Developments in Mathematics

Dorina Mitrea
Irina Mitrea
Marius Mitrea

Geometric
Harmonic
Analysis III
Integral Representations, Calderón-
Zygmund Theory, Fatou Theorems,
and Applications to Scattering
Developments in Mathematics

Volume 74

Series Editors
Krishnaswami Alladi, Department of Mathematics, University of Florida,
Gainesville, FL, USA
Pham Huu Tiep, Department of Mathematics, Rutgers University, Piscataway, NJ,
USA
Loring W. Tu, Department of Mathematics, Tufts University, Medford, MA, USA

Aims and Scope

The Developments in Mathematics (DEVM) book series is devoted to publishing
well-written monographs within the broad spectrum of pure and applied mathe-
matics. Ideally, each book should be self-contained and fairly comprehensive in
treating a particular subject. Topics in the forefront of mathematical research that
present new results and/or a unique and engaging approach with a potential rela-
tionship to other fields are most welcome. High-quality edited volumes conveying
current state-of-the-art research will occasionally also be considered for publication.
The DEVM series appeals to a variety of audiences including researchers, postdocs,
and advanced graduate students.
Dorina Mitrea · Irina Mitrea · Marius Mitrea

Geometric Harmonic
Analysis III
Integral Representations, Calderón-Zygmund
Theory, Fatou Theorems, and Applications to
Scattering
Dorina Mitrea Irina Mitrea
Department of Mathematics Department of Mathematics
Baylor University Temple University
Waco, TX, USA Philadelphia, PA, USA

Marius Mitrea
Department of Mathematics
Baylor University
Waco, TX, USA

ISSN 1389-2177 ISSN 2197-795X (electronic)

Developments in Mathematics
ISBN 978-3-031-22734-9 ISBN 978-3-031-22735-6 (eBook)
https://doi.org/10.1007/978-3-031-22735-6

Mathematics Subject Classification: 32A, 26B20, 31B, 35J, 42B

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2023
This work is subject to copyright. All rights are solely and exclusively licensed 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
Dedicated with love to our parents
Prefacing the Full Series

The current work is part of a series, comprised of five volumes, [112], [113], [114],
[115], [116]. In broad terms, the principal aim is to develop tools in Real and
Harmonic Analysis, of geometric measure theoretic flavor, capable of treating a
broad spectrum of boundary value problems formulated in rather general geometric
and analytic settings.
In Volume I ([112]), we establish a sharp version of Divergence Theorem (aka
Fundamental Theorem of Calculus) which allows for an inclusive class of vector
fields whose boundary trace is only assumed to exist in a nontangential pointwise
sense.
Volume II ([113]) is concerned with function spaces measuring size and/or
smoothness, such as Hardy spaces, Besov spaces, Triebel-Lizorkin spaces, Sobolev
spaces, Morrey spaces, Morrey-Campanato spaces, spaces of functions of Bounded
Mean Oscillations, etc., in general, geometric settings. Work here also highlights
the close interplay between the differentiability properties of functions and singular
integral operators.
The topic of singular integral operators is properly considered in Volume III
([114]), where we develop a versatile Calderón-Zygmund theory for singular integral
operators of convolution type (and with variable coefficient kernels) on uniformly
rectifiable sets in the Euclidean ambient, and the setting of Riemannian manifolds.
Applications to scattering by rough obstacles are also discussed in this volume.
In Volume IV ([115]), we focus on singular integral operators of boundary layer
type which enjoy more specialized properties (compared with generic, garden variety
singular integral operators treated earlier in Volume III). Applications to Complex
Analysis in several variables are subsequently presented, starting from the real-
izations that many natural integral operators in this setting, such as the Bochner-
Martinelli operator, are actual particular cases of double layer potential operators
associated with the complex Laplacian.
In Volume V ([116]), where everything comes together, finer estimates for a
certain class of singular integral operators (of chord-dot-normal type) are produced
in a manner which indicates how their size is affected by the (infinitesimal and global)
flatness of the “surfaces” on which they are defined. Among the library of double

vii
viii Prefacing the Full Series

layer potential operators associated with a given second-order system, we then iden-
tify those double layers which fall under this category of singular integral operators.
It is precisely for this subclass of double layer potentials that Fredholm theory may
then be implemented assuming the underlying domain has a compact boundary,
which is sufficiently flat at infinitesimal scales. For domains with unbounded bound-
aries, this very category of double layer potentials may be outright inverted, using
a Neumann series argument, assuming the “surface” in question is sufficiently flat
globally. In turn, this opens the door for solving a large variety of boundary value
problems for second-order systems (involving boundary data from Muckenhoupt
weighted Lebesgue spaces, Lorentz spaces, Hardy spaces, Sobolev spaces, BMO,
VMO, Morrey spaces, Hölder spaces, etc.) in a large class of domains which, for
example, are allowed to have spiral singularities (hence more general than domains
locally described as upper-graphs of functions). In the opposite direction, we show
that the boundary value problems formulated for systems lacking such special layer
potentials may fail to be Fredholm solvable even for really tame domains, like the
upper half-space, or the unit disk. Save for the announcement [111], all principal
results appear here in print for the first time.
We close with a short epilogue, attempting to place the work undertaken in this
series into a broader picture. The main goal is to develop machinery of geometric
harmonic analysis flavor capable of ultimately dealing with boundary value problems
of a very general nature. One of the principal tools (indeed, the piecè de résistance)
in this regard is a new and powerful version of the Divergence Theorem, devised in
Volume I, whose very formulation has been motivated and shaped from the outset
by its eventual applications to Harmonic Analysis, Partial Differential Equations,
Potential Theory, and Complex Analysis. The fact that its footprints may be clearly
recognized in the makeup of such a diverse body of results, as presented in Volumes
II-V, serves as a testament to the versatility and potency of our brand of Divergence
Theorem. Alas, our enterprise is multifaceted, so its success is crucially dependent
on many other factors. For one thing, it is necessary to develop a robust Calderón-
Zygmund theory for singular integrals of boundary layer type (as we do in Volumes
III-IV), associated with generic weakly elliptic systems, capable of accommodating
a large variety of function spaces of interest considered in rather inclusive geometric
settings (of the sort discussed in Volume II). This renders these (boundary-to-domain)
layer potentials useful mechanisms for generating lots of null-solutions for the given
system of partial differential operators, whose format is compatible with the demands
in the very formulation of the boundary value problem we seek to solve. Next, in
order to be able to solve the boundary integral equation to which matters are reduced
in this fashion, the success of employing Fredholm theory hinges on the ability to
suitably estimate the essential norms of the (boundary-to-boundary) layer potentials.
In this vein, we succeed in relating the distance from such layer potentials to the
space of compact operators to the flatness of the boundary of the domain in question
(measured in terms of infinitesimal mean oscillations of the unit normal) in a desirable
manner which shows that, in a precise quantitative fashion, the flatter the domain the
smaller the proximity to compact operators. This subtle and powerful result, bridging
Prefacing the Full Series ix

between analysis and geometry, may be regarded as a far-reaching extension of the

pioneering work of Radon and Carleman in the early 1900’s.
Ultimately, our work aligns itself with the program stemming from A.P. Calderón’s
1978 ICM plenary address in which he advocates the use of layer potentials “for much
more general elliptic systems [than the Laplacian]” – see [15, p. 90], and may be
regarded as an optimal extension of the pioneering work of E.B. Fabes, M. Jodeit,
and N.M. Rivière in [41] (where layer potential methods have been first used to
solve boundary value problems for the Laplacian in bounded C 1 domains). In this
endeavor, we have been also motivated by the problem1 posed by A.P. Calderón on
[15, p. 95], asking to identify the function spaces on which singular integral operators
(of boundary layer type) are well defined and continuous. This is relevant since, as
Calderón mentions, “A clarification of this question would be very important in the
study of boundary value problems for elliptic equations [in rough domains]. The
methods employed so far seem to be insufficient for the treatment of these problems.”
We also wish to mention that our work is also in line with the issue raised as an
open problem by C. Kenig in [79, Problem 3.2.2, pp. 116–117], where he asked
whether operators of layer potential type may be inverted on appropriate Lebesgue
and Sobolev spaces in suitable subclasses on NTA domains with compact Ahlfors
regular boundaries.
The task of making geometry and analysis work in unison is fraught with diffi-
culties, and only seldom can a two-way street be built on which to move between
these two worlds without loss of information. Given this, it is actually surprising
that, in many instances, we come very close to having optimal hypotheses, almost
an accurate embodiment of the slogan if it makes sense to write it, then it’s true.
Acknowledgments: The authors gratefully acknowledge partial support from the
Simons Foundation (through grants # 426669, # 958374, # 318658, # 616050, and
# 637481), as well as NSF (grant # 1900938). Portions of this work have been
completed at Baylor University in Waco, Temple University in Philadelphia, the
Institute for Advanced Study in Princeton, MSRI in Berkeley, and the American
Institute of Mathematics in San Jose. We wish to thank these institutions for their
generous hospitality. Last, but not least, we are grateful to Michael E. Taylor for
gently yet persistently encouraging us over the years to complete this project.

Waco, USA Dorina Mitrea

Philadelphia, USA Irina Mitrea
Waco, USA Marius Mitrea

1 In the last section of [15], simply titled “Problems,” Calderón singles two directions for further
study. The first one is the famous question whether the smallness condition on a   L ∞ (the Lipschitz
constant of the curve {(x, a(x)) : x ∈ R} on which he proved the L 2 -boundedness of the Cauchy
operator) may be removed (as is well known, this has been solved in the affirmative by Coifman,
McIntosh, and Meyer in [20]). We are referring here to the second (and final) problem formulated
by Calderón on [15, p. 95].
Description of Volume III

From the outset, the very formulation of our versions of the Divergence Theorem
from Volume I (cf. [112, §1.2–§1.12]) has been motivated and shaped by potential
applications to Harmonic Analysis, Partial Differential Equations, Function Space
Theory, and Complex Analysis. We have envisioned these versions of the Divergence
Theorem not as end-products, in and of themselves, but as effective tools to further
progress in these areas of mathematics. This has already become apparent in Volume
II ([113]), when dealing with function spaces measuring smoothness of Sobolev type
on the geometric measure theoretic boundaries of sets of locally finite perimeter.
In the opening chapter of the present volume (Chapter 1, titled “Integral Repre-
sentations and Integral Identities”), we further elaborate on this vision. We begin
in §1.1 by revisiting the classical Cauchy-Pompeiu integral representation formula
in open sets  ⊆ C with a lower Ahlfors regular boundary and whose arc-length
σ := H1 ∂ is a doubling measure. Our Divergence Theorem specialized to this
setting then permits us to identify a very general class of functions for which the
Cauchy-Pompeiu integral representation formula is valid. By means of counterexam-
ples, we show that the analytic conditions imposed are in the nature of best possible.
A very general version of the Cauchy integral representation formula, allowing one
to recover a holomorphic function from its (nontangential) boundary trace via the
(boundary-to-domain) Cauchy integral operator, is then obtained as a corollary. In the
same spirit, generalizations of the classical Morera Theorem and Residue Theorem
are established. Variants with no explicit lower Ahlfors regularity assumptions made
on the topological boundary are also discussed.
This line of work continues in §1.2 where higher-dimensional versions of some
of the main results from §1.1 are extended to open subsets of Rn with n ∈ N, n ≥ 2
arbitrary, now involving the Clifford algebra Cn (cf. the discussion in [112, §6.4]) in
place of the field of complex numbers C, and the Dirac operator D (from (A.0.37))
in lieu of the Cauchy-Riemann operator ∂. Once again, all integral representation
formulas in Clifford Analysis derived here make essential use of our brand of the
Divergence Theorem from Volume I ([112]). A more general point of view is adopted
in §1.6 where integral representation formulas are derived, under rather inclusive
geometric and analytic assumptions, for injectively elliptic first-order systems.

xi
xii Description of Volume III

Granted the availability of a fundamental solution for second-order, homogeneous,

constant (complex) coefficient, weakly elliptic systems, which exhibits a number of
desirable properties (as discussed in [109, Theorem 11.1, pp. 393–395]), in §1.5 and
§1.7, we derive boundary layer potential representations and Green-type formulas for
such systems, under minimal geometric and analytic assumptions. Chapter 1 ends
with a section (§1.8) on Rellich-type identities in open sets with Ahlfors regular
boundaries.
Chapter 2 contains a powerful and nuanced Calderón-Zygmund theory for singular
integral operators on uniformly rectifiable sets. The results here, which build on the
work of many predecessors, including Calderón, Zygmund, Mikhlin, Coifman, McIn-
tosh, Meyer, David, Jerison, Kenig, Semmes, and Stein, among many others, enhance,
refine, and sharpen those in [63] by placing more economical demands on the ambient
geometry. As a preamble, in §2.1, we consider integral operators on Hölder spaces on
upper Ahlfors regular sets, and in §2.2, we deal with singular integrals on Lebesgue
spaces on Ahlfors regular quasi-metric spaces. The latter section contains a number
of results in the spirit of Calderón-Zygmund theory, and a rather versatile version of
Cotlar’s Lemma. Principal-value singular integral operators on uniformly rectifiable
sets are treated in earnest in §2.3, while, in §2.4, we consider boundary-to-domain
integral operators on open sets with uniformly rectifiable boundaries.
The results established so far are combined in §2.5to produce a very versatile
jump-formula, identifying the nontangential trace of a boundary-to-domain inte-
gral operator in an open set with a uniformly rectifiable boundary as a jump-term
(involving the Fourier transform of the integral kernel function) and a principal-value
singular integral operator on the boundary (of the sort alluded to earlier). In §2.9,
we present some applications to integral operators whose kernels are fashioned by
applying a suitable number of derivatives to the fundamental solutions of certain
elliptic differential operators. In §2.8, the last section in Chapter 2, we extend the
scope of the Calderón-Zygmund theory developed so far by considering singular
integral operators on uniformly rectifiable sets, with variable coefficient kernels that
are actually the Schwartz kernels of classical pseudodifferential operators of order
−1 with odd principal symbol, acting between vector bundles over a Riemannian
manifold.
The main goal in Chapter 3 is to derive quantitative Fatou-type theorems in
uniformly rectifiable domains. The distinctive format of a Fatou-type theorem is
that size/integrability properties of the nontangential maximal operator for a null-
solution of an elliptic equation in a certain domain implies the a.e. existence of
the pointwise nontangential boundary trace of the function in question. The quan-
titative aspect is reflected in the fact that not only the boundary trace exists but
also contains considerable information concerning the size and regularity of the
original function. In §3.1, one of the central results is a quantitative Fatou-type
theorem for null-solutions of injectively elliptic first-order (homogeneous, constant
complex coefficient) systems of differential operators in arbitrary uniformly rectifi-
able domains. In turn, this theorem has a large spectrum of applications, including
the theory of Hardy spaces associated with injectively elliptic first-order systems
in UR domains discussed in §3.2. Fatou-type theorem for second-order systems is
Description of Volume III xiii

subsequently established in §3.3 where, among other things, we prove a quantitative

Fatou-type theorem for the gradient of null-solutions of second-order systems in UR
domains, along with a version in which both the trace of the function and its gradient
are shown to exist. Lastly, in §3.4, we show that for a null-solution of a weakly
elliptic system, having sufficient regularity on the Sobolev/Besov/Triebel-Lizorkin
scales in a given bounded two-sided NTA domain with an Ahlfors regular boundary
guarantees the existence of its nontangential pointwise trace on the boundary of the
domain in question.
Chapter 4 is primarily reserved for examining the role of Green functions in
establishing uniqueness for the Dirichlet Problem for weakly elliptic homogeneous,
constant (complex) coefficient, second-order systems, formulated in broad geometric
settings. In this vein, see Theorem 4.1.1 and Theorem 4.1.6, which may be regarded
as sharp embodiments of the heuristic principle asserting that uniqueness holds in the
Dirichlet/Regularity Problem for a given system L and domain  granted the exis-
tence of Green functions which “pair well” with the null-solutions of the boundary
value problem in question. A sensible feature of these results is the absence of
concrete function spaces in their formulation, which makes them directly applicable
to a wide range of specific cases of interest. Basically, all results in this chapter make
essential use of the brand of Divergence Theorem developed earlier, in Volume I
([112]).
Chapter 5 continues the discussion initiated in Chapter 4 by further specializing
matters to the case of the Laplace operator. Here, we are concerned with issues of
a potential theoretic nature such as: When is the Poisson kernel associated with a
domain  ⊆ Rn (as the Radon-Nikodym derivative of the harmonic measure with
respect to the surface measure) expressible as the normal derivative of the Green
function? When is said Poisson kernel bounded from below away from zero? When
can one represent the solution of the Dirichlet Problem with L p data as an integral
involving the Poisson kernel? Throughout, we are interested in minimal regularity
conditions, a perspective through which such questions have been raised, for example,
by J. Garnett and D.E. Marshall in [49].
In Chapter 6, the last chapter in this volume, we treat certain fundamental aspects
of scattering theory. Concretely, in §6.1, we produce integral representations in exte-
rior domains of a very general geometric nature for null-solutions of the Helmholtz
operator which satisfy Sommerfeld’s radiation condition at infinity. In §6.2–§6.7, we
develop a unified approach to scattering theory, which identifies the broadest possible
family of radiation conditions for null-solutions of the vector Helmholtz oper-
ator. This family contains, as particular cases, the Sommerfeld, Silver-Müller, and
McIntosh-Mitrea radiation conditions corresponding to scattering by acoustic waves,
electromagnetic waves, and null-solutions of perturbed Dirac operators, respectively.
The new idea is that one can associate a radiation condition with each factorization
of the (vector) Helmholtz operator (as a product of two first-order systems).
To close, we wish to remark that all these applications make essential use of our
earlier versions of the Divergence Theorem.
Contents

Prefacing the Full Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

Description of Volume III . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
1 Integral Representations and Integral Identities . . . . . . . . . . . . . . . . . . . 1
1.1 One Variable Complex Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Integral Representation Formulas in Clifford Analysis . . . . . . . . . . . 37
1.3 First and Second-Order Elliptic Systems of Partial Differential
Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
1.4 Fundamental Solutions for Weakly Elliptic Second-Order
Systems of Partial Differential Operators . . . . . . . . . . . . . . . . . . . . . . . 59
1.5 Boundary Layer Potential Representations for Weakly
Elliptic Second-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
1.6 Integral Representation Formulas for Injectively Elliptic
First-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
1.7 Green-Type Formulas for Second-Order Systems . . . . . . . . . . . . . . . . 188
1.8 Rellich Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets . . . . . . . . . . 267
2.1 Integral Operators Acting on Hölder Spaces on Upper Ahlfors
Regular Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces . . . . . . . 272
2.3 Principal Value Singular Integral Operators on Uniformly
Rectifiable Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
2.4 Boundary-to-Domain Integral Operators on Open sets
with Uniformly Rectifiable Boundaries . . . . . . . . . . . . . . . . . . . . . . . . 382
2.5 The Jump-Formula for Boundary-to-Domain Integral
Operators in Open Sets with Uniformly Rectifiable Boundaries . . . 407
2.6 Singular Integrals on Morrey Spaces and Their Pre-Duals . . . . . . . . 514
2.7 Commutator Estimates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531

xv
xvi Contents

2.8 Calderón-Zygmund Theory for Singular Integrals

on Riemannian Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 558
2.9 Some Applications to Singular Integrals Associated
with Elliptic Differential Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . 624
3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains . . . . . . 633
3.1 Quantitative Fatou-Type Theorems in UR Domains
for First-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 634
3.2 Brief Look at Hardy Spaces Associated with Injectively
Elliptic First-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 682
3.3 Quantitative Fatou-Type Theorems in UR Domains
for Second-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 690
3.4 Fatou-Type Theorems in Two-Sided NTA Domains
for Second-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 726
3.5 Fatou-Type Theorems on Riemannian Manifolds . . . . . . . . . . . . . . . . 731
4 Green Functions and Uniqueness for Boundary Problems
for Second-Order Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 769
4.1 The Role of Green Functions in Uniqueness Issues . . . . . . . . . . . . . . 770
4.2 The Reciprocity Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 787
4.3 How to Construct Green Functions and Use Them
in Uniqueness Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 792
4.4 A Sharp Poisson Integral Representation Formula . . . . . . . . . . . . . . . 799
4.5 The Poisson Kernel Associated with a System: A First Look . . . . . . 813
4.6 The Poisson Kernel Associated with a System in the Upper
Half-Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 818
4.7 More on Uniqueness and Poisson Integral Representations . . . . . . . 824
5 Green Functions and Poisson Kernels for the Laplacian . . . . . . . . . . . . 829
5.1 Upper/Lower Semicontinuous Functions and Super/Sub
Harmonic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 829
5.2 The Harmonic Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 836
5.3 The Green Function for the Laplacian . . . . . . . . . . . . . . . . . . . . . . . . . 842
5.4 The Poisson Kernel in NTA Domains . . . . . . . . . . . . . . . . . . . . . . . . . . 853
5.5 Hardy Spaces of Harmonic Functions in NTA Domains . . . . . . . . . . 855
5.6 Boundary Behavior of the Green Function in NTA Domains . . . . . . 857
5.7 The L p Dirichlet Problem for the Laplacian and Integral
Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 861
5.8 The Nature of the Critical Index p and Further Results
on the Green Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 870
6 Scattering by Rough Obstacles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 881
6.1 Integral Representations for Null-Solutions of the Helmholtz
Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 881
6.2 Radiation Conditions: Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 886
6.3 The Family of Radiation Conditions (RC A ) . . . . . . . . . . . . . . . . . . . . 889
Contents xvii

6.4 Single and Double Acoustic Layer Potentials . . . . . . . . . . . . . . . . . . . 897

6.5 L 2 -Finiteness Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 901
6.6 The Principal Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 904
6.7 Coordinate-Free Formalism and Examples . . . . . . . . . . . . . . . . . . . . . 908

Appendix A: Terms and notation used in Volume III . . . . . . . . . . . . . . . . . . 921

References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 953
Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 961
Symbol Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 965
Chapter 1
Integral Representations and Integral Identities

Working in the complex plane, in §1.1 we prove versions of the classical Cauchy-
Pompeiu integral representation formula, Morera’s Theorem, the Residue Theorem,
and the Schwarz-Pompeiu formula under minimal smoothness assumptions. This
line of work continues in §1.2 where higher-dimensional Euclidean spaces are con-
sidered, now working in the context of Clifford algebras. Integral representation
formulas are then derived in domains of a very general geometric nature, first involv-
ing boundary layer potentials associated with weakly elliptic second-order systems
in §1.5, then in relation to injectively elliptic first-order systems in §1.6. Next, Green-
type formulas for second-order systems under optimal assumptions are presented in
§1.7, while the last section in this chapter (§1.8) concerns Rellich-type identities in
a rather inclusive setting.
Throughout, we seek to work in geometric and analytic settings which are fairly
optimal, in light of the conclusions we have in mind. Indeed, the background hy-
potheses we typically adopt are not too far off from the kind of assumptions one
would have to impose to simply have a meaningfully formulated conclusion. In this
sense, our results are an almost accurate embodiment of the slogan if it makes sense
to write it, then it’s true.

1.1 One Variable Complex Analysis

We begin with a brief historical survey of the Cauchy integral operator and relat-
ed topics, designed to highlight a number of major landmarks and breakthroughs.
Cauchy’s integral reproducing formula for holomorphic functions apparently first ap-
peared in 1831 (cf. [17]). Subsequently, in his 1873 dissertation [163], Y.V. Sokhotski
studied the boundary behavior of the Cauchy integral operator and derived jump-
formulas under Hölder regularity assumptions on the density function. Another
significant achievement of Sokhotski’s work was pursuing the study of the Cauchy
integral operator as a topic of independent interest. In 1885, A. Harnack re-derived
Sokhotski’s jump-formulas in [57] by decomposing the Cauchy integral operator

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 1

D. Mitrea et al., Geometric Harmonic Analysis III, Developments in Mathematics 74,
https://doi.org/10.1007/978-3-031-22735-6_1
2 1 Integral Representations and Integral Identities

as the two-dimensional harmonic double layer plus the tangential derivative of the
two-dimensional harmonic single layer, under rather restrictive hypotheses imposed
both on the underlying domain and the density function. However, it is J. Plemelj’s
work in 1908 which has been particularly influential on this topic. In [145] and [146]
a new, more rigorous, derivation of Sokhotski jump-formulas is presented (under
Hölder regularity assumptions on the density function), and the Cauchy integral
operator is used for the first time as a key tool in solving a boundary value problem
(of transmission type). In 1912 D. Pompeiu has discovered an integral representa-
tion formula involving the Cauchy operator along with a double integral taking into
account the failure of the function to be holomorphic. First published in a series of
papers [148, 149, 150] in 1912, and then revisited in 1913 in [151], this basic integral
representation formula has become known as the Cauchy-Pompeiu formula. Other
significant contributions are due to I.I. Privalov who, in 1918 has proved Sokhotski’s
jump-formulas rigorously by today’s standards and established the boundedness of
Cauchy’s operator on Hölder spaces on Lyapunov domains, and N.I. Muskhelishvili
who, starting in 1922, has systematically employed Cauchy type integral operators
to solve boundary value problems in the theory of elasticity. Significantly, it was in
S.G. Mikhlin’s 1948 paper [106] where the action of the Cauchy integral operator
has been first considered on square-integrable functions. Assuming the underlying
curve Σ to be of class 𝒞1+ε , for some ε > 0, Mikhlin has shown that the principal
value Cauchy operator is bounded on L 2 (Σ, H 1 ). Under the same regularity assump-
tion on Σ, this result was subsequently extended to a bound on L p (Σ, H 1 ) for each
p ∈ (1, ∞) by B.V. Khvedelidze in [83] (see also the more timely exposition in [52,
Vol. I, Theorem 2.1, p. 19]). The year 1977 marks A.P. Calderón’s breakthrough, per-
taining to L p -bounds, 1 < p < ∞, for the principal-value Cauchy integral operator
on Lipschitz curves in the plane with a sufficiently small Lipschitz constant, in [14].
Subsequently, in 1982, R. Coifman, A. McIntosh, and Y. Meyer have refined in [20]
Calderón’s result for the Cauchy integral operator by allowing arbitrary Lipschitz
curves. Finally, in 1984 G. David has succeeded in characterized in [30] the class of
H 1 -measurable, connected, sets Σ ⊂ C for which the Cauchy operator is bounded
on L p (Σ, H 1 ) as those for which1, 2
 
H 1 B(z, r) ∩ Σ
sup < ∞. (1.1.1)
z ∈C, r >0 r

Our first application is a very general version of the Cauchy-Pompeiu represen-

tation formula, alluded to above, for functions in the complex plane. To state it,
recall first that the Cauchy-Riemann operator ∂ in the plane and its conjugate are,
respectively, defined as

1 the necessity of condition (1.1.1) is elementary and has been known a little earlier; it appeared in
the paper [143] where the conjecture was made that this is also sufficient for the boundedness of
the Cauchy operator (see also the exposition in [52, Vol. I, Theorem 3.3, pp. 27-28])
2 of course, condition (1.1.1) expresses the fact that Σ is upper Ahlfors regular (also occasionally
referred to as the Carleson condition, particularly in the Russian literature)
1.1 One Variable Complex Analysis 3

∂ := ∂z̄ := 12 (∂x + i∂y ) and ∂ := ∂z := 12 (∂x − i∂y ), (1.1.2)

√
where i := −1 ∈ C. Given any Lebesgue measurable set Ω ⊆ R2 ≡ C of locally
finite perimeter, introduce σ := H 1 ∂Ω and denote by ν the geometric measure
theoretic outward unit normal to Ω. We then define the complex arc-length
measure on ∂Ω as
dζ := iν(ζ) dσ(ζ) on ∂∗ Ω. (1.1.3)
When dζ regarded as a measure in R2 supported on ∂∗ Ω, which is further iden-
tified with a distribution in R2 , from (1.1.3), (1.1.2), [112, (5.6.10)], and the fact
that σ∗ = H n−1 ∂∗ Ω, we conclude that the complex arc-length measure on ∂Ω
satisfies 3
dζ = −2i∂1Ω (1.1.4)
in the sense of distributions in R2 (this should be compared with the result described
in [112, Proposition 5.6.3]).

Theorem 1.1.1 Let Ω ⊆ R2 ≡ C be an open set with a lower Ahlfors regular

boundary, such that σ := H 1 ∂Ω is a doubling measure on ∂Ω. In particular, Ω
is a set of locally finite perimeter, and its geometric measure theoretic outward unit
normal ν is defined σ-a.e. on ∂∗ Ω.
In this context, suppose u : Ω → C is an L 2 -measurable complex-valued function
which, for some κ > 0, satisfies
∫
(Nκ u)(ζ) κ−n.t.
dσ(ζ) < +∞ and u∂Ω exists σ-a.e. on ∂nta Ω. (1.1.5)
∂Ω 1 + |ζ |

Also, with the Cauchy-Riemann operator ∂ taken in the sense of distributions in Ω,

assume that
 dL 2 (ζ) 
∂u ∈ L 1 Ω, . (1.1.6)
1 + |ζ |
κ −n.t.
Then for any κ > 0 the nontangential trace u∂Ω exists σ-a.e. on ∂nta Ω and
is actually independent of κ . Moreover, with the dependence on the parameter κ
dropped,

for each fixed z ∈ Ω, the function defined σ-a.e. on ∂∗ Ω as

 n.t. 
u (ζ)
∂Ω
∂∗ Ω ζ −→ belongs to L 1 (∂∗ Ω, σ), (1.1.7)
ζ−z

and for L 2 -a.e. point z ∈ Ω one has (with dζ denoting the complex arc-length
measure on ∂Ω; cf. (1.1.3))

3 if Ω is a domain of class 𝒞1 then one may check without difficulty that (1.1.4) holds in the sense
of distributions in the plane
4 1 Integral Representations and Integral Identities
∫  n.t.  ∫
1 u∂Ω (ζ) 1 (∂u)(ζ)
u(z) = dζ − dL 2 (ζ), (1.1.8)
2πi ∂∗ Ω ζ−z π Ω ζ−z
provided Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when Ω
is unbounded and ∂Ω is bounded (that is, when Ω is an exterior domain), formula
(1.1.8) holds under the additional assumption that there exists λ ∈ (1, ∞) such that
⨏
|u| dL 2 = o(1) as R → ∞. (1.1.9)
B(0,λ R)\B(0,R)

As a corollary, if actually ∂u = 0 in Ω then one has the following integral

representation formula
∫  n.t. 
1 u∂Ω (ζ)
u(z) = dζ, ∀z ∈ Ω, (1.1.10)
2πi ∂∗ Ω ζ−z

provided Ω is bounded, or ∂Ω is unbounded. In addition, when Ω is unbounded

and ∂Ω is bounded, formula (1.1.10) continues to be valid provided (1.1.9) is also
assumed.

Before presenting the proof of this theorem we make a series of remarks.

Remark 1. Together, [112, Lemma 8.3.1] and the first property in (1.1.5) imply
∞
u ∈ Lloc (Ω, L 2 ). (1.1.11)

In particular, u ∈ Lloc
1 (Ω, L 2 ) which ensures that it is meaningful to consider the

distribution ∂u in (1.1.6), and to speak of Lebesgue points of u (as done in relation

to (1.1.15)).
Remark 2. Regarding the first condition in (1.1.5) observe that
∂Ω upper Ahlfors regular, ⎫
⎪
⎪ ∫
⎬
⎪ (Nκ u)(ζ)
and Nκ u ∈ L p (∂Ω, σ) =⇒ dσ(ζ) < +∞, (1.1.12)
⎪
⎪ ∂Ω 1 + |ζ |
with 1 ≤ p < ∞ ⎭ ⎪

thanks to Hölder’s inequality and [112, Lemma 7.2.1]. Also, obviously,

whenever Ω has a bounded topological boundary, we have

∫ (1.1.13)
(Nκ u)(ζ)
dσ(ζ) < +∞ ⇐⇒ Nκ u ∈ L 1 (∂Ω, σ).
∂Ω 1 + |ζ |

We also wish to note that it is possible to further relax the first condition in (1.1.5)
and replace it by the demand that, for some ε > 0,
1.1 One Variable Complex Analysis 5
∫   κ−n.t.  
 u (ζ)
∂Ω
Nκε u ∈ Lloc
1
(∂Ω, σ) and dσ(ζ) < +∞. (1.1.14)
∂∗ Ω 1 + |ζ |
Such a version would require using [112, Corollary 1.5.2] (in the version recorded
in [112, (1.5.23)]) in place of [112, Theorem 1.2.1], as we do in the proof of
Theorem 1.1.1 given further below.
Remark 3. Concerning the integrability condition on ∂u, [112, Lemma 3.5.7] en-
sures that (1.1.6) is equivalent with having, for L 2 -a.e. point z ∈ C,
∫  
 (∂u)(ζ)  2
  dL (ζ) < +∞. (1.1.15)
Ω ζ −z

Let us also note here that under the assumption

p
∂u ∈ Lloc (Ω, L 2 ) for some p ∈ (1, ∞), (1.1.16)
1, p
elliptic regularity implies, bearing in mind (1.1.11), that u ∈ Wloc (Ω) (for general
results of this nature see [153], [177]). In particular, u is locally Hölder continuous
in Ω whenever (1.1.16) holds with p > 2.
Remark 4. In the case when Ω is an exterior domain, there is a version of the
integral representation formula (1.1.8) which brings to the forefront the behavior
of u at infinity in a manner different than the decay condition imposed in (1.1.9).
Specifically, from [112, (1.4.5)], [112, Proposition 4.7.1] and (1.1.56) it follows that
if Ω is an exterior domain in C and one assumes that
u is continuous in a neighborhood of infinity
⨏ (1.1.17)
and |u(z)| dH 1 (z) = o(R) as R → ∞,
|z |=R

(hence, in particular, if u is continuous and bounded in a neighborhood of infinity)

then the limit
⨏
u∞ := lim u(z) dH 1 (z) exists in C (1.1.18)
R→∞ |z |=R

and for each Lebesgue point z ∈ Ω of u with the property that (1.1.15) holds one has
∫  n.t.  ∫
1 u∂Ω (ζ) 1 (∂u)(ζ)
u(z) = dζ − dL 2 (ζ) + u∞ . (1.1.19)
2πi ∂∗ Ω ζ−z π Ω ζ−z

Remark 5. In the case when Ω is an exterior domain, (1.1.9) is satisfied if we impose

the pointwise decay condition

u(ζ) = o(1) as |ζ | → ∞. (1.1.20)

6 1 Integral Representations and Integral Identities

Remark 6. It turns out that if [112, Theorem 1.5.1] is used in place of [112,
Theorem 1.4.1] in the proof of Theorem 1.1.1, then

the doubling assumption on σ := H 1 ∂Ω may be relaxed to (1.1.21)

simply asking that σ is a locally finite measure on ∂Ω.
The price to pay is having to demand that the aperture parameter κ is sufficiently
large (depending on Ω) to begin with, and we may lose the flexibility of changing it
when considering nontangential boundary traces. Modulo these nuances, the format
of the main result (i.e., the integral representation formula (1.1.8)) remains the same.
This being said, the lower Ahlfors regularity condition for ∂Ω may not be simply
dropped. To see this, consider the open subset Ω := B(0, 1)\{0} of C and the function
u : Ω → C defined as u(ζ) := ζ1 for each ζ ∈ Ω. It is clear that σ := H 1 ∂Ω is a
locally finite measure. Also, the function u is holomorphic in Ω, and the conditions in
 n.t. 
(1.1.5) are satisfied for any κ > 0. In addition, u∂Ω (ζ) = ζ1 for every ζ ∈ ∂B(0, 1).
If we now fix z ∈ Ω and define f (ζ) := ζ (ζ1−z) for ζ ∈ C \ {z, 0}, we have that f is
meromorphic with poles of order one at 0 and z. Hence, the boundary integral in the
right-hand side of (1.1.8) may be computed using the Residue Theorem as
∫ 1 1
f (ζ) dζ = 2πi Res( f , z) + Res( f , 0) = 2πi − = 0. (1.1.22)
∂B(0,1) z z

Given that the solid integral in (1.1.8) is zero, if (1.1.8) were to hold in this case,
we would obtain u(z) = 0 for each z ∈ Ω, a contradiction. This shows that the lower
Ahlfors regularity condition for ∂Ω may not be simply discarded.
Remark 7. Both assumptions in (1.1.5) are necessary.  To see that this is thecase,
consider the open subset of C described as Ω := z ∈ B(0, 1) : z  [0, 1) and
bring in the holomorphic function u : Ω → C given by u(ζ) := ζ1 for each ζ ∈ Ω.
Then the boundary integral in (1.1.8) is just as in (1.1.22), hence zero. Thus, (1.1.8)
becomes u(z) = 0 for each z ∈ Ω, a contradiction.
In this scenario, Ω satisfies all geometric hypotheses stipulated in Theorem 1.1.1,
and u satisfies all but the first condition in (1.1.5).  The
 latter presently fails.
Specifically, since for each fixed κ > 0 we have Nκ u (x) ≈ x −1 uniformly for
x ∈ (0, 1) ⊆ ∂Ω, it follows that

Nκ u ∈ L 1,∞ (∂Ω, σ) but Nκ u  L 1 (∂Ω, σ). (1.1.23)

This shows that the first condition in (1.1.5) is indeed necessary.

As regards the necessity of the second condition in (1.1.5), consider the open
subset of C described as
 
Ω := z ∈ B(0, 1) : z  − 12 , + 12 × {0} , (1.1.24)
 
pick some f ∈ 𝒞0c (− 12 , + 12 ) which is not identically zero, then define the function
1.1 One Variable Complex Analysis 7
∫
1 1/2
f (t)
u(z) := dt, ∀z ∈ Ω. (1.1.25)
2πi −1/2 z−t
Then u is holomorphic in Ω, and
∫
 n.t.  1 1/2
f (t)
u∂Ω (ζ) = dt for each ζ ∈ ∂∗ Ω = ∂B(0, 1). (1.1.26)
2πi −1/2 ζ −t

Consequently, Fubini’s Theorem gives that for each point z ∈ Ω we have

∫  n.t.  ∫
u (ζ) 1 ∫ 1/2
f (t)  dζ
∂Ω
dζ = dt
∂∗ Ω ζ−z 2πi ∂B(0,1) −1/2 ζ −t ζ−z
∫ 1/2  1 ∫ 
dζ
= f (t) dt
−1/2 2πi ∂B(0,1) (ζ − t)(ζ − z)

= 0, (1.1.27)

since the last integral on the unit circle vanishes, by the Residue Theorem. As such,
if the integral representation formula (1.1.8) were to hold, it would presently imply
that u(z) = 0 for each
 z ∈ Ω. However, this is not the case. For example, for L -a.e.
1

point x ∈ − 2 , + 2 we have
1 1

∫ ∫
1 1  x−ε +1/2  f (t)
lim+ u(x ± iy) = ± f (x) + lim+ + dt, (1.1.28)
y→0 2 ε→0 2πi −1/2 x+ε z−t

which implies that

 
there exists a Lebesgue measurable set A ⊆ − 12 , + 12 with L 1 (A) > 0
such that lim+ u(x + iy) − lim+ u(x − iy) = f (x)  0 for each x ∈ A. (1.1.29)
y→0 y→0

In the current setting, Ω satisfies all geometric hypotheses stipulated in Theo-

rem 1.1.1, and u satisfies all but the second condition in (1.1.5). However, (1.1.29)
proves that the latter fails and, even though
κ−n.t.
u∂Ω does exist at σ-a.e. point on ∂∗ Ω, (1.1.30)

the failure of the second condition in (1.1.5) ultimately invalidates (1.1.8).

Remark 8. Another example which highlights the necessity of the first condition in
(1.1.5) is as follows. Consider
1+z
Ω := B(0, 1), u : Ω → C, u(z) := e 1−z , ∀z ∈ Ω. (1.1.31)

Note that u extends to a holomorphic function in C \ {1} (which actually satisfies

|u(z)| ≤ 1 for |z| > 1), and a simple calculation shows that
8 1 Integral Representations and Integral Identities
 sin θ

u(eiθ ) = ei 1−cos θ ∈ ∂B(0, 1), ∀θ ∈ (0, 2π). (1.1.32)

Thus, for each κ > 0 we trivially have

 κ−n.t.
u∂Ω exists σ-a.e. on ∂Ω and
u as in (1.1.31) =⇒ (1.1.33)
belongs to L ∞ (∂Ω, σ) → L 1 (∂Ω, σ).
√
On the other hand, given any κ > 2, it follows that |eiθ −(1− θ)| < (1+ κ)θ provided
θ ∈ (0, 1) is sufficiently small. Hence, 1 − θ ∈ Γκ (eiθ ) for every θ ∈ (0, 1) small, so
 
Nκ u (eiθ ) ≥ |u(1 − θ)| ≈ e2/θ , uniformly for θ ∈ (0, 1) small, (1.1.34)

which goes to show that

u as in (1.1.31) =⇒ Nκ u  L 1 (∂Ω, σ). (1.1.35)

As such, the Cauchy-Pompeiu representation formula (1.1.8), which in this case

 κ−n.t. 
simply becomes u = C u∂Ω where C is the Cauchy operator for the unit disk,
necessarily fails since otherwise one would have (for any choice of p ∈ (1, ∞))
       κ−n.t.   
Nκ u ≤ cp Nκ u L p (∂Ω,σ) = cp Nκ C u∂Ω  L p (∂Ω,σ)
L 1 (∂Ω,σ)
 κ−n.t.   κ−n.t. 
≤ Cp u∂Ω  L p (∂Ω,σ) ≤ Cp u∂Ω  L ∞ (∂Ω,σ) < +∞, (1.1.36)

contradicting (1.1.35).
Remark 9. Consider the case when Ω is a UR domain in C ≡ R2 . In this situation,
we have ∂∗ Ω = ∂Ω, hence ∂nta Ω has full H 1 measure in ∂Ω. As noted later, in
Comment 4 following the statement of Theorem 3.1.6, if u is holomorphic in Ω and
satisfies
Nκ u ∈ L p (∂Ω, σ) for some p ∈ (1, ∞), (1.1.37)
then thanks to [112, (7.2.5)] and the Fatou-type result from Theorem 3.1.6 both
conditions in (1.1.5) are satisfied. If actually Ω ⊂ C ≡ R2 is a bounded NTA domain
with an Ahlfors regular boundary, then the second condition in (1.1.5) is guaranteed
by the Fatou-type result from Proposition 5.5.2.
These observations permit streamlining the assumptions made in relation to the
version of Cauchy’s reproducing formula recorded in (1.1.10) in the geometric
settings mentioned above.
In the proof of Theorem 1.1.1, presented a little further below, we shall need the
following extension of [109, Exercise 7.47, p. 292].

Lemma 1.1.2 Let Ω be an arbitrary open set in R2 ≡ C, and consider a complex-

1 (Ω, L 2 ) with the property that ∂u ∈ L 1 (Ω, L 2 ) (where the
valued function u ∈ Lloc loc
Cauchy-Riemann operator ∂ is applied in the sense of distributions in Ω).
1.1 One Variable Complex Analysis 9

Then for each Lebesgue point z ∈ C of u with the additional property that the
complex-valued functions

u(ζ) (∂u)(ζ)
Ω ζ → and Ω ζ → 1
belong to Lloc (Ω, L 2 ), (1.1.38)
ζ−z ζ−z
one has
 u(ζ)  (∂u)(ζ)
∂ζ = π u(z)δz + in D (Ω), (1.1.39)
ζ−z ζ−z
where δz denotes the Dirac distribution with mass at z in Ω.
In particular, formula (1.1.39) holds for L 2 -a.e. point z ∈ Ω.

Proof Let us check (1.1.39). For this, without loss of generality we may assume that
0 ∈ Ω and z = 0, the origin in C. Select a real-valued function θ ∈ 𝒞∞ (C) with
the property
 that θ = 0 on B(0, 1) and θ = 1 on C \ B(0, 2). For each parameter
ε ∈ 0, 12 dist(0, ∂Ω) define θ ε : C → R by setting θ ε (ζ) := θ(ζ/ε) for every
ζ ∈ C. Then

1 − θ ε ∈ 𝒞∞
c (Ω), supp (∇θ ε ) ⊆ B(0, 2ε) \ B(0, ε),
(1.1.40)
and lim+ θ ε (ζ) = 1 for every ζ ∈ C \ {0}.
ε→0

Also, there exists a constant C ∈ (0, ∞) such that

 
sup (∇ j θ ε )(ζ) ≤ Cε −j for every j ∈ N0 . (1.1.41)
ζ ∈C

To proceed, fix ϕ ∈ 𝒞∞
c (Ω) and write
  u(ζ)    u(ζ) 
D (Ω) ∂ζ ,ϕ D(Ω) = − D (Ω) , ∂ϕ D(Ω)
ζ ζ
∫
u(ζ)
=− (∂ϕ)(ζ) dL 2 (ζ)
Ω ζ
∫
u(ζ)
= − lim+ (∂ϕ)(ζ)θ ε (ζ) dL 2 (ζ) (1.1.42)
ε→0 Ω ζ

where for the last equality in (1.1.42) we have used (1.1.40)-(1.1.41) and Lebesgue’s
Dominated Convergence Theorem. Next, observe that θ ε ϕ ∈ 𝒞∞ c (Ω) and 0 is not
contained in supp (θ ε ϕ). Consequently, the complex-valued function
1
Ω ζ −→ ϕ(ζ)θ ε (ζ) belongs to 𝒞∞
c (Ω), (1.1.43)
ζ
hence
u(ζ)
Ω ζ → ϕ(ζ)θ ε (ζ) belongs to Lcomp
1
(Ω, L 2 ) → ℰ (Ω). (1.1.44)
ζ
10 1 Integral Representations and Integral Identities

Keeping these in mind, in the sense of distributions in Ω we may compute

 u(ζ)  (∂u)(ζ) u(ζ)
∂ζ ϕ(ζ)θ ε (ζ) = ϕ(ζ)θ ε (ζ) + (∂ϕ)(ζ)θ ε (ζ)
ζ ζ ζ
u(ζ)
+ ϕ(ζ)(∂θ ε )(ζ) (1.1.45)
ζ
and, further,
 
 u(ζ) 
0 = ℰ (Ω) ∂ ζ ϕ(ζ)θ ε (ζ) , 1 ℰ(Ω)
ζ
∫ ∫
(∂u)(ζ) u(ζ)
= ϕ(ζ)θ ε (ζ) dL 2 (ζ) + (∂ϕ)(ζ)θ ε (ζ) dL 2 (ζ)
Ω ζ Ω ζ
∫
u(ζ)
+ ϕ(ζ)(∂θ ε )(ζ) dL 2 (ζ). (1.1.46)
Ω ζ

In turn, this implies

∫ ∫
u(ζ) (∂u)(ζ)
(∂ϕ)(ζ)θ ε (ζ) dL 2 (ζ) = − ϕ(ζ)θ ε (ζ) dL 2 (ζ)
Ω ζ Ω ζ
∫
u(ζ)
− ϕ(ζ)(∂θ ε )(ζ) dL 2 (ζ). (1.1.47)
Ω ζ

Thanks to assumptions and Lebesgue’s Dominated Convergence Theorem, we may

compute
∫ ∫
(∂u)(ζ) (∂u)(ζ)
lim ϕ(ζ)θ ε (ζ) dL 2 (ζ) = ϕ(ζ) dL 2 (ζ)
ε→0+ Ω ζ Ω ζ
 (∂u)(ζ) 
= D (Ω) ,ϕ D(Ω) . (1.1.48)
ζ
To deal with the last integral in (1.1.47), split it as I + II where
∫
1 1
I := − [(ϕu)(ζ) − (ϕu)(0)](∂θ)(ζ/ε) dL 2 (ζ),
ε Ωζ
∫ 1 
II := − (∂θ ε )(ζ) dL 2 (ζ) (ϕu)(0). (1.1.49)
Ω ζ

Using the support condition and estimate from (1.1.40)-(1.1.41), the fact that 0 is
a Lebesgue point of u (hence also of ϕu), and the properties of ϕ, term I may be
estimated by
1.1 One Variable Complex Analysis 11
 ⨏  
|I| ≤ C sup |∇θ| (ϕu)(ζ) − (ϕu)(0) dL 2 (ζ) −−−−→ 0. (1.1.50)
C B(0,2ε) ε→0+

Concerning II, observe that for each ε > 0 we have

∫ ∫
1 1
− (∂θ ε )(ζ) dL 2 (ζ) = − ∂[θ ε − 1](ζ) dL 2 (ζ)
Ω ζ Ω ζ
 1 
= D (Ω) ∂ ζ , θ ε − 1 D(Ω)
ζ

= π δ, θ ε − 1 = π(θ ε − 1)(0) = −π, (1.1.51)

by virtue of the fact that 1/(πζ) is a fundamental solution for ∂ ζ (cf., e.g., [109,
Theorem 7.43, p. 289]), and the properties of θ ε . Finally, by combining (1.1.42)
with (1.1.47)-(1.1.51) we arrive at the version of (1.1.39) written for z = 0. This
concludes the proof of the first claim in the statement of the lemma.
There remains to deal with the very last claim in the statement of the lemma. In
this regard, observe that since for every compact set K ⊂ Ω and every ball B ⊂ C
we have
∫ ∫   
 u(ζ)  2
  dL (ζ) dL 2 (z)
B K ζ −z
∫  ∫
dL 2 (z) 
≤C |u(ζ)| dL 2 (ζ) sup < +∞, (1.1.52)
K z ∈K B |ζ − z|

since u ∈ Lloc
1 (Ω, L 2 ). From this we then deduce that for L 2 -a.e. point z ∈ C we

have ∫  
 u(ζ)  2
  dL (ζ) < +∞ for every compact set K ⊆ Ω. (1.1.53)
K ζ −z

Likewise, using the fact that ∂u ∈ Lloc

1 (Ω, L 2 ) we conclude that for L 2 -a.e. point

z ∈ C we have
∫  
 (∂u)(ζ)  2
  dL (ζ) < +∞ for every compact set K ⊆ Ω. (1.1.54)
K ζ−z
Together, (1.1.53)-(1.1.54) imply that

(1.1.38) holds at L 2 -a.e. point z ∈ C. (1.1.55)

Since Lebesgue’s Differentiation Theorem also ensures that L 2 -a.e. z ∈ Ω is a

Lebesgue point for u, from what we have already established in the first part of the
proof we conclude that formula (1.1.39) is valid at L 2 -a.e. point z ∈ Ω. 

At last, we are ready to present the proof of Theorem 1.1.1.

12 1 Integral Representations and Integral Identities

Proof of Theorem 1.1.1 From assumptions and [112, Corollary 8.9.9] it follows that
κ −n.t.
for any κ > 0 the nontangential trace u∂Ω exists σ-a.e. on ∂nta Ω and is actually
independent of κ . Next, the claim in (1.1.7) is seen from (1.1.5), [112, (8.9.8)], [112,
(8.9.44)], and [112, (8.8.52)].
Fix now a Lebesgue point z ∈ Ω of u with the property that (1.1.15) holds, and
define the vector field
 u(ζ) u(ζ) 
Fz (ζ) := ,i for L 2 -a.e. ζ ∈ Ω. (1.1.56)
ζ−z ζ−z
Then from (1.1.56) and (1.1.11) we see that

Fz ∈ Lloc
2
1
(Ω, L 2 ) . (1.1.57)

Also, with δz denoting the Dirac distribution with mass at z, we have

 u(ζ)  (∂u)(ζ)
divFz (ζ) = 2∂ ζ = 2π u(z)δz + 2 in D (Ω), (1.1.58)
ζ−z ζ−z
where the last equality, making use of the fact that z is a Lebesgue point for u and
that (1.1.15) holds, comes from Lemma 1.1.2. In particular, if we now consider
 
K := B z, dist(z, ∂Ω)/2 , then K is a compact set contained in Ω, and (1.1.58)
together with the assumption made in (1.1.15) imply

divFz belongs to ℰK (Ω) + L 1 (Ω, L 2 ). (1.1.59)

Going further, (1.1.56) and [112, (8.3.45)] allow us to estimate

 Ω\K  √  
Nκ Fz (ζ) ≤ 2 Nκ u (ζ) · sup |η − z| −1
η ∈Γκ (ζ )\K

(Nκ u)(ζ)
≤ Cκ for all ζ ∈ ∂Ω. (1.1.60)
|ζ − z|
In turn, from [112, (8.2.26)], (1.1.60), and the first condition in (1.1.5) we conclude
that

NκΩ\K Fz ∈ L 1 (∂Ω, σ). (1.1.61)

κ−n.t.
Moreover, (1.1.56) and the assumptions on u imply that Fz ∂Ω exists σ-a.e. in ∂nta Ω
and in fact
  κ−n.t.   κ−n.t.  
 κ−n.t.  u∂Ω (ζ) u∂Ω (ζ)

Fz  (ζ) = ,i for σ-a.e. ζ ∈ ∂nta Ω. (1.1.62)
∂Ω ζ−z ζ−z

Finally, observe that in the case when Ω is an exterior domain, condition (1.1.9)
implies that
1.1 One Variable Complex Analysis 13
∫
| Fz | dL 2 = o(R) as R → ∞. (1.1.63)
B(0,λ R)\B(0,R)

In summary, we have proved that Fz satisfies all hypotheses in [112, Theo-
rem 1.4.1]. In combination with (1.1.58) and (1.1.62) (and also assuming (1.1.9) in
the case when Ω is an exterior domain), this permits us to write
∫
(∂u)(ζ)  
2π u(z) + 2 dL 2 (ζ) = (𝒞∞b (Ω))∗ divFz , 1 𝒞∞b (Ω)
Ω ζ−z
∫
 n.t. 
= ν, Fz ∂Ω dσ
∂∗ Ω

∫    n.t.   n.t.  
u∂Ω (ζ) u∂Ω (ζ)
= ν(ζ), ,i dσ(ζ)
∂∗ Ω ζ−z ζ−z

∫  n.t. 
u (ζ)
∂Ω
= ν(ζ) dσ(ζ)
∂∗ Ω ζ−z
∫  n.t. 
1 u∂Ω (ζ)
= dζ, (1.1.64)
i ∂∗ Ω ζ−z

where the last equality uses (1.1.3). From this (1.1.8) follows, at each Lebesgue point
z ∈ Ω of u such that (1.1.15) holds, hence at L 2 -a.e. point z ∈ Ω. 

We continue by presenting an integration by parts formula for the Cauchy-

Riemann operator involving singular functions. This result is going to be a key
ingredient in the treatment of the Residue Theorem for rough domains subsequently
discussed in Theorem 1.1.4.

Theorem 1.1.3 Suppose Ω ⊆ R2 ≡ C is an open set with a lower Ahlfors regular

boundary, and with the property that σ := H 1 ∂Ω is a doubling measure on ∂Ω. In
this setting, assume f ∈ D (Ω) has the property that there exists some compact set
K ⊂ Ω and some aperture parameter κ > 0 for which
   
f Ω\K is holomorphic in Ω \ K, NκΩ\K f Ω\K ∈ L 1 (∂Ω, σ),
κ−n.t. (1.1.65)
and f ∂Ω exists (in C) at σ-a.e. point on ∂nta Ω.
κ −n.t.
Then ∂ f ∈ ℰ (Ω), and for any κ > 0 the nontangential trace f ∂Ω exists σ-a.e.
on ∂nta Ω and is actually independent of κ . When regarding the latter function as
being defined σ-a.e. on ∂∗ Ω, this belongs to L 1 (∂∗ Ω, σ) and, with the dependence
on the parameter κ dropped,
∫
  1  n.t. 
ℰ (Ω) ∂ f , 1 = f  (ζ) dζ (1.1.66)
ℰ(Ω) 2i ∂∗ Ω ∂Ω
14 1 Integral Representations and Integral Identities

in the case when either Ω is bounded, or ∂Ω is unbounded. Moreover, if Ω is un-

bounded and ∂Ω is bounded, formula (1.1.66) continues to hold under the additional
assumption that there exists some λ ∈ (1, ∞) such that
∫
| f | dL 2 = o(R) as R → ∞. (1.1.67)
B(0,λ R)\B(0,R)

In the particular case corresponding to having K = , the above result yields

the following version of Cauchy’s vanishing formula. Let Ω and σ be as before, and
assume f is a holomorphic function in Ω with the property that there exists κ > 0
such that
κ−n.t.
Nκ f ∈ L 1 (∂Ω, σ) and f ∂Ω exists (in C) at σ-a.e. point on ∂nta Ω. (1.1.68)
κ −n.t.
Then for any parameter κ > 0 the nontangential trace f ∂Ω exists σ-a.e. on ∂nta Ω
and is actually independent of κ . The latter function, when regarded as being defined
σ-a.e. on ∂∗ Ω, belongs to L 1 (∂∗ Ω, σ) and, with the dependence on the parameter
κ dropped, Cauchy’s vanishing formula
∫
 n.t. 
f ∂Ω (ζ) dζ = 0 (1.1.69)
∂∗ Ω

holds in the case when either Ω is bounded, or ∂Ω is unbounded. Finally, if Ω

is unbounded and ∂Ω is bounded, formula (1.1.69) continues to hold under the
additional assumption made in (1.1.67).
In relation to Theorem 1.1.3 it is worth observing that it is possible to further
 
relax the condition that NκΩ\K f Ω\K ∈ L 1 (∂Ω, σ) in (1.1.68) and replace it by the
demand that, for some sufficiently small ε > 0,
∫
  κ−n.t.  
ε
Nκ f ∈ Lloc (∂Ω, σ) and
1  f (ζ) dσ(ζ) < +∞. (1.1.70)
∂Ω
∂∗ Ω

Such a version would require using [112, Corollary 1.5.2] (in the version recorded
in [112, (1.5.23)]) in place of [112, Theorem 1.4.1] (as we do in the proof of
Theorem 1.1.3 given below).
We also wish to note that in Comment 4, following the statement of Theorem 3.1.6,
a Fatou-type result has been presented according to which, in the case when Ω is
actually a UR domain, assuming the σ-a.e. existence of the nontangential pointwise
κ−n.t.
trace f ∂Ω (i.e., assuming the second condition in (1.1.68)) becomes superfluous.
This observation permits us to streamline the statement of Theorem 1.1.3 in the class
of UR domains.
Proof of Theorem 1.1.3 The assumptions on Ω imply that this is a set of locally finite
perimeter, hence its geometric measure theoretic
 outward unit normal ν = (ν1, ν2 )
is defined σ-a.e. on ∂∗ Ω. The fact that f Ω\K is holomorphic in Ω \ K forces
supp(∂ f ) ⊆ K, hence ∂ f ∈ ℰK (Ω). If we now consider the vector field
1.1 One Variable Complex Analysis 15
 
F := f , i f ∈ D (Ω)
2
(1.1.71)

then
κ−n.t.
the nontangential trace F∂Ω exists (in C2 ) σ-a.e. on ∂nta Ω,
    (1.1.72)
FΩ\K ∈ Lloc
1 (Ω \ K, L 2 ) 2, and N Ω\K F
κ

Ω\K
∈ L 1 (∂Ω, σ).

Also,
divF = ∂x f + i∂y f = 2∂ f in D (Ω), (1.1.73)
hence
divF ∈ ℰK (Ω). (1.1.74)
Moreover, in the case when Ω is an exterior domain, assumption (1.1.67) implies
that F satisfies [112, (1.4.8)]. Finally, we note that at σ-a.e. point on ∂∗ Ω we have
 κ−n.t.   κ−n.t.  κ−n.t.  κ−n.t. 
ν, F∂Ω = ν1 f ∂Ω + iν2 ( f ∂Ω = ν( f ∂Ω . (1.1.75)

Granted all these, formula (1.1.66) now follows from [112, Theorem 1.4.1], bear-
ing in mind [112, (4.6.21)] and (1.1.3). This finishes the proof of Theorem 1.1.3. 

Here is the statement of our version of the Residue Theorem for rough domains
advertised earlier.

Theorem 1.1.4 Let Ω ⊆ R2 ≡ C be an open set with a lower Ahlfors regular

boundary, and with the property that σ := H 1 ∂Ω is a doubling measure on ∂Ω.
Suppose f is a meromorphic function in Ω whose poles are contained in some
compact set K ⊂ Ω. Assume that, for some aperture parameter κ > 0,
   κ−n.t.
NκΩ\K f Ω\K ∈ L 1 (∂Ω, σ) and f ∂Ω
(1.1.76)
exists (in C) at σ-a.e. point on ∂nta Ω.
κ −n.t.
Then for any κ > 0 the nontangential trace f ∂Ω exists σ-a.e. on ∂nta Ω and
is actually independent of κ . When regarding the latter function as defined σ-a.e.
on ∂∗ Ω, this belongs to L 1 (∂∗ Ω, σ) and, with the dependence on the parameter κ
dropped, ∫
  n.t. 
2πi · Res( f , z) = f  (ζ) dζ
∂Ω
(1.1.77)
z pole of f ∂∗ Ω

in the case when either Ω is bounded, or ∂Ω is unbounded. Finally, when Ω is an

exterior domain, formula (1.1.77) continues to hold under
∫ the additional assumption
that there exists some number λ ∈ (1, ∞) such that B(0,λ R)\B(0,R) | f | dL 2 = o(R)
as R → ∞.

Proof Let {z j }1≤ j ≤ N ⊆ K be the distinct poles of the meromorphic function f .

More specifically, assume that for each index j ∈ {1, . . . , N } there exists some open
16 1 Integral Representations and Integral Identities

set O j ⊆ Ω with z j ∈ O j , a holomorphic function g j : O j → C with g j (z j )  0,

and an integer m j ∈ N such that

g j (z)
f (z) = for each z ∈ O j \ {z j }. (1.1.78)
(z − z j )m j

Then for each j ∈ {1, . . . , N } it follows that z j is a pole of order m j for f and

1  ∂  m j −1
Res( f , z j ) = lim (z − z j )m j f (z)
(m j − 1)! z→z j ∂z
1 (m −1)
= g j (z j ). (1.1.79)
(m j − 1)! j

Without loss of generality we may assume that the sets {O j }1≤ j ≤ N are mutually
disjoint. Let us also introduce O0 := Ω \ {z1, . . . , z N }. Hence,
N
Ω= O j and O j1 ∩ O j2 =  if 1 ≤ j1  j2 ≤ N. (1.1.80)
j=0
 
We wish to extend the function f ∈ 𝒞∞ Ω \ {z1, . . . , z N } to a distribution u in
Ω. To do so, set 
u0 := f  O0 ∈ Lloc
1
(O0, L 2 ) ⊆ D (O0 ), (1.1.81)
and for each j ∈ {1, . . . , N } introduce

(−1)m j −1  ∂  m j −1  1 
u j := gj ∈ D (O j ), (1.1.82)
(m j − 1)! ∂z z − zj

where the expression in the brackets is regarded as a locally integrable function (in
the variable z) and the subsequent iterated derivatives (in z) are taken in the sense
of distributions in O j . Since, as seen from (1.1.82) and (1.1.78), we have
 
u j  O j \{z j } = f  O j \{z j } for each j ∈ {1, . . . , N }, (1.1.83)

it follows that
 
u j1  O j ∩O j2
= u j2  O j ∩O j2
whenever 0 ≤ j1  j2 ≤ N. (1.1.84)
1 1

As such, there exists a unique distribution u ∈ D (Ω) with the property that

u O j = u j in O j for each j ∈ {0, 1, . . . , N }. (1.1.85)

See, e.g., [109, Exercise 2.156, p. 93]. Thus,

 
u ∈ D (Ω) satisfies uΩ\{z1,...,z N } = f Ω\{z1,...,z N } (1.1.86)
1.1 One Variable Complex Analysis 17

as wanted. In particular, (1.1.76) implies that

   κ−n.t.
NκΩ\K uΩ\K ∈ L 1 (∂Ω, σ) and u∂Ω
(1.1.87)
exists (in C) at σ-a.e. point on ∂nta Ω.

To proceed, observe that since f is holomorphic in Ω \ {z1, . . . , z N } we have

∂u0 = 0 in D (O0 ). (1.1.88)

Also, since for each j ∈ {1, . . . , N } we have ∂[1/(z − z j )] = πδz j (see, e.g., [109,
Theorem 7.43, p. 289]), for each j ∈ {1, . . . , N } we may write

π(−1)m j −1  ∂  m j −1
∂u j = gj δz j
(m j − 1)! ∂z
m j −1 ! "  ∂ k
π(−1)m j −1  m j − 1 (m −1−k)
= (−1)m j −1+k g j j (z j ) δz j
(m j − 1)! k=0 k ∂z


m j −1 ! "  ∂ k
m j − 1 π(−1)k (m j −1−k)
= gj (z j ) δz j ∈ ℰ (O j ) (1.1.89)
k=0
k (m j − 1)! ∂z

where the second equality uses [109, Exercise 2.45, p. 34]. Consider now the distri-
bution

N mj −1 ! "  ∂ k
π(−1)k m j − 1 (m j −1−k)
w := gj (z j ) δz j ∈ ℰ (Ω) (1.1.90)
j=1 k=0
(m j − 1)! k ∂z

and note that, as seen from (1.1.88)-(1.1.90), we have


w  O j = ∂u j in D (O j ) for each j ∈ {0, 1, . . . , N }. (1.1.91)

From (1.1.80), (1.1.91), and (1.1.85) we then conclude that

∂u = w in D (Ω). (1.1.92)

Based on (1.1.92), (1.1.90), and (1.1.79) we may now compute

18 1 Integral Representations and Integral Identities
   
ℰ (Ω) ∂u, 1 ℰ(Ω)
= ℰ (Ω) w, 1 ℰ(Ω)

N mj −1 ! "  ∂ k 
π(−1)k m j − 1 (m j −1−k)
= gj (z j ) ℰ (Ω) δz , 1
j=1 k=0
(m j − 1)! k ∂z j
ℰ(Ω)


N
π (m −1)
= g j (z j )
j=1
(m j − 1)! j

=π· Res( f , z). (1.1.93)
z pole of f

Having established these, Theorem 1.1.3 applies to the complex distribution u

(playing the role of f ) and (1.1.66) yields (1.1.77) on account of (1.1.93). 
We wish to remark that, using the version of Theorem 1.1.3 mentioned just after its
 
statement, it is possible to relax the condition NκΩ\K f Ω\K ∈ L 1 (∂Ω, σ) appearing
in (1.1.76) and replace it by the demand that
∫
  κ−n.t.  
Nκε f ∈ Lloc
1
(∂Ω, σ) and  f (ζ) dσ(ζ) < +∞. (1.1.94)
∂Ω
∂∗ Ω

In particular, the first condition above is satisfied if f Ω\K has a continuous extension
to Ω \ K (a hypothesis which also takes care of the second condition in (1.1.76)).
Hence, in such a scenario, if it is meaningful to write it4 then the residue formula
(1.1.77) is true.
This version of the Residue Theorem is actually more efficient than the standard
technology (based on choosing a suitable contour of integration, evaluating various
integrals, and passing to limit) even in such mundane scenarios as the task of showing
that ∫ +∞ ix
e π
dx = . (1.1.95)
−∞ x + 1
2 e
Specifically, choosing Ω := R2+ and considering the meromorphic function

eiz
f (z) := for z ∈ Ω, (1.1.96)
z2 +1

then f has a simple pole at z = i, with residue (2ie) −1

∫ , the function
∫ f may be
extended continuously to a neighborhood of ∂Ω, and ∂Ω | f | dσ = R xdx 2 +1 < +∞.
Thus, (1.1.77) holds and this gives (1.1.95).
Incidentally, it is not much harder to see that the first condition in (1.1.76) is true
in this case. Indeed, if K := B(i, 1/2) then for each κ > 0 fixed the last part in [112,
Lemma 8.3.7] shows (bearing in mind that |eiz | ≤ 1 for each z ∈ Ω and |eiz | = 1 for
each z ∈ ∂Ω) that
4 i.e., if the integral in the right side of (1.1.77) is absolutely convergent
1.1 One Variable Complex Analysis 19
    1
NκΩ\K f Ω\K (x) ≈ , uniformly for x ∈ R ≡ ∂Ω, (1.1.97)
x2 + 1
from which the desired conclusion follows.
Compared with Theorem 1.1.1, one remarkable aspect of the theorem below is
that no explicit lower Ahlfors regularity assumptions are made on the boundary of
∂Ω.
Theorem 1.1.5 Let Ω be a nonempty bounded open subset of R2 with the property
that ∂Ω has finitely many connected components and H 1 (∂Ω) < +∞. Abbreviate
σ := H 1 ∂Ω and denote by ν the geometric measure theoretic outward unit normal
to Ω. In this context, suppose
p
u ∈ L 2 (Ω, L 2 ) ∩ Lloc (Ω, L 2 ) with p > 2, (1.1.98)

is a complex-valued function which, for some sufficiently large aperture parameter

κ = κΩ ∈ (0, ∞) and some small truncation parameter ε > 0, satisfies
∫
κ−n.t.
Nκε u dσ < +∞ and u∂Ω exists σ-a.e. on Aκ (∂Ω). (1.1.99)
∂Ω

Also, with the Cauchy-Riemann operator ∂ taken in the sense of distributions in Ω,

assume that
∂u ∈ Lloc 1
(Ω, L 2 ). (1.1.100)
κ−n.t.
Then the nontangential trace u∂Ω exists σ-a.e. on ∂∗ Ω and, as a function,
belongs to L 1 (∂∗ Ω, σ). Moreover, for each Lebesgue point z ∈ Ω of u with the
property that
∫  
 (∂u)(ζ)  2
  dL (ζ) < +∞ (1.1.101)
Ω ζ −z
one has (with dζ denoting the complex arc-length measure on ∂Ω; cf. (1.1.3))
∫  κ−n.t.  ∫
1 u∂Ω
(ζ) 1 (∂u)(ζ)
u(z) = dζ − dL 2 (ζ). (1.1.102)
2πi ∂∗ Ω ζ−z π Ω ζ−z

Proof Pick a Lebesgue point z ∈ Ω of u with the property that (1.1.101) holds, and
define the vector field
 u(ζ) u(ζ) 
Fz (ζ) := ,i for L 2 -a.e. ζ ∈ Ω. (1.1.103)
ζ−z ζ−z

Choose a compact neighborhood Ko ⊆ Ω of z, select a cutoff function ψ ∈ 𝒞∞ c (Ω)

with ψ ≡ 1 in Ko , and decompose Fz = (1 − ψ)Fz + ψ Fz . Since from (1.1.103) and
(1.1.98) we have
(1 − ψ)Fz ∈ L 2 (Ω, L 2 )
2
(1.1.104)
and (keeping in mind that p > 2)
20 1 Integral Representations and Integral Identities

ψ Fz ∈ L 1 (Ω, L 2 ) and supp (ψ Fz ) ⊆ Ko,

2
(1.1.105)

we conclude that
Fz ∈ L 2 (Ω, L 2 ) + ℰKo (Ω) .
2
(1.1.106)
In addition, much as in (1.1.58), we presently have

(∂u)(ζ)
divFz (ζ) = 2π u(z)δz + 2 in D (Ω), (1.1.107)
ζ−z
where δz denotes the Dirac distribution with mass at z in Ω, hence

divFz ∈ L 1 (Ω, L 2 ) + ℰKo (Ω), (1.1.108)

thanks to (1.1.101). Finally, as is apparent from (1.1.103) and (1.1.99), for each
sufficiently small truncation parameter ε ∈ 0, dist(Ko, ∂Ω) we have
∫
Nκε Fz dσ < +∞, (1.1.109)
∂Ω
κ−n.t.
and the nontangential trace Fz ∂Ω exists σ-a.e. on Aκ (∂Ω).
Granted these properties, [112, Corollary 1.6.5] applies. In particular, this implies
κ−n.t.
that the nontangential trace u∂Ω exists σ-a.e. on ∂∗ Ω and, as a function, belongs to
L 1 (∂∗ Ω, σ). Moreover, [112, (1.6.17)] yields (1.1.102) by reasoning as in (1.1.64).

Here is a version of Cauchy’s reproducing formula in open subsets of the plane for
which no explicit lower Ahlfors regularity assumptions are made on theirs bound-
aries.

Corollary 1.1.6 Let Ω be a nonempty bounded open subset of R2 with the property
that ∂Ω has finitely many connected components and H 1 (∂Ω) < +∞. Suppose

u ∈ L 2 (Ω, L 2 ) (1.1.110)

is a holomorphic function which, for some sufficiently large aperture parameter

κ = κΩ ∈ (0, ∞), satisfies
∫
κ−n.t.
Nκ u dH 1 < +∞ and u∂Ω exists H 1 -a.e. on Aκ (∂Ω). (1.1.111)
∂Ω
κ−n.t.
Then the nontangential trace u∂Ω exists at H 1 -a.e. point on ∂∗ Ω and, as a
function, belongs to L 1 (∂∗ Ω, H 1 ). Moreover,
∫  κ−n.t. 
1 u∂Ω
(ζ)
u(z) = dζ for each z ∈ Ω. (1.1.112)
2πi ∂∗ Ω ζ−z

Proof This is a direct consequence of Theorem 1.1.5. 

1.1 One Variable Complex Analysis 21

We continue with a general version of Morera’s theorem, of the sort described in

the corollary below. Once again, a remarkable feature is the lack of an explicit lower
Ahlfors regularity assumption.

