Introduction To Nonextensive Statistical

Mechanics Approaching A Complex World 2nd

Edition Constantino Tsallis download

https://ebookbell.com/product/introduction-to-nonextensive-
statistical-mechanics-approaching-a-complex-world-2nd-edition-
constantino-tsallis-47633060

Explore and download more ebooks at ebookbell.com

Here are some recommended products that we believe you will be
interested in. You can click the link to download.

Introduction To Nonextensive Statistical Mechanics Approaching A

Complex World 1st Edition Constantino Tsallis

https://ebookbell.com/product/introduction-to-nonextensive-
statistical-mechanics-approaching-a-complex-world-1st-edition-
constantino-tsallis-11855310

Introduction To Modern Analysis 2nd Edition 2nd Kantorovitz

https://ebookbell.com/product/introduction-to-modern-analysis-2nd-
edition-2nd-kantorovitz-44870612

Introduction To The Speechmaking Process 15th Edition Diana K Leonard

Raymond S Ross

https://ebookbell.com/product/introduction-to-the-speechmaking-
process-15th-edition-diana-k-leonard-raymond-s-ross-44874488

Introduction To Construction Management 2nd Edition 2nd Fred Sherratt

https://ebookbell.com/product/introduction-to-construction-
management-2nd-edition-2nd-fred-sherratt-44899008
Introduction To Analysis With Complex Numbers Irena Swanson

https://ebookbell.com/product/introduction-to-analysis-with-complex-
numbers-irena-swanson-44912170

Introduction To Quantitative Methods In Business With Applications

Using Microsoft Office Excel 1st Edition Bharat Kolluri

https://ebookbell.com/product/introduction-to-quantitative-methods-in-
business-with-applications-using-microsoft-office-excel-1st-edition-
bharat-kolluri-44915766

Introduction To Biostatistical Applications In Health Research With

Microsoft Office Excel And R 2nd Edition Robert P Hirsch

https://ebookbell.com/product/introduction-to-biostatistical-
applications-in-health-research-with-microsoft-office-excel-and-r-2nd-
edition-robert-p-hirsch-44915830

Introduction To Strategies For Organic Synthesis 2nd Edition Laurie S

Starkey

https://ebookbell.com/product/introduction-to-strategies-for-organic-
synthesis-2nd-edition-laurie-s-starkey-44915846

Introduction To Hydrogen Technology 2nd Edition K S V Santhanam

https://ebookbell.com/product/introduction-to-hydrogen-technology-2nd-
edition-k-s-v-santhanam-44916142
Constantino Tsallis

Introduction
to Nonextensive
Statistical
Mechanics
Approaching a Complex World
Second Edition
Introduction to Nonextensive Statistical Mechanics
Constantino Tsallis

Introduction to Nonextensive
Statistical Mechanics
Approaching a Complex World
Second Edition
Constantino Tsallis
Centro Brasileiro de Pesquisas Fisicas
Rio de Janeiro, Brazil

ISBN 978-3-030-79568-9 ISBN 978-3-030-79569-6 (eBook)

https://doi.org/10.1007/978-3-030-79569-6

1st edition: © Springer Science+Business Media, LLC 2009

2nd edition: © Springer Nature Switzerland AG 2023
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
To my family,
whose love is the roots of my dreams
Preface

In 1902, after three decades that Ludwig Eduard Boltzmann formulated the first
version of standard statistical mechanics, Josiah Willard Gibbs shares, in the Preface
of his superb Elementary Principles in Statistical Mechanics [1, 2, 3]: “Certainly,
one is building on an insecure foundation ...”. After such words by Gibbs, it is,
still today, uneasy to feel really comfortable regarding the foundations of statistical
mechanics from first principles. Since the time when I took the decision to write the
present book, I would certainly second his words. Several interrelated facts contribute
to this inclination.
First, the verification of the notorious fact that all branches of physics deeply
related to the theory of probabilities, such as statistical mechanics and quantum
mechanics, have exhibited, along history and up to now, endless interpretations,
reinterpretations, and controversies. All this is fully complemented by philosophical
and sociological considerations. As one among many evidences, let us mention the
eloquent words by Gregoire Nicolis and David Daems [4]: “It is the strange privilege
of statistical mechanics to stimulate and nourish passionate discussions related to
its foundations, particularly in connection with irreversibility. Ever since the time of
Boltzmann, it has been customary to see the scientific community vacillating between
extreme, mutually contradicting positions.”
Second, I am inclined to think that, together with the central geometrical concept
of symmetry, virtually nothing more basically than energy and entropy deserves
the qualification of pillars of modern physics. Both concepts are amazingly subtle.
However, energy has to do with possibilities, whereas entropy with the probabilities
of those possibilities. Consequently, the concept of entropy is, epistemologically
speaking, one step further. One might remember, for instance, the illustrative dialog
that Claude Elwood Shannon had with John von Neumann [5, 6]: “My greatest
concern was what to call it. I thought of calling it ‘information’, but the word was
overly used, so I decided to call it ‘uncertainty’. When I discussed it with John von
Neumann, he had a better idea. Von Neumann told me, ‘You should call it entropy, for
two reasons. In the first place, your uncertainty function has been used in statistical
mechanics under that name, so it already has a name. In the second place, and more
important, nobody knows what entropy really is, so in a debate, you will always

vii
viii Preface

have the advantage.’”. Frequently we hear and read diversified opinions about what
should and what should not be considered as “the physical entropy”, its connections
with heat, information, and so on.
Third, the dynamical foundations of the standard, Boltzmann-Gibbs (BG) statis-
tical mechanics are, mathematically speaking, not yet fully established. It is known
that, for classical systems, exponentially diverging sensitivity to the initial condi-
tions (i.e., positive Lyapunov exponents almost everywhere, which typically imply
mixing and ergodicity, properties that are consistent with Boltzmann’s celebrated
Stoszzahl Ansatz, “molecular chaos hypothesis”) is a sufficient property for having a
meaningful statistical theory. More precisely, one expects that this property implies,
for many-body Hamiltonian systems attaining thermal equilibrium, central features
such as the celebrated exponential weight, introduced and discussed in the 1870s
by Ludwig Boltzmann (very especially in his 1872 [7] and 1877 [8] papers)1 in the
so-called μ-space, thus recovering, as particular instance, the velocity distribution
published in 1860 by James Clerk Maxwell [11]. More generally, the exponential
divergence typically leads to the exponential weight in the full phase space, the
so-called -space first proposed by Gibbs. However, hypothesis such as this expo-
nentially diverging sensitivity, are they necessary? In the first place, are they, in
some appropriate logical chain, necessary for having BG statistical mechanics? I
would say yes. But are they also necessary for having a valid statistical mechanical
description at all for any type of thermodynamic-like systems?.2 I would say no. In
any case, it is within this belief that I write the present book. All in all, if such is
today the situation for the successful, undoubtedly correct for a very wide class of
systems, universally used, and centennial BG statistical mechanics and its associated
thermodynamics (“a science with secure foundations, clear definitions, and distinct
boundaries” as so well characterized by James Clerk Maxwell), what can we then
expect for its possible generalization only a few decades after its first proposal, in
1988?
Fourth,—last but not least—no logical-deductive mathematical procedure exists,
nor will presumably ever exist, for proposing a new physical theory or for general-
izing a pre-existing one. It is enough to think about Newtonian mechanics, which has
already been generalized along at least two completely different (and compatible)
paths, which eventually led to the theory of relativity and to quantum mechanics.
This fact is consistent with the evidence that there is no unique way of generalizing
a coherent set of axioms. Indeed, the most obvious manner of generalizing it is to
replace one or more of its axioms with weaker ones. And this can be done in more
than one manner, sometimes in infinite manners. So, if the prescriptions of logics
and mathematics are helpful only for analyzing the admissibility of a given general-
ization, how do generalizations of physical theories, or even scientific discoveries in

1 English translation in [9]; see also [10].

2 For example, we can read in a recent paper by Giulio Casati and Tomaz Prosen [12] the following
sentence: “While exponential instability is sufficient for a meaningful statistical description, it is
not known whether or not it is also necessary.”
Preface ix

general, occur? Through all types of heuristic procedures, but mainly—I would say—
through methaphors [13]. Indeed, theoretical and experimental scientific progress
occurs all the time through all types of logical and heuristic procedures, but the
particular progress involved in the generalization of a physical theory immensely, if
not essentially, relies on some kind of metaphor.3 Well-known examples are the idea
of Erwin Schroedinger of generalizing Newtonian mechanics through a wave-like
equation inspired by the phenomenon of optical interference, and the discovery by
Friedrich August Kekule of the cyclic structure of benzene inspired by the shape of
the mythological Ouroboros. In other words, generalizations not only use the clas-
sical logical procedures of deduction and induction, but also, and overall, the specific
type of inference referred to as abduction (or abductive reasoning), which plays the
most central role in Charles Sanders Peirce’s semiotics. The procedures for theo-
retically proposing a generalization of a physical theory somehow crucially rely on
the construction of what one may call a plausible scenario. The scientific value and
universal acceptability of any such a proposal are of course ultimately dictated by
its successful verifiability in natural and/or artificial and/or social systems. Having
made all these considerations, I hope that it must by now be very transparent for
the reader why, in the beginning of this Preface, I evoked Gibbs’ words about the
fragility of the basis on which we are founding.
Newton’s decomposition of white light into rainbow colors, not only provided a
deeper insight into the nature of what we know today to classically be electromag-
netic waves, but also opened the door to the discovery of infrared and ultraviolet.
While trying to follow the methods of this great master, it is my cherished hope that
the present, nonextensive generalization of Boltzmann–Gibbs statistical mechanics,
may provide a deeper understanding of the standard theory, in addition to proposing
some extension of the domain of applicability of the powerful methods of statistical
mechanics. The book is written at a graduate course level, and some basic knowl-
edge of quantum and standard statistical mechanics, as well as thermodynamics, is
assumed. The style is, however, slightly different from a conventional textbook, in
the sense that not all the results that are presented are proved. The quick ongoing
development of the field does not yet allow for such an ambitious task. Various points
of the theory are presently only partially known and understood. So, here and there
we are obliged to proceed with heuristic arguments. The book is unconventional also
in the sense that here and there historical and other side remarks are included as well.
Some sections of the book, the most basic ones, are presented with all details and
intermediate steps; some others, more advanced or quite lengthy, are presented only
through their main results, and the reader is referred to the original publications to
know more. We hope, however, that a unified perception of statistical mechanics, its
background, and its basic concepts does emerge.
The book is organized into four parts, namely Part I—Basics or How the Theory
Works, Part II—Foundations or Why the Theory Works, Part III—Applications or
What for the Theory Works, and Part IV—Last (But Not Least). The first part consti-
tutes a pedagogical introduction to the theory and its background (Chaps. 1–3). The

3 I was first led to think about this by Roald Hoffmann in 1995.

x Preface

second part contains the state of the art in its dynamical foundations, in particular how
the index (indices) q can be obtained, in some paradigmatic cases, from microscopic
first principles or, alternatively, from mesoscopic principles (Chaps. 4–6). The third
part is dedicated to listing brief presentations of typical applications of the theory and
its concepts, or at least of its functional forms, as well as possible extensions existing
in the literature (Chap. 7). Finally, the fourth part constitutes an attempt to place the
present—intensively evolving, open to further contributions and insights4 —theory
into contemporary science, by addressing some frequently asked or still unsolved
current issues (Chap. 8). An Appendix with useful formulae has been added at the
end, as well as another one discussing escort distributions and q-expectation values.
It may be useful to point out at this stage that this book can be quite conve-
niently read along two possible tracks. The first track concerns readers at a graduate
student level. It basically consists of reading Chaps. 1, 2, 3, 5, and 8. The second
track concerns readers at the research level with some practice in standard statistical
mechanical methods. It basically consists of reading Chaps. 3–8. For both tracks, let
us emphasize that Chap. 7 contains a large amount of diversified applications. The
reader may focus on those of his/her main preference. In the present new edition,
we have refined some concepts, included some recent analytical discussions, and
added a considerable number of applications as well as experimental and numerical
verifications. Let us mention also that it has been unfortunately impossible to unify
the notations along all the chapters of the book, due to the fact that very many figures
have been reproduced from a large number of papers in the literature.
Toward this end, it is a genuine pleasure to warmly acknowledge the contributions
of M. Gell-Mann, maître à penser, with whom I have had frequent and delightfully
deep conversations on the subject of nonextensive statistical mechanics ... as well as
on many others. Many other friends and colleagues have substantially contributed to
the ideas, results, and figures presented in this book. Those contributions range from
insightful questions or remarks—sometimes fairly critical—to entire mathematical
developments and seminal ideas. Their natures are so diverse that it becomes an
impossible task to duly recognize them all. So, faute de mieux, I decided to name
them in alphabetical order, being certain that I am by no means doing justice to their
enormous and diversified intellectual importance. In all cases, my gratitude could not
be deeper. They are S. Abe, F. C. Alcaraz, R. F. Alvarez-Estrada, S. Amari, G. F. J.
Ananos, C. Anteneodo, V. Aquilanti, N. Ay, G. Baker Jr., F. Baldovin, M. Baranger,
G. G. Barnafoldi, C. Beck, I. Bediaga, G. Bemski, T. S. Biro, A. R. Bishop, H. Blom,
B. M. Boghosian, E. Bonderup, J. P. Boon, E. P. Borges, L. Borland, T. Bountis, E.
Brezin, A. Bunde, L. F. Burlaga, B. J. C. Cabral, M. O. Caceres, S. A. Cannas, A.
Carati, F. Caruso, M. Casas, G. Casati, N. Caticha, A. Chame, P.-H. Chavanis, C. E.
Cedeño, L. J. L. Cirto, J. Cleymans, E. G. D. Cohen, A. Coniglio, M. Coutinho Filho,
E. M. F. Curado, S. Curilef, J. S. Dehesa, S. A. Dias, A. Deppman, A. Erzan, L. R.
Evangelista, J. D. Farmer, R. Ferreira, M. A. Fuentes, L. Galgani, J. P. Gazeau, P.-G.
de Gennes, A. Giansanti, A. Greco, P. Grigolini, D. H. E. Gross, G. R. Guerberoff,
E. Guyon, M. Hameeda, R. Hanel, H. J. Haubold, R. Hersh, H. J. Herrmann, H. J.

4 For a regularly updated bibliography of the subject see http://tsallis.cat.cbpf.br/biblio.htm.

Preface xi

Hilhorst, R. Hoffmann, G. ‘t Hooft, K. Huang, M. Jauregui, H. J. Jensen, P. Jizba, L.

P. Kadanoff, G. Kaniadakis, T. A. Kaplan, S. Kawasaki, J. Korbel, D. Krakauer, P. T.
Landsberg, V. Latora, C. M. Lattes, E. K. Lenzi, S. V. F. Levy, H. S. Lima, J. A. S.
Lima, M. L. Lyra, S. D. Mahanti, A. M. Mariz, J. Marsh, A. M. Mathai, R. Maynard,
G. F. Mazenko, E. Megias, R. S. Mendes, L. C. Mihalcea, L. G. Moyano, J. Naudts,
K. P. Nelson, G. Nicolis, F. D. Nobre, J. Nogales, F. A. Oliveira, P. M. C. Oliveira, I.
Oppenheim, A. W. Overhauser, G. Parisi, R. Pasechnik, G. P. Pavlos, R. Piasecki, A.
Plastino, A. R. Plastino, A. Pluchino, D. Prato, P. Quarati, S. M. D. Queiros, A. K.
Rajagopal, A. Rapisarda, M. A. Rego-Monteiro, M. S. Ribeiro, A. Robledo, M. C.
Rocca, A. Rodriguez, S. Ruffo, G. Ruiz, M. Rybczynski, S. R. A. Salinas, Y. Sato, V.
Schwammle, L. Silva, L. R. da Silva, R. N. Silver, I. D. Soares, A. M. C. Souza, H.
E. Stanley, D. A. Stariolo, D. Stauffer, S. Steinberg, R. Stinchcombe, H. Suyari, F.
Vallianatos, H. L. Swinney, F. A. Tamarit, P. Tempesta, W. J. Thistleton, S. Thurner,
U. Tirnakli, L. Tisza, F. Topsoe, R. Toral, A. C. Tsallis, A. F. Tsallis, E. L. P. Tsallis, F.
Tsallis Viegas, S. Umarov, P. Van, M. E. Vares, M. C. S. Vieira, C. Vignat, J. Villain,
A. F. Vinas, S. Vinciguerra, G. Viswanathan, R. S. Wedemann, B. Widom, G. Wilk,
H. O. Wio, Z. Wlodarczyk, D. J. Zamora, and I. I. Zovko. Unavoidably, I must have
forgotten to mention some—this idea started developing over three decades ago!—to
them my most genuine apologies and gratitude. Finally, as in virtually all the fields of
science and very especially during the first stages of any new development, there are
also a few colleagues whose intentions have not been—I confess—very transparent
to me. But they have nevertheless—perhaps even unwillingly—contributed to the
progress of the ideas that are presented in this book. They surely know who they are.
My gratitude goes to them as well: it belongs to human nature to generate fruitful
ideas through all types of paths.
Along the years, I have relevantly benefited from the partial financial support
of various Agencies, especially the Brazilian CNPq, FAPERJ, PRONEX/MCT, and
CAPES, the USA NSF, SFI, SI International, AFRL, and John Templeton Foundation,
the Italian INFN and INFM, among others. I am indebted to all of them.
Finally, some of the figures that are presented in the present book have been repro-
duced from various publications indicated case by case. I gratefully acknowledge the
gracious authorization to do so from their authors.
Rio de Janeiro, Brazil through the periods 2004–2009 and 2018–2023

