specifically the D♭-​major collection dpc(−5)—​or rotated clockwise by anything less than
1 semitone, the ME set resulting from
more properly the enharmonic diatonic pitch-​ 7

class space edpc(7) as defined in Chapter 13. The the rounding-​down process in the figure
process may be regarded as a kind of quantiza- will not change, but for larger rotations
tion, whereby continuously varying quantities it will. Starting with a perfectly even set
(pitch classes in cpc) are restricted to a discrete whose first two elements are 1 7 and 16 7 , and
set of possible values (pitch classes in pc). The rounding fractions down as before, verify
perfectly even seven-​note division shown in the that the resulting ME set is again diatonic.
inner circle of the figure cannot be realized in the What if the perfectly even set contains the
discrete space of the outer circle, so each note number 1?
(except 0) must travel a short distance counter- (d) Suppose the beams from the inner circle of
clockwise until an available note is encountered. Figure 14.2.3 fire one at a time, starting with

600 • E x p l o r i n g M u sical S p aces

 the beam in the 0 position and proceeding
clockwise. The result will be an ordered set
of pitch classes on the outer circle—​a C
locrian scale. Now suppose the inner circle
is rotated clockwise in increments of 1 7
semitone, the beams always firing in the
same order. Show that the scales produced
are C phrygian, C aeolian, C dorian, …, in
the same order generated by the signature
  transformation s1.22

Exercise 14.2.6 This exercise outlines a

different procedure for constructing ME sets.
Start by constructing perfectly even sets of
cardinalities 5 and 7 in cpc, coloring the notes FIGURE 14.2.7 Generic and specific intervals in
of the five-​note set black and those of the an octatonic scale
seven-​note set white. Draw both sets in one
circle, rotating each of them to any desired
position, subject only to the constraint that scales in Section 13.4. Myhill’s property (cardi-
no black note and white note are in precisely nality equals variety for lines of cardinality 2)
the same location. Finally, redraw the picture, ensures that every generic interval in a diatonic
preserving the order of black and white notes scale comes in precisely two species; thirds, for
but spacing the twelve notes evenly around example, may be major or minor. Myhill’s prop-
the circle. Verify that the resulting sets are erty does not hold for octatonic scales—​ for
the familiar ME pentatonic and diatonic example, two octatonic scale-​steps always span
collections.
   23
a minor third, so generic interval 2 corresponds
only to specific interval 3, never to anything
Another important characterization of maxi- else—​but generic interval 3 is realized in two
mal evenness presents the first occasion on different specific intervals, as Figure 14.2.7 con-
which it becomes useful to broaden our defi- firms. Clough and Douthett’s insight is this: a
nition of the word generic, allowing any non- set is maximally even if and only if every generic
empty proper subset of the chromatic scale to interval is realized either in a single specific
be regarded as a generic scale. Following Clough interval or in two specific intervals which are
and Douthett, we write d for the cardinality of consecutive integers. The octatonic scale illus-
the generic scale and c for the cardinality of the trates both possibilities: if intg(x, y) =​2, then
specific scale (borrowing the initial letters of the ints(x, y) is always 3, while if intg(x, y) =​3, then
words diatonic and chromatic for this purpose). ints(x, y) may be either 4 or 5.
We are now interested in maximally even sets of
various generic cardinalities d, not just d =​7. For Exercise 14.2.8
the moment c remains 12, but soon we will con- (a) Write out all generic intervals and the
sider other specific cardinalities as well. corresponding specific intervals for a
For example, we may regard an octatonic diatonic scale, an octatonic scale, and a
collection as a generic scale of cardinality d =​ whole-​tone scale, and confirm that they
8 embedded within the usual chromatic scale. satisfy the intervallic condition for maximal
Figure 14.2.7 shows that within such a scale, we evenness described above.
may measure intervals either generically or spe- (b) Examine the melodic minor, harmonic
cifically; the two interval functions are denoted minor, and super–​whole-​tone scales from
intg and ints. Three steps of the scale separate Figure 14.2.2. For each of these scales, show
the notes C and E, so intg(C, E) =​3; the inter- that some generic interval comes in three
val intg(C♯, F♯) is likewise 3. The specific interval different specific sizes, in violation of the
function ints is the usual interval function in pc, required condition for maximal evenness.
so ints(C, E) =​4 while ints(C♯, F♯) =​5. (c) Consider the hexatonic collection {0, 1, 4,
The idea that a single generic interval may 5, 8, 9} as a generic scale. Show that for
correspond to two different specific intervals this scale, every generic interval comes in
was encountered in the context of diatonic either one or two specific sizes, but in the

Beyond the Diatonic • 601

 latter case the two specific sizes are not starting at bc 7, the result is ,
consecutive integers as required by the a rhythm associated with Steve Reich because
  maximal evenness condition. of its appearance in Clapping Music and several
other works.25
An application of maximally even sets with
chromatic cardinalities other than c =​12 arises —​—​—​
in the realm of rhythm, in the form of maximally
even beat-​class sets in various meters.24 Figure Our knowledge of the OPTIC spaces and distance
14.2.9 shows four examples. The 3 +​3 +​2 tresillo functions from Part Three offers another way to
rhythm, familiar in music of Latin America and conceive of maximal evenness, one that was not
many other cultures, is a maximally even beat-​ available to Clough and Douthett in 1991, and
class set with d =​3 and c =​8. The 2 +​2 +​2 +​3 which enables us to quantify just how “even” any
rhythm in the second example, a 4-​in-​9 ME set, set is. In three-​voice OP-​space, the central axis
occurs in examples as varied as the first move- consists of augmented triads—​ perfectly even
ment of Tchaikovsky’s Symphony No. 4, Ravel’s three-​part divisions of the octave. (Preliminary
“Alborada del gracioso,” Bartók’s Mikrokosmos forms of this property were noted in Chapter
No. 152, and, famously, Brubeck’s Blue Rondo à la 10; for the complete space see Figure 11.3.4.)
Turk. The rhythmic manifestation of the 7-​in-​12 Analogously, perfectly even diminished sev-
diatonic collection, common in Cuban and sub-​ enth chords lie on the central axis of four-​voice
Saharan African music, is sometimes known in OP-​space (Figures 11.5.3 and 11.5.9). Diatonic
the rhythm literature as the bembé rhythm or the scales and other seven-​note scales inhabit seven-​
standard pattern. The second strain of Joplin’s voice OP-​ space. Geometric relationships in
Maple Leaf Rag features a 5-​in-​16 ME rhythm in seven dimensions are difficult to visualize, to say
sixteenth notes, spanning two measures. the least, but this space too has a central axis, a
Rhythms such as these often repeat in the line consisting of perfectly even divisions of the
manner of an ostinato. They may be rotated so octave into seven parts, such as the set we con-
that the downbeat falls at various points in the structed in the inner circle of Figure 14.2.3 as a
pattern. The 3 +​2 +​2 +​2 rotation of the 4-​in-​ first step toward obtaining an ME set via quanti-
9 rhythm occurs occasionally in the first move- zation. That process—​particularly the rounding
ment of Brahms’s Quintet in G Major, Op. 111; method described in Exercise 14.2.5b—​is tan-
the Maple Leaf Rag pattern is rotated so that no tamount to taking a perfectly even point on the
articulation falls on the first downbeat. Maximal central axis of this continuous space and locating
evenness implies inversional symmetry, appar- the nearest lattice point. ME sets, that is, are the
ent in all the diagrams in Figure 14.2.9; conse- closest integer-​valued points to the central axis
quently the retrograde of any of these rhythms of OP-​space.
always matches some rotation, as the just-​noted Armed with our understanding of distance
Brahms example illustrates. If an attack is added functions from Chapter 12, we can calculate
to the diatonic rhythm at beat class 8 (the eight exactly how far a point is from the central axis
o’clock position), the resulting rhythm is no lon- and use the result as a measure of the set’s even-
ger ME but remains symmetrical because bc 8 lies ness.26 The following exercise explores this idea.
on the axis of symmetry; if this pattern is played For any set in OP-​space, the nearest point on

FIGURE 14.2.9 Four maximally even rhythms

602 • E x p l o r i n g M u sical S p aces

the central axis is always a set of the same sum representing those collections. As described
class as the original set. A perfectly even set of above, a lattice of seven-​ note scales may be
the desired sum class may be calculated; alterna- conceived as a discrete subset of seven-​voice
tively, we may transpose the original set so that OP-​space cpitch7/​~OP, but there are ways to con-
its sum class is 0 and compare it to a perfectly struct some such lattices without confronting
even set of sum class 0. The exercise adopts the the complexities of seven-​ dimensional geom-
latter approach. etry. Figure 14.2.11 shows relationships among
several diatonic, melodic minor, harmonic
Exercise 14.2.10 minor, and harmonic major scales.28 This appar-
(a) Explain why, whenever a seven-​note set ently three-​ dimensional picture is potentially
is transposed by perfect fifth (T7), its sum misleading as a depiction of the seven-​dimen-
class increases by 1. sional space,29 but it accurately captures single-​
(b) As a seven-​note pitch-​class set, which semitone voice-​ leading relationships between
diatonic collection is of sum class 0? scales that are adjacent in the figure.
Write this set as a 7-​tuple β of pitch-​class The lattice structure is determined by the
numbers, starting with 0.27 scales as pitch-​class sets; which note is the tonic
(c) Take the diatonic 7-​tuple β that you is of no consequence, but familiar scale names
identified in (b), and compare it with the are provided for convenience in the form of
perfectly even 7-​tuple α =​(0, 15 7 , 33 7 , spelled heptachord labels. The label (F♯)hmaj
51 7 , 66 7 , 84 7, 102 7 ), whose sum class is (+​5), for example, refers to the heptachord
also 0. This α is the point on the central hmaj(+​5), the F♯ harmonic major scale. Spelling
axis of OP-​space that lies closest to β. is irrelevant in OP-​ space, so the same point
Construct the displacement multiset as also represents G♭ harmonic major, hmaj(−7).
described in Section 12.4—​a list of seven The scales are therefore theoretically the enhar-
numbers Δx1, …, Δx7, the absolute values monic versions of the spelled heptachords, and
of the differences between corresponding so would more properly be labeled ehmaj(5)
numbers in the two tuples. and so on—​a technicality that the figure ignores.
(d) Verify that the voice-​leading distance Pitch classes are also shown for each scale. The
distVL(α, β) =​ Δx1 +​ ⋯ +​ Δx7 is equal to 15 7 . strings of pitch classes stand for OP-​classes, but
(e) Verify that the Euclidean distance in order to clarify voice-​ leading relationships
distEuc(α, β) =​[(Δx1)2 +​ ⋯ +​(Δx7)2]1  ⁄  2 is with neighboring scales they have not been
equal to 28 / 7 ≈ 0.76. rotated to give OP normal forms (the OP normal
(f) The distance values in (d) and (e) are form for hmaj(+​5) is [5, 6, 8, 10, 11, 1, 2]).
difficult to interpret out of context, but
they become meaningful when compared Exercise 14.2.12 Calculate the number of
with similar calculations for other sets. integer-​valued lattice points in seven-​voice OP-​
Consider a melodic minor scale, a harmonic space in the two different ways described below,
minor scale, and a super–​whole-​tone and verify that both methods give the same
collection as in Figure 14.2.2. For each set, result.
determine a transposition of sum class 0. (a) Each lattice point represents a seven-​note
Compare all of these sets with the perfectly subset of pc, so the number of lattice points
even set α from (c), determining the voice-​ must be the number of seven-​element
leading and Euclidean distance in each subsets in a twelve-​element set. Calculate
case. Your calculations should confirm that this number.30
none of these sets is as close to α as the (b) Each lattice point also represents one of
diatonic collection is. (Before undertaking the twelve transpositions of one of the 66
these calculations, you may wish to guess spelled heptachord types. Calculate this
which of the three collections appear most   number.
and least even to you, and see if the results
  confirm your intuition.) There are some resemblances between the
scale lattice in Figure 14.2.11 and the chain of
Tymoczko has extensively studied relation- hexatonic triad cubes at the center of three-​voice
ships among scales and other nearly even col- OP-​space (recall Figures 10.2.1 and 11.3.4) or
lections of various cardinalities, constructing an the analogous chain of octatonic seventh-​chord
assortment of “voice-​leading lattices” of points hypercubes in four-​ voice OP-​space. The scale

Beyond the Diatonic • 603

FIGURE 14.2.11 A partial lattice of seven-​note scales (after Tymoczko)

lattice, however, appears more complex, partly depending on whether this divisibility relation is
because there are no perfectly even seven-​note satisfied, and shows that the features just noted
sets comparable to the augmented triads and are characteristic of the two types: if d divides c,
diminished seventh chords that lie on the cen- then the most nearly even set types form a sim-
tral axes of their respective spaces. To include ple chain of d-​dimensional cubes linked by single
not only the diatonic scales but other nearby vertices, whereas if d and c are coprime, the lat-
scale types as well, a single chain of cubes, joined tice configuration is more complex, involving
at the corners as in the three-​and four-​voice cubes sharing faces or facets of some intermedi-
spaces, will not suffice; instead Figure 14.2.11 ate dimension.31
shows cubes attached by their faces in a more Another situation of interest in which d and c
complex pattern. The chain of fifth-​related dia- are coprime is the case in which d =​3 and c =​7.
tonic scales, shown in heavy outline, spirals That is, the sets under consideration are three-​
around the central axis of OP-​space, always lying note chords in mod-​7 space. This mod-​7 space is
closer to that axis than the points represent- what we know as generic pitch-​class space gpc—​
ing any of the other scale types, in a way that but it now plays the role of the specific scale, to
the three-​dimensional view cannot adequately be examined in relation to its subsets. Figure
capture. 14.2.13a shows the appropriate quantization
The reason why seven-​voice space lacks per- process, in which the perfectly even 3-​in-​7 set {0,
fectly even sets is simple: the cardinality of the 21 3 , 42 3 } yields the maximally even set {0, 2, 4}. A
sets under consideration (7) is not a divisor of mod-​7 set of OPTC or OPTIC normal form [0, 2,
the cardinality of the underlying specific scale 4] is a triad; triads and their four-​note comple-
(12). Tymoczko describes two families of lattices ments, seventh chords, are ME sets in generic

604 • E x p l o r i n g M u sical S p aces

FIGURE 14.2.13 A triad as (a) a maximally even set in generic space; (b) a second-​order maximally
even set in chromatic space

space. This observation reminds us that maximal (d) What chord progression results if the
evenness is a property of a set not in isolation inner circle is rotated continuously in the
but in relation to a larger scale: while it is com- clockwise direction?
mon to speak of a “maximally even set,” the term (e) What chord progression results if the inner
“maximally even subset” is more precise. and outer circles are fixed as shown in the
If the mod-​7 space in Figure 14.2.13a is inter- figure, and the middle circle is rotated
preted as a diatonic scale, then its 024 subsets continuously in the counterclockwise
include major, minor, and diminished triads—​ direction?33
none of which are maximally even in chromatic (f) Construct a 3-​in-​8-​in-​12 filtered point-​
space pc, where only augmented triads are ME. symmetry system in which all three circles
Major, minor, and diminished triads can arise in are initially aligned on pitch class 0. What
chromatic space, however, as second-​order maxi- chord progression results if the inner circle
mally even sets: maximally even subsets of maxi-   is rotated clockwise?34
mally even subsets. Figure 14.2.13b illustrates.
This filtered point-​symmetry model has three A particular alignment of the circles in a
circles, with three notes equally spaced around filtered point-​symmetry system, called a con-
the inner circle, seven in the middle circle, and figuration, may be specified by coordinates that
twelve in the outer circle. The 3-​in-​7 quantiza- record the rotations of each circle. Rotations of
tion yields a generic triad as in part (a) of the the circles generate transformation groups act-
figure; the circles in (b) are oriented so that the ing on the space of possible configurations. Each
7-​in-​12 quantization applied to this triad then configuration determines a chord, scale, or other
yields the C-​major triad {0, 4, 7}. pitch-​class set of the appropriate type via the
filtering process.35 Plotkin and Douthett show
Exercise 14.2.14 ways to trace paths in configuration spaces, sug-
(a) Numerical calculations are not shown gesting relationships with tonnetz diagrams and
in Figure 14.2.13b. Show the details of other techniques we have studied.
calculations yielding the second-​order ME We conclude this section by returning briefly
set {0, 4, 7}.32 to the observation that the cardinalities 3 and
(b) When the 7-​in-​12 quantization process 7 in Figure 14.2.13a are coprime. Our remarks
is applied to all seven notes of the mod-​7 above suggest that a lattice of nearly even
circle in the figure, which diatonic set is the trichords in generic space could be constructed,
result? and that such a lattice may share some of the
(c) Show that by rotating only the inner circle characteristics of the 7-​ in-​12 case shown in
in Figure 14.2.13b to different orientations, Figure 14.2.11. In fact, the 3-​in-​7 lattice looks
triads of all three diatonic species (major, very much like the 7-​in-​12, as the following exer-
minor, and diminished) can be produced. cise shows.

Beyond the Diatonic • 605

Exercise 14.2.15 Redraw or trace the lattice subscript. For example, the usual C-​major scale
in Figure 14.2.11, omitting the vertex labels. may be written {0, 2, 4, 5, 7, 9, 11}12. For many
Then label the vertices with three-​note chords purposes transpositionally equivalent generic
in generic space gpc, using mod-​7 OP normal scales may be considered the same; a T-​class
forms. The points on the heavy line, originally of scales may be labeled with its mod-​c OPTC
fifth-​related diatonic scales, are now third-​ normal form, the mod-​c equivalent of a d-​note
related generic triads; place ACE at the bottom, chord species. For the diatonic scale in the
in the position originally occupied by the canonical system the OPTC normal form is [0, 1,
B♭-​major scale, and FAC above it, in place of the 3, 5, 6, 8, 10](OPTC
12 )
, which we will write in the sim-
F-​major scale. Other nearby points should be pler form [0, 1, 3, 5, 6, 8, 10]12 as long as confu-
labeled with other types of trichords, less even sion with other normal forms is unlikely.
than triads; for example, the points originally The interval between any two notes of C may
labeled mmin(−1) and hmin(−1) are now DEA be measured using the specific interval function
and BCE respectively. Indicate the mod-​7 sum ints: C × C → ℤc. Because D is a subset of C, inter-
classes. The original figure is large enough so vals between notes of D may be measured with
that two of the trichords at the bottom will be ints in this way, but they may also be measured
duplicated
   at the top.36 with the generic interval function intg: D × D →
ℤd, which counts steps from note to note within
14.3 BEYOND THE D regardless of how far apart those notes may be
CHROMATIC: OTHER in C. Corresponding to these two interval func-
tions are transposition operators: any note of C
SPECIFIC CARDINALITIES may be transposed by the specific transposition
Our study of scales has focused on the most operator Ti: C → C, where i is an integer mod c,
familiar examples, especially the twelve-​ note and notes of D may also be transposed by the
chromatic and the seven-​note diatonic. In the generic transposition operator tj: D → D, where j
preceding sections we have considered some is an integer mod d.
non-​diatonic pitch-​class sets as possible “scales,” Many of the concepts defined in Chapter 13
but still in the context of the twelve-​note space and previously in this chapter extend in straight-
pc. In principle there are no theoretical barriers forward ways to this more general setting. D is
to constructing scale systems with more than a generated scale if its notes may be arranged in
twelve notes, though there are practical hurdles a generated order starting with some note x and
to overcome in composing, notating, and per- continuing with successive transpositions Tg(x),
forming music in such scales. In this final sec- T2g(x), …, T(d−1)g(x) (mod c) through some gener-
tion of the chapter we explore some approaches ating interval g (defined as a specific interval). D
to generalizing what we know about diatonic and is well-​formed if it is generated and every instance
chromatic scales to scales of other cardinalities. of the generating interval spans the same generic
As noted at the outset of the chapter, this discus- interval, or equivalently if the mapping from
sion will be far from exhaustive. scalar order to generated order is an automor-
A specific scale C of cardinality c may be iden- phism of ℤd. Specific intervals determine interval
tified with ℤc, the integers mod c, and inter- classes (actually distances, the mod-​c absolute
preted as a c-​note pitch-​class space analogous to values of the specific interval function), which
pc. As usual we will not devote much attention range from 0 to ⎣c/​2⎦ (the greatest integer less
to the question of tuning; for our purposes we than or equal to c/​2), which is equal to c/​2 if c
may assume c-​note equal temperament, a divi- is even or to (c − 1)/​2 if c is odd. D is a deep scale
sion of the octave into c equal intervals of fre- if every interval class occurs a different number
quency ratio 21 ⁄c. We sometimes refer to C as the of times between pairs of notes in D. A chord
chromatic universe. Any nonempty proper subset genus is a mod-​d OPTC equivalence class of sub-
D of C, of cardinality d < c, may be regarded as sets of D (where T equivalence refers to generic
a generic scale. A generic scale D, together with transposition tj); a chord species is a mod-​c OPTC
the specific scale C of which it is a subset, will be equivalence class of subsets of C (where T equiv-
called a scale system; the scale system consisting alence refers to specific transposition Ti). A line
of the usual 7-​note diatonic and 12-​note chro- genus is a mod-​d OT-​class of tuples in D; a line
matic scales is the canonical system. The elements species is a mod-​c OT-​class of tuples in C. D sat-
of D may be listed in ordinary set notation; it is isfies cardinality equals variety for chords if every
useful to indicate the specific cardinality c with a chord genus of cardinality k is realized in exactly

606 • E x p l o r i n g M u sical S p aces

k species for every k with 1 ≤ k ≤ d − 1; D satis- Exercise 14.3.1 Seven generic scales in the
fies cardinality equals variety for lines if every line usual mod-​12 chromatic scale are listed below.
genus of cardinality k is realized in exactly k spe- Determine which, if any, of the following five
cies for every k with 1 ≤ k ≤ d. D has Myhill’s prop- properties each scale satisfies: (1) generated,
erty if every generic interval is realized in exactly (2) well-​formed, (3) deep, (4) C =​V for lines, (5)
two species; this is the special case of C =​V for maximally even.38
lines of cardinality k =​2. D is maximally even if it
may be obtained by quantizing a perfectly even (a) The pentatonic scale [0, 2, 4, 7, 9]12
set of cardinality d {x, x +​ c/​d, x +​2c/​d, …, x +​ (b) The whole-​tone scale [0, 2, 4, 6, 8, 10]12
(d − 1)c/​d}c in mod-​c space, or equivalently if (c) The hexatonic scale [0, 1, 4, 5, 8, 9]12
every generic interval is realized either in a sin- (d) The diatonic hexachord [0, 2, 4, 5, 7, 9]12
gle specific interval or in two specific intervals (e) The harmonic minor scale [0, 1, 3, 4, 6, 8, 9]12
which are consecutive integers. For any given (f) The chromatic heptachord [0, 1, 2, 3, 4, 5, 6]12
cardinalities c and d, a maximally even d-​in-​c set (g)  The octatonic scale [0, 1, 3, 4, 6, 7, 9, 10]12
may always be constructed by quantization, and
it is essentially unique in the sense that all such Exercise 14.3.2 Consider the same properties
sets are related by specific transposition. as in the previous exercise for the following
Relationships among these properties have scales in chromatic universes with values of c
been studied extensively. Clough, Engebretsen, other than 12.39
and Kochavi 1999 offer many details, a few of
which we shall summarize quickly; Exercises (a) [0, 1, 2, 4]6
14.3.1 and 14.3.2 below offer a number of exam- (b) [0, 1, 3, 4]7
ples. Notably, all of the properties enumerated (c) [0, 1, 3, 5, 6, 8]10
above imply inversional symmetry: an asym- (d) [0, 1, 4, 7, 8, 11, 14]17
metric scale D cannot be generated, deep, well-​ (e) [0, 2, 4, 6, 8, 11, 13, 15, 17]20
formed, or ME, and cannot satisfy any form of (f) [0, 1, 3, 5, 7, 9, 10, 12, 14, 16, 18]20
C =​V. While Myhill’s property is ostensibly a (g)  [0, 2, 6, 10, 12, 16, 20]24
special case of C =​V for lines, the special case
turns out to imply the general: if D has Myhill’s The next exercise provides a simple reason
property, then C =​V for lines is always satisfied. why some of the scales in the two preceding exer-
The full generality of C =​V for chords does not cises cannot be deep: they are not big enough.
necessarily follow, however; the property of C =​
V for chords is generally more difficult to charac- Exercise 14.3.3
terize than C =​V for lines.37 Every scale satisfy- (a) If n is a positive integer, the sum 1 +​2 +​⋯
ing C =​V for either lines or chords is well-​formed +​ (n − 1) is equal to n(n − 1)/​2, which is also
(and therefore generated). Clough and Myerson the number of 2-​element subsets of a set of
(1985, 263–​ 64) show that Myhill’s property cardinality n. Use these properties to show
implies the existence of a “generalized circle of that for a deep scale to exist, the specific
fifths”: in the generated order of the scale D, the and generic cardinalities c and d must
wraparound interval from the last note to the satisfy d ≥ ⎣c/​2⎦.40
first is generically the same as the generating (b) Call a set internally deep if all the nonzero
interval, but its specific size is different (gen- entries in its interval-​class vector are
eralizing the situation in the canonical system, different—​that is, if every interval that
where the generating interval is a perfect fifth actually occurs in the set occurs a unique
and the wraparound interval is a diminished number of times. Of the non-​deep scales
fifth). While C =​V implies well-​formedness, the examined in Exercises 14.3.1 and 14.3.2,
converse is not true: perfectly even scales such four are internally deep. Which scales
as a whole-​tone scale are well-​formed but do not   are these?
satisfy C =​V for lines or chords. The intervallic
condition for maximal evenness suggests a rela- Evidently the distinctive nature of the usual
tionship with Myhill’s property and therefore 7-​in-​12 diatonic scale is not encapsulated by a
with C =​V, but in general neither maximal even- single easily defined property, but rather is char-
ness nor C =​V implies the other. If c and d are acterized by a large assortment of attributes
coprime, however, a maximally even scale satis- like those discussed above. Exercises 14.3.1 and
fies Myhill’s property and C =​V for lines. 14.3.2 suggest, accurately enough, that while

Beyond the Diatonic • 607

each of these properties is satisfied by a variety phenomena. We are accustomed to being able
of scales, there are few scales that satisfy all the to divide an octave equally in several different
properties. Each property, therefore, provides ways, including in half, which requires that c be
its own generalization of the concept of diato- even (property (b)). Property (c), the divisibility
nicity: the aim of generalizing the diatonic scale of c by 3, will be discussed further below in rela-
to other cardinalities may be approached in a tion to (h). Both (b) and (c) are properties of the
variety of ways, depending on which features are chromatic scale only, not involving the diatonic.
considered most desirable. Properties (d) and (e), the prime cardinality of
In this discussion of some rather technical the diatonic scale and the coprime nature of c
aspects of scales, we should not lose sight of and d, have been noted many times in this and
some simple arithmetical properties of the num- the preceding chapter. (If d is prime, then c and
bers 7 and 12 that come into play in the canoni- d are automatically coprime unless c is divisible
cal system in various ways. Eight such properties by d.) The alternative formulation of (e)—​c and
are listed below. Here c and d denote, as usual, g are coprime—​ensures that the interval g that
the cardinalities of the specific and generic scales generates D also generates the entire specific
C and D. We assume that D is a generated scale, scale C, so that C may be arranged in a “circle of
and we write g for the specific size of the gener- g’s”—​the circle of fifths in the familiar setting—​
ating interval. It is a simple matter to verify that of which D forms a contiguous segment.
the properties in this list hold in the familiar set- Of the three formulations of property (f), the
ting with c =​12, d =​7, and g =​7; below we will greatest musical importance perhaps attaches
consider the implications of these properties for to the statement dg =​1 (mod c). Consider D as
other values of c, d, and g. a contiguous segment of the circle of g’s as just
described, referring back if necessary to the pic-
(a) d =​ g. ture of dpc(−3) as a subset of fifth at the lower
(b) c is an even number, and in particular a right corner of Figure 13.2.1. Adjacent notes in
composite number (not prime). this circle are related by specific interval g. If D
(c) c is divisible by 3. is rotated one position clockwise, the resulting
(d) d is an odd prime number. set is a Tg transposition of the scale D; in Figure
(e) c and d are coprime. In light of (a), 13.2.1, dpc(−3) becomes dpc(−2). These two
therefore, c and g are coprime. transpositions of the scale have d − 1 notes in
(f) d 2 =​1 (mod c). In light of (a), therefore, common. The new note (A) is d places clockwise
g2 =​1 (mod c) and dg =​1 (mod c). from the note it replaces (A♭), so the mod-​c spe-
(g) d =​ c/​2 +​ 1. cific interval from the old note to the new (from
(h) d is the sum of two consecutive integers A♭ to A) is dg. To say that this interval is 1 ensures
whose product is c. that the new scale is related to the old by the
smoothest possible voice leading: only one note
Property (a) reminds us that we have encoun- changes, and it changes by the smallest possible
tered the number 7 in two different roles, as the interval—​as is the case, of course, in moving
cardinality of the generic scale and also as the from one diatonic scale to a closely related one.
specific size of the generating interval, the per- Transpositions by the complementary genera-
fect fifth. There is no obvious reason why these tor g′ =​ c − g are just as smooth as transpositions
two numbers should be the same, and it is diffi- by g, because g′ =​−g (mod c) and therefore dg′
cult to make a case for this property as an impor- =​−1 (mod c). This property ensures that closely
tant determinant of diatonic structure. If interval related diatonic scales are adjacent in the lattice
g generates a scale, so does the complementary of Figure 14.2.11; it allows diatonic scales to par-
interval g′ =​ c − g (in the canonical case g′ =​12 − 7 ticipate in “maximally smooth cycles” as defined
=​ 5); d may be equal to either g or g′ but not both. in Cohn 1996, 15–​ 16; and its importance is
Also, the scale [0, 2, 6, 10, 12, 16, 20]24 from noted also in Balzano 1980, 70–​71, where it is
Exercise 14.3.2g may be interpreted as the usual dubbed the “F–​F♯ property” and described as
diatonic scale situated within a specific scale that essential to the behavior of key signatures.
has been enlarged to contain quarter-​tones (so When a circle of g’s exists, the multiplication
that the usual semitone is specific interval 2); for operator Mg (M7 in the canonical system) con-
this scale d remains 7 but g is 14. verts the chromatic (scalar) arrangement of C
Some of the other listed properties, how- to the circle of g’s. The statement g2 =​1 (mod c),
ever, are essential to many familiar musical another formulation of property (f), guarantees

608 • E x p l o r i n g M u sical S p aces

that the same operator Mg maps also in the other d are implied. Cohn’s condition (c), meanwhile,
direction, from the circle of g’s to the chromatic disallows c =​20 entirely. The case c =​24 is consis-
circle, as discussed previously in conjunction tent with (c) and (g), but not (h). If c =​24, how-
with M7 in Figure 13.2.1. ever, then (g) implies that d =​13, so this generic
Properties (g) and (h) describe particular scale is not the usual diatonic—​even though, as
arithmetical relationships between the cardi- we have suggested, a 24-​note quarter-​tone scale
nalities c and d. Property (g), d =​ c/​2 +​1, requires contains the usual 12-​note chromatic and there-
that c be even. If we also assume that d is odd, fore also the usual 7-​note diatonic scale among
then (g) implies that c is a multiple of 4, that c its subsets.43
and d are coprime, and that d 2 =​1 (mod c) (see Moreover, we have paid no heed to tuning,
Exercise 14.3.4a below). Agmon 1989 defines a a primary consideration for many who study
family of “diatonic systems” which turn out to microtonal scales. Twelve-​ note equal tem-
be precisely the maximally even d-​in-​c sets for perament is often described as an acoustical
which d =​ c/​2 +​1 and d is odd; the next possibili- compromise, appealing because it offers close
ties for such systems after the canonical 7-​in-​12 approximations to the “pure” intervals of the
are 9-​in-​16 and 11-​in-​20 (the scale in Exercise perfect fifth and major third as determined by
14.3.2f).41 Clough, Engebretsen, and Kochavi small integer ratios from the harmonic series.44
1999 show that every scale with this combina- Neither c =​18 nor c =​20 approximates these
tion of properties is deep and well-​formed, and intervals well, but c =​19 does. A 19-​note uni-
satisfies C =​V for lines. verse, though it violates many of the arithmeti-
Balzano 1980 calls attention to property (h), cal properties listed above, is of great interest,
which in the canonical case is the observation and we will study it briefly at the end of this
that 7 =​3 +​4 while 12 =​3 ∙ 4. In the general section.
setting this property says that for some inte-
ger k, d =​2k +​1 and c =​ k(k +​1). We know from Exercise 14.3.4
Section 6.4 that the factorization 12 =​3 ∙ 4 is (a) Assume that d =​ c/​2 +​1 (property (g)),
important in the group structure underlying the and assume that d is odd. Show that c is a
neo-​Riemannian tonnetz, and indeed Balzano’s multiple of 4, that c and d are coprime, and
motivation is to generalize this aspect of the that d 2 =​1 (mod c).45
tonnetz. The assumption d =​ g is also implicit in (b) Show that the only numbers c and d
Balzano’s argument.42 After 7 and 12, the next satisfying both property (g) (d =​ c/​2 +​ 1)
numbers satisfying Balzano’s condition are d =​ and property (h) (d =​2k +​1 and c =​ k(k +​1))
4 +​5 =​9 and c =​4 ∙ 5 =​20. This 9-​in-​20 system   are c =​12 and d =​7.46
is the one in Exercise 14.3.2e above; for a corre-
sponding tonnetz see Balzano 1980, 76. Zweifel Even if a generic scale shares many structural
1996 argues that Balzano’s 9-​in-​20 scale is more properties with the usual diatonic, it is reason-
closely analogous to the pentatonic scale than to able to wonder how closely musical behavior
the diatonic, and proposes its complement, the can resemble familiar diatonic behavior if the
11-​in-​20 scale from 14.3.2f, as a better analog of generic cardinality d is anything other than 7.
the diatonic. Either g =​9 or g =​11 generates both This is a difficult question to answer: musical
of these scales, but d =​9 in one case while d =​11 examples composed in such scales are scarce,
in the other. partly because the complexities of notation are
Another approach to generalizing the tonnetz considerable. A simpler situation of interest,
is that of Cohn 1997. With the goal of recaptur- however, is the case in which d remains 7 but
ing the parsimonious voice-​leading behavior of c is something other than 12. In this case the
the familiar transformations P, R, and L, Cohn notes of the generic scale may be given the usual
arrives at the stipulation that c be divisible by 3 seven letter names and even written in tradi-
(property (c)); he constructs tonnetz variants for tional staff notation with some modifications
c =​18 and c =​24. in the treatment of accidentals, and a composer
These properties, clearly, do not always accord may choose to mimic familiar diatonic patterns
with each other. In fact, no pairs of integers c within this scale even if the tuning is unfamil-
and d other than the familiar 12 and 7 satisfy iar and the specific intervals “wrong.” One com-
both (g) and (h) (see Exercise 14.3.4b below). poser to have pursued this approach is Easley
As noted above, c =​20 is consistent with either Blackwood, whose Microtonal Etudes will be dis-
(g) or (h), but two different generic cardinalities cussed below.