Corollary 1.1.7 Let Ω be a nonempty bounded open subset of R2 with the property
that ∂Ω has finitely many connected components and H 1 (∂Ω) < +∞. Assume

f ∈ L ∞ (Ω, L 2 ) is such that ∂¯ f ∈ L 1 (Ω, L 2 ) and

(1.1.113)
fb (ζ) := lim Ω z→ζ f (z) exists at H 1 -a.e. ζ ∈ ∂Ω.

Then fb belongs to L 1 (∂∗ Ω, H 1 ) and

∫ ∫
fb (ζ) dζ = 2i ∂¯ f dL 2 . (1.1.114)
∂∗ Ω Ω

Proof This follows by applying [112, Corollary 1.6.4] to F := ( f , i f ). 

The subsequent discussion serves a preamble and motivation for the statement
of a sharp version of the Schwarz-Pompeiu formula in the unit disk in the complex
plane, formulated in Theorem 1.1.8.
Given a simply connected UR domain Ω ⊆ R2 ≡ C, abbreviate σ := H 1 ∂Ω.
In addition, fix an aperture parameter κ > 0 together with an integrability exponent
p ∈ (1, ∞). Recall that L p (∂Ω, σ) stands for the space of σ-measurable p-th power
p
integrable complex-valued functions defined on ∂Ω, and denote by LR (∂Ω, σ) the
subspace of L (∂Ω, σ) consisting of real-valued functions. Next, bring in the Hardy
p

space in Ω associated with the Cauchy-Riemann operator, i.e.,

H p (Ω), the collection of all holomorphic functions in Ω satisfying

(1.1.115)
Nκ u ∈ L p (∂Ω, σ), and also u(∞) = 0 when Ω is an exterior domain.
(See §3.2 for a more general point of view.) Then Fatou’s theorem (cf. Theorem 3.1.6)
guarantees that the boundary Hardy space
 κ−n.t. 
H p (∂Ω, σ) := u∂Ω : u ∈ H p (Ω) (1.1.116)

is a well-defined subspace of L p (∂Ω, σ), which is independent of the aperture

parameter κ ∈ (0, ∞).
In this context, consider the question of whether
p
Re H p (∂Ω, σ) = LR (∂Ω, σ), (1.1.117)

i.e., if the collection of all real parts of functions in the boundary Hardy space
p
H p (∂Ω, σ) is precisely the “real” Lebesgue space LR (∂Ω, σ). Of course, the dif-
p
ficult direction is to checking whether for any given f ∈ LR (∂Ω, σ) there exists
κ−n.t.
u ∈ H p (Ω) with Re u∂Ω = f at σ-a.e. point on ∂Ω. (1.1.118)
22 1 Integral Representations and Integral Identities

In this regard, a couple of lines of attack present themselves. One approach relies on
the following two ingredients5:
(i) the solvability of the L p Dirichlet Problem (with nontangential maximal function
control, boundary trace taken in a nontangential pointwise sense, and decay at
infinity in exterior domains) for the Laplacian in the set Ω;
(ii) the L p -comparability of the nontangential maximal function with the Lusin
area-function, in a uniform fashion, for harmonic functions in Ω (normalized by
requiring that they vanish at a fixed point, or at infinity).
More specifically, use item (i) to find a solution w1 of the aforementioned Dirichlet
p
Problem with datum f ∈ LR (∂Ω, σ), Without loss of generality, this solution may be
assumed itself to be real-valued. In view of the fact that Ω is simply connected, there
exists a real-valued harmonic function w2 in Ω with the property that u := w1 + iw2
is holomorphic in Ω (cf. [112, Proposition 5.8.1]). Thanks to the Cauchy-Riemann
equations, |∇w1 | = |∇w2 | hence, in particular, w1 and w2 have the same Lusin area-
function (cf. (2.4.31)). Since the nontangential maximal function of w1 belongs to
L p (∂Ω, σ), we may then invoke item (ii) to conclude that the nontangential maximal
function of w2 also belongs to L p (∂Ω, σ). These properties eventually place u in the
n.t.
Hardy space H p (Ω) (cf. (1.1.115)) and Re u∂Ω = f , as wanted.
To describe another approach to finding a function u as in (1.1.118), denote by
KΔ the principal-value harmonic double layer on ∂Ω (cf. (A.0.62)), and denote by
𝒞 the boundary-to-domain Cauchy integral operator in Ω. The approach in question
requires that the operator
p
1
2I + KΔ is invertible on LR (∂Ω, σ) or, equivalently, on L p (∂Ω, σ), (1.1.119)

a scenario in which (the theory developed in [115, Chapter 1] ensures that) a solution
of (1.1.118) is given by
  −1 
u := 𝒞 12 I + KΔ f in Ω. (1.1.120)

If a uniqueness result, modulo additive (purely imaginary) constants, accompanies

(1.1.118), then we may recast (1.1.120) as the integral representation formula
  −1  κ−n.t.  
u = 𝒞 12 I + KΔ Re u∂Ω + ic in Ω, (1.1.121)

valid for any function u ∈ H p (Ω), where c ∈ R is a constant which depends on u.

This raises the prospect of recovering (up to additive purely imaginary constants) any
function in H p (Ω) from only knowing the nontangential pointwise boundary trace
of its real part (compare with the Cauchy reproducing formula (1.1.10) which uses
the entire boundary trace). One obstacle in the way of making (1.1.121) a reality is
accommodating the demand formulated in (1.1.119). As shown in [116, §4.2, §7.2],
this may be arranged provided Ω is sufficiently flat, globally or infinitesimally. Ideal
5 We shall return to item (i), in a more general, higher-dimensional context, in [116, §8.1]. For item
(ii), see [27]
1.1 One Variable Complex Analysis 23

prototypes of such domains include: the unit disk D in the plane, the complement of
its closure, and the upper/lower half-planes C± . For example, what is special about
  −1
the unit disk D is that not only 12 I + KΔ exists, but it has a very simple form,
namely
⨏
1  −1
2 I + KΔ f = 2 f − f dH 1 for each f ∈ L p (∂D, H 1 ). (1.1.122)
∂D

As such, for Ω := D, formula (1.1.121) becomes (bearing in mind that the Cauchy
operator maps constants to constants)
  ⨏
κ−n.t.    κ−n.t. 

u(z) = 2 𝒞 Re u ∂D (z) − Re u∂D (ζ) dH 1 (ζ) + ic
∂D
∫ ∫
1  κ−n.t.  2 1  κ−n.t.  1
= Re u∂D (ζ) dζ − Re u∂D (ζ) dζ + ic
2πi ∂D ζ−z 2πi ∂D ζ
∫
1  κ−n.t.  ζ+z
= Re u∂D (ζ) dζ + ic for all z ∈ D, (1.1.123)
2πi ∂D (ζ − z)ζ

where we have used dζ = iζ dH 1 (ζ) and ζ −z 2

− ζ1 = (ζζ−z)ζ
+z
. Specializing (1.1.123)
to the case when z = 0 forces
∫
1  κ−n.t.  1
u(0) = Re u∂D (ζ) dζ + ic
2πi ∂D ζ
∫
1  κ−n.t. 
= Re u∂D (ζ) dH 1 (ζ) + ic
2π ∂D
⨏
 κ−n.t. 
= Re u∂D (ζ) dH 1 (ζ) + ic = (Re u)(0) + ic, (1.1.124)
∂D

where the last step uses (a slight variant of) the classical Mean Value Formula for
the harmonic function Re u in D. In turn, (1.1.124) forces c = (Im u)(0) which, when
used back in (1.1.123), leads to the conclusion that for each function u ∈ H p (D) we
have
∫  κ−n.t. 
1 Re u∂D (ζ) ζ + z
u(z) = dζ + i(Im u)(0) for all z ∈ D. (1.1.125)
2πi ∂D ζ ζ−z
In particular, this implies the classical Schwarz formula stating that for each function
u ∈ 𝒞0 (D) which is holomorphic in D we have
∫  
1 Re u (ζ) ζ + z
u(z) = dζ + i(Im u)(0) at each z ∈ D. (1.1.126)
2πi ∂D ζ ζ−z
The integral representation formula (1.1.125) is naturally associated with solutions
of the boundary value problem (formulated using the Cauchy-Riemann operator ∂
24 1 Integral Representations and Integral Identities

in the complex plane)

⎧
⎪ u ∈ 𝒞∞ (Ω),
⎪
⎪
⎪
⎪
⎨ ∂u = 0 in Ω,
⎪
(1.1.127)
⎪
⎪ Nκ u ∈ L p (∂Ω, σ),
⎪
⎪
⎪ Re uκ−n.t. = f at σ-a.e. point on ∂Ω,
⎪
⎩ ∂Ω

in the case when Ω := D, the unit disk in the complex plane. With stronger demands,
such as u ∈ 𝒞0 (Ω), replacing the nontangential maximal function condition in the
third line above, this boundary value problem6 this problem has been treated by
Hermann Amandus Schwarz in [157]. It is therefore appropriate that we refer to
(1.1.126) as the L p Schwarz Problem7.
Here our goal is to extend the class of functions for which integral representation
formulas in the spirit of (1.1.125) are valid, by going beyond Hardy spaces. In partic-
ular, we are interested in allowing functions u which are not necessarily holomorphic,
a scenario in which (1.1.125) should be augmented through the consideration of a
solid integral involving ∂z̄ u, much the manner in which Cauchy’s integral representa-
tion formula (1.1.10) has been generalized by the Cauchy-Pompeiu formula (1.1.8).
In view of this, it is natural to call resulting extension the Schwarz-Pompeiu formula.
In the case of the unit disk, this is formally stated in the theorem below.
 
Theorem 1.1.8 Denote by D := z ∈ C : |z| < 1 the open unit disk in the complex
plane, and abbreviate σ := H 1 ∂D. Also, recall that ∂ = ∂z̄ = 12 (∂x +i∂y ) stands for
the Cauchy-Riemann operator8. Having fixed an aperture parameter κ > 0, consider
a Lebesgue-measurable function u : D → C satisfying
 L 2 (ζ) 
Nκ u ∈ L 1 (∂D, σ), ∂u ∈ L 1 D, ,
|ζ | (1.1.128)
κ−n.t.
the nontangential trace (Re u)∂D exists at σ-a.e. point on ∂D,

and 0 ∈ D is a Lebesgue point for u, i.e.,

⨏
u(0) = lim+ u dL 2 . (1.1.129)
r→0 B(0,r)

Then for any other aperture parameter κ > 0 the nontangential boundary trace
κ −n.t.
(Re u)∂D exists at σ-a.e. point on ∂D, is actually independent of the parameter κ ,

6 itself a precursor of the more general Riemann-Hilbert problem, in which a linear combination
of the real and imaginary parts of a holomorphic function is prescribed on the boundary
7 As noted earlier, if (1.1.119) holds (something considered in [116, Chapters 4 and 7]) then a
solution of the L p Schwarz Problem (1.1.127) is given by (1.1.120). Uniqueness, modulo some
additive purely imaginary constant, follows from uniqueness in the L p Dirichlet Problem for the
Laplacian (a topic considered later, in Theorem 5.7.6; see also §4.3 and [116, Chapter 8]).
8 for this, we shall freely use a variety of alternative pieces of notation; e.g., we agree that
∂ = ∂z̄ = ∂ z stand for the Cauchy-Riemann operator in the variable z
1.1 One Variable Complex Analysis 25

and for L 2 -a.e. point z ∈ D one has (with the dependence on the aperture parameter
dropped)
∫  n.t. 
1 (Re u)∂D (ζ) ζ + z
u(z) = dζ + iIm u(0)
2πi ∂D ζ ζ−z
∫  &
1 (∂u)(ζ) ζ + z (∂u)(ζ) 1 + zζ
− + dL 2 (ζ). (1.1.130)
2π D ζ ζ−z ζ 1 − zζ

Before presenting the proof of this theorem, few comments are in order. First, the
fact that u is Lebesgue-measurable permits us to meaningfully define the nontangen-
tial maximal function Nκ u. Second, from [112, Lemma 8.3.1] and the first property
in (1.1.128) we see that
∞
u ∈ Lloc (D, L 2 ). (1.1.131)
As a consequence, u ∈ Lloc 1 (D, L 2 ) hence it is meaningful to speak of Lebesgue

points of u and to consider the distribution ∂u. The latter is further assumed
 to 2be of

function-type9, and to actually belong to the weighted Lebesgue space L 1 D, L|ζ(ζ| ) .
Third, as is apparent from the proof given below, formula (1.1.130) is actually
valid at each z ∈ D which is a Lebesgue point for u and has the property that
∫ |(∂u)(ζ ) |
D |ζ −z |
dL 2 (ζ) < +∞.
As a corollary of (1.1.130) and Fatou’s theorem, corresponding to the particular
case in which the function in question is annihilated by the Cauchy-Riemann operator,
one has
∫  n.t. 
1 (Re u)∂D (ζ) ζ + z
u(z) = dζ + iIm u(0) for each z ∈ D,
2πi ∂D ζ ζ−z (1.1.132)
whenever u is a holomorphic function in D with Nκ u ∈ L 1 (∂D, σ).

We also want to remark that, in contrast with the Cauchy-Pompeiu formula (1.1.8),
the integral representation formula (1.1.130) reconstructs a function u from knowing
∂u and only the boundary trace of Re u plus the value of Im u at one point10. It is
also possible to utilize the nontangential boundary trace of Im u, instead of that of
Re u. Specifically, if in place of the second line of (1.1.128) we now assume that
κ−n.t.
(Im u)∂D exists at σ-a.e. point on ∂D, then (1.1.130) written for iu in place of u
yields the integral representation formula

9 i.e., the distribution naturally induced by a locally integrable function, via integration
10 This being said, the present considerations are restricted to the unit disk, whereas the Cauchy-
Pompeiu formula (1.1.8) has been established in large classes of domains.
26 1 Integral Representations and Integral Identities

∫  n.t. 
1 (Im u)∂D (ζ) ζ + z
u(z) = dζ + Re u(0)
2π ∂D ζ ζ−z
∫  &
1 (∂u)(ζ) ζ + z (∂u)(ζ) 1 + zζ
− − dL 2 (ζ) (1.1.133)
2π D ζ ζ−z ζ 1 − zζ

valid at L 2 -a.e. point z ∈ D.

Lastly, we wish to remark that our version of the Schwarz-Pompeiu formula in the
above theorem is a generalization of the classical Schwarz formula (1.1.126) which
is fairly optimal. Indeed, the hypotheses in Theorem 1.1.8 are not too far off from the
kind of assumptions one would have to impose simply to make sense of the integral
representation formula (1.1.130). Thus, in this sense, this result is almost an accurate
embodiment of the slogan if you can write it, then it’s true.
To illustrate this, we note that11
the Schwarz-Pompeiu formula recorded in (1.1.130) may fail even in
the class of holomorphic functions in the unit disk if in the first mem- (1.1.134)
bership in (1.1.128) the space L 1 (∂D, σ) is replaced by L 1,∞ (∂D, σ).

In a nutshell, the margin between L 1 and weak-L 1 delineates the natural range of
validity for the Schwarz-Pompeiu formula (1.1.130). An example of the sort alluded
to in (1.1.134) is offered by the function
1−z
u : D −→ C, u(z) := for each z ∈ D, (1.1.135)
1+z
which satisfies
 
u ∈ 𝒞∞ D \ {−1} , ∂u = 0 in D, Im u(0) = 0, (1.1.136)

and
Im ζ
u(ζ) = −2i for each ζ ∈ ∂D \ {−1},
|1 + ζ | 2 (1.1.137)
n.t.
hence (Re u)∂D = 0 at L 1 -a.e. point on ∂D.
Also, [112, Lemma 8.3.7] tells us that, for any fixed aperture parameter κ > 0,
  1
Nκ u (ζ) ≈ , uniformly for ζ ∈ ∂D \ {−1}, (1.1.138)
|1 − ζ |
from which we conclude (see the discussion in [112, Example 6.2.2])

Nκ u ∈ L 1,∞ (∂D, σ) but Nκ u  L 1 (∂D, σ). (1.1.139)

11 in the scenario described in (1.1.134) the local version of Fatou theorem for holomorphic (even
harmonic) functions in the unit disk guarantees the existence of the nontangential boundary trace
almost everywhere
1.1 One Variable Complex Analysis 27

Clearly, the Schwarz-Pompeiu formula (1.1.130) does not hold for this function u (as
the right side is identically zero), and the only hypothesis in Theorem 1.1.8 which is
presently violated is the failure of Nκ u to be in L 1 (∂D, σ).
We next take up the task of providing the proof of Theorem 1.1.8.
Proof of Theorem 1.1.8 First, from assumptions and [112, Proposition 8.9.8] we see
κ −n.t.
that for any other aperture parameter κ > 0 the nontangential trace (Re u)∂D exists
at σ-a.e. point on ∂D, and is actually independent of the parameter κ . As regards the
integral representation formula (1.1.130), the idea is to put ourselves in a position
where we can invoke [112, Corollary 1.4.3]. This requires some preparations. First,
from assumptions and [112, Lemma 3.5.7] it follows that for L 2 -a.e. point z ∈ C we
have ∫
|(∂u)(ζ)|
dL 2 (ζ) < +∞. (1.1.140)
D |ζ − z|
Fix some z ∈ D which is a Lebesgue point for the function u and such that (1.1.140)
holds (we know that L 2 -a.e. point in D enjoys these properties). For each point
ζ ∈ D \ {0, z} introduce
 &
1 u(ζ) ζ + z u(ζ) 1 + zζ
φ(ζ) := + (1.1.141)
2π ζ ζ−z ζ 1 − zζ

and
 &
1 u(ζ) ζ + z u(ζ) 1 + zζ
ψ(ζ) := − , (1.1.142)
2π ζ ζ−z ζ 1 − zζ

then use these scalar-valued functions to define the vector field (with complex-valued
components)  

F(ζ) := φ(ζ), iψ(ζ) for each ζ ∈ D \ {0, z}. (1.1.143)
Based on (1.1.131), (1.1.141)-(1.1.142), and (1.1.143) we see that

F ∈ Lloc
2
1
(D, L 2 ) . (1.1.144)
 
Next, bring in K := B(0, R) for some fixed R ∈ |z|, 1 , and note that K is a
compact subset of the unit disk such that both 0 and z are contained in its interior.
Then (1.1.141)-(1.1.143) allow us to estimate
 D\K   
Nκ F (ζ) ≤ CK Nκ u (ζ) for all ζ ∈ ∂D. (1.1.145)

In turn, from [112, (8.2.26)], (1.1.145), and the first condition in (1.1.128) we
conclude that

NκD\K F ∈ L 1 (∂D, σ). (1.1.146)

28 1 Integral Representations and Integral Identities

Denote by ∂ the conjugate of the Cauchy-Riemann operator, and agree to let ∂ζ

indicate that this acts in the variable ζ. The vector field F has been designed so that

divF = ∂x φ + i∂y ψ

1  u(ζ) ζ + z  1  u(ζ) 1 + zζ 
= ∂ζ + ∂ζ in D (D). (1.1.147)
π ζ ζ−z π ζ 1 − zζ
To be able to continue, write
ζ+z 2 1
= − (1.1.148)
ζ(ζ − z) ζ − z ζ
and use this together with Lemma 1.1.2 twice (mindful of the fact that both 0 and z
are Lebesgue points for u) to compute
 u(ζ) ζ + z    2 1   2u(ζ)   u(ζ) 
∂ζ = ∂ ζ u(ζ) − = ∂ζ − ∂ζ
ζ ζ−z ζ−z ζ ζ−z ζ

(∂u)(ζ) ζ + z
= + 2πu(z)δz − πu(0)δ0 in D (D). (1.1.149)
ζ ζ−z
Also, keeping in mind that the function

1 + zζ
D ζ −→ ∈C (1.1.150)
1 − zζ
is holomorphic, we may once again invoke Lemma 1.1.2 (once more using (1.1.129))
to write
 u(ζ) 1 + zζ   u(ζ) 1 + zζ  (∂u)(ζ) 1 + zζ
∂ζ = ∂ζ = + πu(0)δ0
ζ 1 − zζ ζ 1 − zζ ζ 1 − zζ

(∂u)(ζ) 1 + zζ
= + πu(0)δ0 in D (D). (1.1.151)
ζ 1 − zζ
 
Upon noting that −πu(0)δ0 + πu(0)δ0 = −2πi Im u(0) δ0 , from (1.1.147)-(1.1.151)
we therefore obtain

1 (∂u)(ζ) ζ + z 1 (∂u)(ζ) 1 + zζ  

(divF)(ζ) = + + 2u(z)δz − 2i Im u(0) δ0
π ζ ζ−z π ζ 1 − zζ
(1.1.152)

in the sense of distributions in D, in the variable ζ. As such, from (1.1.152), (1.1.140),

and the last assumption in the first line of (1.1.128) we see that
1.1 One Variable Complex Analysis 29

divF belongs to ℰ (D) + L 1 (D, L 2 ). (1.1.153)

Towards the goal of eventually employing [112, Corollary 1.4.3], the next step is
to check the hypothesis [112, (1.4.36)]. Guided by [112, (1.4.35)] and (1.1.143), for
each ζ = x + iy ∈ D \ K we compute
 &  &
1 u(ζ) ζ + z 1 u(ζ) 1 + zζ
f (ζ) := xφ(ζ) + iyψ(ζ) = ζ+ ζ
2π ζ ζ−z 2π ζ 1 − zζ

1 ζ+z 1 1 + zζ
= u(ζ) + u(ζ)
2π ζ − z 2π 1 − zζ
 &
1  ζ +z 1 1 + zζ ζ + z
= u(ζ) + u(ζ) + u(ζ) −
2π ζ − z 2π 1 − zζ ζ − z

1 ζ+z 1 z(1 − |ζ | 2 )
= (Re u)(ζ) − u(ζ) , (1.1.154)
π ζ−z π (1 − zζ)(ζ − z)

where the first equality defines f (ζ), and where we have used (1.1.141)-(1.1.142) in
the second equality. From the second line of (1.1.128) we know that the nontangential
κ−n.t.
boundary trace (Re u)∂D exists at σ-a.e. point on ∂D. Also, the first property in
the first line of (1.1.128) implies that for σ-a.e. point ζ∗ ∈ ∂D the function u is
essentially bounded in the nontangential approach region Γκ (ζ∗ ). Mindful of the fact
that 1 − |ζ | 2 vanishes for ζ ∈ ∂D, we conclude from this that
the nontangential pointwise boundary trace of the function
(D \ K) ζ −→ u(ζ) z(1− |ζ | ) ∈ C is zero σ-a.e. on ∂D. (1.1.155)
2

(1−zζ )(ζ −z)

All things considered, the conclusion is that for f : D\K → C defined as in (1.1.154)
we have
 κ−n.t.  1 κ−n.t.  ζ + z
f ∂D (ζ) = (Re u)∂D (ζ) at σ-a.e. ζ ∈ ∂D. (1.1.156)
π ζ−z
Consequently,
∫  ∫
κ−n.t.  1  κ−n.t.  1
f ∂D (ζ) dσ(ζ) = f ∂D (ζ) dζ
∂D i ∂D ζ
∫  κ−n.t. 
1 Re u∂D (ζ) ζ + z
= dζ, (1.1.157)
πi ∂D ζ ζ−z

since dζ = iν(ζ) dσ(ζ) = iζ dσ(ζ).

At this stage, we have checked that F satisfies all hypotheses in [112, Corol-
lary 1.4.3]. Together with (1.1.152) and (1.1.157), the Divergence Formula [112,
30 1 Integral Representations and Integral Identities

(1.4.38)] then allows us to write

∫  &
1 (∂u)(ζ) ζ + z (∂u)(ζ) 1 + zζ
2u(z) − 2iIm u(0) + + dL 2 (ζ)
π D ζ ζ −z ζ 1 − zζ
∫  κ−n.t. 
 
 1 𝒞∞ (D) =
= (𝒞∞b (D))∗ divF, b
f ∂D (ζ) dσ(ζ)
∂D
∫  κ−n.t. 
1 (Re u)∂D (ζ) ζ + z
= dζ . (1.1.158)
πi ∂D ζ ζ−z
From this (1.1.130) readily follows, at each Lebesgue point z ∈ D of u such that
(1.1.140) holds, hence also at L 2 -a.e. point z ∈ D. 

There is also a version of the Schwarz-Pompeiu formula in the complement of

the closed unit disk in the plane, of the sort described in our next theorem.

Theorem 1.1.9 Denote by D the open unit disk in the complex plane, and abbreviate
σ := H 1 ∂D. Also, let ∂ stand for the Cauchy-Riemann operator in the complex
plane. Consider a Lebesgue-measurable function u : C \ D → C satisfying, for some
aperture parameter κ > 0,
 L 2 (ζ) 
Nκ u ∈ L 1 (∂D, σ), ∂u ∈ L 1 C \ D, ,
|ζ | (1.1.159)
κ−n.t.
the nontangential trace (Re u)∂D exists at σ-a.e. point on ∂D,

and such that, for some λ ∈ (1, ∞),

⨏
|u| dL 2 = o(1) as R → ∞. (1.1.160)
B(0,λR)\B(0,R)

Then for any other aperture parameter κ > 0 the nontangential boundary trace
κ −n.t.
(Re u)
∂D
exists at σ-a.e. point on ∂D, is actually independent of κ , and for
L 2 -a.e. point z ∈ C \ D one has (with the dependence on the aperture parameter
dropped)
∫ n.t. 
1 (Re u)∂D (ζ) ζ + z
u(z) = − dζ
2πi ∂D ζ ζ−z
∫  &
1 (∂u)(ζ) ζ + z (∂u)(ζ) 1 + zζ
− + dL 2 (ζ). (1.1.161)
2π C\D ζ ζ−z ζ 1 − zζ

As with (1.1.134), this result is sharp in the sense that the Schwarz-Pompeiu for-
mula (1.1.161) may fail even in the class of holomorphic functions in the complement
of the closed unit disk in the plane if in the first membership in (1.1.159) the space
1.1 One Variable Complex Analysis 31

L 1 (∂D, σ) is replaced by its weak version, L 1,∞ (∂D, σ). To see this, consider the
function
1
u : C \ D → C defined by u(z) := for each z ∈ C \ D. (1.1.162)
1+z
Then
 
u ∈ 𝒞∞ C \ D , ∂u = 0 in C \ D,
 n.t. 
u∂(C\D) (z) exists for all z ∈ ∂D \ {−1}, (1.1.163)

u(z) = O(|z| −1 ) as |z| → ∞, and Nκ u ∈ L 1,∞ (∂D, σ),

with the last property a consequence of [112, Lemma 8.3.7] (and the discussion in
[112, Example 6.2.2]). In addition, for each z = x + iy ∈ ∂D \ {−1} we have
 n.t.   1+z  1+x 1+x 1
(Re u)∂D (z) = Re = = = . (1.1.164)
|1 + z| 2 (1 + x)2 + y 2 2 + 2x 2

Hence, if z ∈ C \ D, then
∫  n.t.  ∫ ∫
1 (Re u)∂D (z)(ζ + z) 1 2 1 1
dζ = dζ − dζ
2πi ∂D ζ(ζ − z) 4πi ∂D ζ−z 4πi ∂D ζ
1 1
=0− =− , (1.1.165)
2 2
where the penultimate equality uses the Residue Theorem. As such, formula (1.1.161)
fails for this function u (as the right side is a constant), and the only hypothesis in
Theorem 1.1.9 which is presently violated is the membership of Nκ u to L 1 (∂D, σ).
We now give the proof of Theorem 1.1.9.
Proof of Theorem 1.1.9 The claims pertaining to the nontangential boundary trace
of Re u follow from assumptions and [112, Proposition 8.9.8]. To justify (1.1.161),
we reason as in the proof of Theorem 1.1.9, now making use of [112, Corollary 1.4.4]
in place of [112, Corollary 1.4.3], and otherwise natural adjustments. Specifically,
we now fix a Lebesgue point z ∈ C\D for the function u, with the additional property
that ∫
|(∂u)(ζ)|
dL 2 (ζ) < +∞, (1.1.166)
C\D |ζ − z|

and once again consider a vector field F defined as in (1.1.141)-(1.1.143) but for
ζ ∈ C \ D. The fact that we now work in the complement of the closed unit disk
affects the nature of divF. Specifically, since C \ D ζ → 1 ∈ C is a holomorphic
ζ
function, in place of (1.1.149) we now have
32 1 Integral Representations and Integral Identities
 u(ζ) ζ + z    2 1   2u(ζ)   u(ζ) 
∂ζ = ∂ ζ u(ζ) − = ∂ζ − ∂ζ
ζ ζ−z ζ−z ζ ζ−z ζ

(∂u)(ζ) ζ + z  
= + 2πu(z)δz in D C \ D . (1.1.167)
ζ ζ−z
Moreover, in view of the fact that the functions
1 1 + zζ
C\D ζ −→ ∈ C and C \ D ζ −→ ∈C (1.1.168)
ζ 1 − zζ
are holomorphic, in place of (1.1.151) we now obtain

 u(ζ) 1 + zζ   u(ζ) 1 + zζ  (∂u)(ζ) 1 + zζ

∂ζ = ∂ζ =
ζ 1 − zζ ζ 1 − zζ ζ 1 − zζ

(∂u)(ζ) 1 + zζ  
= in D C \ D . (1.1.169)
ζ 1 − zζ

Ultimately, these are responsible for the absence of the term iIm u(0) in (1.1.161),
compared to (1.1.130). Finally, if f (ζ) is defined as in (1.1.154) but now for ζ ∈ C\D
with ζ  z, we deduce from (1.1.160) and the final formula in (1.1.154) that
∫
| f | dL 2 = o(R2 ) as R → ∞. (1.1.170)
B(0,λR)\B(0,R)

Granted these properties, we may now invoke [112, Corollary 1.4.4] and conclude
from [112, (1.4.54)] (which is responsible for the change in sign for the boundary
integral in (1.1.161), compared to (1.1.130)) that (1.1.161) holds. 
We conclude by presenting a companion result to Theorem
 1.1.8, dealing
 with a
sharp form of the Schwarz-Pompeiu formula for C+ := z ∈ C : Im z > 0 . The goal
is now to recover a given complex-valued function defined in C+ from the knowledge
of the nontangential boundary trace of its real part on R ≡ ∂C+ , and the action of
the Cauchy-Riemann operator ∂ on said function in C+ . This is made precise in the
theorem below, where an optimal class of functions for which such a goal may be be
accomplish is identified.
Theorem 1.1.10 Consider a Lebesgue-measurable function u : C+ → C which, for
some aperture parameter κ ∈ (0, ∞), satisfies
 L 1 (t) 
Nκ u ∈ L 1 R, , ∂u ∈ L 1 (C+, L 2 ), and
1 + |t| (1.1.171)
κ−n.t.
the nontangential trace (Re u)∂C+ exists L 1 -a.e. on R ≡ ∂C+ .

Then for any other aperture parameter κ > 0 the nontangential boundary trace
κ −n.t.
(Re u)∂C+
exists at L 1 -a.e. point on R ≡ ∂C+ , is actually independent of κ , and
1.1 One Variable Complex Analysis 33

for L 2 -a.e. point z ∈ C+ one has (with the dependence on the aperture parameter
dropped)

∫ (Re u)n.t.  (t) ∫  &

1 ∂C+ 1 (∂u)(ζ) (∂u)(ζ)
u(z) = dt − − dL 2 (ζ).
2πi R t−z π C+ ζ−z ζ−z
(1.1.172)

Before presenting its proof, we wish to make a number of comments designed to

shed further light on the nature and scope of Theorem 1.1.10. First, the hypothesis
that u is Lebesgue-measurable allows us to meaningfully define the nontangential
maximal function Nκ u. In turn, from [112, Lemma 8.3.1] and the first property listed
in (1.1.171) we deduce that
∞
u ∈ Lloc (C+, L 2 ). (1.1.173)
In particular, u belongs to Lloc1 (C , L 2 ), which makes it possible consider the distri-
+
bution ∂u in C+ . This distribution is further assumed to be of function-type, and to
actually belong to the Lebesgue space L 1 (C+, L 2 ). Also, as may be seen from the
proof given below, formula (1.1.172) is in fact true for each Lebesgue point z ∈ C+
∫
of u satisfying C |(∂u)(ζ )|
|ζ −z | dL (ζ) < +∞.
2
+
A special case of (1.1.172), corresponding to the scenario in which the function
in question is annihilated by the Cauchy-Riemann operator, reads12

∫ (Re u)n.t.  (t)

1 ∂C+
u(z) = dt at each point z ∈ C+,
2πi R t−ζ (1.1.174)
 L 1 (t) 
provided u is holomorphic in C+ and Nκ u ∈ L R, 1
1+ |t | .

In contrast with what the Cauchy-Pompeiu formula (1.1.8) would give when
Ω := C+ , the integral representation formula (1.1.172) reconstructs a function u from
knowing ∂u and only the boundary trace of Re u. There is also a version of (1.1.172)
emphasizing the use of the imaginary part in place of the real part. Concretely, if
κ−n.t.
in place of the second line of (1.1.171) it is now assumed that (Im u)∂C+ exists at
L 1 -a.e. point on R ≡ ∂C+ , then (1.1.172) written for iu in place of u yields the
integral representation formula

12 here the classical Fatou theorem for holomorphic functions in the upper half-plane is used to
ensure that the nontangential boundary trace exists
34 1 Integral Representations and Integral Identities

∫ (Im u)n.t.  (t)

1 ∂C+
u(z) = dt
2π R t−ζ
∫  &
1 (∂u)(ζ) (∂u)(ζ)
− + dL 2 (ζ) at L 2 -a.e. point z ∈ C+ . (1.1.175)
π C+ ζ−z ζ−z

It is of significance to note that our version of the Schwarz-Pompeiu formula in

the above theorem is in the nature of best possible. Indeed,13
the Schwarz-Pompeiu formula recorded in (1.1.172) may fail even in
the class of holomorphic functions in the upper
 half-plane
 if in the
L 1 (t)
first membership in (1.1.171) the space L R, 1+ |t | is replaced by
1 (1.1.176)
 
its weak version, L 1,∞ R, L
1 (t)
1+ |t | .

To see that this is the case, consider the function

i
u : C+ −→ C, u(z) := for each z ∈ C+ . (1.1.177)
z
This satisfies  
u ∈ 𝒞∞ C+ \ {0} , ∂u = 0 in C+, (1.1.178)
and
u(t) = i/t for each t ∈ R \ {0}, hence
n.t. (1.1.179)
(Re u)∂C+ = 0 at L 1 -a.e. point on R ≡ ∂C+ .
In addition, from [112, Lemma 8.3.7] we see that, for any fixed aperture parameter
κ > 0,
  1
Nκ u (t) ≈ , uniformly for t ∈ R \ {0}, (1.1.180)
|t|
from which we deduce, based on straightforward calculus, that
 L 1 (t)   L 1 (t) 
Nκ u ∈ L 1,∞ R, but Nκ u  L 1 R, . (1.1.181)
1 + |t| 1 + |t|
The bottom line is that the Schwarz-Pompeiu formula (1.1.172) does not hold for
this function u (as the right side is identically zero), and the only hypothesis
 in

Theorem 1.1.10 which is presently violated is the failure of Nκ u to be in L 1 R, L
1 (t)
1+ |t | .

The proof of Theorem 1.1.10 is given next.

13 in the scenario described in (1.1.176) the local version of Fatou theorem for holomorphic (even
harmonic) functions in the upper half-plane guarantees the existence of the nontangential boundary
trace almost everywhere
1.1 One Variable Complex Analysis 35

Proof of Theorem 1.1.10 The claims regarding the nontangential boundary trace
of Re u are seen from assumptions and [112, Corollary 8.9.9]. The bulk of the
argument concerns the integral representation formula (1.1.172). The strategy is to
arrange matters so that we can eventually employ the two-dimensional version of
[112, Corollary 1.4.2]. To set the stage, from assumptions and [112, Lemma 3.5.7]
(with m := 0) we see that for L 2 -a.e. point z ∈ C+ we have
∫
|(∂u)(ζ)|
dL 2 (ζ) < +∞. (1.1.182)
C+ |ζ − z|

Henceforth, we fix some z ∈ C+ which is a Lebesgue point for the function u and

which satisfies (1.1.182) (these properties are enjoyed by L 2 -a.e. point in C+ ). Define
the vector field
F = (F1, F2 ) : C+ \ {z} −→ C2 (1.1.183)
by setting
 
 u(ζ) u(ζ) u(ζ) u(ζ)
F(ζ) := − ,i +i at each ζ ∈ C+ \ {z}. (1.1.184)
ζ−z ζ−z ζ−z ζ−z

In particular, (1.1.173) ensures that

1   2
F ∈ Lloc C+, L 2 . (1.1.185)
 
Bring in K := B z, 2−1 Im z , which a compact neighborhood of z, contained in
C+ . Then (1.1.184) together with [112, Lemma 8.3.7] imply
 
 C+ \K  Nκ u (t)
Nκ 
F (t) ≤ Cz for all t ∈ R ≡ ∂C+ . (1.1.186)
1 + |t|
In turn, from [112, (8.2.26)], (1.1.186), and the first condition in (1.1.171) we
conclude that

NκC+ \K F ∈ L 1 (R, L 1 ). (1.1.187)

To continue, note that the last scalar component of F is given by

 
u(ζ) u(ζ) u(ζ) + u(ζ) i i
F2 (ζ) = i +i =i + u(ζ) −
ζ−z ζ−z ζ−z ζ−z ζ−z

(Re u)(ζ) 2 Im ζ
= 2i − u(ζ) at each ζ ∈ C+ \ {z}. (1.1.188)
ζ−z (ζ − z)(ζ − z)
 κ−n.t. 
The second line of (1.1.171) gives that the nontangential trace (Re u)∂C+ (t) exists
at L 1 -a.e. point t ∈ R ≡ ∂C+ . Also, the first property in the first line of (1.1.171)
36 1 Integral Representations and Integral Identities

guarantees that the function u is essentially bounded in the nontangential approach

region Γκ (t) for L 1 -a.e. point t ∈ R ≡ ∂C+ . In view of the fact that Im ζ vanishes
for ζ ∈ ∂C+ ≡ R, this ultimately leads to the conclusion that
the nontangential pointwise boundary trace of the function
(C+ \ {z}) ζ → u(ζ) 2 Im ζ ∈ C is zero L 1 -a.e. on R. (1.1.189)
(ζ −z)(ζ −z)

As a consequence, we obtain
 κ−n.t. 
 κ−n.t.  (Re u)∂C+ (t)
F2 ∂C+ (t) = 2i at L 1 -a.e. point t ∈ R ≡ ∂C+ . (1.1.190)
t−z
Going further, if δz denotes the Dirac distribution with mass at z in C+ , then
Lemma 1.1.2 gives
 u(ζ)  (∂u)(ζ)
∂ζ = π u(z)δz + in D (C+ ). (1.1.191)
ζ−z ζ−z
Denote by ∂ the conjugate of the Cauchy-Riemann operator, and agree to let ∂ζ
indicate that this acts in the variable ζ. Then, keeping in mind that the function
1
C+ ζ −→ ∈C (1.1.192)
ζ−z
 
1 C , L 2 , we see that
is holomorphic, and that ∂u ∈ Lloc +

 u(ζ)   u(ζ)  (∂u)(ζ) (∂u)(ζ)

∂ζ = ∂ζ = = in D (C+ ). (1.1.193)
ζ−z ζ−z ζ−z ζ−z
Relying on (1.1.184), (1.1.191), and (1.1.193) we may now compute (with the
divergence considered in the sense of distributions in the variable ζ ∈ C+ )

   u(ζ)   u(ζ) 
divF (ζ) = 2∂ ζ − 2∂ζ
ζ−z ζ−z

2(∂u)(ζ) 2(∂u)(ζ)
= 2π u(z)δz + − in D (C+ ). (1.1.194)
ζ−z ζ−z
As such, from (1.1.194), (1.1.182), and the last assumption in the first line of (1.1.171)
we see that
divF belongs to ℰ (C+ ) + L 1 (C+, L 2 ). (1.1.195)
At this stage, all hypotheses of [112, Corollary 1.4.2] (with n := 2) have been
verified, including [112, (1.4.26)]. We may therefore invoke the version of the Di-
vergence Formula recorded in [112, (1.4.24)] which presently gives
1.2 Integral Representation Formulas in Clifford Analysis 37

∫  &
(∂u)(ζ) (∂u)(ζ)
2π u(z)+2 − dL 2 (ζ)
C+ ζ−z ζ−z
∫ 
  κ−n.t. 

= (𝒞∞ (C+ ))∗ divF, 1 𝒞∞b (C+ ) = − F2 ∂C+ (t) dL 1 (t)
b R

∫ (Re u)κ−n.t.  (t)

∂C+
= −2i dt, (1.1.196)
R t−z
thanks also to (1.1.194) and (1.1.190). Having established this, we conclude that
(1.1.172) holds at each Lebesgue point z ∈ C+ of u satisfying (1.1.182) holds, hence
also at L 2 -a.e. point z ∈ C+ . 

1.2 Integral Representation Formulas in Clifford Analysis

There is a higher-dimensional version of Theorem 1.1.1 for subsets Ω of Rn with

n ∈ N, n ≥ 2 arbitrary, involving the Clifford algebra Cn (cf. the discussion in [112,
§6.4]) in place of the field of complex numbers C, and the Dirac operator D from
(A.0.37) in lieu of the Cauchy-Riemann operator ∂ from (1.1.2), presented a little
later in Theorem 1.2.2. To facilitate the proof of the latter theorem, we first take care
of the technical result described in the lemma below.

Lemma 1.2.1 Assume Ω is an open set in Rn and suppose u ∈ Lloc 1 (Ω, L n ) ⊗ C is

n
such that Du ∈ Lloc (Ω, L ) ⊗ Cn (with the Dirac operator D taken in the sense of
1 n

distributions in Ω). For each fixed point x ∈ Rn consider the Clifford algebra-valued
functions
1 x−y
Φx (y) :=  (Du)(y) for L n -a.e. y ∈ Ω, (1.2.1)
ωn−1 |x − y| n
and, for j ∈ {1, . . . , n},

(j) 1 x−y
Ψx (y) :=  e j  u(y) for L n -a.e. y ∈ Ω. (1.2.2)
ωn−1 |x − y| n
Then for each Lebesgue point x ∈ Ω of the function u with the additional property
that
1 (Ω, L n ) ⊗ C and
Φx belongs to the space Lloc n
(j)
(1.2.3)
Ψx ∈ Lloc (Ω, L ) ⊗ Cn for each j ∈ {1, . . . , n},
1 n

one has

n
(j)
∂j Ψx = u(x)δx + Φx in D (Ω) ⊗ Cn, (1.2.4)
j=1

where δx denotes the Dirac distribution with mass at x in Ω.

38 1 Integral Representations and Integral Identities

Furthermore, for every x ∈ Rn \ Ω the conditions in (1.2.3) are satisfied and, in

place of (1.2.4), one now has

n
(j)
∂j Ψx = Φx in D (Ω) ⊗ Cn . (1.2.5)
j=1

Of course, thanks to Lebesgue’s Differentiation Theorem, we see that

formula (1.2.4) is valid for L n -a.e. point x ∈ Ω
(1.2.6)
such that the memberships listed in (1.2.3) hold.
We also wish to comment on the nature of the very last assumption made in (1.2.3).
Our first observation in this regard is that since for every compact set K ⊂ Ω and
every ball B ⊂ Rn we have
∫ ∫ 
n 
(j)
|Ψx (y)| dy dx
B K j=1

∫  ∫
dx 
≤ Cn |u(y)| dy sup < +∞, (1.2.7)
y ∈K |x − y| n−1
K B

it follows that
L n -a.e. point x ∈ Ω has the property that
(j) (1.2.8)
Ψx ∈ Lloc 1 (Ω, L n ) ⊗ C for 1 ≤ j ≤ n.
n

Our second observation is that, as seen from Hölder’s inequality,

p
if the function u actually belongs to the space Lloc (Ω, L n ) ⊗ Cn for
some exponent p ∈ (n, ∞] then for every point x ∈ Ω we have that (1.2.9)
(j)
Ψx ∈ Lloc1 (Ω, L n ) ⊗ C for each exponent j ∈ {1, . . . , n}.
n

We now turn to the proof of Lemma 1.2.1.

Proof of Lemma 1.2.1 Consider the claim made in (1.2.4). Without loss of gener-
ality it may be assumed that 0 ∈ Ω and x = 0, the origin in Rn . Pick a scalar-
valued function θ ∈ 𝒞∞ (Rn ) with  the property that θ = 0 on B(0, 1) and θ = 1
on Rn \ B(0, 2). For each ε ∈ 0, 12 dist(0, ∂Ω) define θ ε : Rn → R by setting
θ ε (y) := θ(y/ε) for every y ∈ Rn . Then

1 − θ ε ∈ 𝒞∞
c (Ω), supp (∇θ ε ) ⊆ B(0, 2ε) \ B(0, ε),
(1.2.10)
and lim+ θ ε (y) = 1 for every y ∈ Rn \ {0},
ε→0

and there exists a constant C ∈ (0, ∞) such that

 j 
(∇ θ ε )(y) ≤ Cε −j for every y ∈ Rn and every j ∈ N0 . (1.2.11)

Next, fix ϕ ∈ 𝒞∞
c (Ω) and write
1.2 Integral Representation Formulas in Clifford Analysis 39


n 
(j)
D (Ω)⊗ C n ∂j Ψ0 , ϕ D(Ω)
j=1


n
 
(j)
=− D (Ω)⊗ C n Ψ0 , ∂j ϕ D(Ω)
j=1


n ∫
1 y
=  e j  u(y)(∂j ϕ)(y) dy
j=1
ωn−1 |y| n
Ω
∫
1 y
= lim+  (D ϕ)(y)  u(y)θ ε (y) dy, (1.2.12)
ε→0 ωn−1 |y| n
Ω

where for the last equality in (1.2.12) we have used (1.2.10)-(1.2.11) and Lebesgue’s
Dominated Convergence Theorem. To proceed, observe that θ ε ϕ ∈ 𝒞∞ c (Ω) and
0  supp (θ ε ϕ). Consequently,
1 y
 e j  u(y)(θ ε ϕ)(y) ∈ Lcomp
1
(Ω, L n ) ⊗ Cn → ℰ (Ω) ⊗ Cn . (1.2.13)
ωn−1 |y| n
Recall that

1 
n  y 
∂j  e j = −δ in D (Ω) ⊗ Cn . (1.2.14)
ωn−1 j=1
|y| n

In particular,

1 
n  y 
∂j  e j = 0 for each y ∈ Ω \ {0}. (1.2.15)
ωn−1 j=1
|y| n

Bearing this in mind, in the sense of distributions in Ω we may compute

1 
n  y  1 y
∂j  e j  u θ ε ϕ = −Φ0 θ ε ϕ +  (D ϕ)  u θ ε
ωn−1 j=1
|y| n ωn−1 |y| n

1 y
+  (Dθ ε )  u ϕ. (1.2.16)
ωn−1 |y| n
Together with (1.2.13) this implies
40 1 Integral Representations and Integral Identities

 1  n  y  
0 = ℰ (Ω)⊗ C n ∂j  e j  u θ ε ϕ , 1 ℰ(Ω)
ωn−1 j=1 |y| n
∫ ∫
1 1 y
=− Φ0 θ ε ϕ dL n +  (D ϕ)(y)  u(y)θ ε (y) dy
ωn−1 Ω ωn−1 Ω |y| n
∫
1 y
+  (D θ ε )(y)  (ϕu)(y) dy, (1.2.17)
ωn−1 Ω |y| n
hence
∫
1 y
 (D ϕ)(y)  u(y)θ ε (y) dy
ωn−1 Ω |y| n
∫ ∫
1 y
= Φ0 θ ε ϕ dL n −  (D θ ε )(y)  (ϕu)(y) dy. (1.2.18)
Ω ωn−1 Ω |y| n
Thanks to assumptions and Lebesgue’s Dominated Convergence Theorem, we may
compute
∫ ∫
 
lim+ Φ0 θ ε ϕ dL n = Φ0 ϕ dL n = D (Ω)⊗ C n Φ0, ϕ D(Ω) . (1.2.19)
ε→0 Ω Ω

To handle the last integral in (1.2.18), split

∫
1 y
−  (D θ ε )(y)  u(y)ϕ(y) dy = I + I I, (1.2.20)
ωn−1 Ω |y| n
where
∫
1 y
I := −  (D θ)(y/ε)  [(ϕu)(y) − (ϕu)(0)] dy
ωn−1 ε ∫Ω |y| n
1   (1.2.21)
y
I I := −  (D θ ε )(y) dy  (ϕu)(0).
ωn−1 Ω |y| n
Using the support condition and estimate from (1.2.10)-(1.2.11), the fact that 0 is
a Lebesgue point of u (hence also of ϕu), and the properties of ϕ, term I may be
estimated by
 ⨏  
|I | ≤ Cn sup |∇θ| (ϕu)(y) − (ϕu)(0) dy −−−−→ 0. (1.2.22)
+ ε→0
Rn B(0,2ε)

As for I I, observe that for each ε > 0 we have

1.2 Integral Representation Formulas in Clifford Analysis 41
∫
1 y
−  (D θ ε )(y) dy
ωn−1 Ω |y| n

∫
1 y
=−  D [θ ε − 1](y) dy
ωn−1 Ω |y| n
 1  n  y  
= D (Ω)⊗ C n ∂j  e j , θ ε − 1 D(Ω)
ωn−1 j=1 |y| n


= − δ, θ ε − 1 = −(θ ε − 1)(0) = 1, (1.2.23)

by virtue of (1.2.14) and the properties of θ ε . Finally, by combining (1.2.12) with

(1.2.18)-(1.2.23) we arrive at the version of (1.2.4) written for x = 0.
Finally, consider the claims in the very last portion of the statement of the lemma.
The new aspect is that having fixed an arbitrary point x ∈ Rn \ Ω now renders
the function Ω y → (x − y)/|x − y| n of class 𝒞∞ in Ω. Granted this, and
keeping in mind the hypotheses on u, it follows that the conditions in (1.2.3) are
satisfied. Moreover, (1.2.5) is seen by a direct calculation making use of the product
formula for distributional derivatives (in a product involving a smooth function and
a distribution), since now we have

n  x−y 
∂y j  e j = 0 for each y ∈ Ω. (1.2.24)
j=1
|x − y| n

This finishes the proof of the lemma. 

Here is the higher-dimensional version of Theorem 1.1.1 promised earlier.

Theorem 1.2.2 Assume Ω ⊆ Rn is an open set with a lower Ahlfors regular bound-
ary, and with the property that σ := H n−1 ∂Ω is a doubling measure on ∂Ω. In
particular, Ω is a set of locally finite perimeter, and its geometric measure theoretic
outward unit normal ν is defined σ-a.e. on ∂∗ Ω.
In this context, suppose u : Ω → Cn is an L n -measurable Clifford algebra-
valued function which, for some κ > 0, satisfies
∫
(Nκ u)(x) κ−n.t.
dσ(x) < +∞ and u∂Ω exists σ-a.e. on ∂nta Ω. (1.2.25)
∂Ω 1 + |x|
n−1

Furthermore, with the Dirac operator D taken in the sense of distributions in Ω,

assume that  
dy
Du ∈ L 1 Ω, . (1.2.26)
1 + |y| n−1

κ −n.t.
Then for any κ > 0 the nontangential trace u∂Ω exists σ-a.e. on ∂nta Ω and
is actually independent of κ . Moreover, with the dependence on the parameter κ
dropped,
42 1 Integral Representations and Integral Identities

for each fixed x ∈ Ω, the Cn -valued function defined σ-a.e. on ∂∗ Ω as

x−y  n.t. 
∂∗ Ω y →  ν(y)  u∂Ω (y) belongs to L 1 (∂∗ Ω, σ) ⊗ Cn,
|x − y| n

(1.2.27)
and for L -a.e. point x ∈ Ω one has
n

∫  n.t. 
1 x−y
u(x) =  ν(y)  u∂Ω (y) dσ(y)
ωn−1 ∂∗ Ω |x − y| n

∫
1 x−y
−  (Du)(y) dy, (1.2.28)
ωn−1 Ω |x − y| n
when either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when
Ω is unbounded and ∂Ω is bounded, formula (1.2.28) continues to hold under the
additional assumption that there exists λ ∈ (1, ∞) such that
∫
|u(x)| dL n = o(Rn ) as R → ∞. (1.2.29)
[B(0,λ R)\B(0,R)]∩Ω

As a corollary, if actually Du = 0 in Ω then one has the following integral

representation formula
∫  n.t. 
1 x−y
u(x) =  ν(y)  u∂Ω (y) dσ(y), ∀x ∈ Ω, (1.2.30)
ωn−1 ∂∗ Ω |x − y| n

provided either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when

Ω is unbounded and ∂Ω is bounded, formula (1.2.30) continues to be true under the
additional assumption that there exists λ ∈ (1, ∞) such that (1.2.29) holds.

Before proving this theorem we wish to remark that [112, Lemma 8.3.1] and the
first property in (1.2.25) imply
∞
u ∈ Lloc (Ω, L n ) ⊗ Cn . (1.2.31)

In particular, u ∈ Lloc
1 (Ω, L n ) ⊗ C which ensures that it is meaningful to speak of
n
the Clifford algebra-valued distribution Du in (1.2.26).
Concerning the first condition in (1.2.25) it is worth noting that, thanks to Hölder’s
inequality and [112, Lemma 7.2.1],
& ∫
∂Ω upper Ahlfors regular, and (Nκ u)(x)
=⇒ dσ(x) < +∞, (1.2.32)
Nκ u ∈ L p (∂Ω, σ) with 1 ≤ p < ∞ ∂Ω + |x|
1 n−1

and, obviously,

whenever ∂Ω is bounded, we have

∫ (1.2.33)
(Nκ u)(x)
dσ(x) < +∞ ⇐⇒ Nκ u ∈ L 1 (∂Ω, σ).
∂Ω 1 + |x| n−1
1.2 Integral Representation Formulas in Clifford Analysis 43

Let us also remark that it is possible to further relax the first condition in (1.2.25) to
demanding instead that
∫   κ−n.t.  
 u (x)
∂Ω
Nκ u ∈ Lloc
1
(∂Ω, σ) and dσ(y) < +∞. (1.2.34)
∂∗ Ω 1 + |x| n−1

This would require using [112, Corollary 1.5.2] (in the version recorded in [112,
(1.5.23)]) in place of [112, Theorem 1.4.1] (as we do in the proof of Theorem 1.2.2
given further below). We also wish to point out that, if [112, Theorem 1.5.1] is
employed in place of [112, Theorem 1.4.1] in the proof of Theorem 1.2.2, then the
doubling assumption on σ := H n−1 ∂Ω may be relaxed to simply demanding that
this is a locally finite measure. In such a scenario, we need to impose the condition
that the aperture parameter κ is sufficiently large (depending on Ω), and the ability
of changing this parameter when considering nontangential boundary traces is lost.
However, modulo these adjustments, the format of the main result (i.e., the integral
representation formula (1.2.28)) stays the same.
As regards the integrability condition imposed on Du, [112, Lemma 3.5.7] guar-
antees that (1.2.26) is equivalent with having, for L n -a.e. point x ∈ Rn ,
∫
|(Du)(y)|
dy < +∞. (1.2.35)
|x − y| n−1
Ω

In the case when Ω is an exterior domain, there is a version of the integral

representation formula (1.2.28) which brings to the forefront the behavior of u
at infinity in a manner different than the growth condition imposed in (1.2.29).
Specifically, [112, (1.4.5)] and [112, Proposition 4.7.1] imply (on account of (1.2.43)
and (1.2.48)) that if Ω is an exterior domain in Rn and one assumes that

u is continuous in a neighborhood of infinity and

⨏ (1.2.36)
|u(y)| dH n−1 (y) = o(R) as R → ∞,
|y |=R

(hence, in particular, if u is continuous and bounded in a neighborhood of infinity)

then the limit
⨏
u∞ := lim u(y) dH n−1 (y) exists in Cn (1.2.37)
R→∞ |y |=R

and for L n -a.e. point x ∈ Ω one has

∫  n.t. 
1 x−y
u(x) =  ν(y)  u∂Ω (y) dσ(y)
ωn−1 ∂∗ Ω |x − y| n
∫
1 x−y
−  (Du)(y) dy + u∞ . (1.2.38)
ωn−1 Ω |x − y| n
44 1 Integral Representations and Integral Identities

In light of the last part in Lemma 1.2.1, our proof of Theorem 1.2.2 gives more,
namely that for each x ∈ Rn \ Ω one has
∫  n.t. 
1 x−y
0=  ν(y)  u∂Ω (y) dσ(y)
ωn−1 ∂∗ Ω |x − y| n
∫
1 x−y
−  (Du)(y) dy, (1.2.39)
ωn−1 Ω |x − y| n
provided either Ω is bounded, or ∂Ω is unbounded. Moreover, formula (1.2.39)
continues to be valid in the remaining case, i.e., when Ω is unbounded and ∂Ω
is bounded, under the additional assumption that there exists λ ∈ (1, ∞) such that
(1.2.29) holds.
After this digression, we are ready to present the proof of Theorem 1.2.2.
κ −n.t.
Proof of Theorem 1.2.2 That for any κ > 0 the nontangential trace u∂Ω exists
σ-a.e. on ∂nta Ω and is actually independent of κ follows from assumptions and [112,
Corollary 8.9.9]. Also, the claim in (1.2.27) is a consequence of (1.2.25) and [112,
(6.4.34), (8.8.52), (8.9.8), (8.9.44)].
To proceed, for each point x ∈ Ω with the property that (1.2.35) holds consider
the Clifford algebra-valued function defined as
1 x−y
Φx (y) :=  (Du)(y) for L n -a.e. y ∈ Ω. (1.2.40)
ωn−1 |x − y| n
In view of (1.2.26), the function Φx is L n -measurable and satisfies
∫ ∫
1 |(Du)(y)|
|Φx (y)| dy ≤ dy < +∞, (1.2.41)
Ω ωn−1 |x − y| n−1
Ω

thanks to [112, (6.4.35)] and (1.2.35). Hence, for each x ∈ Ω such that (1.2.35) holds
we have
Φx ∈ L 1 (Ω, L n ) ⊗ Cn . (1.2.42)
Moving on, if for each fixed x ∈ Ω we define
 &
1 x − y
Fx (y) :=  e j  u(y) for L n -a.e. y ∈ Ω, (1.2.43)
ωn−1 |x − y| n
1≤ j ≤n

then thanks to (1.2.31) we have

Fx ∈ Lloc
1
(Ω, L n ) ⊗ Cn, ∀x ∈ Ω. (1.2.44)

In addition, from (1.2.43) and the last property in (1.2.25) we conclude that, for each
κ−n.t.
fixed x ∈ Ω, the nontangential trace Fx ∂Ω exists at σ-a.e. point on ∂nta Ω and, in
fact,
1.2 Integral Representation Formulas in Clifford Analysis 45
 &
 κ−n.t.  1 x−y  κ−n.t. 
Fx ∂Ω (y) =  e j  u∂Ω (y) (1.2.45)
ωn−1 |x − y| n
1≤ j ≤n

for σ-a.e. y ∈ ∂nta Ω.

 
Going further, introduce K := B x, dist(x, ∂Ω)/2 which is a compact subset of
Ω. Then (1.2.43) and [112, (6.4.35), (8.3.45)] allow us to estimate
 Ω\K   
Nκ Fx (y) ≤ Cn NκΩ\K u (y) · sup |x − z| −(n−1)
z ∈Γκ (y)\K

(Nκ u)(y)
≤ Cx,n,κ for all y ∈ ∂Ω. (1.2.46)
|x − y| n−1
Collectively, [112, (8.2.26)], (1.2.46), and the first condition in (1.2.25) prove that
for each fixed x ∈ Ω we have

NκΩ\K Fx ∈ L 1 (∂Ω, σ). (1.2.47)

Granted (1.2.42) and (1.2.44), we may also invoke Lemma 1.2.1. This gives that
for each Lebesgue point x ∈ Ω of u such that (1.2.35) holds we have

divFx = u(x)δx + Φx ∈ ℰ (Ω) ⊗ Cn + L 1 (Ω, L n ) ⊗ Cn, (1.2.48)

where the divergence has been taken in the sense of distributions in Ω. Finally, we
wish to note that in the case when Ω is an exterior domain, condition (1.2.29) implies
that for each fixed x ∈ Ω we have

the vector field Fx satisfies the growth condition [112, (1.4.8)]. (1.2.49)

Having established (1.2.44), (1.2.45), (1.2.47), (1.2.48), (1.2.49), [112, Theo-

rem 1.4.1] applies and, with the dependence on the parameter κ dropped, formula
[112, (1.4.6)] presently gives that for each Lebesgue point x ∈ Ω of u such that
(1.2.35) holds we have
∫
 
u(x) + Φx (y) dy = (𝒞∞ (Ω))∗ divFx, 1 𝒞∞b (Ω)
Ω b

∫ 
n
 n.t. 
= ν j (y) Fx ∂Ω j (y) dσ(y)
∂∗ Ω j=1

∫ n   n.t. 
1 x−y
=  ν j (y)e j  u∂Ω (y) dσ(y)
ωn−1 |x − y| n j=1
∂∗ Ω
∫  n.t. 
1 x−y
=  ν(y)  u∂Ω (y) dσ(y). (1.2.50)
ωn−1 |x − y| n
∂∗ Ω
46 1 Integral Representations and Integral Identities

In view of (1.2.40), this proves (1.2.28) (also assuming (1.2.29) in the case when Ω
is an exterior domain) at each Lebesgue point x ∈ Ω of u such that (1.2.35) holds.

In the last part of this section we discuss an application to vector fields satisfying
the so called generalized Cauchy-Riemann system in the sense of Fefferman-Stein
(cf. [45, (8.1), p. 168]). Given an arbitrary open set Ω ⊆ Rn , recall that a vector
n
field u = (u j )1≤ j ≤n ∈ 𝒞∞ (Ω) is said to satisfy the Moisil-Teodorescu system (cf.
[136], [137], [138], [178]), or generalized Cauchy-Riemann equations14, if

⎧
⎪ ∂u j ∂uk
⎪
⎪ = in Ω for 1 ≤ j, k ≤ n,
⎪
⎪ ∂ x ∂ xj
⎨
⎪ k
(MT/GCR) (1.2.51)
⎪
⎪ n
∂u j
⎪
⎪ = 0 in Ω.
⎪
⎪ ∂ xj
⎩ j=1

If this is the case, then identifying the given vector field u with the Clifford algebra-
valued function
n
u := u j e j : Ω → Cn (1.2.52)
j=1

permits us to express (1.2.51) concisely as

Du = 0 in Ω. (1.2.53)

Let us now strengthen the hypotheses on Ω by also assuming that ∂Ω is a lower

Ahlfors regular set and that σ := H n−1 ∂Ω is a doubling measure on ∂Ω. Also, sup-
pose that either Ω is bounded, or ∂Ω is unbounded and n ≥ 2. In this setting, suppose
n
the vector field u = (u j )1≤ j ≤n ∈ 𝒞∞ (Ω) satisfies the Moisil-Teodorescu system
(aka the generalized Cauchy-Riemann equations) (1.2.51) in Ω and, additionally, for
each index j ∈ {1, . . . , n} we have
∫
(Nκ u j )(x) n.t.
dσ(x) < +∞ and u j ∂Ω exists σ-a.e. on ∂nta Ω. (1.2.54)
∂Ω 1 + |x|
n−1

Then Theorem 1.2.2 implies that

∫  n.t. 
1 x−y
u(x) =  ν(y)  u∂Ω (y) dσ(y), ∀x ∈ Ω. (1.2.55)
ωn−1 ∂∗ Ω |x − y| n

Taking the vector part of both sides yields, on account of [112, (6.4.87)], that for
each x ∈ Ω we have

14 a piece of terminology used in [165], [170] (both of these papers cite [138])
1.3 First and Second-Order Elliptic Systems of Partial Differential Operators 47
∫
1 y − x, ν(y)  n.t. 
u(x) = u ∂Ω (y) dσ(y)
ωn−1 ∂∗ Ω |x − y| n
  n.t.  
∫ x − y , u∂Ω (y)
1
+ ν(y) dσ(y)
ωn−1 ∂∗ Ω |x − y| n
∫   n.t.   x − y
1
− ν(y) , u∂Ω (y) dσ(y). (1.2.56)
ωn−1 ∂∗ Ω |x − y| n

Anticipating notation to be introduced later, in terms of the harmonic double layer D

from (2.5.202) and the operators R jk defined in (2.5.256), for each j ∈ {1, . . . , n} this
further implies the following integral representation formula for the j-th component
of the vector field u satisfying the generalized Cauchy-Riemann equations in Ω as
well as (1.2.54):
 n.t.  n  n.t. 
u j = D u j ∂Ω − R jk uk ∂Ω in Ω. (1.2.57)
k=1

Finally, we wish to note that formula (1.2.57) also holds if Ω is unbounded and either
∂Ω is bounded or n = 1, provided the decay condition (1.2.29) is satisfied.

1.3 First and Second-Order Elliptic Systems of Partial

Differential Operators

We begin by introducing the notion of weak ellipticity, in relation to second-order

systems, which plays a basic role in this work.

Definition 1.3.1 Fix n, M ∈ N, with n ≥ 2, and let

 αβ 
L = ar s ∂r ∂s 1≤α,β ≤M (1.3.1)

be a homogeneous M × M second-order system in Rn , with complex constant coef-

ficients. Call L weakly elliptic provided its M × M symbol (or characteristic)
matrix15  αβ 
L(ξ) := − ar s ξr ξs 1≤α,β ≤M , ∀ξ = (ξr )1≤r ≤n ∈ Rn, (1.3.2)
is an invertible linear mapping for each ξ ∈ Rn \ {0}, i.e., it satisfies

det L(ξ)  0, ∀ξ ∈ Rn \ {0}. (1.3.3)

15 The reader is alerted to the fact that our present definition of L(ξ) differs by a minus sign from
what [109, (11.3.2), p. 391] would give for second-order systems
48 1 Integral Representations and Integral Identities

In certain segments of the literature on partial differential equations, a system

satisfying (1.3.3) is referred to as being elliptic in the sense of Petrowskiĭ (cf., e.g.,
[107, p. 275]). The notion of weak ellipticity for a second-order system brought forth
in Definition 1.3.1 is ubiquitous in this work. The present terminology is suggested
by the fact that the demand imposed in (1.3.3) is weaker than ellipticity considered
in a strong sense (à la Legendre-Hadamard) which requires the characteristic matrix
(evaluated at points in Rn \ {0}) to be a strictly positive definite matrix. Here is a
formal definition of the latter brand of ellipticity:
Definition 1.3.2 Fix M, n ∈ N and let L be a homogeneous, second-order, con-
stant complex coefficient, M × M system in Rn . Say that L satisfies the Legendre-
Hadamard (strong) ellipticity condition provided

there exists a constant c ∈ (0, ∞) with the property that

  (1.3.4)
Re − L(ξ)ζ , ζ ≥ c |ξ | 2 |ζ | 2 for all ξ ∈ Rn and ζ ∈ C M .

It is then clear that, in the context of the above definition,

L is Legendre-Hadamard (strongly) elliptic if and only if

(1.3.5)
L  is Legendre-Hadamard (strongly) elliptic.

Of course, any Legendre-Hadamard (strongly) elliptic second-order system is

weakly elliptic, but the latter label applies to a larger class. In this vein, it is instructive
to contrast the two notions of ellipticity discussed above by consider the case of the
complex Lamé system, defined in relation to any two complex parameters λ, μ ∈ C
as
n
Lλ,μ u := μ Δu + (λ + μ)∇divu, ∀u = (u1, . . . , un ) ∈ D (Rn ) , (1.3.6)

where Δ := ∂12 + · · · + ∂n2 , the Laplacian in Rn , acts on the individual components of

u. If for any a = (a1, . . . , a N ) ∈ C N and b = (b1, . . . , b M ) ∈ C M , where N, M ∈ N,
we agree to define a ⊗ b to be the N × M matrix
 
a ⊗ b := a j bk 1≤ j ≤ N (1.3.7)
1≤k ≤M

then the characteristic matrix for the complex Lamé system (1.3.6) may be expressed
as
Lλ,μ (ξ) = −μ|ξ | 2 In×n − (λ + μ)ξ ⊗ ξ, ∀ξ ∈ Rn . (1.3.8)
In relation to (1.3.8), the following result appears in [109, Proposition 10.14, p. 366].
Proposition 1.3.3 Given any complex numbers μ, λ ∈ C, it follows that Lλ,μ (ξ) is
invertible for some (or every) ξ ∈ Rn \ {0} if and only if μ  0 and λ + 2μ  0. In
particular,
the complex Lamé system (1.3.6) is weakly elliptic
(1.3.9)
if and only if one has μ  0 and λ + 2μ  0.
Moreover, if μ  0 and λ + 2μ  0, then for each ξ ∈ Rn \ {0} one has
1.3 First and Second-Order Elliptic Systems of Partial Differential Operators 49

−1 −1  λ+μ ξ ξ 
Lλ,μ (ξ) = In×n − ⊗ . (1.3.10)
μ|ξ | 2 λ + 2μ |ξ | |ξ |

As far as strong ellipticity is concerned, we claim that

the Lamé system Lλ,μ = μ Δ + (λ + μ)∇div, with moduli

λ, μ ∈ C, is Legendre-Hadamard strongly elliptic if and only (1.3.11)
if Re μ > 0 and Re (2μ + λ) > 0.
Indeed, given any λ, μ ∈ C, this a consequence of the fact that
   
inf
n n
Re − L λ,μ (ξ)ζ, ζ = min Re μ, Re (λ + 2μ) (1.3.12)
ξ ∈R ,ζ ∈C
|ξ |=1, |ζ |=1

plus rescaling. To justify (1.3.12), we proceed by “double inequality.” In one direc-

tion, for each pair of vectors ξ ∈ Rn , ζ ∈ Cn with |ξ | = 1 and |ζ | = 1 write
 
Re − Lλ,μ (ξ)ζ, ζ = Re μ + Re (λ + μ)|ξ, ζ| 2
 
= Re μ 1 − |ξ, ζ| 2 + Re (λ + 2μ)|ξ, ζ| 2
   
≥ min Re μ, Re (λ + 2μ) · 1 − |ξ, ζ| 2
 
+ min Re μ, Re (λ + 2μ) · |ξ, ζ| 2
 
= min Re μ, Re (λ + 2μ) , (1.3.13)

where we have used the fact that 1 − |ξ, ζ| 2 ≥ 0, by Cauchy-Schwarz’s inequality.
Taking infimum establishes one of the desired inequalities. The opposite inequality
is obtained by simply specializing the first line in (1.3.13) to the case when ζ ∈ Cn
is a unit vector orthogonal to ξ ∈ S n−1 and to the case when ζ := ξ ∈ S n−1 . This
proves (1.3.12) from which (1.3.11) follows.
Moving on, we note that for future endeavors is going to be important to adopt an
alternative point of view, namely start with a given coefficient tensor and associate
it to it a homogeneous second-order system. Specifically, assume n, M ∈ N and
consider a coefficient tensor
 
αβ
A = ar s 1≤r,s ≤n (1.3.14)
1≤α,β ≤M

with complex entries. To this, we agree to associate the M × M second-order system

(as usual, the summation convention over repeated indices is in effect)
 
αβ
L = L A := ar s ∂r ∂s 1≤α ≤M . (1.3.15)
1≤β ≤M
50 1 Integral Representations and Integral Identities

We continue by introducing the class of injectively elliptic first-order systems in

Rn .

Definition 1.3.4 Consider an arbitrary N × M homogeneous first-order system with

constant complex coefficients in Rn (where N, M ∈ N)

n 
αβ
D= a j ∂j 1≤α ≤ N (1.3.16)
j=1 1≤β ≤M

and recall that its principal symbol is defined as the N × M matrix


n 
αβ
Sym(D; ξ) := i aj ξj 1≤α ≤ N for each ξ = (ξ1, . . . , ξn ) ∈ Rn . (1.3.17)
j=1 1≤β ≤M

Then D is injectively elliptic, i.e., it has the property that the mapping

Sym(D; ξ) : C M −→ C N is injective whenever ξ ∈ Rn \ {0}. (1.3.18)

Various equivalent characterizations of the class of injectively elliptic first-order

systems in Rn are contained in the lemma below.

Lemma 1.3.5 Fix N, M ∈ N and let


n 
αβ
D= a j ∂j 1≤α ≤ N (1.3.19)
j=1 1≤β ≤M

be some N × M homogeneous first-order system with constant complex coefficients

in Rn . Then the following statements are equivalent:

(i) The first-order system D is injectively elliptic (in the sense of Definition 1.3.4).
(ii) With ‘bar’ denoting complex conjugation, the homogeneous, constant (complex)
coefficient second-order M × M system in Rn given by

L := −D∗ D = −(D) D (1.3.20)

is Legendre-Hadamard strongly elliptic (hence also weakly elliptic, in the sense

of Definition 1.3.1).
(iii) There exists some M × N homogeneous first-order system D ' with constant
complex coefficients in Rn with the property that the composition

' is a weakly elliptic homogeneous, constant (com-

L := DD (1.3.21)
plex) coefficient, second-order M × M system in Rn .
Proof Assume first that D is injectively elliptic. Since from (1.3.19) and (A.0.38)
we have
1.3 First and Second-Order Elliptic Systems of Partial Differential Operators 51


n 
αβ
D = − a j ∂j 1≤β ≤M , (1.3.22)
j=1 1≤α ≤ N

it follows that the system L := D∗ D (cf. (1.3.20)) may be written explicitly as

 
βα βγ
L = − a j ak ∂j ∂k . (1.3.23)
1≤α,γ ≤M

Then for each two vectors ξ ∈ Rn and w ∈ C M the characteristic matrix of L (cf.
(1.3.2)) satisfies
 2
L(ξ)w, w = a j ak ξ j ξk wα wγ = Sym(D; ξ)w  ,
βα βγ
(1.3.24)

so, in particular,

L(ξ)w, w ∈ (0, ∞) for all vectors

(1.3.25)
ξ ∈ Rn and w ∈ C M with |ξ | = |w| = 1.