Rio de Janeiro, Brazil Constantino Tsallis

Contents

Part I Basics or How the Theory Works

1 Historical Background and Physical Motivations . . . . . . . . . . . . . . . . . . 3
1.1 An Overall Perspective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 Background and Indications in the Literature . . . . . . . . . . . . . . . . . . 11
1.4 Symmetry, Energy, and Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5 A Few Words on the Foundations of Statistical Mechanics . . . . . . . 20
2 Learning with Boltzmann-Gibbs Statistical Mechanics . . . . . . . . . . . . . 27
2.1 Boltzmann–Gibbs Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.1 Entropic Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2 Kullback–Leibler Relative Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3 Constraints and Entropy Optimization . . . . . . . . . . . . . . . . . . . . . . . . 39
2.3.1 Imposing the Mean Value of the Variable . . . . . . . . . . . . . . 39
2.3.2 Imposing the Mean Value of the Squared Variable . . . . . . 40
2.3.3 Imposing the Mean Values of Both the Variable
and Its Square . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.3.4 Others . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4 Boltzmann–Gibbs Statistical Mechanics
and Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.4.1 Isolated System—Microcanonical Ensemble . . . . . . . . . . . 44
2.4.2 In the Presence of a Thermostat—Canonical
Ensemble . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.4.3 Others . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3 Generalizing What We Learnt: Nonextensive Statistical
Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.1 Playing with Differential Equations—A Metaphor . . . . . . . . . . . . . 47
3.2 Nonadditive Entropy Sq . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

xiii
xiv Contents

3.2.2 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.3 Correlations, Occupancy of Phase Space, and Extensivity
of Sq . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.3.1 The Thermodynamical Limit . . . . . . . . . . . . . . . . . . . . . . . . 65
3.3.2 Einstein Likelihood Principle . . . . . . . . . . . . . . . . . . . . . . . . 72
3.3.3 The q-Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3.4 The q-Sum and Related Issues . . . . . . . . . . . . . . . . . . . . . . . 78
3.3.5 Extensivity of Sq —Effective Number of States . . . . . . . . . 81
3.3.6 Extensivity of Sq —Binary Systems . . . . . . . . . . . . . . . . . . . 85
3.3.7 Extensivity of Sq —Physical Realizations . . . . . . . . . . . . . . 96
3.3.8 An Epistemological Analogy . . . . . . . . . . . . . . . . . . . . . . . . 106
3.4 q-Generalization of the Kullback–Leibler Relative Entropy . . . . . . 106
3.5 Constraints and Entropy Optimization . . . . . . . . . . . . . . . . . . . . . . . . 110
3.5.1 Imposing the Mean Value of the Variable . . . . . . . . . . . . . . 110
3.5.2 Imposing the Mean Value of the Squared Variable . . . . . . 111
3.5.3 Imposing the Mean Values of Both the Variable
and Its Square . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
3.5.4 Others . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
3.6 Nonextensive Statistical Mechanics and Thermodynamics . . . . . . . 114
3.7 About the Escort Distribution and the q-Expectation Values . . . . . 122
3.8 About Universal Constants in Physics . . . . . . . . . . . . . . . . . . . . . . . . 126
3.9 Comparing Various Entropic Forms . . . . . . . . . . . . . . . . . . . . . . . . . . 129

Part II Foundations or Why the Theory Works

4 Probabilistic and Stochastic Dynamical Foundations
of Nonextensive Statistical Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.2 Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.2.1 Normal Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.2.2 Anomalous Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.3 Stable Solutions of Fokker–Planck-Like Equations . . . . . . . . . . . . . 150
4.4 Connection Between Entropic Functionals, Fokker–Planck
Equations and H -theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.5 Many-Body Interacting Systems with Overdamping . . . . . . . . . . . . 153
4.5.1 Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
4.5.2 Thermodynamics, Including Clausius Relation,
First and Second Principles, Equation of States,
and Carnot Cycle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
4.5.3 Zeroth Principle of Thermodynamics . . . . . . . . . . . . . . . . . 160
4.5.4 An Extended Model with Repulsive Power-Law
Short-Ranged Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
4.5.5 Model with Direction-Depending Drag . . . . . . . . . . . . . . . . 164
4.5.6 Plasma in Spherical Capacitor with Overdamping . . . . . . 166
4.6 Generalizing the Langevin Equation . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Contents xv

4.7 Other Complex Dissipative Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 171

4.7.1 Coherent Noise Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
4.7.2 Ehrenfest Dog-Flea Model . . . . . . . . . . . . . . . . . . . . . . . . . . 172
4.7.3 Random Walk Avalanches . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
4.7.4 Kuramoto Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
4.8 Probabilistic Models with Strong Correlations—Numerical
and Analytical Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
4.8.1 The MTG Model and Its Numerical Approach . . . . . . . . . 182
4.8.2 The TMNT Model and Its Numerical Approach . . . . . . . . 187
4.8.3 Analytical Approach of the MTG and TMNT
Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
4.8.4 The MFMT Models and Their Numerical Approach . . . . 193
4.8.5 The RST1 Model and Its Analytical Approach . . . . . . . . . 195
4.8.6 The RST2 Model and Its Numerical Approach . . . . . . . . . 199
4.8.7 The HTT Model and Its Analytical Approach . . . . . . . . . . 200
4.8.8 Comparative Considerations . . . . . . . . . . . . . . . . . . . . . . . . . 202
4.9 Central Limit Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
4.10 Large Deviation Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
4.10.1 Purely Probabilistic Models . . . . . . . . . . . . . . . . . . . . . . . . . 214
4.10.2 Dynamical Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
4.11 Time-Dependent Ginzburg–Landau d-Dimensional O(n)
Ferromagnet with n = d . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
5 Deterministic Dynamical Foundations of Nonextensive
Statistical Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
5.1 Dissipative Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
5.1.1 One-Dimensional Dissipative Maps . . . . . . . . . . . . . . . . . . 229
5.1.2 Two-Dimensional Dissipative Maps . . . . . . . . . . . . . . . . . . 245
5.1.3 High-Dimensional Dissipative Maps . . . . . . . . . . . . . . . . . . 245
5.2 Conservative Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
5.2.1 Strongly Chaotic Two-Dimensional Conservative
Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
5.2.2 Weakly Chaotic Two-Dimensional Conservative
Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
5.2.3 Strongly Chaotic Four-Dimensional Conservative
Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
5.2.4 High-Dimensional Conservative Maps . . . . . . . . . . . . . . . . 260
5.3 Many-Body Long-Range-Interacting Hamiltonian Systems . . . . . . 263
5.3.1 Planar Rotators (Inertial XY-Like Ferromagnetic
Model) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
5.3.2 Three-Dimensional Rotators (Inertial
Heisenberg-Like Ferromagnetic Model) . . . . . . . . . . . . . . . 285
5.3.3 Fermi-Pasta-Ulam-Like Model for Coupled
Oscillators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
5.4 The q-Triplet and More . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
xvi Contents

5.5 Connection with Critical Phenomena . . . . . . . . . . . . . . . . . . . . . . . . . 315

5.6 A Conjecture on the Time and Size Dependences of Entropy . . . . . 316
6 Generalizing Nonextensive Statistical Mechanics . . . . . . . . . . . . . . . . . . 317
6.1 Crossover Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
6.2 Further Generalizing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
6.2.1 Spectral Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
6.2.2 Beck-Cohen Superstatistics . . . . . . . . . . . . . . . . . . . . . . . . . 325

Part III Applications or What for the Theory Works

7 Thermodynamical and Nonthermodynamical Applications . . . . . . . . . 331
7.1 Physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
7.1.1 Cold Atoms in Optical Lattices . . . . . . . . . . . . . . . . . . . . . . 333
7.1.2 Trapped Ions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
7.1.3 Granular Matter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
7.1.4 High-Energy Physics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
7.1.5 Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
7.1.6 Fingering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354
7.1.7 Condensed Matter Physics . . . . . . . . . . . . . . . . . . . . . . . . . . 355
7.1.8 Plasma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
7.1.9 Astrophysics and Astronomy . . . . . . . . . . . . . . . . . . . . . . . . 360
7.1.10 Geophysics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
7.1.11 Quantum Chaos . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
7.1.12 Quantum Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
7.1.13 Random Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 383
7.1.14 Nonlinear Quantum Mechanics . . . . . . . . . . . . . . . . . . . . . . 384
7.2 Chemistry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
7.2.1 Generalized Arrhenius Law and Anomalous
Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
7.2.2 Lattice Lotka–Volterra Model for Chemical
Reactions and Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
7.2.3 Re-association in Folded Proteins . . . . . . . . . . . . . . . . . . . . 392
7.2.4 Ground State Energy of the Chemical Elements
(Mendeleev’s Table) and of Doped Fullerenes . . . . . . . . . . 394
7.2.5 Compton Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
7.3 Economics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 399
7.3.1 Inter-Occurrence Times in Economics,
Geophysics, Genome Structure, Civil Engineering,
and Turbulence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
7.4 Computer Sciences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
7.4.1 Optimization Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
7.4.2 Analysis of Time Series and Signals . . . . . . . . . . . . . . . . . . 414
7.4.3 Analysis of Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 415
7.4.4 Ping Internet Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . 418
7.5 Biosciences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
Contents xvii

7.6 Cellular Automata . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427

7.7 Self-Organized Criticality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428
7.8 Asymptotically Scale-Free Networks . . . . . . . . . . . . . . . . . . . . . . . . . 431
7.8.1 The Natal Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
7.8.2 Albert–Barabási Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
7.8.3 Non-growing Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
7.8.4 Connection Between Asymptotically
Scale-Invariant Networks with Weighted Links
and q-Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
7.8.5 Lennard-Jones Cluster . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456
7.9 Linguistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
7.10 Other Sciences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 458

Part IV Last (But Not Least)

8 Final Comments and Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
8.1 Falsifiable Predictions and Conjectures, and Their
Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
8.1.1 The Scaling Relation γ = 3−q 2
....................... 471
8.1.2 q-Gaussian Distributions of Velocities . . . . . . . . . . . . . . . . 472
8.1.3 Generalized Central Limit Theorem Leading
to Stable q-Gaussian Distributions . . . . . . . . . . . . . . . . . . . 473
8.1.4 Existence of q, λq and K q , and the Identity K q = λq . . . . 473
8.1.5 Scaling with N ∗ for Long-Range-Interacting
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473
8.1.6 Vanishing Lyapunov Spectrum for Classical
Long-Range-Interacting Many-Body Hamiltonian
Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474
8.1.7 Nonuniform Convergence
for Long-Range Hamiltonians Associated
with the (N , t) → (∞, ∞) Limits . . . . . . . . . . . . . . . . . . . . 474
8.1.8 q-Duplets, q-Triplets and q-N plets . . . . . . . . . . . . . . . . . . . 475
8.1.9 Degree Distributions of the q-Exponential Type
for Scale-Free Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . 475
8.2 Frequently Asked Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 476
8.3 Open Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497

Appendix A: Useful Mathematical Formulae . . . . . . . . . . . . . . . . . . . . . . . . . 499

Appendix B: Escort Distributions and q-Expectation Values . . . . . . . . . . 509
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 517
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 567
 Part I
Basics or How the Theory Works
Chapter 1
Historical Background and Physical
Motivations

Beauty is the first test:

there is no permanent place in the world for ugly mathematics.
G.H. Hardy
(A Mathematician’s Apology, 1941)

1.1 An Overall Perspective

Let us start by placing the content of the present book in the scenario of contempo-
rary physics. Its present theoretical pillars are usually considered to be mechanics—
classical, special relativity, general relativity, and quantum—Maxwell electromag-
netism, and Boltzmann–Gibbs (BG) statistical mechanics, which essentially incor-
porates theory of probabilities within the realm of mechanics and electromagnetism.
Moreover, the BG theory connects, for large systems, to classical thermodynamics.
In what concerns mechanics, the corresponding regions of physical applicability
of those theories are as follows.

– Systems with masses m neither too small nor too large and velocities v small
compared to the speed of light c (e.g., a falling apple) obey Newton’s classical
mechanics.
– Systems with masses neither too small nor too large and velocities v comparable
to c (e.g., the Earth’s orbit around the Sun, the Solar system’s orbit around the
galaxy, the surface of the Sun) obey Einstein’s special relativity, which recovers
classical mechanics as the limiting case v/c → 0.
– Systems with large and compact masses and velocities comparable to c (e.g.,
neutron star, black hole) obey Einstein’s general relativity, which recovers special
relativity as the limiting case Gm/(c2 R) → 0, where G and R are, respectively,
Newton’s gravitation constant and its (linear) size [14].
– Systems with masses very small and velocities small compared to c (e.g., an
atomic electron in orbits distant from the nucleus) obey nonrelativistic quantum
© Springer Nature Switzerland AG 2023 3
C. Tsallis, Introduction to Nonextensive Statistical Mechanics,
https://doi.org/10.1007/978-3-030-79569-6_1
4 1 Historical Background and Physical Motivations

mechanics, namely Schroedinger equation, which recovers classical mechanics as

the limiting case h/mv → 0, where h is Planck’s constant.
– Systems with masses very small and velocities comparable to c (e.g., an atomic
electron in orbits close to the nucleus) obey relativistic quantum mechanics, namely
Klein–Gordon and Dirac equations, for spinless particles and particles with spin,
respectively.

Let us focus now on statistical mechanics.

– Thermostatistical systems with local space-time correlations obey Boltzmann–

Gibbs statistical mechanics, constructed upon the Boltzmann–Gibbs-von
Neumann-Shannon additive entropy S BG . Important examples include distinguish-
able particles with short-range interactions, indistinguishable particles with half-
integer spin (Fermi–Dirac statistics), indistinguishable particles with integer spin
(Bose–Einstein statistics), strongly chaotic nonlinear dynamical systems, second-
order critical phenomena close to Tc , random geometrical systems such as bond
percolation (Kasteleyn and Fortuin theorem [15]) and self-avoiding random walk
(de Gennes isomorphism [16]).
– Thermostatistical systems with nonlocal (sometimes referred to as global) space-
time correlations obey non-Boltzmann–Gibbs statistical mechanics, e.g., nonex-
tensive statistical mechanics (q-statistics for short), constructed upon nonadditive
entropies such as Sq with S1 = S BG . Typical examples include distinguishable
particles with long-range interactions (e.g., gravitation, plasmas), weakly chaotic
nonlinear dynamical systems, second-order critical phenomena at Tc , random geo-
metrical systems such as (asymptotically) scale-invariant networks. The present
book is dedicated to this generalization, which recovers BG statistical mechanics
as the limiting case (1 − q)/k B → 0, where k B is Boltzmann’s constant.

The word “nonextensive” that—after some hesitation—I eventually adopted, in

the title of the book and elsewhere, to refer to the present specific generalization of
BG statistical mechanics may—and occasionally does—cause some confusion, and
surely deserves clarification.
The whole theory is based on a single concept, namely the entropy noted Sq which,
for the entropic index q equal to unity, reproduces the standard BG entropy.
The traditional functional S BG is said to be additive [17]. Indeed, for a system
composed of any two (probabilistically) independent subsystems, the entropy S BG
of the sum coincides with the sum of the entropies. The entropy Sq (q = 1) violates
this property, and is therefore nonadditive. As we see, entropic additivity depends,
from its very definition, only on the functional form of the entropy in terms of
probabilities. The situation is generically quite different for the thermodynamical
concept of extensivity. An entropy of a system or of a subsystem is said extensive
if, for a large number N of its elements (probabilistically independent or not), the
entropy is (asymptotically) proportional to N . Otherwise, it is nonextensive. This is to
say, extensivity depends on both the mathematical form of the entropic functional and
the correlations possibly existing within the elements of the system. Consequently,
1.1 An Overall Perspective 5

for a (sub)system whose elements are either independent or locally correlated, the
additive entropy S BG is extensive, whereas the nonadditive entropy Sq (q = 1) is
nonextensive. In contrast, however, for a (sub)system whose elements are generically
nonlocally correlated, the additive entropy S BG is typically nonextensive, whereas the
nonadditive entropy Sq (q = 1) can be extensive for a special value of q. Probabilistic
systems exist such that Sq is not extensive for any value of q, either q = 1 or q = 1.
All these statements are further detailed in the body of the book1 .
We shall also see that, consistently, the index q appears to characterize universality
classes of nonadditivity, by phrasing this concept similarly to what is done in the
standard theory of critical phenomena. Within each class, one expects to find infinitely
many dynamical systems.
Coming back to the name nonextensive statistical mechanics, would it not be
more appropriate to call it nonadditive statistical mechanics? Certainly yes if one
focuses on the entropy that is being used. However, there is on one hand the fact that
the expression nonextensive statistical mechanics is by now spread in thousands of
papers. There is, on the other hand, the fact that important systems whose approach
benefits from the present generalization of the BG theory are long-range-interacting
many-body Hamiltonian systems. For such systems, the total energy is well known
to be nonextensive, even if the extensivity of the entropy can be preserved by con-
veniently choosing the value of the index q. Summarizing, since Gibbs coined the
expression, this branch of contemporary theoretical physics is referred to through the
composition of two words, namely statistical and mechanics. The word statistical
definitively refers to probabilities, and consistently to entropy; the word mechanics
refers instead to mechanical concepts, hence, inevitably, to energy. For naming the
present generalization, we could focus on entropy and call it nonadditive statisti-
cal mechanics, or alternatively focus on energy and call it nonextensive statistical
mechanics. For historical reasons the latter prevailed.
Still at the linguistic and semantic levels, should we refer to Sq as an entropy
or rather as an entropic functional or entropic form? And, even before that, why
should such a minor-looking point have any relevance in the first place? The point is
that, in physics, since more than one century, only one entropic functional has been
unanimously considered “physical” in the thermodynamical sense, namely the BG
one. In other areas, such as cybernetics, control theory, nonlinear dynamical systems,
information theory, many other (by now over fifty!) entropic functionals differing
from the Shannon one (which precisely coincides with the BG one) have been pro-
posed, studied, and occasionally used. In the physical community only the BG form
is unquestionably admitted as physically meaningful because of its deep connections

1 During more than one century, physicists have primarily addressed locally interacting systems,
and therefore the entropic form which satisfies the thermodynamical requirement of extensivity
is S BG . A regretful consequence of this fact is that entropic additivity and extensivity have been
practically—and wrongly—considered as synonyms in many communities, thus generating all
kinds of confusions and inadvertences. For example, by mere inadvertence, our own book enti-
tled Nonextensive Entropy—Interdisciplinary Applications [18] should definitively have been more
appropriately entitled Nonadditive Entropy—Interdisciplinary Applications! Indeed, already in its
first chapter, an example is shown where the nonadditive entropy Sq (q = 1) is extensive.
6 1 Historical Background and Physical Motivations

with classical thermodynamics. So, what about Sq in this specific context? A variety
of thermodynamical arguments—extensivity, Clausius inequality, zeroth, first and
second principles of thermodynamics, Carnot cycle, and others—that are presented
later on, definitively point Sq as a physical entropy in a strongly analogous sense
that S BG surely is. This issue is deeply related to the Einstein principle of likelihood
factorization, to the Legendre transformation structure of thermodynamics, and to
the large deviation theory, as we shall further elaborate in the body of the book.
Complexity is nowadays a frequently used yet poorly defined—at least quantita-
tively speaking—concept. It tries to embrace a great variety of scientific and tech-
nological approaches of all types of natural, artificial, and social systems. A name,
plectics, has been coined by Murray Gell-Mann to refer to this emerging science [19].
One of the main—necessary but by no means sufficient—features of complexity has
to do with the fact that both very ordered and very disordered systems are, in the
sense of plectics, considered to be simple, not complex. Ubiquitous phenomena, such
as the origin of life and languages, the growth of cities and computer networks, cita-
tions of scientific papers, co-authorships and co-actorships, displacements of living
beings, financial fluctuations, turbulence, are frequently considered to be complex
phenomena. They all seem to occur close, in some sense, to the frontier between
order and disorder. Most of their basic quantities exhibit nonexponential behaviors,
very frequently power-laws. It happens that the distributions and other relevant quan-
tities that emerge naturally within the frame of nonextensive statistical mechanics
are precisely of this type, becoming of the exponential type in the q = 1 limit. One
of the most typical dynamical situations has to do with the edge of chaos, occurring
in the frontier between regular motion and standard chaos. Since these two typi-
cal regimes would clearly be considered “simple” in the sense of plectics, one is
strongly tempted to consider as “complex” the regime in between, which has some
aspects of the disorder associated with strong chaos but also some of the order lurking
nearby.2 Nonextensive statistical mechanics turns out to be appropriate precisely for
that intermediate region, thus suggesting that the entropic index q could be a conve-
nient manner for quantifying some relevant aspects of complexity, surely not in all
cases but probably so for vast classes of systems. Regular motion and chaos are time
analogs for the space configurations occurring, respectively, in crystals and fluids.
In this sense, the edge of chaos would be the analog of quasi-crystals, glasses, spin-
glasses, and other amorphous, typically metastable, structures. One does not expect
statistical concepts to be intrinsically useful for regular motions and regular struc-
tures. On the contrary, one naturally tends to use probabilistic concepts for chaos
and fluids. These probabilistic concepts and their associated entropy, S BG , would
typically be the realm of BG statistical mechanics and standard thermodynamics.
It appears that, in the marginal cases, or at least in very many of them, between