Beyond the Diatonic • 609

FIGURE 14.3.5 An enharmonic system

Enharmonic systems, introduced in Hook desired, so that c may be much larger than 12.
2007c, model scale structures of exactly this It is advantageous to list the seven step sizes in
kind. As shown in Figure 14.3.5, an enharmonic circle-​of-​fifths order. The 7-​tuple Θ =​(θF, θC, θG,
system consists of seven “white notes,” identi- θD, θA, θE, θB) is called the enharmonic equivalence
fied by the usual letter names and separated by vector, or EE vector, of the system, so the EE vec-
variable numbers of “black notes.” The specific tor of the canonical system is (2, 2, 2, 2, 2, 1, 1).
interval from C to D is denoted θC; the θC − 1 When the numbers are interpreted as accidentals
black notes between C and D are C♯, C♯ ♯, …, C(θC−1)♯ assigned to the seven generic pitch classes in the
(the last of these is to be read as a C with θC − same circle-​of-​fifths order, this vector specifies
1 sharps). It is understood that CθC ♯, a C with θC the spelled heptachord {F𝄪, C𝄪, G𝄪, D𝄪, A𝄪, E♯, B♯},
sharps, is the same note as D; the system thus which we may recognize as dpc(+​12), the B♯-​
provides a new definition of enharmonic equiv- major scale. More generally, an EE vector Θ may
alence. The symbol ♭ may be used for chromatic be regarded as a generalized spelled heptachord
inflections in the descending direction, so that describing a scale enharmonically equivalent to
D♭ is the same as C(θC−1)♯. The numbers θC, θD, …, the white-​key scale in a sharper spelling.
θB are the specific interval sizes associated with An EE vector completely determines the
the generic steps C–​D, D–​E, …, B–​C; the relative enharmonic equivalence relation, and indeed
sizes of these numbers determine the intervallic the entire structure of the enharmonic system;
structure of the white-​note scale. Enharmonic the enharmonic system defined by the vector Θ
equivalence may be understood as an equivalence is denoted enhΘ. The sum of the components in
relation on signed letter space sl from Figure Θ, θF +​ θC +​ ⋯ +​ θB, is called the accidental index
1.2.11a, generalizing Exercise 1.2.12, which of the vector (a name consistent with our use of
investigated the usual enharmonic equivalence the same term for spc sets), and is equal to c, the
relation in this space. The step size θC, for exam- total number of notes in the system. The specific
ple, determines how the notes in the C column of scale consisting of all the notes of enhΘ is a mod-​
Figure 1.2.11a are to be identified with the notes c pitch-​class space in which a specific interval
in the D column. function and specific transposition operators are
The canonical system, as an enharmonic sys- available; the system also contains in its white
tem consisting of the usual 7-​in-​12 scales, will be notes a copy of generic pitch-​class space gpc
denoted enh12. In this system θE =​ θB =​1, as there with its own mod-​7 interval structure.
are no black notes between E and F or between The canonical EE vector (2, 2, 2, 2, 2, 1, 1)
B and C, while the other five θs are all equal to is a standard vector, meaning that it is con-
2. The general definition of an enharmonic sys- stant through the first several components,
tem, however, allows the θs to be as large as then steps down by 1. This vector is standard of

610 • E x p l o r i n g M u sical S p aces

step position 5; the step position is the number Nonstandard key signatures may be employed as
of equal components before the step. A vector well, even in the canonical system; for instance
whose components are all the same is stan- the nonstandard vector (0, 1, 0, 0, 0, 0, −1) of
dard of step position 0. The step position 5 of accidental index 0 denotes a key signature with
the canonical EE vector derives from the five C♯ and B♭, corresponding to a D harmonic minor
whole steps and two half steps in the diatonic scale. Every signature vector of accidental index
scale. Slightly more general than standard vec- n represents some spelled heptachord whose
tors are regular vectors: a vector is regular of accidental index is n (but not necessarily a proper
step position k if its first k components are the spelled heptachord; doublings and crossings are
same and the remaining components are the possible).
same (but the difference between the two values
need not be 1). The properties of the diatonic Exercise 14.3.7
scale explored earlier in this chapter may be (a) For each of the five accidental indices n =​
checked for the white-​note scale in enhΘ; it can 5, −2, −7, 12, and −12, write the standard
be shown that standard or regular EE vectors of vector Φn and identify the key signature
certain step positions guarantee certain proper- and the traditional major key to which
ties of the scale. Standard vectors of step posi- it corresponds. For any number n, what
tion 5 are particularly well-​behaved: if Θ is any can you say about the step position of the
such vector, then the white-​note scale in enhΘ vector Φn? The key signatures of accidental
is generated, well-​formed, and maximally even, indices 12 and −12 are not exactly
and satisfies Myhill’s property and therefore C =​ “traditional,” but what are they, and what
V for lines (Hook 2007c, 109).47 are they enharmonically equivalent to in
the canonical system? Where have we seen
Exercise 14.3.6 the vector Φ12 before?
(a) Write a regular EE vector of step position (b) Signature vectors corresponding to
5 for a 17-​note enharmonic system whose harmonic minor scales are nonstandard,
white-​note subset is some transposition but as with standard vectors there is exactly
of the scale [0, 1, 4, 7, 8, 11, 14]17 from one such vector of every possible accidental
Exercise 14.3.2d. index. Write the harmonic minor signature
(b) Write a similar EE vector for a 22-​note vectors of accidental index −1, 0, 1, and 7,
enharmonic system. and identify the harmonic minor scales to
(c) What EE vector describes a diatonic scale   which they correspond.49
in a quarter-​tone (24-​note) enharmonic
system? Is this vector standard? Is it Musical behavior in an enharmonic system
  regular? Is the scale maximally even?48 with a key signature depends on the interaction
of two vectors, the EE vector defining the system
Because enharmonic systems admit the use and the signature vector that determines the
of standard pitch notation with sharps and flats, key signature. The key signatures of accidental
they also admit the use of key signatures. A key index 5 and −7 in the above exercise determine
signature may be written as a 7-​place signature keys that are enharmonically equivalent in the
vector (λF, λC, λG, λD, λA, λE, λB) resembling an EE canonical system, but in some other systems the
vector. Each λ is an integer indicating the acci- keys with those signatures would not be equiva-
dental applied to the corresponding note, with lent at all. The problem of determining when
sharps denoted 1 and flats −1. The signature vec- two signature vectors define enharmonically
tor for the traditional three-​sharp key signature equivalent scales (possibly even if the vectors are
is (1, 1, 1, 0, 0, 0, 0), while the vector for a five-​ nonstandard) is one of many questions that can
flat signature is (0, 0, −1, −1, −1, −1, −1). The first be addressed, either in a particular enharmonic
of these vectors is of accidental index 3 and is system or more generally. Another question of
a standard vector of step position 3, as defined interest is whether familiar diatonic interval
above; the second is of accidental index −5 and names may be used in consistent ways. In some
standard of step position 2. All traditional dia- systems, for example, the intervals C–​E and E♭–​G
tonic key signatures have standard signature may be of different specific sizes, so it is not clear
vectors; in fact, there is a unique standard vec- whether labels such as “major third” are mean-
tor Φn of any given accidental index n, and it cor- ingful. Once again, standard EE vectors of step
responds to the usual key signature for dpc(n). position 5 ensure the most familiar results: if Θ

Beyond the Diatonic • 611

is such a vector, then diatonic interval names in
enhΘ are coherent (Hook 2007c, 114).
Easley Blackwood’s Twelve Microtonal Etudes
for electronic synthesizer, Op. 28 (1980),
explore all equal-​ tempered tunings from c =​
13 to c =​24, using traditional notation some-
times supplemented with additional accidentals
besides sharps and flats. That Blackwood sought
when possible to recapture familiar features of
diatonic scales and common triads is evident in
the notation, in the musical structure, in his own
descriptions of the pieces, and in the title of his
subsequent book The Structure of Recognizable
Diatonic Tunings (1985). He writes, for example,
that 15-​note tuning “contains triads that are
sufficiently in tune to serve as the final har-
mony in cadences. The major scale, however, is
so strange that even the most common diatonic
progressions are disturbing unless precautions
are taken” (Blackwood 1994). This composi-
tional approach is well suited to modeling by
enharmonic systems, and eleven of Blackwood’s
twelve Etudes are notated in ways that are essen-
tially consistent with enharmonic systems.50
After Φ12 =​(2, 2, 2, 2, 2, 1, 1), the next stan-
dard vector of step position 5 is Φ19 =​(3, 3, 3,
3, 3, 2, 2), suggesting that the enharmonic sys-
tem enhΦ19, which we abbreviate to enh19, may
have many familiar properties. In fact, a scale
with 19 notes per octave has enjoyed occasional
advocacy, especially in the Renaissance and
in the early twentieth century, because of the
familiar sound of its diatonic scale and its simi-
larities with—​some might say improvements
upon—​the 12-​note system enh12. The EE vec-
tor Φ19 specifies three specific steps in every
major second in the diatonic scale (for example,
C–​C♯–​D♭–​D), and two in each of the two minor FIGURE 14.3.8 (a) A keyboard for a 19-​note
seconds (E–​E♯–​F, with E♯ enharmonically equiva- scale; (b) the 19-​note enharmonic system enh19;
lent to F♭). Figure 14.3.8a shows a keyboard lay- (c) the 19-​note circle of fifths
out for this scale—​a not entirely hypothetical
construction, as a number of 19-​note keyboards
of similar design were constructed in the six- for example, that every traditionally notated
teenth century. Part (b) of the figure shows the perfect fifth such as C–​G, B–​F♯, or A♭–​E♭ spans
scalar arrangement of the enharmonic system, eleven chromatic steps, while major and minor
and (c) shows the generated order, the 19-​note thirds always span six and five chromatic steps,
circle of fifths.51 respectively. Major and minor scales may be
The standard EE vector of step position 5, as notated traditionally, with the usual key signa-
we have seen, ensures that many of the usual tures, and all such scales are specific transposi-
properties of the canonical system are satisfied tions of the white-​note scale, so we may apply
by enh19. In particular, the 7-​in-​19 white-​note the name diatonic to any of them. There are more
scale is generated, well-​formed, and maximally specific interval classes in enh19 than in enh12,
even, and it satisfies Myhill’s property and there- and according to Exercise 14.3.3a, a 7-​in-​19 scale
fore C =​V. Diatonic interval names are consis- does not have enough notes to be a deep scale;
tent throughout the scale: the reader may verify, the following exercise shows, however, that the

612 • E x p l o r i n g M u sical S p aces

diatonic scale in enh19 possesses a property should make us think twice before relying on
nearly as strong. aspects of the 12-​note chromatic scale in ana-
lyzing diatonic behavior: any credible analysis
Exercise 14.3.9 What is the interval-​class of diatonic music, it seems, should apply equally
vector of the white-​note scale in enh19? How well to the diatonic scale in enh19.
does it resemble the vector of the usual 7-​in-​ There are other ways, however, in which enh19
12 diatonic collection? Verify that the 7-​in-​19 differs markedly from the canonical system.
diatonic scale is internally deep as defined in Earlier in this section we identified eight arith-
Exercise
   14.3.3b. metical conditions (a)–​(h) satisfied by the num-
bers c =​12, d =​7, and g =​7 in the 7-​in-​12 scale.
The acoustical sizes of common intervals in The corresponding numbers in enh19, c =​19, d =​
19-​note tuning may be considered either better 7, and g =​11, violate six of these conditions as we
or worse than those in 12-​note tuning, depend- presented them: (a), (b), (c), (f), (g), and (h). An
ing on one’s point of view: the following exer- especially significant difference is that because
cise shows that measured against interval ratios the cardinality c =​19 is a prime number, equal
from the harmonic series, the perfect fifth is bet- divisions of the octave in enh19 are impossible.
ter in tune in enh12, while both major and minor Every nonzero interval is coprime to 19 and
thirds are better in enh19. therefore generates a complete cycle through the
19 pitch classes; in this respect the 19-​note scale
Exercise 14.3.10 resembles a 7-​note generic scale more closely
(a) To calculate the tuning error of the perfect than a 12-​note chromatic scale. The chromatic
fifth in 12-​note equal temperament, first circle (Figure 14.3.8b, generated by specific
express the pure interval as a fraction of an interval 1 or 18) and the circle of fifths (14.3.8c,
octave by taking the base-​2 logarithm of generated by 8 or 11) are no longer the only com-
the 3 2 frequency ratio, and then convert to plete interval cycles; there are seven other such
cents by multiplying by 1200; you should cycles, all essentially different. A diminished sev-
find that a pure perfect fifth is about enth chord such as B–​D–​F–​A♭ divides the octave
702¢. A perfect fifth in 12-​note tuning is of enh19 into specific intervals in the pattern
exactly 7 12 of an octave, or 700¢, which is 5–​5–​5–​4: three minor thirds and one smaller
therefore about 2¢ too small. Perform the augmented second. A diminished seventh chord
corresponding error calculation for the therefore no longer lends itself to enharmonic
perfect fifth in an equal-​tempered 19-​note reinterpretation; instead of just four different
scale (where a perfect fifth is 11 specific diminished seventh chords, there are nineteen.
steps, or 11 19 of an octave). Similar considerations apply to augmented tri-
(b) In a similar way, calculate the tuning errors ads (6–​6–​7) and whole-​tone scales (3–​3–​3–​3–​
of the major third and minor third in enh12 3–​4). Each of these sets is maximally even but
  and enh19.52 no longer perfectly even. Transpositional sym-
metry, in fact, is impossible, by the same reason-
It is likely because of the familiar diatonic ing as in Exercise 13.1.23d for the mod-​7 case: in
properties of enh19 that Blackwood places his enh19, with the trivial exceptions of the empty
19-​note Etude last in the set of twelve. This set and the 19-​note aggregate, every pc set has
Etude begins with a section entirely diatonic in nineteen different transpositions.
C major, featuring familiar functional progres- The diatonic cardinality d =​7 and genera-
sions and diatonic sequences. Although differ- tor g =​11 do not match in enh19, as property
ences from 12-​note tuning are audible, most (a) would require, but both remain coprime to
listeners are struck by the perception that these c =​19, satisfying property (e). Regarding (f),
common diatonic processes “work” in the 19-​ while neither d 2 =​1 nor g2 =​1 (mod c) holds,
note scale just as they do in the 12-​note scale. the important form dg =​1 (mod c) remains true
One could perform in enh19, with musically (7 ∙ 11 =​1 (mod 19)), ensuring, as noted above,
satisfying results, a piece such as the variations that closely related diatonic scales are related by
from Mozart’s A-​Major Sonata, K. 331, which smooth voice leading.
involves only brief decorative chromaticism.
A Schenkerian analysis of such a piece, pre- Exercise 14.3.11
pared with the 12-​note system in mind, would (a) Construct a circle of major thirds and a
be equally valid in enh19. These observations circle of minor thirds in enh19.

Beyond the Diatonic • 613

(b) Arrange the nineteen diminished seventh nothing “hexatonic” about this larger cycle;
chords in a maximally smooth cycle. Each indeed, the word has no clear meaning in enh19,
chord should be related to the next by voice-​ where no set exhibits the distinctive symmetry
leading distance 1 in enh19 (motion of one of the usual hexatonic collection. In the 19-​note
voice through a single step of the specific version of three-​voice OP-​space, the chain of
  scale). cubes containing major and minor triads is not
like the chain of hexatonic triad cubes, joined at
A short segment of the circle of fifths in single vertices, that lies along the central axis of
Figure 14.3.8c or of one of the circles of thirds the 12-​note space (Figures 10.2.1 and 11.3.4);
in Exercise 14.3.11 resembles the mod-​12 case, instead, it resembles the scale lattice of Figure
but the redefinition of enharmonic equivalence 14.2.11, in which cubes share faces. Triads such
prevents these circles from closing in the accus- as C major and E major, related by major-​third
tomed ways. Something similar is true of the transposition, do not share a sum class in enh19
19-​note tonnetz in Figure 14.3.12, which may as they do in enh12. Because enh19 differentiates
be compared with the mod-​ 12 tonnetz from a chromatic semitone (specific interval 1) from a
Figure 1.4.2. Pitch classes remain organized diatonic semitone (specific interval 2), the trans-
by minor thirds horizontally and major thirds formations P and L are no longer equally smooth,
vertically, but the familiar 3-​by-​4 tiles (one of and the PL-​chain is not maximally smooth: in P
which is shown in light outline) do not fit the the moving voice traverses one specific step, but
19-​note version of enharmonic equivalence. One in L it traverses two.53 In an imagined perfor-
possible tiling consists of the parallelograms mance in 19-​note tuning of Liszt’s Consolation
shown (no rectangular tile is possible in a prime No. 3 (analyzed in Exercise 4.2.13b), the L and
cardinality). P transformations might retain their musi-
The enh19 tonnetz may be triangulated in the cal effect, but the piece would ascend by major
manner of Figure 1.4.5, each triangle represent- thirds from D♭ through F and A to C♯, ending off
ing a major or minor triad, and adjacent triads the original tonic. In contrast with the Mozart
retain musically familiar relationships that we example mentioned previously, a Schenkerian
may still label P, R, and L. But the algebra of these analysis of this Liszt work, if it shows the final
transformations is evidently different now: a tonic as a prolongation of the initial tonic, can-
PL-​chain in enh19, for example, cycles through not be valid in enh19.
all 38 major and minor triads before repeating,
rather than returning to its starting point after Exercise 14.3.13 Review the chains of P, R, and
only six transformations like the hexatonic PL-​ L transformations in the examples from Exercises
cycles familiar to us since Section 3.1. There is 4.2.13a (Beethoven), 4.2.13c (Smetana), 4.2.14

FIGURE 14.3.12 A tonnetz in enh19

614 • E x p l o r i n g M u sical S p aces

(Verdi), and 4.2.15 (Schubert). Assume that the Exercise 14.3.15 Which of the following
Verdi example returns to F minor at the end. In statements are true in enh12, and which are
enh12 each of these four progressions is then true in enh19?
tonally closed, ending on the starting triad. If the
same chains of transformations are performed (a) An augmented second is enharmonically
in enh19, only one of these progressions will be equivalent to a minor third.
tonally closed. Which one? Where will each of the (b) Two augmented seconds add up to a doubly
others
   end up? augmented third.
(c) A doubly augmented third is enharmonically
Blackwood’s 19-​note Etude is a sonata form   equivalent to a perfect fourth.
that contrasts a diatonic first theme with a chro-
matic second theme. The second theme has a Exercise 14.3.16
clear tonal center, E major—​Blackwood employs (a) The development section of Blackwood’s
the traditional four-​sharp key signature—​and it 19-​note Etude (mm. 43–​63) modulates
continues to make ample use of major and minor loosely around the 19-​note circle of fifths,
triads. These triads are juxtaposed, however, in following this succession of tonal centers:54
intervallic relationships that are like nothing
available in 12-​tone tuning and sound utterly E –​F♯ –​G♯ –​B♯ (=​C♭) –​F♭ –​D♭ –​A♭ –​C
alien to many listeners. In particular, Blackwood
exploits the division of enh19’s perfect fourth In enh12, does the same chain of diatonic
(specific interval 8) into two equal augmented intervals, starting on E, end on C, or does it
seconds. At the beginning of the second theme, lead somewhere else?
major triads move in parallel through this inter-
val: E–​G♭–​A–​G♭–​E–​D♭–​B. The melodic intervals (b) At m. 55, Blackwood briefly visits the key
are all equal; the effect cannot be captured by of F♭ minor. There is no key signature here,
playing the same notated chords in enh12. A few but theoretically what is the key signature
measures later, in the closing section of the expo- for F♭ minor in enh19? What is its relative
sition, the E–​D♭–​B progression is expanded in the major? What is the key signature for the
sequence shown in Figure 14.3.14, which moves enharmonically equivalent key of E♯ minor?
from tonic to dominant via two augmented sec- Are these key signatures the same as the
onds. In the first measure of this sequence, an key signatures for F♭ minor and E♯ minor
augmented-​sixth chord gives way to a dimin-   in enh12?
ished seventh in a way that cannot be matched
in enh12. The augmented sixth and diminished Exercise 14.3.17
seventh are enharmonically equivalent intervals (a) The 19-​note system, like 7-​note generic
in enh19, and the chords thus share two common space or any other pitch-​class space of
tones, C and A♯/​B♭ ♭. Both chords relate to the sur- prime cardinality, has the property that
rounding major and minor triads in traditional every interval is divisible by every other
ways, but the effect of the sequence is neverthe- interval. In mm. 112–​13 in the coda of the
less quite novel. 19-​note Etude, Blackwood moves rapidly

FIGURE 14.3.14 A chromatic sequence in Blackwood, Etude in 19 Notes, Op. 28, No. 12, mm. 35–38

Beyond the Diatonic • 615

 from the tonic C to the subdominant F via a the melodic minor scale, with raised 6̂ and 7.̂ Every
sequence of descending minor thirds. How melodic minor scale is also an acoustic scale on a dif-
many minor thirds are required? ferent tonic.
(b) How many ascending whole steps are 6. The name Neapolitan minor is used elsewhere as
required to move from tonic to dominant in we use it here. Some sources define Neapolitan major
  enh19? How many descending whole steps? to be a scale such as CD♭E♭FGAB, identified here as a
super–​whole-​tone heptachord. The use of the word
Exercise 14.3.18 Although major and minor major for a scale with ♭3̂ is counterintuitive; the ter-
triads cannot form maximally smooth cycles minology used here also has the advantage that nmin
in the 19-​note universe, there are more and nmaj are inversionally related in spc and in pc.
possibilities for such cycles in enh19 than in 7. For previous analytical applications of some of
enh12. The cycle of diminished seventh chords these scale types, see Tymoczko 2004 (melodic minor/​
in Exercise 14.3.11b provides one example acoustic), Riley 2004 (harmonic major), and Chapter
without a counterpart in enh12. In fact, every 7 of Rings 2011b (double harmonic). Werker 1898
maximally even set type, of any cardinality makes sweeping claims for the Gypsy scale as the
from 1 to 18, can form a maximally smooth source of a wide range of chromatic phenomena in the
cycle. The maximally even set of cardinality music of many cultures.
5 in enh19 is a set with no close analog in 8. Hint for Exercise 14.1.9b: Calculate the sum class
enh12. Describe this set and write a complete of dpc(0) as a pitch-​class set, and consider what hap-
maximally smooth cycle of sets of this type in pens to the sum class as accidentals are adjusted.
musical
   notation. 9. Hints for Exercise 14.1.10: The fact that the set’s
cardinality is odd is crucial for (a). Two notes can lie on
one axis of symmetry in pc, but with the qualification
“more than one note of the set” included in (d) as writ-
NOTES ten, the answers to (c) and (d) are the same.
1. The presentation of spelled heptachords here is 10. The 66 τ-​classes of proper spelled heptachords
revised from Hook 2011. are implicit in the structure of the diatonic bell, a
2. Hints for Exercise 14.1.3: Examples for (b) can be graphical representation of seven-​note scales devised
found among the sets in Table 14.1.2. For (c), the acci- by Pierre Audétat in 2006. See www.cloche-​diatonique.
dental index changes under τ1 transposition for five ch; also see Junod et al. 2009 for further applications.
of the eight sets in the table. The answer to (e) is yes, 11. Hint for Exercise 14.1.12: Three of the hepta-
though only some values of α are possible. For (f), try chords in this exercise require one or more double
writing both sets in scalar order. sharps or flats in their spelling.
3. For diatonic scales we have distinguished the 12. Hint for Exercise 14.1.14: The heptachord is
pitch space dpitch(n) from the pitch-​ class space symmetric, so Exercise 14.1.10a is relevant. This
dpc(n); in Chapter 13 we also introduced enhar- heptachord and a few others of large span have the
monic spaces edpc(n). In principle, these distinc- property that some transpositions of the pc set can-
tions could be made for every spelled heptachord; for not be written as spelled heptachords without the use
harmonic minor scales, for example, we could define of triple sharps or flats, though the balanced form
spaces hminpitch(n) (an infinite set of pitches) and always can.
hminpc(n) (a seven-​note spc set) for every integer n, 13. Hint for Exercise 14.1.20a: The index k in τk is
and ehminpc(n) (a seven-​note pc set) for every mod-​ an integer, but when the same k appears in tk (or in
12 integer n. Apart from brief references to the enhar- this case t3k), it may be regarded as an integer mod 7
monic spaces in Exercise 14.1.19 and Section 14.2, we (because t7 =​ t0). For example, if k =​4, then t3k is the
will not find it necessary to make these distinctions; same as t5.
we generally understand that hmin(n) refers to the 14. A related perspective is the chord-​scale theory
seven-​note space hminpc(n). familiar to many jazz musicians, originating in work
4. Hints for Exercise 14.1.7: Several possible exam- of George Russell (2001) dating to the 1950s.
ples for this exercise may be found among the spc sets 15. Satyendra 1992, 310–​18, offers a more exten-
in Table 14.1.2. Try sets of cardinalities both smaller sive study of inflected repetition in this waltz.
and larger than 7. 16. While dismissing selected notes as foreign to
5. References to melodic minor here always refer to the heptachord is often appropriate, it is sometimes
what is traditionally known as the ascending form of not difficult to identify another heptachord from