From this, the Extreme Value Theorem (for continuous functions on compact sets),
and homogeneity we then conclude that there exists some c ∈ (0, ∞) with the property
that
Re L(ξ)w, w ≥ c|ξ | 2 |w| 2 for all ξ ∈ Rn and w ∈ C M . (1.3.26)
In light of Definition 1.3.2, this shows that the system −L satisfies the Legendre-
Hadamard (strong) ellipticity condition. The veracity of the implication (i) ⇒ (ii) is
therefore established.
Next, choosing D' := −D∗ proves that we also have (ii) ⇒ (iii). To verify that
(iii) ⇒ (i), assume that there exists some M × N homogeneous first-order system
' with constant complex coefficients in Rn such that (1.3.21) holds. Then the
D
characteristic matrix L(ξ) of L is invertible for each vector ξ ∈ Rn \ {0}, and since
' ξ)Sym(D; ξ) for each ξ ∈ Rn,
L(ξ) = Sym( D; (1.3.27)

it follows that (1.3.18) holds. The proof of the lemma is therefore complete. 

This lemma suggests making the following definition.

Definition 1.3.6 Let D be a homogeneous first-order N × M system with constant

complex coefficients in Rn (where N, M ∈ N). Say that a homogeneous first-order
M × N system D ' with constant complex coefficients in Rn complements D provided
(1.3.21) holds.

In this piece of terminology, Lemma 1.3.5 guarantees that

a given homogeneous constant coefficient first-order system D
has a complement if and only if D is injectively elliptic, and if (1.3.28)
D is injectively elliptic then D∗ complements D.
52 1 Integral Representations and Integral Identities

It is worth mentioning that there could be infinitely many homogeneous first-order

systems D ' complementing a given homogeneous first-order injectively elliptic sys-
tem D in Rn . For example any operator of the form divA, where A ∈ Rn×n is a
strictly positive-definite matrix, complements the gradient operator in Rn .
Several notable examples of injectively elliptic first-order systems are discussed
below.
Example 1.3.7 The Cauchy-Riemann operator ∂ from (1.1.2) and the Dirac operator
D from (A.0.37) are injectively elliptic first-order systems. A moment’s reflection
shows that the gradient operator ∇ is also an injectively elliptic first-order system
in Rn . More generally, the Jacobian (A.0.48), i.e., the operator acting on each
M
vector-valued distribution u = (u1, . . . , u M ) ∈ D (Rn ) (for some fixed M ∈ N)
according to
∂1 u1 · · · ∂n u1
  ( .. +, .
∇u := ∂k u j 1≤ j ≤M = )) ... ..
. . , (1.3.29)
1≤k ≤n
* ∂1 u M · · · ∂n u M -
is injectively elliptic. To justify the latter, observe that for each ξ = (ξk )1≤k ≤n ∈ Rn
and each u = (u j )1≤ j ≤M ∈ C M we have
 
(−i)Sym(∇; ξ)u = ξk u j 1≤ j ≤M = u ⊗ ξ. (1.3.30)
1≤k ≤n

This further implies

 
(−i)Sym(∇; ξ)u ξ = (u ⊗ ξ)ξ = |ξ | 2 u, (1.3.31)

from which it is then clear that the Jacobian operator (1.3.29) is indeed injectively
elliptic.
Example 1.3.8 The Moisil-Teodorescu system (aka the generalized Cauchy-Riemann
equations) acting on a vector-valued distribution u = (u1, . . . , un ) in Rn according
to
  
n 
Du := ∂j uk − ∂k u j 1≤ j,k ≤n, ∂j u j (1.3.32)
j=1

is injectively elliptic. To justify this, observe first that

  . 
Sym(D; ξ)u := i ξ j uk − ξk u j 1≤ j,k ≤n, nj=1 ξ j u j
(1.3.33)
for each u = (u j )1≤ j ≤n and ξ := (ξ1, . . . , ξn ) ∈ Rn .

If ξ  0 and yet Sym(D; ξ)u = 0, multiplying each ξ j uk − ξk u j by ξ j , summing

.
in j, then using nj=1 ξ j u j = 0, leads to the conclusion that |ξ | 2 uk = 0 for each
k ∈ {1, . . . , n}. Hence, u = 0 as desired.
Example 1.3.9 Consider a matrix A = (a jk )1≤ j,k ≤n ∈ Cn×n which is weakly elliptic,
i.e., satisfies
1.3 First and Second-Order Elliptic Systems of Partial Differential Operators 53

Aξ, ξ = a jk ξ j ξk  0 for all ξ ∈ Rn \ {0}. (1.3.34)

Then the modified Moisil-Teodorescu system, associated with A, acting on a vector-
valued distribution u = (u1, . . . , un ) in Rn according to
  
Du := ∂j uk − ∂k u j 1≤ j,k ≤n, div Au (1.3.35)

is injectively elliptic. Indeed,

  
Sym(D; ξ)u := i ξ j uk − ξk u j 1≤ j,k ≤n, a njk ξ j uk
(1.3.36)
for each u = (u j )1≤ j ≤n and ξ := (ξ1, . . . , ξn ) ∈ Rn .

Assume ξ  0 yet Sym(D; ξ)u = 0. For an arbitrary  ∈ {1, . . . , n}, multiply the
equality a njk ξ j uk = 0 by ξ then use the fact that ξ uk = ξk u to obtain a jk ξ j ξk u = 0.
In view of the fact that (1.3.34) ensures a jk ξ j ξk  0, this forces u = 0 for each
 ∈ {1, . . . , n}. Thus, u = 0 as wanted.
 
Example 1.3.10 The first-order system D := ∂z̄ j 1≤ j ≤n in Cn ≡ R2n is injectively
 
elliptic. Indeed, since for each ξ ∈ Cn ≡ R2n we have Sym(D; ξ)u = 12 |ξ ||u|, the
desired conclusion readily follows.
Example 1.3.11 With the summation convention over repeated indices in effect,
suppose  
αβ
L = ar s ∂r ∂s 1≤α ≤M . (1.3.37)
1≤β ≤M

is a weakly elliptic homogeneous constant (complex) coefficient M × M second-order

system in Rn . Then the first-order, homogeneous, constant coefficient, (n · M) × M
system  αβ 
Du := A∇u := ar s ∂s uβ 1≤α ≤M (1.3.38)
1≤r ≤n
   
is injectively elliptic. Indeed, we have iξr Sym(D; ξ)u α,r = L(ξ)u α for each
ξ ∈ Rn , u ∈ C M , and α ∈ {1, . . . , M }. Since the characteristic matrix L(ξ) is
invertible whenever ξ ∈ Rn \ {0}, it follows that Sym(D; ξ) is, as claimed, injective
whenever ξ ∈ Rn \ {0}.
Example 1.3.12 Suppose L is a weakly elliptic, second-order, homogeneous, con-
stant (complex) coefficient M × M system associated with some coefficient tensor
αβ 
A = ar s 1≤r,s ≤n as in (1.3.15). In relation to this, introduce the homogeneous
1≤α,β ≤M
constant coefficient first-order differential operator D A, mapping distributions in Rn
 β n×M
with n × M components w = ws 1≤s ≤n ∈ D (Rn ) into
1≤β ≤M

 β β
( ∂r ws − ∂s wr 1≤r,s ≤n +
D A w := )   1≤β ≤M , . (1.3.39)
αβ β
* ar s ∂r ws 1≤α ≤M -
Then
54 1 Integral Representations and Integral Identities
 
as a linear operator, the matrix Sym D A; ξ is injective for each
vector ξ ∈ Rn \ {0}, hence the homogeneous constant coefficient (1.3.40)
first-order system D A is injectively elliptic,
and
 
Ker L u = (uβ )1≤β ≤M −→ ∇u = ∂s uβ 1≤s ≤n ∈ Ker D A
1≤β ≤M (1.3.41)
is a well-defined linear mapping.

To see that this is the case, observe that the principal symbol
 of D A at each point
β
ξ = (ξr )1≤r ≤n ∈ Rn is the linear mapping acting on w = ws 1≤s ≤n ∈ Cn×M
1≤β ≤M
according to
 β β
 ( ξr ws − ξs wr 1≤r,s
 ≤n +
Sym D A; ξ w = i )  1≤β ≤M , . (1.3.42)
αβ β
* ar s ξr ws 1≤α ≤M -
 β
If ξ = (ξr )1≤r ≤n ∈ Rn \ {0} and w = ws 1≤s ≤n ∈ Cn×M are such that
  1≤β ≤M
Sym D A; ξ w = 0, then
β β
ξr ws − ξs wr = 0 for 1 ≤ r, s ≤ n and 1 ≤ β ≤ M,
(1.3.43)
αβ β
ar s ξr ws = 0 whenever 1 ≤ α ≤ M.

Since ξ  0, there exists some r∗ ∈ {1, . . . , n} such that ξr∗  0. If we now define
β
wr∗
ηβ := ∈ C for each β ∈ {1, . . . , M }, (1.3.44)
ξr∗

then the first line in (1.3.43) implies that, for each β ∈ {1, . . . , M } and for each
s ∈ {1, . . . , n}, we have
β
β β wr∗
ξr∗ ws = ξs wr∗ = ξs ξr∗ = ξs ξr∗ ηβ . (1.3.45)
ξr∗

Thus,
β
ws = ξs ηβ for each β ∈ {1, . . . , M } and s ∈ {1, . . . , n}. (1.3.46)

In concert with the second line in (1.3.43), this permits us to write

αβ
ar s ξr ξs ηβ = 0 for each α ∈ {1, . . . , M }. (1.3.47)

In view of the weak ellipticity condition (1.3.3), we conclude from (1.3.47) that each
β
ηβ is zero which, in concert with (1.3.46), ultimately proves that ws = 0 whenever
1 ≤ s ≤ n and 1 ≤ β ≤ M. Bearing (1.3.18) in mind, the claims in (1.3.40) follow.
Finally, (1.3.41) is clear from definitions.
1.3 First and Second-Order Elliptic Systems of Partial Differential Operators 55

In relation to Example 1.3.12, it is worth noting that

αβ
if M := 1, L := Δ, and ar s = δαβ δr s then the first-order differ-
ential operator D A from (1.3.39) reduces to the Moisil-Teodorescu (1.3.48)
system (aka the generalized Cauchy-Riemann equations) discussed
in Example 1.3.8.

Example 1.3.13 The deformation tensor in Rn , given by

 
Def u := 12 ∂j uk + ∂k u j 1≤ j,k ≤n
n (1.3.49)
for each u = (u1, . . . , un ) ∈ D (Rn ) ,

is an injectively elliptic, first-order homogeneous, constant (real) coefficient, n2 × n

system in Rn .

To justify this, observe that for every ξ = (ξ1, . . . , ξn ) ∈ Rn we have

 
(−i)Sym(Def; ξ) u = 12 ξ j uk + ξk u j 1≤ j,k ≤n
(1.3.50)
for each u = (u1, . . . , un ) ∈ Rn .

As a consequence, if ξ = (ξ1, . . . , ξn ) ∈ Rn \ {0} is such that (−i)Sym(Def; ξ) u = 0

for some vector u = (u1, . . . , un ) ∈ Rn then ξ j uk + ξk u j = 0 for each integers
j, k ∈ {1, . . . , n}. In particular, for each fixed index k ∈ {1, . . . , n} we have the
equality 0 = ξ j (ξ j uk + ξk u j ) = |ξ | 2 uk + ξk ξ, u, hence

ξ, u
u = − ξ. (1.3.51)
|ξ | 2
After taking the dot product with ξ this further implies

ξ, u 2
ξ, u = − |ξ | = −ξ, u, (1.3.52)
|ξ | 2

from which we conclude that ξ, u = 0. In light of (1.3.51), this forces u = 0, which
ultimately shows that the symbol mapping Sym(Def; ξ) is indeed injective.

Example 1.3.14 Employing the formalism associated with differential forms in Rn ,

with d and δ denoting the exterior derivative operator and its formal transpose,
respectively, consider the first-order homogeneous, constant (complex) coefficient
differential operator

Dα,β := αd + βδ where α, β ∈ C, (1.3.53)

acting on differential forms (of mixed, arbitrary degrees). Since, in view of the fact
d2 = δ2 = 0, we have
2
Dα,β = (αβ)(dδ + δd) = −(αβ)Δ, (1.3.54)
56 1 Integral Representations and Integral Identities

where Δ is the Laplacian in Rn , we conclude from Lemma 1.3.5 that Dα,β is an

injectively elliptic first-order operator for each α, β ∈ C \ {0}.

As far as weak ellipticity for second-order systems is concerned, the first example
that comes to mind is that of the Laplace operator Δ in Rn . In fact, we have already
noted that any Legendre-Hadamard strongly elliptic second-order system is also
weakly elliptic, and Lemma 1.3.5 tells us that

the second-order system L := −D∗ D is Legendre-Hadamard strongly

elliptic, hence also weakly elliptic, whenever D is a homogeneous (1.3.55)
(complex) constant coefficient injectively elliptic first-order system.
Consequently,

the second-order system L := −D∗ D is Legendre-Hadamard strongly

elliptic, hence also weakly elliptic, whenever D is any of the first-order (1.3.56)
systems from Examples 1.3.7-1.3.14.
Thus, among other things,

if D is the Jacobian operator

 (1.3.29),
 then its transpose D is given

by the formula D w = − ∂k w jk 1≤ j ≤M for each matrix-valued dis-
M×n
tribution w = (w jk )1≤ j ≤M ∈ D (Rn ) , a scenario in which the (1.3.57)
1≤k ≤n
second-order system L := −D∗ D
becomes −∇ ∇
= Δ, the Laplacian
in Rn acting on the M components of vector-valued distributions,
and
if D is the deformation tensor (1.3.49), then its 

 transpose D is given
by the formula D w = − 2 ∂k w jk + ∂k wk j 1≤ j ≤n for each matrix-
1
n×n
valued distribution w = (w jk )1≤ j,k ≤n ∈ D (Rn ) , so in this case (1.3.58)
 
the system L := −D∗ D takes the form −Def  Def = 12 Δ + ∇div on
vector-valued distributions in Rn .
The above discussion suggests the consideration of a family of second-order
systems, indexed by triplets of complex numbers a, b, c ∈ C, acting on vector-valued
distributions in Rn according to

La,b,c := −2a Def  Def + b∇div − c∇ ∇. (1.3.59)

In relation to this, we have the following results:

the system La,b,c is weakly elliptic

(1.3.60)
if and only if a + c  0 and 2a + b + c  0,

and
1.3 First and Second-Order Elliptic Systems of Partial Differential Operators 57
the system La,b,c is Legendre-Hadamard strongly elliptic
(1.3.61)
if and only if Re (a + c) > 0 and Re (2a + b + c) > 0.
Indeed, from (1.3.57), (1.3.58), and (1.3.6) we see that
the system La,b,c defined in (1.3.59) coincides with the complex Lamé
(1.3.62)
system Lλ,μ = μ Δ+(λ+ μ)∇div for the moduli λ := b−c and μ := a+c,

so the claims in (1.3.60)-(1.3.61) become consequences of (1.3.9) and (1.3.11).

Other relevant examples of weakly elliptic and strongly elliptic systems are dis-
cussed below.

Example 1.3.15 Work in R2 ≡ C. Bring in the Cauchy-Riemann operator ∂z̄ and its
complex conjugate ∂z from (1.1.2). For each parameter λ ∈ C then define the scalar,
second-order, homogeneous, differential operator

Lλ := ∂z̄2 − λ2 ∂z2 . (1.3.63)

Then for each ξ = (ξ1, ξ2 ) ∈ R2 we have

Lλ (ξ) = − 14 (ξ1 + iξ2 )2 − λ2 (ξ1 − iξ2 )2 (1.3.64)

If ξ ∈ R2 \ {0} is such that Lλ (ξ) = 0, then (1.3.64) implies

ξ1 + iξ2  ξ + iξ  2
1 2
λ=± =± , (1.3.65)
ξ1 − iξ2 |ξ |

which ultimately forces |λ| = 1. This analysis ultimately shows that

the differential operator Lλ is weakly elliptic

(1.3.66)
if and only if |λ|  1.

For the scalar operator Lλ defined in (1.3.63), the Legendre-Hadamard (strong)

ellipticity condition (1.3.4) amounts to the demand that

Re [−Lλ (ξ)] > 0 for each ξ ∈ S 1 . (1.3.67)

√
Testing this for the choices ξ = (1, ±1)/ 2 forces, in light of (1.3.64),

0 < Re (1 ± i)2 − λ2 (1 ∓ i)2 = Re ± 2i ± 2iλ2 = ±Re 2i + 2iλ2 (1.3.68)

an obvious impossibility, irrespective of λ ∈ C. This proves that the differential

operator
Lλ is not Legendre-Hadamard (strongly) elliptic
(1.3.69)
for any choice of λ ∈ C.
Our next example pertains to a family of operators acting on differential forms of
arbitrary degrees, which contains the complex Lamé system (1.3.6) as a particular
case.
58 1 Integral Representations and Integral Identities

Example 1.3.16 We shall employ the formalism associated with differential forms
in Rn . In particular, d and δ denote, respectively, the exterior derivative operator
and its formal transpose. For α, β ∈ C, consider the second-order homogeneous,
constant (complex) coefficient differential operator

Lα,β := αdδ + βδd (1.3.70)

acting on differential forms (of mixed, arbitrary degrees). Then

the differential operator Lα,β is weakly elliptic

(1.3.71)
for each α, β ∈ C \ {0},

and
Lα,β is Legendre-Hadamard (strongly) elliptic
(1.3.72)
if and only if Re α > 0 and Re β > 0.

Let us justify (1.3.71). In view of the fact

d2 = δ2 = 0 and dδ + δd = −Δ (1.3.73)

where Δ is the Laplacian in Rn (acting on differential forms componentwise), we

have
Lβ,α Lα,β = (αβ)Δ2 = Lα,β Lβ,α (1.3.74)
At the level of principal symbols, this translates into16

Lβ,α (ξ)Lα,β (ξ) = (αβ)|ξ | 4 = Lα,β (ξ)Lβ,α (ξ) for all ξ ∈ Rn . (1.3.75)

Hence, if αβ  0 we conclude that Lα,β (ξ) is invertible for each ξ ∈ Rn \ {0}, which
ultimately proves (1.3.71).
As far as (1.3.72) is concerned, via homogeneity this becomes a consequence of
   
inf Re − Lα,β (ξ)u, u = min Re α, Re β (1.3.76)
|ξ |=1, |u |=1

which, in turn, is justified by “double inequality.” To set the stage, observe that for
each ξ ∈ Rn and each differential form u, formula (1.3.70) entails
 
Lα,β (ξ)u = − αξ ∧ (ξ ∨ u) + βξ ∨ (ξ ∧ u) . (1.3.77)

Next, if ξ is a vector in Rn and u is a differential form in Rn with |ξ | = 1 and |u| = 1,

we may write (based on (1.3.77), item (3) in [112, Lemma 6.4.7], and the second
formula in [112, (6.4.133)])

16 suppressing the identity operator (acting on differential forms) in the middle term
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 59
      
Re − Lα,β (ξ)u, u = Re α ξ ∧ (ξ ∨ u), u + β ξ ∨ (ξ ∧ u), u
 
= Re α|ξ ∨ u| 2 + β|ξ ∧ u| 2 = (Re α)|ξ ∨ u| 2 + (Re β)|ξ ∧ u| 2
  
≥ min Re α, Re β · |ξ ∨ u| 2 + |ξ ∧ u| 2 )
 
= min Re α, Re β , (1.3.78)

which is one of the desired inequalities. The inequality in the opposite direction may
be seen by specializing
  
Re − Lα,β (ξ)u, u = (Re α)|ξ ∨ u| 2 + (Re β)|ξ ∧ u| 2 (1.3.79)

(itself implicit in (1.3.78)) to the case when u := ξ ∈ S n−1 and to the case when
u := ∗ξ with ξ ∈ S n−1 (where ‘star’ denotes the Hodge isomorphism). Thus, (1.3.72)
is established.
The manner in which the operator Lα,β defined in (1.3.70) ties up with the
complex Lamé system from (1.3.6) is as follows. From (1.3.70) and the last formula
in (1.3.73) we have

Lα,β = αdδ + β(−Δ − dδ) = −βΔ + (α − β)dδ (1.3.80)

hence,
when acting on vector fields, the operator Lα,β agrees with the complex
Lamé system (1.3.6) written for the Lamé moduli λ, μ chosen such that (1.3.81)
μ = −β and λ + μ = −(α − β), i.e., for λ := −α + 2β and μ := −β.
In this vein, it is reassuring to note that the demands on α, β in (1.3.71) agree with
the conditions imposed on λ, μ in (1.3.9), while the demands on α, β in (1.3.72)
agree with the conditions imposed on λ, μ in (1.3.11).

1.4 Fundamental Solutions for Weakly Elliptic Second-Order

Systems of Partial Differential Operators

Consider a function satisfying

∫
Θ ∈ 𝒞0 (Rn \ {0}), Θ(ξ) dH n−1 (ξ) = 0,
S n−1 (1.4.1)
and Θ is positive homogeneous of degree −n.
1 (Rn, L n ), so associating a distribution
Typically, such a function fails to belong to Lloc
with Θ requires some care. Specifically, with Θ from (1.4.1) we associate the principal
60 1 Integral Representations and Integral Identities

value distribution P.V. Θ ∈ 𝒮 (Rn ) defined as

∫
 
P.V. Θ, ϕ := lim+ Θ(x)ϕ(x) dx, ∀ϕ ∈ 𝒮(Rn ). (1.4.2)
ε→0 x ∈R n, |x |>ε

Then this is a well-defined tempered distribution in Rn with the property that

  
P.V. Θ Rn \{0} = ΘRn \{0} in D (Rn \ {0}). (1.4.3)

In addition, for each ϕ ∈ 𝒮 (Rn ) one has

 
P.V. Θ ∗ ϕ ∈ 𝒮 (Rn ) ∩ 𝒞∞ (Rn ) (1.4.4)

and
  ∫

P.V. Θ ∗ ϕ (x) = lim+ Θ(x − y)ϕ(y) dy, ∀x ∈ Rn . (1.4.5)
ε→0 y ∈R n, |x−y |>ε

Functions satisfying (1.4.1) arise naturally in applications. To give one basic

example, suppose some function Φ ∈ 𝒞1 (Rn \ {0}) which is positive homogeneous
of degree 1 − n has been given, and some index j ∈ {1, . . . , n} has been fixed.
Then ∂j Φ satisfies the conditions in (1.4.1) and, as such, P.V.(∂j Φ) is a well-defined
tempered distribution. Moreover, Φ ∈ Lloc 1 (Rn, L n ), hence Φ may be canonically

regarded as a distribution in Rn , and for each j ∈ {1, . . . , n} the distributional

derivative ∂j Φ may be computed as follows:
∫ 
∂j Φ = Φ(ξ)ξ j dH n−1 (ξ) δ + P.V.(∂j Φ) in 𝒮 (Rn ), (1.4.6)
S n−1

where δ is Dirac’s distribution in Rn with mass at the origin.

One remarkable aspect of the proposition below is that it provides a description
of a fundamental solution for a second-order system purely in terms of ordinary
calculus (i.e., without any reference to the theory of distributions).

Proposition 1.4.1 Pick n, M ∈ N with n ≥ 2, and consider the homogeneous,

second-order, M × M system, with constant complex coefficients,
 αβ 
L = ar s ∂r ∂s 1≤α,β ≤M (1.4.7)

(with the summation convention over repeated indices in effect). Also, consider
a matrix-valued function E : Rn \ {0} → C M×M whose entries belong to the
n/(n−1) n
intersection 𝒞2 (Rn \ {0})∩ Lloc (R , L n ) and have first-order derivatives positive
homogeneous of degree 1 − n in R \ {0}.
n

Then the following statements are equivalent:

1 (Rn, L n ) M×M
(1) When viewed in Lloc , the function E is a fundamental solution
for the system L in Rn .
(2) Each column of E is a (pointwise) null-solution of the system L in Rn \ {0}, and
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 61
∫
αγ
ξr ar s (∂s Eγβ )(ξ) dH n−1 (ξ) = δαβ
S n−1 (1.4.8)
for every α, β ∈ {1, . . . , M }.
αβ
Let us organize the complex numbers
 αβ  ar s used in the writing of L in (1.4.7) as a
“tensor” of coefficients A = ar s 1≤r,s ≤n . Then associate with coefficient tensor
1≤α,β ≤M
A the (co-)directional derivative operator along a vector ξ = (ξ1, . . . , ξn ) ∈ Rn by
setting  
αβ
∂ξAu := ξr ar s ∂s uβ , (1.4.9)
1≤α ≤M
for every vector-valued function u = (uβ )1≤β ≤M . The upshot is that this piece of
notation allows us to recast (1.4.8) succinctly as
∫
∂ξA E(ξ) dH n−1 (ξ) = I M×M , (1.4.10)
S n−1

where the conormal derivative operator ∂ξA acts on the columns of E.

Proof of Proposition 1.4.1 Start by writing (Eγβ )1≤γ,β ≤M for the scalar components
of the given matrix-valued function E. Fix an arbitrary vector-valued test function
ϕ = (ϕβ )1≤β ≤M , with components in 𝒞∞ c (R ). Then, with L acting on the columns
n

of E, making use of (1.4.6), [112, (4.5.8)], and (1.4.5) for each α ∈ {1, . . . , M } we
may write
  αγ  
(LE) ∗ ϕ α = ar s ∂r (∂s Eγβ ) ∗ ϕβ
∫ 
αγ
= ar s (∂s Eγβ )(ξ)ξr dH n−1 (ξ) (δ ∗ ϕβ )
S n−1
 
αγ
+ ar s P.V.(∂r ∂s Eγβ ) ∗ ϕβ
∫ 
αγ
= ξr ar s (∂s Eγβ )(ξ) dH n−1 (ξ) ϕβ (1.4.11)
S n−1
∫
αγ
+ lim+ ar s (∂r ∂s Eγβ )(· − y)ϕβ (y) dy in D (Rn ),
ε→0 |y−· |>ε

where we have also used the fact that δ ∗ ϕβ = ϕβ for each β ∈ {1, . . . , M }. Having
established this, we now turn in earnest to the proof of the equivalence claimed in
the statement of proposition.
First, assume that E isa fundamental
 solution for L in Rn . Then LE = δI M×M in

D (R ) implies that L E Rn \{0} = 0 in D (Rn ). Since by assumption the entries in
n

E Rn \{0} belong to 𝒞2 (Rn \ {0}), we arrive at the conclusion that LE = 0 pointwise
in Rn \ {0}. Explicitly, for each α, β ∈ {1, . . . , M },
αγ 
ar s ∂r ∂s Eγβ )(x) = 0, ∀x ∈ Rn \ {0}. (1.4.12)
62 1 Integral Representations and Integral Identities

This proves the first claim in item (2). Next, for each arbitrary C M -valued test
function ϕ = (ϕβ )1≤β ≤M , with components in the space 𝒞∞ c (R ), we may write
n

ϕ = (δI M×M ) ∗ ϕ = (LE) ∗ ϕ in D (Rn ). In light of (1.4.11) and (1.4.12), this forces
∫ 
αγ
ϕα = ξr ar s (∂s Eγβ )(ξ) dH n−1 (ξ) ϕβ, ∀α ∈ {1, . . . , M }. (1.4.13)
S n−1

Since ϕβ ∈ 𝒞∞ c (R ) has been arbitrarily chosen for each β ∈ {1, . . . , M }, identity

n

(1.4.8) follows from (1.4.13). This finishes the proof of (1) ⇒ (2).
Conversely, suppose LE = 0 ∈ C M×M pointwise in Rn \ {0}, and assume (1.4.8)
holds. Granted these, formula (1.4.11) simply reduces to (LE) ∗ ϕ = ϕ for each
ϕ = (ϕβ )1≤β ≤M with components in 𝒞∞ c (R ). In turn, this readily implies that
n

LE = δI M×M in D (R ), as wanted.
n 
Our next theorem provides a canonical way of associating a suitable fundamental
solution for any given second-order, homogeneous, weakly elliptic system, with
constant (complex) complex coefficients. In various degrees of generality, this issue
has been considered in a number of sources, including [74, pp. 72-76], [66, p. 169],
[142, Proposition 5.2.3, p. 349], [122], [108], [139, p. 104], [161]. The theorem
below is a special case of [109, Theorem 11.1, pp. 393-395], which treats weakly
elliptic, homogeneous, constant coefficient systems of arbitrary (even) order. To state
it, we make two conventions. The first convention regards the normalization of the
Fourier transform in Rn ; the reader is alerted to the fact that the normalization we
shall consistently employ throughout is different from those employed in [175], or
[166]. Specifically, we define the Fourier transform φ/ of any given complex-valued
Schwartz function φ in Rn as
∫
φ/(ξ) := e−i x,ξ  φ(x) dx, ξ ∈ Rn . (1.4.14)
Rn

The action of the Fourier transform then extends to tempered distributions via duality,
in the usual fashion (cf., e.g., [66], [108], [109]). The second convention is that,
unless explicitly stated otherwise, the summation convention over repeated indices
is in effect.

Theorem 1.4.2 Fix n, M ∈ N, with n ≥ 2, and let

 αβ 
L = ar s ∂r ∂s 1≤α,β ≤M (1.4.15)

be a second-order homogeneous M × M system in Rn , with complex constant co-

efficients, which is weakly elliptic in the sense of Definition 1.3.1. That is, the
characteristic matrix17 of L, given by
 αβ 
L(ξ) := − ar s ξr ξs 1≤α,β ≤M , ∀ξ = (ξr )1≤r ≤n ∈ Rn, (1.4.16)

17 Once again, the reader is alerted to the fact that our present definition of L(ξ) differs by a minus
sign from what [109, (11.3.2), p. 391] would give for second-order systems
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 63

satisfies
det L(ξ)  0 for each ξ ∈ Rn \ {0}. (1.4.17)
 
Consider the M × M matrix-valued function E = Eαβ 1≤α,β ≤M defined at each
x ∈ Rn \ {0} by18
 ∫ &
⎧
⎪ Δ (n−1)/2  
⎪
⎪ − x −1
 x, ξ  L(ξ) dH n−1 (ξ) if n is odd,
⎪
⎪
⎪
⎪ 4(2πi)n−1
⎪
⎪ S n−1
⎨
⎪
E(x) := (1.4.18)
⎪
⎪  &
⎪ Δ(n−2)/2 ∫
⎪  
⎪ −1
⎪
⎪
⎪
x
ln  x, ξ  L(ξ) dH n−1 (ξ) if n is even.
⎪
⎪ (2πi) n
⎩ S n−1

In relation to the C M×M -valued function (1.4.18), the following properties hold:
(1) For each α, β ∈ {1, . . . , M } one has

Eαβ ∈ 𝒞∞ (Rn \ {0}) ∩ Lloc

1 (Rn, L n ),
(1.4.19)
Eαβ (−x) = Eαβ (x) for all x ∈ Rn \ {0}.

In fact, each entry Eαβ is a real-analytic function in Rn \ {0}. Moreover, each Eαβ
is an even tempered distribution in Rn (induced via integration against Schwartz
functions).
(2) If for each y ∈ Rn one denotes by δy Dirac’s delta distribution with mass at
y in Rn , then in the sense of tempered distributions in Rn one has

Lx E(x − y) = δy (x) I M×M , ∀y ∈ Rn, (1.4.20)

where I M×M is the M × M identity matrix, and the subscript x indicates that the
operator L in (1.4.20) is applied to each column of the matrix E(x − y) in the variable
x. Componentwise, for each α, β ∈ {1, . . . , M } one therefore has

αγ 0 if α  β,
ar s ∂xr ∂xs Eγβ (x − y) = δαβ δy (x) = x, y ∈ Rn . (1.4.21)
δy (x) if α = β,

(3) For each multi-index γ ∈ N0n with |γ| > 0, the tempered distribution ∂γ E
is positive homogeneous of degree 2 − n − |γ| in Rn . This is also true for |γ| = 0
provided n ≥ 3, i.e., the tempered distribution E is positive homogeneous of degree
2 − n in Rn if n ≥ 3.
Finally, corresponding to n = 2 and |γ| = 0, one may express
∫
ln |x| −1
E(x) = Φ(x) − L(ξ) dH 1 (ξ), ∀x ∈ R2 \ {0}, (1.4.22)
4π 2 S 1

18 for each m ∈ N, we let Δ m

x denote the m-fold application of the Laplacian in the variable x
64 1 Integral Representations and Integral Identities

where Φ : R2 \ {0} → C M×M , given by

∫  x 
1   −1
Φ(x) := − 2 ln  , ξ  L(ξ) dH 1 (ξ), ∀x ∈ R2 \ {0}, (1.4.23)
4π |x|
S1

is a function of class 𝒞∞ and positive homogeneous of degree 0 in R2 \ {0}.

(4) For each γ ∈ N0n there exists a finite constant Cγ > 0 such that for each
x ∈ Rn \ {0} one has
⎧
⎪ Cγ
⎪
⎪ if either n ≥ 3, or n = 2 and |γ| > 0,
⎪
⎪ |x| n+ |γ |−2
 γ  ⎪
⎨
⎪ ∫ −1
(∂ E)(x) ≤ or n = 2 and S 1 L(ξ) dH 1 (ξ) = 0, (1.4.24)
⎪
⎪
⎪
⎪
⎪
⎪   
⎪ C0 1 +  ln |x|  whenever n = 2 and |γ| = 0.
⎩
(5) For each indexes α, β ∈ {1, . . . , M } one has19
∫
αγ
ξr ar s (∂s Eγβ )(ξ) dH n−1 (ξ)
S n−1
∫
γα
= ξr asr (∂s Eβγ )(ξ) dH n−1 (ξ) = δαβ . (1.4.25)
S n−1

In fact, given any vector ω ∈ S n−1 , for each α, β ∈ {1, . . . , M } one has
∫
αγ
ξr ar s (∂s Eγβ )(ξ) dH n−1 (ξ)
ξ ∈S n−1
ξ,ω>0
∫
γα 1
= ξr asr (∂s Eβγ )(ξ) dH n−1 (ξ) = δαβ . (1.4.26)
2
ξ ∈S n−1
ξ,ω>0

(6) Let ‘hat’ denote the Fourier transform in Rn (originally defined on Schwartz
functions as in (1.4.14), then extended to tempered distributions via duality). Then
E/ is a tempered distribution in Rn (which is positive homogeneous of degree −2 if
n ≥ 3), whose restriction to Rn \ {0} is a (matrix-valued) function of class 𝒞∞ . In
fact,

19 The format of the identities in (1.4.25) strongly depends on the choice of a coefficient tensor to
represent the system L as in (1.3.1). Later on, in Definition 1.7.1, we shall recognize the integrands
in (1.4.25) as being conormal derivatives of columns and rows of E on the surface S n−1 . Bearing
this in mind, it follows from Proposition 1.7.4 that any two coefficient tensors used to write the
system L as in (1.3.1) produce conormal derivatives which differ by a linear combination of
tangential derivatives. Granted this, the independence of (1.4.25) on the choice of the coefficient
tensor used to represent the original system L is then implied by the integration by parts on the
boundary formula from [113, (11.1.7)].
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 65
−1
/ = L(ξ)
E(ξ) for every ξ ∈ Rn \ {0}, (1.4.27)
and
−1
/ = L(·)
if n ≥ 3 then E as (tempered) distributions in Rn, (1.4.28)

which also implies

0 −1
if n ≥ 3 then E = (2π)−n L(·) as (tempered) distributions in Rn . (1.4.29)

More generally, given any γ ∈ N0n it follows that the tempered distribution ∂1
γ E is
∞
a function of class 𝒞 when restricted to R \ {0} which, regarded as such, satisfies
n

−1
∂1
γ E(ξ) = i |γ | ξ γ L(ξ) for every ξ ∈ Rn \ {0}, (1.4.30)

and
if γ ∈ N0n then ∂1γ E = i |γ | ξ γ L(ξ) −1 as tempered
(1.4.31)
distributions in R when either |γ| > 0 or n ≥ 3.
n

(7) Writing E L in place of E to emphasize the dependence on L, the fundamental

solution E L defined in (1.4.18) satisfies
     ∗
EL = EL , E L = E L, EL = EL∗ ,
(1.4.32)
as well as EλL = λ−1 E L for each λ ∈ C \ {0}.

where , ·, ∗ denote, respectively, transposition, complex conjugation, and complex

(or Hermitian) adjunction. In concert with (1.4.21), the first formula in (1.4.32) also
implies that for each indexes α, β ∈ {1, . . . , M } one has

γα 0 if α  β,
ar s ∂xr ∂xs Eβγ (x − y) = δαβ δy (x) = x, y ∈ Rn . (1.4.33)
δy (x) if α = β,

A more general version of the last formula in (1.4.32), which is visible from
(1.4.18), is that for each invertible matrix C ∈ C M×M one has20

E LC = C −1 E L and EC L = E L C −1 . (1.4.34)

Finally, pick an arbitrary unitary transformation R in Rn (i.e., a matrix R ∈ Rn×n

with the property that R R = In×n ), and consider the second-order homogeneous
complex constant coefficient M × M system in Rn formally written as
 
αβ
L ◦ R := ar s (R∇)r (R∇)s . (1.4.35)
1≤α,β ≤M

Then the characteristic matrix of L ◦ R satisfies

20 with the systems LC and C L interpreted in the sense of multiplication of M × M matrices

66 1 Integral Representations and Integral Identities

(L ◦ R)(ξ) = L(Rξ) for each ξ ∈ Rn, (1.4.36)

so the system L ◦ R is weakly elliptic, and from (1.4.18), the invariance of integration
on S n−1 under unitary transformations (cf., e.g., [109, (14.9.11), p. 578]), and the
invariance of the action of the Laplacian under unitary transformations (cf., e.g.,
[109, Exercise 7.77, p. 319]) one obtains

E L◦R = E L ◦ R in Rn \ {0}. (1.4.37)

(8) Any M × M matrix with tempered distribution entries E ' which is a fundamental
'
solution of the system L in R is of the form E = E + Q where E is as in (1.4.18) and
n

Q is an M × M matrix whose entries are polynomials in Rn and whose columns, Q k

with k ∈ {1, . . . , M }, satisfy the pointwise equations LQ k = 0 ∈ C M in Rn for each
k ∈ {1, . . . , M }.

We wish to make five comments in relation to Theorem 1.4.2. Our first comment
concerns the proof of the integral identities recorded in item (5), since these are not
explicitly stated in [109, Theorem 11.1, pp. 393-395]. Specifically, having fixed two
arbitrary indexes α, β ∈ {1, . . . , M }, consider the vector field
 αγ  1     n
F := ar s ∂s Eγβ 1≤r ≤n ∈ Lloc B(0, 1), L n ∩ 𝒞∞ B(0, 1) \ {0} , (1.4.38)

and notice that, thanks to (1.4.21), we have

αγ  
divF = ar s ∂r ∂s Eγβ = δαβ δ0 ∈ ℰ B(0, 1) . (1.4.39)

Consequently, the Divergence Formula in the version stated in [112, (1.4.17)] applies
and presently gives
∫ ∫
αγ  dH n−1 (ξ)
ξr ar s (∂s Eγβ )(ξ) dH (ξ) =
n−1
ξ · F(ξ)
S n−1 S n−1
 
 1 ℰ(B(0,1))
= ℰ (B(0,1)) divF,
 
= ℰ (B(0,1)) δαβ δ0, 1 ℰ(B(0,1))

= δαβ . (1.4.40)

Likewise, since the vector field

 γα  1    
G := asr ∂s Eβγ 1≤r ≤n ∈ Lloc B(0, 1), L n ∩ 𝒞∞ B(0, 1) \ {0}
n
(1.4.41)

satisfies (cf. (1.4.33))

γα  
divG = asr ∂r ∂s Eβγ = δαβ δ0 ∈ ℰ B(0, 1) , (1.4.42)

we may once again invoke the Divergence Formula in the version stated in [112,
(1.4.17)] to conclude that
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 67
∫ ∫
γα  dH n−1 (ξ)
ξr asr (∂s Eβγ )(ξ) dH n−1 (ξ) = ξ · G(ξ)
S n−1 S n−1
 
 1 ℰ(B(0,1))
= ℰ (B(0,1)) divG,
 
= ℰ (B(0,1)) δαβ δ0, 1 ℰ(B(0,1))

= δαβ . (1.4.43)

Together, (1.4.40) and (1.4.43) prove (1.4.25). Once this has been established, the
formulas claimed in (1.4.26) become consequences of (1.4.25) taking into account
the fact that, as seen from (1.4.19), the integrands are even functions.
Our second comment is that for each ε > 0 we have
2 .    M×M
E∈ 𝒞α Rn \ B(0, ε) . (1.4.44)
0<α<1

Indeed, when n ≥ 3, this is a consequence of the fact that, as seen from (1.4.24),
the matrix-valued function E is both bounded and Lipschitz in Rn \ B(0, ε). When
n = 2, the function Φ continues to be both bounded and Lipschitz in Rn \ B(0, ε)
(since it is smooth and positive homogeneous of degree 0 in R2 \ {0}), but this is not
the case for the logarithmic term in (1.4.22). This being said, for each given pair of
points x, y ∈ Rn \ B(0, ε) with |x| > |y| we have
      
 ln |x| − ln |y|  =  ln |x|  = ln 1 + |x| − |y|
|y| |y|
 |x| − |y|  α
≤ Cα ≤ Cα ε −α |x − y| α . (1.4.45)
|y|
Above, the first inequality is a consequence of the elementary estimate

ln(1 + t) ≤ Cα t α for each t ∈ (0, ∞) and α ∈ (0, 1), (1.4.46)

and the final inequality in (1.4.45) is implied by the triangle inequality. Now, (1.4.44)
readily follows from (1.4.45) and (1.4.22).
Our third comment is that
if n = 2 then, for any given vector c ∈ C M , the limit
lim E(x)c exists (in C M ) if and only if c = 0 ∈ C M , (1.4.47)
x→∞

and
if n = 2 then, for any given vector c ∈ C M , having E(x)c = o(ln |x|)
as |x| → ∞ is equivalent to having
 ∫ E(x)c = O(1) as |x| → ∞, which (1.4.48)
−1
is further equivalent to having S 1 L(ξ) dH 1 (ξ) c = 0.
68 1 Integral Representations and Integral Identities

To justify (1.4.47), assume n = 2 and c ∈ C M is such that a := lim E(x)c exists.

x→∞
Then (1.4.22) implies that
 ∫ &
ln |x|  −1

a = lim Φ(x)c − L(ξ) dH (ξ) c
1
x→∞ 4π 2 S1
 ∫ &
Φ(x)c 1  −1

= lim ln |x| − 2 L(ξ) dH (ξ) c .
1
(1.4.49)
x→∞ ln |x| 4π S1

Φ(x)c
Since lim = 0 given that the matrix-valued function Φ is continuous and
x→∞ ln |x |
positive homogeneous of degree 0 in R2 \ {0} (hence bounded), we see that (1.4.49)
forces ∫ 
−1
L(ξ) dH 1 (ξ) c = 0. (1.4.50)
S1
Re-writing the first equality in (1.4.49) in light of this piece of information leads to
the conclusion that  
a = lim Φ(x)c . (1.4.51)
x→∞

In particular, for each fixed x ∈ R2

\ {0} we have, using again the fact that Φ is
positive homogeneous of degree 0 in R2 \ {0},
 
Φ(x)c = lim Φ(t x)c = a. (1.4.52)
t→∞

This goes to show that Φ(x)c = a for each x ∈ R2 \ {0}, hence E(x)c = a for each
x ∈ R2 \ {0}, thanks to this, (1.4.22), and (1.4.50). In particular, E(·)c regarded as
a locally integrable vector-valued function in R2 is actually constant. Applying the
system L and recalling (1.4.20) then yields δc = 0 in the sense of distributions,
which ultimately shows that c = 0 ∈ C M , as wanted. This establishes the direct
implication in (1.4.47), and the opposite implication is trivial. The justification of
(1.4.48) uses similar ideas, the crux of the matter being the sequence of implications
(regarding the behavior as |x| → ∞)
∫ 
−1
E(x)c = o(ln |x|) =⇒ L(ξ) dH 1 (ξ) c = 0
S1

=⇒ E(x)c = Φ(x)c = O(1). (1.4.53)

Our fourth comment is that, in the two-dimensional setting, it is clear from (1.4.22)
that
E is positive homogeneous of degree 0 in R2 \ {0}
∫ (1.4.54)
−1
if and only if L(ξ) dH 1 (ξ) = 0.
S1
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 69

Our fifth (and final) comment is that a suitable version of (1.4.29) continues to
be true in the case when n = 2. This requires extending the matrix-valued function
M×M
[L(·)]−1 ∈ 𝒞∞ (R2 \ {0}) , which is homogeneous of degree −2, to a matrix-
valued tempered distribution in R2 , through the consideration of its finite part. See,
e.g., [103, Theorem 6.8, p. 198] for details in the scalar case.
Moving on, we wish to further elaborate on the manner in which a linear change
of variables in the Euclidean space affects a given weakly elliptic system and its
fundamental solution constructed as in Theorem 1.4.2.

 αβn, M ∈ N, with n ≥ 2, have been fixed.

Proposition 1.4.3 Assume that two integers
Given a (complex) coefficient tensor A = a jk 1≤ j,k ≤n , associate the second-order,
1≤α,β ≤M
homogeneous, complex constant coefficient, M × M system in Rn
 αβ 
L = L A := ar s ∂r ∂s 1≤α,β ≤M . (1.4.55)

Also, for any matrix W = (w jk )1≤ j,k ≤n ∈ Cn×n define two new coefficient tensors,
namely  αβ 
W ◦ A := w j a k 1≤ j,k ≤n (1.4.56)
1≤α,β ≤M

and  αβ 
A ◦ W := a j w k 1≤ j,k ≤n . (1.4.57)
1≤α,β ≤M

In addition, generalize (1.4.35) by defining the second-order, homogeneous, complex

constant coefficient, M × M system in Rn , formally written as
 
αβ
L ◦ W := a jk (W∇) j (W∇)k . (1.4.58)
1≤α,β ≤M

Then, in relation to these, the following statements are true:

(i) For any matrix W ∈ Rn×n one has

(L A) ◦ W = LW  ◦A◦W (1.4.59)

and
LW  ◦A◦W (ξ) = L A(Wξ) for all ξ ∈ Rn . (1.4.60)
In particular,

if the real n × n matrix W is invertible then

(1.4.61)
LW  ◦A◦W is weakly elliptic if and only if L A is weakly elliptic.

Also,

if the real n × n matrix W is invertible then

LW  ◦A◦W is Legendre-Hadamard (strongly) elliptic (1.4.62)
if and only if L A is Legendre-Hadamard (strongly) elliptic.
70 1 Integral Representations and Integral Identities

(ii) For any invertible matrix W ∈ Rn×n one has

  
L A ◦ (W −1 ) u ◦ W) = LW −1 ◦A◦(W −1 ) u ◦ W) = (L Au) ◦ W
M (1.4.63)
for any vector distribution u ∈ D (Rn ) ,

where the composition of a distribution with an invertible real matrix is under-

stood in the usual sense21 (cf., e.g., [109, Proposition 4.33, p. 140]).
(iii) If L A is weakly elliptic then for any invertible matrix W ∈ Rn×n one has

E L◦W = |det W | −1 E L ◦ (W −1 ) if n ≥ 3,
(1.4.64)
as (matrix) tempered distributions,

and
the (matrix) tempered distributions
E L◦W and |det W | −1 E L ◦ (W −1 ) (1.4.65)
differ by a constant (in C M×M ) if n = 2.

Proof To justify (1.4.59), use (1.4.56)-(1.4.58) to write

 
αβ
(L A) ◦ W = a jk (W∇) j (W∇)k
1≤α,β ≤M
 
αβ
= a jk w jr ∂r wks ∂s = LW  ◦A◦W , (1.4.66)
1≤α,β ≤M

as wanted. Next, for each ξ = (ξ1, . . . , ξn ) ∈ Rn we have

 
αβ
LW  ◦A◦W (ξ) = − a jk w jr ξr wks ξs
1≤α,β ≤M
 
αβ
= − a jk (Wξ) j (Wξ)k = L A(Wξ), (1.4.67)
1≤α,β ≤M

proving (1.4.60). The claim in (1.4.61) is then a consequence of (1.4.60) and Defini-
tion 1.3.1. Also, if L A is Legendre-Hadamard (strongly) elliptic (cf. Definition 1.3.2)
then for all ξ ∈ Rn and ζ ∈ C M we may write (bearing in mind that W is invertible)
   
Re − LW  ◦A◦W (ξ)ζ , ζ = Re − L A(Wξ)ζ , ζ

≥ c |Wξ | 2 |ζ | 2 ≥ '
c |ξ | 2 |ζ | 2 (1.4.68)

for some c, '

c ∈ (0, ∞) independent of ξ and ζ. Ultimately, this establishes (1.4.62).
Going further, the first equality in (1.4.63) is seen directly from (1.4.59). To show
M
the second equality in (1.4.63), assume first that u ∈ 𝒞∞ (Rn ) . Then

21 i.e., u ◦ W, ϕ  := |det W | −1 u, ϕ ◦ W −1  for each test function ϕ in R n

1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 71
     
αβ
LW −1 ◦A◦(W −1 ) u ◦ W) = a jk (W −1 )r j (W −1 ) ks ∂r ∂s uβ ◦ W
1≤α ≤M
   
αβ
= a jk (W −1 )r j (W −1 )sk (∂ ∂m uβ ) ◦ W w r wms
1≤α ≤M
  
αβ
= a jk δ j δmk (∂ ∂m uβ ) ◦ W
1≤α ≤M
 
αβ
= a jk ∂j ∂k uβ ◦ W = (L Au) ◦ W, (1.4.69)
1≤α ≤M

by Chain Rule and definitions. Having proved (1.4.69), the second equality in (1.4.63)
follows from the sequential density of 𝒞∞ c (R ) in D (R ) (cf., e.g., [109, The-
n n

orem 2.106, p. 75]) and the continuity of partial differentiation on the space of
distributions.
Consider now the claims made in item (iii). In all dimensions n ≥ 2 we may
compute, in the sense of distributions in Rn ,
   
(L ◦ W) E L ◦ (W −1 ) = LW  ◦A◦W E L ◦ (W −1 ) = (LE L ) ◦ (W −1 )

= (δ ◦ (W −1 ) ) I M×M = |det W |δ I M×M , (1.4.70)

with the first equality coming from (1.4.59), the second equality implied by (1.4.63),
the third equality a consequence of (1.4.20), and the last equality seen from [109,
Proposition 4.33, p. 140]. This shows that in all dimensions n ≥ 2 the tempered
distribution |det W | −1 E L ◦ (W −1 ) is a fundamental solution for the weakly elliptic
operator L ◦ W. Based on this and items (8) and (4) in Theorem 1.4.2, we then
conclude that the claims made in (1.4.64)-(1.4.65) are true. 
It is also of interest to specialize Theorem 1.4.2 to the case of the complex Lamé
system Lλ,μ in the regime of Lamé molduli λ, μ, ensuring its weak ellipticity.

Proposition 1.4.4 Let Lλ,μ be the complex Lamé system (1.3.6) associated to two
Lamé moduli μ, λ ∈ C satisfying

μ  0 and λ + 2μ  0 (1.4.71)

(thus ensuring the weak ellipticity of Lλ,μ ; cf. (1.3.9)). Then the fundamental solution
E of Lλ,μ from (1.4.18) takes the explicit form E = (E jk )1≤ j,k ≤n , a matrix whose
( j, k) entry is defined at each point x = (x1, . . . , xn ) ∈ Rn \ {0} according to22

22 at least in lower dimensions and for real Lamé moduli, this goes back to the work of Lord Kelvin
(aka William Thomson)
72 1 Integral Representations and Integral Identities
3 4
⎧
⎪ −1 (3μ + λ)δ jk x j xk
⎪
⎪ + (μ + λ) n if n ≥ 3,
⎪
⎨ 2μ(2μ + λ)ωn−1 (n − 2)|x|
⎪ |x|
n−2
⎪
E jk (x) =
⎪
⎪ 3 4
⎪
⎪ 1 x j xk
⎪
⎪ 4π μ(2μ + λ) (3μ + λ)δ jk ln |x| − (μ + λ) + cμ,λ δ jk if n = 2,
⎩ |x| 2
(1.4.72)
for every j, k ∈ {1, . . . , n}, where cμ,λ ∈ C is the constant given by

(1 + ln 4)(μ + λ) ln 2 (1 + ln 4)(3μ + λ) 1
cμ,λ := − =− + . (1.4.73)
8π μ(2μ + λ) 2π μ 8π μ(2μ + λ) 4π μ
Proof Assume first that n ≥ 3. Then the n × n matrix whose entries are as in the top
line of (1.4.72) is known to be a matrix-valued tempered distribution which is a fun-
damental solution for the Lamé system (1.3.6) in Rn (see, e.g., [108, Theorem 10.15
on p. 323, Exercise 10.32 on p. 339], [87, (9.2)], [86, Chapter 9]). Denote by Q the
difference between this fundamental solution and what E from (1.4.18) becomes
when L := Lλ,μ . Then the uniqueness result in item (8) of Theorem 1.4.2 ensures
that Q is a polynomial in Rn . In view of (1.4.24) (with γ = 0), the fact that we are
presently assuming n ≥ 3, and the first line in (1.4.72), said polynomial decays at
infinity, hence actually Q = 0. The desired conclusion is therefore established when
n ≥ 3.
To treat the two-dimensional case, fix λ, μ ∈ C such that μ  0 and λ + 2μ  0.
Corresponding to n = 2, in light of (1.3.10) formula (1.4.18) becomes
∫ &
1  
E(x) = 2 ln  x, ξ  dH 1 (ξ) I2×2
4π μ S 1
∫
λ+μ   
− 2 ln  x, ξ  ξ ⊗ ξ dH 1 (ξ), (1.4.74)
4π μ(λ + 2μ) S1

for each x ∈ R2 \ {0}. Using the rotation invariance of integrals over the unit sphere,
for each fixed x ∈ R2 \ {0} we may compute
∫ ∫
   
ln  x, ξ  dH 1 (ξ) = ln  |x|e1, ξ  dH 1 (ξ)
S1 S1
∫
= 2π ln |x| + ln |ξ1 | dH 1 (ξ)
S1
∫ 2π
= 2π ln |x| + ln | cos θ| dθ
0

= 2π ln |x| − 2π ln 2, (1.4.75)

where the last equality uses the fact that

1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 73
∫ 2π
ln | cos θ| dθ = −2π ln 2. (1.4.76)
0

We shall establish this identity later, in the very last portion of the proof, where we
will also show that for each fixed x ∈ R2 \ {0} we have
∫  π 
   x⊗x
ln  x, ξ  ξ ⊗ ξ dH 1 (ξ) = − − π ln 2 + π ln |x| I2×2 + π . (1.4.77)
S1 2 |x| 2
Granted these, it is apparent from (1.4.75)-(1.4.77) that formula (1.4.74) reduces
precisely to the n = 2 version of (1.4.72) with cμ,λ ∈ C as in (1.4.73).
At this stage, there remains to prove (1.4.76) and (1.4.77). As regards the first
identity, we note that breaking up the interval of integration at π, then making a
natural change of variables in the integral over [π, 2π] yields
∫ ∫ π ∫ π
2π  
ln | cos θ| dθ = 2 ln | cos θ| dθ = ln cos2 θ dθ. (1.4.78)
0 0 0

To proceed, we compute
∫ π ∫ π/2 ∫ π
     
ln sin2 θ dθ = ln sin2 θ dθ + ln sin2 θ dθ
0 0 π/2
∫ π/2 ∫ π/2
   
= ln sin2 θ dθ + ln cos2 θ dθ
0 0
∫ 
π/2 2
= ln 2−1 sin(2θ) dθ
0
∫ π/2  
= −π ln 2 + ln sin2 (2θ) dθ
0
∫ π
1  
= −π ln 2 + ln sin2 θ dθ, (1.4.79)
2 0

which implies
∫ π  
ln sin2 θ dθ = −2π ln 2. (1.4.80)
0

A similar computation further gives

74 1 Integral Representations and Integral Identities
∫ π ∫ π/2 ∫ π/2
     
ln cos2 θ dθ = ln cos2 θ dθ + ln sin2 θ dθ
0 0 0
∫ 
π/2 2
= ln 2−1 sin(2θ) dθ
0
∫ π/2  
= −π ln 2 + ln sin2 (2θ) dθ
0
∫ π
1  
= −π ln 2 + ln sin2 θ dθ = −2π ln 2, (1.4.81)
2 0

where the last step uses (1.4.80). Then (1.4.76) follows from (1.4.78) and (1.4.81).
Turning our attention to (1.4.77), we first note that a variant of [108, Proposi-
tion 13.47, p. 440] gives that if x ∈ Rn \ {0} is fixed, then for each j, k ∈ {1, . . . , n}
one has
∫
  x j xk ωn−1
ln |x, ξ| ξ j ξk dH n−1 (ξ) = an δ jk + bn 2 + ln |x|δ jk , (1.4.82)
S n−1 |x| n

where ∫
1  
an := ln |ξ1 | (1 − ξ12 ) dH n−1 (ξ),
n − 1 S n−1
∫ (1.4.83)
1  
bn := ln |ξ1 | (nξ12 − 1) dH n−1 (ξ).
n − 1 S n−1
Specializing this to the case n = 2 gives that for each x ∈ R2 \ {0} we have
∫  
   x⊗x
ln  x, ξ  ξ ⊗ ξ dH 1 (ξ) = a2 + π ln |x| I2×2 + b2 , (1.4.84)
S1 |x| 2
and we claim that
π
a2 = − − π ln 2 and b2 = π. (1.4.85)
2
In this regard, we first note that
∫ ∫ π
1 2π    
a2 = (sin θ) ln cos θ dθ =
2 2
(sin2 θ) ln cos2 θ dθ, (1.4.86)
2 0 0

and we find it useful to also consider

∫ π
 
c2 := (cos2 θ) ln cos2 θ dθ. (1.4.87)
0

Then an integration by parts permits us to compute

1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 75
∫ π  
c2 = (sin θ) (cos θ) ln cos2 θ dθ
0
∫
   π
 π
= (sin θ)(cos θ) ln cos θ  + a2 + 2 2
sin2 θ dθ, (1.4.88)
0 0

hence c2 = a2 + π. Since, by (1.4.86), (1.4.87), and (1.4.81) we also have,

∫ π
 
a2 + c2 = ln cos2 θ dθ = −2π ln 2, (1.4.89)
0

we ultimately conclude that

π π
a2 = − − π ln 2 and c2 = − π ln 2. (1.4.90)
2 2
Consequently, from (1.4.83) with n = 2, (1.4.87), (1.4.90), and (1.4.81), we deduce
that
∫ ∫ π
1 2π  2   
b2 = ln cos θ (2 cos θ − 1) dθ =
2
ln cos2 θ (2 cos2 θ − 1) dθ
2 0 0
∫ π
 
= 2c2 − ln cos2 θ dθ = (π − 2π ln 2) + 2π ln 2 = π. (1.4.91)
0

Together with (1.4.90), this finishes the proof of (1.4.85).

Having established this, formula (1.4.77) follows by combining (1.4.84) with
(1.4.85). The proof of Proposition 1.4.4 is therefore complete. 
Here is what Theorem 1.4.2 yields when specialized to the case of the operator
Def  Def.
Corollary 1.4.5 Fix n ∈ N with n ≥ 2 are recall the deformation tensor Def in Rn
defined in (1.3.49). Then the fundamental solution associated as in Theorem 1.4.2
with the homogeneous, constant coefficient, weakly elliptic, second-order, n × n
system L := Def  Def in Rn is the matrix-valued function E = (E jk )1≤ j,k ≤n with
entries defined at each point x = (x1, . . . , xn ) ∈ Rn \ {0} as
3 4
⎧
⎪ 1 3δ jk x j xk
⎪
⎪ + if n ≥ 3,
⎪ |x| n
⎨ 2ωn−1 (n − 2)|x|
⎪ n−2
⎪
E jk (x) := (1.4.92)
⎪
⎪ 3 4
⎪
⎪ 1 x j x k 3 ln 4 − 1
⎪
⎪ −3δ jk ln |x| + + δ jk if n = 2,
⎩ 4π |x| 2 8π

for every j, k ∈ {1, . . . , n}.

Proof Since from (1.3.58) we know that, when
 acting
 on vector-valued distributions
in Rn , the operator L = Def  Def = − 12 Δ + ∇div agrees with the Lamé system
Lμ,λ for μ := −1/2 and λ := 0, formula (1.4.92) is implied by (1.4.72) specialized
to these values of the Lamé moduli (which satisfy (1.4.71)). 
76 1 Integral Representations and Integral Identities

Our next result explicitly identifies the fundamental solution produced by Theo-
rem 1.4.2 for the family of “generalized Lamé” operators Lα,β introduced earlier, in
Example 1.3.16. See also the discussion in (1.3.81), according to which the afore-
mentioned result contains Proposition 1.4.4 as a particular case. To state it, the reader
is reminded that the standard fundamental solution for the Laplacian Δ in Rn has been
defined in (1.5.56). Also, the standard fundamental solution for the bi-Laplacian Δ2
in Rn (with N satisfying n ≥ 2) is given by

⎧
⎪ 1 1
⎪
⎪ if n ≥ 3 and n  4,
⎪
⎪ 2(n − 2)(n − 4)ωn−1 |x| n−4
⎪
⎪
⎪
⎨ 1
⎪
EΔ2 (x) := ln |x| if n = 4, (1.4.93)
⎪
⎪ 8π 2
⎪
⎪
⎪
⎪
⎪ 1
⎪
⎪
⎩ 8π |x| ln |x| if n = 2,
2

for each x ∈ Rn \ {0} (see, e.g., [109, Theorem 7.21, pp. 251–252]).

Proposition 1.4.6 Fix n ∈ N with n ≥ 2. Select two complex numbers α, β ∈ C \ {0}

along with a degree  ∈ {0, 1, . . . , n} and consider the second-order homogeneous,
constant (complex) coefficient weakly elliptic23 differential operator

Lα,β := αdδ + βδd (1.4.94)

acting on differential forms of degree  in Rn . Then, if EΔ and EΔ2 are as in (1.5.56)

and (1.4.93), respectively, it follows that the fundamental solution of Lα,β from
Theorem 1.4.2 is the double form24 defined at each z ∈ Rn \ {0} according to25


Eα,β (z) = − α−1 EΔ (z)δI J (1.4.95)
|I |= |J |=
&

n
− (β−1 − α−1 ) ∂zi ∂z j EΔ2 (z) ε iI
jJ + cI J d x ⊗ dy
I J

i, j=1

where cI J is a complex constant if n = 2, and actually cI J = 0 if n ≥ 3. More

explicitly, if n ≥ 3 then for each z ∈ Rn \ {0} one has

23 see (1.3.71)
24 a double form is the algebraic equivalent of a matrix-valued function in the present context
.
25 as in [112, §6.4], the symbol indicates that the sum is performed only over strictly increasing
multi-indices
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 77

  n− 1 1
Eα,β (z) = − α−1 + (β−1 − α−1 ) δI J
2 (2 − n)ωn−1 |z| n−2
|I |= |J |=
&

n
1 zi z j iI
−1 −1
− (β −α ) ε dx I ⊗ dy J (1.4.96)
i, j=1
2ω n−1 |z| n jJ

whereas if n = 2 then corresponding to  = 0 one has

ln |z| −1 ln 4 −1
0
Eα,β (z) = − β + β for each z ∈ R2 \ {0}, (1.4.97)
2π 4π
corresponding to  = 1 one has

 2
1 −1
Eα,β (z) =
1
− (α + β−1 )δi j ln |z|
i, j=1
4π
&
1 −1 −1 zi z j
− (α − β ) 2 + ci j dxi ⊗ dy j
1
(1.4.98)
4π |z|

for each z ∈ R2 \ {0} with

1 + ln 4 −1 1 −1
ci1j := (α + β−1 )δi j − β δi j , (1.4.99)
8π 4π
and corresponding to  = 2 one has
 ln |z| ln 4 −1 
2
Eα,β (z) = − α−1 + α (dx1 ∧ dx2 ) ⊗ (dy1 ∧ dy2 ) (1.4.100)
2π 4π
for each z ∈ R2 \ {0}.

Proof By virtue of (1.3.74) it follows that the double form

 
' := Lα−1,β −1
E EΔ2 dx I ⊗ dy I (1.4.101)
α,β
|I |=

is a fundamental solution for the operator Lα,β acting on differential forms of degree
 in Rn . Use the last formula in (1.3.73) to express

Lα−1,β −1 = α−1 dδ + β−1 δd = α−1 dδ + δd) + (β−1 − α−1 )δd

= −α−1 Δ + (β−1 − α−1 )δd. (1.4.102)

Keeping in mind that ΔEΔ2 = EΔ , we first see that

    
−α−1 Δ EΔ2 dx I ⊗ dy I = −α−1 EΔ δI J dx I ⊗ dy J . (1.4.103)
|I |= |I |= |J |=
78 1 Integral Representations and Integral Identities

Second, based on definitions (cf. [112, (6.6.141), (6.4.142)]) and item (ii) in [112,
Lemma 6.4.6] we compute
 
    n 
jI
δd EΔ2 dx ⊗ dy = δ
I I
∂j EΔ2 εK dx ⊗ dy
K I

|I |= |K |= +1 j=1 |I |=

 
n    
jI K
=− ∂i ∂j EΔ2 εK εiJ dx J ⊗ dy I
|J |= i, j=1 |I |= |K |= +1

  
n
jI
=− ∂i ∂j EΔ2 εiJ dx J ⊗ dy I
|I |= |J |= i, j=1

 
n
=− ∂i ∂j EΔ2 ε iI
jJ dx ⊗ dy ,
I J
(1.4.104)
|I |= |J |= i, j=1

where in the last equality we have simply intertwined I and J. We can then deduce
from (1.4.101) and (1.4.103)-(1.4.104) that the double form defined at each point
z ∈ Rn \ {0} by


E' (z) = − α−1 EΔ (z)δI J (1.4.105)
α,β
|I |= |J |=
&

n
−1 −1
− (β −α ) ∂zi ∂z j EΔ2 (z) ε iI
jJ dx I ⊗ dy J
i, j=1

is a fundamental solution for the operator Lα,β acting on -forms in Rn with n ≥ 2.

Using (1.5.56) and (1.4.93), we may write E ' explicitly. Concretely, if n ≥ 3 then
α,β
for each z ∈ R \ {0} we have
n


  n − 1 1
' (z) =
E − α−1 + (β−1 − α−1 ) δI J (1.4.106)
α,β
2 (2 − n)ωn−1 |z| n−2
|I |= |J |=
&
 n
1 zi z j iI
−1 −1
− (β − α ) ε dx I ⊗ dy J
i, j=1
2ω n−1 |z| n jJ

while if n = 2 then for each z ∈ R2 \ {0} we have

1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 79


  2− 1
' (z) =
E − α−1 + (β−1 − α−1 ) ln |z|δI J (1.4.107)
α,β
2 2π
|I |= |J |=
&
1  2
z i z j
− (β−1 − α−1 ) ε iI + '
cI J dx I ⊗ dy J
4π i, j=1 |z| 2 jJ

where

1 −1  2
2 −  −1
'
cI J = − (β − α−1 ) δi j ε iI
jJ = − (β − α−1 )δI J . (1.4.108)
8π i, j=1
8π

Next, if E Lα, β denotes the fundamental solution of Lα,β acting on -forms from
Theorem 1.4.2, then the uniqueness result from item (8) in Theorem 1.4.2 guarantees
the existence of a polynomial Q in Rn with the property that
'α,β + Q.
E Lα, β = E (1.4.109)

From this, (1.4.24), and (1.4.106)-(1.4.108) we then deduce that Q is constant if

n = 2, and vanishes identically if n ≥ 3. Having established that, the claims pertaining
to (1.4.95) and (1.4.96) follow.
There remains to identify E Lα, β in the manner specified in (1.4.97)-(1.4.100),
in the two-dimensional setting. To set the stage, we first make a couple of general
observations. From definitions we see that

Lα,β (ξ) = αξ ∧ (ξ ∨ ·) + βξ ∨ (ξ ∧ ·) for all ξ ∈ Rn . (1.4.110)

In concert with (1.3.75), this shows that for each ξ ∈ Rn \ {0} we have
−1
Lα,β (ξ) = (αβ)−1 |ξ | −4 Lβ,α (ξ) = |ξ | −4 Lα−1,β −1 (ξ)

= α−1 |ξ | −4 ξ ∧ (ξ ∨ ·) + β−1 |ξ | −4 ξ ∨ (ξ ∧ ·) (1.4.111)

which, for ξ = (ξ1, . . . , ξn ) ∈ S n−1 , reduces to

−1
Lα,β (ξ) = α−1 ξ ∧ (ξ ∨ ·) + β−1 ξ ∨ (ξ ∧ ·)

n  
= ξi ξ j α−1 ei ∧ (e j ∨ ·) + β−1 ei ∨ (e j ∧ ·) . (1.4.112)
i, j=1

Let us now work in the two-dimensional setting. If E Lα, β (x) is regarded as a linear
map on the space of differential forms of degree , then from (1.4.18) (used with
n = 2), (1.4.112), and (1.4.77) we deduce that
80 1 Integral Representations and Integral Identities
∫
1   −1
E Lα, β (z) = − ln  z, ξ  Lα,β (ξ) dH 1 (ξ) (1.4.113)
4π 2
S1

∫
1     
2
=− 2 ln  z, ξ  ξi ξ j dH 1 (ξ) ×
4π i, j=1
S1
 
× α−1 ei ∧ (e j ∨ ·) + β−1 ei ∨ (e j ∧ ·)

1 + ln 4 − 2 ln |z| 
2  
= δi j α−1 ei ∧ (e j ∨ ·) + β−1 ei ∨ (e j ∧ ·)
8π i, j=1

1  zi z j  −1 
2
−1
− α ei ∧ (e j ∨ ·) + β ei ∨ (e j ∧ ·)
4π i, j=1 |z| 2

for each z = (z1, z2 ) ∈ R2 \ {0}. Next, a computation based on [112, (6.4.126)]

and [112, Lemma 6.4.7] shows that, as linear maps on differential forms of degree
 ∈ {0, 1, 2} in R2 ,
⎧
⎪ 2β−1 Id if  = 0,
  ⎪
⎪ 
2 ⎨
⎪
δi j α−1 ei ∧ (e j ∨ ·) + β−1 ei ∨ (e j ∧ ·) = (α−1 + β−1 )Id if  = 1, (1.4.114)
⎪
⎪
i, j=1 ⎪
⎪ 2α−1 Id
⎩ if  = 2,

where Id is the identity. In a similar fashion, the linear map

 2
zi z j  −1 
α ei ∧ (e j ∨ ·) + β−1 ei ∨ (e j ∧ ·) (1.4.115)
i, j=1
|z| 2

acting on -forms may be canonically identified with

⎧
⎪ β−1 Id if  = 0,
⎪
⎪
⎨
⎪ z⊗z
⎪ (α−1 − β−1 ) 2 + β−1 Id if  = 1, (1.4.116)
⎪
⎪ |z|
⎪ α−1 Id if  = 2,
⎩
where Id is once again the identity, and the middle case entails a matrix regarded
as a linear map on 1-forms (themselves canonically identified with vectors). Finally,
from (1.4.113)-(1.4.116) we conclude that (1.4.97)-(1.4.100) holds. 