2 It is frequently encountered nowadays the belief that complexity emerges typically at the edge of
chaos. For instance, the final words of the Abstract of a lecture delivered in September 2005 by
Leon O. Chua at the Politecnico di Milano were “Explicit mathematical criteria will be given to
identify a relatively small subset of the locally active parameter region called the edge of chaos
where most of the complex phenomena emerge” [20].
1.1 An Overall Perspective 7

great order and great disorder, the statistical procedures can still be used. However,
the associated natural entropy would not anymore be the BG one, but typically Sq
with q = 1. It then appears quite naturally the scenario within which BG statisti-
cal mechanics is the microscopic thermodynamical description properly associated
with smooth geometry (typically, although not necessarily, the plain Euclidean one),
whereas nonextensive statistical mechanics would be the proper counterpart which
has privileged connections with (multi)fractal and similar, hierarchical, statistically
scale-invariant structures (at least asymptotically speaking). As already mentioned,
a paradigmatic case would be those nonlinear dynamical systems whose largest Lya-
punov exponent is neither negative (easily predictable systems) nor positive (strong
chaos) but vanishing instead, i.e., the edge of chaos (weak chaos3 ). Standard, equi-
librium critical phenomena also deserve a special comment. Indeed, I have always
liked to think and say that “criticality is a little window through which one can see
the nonextensive world”. Many people have certainly had similar insights. Alberto
Robledo, Filippo Caruso, and myself have exhibited some rigorous evidences—to be
discussed later on—along this line. Not that there is anything wrong with the usual
and successful use of BG concepts to discuss the neighborhood of criticality in many
cooperative systems at thermal equilibrium! But, if one wants to make a delicate
quantification and classification of some of the physical concepts precisely at the
critical point, the nonextensive language appears to be a privileged one for this task.
It may be so for many anomalous systems. Paraphrasing Angel Plastino’s (A. Plas-
tino Sr.) last statement in his lecture at the 2003 Villasimius meeting, “for different
sizes of screws one must use different screwdrivers”! We cannot avoid remembering,
at this point, Sadi Carnot’s famous words “When a hypothesis no longer suffices to
explain phenomena, it should be abandoned”.
A proposal of a generalization of the BG entropy as the physical basis for dealing,
in statistical-mechanical terms, with some wide classes of complex systems might—
in the view of many—in some sense imply in a new paradigm, whose validity may
be further consolidated by future progress and verifications. Indeed, we shall argue
along the entire book that q is, for time evolving systems, determined a priori by
the microscopic (or associated mesoscopic) dynamics of the system. This is in some
sense less innocuous than it looks at first sight. Indeed, this means that the entropy
to be used for thermostatistical purposes would be not universal but would depend
on the system or, more precisely, on the nonadditive universality class to which the
system belongs4 . Whenever a new scientific viewpoint is proposed, either correct
or wrong, it usually attracts quite extreme opinions. One of the questions that is
regularly asked is the following: “Do I really need this? Is it not possible to work

3 In the present book, the expression “weak chaos” is used in the sense of a sensitivity to the initial
conditions diverging with time slower than exponentially, and not in other senses occasionally used
in the theory of nonlinear dynamical systems.
4 Leo Tolstoy’s 1877 novel Anna Karenina begins: All happy families are alike; each unhappy

family is unhappy in its own way. This dramatic sentence can be used as a transparent metaphor for
the present generalization of the BG entropy. Indeed, there is only one manner of elements being
essentially independent (q = 1), whereas there are infinitely many manners of being nontrivially
correlated (q = 1).
8 1 Historical Background and Physical Motivations

all this out just with the concepts that we already have, and that have been lengthily
tested?”. This type of question is rarely easy to answer, because it involves the proof
without ambiguity that some given result can by no means be obtained within the
traditional theory.
However, let me present an analogy, basically due to Michel Baranger [21, 22],
in order to clarify at least one of the aspects that are relevant for this nontrivial prob-
lem. Suppose one only knows how to handle straight lines and segments and wants to
calculate areas delimited by curves. Does one really need the Newton-Leibnitz dif-
ferential and integral calculus? Well, one might approach the result by approximating
the curve with polygonals, and that works reasonably well in most cases. However,
if one wants to better approach reality, one would consider more and more, shorter
and shorter, straight segments. But one would ultimately want to take an infinity of
such infinitely small segments. If one does so, then one has precisely jumped into
the standard differential and integral calculus! How big, epistemologically speaking,
was that step is a matter of debate, but its practicality is out of question. The curve
that is handled might, in particular, be a straight line itself (or a finite number of
straight pieces). In this case, there is of course no need to do the limiting process.
Let me present a second analogy, this one primarily due to Angel Ricardo Plas-
tino (A. Plastino Jr.). It was known by ancient astronomers that the apparent orbits
of stars are circles, a form that was considered geometrically “perfect”. The prob-
lematic orbits were those of the planets, for instance that of Mars. Ptolemy proposed
a very ingenious way out, the epicycles, i.e., circles turning around circles. The pre-
dictions became of great precision, and astronomers along centuries developed, with
sensible success, the use of dozens of epicycles, each one on top of the previous one.
It remained so until the proposal of Johannes Kepler: the orbits are well described
by ellipses, a form which generalizes the circle by having an extra parameter, the
eccentricity. The eccentricities of the various planets were determined through fitting
with the observational data. Today we know, through Newtonian mechanics, that it
is possible to compute numerically the planetary orbits, including the changes in the
eccentricities, and in the other orbital elements, arising from the gravitational inter-
action between the planets. To do that, one needs to know the masses of the planets,
and their positions and velocities at a given time (in practice, all these quantities can
be regarded as fitting parameters to be determined from observations). Kepler and
his immediate followers, lacking the necessary observational, theoretical, and com-
putational resources, just fitted the orbital elements, using, however, the appropriate
functional forms, i.e., the Keplerian ellipses. In few years, virtually all European
astronomers abandoned the use of the complex Ptolemaic epicycles and adopted the
simple Keplerian orbits. We know today, through Fourier transform, that the periodic
motion on one ellipse is totally equivalent to an infinite number of specific circular
epicycles. So we can proceed either way. It is clear, however, that an ellipse is by far
more practical and concise, even if in principle it can be thought as infinitely many
circles. We must concomitantly “pay the price” of an extra parameter, the elliptical
eccentricity.
1.2 Introduction 9

1.2 Introduction

Let us consider the free surface of a glass covering a table. And let us idealize
it as being planar. What is its volume? Clearly zero since it has no height. An
uninteresting answer to an uninteresting question. What is its length? Clearly infinity.
One more uninteresting answer to another uninteresting question. Now, if we ask
what is its area, we will have a meaningful answer, say 2 m2 . A finite answer. Not
zero, not infinity—correct but poorly informative features. A finite answer for a
measurable quantity, as expected from good theoretical physics, good experimental
physics, and good mathematics. Who “told” us that the interesting question for this
problem was the area? The system did! Its planar geometrical nature did. If we
were focusing on a fractal, the interesting question would of course be its measure
in d f dimensions, d f being the corresponding fractal or Hausdorff dimension. Its
measure in any dimension d larger than d f is zero, and in any dimension smaller than
d f is infinity. Only the measure at precisely d f dimensions yields a finite number.
For instance, if we consider an ideal 10 cm long straight segment, and we proceed
through the celebrated Cantor-set construction (i.e., eliminate the central third of
the segment, and then also eliminate the central third of each of the two remaining
thirds, and hypothetically continue doing this for ever) we will ultimately arrive to
a remarkable geometrical set—the triadic Cantor-set—which is embedded in a one-
dimensional space but whose Lebesgue measure is zero. The fractal dimension of
this set is d f = ln 2/ ln 3 = 0.6309... Therefore the interesting information about our
present hypothetical system is that its measure is (10 cm)0.6309...  4.275 cm0.6309 .
And, interestingly enough, the specific nature of this valuable geometric information
was mandated by the system itself!
This entire book is written within precisely this philosophy: it is the natural (or
artificial or social) system itself which, through its geometrical-dynamical properties,
determines the specific informational tool—entropy—to be meaningfully used for
the study of its (thermo) statistical properties. The reader surely realizes that this
epistemological standpoint somehow involves what some consider as a kind of new
paradigm for statistical mechanics and related areas. Indeed, the physically important
entropy—a crucial concept—is not thought as being an universal functional that is
given once for ever, but it rather is a delicate and powerful concept to be carefully
constructed for classes of systems. In other words, we adopt here the viewpoint that
the—simultaneously aesthetic and fruitful—way of thinking about this issue is the
existence of universality classes of systems. These systems share the same functional
connection between the entropy and the set of probabilities of their microscopic
states. The most known such universality class is that which we shall refer to as the
Boltzmann–Gibbs (BG) one. Its associated entropy is given (for a set of W discrete
states) by
W
S BG = −k pi ln pi , (1.1)
i=1
10 1 Historical Background and Physical Motivations

with

W
pi = 1 , (1.2)
i=1

and where k is some conventional positive constant. This constant is taken to be

Boltzmann constant k B in thermostatistics, and is usually taken equal to unity for
informational or computational purposes. In this book we shall use, without further
clarification, one or the other of these two conventions, depending on the particular
convenience. The reader will unambiguously detect which convention we are using
within any specific context. For the particular case of equal probabilities (i.e., pi =
1/W , ∀i), Eq. (1.1) becomes
S BG = k ln W, (1.3)

which is carved on Boltzmann’s grave in Vienna by suggestion of Planck. This cele-

brated expression, as well as Eq. (1.1), have been used in a variety of creative manners
by Planck, Einstein, von Neumann, Shannon, Szilard, Jaynes, Tisza, and others. Eq.
(1.1) has the following remarkable property. If we compose two probabilistically
independent subsystems A and B (with numbers of states, respectively, denoted by
W A and W B ), i.e., such that the joint probabilities factorize, piA+B
j = piA p Bj (∀(i, j)),
the entropy S BG is additive5 [17]. By this we mean that

S BG (A + B) = S BG (A) + S BG (B), (1.4)

where


WA 
WB
S BG (A + B) ≡ −k piA+B
j ln piA+B
j (with W A+B = W A W B ), (1.5)
i=1 j=1


WA
S BG (A) ≡ −k piA ln piA , (1.6)
i=1

and

WB
S BG (B) ≡ −k p Bj ln p Bj . (1.7)
j=1

Expression (1.1) was first proposed (in its form for simple continuous systems)
by Boltzmann [7, 8] in the 1870s, and was then refined by Gibbs [1–3] for more
general systems. It is the basis of the usual, BG statistical mechanics. In particular,
its optimization under appropriate constraints (that we shall describe later on) yields,
for a system in thermal equilibrium with a thermostat at temperature T , the celebrated
BG factor or weight, namely

5 The important mathematical distinction between additive and extensive is addressed later on.
1.3 Background and Indications in the Literature 11

e−β Ei
pi = (1.8)
Z BG

with
β ≡ 1/kT, (1.9)


W
Z BG ≡ e−β E j , (1.10)
j=1

and where {E i } denotes the energy spectrum of the system, i.e., the eigenvalues of
the Hamiltonian of the system with the adopted boundary conditions; Z BG is referred
to as the partition function.
Expressions (1.1) and (1.8) are the landmarks of BG statistical mechanics, and
are vastly and successfully used in physics, chemistry, mathematics, computational
sciences, engineering, and elsewhere. Since their establishment, over 140 years ago,
they constitute fundamental pieces of contemporary physics. Though notoriously
applicable in very many systems and situations, we believe that they need to be mod-
ified (generalized) in others, in particular in most of the so-called complex systems
(see, for instance, [19, 23–26]). In other words, there are nowadays strong reasons
to believe that they are not universal, as somehow implicitly (or explicitly) thought
until not long ago by many physicists. They must have in fact a restricted domain
of validity, as any other human intellectual construct; as Newtonian mechanics,
nonrelativistic quantum mechanics, special relativity, Maxwell electromagnetism,
and all others. The basic purpose of this book is precisely to explore—the best that
our present knowledge allows—for what systems and conditions the BG concepts
become either inefficiently applicable or nonapplicable at all, and what might be
done in such cases, or at least in (definitively wide) classes of them. The possibility
of some kind of generalization of BG statistical concepts, or at least some intuition
about the restricted validity of such concepts, already emerged along the years, in
one way or another, in the mind of various physicists or mathematicians. This is, at
least, what one is led to think from the various statements that we reproduce in the
next Section.

1.3 Background and Indications in the Literature

We recall here interesting points raised along the years by various thinkers on the
theme of the foundations and domain of validity of the concepts that are currently
used in standard statistical mechanics.
Boltzmann himself wrote, in his 1896 Lectures on Gas Theory [27], the following
words: (The bold faces in this and subsequent quotations are mine.)
When the distance at which two gas molecules interact with each other noticeably is vanish-
ingly small relative to the average distance between a molecule and its nearest neighbor—or,
12 1 Historical Background and Physical Motivations

as one can also say, when the space occupied by the molecules (or their spheres of action)
is negligible compared to the space filled by the gas—then the fraction of the path of each
molecule during which it is affected by its interaction with other molecules is vanishingly
small compared to the fraction that is rectilinear, or simply determined by external forces.
[...] The gas is “ideal” in all these cases.

Boltzmann is here referring essentially to the hypothesis of ideal gas. It shows nev-
ertheless how clear it was in his mind the relevance of the range of the interactions
for the thermostatistical theory he was putting forward.

Gibbs, in the Preface of his celebrated 1902 Elementary Principles in Statisti-

cal Mechanics—Developed with Especial Reference to the Rational Foundation of
Thermodynamics [1–3], wrote the following touching words:
Certainly, one is building on an insecure foundation, who rests his work on hypotheses
concerning the constitution of matter.

Difficulties of this kind have deterred the author from attempting to explain the mys-
teries of nature, and have forced him to be contented with the more modest aim of deducing
some of the more obvious propositions relating to the statistical branch of mechanics.

In these lines, Gibbs not only shares with us his epistemological distress about the
foundations of the science he, Maxwell, and Boltzmann are founding. He also gives a
precious indication that, in his mind, this unknown foundation would certainly come
from mechanics. Everything indicates that this was also the ultimate understanding
of Boltzmann, who—unsuccessfully—tried his entire life (the so-called Boltzmann’s
program) to derive statistical mechanics from Newtonian mechanics. In fact, Boltz-
mann’s program remains unconcluded until today!

As we see next, the same understanding permeates in the words of Einstein that
we cite next, from his 1910 paper [28]:

Usually W is set equal to the number of ways (complexions) in which a state, which is incom-
pletely defined in the sense of a molecular theory (i.e., coarse grained), can be realized. To
compute W one needs a complete theory (something like a complete molecular-mechanical
theory) of the system. For that reason it appears to be doubtful whether Boltzmann’s prin-
ciple alone, i.e., without a complete molecular-mechanical theory (Elementary theory) has
any real meaning. The equation S = k log W + const. appears [therefore], without an
Elementary theory—or whatsoever one wants to call it—devoid of any meaning from a
phenomenological point of view.6

6 Most of the translation is due to E.G.D. Cohen [29]. A slightly different translation also is avail-
able: [“Usually W is put equal to the number of complexions... In order to calculate W , one needs
a complete (molecular-mechanical) theory of the system under consideration. Therefore it is dubi-
ous whether the Boltzmann principle has any meaning without a complete molecular-mechanical
theory or some other theory which describes the elementary processes. S = N R
log W + const.
seems without content, from a phenomenological point of view, without giving in addition such
an Elementartheorie.” (Translation: Abraham Pais, Subtle is the Lord..., Oxford University Press,
1982)].
1.3 Background and Indications in the Literature 13

By Boltzmann’s principle—expression coined apparently by Einstein himself—

the author refers precisely to the logarithmic form for the entropy that he explicitly
writes down a few words later. It is quite striking the crucial role that Einstein
attributes to microscopic dynamics for giving a clear sense to that particular form
for the entropy.

Coming back to Gibbs’s book [1–3], in page 35 he writes:

In treating of the canonical distribution, we shall always suppose the multiple integral in
equation (92) [the partition function, as we call it nowadays] to have a finite value, as
otherwise the coefficient of probability vanishes, and the law of distribution becomes
illusory. This will exclude certain cases, but not such apparently, as will affect the value of
our results with respect to their bearing on thermodynamics. It will exclude, for instance,
cases in which the system or parts of it can be distributed in unlimited space [...]. It also
excludes many cases in which the energy can decrease without limit, as when the system
contains material points which attract one another inversely as the squares of their
distances. [...]. For the purposes of a general discussion, it is sufficient to call attention to
the assumption implicitly involved in the formula (92).

Clearly, Gibbs is well aware that the theory he is developing has limitations. It does
not apply to anomalous cases such as gravitation.

Enrico Fermi, in his 1936 Thermodynamics [30] wrote:

The entropy of a system composed of several parts is very often equal to the sum of the
entropies of all the parts. This is true if the energy of the system is the sum of the energies of
all the parts and if the work performed by the system during a transformation is equal to the
sum of the amounts of work performed by all the parts. Notice that these conditions are not
quite obvious and that in some cases they may not be fulfilled. Thus, for example, in the
case of a system composed of two homogeneous substances, it will be possible to express
the energy as the sum of the energies of the two substances only if we can neglect the surface
energy of the two substances where they are in contact. The surface energy can generally be
neglected only if the two substances are not very finely subdivided; otherwise, it can play
a considerable role

So, Fermi says “very often”, which virtually implies “not necessarily so”!

Ettore Majorana, mysteriously missing since 25 March 1938, wrote [31]:

This is mainly because entropy is an addditive quantity as the other ones. In other words, the
entropy of a system composed of several independent parts is equal to the sum of entropy of
each single part. [...] Therefore one considers all possible internal determinations as equally
probable. This is indeed a new hypothesis because the universe, which is far from being in
the same state indefinitively, is subjected to continuous transformations. We will therefore
admit as an extremely plausible working hypothesis, whose far consequences could
sometime not be verified, that all the internal states of a system are a priori equally
probable in specific physical conditions. Under this hypothesis, the statistical ensemble
associated to each macroscopic state A turns out to be completely defined.