616 • E x p l o r i n g M u sical S p aces

which such notes may have been borrowed. A source discrete jumps; the resulting pitch-​class set, however,
for the G𝄪 at m. 22, for example, could be nmin(+​7) =​ changes only at multiples of 1 7 semitone.
C♯D♯E♯F♯G𝄪A♯B. 23. For the procedure in Exercise 14.2.6, see
17. Most of the chords in Figure 14.1.29 are sev- Clough and Douthett 1991, 152, Theorem 3.5.
enth chords. The occasional links between triads and 24. London (2012, 128–​29) includes a maximal
seventh chords may be modeled with the function evenness condition in a list of well-​formedness con-
sup* from Section 13.1. straints for meter as a way of ensuring regularity with-
18. Hint for Exercise 14.1.31d: Schubert’s enhar- out requiring isochrony (perfect evenness). Toussaint
monic notation in mm. 79–​80, respelled by ε1, con- (2020, 129–​38) examines some alternate character-
verts the spelled heptachord for the minor Neapolitan izations of maximal evenness in a rhythmic context.
from hmin(−13) to hmin(−1)—​remarkably still in flat The rhythms in Figure 14.2.9 are all generated sets, a
territory even after the sharpwise respelling. property explored in Cohn 2016.
19. Hints for Exercise 14.1.32: Apart from the 25. Time signatures are not displayed in Figure
initial tonic and the concluding pre-​dominant and 14.2.9 because each of these patterns may be
dominant, the remaining chords (shown previously heard metrically in a variety of ways. There are
in Figure 11.5.12a) form a generic descending-​thirds significant perceptual differences, for example,
sequence, discussed in Hook [2014] 2020. Almost between hearing the tresillo rhythm as a synco-
every chord may be assigned to a spelled heptachord pation against an underlying  meter and hear-


with index number +​1 or +​2. Assume that Chopin’s ing three unequal beats per measure, as Bartók’s
E♭ in mm. 2–​3 is a D♯. The diminished seventh at the  +  + notation in Mikrokosmos No. 153 implies.
end of m. 4 presents the greatest challenge; you may 26. The principle that maximally even sets are the
consider respelling it to clarify a connection with the closest integer-​valued points to perfectly even sets
preceding or following chord. The generic voice lead- generally holds if distance is measured using either
ing in this example illustrates Exercise 13.1.20a and the Euclidean distance or the voice-​leading distance.
may be visualized in Figure 13.1.14. See Tymoczko 2013 for a more general discussion of
20. Maximal evenness is fundamentally a math- this principle.
ematical concept, not an intrinsically musical one, 27. Hint for Exercise 14.2.10b: Part (a) may be help-
and many non-​musical manifestations may be imag- ful. Alternatively, recall the relationship between sum
ined. The set {Monday, Wednesday, Friday} is a maxi- class and accidental index for spelled heptachords
mally even subset both of the mod-​7 week and of the from Exercise 14.1.9b.
mod-​5 week (excluding weekends). The seven 31-​day 28. Figure 14.2.11 is modeled on Tymoczko 2011,
months form a maximally even subset—​ indeed, a 111, Figure 3.11.9. See also Tymoczko 2012 for fur-
diatonic subset—​of the twelve months of the year. If ther commentary on the structure of scale lattices.
equally charged particles are somehow constrained to 29. In seven-​voice OP-​space, for example, the
occupy discrete positions around a circle, maximally F major, C major, and A harmonic minor scales do
even configurations will be the most stable. A related not lie on a straight line, as they appear to in Figure
situation arises in the Ising model, a celebrated math- 14.2.11.
ematical model of ferromagnetism involving particles 30. Hint for Exercise 14.2.12a: The answer is the
with spins in certain orientations, in which energy number of “combinations of 12 things 7 at a time.”
is minimized for maximally even arrangements. See The number of ordered seven-​element sets (“permu-
Douthett and Krantz 2007 for more discussion of the tations of 12 things 7 at a time”) is 12 ∙ 11 ∙ 10 ∙ 9 ∙
Ising model and other examples of maximal evenness 8 ∙ 7 ∙ 6, but this number counts every unordered set
beyond the musical realm. 7! times.
21. A fraction ending in 1 2 will never appear when 31. See Tymoczko 2011, 103–​12, and Tymoczko
the fractions are multiples of 1 7 , but in some quanti- 2012, 15–​16. There is theoretically a third case in
zation procedures half-​integers are possible. In such which d is neither a divisor of c nor coprime to c, but
cases one must observe a consistent convention, this case rarely arises in familiar musical situations.
either rounding all half-​integers up or rounding them 32. Hint for Exercise 14.2.14a: All the calculations
all down. may be done mod 12. The two inner circles contain
22. A 1 7 -​semitone rotation is exactly 1 84 of a circle, three-​and seven-​note perfectly even sets in continu-
consistent with the observation from Section 13.3 ous mod-​12 space. These two circles are rotated, how-
that s1 generates a cyclic group of order 84. The inner ever, so that neither of them contains the number 0.
circle may be rotated continuously rather than in To determine the numbers in these sets, examine the

Beyond the Diatonic • 617

figure carefully, noting which points align with points symmetric designs must be excluded from the state-
in other circles. ment of the theorem or accounted for separately.
33. Hint for Exercise 14.2.14e: The progression is one 38. Hints for Exercise 14.3.1: Some properties may
that we have encountered several times previously. You be somewhat tedious to check; the work can be sim-
will obtain the correct result if you rotate the middle plified by relying on some of the logical relationships
circle in increments of 2 7 semitone at a time. After mentioned in the paragraph before the exercise. You
six such rotations the circle will have moved through should find that two of the given scales have four of
exactly one-​seventh of an octave, so it will appear as the five properties, one scale has three properties, one
it did initially; the progression therefore repeats after has two, one has one, and two have none of the prop-
six chords. Each time a note on the middle circle moves erties at all. Because the usual diatonic scale satisfies
past a stationary note on the outer circle, one voice in C =​V, the properties of the chromatic heptachord in
the chord moves down by semitone, but when a note (f) may appear to contradict Clough and Myerson’s
on the middle circle passes a note on the inner circle, demonstration (1985, 266) that for given c and d, a
a voice moves up instead. This system is depicted in scale satisfying C =​V is “essentially unique.” This theo-
Douthett 2008, 81, with the progression traversed in rem applies only to what Clough and Myerson call
the reverse order from the description here. semireduced scales, in which the two specific intervals
34. Hint for Exercise 14.2.14f: Because the outer corresponding to generic interval 1 are consecutive
two circles are fixed, this progression remains within integers. The chromatic heptachord is not semire-
one octatonic scale. Even if the inner circle rotates duced; it may be converted to a semireduced scale by
continuously, voices sometimes move by whole step removing four of the five notes of C that are not in the
rather than by semitone. See Plotkin and Douthett heptachord. In this sense the chromatic heptachord
2013, 108. is “essentially” the same as [0, 1, 2, 3, 4, 5, 6]8, the
35. The transformation groups act on configura- unique C =​V scale with c =​8 and d =​7.
tions, which determine chords; the groups do not act 39. Hints for Exercise 14.3.2: One of these scales
directly on the space of chords. For example, write r has all five properties; one scale has four properties,
for a counterclockwise rotation of the middle circle in two scales have three, one has two, and two have one.
Figure 14.2.13b by 1 7 semitone. In the configuration Clough, Engebretsen, and Kochavi 1999 show that
shown, a single application of r does not change the the scale in (a) is the unique scale with its particular
chord, but the chord does change when r is applied a combination of attributes. Note the similarities of the
second time. If X denotes the C-​major triad, it seems seven-​note scales in (d) and (g) to the usual diatonic
to follow that r(X) =​ X but r(r(X)) ≠ X, which is clearly [0, 1, 3, 5, 6, 8, 10]12: the usual half and whole steps
impossible. The problem is that r is not a well-​defined have been replaced by intervals of specific sizes 1 and
function on chords: every application of r changes the 3 in (d), 2 and 4 in (g). Scales (e) and (f) are comple-
configuration, but different configurations may deter- ments in the mod-​20 universe; complementary scales
mine the same chord. often share many, but not all, properties.
36. See Tymoczko 2011, 105, Figure 3.11.2b for 40. Hint for Exercise 14.3.3a: If n =​ ⎣c/​2⎦, then an
a version of the lattice described in Exercise 14.2.15. interval-​class vector consists of n numbers, which for
While Figure 14.2.11 was somewhat misleading as a a deep set must all be different (though one of them
depiction of seven-​voice mod-​12 OP-​space, this new may be 0).
lattice accurately shows a central portion of three-​ 41. See also Agmon 1996, which examines alterna-
voice mod-​7 OP-​space (see Exercise 13.1.28), because tive formulations of the concept of “diatonic system”
there are now as many dimensions as voices. Frederick in relation to the work of Clough and others.
2018 examines the relationship between the two fig- 42. For Balzano, k and k +​1 are the generat-
ures in greater detail, and comments more generally ing intervals of the tonnetz, and their sum is g,
on Tymoczko’s two types of lattices. the generating interval of the scale—​but this sum
37. For a thorough study of C =​V for chords see must also equal d in order for the scale to match the
Clampitt 2008. The exclusion of trivial genera in appearance of the diatonic scale in the usual ton-
Theorem 13.4.2 suggests a reason why C =​V for netz. Recall Figure 1.4.12, based on Balzano’s ton-
chords is more difficult than for lines: if d is not prime, netz, in which the diatonic scale spans one 3-​by-​4
many chords may be transpositionally symmetric. The block diagonally.
word trivial is not appropriately applied to transposi- 43. Douthett and Hook 2009 show that a quarter-​
tional symmetry in such cases, but transpositionally tone system admits a consistent system of diatonic

618 • E x p l o r i n g M u sical S p aces

interval names, extending the usual system to include out to be precisely those whose EE vectors are regular
intervals that are traditionally forbidden; for example, of index 5, of the form (a, a, a, a, a, b, b) with a > b > 0
an interval of 31 2 semitones is a “perfect third,” and (Hook 2007c, 119).
51 2 semitones is a “major fourth.” 51. Mandelbaum 1961 offers a wealth of detail on
44. The approximations are described by the the history and theory of 19-​note scales and tunings.
approximate equalities 27⁄ 12 ≈ 3 2 for the perfect fifth See also Yasser 1932, 281, for a keyboard diagram
and 24 ⁄ 12 ≈ 5 4 for the major third. These formulas may essentially matching Figure 14.3.8a.
be recast in forms describing near-​matches between 52. Hints for Exercise 14.3.10: The frequency ratio
powers of 2 and powers of 3 and 5, namely 219 ≈ 312 of the pure major third is 5 4 ; of the minor third, 6 5.
and 27 ≈ 53. In the general case the latter formulas Base-​2 logarithms may be calculated if necessary by
become 2c+​g ≈ 3c and 22c+​h ≈ 5c, where g and h are the the formula log2(x) =​log10(x)/​log10(2). You should
specific intervals approximating the perfect fifth and find that the minor third in 19-​note tuning is almost
major third, respectively. perfectly in tune. Notwithstanding the results of this
45. Hints for Exercise 14.3.4a: If d is odd, then d =​ exercise, for many first-​time listeners to 19-​note tun-
2k +​1 for some integer k. Write c and d 2 in terms of k. ing it is the major thirds that sound noticeably odd,
To show that c and d are coprime, calculate 2d − c, and because they are so much smaller than the large
observe that any number that is a factor of both c and major thirds we are accustomed to hearing in the 12-​
d must be a factor of this number. note scale.
46. Hint for Exercise 14.3.4b: Use the formulas from 53. In the transformation R, the moving voice
(h) to rewrite (g) in terms of k, and solve for k. moves through a whole step, which is now three
47. In Hook 2007c the accidental index of a vec- specific steps. Cohn’s requirement that the car-
tor is called its weight, and the step position is called dinality c be a multiple of 3 (Cohn 1997, 13; our
its index. The terminology is adjusted here to be more property (c) above) is derived from the stipula-
consistent with spelled heptachords. tion that transformations analogous to P, R, and L
48. Hints for Exercise 14.3.6: The two small steps entail no more than two specific steps of voice lead-
in the white-​note scale in the 17-​note system, and ing, thus excluding R in enh19 from consideration.
also in the 22-​note system, should correspond to the Acoustically, however, R is a smoother transforma-
intervals B–​C and E–​F. In the 24-​note system, because tion in enh19 than in enh12, as 3 19 is a smaller frac-
two specific steps add up to one traditional semitone, tion of an octave than 2 12 .
the interpretation of the accidentals ♯ and ♭ is poten- 54. All the pitch classes indicated are local ton-
tially confusing. If we write ↑ and ↓ rather than ♯ and ics with the exception of A♭, which arrives at m. 58
♭ for the incremental (quarter-​tone) modifiers in this as a dominant of D♭, then resolves as an augmented
enharmonic system, then by defining ♯ =​ ↑↑ and ♭ =​ ↓↓ sixth in C.
we may preserve the usual meanings of the symbols
♯ and ♭.
49. Hint for Exercise 14.3.7b: See Table 14.1.6.
SUGGESTED READING
50. See Hook 2007c, 116–​18. In some cases acci- Spelled heptachords were introduced in Hook
dentals must be redefined; in Blackwood’s notation 2011. Clough and Douthett’s 1991 article on
for the 17-​note Etude, for example, a ♯ alters a note by maximally even sets, while long and technical,
two specific steps. The scale used in the 13-​note Etude explores many details not considered here, and
is not a 13-​note enharmonic system but comprises Douthett and Krantz 2007 describe non-​musical
every other note of a 26-​note system whose EE vec- applications of maximal evenness. Filtered point-​
tor Φ26 =​(4, 4, 4, 4, 4, 3, 3) is standard of index 5. The symmetry models made their first appearance
only Etude whose notation is not consistent with any in Douthett 2008, with more applications and
enharmonic system is the Etude in 23 notes, a scale further theoretical development in Plotkin and
that Blackwood describes as “a particular challenge.” Douthett 2013 and Plotkin 2019. Tymoczko 2012
Although tuning is not a criterion in the definition examines relationships between scale lattices and
of enharmonic systems, some aspects of enharmonic tonnetzes, and Tymoczko 2013 studies maximal
systems align well with properties that Blackwood evenness in relation to signature transformations
formulates in terms of tuning. For example, the equal and considerations of voice leading and geometry.
tunings that contain what Blackwood calls “recog- Toussaint 2020 diagrams many rhythms in beat-​
nizable” diatonic scales (defined in Blackwood 1985, class circles like those in Figure 14.2.9 and studies
195–​99, on the basis of acoustic interval sizes) turn their mathematical properties.

Beyond the Diatonic • 619

For logical relationships among many properties cardinalities) as well as Agmon 1989 and 1996.
of scales of various specific and generic cardinali- Žabka 2014 proposes a theory of generated
ties like those studied in the first part of Section scales modeled via interval lattices, with con-
14.3, see Clough, Engebretsen, and Kochavi nections to spelled heptachords and other con-
1999. For other approaches to the problem of cepts discussed here. For enharmonic systems
identifying scales that mimic some aspects of and a tabulation of the systems employed in
the canonical 7-​in-​12 system, see Balzano 1980 Blackwood’s Microtonal Etudes, see Hook 2007c.
and Cohn 1997 (which propose two very dif- For a wealth of background on 19-​note tuning
ferent ways to generalize the tonnetz to other see Mandelbaum 1961.

620 • E x p l o r i n g M u sical S p aces

 Appendix 1

List of Musical Spaces

The most important musical spaces described in QS: Quotient space (2.4, 7.4)
this book are listed on the following pages. The ~C: Cardinality equivalence (2.4, 10.3)
following abbreviations are used: ~E: Enharmonic equivalence (1.2, 2.4, 2.6)
~G: Generic equivalence (2.4.5, 2.6)
Reference: Section, figure, and exercise ~I: Inversional equivalence (10.3)
numbers in this book ~O: Octave equivalence (1.2, 2.4, 10.3)
IS, IG, IF: Interval space, interval group, ~P: Permutational equivalence (2.4, 10.3)
interval function (5.4) ~T: Transpositional equivalence (2.4, 10.3)
DS, DF: Distance space, distance ≅: Isomorphic interval spaces (7.2)
function (12.2) or isometric distance spaces (12.2)
L AB E L NA M E REFERENCE N OT E S

bc(n) mod-​n beat-​class 2.5 Finite modular IS of cardinality n with IG

space (ℤn, +​); QS of beat modulo mod-​n beat
equivalence; finite DS
beat beat space 1.4 Infinite IS with IG (ℤ, +​), ≅ pitch; DS ≅ ℤ
cpc continuous pitch-​ 1.2.3 Continuous modular IS with IG (ℝ12, +​); QS
class space cpitch ∕~O; DS with DF defined by | |12 (12.2)
cpcn n-​voice pitch-​ 10.3 Cartesian product cpc × ⋯ × cpc (n copies),
class–​tuple space ordered n-​tuples of elements of cpc;
(O-​space) topologically 𝕋 n; IS with IG ℝ12n; DS with
max, Euclidean, or voice-​leading DF; QS
cpitchn/​~O
cpitch continuous pitch 1.1.1 Infinite IS with IG (ℝ, +​); DS ≅ ℝ
space
cpitchn n-​voice pitch-​tuple 10.3 Cartesian product cpitch × ⋯ × cpitch
space (n copies), ordered n-​tuples of elements of
cpitch; topologically ℝn; IS with IG ℝn; DS
with max, Euclidean, or voice-​leading DF;
OPTIC spaces are QSs of cpitchn modulo
combinations of ~O, ~P, ~T, ~I, ~C
cpitch≤n pitch-​tuple space 10.3 Union of pitch-​tuple spaces up to n voices,
up to n voices cpitch1 ∪ cpitch2 ∪ ⋯ ∪ cpitchn
cpitch* general pitch-​tuple 10.3 Infinite union of all pitch-​tuple spaces,
space cpitch1 ∪ cpitch2 ∪ cpitch3 ∪ ⋯
dfifth(n) diatonic (mod-​7) 13.2.1 Same set as dpc(n); ≅ dpc(n) as IS but
fifths space with different IF
index n
dharm(n) double harmonic 14.1.8 Double harmonic (Gypsy) spelled
heptachord of accidental index n
dpc(n) diatonic (mod-​7) 1.2.9, 13.2.1 n ∈ ℤ is central note of diatonic collection
pitch-​class space in spc, also indicates number of sharps or
with index n flats in scale; 7-​note modular IS with IG
(ℤ7, +​), ≅ gpc; QS dpitch(n) ∕ ~O; finite DS;
may be identified with a subset of spc; via
~E becomes edpc(n), a subset of pc; also a
spelled heptachord (14.1.8)
dpitch(n) diatonic pitch space 1.1.7 n ∈ ℤ is central note of diatonic collection in
with index n spc, also indicates number of sharps or flats
in scale; infinite IS with IG (ℤ, +​), ≅ gpitch
and pitch; DS ≅ ℤ
edpc(n) enharmonic 13.2 n ∈ ℤ12; space formed by ~E from all dpc(m)
diatonic pitch-​class with m ≡ n (mod 12); subset of pc; 7-​note
space with index n modular IS with IG (ℤ7, +​), ≅ gpc; finite DS
edpc* unified enharmonic 13.3 Union of all 12 spaces edpc(n); finite IS of
space cardinality 84 with IG ≅ ℤ84

622 • A ppe n di x 1
L AB E L NA ME REFERENCE N OT E S

enhΘ enharmonic system 14.3 Modular IS with IG (ℤc, +​), where c is the
with vector Θ accidental index of vector Θ; contains a copy
of gpc in the 7 white notes
enhn standard 14.3 enhΘ, where Θ =​ Φn (standard vector of
enharmonic system accidental index n); modular IS with IG
of accidental index n (ℤn, +​); contains a copy of gpc in the 7
white notes; enh12 is the same as pc
enn012, enneatonic pitch or 3.1.17 ennabc is the enneatonic collection
enn123, pitch-​class space containing pcs a, b, c; ennabc may refer to a
enn234, pitch space (infinite IS with IG (ℤ, +​), ≅ pitch
enn345 and gpitch; DS ≅ ℤ) or to a pitch-​class space
(9-​note modular IS with IG (ℤ9, +​); finite DS)
fifth enharmonically 1.3.1 12-​note modular IS with IG (ℤ12, +​); same
conformed fifths set as pc; ≅ pc as IS but different IF; QS
space (the circle of spc ∕ ~E; finite DS
fifths)
freq frequency space 1.1.5 Same notes as cpitch, interpreted as
frequencies; infinite IS with IG (ℝ>0, ∙);
≅ cpitch as IS, but different IF

gfifth generic (mod-​7) 1.3.8 Same set as gpc; ≅ gpc as IS but different
fifths space IF; QS spc ∕~G; finite DS
gpc generic (mod-​7) 1.2.8 7-​note modular IS with IG (ℤ7, +​); QS gpitch ∕~O;
pitch-​class space finite DS with DF defined by | |7 (12.2)
gpitch generic pitch space 1.1.6 Infinite IS with IG (ℤ, +​), ≅ pitch; DS ≅ ℤ
gsev generic seventh-​ 13.1 Space of the 7 generic seventh chords
chord space (CEGB, DFAC, …); ≅ gpc as IS
gthird generic (mod-​7) 1.3.12 Same set as gpc; ≅ gpc as IS but different IF
thirds space
gtriad generic triad space 4.3.1, 13.1 Space of the 7 generic triads (CEG, DFA, …);
≅ gpc as IS

hex01, hexatonic pitch or 4.2 hexab is the hexatonic collection containing

hex12, pitch-​class space pcs a, b; hexab may refer to a pitch space (IS
hex23, with IG (ℤ, +​), ≅ pitch and gpitch; DS ≅ ℤ)
hex34 or to a pitch-​class space (6-​note modular IS
with IG (ℤ6, +​); finite DS)
hmaj(n), harmonic major, 14.1.8 Harmonic major and harmonic minor
hmin(n) harmonic minor spelled heptachords of accidental index n
just just pitch (or 2.3.26, 7.3.7, Infinite IS with IG {2a · 3b · 5c | a, b,
frequency) space 7.4 c ∈ ℤ} ≅ ℤ × ℤ × ℤ; subspace of freq;
just frequency-​class space just ∕~O
(Riemannian tonnetz space) is infinite IS
with IG {3b · 5c | b ∈ ℤ}
key key space 4.1 Same elements as triad, interpreted as
keys; various possible IS and DS structures

A ppe n di x 1 • 623
L AB E L NA M E REFERENCE N OT E S

mmin(n) melodic minor 14.1.8 Melodic minor (or acoustic) spelled

heptachord of accidental index n
nmaj(n), Neapolitan major, 14.1.8 Neapolitan major and Neapolitan minor
nmin(n) Neapolitan minor spelled heptachords of accidental index n
nrtriad neo-​Riemannian 4.2.5 Same set as triad; 24-​element DS with
triad space DF defined by neo-​Riemannian distance
distNR (12.3); noncommutative IS with
IG ℛ (Riemann group)
oct01, octatonic pitch or 2.1.5, 7.5.11 octab is the octatonic collection containing
oct12, pitch-​class space pcs a, b; octab may refer to a pitch space
oct23 (IS with IG (ℤ, +​), ≅ pitch and gpitch;
DS ≅ ℤ) or to a pitch-​class space (8-​note
modular IS with IG (ℤ8, +​); finite DS)
pc discrete chromatic 1.2.3 12-​note modular IS with IG (ℤ12, +​);
(mod-​12) pitch-​class QS pitch ∕~O; finite DS with DF defined
space by | |12 (12.2)
pcset set of all pitch-​class 2.3 No standard IS or DS structure on the
sets full set pcset; sets of ≤ n pcs form QS
pitchn/​~OPC
pent(n) pentatonic pitch or 7.5.14 Index number is central note of pentatonic
pitch-​class space collection in spc; label may refer to a pitch
with index n space (IS with IG (ℤ, +​), ≅ pitch and gpitch;
DS ≅ ℤ) or to a pitch-​class space (5-​note
modular IS with IG (ℤ5, +​); finite DS)
pitch discrete chromatic 1.1.1 Infinite IS with IG (ℤ, +​); DS ≅ ℤ
pitch space
pyth Pythagorean pitch 2.3.25, 7.3, Infinite IS with IG {2a · 3b | a, b ∈ ℤ} ≅ ℤ × ℤ;
(or frequency) 7.4 subspace of freq; Pythagorean frequency-​
space class space pyth ∕~O is infinite IS with IG
{3b | b ∈ ℤ}, ≅ spc
row 12-​tone row space 6.5, 9.5 Complete space contains 12! rows; 48 forms
of one row form a noncommutative IS with
IG 𝒮er12 ≅ 𝒯ℐ12 × 𝒞2; other IS structures
possible (4.1.12, 4.1.16, 9.5)
sd scale-​degree space 1.2.10 7-​note modular IS with IG (ℤ7, +​), ≅ gpc and
dpc(n); finite DS
sev dominant/​half-​ 8.3.19 24 dominant and half-​diminished seventh
diminished seventh chords; various IS structures similar to
chord space triad
shexa(n), superhexatonic 14.1.8 2 superhexatonic spelled heptachords of
shexb(n) accidental index n
sl signed letter space 1.2.11 Infinite IS with IG ℤ7 × ℤ; QS srl ∕~O; same
set as spc, but different IF, ≇ spc as IS

624 • A ppe n di x 1
L AB E L NA ME REFERENCE N OT E S

socta(n), suboctatonic 14.1.8 2 suboctatonic spelled heptachords of

soctb(n) accidental index n
spc spelled pitch-​class 1.3.3, 2.6.1 Infinite IS with IG ℤ, ≅ pitch; DS ≅ ℤ
space (the line of
fifths)
srl signed registral 1.1.11 Infinite IS with IG ℤ × ℤ
letter space
swt(n) super–​whole-​tone 14.1.8 Super–​whole-​tone spelled heptachord of
accidental index n
time temporal space 1.4 Infinite IS with IG (ℝ, +​), ≅ cpitch; DS ≅ ℝ
tonn conformed 1.4.2–​1.4.10, Same set as pc; 12-​note IS with IG ℤ4 × ℤ3;
(neo-​Riemannian) 3.3.10, 6.4.4 ≅ pc as IS, but different IF
tonnetz space
triad consonant triad 4.1–​4.2, 8.1 Set of cardinality 24; various possible IS
space (major, minor and DS structures, including circle-​of-​fifths
triads) distance, nrtriad, weber (4.1–​4.2, 12.3)
weber Weber space 4.2.3 Same set as triad; DS with DF defined by
Weber distance (4.2, 12.3)
wt0, wt1 whole-​tone pitch or 2.1.5 wta is the whole-​tone collection containing
pitch-​class space pc a; wta may refer to a pitch space (IS with
IG (ℤ, +​), ≅ pitch and gpitch; DS ≅ ℤ) or to
a pitch-​class space (6-​note modular IS with
IG (ℤ6, +​); finite DS)

A ppe n di x 1 • 625
 Appendix 2

List of Sets and Groups

Many sets and groups in this list are familiar are customized for the musical applications in
to mathematicians, and many of the labels are this book.
standard in the mathematical literature; others
L AB E L NA M E REFERENCE N OT E S

𝒜n alternating group on n 6.1 Even permutations; a group of order n!/​2;

objects noncommutative for n ≥ 4
𝒜𝑓𝑓n affine group on n objects 6.1 Generated by transpositions and
multiplication operators Ma with a
coprime to n; a group of order n ∙ φ(n)
𝒞n cyclic group on n objects 5.3 General name for a cyclic group of order n,
generated by a single element; rotations of
an n-​sided polygon
𝒞∞ infinite cyclic group 5.3 General name for an infinite cyclic group,
generated by a single element; ≅ (ℤ, +​)
𝒟n dihedral group on n 5.5 General name for a dihedral group, a
objects noncommutative group of order 2n;
rotations and reflections of an n-​sided
polygon
𝒟∞ infinite dihedral group 5.5 General name for an infinite dihedral
group; translations and reflections of ℤ
Kn complete graph on n 3.3 A graph with n vertices, each pair joined by
vertices one edge
ℳ mediant group 8.3 Triadic transformation group generated by
the mediant transformation M; ≅ 𝒞24
ℳn multiplication group 6.1 Mod-​n multiplication operators Ma with a
on ℤn coprime to n; group of order φ(n), ≅ ℤn×
ℚ the rational numbers 2.1 An infinite additive group; ℚ≠0 and ℚ>0 are
multiplicative groups
𝒬 QTT group 8.3 All quasi-​uniform triadic transformations;
a group of order 1152, ≅ 𝒟12 ≀ 𝒞2
ℝ the real numbers 2.1 An infinite additive group; ℝ≠0 and ℝ>0 are
multiplicative groups
ℝn the real numbers mod n 2.5 An infinite additive group; quotient group
of ℝ by congruence mod n
ℝn n-​dimensional Euclidean 2.2 The Cartesian product ℝ × ℝ × ⋯ × ℝ
space (n copies)
ℛ Riemann group 8.2 Riemannian triadic transformations (PLR
group); a noncommutative group of order
24, ≅ 𝒟12, generated by L and R
𝕊n n-​dimensional sphere 11.1 𝕊2 is (the surface of) the usual sphere; 𝕊1 is
a circle
𝒮n symmetric group on n 6.1 All permutations; a group of order n!
objects
𝒮e𝑟n serial group on 6.2 Generated by transposition, inversion, and
permutations of n retrograde; a noncommutative group of
objects order 4n, ≅ 𝒟n × 𝒞2; familiar example is 𝒮e𝑟12

628 • A ppe n di x 2
L AB E L NA M E REFERENCE N OT E S

𝒮e𝑟12+​ extended serial group 9.6 Generated by transposition, inversion,

retrograde, and rotation; a noncommutative
group of order 576; ≅ 𝒟12 × 𝒟12
𝒮e𝑟12X extended serial group 9.6 Generated by 𝒮e𝑟12+​ and X (order/​pitch-​
with order/​pitch-​class class exchange); a noncommutative group
exchange of order 1152; ≅ 𝒟12 ≀ 𝒞2; also ≅ 𝒬
𝒮𝑖𝑔𝑛 sign group 5.2 Multiplicative group {1, −1} or {+​, −}; ≅ 𝒞2
𝕋n n-​dimensional torus 1.4, 10.1 𝕋 2 is (the surface of) the usual torus; 𝕋 1 is
a circle
𝒯n transposition group on 5.3 A cyclic group of order n, ≅ 𝒞n; familiar
mod-​n pitch-​class space examples are 𝒯12 (acting on pc) and 𝒯7 (on
gpc); acts primarily on pitch classes but
also on pc sets, rows
𝒯S general transposition 5.6 Group of transposition operators Transi on
group an IS S; ≅ the IG of S
𝒯∞ transposition group on 5.1, 5.3 An infinite group, ≅ 𝒞∞
discrete pitch space
𝒯ℐn transposition/​inversion 5.5 A dihedral group of order 2n, ≅ 𝒟n; familiar
group on mod-​n pitch-​ example is 𝒯ℐ12; acts primarily on pitch
class space classes but also on pc sets, rows
𝒯ℐ∞ transposition/​inversion 5.5 An infinite group, ≅ 𝒟∞
group on discrete pitch
space
𝒰 UTT group 8.1 All uniform triadic transformations; a
group of order 288, ≅ 𝒞12 ≀ 𝒞2
𝒱4 Klein group 6.3 A group of order 4, ≅ 𝒞2 × 𝒞2
𝒲 Weber group 8.3 Subgroup of 𝒰 of order 48, generated by
P, R, and D
ℤ the integers 2.1 An infinite additive group, ≅ 𝒞∞
ℤn the integers mod n 2.5 A cyclic group of order n, ≅ 𝒞n; quotient
group of ℤ by congruence mod n
ℤn× group of units mod n 5.2 Integers mod n that are coprime to n; a
multiplicative group of order φ(n)