It has been noted earlier that Proposition 1.4.6 contains Proposition 1.4.4 as a
particular case. While this is a byproduct of (1.3.81), it is instructive to check this
directly. Specifically, fix two numbers α, β ∈ C \ {0} then define
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 81

λ := −α + 2β and μ := −β. (1.4.117)

As a consequence, α = −2μ − λ and β = −μ, so

3μ + λ μ+λ
α−1 + β−1 = − and α−1 − β−1 = . (1.4.118)
μ(2μ + λ) μ(2μ + λ)

Keeping these in mind and also observing that for any two indices I, J ∈ {1, . . . , n}
we have
 n
zi z j iI δI J z I zJ
ε jJ = n−2 − for all z ∈ Rn \ {0}, (1.4.119)
i, j=1
|z| n |z| |z| n

one can verify without difficulty that if n ≥ 3 then the fundamental solution Eα,β 1

defined as in (1.4.96) with  := 1 may be naturally identified with the matrix-valued

fundamental solution for the Lamé system Lμ,λ whose entries are defined as in
(1.4.72) with n ≥ 3.
In addition, it is apparent from (1.4.118) and definitions that if n = 2 then the
fundamental solution Eα,β 1 from (1.4.98)-(1.4.99) reduces precisely to the matrix-

valued fundamental solution for the Lamé system Lμ,λ whose entries are defined as
in (1.4.72) with n = 2.
Scalar Weakly Elliptic Operators
In the remaining portion of this section we consider in greater detail the scalar
case, with the goal of providing fully explicit formulas for the fundamental solutions
of weakly elliptic scalar operators (originally constructed in Theorem 1.4.2 via
integral expressions). Henceforth we take M = 1 in (1.3.1). In this scenario, we
agree to drop the dependence on M and simply write L = ar s ∂r ∂s and introduce the
n × n matrix of coefficients A := (ar s )1≤r,s ≤n ∈ Cn×n .
Let us re-visit the two notions of ellipticity, discussed earlier for general systems,
and re-adjust them as to better reflect the fact that we are now considering scalar
operators. First, the weak ellipticity condition for L from (1.3.3) may be equivalently
expressed in the current case as

Aξ, ξ  0 for every ξ ∈ Rn \ {0}, (1.4.120)

which may be further recast as

Aξ, ξ ∈ C \ {0} for every ξ ∈ S n−1, (1.4.121)

and also equivalently rephrased as

 
inf  Aξ, ξ  > 0. (1.4.122)
ξ ∈S n−1

Whenever either of these equivalent descriptions happens, we shall call A ∈ Cn×n a

weakly elliptic matrix.
82 1 Integral Representations and Integral Identities

Second, the Legendre-Hadamard ellipticity condition from Definition 1.3.2 for

the operator L = ar s ∂r ∂s becomes equivalent to the demand that the coefficient
matrix A = (ar s )1≤r,s ≤n has the property that there exists some κ > 0 such that

ReAξ, ξ = Re ar s ξr ξs ≥ κ|ξ | 2, ∀ξ = (ξ1, . . . , ξn ) ∈ Rn, (1.4.123)

which, in turn, may be rephrased as

 
Aξ, ξ ∈ z ∈ C : Re z ≥ κ for every ξ ∈ S n−1 . (1.4.124)

Whenever this happens, we shall say that the matrix A ∈ Cn×n is Legendre-Hadamard
elliptic. Another equivalent way of express the Legendre-Hadamard ellipticity of a
given complex matrix A = (ar s )1≤r,s ≤n is to simply demand that

inf Re Aξ, ξ = inf Re ar s ξr ξs > 0. (1.4.125)

ξ ∈S n−1 ξ ∈S n−1

As is apparent from definitions,

if A ∈ Cn×n is Legendre-Hadamard elliptic,

(1.4.126)
then the matrix A is also weakly elliptic.
On a related note, recall that the (strict) positive definiteness of a complex matrix
A = (ar s )1≤r,s ≤n amounts to the existence of a real number κ > 0 such that
 
Re Aζ, ζ  = Re ar s ζr ζs ≥ κ|ζ | 2, ∀ζ = (ζr )1≤r ≤n ∈ Cn . (1.4.127)

With A∗ := (Ac ) and In×n ∈ Cn×n denoting the identity matrix, this is further
equivalent to the demand that

A∗ ) ≥ κIn×n in Cn, i.e.,

2 (A +
1

1 ∗
 (1.4.128)
2 (A + A )ζ, ζ ≥ κ|ζ | , ∀ζ ∈ Cn .
2

Obviously,

if A ∈ Cn×n is (strictly) positive definite then the

(1.4.129)
matrix A is also Legendre-Hadamard elliptic,
and
if A ∈ Cn×n is (strictly) positive definite and
(1.4.130)
invertible, then so is the inverse matrix A−1 .
Complex matrices that are symmetric and Legendre-Hadamard elliptic have a
number of other useful properties. Among other things, they turn out to be invertible,
as indicated in our next lemma.

Lemma 1.4.7 Fix a matrix A ∈ Cn×n which is symmetric and Legendre-Hadamard

elliptic, i.e., satisfies
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 83

A = A and κ := inf Re Aξ, ξ > 0. (1.4.131)

ξ ∈S n−1

Then both A and Re A are positive definite. In fact,

Re Aζ, ζ  ≥ κ|ζ | 2, ∀ζ ∈ Cn, (1.4.132)

and  
(Re A)ζ, ζ ≥ κ|ζ | 2, ∀ζ ∈ Cn . (1.4.133)
Also,
| Aζ | ≥ κ|ζ | and |(Re A)ζ | ≥ κ|ζ | for every ζ ∈ Cn . (1.4.134)
Finally,
det A  0 and det (Re A)  0, (1.4.135)
hence both A and Re A are invertible.

Proof For each ζ ∈ Cn we may write

 
Re Aζ, ζ  = Re A(Re ζ + iIm ζ), Re ζ − iIm ζ
   
= Re A(Re ζ), Re ζ + A(Im ζ), Im ζ

≥ κ |Re ζ | 2 + κ |Im ζ | 2 = κ |ζ | 2 . (1.4.136)

The second equality in (1.4.136) uses the fact that A is symmetric, while the second
property in (1.4.131) is used for the inequality in (1.4.136). This establishes (1.4.132).
Combining (1.4.132) with (1.4.127) then shows that A is positive definite. The
symmetry of A implies Re A = (Re A) and, further,
 
(Re A)ζ, ζ ∈ R for each ζ ∈ Cn . (1.4.137)

Hence, in a similar fashion to (1.4.136), for each ζ ∈ Cn we have

 
(Re A)ζ, ζ  = (Re A)(Re ζ + iIm ζ), Re ζ − iIm ζ
   
= (Re A)(Re ζ), Re ζ + (Re A)(Im ζ), Im ζ
   
= Re A(Re ζ), Re ζ + Re A(Im ζ), Im ζ

≥ κ |Re ζ | 2 + κ |Im ζ | 2 = κ |ζ | 2, (1.4.138)

proving (1.4.133). Collectively, (1.4.133), (1.4.127), and (1.4.137) then prove that
Re A is also positive definite. Next, combining (1.4.132) with the Cauchy-Schwarz
inequality justifies (1.4.134). In turn, the estimates in (1.4.134) prove that the linear
maps A : Cn → Cn and Re A : Cn → Cn are injective, thus invertible. In particular,
(1.4.135) holds. 
84 1 Integral Representations and Integral Identities

Remark 1.4.8 As a corollary of Lemma 1.4.7 we have:

sym A := 12 (A + A ) is invertible whenever the matrix

(1.4.139)
A ∈ Cn×n happens to be Legendre-Hadamard elliptic.
Other significant consequences of Lemma 1.4.7 are noted below.
Corollary 1.4.9 For any symmetric matrix A ∈ Cn×n , being Legendre-Hadamard
elliptic is actually equivalent to being (strictly) positive definite.
Moreover, any symmetric Legendre-Hadamard elliptic matrix A ∈ Cn×n is in-
vertible, and A−1 is symmetric and strictly positive definite (hence also Legendre-
Hadamard elliptic and weakly elliptic).
Proof In the first claim, the left-to-right implication is contained in Lemma 1.4.7
(cf. (1.4.132)), while the right-to-left implication has been noted in (1.4.129). The
second claim in the statement is seen from Lemma 1.4.7 (cf. (1.4.135)), the fact
that the transpose commutes with the inverse, the first claim in the statement of the
current corollary, (1.4.130), and (1.4.126). 
As remarked earlier, strict positive definiteness (cf. (1.4.127)) implies Legendre-
Hadamard ellipticity (cf. (1.4.123)) which, in turn, implies weak ellipticity (cf.
(1.4.120)). Regarding this hierarchy, we wish to propose yet another brand of el-
lipticity, sitting in between weak ellipticity and Legendre-Hadamard ellipticity, as
defined below.
Definition 1.4.10 Call a matrix A ∈ Cn×n mildly elliptic provided there exists
some z ∈ C \ {0} such that26
 
Ax, x  z t : t ∈ [0, ∞) for every x ∈ Rn \ {0}. (1.4.140)

Equivalently,

 A∈C
a matrix n×n is mildly elliptic if A is weakly elliptic and
 (1.4.141)
the set Ax, x : x ∈ Rn \ {0} does not coincide with C \ {0}.

As is clear from (1.4.124), (1.4.140), and (1.4.121), for every A ∈ Cn×n we have

A is Legendre-Hadamard elliptic =⇒ A is mildly elliptic

=⇒ A is weakly elliptic. (1.4.142)

Later on we shall need the following lemma:

Lemma 1.4.11 Fix n ∈ N, n ≥ 2, and let B = (b jk )1≤ j,k ≤n be an n × n matrix with
complex entries which is symmetric and mildly elliptic. Then the n × n matrix
∫
(Bξ) ⊗ ξ
dH n−1 (ξ) is symmetric. (1.4.143)
S n−1 Bξ, ξ
 
26 recall that the numerical range of the matrix A is the set Aζ, ζ c  : ζ ∈ C n, |ζ | = 1 , where
the superscript ‘c’ indicates complex conjugation
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 85

Proof From Definition 1.4.10 it follows that there exists a simply connected open
set O ⊆ C \ {0} such that

Bξ, ξ ∈ O for every ξ ∈ S n−1 . (1.4.144)

The simple connectivity of the set O implies (see [112, Proposition 5.8.1]) the
existence of
h : O → C \ {0} holomorphic function, with the
(1.4.145)
property that h (z) = 1/z at every point z ∈ O.

In particular, for each differentiable function g : Rn → O and each j ∈ {1, . . . , n},

the Cauchy-Riemann equations and Chain Rule give

(∂j g)(ξ)
∂ξ j h(g(ξ)) = h (g(ξ))(∂j g)(ξ) = , ∀ξ ∈ Rn . (1.4.146)
g(ξ)

To proceed, define f (ξ) := h(Bξ, ξ) for each ξ near S n−1 . Thanks to (1.4.144),
f is a well-defined, smooth, complex-valued function. Having fixed two arbitrary
indices j, k ∈ {1, . . . , n}, we may then compute
∫ ∫
∂τ j k Bξ, ξ
0= ∂τ j k f (Bξ, ξ) dH n−1 (ξ) = dH n−1 (ξ)
S n−1 S n−1 Bξ, ξ
∫
ξ j ∂ξk Bξ, ξ − ξk ∂ξ j Bξ, ξ
= dH n−1 (ξ)
S n−1 Bξ, ξ
∫
2ξ j (Bξ)k − 2ξk (Bξ) j
= dH n−1 (ξ)
S n−1 Bξ, ξ
!∫ "
(Bξ) ⊗ ξ
=2 dH n−1 (ξ)
S n−1 Bξ, ξ kj
!∫ "
(Bξ) ⊗ ξ
−2 dH (ξ) .
n−1
(1.4.147)
S n−1 Bξ, ξ jk

Above, the first equality is a very particular case of the integration by parts formula on
the boundary of the unit ball. See [113, Lemma 11.1.1], used here with Ω := B(0, 1),
ϕ := f , and some ψ ∈ 𝒞1c (Rn ) which is identically one near S n−1 . The second
equality in (1.4.147) is simply a consequence of (1.4.146), while the third equality
follows from (A.0.115). The fourth equality in (1.4.147) uses the fact that B is
symmetric, and the last equality in (1.4.147) is seen from (1.3.7). In turn, (1.4.147)
readily implies (1.4.143), finishing the proof of the lemma. 

Next we elaborate on the nature of the square-root determinant function consid-

ered on the set of symmetric, Legendre-Hadamard elliptic, complex matrices. Our
main result in this regard is stated in Proposition 1.4.12 below. To set the stage,
consider
86 1 Integral Representations and Integral Identities
 
Mn := A ∈ Cn×n : A is Legendre-Hadamard elliptic and symmetric
 
= A ∈ Cn×n : A is strictly positive definite and symmetric , (1.4.148)

where the second equality comes from Corollary 1.4.9. Observe (see (1.4.139)) that

det A  0 for every A ∈ Mn . (1.4.149)

Since any n × n symmetric matrix with complex entries is uniquely determined by

its entries lying on or below the main diagonal, we may naturally identify Mn with
an open convex subset of Cn(n+1)/2 (in fact, with an open convex conical27 set with
vertex at the origin).
Proposition 1.4.12 Under the above identification of Mn with an open subset of
Cn(n+1)/2 , there exists a unique separately holomorphic function (i.e., holomorphic
in each scalar complex component in Cn(n+1)/2 )
1
Mn A −→ (det A) 2 ∈ C (1.4.150)

with the property that

1
(det A) 2 ∈ (0, ∞) whenever A ∈ Mn has real entries. (1.4.151)

Thus (det A)− 2 is unambiguously defined as a number in C \ {0} for each A ∈ Mn

1

and, in fact,
∫
− 12 −n/2
(det A) = π e− Aξ,ξ  dξ, ∀A ∈ Mn . (1.4.152)
Rn

As a preamble, we record the following unique continuation result which appears

in [108, Lemma 7.53, p. 268].
Lemma 1.4.13 Let N ∈ N and assume O is an open and convex subset of C N with
the property that O ∩ R N   (where R N is canonically embedded into C N ). Also,
suppose g1, g2 : O → C are two functions which are separately holomorphic (i.e.,
in each scalar complex component in C N ) such that
 
g1  O∩R N = g2  O∩R N . (1.4.153)

Then g1 = g2 everywhere in O.
We are ready to present the proof of Proposition 1.4.12.
Proof of Proposition 1.4.12 For starters, we claim that
∫ 2
− Aξ,ξ 
(det A) e dξ = π n for each A ∈ Mn . (1.4.154)
Rn

27 i.e., a set stable under multiplication by strictly positive numbers

1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 87

Before proving (1.4.154) we wish to note that the integral appearing above is ab-
solutely convergent. Indeed, if A ∈ Mn then there exists κ ∈ (0, ∞) such that
Re Aξ, ξ ≥ κ|ξ | 2 for every ξ ∈ Rn , hence
∫   ∫ ∫
 − Aξ,ξ  
e−Re Aξ,ξ  dξ ≤ e−κ |ξ | dξ = (π/κ)n/2,
2
e  dξ = (1.4.155)
Rn Rn Rn

in view of the well-known formula

∫ +∞
e−κ t dt = (π/κ)1/2 .
2
(1.4.156)
−∞

In particular, the mapping g : Mn → C given by

∫ 2
− Aξ,ξ 
g(A) := (det A) e dξ , ∀A ∈ Mn, (1.4.157)
Rn

is well defined. Moreover, a moment’s reflection shows that g is holomorphic in each

variable separately. In concert with (1.4.156)-(1.4.157), a diagonalization process
(for real symmetric matrices, involving an orthogonal transformation in Rn ) then
shows that

g(A) = π n whenever A ∈ Mn has real entries. (1.4.158)

With this in hand, Lemma 1.4.13 applies (here is where the convexity of Mn is first
relevant) and yields g(A) = π n for every A ∈ Mn . This finishes the proof of the
claim made in (1.4.154).
We next claim that
∫  −1
f (A) := e− Aξ,ξ  dξ , ∀A ∈ Mn, (1.4.159)
Rn

is the unique mapping f : Mn → C satisfying

f is separately holomorphic,
f (A)2 = det A for every matrix A ∈ Mn, (1.4.160)
f (A) ∈ (0, ∞) whenever A ∈ Mn has real entries.

Indeed, from (1.4.154), (1.4.155), and (1.4.149), we conclude that the mapping
(1.4.159) is well-defined and enjoys all properties stipulated in (1.4.160). To establish
the uniqueness of such a mapping, it suffices to show that if f1, f2 : Mn → C are two
separately holomorphic functions satisfying

f1 (A)2 = f2 (A)2 for every matrix A ∈ Mn,

(1.4.161)
f1 (A), f2 (A) ∈ (0, ∞) whenever A ∈ Mn has real entries,
88 1 Integral Representations and Integral Identities

then they necessarily coincide on Mn . To see that this is the case, observe first that
the properties in (1.4.161) imply

f1 (A) = f2 (A) whenever A ∈ Mn has real entries (1.4.162)

which, thanks to Lemma 1.4.13 (whose applicability is ensured by the convexity

of Mn ), ultimately forces f1 = f2 in Mn . This finishes the proof of the fact that
(1.4.159) is the unique complex-valued map defined on Mn satisfying (1.4.160).
At this stage, defining
1
Mn A −→ (det A) 2 := f (A) ∈ C (1.4.163)

all desired conclusions follow. 

Remark 1.4.14 For any matrix A ∈ Cn×n satisfying (1.4.131), formula (1.4.152)
actually implies (in view of (1.4.155)) the following stronger version of (1.4.135):

|det A| ≥ κ n and |det (Re A)| ≥ κ n . (1.4.164)

Remark 1.4.15 A related formula to (1.4.152), to the effect that

∫
− 12 1 1
(det A) =   n2 dH (ξ),
n−1
∀A ∈ Mn, (1.4.165)
ωn−1 Aξ, ξ
n−1
S

has been proved in [108, (7.8.50), p. 270] (see also [109, (7.12.47), p. 314] and [109,
§7.12, p. 305] for related matters). Here we wish to note that this may be used with
the same effectiveness for the purpose of establishing (1.4.150)-(1.4.151).

Remark 1.4.16 In view of the fact that for each A ∈ Cn×n we have
   
Aξ, ξ = (sym A)ξ, ξ , ∀ξ ∈ Rn, (1.4.166)

it follows from (1.4.152) that

∫
− 12
det (sym A) = π −n/2 e− Aξ,ξ  dξ,
Rn (1.4.167)
for every A ∈ Cn×n Legendre-Hadamard elliptic.

Formula (1.4.165) is useful in dealing with the last integral in (1.4.22) for scalar
Legendre-Hadamard elliptic operators. Specifically, for every Legendre-Hadamard
elliptic matrix A ∈ C2×2 the case n = 2 of (1.4.165) gives (in light of (1.4.166), and
the fact that, when M = 1 and n = 2, we have L(ξ) = −Aξ, ξ for each ξ ∈ R2 )
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 89
∫ ∫
1
− [L(ξ)]−1 dH 1 (ξ) =   dH 1 (ξ)
Aξ, ξ
S1 S1
∫
1
=   dH 1 (ξ)
(sym A)ξ, ξ
S1

− 12
= 2π det (sym A) . (1.4.168)

Consequently, when n = 2 and M = 1 formula (1.4.22) becomes

∫ ln   x , ξ  
1 |x | ln |x|
E(x) = 2   dH 1 (ξ) + 5
4π Aξ, ξ 2π det (sym A) (1.4.169)
S1
for all x ∈ R2 \ {0}.

In particular, this raises the prospect of finding a fully explicit formula28 for the
fundamental solution E associated with any second-order, scalar, constant (complex)
coefficient, Legendre-Hadamard elliptic operator L in Rn with n ≥ 2, something
accomplished in the proposition below.

Proposition 1.4.17 Consider the special case of Legendre-Hadamard elliptic scalar

operators in Rn with n ≥ 2. That is, assume M = 1, in which scenario L = ar s ∂r ∂s ,
and suppose the coefficient matrix A = (ar s )1≤r,s ≤n ∈ Cn×n is Legendre-Hadamard
elliptic; as noted in (1.4.123), this amounts to the existence of some κ > 0 such that

Re ar s ξr ξs ≥ κ|ξ | 2, ∀ξ = (ξ1, . . . , ξn ) ∈ Rn . (1.4.170)

Then the fundamental solution E of L from (1.4.18) takes the explicit form
⎧
⎪ 1   2−n
⎪
⎪
⎪ − 5 (sym A)−1 x, x 2 if n ≥ 3,
⎪
⎨ (n − 2)ωn−1 det (sym A)
⎪
E(x) = (1.4.171)
⎪
⎪ 1  
⎪
⎪ 5 log (sym A)−1 x, x + c A if n = 2,
⎪
⎪ 4π det (sym A)
⎩

at every x ∈ Rn \ {0}. Here, sym A := 12 (A + A ) is the symmetric part of the

5
coefficient matrix A, and the square-root det (sym A) is understood in the sense of
Proposition 1.4.12. Also, c A is a complex constant which depends solely on A, while
log denotes the principal branch of the complex logarithm function (defined by the
requirement that log z = ln z if z ∈ (0, ∞); see the discussion in [112, Remark 5.8.2]).
Finally, ωn−1 denotes the surface measure of the unit sphere in Rn , and

28 i.e., a formula lacking integrals

90 1 Integral Representations and Integral Identities


n
a, b := a j b j for any generic
j=1
(1.4.172)
a = (a j )1≤ j ≤n ∈ Cn and b = (b j )1≤ j ≤n ∈ Cn .

The presence of the additive constant c A in (1.4.171) when n = 2 is expected

given that assignment (acting from the set of all complex 2 × 2 Legendre-Hadamard
elliptic matrices)
1  
A −→ E A(x) := 5 log (sym A)−1 x, x (1.4.173)
4π det (sym A)

does not satisfy the dilation law EλA = λ−1 E A for each λ ∈ (0, ∞) predicted by
(1.4.32).
Proof of Proposition 1.4.17 Let us temporarily denote by E0 the function defined
as in (1.4.171) for the choice c A := 0. From [108, Theorem 7.54, p. 270] we know
that
E0 ∈ Lloc
1
(Rn, L n ) ∩ 𝒞∞ (Rn \ {0}) (1.4.174)
is a tempered distribution in Rn which is a fundamental solution for the weakly
elliptic differential operator L = ar s ∂r ∂s . Moreover, a cursory inspection reveals
that there exists a finite constant C > 0 such that for each x ∈ Rn \ {0} we have

⎧
⎪ C
⎪
⎨ |x| n−2
⎪ if n ≥ 3,
|E0 (x)| ≤ (1.4.175)
⎪
⎪ C 1 +  ln |x|  if n = 2.
⎪
⎩
With E denoting the fundamental solution of L = ar s ∂r ∂s from (1.4.18), the unique-
ness result in item (8) of Theorem 1.4.2 applies and gives that Q := E − E0 is a
polynomial in Rn . In view of (1.4.24) (with γ = 0) and (1.4.175), said polynomial
decays at infinity if n ≥ 3, and grows at most logarithmically if n = 2. This forces Q
to vanish identically when n ≥ 3, and be constant (say, Q(x) = c A for every x ∈ R2 )
when n = 2. The desired conclusion now follows. 

We continue pursuing the goal of producing fully explicit formulas for the funda-
mental solutions of scalar weakly elliptic operators (originally constructed in Theo-
rem 1.4.2 via integral expressions). Henceforth, with each given a matrix of complex
coefficients A = (ar s )1≤r,s ≤n ∈ Cn×n , we agree to associate the second-order scalar
operator in Rn given by
L A := ∇ · A∇ = ar s ∂r ∂s . (1.4.176)
In particular, the “characteristic 1 × 1 matrix” of L A is given by

L A(ξ) = −ar s ξr ξs = −Aξ, ξ for each ξ = (ξ1, . . . , ξn ) ∈ Rn . (1.4.177)

1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 91

Fundamental Solutions for Weakly Elliptic Scalar Operators when n = 2

To set the stage for the subsequent discussion, let us define
 
M2 := A ∈ C2×2 : A = A and A is weakly elliptic . (1.4.178)

In relation to this, we have the following structural result.

Lemma 1.4.18 A matrix A ∈ C2×2 belongs to M2 if and only if A has the form
λ1 +λ2
( 1 − 2 +
A = a) , with a ∈ C \ {0} and λ1, λ2 ∈ C \ R, (1.4.179)
λ1 +λ2
*− 2 λ1 λ2 -

or, equivalently, A is symmetric and

Aξ, ξ = a(ξ1 − λ1 ξ2 )(ξ1 − λ2 ξ2 ), ∀ξ = (ξ1, ξ2 ) ∈ R2 . (1.4.180)

Proof Simple algebra shows that (1.4.179) is equivalent with A being a symmetric
matrix satisfying (1.4.180). To proceed, suppose A is as in (1.4.179). Then clearly A
is symmetric. Also, thanks to (1.4.180), having Aξ, ξ = 0 for some ξ ∈ R2 \ {0}
forces either a = 0, or ξ1 = λ1 ξ2 , or ξ1 = λ2 ξ2 . However, none of these possibilities
materialize since, by assumption, a ∈ C\ {0} and λ1, λ2 ∈ C\R. In view of (1.4.120)
and (1.4.178) this proves that A ∈ M2 .
To prove the converse implication, fix an arbitrary matrix A ∈ M2 and write
 
ab
A= , (1.4.181)
bc

where a, b, c ∈ C. The fact that A is weakly elliptic entails (cf. (1.4.120))

a = Ae1, e1  ∈ C \ {0}. (1.4.182)

Denote by λ1, λ2 ∈ C the roots of the quadratic equation az 2 + 2bz + c = 0. Then

az 2 + 2bz + c = a(z − λ1 )(z − λ2 ), (1.4.183)

which, in view of the equivalence between (1.4.179) and (1.4.180), further implies
  λ1 +λ2
( 1 − 2 +
ab
A= = a) ,. (1.4.184)
bc λ1 +λ2
* − 2 λ 1 λ 2 -
To complete the proof we only need to see that λ1, λ2 ∈ C \ R. Otherwise, if for
instance λ1 ∈ R, we could take ξ = (λ1, 1) ∈ R2 \ {0} and obtain that

Aξ, ξ = aλ12 + 2bλ1 + c = 0, (1.4.185)

92 1 Integral Representations and Integral Identities

which contradicts the assumption that A is weakly elliptic. The desired conclusion
follows. 

In view of Lemma 1.4.18, from now on we shall assume that every A ∈ M2 is

expressed as in (1.4.179) and decompose

M2 = M20 M2+ M2−, (1.4.186)

where
 
M20 := A ∈ M2 : Im λ1 · Im λ2 < 0 , (1.4.187)
 
M2+ := A ∈ M2 : Im λ1 > 0 and Im λ2 > 0 , (1.4.188)
 
M2− := A ∈ M2 : Im λ1 < 0 and Im λ2 < 0 . (1.4.189)

For further use we note that, since I2×2 can be written as in (1.4.179) with a := 1,
λ1 := i, and λ2 := −i,

the identity matrix I2×2 belongs to M20 . (1.4.190)

Lemma 1.4.19 Let A ∈ C2×2 be an arbitrary weakly elliptic matrix. Then its sym-
metric part sym A := 12 (A + A ) belongs to M2 . Moreover, if sym A ∈ M2± then
∫
1
dH 1 (ξ) = 0. (1.4.191)
S 1 Aξ, ξ

Finally, if sym A ∈ M20 , then expressing sym A as in (1.4.179) one has

∫
1 4πi · sgn (Im λ1 )
dH 1 (ξ) = . (1.4.192)
S 1 Aξ, ξ a(λ1 − λ2 )
Proof That sym A belongs to M2 is clear from assumptions, definitions, and
(1.4.166). Express sym A in its canonical form as in (1.4.179). Then based on
(1.4.166) and (1.4.180) we may write
∫ ∫
1 1 dH 1 (ξ)
dH 1 (ξ) = . (1.4.193)
S 1 Aξ, ξ a S 1 (ξ1 − λ1 ξ2 )(ξ1 − λ2 ξ2 )

Observe that for each f ∈ L 1 (S 1, H 1 ) we have

∫ ∫ ∞  t 1  dt
f (ξ) dH (ξ) = 2
1
feven √ ,√ , (1.4.194)
1 + t2 1 + t2 1 + t
2
S1 −∞

where feven is the even part of f , defined as

 
feven (ξ) := 12 f (ξ) + f (−ξ) , ξ ∈ S1 . (1.4.195)
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 93
√
Indeed, (1.4.194)
√ follows by making the changes of variables ξ1 = ±t/ 1 + t 2 and
ξ2 = ±1/ 1 + t with t ∈ R in (1.4.194). See also [116, §1.6] for a more general
2

version of (1.4.194).
Note that the last integrand in (1.4.193) is even. Bearing this in mind and applying
(1.4.194), we may further recast the right-hand side of (1.4.193) as
∫ ∫
2 ∞ dt
= g(t) dt, (1.4.196)
a −∞ (t − λ1 )(t − λ2 ) R

where g is the meromorphic function

2
g(z) := for each z ∈ C \ {λ1, λ2 }. (1.4.197)
a(z − λ1 )(z − λ2 )

If sym A ∈ M2± then either

∫ λ1, λ2 ∈ C+ , or λ1, λ2 ∈ C− , and in either scenario the
Residue Theorem gives R g(t) dt = 0. This justifies (1.4.191). On the other hand, if
sym A ∈ M20 then either λ1 ∈ C+ and λ2 ∈ C− , or λ1 ∈ C− and λ2 ∈ C+ , so another
application of the Residue Theorem then justifies (1.4.192). 

With the piece of terminology introduced in Definition 1.4.10, the lemma below
(which complements Lemma 1.4.24) implies that any matrix in M20 is mildly elliptic.

Lemma 1.4.20 Let A = (ar s )1≤r,s ≤2 ∈ M20 be arbitrary. Then

Ax, x  a11 (−∞, 0] for every x ∈ R2 \ {0}. (1.4.198)

Proof In view of Lemma 1.4.18, we may assume that the matrix A is written as in
(1.4.179). In particular, a11 = a ∈ C \ {0}. Seeking a contradiction, assume that
there exist some t ∈ (−∞, 0] along with some x = (x1, x2 ) ∈ R2 \ {0} such that

at = Ax, x = a(x1 − λ1 x2 )(x1 − λ2 x2 ), (1.4.199)

where the second equality comes from (1.4.180). In view of the fact that a  0, this
implies
t = (x1 − λ1 x2 )(x1 − λ2 x2 ). (1.4.200)
If x2 = 0 then x1  0 and (1.4.200) forces t = x12 , contradicting the fact that t ≤ 0.
If x2  0, using (1.4.199) we may write
! "
t x2 Im λ2 Im (t x1 − t x2 λ2 ) t
= = Im
|x1 − λ2 x2 | 2 |x1 − λ2 x2 | 2 x1 − λ2 x2

= Im (x1 − λ1 x2 ) = −x2 Im λ1, (1.4.201)

hence
t Im λ2
= −Im λ1 . (1.4.202)
|x1 − λ2 x2 | 2
94 1 Integral Representations and Integral Identities

Using that t ≥ 0 and bearing in mind that A ∈ M20 , it follows that

tIm λ1 · Im λ2
0≤ = −(Im λ1 )2 < 0, (1.4.203)
|x1 − λ2 x2 | 2
which is a contradiction. 
Recall the notion of mildly elliptic matrix introduced in Definition 1.4.10 (cf. also
(1.4.141)).

Lemma 1.4.21 Consider an arbitrary symmetric matrix A ∈ C2×2 . Then A is mildly

elliptic if and only if A ∈ M20 .

Proof That any matrix A ∈ M20 is mildly elliptic follows from Lemma 1.4.20 and
Definition 1.4.10. Suppose next that A ∈ C2×2 is a symmetric mildly elliptic matrix.
From (1.4.142) and (1.4.178) we then see that A ∈ M2 . Bearing this in mind together
with (1.4.186) and (1.4.141), the goal is to prove that
 
Ax, x : x ∈ R2 \ {0} = C \ {0} whenever A ∈ M2± . (1.4.204)

Fix an arbitrary A ∈ M2± , and express it as in (1.4.179). Consider the semi-circles

 
S+1 := ξ = (ξ1, ξ2 ) ∈ S 1 : ξ2 ≥ 0 ,
  (1.4.205)
S−1 := ξ = (ξ1, ξ2 ) ∈ S 1 : ξ2 ≤ 0 .

and note that since −ξ ∈ S−1 for each ξ ∈ S+1 , we have

   
Aξ, ξ : ξ ∈ S 1 = Aξ, ξ : ξ ∈ S+1 . (1.4.206)

Bearing this in mind, via dilation, it suffices to show that the closed curve29
 
γ := Aξ, ξ : ξ ∈ S+1 ⊆ C \ {0} (1.4.207)

goes around the origin 0 ∈ C. This, in turn, follows if we succeed in showing that
the winding number of the curve γ around the point 0 ∈ C is nonzero, i.e.,
∫
1 dz
Indγ (0) :=  0. (1.4.208)
2πi γ z

To this end, use the parametrization

29 indeed, the fact that the end-points of S+1 are diametrically opposite implies that the end-points
of γ coincide
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 95
  −t 1   −t 1 
γ(t) := A √ ,√ , √ ,√
1 + t2 1 + t2 1 + t2 1 + t2
 −t λ1   −t λ2 
=a √ −√ √ −√
1 + t2 1 + t2 1 + t2 1 + t2
(t + λ1 )(t + λ2 )
=a , −∞ < t < +∞, (1.4.209)
1 + t2
the second equality being a consequence of (1.4.180). Since
 t 2 + (λ + λ )t + λ λ 
1 2 1 2
γ (t) = a
1 + t2
−(λ1 + λ2 )(t 2 − 1) + 2(1 − λ1 λ2 )t
=a (1.4.210)
(1 + t 2 )2
for each t ∈ R, we have
∫ ∫ +∞
1 dz 1 −(λ1 + λ2 )(t 2 − 1) + 2(1 − λ1 λ2 )t
=± dt
2πi γ z 2πi −∞ (t + λ1 )(t + λ2 )(1 + t 2 )
∫
1
=± f (t) dt, (1.4.211)
2πi R
where the ± is the sign of the parametrization (1.4.209) (“plus” if the parametrization
covers the curve γ counterclockwise, and “minus” if the parametrization covers the
curve γ clockwise), and where f is the meromorphic function

−(λ1 + λ2 )(z 2 − 1) + 2(1 − λ1 λ2 )z

f (z) :=
(z + λ1 )(z + λ2 )(z + i)(z − i) (1.4.212)
defined for each z ∈ C \ {λ1, λ2, ±i}.

Given that A ∈ M2± , we either have λ1, λ2 ∈ C+ , or λ1, λ2 ∈ C− . In the first

scenario, the Residue Theorem gives (treating R as the boundary of C+ )
∫
1 2(λ1 + λ2 ) + 2(1 − λ1 λ2 )i
f (t) dt = Res( f , i) =
2πi R (i + λ1 )(i + λ2 )(2i)
−(λ1 + λ2 )i + (1 − λ1 λ2 ) −(i + λ1 )(i + λ2 )
= =
(i + λ1 )(i + λ2 ) (i + λ1 )(i + λ2 )

= −1, (1.4.213)

where the last step uses the fact that (i + λ1 )(i + λ2 )  0 since we are presently
assuming λ1, λ2 ∈ C+ .
Finally, if λ1, λ2 ∈ C− then another application of the Residue Theorem gives
(now treating R as the boundary of C− , with a negative orientation)
96 1 Integral Representations and Integral Identities
∫
1 2(λ1 + λ2 ) − 2(1 − λ1 λ2 )i
f (t) dt = −Res( f , −i) = −
2πi R (−i + λ1 )(−i + λ2 )(−2i)
(λ1 + λ2 )i + (1 − λ1 λ2 ) −(−i + λ1 )(−i + λ2 )
=− =−
(−i + λ1 )(−i + λ2 ) (−i + λ1 )(−i + λ2 )

= 1, (1.4.214)

where the last step relies on the observation that (−i + λ1 )(−i + λ2 )  0, given that
we now assume λ1, λ2 ∈ C− . Thus, (1.4.208) holds in all cases, finishing the proof.

Going further, we briefly digress for the purpose of introducing notation and
terminology which will become useful shortly. Fix an integer m ∈ N and denote by
Pm the set of all permutation of {1, . . . , m}. Given an arbitrary set E, denote by E m
the m-fold Cartesian product of E with itself. On this set, consider the equivalence
relation ∼ stipulating, for each (a1, . . . , am ) ∈ E m and (b1, . . . , bm ) ∈ E m , that

(a1, . . . , am ) ∼ (b1, . . . , bm ) if and only if there exists

(1.4.215)
σ ∈ Pm such that b j = aσ(j) for each j ∈ {1, . . . , m}.

For each (a1, . . . , am ) ∈ E m denote by (a1, . . . , am ) its equivalence class modulo

the equivalence relation ∼. The symbol E m /∼ stands for the set of all such equiva-
lence classes. We shall refer to elements in E m /∼ as being m-families (with elements
in E), and use the notation

{a1, . . . , am }∗ := (a1, . . . , am ) ∈ E m /∼
(1.4.216)
for each (a1, . . . , am ) ∈ E m .

Thus, informally speaking, m-families are “sets of cardinality m, in which repetitions

are allowed.” If | · − · | is a metric on E, then we equip E m /∼ with the distance30
 
Dm {a1, . . . , am }∗, {b1, . . . , bm }∗ := min max |a j − bσ(j) |. (1.4.217)
σ ∈ Pm 1≤ j ≤m

For example, corresponding to m = 2, we have the following distance

 
'1, λ
D2 {λ1, λ2 }∗,{ λ '2 }∗
    
'1 |, |λ2 − λ
:= min max |λ1 − λ '2 | , max |λ1 − λ
'2 |, |λ2 − λ
'1 |

'1, λ
for all λ1, λ2, λ '2 ∈ C, (1.4.218)

on the set of all 2-families with elements in C. As is apparent from the above
definition, D2 is translation invariant, and commutes with positive dilations.

30 It may be checked, based on definitions, that D m is indeed a distance function on E m /∼.

1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 97

Informally speaking, our next result asserts that the location of the two (possibly
identical) roots of a quadratic equation, considered together as a 2-family, changes
continuously with respect to the coefficients of the quadratic polynomial.
Lemma 1.4.22 For each a, b, c ∈ C with a  0, denote by λ1 (a, b, c) ∈ C and
λ2 (a, b, c) ∈ C the two roots of the quadratic equation

az 2 + 2bz + c = 0. (1.4.219)

Then for each fixed triplet ao, bo, co ∈ C with ao  0 it follows that
 ∗  ∗ 
D2 λ1 (a, b, c), λ2 (a, b, c) , λ1 (ao, bo, co ), λ2 (ao, bo, co )
(1.4.220)
converges to 0+ as (a, b, c) → (ao, bo, co ).

The conclusion in (1.4.220) may be rephrased as saying

 that the function assign-
∗
ing to each triplet a, b, c ∈ C with a  0 the 2-family λ1 (a, b, c), λ2 (a, b, c) is
continuous when regarded as a mapping
 defined from an open subset of C3 taking
values in the metric space C2 /∼, D2 .
Proof of Lemma 1.4.22 Since a  0, equation (1.4.219) has the same roots as

z 2 + 2(b/a)z + (c/a) = 0. (1.4.221)

In turn, the roots of (1.4.221) are given by the quadratic formula which, in view of
(1.4.216), implies
 6 6 &∗
 ∗ b2 b2
λ1 (a, b, c), λ2 (a, b, c) = − a + a2 − a , − a − a2 − a .
b c b c
(1.4.222)

√
Above, we agree to define the square-root of z ∈ C as z := |z| 1/2 eiθ/2 where
θ ∈ [0, 2π) is such that z = |z|e (which determines θ uniquely if z  0). The
iθ
√
square-root function thus defined C z → z ∈ C is continuous in C \ (0, ∞) and
at each real number zo ∈ (0, ∞) we have
√ √ √ √
lim z = zo whereas lim z = − zo . (1.4.223)
C+ z→z o C− z→z o

While this change in sign renders the square-root function discontinuous at each
√(0, ∞),
point on the real semi-axis it does not affect the continuity of the 2-family
√ ∗
valued function C z → z, − z , in the sense that for each zo ∈ C we have
 √ √  ∗ √ √ ∗ 
D z, − z , zo, − zo −→ 0+ as z → zo . (1.4.224)

Then the claim made in (1.4.220) follows from this and (1.4.222), bearing in mind
that D2 is translation invariant. 
Moving on, since any 2 × 2 symmetric matrix with complex entries is uniquely
determined by its entries lying on or below the main diagonal, we may naturally
98 1 Integral Representations and Integral Identities

identify M2 with a subset of C3 (stable under multiplication by real numbers). With

this identification in mind, we have the following result pertaining to the topology
of M2 .

Lemma 1.4.23 When viewed as subsets in C3 (as indicated above), the sets M20 , M2+ ,
and M2− are open, pathwise connected, and mutually disjoint. As such, these sets are
the connected components of the open set M2 (regarded as a subset of C3 ).

Proof That M2 is an open subset of C3 (under the identification described above) is

clear from (1.4.122). Note that M20 is canonically identified with
 
(a, b, c) ∈ C3 : one root of az 2 + 2bz + c = 0 is in C+ and one in C− . (1.4.225)

From this and Lemma 1.4.22 we then see that M20 is an open subset of C3 . Likewise,
since M2± are canonically identified with
 
(a, b, c) ∈ C3 : both roots of az 2 + 2bz + c = 0 are in C± , (1.4.226)

from Lemma 1.4.22 we conclude that they are open subsets of C3 .

Next, fix two arbitrary matrices A, B ∈ M20 . From (1.4.187) and Lemma 1.4.18
we know that we can write
 
( 1 − 2 +
λ1 +λ2
1 − γ1 +γ2
2

A = a) ,, B=b , (1.4.227)
λ1 +λ2 γ1 +γ2
* − 2 λ 1 λ 2 - − 2 γ1 γ 2

with a, b ∈ C \ {0}, Im λ1 · Im λ2 < 0 and Im γ1 · Im γ2 < 0. Relabeling if needed

we may assume that Im λ1, Im γ1 > 0 and Im λ1, Im γ2 < 0. It is useful to express
a, b as

a = |a|eiθa and b = |b|eiθb for some θ a, θ b ∈ [0, 2π]. (1.4.228)

For each t ∈ [0, 1] we then define

c(t) := |a| 1−t |b| t ei(1−t)θa +itθb ,

ρ1 (t) := (1 − t)λ1 + tγ1, (1.4.229)
ρ2 (t) := (1 − t)λ2 + tγ2 .

Clearly, c(t) ∈ C \ {0}, Im ρ1 (t) > 0, Im ρ2 (t) < 0 for every t ∈ [0, 1]. Hence, if we
set
( 1 − ρ1 (t)+ρ
2
2 (t)
+
C(t) := c(t) ) , (1.4.230)
ρ1 (t)+ρ2 (t)
* − 2 ρ 1 (t)ρ 2 (t) -
it follows that the matrix C(t) belongs to M20 for every t ∈ [0, 1]. We have therefore
constructed a function [0, 1] t → C(t) ∈ C2×2 which is continuous, its image
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 99

is contained in M20 , and satisfies C(0) = A and C(1) = B. Hence, M20 is pathwise
connected.
In the case when A, B ∈ M2+ (respectively, M2− ) the same type of construction
works. This time, (1.4.229) ensures that if Im λ1 , Im λ2 , Im γ1 , Im γ2 are all strictly
positive (respectively, strictly negative) then Im ρ1 (t), Im ρ2 (t) are strictly positive
(respectively, strictly negative). Consequently, C(t) as define in (1.4.230) belongs to
M2+ (respectively, M2− ).
In summary, the above argument shows that M20 , M2+ , M2− are all pathwise con-
nected. Since, by design, M20 , M2+ , and M2− are also mutually disjoint open subsets
of C3 , whose union is M2 , it follows that these sets are precisely the connected
components of the open set M2 . 
The prototypes of all weakly elliptic scalar second-order differential operators in
the two dimensional setting, namely the Laplacian Δ = ∂x21 + ∂x22 , Bitsadze’s operator
4 (∂x1 −i∂x2 ) , and the square of the Cauchy-Riemann operator 4 (∂x1 +i∂x2 ) , may be
1 2 1 2
0 + −
expressed using matrices of coefficients belonging to M2 , M2 , and M2 , respectively.
Of course, Δ may be written using the identity as its matrix of coefficients, which
is actually Legendre-Hadamard elliptic. Our next lemma sheds further light on this
phenomenon.

Lemma 1.4.24 The subset of M2 described as

 
M2LH := A ∈ C2×2 : A = A and A is Legendre-Hadamard elliptic (1.4.231)

is included in M20 .

It should be noted that the above inclusion is strict. For example,

 
1 −1
A := (1.4.232)
−1 1 + 2i

is a complex 2 × 2 matrix which may be written as in (1.4.179) with

a := 1 ∈ C \ {0} and λ1 := i ∈ C \ R, λ2 := 2 + i ∈ C \ R. (1.4.233)

As such, Lemma 1.4.18 ensures that the matrix A belongs to M2 . In addition, for
each ξ = (ξ1, ξ2 ) ∈ R2 we have
 
ReAξ, ξ = Re ξ12 − 2ξ1 ξ2 + (1 + 2i)ξ22 = ξ12 − 2ξ1 ξ2 + ξ22 = (ξ1 − ξ2 )2 . (1.4.234)

In particular, ReAξ, ξ vanishes whenever ξ belongs to the diagonal in R × R, so A

fails to be Legendre-Hadamard elliptic.
Proof of Lemma 1.4.24 Recall from (1.4.190) that I2×2 ∈ M20 . Fix A ∈ M2LH and
consider the continuous path

[0, 1] t → C(t) = (1 − t)I2×2 + t A ∈ C2×2 . (1.4.235)

100 1 Integral Representations and Integral Identities

Then C(0) = I2×2 , C(1) = A, and C(t) = C(t) for every t ∈ [0, 1]. Moreover for
each ξ ∈ R2

ReC(t)ξ, ξ = (1 − t)|ξ | 2 + tReAξ, ξ ≥ ((1 − t) + κ t)|ξ | 2 (1.4.236)

where we have used that A is Legendre-Hadamard elliptic with constant κ > 0 (cf.
(1.4.123)). Thus, for every t ∈ [0, 1] the matrix C(t) is Legendre-Hadamard elliptic
with constant (1 − t) + κ t > 0. Consequently, C(t) ∈ M2LH ⊂ M2 for every t ∈ [0, 1].
Moreover, since M20 is a connected component of M2 and I2×2 ∈ M20 it follows that
C(t) ∈ M20 for every t ∈ [0, 1]. Therefore, A = C(1) ∈ M20 as desired. 

Here is another example, of a different nature than (1.4.232), which will play a
role later on, in [115, §8.2], in the treatment of the operators Lλ defined in (1.3.63).

Example 1.4.25 For any a, b ∈ C, a matrix of the form

 
1 b
A=a (1.4.237)
b −1

belongs to M2 if and only if

a ∈ C \ {0} and b ∈ C \ R. (1.4.238)

In addition,
whenever (1.4.238) holds then
(1.4.239)
A∈ M2+ if Im b < 0, and A ∈ M2− if Im b > 0.
As a consequence of this, (1.4.186), and Lemma 1.4.24, any matrix A as in (1.4.237)-
(1.4.238) is not Legendre-Hadamard elliptic.

Indeed, if λ1, λ2 ∈ C solve − λ1 +λ

2
2
= b and λ1 λ2 = −1, then having λ1, λ2 ∈ C \ R
becomes equivalent to having b ∈ C \ R. According to Lemma 1.4.18, the conditions
in (1.4.238) are then equivalent to the membership of A to M2 . In addition, having
λ1
λ1 λ2 = −1 guarantees that λ1, λ2  0, so we may express λ2 = −1 λ1 = − |λ | 2 , which
1
λ1
further implies Im λ2 = Im |λ1 | 2
. As a consequence, Im λ1 and Im λ2 have the same
sign, which is actually the opposite of Im b. The claims in (1.4.239) now follow from
this and (1.4.186)-(1.4.189).
Let us move on. The following residue computation is going to be relevant shortly
(see the proof of Lemma 1.4.28).

Lemma 1.4.26 For every λ ∈ C \ R one has

∫
ξ1 − λ ξ2
Iλ := dH 1 (ξ) = −2πi · sign (Im λ). (1.4.240)
S ξ2 + λ ξ1
1
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 101

Proof Upon noting that the integrand in Iλ is even, it follows from (1.4.194) that
∫ ∞
t − λ dt
Iλ = 2 . (1.4.241)
−∞ 1 + λt 1 + t
2

Assume now that Im λ > 0 and consider the meromorphic function

(z − λ)
f (z) := for each z ∈ C− . (1.4.242)
(1 + λz)(z − i)(z + i)
Then, according to the Residue Theorem,
−i − λ
Iλ = −4πi · Res ( f , −i) = −4πi = −2πi, (1.4.243)
(1 − iλ)(−2i)
which is in agreement with (1.4.240). When Im λ < 0, we considering the meromor-
phic function

(z − λ)
f (z) := for each z ∈ C+ . (1.4.244)
(1 + λz)(z − i)(z + i)
and use the Residue Theorem to compute
i−λ
Iλ = 4πi · Res ( f , i) = 4πi = 2πi, (1.4.245)
(1 + iλ)2i
as wanted. 

The subsequent discussion branches out into a number of cases, focused on special
classes of matrices.
Case I: Matrices in M20 .
Fix A ∈ M20 , that is,
λ1 +λ2
( 1 − 2 +
A = a) , with a ∈ C \ {0} and Im λ1 · Im λ2 < 0. (1.4.246)
λ1 +λ2
*− 2 λ1 λ2 -

Let L = L A := ∇ · A∇ be the second-order differential operator associated with A as

in (1.4.176). In the present case this takes the form
 λ1 + λ2 λ1 + λ2 
L = a ∂12 − ∂1 ∂2 − ∂2 ∂1 + λ1 λ2 ∂22 , (1.4.247)
2 2
hence also
L = a(∂1 − λ1 ∂2 )(∂1 − λ2 ∂2 ). (1.4.248)
Recall the piece of notation introduced in (1.4.9) and note that, in the current setting,
this becomes
102 1 Integral Representations and Integral Identities
 λ1 + λ2 λ1 + λ2 
∂ξA = a ξ1 ∂1 − ξ1 ∂2 − ξ2 ∂1 + λ1 λ2 ξ2 ∂2 (1.4.249)
2 2
for each ξ = (ξ1, ξ2 ) ∈ R2 . Bring in
λ1 +λ2
( λ1 λ2 2 +
A ad j
:= a ) ,. (1.4.250)
λ1 +λ2
* 2 1 -
Finally, define an auxiliary function ϕ A : R2 → C by setting

ϕ A(x) := Aad j x, x = a[λ1 λ2 x12 + (λ1 + λ2 )x1 x2 + x22 ]

= a(x2 + λ1 x1 )(x2 + λ2 x1 ) for each x = (x1, x2 ) ∈ R2 . (1.4.251)

Since, generally speaking,

for each fixed number λ ∈ C \ R we have

(1.4.252)
|x1 + λx2 | ≈ |x| uniformly for x = (x1, x2 ) ∈ R2,

it follows that ϕ A(x) = 0 if and only if x = 0.

In the next two lemmas we take a closer look at the function log ϕ A in R2 \ {0}.

Lemma 1.4.27 Employing notation introduced above in relation to a matrix A ∈ M20 ,

one has ϕ A(x)  at for each t ∈ (−∞, 0] and x ∈ R2 \ {0}. Hence, if log is a
holomorphic branch of the complex logarithm associated with the simply connected
domain C \ {at : t ∈ (−∞, 0]} (cf. [112, Remark 5.8.2]), it follows that log ϕ A is a
well-defined function in 𝒞∞ (R2 \ {0}).

Proof Note that by (1.4.250) we may express

λ−1 +λ−1
( 1 1 2
+
Aad j )
= aλ1 λ2 )
2
,, (1.4.253)
−1 −1
λ1 +λ2
,
−1 −1
λ1 λ2 -
* 2
with (aλ1 λ2 ) ∈ C \ {0} and
Im λ1 · Im λ2
Im (λ1−1 ) · Im (λ2−1 ) = < 0. (1.4.254)
|λ1 λ2 | 2

Hence Aad j belongs to M20 , so Lemma 1.4.20 then guarantees that ϕ A(x)  at
whenever t ≤ 0 and x ∈ R2 \ {0}. 

Here is the second lemma alluded to above, designed to facilitate eventually

invoking Proposition 1.4.1 later on.

Lemma 1.4.28 Using previous notation introduced above in relation to the matrix
A ∈ M20 (cf. (1.4.246) and (1.4.251)), one has
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 103
∫
∂ξA[log ϕ A](ξ) dH 1 (ξ) = −2πia(λ1 − λ2 ) · sign (Im λ1 ). (1.4.255)
S1

Proof Fix an arbitrary ξ = (ξ1, ξ2 ) ∈ R2 \ {0}. Then (1.4.251) and [112, (5.8.6)]
imply
 λ1 λ2 1 1 
∇[log ϕ A](ξ) = + , + , (1.4.256)
ξ2 + λ1 ξ1 ξ2 + λ2 ξ1 ξ2 + λ1 ξ1 ξ2 + λ2 ξ1
so on account of (1.4.249) we may compute

∂ξA[logϕ A](ξ)
 λ ξ λ2 ξ1 λ1 + λ2 ξ1 λ1 + λ2 ξ1
1 1
=a + − −
ξ2 + λ1 ξ1 ξ2 + λ2 ξ1 2 ξ2 + λ1 ξ1 2 ξ2 + λ2 ξ1
λ1 + λ2 λ1 ξ2 λ1 + λ2 λ2 ξ2 λ1 λ2 ξ2 λ1 λ2 ξ2 
− − + +
2 ξ2 + λ1 ξ1 2 ξ2 + λ2 ξ1 ξ2 + λ1 ξ1 ξ2 + λ2 ξ1

λ1 − λ2 ξ1 − λ1 ξ2 ξ1 − λ2 ξ2 
=a − . (1.4.257)
2 ξ2 + λ1 ξ1 ξ2 + λ2 ξ1
This along with Lemma 1.4.26 give
∫
λ1 − λ 2
∂ξA[log ϕ A](ξ) dH 1 (ξ) = a (Iλ1 − Iλ2 )
S1 2
 
= −πia(λ1 − λ2 ) sign (Im λ1 ) − sign (Im λ2 )
= −2πia(λ1 − λ2 )sign (Im λ1 ), (1.4.258)

where we have used that Im λ1 and Im λ2 have opposite signs, since A ∈ M20 . 

We are now in a position to state the main result concerning an explicit formula
for the fundamental solution constructed in Theorem 1.4.2 in relation to differential
operators of the form L A := ∇ · A∇ for a weakly elliptic matrix A whose symmetric
part belongs to M20 .

Proposition 1.4.29 Let A ∈ C2×2 be a weakly elliptic matrix with the property that
its symmetric part, sym A := 12 (A + A ), belongs to M20 . Express the matrix sym A
as in (1.4.246) and for each point x = (x1, x2 ) ∈ R2 \ {0} consider

sign (Im λ1 )
E A(x) := − log a(x2 + λ1 x1 )(x2 + λ2 x1 ) , (1.4.259)
2πia(λ1 − λ2 )
where log is the logarithm branch from Lemma 1.4.27.
Then E A ∈ Lloc 2 (R2, L 2 ) and, when regarded as a distribution in R2 , this is

a fundamental solution of L A := ∇ · A∇ (the second-order differential operator

associated with A as in (1.4.176)). Finally, there exists some C A ∈ C such that
104 1 Integral Representations and Integral Identities

E = E A + C A where E is the canonical fundamental solution of L = L A constructed

in Theorem 1.4.2.

Proof From (1.4.259), Lemma 1.4.27, and (1.4.256) we conclude that

E A ∈ 𝒞∞ (R2 \ {0}) ∩ Lloc

2
(R2, L 2 ) ∩ 𝒮 (R2 ). (1.4.260)

From (1.4.256) we also see that the first-order derivatives of E A are positive homo-
geneous of degree −1 in R2 \ {0}. In addition, it is straightforward to check that
L A E A(x) = 0 at every point x ∈ R2 \ {0}. These properties and Proposition 1.4.1
prove that E A is indeed a fundamental solution of L A. Granted this, if E is the
canonical fundamental solution associated in relation to the current operator L as in
Theorem 1.4.2, then item (8) in Theorem 1.4.2 gives that the difference Q := E A − E
is a polynomial. Since both E and E A have at most logarithmic growth, it follows
that Q is a polynomial with at most logarithmic growth. Hence, Q must be constant,
call it C A ∈ C. Ultimately, E = E A + C A as desired. 
Pressing on, we now consider:
Case II: Assume A ∈ M2± with λ1  λ2 .
Fix A ∈ M2± . From the structural result proved in Lemma 1.4.18 and (1.4.189)-
(1.4.189) we know that this may be expressed as
λ1 +λ2
( 1 − 2 +
A = a) , with a ∈ C \ {0} and Im λ1 · Im λ2 > 0. (1.4.261)
λ1 +λ2
* − 2 λ 1 λ 2 -
Here we work under the additional assumption that λ1  λ2 .
Much as before we consider the differential operator L A := ∇ · A∇ associated
with A (cf. also (1.4.247)), and recall the definition of ∂ξA given in (1.4.249). In this
case, we define the auxiliary function
x2 + λ1 x1
ϕ A(x) := a , x ∈ R2 \ {0}. (1.4.262)
x2 + λ2 x1

Observe that the denominator never vanishes for x ∈ R2 \ {0} since λ2 ∈ C \ R,

hence ϕ A is well defined and 𝒞∞ smooth in R2 \ {0}. Moreover, ϕ A is homogeneous
of degree zero, and from (1.4.252) we also conclude that ϕ A stays away from zero
and infinity.

Lemma 1.4.30 The auxiliary function ϕ A associated with A as in (1.4.262) satisfies

ϕ A(x)  at whenever t ≤ 0 and x ∈ R2 \ {0}. Hence, with log denoting a branch
of the complex logarithm defined as a holomorphic function the simply connected
domain C \ {at : t ∈ (−∞, 0]} (cf. [112, Remark 5.8.2]), it follows that log ϕ A is a
well-defined function, belonging to 𝒞∞ (R2 \ {0}).

Proof Consider the case when A ∈ M2+ (the case when A ∈ M2− is completely
analogous). Thus, we have Im λ1 > 0 and Im λ2 > 0. Fix x = (x1, x2 ) ∈ R2 \ {0}
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 105

arbitrary. If x1 > 0 (respectively, if x1 < 0), then (x2 + λ1 x1 ), (x2 + λ1 x1 ) belong

to C+ (respectively, to C− ). As a consequence, whenever x1  0 it follows that the
argument of (x2 + λ1 x1 )/(x2 + λ2 x1 ) lies in the open interval (−π, π). Hence, in this
scenario we have
x2 + λ1 x1
ϕ A(x) = a  at whenever t ≤ 0. (1.4.263)
x2 + λ2 x1
Finally, if x1 = 0 then x2  0 from which we conclude that ϕ A(x) = a  at whenever
t ≤ 0. 

As in the past, we next consider the conormal derivative of log ϕ A on the unit
sphere.

Lemma 1.4.31 Employing previous notation introduced above in relation to the

matrix A ∈ M2± (cf. (1.4.261) and (1.4.262)), one has
∫
∂ξA[log ϕ A](ξ) dH 1 (ξ) = −2πia · sign (Im λ1 )(λ1 − λ2 ). (1.4.264)
S1

Proof Pick some arbitrary ξ = (ξ1, ξ2 ) ∈ R2 \ {0}. Then from (1.4.262) and [112,
(5.8.6)] we deduce that
 λ1 λ2 1 1 
∇[log ϕ A](ξ) = − , − , (1.4.265)
ξ2 + λ1 ξ1 ξ2 + λ2 ξ1 ξ2 + λ1 ξ1 ξ2 + λ2 ξ1
which in concert with (1.4.249) yields

∂ξA[logϕ A](ξ)
 λ ξ λ2 ξ1 λ1 + λ2 ξ1 λ1 + λ2 ξ1
1 1
=a − − +
ξ2 + λ1 ξ1 ξ2 + λ2 ξ1 2 ξ2 + λ1 ξ1 2 ξ2 + λ2 ξ1
λ1 + λ2 λ1 ξ2 λ1 + λ2 λ2 ξ2 λ1 λ2 ξ2 λ1 λ2 ξ2 
− + + −
2 ξ2 + λ1 ξ1 2 ξ2 + λ2 ξ1 ξ2 + λ1 ξ1 ξ2 + λ2 ξ1
λ1 − λ2 ξ1 − λ1 ξ2 ξ1 − λ2 ξ2 

=a + . (1.4.266)
2 ξ2 + λ1 ξ1 ξ2 + λ2 ξ1
Along with Lemma 1.4.26, this permits us to compute
∫
λ1 − λ 2
∂ξA[log ϕ A](ξ) dH 1 (ξ) = a (Iλ1 + Iλ2 )
S1 2
 
= −πia(λ1 − λ2 ) sign (Im λ1 ) + sign (Im λ2 )
= −2πia(λ1 − λ2 )sign (Im λ1 ), (1.4.267)

where we have used the fact that Im λ1 and Im λ2 have the same sign, given that
A ∈ M2± . 
106 1 Integral Representations and Integral Identities

We can now provide an explicit formula for the fundamental solution constructed
in Theorem 1.4.2 for differential operators of the form L A := ∇ · A∇ for a weakly
elliptic matrix A ∈ C2×2 satisfying sym A ∈ M2± with λ1  λ2 .

Proposition 1.4.32 Let A ∈ C2×2 be a weakly elliptic matrix with the property that
its symmetric part, sym A := 12 (A + A ), belongs to M2± . Write sym A as in (1.4.261)
and assume further that λ1  λ2 . With log denoting the logarithm branch from
Lemma 1.4.30, for each x = (x1, x2 ) ∈ R2 \ {0}, define

sign (Im λ1 )  x +λ x 
2 1 1
E A(x) := − log a . (1.4.268)
2πia(λ1 − λ2 ) x2 + λ2 x1

Then E A ∈ Lloc 2 (R2, L 2 ) and, when regarded as a distribution in R2 , this is

a fundamental solution of L A := ∇ · A∇ (the second-order differential operator

associated with A as in (1.4.176)). Moreover, E A is positive homogeneous of degree
zero (in particular, E A is bounded) in R2 \ {0}, and for each multi-index α ∈ N20 one
has  
(∂ α E A)(x) = O |x| − |α | as |x| → ∞. (1.4.269)
Finally, there exists C A ∈ C such that E = E A + C A where E is the canonical
fundamental solution of L = L A constructed in Theorem 1.4.2.

Compared to the fundamental solutions for “generic” weakly elliptic systems

constructed in Theorem 1.4.2, a significant feature of E A from (1.4.268) is that this
function is positive homogeneous of degree zero (in particular, E A is bounded in
R2 \ {0}). In view of (1.4.177) and Lemma 1.4.19, this is in line with the prediction
made in (1.4.54).
Also, E A from (1.4.268) is singular at the origin, but otherwise 𝒞∞ smooth and
bounded in R2 \ {0}. This stands in sharp contrast with the nature of the fundamental
solution for the Laplacian (in this vein, recall that such singularities are removable
for harmonic functions).
Proof of Proposition 1.4.32 As seen from (1.4.268), Lemma 1.4.30, and (1.4.265),

E A ∈ 𝒞∞ (R2 \ {0}) ∩ Lloc

2
(R2, L 2 ) ∩ 𝒮 (R2 ). (1.4.270)

That E A is positive homogeneous of degree zero and satisfies (1.4.269) is clear from
(1.4.268) and (1.4.252). In particular, the first-order derivatives of E A are positive
homogeneous of degree −1 in R2 \ {0} (this is also apparent from (1.4.265)). In
addition, it is straightforward to check that L A E A(x) = 0 for every x ∈ R2 \ {0}.
Granted these properties, Proposition 1.4.1 guarantees that E A is a fundamental
solution when viewed as a distribution in R2 . Having established this, item (8) in
Theorem 1.4.2 then shows that Q := E A −E is a polynomial, where E is the canonical
fundamental solution associated with L = L A as in Theorem 1.4.2. Finally, since
E has at most logarithmic growth, while E A is bounded, Q must be constant, thus
E = E A + C A for some C A ∈ C. 
Finally, we consider:
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 107

Case III: Assume A ∈ M2± with λ1 = λ2 = λ.

Fix A ∈ M2± , such that
 
1 −λ
A=a for some a ∈ C \ {0} and λ ∈ C \ R. (1.4.271)
−λ λ2

Much as before, associate with A the differential operator L A := ∇ · A∇ (see also

(1.4.247)), and recall ∂ξA introduced in (1.4.249). In the present case, we consider
the auxiliary function ϕ A defined as
x1
ϕ A(x) := a , ∀x = (x1, x2 ) ∈ R2 \ {0}. (1.4.272)
x2 + λx1

Since the denominator never vanishes for x ∈ R2 \ {0}, given that we presently
assume λ ∈ C \ R, it follows that ϕ A is well defined and belongs to 𝒞∞ (R2 \ {0}).

Lemma 1.4.33 With ϕ A defined as in (1.4.272) one has

∫
(∂ξA ϕ A)(ξ) dH 1 (ξ) = −2πia2 · sign (Im λ). (1.4.273)
S1

Proof Fix an arbitrary ξ = (ξ1, ξ2 ) ∈ R2 \ {0}. As seen from (1.4.272) we then have
 ξ2 −ξ1 
∇ϕ A(ξ) = a , . (1.4.274)
(ξ2 + λξ1 )2 (ξ2 + λξ1 )2
In concert with (1.4.249) this implies
 ξ1 ξ2 λξ12 λξ22 λ2 ξ1 ξ2 
(∂ξA ϕ A)(ξ) = a2 + − −
(ξ2 + λξ1 )2 (ξ2 + λξ1 )2 (ξ2 + λξ1 )2 (ξ2 + λξ1 )2
 ξ1 ξ2 (1 − λ2 ) + (ξ 2 − ξ 2 )λ 
1 2
= a2
(ξ2 + λξ1 )2
ξ1 − λξ2
= a2 , (1.4.275)
ξ2 + λξ1

where the last inequality holds since ξ2 + λξ1  0 given that ξ = (ξ1, ξ2 ) ∈ R2 \ {0}
and λ ∈ C \ R. Together with Lemma 1.4.26, this permits us to compute
∫
(∂ξA ϕ A)(ξ) dH 1 (ξ) = a2 Iλ = −2πia2 · sign (Im λ), (1.4.276)
S1

completing the proof. 

Here is the result giving an explicit formula for the fundamental solution con-
structed in Theorem 1.4.2 for differential operators of the form L A := ∇ · A∇ for a
weakly elliptic matrix A ∈ C2×2 satisfying sym A ∈ M2± with λ1 = λ2 .
108 1 Integral Representations and Integral Identities

Proposition 1.4.34 Let A ∈ C2×2 be a weakly elliptic matrix with the property that
its symmetric part, sym A := 12 (A + A ), belongs to M2± and can be written as in
(1.4.271). In relation to this, define

sign (Im λ) x1
E A(x) := − , x = (x1, x2 ) ∈ R2 \ {0}. (1.4.277)
2πia x2 + λx1
∞ (R2, L 2 ) and, when viewed as a distribution in R2 , this is a funda-
Then E A ∈ Lloc
mental solution of L A := ∇ · A∇ (the second-order differential operator associated
with A as in (1.4.176)). Additionally, E A is positive homogeneous of degree zero (in
particular, E A is bounded) in R2 \ {0}, and for each multi-index α ∈ N20 one has
 
(∂ α E A)(x) = O |x| − |α | as |x| → ∞. (1.4.278)

Finally, there exists C A ∈ C such that E = E A + C A, where E is the canonical

fundamental solution of L = L A constructed in Theorem 1.4.2.

In relation to the fundamental solutions for “generic” weakly elliptic systems

constructed in Theorem 1.4.2, a significant feature of E A from (1.4.277) is that this
function is positive homogeneous of degree zero (as a consequence, E A is bounded
in R2 \ {0}). In light of (1.4.177) and Lemma 1.4.19, this is consistent with the
prediction made in (1.4.54). Note that E A from (1.4.277) is singular at the origin,
but otherwise 𝒞∞ smooth and bounded in R2 \ {0}. As with E A from (1.4.268), this
contrasts markedly with the nature of the fundamental solution for the Laplacian (in
fact such singularities are removable for harmonic functions).
In relation to Proposition 1.4.34 it is also instructive to consider the special case
 
1/4 i/4
A := , (1.4.279)
i/4 −1/4

in which scenario
 
L A := ∇ · A∇ = 1
4 ∂x21 + i∂x1 ∂x2 + i∂x2 ∂x1 − ∂x22 = ∂z2 (1.4.280)
 
becomes the square of the Cauchy-Riemann operator ∂z = 1
2 ∂x1 + i∂x2 . All funda-
mental solutions of ∂z2 are of the form

z
+ c for some c ∈ C. (1.4.281)
πz
Under the canonical identification z = x1 + ix2 and bearing in mind that we presently
have a = 1/4 and λ = −i, the fundamental solution E A from (1.4.277) takes the
format predicted in (1.4.281) with c = −1/π.
Proof of Proposition 1.4.34 First observe from (1.4.277) that

E A ∈ 𝒞∞ (R2 \ {0}) ∩ Lloc

∞
(R2, L 2 ) ∩ 𝒮 (R2 ). (1.4.282)
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 109

Also, as seen from (1.4.272) and (1.4.274), the first-order derivatives of E A are
positive homogeneous of degree −1 in R2 \ {0}. In addition, elementary calculus
shows that L A E A(x) = 0 for every x ∈ R2 \ {0}. These and Proposition 1.4.1 then
imply that E A is a fundamental solution of L A. Granted this, item (8) in Theorem 1.4.2
gives that Q := E A −E is a polynomial, where E is the canonical fundamental solution
associated with L = L A as in Theorem 1.4.2. Note that E has at most logarithmic
growth, and that E A is bounded. As such, Q must be constant, hence E = E A + C A
for some constant C A ∈ C. Finally, that the function E A is positive homogeneous of
degree zero and satisfies (1.4.278) is clear from (1.4.277) and (1.4.252). 

This completes the task of providing explicit formulas for the fundamental solution
constructed in Theorem 1.4.2 for differential operators of the form L A := ∇ · A∇
with A ∈ M2 . Relying on what we have proved so far, we next carry out a similar
program in higher dimensions.

Fundamental Solutions for Weakly Elliptic Scalar Operators when n ≥ 3

For each n ∈ N define
 
Mn := A ∈ Cn×n : A = A and A is weakly elliptic . (1.4.283)

In the proof of the lemma below we employ ideas from the proof of [88, Propo-
sition 1.1, p. 109] (cf. also [142, pp. 159–160]).

Lemma 1.4.35 Fix an integer n ≥ 3 and pick an arbitrary matrix A ∈ Mn . Then for
each given pair of linearly independent vectors ξ, η ∈ Rn the 2 × 2 matrix
 
Aξ, ξ Aξ, η
Bξ,η := belongs to M20 . (1.4.284)
Aξ, η Aη, η

As a corollary of this membership and Lemma 1.4.20 (applied to the matrix Bξ,η in
M20 and the vector e2 ∈ R2 \ {0}), one has

Aη, η  tAξ, ξ for every t ∈ (−∞, 0] and

(1.4.285)
each pair of linearly independent vectors ξ, η ∈ Rn .

Proof For each ξ, η ∈ Rn \ {0} define the quadratic polynomial

Q ξ,η (τ) := A(ξ + τη), ξ + τη, τ ∈ C, (1.4.286)

and note that

Q−ξ,η (τ) = Q ξ,η (−τ) for each ξ, η ∈ Rn \ {0} and τ ∈ C. (1.4.287)

Henceforth fix η ∈ Rn \ {0}. For every ξ ∈ Rn \ Rη denote by τ1 (ξ), τ2 (ξ) the

complex roots of the quadratic equation Q ξ,η (τ) = 0. From weakly ellipticity and
110 1 Integral Representations and Integral Identities

(1.4.120) we conclude that τ1 (ξ), τ2 (ξ) ∈ C \ R. Thanks to Lemma 1.4.22 and the
fact that the set Rn \ Rη is connected (since we now assume n ≥ 3), it follows that
the number of roots in, say, C+ is independent of ξ ∈ Rn \ Rη (as a continuous
integer-valued function defined on a connected set). Note (1.4.287) entails that for
each ξ ∈ Rn \ Rη we have (with the piece of notation introduced in (1.4.216))
 ∗  ∗
τ1 (−ξ), τ2 (−ξ) = − τ1 (ξ), −τ2 (ξ) . (1.4.288)

Collectively, these observations imply that for each ξ ∈ Rn \ Rη we have precisely

one root in C+ and one root in C− . As a consequence,

Im τ1 (ξ) · Im τ2 (ξ) < 0 for every ξ ∈ Rn \ Rη. (1.4.289)

After this preamble, we are ready to proceed with the proof of (1.4.284). To set
the stage, fix η ∈ Rn \ {0} together with ξ ∈ Rn \ Rη, and abbreviate B := Bξ,η . We
shall first prove that B ∈ M2 . To this end, note that B ∈ C2×2 and B = B, by design.
Suppose that there exists x = (x1, x2 ) ∈ R2 \ {0} such that Bx, x = 0. To proceed,
consider the case when x2  0. Set τ := x1 /x2 ∈ R and define x̃ := (1, τ) ∈ R2 .
Then, since A = A,

0 = B x̃, x̃ = Aξ, ξ + 2 Aξ, ητ + Aη, ητ 2

= A(ξ + τη), ξ + τη. (1.4.290)

Since A is weakly elliptic, this forces ξ + τη = 0 (cf. (1.4.120)) which, in turn,

contradicts the assumption that ξ ∈ Rn \ Rη.
Consider next the case when x1  0. In this scenario, let τ := x2 /x1 ∈ R and
define x̃ := (τ, 1). Much as before, we have

0 = B x̃, x̃ = A(η + τξ), η + τξ, (1.4.291)

from which we conclude that η + τξ = 0, thanks to the weak ellipticity of A and

(1.4.120). Again, this contradicts the assumption that ξ ∈ Rn \ Rη. Altogether, the
above analysis shows that if Bx, x  0, hence B is weakly elliptic (cf. (1.4.120)).
This concludes the proof of the fact that B ∈ M2 .
We are left with checking that B ∈ M2 possesses the property stipulated in
(1.4.187) for membership to the class M20 . With this goal in mind, we first apply
Lemma 1.4.18 to B to conclude that there exist aB ∈ C \ {0} and λ1B, λ2B ∈ C \ R
such that

Bx, x = aB (x1 − λ1B x2 )(x1 − λ2B x2 ), ∀x = (x1, x2 ) ∈ C2 . (1.4.292)

Applying this to x := (1, τ) with τ ∈ C arbitrary leads to the conclusion that

aB (1 − λ1B τ)(1 − λ2B τ) = Bx, x = Aξ, ξ + 2 Aξ, ητ + Aη, ητ 2

= A(ξ + τη), ξ + τη = Q ξ,η (τ). (1.4.293)
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 111

Recall that we have shown that the imaginary parts of the roots of the quadratic
polynomial Q ξ,η (τ) have opposite signs. Hence, Im (1/λ1B ) · Im (1/λ2B ) < 0 which
in turns implies Im (λ1B ) · Im (λ2B ) < 0, ultimately proving that B ∈ M20 , as desired.