Like Fermi, Majorana leaves the door open to other, nonstandard, possibilities, which
could be not inconsistent with the methods of statistical mechanics.
14 1 Historical Background and Physical Motivations

Claude Elwood Shannon, in his 1948/1949 The Mathematical Theory of Commu-

nication [32], justified the logarithmic entropy k ln W in these plain terms:

It is practically more useful. [...] It is nearer to our intuitive feeling as to the proper measure.
[...] It is mathematically more suitable. [...].

And, after stating the celebrated axioms that yield, as unique answer, the entropy
(1.1), he wrote:
This theorem and the assumptions required for its proof, are in no way necessary for the
present theory. It is given chiefly to lend a certain plausibility to some of our later defini-
tions. The real justification of these definitions, however, will reside in their implications.

It is certainly remarkable how wide Shannon leaves the door open to other
entropies than the one that he brilliantly discussed.

Laszlo Tisza wrote, in 1961, in his Generalized Thermodynamics [33]:

The situation is different for the additivity postulate P a2, the validity of which can-
not be inferred from general principles. We have to require that the interaction energy
between thermodynamic systems be negligible. This assumption is closely related to the
homogeneity postulate P d1. From the molecular point of view, additivity and homogeneity
can be expected to be reasonable approximations for systems containing many particles,
provided that the intermolecular forces have a short-range character.

Peter Landsberg wrote, in 1978/1990, in his Thermodynamics and Statistical

Mechanics [34]:
The presence of long-range forces causes important amendments to thermodynamics,
some of which are not fully investigated as yet.

And in 1984 he added [35]

[...] in the case of systems with long-range forces and which are therefore nonextensive (in
some sense) some thermodynamic results do not hold. [...] The failure of some thermody-
namic results, normally taken to be standard for black hole and other nonextensive systems
has recently been discussed. [...] If two identical black holes are merged, the presence of
long-range forces in the form of gravity leads to a more complicated situation, and the
entropy is nonextensive.

Nico van Kampen, in his 1981 Stochastic Processes in Physics and Chemistry
[36], wrote:
Actually an additional stability criterion is needed, see M.E. Fisher, Archives Rat. Mech.
Anal. 17, 377 (1964); D. Ruelle, Statistical Mechanics: Rigorous Results (Benjamin, New
York 1969). A collection of point particles with mutual gravitation is an example where
this criterion is not satisfied, and for which therefore no statistical mechanics exists.

L.G. Taff writes in his 1985 Celestial Mechanics [37]:

1.3 Background and Indications in the Literature 15

[...] This means that the total energy of any finite collection of self-gravitating mass points
does not have a finite, extensive (e.g., proportional to the number of particles) lower bound.
Without such a property there can be no rigorous basis for the statistical mechanics of
such a system (Fisher and Ruelle 1966). Basically it is that simple. One can ignore the fact
that one knows that there is no rigorous basis for one’s computer manipulations; one can try
to improve the situation, or one can look for another job.

Needless to say that the very existence of the present book constitutes but an attempt
to improve the situation!

The same point is addressed by W.C. Saslaw in his 1985 Gravitation Physics of
Stellar and Galactic Systems [38]:
When interactions are important the thermodynamic parameters may lose their simple
intensive and extensive properties for subregions of a given system. [...] Gravitational
systems, as often mentioned earlier, do not saturate and so do not have an ultimate equi-
librium state.

Radu Balescu, in his 1975 Equlibrium and Nonequilibrium Statistical Mechanics

[39], wrote:
It therefore appears from the present discussion that the mixing property of a mechanical
system is much more important for the understanding of statistical mechanics than the
mere ergodicity. [...] A detailed rigorous study of the way in which the concepts of mixing
and the concept of large numbers of degrees of freedom influence the macroscopic laws
of motion is still lacking.

David Ruelle writes in page 1 of his 1978 Thermodynamical Formalism [40] (and
maintains in page 1 of his 2004 Edition [41]):
The formalism of equilibrium statistical mechanics—which we shall call thermodynamic
formalism—has been developed since J.W. Gibbs to describe the properties of certain
physical systems. [...] While the physical justification of the thermodynamic formalism
remains quite insufficient, this formalism has proved remarkably successful at explaining
facts.

as well as
The mathematical investigation of the thermodynamic formalism is in fact not com-
pleted: the theory is a young one, with emphasis still more on imagination than on technical
difficulties. This situation is reminiscent of pre-classic art forms, where inspiration has not
been castrated by the necessity to conform to standard technical patterns. We hope that some
of this juvenile freshness of the subject will remain in the present monograph!

He writes also, in page 3:

The problem of why the Gibbs ensemble describes thermal equilibrium (at least for
“large systems”) when the above physical identifications have been made is deep and incom-
pletely clarified.

The basic identification he is referring to is between β and the inverse temperature.

Consistently, the first equation in both editions (page 3) is dedicated to define the
entropy to be associated with a probability measure. The BG form is introduced
16 1 Historical Background and Physical Motivations

after the words “we define its entropy” without any kind of justification or physical
motivation.

The same theme is retaken by Floris Takens in the 1991 Structures in Dynamics
[42]. Takens writes:
The values of pi are determined by the following dogma: if the energy of the system in the
−E i /kT /Z (T ), where
ith state   pi = e
is E i and if the temperature of the system is T then:
−E /kT
Z (T ) = i e i , (this last constant is taken so that i pi = 1). This choice of pi is
called the Gibbs distribution. We shall give no justification for this dogma; even a physicist
like Ruelle disposes of this question as “deep and incompletely clarified”.

We know that mathematicians sometimes use the word “dogma” when they do not
have the theorem. Indeed, this is not widely known, but still today no theorem exists,
to the best of our knowledge, stating the necessary and sufficient microscopic con-
ditions for being legitimate the use of the celebrated BG weight!

Roger Balian wrote in his 1982/1991 From Microphysics to Macrophysics [43]:

These various quantities are connected with one another through thermodynamic relations
which make their extensive or intensive nature obvious, as soon as one postulates, for
instance, for a fluid, that the entropy, considered as a function of the volume  and of
the constants of motion such as U and N , is homogeneous of degree 1: S(x, xU, x N ) =
x S(, U, N ) (Eq. 5.43). [...] Two counter-examples will help us to feel why extensivity
is less trivial than it looks. [...] A complete justification of the Laws of thermodynamics,
starting from statistical physics, requires a proof of the extensivity (5.43), a property
which was postulated in macroscopic physics. This proof is difficult and appeals to special
conditions which must be satisfied by the interactions between the particles.

John Maddox wrote in 1993 an article suggestively entitled When entropy does
not seem extensive [44]. He focused on a paper by Mark Srednicki [45] where the
entropy of a black hole is addressed. Maddox writes:
Everybody who knows about entropy knows that it is an extensive property, like mass or
enthalpy. [...] Of course, there is more than that to entropy, which is also a measure of dis-
order. Everybody also agrees on that. But how is disorder measured? [...] So why is the
entropy of a black hole proportional to the square of its radius, and not to the cube of it? To
its surface area rather than to its volume?
These comments and questions are of course consistent with the so-called black-hole Hawk-
ing entropy, whose value per unit area equals 1/(4 Gk −1 −3
B c ), a remarkable combination of
four universal constants. Equivalently S BG = k4B AlB2 H , where A B H is the area of the horizon
P
of events of the black hole, and l P is Planck length, as defined in Eq. (3.276).

A suggestive paper by A.C.D. van Enter, R. Fernandez and A.D. Sokal appeared
in 1993. It is entitled Regularity Properties and Pathologies of Position-Space
Renormalization-Group Transformations: Scope and Limitations of Gibbsian Theory
[46]. We transcribe here a few fragments of its content. From its Abstract:
We provide a careful, and, we hope, pedagogical, overview of the theory of Gibssian measures
as well as (the less familiar) non-Gibbsian measures, emphasizing the distinction between
these two objects and the possible occurrence of the latter in different physical situations.
1.3 Background and Indications in the Literature 17

And from its Sect. 6.1.4 of [46] Toward a Non-Gibbsian Point of View:
Let us close with some general remarks on the significance of (non-)Gibbsianness and
(non)quasilocality in statistical physics. Our first observation is that Gibbsianness has
heretofore been ubiquitous in equilibrium statistical mechanics because it has been
put in by hand: nearly all measures that physicists encounter are Gibbsian because
physicists have decided to study Gibbsian measures! However, we now know that natural
operations on Gibbs measures can sometimes lead out of this class. [...] It is thus of great
interest to study which types of operations preserve, or fail to preserve, the Gibbsian-
ness (or quasilocality) of a measure. This study is currently in its infancy.
[...] More generally, in areas of physics where Gibbsianness is not put in by hand, one
should expect non-Gibbsianness to be ubiquitous. This is probably the case in nonequi-
librium statistical mechanics.
Since one cannot expect all measures of interest to be Gibbsian, the question then arises
whether there are weaker conditions that capture some or most of the “good” physical
properties characteristic of Gibbs measures. For example, the stationary measure of the
voter model appears to have the critical exponents predicted (under the hypothesis of Gibb-
sianness) by the Monte Carlo renormalization group, even though this measure is probably
non-Gibbsian.
One may also inquire whether there is a classification of non-Gibbsian measures accord-
ing to their “degree of non-Gibbsianness”.

The authors make in this paper no reference whatsoever to nonextensive statistical

mechanics (proposed in fact 5 years earlier [47]). It will nevertheless become evident
that, interestingly enough, several among their remarks neatly apply to the content
of the present book. Particularly, it will become obvious that (q − 1) represents a
possible measure of “non-Gibbsianness”, where q denotes the entropic index to be
soon introduced.

From the viewpoint of the dynamical foundations of statistical mechanics, a recent

remark (already quoted in the Preface of this book) by Giulio Casati and Tomaz
Prosen [12] is worthy to be reproduced at this point:
While exponential instability is sufficient for a meaningful statistical description, it is not
known whether or not it is also necessary.

Let us antecipate that it belongs to the aim of the present book to convince the
reader precisely that it is not necessary: power-law instability appears to do the job
similarly well, if we consistently adopt the appropriate entropy.

In his 2004 Boltzmann Award lecture Boltzmann and Einstein: Statistics and
dynamics—An unsolved problem, E.G.D. Cohen wrote [29] (see also [48]):
He (Boltzmann) used both a dynamical and a statistical method. However, Einstein strongly
disagreed with Boltzmann’s statistical method, arguing that a statistical description of a sys-
tem should be based on the dynamics of the system. This opened the way, especially for
complex systems, for other than Boltzmann statistics. […] It seems that perhaps a com-
bination of dynamics and statistics is necessary to describe systems with complicated
dynamics.

And he concludes by quoting Boltzmann’s Lecture On recent developments of the

methods of theoretical physics in 1899 in München, Germany:
18 1 Historical Background and Physical Motivations

“Will the old (classical) mechanics with the old forces ... in its essence remain, or live on,
one day, only in history ...superseded by entirely different notions?” […] “Indeed interesting
questions! One almost regrets to have to die long before they are settled. O! immodest mortal!
Your fate is the joy of watching the ever-shifting battle!” (not to see the outcome).

Many more statements exist in the literature along similar lines. But we believe
that the ones that we have selected are enough (both in quality and quantity!) for
depicting, at least in a “impressionistic” way, the epistemological scenario within
which we are evolving. A few basic interrelated points that emerge include:

(i) No strict physical or mathematical reason appears to exist for not exploring the
possible generalization of the BG entropy and its consequences.
(ii) The BG entropy and any of its possible generalizations should conform to the
microscopical dynamical features of the system, very specifically to properties
such as sensitivity to the initial conditions and mixing. The relevant rigorous
necessary and sufficient conditions are still unknown. The ultimate justifica-
tion of any physical entropy on theoretical grounds is expected to come from
microscopic dynamics and detailed geometrical conditions.
(iii) No physical or mathematical reason appears to exist for not exploring, in natu-
ral, artificial and even social systems, distributions differing from the BG one,
very specifically for stationary or quasi-stationary states differing from thermal
equilibrium, such as metastable states, and other nonequilibrium ones.
(iv) Long-range microscopic interactions (and long-range microscopic memory), as
well as interactions exhibiting severe (e.g., nonintegrable) singularities at the
origin, appear as a privileged field for the exploration and understanding of
anomalous thermostatistical behavior.

1.4 Symmetry, Energy, and Entropy

At this point, let us focus on some connections between three key concepts of physics,
namely symmetry, energy and entropy: see Fig. 1.1. According to Plato, symmetry
sits in Topos Ouranos (heavens), where sit all branches of mathematics—science of
structures. In contrast, energy, and entropy should be assumed to sit in Physis (nature).
Energy deals with the possibilities of the system; entropy deals with the probabilities
of those possibilities. It is fair to say that energy is a quite subtle physical concept.
Entropy is based upon the ingredients of energy, and therefore it is, epistemologically
speaking, one step further. It is most likely because of this feature that entropy
emerged, in the history of physics, well after energy. A coin has two faces, and can
therefore fall in two possible manners, head and tail (if we disconsider the very
unlike possibility that it falls on its edge). This is the world of the possibilities for
this simple system. The world of its probabilities is more delicate. Before throwing
the coin (assumed fair for simplicity), the entropy equals k ln 2. After throwing it, it
still equals k ln 2 for whoever does not know the result, whereas it equals zero for
1.4 Symmetry, Energy, and Entropy 19

Fig. 1.1 Connections between symmetry, energy and entropy

whoever knows it. This example neatly illustrates the informational nature of the
concept.
Let us now address the connections. Those between symmetry and energy are
long and well known. Galilean invariance of the equations is central to Newtonian
mechanics. Its simplest form of energy can be considered to be the kinetic one of a
point particle, namely p 2 /2m, p being the linear momentum, and m the mass. This
energy, although having a unique form, is not universal; indeed it depends on the
mass of the system. If we replace now the Galilean invariance by the Lorentzian one,
this drastically changes the form itself of the kinetic energy, which now becomes
( p 2 c2 + m 20 c4 )1/2 , c being the speed of light in vacuum, and m 0 the mass at rest. In
other words, this change of symmetry is far from innocuous; it does nothing less than
changing Newtonian mechanics into special relativity! Maxwell electromagnetism is,
as well known, deeply related to this same Lorentzian invariance, as well as to gauge
invariance. The latter plays, in turn, a central role in quantum electrodynamics and
quantum field theory. Quantum chromodynamics also is deeply related to symmetry
properties. And so is expected to be quantum gravity, whenever it becomes reality.
Summarizing, the deep connections between symmetry and energy are standard
knowledge in contemporary physics. Changes in one of them are concomitantly
followed by changes in the other one.
20 1 Historical Background and Physical Motivations

What about the connections between energy and entropy? Well, also these are
quite known. They naturally emerge in thermodynamics (e.g., the possibility and
manners for transforming work into heat, and, partially, the other way around). This
obviously reflects on BG statistical mechanics itself.
But, what can we say about the possible connections between symmetry (its nature
and evolution) and entropy? This topic has remained basically unchanged and quite
unexplored during more than one century! Why? Hard to know. However, it is allowed
to suspect that this intellectual lethargy comes, on one hand, from the “slopiness” of
the concept of entropy, and, on the other one, from the remarkable fact that the unique
functional form that has been used in physics is the BG one (Eq. (1.1) and its con-
tinuum or quantum analogs), which depends on only one of the universal constants,
namely Boltzmann constant k B . Within this intellectual landscape, generation after
generation, the idea installed in the mind of very many physicists that the physical
entropic functional itself must be universal, and that it is of course the BG one. In
the present book, we try to convince the reader that it is not so, that many types of
entropy can be physically and mathematically meaningful. Moreover, we shall argue
that dynamical concepts such as the time dependence of the sensitivity to the initial
conditions, mixing, and the associated occupancy and visitation networks they may
cause in phase space, have so strong effects, that even the functional form of the
entropy must be modified in order to conform to the empirical world. Naturally, the
BG entropy will then still have a highly privileged position. It surely is the correct
one when the microscopic nonlinear dynamics is controlled by positive Lyapunov
exponents, hence strong chaos. If the system is such that strong chaos is absent (typ-
ically because the maximal Lyapunov exponent vanishes), then the physical entropy
to be used appears to be a different one.

1.5 A Few Words on the Foundations of Statistical

Mechanics

A mechanical foundation of statistical mechanics from first principles should essen-

tially include, in one way or another, the following main steps [49].