A ppe n di x 2 • 629
 References

Adams, John. 2008. Hallelujah Junction: Composing an Attas, Robin. 2009. “Metaphors in Motion: Agents
American Life. New York: Farrar, Straus and Giroux. and Representation in Transformational
Agmon, Eytan. 1989. “A Mathematical Model of Analysis.” Music Theory Online 15 (1).
the Diatonic System.” Journal of Music Theory 33 Babbitt, Milton. [1960] 2003. “Twelve-​Tone
(1): 1–​25. Invariants as Compositional Determinants.”
Agmon, Eytan. 1995. “Functional Harmony In The Collected Essays of Milton Babbitt, edited
Revisited: A Prototype-​Theoretic Approach.” by Stephen Peles, pp. 55–​69. Princeton, NJ:
Music Theory Spectrum 17 (2): 196–​214. Princeton University Press. Original publication,
Agmon, Eytan. 1996. “Coherent Tone-​Systems: A Musical Quarterly 46 (2): 246–​59.
Study in the Theory of Diatonicism.” Journal of Babbitt, Milton. [1961] 2003. “Set Structure as a
Music Theory 40 (1): 39–​59. Compositional Determinant.” In The Collected
Alegant, Brian. 1993. “The Seventy-​Seven Partitions Essays of Milton Babbitt, edited by Stephen Peles,
of the Aggregate: Analytical and Theoretical pp. 86–​108. Princeton, NJ: Princeton University
Implications.” Ph.D. dissertation, Eastman Press. Original publication, Journal of Music
School of Music. Theory 5 (1): 72–​94.
Alegant, Brian. 1999. “When Even Becomes Odd: Babbitt, Milton. [1965] 2003. “The Structure and
A Partitional Approach to Inversion.” Journal of Function of Musical Theory.” In The Collected
Music Theory 43 (2): 193–​230. Essays of Milton Babbitt, edited by Stephen
Anonymous. 1995. Musica enchiriadis and Scolica Peles, pp. 191–​201. Princeton, NJ: Princeton
enchiriadis. Translated by Raymond Erickson. University Press. Original publication, College
New Haven, CT: Yale University Press. Music Symposium 5: 49–​60.
Babbitt, Milton. 1992. “The Function of Set Blackwood, Easley. 1994. Liner notes to Microtonal
Structure in the Twelve-​Tone System.” Ph.D. Compositions, compact disc recording, Cedille
dissertation, Princeton University. 90000 018.
Bailey, Kathryn. 1991. The Twelve-​Note Music of Blizard, Wayne D. 1989. “Multiset Theory.” Notre
Anton Webern: Old Forms in a New Language. Dame Journal of Formal Logic 30 (1): 36–​66.
Cambridge, UK: Cambridge University Press. Bondy, J. A., and U. S. R. Murty. 2008. Graph Theory.
Balzano, Gerald J. 1980. “The Group-​Theoretic New York: Springer.
Description of 12-​Fold and Microtonal Pitch Brinkman, Alexander R. 1986. “A Binomial
Systems.” Computer Music Journal 4 (4): 66–​84. Representation of Pitch for Computer Processing
Bamberger, Jeanne. 1986. “Cognitive Issues in the of Musical Data.” Music Theory Spectrum 8: 44–​57.
Development of Musically Gifted Children.” In Brower, Candace. 2008. “Paradoxes of Pitch Space.”
Conceptions of Giftedness, edited by Robert J. Music Analysis 27 (1): 51–​106.
Sternberg and Janet E. Davidson, pp. 388–​413. Brown, Stephen C. 2003. “Dual Interval Space
New York: Cambridge University Press. in Twentieth-​Century Music.” Music Theory
Bard-​Schwarz, David, and Richard Cohn, eds. Spectrum 25 (1): 35–​57.
2015. David Lewin’s Morgengruß: Text, Context, Browne, Richmond. 1981. “Tonal Implications of
Commentary. New York: Oxford University Press. the Diatonic Set.” In Theory Only 5 (6–​7): 3–​21.
Barker, Andrew, ed. 1989. Greek Musical Writings. Buchler, Michael. 2007. “Reconsidering
Cambridge, UK: Cambridge University Press. Klumpenhouwer Networks.” Music Theory
Bass, Richard. 1988. “Prokofiev’s Technique of Online 13 (2).
Chromatic Displacement.” Music Analysis 7 (2): Buchler, Michael. 2016. “Are There Any Bad (or
197–​214. Good) Transformational Analyses?” Intégral
Bass, Richard. 1996. “From Gretchen to Tristan: The 30: 41–​51.
Changing Role of Harmonic Sequences in the Budden, F. J. 1972. The Fascination of Groups.
Nineteenth Century.” 19th-​Century Music 19 (3): Cambridge, UK: Cambridge University Press.
263–​85. Bulger, David, and Richard Cohn. 2016.
Bates, Ian. 2012. “Vaughan Williams’s Five Variants “Constrained Voice-​Leading Spaces.” Journal of
of ‘Dives and Lazarus’: A Study of the Composer’s Mathematics and Music 10 (1): 1–​17.
Approach to Diatonic Organization.” Music Burago, Dmitri, Yuri Burago, and Sergei Ivanov.
Theory Spectrum 34 (1): 34–​50. 2001. A Course in Metric Geometry. Providence,
Baur, Steven. 1999. “Ravel’s ‘Russian’ Period: RI: American Mathematical Society.
Octatonicism in His Early Works, 1893–​1908.” Burkhart, Charles. 1973–​74. “Schoenberg’s Farben:
Journal of the American Musicological Society 52 An Analysis of Op. 16, No. 3.” Perspectives of New
(3): 531–​92. Music 12 (1–​2): 141–​72.
Bazayev, Inessa. 2013. “Triple Sharps, Qnt Burstein, L. Poundie, and Joseph N. Straus. 2020.
Relations, and Symmetries: Orthography in Concise Introduction to Tonal Harmony. 2nd ed.
the Music of Nicolai Roslavets.” Music Theory New York: W. W. Norton.
Spectrum 35 (1): 111–​31. Callender, Clifton. 1998. “Voice-​Leading Parsimony
Beach, David. 1994. “Harmony and Linear in the Music of Alexander Scriabin.” Journal of
Progression in Schubert’s Music.” Journal of Music Theory 42 (2): 219–​33.
Music Theory 38 (1): 1–​20. Callender, Clifton. 2004. “Continuous
Benjamin, William E. 1984. “A Theory of Musical Transformations.” Music Theory Online 10 (3).
Meter.” Music Perception 1 (4): 355–​413. Callender, Clifton, Ian Quinn, and Dmitri
Benson, David J. 2007. Music: A Mathematical Tymoczko. 2008. “Generalized Voice-​Leading
Offering. Cambridge, UK: Cambridge Spaces.” Science 320 (5874): 346–​48.
University Press. Capuzzo, Guy. 1999. “Variety within Unity:
Bernard, Jonathan W. 1999. “Ligeti’s Restoration Expressive Means and Their Technical Ends in
of the Interval and Its Significance for His Later the Music of Elliott Carter, 1983–​1994.” Ph.D.
Works.” Music Theory Spectrum 21 (1): 1–​31. dissertation, Eastman School of Music.
Berry, Wallace. [1976] 1987. Structural Functions in Capuzzo, Guy. 2004. “Neo-​Riemannian Theory and
Music. New York: Dover. Original publication, the Analysis of Pop-​Rock Music.” Music Theory
Englewood Cliffs, NJ: Prentice-​Hall. Spectrum 26: 177–​99.
Blackwood, Easley. 1985. The Structure of Carey, Norman, and David Clampitt. 1989. “Aspects
Recognizable Diatonic Tunings. Princeton, NJ: of Well-​Formed Scales.” Music Theory Spectrum
Princeton University Press. 11 (2): 187–​206.

632 • R efere n ces

Carey, Norman, and David Clampitt. 1996. Clough, John. 1998. “A Rudimentary Geometric
“Regions: A Theory of Tonal Spaces in Early Model for Contextual Transposition and
Medieval Treatises.” Journal of Music Theory 40 Inversion.” Journal of Music Theory 42 (2):
(1): 113–​47. 297–​306.
Carpenter, Patricia. 1983. “Grundgestalt as Tonal Clough, John. 2008. “Flip-​Flop Circles and Their
Function.” Music Theory Spectrum 5: 15–​38. Groups.” In Music Theory and Mathematics:
Carter, Nathan C. 2009. Visual Group Theory. Chords, Collections, and Transformations, edited
Washington, DC: Mathematical Association of by Jack Douthett, Martha M. Hyde, and Charles
America. J. Smith, pp. 23–​48. Rochester, NY: University of
Chew, Elaine. 2014. Mathematical and Computational Rochester Press.
Modeling of Tonality: Theory and Applications. New Clough, John, and Jack Douthett. 1991. “Maximally
York: Springer. Even Sets.” Journal of Music Theory 35 (1–​2):
Childs, Adrian P. 1998. “Moving beyond 93–​173.
Neo-​Riemannian Triads: Exploring a Clough, John, Nora Engebretsen, and Jonathan
Transformational Model for Seventh Chords.” Kochavi. 1999. “Scales, Sets, and Interval Cycles:
Journal of Music Theory 42 (2): 181–​93. A Taxonomy.” Music Theory Spectrum 21 (1):
Childs, Adrian P. 2006. “Structural and 74–​104.
Transformational Properties of All-​Interval Clough, John, and Gerald Myerson. 1985. “Variety
Tetrachords.” Music Theory Online 12 (4). and Multiplicity in Diatonic Systems.” Journal of
Chrisman, Richard. 1971. “Identification and Music Theory 29 (2): 249–​70.
Correlation of Pitch-​Sets.” Journal of Music Clough, John, and Gerald Myerson. 1986. “Musical
Theory 15 (1–​2): 58–​83. Scales and the Generalized Circle of Fifths.”
Christensen, Thomas. 1993. Rameau and Musical American Mathematical Monthly 93 (9): 695–​701.
Thought in the Enlightenment. Cambridge, UK: Cogan, Robert, and Pozzi Escot. 1976. Sonic Design:
Cambridge University Press. The Nature of Sound and Music. Englewood Cliffs,
Clampitt, David. 1998. “Alternative Interpretations NJ: Prentice-​Hall.
of Some Measures from Parsifal.” Journal of Cohn, Richard. 1989. Review of David Lewin,
Music Theory 42 (2): 321–​34. Generalized Musical Intervals and Transformations.
Clampitt, David. 2008. “‘Cardinality Equals Variety Journal of Musicological Research 9 (1): 47–​60.
for Chords’ in Well-​Formed Scales, with a Note Cohn, Richard. 1992a. “The Dramatization of
on the Twin Primes Conjecture.” In Music Hypermetric Conflicts in the Scherzo of
Theory and Mathematics: Chords, Collections, Beethoven’s Ninth Symphony.” 19th-​Century
and Transformations, edited by Jack Douthett, Music 15 (3): 188–​206.
Martha M. Hyde, and Charles J. Smith, pp. 9–​22. Cohn, Richard. 1992b. “Transpositional
Rochester, NY: University of Rochester Press. Combination of Beat-​Class Sets in Steve Reich’s
Clampitt, David. 2009. “Webern’s Music for String Phase-​Shifting Music.” Perspectives of New Music
Quartet.” In Intimate Voices: The Twentieth-​ 30 (2): 146–​77.
Century String Quartet, edited by Evan Jones, Cohn, Richard. 1996. “Maximally Smooth Cycles,
vol. 1, pp. 195–​219. Rochester, NY: University of Hexatonic Systems, and the Analysis of Late-​
Rochester Press. Romantic Triadic Progressions.” Music Analysis
Clark, Suzannah. 2011a. Analyzing Schubert. 15 (1): 9–​40.
Cambridge, UK: Cambridge University Press. Cohn, Richard. 1997. “Neo-​Riemannian Operations,
Clark, Suzannah. 2011b. “On the Imagination Parsimonious Trichords, and Their Tonnetz
of Tone in Schubert’s Liedesend (D473), Trost Representations.” Journal of Music Theory 41
(D523), and Gretchens Bitte (D564).” In The (1): 1–​66.
Oxford Handbook of Neo-​Riemannian Music Cohn, Richard. 1998a. “Introduction to Neo-​
Theories, edited by Edward Gollin and Alexander Riemannian Theory: A Survey and a Historical
Rehding, pp. 294–​321. New York: Oxford Perspective.” Journal of Music Theory 42 (2):
University Press. 167–​80.
Clough, John. 1979. “Aspects of Diatonic Sets.” Cohn, Richard. 1998b. “Square Dances with Cubes.”
Journal of Music Theory 23 (1): 45–​61. Journal of Music Theory 42 (2): 283–​96.
Clough, John. 1979–​80. “Diatonic Interval Sets and Cohn, Richard. 1999. “As Wonderful as Star
Transformational Structures.” Perspectives of Clusters: Instruments for Gazing at Tonality in
New Music 18 (1–​2): 461–​82. Schubert.” 19th-​Century Music 22 (3): 213–​32.

R efere n ces • 633

Cohn, Richard. 2000. “Weitzmann’s Regions, My Descartes, René. [1653] 2013. “Compendium
Cycles, and Douthett’s Dancing Cubes.” Music musicae.” Translated by Walter Charleton. In
Theory Spectrum 22 (1): 89–​103. The Compendium Musicae of René Descartes:
Cohn, Richard. 2001. “Complex Hemiolas, Ski-​Hill Early English Responses, edited by Benjamin
Graphs and Metric Spaces.” Music Analysis 20 Wardhaugh, pp. 1–​84. Turnhout, Belgium:
(3): 295–​326. Brepols. Original publication, Utrecht.
Cohn, Richard. 2003. “A Tetrahedral Graph of Dixon, John D., and Brian Mortimer. 1996.
Tetrachordal Voice-​Leading Space.” Music Theory Permutation Groups. New York: Springer.
Online 9 (4). Douthett, Jack. 2008. “Filtered Point-​Symmetry
Cohn, Richard. 2004. “Uncanny Resemblances: and Dynamical Voice-​Leading.” In Music
Tonal Signification in the Freudian Age.” Journal Theory and Mathematics: Chords, Collections,
of the American Musicological Society 57 (2): and Transformations, edited by Jack Douthett,
285–​323. Martha M. Hyde, and Charles J. Smith,
Cohn, Richard. 2011. “Tonal Pitch Space and the pp. 72–​106. Rochester, NY: University of
(Neo-​)Riemannian Tonnetz.” In The Oxford Rochester Press.
Handbook of Neo-​Riemannian Music Theories, Douthett, Jack, and Julian Hook. 2009. “Formal
edited by Edward Gollin and Alexander Rehding, Diatonic Intervallic Notation.” In Mathematics
pp. 322–​48. New York: Oxford University Press. and Computation in Music, Proceedings of the
Cohn, Richard. 2012. Audacious Euphony: Second International Conference of the Society
Chromaticism and the Triad’s Second Nature. New for Mathematics and Computation in Music and
York: Oxford University Press. John Clough Memorial Conference, New Haven,
Cohn, Richard. 2016. “A Platonic Model of Funky CT, June 2009, edited by Elaine Chew, Adrian
Rhythms.” Music Theory Online 22 (2). Childs, and Ching-​Hua Chuan, pp. 104–​14.
Coker, Jerry. 1975. The Jazz Idiom. Englewood Cliffs, Berlin: Springer.
NJ: Prentice-​Hall. Douthett, Jack, and Richard Krantz. 2007.
Conway, John H., Heidi Burgiel, and Chaim “Maximally Even Sets and Configurations:
Goodman-​Strauss. 2008. The Symmetries of Common Threads in Mathematics, Physics, and
Things. Wellesley, MA: A. K. Peters. Music.” Journal of Combinatorial Optimization
Cook, Nicholas. 1996. Review of David Lewin, 14 (4): 385–​410.
Musical Form and Transformation. Music & Letters Douthett, Jack, and Peter Steinbach. 1998.
77 (1): 143–​46. “Parsimonious Graphs: A Study in Parsimony,
Cook, Robert C. 2005. “Parsimony and Extravagance.” Contextual Transformations, and Modes of
Journal of Music Theory 49 (1): 109–​40. Limited Transposition.” Journal of Music Theory
Cook, Scott Alexander. 2009. “Moving through 42 (2): 241–​63.
Triadic Space: An Examination of Bryars’s du Sautoy, Marcus. 2008. Symmetry: A Journey into
Seemingly Haphazard Chord Progressions.” the Patterns of Nature. New York: HarperCollins.
Music Theory Online 15 (1). Dumitrescu, Theodor. 2007. “Constructing a
Dahlhaus, Carl. [1978] 1986. “Sonata Form in Canonic Pitch Spiral: The Case of Salve radix.”
Schubert: The First Movement of the G-​Major In Canons and Canonic Techniques, 14th–​16th
String Quartet, Op. 161 (D. 887).” In Schubert: Centuries: Theory, Practice, and Reception History,
Critical and Analytical Studies, edited by Walter edited by Katelijne Schiltz and Bonnie J.
Frisch, pp. 1–​12. Lincoln: University of Nebraska Blackburn, pp. 141–​70. Leuven: Peeters.
Press. Original publication, Musica 32: 125–​30. Dummit, David S., and Richard M. Foote. 2004.
Davis, Philip J., and Reuben Hersh. 1981. The Abstract Algebra. 3rd ed. Hoboken, NJ: John
Mathematical Experience. Boston: Birkhäuser. Wiley and Sons.
Dawkins, Richard. 1998. Unweaving the Rainbow: Engebretsen, Nora. 2008. “The ‘Over-​Determined’
Science, Delusion and the Appetite for Wonder. Triad as a Source of Discord: Nascent Groups and
Boston: Houghton Mifflin. the Emergent Chromatic Tonality in Nineteenth-​
de la Motte, Diether. [1976] 1991. The Study of Century German Harmonic Theory.” In Music
Harmony: An Historical Perspective. Translated by Theory and Mathematics: Chords, Collections, and
Jeffrey L. Prater. Dubuque, IA: Wm. C. Brown. Transformations, edited by Jack Douthett, Martha
Original publication, Kassel: Bärenreiter. M. Hyde, and Charles J. Smith, pp. 107–​36.
De Souza, Jonathan. 2017. Music at Hand: Rochester, NY: University of Rochester Press.
Instruments, Bodies, and Cognition. New York: Euler, Leonhard. [1739] 1968. Tentamen novae
Oxford University Press. theoriae musicae ex certissimis harmoniae principiis

634 • R efere n ces

 dilucide expositae. New York: Broude. Original International Conference of the Society for
publication, St. Petersburg. Mathematics and Computation in Music, Berlin,
Euler, Leonhard. [1773] 1960. “De harmoniae veris May 2007, edited by Timour Klouche and
principiis per speculum musicum repraesenta- Thomas Noll, pp. 340–​46. Berlin: Springer.
tis.” In Leonhardi Euleri Opera omnia sub auspiciis Grossman, Israel, and Wilhelm Magnus. 1964.
Societatis scientiarum naturalium Helveticae, Groups and Their Graphs. Washington, DC:
Series 3, No. 11 (Zurich: Orell Füssli). Original Mathematical Association of America.
publication, Novi commentarii academiae scien- Hall, Rachel Wells. 2008. “Geometrical Music
tiarum Petropolitanae 18: 330–​53. Theory.” Science 320 (5874): 328–​29.
Fiore, Thomas M., and Ramon Satyendra. 2005. Hall, Rachel Wells, and Dmitri Tymoczko. 2012.
“Generalized Contextual Groups.” Music Theory “Submajorization and the Geometry of
Online 11 (3). Unordered Collections.” American Mathematical
Fitzpatrick, Michael. 2011. “Models and Music Monthly 119 (4): 263–​83.
Theory: Reconsidering David Lewin’s Graph of Harary, Frank. 1969. Graph Theory. Reading, MA:
the 3-​2 Cohn Cycle.” Music Theory Online 17 (1). Addison-​Wesley.
Forte, Allen. 1964. “A Theory of Set-​Complexes for Hardy, G. H., and E. M. Wright. 2008. An
Music.” Journal of Music Theory 8 (2): 136–​83. Introduction to the Theory of Numbers. 6th
Forte, Allen. 1973. The Structure of Atonal Music. ed. Revised by D. R. Heath-​Brown and J. H.
New Haven, CT: Yale University Press. Silverman. Oxford: Oxford University Press.
Forte, Allen. 1988. “Pitch-​Class Set Genera and the Harrison, Daniel. 1988. “Some Group Properties
Origin of Modern Harmonic Species.” Journal of of Triple Counterpoint and Their Influence on
Music Theory 32 (2): 187–​270. Compositions by J. S. Bach.” Journal of Music
Frederick, Leah. 2018. “Diatonic Voice-​Leading Theory 32 (1): 23–​49.
Transformations.” Paper presented at the annual Harrison, Daniel. 1994. Harmonic Function in
meeting of the Society for Music Theory, San Chromatic Music: A Renewed Dualist Theory and an
Antonio, TX. Account of Its Precedents. Chicago: University of
Frederick, Leah. 2019. “Generic (Mod-​7) Voice-​ Chicago Press.
Leading Spaces.” Journal of Music Theory 63 (2): Harrison, Daniel. 2002. “Nonconformist Notions
167–​207. of Nineteenth-​Century Enharmonicism.” Music
Fyr, Kyle. 2011. “Proportion, Temporality, and Analysis 21 (2): 115–​60.
Performance Issues in Piano Works of John Hauptmann, Moritz. 1853. Die Natur der Harmonik
Adams.” Ph.D. dissertation, Indiana University. und der Metrik. Leipzig: Breitkopf und Härtel.
Gamer, Carlton. 1967. “Some Combinatorial Heinichen, Johann David. [1728] 1969. Der General-​
Resources of Equal-​Tempered Systems.” Journal Baß in der Composition. Hildesheim: Georg Olms.
of Music Theory 11 (1): 32–​59. Original publication, Dresden.
Gauldin, Robert. 2004. Harmonic Practice in Tonal Hermann, Richard. 2009. “Becoming Berio:
Music. 2nd ed. New York: W. W. Norton. Evidence from His First Three String Quartets.”
Gillespie, Jeffrey L. 1992. “Motivic Transformations In Intimate Voices: The Twentieth-​Century
and Networks in Schoenberg’s ‘Nacht’ from String Quartet, edited by Evan Jones, vol.
Pierrot lunaire.” Intégral 6: 34–​65. 2, pp. 99–​137. Rochester, NY: University of
Goldenberg, Yosef. 2007. “Schenkerian Voice-​Leading Rochester Press.
and Neo-​Riemannian Operations: Analytical Hisama, Ellie M. 1995. “The Question of Climax
Integration without Theoretical Reconciliation.” in Ruth Crawford’s String Quartet, Mvt. 3.” In
Journal of Schenkerian Studies 2: 65–​84. Concert Music, Rock, and Jazz since 1945: Essays
Gollin, Edward. 1998. “Some Aspects of Three-​ and Analytical Studies, edited by Elizabeth West
Dimensional Tonnetze.” Journal of Music Theory Marvin and Richard Hermann, pp. 285–​312.
42 (2): 195–​206. Rochester, NY: University of Rochester Press.
Gollin, Edward. 2000. “Representations of Hook, Julian. 2002. “Uniform Triadic
Space and Conceptions of Distance in Transformations.” Journal of Music Theory 46
Transformational Music Theories.” Ph.D. disser- (1–​2): 57–​126.
tation, Harvard University. Hook, Julian. 2006. “Exploring Musical Space.”
Gollin, Edward. 2009. “Form, Transformation Science 313 (5783): 49–​50.
and Climax in Ruth Crawford Seeger’s String Hook, Julian. 2007a. “Cross-​Type Transformations
Quartet, Mvmt. 3.” In Mathematics and and the Path Consistency Condition.” Music
Computation in Music, Proceedings of the First Theory Spectrum 29 (1): 1–​39.

R efere n ces • 635

Hook, Julian. 2007b. “David Lewin and the Hyland, Anne M. 2016. “In Search of Liberated
Complexity of the Beautiful” (review of Time, or Schubert’s Quartet in G Major, D. 887:
David Lewin, Generalized Musical Intervals Once More between Sonata and Variation.”
and Transformations and Musical Form and Music Theory Spectrum 38 (1): 85–​108.
Transformation). Intégral 21: 155–​90. Isaacson, Eric J. Forthcoming. Visualizing Music.
Hook, Julian. 2007c. “Enharmonic Systems: Bloomington: Indiana University Press.
A Theory of Key Signatures, Enharmonic Jablonsky, Stephen. 2003. “Introduction aux
Equivalence and Diatonicism.” Journal of graphes de tonalité: Le cercle chromatique des
Mathematics and Music 1 (2): 99–​120. quintes.” Musurgia 10 (3–​4): 91–​104.
Hook, Julian. 2007d. “Why Are There Twenty-​Nine Johnson, Timothy A. 2011. John Adams’s Nixon in
Tetrachords? A Tutorial on Combinatorics and China: Musical Analysis, Historical and Political
Enumeration in Music Theory.” Music Theory Perspectives. Farnham, UK: Ashgate.
Online 13 (4). Johnston, Ben. [1976–​77] 2006. “Rational
Hook, Julian. 2008. “Signature Transformations.” Structure in Music.” In Maximum Clarity and
In Music Theory and Mathematics: Chords, Other Writings on Music, edited by Bob Gilmore,
Collections, and Transformations, edited by Jack pp. 62–​76. Urbana: University of Illinois
Douthett, Martha M. Hyde, and Charles J. Press. Original publication, American Society of
Smith, pp. 137–​60. Rochester, NY: University of University Composers Proceedings 11–​12, 102–​18.
Rochester Press. Junod, Julien, Pierre Audétat, Carlos Agon, and
Hook, Julian. 2011. “Spelled Heptachords.” Moreno Andreatta. 2009. “A Generalisation
In Mathematics and Computation in Music, of Diatonicism and the Discrete Fourier
Proceedings of the Third International Transform as a Mean for Classifying and
Conference of the Society for Mathematics and Characterising Musical Scales.” In Mathematics
Computation in Music, Paris, June 2011, edited and Computation in Music, Proceedings of the
by Carlos Agon, Emmanuel Amiot, Moreno Second International Conference of the Society
Andreatta, and Gerard Assayag, pp. 84–​97. for Mathematics and Computation in Music and
Heidelberg: Springer. John Clough Memorial Conference, New Haven,
Hook, Julian. 2013. “Contemporary Methods in CT, June 2009, edited by Elaine Chew, Adrian
Mathematical Music Theory: A Comparative Childs, and Ching-​Hua Chuan, pp. 166–​79.
Case Study.” Journal of Mathematics and Music 7 Berlin: Springer.
(2): 89–​102. Kaarela, Edwin E. 1992. Letter to the Editor. Journal
Hook, Julian. [2014] 2020. “Generic Sequences and of Music Theory 36 (1): 221.
the Generic Tonnetz.” Journal of Music Theory Karg-​Elert, Sigfrid. [1930] 2004. Akustische Ton-​ Klang-​
64 (1): 63–​103. Original publication, Oxford und Funktionsbestimmung. Paderborn, Germany:
Handbooks Online. Peter Ewers. Original publication, Leipzig.
Hook, Julian. 2015. “Enharmonic Equivalence as Kavanagh, Donncha. 2004. “Ocularcentrism and
an Equivalence Relation.” Paper presented at the Its Others: A Framework for Metatheoretical
annual meeting of the Society for Music Theory, Analysis.” Organization Studies 25 (3): 445–​64.
St. Louis. Keller, Wilhelm. 1976. “Die Modulationen in
Hook, Julian, and Jack Douthett. 2008. “Uniform Mozarts Fantasie KV 475.” In Festschrift Erich
Triadic Transformations and the Twelve-​Tone Valentin, edited by Günther Weiss, pp. 79–​88.
Music of Webern.” Perspectives of New Music 46 Regensburg: Gustav Bosse.
(1): 91–​151. Kellner, David. [1737] 1979. Treulicher Unterricht im
Humphreys, John F. 1996. A Course in Group Theory. General-​Baß. Hildesheim: Georg Olms. Original
New York: Oxford University Press. publication, Hamburg.
Hunter, David J., and Paul T. von Hippel. 2003. “How Klumpenhouwer, Henry. 1991. “Aspects of Row
Rare Is Symmetry in Musical 12-​Tone Rows?” Structure and Harmony in Martino’s Impromptu
American Mathematical Monthly 110 (2): 124–​32. Number 6.” Perspectives of New Music 29 (2):
Hyer, Brian. 1995. “Reimag(in)ing Riemann.” 318–​54.
Journal of Music Theory 39 (1): 101–​38. Klumpenhouwer, Henry. 1994. “Some Remarks on
Hyer, Brian. 2011. “What Is a Function?” In The the Use of Riemann Transformations.” Music
Oxford Handbook of Neo-​Riemannian Music Theory Online 0 (9).
Theories, edited by Edward Gollin and Alexander Klumpenhouwer, Henry. 2006. “In Order to Stay
Rehding, pp. 92–​139. New York: Oxford Asleep as Observers: The Nature and Origins of
University Press. Anti-​Cartesianism in Lewin’s Generalized Musical

636 • R efere n ces

 Intervals and Transformations.” Music Theory Levine, Mark. 1995. The Jazz Theory Book. Petaluma,
Spectrum 28 (2): 277–​89. CA: Sher Music.
Kochavi, Jonathan. 2008. “A Parsimony Metric Lewin, David. 1959. “Re: Intervallic Relations
for Diatonic Sequences.” In Music Theory between Two Collections of Notes.” Journal of
and Mathematics: Chords, Collections, and Music Theory 3 (2): 298–​301.
Transformations, edited by Jack Douthett, Lewin, David. 1960. “Re: The Intervallic Content
Martha M. Hyde, and Charles J. Smith, pp. 174–​ of a Collection of Notes, Intervallic Relations
96. Rochester, NY: University of Rochester Press. between a Collection of Notes and Its
Kolman, Oren. 2004. “Transfer Principles for Complement: An Application to Schoenberg’s
Generalized Interval Systems.” Perspectives of Hexachordal Pieces.” Journal of Music Theory 4
New Music 42 (1): 150–​90. (1): 98–​101.
Kopp, David. 2002. Chromatic Transformations Lewin, David. 1967a. “Moses und Aron: Some
in Nineteenth-​Century Music. Cambridge, UK: General Remarks, and Analytic Notes for Act I,
Cambridge University Press. Scene 1.” Perspectives of New Music 6 (1): 1–​17.
Kostka, Stefan, Dorothy Payne, and Byron Almén. Lewin, David. 1967b. “A Study of Hexachordal
2018. Tonal Harmony: With an Introduction Levels in Schoenberg’s Violin Fantasy.”
to Post-​Tonal Music. 8th ed. New York: Perspectives of New Music 6 (1): 18–​32.
McGraw-​Hill. Lewin, David. 1977. “A Label-​Free Development for
Krebs, Harald. 1999. Fantasy Pieces: Metrical 12-​Pitch-​Class Systems.” Journal of Music Theory
Dissonance in the Music of Robert Schumann. New 21 (1): 29–​48.
York: Oxford University Press. Lewin, David. 1981. “On Harmony and Meter in
Krumhansl, Carol. 1983. “Perceptual Structures for Brahms’s Op. 76, No. 8.” 19th-​Century Music 4
Tonal Music.” Music Perception 1 (1): 28–​62. (3): 261–​65.
Krumhansl, Carol. 1990. Cognitive Foundations Lewin, David. [1984] 2006. “Amfortas’s Prayer to
of Musical Pitch. New York: Oxford Titurel and the Role of D in Parsifal: The Tonal
University Press. Spaces of the Drama and the Enharmonic C♭/​B.”
Krumhansl, Carol. 1998. “Perceived Triad Distance: In Studies in Music with Text, pp. 183–​200. New
Evidence Supporting the Psychological Reality York: Oxford University Press. Original publica-
of Neo-​Riemannian Transformations.” Journal of tion, 19th-​Century Music 7 (3): 336–​49.
Music Theory 42 (2): 265–​81. Lewin, David. [1986] 2006a. “Auf dem Flusse: Image
Laitz, Steven G. 2016. The Complete Musician: and Background in a Schubert Song.” In Studies
An Integrated Approach to Theory, Analysis, in Music with Text, pp. 109–​33. New York:
and Listening. 4th ed. New York: Oxford Oxford University Press. Original publication in
University Press. Schubert: Critical and Analytical Studies, edited by
Lam, Nathan L. 2020. “Modal Spelled Pitch Walter Frisch (Lincoln: University of Nebraska
Class, La-​Minor Solfège, and Schubert’s Third Press), pp. 126–​52.
Relations.” Journal of Music Theory 64 (2): Lewin, David. [1986] 2006b. “Music Theory,
241–​81. Phenomenology, and Modes of Perception.” In
Lay, David C., Stephen R. Lay, and Judi J. Studies in Music with Text, pp. 53–​108. New York:
McDonald. 2016. Linear Algebra and Its Oxford University Press. Original publication,
Applications. 5th ed. Essex, UK: Pearson. Music Perception 3 (4), 327–​92.
Lendvai, Ernő. 1971. Béla Bartók: An Analysis of His Lewin, David. [1987] 2007. Generalized Musical
Music. London: Kahn & Averill. Intervals and Transformations. New York: Oxford
Leong, Daphne. 2007. “Humperdinck and Wagner: University Press. Original publication, New
Metric States, Symmetries, and Systems.” Haven, CT: Yale University Press.
Journal of Music Theory 51 (2): 211–​43. Lewin, David. 1990. “Klumpenhouwer Networks
Lerdahl, Fred. 2001. Tonal Pitch Space. New York: and Some Isographies That Involve Them.” Music
Oxford University Press. Theory Spectrum 12 (1): 83–​120.
Lerdahl, Fred, and Ray Jackendoff. 1983. A Lewin, David. 1991. “Some Problems and Resources
Generative Theory of Tonal Music. Cambridge, MA: of Music Theory.” Journal of Music Theory
MIT Press. Pedagogy 5 (2): 111–​32.
Lester, Joel. 1992. Compositional Theory in the Lewin, David. [1992] 2006. “Some Notes on
Eighteenth Century. Cambridge, MA: Harvard Analyzing Wagner: The Ring and Parsifal.” In
University Press. Studies in Music with Text, pp. 201–​11. New York:

R efere n ces • 637

 Oxford University Press. Original publication, Losada, C. Catherine. 2014. “Complex
19th-​Century Music 16 (1): 49–​58. Multiplication, Structure, and Process: Harmony
Lewin, David. [1993] 2007. Musical Form and and Form in Boulez’s Structures II.” Music Theory
Transformation: Four Analytic Essays. New York: Spectrum 36 (1): 86–​120.
Oxford University Press. Original publication, Mandelbaum, M. Joel. 1961. “Multiple Division of
New Haven, CT: Yale University Press. the Octave and the Tonal Resources of 19-​Tone
Lewin, David. 1994. “A Tutorial on Klumpenhouwer Temperament.” Ph.D. dissertation, Indiana
Networks, Using the Chorale in Schoenberg’s University.
Opus 11, No. 2.” Journal of Music Theory 38 (1): Marvin, Elizabeth West. 1987. “Tonpsychologie and
79–​101. Musikpsychologie: Historical Perspectives on the
Lewin, David. 1995. “Generalized Interval Systems Study of Music Perception.” Theoria 2: 59–​84.
for Babbitt’s Lists, and for Schoenberg’s String Masin, Sergio Cesare, Verena Zudini, and Mauro
Trio.” Music Theory Spectrum 17 (1): 81–​118. Antonelli. 2009. “Early Alternative Derivations
Lewin, David. 1998. “Notes on the Opening of the of Fechner’s Law.” Journal of the History of the
F♯ Minor Fugue from WTC I.” Journal of Music Behavioral Sciences 45 (1): 56–​65.
Theory 42 (2): 235–​39. Mattheson, Johann. [1735] 1980. Kleine General-​
Lewin, David. 2002. “Thoughts on Klumpenhouwer Baß Schule. Laaber, Germany: Laaber-​Verlag.
Networks and Perle-​Lansky Cycles.” Music Theory Original publication, Hamburg.
Spectrum 24 (2): 196–​230. Mazzola, Guerino. 2002. The Topos of Music:
Lewin, David. 2008. “Transformational Geometric Logic of Concepts, Theory, and
Considerations in Schoenberg’s Opus 23, Performance. Basel: Birkhäuser.
Number 3.” In Music Theory and Mathematics: McCreless, Patrick. 1991. “Syntagmatics and
Chords, Collections, and Transformations, edited Paradigmatics: Some Implications for the
by Jack Douthett, Martha M. Hyde, and Charles Analysis of Chromaticism in Tonal Music.” Music
J. Smith, pp. 197–​221. Rochester, NY: University Theory Spectrum 13 (2): 147–​78.
of Rochester Press. Mead, Andrew. 1985. “Large-​Scale Strategy in
Leyton, Michael. 2001. A Generative Theory of Shape. Arnold Schoenberg’s Twelve-​Tone Music.”
Berlin: Springer. Perspectives of New Music 24 (1): 120–​57.
Lind, Stephanie, and John Roeder. 2009. Mead, Andrew. 1988. “Some Implications of
“Transformational Distance and Form in Berg’s the Pitch Class/​O rder Number Isomorphism
‘Schlafend trägt man mich.’” Music Theory Online Inherent in the Twelve-​Tone System: Part
15 (1). One.” Perspectives of New Music 26 (2):
Lindley, Mark. 2001. “Just Intonation.” In The New 96–​163.
Grove Dictionary of Music and Musicians, edited Mead, Andrew. 1989. “Some Implications of the
by Stanley Sadie, 2nd ed., vol. 13, pp. 290–​95. Pitch-​Class/​Order-​Number Isomorphism
London: Macmillan. Inherent in the Twelve-​Tone System: Part Two.
Lippius, Johannes. [1612] 1977. Synopsis of New The Mallalieu Complex: Its Extensions and
Music. Translated by Benito V. Rivera. Colorado Related Rows.” Perspectives of New Music 27 (1):
Springs: Colorado College Music Press. Original 180–​233.
publication, Strasbourg: Karl Kieffer. Mendelson, Elliott. 2015. Introduction to
London, Justin. 2002. “Some Non-​Isomorphisms Mathematical Logic. 6th ed. Boca Raton, FL:
between Pitch and Time.” Journal of Music Theory CRC Press.
46 (1–​2): 127–​51. Messiaen, Olivier. [1944] 1956. The Technique
London, Justin. 2012. Hearing in Time: Psychological of My Musical Language. Translated by John
Aspects of Musical Meter. 2nd ed. New York: Satterfield. Paris: Alphonse Leduc. Original pub-
Oxford University Press. lication, Paris: Alphonse Leduc.
Longuet-​Higgins, H. C. [1962] 1987. “Two Letters Mooney, Michael Kevin. 1996. “The ‘Table of
to a Musical Friend.” In Mental Processes: Studies Relations’ and Music Psychology in Hugo
in Cognitive Science, edited by H. Christopher Riemann’s Harmonic Theory.” Ph.D. dissertation,
Longuet-​Higgins, pp. 64–​81. Cambridge, MA: Columbia University.
MIT Press. Original publication, The Music Morris, Robert D. 1987. Composition with Pitch-​
Review 23 (3): 244–​48, 271–​80. Classes: A Theory of Compositional Design. New
Losada, C. Catherine. 2007. “K-​Nets and Haven, CT: Yale University Press.
Hierarchical Structural Recursion: Further Morris, Robert D. 1991. Class Notes for Atonal Music
Considerations.” Music Theory Online 13 (3). Theory. Lebanon, NH: Frog Peak Music.

638 • R efere n ces

Morris, Robert D. 1993. “New Directions in the Pearce, Trevor. 2008. “Tonal Functions and Active
Theory and Analysis of Musical Contour.” Music Synthesis: Hugo Riemann, German Psychology,
Theory Spectrum 15 (2): 205–​28. and Kantian Epistemology.” Intégral 22: 81–​116.
Morris, Robert D. 1995. “Compositional Spaces and Peck, Robert. 2003. “Klein-​Bottle Tonnetze.” Music
Other Territories.” Perspectives of New Music 33 Theory Online 9 (3).
(1–​2): 328–​58. Peck, Robert. 2004–​05. “Aspects of Recursion in M-​
Morris, Robert D. 1997. “K, Kh, and Beyond.” In Inclusive Networks.” Intégral 18–​19: 25–​70.
Music Theory in Concept and Practice, edited by Peck, Robert. 2010. “Imaginary Transformations.”
James M. Baker, David W. Beach, and Jonathan Journal of Mathematics and Music 4 (3):
W. Bernard, pp. 275–​306. Rochester, NY: 157–​71.
University of Rochester Press. Piston, Walter. 1987. Harmony. 5th ed. Revised and
Morris, Robert D. 1998. “Voice-​Leading Spaces.” expanded by Mark DeVoto. New York: W. W.
Music Theory Spectrum 20 (2): 175–​208. Norton.
Morris, Robert D. 2001. Class Notes for Advanced Plotkin, Richard. 2019. “Chord Proximity,
Atonal Music Theory. 2 vols. Lebanon, NH: Frog Parsimony, and Analysis with Filtered Point-​
Peak Music. Symmetry.” Music Theory Online 25 (2).
Morris, Robert. 2007. “Mathematics and the Plotkin, Richard, and Jack Douthett. 2013.
Twelve-​Tone System: Past, Present, and Future.” “Scalar Context in Musical Models.” Journal of
Perspectives of New Music 45 (2): 76–​107. Mathematics and Music 7 (2): 103–​25.
Morris, Robert D., and Brian Alegant. 1988. “The Pople, Anthony. 2004. “Using Complex Set Theory
Even Partitions in Twelve-​Tone Music.” Music for Tonal Analysis: An Introduction to the
Theory Spectrum 10: 74–​101. Tonalities Project.” Music Analysis 23 (2–​3):
Moseley, Brian. 2018. “Cycles in Webern’s Late 153–​94.
Music.” Journal of Music Theory 62 (2): 165–​204. Quinn, Ian. 2006. “General Equal-​Tempered
Munkres, James. 2018. Topology. 2nd ed. New York: Harmony (Introduction and Part I).” Perspectives
Pearson. of New Music 44 (2): 114–​58.
Murphy, Scott. 2009. “Metric Cubes in Some Music Quinn, Ian. 2007. “General Equal-​Tempered
of Brahms.” Journal of Music Theory 53 (1): 1–​56. Harmony: Parts 2 and 3.” Perspectives of New
Neidhöfer, Christoph. 2005. “A Theory of Harmony Music 45 (1): 4–​63.
and Voice Leading for the Music of Olivier Rahn, John. 1980. Basic Atonal Theory. New York:
Messiaen.” Music Theory Spectrum 27 (1): 1–​34. Schirmer Books.
Newton, Isaac. [1665] 2013. “Isaac Newton’s Ratner, Leonard G. 1970. “Ars combinatoria:
Musical Notes.” In The Compendium Musicae of Chance and Choice in Eighteenth-​Century
René Descartes: Early English Responses, edited by Music.” In Studies in Eighteenth-​Century Music:
Benjamin Wardhaugh, pp. 85–​113. Turnhout, A Tribute to Karl Geiringer on His Seventieth
Belgium: Brepols. Birthday, edited by H. C. Robbins Landon and
Nolan, Catherine. 1995. “Structural Levels and Roger E. Chapman, pp. 343–​63. London: Allen
Twelve-​Tone Music: A Revisionist Analysis of the and Unwin.
Second Movement of Webern’s Piano Variations Reed, Jacob, and Matthew Bain. 2007. “A Tetrahelix
Op. 27.” Journal of Music Theory 39 (1): 47–​76. Animates Bach: Revisualization of David Lewin’s
Ó Searcóid, Mícheál. 2007. Metric Spaces. London: Analysis of the Opening of the F♯ Minor Fugue
Springer. from WTC I.” Music Theory Online 13 (4).
O’Connell, Walter. 1968. “Tone Spaces.” Die Reihe Regener, Eric. 1973. Pitch Notation and Equal
8: 35–​67. Temperament: A Formal Study. Berkeley, CA:
O’Donnell, Shaugn. 1998. “Klumpenhouwer University of California Press.
Networks, Isography, and the Molecular Regener, Eric. 1974. “On Allen Forte’s Theory
Metaphor.” Intégral 12: 53–​80. of Chords.” Perspectives of New Music 13 (1):
Oettingen, Arthur von. 1866. Harmoniesystem 191–​212.
in dualer Entwickelung: Studien zur Theorie der Ricci, Adam. 2002. “A Classification Scheme for
Musik. Leipzig: W. Gläser. Harmonic Sequences.” Theory and Practice
Palisca, Claude V. 1985. Humanism in Italian 27: 1–​36.
Renaissance Musical Thought. New Haven, CT: Riemann, Hugo. 1880. Skizze einer neuen Methode
Yale University Press. der Harmonielehre. Leipzig: Breitkopf und Härtel.
Parks, Richard S. 1989. The Music of Claude Debussy. Riemann, Hugo. 1900. Musik-​Lexikon. 5th ed.
New Haven, CT: Yale University Press. Leipzig: Max Helle.

R efere n ces • 639

Riemann, Hugo. [1914–​15] 1992. “Ideas for a Study Scherzo, Op. 14, No. 2, Measures 1–​32.” Music
‘On the Imagination of Tone’.” Journal of Music Theory Online 15 (1).
Theory 36 (1): 81–​117. Original publication, Roig-​Francolí, Miguel A. 2008. Understanding
Jahrbuch der Musikbiblothek Peters 21: 1–​26. Post-​Tonal Music. New York: McGraw-​Hill.
Riley, Matthew. 2004. “The ‘Harmonic Major’ Mode Rothgeb, John. 1987. Review of Walter Frisch,
in Nineteenth-​Century Theory and Practice.” Brahms and the Principle of Developing Variation
Music Analysis 23 (1): 1–​26. (University of California Press, 1984). Music
Rimsky-​Korsakov, Nikolay. [1886] 1930. Practical Theory Spectrum 9: 204–​15.
Manual of Harmony. Translated by Joseph Russell, Bertrand. 1901. “Recent Work on the
Achron. New York: Carl Fischer. Original publi- Principles of Mathematics.” International
cation, St. Petersburg. Monthly 4: 83–​101.
Rimsky-​Korsakov, Nikolay. [1909] 1942. My Russell, George. 2001. George Russell’s Lydian
Musical Life. Translated by Judah A. Joffe. New Chromatic Concept of Tonal Organization. 4th
York: Alfred A. Knopf. Original publication, St. ed. Vol. 1: The Art and Science of Tonal Gravity.
Petersburg. Brookline, MA: Concept Publishing.
Rimsky-​Korsakov, Nikolay. [1913] 1964. Principles Saariaho, Kaija. 1987. “Timbre and Harmony:
of Orchestration: With Musical Examples Drawn Interpolations of Timbral Structures.”
from His Own Works. Translated by Edward Contemporary Music Review 2 (1): 93–​133.
Agate. New York: Dover. Original publication, St. Sallmen, Mark. 2016. “Transposition Networks and
Petersburg. Network Chains in Schoenberg’s Sechs kleine
Rings, Steven. 2006. Review of David Lewin, Klavierstücke, Op. 19.” Intégral 30: 13–​30.
Generalized Musical Intervals and Transformations, Salzer, Felix, and Carl Schachter. 1969. Counterpoint
Musical Form and Transformation, and Studies in in Composition. New York: Columbia University
Music with Text. Journal of Music Theory 50 (1): Press.
111–​27. Satyendra, Ramon. 1992. “Chromatic Tonality
Rings, Steven. 2007. “Perspectives on Tonality and and Semitonal Relationships in Liszt’s Late
Transformation in Schubert’s Impromptu in E-​ Style.” Ph.D. dissertation, University of
flat, D. 899, No. 2.” Journal of Schenkerian Studies Chicago.
2: 33–​63. Satyendra, Ramon. 2004. “An Informal Introduction
Rings, Steven. 2011a. “Riemannian Analytical to Some Formal Concepts from Lewin’s
Values, Paleo-​and Neo-​.” In The Oxford Handbook Transformational Theory.” Journal of Music
of Neo-​Riemannian Music Theories, edited by Theory 48 (1): 99–​141.
Edward Gollin and Alexander Rehding, pp. 487–​ Schenker, Heinrich. [1926] 1996. “The Organic
511. New York: Oxford University Press. Nature of Fugue as Demonstrated in the C Minor
Rings, Steven. 2011b. Tonality and Transformation. Fugue from Bach’s Well-​Tempered Clavier, Book
New York: Oxford University Press. I.” Translated by Hedi Siegel. In The Masterwork
Robinson, Thomas. 2009. “Pitch-​Class Multisets.” in Music: A Yearbook, vol. II, edited by William
Ph.D. dissertation, City University of New York. Drabkin, pp. 31–​54. Cambridge, UK: Cambridge
Rockwell, Joti. 2009. “Banjo Transformations and University Press. Original publication, Munich:
Bluegrass Rhythm.” Journal of Music Theory 53 Drei Masken Verlag.
(1): 137–​62. Schenker, Heinrich. [1935] 1979. Free Composition.
Roeder, John. 1987. “A Geometric Representation Translated by Ernst Oster. Vol. 2. Hillsdale,
of Pitch-​Class Series.” Perspectives of New Music NY: Pendragon. Original publication, Vienna:
25 (1–​2): 362–​409. Universal Edition.
Roeder, John. 1994. “Voice Leading as Schillinger, Joseph. 1946. The Schillinger System
Transformation.” In Musical Transformation of Musical Composition. 2 vols. New York: Carl
and Musical Intuition: Eleven Essays in Honor Fischer.
of David Lewin, edited by Raphael Atlas and Schoenberg, Arnold. [1947] 1975. “Brahms the
Michael Cherlin, pp. 41–​58. Dedham, MA: Progressive.” In Style and Idea: Selected Writings of
Ovenbird Press. Arnold Schoenberg, edited by Leonard Stein, pp.
Roeder, John. 2003. “Beat-​Class Modulation in 398–​441. Berkeley: University of California Press.
Steve Reich’s Music.” Music Theory Spectrum 25 Schoenberg, Arnold. [1954] 1969. Structural
(2): 275–​304. Functions of Harmony. Rev. ed. New York:
Roeder, John. 2009. “Constructing Transformational W. W. Norton. Original publication, New York:
Signification: Gesture and Agency in Bartók’s W. W. Norton.

640 • R efere n ces

Segall, Christopher. 2014. “Sergei Taneev’s Vertical-​ Temperley, David. 2000. “The Line of Fifths.” Music
Shifting Counterpoint: An Introduction.” Music Analysis 19 (3): 289–​319.
Theory Online 20 (3). Temperley, David. 2001. The Cognition of Basic
Segall, Christopher. 2017. “Alfred Schnittke’s Musical Structures. Cambridge, MA: MIT Press.
Triadic Practice.” Journal of Music Theory 61 (2): Thomassen, Carsten. 1983. “Infinite Graphs.” In
243–​87. Selected Topics in Graph Theory 2, edited by Lowell
Sethares, William A. 2005. Tuning, Timbre, W. Beineke and Robin J. Wilson, pp. 129–​60.
Spectrum, Scale. 2nd ed. London: Springer. London: Academic Press.
Shepard, Roger N. 1982. “Structural Thurston, William P. 1980. The Geometry and
Representations of Musical Pitch.” In The Topology of Three-​Manifolds. Princeton University
Psychology of Music, edited by Diana Deutsch, Lecture Notes. Available at http://​libr​ary.msri.
pp. 343–​90. New York: Academic Press. org/​books/​gt3m/​.
Silverman, Danielle, and Jim Wiseman. 2006. Toussaint, Godfried. 2020. The Geometry of Musical
“Noting the Difference: Musical Scales and Rhythm: What Makes a “Good” Rhythm Good? 2nd
Permutations.” American Mathematical Monthly ed. Boca Raton, FL: CRC Press.
113: 648–​51. Tovey, Donald Francis. 1931. A Companion to “The
Smith, Charles J. 2003. “Functional Fishing with Art of Fugue” (Die Kunst der Fuge). London:
Tonnetz: A Dualist Syntax of Transformations.” Oxford University Press.
Paper presented at the annual meeting of the Tufte, Edward. 2001. The Visual Display of
Society for Music Theory, Madison, WI. Quantitative Information. 2nd ed. Cheshire, CT:
Solomon, Ronald. 2001. “A Brief History of the Graphics Press.
Classification of the Finite Simple Groups.” Tymoczko, Dmitri. 2002. “Stravinsky and the
Bulletin of the American Mathematical Society 38 Octatonic: A Reconsideration.” Music Theory
(3): 315–​52. Spectrum 24 (1): 68–​102.
Spitzer, Michael. 2003. “The Metaphor of Musical Tymoczko, Dmitri. 2004. “Scale Networks and
Space.” Musicae Scientiae 7 (1): 101–​20. Debussy.” Journal of Music Theory 48: 219–​94.
Starr, Daniel. 1978. “Sets, Invariance and Tymoczko, Dmitri. 2006. “The Geometry of Musical
Partitions.” Journal of Music Theory 22 (1): 1–​42. Chords.” Science 313 (5783): 72–​74.
Stoll, Robert R. 1979. Set Theory and Logic. New Tymoczko, Dmitri. 2009a. “Generalizing Musical
York: Dover. Intervals.” Journal of Music Theory 53 (2): 227–​54.
Straus, Joseph N. 2003. “Uniformity, Balance, and Tymoczko, Dmitri. 2009b. “Three Conceptions
Smoothness in Atonal Voice Leading.” Music of Musical Distance.” In Mathematics and
Theory Spectrum 25 (2): 305–​52. Computation in Music, Proceedings of the Second
Straus, Joseph N. 2005a. Introduction to Post-​Tonal International Conference of the Society for
Theory. 3rd ed. Upper Saddle River, NJ: Pearson Mathematics and Computation in Music and
Prentice Hall. John Clough Memorial Conference, New Haven,
Straus, Joseph N. 2005b. “Voice Leading in Set-​ CT, June 2009, edited by Elaine Chew, Adrian
Class Space.” Journal of Music Theory 49 (1): Childs, and Ching-​Hua Chuan, pp. 258–​72.
45–​108. Berlin: Springer.
Straus, Joseph N. 2011. “Contextual-​Inversion Tymoczko, Dmitri. 2011. A Geometry of Music:
Spaces.” Journal of Music Theory 55 (1): 43–​88. Harmony and Counterpoint in the Extended
Straus, Joseph N. 2016. Introduction to Post-​Tonal Common Practice. New York: Oxford
Theory. 4th ed. New York: W. W. Norton. University Press.
Swift, Richard G. 1960. Letter to the Editor. Journal Tymoczko, Dmitri. 2012. “The Generalized
of Music Theory 4 (1): 128. Tonnetz.” Journal of Music Theory 56 (1): 1–​52.
Taneyev, Sergey. [1909] 1962. Convertible Tymoczko, Dmitri. 2013. “Geometry and the
Counterpoint in the Strict Style. Translated by Quest for Theoretical Generality.” Journal of
G. Ackley Brower. Boston: Bruce Humphries. Mathematics and Music 7 (2): 127–​44.
Original publication, Moscow and Leipzig: van der Waerden, B. L. 1985. A History of Algebra: From
Beliaev. Al-​Khwārizmī to Emmy Noether. Berlin: Springer.
Taruskin, Richard. 1985. “Chernomor to Kashchei: Vial, François-​Guillaume. c. 1767. Arbre généalogique
Harmonic Sorcery; or, Stravinsky’s ‘Angle.’” de l’harmonie. Paris.
Journal of the American Musicological Society 38 Vuza, Dan Tudor. 1988. “Some Mathematical
(1): 72–​142. Aspects of David Lewin’s Book Generalized

R efere n ces • 641

 Musical Intervals and Transformations.” Weyl, Hermann. 1952. Symmetry. Princeton, NJ:
Perspectives of New Music 26 (1): 258–​87. Princeton University Press.
Waller, Derek A. 1978. “Some Combinatorial Aspects of White, Arthur T. 1996. “Fabian Stedman: The First
the Musical Chords.” Mathematical Gazette 62: 12–​15. Group Theorist?” American Mathematical Monthly
Wardhaugh, Benjamin. 2008. Music, Experiment and 103: 771–​78.
Mathematics in England, 1653–​1705. Farnham, White, Christopher Wm., and Ian Quinn. 2018.
UK: Ashgate. “Chord Context and Harmonic Function in
Wason, Robert W., and Elizabeth West Marvin. Tonal Music.” Music Theory Spectrum 40 (2):
1992. “Riemann’s ‘Ideen zu einer “Lehre von den 314–​37.
Tonvorstellungen” ’: An Annotated Translation.” Wussing, Hans. 1984. The Genesis of the Abstract
Journal of Music Theory 36 (1): 69–​117. Group Concept: A Contribution to the History
Weber, Gottfried. 1817–​21. Versuch einer geordneten of the Origin of Abstract Group Theory.
Theorie der Tonsetzkunst. Mainz: B. Schott. Translated by Abe Shenitzer. Cambridge, MA:
Weeks, Jeffrey R. 2020. The Shape of Space. 3rd ed. MIT Press.
Boca Raton, FL: CRC Press. Yasser, Joseph. 1932. A Theory of Evolving Tonality.
Weitzmann, Carl Friedrich. 1853. Der übermässige New York: American Library of Musicology.
Dreiklang. Berlin: T. Trautweinschen. Yust, Jason. 2015. “Distorted Continuity:
Werker, Wilhelm. 1898. Die Theorie der Tonalität: Chromatic Harmony, Uniform Sequences,
Ein Beitrag zur Gründung eines konsequenten and Quantized Voice Leadings.” Music Theory
Ton-​ und Musiktheorie-​Systems. Norden, Spectrum 37 (1): 120–​43.
Germany: Braams. Žabka, Marek. 2014. “Dancing with the Scales:
Westergaard, Peter. 1963. “Webern and ‘Total Subchromatic Generated Tone Systems.” Journal
Organization’: An Analysis of the Second of Music Theory 58 (2): 179–​233.
Movement of Piano Variations, Op. 27.” Zbikowski, Lawrence M. 2002. Conceptualizing
Perspectives of New Music 1 (2): 107–​20. Music: Cognitive Structure, Theory, and Analysis.
Westergaard, Peter. 1975. An Introduction to Tonal New York: Oxford University Press.
Theory. New York: W. W. Norton. Zweifel, Paul. 1996. “Generalized Diatonic and
Westergaard, Peter. 1996. “Geometries of Sounds in Pentatonic Scales: A Group-​Theoretic Approach.”
Time.” Music Theory Spectrum 18 (1): 1–​21. Perspectives of New Music 34 (1): 140–​61.

642 • R efere n ces

 Index

Page numbers in italics indicate definitions, primary discussions, or musical examples.

Abel, Niels Henrik, 202n9 algebraic equations (in groups), 179–​80

abelian (group). See group (commutative) Alkan, Valentin
absolute value (|x|), 51 Etude (Op. 39/​10), 34n18
mod-​n (|x|n), 490 all-​interval tetrachord, 357, 362–​64, 382n39
abstract subset relation, 99–​100 alternating group (𝒜n), 213–​14, 628
accidental index Amiot, Emmanuel, xv(n12)
of an enharmonic equivalence vector, 610 angle brackets 〈 〉
of an spc, 69–​71 for generated groups, 181–​83
of an spc set, 584 for UTTs, 301–​02, 334n2
of a spelled heptachord, 585–​86 anti-​automorphism, 247
and sum class (for spelled heptachords), 588 anti-​homomorphism, 247
accidentals anti-​isomorphism, 247, 285–​86, 309, 328, 329,
integer notation for, 585 334–​35n10
nonstandard, 612, 619n48/​50 anti-​uniform (triadic transformation), 327
in spelled heptachords, 585–​86 antisymmetric (binary relation), 98
acoustic scale, 298n37, 587, 616n7 argument (of a function), 46
See also melodic minor scale array (serial), 378–​80, 383–​84n63, 384
action. See group action See also matrix
Adams, John, 335n19, 338 arrow
China Gates, 152, 165n60 for boundary identification, 23–​24
Phrygian Gates, 149–​52, 165n60/​62, 551 in a directed graph, 96–​97, 117n36
additive group. See group in function notation, 45–​46, 74n17
adjacent (vertices), 80 for inflections, 591
admissible move, 494, 497 for intervals vs. transformations, 171
admissible path, 525n17 for involutions, 111, 128, 221, 341
affine group (𝒜ffn), 213–​14, 223, 248–​49n21, 249n22, 381n27, for pivot transformations, 567
628 for straight-​line voice leadings, 433
associative property Balzano, Gerald, 27, 251, 608–​09, 618n42
of addition, 170 banjo pick permutation, 247n5
of composition of functions, 172 Bartók, Béla, 165–​66n71
generalized, 179 Mikrokosmos: No. 109, “From the Island of Bali,” 238;
group axiom, 173 No. 152, Dance in Bulgarian Rhythm, 602; No. 153,
asymmetrical, 236 Dance in Bulgarian Rhythm, 617n25
Audétat, Pierre, 616n10 Music for String Instruments, Percussion, and Celesta, 32, 282
augmented-​sixth chord Beach, David, 157
French, 124, 161–​62n13, 199, 236, 461 beacon. See filtered point-​symmetry
German, 73, 583–​84, 590 beat-​class space (bc(n)), 68, 77n55, 619, 622
augmented triad, 336n25, 478n30, 513 as an interval space, 270
in 3-​voice OP-​space, 442–​43 maximally even sets in, 602
in a 19-​note scale, 613 transposition in, and serial rotation, 383n54
and hexatonic bridge families, 314–​15 beat space (beat), 31, 68, 622
in hexatonic triad cubes, 79–​87, 388–​401 as a distance space, 489
maximal evenness of, 599 as an interval space, 264
notation for, 114n1 product with pitch space (pitch × beat), 264–​65
symmetry of, 236 Beethoven, Ludwig van, 161n11, 336n28
and tonnetzes, 23, 25, 132, 336n27, 404 Quartet, string (Op. 18/​6, B♭), 110, 321, 324–​25
and transformations, 112, 314–​16, 336n27 Sonata, piano (Op. 53, C, “Waldstein”), 157, 166n75/​77
automorphism Sonata, piano (Op. 57, f, “Appassionata”), 161n10, 380n11
vs. anti-​automorphism, 247 Sonata, piano (Op. 111, c), 104
of a group, 239–​40, 352, 550, 579n41 Sonata, violin (Op. 24, F, “Spring”), 133, 134–​35, 162n28,
inner, 246 163n33, 614–​15
of an interval space, 255, 352 Sonata, violin (Op. 30/​3, G), 534–​35
of ℤ7, 240 Symphony No. 1 (C), 413, 420
of ℤ12, 239 Symphony No. 3 (E♭, “Eroica”), 413, 420–​21
automorphism group, 239–​40 Symphony No. 4 (B♭), 264–​65
axiom system (mathematical logic), 381n18 Symphony No. 5 (c), 413
axiomatic definition, 168 Symphony No. 7 (A), 577n8
of distance spaces, 487 Symphony No. 9 (d), 104, 117n49, 133, 135, 162–​63n28,
of groups, 173 413
of interval spaces, 183–​84 bembé rhythm. See diatonic rhythm
axis of inversion Berg, Alban
in pitch-​class space, 20–​21, 161n3, 192–​93, 214 Concerto, violin, 375
in pitch space, 191–​93, 409–​10, 423n41 Lyric Suite, 365
axis of symmetry “Schlafend trägt man mich” (Op. 2/​2), 413
in generic space, 541–​42 Bering Strait flaw, 35n27
in OPTIC spaces, 451–​55 Berio, Luciano
in pitch space and pc space, 237–​38, 548–​49 Quartetto per archi, 75n21
in spc space, 238–​39, 588 Berlioz, Hector
axis of transposition. See line of transposition Symphonie fantastique, 137–​38
bijection, 50
Babbitt, Milton, 167, 201n2, 298n28, 364, 383n54, 579n38 binary operation. See operation
Composition for Four Instruments, 375 binary relation. See relation
Du, 379 bipartite (graph), 84, 129, 396, 461
Babbitt hexachord theorem. See hexachord theorem Blackwood, Easley, 609
Bach, Johann Sebastian, 161n11, 254, 335n19 Twelve Microtonal Etudes (Op. 28), 612, 613, 615–​16,
Chorale No. 356, “Jesu, meine Freunde,” 105–​06 619n50, 620
Fantasia and Fugue (BWV 542, g), 319, 325, 336n33 Boolean algebra, 114n11
Goldberg Variations, 104 Borwick, Leonard, 297n20
Musical Offering, 166n78 boundary
Sinfonia (BWV 789, D), 248n13 false, 23–​24, 432, 458–​59, 473–​74, 476
Sinfonia (BWV 795, f), 217, 248n13 mirror, 427, 435–​36, 441–​42, 451, 453–​55, 473–​74, 476
Well-​Tempered Clavier, Book I: Fugue No. 2 (c), 217, 248n12; See also fundamental region; orbifold; singularity
Fugue No. 14 (f♯), 28, 164n52; Fugue No. 15 (G), 290–​91; bounded (space), 407
Fugue No. 21 (B♭), 215–​17 braces { } (for unordered sets), 39–​40
Well-​Tempered Clavier, Book II: Fugue No. 9 (E), 47–​48; brackets [ ]‌(for normal forms), 416
Prelude No. 20 (a), 336n35 Brahms, Johannes, 117n41, 249n30
Well-​Tempered Clavier, key sequence, 325 Capriccio (Op. 76/​8, C), 104, 117n48
Bach, P. D. Q. Concerto, piano, No. 2 (B♭), 104
Short-​Tempered Clavier, 164n58 Concerto, violin (Op. 77, D), 594
BACH form (tetrachord), 148–​49, 331, 370–​71, 379 Concerto, violin and cello (Op. 102, a), 81, 114n5, 132, 134,
Bailey, Kathryn, 367–​68 135, 136, 354
balanced form Ein deutsches Requiem, 335n23
of a spelled heptachord, 586 Intermezzo (Op. 116/​4, E), 104
of a T-​class, 437, 449, 469, 478n19 Quartet, string (Op. 51/​1, c), 104