Given that any n×n symmetric matrix with complex entries is uniquely determined
by its entries lying on or below the main diagonal, we may naturally identify Mn with
an open convex subset of Cn(n+1)/2 . The following result stands in sharp contrast to
its two-dimensional counterpart, discussed in Lemma 1.4.23.

Lemma 1.4.36 Assume n ≥ 3. Then the set Mn is open (in Cn(n+1)/2 ) and path-
connected.

Proof That Mn may be canonically identified with an open subset of Cn(n+1)/2

(under the identification described above) is clear from (1.4.122). To prove the path-
connectivity of Mn , pick an arbitrary matrix A = (a jk )1≤ j,k ≤n ∈ Mn . Then, by weak
ellipticity (cf. (1.4.120)), we have a11 = Ae1, e1  ∈ C \ {0}. For every t ∈ [0, 1]
define A(t) := (1 − t)A + ta11 In×n . This is a continuous path with A(0) = A and
A(1) = a11 In×n . By design, A(t) = A(t) for every t ∈ [0, 1]. Observe that A(t) is
weakly elliptic for every t ∈ [0, 1]. Indeed, if this were not to be the case, we would
be able to find t ∈ [0, 1] and η ∈ Rn \ {0} such that A(t)η, η = 0. On the one hand,
t cannot be 0 or 1, since A and a11 In×n are weakly elliptic. On the other hand, for
each t ∈ (0, 1) we may write

0 = A(t)η, η = (1 − t)Aη, η + ta11 |η| 2 = (1 − t)Aη, η + t|η| 2 Ae1, e1  (1.4.294)

which implies that Aη, η = −(t|η| 2 /(1 − t))Ae1, e1 . Thanks to Lemma 1.4.35, this
further entails η = λe1 for some λ ∈ R \ {0}. But in that case A(t)η, η = λ2 a11  0,
which is a contradiction. We have therefore shown that A(t) ∈ Mn for every t ∈ [0, 1].
To complete the proof, express a11 ∈ C \ {0} as a11 = ρeiθ for some ρ ∈ (0, ∞)
' := ρ1−t eiθ(1−t) In×n for each t ∈ [0, 1],
and θ ∈ [0, 2π]. This permits us to define A(t)
'
which is a continuous path joining A(0) = a11 In×n with A(1) ' = In×n . In addition,
'
A(t) ∈ Mn for every t ∈ [0, 1].
Concatenating the first path with the second paths enable us to connect within Mn
any A ∈ Mn with In×n . This ultimately shows that Mn is indeed path-connected. 

The next lemma elaborates on further algebraic properties of matrices in Mn .

Lemma 1.4.37 Assume n ≥ 3. If A ∈ Mn then A is invertible and A−1 ∈ Mn .

Moreover, if A ∈ Mn is Legendre-Hadamard elliptic then A−1 is also Legendre-
Hadamard elliptic.

Proof Let A ∈ Mn and let ζ = ξ + iη ∈ Cn with ξ, η ∈ Rn . Notice that since A = A

we have

Aζ, ζ = Aξ, ξ + iAη, ξ − iAξ, η + Aη, η = Aξ, ξ + Aη, η. (1.4.295)

This and Lemma 1.4.35 imply that

112 1 Integral Representations and Integral Identities

Aζ, ζ  0 for every ζ ∈ Cn \ {0}. (1.4.296)

The invertibility of A becomes a consequence of (1.4.296). Since A is symmetric,

(A−1 ) = A−1 . Let us now show that A−1 is weakly elliptic. Seeking a contradiction,
suppose the vector ξ ∈ Rn \ {0} satisfies A−1 ξ, ξ = 0. Set ζ := A−1 ξ ∈ Cn \ {0}
and observe that
Aζ, ζ = ξ, A−1 ξ = A−1 ξ, ξ = 0. (1.4.297)
After taking a complex conjugation, this yields Aζ, ζ = 0, in contradiction with
(1.4.296).
The last claim in the statement, to the effect that if A ∈ Mn is Legendre-Hadamard
elliptic then A−1 is also Legendre-Hadamard elliptic, is seen from Corollary 1.4.9.
We next study the numerical range of a matrix A ∈ Mn , when acting on real
vectors.

Lemma 1.4.38 Assume n ≥ 3. Pick some A ∈ Mn and set

7 8
Aξ, ξ
K := : ξ ∈ S n−1 . (1.4.298)
|Aξ, ξ|

Then there exist θ ∈ [0, 2π) and θ ∗ ∈ (0, π/2) such that
 
eiθ K ⊆ eiω : |ω| ≤ θ ∗ . (1.4.299)

Proof We begin by observing that, by definition, K ⊆ S 1 . Since the function

S n−1 ξ −→ Aξ, ξ/|Aξ, ξ| ∈ C (1.4.300)

is continuous (given that A is weakly elliptic), and S n−1 is a compact connected

subset of Rn , it follows that K is a connected compact subset of S 1 . Hence, K is a
closed arc in the unit circle S 1 .
Let us now prove that K does not contain diametrically opposite points. Otherwise,
there would be ξ, η ∈ S n−1 such that

Aξ, ξ Aη, η
=− , (1.4.301)
|Aξ, ξ| |Aη, η|

hence Aη, η = t Aξ, ξ for some t < 0. This and Lemma 1.4.35 then imply that
ξ = λη for some λ ∈ R. Since ξ, η ∈ S n−1 this forces ξ = ±η, and in either case
(1.4.301) cannot happen.
The above argument proves that K is a closed arc contained in an open half circle,
hence there exist θ ∈ [0, 2π) and θ ∗ ∈ (0, π/2) such that (1.4.299) holds. 
As opposed to the two-dimensional case, when n ≥ 3 any matrix A ∈ Mn becomes
Legendre-Hadamard elliptic after multiplication with a suitable complex number.

Proposition 1.4.39 Assume n ≥ 3 and fix some A ∈ Mn . Then there exists θ ∈ [0, 2π)
such that the matrix Aθ := eiθ A is Legendre-Hadamard elliptic.
1.4 Fundamental Solutions for Weakly Elliptic Second-Order Systems of PDE 113

To illustrate the failure of the above result when n = 2, consider the 2 × 2 matrix
 
1 1 −i
A := . (1.4.302)
4 −i −1

This corresponds to (1.4.179) with a = 1/4 and λ1 = λ2 = i, hence A ∈ M2+ ⊆ M2 .

From (1.4.180) we also see that

Aξ, ξ = 4−1 (ξ1 − iξ2 )2, ∀ξ = (ξ1, ξ2 ) ∈ R2 . (1.4.303)

Next, pick an arbitrary α ∈ R and define ξα := (cos α, sin α) ∈ S 1 . If θ ∈ [0, 2π)

is such that the matrix Aθ := eiθ A is Legendre-Hadamard elliptic, then (1.4.123)
ensures the existence of some κ > 0 independent of α such that

ReAθ ξα, ξα  ≥ κ|ξα | 2 = κ. (1.4.304)

Since, as seen from (1.4.303),

Aθ ξα, ξα  = 4−1 eiθ (cos α − i sin α)2 = 4−1 eiθ e−2iα = 4−1 ei(θ−2α), (1.4.305)

we have
ReAθ ξα, ξα  = 4−1 cos(θ − 2α). (1.4.306)
Choosing α := θ/2 − π/4 then contradicts (1.4.304). Ultimately, this shows that

for A ∈ M2 defined in (1.4.302), there is no θ ∈ [0, 2π) such

(1.4.307)
that the matrix Aθ := eiθ A is Legendre-Hadamard elliptic.
Finally, we wish to point out that the second-order operator L A := ∇ · A∇ associated
with A from (1.4.302) is precisely the Bitsadze operator ∂z̄2 .
Here is the proof of Proposition 1.4.39.
Proof of Proposition 1.4.39 Define C A := minξ ∈S n−1 |Aξ, ξ|. By weakly elliptici-
ty, C A is a well-defined number in the interval (0, ∞) (cf. (1.4.122)). Lemma 1.4.38
guarantees the existence of some θ ∈ [0, 2π) and θ ∗ ∈ (0, π/2) such that (1.4.299)
holds. This implies that for every ξ ∈ S n−1 we have
! "
ReAθ ξ, ξ Aξ, ξ
C A−1 ReAθ ξ, ξ ≥ = Re eiθ ≥ cos θ ∗ . (1.4.308)
|Aξ, ξ| |Aξ, ξ|

Setting then κ := C A cos θ ∗ > 0 we conclude that ReAθ ξ, ξ ≥ κ|ξ | 2 for all ξ ∈ Rn ,
which goes to show that Aθ is Legendre-Hadamard elliptic. 
We shall also need the following result concerning the numerical range of the
inverse of the symmetric part of a Legendre-Hadamard elliptic matrix A ∈ Cn×n
with n ≥ 3, acting on real vectors.
Lemma 1.4.40 Assume n ≥ 3 and pick a Legendre-Hadamard elliptic matrix A in
Cn×n . Then sym A := 12 (A + A ) is invertible and there exists ϑ∗ ∈ (0, π/2) such that
114 1 Integral Representations and Integral Identities

for every x ∈ Rn \ {0} one has

   
(sym A)−1 x, x ∈ Σϑ∗ := z ∈ C \ {0} : | arg(z)| < ϑ∗ . (1.4.309)
2−n
In particular, since Σϑ∗ z → z n ∈ C may be defined as a holomorphic function
(given that the set Σϑ∗ ⊆ C \ {0} is simply connected; cf. item (9) in [112, Propo-
  2−n
sition 5.8.1]), one can make sense of the expression (sym A)−1 x, x n for every
x ∈ Rn \ {0}.

Proof From definitions and (1.4.166) we see that sym A belongs to Mn . As such,
Lemma 1.4.37 guarantees that sym A is invertible and (sym A)−1 is also Legendre-
Hadamard elliptic. Then there exist κ ∈ (0, ∞) and C ∈ [κ, ∞) such that
   
 (sym A)−1 x, x  ≤ C and Re (sym A)−1 x, x ≥ κ for all x ∈ S n−1 . (1.4.310)

Hence, if ϑ∗ ∈ (0, π/2) is such that ϑ∗ > arccos(κ/C), then (sym A)−1 x, x ∈ Σϑ∗
for every x ∈ S n−1 . By homogeneity, the same membership then extends to all points
x ∈ Rn \ {0}. 

Finally, we are prepared to state and prove our main result concerning an explicit
formula for the fundamental solution constructed in Theorem 1.4.2 in relation to
differential operators of the form L A := divA∇ = ∇ · A∇ for arbitrary weakly elliptic
matrices A ∈ Cn×n with n ≥ 3.

Proposition 1.4.41 Assume n ≥ 3, and pick an arbitrary weakly elliptic matrix

A ∈ Cn×n . Then there exists an angle θ ∈ [0, 2π) such that sym Aθ is invertible and
Legendre-Hadamard elliptic if Aθ := eiθ A. Moreover, the function

−eiθ   2−n
E A(x) := 5 (sym Aθ )−1 x, x 2 , (1.4.311)
(n − 2)ωn−1 det (sym Aθ )

defined for all x ∈ Rn \ {0}, belongs to Lloc 1 (Rn, L n ) and, when regarded as a

distribution in Rn , this is a fundamental solution for the second-order operator

L A := ∇ · A∇. In fact, E A agrees with the canonical fundamental solution associated
to the weakly elliptic operator L := L A = divA∇ as in Theorem 1.4.2.

In the above context, since sym Aθ is Legendre-Hadamard elliptic and symmetric

(which amounts to simply saying that sym Aθ ∈ Mn ), Proposition 1.4.12 ensures that
 − 1
det (sym Aθ ) 2 is an unambiguously defined complex number. In fact, (1.4.167)
gives the concrete representation
∫
 − 1
det (sym Aθ ) 2 = π −n/2 e− Aθ ξ,ξ  dξ. (1.4.312)
Rn

Proof of Proposition 1.4.41 From (1.4.166), hypotheses, and definitions we see that
sym A ∈ Mn . As such, Proposition 1.4.39 guarantees the existence of some number
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 115

θ ∈ [0, 2π) with the property that eiθ sym A is Legendre-Hadamard elliptic. Abbre-
viate Aθ := eiθ A, so sym Aθ = eiθ sym A is symmetric and Legendre-Hadamard
elliptic. Proposition 1.4.17 then tells us that
−1   2−n
Esym Aθ (x) = 5 (sym Aθ )−1 x, x 2 , (1.4.313)
(n − 2)ωn−1 det (sym Aθ )

defined for all x ∈ Rn \ {0}, is the canonical fundamental solution associated as

in Theorem 1.4.2 with the scalar weakly elliptic second-order differential opera-
tor Lsym Aθ = ∇ · (sym Aθ )∇. Comparing (1.4.313) with (1.4.311) reveals that in
fact E A = eiθ Esym Aθ . Then the following equalities hold in the sense of tempered
distribution:
L A E A = e−iθ L Aθ E A = Lsym Aθ Esym Aθ = δ. (1.4.314)
Hence, E A is a fundamental solution of L A. If E is the canonical fundamental
solution associated with L := L A = divA∇ as in Theorem 1.4.2, then item (8) in
Theorem 1.4.2 implies that Q := E A − E is a polynomial in Rn . Use (1.4.311)
together with item (4) in Theorem 1.4.2 to write

|Q(x)| ≤ |E A(x)| + |E(x)| ≤ C |x| 2−n, ∀x ∈ Rn \ {0}, (1.4.315)

Since n ≥ 3, we necessarily have Q ≡ 0, from which we conclude that E A = E, as

wanted. 

1.5 Boundary Layer Potential Representations for Weakly

Elliptic Second-Order Systems

We start by discussing a version of Green’s third identity, dealing with an integral

representation formula of a function in terms of a double layer, a single layer, and a
volume potential, in a very general geometric and analytic setting.

Theorem 1.5.1 Let Ω ⊆ Rn , where31 n ≥ 3, be an open set with a lower Ahlfors

regular boundary, and with the property that σ := H n−1 ∂Ω is a doubling measure
on ∂Ω. In particular, Ω is a set of locally finite perimeter, and its geometric measure
theoretic outward unit normal ν = (ν1, . . . , νn ) is defined σ-a.e. on ∂∗ Ω. With the
summation convention over repeated indices understood throughout, let
 αβ 
L = ar s ∂r ∂s 1≤α,β ≤M (1.5.1)

be a homogeneous, weakly elliptic, second-order M × M system in Rn , with complex

constant coefficients, and denote by E = (Eγβ )1≤γ,β ≤M the matrix-valued funda-
mental solution associated with L as in Theorem 1.4.2. Finally, fix an aperture
parameter κ > 0.

31 see the last part in the statement for the case n = 2

116 1 Integral Representations and Integral Identities
M
Suppose u = (uβ )1≤β ≤M ∈ Lloc 1 (Ω, L n ) is a vector-valued function satisfying
(with all derivatives taken in the sense of distributions in Ω)
  dy  M  
αβ n M×n
Lu ∈ L 1 Ω, and ar s s β 1≤α ≤M ∈ Lloc (Ω, L )
∂ u 1
,
1 + |y| n−2 1≤r ≤n
κ−n.t.  κ−n.t. 
u∂Ω and (ar s ∂s uβ )∂Ω 1≤α ≤M exist at σ-a.e. point on ∂nta Ω.
αβ

1≤r ≤n
(1.5.2)
In addition, assume the following integrability conditions hold:
∫ ∫  αβ 
(Nκ u)(y) Nκ (ar s ∂s uβ ) (y)
dσ(y) < ∞ and dσ(y) < ∞, (1.5.3)
∂Ω 1 + |y| n−1 ∂Ω 1 + |y| n−2

for each α ∈ {1, . . . , M } and r ∈ {1, . . . , n}. 

κ −n.t. κ −n.t. 
Then for any κ > 0 the nontangential traces u∂Ω and (ar s ∂s uβ )∂Ω 1≤α ≤M
αβ

1≤r ≤n
also exist at σ-a.e. point on ∂nta Ω and are actually independent of κ . Moreover, for
L n -a.e. point x ∈ Ω one has (with absolutely convergent integrals, and with the
dependence on the aperture parameter dropped)
 ∫  n.t.  
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
u(x) = −
∂∗ Ω 1≤γ ≤M
∫  n.t.  
Eγα (x − y)νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
−
∂∗ Ω 1≤γ ≤M
!∫ "
 
+ Eγβ (x − y) Lu β (y) dy (1.5.4)
Ω 1≤γ ≤M

if either Ω is bounded, or ∂Ω is unbounded. In the case when Ω is an exterior

domain, the same conclusion holds under the additional assumption that there exists
λ ∈ (1, ∞) such that
⨏
|u| dL n = o(1) as R → ∞. (1.5.5)
B(0,λ R)\B(0,R)

Furthermore, for each x ∈ Rn \ Ω one has

 ∫  n.t.  
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
0= −
∂∗ Ω 1≤γ ≤M
∫  n.t.  
Eγα (x − y)νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
−
∂∗ Ω 1≤γ ≤M
!∫ "
 
+ Eγβ (x − y) Lu β (y) dy (1.5.6)
Ω 1≤γ ≤M
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 117

with the same caveat as above in the case when Ω is an exterior domain.
Finally, similar results are valid in the case when n = 2 provided either
∫
−1
L(ξ) dH 1 (ξ) = 0 ∈ C M×M , (1.5.7)
S1

or otherwise one assumes that ∂Ω is compact, replaces the first membership in

(1.5.2) by ∫
|(Lu)(y)| ln(2 + |y|) dy < +∞, (1.5.8)
Ω
and, in the case when Ω is an exterior domain, replaces (1.5.5) by (compare with
Theorem 1.5.7)
⨏
 
|u| dL 2 = o ln1R as R → ∞. (1.5.9)
B(0,λ R)\B(0,R)

Before presenting the proof of this theorem we make a number of comments.

First, we note that the integral decay condition (1.5.5) is satisfied if one assumes the
pointwise decay condition

u(x) = o(1) as |x| → ∞. (1.5.10)

In the particular case when the function u is a null-solution of the system L in an

exterior domain Ω, the decay condition (1.5.5) implies (see Lemma 1.5.6) that for
each multi-index α ∈ N0n we actually have

(∂ α u)(x) = o(|x| − |α | ) as |x| → ∞. (1.5.11)

See also Theorem 1.5.7 for a relaxation of the decay condition (1.5.9) for null-
solutions of the system L in exterior domains in the two-dimensional setting.
Second, according to [112, Lemma 3.5.7], the first membership in (1.5.2) is
equivalent with having, for L n -a.e. point x ∈ Rn ,
∫
|(Lu)(y)|
dy < +∞. (1.5.12)
Ω |x − y|
n−2

Moreover, from [112, Lemma 6.2.9] (used with m := n − 2) we see that the first
membership in (1.5.2) is satisfied provided the components of Lu belong to the
Lorentz space L n/2,1 (Ω, L n ). Here we also wish to note that, according to [112,
Lemma 3.5.8], having (1.5.8) is equivalent with the demand that, for L 2 -a.e. point
x ∈ R2 , ∫
  
|(Lu)(y)| 1 +  ln |x − y|  dy < +∞. (1.5.13)
Ω
Third, assuming
1,1 M
u ∈ Wloc (Ω) , (1.5.14)
118 1 Integral Representations and Integral Identities

if ∂Ω is upper Ahlfors regular then the integrability conditions recorded in (1.5.3) are
satisfies whenever Nκ u ∈ L p (∂Ω, σ) with p ∈ [1, ∞) and Nκ (∇u) ∈ L q (∂Ω, σ)
with q ∈ [1, n − 1). Moreover, if (1.5.14) holds and ∂Ω is actually compact, then the
integrability conditions in (1.5.3) are implied by

Nκ u, Nκ (∇u) ∈ L 1 (∂Ω, σ). (1.5.15)

In Remark 1.5.3 we shall show that, even when ∂Ω is compact and (1.5.14) holds,
Theorem 1.5.1 fails if (1.5.3) is relaxed to demanding that Nκ u and Nκ (∇u) belong
to L 1 (∂∗ Ω, σ). This being said, it is possible to further relax (1.5.3) to demanding
instead that for each α ∈ {1, . . . , M } and r ∈ {1, . . . , n} we have
αβ
Nκ u ∈ Lloc
1 (∂Ω, σ), Nκ (ar s ∂s uβ ) ∈ Lloc
1 (∂Ω, σ),

∫   κ−n.t.  
u ∂Ω (y)
dσ(y) < ∞, and
1 + |y| n−1 (1.5.16)
∂∗ Ω
∫   αβ κ−n.t.  
(ar s ∂s uβ )∂Ω
(y)
dσ(y) < ∞.
1 + |y| n−2
∂∗ Ω

Such a version would require using [112, Corollary 1.5.2] (in the version recorded
in [112, (1.5.23)]) in place of [112, Theorem 1.4.1] (as we do in the proof of
Theorem 1.5.1 given below).
Fourth, if [112, Theorem 1.5.1] is employed in lieu of [112, Theorem 1.4.1] in the
proof of Theorem 1.5.1, we may relax the doubling assumption on σ := H n−1 ∂Ω
to simply asking that this is a locally finite measure. In such a scenario, we need
to ask that the aperture parameter κ is sufficiently large (depending on Ω), and the
flexibility of changing κ when considering nontangential boundary traces may be
lost. However, modulo these nuances, the format of the main results (i.e., formulas
(1.5.4), (1.5.6)) remains the same. As discussed in Remark 1.5.2, Theorem 1.5.1
fails without the lower Ahlfors regularity assumption on ∂Ω.
Here is the proof of Theorem 1.5.1.
Proof of Theorem 1.5.1 To get started, from (1.5.3) and [112, Lemma 8.3.1] we see
that
∞ M  αβ  ∞ M×N
u ∈ Lloc (Ω, L n ) and ar s ∂s uβ 1≤α ≤M ∈ Lloc (Ω, L n ) . (1.5.17)
1≤r ≤n

Also, from (1.5.2), (1.5.3), and [112, Corollary 8.9.9] we deduce that for any κ > 0
the nontangential traces
κ −n.t.  κ −n.t. 
u∂Ω and (ar s ∂s uβ )∂Ω 1≤α ≤M exist at σ-a.e. point on ∂nta Ω,
αβ

1≤r ≤n (1.5.18)
and are actually independent of the chosen aperture parameter κ > 0.
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 119

Moreover, with the dependence on κ dropped, based on [112, (8.9.8)], [112,

(8.9.44)], and [112, (8.8.52)] we conclude that for each α ∈ {1, . . . , M } and
r ∈ {1, . . . , n} we have
∫   n.t.  
 u (y)
∂Ω
dσ(y) < +∞ and
∂∗ Ω 1 + |y| n−1
(1.5.19)
∫   αβ n.t.  
 (ar s ∂s uβ ) (y)
∂Ω
dσ(y) < +∞.
∂∗ Ω 1 + |y| n−2

In particular, from (1.5.19) and (1.4.24) it follows that, for each fixed point x ∈ Ω,
the boundary integrals in the first two lines of (1.5.4) are absolutely convergent.
To proceed, fix an arbitrary index γ ∈ {1, . . . , M }, pick a Lebesgue point x ∈ Ω
for the function
 u with the property that (1.5.12) holds, and consider the vector field
Fx = Fs 1≤s ≤n with components (recall that, throughout, the summation convention
over repeated indices is in effect)
βα  αβ 
Fs := −ar s (∂r Eγβ )(x − ·)uα − Eγα (x − ·) asr ∂r uβ
(1.5.20)
at L n -a.e. point in Ω.

Then from (1.5.20) and (1.5.17) we see that

Fx ∈ Lloc
n
1
(Ω, L n ) . (1.5.21)

The next goal is to compute divFx in the sense of distributions in Ω. To this end,
fix an arbitrary scalar-valued function ϕ ∈ 𝒞∞ c (Ω) and write
∫
 
D (Ω) div 
Fx , ϕ D(Ω) = − Fs (y)(∂s ϕ)(y) dy
Ω
∫
βα
= ar s (∂r Eγβ )(x − y)uα (y)(∂s ϕ)(y) dy
Ω
∫
 αβ 
+ Eγα (x − y) asr ∂r uβ (y)(∂s ϕ)(y) dy
Ω

=: I + I I. (1.5.22)

Bring in a scalar-valued function θ ∈ 𝒞∞ (R n ) with the property

 that θ = 0 on B(0, 1)
and θ = 1 on Rn \ B(0, 2). For each ε ∈ 0, 12 dist(x, ∂Ω) define θ ε : Rn → R by
setting
y − x
θ ε (y) := θ for every y ∈ Rn . (1.5.23)
ε
Then θ ε ∈ 𝒞∞ (Rn ) is a bounded function satisfying
120 1 Integral Representations and Integral Identities

lim θ ε (y) = 1 for every y ∈ Rn \ {x}, (1.5.24)

ε→0+
 
and there exists some C ∈ (0, ∞) such that for each ε ∈ 0, 12 dist(x, ∂Ω) we have

1 − θ ε ∈ 𝒞∞
c (Ω), θ ε ≡ 0 on B(x, ε),
supp (∇θ ε ) ⊆ B(x, 2ε) \ B(x, ε), and (1.5.25)
|(∇ j θ ε )(y)| ≤ Cε −j for every j ∈ N0 and y ∈ Rn .

In particular, for each γ, β ∈ {1, . . . , M } we have

Eγβ (x − ·)θ ε ∈ 𝒞∞ (Ω) and Eγβ (x − ·)ϕθ ε ∈ 𝒞∞ c (Ω), plus

similar properties for expressions involving partial derivatives (1.5.26)
acting on either of the intervening functions.
Let us focus on term I from (1.5.22). Relying on (1.5.17), Lebesgue’s Dominated
Convergence Theorem, and integrations by parts (keeping in mind (1.5.26) and the
βα
fact that ar s ∂s uα ∈ Lloc1 (Ω, L n ) for each β ∈ {1, . . . , M } and r ∈ {1, . . . , n}; cf.

(1.5.2)), we may write

∫
βα
I = lim+ ar s (∂r Eγβ )(x − y)uα (y)(∂s ϕ)(y)θ ε (y) dy
ε→0 Ω
∫
βα
= lim+ ar s (∂s ∂r Eγβ )(x − y)uα (y)ϕ(y)θ ε (y) dy
ε→0 Ω
∫
 βα 
− (∂r Eγβ )(x − y) ar s ∂s uα (y)ϕ(y)θ ε (y) dy
Ω
∫ &
βα
− ar s (∂r Eγβ )(x − y)uα (y)ϕ(y)(∂s θ ε )(y) dy . (1.5.27)
Ω

Observe that since (1.4.33) gives that, for each α ∈ {1, . . . , M },

βα
ar s (∂s ∂r Eγβ )(x − y) = 0 for each y ∈ Rn \ {x}, (1.5.28)

the first integral in the curly brackets in (1.5.27) vanishes. As regards the contribution
from the very last integral in (1.5.27), we expand
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 121
∫
βα
lim ar s (∂r Eγβ )(x − y)uα (y)ϕ(y)(∂s θ ε )(y) dy
ε→0+ Ω
∫
βα  
= lim+ ar s (∂r Eγβ )(x − y) (ϕuα )(y) − (ϕuα )(x) (∂s θ ε )(y) dy
ε→0 Ω
 ∫ &
βα
+ lim ar s (∂r Eγβ )(x − y)(∂s θ ε )(y) dy (ϕuα )(x). (1.5.29)
ε→0+ Ω

In relation to this, we note that since x is a Lebesgue point for u, we may estimate
(using (1.4.24) and (1.5.25))
∫   
 βα 
lim sup  ar s (∂r Eγβ )(x − y) (ϕuα )(y) − (ϕuα )(x) (∂s θ ε )(y) dy 
ε→0+ Ω
⨏
 
≤ C lim sup (ϕuα )(y) − (ϕuα )(x) dy = 0. (1.5.30)
ε→0+ B(x,2ε)

Also, (1.4.33) permits us to conclude that for each index α ∈ {1, . . . , M } we have
∫
βα
lim+ ar s (∂r Eγβ )(x − y)(∂s θ ε )(y) dy
ε→0 Ω
∫
βα
= lim+ ar s (∂r Eγβ )(x − y)∂s (θ ε − 1)(y) dy
ε→0 Ω
 
βα
= − lim+ D (Ω) ar s ∂r Eγβ (x − ·) , ∂s (θ ε − 1) D(Ω)
ε→0
 
βα
= lim+ D (Ω) ar s ∂s ∂r Eγβ (x − ·) , θ ε − 1 D(Ω)
ε→0
 
= δαγ lim+ D (Ω) δx, θ ε − 1 D(Ω)
ε→0

= δαγ lim+ (θ ε − 1)(x) = −δαγ . (1.5.31)

ε→0

Altogether,
∫
 βα 
I = (ϕuγ )(x) − lim+ (∂r Eγβ )(x − y) ar s ∂s uα (y)ϕ(y)θ ε (y) dy. (1.5.32)
ε→0 Ω

Next, to treat term I I from (1.5.22), we employ Lebesgue’s Dominated Conver-

gence Theorem and integrations by parts (bearing in mind (1.5.26)) to express
122 1 Integral Representations and Integral Identities
∫
 αβ 
I I = lim+ Eγα (x − y) asr ∂r uβ (y)(∂s ϕ)(y)θ ε (y) dy
ε→0 Ω

 
αβ
= lim+ D (Ω) asr ∂r uβ, ∂s Eγα (x − ·)ϕθ ε D(Ω)
ε→0

∫
 αβ 
− Eγα (x − y) asr ∂r uβ (y)ϕ(y)(∂s θ ε )(y) dy
Ω
∫ &
 αβ 
+ (∂s Eγα )(x − y) asr ∂r uβ (y)ϕ(y)θ ε (y) dy . (1.5.33)
Ω

Upon recalling (1.5.26), we compute

 
αβ
lim+ D (Ω) asr ∂r uβ, ∂s Eγα (x − ·)ϕθ ε D(Ω)
ε→0
 
αβ
= − lim+ D (Ω) asr ∂s ∂r uβ, Eγα (x − ·)ϕθ ε D(Ω)
ε→0
 
= − lim+ D (Ω) (Lu)α, Eγα (x − ·)ϕθ ε D(Ω)
ε→0
∫
= − lim+ Eγα (x − y)(Lu)α (y)ϕ(y)θ ε (y) dy
ε→0 Ω
∫
=− Eγα (x − y)(Lu)α (y)ϕ(y) dy, (1.5.34)
Ω

where the last two equalities are based on (1.5.12) and Lebesgue’s Dominated
Convergence Theorem. Since, thanks to (1.4.24) and (1.5.17), we also have
∫  αβ  
 
lim sup Eγα (x − y) asr ∂r uβ (y)ϕ(y)(∂s θ ε )(y) dy  (1.5.35)
ε→0+ Ω
 ∫   
 αβ  
≤ C lim sup ε 1−n  (ar s ∂s uβ )(y) 1≤α ≤M |ϕ(y)| dy = 0,
ε→0+ B(x,2ε) 1≤r ≤n

it follows that
∫
II = − Eγα (x − y)(Lu)α (y)ϕ(y) dy
Ω
∫
 αβ 
+ lim+ (∂s Eγα )(x − y) asr ∂r uβ (y)ϕ(y)θ ε (y) dy. (1.5.36)
ε→0 Ω

Note that by interchanging α with β, and r with s, the limit in (1.5.36) matches the
one (1.5.32). Bearing this in mind, from (1.5.22), (1.5.36), and (1.5.32) we ultimately
conclude that for each Lebesgue point x ∈ Ω for u such that (1.5.12) holds we have
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 123

divFx = uγ (x)δx − Eγβ (x − ·)(Lu)β in D (Ω), (1.5.37)

hence

divFx ∈ ℰ (Ω) + L 1 (Ω, L n ). (1.5.38)

 
Going further, K := B x, 12 dist(x, ∂Ω) is a compact subset of Ω, and based on
(1.5.20), (1.4.24), and [112, Lemma 8.3.7], for each y ∈ ∂Ω, we may estimate
   
NκΩ\K Fx (y) ≤ C Nκ u (y) · sup |x − z| 1−n
z ∈Γκ (y)\K
   
αβ
+ C Nκ asr ∂r uβ α,s (y) · sup |x − z| 2−n
z ∈Γκ (y)\K
  
αβ    &
Nκ u (y) Nκ asr ∂r uβ α,s (y)
≤C + , (1.5.39)
|x − y| n−1 |x − y| n−2

for some constant C = C(Ω, L, n, κ) ∈ (0, ∞). From (1.5.39), (1.5.3), and [112,
(8.2.26)] it follows that

NκΩ\K Fx ∈ L 1 (∂Ω, σ). (1.5.40)

Moreover, from (1.5.20), [112, (8.9.10)-(8.9.11)], and (1.5.18) we conclude that

n.t.

Fx  exists at σ-a.e. point on ∂nta Ω (1.5.41)
∂Ω

and, in fact, for each s ∈ {1, . . . , n} we have

 n.t.   n.t. 

(y) = −ar s (∂r Eγβ )(x − y) uα ∂Ω (y)
βα
Fx 
∂Ω s
 αβ n.t. 
− Eγα (x − y) (asr ∂r uβ )∂Ω (y) (1.5.42)

at σ-a.e. point y ∈ ∂nta Ω.

Collectively, (1.5.21), (1.5.38), and (1.5.41) ensure that, for each Lebesgue point
x ∈ Ω for u such that (1.5.12) holds, the vector field Fx satisfies the hypotheses of
[112, Theorem 1.4.1] in the case when either Ω is bounded, or ∂Ω is unbounded.
Assuming this is the case, on account of [112, (4.6.19)], (1.5.37), and (1.5.42), the
Divergence Formula [112, (1.4.6)] presently yields
124 1 Integral Representations and Integral Identities
∫
 
uγ (x) − Eγβ (x − y) Lu β (y) dy
Ω
∫
   
= ℰ (Ω) uγ (x)δx, 1 ℰ(Ω) − Eγβ (x − y) Lu β (y) dy
Ω
 
= (𝒞∞ (Ω))∗ uγ (x)δx − Eγβ (x − ·)(Lu)β, 1 𝒞∞
b
(Ω)
b
∫
   n.t. 
= (𝒞∞ (Ω)) divFx, 1
∗ 𝒞∞
b
(Ω) = ν · Fx ∂Ω dσ
b ∂∗ Ω
∫
 n.t. 
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
=−
∂∗ Ω
∫
 αβ n.t. 
− Eγα (x − y)νs (y) (asr ∂r uβ )∂Ω (y) dσ(y), (1.5.43)
∂∗ Ω

for each Lebesgue point x ∈ Ω of the function u with the property that (1.5.12)
holds (hence also for L n -a.e. point x ∈ Ω), and each γ ∈ {1, . . . , M }. After a slight
adjustment in notation, this finishes the proof of (1.5.4) in the case when either Ω is
bounded, or ∂Ω is unbounded.
Consider next the case when Ω is an exterior domain, under the additional as-
sumption that there exists λ ∈ (1, ∞) such that (1.5.5) holds. In such a scenario,
bring in a function ξ ∈ 𝒞∞ c (R ) satisfying ξ ≡ 1 on B(0, 1), ξ ≡ 0 on R \ B(0, λ),
n n

and for every R > 0 define ξR := ξ(·/R) in R . For each fixed R > 0, sufficiently
n

large so that x ∈ B(0, R) and ∂Ω ⊆ B(0, R), run the same argument  as before but,
this time, in place of (1.5.20) consider now the vector field FxR = FsR 1≤s ≤n with
components
βα    αβ 
FsR := ar s ∂r ξR Eγβ (x − ·) uα − ξR Eγα (x − ·) asr ∂r uβ (1.5.44)

defined at L n -a.e. point in Ω. Upon observing that

 
βα
FxR = ξR Fx + ar s (∂r ξR )Eγβ (x − ·)uα (1.5.45)
1≤s ≤n

and that

(∂ μ ξR )E(x − ·) ∈ 𝒞c∞ (Ω)

M×M
for any μ ∈ N0n with | μ| > 0, (1.5.46)

it follows that in place of (1.5.37) we now get (after some algebra that involves
canceling those terms which contain first-order derivatives of u)
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 125

divFxR = uγ (x)δx − ξR Eγβ (x − ·)(Lu)β

βα
− ar s (∂r ξR )(∂s Eγβ )(x − ·)uα
βα
− ar s (∂s ξR )(∂r Eγβ )(x − ·)uα
βα
+ ar s (∂r ∂s ξR )Eγβ (x − ·)uα in D (Ω). (1.5.47)

Consequently,

divFxR ∈ ℰ (Ω) + L 1 (Ω, L n ). (1.5.48)

Also, (1.5.20), [112, (8.9.10)-(8.9.11)], and (1.5.18) ensure that

n.t.

FxR  exists at σ-a.e. point on ∂nta Ω (1.5.49)
∂Ω

Since FxR vanishes identically outside of a bounded subset of Ω, [112, Theorem 1.4.1]
is applicable to this vector field and the Divergence Formula [112, (1.4.6)] gives
(bearing in mind (1.5.46))
∫
 
uγ (x)− ξR (y)Eγβ (x − y) Lu β (y) dy
Ω
∫
βα
− ar s (∂r ξR )(y)(∂s Eγβ )(x − y)uα (y) dy
Ω
∫
βα
− ar s (∂s ξR )(y)(∂r Eγβ )(x − y)uα (y) dy
Ω
∫
βα
+ ar s (∂r ∂s ξR )(y)Eγβ (x − y)uα (y) dy
Ω
∫
   n.t. 
= (𝒞∞ (Ω))∗ divFxR, 1 𝒞∞b (Ω) = ν · FxR ∂Ω dσ
b ∂∗ Ω
∫
 n.t. 
νs (y)ar s ξR (y)(∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
=−
∂∗ Ω
∫
 αβ n.t. 
− ξR (y)Eγα (x − y)νs (y) (asr ∂r uβ )∂Ω (y) dσ(y). (1.5.50)
∂∗ Ω

There remains to pass to limit R → ∞. In this regard, thanks to (1.4.24), (1.5.5),

plus the fact that supRn |∇ξR | ≤ CR−1 and supp(∇ξR ) ⊆ B(0, λ R) \ B(0, R) for
each R > 0, we have
126 1 Integral Representations and Integral Identities
∫ 
 βα 
 ar s (∂s ξR )(y)(∂r Eγβ )(x − y)uα (y) dy 
Ω
∫ 
 βα 
+ ar s (∂r ξR )(y)(∂s Eγβ )(x − y)uα (y) dy 
Ω
⨏
≤C |u| dL n = o(1) as R → ∞. (1.5.51)
B(0,λ R)\B(0,R)

Similarly, since supRn |∇2 ξR | ≤ CR−2 and supp(∇2 ξR ) ⊆ B(0, λ R) \ B(0, R) for
each R > 0, the fact that |E(z)| ≤ C|z| 2−n for each z ∈ Rn \ {0} (cf. (1.4.24)) once
again implies (keeping in mind (1.5.5))
∫ 
 βα 
 ar s (∂r ∂s ξR )(y)Eγβ (x − y)uα (y) dy  = o(1) as R → ∞. (1.5.52)
Ω

Also, (1.5.12), (1.4.24), and Lebesgue’s Dominated Convergence Theorem imply

∫ ∫
   
lim ξR (y)Eγβ (x − y) Lu β (y) dy = Eγβ (x − y) Lu β (y) dy. (1.5.53)
R→∞ Ω Ω

In addition, the fact that ξR ≡ 1 on B(0, R) and the radius R is sufficiently large so
that ∂Ω ⊆ B(0, R) ensure that
∫
 n.t. 
νs (y)ar s ξR (y)(∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
∂∗ Ω
∫
 βα n.t. 
= νs (y)(∂r Eγβ )(x − y) (ar s uα )∂Ω (y) dσ(y), (1.5.54)
∂∗ Ω

as well as
∫
 αβ n.t. 
ξR (y)Eγα (x − y)νs (y) (asr ∂r uβ )∂Ω (y) dσ(y)
∂∗ Ω
∫
 αβ n.t. 
= Eγα (x − y)νs (y) (asr ∂r uβ )∂Ω (y) dσ(y). (1.5.55)
∂∗ Ω

On account of (1.5.50)-(1.5.55), formula (1.5.4) follows in the case when Ω is an

exterior domain. Parenthetically, we wish to remark that  the reason for which we
preferred to work with the truncated vector field FxR = FsR 1≤s ≤n (with components
 
as in (1.5.44)) in place of the earlier vector field Fx = Fs 1≤s ≤n with components as
in (1.5.20), is that the later choice would require (via [112, (1.4.8)]) slightly stronger
decay conditions on the original function u than the ones imposed in (1.5.5). This
being said, if u is a nullsolution of L then such stronger decay conditions are
automatically satisfied (see Lemma 1.5.6), so in  this
 scenario the is no need to resort
to employing the truncated vector field FxR = FsR 1≤s ≤n in place of the original F. 
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 127

Moving on, formula (1.5.6) is established in a similar (and simpler) fashion,

keeping in mind that if x ∈ Rn \ Ω then E(x − ·) is already a 𝒞∞ matrix-valued
function in Ω. As such, there is no need to bring in the cutoff function θ ε and, this
time, the Dirac distribution in (1.5.37) is no longer present.
Finally, the same proof works in the case when n = 2 provided either (1.5.7) holds,
or otherwise we assume ∂Ω is compact, replace the first membership in (1.5.2) by
(1.5.8), and also replace (1.5.5) by (1.5.9) (such changes now allow us to justify
(1.5.52), keeping (1.4.24) in mind). In this scenario, we start by considering a point
x ∈ R2 with the property (1.5.13) holds and, in the case of (1.5.4), we also demand
that x ∈ Ω is a Lebesgue point for the function u. 

Remark 1.5.2 Theorem 1.5.1 fails without the lower Ahlfors regularity assumption
on ∂Ω.

To see that this is the case, consider the (scalar) differential operator L := Δ, the
Laplacian in Rn , with fundamental solution
⎧
⎪ 1 1
⎪
⎪ if n ≥ 3,
⎪
⎨ ωn−1 (2 − n) |x| n−2
⎪
EΔ (x) := (1.5.56)
⎪
⎪
⎪
⎪ 1
⎪ ln |x| if n = 2.
⎩ 2π
In particular,
1 x
(∇EΔ )(x) = · for x ∈ Rn \ {0}. (1.5.57)
ωn−1 |x| n
Next, fix n ∈ N with n ≥ 3 and consider the open subset of Rn given by

Ω := B(0, 1) \ {0}. (1.5.58)

Note that the set ∂Ω = ∂B(0, 1) ∪ {0} is upper Ahlfors regular, hence the measure
σ := H n−1 ∂Ω is locally finite on ∂Ω. Also, ∂∗ Ω = ∂B(0, 1) so, in particular,
H n−1 (∂Ω \ ∂∗ Ω) = 0. However, ∂Ω is not lower Ahlfors regular.
Finally, having fixed some j ∈ {1, . . . , n}, let us define the function u : Ω → R
by setting
xj
u(x) := for each x ∈ Ω. (1.5.59)
|x| n
Since u is a constant multiple of ∂j EΔ , it follows that u is harmonic in Ω. Also, a
direct computation gives
δ x j xk 
jk
(∇u)(x) = − n n+2 for x = (x1, . . . , xn ) ∈ Ω. (1.5.60)
|x| n |x| 1≤k ≤n

Based on these observations, it is easy to check that the function u satisfies the
hypotheses of Theorem 1.5.1.
Fix x ∈ Ω. The goal is to compute the integrals in the right-hand side of (1.5.4)
for this choice of Ω, u, and x. To proceed, fix ε ∈ (0, |x|/4) and define the auxiliary
128 1 Integral Representations and Integral Identities

domain
 
Ωε := Ω \ B(0, ε) ∪ B(x, ε) . (1.5.61)

To simplify notation, introduce

∫
 n.t. 
1
ID := νD (y) · ∇y E(x − y) u∂D (y) dH n−1 (y), (1.5.62)
∂D
∫
 n.t. 
2
ID := E(x − y) νD (y) · (∇u)∂D (y) dH n−1 (y), (1.5.63)
∂D

where D is one of the balls B(0, 1), B(0, ε), or B(x, ε), νD is the outward unit normal
to the ball D, and the nontangential boundary traces are taken from within D. Note
that Ωε is smooth and both u and E(x − ·) are harmonic in Ωε . Thus, employing the
notation in (1.5.62)-(1.5.63), the classical Green’s Theorem in Ωε implies
1
IB(0,1) − IB(0,1)
2
− IB(0,ε)
1
+ IB(0,ε)
2
− IB(x,ε)
1
+ IB(x,ε)
2
= 0. (1.5.64)

Consider first the integrals associated with B(0, ε). Making use of (1.5.57) we
may compute
∫
1 y x − y yj
IB(0,ε) = −
1
· dH n−1 (y)
ωn−1 |y |=ε ε |x − y| n ε n
n ∫
1  zk z j (xk − εzk )
=− dH n−1 (z), (1.5.65)
ωn−1 k=1 |z |=1 |x − εz| n

where in the second equality we have made the change of variables y = εz. Relying
on Lebesgue’s Dominated Convergence Theorem and the fact that
∫
δ jk ωn−1
zk z j dH n−1 (z) = , (1.5.66)
|z |=1 n

(see [109, Proposition 14.69, p. 581] for more general results of this nature) we
further obtain
n ∫
1 xk 
1
lim+ IB(0,ε) =− zk z j dH n−1 (z)
ε→0 ωn−1 |x| n k=1 |z |=1

1 n
xk δ jk ωn−1 1
=− = − u(x). (1.5.67)
ωn−1 k=1
|x| n n n

Relying again on (1.5.60) and the change of variables y = εz we also have

1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 129
∫  yk  δ jk y j yk 
n
1 1
2
IB(0,ε) = − n dH n−1 (y)
(2 − n)ωn−1 |y |=ε |x − y| n−2 k=1 ε ε n ε n+2
∫
n−1 1 zj
= dH n−1 (z). (1.5.68)
(n − 2)ωn−1 |z |=1 |x − εz| n−2 ε
∫
Since z
|z |=1 j
dH n−1 (z) = 0, we may further write
∫ 
n−1 1 1  zj
2
IB(0,ε) = − dH n−1 (z). (1.5.69)
(n − 2)ωn−1 |z |=1 |x − εz| n−2 |x| n−2 ε

Based on the Fundamental Theorem of Calculus and the Mean Value Formula for
integrals, we have
∫ ε  
1 1 d 1
− = dt
|x − εz| n−2 |x| n−2
0 dt |x − tz|
n−2
∫ ε
x − tz
= (n − 2)z · dt
0 |x − tz| n
x − tε z
= ε(n − 2)z · , (1.5.70)
|x − tε z| n

for some tε ∈ (0, ε). Combining (1.5.69), (1.5.70), Lebesgue’s Dominated Conver-
gence Theorem, and (1.5.66), we arrive at
∫
n − 1  xk
n
n−1
2
lim+ IB(0,ε) = z j zk dH n−1 (z) = u(x). (1.5.71)
ε→0 ωn−1 k=1 |x| n |z |=1 n

In the next phase, we treat the integrals over ∂B(x, ε). First, (1.5.57) implies
∫
1 y−x y−x
IB(x,ε) =
1
· n u(y) dH n−1 (y)
ωn−1 |y−x |=ε ε ε
∫
1
= n−1 u(y) − u(x) dH n−1 (y) + u(x). (1.5.72)
ε ωn−1 |y−x |=ε

Since |u(y) − u(x)| ≤ Cε for y ∈ ∂B(x, ε), from (1.5.72) we obtain

lim I 1 = u(x). (1.5.73)

ε→0+ B(x,ε)

2
Second, as regards IB(x,ε) , we have
∫
 2  #∇u# L ∞ (B(x, |x |/4)) dH n−1 (y)
I 
B(x,ε) ≤ (n − 2)ωn−1
−−−−→ 0. (1.5.74)
|y−x |=ε ε n−2 ε→0+
130 1 Integral Representations and Integral Identities

On account of (1.5.67), (1.5.71), (1.5.73), and (1.5.74), by passing to limit ε → 0+

in (1.5.64) we arrive at
1
IB(0,1) − IB(0,1)
2
= 0. (1.5.75)
1
Since IB(0,1) − IB(0,1)
2 actually represents the right-hand side of (1.5.4) in the present
setting, from (1.5.75) we ultimately conclude that the integral representation formula
(1.5.4) does not hold for Ω as in (1.5.58) and u as in (1.5.59). The source of this
failure is the lack of lower Ahlfors regularity for ∂Ω.

Remark 1.5.3 Theorem 1.5.1 fails without (1.5.3). In fact, without (1.5.3), formula
1,1 M
(1.5.4) may fail even if ∂Ω is compact, u ∈ Wloc (Ω) and Nκ u, Nκ (∇u) belong to
L (∂∗ Ω, σ) for each κ > 0.
1

To see that this is the case, define Ω := B(0, 1) \ {x j = 0}. Hence, Ω is an open set
with a compact Ahlfors regular boundary, and such that

∂∗ Ω = ∂B(0, 1). (1.5.76)

For this choice of Ω, take u as in (1.5.59). In particular, u has a smooth extension to

1,1 M
Rn \ {0} which, in light of (1.5.76), shows that u ∈ Wloc (Ω) and Nκ u, Nκ (∇u)
belong to L (∂∗ Ω, σ) for each κ > 0. However, the equality in (1.5.75) shows (bearing
1

(1.5.76) in mind) that formula (1.5.4) does not hold for Ω and u as specified.
We continue by presenting an integral representation formula in an exterior do-
main for null-solutions of a weakly elliptic system with control near the boundary
in terms of the truncated nontangential maximal function and sub-linear growth at
infinity.

Corollary 1.5.4 Let Ω ⊆ Rn , where n ≥ 2, be an unbounded open set with a

compact lower Ahlfors regular boundary, and with the property that σ := H n−1 ∂Ω
is a doubling measure on ∂Ω. In particular, Ω is a set of locally finite perimeter,
and its geometric measure theoretic outward unit normal ν = (ν1, . . . , νn ) is defined
σ-a.e. on ∂∗ Ω. With the summation convention over repeated indices understood
throughout, let  αβ 
L = ar s ∂r ∂s 1≤α,β ≤M (1.5.77)
be a homogeneous, weakly elliptic, second-order M ×M system in Rn , with complex
constant coefficients, and denote by E = (Eγβ )1≤γ,β ≤M the matrix-valued funda-
mental solution associated with L as in Theorem 1.4.2. Finally, fix an aperture
parameter κ > 0.
Suppose u = (uβ )1≤β ≤M is a vector-valued function satisfying

u ∈ 𝒞∞ (Ω)
M
and Lu = 0 in Ω,

κ−n.t.  κ−n.t.  (1.5.78)

u∂Ω and (ar s ∂s uβ )∂Ω 1≤α ≤M exist at σ-a.e. point on ∂nta Ω.
αβ

1≤r ≤n
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 131

In addition, assume that for some truncation parameter ε > 0 the following integra-
bility conditions hold:
∫ ∫
αβ
Nκε u dσ < ∞ and Nκε (ar s ∂s uβ ) dσ < ∞, (1.5.79)
∂Ω ∂Ω

for each α ∈ {1, . . . , M } and r ∈ {1, . . . , n}. Finally, assume that there exists
λ ∈ (1, ∞) such that
⨏
|u| dL n = o(R) as R → ∞. (1.5.80)
B(0,λ R)\B(0,R)

Then for any other given

 aperture parameter κ ∈ (0, ∞) the nontangential bound-
κ −n.t. κ −n.t. 
ary traces u and (ar s ∂s uβ )
αβ
∂Ω ∂Ω
also exist at σ-a.e. point on ∂ Ω
1≤α ≤M nta
1≤r ≤n
and are actually independent of κ . Furthermore, there exists a constant c ∈ C M
with the property that for each point x ∈ Ω one has (with absolutely convergent
integrals, and with the dependence on the aperture parameter dropped)
 ∫  n.t.  
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
u(x) = c + −
∂∗ Ω 1≤γ ≤M
∫  n.t.  
Eγα (x − y)νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
− . (1.5.81)
∂∗ Ω 1≤γ ≤M

Proof We largely reason as in the proof of Theorem 1.5.1. Fix a point x0 ∈ Rn \ Ω

along with an index γ ∈ {1, . . . , M }, both arbitrary. This time, in place of (1.5.44),
we shall consider the vector field Fx,x
R := F
0
xR − FxR , whose s-th component is given,
0
at L -a.e. point in Ω, by
n

βα  
ar s ∂r ξR Eγβ (x − ·) − ξR Eγβ (x0 − ·) uα
   αβ 
− ξR Eγα (x − ·) − Eγα (x0 − ·) asr ∂r uβ (1.5.82)

for each s ∈ {1, . . . , n}. Then Fx,x

R is locally integrable in Ω, and in place of (1.5.47)
0
we now get (bearing in mind that both Lu = 0 and L[E(x0 − ·)] = 0 in Ω):

divFx,x
R
0
= divFxR − divFxR0
βα  
= uγ (x)δx − ar s (∂r ξR ) (∂s Eγβ )(x − ·) − (∂s Eγβ )(x0 − ·) uα
βα  
− ar s (∂s ξR ) (∂r Eγβ )(x − ·) − (∂r Eγβ )(x0 − ·) uα
βα  
+ ar s (∂r ∂s ξR ) Eγβ (x − ·) − Eγβ (x0 − ·) uα in D (Ω). (1.5.83)

As a consequence, for large R > 0 we once again have

132 1 Integral Representations and Integral Identities

divFx,x
R
0
∈ ℰ (Ω) + L 1 (Ω, L n ). (1.5.84)

Also, much as before,

n.t.
R 
Fx,x 0
 exists at σ-a.e. point on ∂nta Ω. (1.5.85)
∂Ω

Finally, since Fx,x

R vanishes identically outside of a bounded subset of Ω, from
 
0

(1.5.79), (1.5.82), and [112, (8.2.28)] we see that if K := B x, 12 dist(x, ∂Ω) then

NκΩ\K Fx,x0 ∈ L 1 (∂Ω, σ). (1.5.86)

Hence, [112, Theorem 1.4.1] is applicable to this vector field and the Divergence
Formula [112, (1.4.6)] presently yields, assuming R > 0 is sufficiently large,
∫
βα  
uγ (x) − ar s (∂r ξR )(y) (∂s Eγβ )(x − y) − (∂s Eγβ )(x0 − y) uα (y) dy
Ω
∫
βα  
− ar s (∂s ξR )(y) (∂r Eγβ )(x − y) − (∂r Eγβ )(x0 − y) uα (y) dy
Ω
∫
βα  
+ ar s (∂r ∂s ξR )(y) Eγβ (x − y) − Eγβ (x0 − y) uα (y) dy
Ω
∫
   R n.t. 
= (𝒞∞ (Ω))∗ divFx,x
R
, 1 𝒞∞b (Ω) = ν · Fx,x  dσ
b 0 0 ∂Ω
∂∗ Ω
∫
βα  
=− νs (y)ar s ξR (y) (∂r Eγβ )(x − y) − (∂r Eγβ )(x0 − y) ×
∂∗ Ω

 n.t. 
× uα ∂Ω (y) dσ(y)
∫
 
− ξR (y) Eγα (x − y) − Eγα (x0 − y) νs (y)×
∂∗ Ω

 αβ n.t. 
× (asr ∂r uβ )∂Ω (y) dσ(y).
∫
 n.t. 
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
= cγ −
∂∗ Ω
∫
 αβ n.t. 
− Eγα (x − y)νs (y) (asr ∂r uβ )∂Ω (y) dσ(y), (1.5.87)
∂∗ Ω

where
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 133
∫
 n.t. 
νs (y)ar s (∂r Eγβ )(x0 − y) uα ∂Ω (y) dσ(y)
βα
cγ :=
∂∗ Ω
∫
 αβ n.t. 
+ Eγα (x0 − y)νs (y) (asr ∂r uβ )∂Ω (y) dσ(y) (1.5.88)
∂∗ Ω

is a constant in C. Let us also note that, as may be seen from Theorem 1.4.2 and the
Mean Value Theorem, there exists C = C(L, x, x0 ) ∈ (0, ∞) with the property that
 
|∇ξR | (∇E)(x − ·) − (∇E)(x0 − ·) ≤ CR−n−1 and
  (1.5.89)
|∇2 ξR | E(x − ·) − E(x0 − ·) ≤ CR−n−1 .

Then (1.5.81) with c := (cγ )1≤γ ≤M ∈ C M follows from (1.5.87) after passing to
limit R → ∞ and invoking (1.5.89) and (1.5.80). 
Moving on, we record the following significant result (which will play a role
shortly, in the proof of Theorem 1.5.7, as well as later).

Proposition 1.5.5 Assume that either

n ≥ 2 and Ω ⊆ Rn is an open set of locally finite
(1.5.90)
perimeter with a compact toplogical boundary,
or
n ≥ 3 and Ω ⊆ Rn is an∫ open set with a lower Ahlfors regular
boundary and such that ∂Ω [1 + |z|]3−2n dH n−1 (z) < +∞. (1.5.91)

In all cases, abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the geometric

measure theoretic outward unit normal to Ω. With the summation convention over
repeated indices understood throughout, let
 αβ 
L = ar s ∂r ∂s 1≤α,β ≤M (1.5.92)

be a homogeneous, weakly elliptic, second-order M × M system in Rn , with complex

constant coefficients, and denote by E = (Eγβ )1≤γ,β ≤M the matrix-valued funda-
mental solution associated with L as in Theorem 1.4.2. Finally, select an arbitrary
index δ ∈ {1, . . . , M }.
Then for any two points x, xo ∈ Rn one has
 ∫ 
βα
− νs (y)ar s (∂r Eγβ )(x − y)Eαδ (y − xo ) dσ(y) (1.5.93)
∂∗ Ω 1≤γ ≤M
∫ 
αβ
= Eγα (x − y)νr (y)ar s (∂s Eβδ )(y − xo ) dσ(y)
∂∗ Ω 1≤γ ≤M

provided either both x, xo belong to Ω, or both x, xo belong to Rn \ Ω. Furthermore,

if x ∈ Ω and xo ∈ Rn \ Ω then in place of (1.5.93) one has
134 1 Integral Representations and Integral Identities
 ∫ 
βα
− νs (y)ar s (∂r Eγβ )(x − y)Eαδ (y − xo ) dσ(y)
∂∗ Ω 1≤γ ≤M
∫ 
αβ
= Eγα (x − y)νr (y)ar s (∂s Eβδ )(y − xo ) dσ(y)
∂∗ Ω 1≤γ ≤M

+ E. δ (x − xo ), (1.5.94)

where E. δ = (Eγδ )1≤γ ≤M .

At least under stronger geometric assumptions than presently assumed, say n ≥ 3

and Ω ⊆ Rn an Ahlfors regular domain, and when x, xo ∈ Ω, formula (1.5.93) is
a consequence of (1.5.6) in Theorem 1.5.1, written for the open set Ω− := Rn \ Ω
M
in place of Ω, and for the vector-valued function u := E. δ (· − xo ) ∈ 𝒞∞ (Ω− ) .
Specifically, [112, Lemma 5.10.9] ensures that Ω− is an Ahlfors regular domain,
whose topological boundary coincides with that of Ω, whose geometric measure
theoretic boundary agrees with that of Ω, and whose geometric measure theoretic
outward unit normal is −ν at σ-a.e. point on ∂Ω. In addition, according to The-
orem 1.4.2 and [112, (8.3.47)], for each aperture parameter κ > 0 there exists
C = C(Ω, κ, L) ∈ (0, ∞) such that
  C
NκΩ− u (z) ≤ for each z ∈ ∂Ω− . (1.5.95)
[1 + |z|]n−2
Since also Lu = 0 in Ω− and u decays at infinity if Ω− happens to be an exterior
domain (cf. Theorem 1.4.2), we conclude that all demands in the statement of
Theorem 1.5.1 are presently satisfied. Then (1.5.6) applies and gives (1.5.93). Also,
in the same setting, formula (1.5.94) is implied by (1.5.4) written for the vector-
M
valued function u := E. δ (· − xo ) ∈ 𝒞∞ (Ω) (which satisfies Lu = 0 in Ω, plus a
nontangential maximal inequality similar to (1.5.95)).
The proof below works in for the more general settings specified in (1.5.90)-
(1.5.91).
Proof of Proposition 1.5.5 We distinguish a number of cases. First, assume

n ≥ 2 and Ω ⊆ Rn is a bounded open set of locally finite perimeter. (1.5.96)

In the aforementioned setting, fix two distinct points x, xo ∈ Ω and two arbitrary

indices γ, δ ∈ {1, . . . , M }. Bring in the vector field F = (Fs )1≤s ≤n with components
given at each point y ∈ Rn \ {x, xo } by
βα
Fs (y) := ar s (∂r Eγβ )(x − y)Eαδ (y − xo )1Ω (y)
αβ
+ Eγα (x − y)asr (∂r Eβδ )(y − xo )1Ω (y). (1.5.97)

When regarded as a Cn -valued function defined L n -a.e. in Rn , it follows that

F ∈ Lcomp
n n
1
(Rn, L n ) ⊆ ℰ (Rn ) , (1.5.98)
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 135

since Ω is presently assumed to be bounded, so F is compactly supported.

To proceed, consider the compactly supported distribution

u := Eγδ (x − xo )δxo − Eγδ (x − xo )δx ∈ ℰ (Rn ) (1.5.99)

and the complex Borel measure in Rn given by

βα
μ := −νs ar s (∂r Eγβ )(x − ·)Eαδ (· − xo )σ∗
αβ
− Eγα (x − ·)νr asr (∂r Eβδ )(· − xo ) σ∗ (1.5.100)

where σ∗ := H n−1 ∂∗ Ω. In relation to these, we claim that

divF = μ + u in D (Rn ). (1.5.101)

To justify this claim, pick an arbitrary open subset W of Ω which  contains x but
does not contain xo . Then, as seen from (1.5.97), the restriction FW (regarded as a
locally integrable function in W) is given by


F = − ar s ∂r Eγβ (x − ·) Eαδ (· − xo )
βα
W


αβ
+ Eγα (x − ·)asr ∂r Eβδ (· − xo ) . (1.5.102)
1≤s ≤n

In the computation below, we treat E(x − ·) as a distribution in W and take advantage

of the fact that E(· − xo ) is 𝒞∞ -smooth in W. Bearing these in mind, we may then
compute
  
div FW = −ar s ∂r ∂s Eγβ (x − ·) Eαδ (· − xo )
βα

βα
− ar s ∂r Eγβ (x − ·) ∂s Eαδ (· − xo )
αβ
+ ∂s Eγα (x − ·) asr ∂r Eβδ (· − xo )
αβ
+ Eγα (x − ·)asr ∂r ∂s Eβδ (· − xo )

=: I + II + III + IV. (1.5.103)

Thanks to (1.4.21) and (1.4.33) we have

I = −Eγδ (x − xo )δx and IV = 0, (1.5.104)

since xo does not belong to W. Also, after interchanging α with β and r with s in II,
we see that
II + III = 0. (1.5.105)
136 1 Integral Representations and Integral Identities

Gathering terms then leads to the conclusion that

   
div FW = −Eγδ (x − xo )δx = (μ + u)W in D (W) (1.5.106)

where we have also taken into account that μW = 0 (see (1.5.100)). In a very similar
fashion, for any open subset Wo of Ω which contains xo but does to contain x we
obtain
   
div FWo = Eγδ (x − xo )δxo = (μ + u)Wo in D (Wo ). (1.5.107)

Finally, consider an open subset W∞ of Rn which does not contain xo and x, and such
that ∂Ω ⊆ W∞ . Then, keeping in mind that distributional differentiation commutes
with restriction to open sets and that σ∗ is supported in W∞ , for each s ∈ {1, . . . , n}
we write (regarding 1Ω as a distribution in Rn )
  
∂s 1Ω W∞
= ∂s 1Ω  = − νs σ∗ 
W∞ W∞

= −νs σ∗ regarded as a distribution in W∞ (1.5.108)

with the third equality provided by [112, (5.6.10)]. Based on (1.5.108), (1.5.97),
and the fact that both functions E(x − ·) and E(· − xo ) are 𝒞∞ -smooth in W∞ , we
compute
     
div FW∞ = −ar s (∂r ∂s Eγβ )(x − ·)Eαδ (· − xo ) 1Ω W∞
βα

  
+ ar s (∂r Eγβ )(x − ·)(∂s Eαδ )(· − xo ) 1Ω W∞
βα

βα
− ar s (∂r Eγβ )(x − ·)Eαδ (· − xo )νs σ∗
  
− (∂s Eγα )(x − ·)asr (∂r Eβδ )(· − xo ) 1Ω W∞
αβ

  
+ Eγα (x − ·)asr (∂r ∂s Eβδ )(· − xo ) 1Ω W∞
αβ

αβ
− Eγα (x − ·)asr (∂r Eβδ )(· − xo )νs σ∗

=: I∞ + II∞ + III∞ + IV∞ + V∞ + VI∞ (1.5.109)

in the sense of distributions in W∞ . Appealing to (1.4.21) and (1.4.33) permits us to

conclude that
I∞ = 0 and V∞ = 0. (1.5.110)
while interchanging α with β and r with s in II∞ , allows us to combine

II∞ + IV∞ = 0. (1.5.111)

Lastly, as seen from (1.5.100),

1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 137

III∞ + VI∞ = μW∞ .
(1.5.112)

In view of the fact that uW∞ = 0 (since xo and x are not contained in W∞ ), from
(1.5.109)-(1.5.112) we see that
   
div FW∞ = (μ + u)W∞ in D (W∞ ). (1.5.113)

Finally, (1.5.106), (1.5.107), (1.5.113), and [109, Proposition 2.52, p. 38] prove the
claim made in (1.5.101).
At this stage, observe that on the one hand we have
   
 
ℰ (R n ) div F, 1 ℰ(R n ) = −[ℰ (R n )] n F, ∇1 [ℰ(R n )] n = 0 (1.5.114)

while, on the other hand, (1.5.101) gives

     

ℰ (R n ) div F, 1 ℰ(R n ) = ℰ (R n ) μ, 1 ℰ(R n ) + ℰ (R n ) u, 1 ℰ(R n ) . (1.5.115)

Consequently,    
ℰ (R n ) μ, 1 ℰ(R n ) + ℰ (Rn ) u, 1 ℰ(Rn ) = 0. (1.5.116)
In this vein, let also observe that (1.5.99) implies
 
ℰ (R n ) u, 1 ℰ(R n ) = Eγδ (x − xo ) − Eγδ (x − xo ) = 0, (1.5.117)

while (1.5.100) gives

∫
  βα
ℰ (R n ) μ, 1 ℰ(R n ) =− νs ar s (∂r Eγβ )(x − ·)Eαδ (· − xo ) dσ
∂∗ Ω
∫
αβ
− Eγα (x − ·)νr asr (∂r Eβδ )(· − xo ) dσ. (1.5.118)
∂∗ Ω

In view of the arbitrariness of the index γ ∈ {1, . . . , M }, formulas (1.5.117)-(1.5.118)

readily imply (1.5.93) (in the case described in (1.5.96)), under the additional as-
sumption that the points x, xo are distinct. Via continuity, this further implies the
version of (1.5.93) when the points x, xo ∈ Ω are identical. The scenario in which
x, xo ∈ Rn \ Ω is handled very similarly, the task being made a little simpler by the
fact that in place of (1.5.99) we now have u = 0. Finally, in the case when x ∈ Ω and
xo ∈ Rn \ Ω the same argument leads to (1.5.94), since in place of (1.5.99) we now
have u = −Eγδ (x − xo )δx ∈ ℰ (Rn ).
The next case we consider is the situation when
n ≥ 2 and Ω ⊆ Rn is a unbounded open set of locally
(1.5.119)
finite perimeter with a compact topological boundary.
In particular, it follows that Ω is an exterior domain. Once again, fix two arbitrary
indices γ, δ ∈ {1, . . . , M }, along two with points x, xo ∈ Ω (the situations when
x, xo ∈ Rn \ Ω, or when x ∈ Ω and xo ∈ Rn \ Ω, are dealt with analogously).
138 1 Integral Representations and Integral Identities

As before, in view of the conclusions we seek, there is no loss of generality in

assuming that said points are distinct. Next, choose a large (open) ball Bo in Rn with
∂Ω ∪ {x, xo } ⊆ Bo , and define the vector field F = (Fs )1≤s ≤n with components
given at each point y ∈ Rn \ {x, xo } by
βα
Fs (y) := ar s (∂r Eγβ )(x − y)Eαδ (y − xo )1Ω∩Bo (y)
αβ
+ Eγα (x − y)asr (∂r Eβδ )(y − xo )1Ω∩Bo (y). (1.5.120)

Such a vector field continues to satisfy (1.5.98). In addition, writing

1Ω∩Bo = 1Ω − 1Rn \Bo = 1Ω + 1Bo − 1 (1.5.121)

permits us to re-run the computation that has led to (1.5.101) (twice: once for the
current Ω, and then for Rn \ Bo ) and conclude that

divF = μ + u + μo in D (Rn ). (1.5.122)

Above, u, μ are as in (1.5.99) and (1.5.100), respectively, while μo is the complex

Borel measure in Rn given by
βα
μo := −Ns ar s (∂r Eγβ )(x − ·)Eαδ (· − xo )σo
αβ
− Eγα (x − ·)Nr asr (∂r Eβδ )(· − xo ) σo (1.5.123)

where σo := H n−1 ∂Bo and N = (Ns )1≤s ≤n is inward unit normal to Bo . As in

(1.5.114), we once again have
 

ℰ (R n ) div F, 1 ℰ(R n ) = 0. (1.5.124)

In view of (1.5.122)-(1.5.123), the contribution from μo in the context of the left-

hand side above is
∫
βα
Ns (y)ar s (∂r Eγβ )(x − y)Eαδ (y − xo ) dσo (y) (1.5.125)
∂Bo
∫
αβ
+ Eγα (x − y)Ns (y)asr (∂r Eβδ )(y − xo ) dσo (y) = 0,
∂Bo

with the final equality coming from what we have proved in the first case (with Ω
replaced by Bo ). Ultimately, this goes to show that the left-hand side of (1.5.124)
amounts to
∫
  βα
n
ℰ (R ) u + μ, 1 n
ℰ(R ) = − νs ar s (∂r Eγβ )(x − ·)Eαδ (· − xo ) dσ
∂∗ Ω
∫
αβ
− Eγα (x − ·)νr asr (∂r Eβδ )(· − xo ) dσ, (1.5.126)
∂∗ Ω
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 139

thanks to (1.5.117)-(1.5.118). Together, (1.5.126) and (1.5.114) then establish

(1.5.93) in the case considered in (1.5.119). This concludes the treatment of the
scenario described in (1.5.90).
Finally, consider the task of proving (1.5.93) in the setting considered in (1.5.91).
To set the stage, fix two arbitrary indices γ, δ ∈ {1, . . . , M } and two points x, xo ∈ Ω
(the scenario in which x, xo ∈ Rn \ Ω goes through virtually verbatim). As before, in
view of the conclusion we seek, there is no loss of generality in assuming that said
points are distinct. Define the vector field F = (Fs )1≤s ≤n with components given at
each point y ∈ Ω \ {x, xo } by
βα
Fs (y) := ar s (∂r Eγβ )(x − y)Eαδ (y − xo )
αβ
+ Eγα (x − y)asr (∂r Eβδ )(y − xo ). (1.5.127)

As a Cn -valued function defined L n -a.e. in Ω, this satisfies

  n
F ∈ Lloc
1
(Rn, L n ) ∩ 𝒞∞ Ω \ {x, xo } . (1.5.128)

Moreover, if K is a compact neighborhood of {x, xo } contained in Ω, then based on

(1.5.127), (1.4.24), and [112, (8.3.47)] we conclude (also keeping in mind that we
are presently assuming n ≥ 3) that for each aperture parameter κ > 0 there exists
some constant C = C(Ω, κ, L) ∈ (0, ∞) such that
  C
NκΩ\K F (z) ≤ for each z ∈ ∂Ω. (1.5.129)
[1 + |z|]2n−3

In particular, NκΩ\K F ∈ L 1 (∂Ω, σ) by the finiteness condition in (1.5.91). Arguing

as in the first case, we see that

divF = Eγδ (x − xo )δxo − Eγδ (x − xo )δx ∈ ℰ (Ω). (1.5.130)

In concert with [112, (4.6.21)], this implies

 
 1 𝒞∞ (Ω) = Eγδ (x − xo ) − Eγδ (x − xo ) = 0.
∗ div F, (1.5.131)
(𝒞b (Ω))
∞
b

Lastly, in the case when Ω is an exterior domain, (1.5.127) and (1.4.24) give (again,
keeping in mind that n ≥ 3)

F(x) = O(|x| 3−2n ) as |x| → ∞. (1.5.132)

Granted these properties, [112, Theorem 1.5.1] applies and the Divergence Formula
[112, (1.5.11)] written for the present vector field reduces, in view of (1.5.131) and
[112, (8.9.10)], to ∫
  
ν · F  dσ = 0.
∂Ω
(1.5.133)
∂∗ Ω
140 1 Integral Representations and Integral Identities

In turn, (1.5.133) and (1.5.127) readily imply (1.5.93) in the scenario described in
(1.5.91). 