(i) Adopt a microscopic dynamics. This dynamics typically is free from any phe-
nomenological noise or stochastic ingredient, so that the foundation may be
considered as from first principles. This dynamics could be Newtonian, or quan-
tum, or relativistic mechanics (or some other mechanics to be found in future) of
a many-body system composed by say N interacting elements or fields. It could
also be conservative or dissipative coupled maps, or even cellular automata.
Consistently, time t could be continuous or discrete. The same is valid for space.
The quantity which is defined in space-time could itself be continuous or dis-
crete. For example, in quantum mechanics, the quantity is a complex continuous
variable (the wave function) defined in a continuous space-time. On the other
1.5 A Few Words on the Foundations of Statistical Mechanics 21

extreme, we have cellular automata, for which all three relevant variables—time,
space and the quantity therein defined—are discrete. In the case of a Newtonian
mechanical system of particles, we may think of N Dirac delta functions local-
ized in continuous spatial positions which depend on a continuous time.
Langevin-like equations (and associated Fokker–Planck-like equations) are typ-
ically considered not microscopic, but mesoscopic instead. The reason of course
is the fact that they include at their very formulation, i.e., in an essential manner,
some sort of noise. Consequently, they should not be used as a starting point if
we desire the foundation of statistical mechanics to be from first principles. A
schematic ladder going from the microscopic, through the mesoscopic, to the
macroscopic descriptions of the world is presented in Fig. 1.2.
(ii) Then assume some set of initial conditions (either a single one or an ensemble of
many of them) and let the system evolve in time. These initial conditions are, for
say a classical system, defined in the so-called phase space of the microscopic
configurations of the system, for example Gibbs’  space for a Newtonian N -

Fig. 1.2 From the microscopic (first-principle electro-mechanics), through the mesoscopic, to
the macroscopic (thermodynamics) descriptions of nature in contemporary theoretical physics,
including several, though obviously not all, relevant equations. The energy emerges from electro-
mechanics as a constant of motion. The entropic functional emerges from first-principle electro-
mechanics as an adequate function of probabilities directly related to the trajectories in say phase
space or in Hilbert space or in Fock space, and the correlations they involve. The Braun and Hepp
theorem that is mentioned in one of the links is the one proved in [50]. We only depict here a basic
scheme. Further branches and connections do exist which are not detailed here (see for instance
[51–57])
22 1 Historical Background and Physical Motivations

particle system (e.g., the  space for point masses has 2d N dimensions if the
particles live in a d-dimensional space). These initial conditions typically (but
not necessarily) involve one or more constants of motion. For example, if the
system is a conservative Newtonian one of point masses, the initial total energy
and the initial total linear momentum (d dimensional vector) are such constants
of motion. The total angular momentum might also be a constant of motion. It
is quite frequent to use coordinates such that both total linear momentum and
total angular momentum vanish.
If the system consists of conservative coupled maps, the initial hypervolume
of an ensemble of initial conditions near a given one is preserved through time
evolution. By the way, in physics, such coupled maps are frequently obtained
through Poincaré sections of Newtonian dynamical systems.
(iii) After some sufficiently long evolution time (which typically depends on both
N and the spatial range of the interactions), the system might approach some
stationary or quasi-stationary macroscopic state7 . In such a state, the various
regions of phase space are being visited with some probabilities. This set of
probabilities either does not depend anymore on time or depends on it very
slowly. More precisely, if it depends on time, it does so on a scale much longer
(e.g., geological times) than the microscopic time scale. The visited regions
of phase space that we are referring to typically correspond to a partition of
phase space with a degree of (coarse or fine) graining that we adopt for specific
purposes. These probabilities can be either insensitive or, on the contrary, very
sensitive to the ordering in which the t → ∞ (asymptotic) and N → ∞ (ther-
modynamic) limits are taken. This can depend on various things such as the range
of the interactions, or whether the system is on the ordered or on the disordered
side of a continuous phase transition. Generically speaking, the influence of the
ordering of the t → ∞ and N → ∞ limits is typically related to some kind of
breakdown of symmetry, or of ergodicity, or the alike. In some cases, the result
might even depend on the path along which the t → ∞ and N → ∞ limits
are attained. For example, it might depend on whether the simultaneous limit
lim(t,N )→(∞,∞) t/N γ (with γ > 0) is above or below some model-dependent
critical value.
The simplest nontrivial dynamical situation is expected to occur for an isolated
many-body short-range-interacting classical Hamiltonian system (microcanoni-
cal ensemble); later on we shall qualify when an interaction is considered short-
ranged in the present context. In such a case, the typical microscopic (nonlinear)
dynamics is expected to be strongly chaotic, in the sense that the maximal Lya-
punov exponent is positive almost everywhere in phase space. Such a system
would necessarily be mixing, i.e., it would quickly visit virtually all the accessible
phase space (more precisely, very close to almost all the accessible phase space)
for almost any possible initial condition. Furthermore, it would necessarily be
ergodic with respect to some measure in the full phase space, i.e., time averages

7 When the system exhibits some sort of aging, the expression quasi-stationary is preferable to
stationary.
1.5 A Few Words on the Foundations of Statistical Mechanics 23

and ensemble averages would coincide. In most of the cases this measure is
expected to be uniform in phase space, i.e., the hypothesis of equal probabilities
first thought of by Boltzmann would be satisfied.
A slightly more complex situation is encountered for those systems which exhibit
a continuous phase transition. Let us consider the simple case of a ferromagnet
which is invariant under inversion of the hard axis of magnetization, e.g., the
d = 3 X Y classical nearest-neighbor ferromagnetic model on simple cubic lat-
tice. If the system is in its disordered (paramagnetic) phase, the limits t → ∞
and N → ∞ commute, and the entire phase space is expected to be equally
well visited. If the system is in its ordered (ferromagnetic) phase, the situation
is expected to be more subtle. The lim N →∞ limt→∞ set of probabilities is, as
before, equally distributed all over the entire phase space for almost any ini-
tial condition. But this is not expected to be so for the limt→∞ lim N →∞ set of
probabilities. The system probably lives, in this case, only in part of the entire
phase space. Indeed, if the initial condition is such that the initial magnetiza-
tion is nonzero along a given direction in the XY plane, even if infinitesimally
so (for instance, under the presence of a vanishingly small external magnetic
field), then the system is expected to be ergodic but only in the part of phase
space associated with a magnetization along that direction, even if the external
magnetic field becomes vanishingly small. This is usually referred to as a spon-
taneous breakdown of symmetry. This illustrates, already in this simple example,
the importance that the ordering of the t → ∞ and N → ∞ limits can have.
A considerably more complex situation is expected to occur if we consider a long-
range-interacting model, e.g., the same d = 3 X Y classical ferromagnetic model
on simple cubic lattice as before, but now with a coupling constant which decays
with distance as 1/r α , where r is the spin-to-spin distance measured in crystal
units, and α ≥ 0 (the nearest-neighbor model that we just discussed corresponds
to the α → ∞ limit, which is the extreme case of the (quasi) short-ranged domain
α/d > 1). The 0 ≤ α/d ≤ 1 long-ranged model also appears to have a continu-
ous phase transition. In the disordered phase, the system possibly is ergodic over
the entire phase space. But in the ordered phase the result can strongly depend
on the ordering of the two limits, or on the path along which these two limits
are accomplished. The lim N →∞ limt→∞ set of probabilities corresponds to the
system living in the entire phase space. In contrast, the limt→∞ lim N →∞ set of
probabilities for the same (conveniently scaled) total energy might be consider-
ably more complex. It seems that, for this ordering, phase space could exhibit
many macroscopic basins of evolution. One of them leads essentially to part of
the phase space where the system lives in the lim N →∞ limt→∞ ordering, i.e.,
that part of phase space which is associated with a given alignment of the mag-
netization, which coincides with the direction of the initial magnetization along
an external magnetic field. Other basins would correspond to living in a very
complicated, hierarchical-like, geometrical structure. This structure could be a
zero Lebesgue measure one (in the full multidimensional phase space), some-
what similar to that of an airlines company, say Air France, whose central hub is
located in Paris, or British Airways, whose central hub is located in London. The
24 1 Historical Background and Physical Motivations

specific location of the structure in phase space would depend on the particular
initial condition within that special basin, but the geometrical-dynamical nature
of the structure would be virtually the same for a generic initial condition within
phase space. At this point, let us warn the reader that the scenario that we have
depicted here is only conjectural, and remains to be proved. It is however based
on various numerical evidences (see, e.g., [58–62] and references therein, as
well as Sect. 5.3). It is expected to be caused by a possibly vanishing maximal
Lyapunov exponent in the N → ∞ limit. In other words, one would possibly
have, instead of strong, only weak chaos.
(iv) Now let us focus further on the specific role played by the initial conditions. If the
system is strongly chaotic, hence mixing, hence ergodic, this point is irrelevant.
We can make or not averages over initial conditions, we can take almost any
initial condition, the outcome for sufficiently long times will be the same, in
the sense that the set of probabilities in phase space will be the same. But if
the system is only weakly chaotic, the result can drastically change from initial
condition to initial condition. If two initial conditions belong to the same “basin
of evolution”, the difference at the macroscopic level could be quite irrelevant.
If they belong however to different basins, the results can be sensibly different.
For some purposes we might wish to stick to a specific initial condition within a
certain class of initial conditions. For other purposes, we might wish to average
over all initial conditions belonging to a given basin, or even over all possible
initial conditions of the entire phase space. The macroscopic result obtained
after averaging might considerably differ from that corresponding to a single
initial condition. Moreover, the traditional statistical-mechanical classification
of ensemble average (which does not depend on time, as soon as a possible
initial transient is overcome) versus time average (which does not depend on
the chosen single initial initial condition, as long as it is a generic one) might
become insufficient. For example, coming back to our air companies analogy, if
we are within the Air France evolution basin we will observe an hierarchical-like
geometric-dynamical structure centered on Paris, whereas if we are in the British
Airways evolution basin we will observe a similar hierarchical-like structure, but
this time centered on London. For some analysis, it might be interesting to make
averages over the evolution basin of one air company; for other purposes, it
might be useful to make averages over all air companies within a given class
(say inter-continental flights); still for other purposes it might be useful to make
averages over all flights of all air companies; and so on. A really interesting
statistical mechanics should be able, through its various limits (e.g., infinite size,
infinite number of initial conditions, infinite time, infinite precision) and their
orderings, to correctly address at least some of these various relevant averages.
As we shall see along this book, abandoning the traditional but unnecessary
restriction to additive entropic functionals opens a big door toward this goal.
(v) Last but not least, the mathematical form of the entropy functional must be
addressed. Strictly speaking, if we have deduced (from microscopic dynamics)
the probabilities to be associated with every cell in phase space, we can in princi-
ple calculate useful averages of any physical quantity of interest which is defined
1.5 A Few Words on the Foundations of Statistical Mechanics 25

in that phase space. In this sense, we do not need to introduce an entropic func-
tional which is defined precisely in terms of those probabilities. Especially if we
take into account that any set of physically relevant probabilities can be obtained
through extremization (typically maximization) of an infinite number of entropic
functionals
 (monotonically depending one onto the other, such as, for instance,
[− i pi ln pi ] and say its cube), given any set of physically and mathematically
meaningful constraints. However, if we wish to make contact with classical ther-
modynamics, we certainly need to know the mathematical form of such entropic
functional. This functional is expected to match, in the appropriate limits, the
classical, macroscopic, entropy à la Clausius. In particular, one expects it to
satisfy the Clausius property of extensivity, i.e., essentially to be proportional to
the weight or mass of the system. In statistical-mechanical terms, we expect it to
be proportional to N for large N . This simple requirement immediately  admits,
in the realm of BG statistical mechanics, the entropic functional [− i pi ln pi ]
and definitively excludes say its cube.8
The foundations of any valuable statistical mechanics are, as already said, expected
to satisfactorily cover basically all of the above points. There is a wide-spread vague
belief among some physicists that these steps have already been satisfactorily accom-
plished since long for the standard, BG statistical mechanics. This is not so! Not so
surprising after all, given the enormity of the corresponding task! For example, as
already mentioned, at this date there is no available deduction, from and only from
microscopic dynamics, of the celebrated BG exponential weight (1.8). Neither exists
the deduction from microscopic dynamics, without further assumptions, of the BG
entropy (1.1).
For standard systems, there is not a single reasonable doubt about the correctness
of the expressions (1.1) and (1.8), and of their relationships. But, from the logical-
deductive viewpoint, there is still pretty much work to be done! This is, in fact,
kind of easy to notice. Indeed, all the textbooks, without exception, introduce the
BG factor and/or the entropy S BG in some kind of phenomenological manner, or as
self-evident, or within some axiomatic formulation. None of them introduces them
as (and only as) a rational consequence of say Newtonian, or quantum mechanics,
using theory of probabilities. This is in fact sometimes referred to as the Boltzmann
program. Boltzmann himself died without succeeding its implementation. Although
important progress has been accomplished in these last 140 years, Boltzmann pro-
gram still remains in our days as a basic intellectual challenge. Were it not the
genius of scientists like Boltzmann and Gibbs, were we to exclusively depend on

8 Let us anticipate that it has been recently shown [63–66] that, if we impose a Poissonian distribution
for visitation times in phase space, in addition to the first and second principles of thermodynamics,
we obtain the BG functional form for the entropy. If a conveniently deformed Poissonian distribution
is imposed instead, we obtain the Sq functional form. These results in themselves can not be
considered as a justification from first principles of neither the BG nor the nonextensive, statistical
mechanics. Indeed, the visitation distributions are phenomenologically introduced, and the first and
second principles are just imposed. This connection is nevertheless extremely clarifying, and can
help producing a full justification.
26 1 Historical Background and Physical Motivations

mathematically well-constructed arguments, one of the monuments of contempo-

rary physics—BG statistical mechanics—could just not exist, or it would have taken
much longer to emerge!
Many anomalous natural, artificial, and social systems exist for which BG sta-
tistical concepts appear to be inapplicable. Typically because they live in peculiar
stationary or quasi-stationary states that are quite different from thermal equilibrium,
where BG statistics reigns. Nevertheless, as we shall see, not few of them can still
be handled within statistical-mechanical methods, but with a more general entropy,
namely Sq , to be introduced later on [47, 67–69].
It should naturally be clear that, whatever is not yet mathematically justified in
BG statistical mechanics, it is even less justified in the generalization to which the
present book is dedicated. In addition to this, some of the points that are relatively
well understood in the standard theory, can be still unclear in its generalization. In
other words, the theory we are presenting here is still in intense evolution (sets of
reviews can be found in [18, 70–92]).
Chapter 2
Learning with Boltzmann-Gibbs
Statistical Mechanics

α̃n mštron ¥riston

Kleoboulos of Lindos (6th century B.C.)

2.1 Boltzmann–Gibbs Entropy

2.1.1 Entropic Forms

The entropic forms (1.1) and (1.3) that we have introduced in Chap. 1 correspond
to the case where the (microscopic) states of the system are discrete. There are,
however, cases in which the appropriate variables are continuous. For these, the BG
entropy takes the form

S BG = −k d x p(x) ln[σ p(x)], (2.1)

with 
d x p(x) = 1, (2.2)

Ancient and popular Greek expression. A possible translation into English is “All things in their
proper measure are excellent”. The expression is currently attributed to Kleoboulos of Lindos (one of
the seven philosophers of Greek antiquity), although in a more laconic—and logically equivalent—
version, namely Mštron ¥riston. Indeed, it is this abridged form that Clement of Alexandria
(Stromata, 1.14.61) and Diogenes Laertius (Lives of Philosophers, Book 1.93, Loeb Series) attribute
to Kleoboulos. This expression addresses what I consider the basis of all variational principles, in
my opinion the most elegant form in which physical laws can be expressed. The Principle of least
action in mechanics, Fermat’s Principle of least time in optics, and the Optimization of the entropy
in statistical mechanics, are but such realizations.

© Springer Nature Switzerland AG 2023 27

C. Tsallis, Introduction to Nonextensive Statistical Mechanics,
https://doi.org/10.1007/978-3-030-79569-6_2
28 2 Learning with Boltzmann–Gibbs Statistical Mechanics

where x/σ ∈ R D , D ≥ 1 being the dimension of the full space of microscopic states
(called Gibbs  phase space for classical Hamiltonian systems).1 Typically x car-
ries physical units. The constant σ carries the same physical units as x, so that x/σ
is a dimensionless quantity (we adopt from now on the notation [x] = [σ], hence
[x/σ] = 1). For example, if we are dealing with an isolated classical N -body Hamil-
tonian system of point masses interacting among them in d dimensions, we may use
σ =  N d . This standard choice comes of course from the fact that, at a sufficiently
small scale, Newtonian mechanics becomes incorrect and we must rely on quantum
mechanics. In this case, D = 2d N , where each of the d pairs of components of
momentum and position of each of the N particles has been taken into account (we
recall that [momentum][ position] = []). For the case of equal probabilities (i.e.,
p(x) = 1/, where  is the hypervolume of the admissible D-dimensional space),
we have
S BG = k ln(/σ). (2.3)

A particular case of p(x) is the following one:


W
p(x) = pi δ(x − xi ) (W ≡ /σ), (2.4)
i=1

where δ denotes Dirac’s distribution. In this case, Eqs. (2.1), (2.2), and (2.3) precisely
recover Eqs. (1.1), (1.2), and (1.3).
In both discrete and continuous cases that we have addressed until now, we were
considering classical systems in the sense that all physical observables are real quan-
tities and not operators. However, for intrinsic quantum systems, we must generalize
the concept. In that case, the BG entropic form is to be written (as first introduced
by von Neumann [93–95]) in the following manner:

S BG = −k T r ρ ln ρ, (2.5)

with
T r ρ = 1, (2.6)

where ρ is the density matrix acting on a W -dimensional Hilbert vectorial space

(typically associated with the solutions of the Schroedinger equation with the chosen
boundary conditions; in fact, quite frequently we have W → ∞).
A particular case of ρ is when it is diagonal, i.e., the following one:

ρi j = pi δi j , (2.7)

1Strictly speaking, we are using here an oversimplified notation to denote that the vector x ≡

(x1 , x2 , . . . , x W ) is to be divided, component by component, by (σ1 , σ2 , . . . , σW ) with σ = iW σi .
2.1 Boltzmann–Gibbs Entropy 29

where δi j denotes Kroenecker’s delta function. In this case, Eqs. (2.5) and (2.6)
exactly recover Eqs. (1.1) and (1.2).
All three entropic forms (1.1), (2.1), and (2.5) will be generically referred in
the present book as BG-entropy because they all constitute a logarithmic measure of
uncertainty or lack of information. Although we shall use one or the other for specific
purposes, we shall mainly address the simple form expressed in Eq. (1.1).

2.1.2 Properties

2.1.2.1 Non-negativity

If we know with certainty the state of the system, then pi0 = 1, and pi = 0, ∀ i = i 0 .
Then it follows that S BG = 0, where we have taken into account that lim x→0 (x ln x) =
0. In any other case we have pi < 1 for at least two different values of i. We can
therefore write Eq. (1.1) as follows:
 1
S BG = −k ln pi = k ln , (2.8)
pi
W
where · · · ≡ i=1 pi (...) is the standard mean value. Since ln(1/ pi ) > 0 (∀i), it
clearly follows that S BG is positive.

2.1.2.2 Maximal at Equal Probabilities

Energy is a concept which definitively takes into account the physical nature of the
system. Not exactly so, in some sense, the BG entropy.2 This entropy depends of
course on the total number of possible microscopic configurations of the system,
but it is insensitive to its specific physical support; it only takes into account the
(abstract) probabilistic information on the system. Let us make a trivial illustration:
a spin that can be up or down (with regard to some external magnetic field), a coin
that comes head or tail, a computer bit which can be 0 or 1, are all equivalent in what
concerns the concept of entropy. Consequently, entropy is expected to be a functional
which is invariant with regard to any permutations of the states. Indeed, expression
(1.1) exhibits this invariance through the form of a sum. Consequently, if W > 1,
the entropy must have an extremum (maximum or minimum), and this must occur
for equal probabilities. Indeed, this is the unique possibility for which the entropy is
invariant with regard to the permutation of any two states. It is easy to verify that, for

2 This statement is to be revisited for the more general entropy Sq . Indeed, as we shall see, the index
q does depend on some universal aspects of the physical system, e.g., the type of inflection of a
dissipative unimodal map, or, possibly, the type of power-law decay with a distance of two-body
long-range interactions for many-body Hamiltonian systems.
30 2 Learning with Boltzmann–Gibbs Statistical Mechanics

S BG , it is a maximum, and not a minimum. In fact, the identification as a maximum

(and not a minimum) will become obvious when we shall prove, later on, that S BG is
a concave functional. Of course, the expression that S BG takes for equal probabilities
has already been indicated in Eq. (1.3).

2.1.2.3 Expansibility

Adding to a system new possible states with zero probability should not modify the
entropy. This is precisely what is satisfied by S BG if we take into account the already
mentioned property lim x→0 (x ln x) = 0. So, we have that

S BG ( p1 , p2 , ..., pW , 0) = S BG ( p1 , p2 , ..., pW ). (2.9)

2.1.2.4 Additivity

Let O be a physical quantity associated with a given system, and let A and B be two
probabilistically independent subsystems. We shall use the term additive if and only
if O(A + B) = O(A) + O(B). If so, it is clear that if we have N independent equal
systems, then O(N ) = N O(1), where the notation is self-explanatory. A weaker
condition is O(N ) ∼ N  for N → ∞, with 0 < || < ∞, i.e., lim N →∞ O(N )/N
is finite (generically  = O(1)). In this case, the expression asymptotically additive
might be used. Clearly, any observable which is additive with regard to a given
composition law, is asymptotically additive (with  = O(1)), but the opposite is not
necessarily true.
It is straightforwardly verified that, if A and B are independent, i.e., if the joint
probability satisfies piA+B
j = piA p Bj (∀(i j)), then

S BG (A + B) = S BG (A) + S BG (B). (2.10)

Therefore the entropy S BG is additive.