644 • Index
 Quartet, string (Op. 51/​2, a), 104 and Lewin’s theorem, 221
Quartet, string (Op. 67, B♭), 104, 117n47 as a transformation network, 221, 308
Quintet, string (Op. 111, G), 104, 602 Cayley’s theorem, 221, 500
Sextet, string (Op. 18, B♭), 597 center (of a group), 246–​47, 295
Symphony No. 1 (c), 291–​92, 294, 548 center of inversion. See axis of inversion
Symphony No. 2 (D), 102–​03 centralizer, 246–​47, 285–​86, 329
Symphony No. 3 (F), 104 centroid, 146–​47
Symphony No. 4 (e), 535–​38, 577n7/​9, 594 chain technique (for distance calculations in C-​space), 522–​23
Trio, clarinet, cello, and piano (Op. 114, a), 594 change ringing, 201n1
Vier ernste Gesänge (Op. 121), No. 1, “Denn es gehet,” 104 chart of the regions (Schoenberg), 127, 162n18
Britten, Benjamin Chausson, Ernest
Billy Budd, Act III, Scene 2, 152–​53, 165n64 Trio, piano (Op. 3, g), 335n19
Peter Grimes, Passacaglia, 322 Chew, Elaine, 146–​47
Brubeck, Dave chicken-​wire torus (Douthett-​Steinbach), 162n22
Blue Rondo à la Turk, 602 Chinese remainder theorem, 69, 71, 77n58, 580n50
Bruch, Max Chopin, Frédéric, 480n58
Scottish Fantasy, 166n75 Etude (Op. 10/​1, C), 597–​98
Bruckner, Anton Etude (Op. 10/​6, e♭), 466, 468
Psalm 150, 398, 402 Etude (Op. 10/​8, F), 466, 468
Symphony No. 9 (d), 594 Etude (Op. 10/​10, A♭), 466, 468
Bryars, Gavin, 338 Fantasy (Op. 49, f), 134, 162–​63n28, 163n29, 313–​14,
bubble notation (in transformation networks), 108–​09, 335n23
118n55/​56, 266, 278, 281–​82, 320–​21, 332–​33, 341, Mazurka (Op. 68/​4, f), 480n60
349–​52, 356, 366–​71, 380n10, 565–​68 Nocturne (Op. 9/​1, b♭), 594
Nocturne (Op. 37/​1, g), 123–​24
C =​V. See cardinality equals variety Nocturne (Op. 62/​1, B), 597
C equivalence. See cardinality equivalence Prelude (Op. 28/​4, e), 466–​67, 480n60, 597, 617n19
C-​scale order (for spelled heptachords), 585 Prelude (Op. 28/​9, E), 124, 161n11, 164n46
C-​space, 412–​14, 439, 520–​24 Preludes (Op. 28), key sequence, 325
C+​ space, C+​normal form, 416 Prelude (Op. 45, c♯), 480–​81n61
calculus, 297n9, 527n43 Sonata, cello (Op. 65, g), 166n75
calendar, 77n51, 617n20 Sonata, piano, No. 3 (Op. 58, b), 415
Callender, Clifton, xiv, 388, 423n45, 424, 479n33/​34, 481 chord genus, 554, 568–​77
canonical scale system, 606, 610 in non-​canonical scale systems, 606
cardinality chord quality. See chord species
of a design, 570 chord-​scale theory, 616n14
of a genus, 570 chord species (quality), 554, 568–​77
infinite, 40, 73n4 in non-​canonical scale systems, 606
of a multiset, 45 and OPT-​classes, 410, 414–​15
vs. order (of a group), 177 and OPTC-​classes, 413–​15, 554, 569
of an ordered set, 43 and spelled heptachords, 587, 589
of a set, 40 chordal core (Lerdahl), 141–​42, 164n44
of a species, 570 chordal-​regional space (Lerdahl), 136, 142, 153–​54, 164n45/​
cardinality equals variety, 564, 570–​76, 580n57, 581, 618n38 46, 166
for chords, 570, 606–​07, 618n37 chordal space (Lerdahl), 141–​42
in enharmonic systems, 611, 612 chroma circle, 13
for lines, 570, 606–​07 chromatic design. See design
in non-​canonical scale systems, 606–​07, 618n37 chromatic fifths space. See fifths space
and proper spelled heptachords, 592–​93 chromatic heptachord (0123456), 547–​48, 550, 590, 607,
cardinality equivalence (~C), 65, 405–​06, 412–​14, 520–​24 618n38
anomalous behavior of, 413–​14, 520–​24 chromatic inflection. See inflection
in OPTIC spaces, 439, 459–​60, 476, 520–​24 chromatic pitch space. See pitch space
See also C-​space chromatic universe, 606
Carey, Norman, 550, 581 circle of fifths, 18–​21, 121–​24, 127, 161n8, 223
Carter, Elliott in a 19-​note scale, 612, 615
Scrivo in vento, 382n39 chromatic vs. diatonic vs. generic, 545–​50
String Quartet No. 2, 75n21 generalized, 607, 608
Carter, Nathan, 251 See also fifths space
Cartesian product, 43–​45 circle-​of-​fifths distance, 122, 136–​37, 494, 514
of more than two sets, 44–​45 circle of thirds
See also direct product in a 19-​note scale, 613
Castel, Louis-​Bertrand, 36n37 and functional harmony, 142–​43
category theory, 380–​81n12 generic, 22, 142–​43, 538–​41
Cayley, Arthur, 248n17 See also thirds space
Cayley diagram, 218–​24, 248n17, 307–​09, 364 city-​block metric, 501
and distance functions, 497–​500 Clampitt, David, 383n49, 550, 581
and label functions, 254 class (vs. set), 73n1

Index • 645
closure (group property), 170, 173, 175, 189 connected (graph), 81
Clough, John, xiv, 163n41, 337n53, 531, 542, 569–​70, See also strongly connected
580n53, 581, 599, 619 consistency (of an axiom system), 381n18
cluster cycle, 599 consistency properties (of transformation graphs). See path
coarse (equivalence relation), 63, 77n46 consistency; realizable
codomain (of a function), 75n26 consonant triad space (triad), 119–​24, 625
Cohn, Richard, xiii, 313, 315, 335n19, 338 as a commutative interval space, 231–​32, 301, 325, 351–​53
collection, 73n1 distance functions on, 136–​38, 494–​95, 512–​14
See also set as a noncommutative interval space, 199, 232, 254, 261,
combination (vs. permutation), 477n9, 617n30 353
comma. See syntonic comma; Pythagorean comma See also neo-​Riemannian triad space; Weber space
common-​tone distance, 496, 505, 514, 526n36 constant (vs. variable), 74–​75n20
common-​tone graph, 92 contents (of a vertex), 80
common tones, 25–​26, 146 contextual inversion, 120–​21, 161n3, 306, 310
and interval-​class vectors, 549, 579n39 of twelve-​tone rows, 125
between triads, 55, 129–​30, 496, 526n36, 538 continuous function, 47, 75n22
commutative diagram. See diagram continuous pitch-​class space (cpc), 13–​15, 622
commutative group. See group as a distance space, 490–​91
commutative interval space. See interval space as an interval space, 188, 270
commutative property See also pitch-​class–​tuple space
of addition, 170 continuous pitch space (cpitch), 4–​7, 174, 228, 249n31, 622
of binary operations, 172, 380–​81n12 as a distance space, 488–​89
of functions, 53–​54 as an interval space, 186
in groups, 173, 202n8 See also frequency space; pitch-​tuple space
of transposition groups, 172 continuous space, 6–​7, 34n7, 389–​90, 578n22
See also group (commutative); interval space (commutative) continuous voice leading. See voice leading
commutator, 246–​47 contour class, 451
commutator subgroup, 246–​47 contour theory (and OPTIC spaces), 479n35
complement (of a set), 41, 273–​74, 599 contrapuntal configuration, 215
in generic space, 540–​42 control group (in a wreath product), 305
complete bipartite graph, 116n29 Cook, Robert, 335n18, 344, 380n9
complete generic design, 575, 592 coordinate axis subgroup, 249–​50n32
complete graph (Kn), 94–​95, 116n29, 163n40, 628 coordinate system
K5 in neo-​Riemannian tonnetz, 94–​95 in pitch-​tuple space, 388–​89, 421n3, 427, 447–​48, 460, 469
K7 and diatonic tonnetz, 138–​39, 534 for twelve-​tone rows, 365–​70, 382–​83n47
completeness theorem coordinate transformation (for twelve-​tone rows), 367–​70
for first-​order languages, 381n18 coordinates (of an ordered tuple), 43
for transformation graphs, 347 coprime, 69, 230
complex multiplication (in serial theory), 384 coset, 240–​45, 268–​69, 406
complex numbers, 202n13 right vs. left, 244–​45
components cosines, law of, 481n65, 527n51
of a graph, 82 counterpoint
of an ordered tuple, 43 and permutations, 215–​17, 248n10/​13
componentwise, 229, 264 CQT spaces. See OPTIC spaces
composite equivalence relation. See equivalence relation Crawford (Seeger), Ruth
composition of functions, 52–​54, 171–​73 Quartet, string, 248n10
associativity of, 172 cross section
as a binary operation, 56, 171–​73 of OP-​space, 440–​41, 463–​65
for interval-​preserving and interval-​reversing mappings, of a product space, 267
283–​84, 293–​94 cross-​type transformation, 277–​82, 290–​93, 332–​33, 342,
for interval-​space homomorphisms, 261–​63 380n6
orthography for. See orthography crossing (in an spc set), 584
for permutations, 211–​12 crystallography, 202n5
for transpositions and inversions, 193–​95, 282–​84, 293–​94 cube, 4-​dimensional. See hypercube
cone, 479n44 Cube Dance (Douthett), 315–​16, 336n26, 400–​02, 422n22,
configuration space (filtered point-​symmetry), 605, 618n35 480n51, 495
conformance, 19, 27–​28, 36n44, 37n59, 74n13 cube distance, 495, 514
conformed tonnetz. See conformance; tonnetz (neo-​Riemannian) cubic octave, 407, 422n28
congruence class, 66, 269 cubic semitone, 422n28
congruence mod H (subgroup), 242 cycle (graph), 81–​82
congruence mod n, 65–​66, 240–​42 directed vs. undirected, 97, 346–​47
congruence subgroup, 269 cycle (permutation), 210–​17, 422–​23n30
conjoin, 13 cycle type (of a permutation), 210
conjugacy class, 246 cyclic graph, 81–​82
conjugate cyclic group (𝒞n), 181–​82, 214, 224–​26, 628
of a function, 245–​46, 350, 358, 376 Cayley diagram for, 218–​19, 221–​22
of a group element, 180, 245–​46, 251n47 direct product of, 229–​31

646 • Index
 group table for, 217–​18 dilation, 493
infinite (𝒞∞), 182, 228, 628 dimension, 14, 35n25, 74n14
cyclic permutation, 211–​12 and cardinality equivalence, 520–​24
cylinder, 16–​17, 24, 35n30, 146–​47 diminished seventh chord, 124
in a 19-​note scale, 613
Dahlhaus, Carl, 157–​58, 160 maximal evenness of, 599
de la Motte, Diether, 76n32 in octatonic seventh-​chord hypercubes, 460–​62
de Morgan’s laws, 41, 114n11 symmetry of, 235
Debussy, Claude, 161n11, 166n78, 580n53 in a tonnetz, 23, 25–​26
Images I, No. 3, “Mouvement,” 282 diminished triad, 336n27
Prélude à “L’après-​midi d’un faune,” 265–​67, 275, 277, direct product
297n20, 336n35, 345 of groups, 229–​33, 250n36
Sonata, violin, 565–​68 of interval spaces, 263–​68
deep (interval property), 549–​50, 606, 607 See also Cartesian product
degree of symmetry, 236 direct transformation. See transformation
degree of a vertex, 80 directed graph, 96–​106
design, 551–​53, 569–​70 distances in, 493
chromatic, 551–​53 underlying, 109–​10
complete, 575 directionally adjusted (transformational product), 346
diatonic, 551–​52 disconnected (graph), 81–​82
enharmonic, 551–​52 discrete pitch-​class space. See pitch-​class space
generic, 551–​53, 583, 590 discrete pitch space. See pitch space
inflections of, 590–​97 discrete space, 6–​7, 34n7
reduction to a chord, 570 discrete voice leading. See voice leading
spc, 551–​53, 590 disjoint (sets), 40
diagram (commutative vs. noncommutative), 380–​81n12, 579n33 displacement dissonance, 117n45
diatonic bell (Audétat), 616n10 displacement multiset, 501, 603
diatonic collection distance
in a 19-​note scale, 612 between key areas, 136–​38
arithmetical properties of, 608 as shortest path length, 490, 497–​98, 500, 506–​08
and chromatic collection, 545–​50 between triads, 136–​38, 512–​14
maximal evenness of, 599–​601 as an undirected “interval,” 489
in OP-​space, 602–​04 between vertices in a graph, 81, 493
recognizable (Blackwood), 619n50 distance function, 487–​89
as a spelled heptachord (dpc(n)), 585–​86 vs. interval function, 184, 264, 324, 488–​89
symmetry of, 27, 37n54, 236, 291–​92, 548–​49 metric compatibility of, 489, 491, 503
See also diatonic pitch-​class space in a product space, 500–​05
diatonic design. See design in a quotient space, 506–​24
diatonic fifths space (dfifth(n)), 21, 545–​47, 576–​77, 622 and similarity measures, 525n25
diatonic hexachord, 550, 607 See also distance space; specific distance functions by name
diatonic interpretation (δn), 547, 590 (e.g., neo-​Riemannian distance)
diatonic interval. See interval distance-​preserving mapping, 163n38/​39
diatonic mode. See mode See also isometry
diatonic pitch-​class space (dpc(n)), 16, 532–​50, 622 distance space, 264, 487–​524, 528
as a distance space, 491–​92, 532 axioms for, 487–​89
as an interval space, 269, 532 and topology, 524n13
as a spelled heptachord, 585–​86 See also distance function
See also enharmonic diatonic pitch-​class space distribution constraint (for displacement multisets), 526n31
diatonic pitch space (dpitch(n)), 9–​10, 622 distributive laws (in set theory), 41
as a distance space, 489–​90, 532 divisibility lattice, 182
as an interval space, 257, 532 division (in multiplicative groups), 178
diatonic rhythm, 602 dodecahedron, 115n12
diatonic system (Agmon), 609, 618n41 domain (of a function), 46
diatonic tonnetz. See tonnetz, diatonic dominant transformation (D), 122, 161n6, 322,
diatonic triad space, 138–​43 336n37/​38
dice games, 247n5 double-​circle space, 119–​27
diesis, 162n20, 163n33, 272 as a Cayley diagram, 222
difference (of sets), 42 and direct products, 231–​32
difference tone, 189 for generic chords, 539–​41
digraph. See directed graph for twelve-​tone rows, 376–​78
dihedral group (𝒟n), 192–​95, 204n39, 213–​14, 226, 232, for UTTs and QTTs, 305, 327
248n19, 628 double harmonic scale (dharm(n)), 587–​88, 616n7, 622
automorphisms and isomorphisms of, 309, 328–​29, 354–​55, doubly transitive (group action), 199, 360
382n33 Douthett, Jack, 315, 336n26, 382n45, 480n51, 599, 600, 605,
Cayley diagram for, 221–​22, 307–​09 619
generalized (𝒟ℝ), 411 dual graph, 85–​86, 130, 162n24
infinite (𝒟∞), 192–​95, 628 dual-​UTT, 330

Index • 647
duality, 114–​15n11 Euclidean space (ℝn), 35n25, 264, 390, 405, 426, 628
of inversion operators (Invuv , Invvu), 294–​95 Euler, Leonhard, 27, 37n55
in mathematics, 114–​15n11 See also phi function
of permutations, 383n56 Euler characteristic, 95–​96, 116n32
in physics, 114–​15n11 of a sphere, 95
of pitch-​class and order numbers, 371–​80, 383n53/​55/​56/​57, of a torus, 95–​96
384 See also Euler’s formula
of QTTs, 330 Euler’s formula
Riemannian, 114–​15n11, 302–​03, 334n3/​8, 335n13, 366 for connected planar graphs, 85, 95
of TI group and Riemann group, 329–​30, 337n55/​59 for disconnected graphs, 85
of transpositions and interval-​preserving mappings, 284–​86, for toroidal graphs, 95–​96, 126, 128, 129, 138
298n39, 329–​30, 374, 383n57, 422–​23n30 See also Euler characteristic
Dukas, Paul even permutation, 212–​13
Sonata, piano (e♭), 594 event network, 113–​14, 118n61, 216–​17, 341–​42, 370
Dvořák, Antonín, 117n41 exchange transformation (X)
Quartet, piano, No. 2 (Op. 87, E♭), 165n70, 595–​96 order/​pitch-​class exchange, 375–​80
Symphony No. 7 (d), 110, 466–​68 triad/​seventh-​chord exchange, 333, 337–​38n62
exponential function, 51–​52
eccentricity, 517, 527n46 exponentiation (and groups), 178
edge (in a graph), 80 extraction (of a sequence), 535
multiple edges, 92, 115–​16n20
edge labeling, 80 face
EE vector. See enharmonic equivalence vector of a cube, 83–​84
embedding, 14, 84–​85 of a hypercube, 460
planar, 84–​85 of a planar graph, 85–​86, 114n9
in a product space, 267 facet (of a hypercube), 87, 460
spherical, 85–​86 factorial (n!), 43
toroidal, 95–​96, 126, 138–​39, 143–​44, 163n20 false boundary. See boundary
empty set, 40 family, 73n1
empty word, 525n21 See also set
Enchiriadis treatises, 34n10 Fauré, Gabriel
enharmonic coordinates, 556 Barcarolle No. 4 (Op. 44, A♭), 396–​98, 422n19
enharmonic design. See design Pavane (Op. 50), 141, 164n43
enharmonic diatonic pitch-​class space (edpc(n)), 546–​49, 556, 622 Quartet, piano, No. 1 (Op. 15, c), 105
enharmonic diesis, 162n20, 163n33, 272 Quartet, string (Op. 121, e), 422n19, 424
enharmonic doubling (in an spc set), 584 Fechner, Gustav Theodor, 75n28
enharmonic equivalence (~E), 19, 34n17, 61–​62, 234, 260, Fechner’s law, 75n28
545–​46 fiber group (in a wreath product), 305
of diatonic designs, 555 field (of spcs), 590
as an equivalence relation, 61–​62 fifth-​distance vector, 549
in non-​canonical scale systems, 610–​16 fifth-​string
of proper spelled heptachords, 588–​89 cyclical, 575
strong, 555, 595 of a diatonic set, 572–​74
weak, 597 as a normal form, 572, 580n58
enharmonic equivalence vector, 610 of an spc set, 572, 583
enharmonic respelling (εk), 595 and species, 574
enharmonic system, 610–​16, 619n50, 620, 623 fifths space (fifth), 18–​19, 623
enneatonic collection, 42, 86, 274, 314, 623 chromatic vs. diatonic vs. generic, 545–​50
maximal evenness of, 599 as a cyclic graph, 81–​82
enneatonic seventh-​chord graph, 86, 92 as a distance space, 492
enumeration techniques, 74n12, 247n5, 250n38, 382n42, as an interval space, 188, 269–​70
477n9, 578n15 vs. pitch-​class space, 18–​19, 188, 260, 532–​33
EPTIC relations (for spelled pitch-​class tuples), 580n58 See also diatonic fifths space; generic fifths space
equal temperament, 34n17, 57, 609 filter (spelled heptachord), 590
equality (as an equivalence relation), 63 filtered point-​symmetry (Douthett), 600–​01, 604–​05, 618n35, 619
equivalence class, 60 fine (equivalence relation), 63, 77n46
equivalence relation, 60–​65 fix, 235
composite, 61–​62, 408–​15, 423n32 fixed-​0 row labeling, 365
induced by a normal subgroup, 269 fixed point, 191–​93, 210, 288, 427
normal, 269, 297n25, 486 See also invariance
and transformation groups, 233–​34 Forte, Allen, 73n2, 77n45, 117n39, 202n5, 580n53
Euclid, 33n5 Fourier properties, xv(n12), 424, 525n25
Euclidean distance Franck, César, 335n19
as a measure of evenness, 602–​03 Quintet, piano (f), 133, 162–​63n28, 344
in a product space, 501–​05 Symphony (d), 594
in a quotient space, 508, 514–​20 Frederick, Leah, 578n21
between triads, 513–​14 free (group action), 204n44

648 • Index
French augmented-​sixth chord. See augmented-​sixth chord generic thirds space (gthird), 22, 240, 532–​33, 538–​41, 623
frequency, 5–​6 as a cyclic graph, 81–​82
frequency-​class space (freq ⁄~O), 271–​72 as a distance space, 492, 532
frequency space (freq), 8, 178, 623 as an interval space, 492, 532
as an interval space, 188–​89, 196–​97 generic tonnetz. See tonnetz
vs. pitch space, 51–​52, 189, 228, 249n31, 258–​59, 263 generic transposition. See transposition
quotients of, 271–​72 generic triad space (gtriad), 138, 154–​55, 538–​41, 623
function (mapping), 45–​58, 74n18 generic voice-​leading distance, 540
common notations, 45–​46 genus, 554, 568–​77, 579n44, 580n53, 581
definition by cases, 75n27 geometry (vs. topology), 6, 14–​15
equality of functions, 46 German augmented-​sixth chord. See augmented-​sixth chord
graph of. See graph Gesualdo, Carlo, 335n19
vs. harmonic function, 74n18 Ginastera, Alberto
of multiple variables, 56 Danzas argentinas No. 1, 280–​82, 284, 292
vs. relation, 58–​59 Glazunov, Alexander
of subsets, 55 Symphony No. 3 (Op. 33, D), 319, 336n31
functional harmony, 96–​97, 116–​17n33 glissando (as a voice leading), 389
and the diatonic circle of thirds, 142–​43, 164n48, 323–​24 Gödel, Kurt, 381n18
See also harmonic function Gogol, Nikolai, 331
fundamental region, 15, 24, 407 Gollin, Edward, 145
boundary points of, 422n29 Górecki, Henryk
vs. normal region, 428 Symphony No. 3, 580n55
fundamental theorem of finite abelian groups, 249n26 Gounod, Charles
fuse function (for pitch tuples), 412, 521–​24 Faust, 596–​97
and voice leading, 423n45 graph (of a function), 47–​48
fuzzy transposition and inversion, 76n38, 421n1 graph (in graph theory), 79–​96
alternate definitions of “graph,” 115n20, 116n26
Galeazzi, Francesco, 526n36 as a distance space, 493–​97
Galois, Évariste, 167 infinite, 92–​94
Gamer, Carlton, 579n38 See also directed graph; transformation graph
generalized interval system (Lewin), 203n26 greatest integer function (⌊x⌋), 51, 509
See also interval space greatest lower bound, 100, 509
generated set, 541–​42, 548–​50, 606, 607, 611–​12, 617n24 Grieg, Edvard, 335n19
generated subgroup, 181 “Illusion” (Op. 57/​3), 478n30
generating interval, 18, 22, 541–​42, 548–​50, 606 Sonata, violin (Op. 45, c), 279–​80
generating set (for a group), 182–​83, 219–​20, 498–​500 group, 167–​68, 173–​83, 201n1/​2
generator (of a group), 176, 180–​83, 203n25, 218–​20, 230–​31 additive, 174–​76
generic, 8–​10, 37–​38n61, 601 axioms for, 173
vs. diatonic, 8–​10, 532 classification of, 225–​27, 249n26/​27
vs. specific, 34n9, 545, 569, 580n53, 583, 601, 606 commutative (abelian), 173, 202n9, 242–​44, 249n26
generic aggregate, 534, 555, 583, 585 modular, 176–​77
generic design. See design multiplicative, 177–​78
generic equivalence (~G), 9, 19, 34n11, 62, 65, 272, 532, 545–​46 noncommutative, 243–​47
as an equivalence relation, 62 trivial, 176
generic fifths space (gfifth), 21, 240, 532–​33, 545–​47, 623 See also interval group; symmetry; transformation group;
and C =​V, 575 specific groups by name (e.g., dihedral group)
as a cyclic graph, 81–​82 group action, 189, 203–​04n35, 406
as a distance space, 492, 532 group congruence, 242, 297n25
as an interval space, 492, 532 Group Explorer (web application), 251
generic interval. See interval group operation, 173–​74
generic interval class. See interval class group table, 217–​18, 248n16
generic inversion. See inversion grouping dissonance, 117n45
generic pitch class, 15–​16 Gypsy scale. See double harmonic scale
generic pitch-​class numbers, 15–​16
generic pitch-​class set, 541–​43 half-​line, 34n6
generic pitch-​class space (gpc), 15–​16, 240, 532–​47, 623 Hamilton, William Rowan, 114n6
as a cyclic graph, 81–​82 Hamiltonian cycle, 114n6
as a distance space, 491–​92, 532 Hanon, Charles-​Louis, 163n41
as an interval space, 188, 257, 269, 532 harmonic function, 36n41, 74n18, 142–​43, 164n48
as a specific scale, 604–​05 See also functional harmony
generic pitch space (gpitch), 8–​11, 623 harmonic major scale (hmaj(n)), 587, 616n7, 623
as a distance space, 489, 532 in OP-​space, 603–​04
as an interval space, 187, 256–​57, 532 harmonic minor scale (hmin(n)), 280, 413, 424n47, 599–​601,
isomorphism with pitch, 10, 92, 187, 256–​57 611, 623
generic projection. See projection as a generic scale, 607
generic set class, 541–​43 in OP-​space, 603–​04
generic seventh-​chord space (gsev), 539–​41, 623 as a spelled heptachord, 585–​87, 590–​92

Index • 649
Harrison, Daniel, 248n10 image
Hasse diagram, 117n38 of an element, 46
Hauptmann, Moritz, 36n41, 114–​15n11, 335n13/​22 of a function, 75n26
Haydn, Joseph, 336n28 of a subset, 55
Symphony No. 94 (G, “Surprise”), 413 in-​degree (of a vertex), 96–​97
Symphony No. 98 (B♭), 163n32 incidence function (in a graph), 82
Hegel, Georg Wilhelm Friedrich, 114–​15n11 incident (vertex and edge), 80
Heinichen, Johann David, 161n8 inclusion lattice, 100, 117n42
helix (pitch space), 12–​13 inclusion mapping, 259, 278
hemiola, 102 independent equations, 392
hereditary (graph properties), 347 index (of inversion), 190–​91, 204n37
hexachord theorem, 273–​74, 298n28, 578n19 index (of a subgroup), 241–​42
hexatonic bridge family, 314–​15, 336n27, 354 indirect transformation. See transformation
hexatonic bridge group, 315, 318–​19 induced. See equivalence relation; homomorphism; quotient
hexatonic collection, 25–​26, 80, 132, 162n26, 310–​12, 314–​ interval space
15, 394, 623 inflection (of generic designs) (X ∕ H), 590–​97
as a generic scale, 601–​02, 607 injection. See one-​to-​one (function)
hexatonic cycle. See PL-​cycle inner automorphism. See automorphism
hexatonic pole, 92, 122, 310–​12, 320–​21, 335n18/​19, 503 instrumental space, 203n30
hexatonic system, 63, 80, 114n2, 118 integers (ℤ), 40, 629
hexatonic triad cube, 82–​87, 233, 388–​94, 518 as an additive group, 173–​74
hexatonic triad cube chain, 394–​402 integers mod n (ℤn), 66–​69, 629
in 3-​voice OP-​space, 442, 513, 603–​04 as a group, 176–​77
and octatonic hypercube chain, 462–​63 integral (calculus), 297n9
and scale lattices, 603–​04 internally deep (interval property), 607, 613
tonnetz in, 403–​04 interscalar mapping, 256, 278–​82, 290–​93
twisted, 400 intersection (of sets), 40–​41
See also Cube Dance interval
hexatonic triad family, 314–​15, 336n27, 354, 370, 396 diatonic, 9–​10, 34n15, 611–​12, 618–​19n43
hexatonic triad graph, 79–​87, 233, 388 directed vs. undirected, 168–​69, 483, 489, 524n1/​10
hexatonic triad group, 223–​24, 237, 249n29, 312, 314–​15, as an equivalence class, 487, 524n5
318–​19 generic, 9–​10, 541–​44, 547, 601, 606
Holst, Gustav in an interval space, 183
The Planets, “Venus,” 594 inverse, 169, 185
homeomorphism, 478n18 and label functions, 253
homogeneity specific, 9–​10, 601, 606
distance-​space property, 493, 503, 509 vs. transformation, 195–​201, 204n50
group property, 176, 202n12 See also interval function; interval space
interval-​space axiom, 183–​84, 484 interval class, 116n22, 490–​91, 524n1
homometric, 202n5 in 2-​voice OP-​space, 431–​32
homomorphism as a distance, 490–​91
vs. anti-​homomorphism, 247 generic, 541–​44, 571, 578n16
of groups, 224–​25 in non-​canonical scale systems, 606
induced, 258 interval-​class space, 91, 116n22, 437–​39
of interval spaces, 254–​63, 296n4 interval-​class vector, 274, 298n29/​30, 549–​50, 579n39
of transformation graphs and networks, 381n26 generic, 543, 578n16
Humperdinck, Engelbert in non-​canonical scale systems, 618n40
Hänsel und Gretel, 249n30 interval-​compatible (mapping), 255–​58
Hungarian minor scale. See double harmonic scale interval configuration, 169–​70, 185
Hyer, Brian, 161n6, 326 interval-​content equivalence (of pitch-​class sets), 63, 250n38
Hyer group (𝒰even), 326–​27, 337n49 interval cycle, 81–​82, 542, 600
hyper-​Tk and hyper-​Ik, 362–​64, 382n38 See also generating interval
hypercube, 87, 460, 480n49 interval equations, solving, 185–​86
hypermeter, 102 interval function, 168–​69, 183–​89
hypernetwork. See network of networks vs. distance function. See distance function
hyperprism, 465, 480n57 inadequacy as a distance measurement, 483–​87
and “size,” 483, 524n2
I-​space, 426–​27 interval group, 170, 183–​89
I+​ space, I+​normal form, 416 vs. transformation group, 171–​73
icosahedron, 115n12 interval-​preserving mapping, 169, 257, 297n13
identify, 13 and transposition, 276–​86, 298n39, 329–​30, 351, 374
identity element (of a group), 170, 173, 175, 179, 202n7 interval-​reversing mapping, 169, 286–​87, 289–​96
group axiom, 173 impossibility in noncommutative interval spaces, 295–​96
vs. identity function, 173 and inversion, 285–​86, 289, 293–​96
identity function, 53, 75n31, 172–​73, 190 interval space, 174, 183–​89, 203n26
identity interval property, 184–​85 axioms for, 183–​84, 299
if and only if, 76n39 commutative, 184