We next include a useful result regarding behavior at infinity for null-solutions of

weakly elliptic systems of any order. This has been mentioned, in passing, in relation
to Theorem 1.5.1 (cf. the first comment made after its statement), and will play a
role shortly, in the proof of Theorem 1.5.7 stated a little further.

Lemma 1.5.6 Fix M, m, n ∈ N with n ≥ 2. Suppose Ω ⊆ Rn is an exterior domain,

and assume L is a weakly elliptic homogeneous M × M system of order 2m, with
constant complex coefficients in Rn . Also, suppose f : (0, ∞) → (0, ∞) is a function
with the property that for each fixed λ ∈ (1, ∞) one has f (λR) ≈ f (R), uniformly
in R ∈ (0, ∞) assumed to be sufficiently large (depending on λ). Finally, pick a
M
function u ∈ 𝒞∞ (Ω) satisfying Lu = 0 in Ω. Then the following conditions are
equivalent:
(1) For some (or any) λ ∈ (1, ∞) one has
⨏
 
|u| dL n = o f (R) as R → ∞. (1.5.134)
B(0,λ R)\B(0,R)

(1’) For some (or any) λ ∈ (1, ∞) one has

⨏
 
|∂ α u| dL n = o R− |α | · f (R) as R → ∞,
B(0,λ R)\B(0,R) (1.5.135)
for each multi-index α ∈ N0n .

(2) One has  

u(x) = o f (|x|) as |x| → ∞. (1.5.136)
(2’) One has
 
(∂ α u)(x) = o |x| − |α | · f (|x|) as |x| → ∞,
(1.5.137)
for each multi-index α ∈ N0n .

(3) One has ⨏

 
|u| dH n−1 = o f (R) as R → ∞. (1.5.138)
∂B(0,R)

(3’) One has

⨏
 
|∂ α u| dH n−1 = o R− |α | · f (R) as R → ∞,
∂B(0,R) (1.5.139)
for each multi-index α ∈ N0n .

Furthermore, similar equivalences hold with little “o” replaced by big “O”
throughout. Finally, the same results are true if u is a null-solution in Ω of an
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 141

injectively elliptic, homogeneous, constant (complex) coefficient, first-order system

D in Rn .

Proof Select some Ro ∈ (0, ∞) large enough so that Rn \ B(0, Ro ) ⊆ Ω. Fix some
λ ∈ (1, ∞), pick x ∈ Ω satisfying |x| > λ Ro , then define

R := 2−1 (1 + λ−1 )|x| and r := 4−1 (1 − λ−1 )|x|. (1.5.140)

These choices ensure that B(x, r) ⊆ B(0, λR)\B(0, R) ⊆ Ω. Based on this and interior
estimates for higher-order weakly elliptic systems (cf. [112, Theorem 6.5.7]), for each
multi-index α ∈ N0n we may write
∫ ⨏
α C C
|(∂ u)(x)| ≤ n+ |α | |u| dL ≤ |α |
n
|u| dL n (1.5.141)
r B(x,r) r B(0,λ R)\B(0,R)

for some constant C ∈ (0, ∞) which depends only on n, L, λ, and α. Since R ≈ |x|
and r ≈ |x|, we conclude from (1.5.141) that (1) ⇒ (2’). Via integration, it is clear
that (2’) ⇒ (3’) and (2) ⇒ (1). Employing the spherical version of Fubini’s Theorem
(cf., e.g., [109, Theorem 14.63, p. 577]) permits us to conclude that (3’) ⇒ (1’) and
(3) ⇒ (1). Lastly, the implications (1’) ⇒ (1), (2’) ⇒ (2), and (3’) ⇒ (3’) are trivial.
That similar equivalences hold with little “o” replaced by big “O” throughout,
is justified in a very similar fashion. Finally, that the same results are true for
null-solutions of injectively elliptic, homogeneous, constant (complex) coefficient
first-order systems, is seen from what we have proved so far and Lemma 1.3.5. 

We shall now use Corollary 1.5.4 as a stepping stone to prove integral represen-
tations in exterior domains for null-solutions of weakly elliptic systems which, in
which in place of sub-linear growth at infinity, are now allowed to have growth of
arbitrary order. The reader is reminded that for each number N ∈ R the symbol [N]
denotes the integer part of N (i.e., the largest integer ≤ N), while

N − 1 if N ∈ Z,
N$ := (1.5.142)
[N] if N ∈ R \ Z,

denotes the largest integer strictly less than N.

Theorem 1.5.7 Let Ω ⊆ Rn , where n ≥ 2, be an unbounded open set with a compact

lower Ahlfors regular boundary, and with the property that σ := H n−1 ∂Ω is a
doubling measure on ∂Ω. In particular, Ω is a set of locally finite perimeter, and
its geometric measure theoretic outward unit normal ν = (ν1, . . . , νn ) is defined
σ-a.e. on ∂∗ Ω. With the summation convention over repeated indices understood
throughout, let  αβ 
L = ar s ∂r ∂s 1≤α,β ≤M (1.5.143)
be a homogeneous, weakly elliptic, second-order M × M system in Rn , with com-
plex constant coefficients, and denote by E = (Eγβ )1≤γ,β ≤M the matrix-valued
fundamental solution associated with L as in Theorem 1.4.2. Finally, fix an aper-
142 1 Integral Representations and Integral Identities

ture parameter κ > 0, along with a truncation parameter ε > 0, a dilation factor
λ ∈ (1, ∞), and an exponent N ∈ [0, ∞).
Suppose u = (uβ )1≤β ≤M is a vector-valued function satisfying

u ∈ 𝒞∞ (Ω)
M
and Lu = 0 in Ω,

κ−n.t.  κ−n.t.  (1.5.144)

u∂Ω and (ar s ∂s uβ )∂Ω 1≤α ≤M exist at σ-a.e. point on ∂nta Ω,
αβ

1≤r ≤n

plus ∫ ∫
αβ
Nκε u dσ < ∞ and Nκε (ar s ∂s uβ ) dσ < ∞, (1.5.145)
∂Ω ∂Ω
for each α ∈ {1, . . . , M } and r ∈ {1, . . . , n}, as well as
⨏
|u| dL n = o(R N ) as R → ∞. (1.5.146)
B(0,λ R)\B(0,R)

Then for any other  given aperture parameter κ > 0 the nontangential boundary
κ −n.t. κ −n.t. 
traces u  αβ
and (ar s ∂s uβ )  also exist at σ-a.e. point on ∂ Ω and
∂Ω ∂Ω 1≤α ≤M nta
1≤r ≤n
are actually independent of the parameter κ . Moreover, there exists some (unique)
C M -valued function P in Rn satisfying

LP = 0 ∈ C M in Rn, (1.5.147)

whose scalar components are polynomials of degree32 ≤ N$ (cf. (1.5.142)) and

with the property that for each point x ∈ Ω one has (with absolutely convergent
integrals, and with the dependence on the aperture parameter dropped)
 ∫  n.t.  
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
u(x) = P(x) + −
∂∗ Ω 1≤γ ≤M
∫  n.t.  
Eγα (x − y)νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
− . (1.5.148)
∂∗ Ω 1≤γ ≤M

In fact,
P(x) = o(|x| N ) as |x| → ∞, if n ≥ 3, (1.5.149)
while in the two-dimensional setting
 ∫ 
ln |x| −1
P(x) = Φ(x) − L(ξ) dH 1 (ξ) c + o(|x| N ) as |x| → ∞, (1.5.150)
4π 2 S 1

32 we adopt the convention that a polynomial has a strictly negative degree if and only if said
polynomial is identically zero
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 143

where Φ : R2 \ {0} → C M×M is given in (1.4.23) and where the constant c ∈ C M is

given by
∫  n.t.  
νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
c := . (1.5.151)
∂∗ Ω 1≤α ≤M

In addition, for each point x ∈ Rn \ Ω one has

∫  n.t.  
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
P(x) =
∂∗ Ω 1≤γ ≤M
∫  n.t.  
Eγα (x − y)νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
+ . (1.5.152)
∂∗ Ω 1≤γ ≤M

Finally, if little “o” is replaced by big “O” in (1.5.146) then all conclusions
remain valid with the understanding that (1.5.149)-(1.5.150) now hold with little
“o” replaced by big “O,” and one now has deg P ≤ [N].

Proof The existence and independence of the boundary traces on the aperture pa-
rameter is a consequence of (1.5.144), (1.5.145), and [112, Corollary 8.9.9]. This
takes care of the first claim in the statement.
There remains to establish the integral representation formula claimed in
 
(1.5.148). Pick an arbitrary point x ∈ Ω and define K := B x, 12 dist(x, ∂Ω) . Also,
select an arbitrary index γ ∈ {1, . . . , M }. For each fixed multi-index η ∈ N0n , we next
 
consider the vector field Fx,η = Fs 1≤s ≤n with components (recall that, throughout,
the summation convention over repeated indices is in effect)
βα  αβ 
Fs := −ar s ∂ η (∂r Eγβ )(x − ·) uα − ∂ η Eγα (x − ·) asr ∂r uβ ∈ D (Ω). (1.5.153)

Above, the package of derivatives ∂ η is applied in the sense of distributions in Ω

to (∂r Eγβ )(x − ·) and Eγα (x − ·) which are locally integrable functions in Ω (hence
M
themselves distributions in Ω). Since u ∈ 𝒞∞ (Ω) , it then makes sense to further
multiply the resulting distributions by smooth functions. The end-result is that Fx,η
is a well-defined vector field with components distributions in Ω, i.e.,

Fx,η ∈ D (Ω) .
n
(1.5.154)

In addition, from (1.5.153) we see that


Fx,η Ω\K ∈ 𝒞∞ (Ω \ K) ⊆ Lloc
n n
1
(Ω \ K, L n ) . (1.5.155)

Let us compute the divergence of the vector field Fx,η in the sense of distributions
in Ω. Specifically, divFx,η ∈ D (Ω) is given by
144 1 Integral Representations and Integral Identities

divFx,η = ∂s Fs
βα βα
= ar s ∂ η ∂r ∂s Eγβ (x − ·) uα + ar s ∂ η ∂r Eγβ (x − ·) (∂s uα )
 αβ 
− ∂ η ∂s Eγα (x − ·) asr ∂r uβ
 αβ 
− ∂ η Eγα (x − ·) asr ∂r ∂s uβ . (1.5.156)

In regard to this, note that

 
βα βα
ar s ∂ η ∂r ∂s Eγβ (x − ·) uα = uα ∂ η ar s ∂r ∂s Eγβ (x − ·)

= uα ∂ η δαγ δx ) = uγ ∂ η δx ∈ ℰ (Ω), (1.5.157)

and for each α ∈ {1, . . . , M } we have

αβ
asr ∂r ∂s uβ = (Lu)α = 0 in Ω. (1.5.158)

Also,
βα  αβ 
ar s ∂ η ∂r Eγβ (x − ·) (∂s uα ) − ∂ η ∂s Eγα (x − ·) asr ∂r uβ = 0, (1.5.159)

as may be seen by interchanging α with β and r with s in the first term. Together,
(1.5.156)-(1.5.159) show that

divFx,η = uγ ∂ η δx ∈ ℰ (Ω). (1.5.160)

It is also apparent from (1.5.153) and assumptions that

n.t.

Fx,η  exists at σ-a.e. point on ∂nta Ω (1.5.161)
∂Ω

and for each s ∈ {1, . . . , n} and σ-a.e. point y ∈ ∂nta Ω we actually have
 n.t.   n.t. 

(y) = −(−1) |η | ar s (∂ η ∂r Eγβ )(x − y) uα ∂Ω (y)
βα
Fx,η 
∂Ω s
 αβ n.t. 
− (−1) |η | (∂ η Eγα )(x − y) (asr ∂r uβ )∂Ω (y). (1.5.162)

Next, from (1.5.146) and Lemma 1.5.6 we conclude that

u(x) = o(|x| N ) and (∇u)(x) = o(|x| N −1 ) as |x| → ∞. (1.5.163)

In view of these properties, the estimates in (1.4.24), and the definition in (1.5.153)
we see that  
Fx,η (x) = o |x| N −n+1− |η | as |x| → ∞. (1.5.164)
In particular,
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 145

Fx,η (x) = o(1) as |x| → ∞, if |η| > 0 and |η| ≥ N − n + 1. (1.5.165)

From this, (1.5.155), (1.5.153), and (1.5.145) we then conclude (bearing in mind
that ∂Ω is compact) that

NκΩ\K Fx,η ∈ L 1 (∂Ω, σ) if |η| > 0 and |η| ≥ N − n + 1. (1.5.166)

Let us now fix an arbitrary multi-index η ∈ N0n with |η| > 0 and |η| ≥ N − n + 1.
Collectively, (1.5.154), (1.5.155), (1.5.160), (1.5.161), and (1.5.166) then permit
us to invoke [112, Theorem 1.4.1], and the Divergence Formula recorded in [112,
(1.4.6)] presently gives
   
(∂ η uγ )(x) = ℰ (Ω) δx, ∂ η uγ ℰ(Ω) = (−1) |η | ℰ (Ω) ∂ η δx, uγ ℰ(Ω)
   
= (−1) |η | ℰ (Ω) uγ ∂ η δx, 1 ℰ(Ω) = (−1) |η | ℰ (Ω) divFx,η, 1 ℰ(Ω)
 
= (−1) |η | (𝒞∞ (Ω))∗ divFx,η, 1 𝒞∞b (Ω)
b
∫  n.t. 
= (−1) |η | ν · Fx,η ∂Ω dσ
∂∗ Ω
∫
 n.t. 
νs (y)ar s (∂ η ∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
=−
∂∗ Ω
∫
 αβ n.t. 
− (∂ η Eγα )(x − y)νs (y) (asr ∂r uβ )∂Ω (y) dσ(y), (1.5.167)
∂∗ Ω

on account of (1.5.160), [112, (4.6.21)], and (1.5.162). In turn, from (1.5.167) and
assumptions we see that if we define the function w : Ω → C M by setting, at each
point x ∈ Ω,
∫  n.t.  
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
w(x) := u(x) +
∂∗ Ω 1≤γ ≤M
∫  n.t.  
Eγα (x − y)νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
+ , (1.5.168)
∂∗ Ω 1≤γ ≤M

then
M
w belongs to 𝒞∞ (Ω) and satisfies Lw = 0 in Ω, as well as
(1.5.169)
∂ η w = 0 in Ω for each η ∈ N0n with |η| > 0 and |η| ≥ N − n + 1.

Expanding w locally into a Taylor series (with remainder), the last line in (1.5.169)
implies that the scalar components of w are locally polynomials in Ω. This and a
standard connectivity argument then ultimately show that the scalar components of
w are polynomials in each of the connected component of Ω. Denote by Ω∞ the
unbounded connected component of Ω and
146 1 Integral Representations and Integral Identities

' to be the (unique) extension of w  to a C M -valued function
define P Ω∞
in Rn whose scalar components (P 'α )1≤α ≤M are polynomials with (1.5.170)
complex coefficients (of degree ≤ |η| − 1 in Rn ).

Then L P ' is a C M -valued function in Rn whose scalar components are polynomials

' vanishes
with complex coefficients in Rn , and the first line in (1.5.169) gives that L P
in the (open, nonempty) set Ω∞ . This implies that
' = 0 ∈ C M in Rn .
LP (1.5.171)

If we now define 
' ' in Ω,
u := u − P (1.5.172)
Ω
if follows that    
uΩ∞ = uΩ∞ − P
' ' = (u − w)
Ω∞ Ω∞
in Ω∞ (1.5.173)
which, together with (1.5.168) and (1.4.24), proves that

O(|x| 2−n ) if n ≥ 3,
'
u(x) = as |x| → ∞. (1.5.174)
O(ln |x|) if n = 2,

In light of (1.5.171), (1.5.172), and the original assumptions on u, we then conclude

that '
u satisfies all hypotheses in Corollary 1.5.4. The latter guarantees the existence
of a constant c ∈ C M with the property that at each point x ∈ Ω we have
 ∫  n.t.  
uα ∂Ω (y) dσ(y)
βα
'
u(x) = c + − νs (y)ar s (∂r Eγβ )(x − y) '
∂∗ Ω 1≤γ ≤M
∫  n.t.  
uβ )∂Ω (y) dσ(y)
αβ
− Eγα (x − y)νr (y) (ar s ∂s ' . (1.5.175)
∂∗ Ω 1≤γ ≤M

Going further, from [112, (5.2.3), (5.6.16)] observe that

Ωext := Rn \ Ω is a compact set of locally finite perimeter with the

property that ∂∗ (Ωext ) = ∂∗ Ω, and the geometric measure theoretic (1.5.176)
outward unit normal to Ωext is −ν at σ-a.e. point on ∂∗ Ω.

Granted these properties, for each index γ ∈ {1, . . . , M } and each fixed point x ∈ Ω
we may now apply the De Giorgi-Federer version of the Divergence Formula
  recalled
in [112, Theorem 1.1.1] to the set Ωext and the vector field G x = G s 1≤s ≤n with
components given by
βα αβ
'α − Eγα (x − ·)asr 'β ∈ 𝒞∞ (Ωext ).
G s := −ar s (∂r Eγβ )(x − ·)P ∂r P (1.5.177)

In view of (1.5.171) and (1.4.33) this is a divergence-free vector field in a neighbor-

hood of Ωext and, bearing in mind (1.5.176), we deduce from [112, (1.1.8)] that for
each x ∈ Ω we have
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 147
 ∫ 
  
−
βα 'α 
νs (y)ar s (∂r Eγβ )(x − y) P (y) dσ(y)
∂∗ Ω 1≤γ ≤M
∂∗ Ω
∫  
−
αβ 'β ) (y) dσ(y)
Eγα (x − y)νr (y)ar s (∂s P = 0. (1.5.178)
∂∗ Ω 1≤γ ≤M
∂∗ Ω

In concert, (1.5.172), (1.5.175), (1.5.178), [112, (8.9.10)], and [112, Proposi-

tion 8.8.6] then imply
 ∫  n.t.  
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
' βα
u(x) = c + P(x) + −
∂∗ Ω 1≤γ ≤M
∫  n.t.  
Eγα (x − y)νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
− (1.5.179)
∂∗ Ω 1≤γ ≤M

at each point x ∈ Ω. If we now define

'
P := c + P, (1.5.180)

then P is a C M -valued function in Rn whose scalar components are polynomials

with complex coefficients in Rn . Also, (1.5.171) implies (1.5.147), while (1.5.179)
implies (1.5.148).
Pressing on, observe that in all dimensions n ≥ 2 the last term in (1.5.148) may
be recast as
∫  n.t.  
νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
−E(x)
∂∗ Ω 1≤γ ≤M
∫    n.t.  
Eγα (x − y) − Eγα (x) νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
−
∂∗ Ω 1≤γ ≤M

= −E(x)c + O(|x| 1−n ) as |x| → ∞, (1.5.181)

where c is as in (1.5.151), thanks to (1.4.22) and the Mean Value Theorem. In turn,
from (1.5.148), (1.5.181), (1.5.146), (1.4.24), and (1.4.22) we see (keeping in mind
that N ≥ 0) that the asymptotic expansions (1.5.149)-(1.5.150) are valid (with c as
in (1.5.151)). Since P is a polynomial, from (1.5.149)-(1.5.150) (while also being
mindful that the function Φ is positive homogeneous of degree 0 in R2 \ {0}) we
then conclude that the degree of P is ≤ N$.
Consider next the task of proving the validity of formula (1.5.152) at points in
M
Rn \ Ω. In the case when n ≥ 3, define the function ω = (ωα )1≤α ≤M ∈ 𝒞∞ (Ω)
by setting
148 1 Integral Representations and Integral Identities

ω(x) : = u(x) − P(x)

 ∫  n.t.  
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
= −
∂∗ Ω 1≤γ ≤M
∫  n.t.  
Eγα (x − y)νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
− (1.5.182)
∂∗ Ω 1≤γ ≤M

for each x ∈ Ω. Then (1.4.24) ensures that

ω(x) = O(|x| 2−n ) as |x| → ∞. (1.5.183)

Collectively, (1.5.183), (1.5.147), plus the original assumptions on u ensure that

the hypotheses of Theorem 1.5.1 are satisfied by the function ω, and (1.5.6) gives
(keeping in mind that Lω = 0) that for each x ∈ Rn \ Ω we have
 ∫  n.t.  
νs (y)ar s (∂r Eγβ )(x − y) ωα ∂Ω (y) dσ(y)
βα
0= −
∂∗ Ω 1≤γ ≤M
∫  n.t.  
Eγα (x − y)νr (y) (ar s ∂s ωβ )∂Ω (y) dσ(y)
αβ
−
∂∗ Ω 1≤γ ≤M
 ∫ 
 n.t. 
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
= −
∂∗ Ω 1≤γ ≤M
∫  n.t.  
Eγα (x − y)νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
−
∂∗ Ω 1≤γ ≤M
 ∫ 
  
νs (y)ar s (∂r Eγβ )(x − y) Pα ∂∗ Ω (y) dσ(y)
βα
− −
∂∗ Ω 1≤γ ≤M
∫  
Eγα (x − y)νr (y)ar s (∂s Pβ )∂∗ Ω (y) dσ(y)
αβ
+ . (1.5.184)
∂∗ Ω 1≤γ ≤M

In relation to the last two lines above, we make the claim that
 ∫    
νs (y)ar s (∂r Eγβ )(x − y) Pα ∂∗ Ω (y) dσ(y)
βα
−
∂∗ Ω 1≤γ ≤M
∫  
Eγα (x − y)νr (y)ar s (∂s Pβ )∂∗ Ω (y) dσ(y)
αβ
−
∂∗ Ω 1≤γ ≤M

= −P(x) (1.5.185)

for each x ∈ Rn \ Ω. To justify (1.5.185), fix an arbitrary point x ∈ Rn \ Ω along with

an arbitrary index γ ∈ {1, . . . , M }. Also, bring back the compact set Ωext ⊆ Rn from
(1.5.176) and consider a large
 (open) ball B ⊆ Rn containing Ωext . Next, introduce
the vector field Hx = Hs 1≤s ≤n with scalar components
βα αβ
Hs := −ar s (∂r Eγβ )(x − ·)Pα 1Ωe x t − Eγα (x − ·)asr (∂r Pβ )1Ωe x t (1.5.186)
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 149

and note that

Hx ∈ Lcomp
n n
1
(Rn, L n ) ⊆ ℰ (Rn ) . (1.5.187)

Moreover, based on (1.5.171), (1.4.33), [112, (5.6.10)], and (1.5.176), the same type
of argument as in the proof of (1.5.99) currently gives

divHx = η + μ in D (Rn ) (1.5.188)

where η is the compactly supported distribution given by

η := Pγ (x)δx ∈ ℰ (Rn ) (1.5.189)

and, if σ∗ := H n−1 ∂∗ Ω, then μ is the complex Borel measure in Rn given by

βα αβ
μ := −νs ar s (∂r Eγβ )(x − ·)Pα σ∗ − Eγα (x − ·)νs asr (∂r Pβ )σ∗ . (1.5.190)

Granted these properties, we may apply [112, Proposition 2.8.6] (with O := Rn and
Ω := B), and the Divergence Formula recorded in [112, (2.8.35)] presently gives
  
 
(𝒞∞b (B))∗ divHx B, 1 𝒞∞b (B) = 0. (1.5.191)

At this stage, the claim made in (1.5.185) becomes a consequence of (1.5.191) and
(1.5.188)-(1.5.190). Lastly, (1.5.184) and (1.5.185) establish (1.5.152) in the case
when n ≥ 3.
There remains to justify (1.5.152) in the two-dimensional case. The novelty in
this scenario is that, in contrast to (1.5.183) when n ≥ 3, now ω may not have
enough decay at infinity to justify using Theorem 1.5.1. To remedy this, we fix some
point xo ∈ Rn \ Ω and, with the constant vector c ∈ C M as in (1.5.151), in place of
M
(1.5.182) now define ω ωα )1≤α ≤M ∈ 𝒞∞ (Ω)
' = (' according to

ω
'(x) := u(x) − P(x) + E(x − xo )c (1.5.192)
 ∫  n.t.  
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
= −
∂∗ Ω 1≤γ ≤M
∫    n.t.  
Eγα (x − y) − Eγα (x − xo ) νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
−
∂∗ Ω 1≤γ ≤M

for each x ∈ Ω. Thanks to (1.4.24) and the Mean Value Theorem we have (recall
that we presently assume n = 2)

'(x) = O(|x| −1 ) as |x| → ∞,

ω (1.5.193)

so Theorem 1.5.1 is applicable to the function ω

'. Specifically, formula (1.5.6) written
for the function ω
' presently implies (in view of (1.5.192) and the fact that L ω
' = 0)
that for each x ∈ Rn \ Ω we have
150 1 Integral Representations and Integral Identities
 ∫ 
 n.t. 
νs (y)ar s (∂r Eγβ )(x − y) uα ∂Ω (y) dσ(y)
βα
0= −
∂∗ Ω 1≤γ ≤M
∫  n.t.  
Eγα (x − y)νr (y) (ar s ∂s uβ )∂Ω (y) dσ(y)
αβ
−
∂∗ Ω 1≤γ ≤M
 ∫ 
  
νs (y)ar s (∂r Eγβ )(x − y) Pα ∂∗ Ω (y) dσ(y)
βα
− −
∂∗ Ω 1≤γ ≤M
∫  
Eγα (x − y)νr (y)ar s (∂s Pβ )∂∗ Ω (y) dσ(y)
αβ
+
∂∗ Ω 1≤γ ≤M
 ∫ 
βα
+ − νs (y)ar s (∂r Eγβ )(x − y)Eαδ (y − xo )cδ dσ(y)
∂∗ Ω 1≤γ ≤M
∫ 
αβ
− Eγα (x − y)νr (y)ar s (∂s Eβδ )(y − xo ) dσ(y) . (1.5.194)
∂∗ Ω 1≤γ ≤M

Much as in (1.5.185) (whose proof continues to be valid in the two-dimensional

setting), the combination of lines three and four above amounts to P(x). Also,
Proposition 1.5.5 applies (since in the current setting fits into (1.5.90)) and (1.5.93)
shows that lines five and six in (1.5.194) amount to zero. All things considered,
(1.5.194) becomes (1.5.152), so the latter formula has now been justified in all
dimensions n ≥ 2. Finally, the very last claim in the statement is implicit in what we
have proved already. 

We next take up the task of elucidating the behavior at infinity of a null-solution

of a weakly elliptic second-order system in an exterior domain. Here is our first result
in this regard.

Corollary 1.5.8 Fix n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an exterior domain.

Let L be a weakly elliptic homogeneous second-order M × M system (for some
M ∈ N) with constant complex coefficients in Rn and denote by E the matrix-valued
fundamental solution associated with L as in Theorem 1.4.2. Finally, consider a
M
function u ∈ 𝒞∞ (Ω) satisfying Lu = 0 in Ω as well as
⨏
|u| dL n = o(R N ) as R → ∞, (1.5.195)
B(0,λ R)\B(0,R)

for some λ ∈ (1, ∞) and N ∈ [0, ∞). Then there exist a C M -valued function P in Rn
satisfying
LP = 0 ∈ C M in Rn, (1.5.196)
whose scalar components are polynomials of degree33 ≤ N$ (cf. (1.5.142)), along
with a constant vector c ∈ C M , such that the following asymptotic expansion holds:

33 we adopt the convention that a polynomial has a strictly negative degree if and only if said
polynomial is identically zero
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 151

(∂ α u)(x) = (∂ α P)(x) + (∂ α E)(x)c + O(|x| 1−n− |α | ) as |x| → ∞,

(1.5.197)
for each multi-index α ∈ N0n .

As a consequence, given a multi-index α ∈ N0n , under the assumption that either

∫ −1
|α| > 0, or n ≥ 3, or n = 2 and S 1 L(ξ) dH 1 (ξ) = 0, one has

O(|x| 2−n− |α | ) if |α| ≥ N$ + 1 or N$ + n ≤ 2,
(∂ α u)(x) = N $− |α | )
(1.5.198)
O(|x| if N$ + n ≥ 2,

as |x| → ∞.
Proof In light of the conclusions we seek, there is no loss of generality in assuming
that Ω is the complement of the closure of a large ball centered at origin in Rn , and
M
that actually u ∈ 𝒞∞ (Ω) . Assume this to be the case. Let ν = (ν1, . . . , νn ) be the
outward unit normal to Ω and abbreviate σ := H n−1 ∂Ω. Also, let P be associated
with the function u as in Theorem 1.5.7, and take c ∈ C M to be the opposite of the
vector defined in (1.5.151). As a consequence of (1.5.148) and (1.5.181) we have

u(x) = P(x) + E(x)c (1.5.199)

 ∫ 
βα
+ − νs (y)ar s (∂r Eγβ )(x − y)u(y) dσ(y)
∂Ω 1≤γ ≤M
∫   
αβ
− Eγα (x − y) − Eγα (x) νr (y)ar s (∂s uβ )(y) dσ(y)
∂Ω 1≤γ ≤M

at each point x ∈ Ω. Then the asymptotic expansion claimed in (1.5.197) is seen by

differentiating both sides of formula (1.5.199), and taking into account (1.4.22) and
the Mean Value Theorem.
To justify the claim made in (1.5.198), fix an arbitrary multi-index α ∈ N0n . Since
deg P ≤ N$, we see that (∂ α P)(x) = O(|x| N $− |α | ) as |x| → ∞, and actually
∂ α P = 0 if |α| ≥ N$ + 1. Furthermore, under the additional assumption that either
|α| > 0, or n ≥ 3, we know from (1.4.24) that (∂ α E)(x) = O(|x| 2−n− |α | ) as |x| → ∞.
Keeping these asymptotic formulas in mind, (1.5.198) follows from (1.5.197). 
The version of Corollary 1.5.8 corresponding to functions with at most linear
growth at infinity (i.e., with N ∈ [0, 1] in (1.5.195)) is looked at in greater detail in
the theorem below.
Theorem 1.5.9 Fix n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an exterior domain.
Also, for some M ∈ N, let L be a weakly elliptic homogeneous second-order M × M
M
system with constant complex coefficients in Rn . Given a function u ∈ 𝒞∞ (Ω)
satisfying Lu = 0 in Ω, consider the following conditions:
(1) For some, or any, λ ∈ (1, ∞) one has
⨏
|u| dL n = o(1) as R → ∞. (1.5.200)
B(0,λ R)\B(0,R)
152 1 Integral Representations and Integral Identities

(2) One has

u(x) = o(1) as |x| → ∞. (1.5.201)
(3) One has ⨏
|u| dH n−1 = o(1) as R → ∞. (1.5.202)
∂B(0,R)

(4) One has

u(x) = O(|x| 2−n ) as |x| → ∞. (1.5.203)
(5) For some λ ∈ (1, ∞) and some N ∈ (0, 1] one has
⨏
|u| dL n = o(R N ) as R → ∞. (1.5.204)
B(0,λ R)\B(0,R)

Then
 
conditions (1), (2), (3) are equivalent and imply (∂ α u)(x) = o |x| − |α |
as |x| → ∞ for each multi-index α ∈ N0n , any of conditions (1)-(4)
(1.5.205)
implies (5) and, in turn, condition (5) implies that, for each
 multi-index
α
α ∈ N0 with |α| > 0, we have (∂ u)(x) = O |x|
n 2−n− |α | as |x| → ∞.

Also,
if n ≥ 3 then conditions (1),
 (2), (3), (4) are equivalent, and any
implies that (∂ α u)(x) = O |x| 2−n− |α | as |x| → ∞ for each multi- (1.5.206)
index α ∈ N0n .

Furthermore, for some constant c ∈ C M ,


O(ln |x|) if n = 2,
condition (5) implies u(x) = as |x| → ∞. (1.5.207)
c + O(|x| 2−n ) if n ≥ 3,

Moreover, in the case when n = 2, the following two conditions are equivalent:

u(x) = o(ln |x|) as |x| → ∞, (1.5.208)

u(x) = O(1) as |x| → ∞. (1.5.209)

Also, the following two conditions

the limit lim u(x) exists (in C M ), (1.5.210)

x→∞

there exists some c ∈ C M such that

(1.5.211)
u(x) = c + O(|x| −1 ) as |x| → ∞
are equivalent and either implies
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 153
 
(∂ α u)(x) = O |x| −1− |α | as |x| → ∞
(1.5.212)
for each multi-index α ∈ N20 with |α| > 0.

In particular,

if n = 2 then u(x) = o(1) as |x| → ∞ if and only if u(x) = O(|x| −1 )

as |x| → ∞, and either of these two conditions
 implies that, for each (1.5.213)
multi-index α ∈ N20 , we have, (∂ α u)(x) = O |x| −1− |α | as |x| → ∞.

Finally, if n = 2 then (1.5.208)-(1.5.209) are equivalent to (1.5.210)-(1.5.211) for

any null-solution u of L in an exterior domain in R2 if and only if
∫
−1
the M × M matrix L(ξ) dH 1 (ξ) is invertible (1.5.214)
S1

(with the latter condition always true if the system L is assumed to actually satisfy
the Legendre-Hadamard strong ellipticity condition, and if M = 1 said condition
holds if and only if L = ∇ · A∇ for some A ∈ M20 ; cf. Lemma 1.4.19 and (1.4.186)).

Before presenting the proof of this theorem we wish to comment on a couple of

concrete examples. Consider first the case of the two-dimensional complex Lamé
system Lλ,μ , defined as in (1.3.6) with μ  0 and λ + 2μ  0. According to (1.3.9),
these conditions ensure that Lλ,μ is weakly elliptic. Then a direct computation based
on (1.3.10) and (1.5.66) shows that
∫
−1 λ + 3μ
Lλ,μ (ξ) dH 1 (ξ) = −π I2×2 . (1.5.215)
S1 μ(λ + 2μ)
Consequently, the 2 × 2 matrix
∫
−1
Lλ,μ (ξ) dH 1 (ξ) is invertible if and only if λ + 3μ  0. (1.5.216)
S1

In particular, this shows that in general (1.5.214) is not expected to hold for generic
weakly elliptic systems.
Let us also note that, in the two-dimensional setting, having λ + 3μ = 0 makes
the logarithmic term in Kelvin’s fundamental solution (1.4.72) disappear. As such,
Proposition 1.4.4 implies that whenever μ ∈ C \ {0} and λ = −3μ the matrix-valued
function E = (E jk )1≤ j,k ≤2 whose ( j, k) entry is defined at each point x = (x1, x2 ) in
R2 \ {0} according to
μ+λ x j xk 1 x j xk
E jk (x) = − = (1.5.217)
4π μ(2μ + λ) |x| 2 2π μ |x| 2

is a fundamental solution for the complex Lamé system Lλ,μ associated with the
2×2
Lamé moduli μ, λ as in (1.3.6). In particular, any column of E ∈ 𝒞∞ (R2 \ {0})
is a bounded null-solution of the complex Lamé system Lλ,μ in R2 \ {0} which does
154 1 Integral Representations and Integral Identities

not possess a limit at infinity. This should be contrasted with the fact that (1.5.209)
implies (1.5.210), if (1.5.214) holds (which is not presently the case; cf. (1.5.215)).
Similar phenomena occur for scalar differential operators in the two-dimensional
setting. To illustrate this, consider the Bitsadze’s operator in the complex plane

L := ∂z̄2 . (1.5.218)

This is a homogeneous, constant (complex) coefficient, weakly elliptic, scalar oper-

ator, for which (1.5.214) fails. Indeed, if L is as in (1.5.218) then

L(ξ) = − 14 (ξ1 + iξ2 )2 for each ξ = (ξ1, ξ2 ) ∈ R2, (1.5.219)

which further implies

∫ ∫ ∫
−1 1 dz
L(z) dH 1 (z) = −4 dH 1 (z) = 4i = 0, (1.5.220)
S1 S1 z2 S1 z3
in violation of (1.5.214). At the same time, the function

z
u(z) := for each z ∈ C \ {0} (1.5.221)
z

(which is, up to normalization, a fundamental solution of ∂z2 in C) is a bounded

null-solution of the Bitsadze’s operator L = ∂z̄2 in C \ {0} which does not have a
limit at infinity. Again, this should be contrasted with the fact that (1.5.209) implies
(1.5.210), if (1.5.214) holds (which is not the case here; see (1.5.220)).
Here is the actual proof of Theorem 1.5.9.
Proof of Theorem 1.5.9 First off,  the fact that conditions (1), (2), (3) are equivalent
and imply (∂ α u)(x) = o |x| − |α | as |x| → ∞ for each multi-index α ∈ N0n is clear
from Lemma 1.5.6. Next, it is clear that any of conditions (4) or (1) (therefore also
(2) or (3)) implies (5). In addition, since N$ = 0 (cf. (1.5.142)), Corollary 1.5.8
guarantees that if condition (5) holds then (∂ α u)(x) = O |x| 2−n− |α | as |x| → ∞ for
any multi-index α ∈ N0n with |α| > 0. This completes the proof of (1.5.205).
Next, the fact that if n ≥ 3 condition (1) implies (4) is seen from Corollary 1.5.8
(cf. (1.5.198) written for α = (0, . . . , 0) ∈ N0n , bearing in mind that 0$ = −1).
The opposite implication is trivial (since n ≥ 3), so ultimately
 conditions
 (1)-(4)
are equivalent if n ≥ 3. That (4) implies (∂ α u)(x) = O |x| 2−n− |α | as |x| → ∞ for
any multi-index α ∈ N0n follows from Lemma 1.5.6. This completes the proof of
(1.5.206).
Going further, the implication in (1.5.207) is a consequence of the asymptotic
formula (1.5.197) (written for α = (0, . . . , 0) ∈ N0n ), the fact that the polynomial P
appearing there is presently a constant (since deg P ≤ N$ = 0 given that N ∈ (0, 1]),
and the estimates in (1.4.24).
Clearly, (1.5.209) implies (1.5.208). To prove the opposite implication, first note
that (1.5.208) entails
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 155

u(x) = o(|x| N ) as |x| → ∞, for each fixed N ∈ (0, 1]. (1.5.222)

This permits us to once again invoke Corollary 1.5.8. Bearing in mind that the
polynomial P appearing there is presently a constant (given that deg P ≤ N$ = 0),
the asymptotic formula (1.5.197) written for α = (0, . . . , 0) ∈ N0n becomes

u(x) = P + E(x)c + O(|x| −1 ) as |x| → ∞. (1.5.223)

From this and (1.5.208) we then conclude that E(x)c = o(ln |x|) as |x| → ∞. In
view of (1.4.48) this forces E(x)c = O(1) as |x| → ∞ which, when used back in
formula (1.5.197), proves (1.5.209). Thus, conditions (1.5.208), (1.5.209) are indeed
equivalent.
The fact that conditions (1.5.210), (1.5.211) are also equivalent is established
in a very similar fashion as above, via (1.5.223), now relying on (1.4.47) in place
of (1.4.48). Moreover, (1.5.211) implies that u − c is a null-solution of L which is
O(|x| −1 ) as |x| → ∞. From this and Lemma 1.5.6 we then see that (1.5.212) holds.
Pressing on, the first claim in (1.5.213) is seen from the equivalence of (1.5.210)
with (1.5.211). The second claim in (1.5.213) follows from (1.5.212) if |α| > 0
and is trivial if |α| = 0. Working in the two-dimensional setting, consider now the
issue of whether (1.5.208)-(1.5.209) are equivalent to (1.5.210)-(1.5.211). In one
direction, (1.5.210) obviously implies (1.5.209). In the opposite direction, assume
u(x) = O(1) as |x| → ∞. Then (1.5.222) holds which, as in the case of (1.5.223),
leads to the conclusion that u has the asymptotic expansion

u(x) = P + E(x)c + O(|x| −1 ) as |x| → ∞, (1.5.224)

with P a constant in C M . From this we then conclude that E(x)c = O(1) as |x| → ∞
which, in light of (1.4.48),
∫ 
−1
L(ξ) dH 1 (ξ) c = 0. (1.5.225)
S1

If the invertibility condition (1.5.214) is assumed, (1.5.225) forces c = 0 ∈ C M , a

scenario in which (1.5.224) reduces to (1.5.210).
Finally, while working in the two-dimensional setting, suppose (1.5.208)-(1.5.209)
are equivalent to (1.5.210)-(1.5.211) for any null-solution u of L in an exterior do-
main in R2 , the goal now being to show that the invertibility property (1.5.214)
holds.
Reasoning by contradiction, assume (1.5.209) implies (1.5.210) for any null-
solution u of L in an exterior domain in R2 , yet (1.5.214) fails. The latter property
guarantees the existence of a non-zero vector c ∈ C M for which (1.5.225) holds. If
we define
u(x) := E(x)c for each x ∈ R2 \ {0}, (1.5.226)
M
then u ∈ 𝒞∞ (R2 \ {0}) and (1.4.20) gives Lu = 0 in R2 \ {0}. Moreover, from
(1.5.225) and (1.4.22)-(1.4.23) we see that u(x) = Φ(x)c for each x ∈ R2 \ {0}, so u is
156 1 Integral Representations and Integral Identities

bounded in R2 \ {0}. In particular, (1.5.209) holds for this function which, according
to the current working hypotheses, implies that (1.5.210) holds for the function u
defined in (1.5.226). Together with (1.4.47), this ultimately forces c = 0 ∈ C M , in
contradiction with the fact that the vector c was non-zero to begin with. This finishes
the proof of the very last claim in the statement. 
Remarkably, in addition to (1.5.4) there exists an integral representation formula
for the gradient of a function involving only at most first-order partial derivatives of
the fundamental solution. This is presented in our next theorem in a general geometric
setting (but the result is new even in the smooth case). A further generalization is
presented later, in Theorem 1.6.6.

Theorem 1.5.10 Let Ω ⊆ Rn , where n ≥ 2, be an open set with a lower Ahlfors

regular boundary, and with the property that σ := H n−1 ∂Ω is a doubling measure
on ∂Ω. In particular, Ω is a set of locally finite perimeter, and its geometric measure
theoretic outward unit normal ν = (ν1, . . . , νn ) is defined σ-a.e. on ∂∗ Ω. With the
summation convention over repeated indices understood throughout, let
 αβ 
L = ar s ∂r ∂s 1≤α,β ≤M (1.5.227)

be a homogeneous, weakly elliptic, second-order M × M system in Rn , with com-

plex constant coefficients, and denote by E = (Eγβ )1≤γ,β ≤M the matrix-valued
fundamental solution associated with L as in Theorem 1.4.2. In this setting, assume
1,1 M
u = (uβ )1≤β ≤M ∈ Wloc (Ω) is a vector-valued function which, for some κ > 0,
satisfies
κ−n.t.
(∇u)∂Ω exists at σ-a.e. point on ∂nta Ω,
∫   (1.5.228)
Nκ (∇u) (y)
and dσ(y) < +∞.
∂Ω 1 + |y|
n−1

With Lu considered in the sense of distributions in Ω, suppose also that

  dy  M
Lu ∈ L 1 Ω, . (1.5.229)
1 + |y| n−1
Then the function u is locally Lipschitz in Ω and for any other aperture parameter
κ −n.t.
κ > 0 the nontangential trace (∇u)∂Ω also exists σ-a.e. on ∂nta Ω and is actually
independent of κ . Moreover, with the dependence on the parameter κ dropped and
having fixed  ∈ {1, . . . , n} along with γ ∈ {1, . . . , M } arbitrary, it follows that for
L n -a.e. point x ∈ Ω one has (with absolutely convergent integrals)
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 157
∫
βα
(∂ uγ )(x) = ar s (∂r Eγβ )(x − y)×
∂∗ Ω
  n.t.   n.t.  
× ν (y) (∂s uα )∂Ω (y) − νs (y) (∂ uα )∂Ω (y) dσ(y)
∫
αβ  n.t. 
− (∂ Eγα )(x − y)νr (y)ar s (∂s uβ )∂Ω (y) dσ(y)
∂∗ Ω
∫
 
+ (∂ Eγβ )(x − y) Lu β (y) dy (1.5.230)
Ω

if either Ω is bounded, or ∂Ω is unbounded. In the case when Ω is an exterior

domain, the same conclusion holds true under the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (1.5.231)
B(0,λ R)\B(0,R)

In addition, for each x ∈ Rn \ Ω one has

∫
βα
0= ar s (∂r Eγβ )(x − y)×
∂∗ Ω
  n.t.   n.t.  
× ν (y) (∂s uα )∂Ω (y) − νs (y) (∂ uα )∂Ω (y) dσ(y)
∫
αβ  n.t. 
− (∂ Eγα )(x − y)νr (y)ar s (∂s uβ )∂Ω (y) dσ(y)
∂∗ Ω
∫
 
+ (∂ Eγβ )(x − y) Lu β (y) dy (1.5.232)
Ω

with the same convention as before in the case when Ω is an exterior domain.

Proof Collectively, (1.5.228) and [112, Lemma 8.3.1] imply

1,∞ M M
u ∈ Wloc (Ω) = Liploc (Ω) . (1.5.233)

Also, from (1.5.228) and [112, Corollary 8.9.9] it follows that for any κ > 0 the
nontangential traces
κ −n.t.
(∇u)∂Ω exists at σ-a.e. point on ∂nta Ω and
(1.5.234)
is actually independent of the parameter κ > 0.

To proceed, recall from [112, Lemma 3.5.7] that the membership in (1.5.229) is
equivalent with having for L n -a.e. point x ∈ Rn the finiteness property
∫
|(Lu)(y)|
dy < +∞. (1.5.235)
Ω |x − y|
n−1
158 1 Integral Representations and Integral Identities

Pick some
x ∈ Ω for which (1.5.235) holds and which is a Lebesgue
(1.5.236)
point for the locally integrable C M×M -valued function ∇u.

Also, fix two arbitrary indexes,  ∈ {1, . . . , n} and γ ∈ {1, . . . , M }. Consider then

the vector field defined at L n -a.e. point in Ω as
βα  
Fx : = ar s (∂r Eγβ )(x − ·) (∂s uα )e − (∂ uα )es
αβ
− (∂ Eγα )(x − ·)ar s (∂s uβ )er . (1.5.237)

In particular, from (1.5.237), (1.5.233), and (1.4.24) we see that

Fx ∈ Lloc
n
1
(Ω, L n ) . (1.5.238)

With the goal of computing divFx in the sense of distributions in Ω, fix an arbitrary

scalar-valued function ϕ ∈ 𝒞∞ c (Ω) and write
∫
  βα

D (Ω) div Fx , ϕ D(Ω) = −ar s (∂r Eγβ )(x − y)(∂s uα )(y)(∂ ϕ)(y) dy
Ω
∫
βα
+ ar s (∂r Eγβ )(x − y)(∂ uα )(y)(∂s ϕ)(y) dy
Ω
∫
αβ
+ ar s (∂ Eγα )(x − ·)(∂s uβ )(∂r ϕ)(y) dy
Ω

=: I + II + III. (1.5.239)

Having chosen a scalar-valued function θ ∈ 𝒞∞ (Rn ) with the property

 that θ = 0 on
B(0, 1) and θ = 1 on Rn \ B(0, 2), for each ε ∈ 0, 12 dist(x, ∂Ω) define θ ε : Rn → R
by setting
y − x
θ ε (y) := θ for every y ∈ Rn . (1.5.240)
ε
Then
θ ε ∈ 𝒞∞ (Rn ) is a bounded function,
lim θ ε (y) = 1 for every y ∈ Rn \ {x},
ε→0+
1 − θ ε ∈ 𝒞∞
c (Ω), θ ε ≡ 0 on B(x, ε), (1.5.241)
supp (∇θ ε ) ⊆ B(x, 2ε) \ B(x, ε), and also
supy ∈Rn |(∇ j θ ε )(y)| ≤ Cε −j for all j ∈ N0,

for some constant C ∈ (0, ∞) independent of ε. In particular, for each pair of indices
α, β ∈ {1, . . . , M } we have
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 159

Eαβ (x − ·)θ ε ∈ 𝒞∞ (Ω) and Eαβ (x − ·)ϕθ ε ∈ 𝒞∞ c (Ω), plus

similar properties for expressions involving partial derivatives (1.5.242)
acting on either of the intervening functions.
To further transform the term I from (1.5.239), we employ (1.5.233), Lebesgue’s
Dominated Convergence Theorem, and integrations by parts (keeping in mind
(1.5.242)), on account of which we may write
∫
βα
I = − lim+ ar s (∂r Eγβ )(x − y)(∂s uα )(y)(∂ ϕ)(y)θ ε (y) dy
ε→0 Ω
 ∫
βα
= lim+ − ar s (∂ ∂r Eγβ )(x − y)(∂s uα )(y)ϕ(y)θ ε (y) dy
ε→0 Ω
 
βα
+ ar s D (Ω) ∂ ∂s uα, (∂r Eγβ )(x − ·)ϕθ ε D(Ω)

∫ 
βα
+ ar s (∂r Eγβ )(x − y)(∂s uα )(y)ϕ(y)(∂ θ ε )(y) dy
Ω
 
= lim+ I(1) (2) (3)
ε + Iε + Iε . (1.5.243)
ε→0

Likewise,
 ∫
βα
II = lim+ ar s (∂s ∂r Eγβ )(x − y)(∂ uα )(y)ϕ(y)θ ε (y) dy
ε→0 Ω
 
βα
− ar s D (Ω) ∂ ∂s uα, (∂r Eγβ )(x − ·)ϕθ ε D(Ω)

∫ 
βα
− ar s (∂r Eγβ )(x − y)(∂ uα )(y)ϕ(y)(∂s θ ε )(y) dy
Ω
 
= lim+ II(1) (2) (3)
ε + IIε + IIε (1.5.244)
ε→0

and
 ∫
αβ
III = lim+ ar s (∂r ∂ Eγα )(x − y)(∂s uβ )ϕ(y)θ ε (y) dy
ε→0 Ω
 
αβ
− D (Ω) ar s ∂r ∂s uβ, (∂ Eγα )(x − ·)ϕθ ε D(Ω)
∫ 
αβ
− ar s (∂ Eγα )(x − y)(∂s uβ )ϕ(y)(∂r θ ε )(y) dy
Ω
 
= lim+ III(1) (2) (3)
ε + IIIε + IIIε . (1.5.245)
ε→0
160 1 Integral Representations and Integral Identities

Observe that, after interchanging α and β in III(1)

ε the resulting expression matches
(1)
−Iε . Hence,

I(1) (1)
ε + IIIε = 0 for each ε. (1.5.246)

It is also clear from definitions that

I(2) (2)
ε + IIε = 0 for each ε. (1.5.247)

Furthermore, (1.4.33) gives

II(1)
ε = 0 for each ε. (1.5.248)

Let us also remark that based on (1.5.229), (1.5.235), (1.4.24), and Lebesgue’s
Dominated Convergence Theorem, we obtain
 
αβ
lim+ III(2)
ε = − lim+ D (Ω) ar s ∂r ∂s uβ, (∂ Eγα )(x − ·)ϕθ ε D(Ω)
ε→0 ε→0
 
= − lim+ D (Ω) (Lu)α, (∂ Eγα )(x − ·)ϕθ ε D(Ω)
ε→0
∫
= − lim+ (∂ Eγα )(x − y)ϕ(y)θ ε (y)(Lu)α (y) dy
ε→0 Ω
∫
=− (∂ Eγα )(x − y)ϕ(y)(Lu)α (y) dy. (1.5.249)
Ω

Turning our attention to I(3)

ε from (1.5.243), we expand
∫
βα  
I(3)
ε = ar s (∂r Eγβ )(x − y) (ϕ ∂s uα )(y) − (ϕ ∂s uα )(x) (∂ θ ε )(y) dy
Ω
 ∫ 
βα
+ ar s (∂r Eγβ )(x − y)(∂ θ ε )(y) dy (ϕ ∂s uα )(x). (1.5.250)
Ω

In relation to the first integral above, we note that since (1.5.236) implies that x is a
Lebesgue point for ϕ∇u we may estimate, using (1.4.24) and (1.5.241),
 ∫ 
 βα   
lim sup ar s (∂r Eγβ )(x − y) (ϕ ∂s uα )(y) − (ϕ ∂s uα )(x) (∂ θ ε )(y) dy 
ε→0+ Ω
⨏
 
≤ C lim sup (ϕ ∇u)(y) − (ϕ ∇u)(x) dy = 0. (1.5.251)
ε→0+ B(x,2ε)

As regards the second term in (1.5.250), observe that for any two fixed indexes
α ∈ {1, . . . , M } and s ∈ {1, . . . , n} we have
1.5 Boundary Layer Potential Representations for Weakly Elliptic Second-Order Systems 161
∫
βα
ar s (∂r Eγβ )(x − y)(∂ θ ε )(y) dy
Ω
∫
βα
= ar s (∂r Eγβ )(x − y)∂ (θ ε − 1)(y) dy
Ω
 
βα
= − D (Ω) ar s ∂r Eγβ (x − ·) , ∂ (θ ε − 1) D(Ω)
 
βα
= D (Ω) ar s ∂ ∂r Eγβ (x − ·) , θ ε − 1 D(Ω) . (1.5.252)

As a result, from (1.5.250), (1.5.251), and (1.5.252) we conclude that

 
βα
I(3)
ε = a
D (Ω) r s ∂ ∂r Eγβ (x − ·) , θ ε − 1 +
D(Ω) (ϕ ∂s uα )(x) + o(1) as ε → 0 .
(1.5.253)

In fact, in a very similar fashion we have

 
αβ
III(3)
ε = − D (Ω) ar s ∂r ∂ Eγα (x − ·) , θ ε − 1 D(Ω) (ϕ ∂s uβ )(x) + o(1) as ε → 0
+

(1.5.254)

which, in concert with (1.5.253) (in which we interchange α and β), implies
 
lim+ I(3) (3)
ε + IIIε = 0. (1.5.255)
ε→0

Implementing once more the same type of argument that has produced (1.5.253)
yields, on account of (1.4.33) and (1.5.241),
 
βα
lim+ II(3)
ε = − lim+ D (Ω) ar s ∂r ∂s Eγβ (x − ·) , θ ε − 1 D(Ω) (ϕ ∂ uα )(x)
ε→0 ε→0

= − lim+ δαγ (θ ε − 1)(x)(ϕ ∂ uα )(x) = ϕ(x)(∂ uγ )(x)

ε→0
 
= D (Ω) (∂ uγ )(x)δx, ϕ D(Ω) . (1.5.256)

At this stage, combining (1.5.239), (1.5.243)-(1.5.249), and (1.5.255)-(1.5.256)

we ultimately conclude that for each point x ∈ Ω as in (1.5.236) we have

divFx = (∂ uγ )(x)δx − (∂ Eγα )(x − ·)(Lu)α in D (Ω). (1.5.257)

In particular,

divFx ∈ ℰ (Ω) + L 1 (Ω, L n ). (1.5.258)

 
If we now consider the compact subset of Ω given by K := B x, 12 dist(x, ∂Ω) ,
then based on (1.5.237), (1.4.24), and [112, Lemma 8.3.7] we may estimate
162 1 Integral Representations and Integral Identities
   
NκΩ\K Fx (y) ≤ C Nκ (∇u) (y) · sup |x − z| 1−n
z ∈Γκ (y)\K
 
Nκ (∇u) (y)
≤C , ∀y ∈ ∂Ω, (1.5.259)
|x − y| n−1

for some constant C = C(Ω, L, n, κ) ∈ (0, ∞). From (1.5.259), (1.5.228), and [112,
(8.2.26)] it follows that

NκΩ\K Fx ∈ L 1 (∂Ω, σ), (1.5.260)

while from (1.5.237), [112, (8.9.10)-(8.9.11)], and (1.5.234) we conclude that

n.t.

Fx  exists at σ-a.e. point on ∂nta Ω (1.5.261)
∂Ω

and, thanks to [112, (8.8.52)], at σ-a.e. point y ∈ ∂∗ Ω we have

 n.t. 

ν(y)· Fx  (y)
∂Ω
  n.t.   n.t.  
= ar s (∂r Eγβ )(x − y) ν (y) (∂s uα )∂Ω (y) − νs (y) (∂ uα )∂Ω (y)
βα

αβ  n.t. 
− (∂ Eγα )(x − y)νr (y)ar s (∂s uβ )∂Ω (y). (1.5.262)

Finally, when Ω is an exterior domain, it follows from (1.5.237) and (1.4.24) that
 
Fx (y) = O |y| 1−n |(∇u)(y)| as |y| → ∞. (1.5.263)

Together, (1.5.238), (1.5.258), (1.5.261) (as well as (1.5.231) and (1.5.263) in the
case when Ω is an exterior domain) ensure that, for each point x ∈ Ω as in (1.5.236),
the vector field Fx satisfies the hypotheses of [112, Theorem 1.4.1]. As such, on
account of [112, (4.6.19)], (1.5.257), and (1.5.262), the Divergence Formula [112,
(1.4.6)] currently gives
1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 163
∫
 
(∂ uγ )(x) − Eγα (x − y) Lu α (y) dy
Ω
∫
   
= ℰ (Ω) (∂ uγ )(x)δx, 1 ℰ(Ω) − Eγα (x − y) Lu α (y) dy
Ω
 
= (𝒞∞ (Ω))∗ (∂ uγ )(x)δx − Eγα (x − ·)(Lu)α, 1 𝒞∞
b
(Ω)
b
∫
   n.t. 
= (𝒞∞ (Ω)) divFx, 1
∗ 𝒞∞
b
(Ω) = ν · Fx ∂Ω dσ
b ∂∗ Ω
∫
βα
= ar s (∂r Eγβ )(x − y)×
∂∗ Ω
  n.t.   n.t.  
× ν (y) (∂s uα )∂Ω (y) − νs (y) (∂ uα )∂Ω (y) dσ(y)
∫
αβ  n.t. 
− (∂ Eγα )(x − y)νr (y)ar s (∂s uβ )∂Ω (y) dσ(y) (1.5.264)
∂∗ Ω

for each point x ∈ Ω as in (1.5.236), in particular for L n -a.e. point x ∈ Ω. This

finishes the proof of (1.5.230).
Finally, the validity of (1.5.232) for each x ∈ Rn \ Ω is established in a similar
fashion (this time, the Dirac delta function no longer appears in (1.5.257)). 

1.6 Integral Representation Formulas for Injectively Elliptic

First-Order Systems

Our first theorem in this section contains a basic integral representation formula for
functions which behave reasonably under the action of a given injectively elliptic
first-order system, have a nontangential boundary trace, and with their nontangential
maximal function belonging to a suitable weighted Lebesgue space. A more general
version is discussed later, in Theorem 1.6.5.

Theorem 1.6.1 Let Ω ⊆ Rn , where n ≥ 2, be an open set with a lower Ahlfors

regular boundary, and such that σ := H n−1 ∂Ω is a doubling measure on ∂Ω. In
particular, Ω is a set of locally finite perimeter, and its geometric measure theoretic
outward unit normal ν is defined σ-a.e. on ∂∗ Ω.
Next, consider a homogeneous, first-order N × M system D with constant complex
coefficients in Rn (where N, M ∈ N) which is injectively elliptic (cf. (1.3.18)), and
suppose D ' is a homogeneous first-order M × N system with constant complex coeffi-
'
cients in Rn which complements D (i.e., (1.3.21) holds). In particular, since L := DD
is a weakly elliptic second-order M × M system in R it is meaningful to consider the
n

matrix-valued fundamental solution E L  associated with L  as in Theorem 1.4.2.

Also, with D' acting on the columns of E L  define
164 1 Integral Representations and Integral Identities

' := D
E ' E L  . (1.6.1)

Finally, fix an aperture parameter κ ∈ (0, ∞) and suppose u : Ω → C M is a

vector-valued function with Lebesgue measurable components satisfying34
 
Nκ u ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
,
 κ−n.t. 
u∂Ω (x) exists for σ-a.e. x ∈ ∂nta Ω, (1.6.2)
  dy   N
and Du ∈ L 1 Ω, .
1 + |y| n−1
κ −n.t.
Then for any κ > 0 the nontangential trace u∂Ω also exists at σ-a.e. point
on ∂nta Ω and is actually independent of κ . Moreover, with the dependence on the
parameter κ dropped, for L n -a.e. x ∈ Rn \ ∂Ω one has (with absolutely convergent
integrals)
∫
   n.t. 
E' (x − y)(−i)Sym D; ν(y) u (y) dσ(y)
∂Ω
∂∗ Ω

∫ 
u(x) if x ∈ Ω,
− ' (x − y)(Du)(y) dy =
E (1.6.3)
Ω 0 if x ∈ Rn \ Ω,

if either Ω is bounded, or ∂Ω is unbounded. In the case when Ω is an exterior

domain, the same conclusion holds true under the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|u| dL n = o(1) as R → ∞. (1.6.4)
B(0,λ R)\B(0,R)

' from (1.6.1) are in order.

A couple of comments pertaining to the nature of E
First, Theorem 1.4.2 implies that
N ×M  N ×M
' ∈ 𝒮 (Rn )
E , ' n
E ∈ 𝒞∞ (Rn \ {0}) ,
R \{0}
' is odd and positive homogeneous of degree 1 − n
E (1.6.5)
'
(in particular, |(∂ α E)(x)| ≤ Cα |x| 1−n− |α | for each α ∈ N0n ).

Second,
' is a fundamental solution for D .
E (1.6.6)

More specifically, for each fixed point x ∈ Rn we have (with all differential operators
acting on columns in the “dot” variable)

34 with Du considered in the sense of distributions in Ω

1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 165

' − x) = D ( D
D E(· ' E L  )(· − x) = (D D
' ) E L  (· − x)

= L  E L  (· − x) = δx I M×M , (1.6.7)

where δx is Dirac’s distribution with mass at x, I M×M is the M × M identity matrix,

and the last equality is comes from (1.4.21).
Here is the proof of Theorem 1.6.1.
Proof of Theorem 1.6.1 With the summation convention over repeated indices as-
sumed throughout, suppose
 βα   αγ 
'= '
D a j ∂j 1≤β ≤M and D = ak ∂k 1≤α ≤ N . (1.6.8)
1≤α ≤ N 1≤γ ≤M

Then the homogeneous second-order M × M system L := DD ' may be written

explicitly as
 
' = Aβγ ∂j ∂k
L = DD
βγ βα αγ
jk 1≤β,γ ≤M where each A jk := '
a j ak . (1.6.9)

Hence, L = L A, the second-order homogeneous M × M system associated with the

coefficient tensor
 βγ  βγ βα αγ
A := A jk 1≤ j,k ≤n with A jk := '
a j ak . (1.6.10)
1≤β,γ ≤M

Let us also observe that

       
'= D ' E L  = − ' γα γα
E ak ∂k E L  γδ 1≤α ≤ N = − 'ak ∂k E L δγ 1≤α ≤ N (1.6.11)
1≤δ ≤M 1≤δ ≤M

where the last equality uses (1.4.32).

Going further, observe that the first property in (1.6.2) and [112, Lemma 8.3.1]
guarantee that
∞ N
u ∈ Lloc (Ω, L n ) , (1.6.12)

while the first two properties in (1.6.2) together with [112, Corollary 8.9.9] imply
that
κ −n.t.
for any κ > 0 the nontangential trace u∂Ω exists at σ-a.e. point (1.6.13)
on ∂nta Ω and is actually independent of the parameter κ > 0.

Let us turn our attention to (1.6.3) in earnest. Recall from [112, Lemma 3.5.7]
that the last membership in (1.6.2) is equivalent with having for L n -a.e. point x ∈ Rn
the finiteness property
∫
|(Du)(y)|
dy < +∞. (1.6.14)
Ω |x − y| n−1
166 1 Integral Representations and Integral Identities

Fix an arbitrary index α ∈ {1, . . . , M }, pick a Lebesgue point x ∈ Ω for the function

u with the property that (1.6.14) holds, and consider the vector field Fx = Fj 1≤ j ≤n
with components (recall that, throughout, the summation convention over repeated
indices is in effect)
γβ  
Fj := −Ak j ∂k (E L )αγ (x − ·)uβ at L n -a.e. point in Ω. (1.6.15)

Collectively, (1.6.15), (1.6.12), and Theorem 1.4.2 ensure that

Fx ∈ Lloc
n
1
(Ω, L n ) . (1.6.16)

The next goal is to compute divFx in the sense of distributions in Ω. To this end,
fix an arbitrary scalar-valued function ϕ ∈ 𝒞∞ c (Ω) and write
∫
 

D (Ω) div Fx , ϕ D(Ω) = − Fj (y)(∂j ϕ)(y) dy
Ω
∫
γβ  
= Ak j ∂k (E L )αγ (x − y)uβ (y)(∂j ϕ)(y) dy. (1.6.17)
Ω

Bring in a scalar-valued function θ ∈ 𝒞∞ (R n ) with the property

 that θ = 0 on B(0, 1)
and θ = 1 on Rn \ B(0, 2). For each ε ∈ 0, 12 dist(x, ∂Ω) define θ ε : Rn → R by
setting
y − x
θ ε (y) := θ for every y ∈ Rn . (1.6.18)
ε
Then
θ ε ∈ 𝒞∞ (Rn ) is a bounded function satisfying
(1.6.19)
lim+ θ ε (y) = 1 for every point y ∈ Rn \ {x},
ε→0
 
and there exists a constant C ∈ (0, ∞) such that for each ε ∈ 0, 12 dist(x, ∂Ω) we
have
1 − θ ε ∈ 𝒞∞
c (Ω), θ ε ≡ 0 on B(x, ε),
supp (∇θ ε ) ⊆ B(x, 2ε) \ B(x, ε), and (1.6.20)
|(∇ θ ε )(y)| ≤ Cε − for every  ∈ N0 and every y ∈ Rn .

In particular,

the entries of E L (x − ·)θ ε belong to 𝒞∞ (Ω) and the entries of

E L (x − ·)ϕθ ε belong to 𝒞∞ c (Ω); moreover, similar properties are
valid for expressions involving partial derivatives acting on either (1.6.21)
of the intervening functions.
1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 167

Relying on (1.6.12), Lebesgue’s Dominated Convergence Theorem, and integra-

tions by parts (keeping in mind (1.5.26)) we may write
∫
  γβ  
D (Ω) div 
Fx , ϕ D(Ω) = lim Ak j ∂k (E L )αγ (x − y)uβ (y)(∂j ϕ)(y)θ ε (y) dy
+ ε→0 Ω
   
γβ
= lim+ D (Ω) Ak j uβ, ∂k (E L )αγ (x − ·)(∂j ϕ)θ ε D(Ω)
ε→0

= I + II + III, (1.6.22)

where
   
γβ
I := − lim+ D (Ω) Ak j ∂j uβ, ∂k (E L )αγ (x − ·)ϕθ ε D(Ω), (1.6.23)
ε→0
   
γβ
II := lim+ D (Ω) Ak j uβ, ∂j ∂k (E L )αγ (x − ·)ϕθ ε D(Ω), (1.6.24)
ε→0
   
γβ
III := − lim+ D (Ω) Ak j uβ, ∂k (E L )αγ (x − y)ϕ(∂j θ ε ) D(Ω) . (1.6.25)
ε→0

Note that
 
μβ γμ  
I = − lim+ D (Ω) a j ∂j uβ, '
ak ∂k (E L )αγ (x − ·)ϕθ ε D(Ω)
ε→0
    
' EL
= lim+ D (Ω) (Du)μ, D (x − ·)ϕθ ε D(Ω)
ε→0 μα

 
'μα (x − ·)ϕθ ε D(Ω)
= lim+ D (Ω) (Du)μ, E
ε→0
∫
 
= lim+ ' (x − y)(Du)(y) ϕ(y)θ ε (y) dy
E
ε→0 α
Ω
∫
 
= ' (x − y)(Du)(y) ϕ(y) dy
E α
Ω
  
= D (Ω)
' (x − ·)(Du) , ϕ D(Ω),
E (1.6.26)
α

thanks to (1.6.10), (1.6.8), (1.6.14), and Lebesgue’s Dominated Convergence Theo-

rem (which takes into account (1.6.19)). Next,
 
γβ  
II = lim+ D (Ω) uβ, Ak j ∂j ∂k (E L )αγ (x − ·)ϕθ ε D(Ω) = 0, (1.6.27)
ε→0

since (1.4.33) gives that for each β ∈ {1, . . . , M } we have

γβ  
Ak j ∂j ∂k (E L )αγ (x − y) = 0 for each y ∈ Rn \ {x}. (1.6.28)

As regards the contribution from the very last term in (1.6.22), we write
168 1 Integral Representations and Integral Identities
∫
γβ  
III = − lim+ Ak j ∂k (E L )αγ (x − y)uβ (y)ϕ(y)(∂j θ ε )(y) dy
ε→0 Ω
∫
γβ    
= lim+ Ak j ∂k (E L )αγ (x − y) (ϕuβ )(x) − (ϕuβ )(y) (∂j θ ε )(y) dy
ε→0 Ω
 ∫ 
γβ  
− lim+ Ak j ∂k (E L )αγ (x − y)(∂j θ ε )(y) dy (ϕuβ )(x). (1.6.29)
ε→0 Ω

In relation to this, we note that since x is a Lebesgue point for u, we may employ
(1.6.66) and (1.6.20) to estimate
∫ 
 γβ     
lim sup  Ak j ∂k (E L )αγ (x − y) (ϕuβ )(x) − (ϕuβ )(y) (∂j θ ε )(y) dy 
ε→0+ Ω
⨏
 
≤ C lim sup (ϕuβ )(y) − (ϕuβ )(x) dy = 0. (1.6.30)
ε→0+ B(x,2ε)

In addition, (1.4.33) eventually permits us to conclude that for each fixed index
β ∈ {1, . . . , M } we have
∫
γβ  
lim+ Ak j ∂k (E L )αγ (x − y)(∂j θ ε )(y) dy
ε→0 Ω
∫
γβ  
= lim+ Ak j ∂k (E L )αγ (x − y)∂j (θ ε − 1)(y) dy
ε→0 Ω
 
γβ
= − lim+ D (Ω) Ak j ∂k (E L )αγ (x − ·) , ∂j (θ ε − 1) D(Ω)
ε→0
 
γβ
= lim+ D (Ω) Ak j ∂j ∂k (E L )αγ (x − ·) , θ ε − 1 D(Ω)
ε→0
 
= δβα lim+ D (Ω) δx, θ ε − 1 D(Ω)
ε→0

= δβα lim+ (θ ε − 1)(x) = −δβα . (1.6.31)

ε→0

Altogether,
 
III = δβα (ϕuβ )(x) = ϕ(x)uα (x) = D (Ω) uα (x)δx, ϕ D(Ω) . (1.6.32)

From (1.6.22), (1.6.26), (1.6.27), and (1.6.32) we ultimately conclude that for
each Lebesgue point x ∈ Ω for u such that (1.6.14) holds we have
  
divFx = uα (x)δx + E ' (x − ·)(Du) in D (Ω).
α (1.6.33)

In particular,

divFx ∈ ℰ (Ω) + L 1 (Ω, L n ). (1.6.34)

1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 169
 
Going further, K := B x, 12 dist(x, ∂Ω) is a compact subset of Ω, and based on
(1.6.15), (1.4.24), and [112, Lemma 8.3.7] we may estimate
   
NκΩ\K Fx (y) ≤ C Nκ u (y) · sup |x − z| 1−n
z ∈Γκ (y)\K
 
Nκ u (y)
≤C , ∀y ∈ ∂Ω, (1.6.35)
|x − y| n−1

' n, κ) ∈ (0, ∞). From (1.6.35), (1.6.2), and [112,

for some constant C = C(Ω, D, D,
(8.2.26)] it follows that

NκΩ\K Fx ∈ L 1 (∂Ω, σ). (1.6.36)

Moreover, from (1.6.15), [112, (8.9.10)-(8.9.11)], and (1.6.13) we conclude that

n.t.