2.1.2.5 Concavity

Let us assume two arbitrary and different probability sets, namely { pi } and { pi }, asso-
ciated with a single system having W states. We define an intermediate probability
set as follows:
pi = λ pi + (1 − λ) pi (∀i; 0 < λ < 1). (2.11)

The functional S BG ({ pi }) (or any other functional in fact) is said concave if and only
if
S BG ({ pi }) > λS BG ({ pi }) + (1 − λ)S BG ({ pi }). (2.12)
2.1 Boltzmann–Gibbs Entropy 31

This is indeed satisfied by S BG . The proof is straightforward. Because of its negative

second derivative, the (continuous) function −x ln x (x > 0) satisfies

− pi ln pi > λ(− pi ln pi ) + (1 − λ)(− pi ln pi ) (∀i; 0 < λ < 1). (2.13)

W
Applying i=1 on both sides of this inequality, we immediately obtain Eq. (2.12),
i.e., the concavity of S BG .

2.1.2.6 Lesche-Stability or Experimental Robustness

An entropic form S({ pi }) (or any other functional of the probabilities, in fact) is said
Lesche-stable or experimentally robust [96]3 if and only if it satisfies the following
continuity property. Two probability distributions { pi } and { pi } are said close if they
satisfy the metric property:


W
D≡ | pi − pi | ≤ d , (2.14)
i=1

where d is a small real number. Then, experimental robustness is verified if, for any
 > 0, a d exists such that D ≤ d implies
 S({ p }) − S({ p }) 
 i i 
R≡  < , (2.15)
Smax

where Smax is the maximal value that the entropic form can achieve (assuming its
extremum corresponds to a maximum, not a minimum). For S BG the maximal value
is of course ln W .
Condition 2.15 should be satisfied under all possible situations, including for
W → ∞. This implies that the condition
 S({ p }) − S({ p }) 
 i i 
lim lim  =0 (2.16)
d →0 W →∞ Smax
 
 S({ p })−S({ p }) 
should also be satisfied, in addition to lim W →∞ limd →0  i Smax i  = 0, which
is of course always satisfied.
What this property essentially guarantees is that similar experiments performed
onto similar physical systems should provide similar results (i.e., small percent-
age discrepancy) for the measured physical functionals. Lesche showed [96] that

3 Lesche himself called stability this property. Two decades later, in personal conversation, I argued

with him that that name could be misleading in the sense that it seems to suggest some relation
with thermodynamical stability, with which it has no mathematical connection (thermodynamical
stability has, in fact, connection with concavity). I suggested the use of experimental robustness
instead. Lesche fully agreed that it is a better name. Consistently, I use it preferentially since then.
32 2 Learning with Boltzmann–Gibbs Statistical Mechanics

0
BG d=0.001
R
-0.2
d=0.01
-0.4
d=0.1
-0.6

-0.8
QC
-1
0 0.1 0.2 0.3 0.4 0.5
1/W

0.1
BG
R d=0.001
0

-0.1 d=0.01

-0.2 d=0.1

-0.3
QEP
0 0.1 0.2 0.3 0.4 0.5
1/W
Fig. 2.1 Illustration of the Lesche-stability of S BG . QC and Q E P stand for quasi-certainty and
quasi-equal-probabilities, respectively: see details in [97, 98]. From [98]

W q
ln p
S BG is experimentally robust, whereas the Renyi entropy SqR ≡ 1−q i=1 i
is not. See
Fig. 2.1.
It is in principle possible to use, as a concept for distance, a quantity different
from that used in Eq. (2.14). We could use for instance the following definition:


W
1/μ
Dμ ≡ | pi − pi |μ (μ > 0). (2.17)
i=1
2.1 Boltzmann–Gibbs Entropy 33

Equation (2.14) corresponds to μ = 1. The Pythagorean metric corresponds to μ = 2.

What about other values of μ? It happens that only for μ ≥ 1 the triangular inequality
is satisfied in addition to the other traditional requirements for distance, and conse-
quently it does constitute a metric. Still, why not using the values of μ > 1? Because
only for μ = 1 the distance D does not depend on W , which makes it convenient for
a generic property [99]. We come back in Sect. 3.2.2 onto this interesting property
introduced by Lesche.

2.1.2.7 Shannon Uniqueness Theorem

Let us assume that an entropic form S({ pi }) satisfies the following properties:

(i) S({ pi }) is a continuous f unction o f { pi }; (2.18)

(ii) S( pi = 1/W, ∀i) monotonically incr eases with the total
number o f possibilities W ; (2.19)
(iii) S(A + B) = S(A) + S(B) if piA+B
j = piA p Bj ∀(i, j), (2.20)

WB
wher e S(A + B) ≡ S({ piA+B
j }), S(A) ≡ S({ piA }) ( piA ≡ piA+B
j ),
j=1


WA
and S(B) ≡ S({ p Bj }) ( p Bj ≡ piA+B
j );
i=1
(iv) S({ pi }) = S( p L , p M ) + p L S({ pi / p L }) + p M S({ pi / p M }) (2.21)
 
with p L ≡ pi , p M ≡ pi ,
L ter ms M ter ms
L + M = W , and p L + p M = 1.

Then and only then [32]


W
S({ pi }) = −k pi ln pi (k > 0). (2.22)
i=1

It is therefore very clear in what sense the functional (2.22) is unique. This neatly
differs from the fallacious, and yet not rare, statement that form (2.22) is the unique
physically admissible entropic functional.4

4 Some authors prefer the notation S(A × B) instead of S(A + B) in order to emphasize the fact
that the phase space of the total system is the tensor product of the space phases of the subsystems
A and B.
34 2 Learning with Boltzmann–Gibbs Statistical Mechanics

2.1.2.8 Khinchin Uniqueness Theorem

Let us assume that an entropic form S({ pi }) satisfies the following properties:

(i) S({ pi }) is a continuous f unction o f { pi }; (2.23)

(ii) S( pi = 1/W, ∀i) monotonically incr eases with the total
number o f possibilities W ; (2.24)
(iii) S( p1 , p2 , ..., pW , 0) = S( p1 , p2 , ..., pW ); (2.25)
(iv) S(A + B) = S(A) + S(B|A) , (2.26)

WB
wher e S(A + B) ≡ S({ piA+B
j }), S(A) ≡ S({ piA }) ( piA ≡ piA+B
j ),
j=1


WA
and the conditional entropy S(B|A) ≡ piA S({ piA+B
j / piA }).
i=1

Then and only then [100]


W
S({ pi }) = −k pi ln pi (k > 0). (2.27)
i=1

It follows then that the Shannon and the Khinchin sets of axioms are equivalent.

2.1.2.9 Composability

A dimensionless entropic form S({ pi }) (i.e., whenever expressed in appropriate con-

ventional units, e.g., in units of k) is said composable [71, 72] (see also [101–104])
if the entropy S(A + B)/k corresponding to a system composed of two probabilis-
tically independent subsystems A and B can be expressed in the form

S(A + B) S(A) S(B)

=F , ; {η} , (2.28)
k k k

where F(x, y; {η}) is a smooth function of (x, y) which depends on a (typically small)
set of universal indices {η} defined in such a way that F(x, y; {0}) = x + y (addi-
tivity), and which satisfies F(x, 0; {η}) = x (null-composability), F(x, y; {η}) =
F(y, x; {η}) (symmetry), F(x, F(y, z; {η}); {η}) = F(F(x, y; {η}), z; {η}) (asso-
ciativity). For thermodynamical systems, this associativity appears to be consistent
with the 0th Principle of Thermodynamics.
In other words, the whole concept of composability is constructed upon the
requirement that the entropy of (A + B) does not depend on the microscopic con-
figurations of A and of B. Equivalently, we are able to macroscopically calculate the
2.1 Boltzmann–Gibbs Entropy 35

entropy of the composed system without any need of entering into the knowledge of
the microscopic states of the subsystems. This property appears to be a natural one
for an entropic form if we desire to use it as a basis for a statistical mechanics which
would naturally connect to thermodynamics.
The entropy S BG is composable since it satisfies Eq. (2.10). In other words, we
have FBG (x, y) = x + y. Being S BG nonparametric, no index exists in FBG .

2.1.2.10 Sensitivity to the Initial Conditions, Entropy Production Per

Unit Time, and a Pesin-Like Identity

For a one-dimensional dynamical system (characterized by the variable x) the sen-

sitivity to the initial conditions ξ is defined as follows:

x(t)
ξ≡ lim , (2.29)
x(0)→0 x(0)

x(t) being the discrepancy, at time t, between two initially close trajectories.
It can be shown [105–110] that ξ paradigmatically satisfies the equation

dξ
= λ1 ξ, (2.30)
dt
whose solution is given by
ξ = e λ1 t . (2.31)

(The meaning of the subscript 1 will become transparent later on). If the Lyapunov
exponent λ1 > 0 (λ1 < 0), the system will be said to be strongly chaotic (regular).
The case where λ1 = 0 is sometimes called marginal or subexponential behavior
and will be extensively addressed later on.
At this point let us briefly review, without proof, some basic notions of nonlinear
dynamical systems. If the system is d-dimensional (i.e., it evolves in a phase space
whose d-dimensional Lebesgue measure is finite), it has d Lyapunov exponents:
d+ of them are positive, d− are negative, and d0 vanish, hence d+ + d− + d0 = d.
(d+ )
Let us order them all from the largest to the smallest: λ(1) (2)
1 ≥ λ1 ≥ ... ≥ λ1 >
(d +1) (d +2) (d +d ) (d +d +1) (d +d +2)
λ1 + = λ1 + = ... = λ1 + 0 = 0 > λ1 + 0 ≥ λ1 + 0 ≥ ... ≥ λ(d)1 . An
infinitely small segment (having then a vanishing one-dimensional Lebesgue mea-
(1)
sure) diverges like eλ1 t ; this precisely is the case focused in Eq. (2.31). An
infinitely small area (having then a vanishing two-dimensional Lebesgue measure)
(1) (2) (1) (2) (3)
diverges like e(λ1 +λ1 ) t . An infinitely small volume diverges like e(λ1 +λ 1 +λ1 ) t ,
d (r )
and so on. An infinitely small d-dimensional hypervolume evolves like e[ r =1 λ1 ] t .
If the system is conservative, i.e., if the infinitely small
 d-dimensional hypervol-
ume remains constant with time, consistently with rd=1 λ(r 1
)
= 0. An important
particular class of conservative systems is constituted by the so-called symplectic
36 2 Learning with Boltzmann–Gibbs Statistical Mechanics