650 • Index
 direct product, 263–​68 Ising model, 617n20
and group structure, 195–​201, 221–​22, 228, 252–​54, isography (of transformation networks), 109–​10, 333, 353–​
268–​69 57, 381n20
label functions for, 252–​54 of K-​nets, 358–​63
mappings of, 254–​63, 273–​96 negative, 359
modular, 188, 490–​91 positive, 358–​59
noncommutative, 199, 254, 285, 286–​87, 294–​96 strong, 109–​10, 112, 118n57, 341, 348–​51, 357, 359,
quotient, 268–​73, 486 381n20
real, 186, 483, 489–​90 isometric (distance spaces), 492
interval-​space homomorphism theorem, 228, 258, 297n8 isometry (of distance spaces), 492
interval-​string notation, 578n16 vs. isomorphism, 493
interval subspace property, 186–​87 isomorphism, 6
generalized, 243 vs. anti-​isomorphism, 247
interval-​sum equation of directed graphs, 98, 117n35, 348, 381n21
in continuous pitch space, 174 generalized definition, 90
in fifths space, 188 of graphs, 87–​90
interval-​space axiom, 184 of groups, 171, 224–​228, 348, 352–​55
in pitch-​class space, 188 of interval spaces, 187, 255–​63, 348, 351
in pitch space, 168–​69 strong (for transformation networks), 348–​51, 367, 381n20
vs. triangle inequality, 488 topological, 478n18
intervallically distinguishable, 237, 548, 575, 581 of transformation graphs, 348, 353–​57, 381n20/​21
invariance, 149, 233–​37 of transformation networks, 109–​10, 347–​57, 367–​70,
See also fixed point; matrix; symmetry 381n22
inverse isomorphism class 225–​26
group axiom, 173
of a group element, 170, 173–​75, 179 Jablonsky, Stephen, 123
of a group element vs. of a function, 173, 190 Joplin, Scott
vs. inversion, 204n36 Maple Leaf Rag, 602
inverse function (f−1), 50–​52, 190 Journal of Mathematics and Music, ix
inverse image (of a subset), 55–​56 just pitch-​class (frequency-​class) space (just ⁄~O), 272
inverse interval property, 185, 488 just pitch (frequency) space (just), 57–​58, 76n37, 268,
inverse mapping (f(x) =​ x−1), 247, 259, 289, 297n11 297n24, 272, 623
inversion just tuning, 37n57, 57–​58
conditions 1 and 2 for, 287, 294–​95
contextual (vs. fixed). See contextual inversion K-​graph, 358, 362
cross-​type “inversion,” 290–​93, 299n45 K-​net. See Klumpenhouwer network
diatonic vs. chromatic, 291–​92 K relation (Forte), 117n39
dual inversions (Invuv , Invvu), 294–​95 Kellner, David, 121–​22, 161n8
generic (in), 538 kernel (of a homomorphism), 244–​45, 307
in an interval space (Inv[u]i, Invuv ), 286–​96, 299 key area, 81, 119–​23, 141–​43
vs. inverse, 204n36 relationships among, 18, 121–​23, 127–​28, 136–​38, 162n18,
as an isometry, 492 494, 549
as a network isomorphism, 349–​50, 352–​53, 356 vs. triad, 135, 503
non-​uniformity of, 301, 310, 327 key signature, 8–​9, 11–​12, 20, 608
in noncommutative interval spaces, 286, 294–​96 in enharmonic systems, 611–​12, 615
and permutation, 438 and signature transformations, 550–​51, 560
in pitch-​class space (Ik), 54, 191–​95 and spelled heptachords, 583, 586, 590
in pitch space, 190–​95 See also diatonic interpretation; signature transformation
in spc space (ɩ), 584–​85, 588–​89 key space (key), 122–​23, 623
and transposition, 189–​95, 293 Klein bottle, 38n64/​66
as a triadic transformation (QTT), 327–​29 Klein group (𝒱4), 226–​27, 229, 240, 364, 629
of twelve-​tone rows, 125, 364–​66, 372–​75 Klumpenhouwer, Henry, 118n57, 357, 381n20
inversion sign (in a twelve-​tone row label), 369 Klumpenhouwer network (K-​net), 352, 357–​64, 381n20, 384
inversional equivalence (~I), 405–​06, 410–​12, 426–​27 consistency properties, 360–​61, 382n36
axis for, 423n41 and paired transpositions, 357–​58, 360–​61
in PT-​space, 451–​53 recursion, 362–​64, 382n41
symmetry group for, 411, 486 Kopp, David, 326–​27, 337n52
vs. TI equivalence, 411, 477n3 Krebs, Harald, 117n45
See also I-​space; TI equivalence Krumhansl, Carol, 127–​28, 136–​37
inversional symmetry, 27, 124, 236–​37 Kuratowski, Kazimierz, 114n8
in generic space, 541–​42
in OPTIC spaces, 451–​53, 479n37 L′ (nebenverwandt), 312–​15, 335n20/​22/​23, 336n28, 354, 455
in spc space, 584 label-​compatible (mapping), 256
involution label consistency (for transformation networks), 107, 340
arrow conventions for, 111, 128, 221, 341 and path consistency, 118n56, 344
function, 51, 202n16 label function, 56–​57
group element, 179, 202n16, 203n18 for an interval space, 252–​54, 296n1, 269, 382–​83n47

Index • 651
label-​preserving mapping, 276, 289 LP-​cycle. See PL-​cycle
label-​reversing mapping, 289 LR-​chain, 127, 130, 322–​24, 332–​33, 334n9, 335n23, 404
Laitz, Steven G., 141 and the circle of fifths, 127
Lam, Nathan, 36n38, 579n35 vs. M-​chain, 322–​24
lattice, 99–​105, 117n38, 182, 419–​20 LR distance, 499, 514
lattice point (in a continuous space), 389 Lutosławski, Witold
least upper bound, 100 Jeux vénitiens, 75n21
left-​to-​right orthography. See orthography Musique funèbre, 29–​31, 38n66, 123, 152, 365
leittonwechsel (L), 36n43, 54–​55, 76n32, 336n37
in 3-​voice OPTI-​space, 455 macroharmony (Tymoczko), 590
in non-​canonical scale systems, 609, 614–​15 Mahler, Gustav, 335n19
vs. relative, mediant, and submediant, 132, 322–​24 manifold, 14, 24, 426, 476n1, 477n4
as a UTT, 301–​02, 310 mapping. See function
Lendvai, Ernő, 165–​66n71 Martin, Frank
length Prelude No. 2, 320–​21, 336n35, 356
of a path (in a graph), 81, 497 Mathematics and Computation in Music (conferences), ix
of a path (in an OPTIC space), 506–​24 Mathieu group, 249n27
of a word, 498 matrix
length function, 525n17 invariance, 384
length space, 525n17 permutation, 209–​11, 371–​72, 383n52, 383–​84n63
Lerdahl, Fred, 35n29, 127, 130, 137, 141, 162n18, 164n44 twelve-​tone, 218, 378, 383n52/​62, 383–​84n63
See also chordal-​regional space See also array (serial)
letter-​distinct (spc set), 584 Mattheson, Johann, 161n8
letter doubling (in an spc set), 584 max distance
letter space. See generic pitch-​class space in a product space, 501–​05
Lewin, David, ix, xiii, xv(n1/​6), 28, 73n1, 157, 166, 205, 209, in a quotient space, 508–​09, 516–​18
297n24 between triads, 513–​14
and group theory, 167, 248n10, 298n39 maximally even set, 597–​606, 617n20, 619
and interval spaces (generalized interval systems), 185, in beat-​class space, 602, 617n24
203n26, 265, 294–​96, 296n1, 297n25, 298–​99n40, 299 in enharmonic systems, 611–​12
and intervals vs. transformations, 171, 183, 195–​201 in non-​canonical scale systems, 607
and inversion, 294–​95 in OP-​space, 603–​06, 617n26
and transformation graphs and networks, 106, 113, second-​order, 605
118n61, 339, 343–​45, 380n1/​2/​3/​8/​11, 381n14, 384 maximally smooth cycle, 310–​11, 335n17, 608, 614, 616
and triadic transformations, 161n6, 322, 336n37/​38, 338 mediant group (ℳ), 324, 336–​37n42, 628
and twelve-​tone theory, 126, 162n16, 364 mediant transformation (M), 132, 162n27, 322–​24, 336n37/​
Lewinian interval system (Tymoczko), 524n9 39
Lewin’s theorem, 199–​201, 221, 228, 254, 285, 324, 329, 345 vs. leittonwechsel, relative, and submediant, 132, 322–​24
lexicographic ordering, 229 M-​chain, 322–​24, 336n40
Lie group, 423n35 melodic minor scale, 587, 599–​601, 616n5/​7, 624
Ligeti, Győrgy, 75n21 in OP-​space, 603–​04
line. See ordered set as a spelled heptachord (mmin(n)), 587
line genus, 554, 568–​77 Mendelssohn, Felix
in non-​canonical scale systems, 606 Song Without Words (Op. 38/​3, E), 105
line of fifths. See spelled pitch-​class space Messiaen, Olivier
line of transposition, 392–​93, 409–​10, 427–​28, 432, 442–​44, modes of limited transposition, 250n41, 274, 298n31
447–​49 non-​retrogradable rhythms, 250n41
line species, 414, 554, 568–​77 Quatre études de rythme, “Île de feu 2,” 249n27
in non-​canonical scale systems, 606 metric, 487
linear algebra, 203n27, 424 See also distance function
linear equation, 391–​92, 421n5/​6/​10, 424 metric dissonance, 117n45
Liszt, Franz, 336n28 metric space, 487, 528
Années de pèlerinage, troisième année, No. 2, “Aux cyprès,” See also distance space
467–​68 metric state (in ski-​hill lattice), 101–​05, 112, 117n46, 227–​28,
Concerto, piano, No. 2 (A), 166n75 249n30
Consolation No. 3 (D♭), 133, 135, 162n28, 614 metrically compatible (interval and distance functions), 489,
Dante Symphony, 321, 336n36 491, 503
Etudes d’exécution transcendante, No. 8, “Wilde Jagd,” 110, microtonal scales, 606–​16
320 microtone, 35n28
Grande fantaisie symphonique über Themen aus Berlioz’ “Lélio,” minor seventh chord, 86–​87, 124, 461, 464
319, 336n31 mirror boundary. See boundary
Valse oubliée No. 1, 467–​68, 595, 616n15 Möbius hyperprism, 465–​68, 480n57
log-​frequency space, 75n28 Möbius prism, 441–​42
logarithm, 51–​52, 75n29 Möbius strip, 28–​30, 38n64, 151–​52
London, Justin, 118n59 as an 016 tonnetz, 28–​30
loop (in a graph), 90–​91, 115–​16n20 as a generic tonnetz, 143, 164n49
Losada, C. Catherine, 382n41 topology of 2-​voice OP-​space, 431–​32, 543

652 • Index
mode neo-​Riemannian triad space (nrtriad), 128–​36, 341–​42, 624
diatonic, 122–​23, 560, 600–​01 alternate representations of, 143–​44, 164n50, 403–​04
of a genus, 575 as a distance space, 494–​95, 497–​98
of a pitch-​class set, 237 as dual graph of a triangulated tonnetz, 130–​31
of a spelled heptachord, 590–​92 and the Riemann group, 306
mode-​preserving (triadic transformation), 133, 302 vs. Weber space, 128–​29, 162n25, 163n35
mode-​reversing (triadic transformation), 133, 302 See also tonnetz
mode space, 122–​23, 161n9, 233, 497 neo-​Riemannian voice-​leading distance, 495, 514
model (of an axiom system), 381n18 network. See transformation network
modular arithmetic, 65–​69, 77n51, 77n52 network of networks, 118n55, 362–​64, 381n22
modular group. See group nineteen-​note scale, 612–​16, 619n51/​52/​53, 620
modular interval space. See interval space diatonic behavior in, 613
modulo, 60, 65 interval sizes in, 613
monochord, 33n5 node. See vertex
monster group, 249n27 node/​arrow system (Lewin), 117n34
Monteverdi, Claudio, 335n19 See also directed graph
Morris, Robert, xv(n6), 248–​49n21, 312, 335n20 Nolan, Catherine, 367–​68
movable-​0 row labeling, 365 non-​C, non-​I, non-​O, non-​P, non-​T, 416
Mozart, Wolfgang, 161n11 Nono, Luigi
Don Giovanni, Overture, 597 Il canto sospeso, 365
Fantasy (K. 475, c), 34–​35n18 norm, 525n26
Quartet, piano (K. 493, E♭), 343, 380n7 normal equivalence relation, 269, 297n25, 486
Requiem (K. 626), Confutatis, 319 normal form (in OPTIC spaces), 415–​21
Sinfonia concertante (K. 364, E♭), 166n75 algorithm for, 416–​18, 424n52/​53
Sonata, piano (K. 331, A), 613, 614 bracket notation [ ]‌, 416
Sonata, piano (K. 533/​494, F), 597 generic (mod-​7), 532, 577n10, 578n17
Symphony No. 39 (K. 543, E♭), 163n32 hierarchy, 419–​20
Symphony No. 40 (K. 550, g), 538 lattice, 419–​20
Symphony No. 41 (K. 551, C, “Jupiter”), 248n10, 413 of an spc set, 572, 580n58
multigraph, 115n20 normal order (OPC normal form of a pc set), 415, 417–​18
multiplication group (ℳn), 213–​14, 628 normal region, 428
multiplication operator vs. fundamental region, 428
in a group (Multa), 221, 248n18 normal subgroup, 244–​45, 255, 268–​69, 297n25, 486
and K-​nets, 359, 382n40 note-​name space, 8
left vs. right, 248n18 number theory, 78
as a network isomorphism, 352–​53, 381n26
in pitch-​class space (Mk), 195, 210–​14, 223 O-​class, 406–​07, 554
in serial theory, 248–​49n21, 373–​74, 384 O equivalence. See octave equivalence
multiplicative group. See group O-​space
multiplicity (of an element in a multiset), 45 1-​voice, 427
multiset, 45, 74n15 2-​voice, 426, 428–​30, 506–​08
in OPTIC spaces, 399, 402, 408–​09, 413–​15, 430 3-​voice, 270, 407, 426, 517–​19
reduction to set, 45 4-​voice, 520
Murphy, Scott, 163n36 as a distance space, 506–​08, 517–​18
musica ficta, 34n8 homogeneity of, 509
musical space. See space as an interval space, 270, 485–​86
Myerson, Gerald, 569–​70, 580n53, 581 as a manifold, 426
Myhill, John, 580n56 O+​ space, O+​normal form, 416
Myhill’s property, 571, 601, 606, 611, 612 obverse (of L, P, R). See L′; P′; R′
OC-​space, 423n44
Neapolitan major scale (nmaj(n)), 587–​88, 616n6, 624 octahedron, 86
Neapolitan minor scale (nmin(n)), 587, 616n6, 624 octatonic bridge family, 315
near-​symmetry, 164n51, 519 octatonic bridge group, 315, 318–​19
nearly even (chord), 396, 442–​45, 466, 474 octatonic collection, 25–​26, 42, 132, 315, 317–​18, 336n29,
nebenverwandt. See L′ 460, 462, 624
negative isography. See isography as a generic scale, 601, 607
neighborhood (in tonnetz), 166n73 as an interval space, 275, 279–​80
neighborhood (topology), 14, 426 maximal evenness of, 599, 601
neo-​Riemannian (vs. Riemannian), 37n58, 55, 162n23, as a pitch space, 11
303 as an spc set, 584–​85
neo-​Riemannian analysis, 132–​36, 162n25, 309–​22, 338 octatonic seventh-​chord graph, 86, 92
See also Riemann group; schritt; uniform triadic octatonic seventh-​chord hypercube, 115n14, 460–​62
transformation; wechsel; specific transformations by name octatonic seventh-​chord hypercube chain, 462–​68,
(e.g., parallel) 520–​21
neo-​Riemannian distance, 136–​37, 163n35, 403, 494, 499, octatonic triad family, 315
514 octatonic triad graph, 86, 92
neo-​Riemannian tonnetz. See tonnetz octatonic triad group, 315, 318–​19, 325, 337n48

Index • 653
octave equivalence (~O), 12–​16, 19, 342–​43, 405–​07 OPTI equivalence, 411–​12
congruence subgroup for, 269, 406 symmetry group for, 412
as an equivalence relation, 60–​62, 269 OPTI-​space
in frequency space, 271 2-​voice, 437–​39
fundamental regions for, 407, 427–​29 3-​voice, 404, 412, 453–​58, 479n44, 484, 516–​19, 545
in a hexatonic triad cube chain, 398–​99 4-​voice, 474–​76, 481n63, 519–​20
normal region for, 428–​29 area and volume calculations in, 519–​20
in PTI-​space, 453–​55 asymmetry of, 484, 517
and quotient groups, 240–​42 boundaries of, 437–​39, 454–​55, 476, 479n43
symmetry group for, 406, 428, 486 as a distance space, 516–​20
See also O-​space fundamental regions for, 454–​55, 475–​76
ocularcentrism, 33n1 generic (mod-​7), 545
odd permutation, 212–​13 inhomogeneity of, 484, 517
Oettingen, Arthur von, 27–​28, 37n56, 335n13, 579n37 isometries in, 517
OI-​space, 1-​voice, 427, 439 normal region for, 454–​55
OI symmetry, 453, 479n38 voice leading in, 455–​58, 476
omnibus progression, 464–​65, 473–​74 OPTI symmetry, 455–​56
one-​to-​one (function), 48 OPTIC relations, 65, 404–​21
one-​to-​one correspondence, 50 OPTIC-​class, 413
onto (function), 48 and prime form, 415
OP-​class, 408–​09 OPTIC spaces, 387–​481, 500–​24
OP equivalence, 408–​09 area and volume calculations in, 518–​20
OP-​space, 408–​09 as distance spaces, 500–​24
2-​voice (Möbius strip), 429–​37, 485, 509–​12, 543–​44 elements of, 414
3-​voice (Möbius prism), 399, 402, 439–​47, 460, 478n22/​26, generic (mod-​7), 543–​45, 578n21
512–​14, 517–​19, 523–​24, 544–​45 inhomogeneity of, 484
4-​voice (Möbius hyperprism), 463–​68, 520–​21 and interval spaces, 484–​87
5-​voice, 521 names for, 388, 414, 424n51
7-​voice, 602–​04, 617n29 voice leading in, 432–​33
in a 19-​note scale, 614 See also specific spaces by name (e.g., OP-​space)
area and volume calculations in, 519–​20 optimize (in normal form calculation), 416
boundaries of, 431–​32, 435–​36, 440–​42, 463, 510–​12 orbifold, 426–​27, 476n1, 477n4/​10, 484
cross section of, 440–​41, 463–​65 orbifold spaces. See OPTIC spaces
as a distance space, 509–​14 orbit (OrbG(x)), 149, 217, 233–​36, 364, 406
fundamental regions for, 429–​32 orbit-​stabilizer theorem, 235–​36, 250n39, 364, 408
generic (mod-​7), 543–​44 order
inhomogeneity of, 485, 512 of a group, 177
label functions for, 429–​30, 440 of a group element, 181–​82
normal region for, 430–​31, 477n7 order numbers (for twelve-​tone rows), 364, 371–​80
as an orbifold, 432, 277n10 order/​pitch-​class exchange transformation (X), 375–​80
voice leading in, 434–​37, 445–​47, 463–​68, 509–​13 order relation (≤), 99, 117n37
OP symmetry, 455–​56 order transformation (vs. pitch-​class transformation), 233,
OPC-​class, 412–​13 372–​80, 383n53/​55/​56/​57
generic (mod-​7), 554 ordered pair, 43–​44
and normal order, 415 ordered set, 43–​45, 405
OPC equivalence, 412–​13 orthography (for composition of functions), 52–​54, 75n30,
OPC-​space, 460, 520–​23 203–​04n35, 248n18
generic (mod-​7), 577n13 OT-​class
operation, 45, 56, 74n16 mod-​7 (line genus), 554
See also function mod-​12 (line species), 554
operator. See function OT-​space
OPI-​space, 439 2-​voice, 438
OPT equivalence, 410 as an interval space, 270, 297–​98n26, 486
OPT-​space OTI-​space, 437–​39
2-​voice, 437–​39 out-​degree (of a vertex), 96–​97
3-​voice, 458–​59, 516–​18
4-​voice, 470–​74, 519–​20 P-​class, 407
area and volume calculations in, 519–​20 P equivalence. See permutational equivalence
boundaries of, 458–​59, 471–​74, 480n47, 516 p-​norm metric, 526n30
as a distance space, 516–​20 P-​space, 408
fundamental regions for, 458–​59, 470, 479–​80n46 P+​ space, P+​normal form, 416
normal region for, 458–​59 P′ (slide), 312–​13, 335n20/​21, 455, 595
voice leading in, 458–​59, 472–​74 Pachelbel, Johann
OPTC-​class Canon (D), 140
mod-​7 (chord genus), 554 packed to the left (in normal form calculation), 415, 418,
mod-​12 (chord species), 554 424n53
OPTC-​space, 459–​60, 480n47 paired paths. See path

654 • Index
pairwise commuting groups, 232, 285–​86, 328–​30, 373–​74, pitch-​class space (pc), 13–​15, 624
377–​78 chromatic vs. diatonic vs. generic, 545–​50
paradigmatics, 118n61 as a cyclic graph, 81–​82
parallel transformation (P), 54–​55, 76n32, 336n37 as a distance space, 490–​91
in 3-​voice OPTI-​space, 455 vs. fifths space, 18–​19, 188, 260
in non-​canonical scale systems, 609, 614–​15 as an interval space, 188, 268–​69, 299
as a UTT, 301–​02, 310 See also continuous pitch-​class space; diatonic pitch-​class
parallel voice leading. See voice leading space; generic pitch-​class space; spelled pitch-​class space
parity, 212 pitch-​class transformation (vs. order transformation), 233,
Parks, Richard, 580n53 372–​80, 383n53/​55/​56/​57
parsimonious (voice leading), 526n35 pitch-​class–​tuple space (pcn, cpcn), 264, 270, 407, 414, 622
See also voice leading pitch multiset, 408, 414
Pärt, Arvo pitch numbers, 4
Cantus in Memory of Benjamin Britten, 105 pitch-​pair space. See pitch-​tuple space (2-​voice)
partial order, 98–​99 pitch retention loop, 166n73
partition of the aggregate, 384 pitch set, 408, 414
path (in a graph), 80–​81, 114n3 pitch space (pitch), 4–​8, 624
directed, 97 as a distance space, 488–​90
length of, 81, 497 as an interval space, 168–​70, 185–​86
paired (directed), 343 label functions for, 253–​54
paired (undirected), 346 as a quotient set (srl ⁄~E), 61–​62, 273
path (in OPTIC spaces), 388–​94, 426–​27, 432–​37, 445–​47, See also continuous pitch space; diatonic pitch space;
451–​53, 456–​59, 466–​68, 472–​74, 497–​98, 506–​13, frequency space; generic pitch space
520–​21 pitch-​triple space. See pitch-​tuple space (3-​voice)
path-​connected, 34n7 pitch tuple, 65, 264, 405, 414, 422n26
path consistency (for transformation graphs and networks), pitch-​tuple space (pitchn, cpitchn), 264, 405, 414, 622
118n56, 317–​18, 343–​47, 380n8/​9/​10/​11, 380–​81n12, 2-​voice, 264, 427–​29
381n14 3-​voice, 405
of K-​nets, 360–​61, 382n36 as a distance space, 500–​05
and label consistency, 118n56, 344 as an interval space, 264, 406, 485
strong, 346–​47 as a manifold, 426
pentatonic collection (pent(n)), 280–​82, 292–​93, 550, 607, voice leading in, 432–​33
624 pivot interval (Rings), 266, 275
indexing convention for, 280, 298n36 pivot transformation, 567
maximal evenness of, 599 PL-​cycle, 81, 165–​66n71, 223–​24, 249n29, 335n19, 455, 565,
as a pitch space, 11 580n51
perfectly even, 599–​600 in a 19-​note scale, 614
periodic, 15, 24 in a tonnetz, 130, 132–​33, 163n30, 310–​12, 404
permutation, 43, 209–​17, 247n3/​5, 248n10/​13, 326, 337–​ as a transformation graph, 111–​12, 345, 354
38n62, 407–​08 PL group. See hexatonic triad group
vs. combination, 617n30 planar (graph), 84–​85, 114n8, 116n29
cycle representation, 210–​17, 407–​08, 422–​23n30 planing, 280–​82
of elements vs. of order positions, 383n56, 422–​23n30 Platonic solids, 86, 115n12
even vs. odd, 212–​13 Plotkin, Richard, 600, 605
and inversion, 438 PLR-​cycle, 134, 156, 312–​13, 354–​55, 404
matrix, 209–​11, 371–​72, 383n52, 383–​84n63 PLR distance. See neo-​Riemannian distance
of triads, 305 PLR group. See Riemann group
and twelve-​tone rows, 371–​72 position-​finding (in a diatonic collection), 548
permutation group, 209–​17 positive isography. See isography
permutational equivalence (~P), 65, 213, 390, 405–​06, 407–​09 positivity (distance-​space axiom), 487–​88
in a hexatonic triad cube chain, 398–​99 power notation (in groups), 181, 203n20, 203n21
symmetry group for, 407–​08, 430, 486 Power Towers (Douthett), 480n51
in T-​space, 450–​51 PR-​cycle, 127, 130, 165–​66n71, 315, 404, 564
See also P-​space Prater, Jeffrey, 76n32
permutational symmetry, 451, 455–​56 precedence relation (in a directed graph), 98–​99
Petrushka chord, 124, 161–​62n13, 235, 237, 250n42 predecessor (in a directed graph), 98
phi function (φ), 178, 223 Presuv. See interval-​preserving mapping
pitch class, 13 prime form (OPTIC normal form of a pc set), 415–​16, 455
See also generic pitch class; spelled pitch class prime numbers, 73n3
pitch-​class multiset as generic/​diatonic cardinality, 68, 532–​33, 542,
and OP-​classes, 408–​09, 414 576, 608
pitch-​class numbers, 13 as group orders, 227
generic, 15–​16 and modular arithmetic, 68, 178
pitch-​class set as specific/​chromatic cardinality, 613, 615–​16
generic, 541–​43 principal homogeneous space, 203n29
and OP-​classes, 409, 414–​15 product (of group elements), 173–​74
and OPC-​classes, 412–​15 product space. See Cartesian product; direct product

Index • 655
projection R′, 312–​13, 335n20, 455
in gpc space (γ), 69–​73, 545–​47, 579n33, 583 Rachmaninoff, Sergei
in pc space (π), 69–​73, 545–​47, 579n33, 583 “Daisies” (Op. 38/​3), 594
from a product space, 63, 267 Rhapsody on a Theme of Paganini, 105
projective geometry, 114n11 Sonata, cello (Op. 19, g), 319
Prokofiev, Sergei Rameau, Jean-​Philippe, 36n37/​41, 539
Cinderella, 165n67 range (of a function), 48–​50, 75n26
Scherzo (Op. 12/​10, a), 320, 564 rational numbers (ℚ), 40, 628
prolongation (vs. transformation), 105–​06, 117n50/​51 as an additive group, 174
promiscuity (of K-​nets), 359–​60 multiplicative groups within, 177
proof by contradiction, 94–​95 Ravel, Maurice, 336n29
proper spc set, 584 Gaspard de la nuit, I, “Ondine,” 316–​17, 354
proper spelled heptachord, 588–​90 Menuet antique, 583, 591
proper subset, 40 Miroirs, No. 4, “Alborada del gracioso,” 602
pseudograph, 115n20 Pavane pour une infante défunte, 141
pseudometric, 526–​27n38 Quartet, string (F), 594
PT-​space Sonatine, 316–​18, 345
2-​voice, 438 real interval space. See interval space
3-​voice, 450–​52, 516–​18 real numbers (ℝ), 40, 628
4-​voice, 470 as an additive group, 174
boundaries of, 451–​52 extended, 524n14
as a distance space, 516–​18 multiplicative groups within, 177–​78
fundamental regions for, 450–​51 real numbers mod n (ℝn), 69, 628
normal region for, 450–​51 as an additive group, 177
voice leading in, 451–​52, 455–​58 realizable (transformation graph), 345–​47
PTI-​space universally, 346–​47
2-​voice, 438 realize
3-​voice, 452–​53, 516–​18 diatonic design realizes a genus, 554, 569–​70
boundaries of, 453 diatonic design realizes a species, 569–​70
as a distance space, 516–​18 species realizes a genus, 569–​70
fundamental regions for, 452 transformation graph realizes a directed graph, 109–​10
normal region for, 452 transformation network realizes a transformation graph,
voice leading in, 452–​53, 455–​58 109–​10, 341
PTI symmetry, 455–​56 reference point (for a label function), 253, 296n1
Puccini, Giacomo, 335n19 reflection factor (for a QTT), 327–​28
punctured plane, 35n30 reflexive (binary relation), 64
Pythagorean comma, 37n59, 74n13, 76n36, 271 Regener, Eric, 77–​78n60
Pythagorean pitch-​class (frequency-​class) space (pyth ⁄~O), Reger, Max
271–​72 Sonata, clarinet (Op. 49/​2, f♯), 34n18
Pythagorean pitch (frequency) space (pyth), 57, 76n36, 267–​68, regional space (Lerdahl), 127, 162n18
496–​97, 624 registral letter space. See generic pitch space
Pythagorean semitone, 271 regular (enharmonic equivalence vector), 611
Pythagorean tuning, 37n57, 57 regular (graph), 80
regular (polyhedron), 86
QTT. See quasi-​uniform triadic transformation Reich, Steve, 383n54
qualia, tonal, 265 Clapping Music, 602
quality. See chord species relation
quantization, 600 binary, 58–​65
quarter-​tones, 608, 609, 611, 618–​19n43 vs. function, 58–​59
quasi-​uniform triadic transformation (QTT), 327–​30, 337n53, ternary, 60, 106
367 unary, 60
duality, 330 relative transformation (R), 54–​55, 76n32, 336n37
group structure, 328–​30, 628 in 3-​voice OPTI-​space, 455
serial applications, 375–​78 vs. leittonwechsel, mediant, and submediant, 132, 322–​24
Quinn, Ian, xiv, xv(n12), 388, 424, 481n69, 481, 525n25, 580n53 in non-​canonical scale systems, 609, 614–​15, 619n53
quotient distance formula, 508, 526n37 as a UTT, 300–​02, 310
quotient distance space, 506–​24 retrograde operation, 88–​89, 111, 124–​25, 222–​23, 227, 233,
quotient graph, 93–​94 364–​75
quotient group, 177, 240–​45, 268 as an order transformation, 373
quotient homomorphism, 244–​45, 255 retrograde sign (in a twelve-​tone row label), 369
quotient interval space, 255, 268–​73 Revuv. See interval-​reversing mapping
quotient interval-​space homomorphism, 269 reverse (of a transformation graph or network), 109, 350
quotient map, 60, 244–​45, 269, 342–​43 reversible (transformation graph or network), 109, 343
quotient set, 60–​61, 268 rhythm
vs. product, 63 generated, 617n24
quotient space, 60–​61, 91, 414, 477n10 maximally even, 602, 617n24
and orbifolds, 426–​27 See also beat-​class space; beat space; ski-​hill lattice