Fx  exists at σ-a.e. point on ∂nta Ω (1.6.37)
∂Ω

and, in fact, at σ-a.e. point y ∈ ∂∗ Ω we have

 n.t.   n.t. 
 γβ  
ν(y) · Fx  (y) = −ν j (y)Ak j ∂k (E L )αγ (x − y) uβ ∂Ω (y)
∂Ω

γμ   μβ  n.t. 
ak ∂k (E L )αγ (x − y)ν j (y)a j uβ ∂Ω (y)
= −'
     n.t.  
= D' EL 
μα (x − y)(−i) Sym(D; ν(y)) u ∂Ω (y) μ
  n.t.  
'μα (x − y)(−i) Sym(D; ν(y)) u (y)
=E ∂Ω μ
  n.t.  
' (x − y)(−i)Sym(D; ν(y)) u (y) ,
= E (1.6.38)
∂Ω α

by (1.6.15), (1.6.10), (1.3.17).

Finally, when Ω is an exterior domain, it follows from (1.6.15) and (1.4.24) that
 
Fx (y) = O |y| 1−n |u(y)| as |y| → ∞ (1.6.39)

which, in light of (1.6.4), implies that there exists λ ∈ (1, ∞) such that
∫

|y · F(y)| dL n (y) = o(R2 ) as R → ∞. (1.6.40)
B(0,λ R)\B(0,R)

Collectively, (1.6.16), (1.6.34), (1.6.37), and (1.6.40) ensure that, for each
Lebesgue point x ∈ Ω for u such that (1.6.14) holds, the vector field Fx satis-
fies the hypotheses of [112, Theorem 1.4.1]. On account of [112, (4.6.19)], (1.6.33),
and (1.6.38), the Divergence Formula [112, (1.4.6)] presently yields
170 1 Integral Representations and Integral Identities
∫  
uα (x) + ' (x − y)(Du)(y)
E dy
Ω α
∫  
 
= ℰ (Ω) uα (x)δx, 1 ℰ(Ω) + ' (x − y)(Du)(y)
E dy
Ω α
    
' (x − ·)(Du) , 1 𝒞∞ (Ω)
= (𝒞∞ (Ω))∗ uα (x)δx + E
b α b

∫
   n.t. 
= (𝒞∞ (Ω)) divFx, 1
∗ 𝒞∞
b
(Ω) = ν · Fx ∂Ω dσ
b ∂∗ Ω
∫   n.t.  
= ' (x − y)(−i)Sym(D; ν(y)) u (y) dσ(y),
E (1.6.41)
∂Ω α
∂∗ Ω

for each Lebesgue point x ∈ Ω of the function u with the property that (1.6.14) holds
(in particular, for L n -a.e. point x ∈ Ω), and each α ∈ {1, . . . , M }. In view of the
arbitrariness of α ∈ {1, . . . , M }, this establishes (1.6.3) in the case when x ∈ Ω.The
version of (1.6.3) corresponding to x ∈ Rn \ Ω is proved similarly, the most notable
difference being the fact that now III = 0, so in place of (1.6.33) we now have
  
divFx = E ' (x − ·)(Du) in D (Ω).
α (1.6.42)

The proof of Theorem 1.6.1 is therefore complete. 

As a consequence of Theorem 1.6.1, in our next corollary we elucidate the

behavior at infinity for a null-solution of an injectively elliptic first-order system.

Corollary 1.6.2 Fix n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an exterior domain.

Also, consider an injectively elliptic35 homogeneous first-order N × M system D with
constant complex coefficients in Rn (where N, M ∈ N). Then for any given function
M
u ∈ 𝒞∞ (Ω) satisfying Du = 0 in Ω the following conditions are equivalent:

(1) For some, or any, λ ∈ (1, ∞) one has

⨏
|u| dL n = o(1) as R → ∞. (1.6.43)
B(0,λ R)\B(0,R)

(2) One has

u(x) = o(1) as |x| → ∞. (1.6.44)
(3) One has ⨏
|u| dH n−1 = o(1) as R → ∞. (1.6.45)
∂B(0,R)

(4) One has

u(x) = O(|x| 1−n ) as |x| → ∞. (1.6.46)

35 i.e., its principal symbol satisfies (1.3.18)

1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 171

Moreover, if any of the conditions (1)-(4) are satisfied then

 
(∂ α u)(x) = O |x| 1−n− |α | as |x| → ∞.
(1.6.47)
for each multi-index α ∈ N0n .

In light of this result, it is natural to make the following definition.

Definition 1.6.3 Given an exterior domain Ω ⊆ Rn , where n ∈ N with n ≥ 2, along

with an injectively elliptic N × M homogeneous first-order system D with constant
M
complex coefficients in Rn , a function u ∈ 𝒞∞ (Ω) satisfying Du = 0 in Ω is
said to vanish at infinity provided any of the (equivalent) conditions (1)-(4) in
Corollary 1.6.2 holds.

Returning to the matter at hand, we now present the proof of Corollary 1.6.2.
Proof of Corollary 1.6.2 Lemma 1.5.6 shows that conditions (1)-(3) are equivalent,
and that condition (4) implies (1.6.47). Trivially, (4) implies any of the conditions
in (1)-(3), so we are left with proving that (1)-(3) imply (4). With this goal in
mind, with Ro ∈ (0, ∞) sufficiently large, introduce Ωo := Rn \ B(0, Ro ). Then
  M
u ∈ 𝒞∞ Ωo satisfies Du = 0 in Ωo , and (1.6.44) ensures that u is bounded in
Ωo . Granted these qualities, we may write the integral representation formula (1.6.3)
for u in Ωo , which in turn guarantees, on account of (1.4.24), that (1.6.46) holds.
This finishes the proof of the fact that (1)-(4) are equivalent. 
' from (1.6.1) is a fundamental solution for D . While
Recall from (1.6.6) that E
in general this does not guarantee that E' is a fundamental solution for −D, in
the proposition below (which refines [92, Proposition 3]) we show that is the case
whenever D and D ' commute.

Proposition 1.6.4 Consider a homogeneous, first-order N × M system D with con-

stant complex coefficients in Rn (where N, M ∈ N) which is injectively elliptic (cf.
(1.3.18)), and suppose D ' is a homogeneous first-order M × N system with constant
complex coefficients in Rn which complements D (in the sense of Definition 1.3.6).
' is a weakly elliptic second-order M × M system in Rn . In
In particular, L := DD
relation to this, make the assumption that

' ξ) = I M×M
Sym(D; ξ)[L(ξ)]−1 Sym( D;
(1.6.48)
for each ξ ∈ Rn \ {0}.

Next, since L  is also a weakly elliptic second-order M × M system in Rn , it is

meaningful to consider the matrix-valued fundamental solution E L  associated with
' acting on the columns of E L  define36
L  as in Theorem 1.4.2. Finally, with D
' := D
E ' E L  . (1.6.49)

' is a fundamental solution for D 

36 from (1.6.6) we know that E
172 1 Integral Representations and Integral Identities

' is a fundamental solution for −D, i.e.,

Then E
' ) = −δI M×M as tempered distributions in Rn,
D(E (1.6.50)

where δ is Dirac’s distribution with mass at the origin in Rn and I M×M is the M × M
identity matrix.
As a corollary, (1.6.50) holds whenever D commutes with D, ' i.e., when37

' = DD
DD ' (1.6.51)

(since this turns out to imply (1.6.48)).

Proof The fact that L = DD ' is a weakly elliptic second-order M × M system in Rn

implies (cf. (1.3.3)) that

L(ξ) is an invertible matrix for each ξ ∈ Rn \ {0}. (1.6.52)

'= D
From (1.6.49) and the first formula in (1.4.32) we see that E ' (E L ) , so if “hat”
denotes the Fourier transform then (1.4.31) gives

9 '0
' =D
E ' ; ξ) L(ξ)
 (E ) = Sym( D −1
L
(1.6.53)
as tempered distributions in Rn .

Thanks to (1.6.53) and [112, (1.7.17)], on the one hand, in the sense of tempered
distributions in Rn , we have
0  
'  = Sym(D; ξ)(:
D(E) '  = Sym(D; ξ) 9
E) '
E

' ξ).
= −Sym(D; ξ)[L(ξ)]−1 Sym( D; (1.6.54)

On the other hand, since Sym(D; ξ)[L(ξ)]−1 Sym( D;' ξ), regarded as a function
defined for L -a.e. ξ ∈ R , is locally integrable and bounded, formula (1.6.48)
n n

may further be interpreted as saying that

' ξ) = I M×M
Sym(D; ξ)[L(ξ)]−1 Sym( D;
(1.6.55)
as tempered distributions in Rn .

By combining (1.6.54) with (1.6.55) we then arrive at

0 '  = −I M×M as tempered distributions in Rn ,

D(E) (1.6.56)

from which (1.6.50) now follows.

' commute. Taking the Fourier
Consider next the special case when D and D
transform in (1.6.51) then yields

' are square systems

37 note that this forces N = M, i.e., D and D
1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 173

' ξ) = Sym( D;
Sym(D; ξ) Sym( D; ' ξ)Sym(D; ξ), ∀ξ ∈ Rn . (1.6.57)
' on the Fourier transform
Also, in view of (1.3.2) and (A.0.118), the identity L = DD
side reads
' ξ)Sym(D; ξ),
L(ξ) = Sym( D; ∀ξ ∈ Rn . (1.6.58)
On account of (1.6.52), (1.6.58), and (1.6.57) we then conclude that

for each given vector ξ ∈ Rn \ {0}, the matrix [L(ξ)]−1

' ξ) and with Sym(D; ξ). (1.6.59)
commutes both with Sym( D;

For each ξ ∈ Rn \ {0} we may then write

' ξ) = Sym(D; ξ)Sym( D;
Sym(D; ξ)[L(ξ)]−1 Sym( D; ' ξ)[L(ξ)]−1

' ξ)Sym(D; ξ)[L(ξ)]−1

= Sym( D;

= L(ξ)[L(ξ)]−1 = I M×M . (1.6.60)

Hence (1.6.48) holds, so the desired conclusion follows. 

We next present a version of Theorem 1.6.1 phrased exclusively in terms of a given
injectively elliptic first-order system D, i.e., without bringing into play an auxiliary
system D ' which complements D, and without any reference to second-order weakly
elliptic systems. Another significant feature is the fact that the present statement
involves a rather generic fundamental solution E ' of D , and not necessarily the
specific fundamental solution constructed in (1.6.1) (we shall put this feature to
good use shortly, in the very last portion of this section).
Theorem 1.6.5 Let Ω ⊆ Rn , where n ≥ 2, be an open set with a lower Ahlfors
regular boundary, and such that σ := H n−1 ∂Ω is a doubling measure on ∂Ω; in
particular, Ω is a set of locally finite perimeter. Denote by ν the geometric measure
theoretic outward unit normal to Ω, and fix an aperture parameter κ ∈ (0, ∞).
Finally, consider a homogeneous, first-order N × M system D with constant complex
coefficients in Rn (where N, M ∈ N) which is injectively elliptic (cf. (1.3.18)).
Then the following statements are true. First, there exists a matrix-valued tem-
pered distribution E' satisfying
N ×M  N ×M
' ∈ 𝒮 (Rn )
E , ' n
E ∈ 𝒞∞ (Rn \ {0}) ,
R \{0}
' is odd and positive homogeneous of degree 1 − n, and
E (1.6.61)
' = δI M×M (i.e., E
D E ' is a fundamental solution for D ),

where δ is Dirac’s distribution with mass at origin in Rn , and I M×M is the M × M

identity matrix.
Second, fix a matrix-valued tempered distribution E ' as in (1.6.61), and suppose
u : Ω → C M is a vector-valued function with Lebesgue measurable components
satisfying
174 1 Integral Representations and Integral Identities
 
Nκ u ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 ,
 κ−n.t. 
u∂Ω (x) exists for σ-a.e. x ∈ ∂nta Ω, (1.6.62)
  dy  N
and Du ∈ L 1 Ω, ,
1 + |y| n−1
(with Du considered in the sense of distributions in Ω). Then for any κ > 0
κ −n.t.
the nontangential trace u∂Ω also exists at σ-a.e. point on ∂nta Ω and is actually
independent of κ . Also, with the dependence on the parameter κ dropped, for
L n -a.e. x ∈ Rn \ ∂Ω one has (with absolutely convergent integrals)
∫
   n.t. 
E' (x − y)(−i)Sym D; ν(y) u (y) dσ(y)
∂Ω
∂∗ Ω

∫ 
u(x) if x ∈ Ω,
− '
E (x − y)(Du)(y) dy = (1.6.63)
Ω 0 if x ∈ Rn \ Ω,

if either Ω is bounded, or ∂Ω is unbounded. In the case when Ω is an exterior

domain, the same conclusion holds true under the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|u| dL n = o(1) as R → ∞. (1.6.64)
B(0,λ R)\B(0,R)

Observe that if Du = 0 then (1.6.63) implies (under the same background as-
sumptions as in the statement of Theorem 1.6.5) that for each x ∈ Ω we have
∫
   n.t. 
u(x) = ' (x − y)(−i)Sym D; ν(y) u (y) dσ(y),
E (1.6.65)
∂Ω
∂∗ Ω

which may be interpreted as a “generalized Cauchy reproducing formula.”

Theorem 1.6.5 may be regarded as a higher-dimensional generalization of the
version of the Cauchy-Pompeiu’s formula presented earlier in Theorem 1.1.1. The-
orem 1.6.5 also contains as a special case the integral representation formulas from
Theorem 1.2.2 in the context of Clifford algebras.
Proof of Theorem 1.6.5 The existence of a fundamental solution E ' for D as in
(1.6.61) is seen from (1.6.1), (1.6.5), and (1.6.6). This takes care of the first claim in
the statement. To deal with the second claim in the statement, fix such a matrix-valued
' In particular,
tempered distribution E.

' Cα
|(∂ α E)(x)| ≤ |α |
for each α ∈ N0n . (1.6.66)
|x| n−1+

The first property in (1.6.62) and [112, Lemma 8.3.1] guarantee that
∞ N
u ∈ Lloc (Ω, L n ) , (1.6.67)
1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 175

while the first two properties in (1.6.62) together with [112, Corollary 8.9.9] imply
that
κ −n.t.
for any κ > 0 the nontangential trace u∂Ω exists σ-a.e. on (1.6.68)
∂nta Ω and is actually independent of the parameter κ ∈ (0, ∞).

Let us establish (1.6.63). Recall from [112, Lemma 3.5.7] that the last membership
in (1.6.62) is equivalent with having for L n -a.e. point x ∈ Rn the finiteness property
∫
|(Du)(y)|
dy < +∞. (1.6.69)
Ω |x − y|
n−1

To fix notation, we agree to write

   
'= E
E 'αβ 1≤α ≤ N and D = aαγ ∂j 1≤α ≤ N (1.6.70)
j
1≤β ≤M 1≤γ ≤M

with the summation convention over repeated indices assumed throughout. In par-
ticular,  αγ 
D = − a j ∂j 1≤γ ≤M . (1.6.71)
1≤α ≤ N
Next, select an arbitrary index α ∈ {1, . . . , M }, pick a Lebesgue point x ∈ Ω for
the function
  u with the property that (1.6.69) holds, and consider the vector field
Fx = Fj 1≤ j ≤n with components

γβ '
γα (x − ·)uβ at L -a.e. point in Ω.
n
Fj := a j E (1.6.72)

Together, (1.6.72), (1.6.67), and Theorem 1.4.2 ensure that

Fx ∈ Lloc
n
1
(Ω, L n ) . (1.6.73)

Let us now compute divFx in the sense of distributions in Ω. With this goal in
mind, fix an arbitrary scalar-valued function ϕ ∈ 𝒞∞ c (Ω) and write
∫
 
D (Ω) div 
Fx , ϕ D(Ω) = − Fj (y)(∂j ϕ)(y) dy
Ω
∫
γβ '
=− aj E γα (x − y)uβ (y)(∂ j ϕ)(y) dy. (1.6.74)
Ω

Consider an auxiliary scalar-valued function θ ∈ 𝒞∞ (Rn ) with the property

 that
θ = 0 on B(0, 1) and θ = 1 on Rn \ B(0, 2). For each ε ∈ 0, 12 dist(x, ∂Ω) define
θ ε : Rn → R as follows:
y − x
θ ε (y) := θ for every y ∈ Rn . (1.6.75)
ε
It follows that
176 1 Integral Representations and Integral Identities

θ ε ∈ 𝒞∞ (Rn ) is a bounded function satisfying

(1.6.76)
lim θ ε (y) = 1 for every point y ∈ Rn \ {x},
ε→0+
 
and there exists a constant C ∈ (0, ∞) such that for each ε ∈ 0, 12 dist(x, ∂Ω) we
have
1 − θ ε ∈ 𝒞∞
c (Ω), θ ε ≡ 0 on B(x, ε),
supp (∇θ ε ) ⊆ B(x, 2ε) \ B(x, ε), and (1.6.77)
|(∇ θ ε )(y)| ≤ Cε − for every  ∈ N0 and y ∈ Rn .

As a consequence,

the entries of E L (x − ·)θ ε belong to 𝒞∞ (Ω) and the entries of

E L (x − ·)ϕθ ε belong to 𝒞∞ c (Ω); moreover, similar properties are
valid for expressions involving partial derivatives acting on either of (1.6.78)
the intervening functions.
Making use of (1.6.67), Lebesgue’s Dominated Convergence Theorem, and inte-
grations by parts (keeping in mind (1.5.26)) we may write
∫
  γβ '

D (Ω) div Fx , ϕ D(Ω) = − lim+ aj E γα (x − y)uβ (y)(∂ j ϕ)(y)θ ε (y) dy
ε→0 Ω
 
γβ 'γα (x − ·)(∂j ϕ)θ ε D(Ω)
= − lim+ D (Ω) a j uβ, E
ε→0

= I + II + III, (1.6.79)

where
 
γβ 'γα (x − ·)(x − ·)ϕθ ε D(Ω),
I := lim+ D (Ω) a j ∂j uβ, E (1.6.80)
ε→0
   
γβ 'γα (x − ·)ϕθ ε D(Ω),
II := − lim+ D (Ω) a j uβ, ∂j E (1.6.81)
ε→0
 
γβ 'γα (x − y)ϕ(∂j θ ε ) D(Ω) .
III := lim+ D (Ω) a j uβ, E (1.6.82)
ε→0

Observe that
1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 177
 
'γα (x − ·)ϕθ ε
I = lim+ D (Ω) (Du)γ, E D(Ω)
ε→0
∫
 
= lim+ ' (x − y)(Du)(y) ϕ(y)θ ε (y) dy
E
ε→0 α
Ω
∫
 
= ' (x − y)(Du)(y) ϕ(y) dy
E α
Ω
  
= D (Ω)
' (x − ·)(Du) , ϕ D(Ω),
E (1.6.83)
α

thanks to (1.6.70), (1.6.69), and Lebesgue’s Dominated Convergence Theorem

(which takes into account (1.6.76)). Next,
 
γβ  ' 
II = − lim+ D (Ω) uβ, a j ∂j E γα (x − ·)ϕθ ε D(Ω) = 0 (1.6.84)
ε→0

since, according to (1.6.70) and the last line in (1.6.61), for each β ∈ {1, . . . , M } we
have
γβ  '  '
a j ∂j Eγα (x − y) = −(D E) βα (x − y) = 0 for each y ∈ R \ {x}.
n
(1.6.85)

Concerning the contribution from the very last term in (1.6.79), we write
∫
γβ '
III = lim+ aj E γα (x − y)uβ (y)ϕ(y)(∂ j θ ε )(y) dy
ε→0 Ω
∫
γβ '  
= lim+ aj E γα (x − y) (ϕuβ )(y) − (ϕuβ )(x) (∂ j θ ε )(y) dy
ε→0 Ω
 ∫ 
γβ '
+ lim+ aj E γα (x − y)(∂ j θ ε )(y) dy (ϕuβ )(x). (1.6.86)
ε→0 Ω

In relation to this, we note that since x is a Lebesgue point for u, we may estimate
(using (1.4.24) and (1.6.77))
∫   
 γβ ' 
lim sup  aj E γα (x − y) (ϕuβ )(x) − (ϕuβ )(y) (∂ j θ ε )(y) dy 
ε→0+ Ω
⨏
 
≤ C lim sup (ϕuβ )(y) − (ϕuβ )(x) dy = 0. (1.6.87)
ε→0+ B(x,2ε)

Also, (1.6.71) together with the property recorded in last line in (1.6.61) allow us to
conclude that for each fixed index β ∈ {1, . . . , M } we have
178 1 Integral Representations and Integral Identities
∫
γβ '
lim aj E γα (x − y)(∂ j θ ε )(y) dy
ε→0+ Ω
∫
γβ '
= lim+ aj E γα (x − y)∂ j (θ ε − 1)(y) dy
ε→0 Ω
 
γβ 'γα (x − ·) , θ ε − 1 D(Ω)
= lim+ D (Ω) a j ∂j E
ε→0
 
= −δβα lim+ D (Ω) δx, θ ε − 1 D(Ω)
ε→0

= −δβα lim+ (θ ε − 1)(x) = δβα . (1.6.88)

ε→0

We may therefore conclude that

 
III = δβα (ϕuβ )(x) = ϕ(x)uα (x) = D (Ω) uα (x)δx, ϕ D(Ω) . (1.6.89)

Based on (1.6.79), (1.6.83), (1.6.84), and (1.6.89) we ultimately see that for each
Lebesgue point x ∈ Ω for u such that (1.6.69) holds we have
  
divFx = uα (x)δx + E ' (x − ·)(Du) in D (Ω).
α (1.6.90)

In particular,

divFx ∈ ℰ (Ω) + L 1 (Ω, L n ). (1.6.91)

 
Pressing on, bring in K := B x, 12 dist(x, ∂Ω) , which is a compact subset of Ω.
Based on (1.6.72), (1.6.61), (1.6.66), and [112, Lemma 8.3.7] we may then estimate
   
NκΩ\K Fx (y) ≤ C Nκ u (y) · sup |x − z| 1−n
z ∈Γκ (y)\K
 
Nκ u (y)
≤C , ∀y ∈ ∂Ω, (1.6.92)
|x − y| n−1

for some constant C = C(Ω, D, n, κ) ∈ (0, ∞). From (1.6.92), (1.6.62), and [112,
(8.2.26)] it follows that

NκΩ\K Fx ∈ L 1 (∂Ω, σ). (1.6.93)

In addition, from (1.6.72), [112, (8.9.10)-(8.9.11)], and (1.6.68) we conclude that

n.t.

Fx  exists at σ-a.e. point on ∂nta Ω (1.6.94)
∂Ω

and, in fact, at σ-a.e. point y ∈ ∂∗ Ω we have

1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 179
 n.t.   n.t. 
 γβ ' 
ν(y) · Fx  (y) = ν j (y)a j Eγα (x − y) uβ ∂Ω (y)
∂Ω
  n.t.  
'γα (x − y)(−i) Sym(D; ν(y)) u (y)
=E ∂Ω γ
  n.t.  
' (x − y)(−i)Sym(D; ν(y)) u (y) ,
= E (1.6.95)
∂Ω α

by (1.6.72) and (1.3.17).

Lastly, when Ω is an exterior domain, it follows from (1.6.72) and (1.6.66) that
 
Fx (y) = O |y| 1−n |u(y)| as |y| → ∞ (1.6.96)

which, in view of (1.6.64), implies that there exists λ ∈ (1, ∞) such that
∫

|y · F(y)| dL n (y) = o(R2 ) as R → ∞. (1.6.97)
B(0,λ R)\B(0,R)

Together, (1.6.73), (1.6.91), (1.6.94), and (1.6.97) ensure that, for each Lebesgue
point x ∈ Ω for u such that (1.6.69) holds, the vector field Fx satisfies the hypotheses
of [112, Theorem 1.4.1]. On account of [112, (4.6.19)], (1.6.90), and (1.6.95), the
Divergence Formula [112, (1.4.6)] presently yields
∫  
uα (x)+ ' (x − y)(Du)(y) dy
E
Ω α
∫  
 
= ℰ (Ω) uα (x)δx, 1 ℰ(Ω) + ' (x − y)(Du)(y)
E dy
Ω α
    
' (x − ·)(Du) , 1 𝒞∞ (Ω)
= (𝒞∞ (Ω))∗ uα (x)δx + E
b α b

∫
   n.t. 
= (𝒞∞ (Ω)) divFx, 1
∗ 𝒞∞
b
(Ω) = ν · Fx ∂Ω dσ
b ∂∗ Ω
∫   n.t.  
= ' (x − y)(−i)Sym(D; ν(y)) u (y) dσ(y),
E (1.6.98)
∂Ω α
∂∗ Ω

for each Lebesgue point x ∈ Ω of the function u with the property that (1.6.69) holds
(in particular, for L n -a.e. point x ∈ Ω), and each α ∈ {1, . . . , M }. Bearing in mind
that α ∈ {1, . . . , M } is arbitrary, this establishes (1.6.63) in the case when x ∈ Ω.
The version of (1.6.63) corresponding to x ∈ Rn \ Ω is proved similarly, the most
notable difference being the fact that now III = 0, so in place of (1.6.90) we now
have
  
divFx = E ' (x − ·)(Du) in D (Ω).
α (1.6.99)

This concludes the proof of Theorem 1.6.5. 

180 1 Integral Representations and Integral Identities

Let us take another look at the first-order differential operator D A from Ex-
ample 1.3.12. The broader context involved a weakly elliptic, second-order, ho-
 αβ  coefficient M × M system L, associated with some
mogeneous, constant complex
coefficient tensor A = ar s 1≤r,s ≤n as in (1.3.15). Introduce
1≤α,β ≤M

' := n × M and N
M ' := (n2 × M) + M. (1.6.100)

Recall from the definition given in (1.3.39) that D A is a homogeneous constant

coefficient first-order N'× M' system, which maps distributions in Rn with n × M
 β n×M
components w = ws 1≤s ≤n ∈ D (Rn ) into
1≤β ≤M

 β β
( ∂r ws − ∂s wr 1≤r,s ≤n +
D A w := )  1≤β ≤M , . (1.6.101)
αβ β
* ar s ∂r ws 1≤α ≤M -
It has been noted in (1.3.40) that the first-order system D A is injectively elliptic.
Moreover, a direct computation based on (1.6.101) shows that the transpose of D A is
the M'×N ' first-order system acting on distributions in Rn according to the formula
 β 
Ur s 1≤r,s ≤n  
 β β αβ
D A   1≤β ≤M = ∂r Usr − ∂r Ur s − ar s ∂r uα 1≤s ≤n . (1.6.102)
uα 1≤α ≤M 1≤β ≤M

If we denote by E = (Eγβ )1≤γ,β ≤M the matrix-valued fundamental solution associ-

ated with the second-order weakly elliptic system L as in Theorem 1.4.2, then for
each  ∈ {1, . . . , n} and γ ∈ {1, . . . , M } we may rely on (1.6.102) and (1.4.33) to
compute   αβ 


a js ∂j Eγα 1≤r,s ≤n  
DA   1≤β ≤M = δδβγ δ s 1≤s ≤n (1.6.103)
− ∂ Eγα 1≤α ≤M 1≤β ≤M

where δ is Dirac’s distribution with mass at origin in Rn . This shows that if we

' the N
denote by E '× M' matrix-valued tempered distribution in Rn whose (, γ)-th
column is given by   αβ 

a js ∂j Eγα 1≤r,s ≤n
  1≤β ≤M (1.6.104)
− ∂ Eγα 1≤α ≤M
' is a fundamental solution for
for each  ∈ {1, . . . , n} and γ ∈ {1, . . . , M } then E

D A, in the sense that
' = δI ; ;
DA E (1.6.105)
M× M
' Based on
in the sense of distributions in Rn , with DA acting on the columns of E.
'
this and Theorem 1.4.2 we then see that E satisfies all properties in (1.6.61) relative
'× M
to the injectively elliptic first-order N ' system D := D A.
1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 181

Granted this observation, Theorem 1.6.5 applies and gives the following result
(with Ω as in the statement of Theorem 1.6.5): for each vector-valued function
 β ;
w = ws 1≤s ≤n : Ω −→ C M (1.6.106)
1≤β ≤M

with Lebesgue measurable components satisfying

 
Nκ w ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
,
 κ−n.t. 
w ∂Ω (x) exists for σ-a.e. x ∈ ∂nta Ω, (1.6.107)
  dy   N'
and D A w ∈ L 1 Ω, ,
1 + |y| n−1

at L n -a.e. point x ∈ Rn \ ∂Ω one has

∫
   κ−n.t. 
' (x − y)(−i)Sym D A; ν(y) w 
E (y) dσ(y)
∂Ω
∂∗ Ω

∫ 
w(x) if x ∈ Ω,
− '
E (x − y)(D A w)(y) dy = (1.6.108)
Ω 0 if x ∈ Rn \ Ω,

if either Ω is bounded, or ∂Ω is unbounded. In the case when Ω is an exterior

domain, the same conclusion holds true under the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|w| dL n = o(1) as R → ∞. (1.6.109)
B(0,λ R)\B(0,R)

We can formally state the result just established above as follows:

Theorem 1.6.6 Let Ω ⊆ Rn , where n ≥ 2, be an open set with a lower Ahlfors regular
boundary, and with the property that σ := H n−1 ∂Ω is a doubling measure on ∂Ω;
in particular, Ω is a set of locally finite perimeter. Denote by ν = (ν1, . . . , νn )
the geometric measure theoretic outward unit normal to Ω. With the summation
convention over repeated indices understood throughout, let
 αβ 
L = ar s ∂r ∂s 1≤α,β ≤M (1.6.110)

be a homogeneous, weakly elliptic, second-order M × M system in Rn , with com-

plex constant coefficients, and denote by E = (Eγβ )1≤γ,β ≤M the matrix-valued
fundamental solution associated with L as in Theorem 1.4.2. In this setting, assume
β 1 (Ω, L 1 ) n·M is a vector-valued function which, for some
w = (ws ) 1≤s ≤n ∈ Lloc
1≤β ≤M
κ > 0, satisfies
182 1 Integral Representations and Integral Identities
κ−n.t.
w ∂Ω exists at σ-a.e. point on ∂nta Ω,
∫   (1.6.111)
Nκ w (y)
and dσ(y) < +∞.
∂Ω 1 + |y|
n−1

With all partial derivatives considered in the sense of distributions in Ω, also suppose
that  
β β
(∂r ws − ∂s wr ) ∈ L 1 Ω, 1+ |ydy| n−1 for all
β ∈ {1, . . . , M }, r, s ∈ {1, . . . , n}, and (1.6.112)
 
αβ β
ar s ∂r ws ∈ L 1 Ω, 1+ |ydy| n−1 for each α ∈ {1, . . . , M }.
κ −n.t.
Then for any other aperture parameter κ > 0 the nontangential trace w ∂Ω
also exists σ-a.e. on ∂nta Ω and is actually independent of κ . Moreover, with the
dependence on the parameter κ dropped and having fixed  ∈ {1, . . . , n} along
with γ ∈ {1, . . . , M } arbitrary, it follows that for L n -a.e. point x ∈ Ω one has (with
absolutely convergent integrals)
∫   β n.t.   β n.t.  
a js (∂j Eγα )(x − y) ν (y) ws ∂Ω (y) − νs (y) w ∂Ω (y) dσ(y)
γ αβ
w (x) =
∂∗ Ω
∫
αβ  β n.t. 
− (∂ Eγα )(x − y)νr (y)ar s ws ∂Ω (y) dσ(y)
∂∗ Ω
∫  
αβ β β
− a js (∂j Eγα )(x − y) (∂ ws )(y) − (∂s w )(y) dy
Ω
∫
αβ β
+ (∂ Eγα )(x − y)ar s (∂r ws )(y) dy (1.6.113)
Ω

if either Ω is bounded, or ∂Ω is unbounded. In the case when Ω is an exterior

domain, the same conclusion holds true under the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|w| dL n = o(1) as R → ∞. (1.6.114)
B(0,λ R)\B(0,R)

In addition, for each x ∈ Rn \ Ω one has

1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 183
∫   β n.t.   β n.t.  
a js (∂j Eγα )(x − y) ν (y) ws ∂Ω (y) − νs (y) w ∂Ω (y) dσ(y)
αβ
0=
∂∗ Ω
∫
αβ  β n.t. 
− (∂ Eγα )(x − y)νr (y)ar s ws ∂Ω (y) dσ(y)
∂∗ Ω
∫  
αβ β β
− a js (∂j Eγα )(x − y) (∂ ws )(y) − (∂s w )(y) dy
Ω
∫
αβ β
+ (∂ Eγα )(x − y)ar s (∂r ws )(y) dy (1.6.115)
Ω

with the same decay condition as before in the case when Ω is an exterior domain.

Proof This is a direct consequence of (1.6.108), bearing in mind (1.6.101) and the
' is described in (1.6.104).
fact that the (, γ)-th column of E 
We make three remarks pertaining to the above theorem. First, Theorem 1.6.6 sub-
sumes our earlier result from Theorem 1.5.10. More specifically, formulas (1.6.113)
(1.6.115) from Theorem 1.6.6 applied to
 
w := ∇u = ∂s uβ 1≤s ≤n , (1.6.116)
1≤β ≤M

the Jacobian of a function u as in the statement of Theorem 1.5.10 reduce precisely

to (1.5.230) and (1.5.232), respectively (in view of the fact that in this special case
β β
we have ∂r ws − ∂s wr = 0 for each β ∈ {1, . . . , M } and each r, s ∈ {1, . . . , n}).
Second, we wish to note that a direct proof of Theorem 1.6.6 may be given  by
applying the Divergence Formula [112, (1.4.6)] to the vector field Fx = Fk 1≤k ≤n
with components given at L n -a.e. point in Ω by
αβ β αβ β
Fk := a js (∂j Eγα )(x − ·)δk ws − a js (∂j Eγα )(x − ·)δks w
αβ β
− ar s (∂ Eγα )(x − ·)δkr ws (1.6.117)

and reasoning as in the proof of Theorem 1.6.5.

Lastly, we observe that in the particular case of the scalar Laplacian, i.e., for
αβ
M := 1, L := Δ, and ar s = δαβ δr s , the integral representation formula (1.6.113)
reduces to the identities established earlier in (1.2.56)-(1.2.57) for vector fields
satisfying the generalized Cauchy-Riemann equations.
In the theorem below we re-visit the “generalized Cauchy reproducing formula”
(1.6.65), the goal now being to derive a similar integral representation in exterior
domains for functions which exhibit power growth at infinity.

Theorem 1.6.7 Let Ω ⊆ Rn , where n ≥ 2, be an unbounded connected open set,

with a compact lower Ahlfors regular boundary, and such that σ := H n−1 ∂Ω is
a doubling measure on ∂Ω; in particular, Ω is a set of locally finite perimeter, and
its geometric measure theoretic outward unit normal ν is defined σ-a.e. on ∂∗ Ω.
Also, assume D is a homogeneous, first-order N × M system with constant complex
184 1 Integral Representations and Integral Identities

coefficients in Rn (where N, M ∈ N) which is injectively elliptic (cf. (1.3.18)), and

' satisfying
consider a matrix-valued tempered distribution E
N ×M  N ×M
' ∈ 𝒮 (Rn )
E , ' n
E ∈ 𝒞∞ (Rn \ {0}) ,
R \{0}
' is odd and positive homogeneous of degree 1 − n, and
E (1.6.118)
' = δI M×M (i.e., E
D E ' is a fundamental solution for D ),

where δ is Dirac’s distribution with mass at origin in Rn , and I M×M is the M × M

identity matrix (the existence of E' is guaranteed by Theorem 1.6.5). Finally, fix an
aperture parameter κ > 0, along with a truncation parameter ε > 0, a dilation
factor λ ∈ (1, ∞), and an exponent m ∈ [0, ∞).
Next, suppose u is a vector-valued function satisfying
∫
∞
Nκε u dσ < ∞,
M
u ∈ 𝒞 (Ω) , Du = 0 in Ω,
∂Ω (1.6.119)
κ−n.t.
the nontangential trace u  exists at σ-a.e. point on ∂ Ω,
∂Ω nta

and ⨏
|u| dL n = O(Rm ) as R → ∞. (1.6.120)
B(0,λ R)\B(0,R)
κ −n.t.
Then for any κ > 0 the nontangential trace u∂Ω also exists at σ-a.e. point on
∂nta Ω and is actually independent of the parameter κ . Moreover, there exists some
(unique) C M -valued function P in Rn whose scalar components are polynomials
and with the property that for each point x ∈ Ω one has (with absolutely convergent
integrals, and with the dependence on the aperture parameter dropped)
∫
   n.t. 
u(x) = P(x) + ' (x − y)(−i)Sym D; ν(y) u (y) dσ(y).
E (1.6.121)
∂Ω
∂∗ Ω

Suppose D' is a homogeneous first-order M × N system with constant complex co-

efficients in Rn which complements D (in the sense of Definition 1.3.6). In particular,
' is a weakly elliptic second-order M × M system in Rn . If the fundamental
L := DD
' for D from (1.6.118) is actually taken to be
solution E
' := D
E ' E L  , (1.6.122)

where E L  is the matrix-valued fundamental solution associated with the weakly

elliptic second-order M × M system L  in Rn as in Theorem 1.4.2, then P also
satisfies
LP = 0 ∈ C N in Rn . (1.6.123)
' is such that (with L := DD
Finally, if D ' as before)
1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 185

' ξ) = I M×M
Sym(D; ξ)[L(ξ)]−1 Sym( D;
(1.6.124)
for each ξ ∈ Rn \ {0},

and the fundamental solution E ' for D from (1.6.118) is taken to be as in (1.6.122),
it follows that P actually satisfies

DP = 0 ∈ C N in Rn . (1.6.125)

Proof The first claim in the statement, regarding the independence of the boundary
trace on the aperture parameter, follows from assumptions and [112, Corollary 8.9.9].
There remains to prove the integral representation formula claimed in (1.6.121). This
requires some preparations. For specificity, we agree to write
   
E'= E 'αβ 1≤α ≤ N and D = aαγ ∂j 1≤α ≤ N (1.6.126)
j
1≤β ≤M 1≤γ ≤M

with the summation convention over repeated indices assumed throughout. Conse-
quently,  αγ 
D = − a j ∂j 1≤γ ≤M . (1.6.127)
1≤α ≤ N
Introduce the notation (uβ )1≤β ≤M for the scalar components of the vector-valued
M
function u ∈ 𝒞∞ (Ω) . Next, fix an arbitrary multi-index η ∈ N0n and select an
arbitrary index α ∈ {1, . . . , M }. Also, pick an arbitrary point x ∈ Ω and define
 
K := B x, 12 dist(x, ∂Ω) , compact subset of Ω. Use these to define the vector field,
whose scalar components are the distributions,
  γβ
Fx,η = Fj 1≤ j ≤n with Fj := a j ∂ η E'γα (x − ·) uβ ∈ D (Ω). (1.6.128)

Here the package of derivatives ∂ η is applied (in the ‘dot’ variable) in the sense of
distributions in Ω to E 'γα (x − ·) which is a locally integrable function in Ω (thus,
itself a distribution in Ω). It is therefore meaningful to further multiply the resulting
distribution by the scalar components of u (which are smooth functions in Ω). All
things considered, Fx,η is a well-defined vector field with components distributions
in Ω, i.e.,
Fx,η ∈ D (Ω) .
n
(1.6.129)
From (1.6.128), (1.6.118), and the first property in (1.6.119) we see that

Fx,η Ω\K ∈ 𝒞∞ (Ω \ K) ⊆ Lloc
n n
1
(Ω \ K, L n ) . (1.6.130)

In addition, divFx,η ∈ D (Ω), the divergence of the vector field Fx,η in the sense of
distributions in Ω is given by
186 1 Integral Representations and Integral Identities
γβ 'γα (x − ·) uβ + aγβ ∂ η E
divFx,η = ∂j Fj = a j ∂ η ∂j E j
'γα (x − ·) (∂j uβ )
   
γβ 'γα (x − ·) aγβ ∂j uβ
'γα (x − ·) + ∂ η E
= uβ ∂ η a j ∂ j E j
 
' − ·)
= −uβ ∂ η D E(x 'γα (x − ·) (Du)γ
+ ∂η E
βα
 
= uβ ∂ η δβα δx = uα ∂ η δx . (1.6.131)

Thus,

divFx,η = uα ∂ η δx ∈ ℰ (Ω). (1.6.132)

Based on assumptions and (1.6.128) we also see that

n.t.

Fx,η  exists at σ-a.e. point on ∂nta Ω (1.6.133)
∂Ω

and for each j ∈ {1, . . . , n} and σ-a.e. point y ∈ ∂nta Ω we actually have
 n.t.   n.t. 
 γβ 'γα )(x − y) uβ  (y).
Fx,η  (y) = −(−1) |η | a j (∂ η E ∂Ω
(1.6.134)
∂Ω j

Next, from (1.6.120) and Lemma 1.5.6 we conclude that

u(x) = O(|x| m ) as |x| → ∞, (1.6.135)

while (1.6.118) implies that

 
'
(∂ η E)(x) = O |x| 1−n− |η | as |x| → ∞. (1.6.136)

In view of these properties and the definition in (1.6.128) we see that

 
Fx,η (x) = O |x| m+1−n− |η | as |x| → ∞. (1.6.137)

In particular,

Fx,η (x) = o(1) as |x| → ∞, if |η| > m − n + 1. (1.6.138)

Collectively, (1.6.138), (1.6.130), (1.6.128), (1.6.118), and (1.6.119) imply (mindful

of the fact that ∂Ω is compact) that

NκΩ\K Fx,η ∈ L 1 (∂Ω, σ) if |η| > m − n + 1. (1.6.139)

Fix now an arbitrary multi-index η ∈ N0n with |η| > m − n + 1. Granted, (1.6.129),
(1.6.130), (1.6.132), (1.6.133), and (1.6.139), we may invoke [112, Theorem 1.4.1],
and the Divergence Formula recorded in [112, (1.4.6)] presently gives
1.6 Integral Representation Formulas for Injectively Elliptic First-Order Systems 187
   
(∂ η uα )(x) = ℰ (Ω) δx, ∂ η uα ℰ(Ω) = (−1) |η | ℰ (Ω) ∂ η δx, uα ℰ(Ω)
   
= (−1) |η | ℰ (Ω) uα ∂ η δx, 1 = (−1) |η | ℰ (Ω) divFx,η, 1 ℰ(Ω)
ℰ(Ω)
∫ 
  n.t. 
= (−1) |η | (𝒞∞ (Ω))∗ divFx,η, 1 𝒞∞b (Ω) = (−1) |η | ν · Fx,η ∂Ω dσ
b ∂∗ Ω
∫
 n.t. 
=−
γβ 'γα )(x − y) uβ  (y) dσ(y),
ν j (y)a j (∂ η E (1.6.140)
∂Ω
∂∗ Ω

on account of (1.6.132), [112, (4.6.21)], (1.6.134), and [112, Proposition 8.8.6].

In turn, from (1.6.140) and assumptions it follows that if we define the function
P : Ω → C M by setting, at each point x ∈ Ω,
∫  n.t.  
P(x) := u(x) +
γβ '
ν j (y)a j E 
γα (x − y) uβ ∂Ω (y) dσ(y) (1.6.141)
∂∗ Ω 1≤α ≤M

then
M
P belongs to 𝒞∞ (Ω) and satisfies ∂ η P = 0 in Ω
(1.6.142)
for each η ∈ N0n with |η| > m − n + 1.
Expanding P locally into a Taylor series (with remainder), (1.6.142) implies that
the scalar components of P are locally polynomials in Ω. This and the fact that
we are currently assuming Ω to be connected then ultimately show that the scalar
components of P are polynomials in Ω. At this stage, (1.6.121) follows from (1.6.141)
and (A.0.118).
' is as in (1.6.122), then since Lu = D(Du)
Next, if E ' = 0 in Ω and L E' = 0
in Rn \ {0}, from (1.6.121) we readily conclude that (1.6.123) holds. Finally, when
the additional hypothesis (1.6.124) is assumed, (1.6.125) follows from (1.6.121) and
(1.6.50). 

In turn, Theorem 1.6.7 allows us to augment the analysis initiated in Corol-

lary 1.6.2, by now including null-solutions of injectively elliptic first-order systems
with at most linear growth at infinity.

Corollary 1.6.8 Pick n ∈ N satisfying n ≥ 2, along with some N, M ∈ N. Suppose

Ω ⊆ Rn is an exterior domain. Also, consider an injectively elliptic homogeneous
first-order N × M system D with constant complex coefficients in Rn . Then for any
M
given function u ∈ 𝒞∞ (Ω) satisfying Du = 0 in Ω the following conditions are
equivalent:

(1) One has u(x) = o(|x|) as |x| → ∞.

(2) One has u(x) = O(1) as |x| → ∞.
(3) The limit lim u(x) exists in C M .
|x |→∞
(4) There exists some c ∈ C M such that u(x) = c + O(|x| 1−n ) as |x| → ∞.
188 1 Integral Representations and Integral Identities

Moreover, if any of the conditions (1)-(4) are satisfied then

 
(∂ α u)(x) = O |x| 1−n− |α | as |x| → ∞.
(1.6.143)
for each multi-index α ∈ N0n with |α| > 0.

Proof For the purposes we have in mind, there is no loss of generality in assuming
that Ω is the complement of the closure of a large ball centered at the origin in Rn (in
  M
particular, Ω is connected), and that actually u ∈ 𝒞∞ Ω . Assume this the case,
and suppose u(x) = o(|x|) as |x| → ∞. We may then invoke Theorem 1.6.7 (with
m := 1) and conclude from (1.6.121) and (1.6.118) that there exists a polynomial
P in Rn with the property that u(x) = P(x) + O(|x| 1−n ) as |x| → ∞. The latter
property forces P(x) = o(|x|) as |x| → ∞ hence, ultimately, P is a constant, say
P ≡ c ∈ C M . This argument proves the implication (1) ⇒ (4). Since the implications
(4) ⇒ (3) ⇒ (2) ⇒ (1) are clear, it follows that conditions (1)-(4) are equivalent.
Finally, that any of conditions (1)-(4) implies (1.6.143) is seen from what we have
proved so far, (1.6.121), and (1.6.118). 

1.7 Green-Type Formulas for Second-Order Systems

Given n, M, N ∈ N, by a complex coefficient tensor (of type (n × n, M × N))

we shall understand any block of the form
 αβ 
A = ar s 1≤r,s ≤n (1.7.1)
1≤α ≤M
1≤β ≤ N

αβ
where each entry ar s is a complex number. There are several things we shall typically
associate with such a complex coefficient tensor A. First, define the transpose of A
as being the complex coefficient tensor (of type (n × n, N × M)) given by
 αβ    βα αβ
A := ar s 1≤s,r ≤n, i.e., A sr := ar s for all α, β, r, s. (1.7.2)
1≤β ≤ N
1≤α ≤M

Hence, in the case of square coefficient tensors (i.e., when N = M),

 αβ   βα 
if A = ar s 1≤r,s ≤n then A := asr 1≤r,s ≤n , in
 αβ≤ N βα
 1≤α,β

1≤α,β ≤ N (1.7.3)
other words A r s := asr for all α, β, r, s.

Second, associate with each complex coefficient tensor A as in (1.7.1) the linear
mapping

A : C N ×n −→ C M×n given by
 αβ β  β (1.7.4)
Aζ := ar s ζs α,r ∈ C M×n for each ζ := (ζs )β,s ∈ C N ×n,
1.7 Green-Type Formulas for Second-Order Systems 189

using the summation convention over repeated indices understood here and else-
N
where. In particular, given any open set Ω ⊆ Rn and any u ∈ D (Ω) , with
n×N
∇u ∈ D (Ω) denoting the Jacobian matrix of first-order distributional partial
derivatives of u in Ω (cf. (A.0.48)), we have
 αβ  M×n
A∇u = ar s ∂s uβ α,r ∈ D (Ω) . (1.7.5)

Third, with a complex

 coefficient tensor A as in (1.7.1), we associated the (real)
bilinear form A·, · defined by
  αβ β
Aζ, η := ar s ζs ηrα for any given
(1.7.6)
β
ζ = (ζs )β,s ∈ C N ×n and η = (ηrα )α,r ∈ C M×n .

Fourth, with each complex coefficient tensor A as in (1.7.1) we may canonically

associate a homogeneous constant (complex) coefficient second-order M × N system
L A in Rn which, with the summation convention over repeated indices in place, is
expressed as  
αβ
L A := ar s ∂r ∂s 1≤α ≤M . (1.7.7)
1≤β ≤ N

Fifth, with each complex coefficient tensor A as in (1.7.1) we also associate a

conormal derivative on the boundary of a given domain of the following sort.

Definition 1.7.1 Suppose Ω ⊆ Rn is an open set with the property that ∂Ω is lower
Ahlfors regular and σ := H n−1 ∂Ω is a doubling measure on ∂Ω. In particular, Ω
is a set of locally finite perimeter, and its geometric measure theoretic outward unit
normal ν = (ν1, . . . , νn ) is defined σ-a.e. on ∂∗ Ω. Given a vector-valued distribution
N n×N
u = (uβ )1≤β ≤ N ∈ D (Ω) , denote by ∇u ∈ D (Ω) the Jacobian matrix
of first-order distributional partial derivatives of u in Ω. In relation to this suppose
that for some complex coefficient tensor A of type (n × n, M × N), expressed as in
(1.7.1), and some aperture parameter κ > 0,
 αβ  1 (Ω, L n ) M×n and
A∇u := ar s ∂s uβ α,r ∈ Lloc
κ−n.t. (1.7.8)
(A∇u)∂Ω exists (in C M×n ) at σ-a.e. point on ∂∗ Ω.

In such a setting, define the (pointwise) conormal derivative of u (with respect

to the coefficient tensor A and the set Ω) as the C M -valued function
   κ−n.t. 
∂νAu := νr ar s ∂s uβ ∂Ω
αβ
at σ-a.e. point on ∂∗ Ω. (1.7.9)
1≤α ≤M

A few clarifications are in order. First, the largest subset of ∂Ω on which one
may even attempt to consider the nontangential pointwise limit in (1.7.8) is Aκ (∂Ω).
The current hypotheses on Ω guarantee the applicability of [112, Proposition 8.8.4]
which, in turn, ensures that ∂∗ Ω is covered by Aκ (∂Ω), up to a σ-nullset. Ultimately,
this makes the assumption in (1.7.8) meaningfully formulated.
190 1 Integral Representations and Integral Identities

Second, if for some other aperture parameter κ > 0 we also have that
κ −n.t.
(A∇u)∂Ω exists (in C M×n ) at σ-a.e. point on ∂∗ Ω, (1.7.10)

then given that for each x ∈ ∂Ω we have

Γκ (x) ⊆ Γκ (x) if κ ≥ κ , and Γκ (x) ⊆ Γκ (x) if κ ≤ κ , (1.7.11)

it follows that we necessarily have

κ−n.t. κ −n.t.
(A∇u)∂Ω = (A∇u)∂Ω at σ-a.e. point on ∂∗ Ω, (1.7.12)

hence
   κ−n.t.     κ −n.t. 
νr ar s ∂s uβ ∂Ω = νr ar s ∂s uβ ∂Ω
αβ αβ
1≤α ≤M 1≤α ≤M (1.7.13)
at σ-a.e. point on ∂∗ Ω.

This ultimately shows that, as the notation already reflects it,

the definition of the conormal derivative ∂νAu in (1.7.9)

(1.7.14)
is actually independent of the aperture parameter κ > 0.
Third, in the context of Definition 1.7.1, [112, Corollary 8.9.6] implies that

the conormal derivative ∂νAu, defined as in (1.7.9),

(1.7.15)
is a σ-measurable (C M -valued) function on ∂∗ Ω.
Fourth, retaining the assumptions on the domain Ω and the coefficient tensor A
from Definition 1.7.1, as a special case we have

(Ω) and ⎫
N
1,1
u ∈ Wloc ⎪
⎪   κ−n.t. 
⎪
∂νAu = νr ar s (∂s uβ )∂Ω
⎬
⎪ αβ
κ−n.t.
(∇u)∂Ω exists ⎪ =⇒ 1≤α ≤M (1.7.16)
⎪
⎪ at σ-a.e. point on ∂∗ Ω.
σ-a.e. on ∂∗ Ω ⎪
⎭
Occasionally, it is convenient to extend the scope of Definition 1.7.1 by allowing
more general domains, though we now restrict to a smaller class of functions.

Convention 1.7.2 Let Ω ⊆ Rn be a Lebesgue measurable set of locally finite perime-

ter. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the geometric measure
theoretic outward unit normal to Ω. Given a C N -valued function u = (uβ )1≤β ≤M
of class 𝒞1 in a neighborhood of ∂∗ Ω, we agree to define the conormal derivative
∂νAu with respect to any given complex coefficient tensor A of type (n × n, M × N),
expressed as in (1.7.1), according to
  
∂νAu := νr ar s (∂s uβ )∂∗ Ω
αβ
(1.7.17)
1≤α ≤M
1.7 Green-Type Formulas for Second-Order Systems 191

at σ-a.e. point on ∂∗ Ω.

Whenever Ω is open, ∂Ω is lower Ahlfors regular, and σ is a doubling measure

on ∂Ω, there is agreement between the brands of conormal derivatives considered
N
in Definition 1.7.1 and Convention 1.7.2, in the class of function 𝒞1 (Ω) (here,
[112, Proposition 8.8.4] is relevant).
Going further, we note the following useful estimate.

Lemma 1.7.3 Let Ω ⊆ Rn be an open set with the property that ∂Ω is lower
Ahlfors regular and σ := H n−1 ∂Ω is a doubling measure on ∂Ω. Also, fix some
complex coefficient tensor A of type (n × n, M × N) along with an aperture parameter
κ > 0. Then there exists some constant C ∈ (0, ∞) such that for every distribution
N
u ∈ D (Ω) satisfying (1.7.8), every truncation parameter ε > 0, and every
exponents p, q ∈ (0, ∞] one has
 A   ε 
∂ u p, q ≤ C N (A∇u) p, q . (1.7.18)
ν [L (∂∗ Ω,σ)] M κ L (∂Ω,σ)

Proof This is seen from (1.7.15) and [112, Proposition 8.2.3, (8.9.8), (6.2.16),
(6.2.17)]. 

In relation to the conormal derivative operator introduced in Definition 1.7.1

we have the following companion to [113, Proposition 10.2.22], proving that the
conormal derivatives associated with any two coefficient tensors producing the same
second-order system actually differ by a linear combination of tangential derivatives.

Proposition 1.7.4 Let Ω be an open nonempty proper subset of Rn with a lower

Ahlfors regular boundary and such that σ := H n−1 ∂Ω is a doubling measure on
∂Ω; in particular, Ω is a set of locally finite perimeter. Denote by ν the geometric
measure theoretic outward unit normal to Ω, and abbreviate σ∗ := H n−1 ∂∗ Ω.
Also, fix an aperture parameter κ ∈ (0, ∞), a truncation parameter ε > 0, and two
integers M, N ∈ N. In addition, consider a C N -valued function u = (uβ )1≤β ≤ N in
Ω satisfying
N
1,1
u ∈ Wloc (Ω) Nκε u, Nκε (∇u) ∈ Lloc
, 1 (∂Ω, σ),

κ−n.t. κ−n.t.
and the nontangential traces u∂Ω , (∇u)∂Ω (1.7.19)

exist at σ-a.e. point on ∂nta Ω.

κ−n.t.
Then, when considered on ∂∗ Ω, the function u∂Ω belongs to L1,loc
1 (∂ Ω, σ ) N
∗ ∗
and for any two complex coefficient tensors of type (n × n, M × N),
 αβ   αβ 
A = ar s 1≤r,s ≤n and B = br s 1≤r,s ≤n (1.7.20)
1≤α ≤M 1≤α ≤M
1≤β ≤ N 1≤β ≤ N

with the property that

192 1 Integral Representations and Integral Identities

L A = LB where
    (1.7.21)
αβ αβ
L A := ar s ∂r ∂s 1≤α ≤M and LB := br s ∂r ∂s 1≤α ≤M ,
1≤β ≤ N 1≤β ≤ N

one has    κ−n.t.  

αβ 
ar s − br s ∂τr s uβ ∂Ω
αβ
∂νAu − ∂νB u = 1
2 1≤α ≤M (1.7.22)
at σ-a.e. point on ∂∗ Ω.
In particular,
 κ−n.t. 
∂νAu = ∂νB u at σ-a.e. point on ∂∗ Ω \ supp u|∂Ω . (1.7.23)

As a corollary,
with Ω as before, whenever A, B are two coefficient tensors for which
L A = LB and the function u satisfies the properties listed in (1.7.19)
κ−n.t. (1.7.24)
as well as u∂Ω = 0 at σ-a.e. point on ∂∗ Ω, it follows that ∂νAu = ∂νB u
at σ-a.e. point on ∂∗ Ω.
κ−n.t.
Proof That u∂Ω ∈ L1,loc 1 (∂ Ω, σ ) N follows from [113, Proposition 11.3.2]
∗ ∗
(used with p = q = 1). As regards the claim made in (1.7.22), with the coefficient
tensors A, B as in (1.7.20)-(1.7.21), define
αβ αβ αβ
cr s := ar s − br s for each α, β, r, s. (1.7.25)
 
When tested on functions of the form δββ eξ ·x 1≤β ≤ N , where β ∈ {1, . . . , N } and
ξ ∈ Rn are arbitrary and fixed, the coincidence of L A with LB forces that for each
α ∈ {1, . . . , M } and β ∈ {1, . . . , N } we have
αβ
cr s ξr ξs = 0 for each ξ = (ξ j )1≤ j ≤n ∈ Rn . (1.7.26)

After differentiating with respect to ξr and ξs , this ultimately implies

αβ αβ
cr s = −csr for each r, s ∈ {1, . . . , n},
(1.7.27)
α ∈ {1, . . . , M }, and β ∈ {1, . . . , N }.

Granted this, at σ-a.e. point on ∂∗ Ω we may compute

1.7 Green-Type Formulas for Second-Order Systems 193
  κ−n.t.    κ−n.t. 
αβ  αβ 
∂νAu − ∂νB u = νr ar s ∂s uβ ∂Ω − νr ar s ∂s uβ ∂Ω
1≤α ≤M 1≤α ≤M
  κ−n.t. 
αβ 
= νr cr s ∂s uβ ∂Ω
1≤α ≤M
  κ−n.t.   
 κ−n.t.
αβ  αβ 
= 2 νr cr s
1
∂s uβ ∂Ω + 
2 νr cr s ∂s uβ ∂Ω
1
1≤α ≤M 1≤α ≤M
  κ−n.t.   
 κ−n.t.
αβ  αβ 
= 2 νr cr s
1
∂s uβ ∂Ω − 
2 νs cr s ∂r uβ ∂Ω
1
1≤α ≤M 1≤α ≤M
  κ−n.t.  κ−n.t.  
αβ   
= 12 cr s νr ∂s uβ ∂Ω − νs ∂r uβ ∂Ω
1≤α ≤M
  κ−n.t.  
= 1 αβ
2 cr s ∂τr s uβ  ∂Ω
, (1.7.28)
1≤α ≤M

where the first equality originates in (1.7.9), the second one comes from (1.7.25),
the third and firth ones involve trivial algebra, the fourth one may be justified by
interchanging r with s (in the second vector) then invoking (1.7.27), and the last one
is implied by [113, Proposition 11.3.2, (11.3.26)]. From (1.7.28), the claim made in
(1.7.22) follows in view of (1.7.25).
Finally, (1.7.23) is a consequence of (1.7.22) and [113, (11.1.38)], while (1.7.24)
is a particular case of (1.7.23). 

Remark 1.7.5 Assume


n  
n 
'= 'αγ γβ
D br ∂r 1≤α ≤M and D = bs ∂s ;
1≤γ ≤ M
(1.7.29)
r=1 ;
1≤γ ≤ M s=1 1≤β ≤ N

are two homogeneous (constant) coefficient first-order systems in Rn with the prop-
erty that
' = 0.
DD (1.7.30)
Also, suppose the set Ω and the function u are as in Proposition 1.7.4. Then the conor-
A;
mal derivative ∂ν D, D u associated with the coefficient tensor (with the summation
convention over repeated indices in effect)
 αβ  αβ αγ γβ
AD,D
' := ar s 1≤r,s ≤n where each ar s := ' br bs (1.7.31)
1≤α ≤M
1≤β ≤ N

κ−n.t.
may be expressed as a linear combination of tangential derivatives of u∂Ω . Specif-
ically, at σ-a.e. point on ∂∗ Ω one has
  κ−n.t.  
∂ν D, D u = 1 ar s ∂τ uβ 
A; αβ
2 rs ∂Ω 1≤α ≤M
  κ−n.t.  
= 1 'αγ γβ
2 br bs ∂τr s uβ  ∂Ω
. (1.7.32)
1≤α ≤M
194 1 Integral Representations and Integral Identities

Indeed, this is seen from (1.7.22) written for A := AD,D

' and B := 0 (so that
L A = 0 = LB ).

There is yet a more flexible approach to defining conormal derivatives which we

wish to elaborate on. To set the stage, fix

Ω ⊆ Rn open set with the property that ∂Ω is lower Ahlfors regular

(1.7.33)
and such that σ := H n−1 ∂Ω is a doubling measure on the set ∂Ω.
In particular, Ω is a set of locally finite perimeter, and its geometric measure theoretic
outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Next, let L be a given homogeneous,
constant (complex) coefficient, second-order M × N system in Rn . Suppose said
system factors as
'
L = DD (1.7.34)
where D' and D are two homogeneous, constant (complex) coefficient, first-order sys-
N
tems in Rn . Then for any given vector-valued distribution u ∈ D (Ω) satisfying
(for some aperture parameter κ > 0)

Du has locally integrable components in Ω

κ−n.t. (1.7.35)
and (Du)∂Ω exists at σ-a.e. point on ∂∗ Ω,

we agree to define the conormal derivative of u associated with the factorization

(1.7.34) as
' κ−n.t.
' ν)(Du)
∂νD,D u := (−i)Sym( D; . (1.7.36)
∂Ω
' D
To shed light on how this point of view relates to Definition 1.7.1, write D,
explicitly as
 n 
' 'αγ
D= br ∂r 1≤α ≤M (1.7.37)
r=1 1≤γ ≤ N

and

n 
γβ
D= bs ∂s 1≤γ ≤ N (1.7.38)
s=1 1≤β ≤M

then define the coefficient tensor (with the summation convention over repeated
indices in effect)
 αβ  αβ αγ γβ
AD,D
' := ar s 1≤α ≤M where each ar s := '
br bs . (1.7.39)
1≤β ≤ N
1≤r,s ≤n

Based on these considerations and (1.7.34), we then see that L may be explicitly
represented as
L = L A D,
;D
, (1.7.40)
i.e., (cf. (1.7.7))  
αβ
L = ar s ∂r ∂s 1≤α ≤M . (1.7.41)
1≤β ≤ N
1.7 Green-Type Formulas for Second-Order Systems 195
N
Moreover, for any vector-valued distribution u = (uβ )1≤β ≤ N ∈ D (Ω) the
demands in (1.7.8) written for the coefficient tensor A := AD,D ' are satisfied if the
properties stipulated in (1.7.35) hold, and the conormal derivative of u with respect
to the coefficient tensor A := AD,D
' as originally defined in (1.7.9) presently becomes
   κ−n.t.    κ−n.t. 
αγ  γβ
u = νr ar s ∂s uβ ∂Ω br bs ∂s uβ ∂Ω
A D, αβ
= νr '
;D
∂ν
1≤α ≤M 1≤α ≤M
κ−n.t. '
' ν)(Du)
= (−i)Sym( D; = ∂νD,D u at σ-a.e. point on ∂∗ Ω. (1.7.42)
∂Ω

This shows that the “new” conormal derivative (from (1.7.36)) is consistent with the
“old” point of view (from Definition 1.7.1).
Let us illustrate how the above philosophy is implemented in the case of the
complex Lamé system Lλ,μ = μ Δ + (λ + μ)∇div, for some given Lamé moduli
μ, λ ∈ C. From (1.3.62) we see that for each ζ ∈ C we may express

Lλ,μ = −2ζ Def  Def + (λ + μ − ζ)∇div − (μ − ζ)∇ ∇. (1.7.43)

For the goals we have in mind, it is convenient to recast this as

'ζ D
Lλ,μ = D (1.7.44)

where

  Def
( +
  )
Dζ := − 2ζ Def , (λ + μ − ζ)∇, −(μ − ζ)∇ and D := ) div ,,
' (1.7.45)
* ∇ -
are homogeneous, constant coefficient, first-order systems in Rn . For each ζ ∈ C let
us agree to abbreviate
Aζ := AD' ζ ,D (1.7.46)

where AD' ζ ,D is the coefficient tensor associated with the systems D 'ζ , D as in
(1.7.39). Assume next that Ω is as in (1.7.33) and consider a vector-valued function
1,1 n
u = (u j )1≤ j ≤n ∈ Wloc (Ω, L n ) such that
κ−n.t.
(∇u)∂Ω exists (in Cn ) at σ-a.e. point on ∂∗ Ω
2
(1.7.47)

(in particular, u satisfies all conditions in (1.7.35), written for D as in (1.7.45)). Then
thanks to (1.7.46), (1.7.42), (1.7.36), and (1.7.45) we may write
196 1 Integral Representations and Integral Identities
AD
; ' ζ ,D κ−n.t.
A
∂ν ζ u = ∂ν ζ,D D
u = ∂ν 'ζ ; ν)(Du)
u = (−i)Sym( D ∂Ω
  κ−n.t. κ−n.t.
= 2ζ Def u ∂Ω ν + (λ + μ − ζ)(divu)∂Ω ν
κ−n.t.
+ (μ − ζ)(∇u)∂Ω ν, (1.7.48)

where the last step also makes use of the formulas for transpose of the deformation
tensor and the Jacobian operator from (1.3.58) and (1.3.57). Classically,
  κ−n.t. κ−n.t.
∂ν u := 2μ Def u ∂Ω ν + λ(divu)∂Ω ν
Trac

 κ−n.t. κ−n.t.  κ−n.t.

= μ (∇u)∂Ω + (∇u) ∂Ω ν + λ(divu)∂Ω ν (1.7.49)

is referred to as the traction conormal derivative (for the Lamé system). Using this,
we may re-frame (1.7.48) as
   κ−n.t. κ−n.t. κ−n.t. 
∂ν ζ u = ∂ν u + (μ − ζ) − 2 Def u ∂Ω ν + (divu)∂Ω ν + (∇u)∂Ω ν (1.7.50)
A Trac

A
so, in particular, the conormal derivative ∂ν ζ u reduces precisely to the traction
Trac
conormal derivative ∂ν u when ζ = μ. It is of interest to note (compare with
Proposition 1.7.4) that, for any ζ ∈ C, the difference between the conormal deriva-
tive associated with the coefficient tensor Aζ and the classical traction conormal
derivative is given by
 κ−n.t. κ−n.t. 
∂ν ζ u − ∂ν u = (μ − ζ) ν j (∂k uk )∂Ω − νk (∂j uk )∂Ω
A Trac
(1.7.51)
1≤ j ≤n

which, at least if u is reasonably behaved near the boundary (cf. [113, Propo-
sition
 11.3.2]), may be further written in terms of tangential derivatives as (μ −
 κ−n.t.  

ζ) ∂τ j k uk ∂Ω . Finally, we wish to note that the pseudo-stress conor-
1≤ j ≤n
mal derivative operator is defined when 3μ + λ  0 as

2μ(μ + λ) κ−n.t. (μ + λ)(2μ + λ) κ−n.t.

(Def u)∂Ω ν + (divu)∂Ω ν
ψ
∂ν u :=
3μ + λ 3μ + λ

2μ2 κ−n.t.
+ (∇u)∂Ω ν, (1.7.52)
3μ + λ
to which (1.7.48) reduces precisely when

μ(μ + λ)
ζ := . (1.7.53)
3μ + λ

Moving on, the goal in the next lemma is to decompose the nontangential trace
of the full gradient of a function as an algebraically “twisted” sum involving the
1.7 Green-Type Formulas for Second-Order Systems 197

conormal derivative and “tangential” derivatives of said function (see also Proposi-
tion 1.7.7 in this regard).
Lemma 1.7.6 Suppose Ω ⊆ Rn is an open set with the property that ∂Ω is lower
Ahlfors regular and σ := H n−1 ∂Ω is a doubling measure on ∂Ω. In particular,
Ω is a set of locally finite perimeter, and its geometric measure theoretic outward
unit normal ν = (ν1, . . . , νn ) is defined
 αβσ-a.e.
 on ∂∗ Ω. Having fixed some M ∈ N,
consider a coefficient tensor A = ar s 1≤r,s ≤n with complex entries, with the
1≤α,β ≤M
property that the M × M homogeneous second-order system L = L A associated
with A in Rn as in (1.3.15) is weakly elliptic (in the sense of (1.3.3)). Finally, fix an
1,1 M
aperture parameter κ > 0 and pick a function u = (uγ )1≤γ ≤M ∈ Wloc (Ω) for
κ−n.t.
which the nontangential trace (∇u)∂Ω exists (in C M×n ) at σ-a.e. point on ∂∗ Ω.
Then (with notation introduced in (1.3.2), (1.3.7), and (1.7.9)) one has
κ−n.t.  
−1
(∇u)∂Ω = − L(ν) ∂νAu ⊗ ν (1.7.54)
 
 κ−n.t. κ−n.t. 
−1
νs (∂j uβ )∂Ω − ν j (∂s uβ )∂Ω
αβ
− L(ν) ν a
γα r r s
1≤γ ≤M
1≤ j ≤n

at σ-a.e. point on ∂∗ Ω.
Proof Based on (1.3.2), (1.3.7), (1.7.9), and (1.7.16), at σ-a.e. point on ∂∗ Ω we may
write
κ−n.t.  −1

−(∇u)∂Ω − L(ν) ∂νAu ⊗ ν
 
κ−n.t.  −1 A

= − (∂j uγ )∂Ω − L(ν) ∂ν u νj
γ 1≤γ ≤M
1≤ j ≤n
 
 κ−n.t. 
−1 −1 A
= − L(ν) L(ν)(∂j u)∂Ω − ν j L(ν) ∂ν u
γ 1≤γ ≤M
1≤ j ≤n
 
 κ−n.t. 
−1 
= L(ν) − L(ν)(∂j u)∂Ω − ν j ∂νAu
γ 1≤γ ≤M
1≤ j ≤n
 
 κ−n.t. κ−n.t.
−1
νr νs ar s (∂j uβ )∂Ω ν j νr ar s (∂s uβ )∂Ω
αβ αβ
= L(ν) γα
−
1≤γ ≤M
1≤ j ≤n
 
 κ−n.t. κ−n.t. 
−1
νs (∂j uβ )∂Ω − ν j (∂s uβ )∂Ω
αβ
= L(ν) ν a
γα r r s
(1.7.55)
1≤γ ≤M
1≤ j ≤n

from which the desired conclusion readily follows. 

198 1 Integral Representations and Integral Identities

Here is a version of [113, Proposition 11.4.4] adapted to vector-valued functions

and weakly elliptic systems.

Proposition 1.7.7 Suppose Ω ⊆ Rn is an open set with the property that ∂Ω is lower
Ahlfors regular and σ := H n−1 ∂Ω is a doubling measure on ∂Ω. In particular,
Ω is a set of locally finite perimeter, and its geometric measure theoretic outward
unit normal ν = (ν1, . . . , νn ) is defined
 αβσ-a.e.
 on ∂∗ Ω. Having fixed some M ∈ N,
consider a coefficient tensor A = ar s 1≤r,s ≤n with complex entries, with the
1≤α,β ≤M
property that the M × M homogeneous second-order system L = L A associated
with A in Rn as in (1.3.15) is weakly elliptic (in the sense of (1.3.3)). Finally, fix an
integrability exponent p ∈ [1, ∞], an aperture parameter κ ∈ (0, ∞), a truncation
1,1 M
parameter ε > 0, and consider a function u = (uβ )1≤β ≤M ∈ Wloc (Ω) with

Nκε u, Nκε (∇u) ∈ Lloc (∂Ω, σ)

p
(1.7.56)

and such that the nontangential traces

κ−n.t. κ−n.t.
uβ ∂Ω and (∂j uβ )∂Ω exist at σ-a.e. point on ∂∗ Ω
(1.7.57)
for each β ∈ {1, . . . , M } and j ∈ {1, . . . , n}.
κ−n.t.
Then u∂Ω ∈ L1,loc (∂∗ Ω, σ) and, with notation introduced in (1.3.2), (1.3.7),
p M

(1.7.9), at σ-a.e. point on ∂∗ Ω one has

κ−n.t.  
−1
(∇u)∂Ω = − L(ν) ∂νAu ⊗ ν
 
−1  κ−n.t. 
uβ 
αβ
− L(ν) ν a ∂
γα r r s τs j ∂Ω
. (1.7.58)
1≤γ ≤M
1≤ j ≤n

Moreover, if one abbreviates

 
  κ−n.t.  
uβ 
αβ
∇tan
A
u := νr ar s ∇tan ∂Ω
(1.7.59)
s
1≤α ≤M

and  
  κ−n.t.  
∇tan u := ∇tan uγ ∂Ω (1.7.60)
j 1≤γ ≤M
1≤ j ≤n

then under the stronger assumption that Ω is a UR domain and that p ∈ (1, ∞) one
has
κ−n.t.  
−1  A A 
(∇u)∂Ω = − L(ν) ∂ν u − ∇tan u ⊗ ν + ∇tan u (1.7.61)

at σ-a.e. point on ∂Ω.