ones. For these, d is an even integer, and the Lyapunov exponents are coupled two
(d ) (d +d +1)
by two as follows: λ(1) (d) (2)
1 = −λ1 ≥ λ1 = −λ1
(d−1)
≥ ... ≥ λ1 + = −λ1 + 0 ≥
(d+ +1) (d+ +d0 )
λ1 = ... = λ1 = 0. Consistently, such systems have d+ = d− and d0 is a
even integer. The most popular illustration of symplectic systems is the Hamilto-
nian system. They are conservative, which precisely is what the classical Liouville
theorem states!
Do all these degrees of freedom contribute, as time evolves, to the erratic explo-
ration of the phase space? No, they do not. Only those associated with the d+ positive
Lyapunov exponents, and some of the d0 vanishing ones, do. Consistently, it is only
these which contribute to our loss of information, as time evolves, about the location
in phase space of a set of initial conditions. As we shall see, these remarks enable an
intuitive understanding of the so-called Pesin identity, that we will soon state.
Let us now address the interesting question of the BG entropy production as
time t increases. More than one entropy production can be defined as a function of
time. Two basic choices are the so-called Kolmogorov–Sinai entropy (or KS entropy
rate or metric entropy) [111], namely one which is based on a single trajectory in
phase space, the other one being associated to the evolution of an ensemble of initial
conditions. We shall preferentially use here the latter, because of its sensibly larger
computational tractability. In fact, except for pathological cases, they both coincide.
Let us schematically describe the KS entropy rate concept [110–112]. We first
partition the phase space into two regions, noted A and B. Then we choose a generic
initial condition (the final result will not depend on this choice) and, applying the
specific dynamics of the system at equal and finite time intervals τ , we generate
a long string (infinitely long in principle), say AB B B A AB B AB A A A.... Then we
analyze words of length l = 1. In our case, there are only two such words, namely
A and B. The analysis consists in running along the string a window whose width
is l, and determining the probabilities p A and p B of the words A and B, respec-
tively. Finally we calculate the entropy S BG (l = 1) = − p A ln p A − p B ln p B . Then
we repeat for words whose length equals l = 2. In our case, there are four such words,
namely A A, AB, B A, B B. Running along the string a l = 2 window letter by let-
ter, we determine the probabilities p A A , p AB , p B A , p B B , hence the entropy S BG (l =
2) = − p A A ln p A A − p AB ln p AB − p B A ln p B A − p B B ln p B B . Then we repeat for
l = 3, 4, ... and calculate the corresponding values for S BG (l). We then choose
another two-partition, say A and B , and repeat the whole procedure. Then we
do in principle for all possible two-partitions. Then we go to three-partitions, i.e.,
the alphabet will be now constituted by three letters, say A, B, and C. We repeat the
previous procedure for l = 1 (corresponding to the words A, B, C), then for l = 2
(corresponding to the words A A, AB, AC, B A, B B, BC, C A, C B, CC), etc. Then
we run windows with l = 3, 4, .... We then consider a different three-partition, say
A , B , and C . Then we consider four-partitions, and so on. Of all these entropies we
retain the supremum. In the appropriate limits of infinitely fine partitions and τ → 0
we obtain finally the largest rate of increase of the BG entropy. This is basically the
Kolmogorov–Sinai entropy rate.
Random documents with unrelated
content Scribd suggests to you:
 THE TOPOGRAPHICAL SURVEY OF INDIA.
At the close of the war with Tippoo Sahib, Major Lambton planned
the triangulation of the country lying between Madras and the
Malabar coast, a district which had been roughly surveyed, during
the progress of the war, by Colonel Mackenzie. The Duke of
Wellington gave his approval to the project, and his brother, the
Governor-General of India, and Lord Clive (son of the great Clive),
Governor of Madras, used their influence to aid Major Lambton in
carrying out his design. The only astronomical instrument made use
of by the first survey party was one of Ramsden’s zenith-sectors,
which Lord Macartney had placed in the hands of Dinwiddie, the
astronomer, for sale. A steel chain, which had been sent with Lord
Macartney’s embassy to the Emperor of China and refused, was the
only apparatus available for measuring.
Thus began the great Trigonometrical Survey of India, a work
whose importance it is hardly possible to over-estimate. Conducted
successively by Colonel Lambton, Sir George Everest, Sir Andrew
Waugh, and Lieut.-Col. Walker (the present superintendent), the
trigonometrical survey has been prosecuted with a skill and accuracy
which renders it fairly comparable with the best work of European
surveyors. But to complete in this style the survey of the whole of
India would be the work of several centuries. The trigonometrical
survey of Great Britain and Ireland has been already more than a
century in progress, and is still unfinished. It can, therefore, be
imagined that the survey of India—nearly ten times the size of the
British Isles, and presenting difficulties a hundredfold greater than
those which the surveyor in England has to encounter—is not a work
which can be quickly completed.
But the growing demands of the public service have rendered it
imperatively necessary that India should be rapidly and completely
surveyed. This necessity led to the commencement of the
Topographical Survey of India, a work which has been pushed
forward at a surprising rate during the past few years. My readers
may form some idea of the energy with which the survey is in
progress, from the fact that Colonel Thuillier’s Report for the season
1866-67 announces the charting of an area half as large as Scotland,
and the preparatory triangulation of an additional area nearly half as
large as England.
In a period of thirty years, with but few surveying parties at first,
and a slow increase in their number, an area of 160,000 square
miles has been completed and mapped by the topographical
department. The revenue surveyors have also supplied good maps
(on a similar scale) of 364,000 square miles of country during the
twenty years ending in 1866. Combining these results, we have an
area of 524,000 square miles, or upwards of four times that of Great
Britain and Ireland. For all this enormous area the surveyors have
the records in a methodical and systematic form, fit for incorporation
in the atlas of India. Nor does this estimate include the older
revenue surveys of the North-western Provinces which, for want of
proper supervision in former years, were never regularly reduced.
The records of these surveys were destroyed in the Mutiny—chiefly
in Hazaumbaugh and the south-western frontier Agency. The whole
of these districts remain to be gone over in a style very superior to
that of the last survey.
The extent of the country which has been charted may lead to the
impression that the survey is little more than a hasty
reconnaissance. This, however, is very far indeed from being the
case. The preliminary triangulation, which is the basis of the
topographical survey, is conducted with extreme care. In the present
Report, for instance, we find that the discrepancies between the
common sides of the triangles-in other words, the discrepancies
between the results obtained by different observers-are in some
cases less than one-tenth of an inch per mile; in others they are
from one inch to a foot per mile; and in the survey of the Cossyah
and Garrow Hills, where observations had to be taken to large
objects, such as trees, rocks, &c., with no defined points for
guidance, the results differ by as much as twenty-six inches per
mile. These discrepancies must not only be regarded as insignificant
in themselves, but must appear yet more trifling when it is
remembered that they are not cumulative, inasmuch as the
preliminary triangulation is itself dependent on the great
trigonometrical survey.
Let us understand clearly what are the various forms of survey
which are or have been in progress in India. There are three forms
to be considered:—(1) The Great Trigonometrical Surveys; (2) The
Revenue Surveys; and (3) the Topographical Surveys.
Great trigonometrical operations are extended in a straight course
from one measured base to another. Every precaution which modern
skill and science can suggest is taken in the measurement of each
base-line, and in the various processes by which the survey is
extended from one base-line to the other. The accuracy with which
work of this sort is conducted may be estimated from the following
instance. During the progress of the Ordnance Survey of Great
Britain and Ireland, a base-line nearly eight miles long was
measured near Lough Foyle, in Ireland, and another nearly seven
miles long on Salisbury Plain. Trigonometrical operations were then
extended from Lough Foyle to Salisbury Plain, a distance of about
340 miles; and the Salisbury base-line was calculated from the
observations made over this long arc. The difference between the
measured and calculated values of the base-line was less than five
inches! As we have stated, the trigonometrical survey of India will
bear comparison with the best work of our surveyors in England.
A revenue survey is prosecuted for the definition of the boundaries
of estates and properties. The operations of such a survey are
therefore carried on conformably to those boundaries.
The topographical survey of a country is defined by Sir A. Scott
Waugh to imply ‘the measurement and delineation of the natural
features of a country, and the works of man thereon, with the object
of producing a complete and sufficiently accurate map. Being free
from the trammels of boundaries of properties, the principal lines of
operations must conform to the features of the country, and objects
to be surveyed.’
The only safe basis for the topographical survey of a country is a
system of accurate triangulation. And where the extent of country to
be surveyed is large, there will always be a great risk of the
accumulation of error in the triangulation itself; which must,
therefore, be made to depend on the accurate results obtained by
the great trigonometrical operations. In order to secure this result,
fixed stations are established in the vicinity of the great
trigonometrical series. Where this plan cannot be adopted, a
network of large symmetrical triangles is thrown over the district to
be surveyed, or boundary series of triangles are carried along the
outline of the district or along convenient internal lines. The former
of these methods is applicable to a hilly district, the latter to a flat
country.
When the district to be surveyed has been triangulated, the work
of filling-in the topographical details is commenced. Each triangle
being of moderate extent, with sides from three to five miles in
length, and the angular points being determined, as we have seen,
with great exactness, it is evident that no considerable error can
occur in filling-in the details. Hence, methods can be adopted in the
final topographical work which would not be suitable for
triangulation. The triangles can either be ‘measured up,’ or the
observer may traverse from trigonometrical point to point, taking
offsets and intersections; or, lastly, he may make use of the plane
table. The two first methods require little comment; but the principle
of plane-tabling enters so largely into Indian surveying, that this
notice would be incomplete without a brief account of this simple
and beautiful method.
The plane-table is a flat board turning on a vertical pivot. It bears
the chart on which the observer is planning the country. Suppose,
now, that two points A and B are determined, and that we require to
mark in the position of a third point C:—It is clear that if we
observed with a theodolite the angles A B C and B A C, we might lay
these down on the chart with a protractor, and so the position of C
would be determined, with an accuracy proportioned to the care
with which the observations were made and the corresponding
constructions applied to the chart. But in ‘plane-tabling’ a more
direct plan is adopted. A ruler bearing sights, resembling those of a
rifle, is so applied that the edge passing through the point A on the
chart (the observer being situated at the real station A) passes
through the point B on the chart, the line of sight passing through
the real station B. The table being fixed in the position thus
obtained, the ruler is next directed so that its edge passes through
A, while the line of sight points to C. A line is now ruled with a pencil
through A towards C. In a similar manner, the table having been
removed to the station B, a pencil line is drawn through the point B
on the chart towards C. The two lines thus drawn determine by their
intersection the place of C on the chart.
The above is only one instance of the modes in which a plane-
table can be applied; there are several others. Usually the magnetic
compass is employed to fix the position of the table in accordance
with the true bearing of the cardinal points. Also the bearings of
several points are taken around each station; and thus a variety of
tests of the correctness of the work become applicable. Into such
details as these I need not here enter. It is sufficient that my readers
should have been enabled to recognise the simple principles on
which plane-tabling depends, and the accuracy with which (when
suitable precautions are taken) it can be applied as a method of
observation subsidiary to the ordinary trigonometrical processes.
‘A hilly country,’ says Sir A. Waugh, ‘offers the fairest field for the
practice of plane-table surveys, and the more rugged the surface the
greater will be the relative advantages and facilities this system
possesses over the methods of actual measurement. On the other
hand, in flat lands the plane-table works at a disadvantage, while
the traverse system is facilitated. Consequently, in such tracts, the
relative economy of the two systems does not offer so great a
contrast as in the former. In closely wooded or jungly tracts, all kinds
of survey operations are prosecuted at a disadvantage; but in such
localities, the commanding points must be previously cleared for
trigonometrical operations, which facilitates the use of the table.’
In whatever way the topographical details have been filled in, a
rigorous system of check must be applied to the work. The system
adopted is that of running lines across ground that has been
surveyed. This is done by the head of the party or by the chief
assistant-surveyor. A sufficient number of points are obtained in this
way for comparison with the work of the detail surveyors; and when
the discrepancies exceed certain limits, the work in which they
appear is rejected. Owing to the extremely unhealthy, jungly, and
rugged nature of the ground in which nearly all the Indian surveys
have been progressing, it has not always been found practicable to
check by regularly chained lines. There are, however, other modes of
testing plane-table surveys, and as these entail less labour and
expense in hilly and jungly tracts, and are quite as effective if
thoroughly carried out, they have been adopted generally, while the
measured routes or check-lines have only been pursued under more
favourable conditions. Colonel Thuillier states that ‘the inspection of
the work of every detailed surveyor in the field has been rigorously
enforced, and the work of the field season is not considered
satisfactory or complete unless this duty has been attended to.’
The rules laid down to insure accuracy in the survey are—first,
that the greatest possible number of fixed points should be
determined by regular triangulation; secondly, that the greatest
possible number of plane-table fixings should be made use of within
each triangle; and lastly, that eye-sketching should be reduced to a
minimum. If these rules are well attended to, the surveyor can
always rely on the value of the work performed by his subordinates.
But all these conditions cannot be secured in many parts of the
ground allotted to the several topographical parties owing to the
quantity of forest land and the extremely rugged nature of the
country. Hence arises the necessity for test-lines to verify the details,
or for some vigorous system of check; and this is more especially
the case where native assistants are employed.
So soon as the country has been accurately planned, the
configuration of the ground has to be sketched up. This process is
the end and aim of all the preceding work.
The first point attended to is the arterial system, or water
drainage, constituting the outfall of the country; whence are
deduced the lines of greatest depression of the ground. Next the
watersheds or ridges of hills are traced in, giving the highest level.
Lastly, the minor or subordinate features are drawn in with the
utmost precision attainable. ‘The outlines of table-land should be
well defined,’ says Sir A. Waugh, ‘and ranges of hills portrayed with
fidelity, carefully representing the watersheds or divortia aquarum,
the spurs, peaks, depressions or saddles, isthmuses or connecting-
links of separate ranges, and other ramifications. The depressed
points and isthmuses are particularly valuable, as being either the
sites of ordinary passes or points which new roads should conform
to.’
And here we must draw a distinction between survey and
reconnaissance. It is absolutely necessary in making a survey that
the outlines of ground as defined by ridges, water-courses, and feet
of hills should be rigorously fixed by actual observation and careful
measurement. In reconnoitring, more is trusted to the eye.
The scale of the Indian topographical survey is that of one inch
per mile; the scale of half an inch per mile being only resorted to in
very densely wooded or jungly country, containing a few inhabitants
and little cultivated, or where the climate is so dangerous that it is
desirable to accelerate the progress of the survey.
On the scale of one inch per mile the practised draughtsman can
survey about five square miles of average country per day. In
intricate ground, intersected by ravines or covered by hills of
irregular formation, the work proceeds much more slowly; on the
other hand, in open and nearly level country, or where the hills have
simple outlines, the work will cost less and proceed more rapidly. On
the scale of one inch per mile all natural features (such as ravines or
watercourses) more than a quarter of a mile in length can be clearly
represented. Villages, towns, and cities can be shown, with their
principal streets and roads, and the outlines of fortifications. The
general figure and extent of cultivated, waste, and forest lands can
be delineated with more or less precision, according to their extent.
Irrigated rice-lands should be distinctly indicated, since they
generally exhibit the contour of the ground.
The relative heights of hills and depths of valleys should be
determined during the course of a topographical survey. These
vertical elements of a survey can be ascertained by trigonometrical
or by barometrical observations, or by a combination of both
methods. ‘The barometer,’ says Sir A. Waugh, ‘is more especially
useful for determining the level of low spots from which the principal
trigonometrical stations are invisible. In using this instrument,
however, in combination with the other operations, the relative
differences of heights are to be considered the quantities sought, so
that all the results may be referable to the original trigonometrical
station. The height above the sea-level of all points coming under
any of the following heads is especially to be determined, for the
purpose of illustrating the physical relief of the country:—
‘1st. The peaks and highest points of ranges.
‘2nd. All obligatory points required for engineering works, such as
roads, drainage, and irrigation, viz.:—the highest points or necks of
valleys; the lowest depressions or passes in ranges; the junctions of
rivers, and débouchements of rivers from ranges; the height of
inundation-level, at moderate intervals of about three miles apart.
‘3rd. Principal towns or places of note.’
 Of the various methods employed to indicate the steepness of
slope, that of eye-contouring seems alone to merit special comment.
In true contouring, regular horizontal lines, at fixed vertical intervals,
are traced over a country, and plotted on to the maps. This is an
expensive and tedious process, whereas eye-contouring is easy,
light, and effective. On this system all that is necessary is that the
surveyor should consider what routes persons moving horizontally
would pursue. He draws lines on his chart approximating as closely
as possible to these imaginary lines. It is evident that when lines are
thus drawn for different vertical elevations, the resulting shading will
be dark or light, according as the slope is steep or gentle. This
method of shading affords scope as well for surveying skill as for
draughtsmanship.
(From Once a Week, May 1, 1869.)
 A SHIP ATTACKED BY A SWORD-FISH.
I have always been puzzled to imagine how the ‘nine-and-twenty
knights of fame,’ described in the ‘Lay of the Last Minstrel,’ managed
to ‘drink the red wine through the helmet barr’d.’ But in nature we
meet with animals which seem almost as inconveniently armed as
those chosen knights, who
. . . quitted not their armour bright,
Neither by day nor yet by night.
Amongst such animals the sword-fish must be recognised as one
of the most uncomfortably-armed creatures in existence. The shark
has to turn on his back before he can eat, and the attitude scarcely
seems suggestive of a comfortable meal. But the sword-fish can
hardly even by that arrangement get his awkwardly projecting snout
out of the way. Yet doubtless this feature, which seems so
inconvenient, is of great value to Xiphias. In some way as yet
unknown it enables him to get his living. Whether he first kills some
one of his neighbours with this instrument, and then eats him at his
leisure, or whether he plunges it deep into the larger sort of fish,
and attaching himself to them in this way, sucks nutriment from
them while they are yet alive, is not known to naturalists. Certainly,
he is fond of attacking whales, but this may result not so much from
gastronomic tastes as from a natural antipathy—envy, perhaps, at
their superior bulk. Unfortunately for himself, Xiphias, though cold-
blooded, seems a somewhat warm-tempered animal; and, when he
is angered, he makes a bull-like rush upon his foe, without always
examining with due care whether he is likely to take anything by his
motion. And when he happens to select for attack a stalwart ship,
and to plunge his horny beak through thirteen or fourteen inches of
planking, with perhaps a stout copper sheathing outside it, he is apt
to find some little difficulty in retreating. The affair usually ends by
his leaving his sword embedded in the side of the ship. In fact, no
instance has ever been recorded of a sword-fish recovering his
weapon (if I may use the expression) after making a lunge of this
sort. Last Wednesday the Court of Common Pleas—rather a strange
place, by-the-bye, for inquiring into the natural history of fishes—
was engaged for several hours in trying to determine under what
circumstances a sword-fish might be able to escape scot-free after
thrusting his snout into the side of a ship, The gallant ship
‘Dreadnought,’ thoroughly repaired, and classed A 1 at Lloyd’s, had
been insured for 3,000l. against all the risks of the seas. She sailed
on March 10, 1864, from Colombo, for London. Three days later, the
crew, while fishing, hooked a sword-fish. Xiphias, however, broke the
line, and a few moments after leaped half out of the water, with the
object, it would seem, of taking a look at his persecutor, the
‘Dreadnought.’ Probably he satisfied himself that the enemy was
some abnormally large cetacean, which it was his natural duty to
attack forthwith. Be this as it may, the attack was made, and at four
o’clock the next morning the captain was awakened with the
unwelcome intelligence that the ship had sprung a leak. She was
taken back to Colombo, and thence to Cochin, where she was hove
down. Near the keel was found a round hole, an inch in diameter,
running completely through the copper sheathing and planking.
As attacks by sword-fish are included among sea risks, the
insurance company was willing to pay the damages claimed by the
owners of the ship, if only it could be proved that the hole had really
been made by a sword-fish. No instance had ever been recorded in
which a sword-fish had been able to withdraw his sword after
attacking a ship. A defence was founded on the possibility that the
hole had been made in some other way. Professor Owen and Mr.
Frank Buckland gave their evidence; but neither of them could state
quite positively whether a sword-fish which had passed its beak
through three inches of stout planking could withdraw without the
loss of its sword. Mr. Buckland said that fish have no power of
‘backing,’ and expressed his belief that he could hold a sword-fish by
the beak; but then he admitted that the fish had considerable lateral
power, and might so ‘wriggle its sword out of a hole.’ And so the
insurance company will have to pay nearly six hundred pounds
because an ill-tempered fish objected to be hooked, and took its
revenge by running full tilt against copper sheathing and oak
planking.
(From the Daily News, December 11, 1868.)
 THE SAFETY-LAMP.
As recent colliery explosions have attracted a considerable amount
of attention to the principle of the safety-lamp, and questions have
arisen respecting the extent of the immunity which the action of this
lamp secures to the miner, it may be well for me briefly to point out
the true qualities of the lamp.
In the Davy lamp a common oil-light is surrounded by a cylinder
of wire-gauze. When the air around the lamp is pure the flame burns
as usual, and the only effect of the gauze is somewhat to diminish
the amount of light given out by the lamp. But so soon as the air
becomes loaded with the carburetted hydrogen gas generated in the
coal-strata, a change takes place. The flame grows larger and less
luminous. The reason of the change is this:—The flame is no longer
fed by the oxygen of the air, but is surrounded by an atmosphere
which is partly inflammable; and the inflammable part of the gas, so
fast as it passes within the wire cylinder, is ignited and burns within
the gauze. Thus the light now given out by the lamp is no longer
that of the comparatively brilliant oil flame, but is the light resulting
from the combustion of carburetted hydrogen, or ‘fire damp,’ as it is
called; and every student of chemistry is aware that the flame of this
gas has very little illuminating power.
So soon as the miner sees the flame thus enlarged and altered in
appearance he should retire. But it is not true that explosion would
necessarily follow if he did not do so. The danger is great because
the flame within the lamp is in direct contact with the gauze, and if
there is any defect in the wire-work, the heat may make for itself an
opening which—though small—would yet suffice to enable the flame
within the lamp to ignite the gas outside. So long, however, as the
wire-gauze continues perfect, even though it become red-hot, there
will be no explosion. No authority is required to establish this point,
which has been proved again and again by experiment; but I quote
Professor Tyndall’s words on the subject to remove some doubts
which have been entertained on the matter. ‘Although a continuous
explosive atmosphere,’ he says, ‘may extend from the air outside
through the meshes of the gauze to the flame within, ignition is not
propagated across the gauze. The lamp may be filled with an almost
lightless flame; still explosion does not occur. A defect in the gauze,
the destruction of the wire at any point by oxidation hastened by the
flame playing against it, would cause explosion;’ and so on. It need
hardly be said, however, that, imprudent as miners have often been,
no miner would remain where his lamp burned with the enlarged
flame indicative of the presence of fire-damp. The lamp should also
be at once extinguished.
But here we touch on a danger which undoubtedly exists, and—so
far as has yet been seen—cannot be guarded against by any amount
of caution. Supposing the miner sought to extinguish the lamp by
blowing it out, an explosion would almost certainly ensue, since the
flame can be forced mechanically through the meshes, though it will
not pass through them when it is burning in the ordinary way. Now
of course no miner who had been properly instructed in the use of
the safety-lamp would commit such a mistake as this. But it
happens, unfortunately, that sometimes the fire-damp itself forces
the flame of the lamp through the meshes. The gas frequently
issues with great force from cavities in the coal (in which it has been
pent up), when the pick of the miner breaks an opening for it. In
these circumstances an explosion is inevitable, if the issuing stream
of gas happen to be directed full upon the lamp. Fortunately,
however, this is a contingency which does not often arise. It is one
of those risks of coal-mining which seem absolutely unavoidable by
any amount of care or caution. It would be well if it were only such
risks as these that the miner had to face.
Another peculiarity sometimes noticed when there is a discharge
of fire-damp is worth mentioning. It happens, occasionally, that the
light will be put out owing to the absolute exclusion of air from the
lamp. This, however, can only happen when the gas issues in so
large a volume that the atmosphere of the pit becomes irrespirable.
With the exception of the one risk which we have pointed out
above, the Davy lamp may be said to be absolutely safe. It is
necessary, however, that caution and intelligence should be exhibited
in its use. On this point Professor Tyndall remarks that unfortunately
the requisite intelligence is not often possessed nor the requisite
caution exercised by the miner, ‘and the consequence is that even
with the safety-lamp, explosions still occur.’ And he suggests that it
would be well to exhibit to the miner in a series of experiments the
properties of the valuable instrument which has been devised for his
security. ‘Mere advice will not enforce caution,’ he says; ‘but let the
miner have the physical image of what he is to expect clearly and
vividly before his mind, and he will find it a restraining and monitory
influence long after the effect of cautioning words has passed away.’
A few words on the history of the invention may be acceptable.
Early in the present century a series of terrible catastrophes in coal
mines had excited the sympathy of enlightened and humane persons
throughout the country. In the year 1813, a society was formed at
Sunderland to prevent accidents in coal mines or at least to diminish
their frequency, and prizes were offered for the discovery of new
methods of lighting and ventilating mines. Dr. William Reid Clanny, of
Bishopwearmouth, presented to this society a lamp which burnt
without explosion in an atmosphere heavily loaded with fire-damp;
for which invention the Society of Arts awarded him a gold medal.
The Rev. Dr. Gray called the attention of Sir Humphry Davy to the
subject, and that eminent chemist visited the coal mines in 1815
with the object of determining what form of lamp would be best
suited to meet the requirements of the coal miners. He invented two
forms of lamp before discovering the principle on which the present
safety-lamps are constructed. This principle—the property, namely,
that flame will not pass through small apertures—had been, we
believe, discovered by Stephenson, the celebrated engineer, some
time before; and a somewhat angry controversy took place
respecting Davy’s claim to the honour of having invented the safety-
lamp. It seems admitted, however, by universal consent, that Davy’s
discovery of the property above referred to was made independently,
and also that he was the first to suggest the idea of using wire-
gauze in place of perforated tin.
In comparing the present frequency of colliery explosions with
what took place before the invention of the safety-lamp, we must
take into consideration the enormous increase in the coal trade since
the introduction of steam machinery. The number of miners now
engaged in our coal mines is far in excess of the number employed
at the beginning of the present century. Thus accidents in the
present day are at once more common on account of the increased
rapidity with which the mines are worked, and when they occur
there are more sufferers; so that the frequency of colliery explosions
in the opening years of the present century and the number of
deaths resulting from them, are in reality much more significant than
they seem to be at first sight. But even independently of this
consideration, the record of the colliery accidents which took place
at that time is sufficiently startling. Seventy-two persons were killed
in a colliery at North Biddick at the commencement of the present
century. Two explosions in 1805, at Hepburn and Oxclose, left no
less than forty-three widows and 151 children unprovided for. In
1808, ninety persons were killed in a coal-pit at Lumley. On May 24,
1812, ninety-one persons were killed by an explosion at Felling
Colliery, near Gateshead. And many more such accidents might
readily be enumerated.
(From the Daily News, December 4, 1868.)
 THE DUST WE HAVE TO BREATHE.
A microscopist, Mr. Dancer, F.R.A.S., has been examining the dust
of our cities. The results are not pleasing. We had always recognised
city dust as a nuisance, and had supposed that it derived the
peculiar grittiness and flintiness of its structure from the constant
macadamizing of city roads. But it now appears that the effects
produced by dust, when, as is usual, it finds its way to our eyes, our
nostrils, and our throats, are as nothing compared with the mischief
it is calculated to produce in a more subtle manner. In every
specimen examined by Mr. Dancer animal life was abundant. But the
amount of ‘molecular activity’—such is the euphuism under which
what is exceedingly disagreeable to contemplate is spoken about—is
variable according to the height at which the dust is collected. And
of all heights which these molecular wretches could select for the
display of their activity, the height of five feet is that which has been
found to be the favourite. Just at the average height of the foot-
passenger’s mouth these moving organisms are always waiting to be
devoured and to make us ill. And this is not all. As if animal
abominations were insufficient, a large proportion of vegetable
matter also disports itself in the light dust of our streets. The
observations show that in thoroughfares where there are many
animals engaged in the traffic, the greater part of the vegetable
matter thus floating about ‘consists of what has passed through the
stomachs of animals,’ or has suffered decomposition in some way or
other. This unpleasing matter, like the ‘molecular activity,’ floats
about at a height of five feet, or thereabouts.
After this, one begins to recognise the manner in which some
diseases propagate themselves. What had been mysterious in the
history of plagues and pestilences seems to receive at least a partial
solution. Take cholera, for example. It has been shown by the
clearest and most positive evidence that this disease is not
propagated in any way save one—that is, by the actual swallowing
of the cholera poison. In Professor Thudichum’s masterly paper on
the subject in the ‘Monthly Microscopical Journal,’ it is stated that
doctors have inhaled a full breathing from a person in the last stage
of this terrible malady without any evil effects. Yet the minutest
atom of the cholera poison received into the stomach will cause an
attack of cholera. A small quantity of this matter drying on the floor
of the patient’s room, and afterwards caused to float about in the
form of dust, would suffice to prostrate a houseful of people. We can
understand, then, how matter might be flung into the streets, and,
after drying, its dust be wafted through a whole district, causing the
death of hundreds. One of the lessons to be learned from these
interesting researches of Mr. Dancer is clearly this, that the watering-
cart should be regarded as one of the most important of our
hygienic institutions. Supplemented by careful scavengering, it might
be effective in dispossessing many a terrible malady which now
holds sway from time to time over our towns.
(From the Daily News, March 6, 1869.)
 PHOTOGRAPHIC GHOSTS.
On the outskirts of the ever-widening circle lighted up by science
there is always a border-land wherein superstition holds sway. ‘The
arts and sciences may drive away the vulgar hobgoblin of darker
days; but they bring with them new sources of illusion. The ghosts
of old could only gibber; the spirits of our day can read and write,
and play on divers musical instruments, and quote Shakespeare and
Milton. It is not, therefore, altogether surprising to learn that they
can take photographs also. You go to have your photograph taken,
we will suppose, desiring only to see your own features depicted in
the carte; and lo! the spirits have been at work, and a photographic
phantom makes its appearance beside you. It is true this phantom is
of a hazy and dubious aspect: the ‘dull mechanic ghost’ is indistinct,
and may be taken for anyone. Still, it is not difficult for the eye of
fancy to trace in it the lineaments of some departed friend, who, it is
to be assumed, has come to be photographed along with you. In
fact, photography, according to the spiritualist, resembles what
Byron called—
The lightning of the mind,
Which out of things familiar, undesigned,
When least we deem of such, calls up to view
The spectres whom no exorcism can bind.
The phenomena of spiritual photography were first observed some
years since, and a set of carte photographs were sent from America
to Dr. Walker, of Edinburgh, in which photographic phantoms were
very obviously, however indistinctly, discernible. More recently an
English photographer noticed a yet stranger circumstance, though
he was too sensible to seek for a supernatural interpretation of it.
When he took a photograph with a particular lens, there could be
seen not only the usual portrait of the sitter, but at some little
distance a faint ‘double,’ exactly resembling the principal image.
Superstitious minds might find this result even more distressing than
the phantom photographic friend. To be visited by the departed
through the medium of a lens, is at least not more unpleasing than
to hold converse with spirits through an ordinary ‘rapping’ medium.
But the appearance of a ‘double,’ or ‘fetch,’ has ever been held by
the learned in ghostly lore to signify approaching death.
Fortunately both one and the other appearance can be very easily
accounted for without calling in the aid of the supernatural. At a
recent meeting of the Photographical Society it was shown that an
image may often be so deeply impressed on the glass that the
subsequent cleaning of the plate, even with strong acids, will not
completely remove the picture. When the plate is used for receiving
another picture, the original image makes its reappearance, and as it
is too faint to be recognisable, a highly susceptible imagination may
readily transform it into the image of a departed friend. The ‘double’
is generated by the well-known property of double refraction,
obtained by a lens under certain circumstances of unequal pressure,
or sometimes by inequalities in the process of annealing. So vanish
two ghosts which might have been more or less troublesome to
those who are ready to see the supernatural in commonplace
phenomena. Will the time ever come when no more such phantoms
will remain to be exorcised?
(From the Daily News, March 2, 1869.)
 THE OXFORD AND CAMBRIDGE ROWING
STYLES.
Whatever opinion we may have of the result of the approaching
contest (1869), there can be no doubt that this year, as in former
years, there is a striking dissimilarity between the rowing styles of
the dark blue and the light blue oarsmen. This dissimilarity makes
itself obvious whether we compare the two boats as seen from the
side, or when the line of sight is directed along the length of either.
Perhaps it is in the latter aspect that an unpractised eye will most
readily detect the difference I am speaking of. Watch the Cambridge
boat approaching you from some distance, or receding, and you will
notice in the rise and fall of the oars, as so seen, the following
peculiarities—a long stay of the oar in the water, a quick rise from
and return to the water, the oars remaining out of the water for the
briefest possible interval of time. In the case of the Oxford boat
quite a different appearance is presented—there is a short stay in
the water, a sharp rise from and return to it, and between these the
oars appear to hang over the water for a perceptible interval. It is,
however, when the boats are seen from the side that the meaning of
these peculiarities is detected, and also that the fundamental
distinction between the two styles is made apparent to the
experienced eye. In the Cambridge boat we recognise the long
stroke and ‘lightning feather’ inculcated in the old treatises on
rowing: in the Oxford boat we see these conditions reversed, and in
their place the ‘waiting feather’ and lightning stroke. By the ‘waiting
feather’ I do not refer to what is commonly understood by slow
feathering, but to a momentary pause (scarcely to be detected when
the crew is rowing hard) before the simultaneous dash of the oars
upon the first grip of the stroke.15 And observing more closely—
which, by the way, is no easy matter—as either boat dashes swiftly
past, we detect the distinctive peculiarities of ‘work’ by which the
two styles are severally arrived at. In the Cambridge crew we see
the first part of the stroke done with the shoulders—precisely
according to the old-fashioned models—the arms straight until the
body has fallen back to an almost upright position; then comes the
sharp drop back of the shoulders beyond the perpendicular, the arms
simultaneously doing their work, so that as the swing back is
finished, the backs of the hands just touch the ribs in feathering. All
these things are quite in accordance with what used to be
considered the perfection of rowing; and, indeed, this style of rowing
has some important good qualities and a very handsome
appearance. The lightning feather, also, which follows the long
sweeping stroke, is theoretically perfect. Now, in the case of the
Oxford crew, we observe a style which at first sight seems less
excellent. As soon as the oars are dashed down and catch their first
hold of the water, the arms as well as the shoulders of each oarsman
are at work.16 The result is, that when the back has reached an
upright position, the arms have already reached the chest, and the
stroke is finished. Thus the Oxford stroke takes a perceptibly shorter
time than the Cambridge stroke; it is also, necessarily, somewhat
shorter in the water. One would, therefore, say it must be less
effective. Especially would an unpractised observer form this opinion,
because the Oxford stroke seems to be much shorter in range than
it is in reality. There we have the secret of its efficiency. It is actually
as long as the Cambridge stroke, but is taken in a perceptibly
shorter time. What does this mean but that the oar is taken more
sharply, and, therefore, much more effectively, through the water?
Much more effectively so far as the actual conditions of the
contest are concerned. The modern racing outrigger requires a sharp
impulse, because it will take almost any speed we can apply to it. It
will also retain that speed between the strokes, a consideration of
great importance. The old-fashioned racing-eights required to be
continually under propulsion. The lightning-feather was a necessity
in their case, for between every stroke the boat would lag terribly
with a slow-feathering crew. I do not say, of course, that the speed
of a light outrigged craft does not diminish between the strokes.
Anyone who has watched a closely contested bumping-race, and
noticed the way in which the sharply cut bow of the pursuing boat
draws up to the rudder of the other as by a succession of impulses,
although either boat seen alone would seem to sweep on with
almost uniform speed, will know that the motion of the lightest boat
is not strictly uniform. But there is an immense difference between
the almost imperceptible loss of way of a modern eight and the dead
‘lag’ in the old-fashioned craft. And hence we get the following
important consideration. Whereas with the old boats it was useless
for a crew to attempt to give a very quick motion to their boat by a
sharp, sudden ‘lift,’ this plan is calculated to be, of all others, the
most effective with the modern racing-eight.
It may seem, at first sight, that, after all, the result of the
Cambridge style should be as effective as that of the other. If arms
and shoulders do their work in both crews with equal energy—which
we may assume to be the case—and if the number of strokes per
minute is equal, the actual propulsive energy ought to be equal
likewise. A little consideration will show that this is a fallacy. If two
men pull at a weight together they will move it farther with a given
expenditure of energy than if first one and then the other apply his
strength to the work. And what is more to the purpose, they will be
able to move it faster. So shoulders and arms working
simultaneously will give a greater propulsive power than when
working separately, even though in the latter case each works with
its fullest energy. And not only so, but by the simultaneous use of
arms and shoulders, that sharpness of motion can alone be given
which is essential to the propulsion of a modern racing-boat.
I have said that the two crews are severally rowing in the style
which has lately been peculiar to their respective Universities. But
the Cambridge crew is rowing in that form of the Cambridge style
which brings it nearest to the requirements of modern racing. The
faults of the style are subdued, so to speak, and its best qualities
brought out effectively. In one or two of the long series of defeats
lately sustained by Cambridge the reverse has been the case. At
present, too, there is a certain roughness about the Oxford crew
which encourages the hopes of the light blue supporters. But it must
be admitted that this roughness is rather apparent than real, great
as it seems, and it will doubtless disappear before the day of
encounter. I venture to predict that the ‘time’ of the approaching
race, taken in conjunction with the state of the tide, will show the
present crews to be at least equal to the average.17
(From the Daily News, April 1869.)
BETTING ON HORSE RACES: OR, THE STATE OF
THE ODDS.
There appears every day in the newspapers an account of the
betting on the principal forthcoming races. The betting on such races
as the Two Thousand Guineas, the Derby, and the Oaks, often
begins more than a year before the races are run; and during the
interval, the odds laid against the different horses engaged in them
vary repeatedly, in accordance with the reported progress of the
animals in their training, or with what is learned respecting the
intentions of their owners. Many who do not bet themselves, find an
interest in watching the varying fortunes of the horses which are
held by the initiated to be leading favourites, or to fall into the
second rank, or merely to have an outside chance of success. It is
amusing to notice, too, how frequently the final state of the odds is
falsified by the event; how some ‘rank outsider’ will run into the first
place, while the leading favourites are not even ‘placed.’
It is in reality a simple matter to understand the betting on races
(or contests of any kind), yet it is astonishing how seldom those who
do not actually bet upon races have any inkling of the meaning of
those mysterious columns which indicate the opinion of the betting
world respecting the probable results of approaching contests,
equine or otherwise.
Let us take a few simple cases of ‘odds,’ to begin with; and,
having mastered the elements of our subject, proceed to see how
cases of greater complexity are to be dealt with.
Suppose the newspapers inform us that the betting is 2 to 1
against a certain horse for such and such a race, what inference are
we to deduce? To learn this let us conceive a case in which the true
odds against a certain event are as 2 to 1. Suppose there are three
balls in a bag, one being white, the others black. Then, if we draw a
ball at random, it is clear that we are twice as likely to draw a black
as to draw a white ball. This is technically expressed by saying that
the odds are 2 to 1 against drawing a white ball; or 2 to 1 on (that
is, in favour of) drawing a black ball. This being understood, it
follows that, when the odds are said to be 2 to 1 against a certain
horse, we are to infer that, in the opinion of those who have studied
the performance of the horse, and compared it with that of the other
horses engaged in the race, his chance of winning is equivalent to
the chance of drawing one particular ball out of a bag of three balls.
Observe how this result is obtained: the odds are 2 to 1, and the
chance of the horse is as that of drawing one ball out of a bag of
three—three being the sum of the two numbers 2 and 1. This is the
method followed in all such cases. Thus, if the odds against a horse
are 7 to 1, we infer that the cognoscenti consider his chance equal
to that of drawing one particular ball out of a bag of eight.
A similar treatment applies when the odds are not given as so
many to one. Thus, if the odds against a horse are as 5 to 2, we
infer that the horse’s chance is equal to that of drawing a white ball
out of a bag containing five black and two white balls—or seven in
all.
We must notice also that the number of balls may be increased to
any extent, provided the proportion between the total number and
the number of a specified colour remains unchanged. Thus, if the
odds are 5 to 1 against a horse, his chance is assumed to be
equivalent to that of drawing one white ball out of a bag containing
six balls, only one of which is white; or to that of drawing a white
ball out of a bag containing sixty balls, of which ten are white-and so
on. This is a very important principle, as we shall now see.
Suppose there are two horses (amongst others) engaged in a
race, and that the odds are 2 to 1 against one, and 4 to 1 against
the other-what are the odds that one of the two horses will win the
race? This case will doubtless remind my readers of an amusing
sketch by Leech, called—if I remember rightly—‘Signs of the
Commission.’ Three or four undergraduates are at a ‘wine,’
discussing matters equine. One propounds to his neighbour the
following question: I say, Charley, if the odds are 2 to 1 against
Rataplan, and 4 to 1 against Quick March, what’s the betting about
the pair?’—‘Don’t know, I’m sure,’ replies Charley; ‘but I’ll give you 6
to 1 against them.’ The absurdity of the reply is, of course, very
obvious; we see at once that the odds cannot be heavier against a
pair of horses than against either singly. Still, there are many who
would not find it easy to give a correct reply to the question. What
has been said above, however, will enable us at once to determine
the just odds in this or any similar case. Thus-the odds against one
horse being 2 to 1, his chance of winning is equal to that of drawing
one white ball out of a bag of three, one only of which is white. In
like manner, the chance of the second horse is equal to that of
drawing one white ball out of a bag of five, one only of which is
white. Now we have to find a number which is a multiple of both the
numbers three and five. Fifteen is such a number. The chance of the
first horse, modified according to the principle explained above, is
equal to that of drawing a white ball out of a bag of fifteen of which
five are white. In like manner, the chance of the second is equal to
that of drawing a white ball out of a bag of fifteen of which three are
white. Therefore the chance that one of the two will win is equal to
that of drawing a white ball out of a bag of fifteen balls of which
eight (five added to three) are white. There remain seven black
balls, and therefore the odds are 8 to 7 on the pair.
To impress the method of treating such cases on the mind of the
reader, let us take the betting about three horses—say 3 to 1, 7 to 2,
and 9 to 1 against the three horses respectively. Then their
respective chances are equal to the chance of drawing (1) one white
ball out of four, one only of which is white; (2) a white ball out of
nine, of which two only are white; and (3) one white ball out of ten,
one only of which is white. The least number which contains four,
nine, and ten is 180; and the above chances, modified according to
the principle explained above, become equal to the chance of
drawing a white ball out of a bag containing 180 balls, when 45, 40,
and 18 (respectively) are white. Therefore, the chance that one of
the three will win is equal to that of drawing a white ball out of a
bag containing 180 balls, of which 103 (the sum of 45, 40, and 18)
are white. Therefore, the odds are 103 to 77 on the three.
One does not hear in practice of such odds as 103 to 77. But
betting-men (whether or not they apply just principles of
computation to such questions, is unknown to me) manage to run
very near the truth. For instance, in such a case as the above, the
odds on the three would probably be given as 4 to 3—that is,
instead of 103 to 77 (or 412 to 308), the published odds would be
equivalent to 412 to 309.
And here a certain nicety in betting has to be mentioned. In
running the eye down the list of odds, one will often meet such
expressions as 10 to 1 against such a horse offered, or 10 to 1
wanted. Now, the odds of 10 to 1 taken may be understood to imply
that the horse’s chance is equivalent to that of drawing a certain ball
out of a bag of eleven. But if the odds are offered and not taken, we
cannot infer this. The offering of the odds implies that the horse’s
chance is not better than that above mentioned, but the fact that
they are not taken implies that the horse’s chance is not so good. If
no higher odds are offered against the horse, we may infer that his
chance is very little worse than that mentioned above. Similarly, if
the odds of 10 to 1 are asked for, we infer that the horse’s chance is
not worse than that of drawing one ball out of eleven; if the odds
are not obtained, we infer that his chance is better; and if no lower
odds are asked for, we infer that his chance is very little better.
Thus, there might be three horses (A, B, and C) against whom the
nominal odds were 10 to 1, and yet these horses might not be
equally good favourites, because the odds might not be taken, or
might be asked for in vain. We might accordingly find three such
horses arranged thus:—