656 • Index
RI-​chaining (rows), 383n50 scale index (in enharmonic coordinates), 556
Riemann, Bernhard, xv(n11) scale lattice. See voice-​leading lattice
Riemann, Hugo, xv(n11) scale system, 606–​16
and dualism, 114–​15n11, 302–​03, 334n3/​8, 366 canonical, 606, 610
and harmonic function, 36n41, 74n18, 336n38 Schenker, Heinrich, 165n65
and the tonnetz, 27–​28, 37n56 Schenkerian theory, xv(n2), 36n41, 165n65, 613, 614
and triadic relationships, 120, 302–​03, 334n3/​8, 335n13 and transformational/​tonnetz analysis, 105–​06, 117n50/​
Riemann group (ℛ), 204n47, 306–​10, 336–​37n42, 628 51, 135–​36, 158, 163n34, 363
Cayley diagram for, 307–​09 Schillinger, Joseph, 298n38
duality with TI group, 329–​30, 337n55/​59 Schnittke, Alfred, 338
and voice leading, 310, 335n14 Schoenberg, Arnold, 167, 222, 373, 535
Riemann space. See neo-​Riemannian triad space chart of the regions, 127, 162n18
Riemannian Fantasy, violin and piano (Op. 47), 162n16, 162–​63n28
vs. neo-​Riemannian, 37n58, 55, 162n23, 303 “Farben” (Op. 16/​3), 248n14
vs. non-​Riemannian, 303 Moses und Aron, 162n16
See also uniform triadic transformation (Riemannian) Piano Piece (Op. 11/​1), 415
Riemannian tonnetz. See tonnetz Piano Piece (Op. 23/​3), 337n59
Riepel, Joseph, 248n10 Piano Piece (Op. 33a), 126–​27
right-​to-​left orthography. See orthography Pierrot lunaire: No. 4, “Eine blasse Wäscherin,” 217; No. 8,
Rimsky-​Korsakov, Nikolai “Nacht,” 32, 144, 415–​16, 420, 517–​18
Christmas Eve, 331–​34, 337n61, 338n63, 345, 351, 466 Quartet, string, No. 4 (Op. 37), 371–​78
Legend for orchestra (Skazka, Op. 29), 322, 336n36 Quintet, wind (Op. 26), 383n54
Scheherazade, 110 Serenade (Op. 24), 365
Rings, Steven, 265, 299, 344 Verklärte Nacht, 313
RL-​chain, RL-​cycle. See LR-​chain schritt, 36n43, 306–​22, 334n8, 335n13
Rodgers, Richard vs. transposition, 310, 317–​18, 320–​21, 336n30, 366
The Sound of Music, 166n75 schritt group (ℛ+​), 306–​08
Roeder, John, 118, 424, 526n35 schritt-​wechsel group. See Riemann group
Romanesca, 140 Schubert, Franz, 137, 166, 336n28, 480n58
rooted graph, 165n66 Fantasy, piano four hands (D. 940, f), 319, 336n32, 341–​42
Roslavets, Nikolai Impromptu (D. 899/​2, E♭), 166
Sonata, piano, No. 1, 34n18 Impromptu (D. 899/​3, G♭), 596–​97, 617n18
rotation (serial), 372–​73, 383n54 Impromptu (D. 935/​1, f), 155–​56
See also serial group (extended) Mass (D. 950, E♭), Sanctus, 319, 336n31
rotational array (Stravinsky), 383n54 Octet (D. 803, F), 319, 336n31
Rothgeb, John, 577n7 Quartet, string (D. 887, G), 156–​60, 166n72/​74/​76/​77/​78
rounding off, 259, 600, 617n21 Quartettsatz (D. 703, c), 160, 166n80
row, 74n11 Quintet, piano (D. 667, A, “Trout”), 134, 135, 309–​10,
See also twelve-​tone row 614–​15
row class, 235, 364 Quintet, string (D. 956, C), 313, 415
See also row space (forms of one row) Scherzo (D. 593/​2, D♭), 141
row/​column permutation property (for group tables), 218, 248n16 Die schöne Müllerin, No. 8, “Morgengruss,” 166, 336n35
row space Sonata, piano (D. 664, A), 597
all twelve-​tone rows (row), 235, 364, 374, 624 Sonata, piano (D. 958, c), 319
forms of one row, 124–​27, 222, 364, 496, 499 Sonata, piano (D. 960, B♭), 160, 163n33, 165–​66n71, 166,
RP-​cycle. See PR-​cycle 550–​51, 553, 561, 579n42
ruler group, 175 Trio, piano, No. 1 (D. 898, B♭), 160
Russell, Bertrand, 201n3 Winterreise, No. 7, “Auf dem Flusse,” 153–​55, 165n65/​68/​
Russell, George, 616n14 69, 336n27
Schumann, Clara
s1-​chain, 572–​76 Ballade (Op. 6/​4, d), 594
s1-​cycle, 574–​76, 581n61 Schumann, Robert
Saariaho, Kaija Dichterliebe, 162n25
Vers le blanc, 447, 451, 459, 479n33, 481, 518 Seeger, Ruth. See Crawford (Seeger), Ruth
Saint-​Saëns, Camille segmental straight-​line voice leading. See voice leading
Symphony No. 3 (c), 321 semantics (vs. syntax), 347, 381n18
scale semidirect product, 334n7
cross-​type mappings, 278–​82, 290–​93 semigroup, 175–​76, 182, 190, 380n1/​3, 381n24, 382n29
as an interval space, 257, 532 semilattice, 106
measuring by step, 10, 187, 256 semireduced (scale), 618n38
microtonal, 606–​16 semiregular (group action), 204n44
See also specific scale types by name (e.g., harmonic minor) sequence
scale-​degree space (sd), 16, 624 “ascending 5–​6,” 48–​49, 106, 141
as a distance space, 532 ascending fifths, 158
as an interval space, 265, 532 chromatic, 106, 110, 132–​34, 157–​60, 319–​22
product with pitch-​class space (sd × pc), 265–​67, 275, 292, descending fifths, 110, 140–​41
299, 335n19, 579n45 diatonic, in a generic tonnetz, 140–​41, 163n42

Index • 657
sequence (cont.) equivalence with interval space. See Lewin’s theorem
extraction of, 535 generalized, 199, 306, 360
generic, 534–​38 and network isography, 356–​57
hierarchy, 159–​60, 534–​38, 577n4 subgroups of the UTT group, 324–​25, 336–​37n42
Laitz notation, 141 singularity, 427
Pachelbel (Romanesca), 140–​41 See also boundary
vs. series (in mathematical terminology), 74n11 sink (in a directed graph), 97–​98
subdivision of, 535 ski-​hill lattice (Cohn), 100–​05, 112, 117n41/​42, 227–​28,
serial group (𝒮𝑒𝑟12), 222, 233, 235, 248n20, 364–​66, 628 249n30
Cayley diagram for, 125, 364 as a distance space, 497, 500
commutative counterpart to (𝒮𝑒𝑟12′), 365–​66, 382n46 slide. See P′
extended (𝒮𝑒𝑟12+​), 373–​75, 629 Slonimsky, Nicolas
extended with exchange (𝒮𝑒𝑟12X), 375–​80, 629 51 Minitudes, 298n38
serial transformation, 222–​23, 235, 239, 364–​80 Smetana, Bedřich
series (vs. sequence, in mathematical terminology), 74n11 The Bartered Bride, Overture, 133, 162–​63n28, 402, 564,
set, 39–​42, 73n1 614–​15
common notations, 39–​41 smooth (voice leading). See voice leading
See also multiset; ordered set solfège index (in enharmonic coordinates), 556
set class solfège syllables, 548, 579n35
generic, 541–​43 source (in a directed graph), 97–​98
and group action, 234 space, x–​xi, 4–​8
and OPTI-​classes, 411–​12, 414–​15, 476 formal/​syntactic vs. physical/​acoustic, 7–​8
and OPTIC-​classes, 413–​15 names/​labels for, 7
and spelled heptachords, 589 See also specific spaces by name (e.g., pitch space)
set-​class space space assignment function (in a cross-​type transformation
2-​voice (dyadic), 91, 439 graph), 380n5
3-​voice (trichordal), 90–​91, 404, 455, 484 span (of an spc set), 572, 583
seventh chord span optimization (in normal form calculation), 418, 424n53
generic, 539–​41 spanning cycle (in a graph), 81, 114n6
inflections of, 590 spatial network, 113–​14, 118n61, 216–​17, 341–​42
as a nearly even set, 466, 474 spc. See spelled pitch class
as an spc set, 583–​84 spc design. See design
transformations on, 330–​34, 337n58, 337–​38n62, 539–​40 spc transposition (τ). See transposition
shearing transformation, 26, 162n21 species, 554, 563–​64, 568–​77, 579n44, 580n53, 581
Shepard, Roger, 35n19 equivalence with fifth-​string for diatonic sets, 574
Shostakovich, Dmitri, 335n19 of proper spelled heptachords, 592
Preludes and Fugues, Op. 87, Fugue No. 1 (C), 571, 580n55 specific (vs. generic), 10, 34n9, 545–​46, 569, 580n53, 583,
Sibelius, Jean 601, 606
Tapiola, 478n30 specific interval. See interval
The Tempest, 478n30 specular reflection, 436
sign group (𝒮ign), 177, 301, 629 spelled (vs. signed), 35n32
signature group, 560–​64 spelled heptachord, 583–​97, 616n1, 619
signature transformation, 150, 161n9, 280, 550–​68, 579n42, 581 and chord species, 587, 589
and chromatic and diatonic transposition, 553, 559–​64 common types, 587–​88
of diatonic scales, 161n9, 559–​60 definition and basic properties, 585–​87, 589
enharmonic compatibility, 558 and enharmonic equivalence, 588–​89, 610
in enharmonic coordinates, 558–​59 enharmonic form of, 593, 603, 616n3
and spc transposition, 551 and inflections, 590–​97
and spelled heptachords, 585–​86 and key signatures, 583, 586, 590
summary of properties, 564 in pc space vs. pitch space, 616n3
See also s1-​chain symmetry of, 589
signature vector, 611–​12 τ-​classes of, 589
signed (letter), 9 τɩ-​classes of, 589
vs. spelled, 35n32 spelled pitch class, 17–​21
signed letter space (sl), 16–​18, 610, 624 spelled pitch-​class numbers, 19–​20, 292
as a distance space, 496, 500 spelled pitch-​class set, 73, 583–​85
as an interval space, 272–​73 spelled pitch-​class space (spc), 17–​21, 234, 238–​39, 625
See also spelled pitch-​class space and chromatic and diatonic spaces, 545–​50
signed registral letter space (srl), 11–​12, 625 as a distance space, 489–​90
as an interval space, 272–​73 as an interval space, 188, 260–​61
signed scale degree, 20–​21 spelled heptachords in, 583–​90
similarity measure, 525n25 spelled pitch space. See signed registral letter space
simple (graph), 115n20 sphere
simple (group), 227, 249n27 in a distance space, 504, 511, 516, 527n39
simplicial coordinates, 421n8 graphs embedded in, 85–​86, 95
simply transitive (group action), 197–​201, 204n44/​45/​47 as a manifold, 426
choice of group in analysis, 324, 336n41 n-​dimensional (𝕊n), 426, 628

658 • Index
 as a quotient space, 476–​77n2, 480n47 sum distance. See voice-​leading distance
vs. torus, 36n45, 95 super–​whole-​tone heptachord (swt(n)), 587–​88, 590,
spiral (pitch space), 12–​13 599–​601, 603, 625
spiral array (Chew), 146–​47 superhexatonic heptachords (shexa(n), shexb(n)), 587–​88, 624
spiral of fifths, 21 superposition (of tertian chords), 539
split (pitch tuple), 412, 521–​24 superset, 40
and voice leading, 423n45 surjection. See onto (function)
sporadic group, 249n27 symmetric (binary relation), 64
stabilize, 235 symmetric difference (of sets), 42
stabilizer (StG(x)), 235–​36 symmetric group (𝒮n), 213–​14, 407–​08, 628
stable segment (of an s1-​chain), 572 action on twelve-​tone rows, 374
standard (enharmonic equivalence vector), 610–​11, 623 symmetry, 205
standard pattern. See diatonic rhythm in beat-​class space, 602
state (in theory of computation), 477n11 degree of, 236
state class (in ski-​hill lattice), 103–​05, 227–​28, 249n30 distance-​space axiom, 487–​88
Stedman, Fabian, 201n1 in generic space, 541–​42
step position (of an enharmonic equivalence vector), 610–​11 of graphs, 115n15
Stockhausen, Karlheinz group property, 176, 202n12
Klavierstück III, 118n61, 337n58 and group theory, 167, 233–​40
straight-​line voice leading. See voice leading and invariance, 233, 235–​39
Straus, Joseph N., 90–​91 in OPTIC spaces, 518–​19
Strauss, Johann II in pc space vs. pitch space, 237–​38
An die schönen blauen Donau, 413 of pitch tuples, 455–​56
Strauss, Richard, 335n19 and scale properties, 607
Sonata, cello (Op. 6, F), 594 in spc space, 238–​39, 584, 588
Till Eulenspiegels lustige Streiche, 594 of spelled heptachords, 588–​89
Stravinsky, Igor, 166n78, 383n54 and stabilizers, 235–​36
Le sacre du printemps, 413, 424n47 of transformation networks, 350
See also Petrushka chord of twelve-​tone rows, 125, 239, 364–​65, 375, 378–​79,
string. See ordered set 382n42, 423n44
string length, 5–​6 See also inversional symmetry; permutational symmetry;
stroboscopic portrait, 38n67 stabilizer; transpositional symmetry
strong isography. See isography symmetry group, 236, 406, 486
strong isomorphism. See isomorphism syntagmatics, 118n61
strong path consistency. See path consistency syntax (vs. semantics), 347, 381n18
strongly connected (graph), 98, 218 syntonic comma, 27–​28, 37n57/​59, 58, 74n13, 272, 336n40
structure implies multiplicity, 576 syntonic drift, 37n60
subdivision system modulation (Lewin), 251n50
of a graph, 94–​95 Szymanowski, Karol
of a sequence, 535 Etude (Op. 4/​3, b♭), 594
subdominant transformation (D−1), 322 Etude (Op. 33/​3), 292–​93, 294, 299n48
subgraph, 80–​81
subgroup, 174–​75 T-​chaining (rows), 383n50
generated, 181–​83 T-​class
subgroup lattice, 182 generic, 541–​43
submediant transformation (M−1), 322–​24 of pitch-​class sets, 234, 414
suboctatonic heptachords (subocta(n), suboctb(n)), 587–​88, 625 of pitch tuples, 409, 414, 437–​38, 448–​49, 469–​70
subposition (of tertian chords), 539 of twelve-​tone rows, 235, 364
subset, 40 T-​complete (K-​net), 361
abstract, 99–​100 T-​connected (K-​net), 361
as a partial order relation, 99 T equivalence. See transpositional equivalence
subset lattice, 99, 419–​20 T-​net, t-​net. See transposition network
subspace, 7 T-​partition (of a K-​net), 358, 360–​62
of a distance space, 490 T-​set (in a K-​net), 358, 360–​62
of an interval space, 186–​87 T-​space, 409–​10
subtraction (in additive groups), 178 1-​voice, 427
successor (in a directed graph), 98 2-​voice, 437–​38, 516
Sudoku puzzle, 218, 248n16 3-​voice, 270, 447–​50, 514–​18
sum axis, 391–​92 4-​voice, 468–​70, 519
sum class, 83 axes of symmetry in, 453–​56
and accidental index (for spelled heptachords), 588 dimension of, 409–​10, 439, 447, 468
and generic transpositions, 533 as a distance space, 514–​18
in hexatonic triad cubes, 389, 391, 396, 422n20 as an interval space, 270, 486
in a hexatonic triad graph, 83 voice leading in, 449, 451–​53, 455–​58
in OP-​space, 431–​32, 443–​44, 463–​65, 603 T type (of a pc set or multiset), 417–​18, 424n54
in pitch-​tuple space, 427 T+​ space, T+​normal form, 416
sum-​class plane, 391, 409–​10, 423n36 Taneyev, Sergei, 34n15

Index • 659
tangent space, 164n45 torus, 24–​25, 126, 390
τ-​class (of spc sets), 584 doughnut-​shaped, 24–​25
τ equivalence (of spelled heptachords), 589 flat, 24–​25
τ-​transposition. See transposition, spc as a manifold, 426
τɩ-​class (of spc sets), 584 n-​dimensional (𝕋 n), 36n46, 146, 390, 407, 426, 629
τɩ equivalence (of spelled heptachords), 589 vs. sphere, 36n45
taxicab metric, 501 See also embedding, toroidal
Tchaikovsky, Pyotr Ilyich total order, 99
Eugene Onegin, 297n19 total transposition (of a UTT), 306, 307, 326
Romeo and Juliet, 47, 594 Tovey, Donald Francis, 248n13
Symphony No. 4 (f), 105, 156, 602 transfer principles (Kolman), 296n4
teleport, 24, 435 transformation, 45, 74n16
temporal space (time), 30–​31, 489, 625 direct vs. indirect, 107–​08, 140–​41, 342
ternary. See relation and intervals, 195–​201, 204n50
tesseract, 480n49 neo-​Riemannian, 54–​55, 74n19
tetrahedron, 115n12 vs. prolongation, 105–​06, 117n50/​51
in a 3-​dimensional tonnetz, 145–​46 Riemannian, 55, 302–​03, 306–​22
4-​voice OPT-​space, 470–​74 on a set, 46
4-​voice OPTI-​space, 474–​76 vs. transposition, 202n6
cross section of 4-​voice OP-​space, 463–​65 See also cross-​type transformation; function; specific
third-​substitution (in sequences), 141, 164n43 transformations by name (e.g., parallel)
thirds space transformation graph, 109–​10, 339–​64, 380n3, 384
vs. fifths space, 532–​33 cross-​type, 342
and functional harmony, 142–​43, 323–​24 graphic conventions for, 341–​42
generic, 22 underlying, 109–​10, 340
vs. pitch-​class space, 532–​33 transformation group, 171, 183, 189–​95
Thurston, William, 476n1 vs. interval group, 171–​73
TI-​class, 414 transformation network, 106–​14, 339–​64, 384
generic, 541–​43 cross-​type, 342, 380n5/​6, 381n23
See also set class graphic conventions for, 341–​42
TI equivalence (~TI), 63, 65, 234–​38 transformational attitude, 200–​01, 204n50, 205
in continuous space, 410–​11, 423n41 transformational voice leading, 421n1
in generic space, 541–​43, 578n17 transitive (binary relation), 64, 204n43
symmetry group for, 411 transitive (group action), 197–​98, 204n43/​44
TI-​space, 437–​38 See also simply transitive
TI type (of a pc set or multiset), 417–​18, 424n54 transitive closure (of a binary relation), 117n36
tile, 23–​24 transposition
See also fundamental region chromatic (Rings), 275
time domain. See temporal space cross-​type “transposition,” 278–​82
tonality graph (Jablonsky), 123–​24 in enharmonic coordinates, 557–​58
tonnetz, 22–​33, 36n43, 130–​36 of a field (spelled heptachord), 591–​93
3-​dimensional (tetrahedral), 145–​46 generic (tk), 72–​73, 138–​41, 163n41, 274, 533–​34,
in a 19-​note scale, 614 547, 606
alternate definitions of “tonnetz,” 25, 131 interscalar, 298n34
analysis in, 131–​35, 140–​41, 316–​20, 338 in an interval space (Transi), 195–​201, 273–​86, 299
diatonic, 138–​41 and inversion, 189–​95, 293
as a direct product, 230–​31, 267–​68, 272 as an isometry, 492
distance in. See neo-​Riemannian distance and label functions, 274
finite vs. infinite, 93–​94 as a network isomorphism, 349, 353
generic, 32–​33, 138–​41, 143, 163n42, 534 octatonic, 275
as a graph, 92–​96 pentatonic, 281–​82, 292–​93
in a hexatonic triad cube chain, 403–​04 in permutation theory, 247n3
neo-​Riemannian (conformed), 23–​28, 92–​96, 130–​36, in pitch-​class space (Tk), 54, 72–​73, 189–​95
310–​20, 496, 500 in pitch space (Tk), 171–​73, 193–​95
and neo-​Riemannian triad space, 130–​31 pivot (Rings), 275
non-​triadic, 28–​32, 144–​46, 164n52 real (Rings), 275
planarity of, 94–​95, 116n28 scalar (intrascalar), 274, 298n31
Riemannian (unconformed), 27–​28, 76n37, 96, 163n33, of sets, 275
496, 500 and signature transformations, 553, 559–​64
triangulated, 25–​27 in spc space (τk), 72–​73, 274–​75, 547, 551
and voice leading, 402–​04, 422n23 specific, 606
See also neo-​Riemannian triad space of spelled heptachords, 590–​93
tonnetz pole, 163n36, 315 vs. transformation, 202n6
tonnetz space (tonn), 25, 96, 231, 625 as a UTT, 301–​02, 334n4
topology (vs. geometry), 6, 14–​15 See also line of transposition
Torke, Michael transposition group (𝒯12, 𝒯∞), 171–​73, 182, 190, 629
The Yellow Pages, 580n48 Cayley diagram for, 221–​22

660 • Index
 continuous, 409 graph (underlying a labeled graph), 80
in an interval space (𝒯S), 195–​201, 204n42, 629 space (underlying a transformation network), 113–​14
transposition/​inversion group (𝒯ℐ12, 𝒯ℐ∞), 192–​95, 204n38, transformation graph (underlying a transformation net-
243–​44, 629 work), 109–​10, 340
action on pitch-​class sets, 234–​38 unified enharmonic space (edpc*), 556, 622
action on pitch classes vs. order numbers, 373–​75 action of signature group on, 556–​64
Cayley diagram for, 221–​22 as an interval space, 556
duality with Riemann group, 329–​30, 337n55/​59 uniform flip-​flop circle (Clough), 336n24
transposition network (T-​net), 106–​11, 171, 352, 358 uniform prismatic polytope, 480n57
diatonic (t-​net), 140–​41 uniform triadic transformation, 300–​10, 338
transpositional equivalence (~T), 63, 65, 234–​37, 270 action of, 302–​04
in continuous space, 405–​06, 409–​10, 423n34, 447–​49, dual, 330
468–​70 even vs. odd, 326–​27, 337n50/​51
in generic space, 541–​43, 578n17 group structure, 304–​05, 325–​27, 337n47, 629
normal region for, 437 inverse of, 302–​04
symmetry group for, 409–​10, 423n35, 486 mod-​n applications, 331, 337n60
See also T-​space non-​triadic applications, 330–​34, 337n56/​57/​58/​59
transpositional-​inversional equivalence. See TI equivalence order-​24, 324–​25
transpositional symmetry, 236–​37 ordered-​triple notation for, 301–​02
in generic space, 542 as a permutation, 326
tresillo, 602, 617n25 product (composition) of, 302–​04
triad Riemannian, 303, 306–​22
generic, 538–​41 serial applications, 331, 366–​71, 382n45
as a nearly even set, 396, 442, 444–​45 and voice leading, 303, 334n4
ordered-​pair notation for, 301 uniformity
as a second-​order maximally even set, 604–​05 of a triadic transformation, 301
triad space. See consonant triad space; diatonic triad space; generic of a voice leading, 526n31
triad space; neo-​Riemannian triad space; Weber space union (of sets), 40–​42
triadic transformation, 54–​55, 132–​34, 300–​34 union-​cardinality law, 41–​42
See also uniform triadic transformation unit (mod n), 178, 202n15, 203n23
triangle inequality (distance-​space axiom), 487–​89, 491 units group (ℤn×), 178, 213, 223, 227, 239, 629
vs. interval-​sum equation, 488 universal equivalence, 63
in product spaces, 502, 525n27 universal set, 40
triangulation (of a graph), 95 universally realizable (transformation graph). See realizable
trichordal array (Babbitt), 375, 383n54 unsigned (letters), 8–​9
triple sharp, triple flat, 11–​12, 34–​35n18 UTT. See uniform triadic transformation
Tristan chord, 413, 415, 420, 502
trivial (design), 555, 618n37 v-​unit (distance in T-​space), 481n64, 515, 519
trivial (genus), 555, 564, 569 value (of a function), 46
trivial (group), 176 variable, 46–​47
Tufte, Edward, 33n1, 38, 165n63 vs. constant, 74–​75n20
tuning, 7–​8 Vaughan Williams, Ralph
as a mapping, 8, 57–​58 A Sea Symphony, 335n19
in microtonal scales, 609, 613, 619n44/​50/​52 vector, 43, 184, 203n27, 392–​93, 421n10/​13, 424
tuple. See ordered set See also enharmonic equivalence vector; interval-​class vec-
twelve-​tone area, 126–​27, 162n16, 371 tor; ordered set; signature vector
twelve-​tone array. See array vector space, 203n27
twelve-​tone matrix. See matrix Venn diagram, 41, 73n6
twelve-​tone operator (Morris), 248–​49n21 Verdi, Giuseppe
twelve-​tone row, 88, 214, 233 La traviata, “Si ridesta,” 75n25, 325
labeling, 115n16, 126, 365–​70 Il trovatore, “Ah sì, ben mio, coll’essere,” 134, 163n31,
in OC-​space, 423n44 163n33, 614–​15
as an order-​pc bijection, 371–​72 vertex (in a graph), 80
as a permutation, 371–​72 vertex labeling, 80
transformations on, 222–​23, 235, 239, 331, 364–​80 and graph isomorphism, 88
twelve-​tone row space. See row space Vial, François-​Guillaume, 162n17
Tymoczko, Dmitri, xv(n6), 590 VII7–​I design, 593–​94
and geometry, xiv, 388, 424, 478n22/​24, 481 voice crossing, 477n12, 510–​12
and scales, 603 voice exchange, 398–​99, 435–​36, 445
and the tonnetz, 422n23 voice leading
and voice leading, 421n12, 466, 480n58 of 014 trichords, 144
arrow notation for, 433
u-​unit (distance in T-​space), 515 continuous vs. discrete, 432–​33
unary. See operation; relation as a function of time, 47–​49, 447, 477n13
unconformed. See conformance in generic space, 538–​40, 577n13
underlying and interval spaces, 387
directed graph (underlying a graph or network), 109–​10 of nearly even chords, 444–​45, 466

Index • 661
voice leading (cont.) wechsel, 36n43, 120, 306–​22, 334n8, 335n13
oblique, 435 vs. inversion, 310, 335n16, 366
parallel, 392–​94, 421n12, 435 as a relation, 310, 335n15
as a path in a space, 14, 388, 432–​33 and transpositions, 310, 317–​18, 320–​21, 324
in pitch-​tuple space and OPTIC spaces, 432–​33 Weitzmann, Carl Friedrich, 36n49, 313, 335n22, 336n25/​26
segmental straight-​line, 433 Weitzmann region. See hexatonic bridge family
smoothness, 387, 505, 526n35 well-​formed scale, 550, 579n40/​41, 606–​07, 611, 612
straight-​line, 433 Westergaard, Peter, 367–​68
and transformations, 123, 129–​30, 144, 303, 310, 333, 387, Westminster Quarters, 43, 65, 213
421n1, 433 Whitman, Walt, xiv–​xv
zero-​sum, 393–​94, 421n12, 435, 463–​64 whole-​tone collection, 25–​26, 625
voice-​leading distance, 403, 422n24 in a 19-​note scale, 613
generic, 539–​40 as a generic scale, 607
as a measure of evenness, 603 maximal evenness of, 599, 601
in a product space, 501–​05 as a pitch space, 11
in a quotient space, 508, 514–​18 as an spc set, 584–​85
between seventh chords, 539–​40 symmetry of, 236
between triads, 513–​14, 539–​40 whole-​tone heptachord (improper), 588
as “work,” 500, 525n24 whole-​tone-​plus-​one collection. See super–​whole-​tone
voice-​leading lattice, 603–​06, 608, 617n28/​31, 618n36, heptachord
619–​20 Winograd, Terry, 579n38
voice-​leading space, 387–​481 Wolf, Hugo
Mörike-​Lieder, No. 22, “Seufzer,” 402
Wagner, Richard, 380n11 word (in group theory), 249n23, 497–​500, 525n19
Parsifal, Grail theme, 34–​35n18, 164n49, 249n30, 311–​12, equivalence of, 498
314, 335n19 on a set of generators, 498
Das Rheingold, Scene 4, 311–​12 word-​length distance, 499
Tristan und Isolde, Preludes to Acts I and III, 278–​79 wormhole, 24, 435, 460
See also Tristan chord wraparound interval (for a diatonic spc set), 575
walk (in a graph), 114n3 wreath product, 304–​05, 328, 334n7, 337–​38n62, 376
Weber, Gottfried, 27, 127 wxy, wxz, wyz forms (of a 4-​voice T-​class), 469
Weber distance, 136–​38, 163n35, 494, 499, 514
Weber group (𝒲), 325–​26, 499, 629 x form (of a 2-​voice T-​class), 437
Weber space (weber), 127–​29, 142, 155, 166, 162n17/​20, xy, xz forms (of a 3-​voice T-​class), 449
325, 625 xyz form (of a 4-​voice T-​class), 469
as a distance space, 494–​95, 525n18
vs. neo-​Riemannian triad space, 128–​29, 162n25, 163n35 y form (of a 2-​voice T-​class), 437
Webern, Anton, 123, 365 Yust, Jason, xv(n12)
Concerto for Nine Instruments (Op. 24), 365, 382n45 yz form (of a 3-​voice T-​class), 270, 449, 454–​55
“Herr Jesus mein” (Op. 23/​3), 88–​89, 112–​14, 115n16, 217
Quartet, string (Op. 28), 148–​49, 164n59, 331, 370–​71, Z relation (Forte), 170, 202n5, 357, 543
379, 383n49/​50/​51 zeitnetz, 118n59
Symphony (Op. 21), 365 zero-​sum voice leading. See voice leading
Variations, piano (Op. 27), 265, 366–​70, 383n48 zone of disagreement (for distance functions), 503–​05

662 • Index