Odds.
A 10 to 1 (wanted).
 B 10 to 1 (taken).
C 10 to 1 (offered).

Or these different stages might mark the upward or downward

progress of the same horse in the betting. In fact, there are yet
more delicate gradations, marked by such expressions respecting
certain odds, as—offered freely, offered, offered and taken (meaning
that some offers only have been accepted), taken, taken and
wanted, wanted, and so on.
As an illustration of some of the principles I have been
considering, let us take from the day’s paper,18 the state of the odds
respecting the ‘Two Thousand Guineas.’ It is presented in the
following form:—
TWO THOUSAND GUINEAS.

7 to 2 against Rosicrucian (off.).

6 to 1 against Pace (off.; 7 to 1 w.).
10 to 1 against Green Sleeve (off.).
100 to 7 against Blue Gown (off.).
180 to 80 against Sir J. Hawley’s lot (t.).

This table is interpreted thus: bettors are willing to lay the same
odds against Rosicrucian as would be the true mathematical odds
against drawing a white ball out of a bag containing two white and
seven black balls; but no one is willing to back the horse at this rate;
on the other hand, higher odds are not offered against him. Hence it
is presumable that his chance is somewhat less than that above
indicated. Again, bettors are willing to lay the same odds against
Pace as might fairly be laid against drawing one white ball out of a
bag of seven, one only of which is white; but backers of the horse
consider that they ought to get the same odds as might be fairly laid
against drawing the white ball when an additional black ball had
been put into the bag. As respects Green Sleeve and Blue Gown,
bettors are willing to lay the odds which there would be,
Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.

More than just a book-buying platform, we strive to be a bridge

connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.

Join us on a journey of knowledge exploration, passion nurturing, and

personal growth every day!

ebookbell.com