1.7 Green-Type Formulas for Second-Order Systems 199

Proof Formula (1.7.58) is a consequence of Lemma 1.7.6 and [113, Proposi-

tion 11.3.2]. Assuming that Ω is actually a UR domain and p ∈ (1, ∞), formula
(1.7.61) follows from (1.7.58), [113, (11.4.8)], (1.7.59), and (1.7.60). 

We continue by deriving a formula relating the boundary integral of the conormal

derivative of a given function to the solid integral of the action of the system
(associated with the coefficient tensor which has produced the conormal derivative
in the first place) on said function.

Proposition 1.7.8 Assume Ω ⊆ Rn is an open set with a lower Ahlfors regular

boundary and with the property that σ := H n−1 ∂Ω is a doubling measure. In
particular, Ω is a set of locally finite perimeter, and its geometric measure theoretic
outward unit normal ν is defined σ-a.e. on ∂∗ Ω.
Having fixed some M, N ∈ N, suppose A is a complex coefficient tensor of type
(n × n, M × N) and denote by L A the homogeneous constant (complex) coefficient
second-order M × N system in Rn associated with A as in (1.7.7). Finally, fix an
aperture parameter κ > 0 and consider a complex vector-valued function
N
u ∈ Lloc
1
(Ω, L n ) (1.7.62)

satisfying (with all derivatives taken in the sense of distributions in Ω):

M ·n
A∇u ∈ Lloc
1 (Ω, L n ) , Nκ (A∇u) ∈ L 1 (∂Ω, σ),
M
L Au ∈ L 1 (Ω, L n )
, and the nontangential (1.7.63)
κ−n.t.
boundary trace (A∇u)∂Ω exists σ-a.e. on ∂nta Ω.

Then for any other specified aperture parameter κ ∈ (0, ∞) the nontangential
κ −n.t.
boundary trace (A∇u)∂Ω exists at σ-a.e. point on ∂nta Ω and is actually indepen-
dent of κ . Moreover, with the dependence on the aperture parameter dropped and
with the conormal derivative considered in the sense of (1.7.8)-(1.7.9), one has
∫ ∫
∂νAu dσ = L Au dL n (1.7.64)
∂∗ Ω Ω

when either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when

Ω is unbounded and ∂Ω is bounded, formula (1.7.64) continues to hold under the
additional assumption that there exists λ ∈ (1, ∞) such that38
∫
| A∇u| dL n = o(R) as R → ∞. (1.7.65)
[B(0,λ R)\B(0,R)]∩Ω

As a corollary, if the last property in the first line of (1.7.63) is replaced by

L Au = 0 in the sense of distributions in Ω, (1.7.66)

38 for example, (1.7.65) holds if | A(∇u)(x)| = o(|x | 1−n ) as |x | → ∞

200 1 Integral Representations and Integral Identities

then (1.7.64) becomes

∫
∂νAu dσ = 0. (1.7.67)
∂∗ Ω

Proof Assume the coefficient tensor A is explicitly expressed as in (1.7.1). Fix an

arbitrary index α ∈ {1, . . . , M } and define the vector field
 
αβ
Fα := ar s ∂s uβ in Ω, (1.7.68)
1≤r ≤n

where (uβ )1≤β ≤ N are the scalar components of the vector-valued function u. Then
[112, Theorem 1.2.1] applies to Fα and gives, under the conditions stipulated in the
current statement, that
∫ ∫
 A 
∂ν u α dσ = (L Au)α dL n . (1.7.69)
∂∗ Ω Ω

In view of the arbitrariness of α, all desired conclusions follow. 

Theorem 1.7.10, stated a little below, establishes a version of Green’s second

identity for second-order systems in a very general geometric and analytic setting.
As a preamble, this requires that we make one quick definition.

Definition 1.7.9 Suppose

Σ ⊆ Rn is a H n−1 -measurable set satisfying

(1.7.70)
H n−1 (Σ ∩ K) < ∞ for each compact K ⊂ Rn .

As in the past, abbreviate σ := H n−1 Σ and denote by L 0 (Σ, σ) the space of all
μ-measurable functions f on Σ with the property that | f | < +∞ at σ-a.e. on Σ. In
this setting, call (X, Y) a dual system on (Σ, σ) provided

∫X, Y are subsets of L (Σ, σ) with the property that

0
(1.7.71)
Σ
| f ||g| dσ < ∞ for any given f ∈ X and g ∈ Y.

Finally, call a dual system (X, Y) on (Σ, σ) regular if, in fact, X, Y are subsets of
< p
Lloc (Σ, σ).
0<p ≤∞

Basic examples of regular dual systems are offered by: pairs of dual Lebesgue
spaces, i.e.,

let Σ be as in (1.7.70) and, with σ := H n−1 Σ, take X := L p (Σ, σ) and

(1.7.72)
Y := L p (Σ, σ) with p, p ∈ [1, ∞] satisfying 1/p + 1/p = 1,
pairs of dual Lorentz spaces (cf. O’Neil’s inequality recorded in [112, (6.2.61)]),
i.e.,
1.7 Green-Type Formulas for Second-Order Systems 201

let Σ be as in (1.7.70) and, with σ := H n−1 Σ, take X := L p,q (Σ, σ)

together with Y := L p ,q (Σ, σ) where the indices p, q, p , q ∈ [1, ∞] (1.7.73)
are assumed to satisfy 1/p + 1/p = 1 as well as 1/q + 1/q = 1,
pairs of dual Muckenhoupt weighted Lebesgue spaces (cf. item (2) in [112,
Lemma 7.7.1]), i.e.,

let Σ be a closed Ahlfors regular set in Rn and, with σ := H n−1 Σ,

take X := L p (Σ, wσ) and Y := L p (Σ, w −p /p σ) with p, p ∈ (1, ∞)
(1.7.74)
satisfying 1/p + 1/p = 1, and the weight w in Muckenhoupt’s class
Ap (Σ, σ),

the pair consisting of a Morrey space and its pre-dual (cf. [113, Proposition 6.2.8,
(6.2.78)]), i.e.,

let Σ be a closed Ahlfors regular set in Rn and, with σ := H n−1 Σ, take

X := M p,λ (Σ, σ) and Y := B q,λ (Σ, σ) with exponents p, q ∈ (1, ∞) (1.7.75)
satisfying 1/p + 1/q = 1, and the parameter λ belonging to (0, n − 1).
While in all these examples it happens that X and Y are linear spaces, we emphasize
that this is not a requirement in the original definition of a dual system. In fact, a dual
system (X, Y) may simply be a pair of singletons (which is actually an important
particular case).
With this preamble taken care of, here is the general version of Green’s second
identity promised earlier.

Theorem 1.7.10 Assume Ω ⊆ Rn is an open set with a lower Ahlfors regular bound-
ary and with the property that σ := H n−1 ∂Ω is a doubling measure. In particular,
Ω is a set of locally finite perimeter, and its geometric measure theoretic outward
unit normal ν is defined σ-a.e. on ∂∗ Ω.
Having fixed some M, N ∈ N, suppose A is a complex coefficient tensor of type
(n × n, M × N) and denote by L A the homogeneous constant (complex) coefficient
second-order M × N system in Rn associated with A as in (1.7.7). Finally, fix
κ, κ > 0, assume (X0, Y0 ) and (X1, Y1 ) are two regular dual systems on (∂Ω, σ)
(in the sense of Definition 1.7.9), and consider two complex vector-valued functions
N M
u ∈ Lloc
1
(Ω, L n ) , w ∈ Lloc
1
(Ω, L n ) , (1.7.76)

satisfying (with all derivatives taken in the sense of distributions in Ω):

202 1 Integral Representations and Integral Identities
M ·n
Nκ u ∈ X0, A∇u ∈ Lloc
1 (Ω, L n ) , Nκ (A∇u) ∈ X1,
N ·n
Nκ w ∈ Y1, A ∇w ∈ Lloc
1 (Ω, L n ) , Nκ (A ∇w) ∈ Y0,
κ−n.t. κ−n.t. κ −n.t. κ −n.t.
u∂Ω , (A∇u)∂Ω , w ∂Ω , (A ∇w)∂Ω exist at σ-a.e. point on ∂nta Ω, (1.7.77)
M N
L Au ∈ Lloc1 (Ω, L n ) , LA w ∈ Lloc (Ω, L )
1 n ,
   
L Au, w ∈ L 1 (Ω, L n ), u, L 
A w ∈ L (Ω, L ).
1 n

Then for any other specified aperture parameter κ ∈ (0, ∞) the nontangential
κ −n.t. κ −n.t. κ −n.t. κ −n.t.
boundary traces u∂Ω , (A∇u)∂Ω , w ∂Ω , (A ∇w)∂Ω exist at σ-a.e. point
on ∂nta Ω and are independent of κ . Moreover, with the dependence on the aperture
parameter dropped and with the conormal derivatives interpreted in the sense of
(1.7.8)-(1.7.9), the following Green type formula (involving absolutely convergent
integrals) holds
∫ ∫
   
L Au, w dL − n
u, L 
A w dL
n
Ω Ω
∫ ∫
 n.t.   n.t. A 
= ∂νAu, w ∂Ω dσ − u , ∂ w dσ,
∂Ω ν (1.7.78)
∂∗ Ω ∂∗ Ω

when either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when

Ω is unbounded and ∂Ω is bounded, formula (1.7.78) continues to hold under the
additional assumption that there exists λ ∈ (1, ∞) such that39
∫
 
| A∇u||w| + |u|| A ∇w| dL n = o(R) as R → ∞. (1.7.79)
[B(0,λ R)\B(0,R)]∩Ω

As a corollary, if the last two lines in (1.7.77) are replaced by

L Au = 0 and L 
A w = 0 in the sense of distributions in Ω, (1.7.80)

then (1.7.78) becomes

∫ ∫
 n.t.   n.t. A 
∂νAu, w ∂Ω dσ = u , ∂ w dσ.
∂Ω ν (1.7.81)
∂∗ Ω ∂∗ Ω

Finally, similar results are valid if (X0, Y0 ) and (X1, Y1 ) are two dual systems
on (∂Ω, σ) which are not necessarily regular, with the understanding that, in such
a case, all nontangential boundary traces are taken with respect to an aperture
parameter which is < min{κ, κ }.

Before presenting the proof of this theorem we wish to note that a related version
holds with the doubling assumption on σ := H n−1 ∂Ω relaxed to simply asking

39 e.g., (1.7.79) holds if | A(∇u)(x)| |w(x)| = o(|x | 1−n ) and |u(x)| | A (∇w)(x)| = o(|x | 1−n )
1.7 Green-Type Formulas for Second-Order Systems 203

that this is a locally finite measure. In such a scenario, the aperture parameter κ is
assumed to be sufficiently large (depending on Ω), and the flexibility of changing
κ when considering nontangential boundary traces may be lost. Other than these
nuances, the format of the main result (i.e., formula (1.7.78)) remains unchanged.
The proof of this version of Theorem 1.7.10 requires that [112, Theorem 1.5.1] is
employed in place of [112, Theorem 1.4.1] (which is used below).
In the important particular case when each of the duals systems (X0, Y0 ) and
(X1, Y1 ) is a pair of singletons, the nontangential maximal function memberships in
the first two lines of (1.7.77) simply become the integrability conditions
∫ ∫
Nκ u · Nκ (A ∇w) dσ < +∞ and Nκ (A∇u) · Nκ w dσ < +∞. (1.7.82)
∂Ω ∂Ω

Another basic particular choice is having (X0, Y0 ) and (X1, Y1 ) pairs of dual
Lebesgue spaces on (∂Ω, σ). Specifically, with the exponents p, q, p , q ∈ [1, ∞]
such that 1/p + 1/p = 1 = 1/q + 1/q , this amounts to replacing the nontangential
maximal function memberships in the first two lines of (1.7.77) by

Nκ u ∈ L p (∂Ω, σ), Nκ (A∇u) ∈ L q (∂Ω, σ),

(1.7.83)
Nκ w ∈ L q (∂Ω, σ), Nκ (A ∇w) ∈ L p (∂Ω, σ).

Assuming this is the case, we wish to further single out a natural scenario when
the integrability conditions in the very last line in (1.7.77) are naturally satisfied.
1 (Ω, L n ) M and L  w ∈ L 1 (Ω, L n ) N ,
Concretely, if the demands that L Au ∈ Lloc A loc
made in the penultimate line in (1.7.77), are strengthened to
M
L Au ∈ L nq /(nq −n+1) (Ω, L n ) and
(1.7.84)
N
L
Aw ∈ L
np/(np−n+1) (Ω, L n ) ,

then the two integrability conditions in the last line of (1.7.77) are automatically
satisfied when Ω is bounded or ∂Ω is unbounded, since in such a scenario [112,
Proposition 8.6.3] and [112, Lemma 8.3.2] permit us to conclude that
 ∗
 np N np N
Nκ u ∈ L p (∂Ω, σ) ⇒ u ∈ L n−1 (Ω, L n ) = L n p−n+1 (Ω, L n ) ,
 ∗
 nq M nq
M
Nκ w ∈ L (∂Ω, σ) ⇒ w ∈ L n−1 (Ω, L )
q n = L nq −n+1 (Ω, L )n .

(1.7.85)
Let us also take a look at the case when Ω is an exterior domain, under the assumption
that the functions u, w satisfy

u(x) = O(|x| 1−n ) and w(x) = O(|x| 1−n ) as |x| → ∞,

(1.7.86)
(∇u)(x) = o(1) and (∇w)(x) = o(1) as |x| → ∞,
204 1 Integral Representations and Integral Identities

(which, incidentally, implies that condition (1.7.79) holds). Granted this, and further
assuming that p > 1, q < ∞, and that (1.7.84) replaces the penultimate line in
(1.7.77), it follows that the two integrability conditions in the last line of (1.7.77) are
also satisfied. Indeed, while in this case in place of (1.7.85) we only have (recall that
p
Lbdd has been introduced in (A.0.65))

np/(n−1) N
Nκ u ∈ L p (∂Ω, σ) =⇒ u ∈ Lbdd (Ω, L n ) ,
(1.7.87)
nq /(n−1) M
Nκ w ∈ L q (∂Ω, σ) =⇒ w ∈ Lbdd (Ω, L n ) ,

the decay conditions in the first line of (1.7.86) do ensure that both |u| np/(n−1) and
|w| nq /(n−1) are integrable in a neighborhood of infinity so that, ultimately, we still
have
 
N N ∗
u ∈ L np/(n−1) (Ω, L n ) = L np/(np−n+1) (Ω, L n ) ,
  (1.7.88)
M M ∗
w ∈ L nq /(n−1) (Ω, L n ) = L nq /(nq −n+1) (Ω, L n ) ,

in this case as well.

Here is the actual proof of Theorem 1.7.10.
Proof of Theorem 1.7.10 From assumptions and [112, Corollary 8.9.9] it follows
that there exists an exponent p ∈ (0, ∞] with the property that for any given aperture
parameter κ > 0 we have

Nκ u, Nκ (A∇u), Nκ w, Nκ (A ∇w) ∈ Lloc (∂Ω, σ),

p
(1.7.89)

and the nontangential boundary traces

κ −n.t. κ −n.t. κ −n.t. κ −n.t.
u∂Ω , (A∇u)∂Ω , w ∂Ω , (A ∇w)∂Ω exist σ-a.e. on ∂nta Ω
(1.7.90)
and agree with the respective boundary traces in (1.7.77).

To proceed, note that (1.7.76), the first two lines in (1.7.77) (or (1.7.89)), and
[112, Lemma 8.3.1] imply

∞ (Ω, L n ) N ∞ (Ω, L n ) M ·n
u ∈ Lloc , A∇u ∈ Lloc ,
(1.7.91)
∞ (Ω, L n ) M N ·n
w ∈ Lloc , A ∇w ∈ Lloc
∞ (Ω, L n ) .

Consider now the vector field F = (Fr )1≤r ≤n : Ω → Cn with components given
(with the summation convention over repeated indices enforced throughout the proof)
by
1.7 Green-Type Formulas for Second-Order Systems 205

Fr := (A∇u)α,r wα − uβ (A ∇w)β,r

αβ αβ ∞
= (ar s ∂s uβ )wα − uβ (asr ∂s wα ) ∈ Lloc (Ω, L n ), (1.7.92)

where the final membership is ensured by (1.7.91). In particular,

F ∈ Lloc
n
1
(Ω, L n ) . (1.7.93)

Let us compute divF in the sense of distributions in Ω. With this goal in mind, fix
an arbitrary scalar-valued function ϕ ∈ 𝒞∞c (Ω) and write
∫
 

D (Ω) div F, ϕ D(Ω) = − Fr ∂r ϕ dL n
Ω
∫
αβ
=− (ar s ∂s uβ )wα ∂r ϕ dL n
Ω
∫
αβ
+ uβ (asr ∂s wα )∂r ϕ dL n
Ω

=: I + I I. (1.7.94)
 
For each sufficiently small ε > 0 consider Ωε := x ∈ Ω : dist(x, ∂Ω) > ε . Using
N
a Friedrichs mollifier, we may construct a sequence uε = (uβε )1≤β ≤ N ∈ 𝒞∞ (Ωε )
such that
uε −−−−→
+
u at L n -a.e. point in Ω, and
ε→0
A∇uε −−−−→
+
A∇u at L n -a.e. point in Ω;
ε→0
moreover, for each fixed compact set K ⊂ Ω
M (1.7.95)
we have L Auε −−−−→
+
L Au in L 1 (K, L n )
ε→0
and there exists some small εK > 0 such that
   
sup #uε #[L ∞ (K, L n )] N +  A∇uε [L ∞ (K, L n )] M ·n < ∞,
0<ε<ε K

M
and also a sequence w ε = (wαε )1≤α ≤M ∈ 𝒞∞ (Ωε ) such that
206 1 Integral Representations and Integral Identities

w ε −−−−→
+
w at L n -a.e. point in Ω, and
ε→0
A ∇w ε −−−−→
+
A ∇w at L n -a.e. point in Ω;
ε→0
in addition, for each given compact set K ⊂ Ω
N (1.7.96)
we have L  ε
A w −−−−→
+
L
A w in L (K, L )
1 n
ε→0
and there exists some small εK > 0 such that
   
sup #w ε #[L ∞ (K, L n )] M +  A ∇w ε [L ∞ (K, L n )] N ·n < ∞.
0<ε<ε K

In relation to these, let us make a useful elementary observation of general nature,

to the that effect that
if (X, μ) is a measure space, { f j } j ∈N ⊆ L 1 (X, μ) is a convergent
sequence in L 1 (X, μ) to some f ∈ L 1 (X, μ), and {g j } j ∈N is a bounded
sequence in L ∞ (X, μ) such that the pointwise limit g := lim g j exists (1.7.97)
∫ ∫
j→∞
μ-a.e. in X, then f g ∈ L 1 (X, μ) and X f j g j dμ −→ X f g dμ as
j → ∞.
Then, based on (1.7.91), (1.7.95), (1.7.96), the last two lines in (1.7.77), Lebesgue’s
Dominated Convergence Theorem, and (1.7.97) we may compute
∫
αβ
I = − lim+ (ar s ∂s uβε )wαε ∂r ϕ dL n
ε→0 Ω
∫ ∫
αβ αβ
= − lim+ (ar s ∂s uβε )∂r (wαε ϕ) dL n + lim+ (ar s ∂s uβε )(∂r wαε )ϕ dL n
ε→0 Ω ε→0 Ω
∫ ∫
 
= lim+ (L Auε )α wαε ϕ dL n + lim+ A∇uε, ∇w ε ϕ dL n
ε→0 Ω ε→0 Ω
∫ ∫
   
= L Au, w ϕ dL n + lim+ A∇uε, ∇w ε ϕ dL n . (1.7.98)
Ω ε→0 Ω

In a similar fashion,
∫ ∫
αβ
I I = − uβ (L A w)β ϕ dL − lim+ (∂r uβε )(asr ∂s wαε )ϕ dL n
 n
Ω ε→0 Ω
∫ ∫
   
=− u, L 
A w ϕ dL − lim+
n
A∇uε, ∇w ε ϕ dL n . (1.7.99)
Ω ε→0 Ω

At this stage, from (1.7.94) and (1.7.98)-(1.7.99) we conclude that

   
divF = L Au, w − u, L  A w in D (Ω). (1.7.100)

In particular, the last line in (1.7.77) ensures that

1.7 Green-Type Formulas for Second-Order Systems 207

divF ∈ L 1 (Ω, L n ). (1.7.101)

 
Also, if we choose an aperture parameter κ∗ ∈ 0, min{κ, κ } , from (1.7.92) and
[112, (8.2.25)] we see that
 
Nκ∗ F ≤ C Nκ (A∇u)Nκ w + Nκ uNκ (A ∇w) on ∂Ω, (1.7.102)

for some constant C ∈ (0, ∞) which depends only on the coefficient tensor A. In turn,
from (1.7.102), [112, (8.2.26)], the nontangential maximal function memberships in
the first two lines of (1.7.77), and (1.7.71) we conclude that

Nκ∗ F ∈ L 1 (∂Ω, σ). (1.7.103)

Going further, observe that (1.7.92) and (1.7.90) imply that the nontangential
boundary trace
κ −n.t.

F exists at σ-a.e. point on ∂nta Ω. (1.7.104)
∂Ω
In fact, with the dependence on κ dropped, for each r ∈ {1, . . . , n} we have
 n.t.   αβ n.t.   n.t.   n.t.   αβ n.t. 

F = (ar s ∂s uβ )∂Ω wα ∂Ω − uβ ∂Ω (asr ∂s wα )∂Ω (1.7.105)
∂Ω r

at σ-a.e. point on ∂nta Ω. As a consequence of this, (1.7.9), and [112, Proposi-

tion 8.8.6], we obtain
 n.t.   n.t.   n.t.  
ν · F ∂Ω = ∂νAu, w ∂Ω − u∂Ω, ∂νA w at σ-a.e. point on ∂∗ Ω. (1.7.106)

Finally, (1.7.79) ensures that the vector field F satisfies the growth condition [112,
 [112, Theorem 1.2.1] applies and the Diver-
(1.2.3)]. Granted these properties of F,
gence Formula [112, (1.2.2)] presently yields (1.7.78) on account of (1.7.100) and
(1.7.106). 

Other versions of Theorem 1.7.10 of practical interest are valid. To state such
 a re-
sult, given an open set Ω ⊆ Rn , the reader is reminded that the pairing ℰ (Ω) ·, · ℰ(Ω)
has been defined in (A.0.104).

Theorem 1.7.11 Suppose Ω ⊆ Rn is an open set with a lower Ahlfors regular

boundary and with the property that σ := H n−1 ∂Ω is a doubling measure. Hence,
Ω is a set of locally finite perimeter, and its geometric measure theoretic outward
unit normal ν is defined σ-a.e. on ∂∗ Ω.
Given some arbitrary M, N ∈ N, suppose A is a complex coefficient tensor of type
(n × n, M × N) and denote by L A the homogeneous constant (complex) coefficient
second-order M × N system in Rn associated with A as in (1.7.7). Also, pick some
aperture parameters κ, κ > 0, and assume (X0, Y0 ) and (X1, Y1 ) are two regular
dual systems on (∂Ω, σ) (in the sense of Definition 1.7.9). Finally, consider two
complex vector-valued functions
208 1 Integral Representations and Integral Identities

1,1 N 1,1 M
u ∈ Wloc (Ω) , w ∈ Wloc (Ω) , (1.7.107)

satisfying (with all derivatives taken in the sense of distributions in Ω):

Nκ u ∈ X0 and Nκ (∇u) ∈ X1,

Nκ w ∈ Y1 and Nκ (∇w) ∈ Y0,
κ−n.t. κ−n.t. κ −n.t. κ −n.t.
u∂Ω , (∇u)∂Ω , w ∂Ω , (∇w)∂Ω exist σ-a.e. on ∂nta Ω, (1.7.108)
M
L Au ∈ ℰ (Ω) and w is of class 𝒞∞ near supp (L Au),
N
L
A w ∈ ℰ (Ω) and u is of class 𝒞∞ near supp (L A w).

Then for any other specified aperture parameter κ ∈ (0, ∞) the nontangential
κ −n.t. κ −n.t. κ −n.t. κ −n.t.
boundary traces u∂Ω , (∇u)∂Ω , w ∂Ω , (∇w)∂Ω exist at σ-a.e. point on
∂nta Ω and are independent of κ . In addition, with the dependence on the aperture
parameter dropped and with the conormal derivatives interpreted in the sense of
(1.7.8)-(1.7.9), the following Green-type formula holds
   
[ℰ (Ω)] M L Au, w [ℰ(Ω)] M − [ℰ (Ω)] N L A w, u [ℰ(Ω)] N
∫ ∫
 A n.t.   n.t. A 
= 
∂ν u, w ∂Ω dσ − u∂Ω, ∂ν w dσ, (1.7.109)
∂∗ Ω ∂∗ Ω

when either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when Ω

is unbounded and ∂Ω is bounded, formula (1.7.109) continues to be true under the
additional assumption that there exists λ ∈ (1, ∞) such that (1.7.79) holds.
Finally, similar results hold when (X0, Y0 ) and (X1, Y1 ) are two dual systems
on (∂Ω, σ) which are not necessarily regular, with the understanding that, in such
a case, all nontangential boundary traces are taken with respect to an aperture
parameter which is < min{κ, κ }.

Proof We proceed along the lines of the argument in the proof of Theorem 1.7.10, so
we will only indicate the main changes. Recall the 𝒞∞ -singular support of a distribu-
tion from [112, Definition 2.8.5]. Pick ψ ∈ 𝒞∞ c (Ω) with ψ ≡ 1 near supp (L Au) and
whose support is contained in 𝒞∞ -singsup u, along with ξ ∈ 𝒞∞ c (Ω) with ξ ≡ 1 near
supp (L A w) and whose support is contained in 𝒞∞ -singsup w. Using a Friedrichs
N
mollifier, construct a sequence uε = (uβε )1≤β ≤ N ∈ 𝒞∞ (Ωε ) satisfying

uε −−−−→
+
u and ∇uε −−−−→
+
∇u at L n -a.e. point in Ω,
ε→0 ε→0
ψuε −−−−→
+
ψu in the topology of test functions in Ω,
ε→0
for each compact K ⊂ Ω there exists εK > 0 such that (1.7.110)
   
sup #uε #[L ∞ (K, L n )] N + ∇uε [L ∞ (K, L n )] N ·n < +∞,
0<ε<ε K
1.7 Green-Type Formulas for Second-Order Systems 209
M
along with a sequence w ε = (wαε )1≤α ≤M ∈ 𝒞∞ (Ωε ) such that

w ε −−−−→
+
w and ∇w ε −−−−→
+
∇w at L n -a.e. point in Ω,
ε→0 ε→0
ξw ε −−−−→
+
ξw in the topology of test functions in Ω,
ε→0
for each compact K ⊂ Ω there exists εK > 0 such that (1.7.111)
   
sup #w ε #[L ∞ (K, L n )] M + ∇w ε [L ∞ (K, L n )] M ·n < +∞.
0<ε<ε K

In particular,
N
uε −−−−→
+
1 (Ω, L n )
u in Lloc
ε→0
M (1.7.112)
and w ε −−−−→ w in Lloc
1 (Ω, L n ) ,
ε→0+

hence
M
L Auε −−−−→
+
L Au in D (Ω)
ε→0
N (1.7.113)
and L A w ε −−−−→ L A w in D (Ω) .
ε→0+

Let us also recall (cf. [66, Theorems 2.1.8, pp. 38-39]) that

if {ω j } j ∈N is a sequence of distributions in Ω converging to some

ω ∈ D (Ω) in D (Ω), then for each sequence {φ j } j ∈N ⊆ 𝒞∞ c (Ω)
converging to some φ ∈ 𝒞∞ c (Ω) in the topology of test functions in Ω
(1.7.114)
we have lim ω j , φ j  = ω, φ.
j→∞

Then, in place of (1.7.98) we now have

∫ ∫
 
I = lim+ (L Auε )α wαε ϕ dL n + lim+ A∇uε, ∇w ε ϕ dL n
ε→0 Ω ε→0 Ω
∫
   
= lim+ [D (Ω)] M (L Auε )α, ϕ (ψ w ε )α [D(Ω)] M + lim+ A∇uε, ∇w ε ϕ dL n
ε→0 ε→0 Ω
∫
   
= [D (Ω)] M (L Au)α, ϕ (ψ w)α [D(Ω)] M + lim+ A∇uε, ∇w ε ϕ dL n
ε→0 Ω
∫
   
= D (Ω) (ψwα )(L Au)α, ϕ D(Ω) + lim+ A∇uε, ∇w ε ϕ dL n, (1.7.115)
ε→0 Ω

thanks to (1.7.110)-(1.7.113) and (1.7.114). Similarly, in lieu of (1.7.99) we presently

obtain
∫
   
I I = − D (Ω) (ξuβ )(L A w)β, ϕ D(Ω) − lim+ A∇uε, ∇w ε ϕ dL n . (1.7.116)
ε→0 Ω

From (1.7.115)-(1.7.116), in place of (1.7.100) we now arrive at the conclusion that

210 1 Integral Representations and Integral Identities

divF = (ψwα )(L Au)α − (ξuβ )(L A w)β ∈ ℰ (Ω). (1.7.117)

Consequently,
 

(𝒞∞b (Ω))∗ divF, 1 𝒞∞b (Ω) (1.7.118)
   
= ℰ (Ω) (ψwα )(L Au)α, 1 ℰ(Ω) − ℰ (Ω) (ξuβ )(L A w)β, 1 ℰ(Ω)
   
= ℰ (Ω) (L Au)α, ψwα ℰ(Ω) − ℰ (Ω) (L A w)β, ξuβ ℰ(Ω)
   
= [ℰ (Ω)] M L Au, w [ℰ(Ω)] M − [ℰ (Ω)] N L A w, u [ℰ(Ω)] N .

At this stage, [112, Theorem 1.4.1] applies, and the Divergence Formula [112,
(1.4.6)] establishes (1.7.109), on account of (1.7.118). 
Moving on, we augment Theorem 1.7.10 by dealing with Green’s first identity for
second-order systems in the theorem below.

Theorem 1.7.12 Suppose Ω ⊆ Rn is an open set with a lower Ahlfors regular

boundary and with the property that σ := H n−1 ∂Ω is a doubling measure. Hence,
Ω is a set of locally finite perimeter, and its geometric measure theoretic outward
unit normal ν is defined σ-a.e. on ∂∗ Ω.
Next, assume A is a complex coefficient tensor of type (n × n, M × N), for some
M, N ∈ N, and denote by L A the homogeneous constant (complex) coefficient
second-order M × N system in Rn associated with A as in (1.7.7). Finally, pick
some κ, κ > 0, assume (X, Y) is a regular dual system on (∂Ω, σ) (in the sense of
Definition 1.7.9), and consider two complex vector-valued functions
N 1,1 M
u ∈ Lloc
1
(Ω, L n ) , w ∈ Wloc (Ω) , (1.7.119)

satisfying (with all derivatives taken in the sense of distributions in Ω):

M ·n
A∇u ∈ Lloc
1 (Ω, L n ) , Nκ (A∇u) ∈ X, Nκ w ∈ Y,
κ−n.t. κ −n.t.
the nontangential traces (A∇u)∂Ω and w ∂Ω exist σ-a.e. on ∂nta Ω,
(1.7.120)
 
A∇u, ∇w belongs to L (Ω, L ), and
1 n

1 (Ω, L n ) M ,
 
L Au ∈ Lloc L Au, w ∈ L 1 (Ω, L n ).

Then for any other specified aperture parameter κ ∈ (0, ∞) the nontangential
κ −n.t. κ −n.t.
boundary traces (A∇u)∂Ω and w ∂Ω exist at σ-a.e. point on ∂nta Ω and are
independent of κ . Moreover, with the dependence on the parameter κ dropped,
the following Green type formula (involving absolutely convergent integrals) holds
∫ ∫ ∫
     A n.t. 
L Au, w dL +
n
A∇u, ∇w dL = n
∂ν u, w ∂Ω dσ, (1.7.121)
Ω Ω ∂∗ Ω
1.7 Green-Type Formulas for Second-Order Systems 211

when either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when Ω

is unbounded and ∂Ω is bounded, formula (1.7.121) continues to be true under the
additional assumption that there exists λ ∈ (1, ∞) such that
∫
| A∇u||w| dL n = o(R) as R → ∞. (1.7.122)
[B(0,λ R)\B(0,R)]∩Ω

In particular, corresponding to the case when the last line in (1.7.120) is strength-
ened to

L Au = 0 in the sense of distributions in Ω, (1.7.123)

formula (1.7.121) becomes

∫ ∫
   n.t. 
A∇u, ∇w dL =
n
∂νAu, w ∂Ω dσ. (1.7.124)
Ω ∂∗ Ω

Finally, similar results are valid if (X, Y) is a dual system on (∂Ω, σ) which is
not necessarily regular, with the understanding that, in such a case, all nontan-
gential boundary traces are taken with respect to an aperture parameter which is
< min{κ, κ }.

At least formally, subtracting from (1.7.121) its own version written with the roles
of u and w reversed and with A in place of A would produce (1.7.78). However,
such a proof of Green’s identity recorded in (1.7.78) is far from economical since
this would require imposing conditions on u, w (in the spirit of (1.7.119)-(1.7.120))
that ensure the validity of both (1.7.121) as well as its version with u and w switched
and A replaced by A . In the proof of Theorem 1.7.12 given below, we avoid this
issue by choosing to derive (1.7.121) directly, by applying our version of Divergence
Theorem to a suitably chosen vector field.
Proof of Theorem 1.7.12 In concert with [112, Corollary 8.9.9], the present as-
sumptions imply that there exists some p ∈ (0, ∞] with the property for any given
aperture parameter κ > 0 we have
p
Nκ (A∇u), Nκ w ∈ Lloc (∂Ω, σ), (1.7.125)

and the nontangential boundary traces

κ −n.t. κ −n.t.
(A∇u)∂Ω and w ∂Ω , both exist at σ-a.e. point on ∂nta Ω,
(1.7.126)
and agree with the nontangential boundary traces in (1.7.120).

Observe that (1.7.119), the first line in (1.7.120) (or (1.7.125)), and [112,
Lemma 8.3.1], imply
∞ M ·n ∞ M
A∇u ∈ Lloc (Ω, L n ) and w ∈ Lloc (Ω, L n ) . (1.7.127)
212 1 Integral Representations and Integral Identities

With the summation convention over repeated indices in effect throughout the proof,
consider now the vector field
   
αβ
F := (A∇u)α,r wα ∞ n
= (ar s ∂s uβ )wα ∈ Lloc (Ω, L n ) (1.7.128)
1≤r ≤n 1≤r ≤n

where the membership is a consequence (1.7.127). In particular,

F ∈ Lloc
n
1
(Ω, L n ) . (1.7.129)

To compute the divergence of F,  in the sense of distributions in Ω, fix an arbitrary

∞
scalar-valued function ϕ ∈ 𝒞c (Ω) and write
∫ ∫
  αβ
D (Ω) div 
F, ϕ D(Ω) = − F ∂
r r ϕ dL n
= − (ar s ∂s uβ )wα ∂r ϕ dL n . (1.7.130)
Ω Ω
 
Next, for each sufficiently small ε > 0 let Ωε := x ∈ Ω : dist(x, ∂Ω) > ε .
Using a Friedrichs mollifier, it is possible to construct a sequence of functions
M
w ε = (wαε )1≤α ≤M ∈ 𝒞∞ (Ωε ) such that

for each compact set K ⊂ Ω we have

M
w ε −−−−→
+
w in L 1 (K, L n ) and
ε→0
ε n×M
∇w −−−−→ ∇w in L 1 (K, L n ) , (1.7.131)
ε→0+
and there exists some number εK > 0
so that sup0<ε<εK #w ε #[L ∞ (K, L n )] M < ∞.

By eventually passing to a subsequence there is no loss of generality in assuming

that we also have

w ε −−−−→
+
w at L n -a.e. point in Ω. (1.7.132)
ε→0

On account of (1.7.130), (1.7.127), (1.7.131)-(1.7.132), the last two lines in (1.7.120)

and, finally, Lebesgue’s Dominated Convergence Theorem, we may then compute
1.7 Green-Type Formulas for Second-Order Systems 213
 
D (Ω)
 ϕ
divF, D(Ω)
∫
αβ
= − lim+ (ar s ∂s uβ )wαε ∂r ϕ dL n
ε→0 Ω
∫ ∫
αβ αβ
= − lim+ (ar s ∂s uβ )∂r (wαε ϕ) dL n + lim+ (ar s ∂s uβ )(∂r wαε )ϕ dL n
ε→0 Ω ε→0 Ω
  ∫
αβ αβ
= − lim+ D (Ω) ar s ∂s uβ, ∂r (wαε ϕ) D(Ω) + (ar s ∂s uβ )(∂r wα )ϕ dL n
ε→0 Ω
  ∫
αβ  
= lim+ D (Ω) ar s ∂r ∂s uβ, wαε ϕ D(Ω) + A∇u, ∇w ϕ dL n
ε→0 Ω
  ∫
 
= lim+ D (Ω) (L Au)α, wαε ϕ D(Ω) + A∇u, ∇w ϕ dL n
ε→0 Ω
∫ ∫
 
= lim+ (L Au)α wαε ϕ dL n + A∇u, ∇w ϕ dL n
ε→0 Ω Ω
∫ ∫
 
= (L Au)α wα ϕ dL n + A∇u, ∇w ϕ dL n
Ω Ω
∫ ∫
   
= L Au, w ϕ dL n + A∇u, ∇w ϕ dL n . (1.7.133)
Ω Ω

Having established this, we conclude that

   
divF = L Au, w + A∇u, ∇w in D (Ω), (1.7.134)

so the last line in (1.7.120) implies

divF ∈ L 1 (Ω, L n ). (1.7.135)

 
In addition, if we choose an aperture parameter κ∗ ∈ 0, min{κ, κ } , from (1.7.128)
and [112, (8.2.25)] we see that there exists a constant C A ∈ (0, ∞) such that

Nκ∗ F ≤ C A Nκ (A∇u)Nκ w on ∂Ω. (1.7.136)

From (1.7.136), [112, (8.2.26)], the nontangential maximal function memberships

in the first line of (1.7.120), and (1.7.71) we conclude that

Nκ∗ F ∈ L 1 (∂Ω, σ). (1.7.137)

In addition, (1.7.126) and (1.7.128) imply that the nontangential boundary trace
κ −n.t.

F exists at σ-a.e. point on ∂nta Ω. (1.7.138)
∂Ω
214 1 Integral Representations and Integral Identities

Specifically, with the dependence on κ dropped, we have

n.t.   n.t.   n.t.  

F = (ar s ∂s uβ )∂Ω wα ∂Ω
αβ
at σ-a.e. point on ∂nta Ω. (1.7.139)
∂Ω 1≤r ≤n

In concert with (1.7.9) and [112, Proposition 8.8.6] this implies

 n.t.   n.t. 
ν · F ∂Ω = ∂νAu, w ∂Ω at σ-a.e. point on ∂∗ Ω. (1.7.140)

Let us also observe that (1.7.122) ensures the validity of the integral growth condition

[112, (1.2.3)] for the present vector field F.
Having established the aforementioned properties of F,  [112, Theorem 1.2.1]
applies and the Divergence Formula [112, (1.2.2)] currently yields (1.7.121) thanks
to (1.7.134) and (1.7.140). 
It turns out that there yet another integral identity for second-order systems relating
the bulk with the boundary of a region.

Theorem 1.7.13 Let Ω ⊆ Rn be an open set with a lower Ahlfors regular boundary
and with the property that σ := H n−1 ∂Ω is a doubling measure. Hence, Ω is a set
of locally finite perimeter, and its geometric measure theoretic outward unit normal
ν = (ν1, . . . , νn ) is defined σ-a.e. on ∂∗ Ω.
Given some arbitrary M, N ∈ N, suppose A is a complex coefficient tensor of type
(n × n, M × N) and denote by L A the homogeneous constant (complex) coefficient
second-order M × N system in Rn associated with A as in (1.7.7). Also, fix two
aperture parameters κ, κ > 0, assume (X, Y) is a regular dual systems on (∂Ω, σ)
(in the sense of Definition 1.7.9), and select some  ∈ {1, . . . , n}. Finally, consider
two complex vector-valued functions
2,1 N 2,1 M
u = (uα )1≤α ≤ N ∈ Wloc (Ω) , w = (wβ )1≤β ≤M ∈ Wloc (Ω) , (1.7.141)

satisfying (with all derivatives taken in the sense of distributions in Ω):

Nκ (∇u) belongs to X, Nκ (∇w) belongs to Y,

κ−n.t. κ −n.t.
(∇u)∂Ω and (∇w)∂Ω exist σ-a.e. on ∂nta Ω,
(1.7.142)
L Au ∈ 1 (Ω, L n ) M , L 1 (Ω, L n ) N ,
Lloc Aw ∈ Lloc
   
L Au, ∂ w ∈ L 1 (Ω, L n ), ∂ u, L 
A w ∈ L (Ω, L ).
1 n

Then for any other specified aperture parameter κ ∈ (0, ∞) the nontangential
κ −n.t. κ −n.t.
boundary traces (∇u)∂Ω , (∇w)∂Ω exist at σ-a.e. point on ∂nta Ω and are in-
dependent of κ . Moreover, with the dependence on the parameter κ dropped, the
following identity (involving absolutely convergent integrals) holds
1.7 Green-Type Formulas for Second-Order Systems 215
∫ ∫
   
L Au, ∂ w dL n + ∂ u, L 
A w dL
n
Ω Ω
∫ ∫
 n.t.   n.t.  
= ∂νAu, (∂ w)∂Ω dσ + (∂ u)∂Ω, ∂νA w dσ
∂∗ Ω ∂∗ Ω
∫
n.t. n.t.
ν ar s (∂r wβ )∂Ω (∂s uα )∂Ω dσ,
βα
− (1.7.143)
∂∗ Ω

when either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when Ω

is unbounded and ∂Ω is bounded, formula (1.7.143) continues to be true under the
additional assumption that there exists λ ∈ (1, ∞) such that
∫
|∇u||∇w| dL n = o(R) as R → ∞. (1.7.144)
[B(0,λ R)\B(0,R)]∩Ω

As a corollary, if the last two lines in (1.7.142) are replaced by

L Au = 0 and L 
A w = 0 in the sense of distributions in Ω, (1.7.145)

then (1.7.143) becomes

∫ ∫
 A n.t.   n.t.  

∂ν u,(∂ w) ∂Ω dσ + (∂ u)∂Ω, ∂νA w dσ
∂∗ Ω ∂∗ Ω
∫
n.t. n.t.
ν ar s (∂r wβ )∂Ω (∂s uα )∂Ω dσ.
βα
= (1.7.146)
∂∗ Ω

Finally, similar results are valid if (X, Y) is a dual system on (∂Ω, σ) which is
not necessarily regular, with the understanding that, in such a case, all nontan-
gential boundary traces are taken with respect to an aperture parameter which is
< min{κ, κ }.

Proof Based on assumptions and [112, Corollary 8.9.9] we conclude that there exists
p ∈ (0, ∞] with the property that for any aperture parameter κ > 0 we have
p
Nκ (∇u), Nκ (∇w) ∈ Lloc (∂Ω, σ), (1.7.147)

and the nontangential boundary traces

κ −n.t. κ −n.t.
(∇u)∂Ω , (∇w)∂Ω exist σ-a.e. on ∂nta Ω and agree with
(1.7.148)
the respective nontangential boundary traces in (1.7.142).

Also, (1.7.141), the first line in (1.7.142) (or (1.7.147)), and [112, Lemma 8.3.1]
imply
216 1 Integral Representations and Integral Identities

2,1 N N
u ∈ Wloc (Ω) ∩ Liploc (Ω) ,
(1.7.149)
2,1 M M
w∈ Wloc (Ω) ∩ Liploc (Ω) .

To proceed, consider the vector field which, with the summation convention over
repeated indices in effect, is given by
βα βα αβ
F := ar s (∂r wβ )(∂s uα )e − ar s (∂r wβ )(∂ uα )es − ar s (∂ wα )(∂s uβ )er . (1.7.150)

Hence, (1.7.150) and (1.7.149) give

F ∈ Lloc
∞ n n
(Ω, L n ) ⊂ Lloc
1
(Ω, L n ) . (1.7.151)

Next, based on (1.7.149), (1.7.150), and Leibniz’s product formula for weak deriva-
tives from [112, Proposition 4.3.1], we may compute div F in the sense of distributions
in Ω as
βα βα
divF = ar s (∂ ∂r wβ )(∂s uα ) + ar s (∂r wβ )(∂ ∂s uα )
βα βα
− ar s (∂s ∂r wβ )(∂ uα ) − ar s (∂r wβ )(∂s ∂ uα )
αβ αβ
− ar s (∂r ∂ wα )(∂s uβ ) − ar s (∂ wα )(∂r ∂s uβ ). (1.7.152)

After observing that, in the right-hand side above, the first term cancels the fifth and
the second term cancels the fourth, we conclude that
   
divF = − ∂ u, L A w − L Au, ∂ w in D (Ω). (1.7.153)

Together with the last line in (1.7.142) this gives

divF ∈ L 1 (Ω, L n ). (1.7.154)

 
Also, if we choose an aperture parameter κ∗ ∈ 0, min{κ, κ } , from (1.7.150) and
[112, (8.2.25)] we see that

Nκ∗ F ≤ C · Nκ (∇u)Nκ (∇w) on ∂Ω, (1.7.155)

for some constant C ∈ (0, ∞) which depends only on the coefficient tensor A.
In turn, (1.7.155), (1.7.147), [112, (8.2.26)], the nontangential maximal function
memberships in the first line of (1.7.142), and (1.7.71) ensure that

Nκ∗ F ∈ L 1 (∂Ω, σ). (1.7.156)

Moving on, observe that (1.7.150) and (1.7.90) imply that the nontangential
boundary trace
κ −n.t.

F exists at σ-a.e. point on ∂nta Ω. (1.7.157)
∂Ω
In fact, with the dependence on κ dropped, we have
1.7 Green-Type Formulas for Second-Order Systems 217
n.t. n.t.   n.t.  n.t.   n.t. 
 βα  βα 
F = ar s (∂r wβ )∂Ω (∂s uα )∂Ω e − ar s (∂r wβ )∂Ω (∂ uα )∂Ω es
∂Ω

αβ  n.t.   n.t. 
− ar s (∂ wα )∂Ω (∂s uβ )∂Ω er at σ-a.e. point on ∂nta Ω. (1.7.158)

From (1.7.158), (1.7.9), and [112, Proposition 8.8.6], we therefore see that at σ-a.e.
point on ∂∗ Ω we have
 n.t.  βα  n.t.   n.t. 
ν · F ∂Ω = ν ar s (∂r wβ )∂Ω (∂s uα )∂Ω
 n.t.    n.t. 
− (∂ u)∂Ω, ∂νA w − ∂νAu, (∂ w)∂Ω . (1.7.159)

Finally, (1.7.144) guarantees that the vector field F satisfies the growth condition
 we may invoke [112, Theorem 1.2.1]
[112, (1.2.3)]. Granted these properties of F,
and the Divergence Formula [112, (1.2.2)] currently produces (1.7.143), thanks to
(1.7.153) and (1.7.159). 

A special case of Theorem 1.7.12 which is particularly useful in applications is

singled out in our next corollary. In particular, this is the key ingredient in the proof
of the uniqueness result for the mixed boundary value problem stated in [115, §1.7].

Corollary 1.7.14 Assume Ω ⊆ Rn , where n ≥ 2, is an open set with a lower Ahlfors

regular boundary and with the property that σ := H n−1 ∂Ω is a doubling measure.
In particular, Ω is a set of locally finite perimeter, and its geometric measure theoretic
outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Having fixed some M ∈ N, let A be a
complex coefficient tensor of type (n × n, M × M) and denote by L A the homogeneous
constant (complex) coefficient second-order M × M system in Rn associated with A
as in (1.7.7). In this context, consider a complex vector-valued function
1,1 M
u ∈ Wloc (Ω) (1.7.160)

with the property that, for some κ > 0 and p, q, r ∈ [1, ∞], one has:

Nκ (∇u) ∈ L p (∂Ω, σ), Nκ u ∈ L q (∂Ω, σ),

κ−n.t. κ−n.t.
u∂Ω and (∇u)∂Ω exist σ-a.e. on ∂nta Ω, (1.7.161)
M
and L Au belongs to the space L r (Ω, L n ) .

Then for any other specified aperture parameter κ ∈ (0, ∞) the nontangential
κ −n.t. κ −n.t.
boundary traces u∂Ω and (∇u)∂Ω exist at σ-a.e. point on ∂nta Ω and are in-
dependent of κ . Moreover, with the dependence on the parameter κ dropped, the
formula
∫ ∫ ∫
     A n.t. 
L Au, u dL n + A∇u, ∇u dL n = ∂ν u, u∂Ω dσ (1.7.162)
Ω Ω ∂∗ Ω
218 1 Integral Representations and Integral Identities

holds (with all integrals involved absolutely convergent) in any of the following
scenarios:

(i) ∂Ω is unbounded, p = 2(n − 1)/n, q = 2(n − 1)/(n − 2), and r = 2n/(n + 2);
(ii) Ω is bounded, p ≥ 2(n − 1)/n, q ≥ p/(p − 1), and r ≥ np/(n + p − 1);
(iii) Ω is an exterior domain, p ≥ 2(n − 1)/n, q ≥ p/(p − 1), and there exist two
real numbers a, b such that

(∇u)(x) = O(|x| −a ) and u(x) = O(|x| −b ) as |x| → ∞, (1.7.163)

 
a > n/2, b > max n − 1 − a, (n − 1)/q , (1.7.164)

and 
n/(n − b) if b < n,
nq/(nq − n + 1) ≤ r < (1.7.165)
∞ if b ≥ n.
Moreover, if actually L Au = 0 in the sense of distributions in Ω, then (1.7.165)
may be ignored and, in place of (1.7.164), one may simply ask that

a > n/2 and b > n − 1 − a. (1.7.166)

Proof Consider first the case when ∂Ω is unbounded, and p, q, r are as in item (i).
Then [112, Proposition 8.6.3] and [112, Lemma 8.3.2] permit us to conclude that
 
M M ∗
u ∈ L 2n/(n−2) (Ω, L n ) = L 2n/(n+2) (Ω, L n ) ,
(1.7.167)
n×M
and ∇u ∈ L 2 (Ω, L n ) .

Consequently,
   
both A∇u, ∇u and L Au, u belong to L 1 (Ω, L n ). (1.7.168)

Given that p = 2(n − 1)/n and q = 2(n − 1)/(n − 2) are Hölder conjugate exponents,
all desired conclusions, including formula (1.7.162), are in this case consequences
of Theorem 1.7.12 (presently used with w := u and p := 2(n − 1)/n).
Let us now consider the case when Ω is bounded, and p, q, r are as in item (ii).
Then [112, Proposition 8.6.3] and [112, Lemma 8.3.2] permit us to conclude that
M M
u ∈ L nq/(n−1) (Ω, L n ) ⊆ L np/[(n−1)(p−1)] (Ω, L n )
(1.7.169)
n×M n×M
and ∇u ∈ L np/(n−1) (Ω, L n ) ⊆ L 2 (Ω, L n ) .

Also,
 ∗
 np M np
M M
L Au ∈ L (Ω, L )
r n
⊆ L n+p−1 (Ω, L ) n
= L (n−1)(p−1) (Ω, L )
n
.

(1.7.170)
1.7 Green-Type Formulas for Second-Order Systems 219

It is then apparent from (1.7.169)-(1.7.170) that (1.7.168) holds in this case as well.
Given that, with p := p/(p − 1) denoting the Hölder conjugate exponent of p, we
also have

Nκ (∇u) ∈ L p (∂Ω, σ) and Nκ u ∈ L q (∂Ω, σ) ⊆ L p (∂Ω, σ), (1.7.171)

since σ(∂Ω) < +∞, all desired conclusions are once again seen from Theorem 1.7.12
(employed with w := u, and the current exponent p).
To deal with case (iii), suppose Ω is an exterior domain, p ≥ 2(n − 1)/n,
q ≥ p/(p − 1), and there exist a, b ∈ R such that (1.7.163)-(1.7.165) hold. In
particular, a, b > 0 and, since a + b > n − 1, we also have

|u(x)||(∇u)(x)| = o(|x| 1−n ) as |x| → ∞. (1.7.172)

In such a scenario, starting with Nκ (∇u) ∈ L p (∂Ω, σ) [112, Proposition 8.6.3] now
only guarantees that
np/(n−1) n×M np/(n−1) n×M
∇u ∈ Lbdd (Ω, L n ) ⊆ Lbdd (Ω, L n ) . (1.7.173)

However, this may be augmented with the pointwise decay property of ∇u at infinity.
Since a > n/2, the latter ensures that

|∇u| 2 is Lebesgue integrable in a neighborhood of infinity. (1.7.174)

Together, (1.7.173)-(1.7.174) ultimately prove that we still have

n×M
∇u ∈ L 2 (Ω, L n ) . (1.7.175)

In particular,  
A∇u, ∇u belongs to L 1 (Ω, L n ). (1.7.176)
Likewise, starting with Nκ u ∈ L q (∂Ω, σ), [112, Proposition 8.6.3] now gives
nq/(n−1) M ns/(n−1) M
u ∈ Lbdd (Ω, L n ) ⊆ Lbdd (Ω, L n )
(1.7.177)
for each s ∈ (0, q],

while the pointwise decay property of u at infinity permits us to deduce that

the function |u| ns/(n−1) is Lebesgue integrable in a

(1.7.178)
neighborhood of infinity whenever s > (n − 1)/b.
From (1.7.177)-(1.7.178) we then conclude that
M
u ∈ L ns/(s−1) (Ω, L n ) whenever (n − 1)/b < s ≤ q. (1.7.179)
M
Bearing in mind that L Au belongs to L r (Ω, L n ) with r as in (1.7.165), we finally
see from (1.7.179) that
220 1 Integral Representations and Integral Identities
 
L Au, u belongs to L 1 (Ω, L n ). (1.7.180)

With (1.7.176), (1.7.180), and (1.7.172) in hand, and keeping in mind that (1.7.171)
continues to be valid, all desired conclusions are presently implied by Theorem 1.7.12
(used with w := u and the original exponent p).
Finally, in the case when L Au = 0 the same type of analysis as above applies, this
time only making use of a > n/2 and a + b > n − 1. 

While by now we have established a number of general Green type formulas,

when faced with the task of proving a specific identity of this nature, depending
on circumstances, one may still find it useful to go back to our original Divergence
Theorems in order to ensure that the information available is used in an optimal
manner. A concrete scenario is as follows.

Proposition 1.7.15 Fix n ∈ N with n ≥ 2 and assume u is a harmonic function in

R+n which, for some κ > 0, satisfies

Nκ u, Nκ (∇u) ∈ L 2 (Rn−1, L n−1 ), (1.7.181)

where Nκ denotes the nontangential maximal operator (with aperture parameter κ)

relative to the open set R+n . Then
κ−n.t. κ−n.t.
u∂Rn , (∇u)∂Rn exist L n−1 -a.e. on Rn−1 ≡ ∂R+n (1.7.182)
+ +

and for each upper-graph Lipschitz domain R ⊆ R+n one has the Green formula
∫ ∫  ∂|u| 2 ∂t 
t|(∇u)(x , t)| 2 dx dt = t − |u| 2 dσ, (1.7.183)
R ∂R ∂ν ∂ν

where ν = (ν1, . . . , νn ) is the outward unit normal to R, and σ := H n−1 ∂R.

Above, at H n−1 -a.e. point (x , t) ∈ ∂R ∩ ∂R+n , the expression in the curly brackets is
interpreted as
 κ−n.t.   κ−n.t.    κ−n.t.  2
 
2t u∂R (x , t)ν(x , t) · (∇u)∂R (x , t) −  u∂R (x , t) νn (x , t). (1.7.184)

This result justifies the claim made by C. Fefferman and E. Stein at the bottom of
page 162 in [45] where they take R to be the sawtooth region in the upper half-space
associated with a given closed set E ⊆ Rn−1 ≡ ∂R+n and some κ > 0. Specifically,

R := Γκ (x), (1.7.185)
x ∈E

where, for each x ∈ ∂R+n , we have denoted by Γκ (x) the nontangential approach
region with aperture parameter κ and vertex at x, relative to the domain R+n . The
uniform cone property satisfied by R ensures that this set is the upper-graph of a
non-negative Lipschitz function (see [62]).
1.7 Green-Type Formulas for Second-Order Systems 221

Harmonic functions in the upper half-space satisfying (1.7.181) arise naturally.

To offer a concrete example, recall that the Poisson kernel for the Laplacian in R+n is
given by
2 1
PΔ (x ) = n , ∀x ∈ Rn−1 .
ωn−1 1 + |x | 2  2
(1.7.186)

If for each x ∈ Rn−1 and t > 0 we set PtΔ (x ) := t 1−n PΔ (x /t), then given any
f ∈ W 1,2 (Rn−1 ) the function u : R+n → R defined as u(x , t) := (PtΔ ∗ f )(x ) for each
(x , t) ∈ R+n satisfies (1.7.181) (cf. [94]).
To give another example (and this is the case considered in [45]), start with a
harmonic function u in R+n with the property that Nκ u ∈ L p (Rn−1, L n−1 ) for some
p ∈ (0, 2] and κ > 0 and, for each fixed ε > 0, define uε (x , t) := u(x , t + ε) for
every (x , t) ∈ R+. Picknan arbitrary point (x , t) ∈ R+ . Using interior estimate in the
n n

ball B (x , t), t/2 ⊂ R+ , for each k ∈ N0 we may estimate

⨏
 k 
(∇ u)(x , t) ≤ C |u(y , s)| dy ds
t k B((x ,t),t/2)
∫ 3t/2 ∫
C
≤ k+n |u(y , s)| dy ds
t t/2 B n−1 (x ,t/2)
⨏
C
≤ k (Nκ u)(y ) dy
t B n−1 (x ,t/2)

C  
≤ k
M Nκ u (x ), (1.7.187)
t
where Bn−1 (a , r) denotes the (n − 1)-dimensional ball centered at a ∈ Rn−1 of
radius r, and M is the Hardy-Littlewood maximal operator in Rn−1 . Also, using
reverse Hölder estimates for harmonic functions yields
!⨏ " 1/p
 
u(x , t) ≤ C |u(y , s)| p dy ds
B((x ,t),t/2)
!∫ 3t/2 ∫ " 1/p
C
≤ |u(y , s)| p dy ds
t n/p t/2 B n−1 (x , t/2)
!∫ " 1/p
C
≤ (Nκ u)(y ) dy
p
t (n−1)/p B n−1 (x , t/2)

≤ C t −(n−1)/p #Nκ u# L p (Rn−1, L n−1 ) . (1.7.188)

When used with t + ε in place of t, estimate (1.7.188) gives

|uε (x , t)| ≤ Cε −(n−1)/p #Nκ u# L p (Rn−1, L n−1 ), ∀(x , t) ∈ R+n, (1.7.189)

hence uε ∈ L ∞ (R+n, L n ) which further implies

222 1 Integral Representations and Integral Identities
2
Nκ uε ∈ L q (Rn−1, L n−1 ) for each ε > 0. (1.7.190)
p ≤q ≤∞

Writing (1.7.187) for uε/2 in place of u while simultaneously replacing t by t + ε/2

yields
 k   
(∇ uε )(x , t) ≤ C M Nκ uε/2 (x ), ∀(x , t) ∈ R+n . (1.7.191)
ε k

Bearing in mind that p ≤ 2, in view of (1.7.190) and the boundedness of M, from

this we obtain
 
sup (∇k uε )(·, t) L 2 (Rn−1, L n−1 ) < +∞, for each k ∈ N0 . (1.7.192)
t>0

Finally, it is well known (see, e.g., [166, Corollary 1.2.1, p. 200]) that (1.7.192)
implies the conditions in (1.7.181) for the function uε . The above argument may be
summarized as follows:
given a harmonic function u in R+n with Nκ u ∈ L p (Rn−1, L n−1 ) for
some exponent p ∈ (0, 2] and apperture parameter κ > 0, if for ε > 0
arbitrary we define uε (x , t) := u(x , t + ε) for each (x , t) ∈ R+n , then (1.7.193)
uε is a harmonic function in R+n which satisfies the conditions listed in
(1.7.181).
After this detour, we are ready to present the proof of Proposition 1.7.15.
Proof of Proposition 1.7.15 That any harmonic function u in R+n which satisfies
(1.7.181) also satisfies (1.7.182) is a classical result (see, e.g., [166, Chapter VII]).
Next, by assumption, the set ∂R is the graph of some non-negative Lipschitz function
φ : Rn−1 → R and we find it useful to abbreviate M := #∇φ# L ∞ (Rn−1, L n−1 ) . In
 √ 
particular, if we fix 0 < κ < − M + 1 + M 2 /M then

x + Γκ (0) ⊆ R for each x ∈ ∂R. (1.7.194)

Let us introduce the vector field

 , t) := tu(x , t)(∇u)(x , t) − 1 |u(x , t)| 2 en,
G(x ∀(x , t) ∈ R+n . (1.7.195)
2

Hence,
G ∈ 𝒞∞ (R+n ) .
n
(1.7.196)
Simple geometry shows that we may choose κ > κ with the property that
B((y , t), t/2) ⊆ Γκ (x) for each x ∈ ∂R+n and each (y , t) ∈ Γκ (x). Based on this ob-
servation and interior estimates, for each x = (x , 0) ∈ ∂R+n and each (y , t) ∈ Γκ (x)
we may write
⨏
 
t|(∇u)(y , t)| ≤ C |u| dL n ≤ C Nκ u (x ), (1.7.197)
B((y , t), t/2)
1.7 Green-Type Formulas for Second-Order Systems 223

hence    
sup t|(∇u)(y , t)| ≤ C Nκ u (x ). (1.7.198)
(y ,t)∈Γκ (x)

From this we then conclude that there exists some dimensional constant C ∈ (0, ∞)
with the property that
 2
Nκ G ≤ C Nκ u everywhere on Rn−1 ≡ ∂R+n . (1.7.199)

To proceed, consider the vector field


F := G  R ∈ 𝒞∞ (R)
n
(1.7.200)

and (keeping (1.7.194) in mind) define

 
N̊κ F (x) := sup | F(y)|
 for each x ∈ ∂R. (1.7.201)
y ∈x+Γκ (0)

Then (1.7.199) implies that for each x = (x , t) ∈ ∂R we have

     2
N̊κ F (x) ≤ sup 
| G(y)| = Nκ G (x ) ≤ C Nκ u (x ). (1.7.202)
y ∈Γκ (x ,0)

Bearing in mind that ∂R is the graph of the Lipschitz function φ : Rn−1 → R, from
(1.7.202) we conclude that
  5  
N̊κ F 1 ≤ C 1 + M 2 N u2
κ < +∞. (1.7.203)
L (∂R,σ) L 2 (R n−1, L n−1 )

In particular, this forces N̊κ F ∈ L 1 (∂R, σ). Having established this, [122, Proposi-
tion 2.2, p. 24] applies and gives that

NR,κ F ∈ L 1 (∂R, σ) (1.7.204)

where NR,κ denotes the nontangential maximal operator (with aperture parameter κ)
relative to the domain R. Moreover, if for each x ∈ ∂R we let Γ R,κ (x) stand for
the nontangential approach region (with aperture parameter κ) relative to the domain
R, the fact that

Γ R,κ (x) ⊆ Γκ (x) for each point x ∈ ∂R ∩ ∂R+n, (1.7.205)

together with (1.7.200), (1.7.195)-(1.7.196), and (1.7.182), implies that

κ−n.t.
F∂R exists σ-a.e. on ∂R. (1.7.206)

In addition, a direct computation which takes into account that the function u is
harmonic (hence, Δ(|u| 2 ) = 2|∇u| 2 ) gives
 , t) = t|(∇u)(x , t)| 2 for each (x , t) ∈ R.
(divF)(x (1.7.207)
224 1 Integral Representations and Integral Identities

On the other hand, from [166, Corollary 1.2.1, p. 200] and standard L 2 -square
function estimates for derivatives of the Poisson kernel (see, e.g., [166, Theorem 1,
p. 82] when p = 2) we see that
∫
t|(∇u)(x , t)| 2 dx dt < +∞. (1.7.208)
R+n

Collectively, (1.7.207)-(1.7.208) prove that

divF ∈ L 1 (R, L n ). (1.7.209)

Granted (1.7.200), (1.7.204), (1.7.206), and (1.7.209), [112, Theorem 1.2.1] applies
and the Divergence Formula recorded in [112, (1.2.2)] yields precisely the identity
in (1.7.183) (interpreting the integrand in the right-hand side as in (1.7.184) for any
given parameter κ > 0). 

In the last part of this section we revisit the Green type formulas for second-order
systems from Theorem 1.7.10 and Theorem 1.7.12 and adopt a more general point
of view in which the system in question is written as a composition of two first-order
differential operators D and D. ' This adds more flexibility in applications and also
leads to an invariant formulation of Green formulas which carries over to the setting
of manifolds. We take up the latter task in Theorem 1.7.19 and Theorem 1.7.20. For
now, we begin with the following generalization of Theorem 1.7.10.

Theorem 1.7.16 Suppose Ω ⊆ Rn is an open set with a lower Ahlfors regular

boundary and with the property that σ := H n−1 ∂Ω is a doubling measure. In
particular, Ω is a set of locally finite perimeter, and its geometric measure theoretic
outward unit normal ν is defined σ-a.e. on ∂∗ Ω.
Having fixed M, N, N ∈ N, consider a first-order M × N system D with constant
complex coefficients in Rn , along with a first-order N × M system D ' with constant
complex coefficients in Rn . Then define the constant (complex) coefficient second-
order N × N system in Rn given by
'
L := DD. (1.7.210)

Finally, fix some κ, κ > 0, assume (X0, Y0 ) and (X1, Y1 ) are two regular dual
systems on (∂Ω, σ) (in the sense of Definition 1.7.9), and consider two complex
vector-valued functions
N N
u ∈ Lloc
1
(Ω, L n ) , w ∈ Lloc
1
(Ω, L n ) , (1.7.211)

satisfying (with all derivatives taken in the sense of distributions in Ω):

1.7 Green-Type Formulas for Second-Order Systems 225

Du ∈ Lloc
1 (Ω, L n ) M
, ' w ∈ L 1 (Ω, L n )
D
M
,
loc

Nκ u ∈ X0, Nκ (Du) ∈ X1, Nκ w ∈ Y1, Nκ ' w)

(D ∈ Y0,
κ−n.t. κ−n.t. κ −n.t. κ −n.t.
u∂Ω , (Du)∂Ω , w ∂Ω , ( D' w)
∂Ω
exist σ-a.e. on ∂nta Ω, (1.7.212)

N N
1 (Ω, L n )
Lu ∈ Lloc , L  w ∈ Lloc
1 (Ω, L n ) ,
   
Lu, w − u, L  w belongs to L 1 (Ω, L n ).

Then for any other specified aperture parameter κ ∈ (0, ∞) the nontangential
κ −n.t. κ −n.t. κ −n.t. κ −n.t.
boundary traces u∂Ω , (Du)∂Ω , w ∂Ω , ( D ' w)
∂Ω
exist at σ-a.e. point on
∂nta Ω and are independent of κ . Moreover, with the dependence on the parameter
κ dropped, the following Green type formula (involving absolutely convergent
integrals) holds
∫ 
   
Lu, w − u, L  w dL n
Ω
∫   n.t.  n.t. 
= ' ν) (Du) , w 
(−i) Sym( D; dσ
∂Ω ∂Ω
∂∗ Ω
∫  n.t.   n.t.  
− u∂Ω, (−i) Sym(D ; ν) ( D
' w)
∂Ω
dσ, (1.7.213)
∂∗ Ω

when either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when Ω

is unbounded and ∂Ω is bounded, formula (1.7.213) continues to be true under the
additional assumption that there exists λ ∈ (1, ∞) such that
∫  
' w| + |w||Du| dL n = o(R) as R → ∞.
|u|| D (1.7.214)
[B(0,λ R)\B(0,R)]∩Ω

As a consequence, if the last condition in (1.7.212) is replaced by

Lu = 0 and L  w = 0 in the sense of distributions in Ω, (1.7.215)

then (1.7.213) reduces to

∫ 
 n.t.  n.t. 
' ν) (Du) , w 
(−i) Sym( D; dσ
∂Ω ∂Ω
∂∗ Ω
∫  n.t.   n.t.  
= u∂Ω, (−i) Sym(D ; ν) ( D
' w)
∂Ω
dσ. (1.7.216)
∂∗ Ω

Finally, similar results are valid if (X0, Y0 ) and (X1, Y1 ) are two dual systems
on (∂Ω, σ) which are not necessarily regular, with the understanding that, in such
a case, all nontangential boundary traces are taken with respect to an aperture
parameter which is < min{κ, κ }.
226 1 Integral Representations and Integral Identities

Proof From assumptions and [112, Corollary 8.9.9] it follows that there exists
p ∈ (0, ∞] with the property that for any given aperture parameter κ > 0 we have
' w) ∈ L p (∂Ω, σ),
Nκ u, Nκ (Du), Nκ w, Nκ ( D (1.7.217)
loc

and the nontangential boundary traces

κ −n.t. κ −n.t. κ −n.t. κ −n.t.
u∂Ω , (Du)∂Ω , w ∂Ω , ( D ' w)
∂Ω
exist
at σ-a.e. point on ∂nta Ω and agree with the respective (1.7.218)

nontangential boundary traces in (1.7.212).

To proceed, note that (1.7.211), the first two lines in (1.7.212), and [112,
Lemma 8.3.1] imply

∞ (Ω, L n ) N ∞ (Ω, L n ) M
u ∈ Lloc , Du ∈ Lloc ,
(1.7.219)
∞ (Ω, L n )
w ∈ Lloc
N
, ' w ∈ L ∞ (Ω, L n )
D
M
.
loc

Consider now the vector field F : Ω → Cn implicitly defined by the requirement

that
 

ξ · F(x) ' ξ)(Du)(x), w(x)
= (−i) Sym( D;
 
' w)(x)
− u(x), (−i) Sym(D ; ξ)( D

for each ξ ∈ Rn and L n -a.e. x ∈ Ω. (1.7.220)

Then (1.7.219) ensures that

F ∈ Lloc
∞ N n
(Ω, L n ) ⊆ Lloc
1
(Ω, L n ) . (1.7.221)

Let us compute divF in the sense of distributions in Ω. With this goal in mind, fix
an arbitrary real-valued function ϕ ∈ 𝒞∞
c (Ω) and write
∫
 

D (Ω) div F, ϕ D(Ω) = − ∇ϕ · F dL n
Ω
∫  
=− ' ∇ϕ)Du, w dL n
(−i) Sym( D;
Ω
∫  
+ ' w dL n
u, (−i) Sym(D ; ∇ϕ) D
Ω

=: I + I I, (1.7.222)
1.7 Green-Type Formulas for Second-Order Systems 227

 on (1.7.220) with ξ :=∇ϕ. For each sufficiently

where the second equality is based
small ε > 0 consider now Ωε := x ∈ Ω : dist(x, ∂Ω) > ε . Using a Friedrichs mol-
N
lifier, we may construct a sequence of functions w ε = (wαε )1≤α ≤ N ∈ 𝒞∞ (Ωε )
such that
w ε −−−−→
+
w uniformly on compact subsets of Ω,
ε→0
' w ε −−−−→ D
D ' w at L n -a.e. point in Ω and, also,
+
ε→0
(1.7.223)
for each compact set K ⊂ Ω there exists some εK > 0
  ε
with the property that sup  D' w  ∞
[L (K, L n )] M
< ∞.
0<ε<ε K

Then, based on [112, (1.7.17), (1.7.20)], (1.7.219), (1.7.223), the penultimate line in
(1.7.212), and Lebesgue’s Dominated Convergence Theorem, we may compute
∫  
I = − lim+ ' ∇ϕ)Du, w ε dL n
(−i) Sym( D;
ε→0 Ω
∫  
= lim+ ' ; ∇ϕ)w ε dL n
Du, (−i) Sym( D
ε→0 Ω
∫ ∫
   
= lim+ ' (ϕw ε ) dL n − lim
Du, D ' w ε dL n
Du, ϕ D
ε→0 + ε→0
Ω Ω
  ∫
 
= lim+ [D (Ω)] M ' (ϕw ε )
Du, D [D(Ω)] M − ' w ϕ dL n
Du, D
ε→0 Ω
  ∫
 
' , ϕw ε
= lim+ [D (Ω)] N DDu [D(Ω)] N − ' w ϕ dL n
Du, D
ε→0 Ω
∫
   
= lim+ [D (Ω)] N Lu , ϕw ε [D(Ω)] N − ' w ϕ dL n
Du, D
ε→0 Ω
∫ ∫
   
= lim+ Lu, ϕw ε dL n − ' w ϕ dL n
Du, D
ε→0 Ω Ω
∫ ∫
   
= Lu, ϕw dL n − ' w ϕ dL n
Du, D
Ω Ω
∫ ∫
   
= Lu, w ϕ dL n − ' w ϕ dL n .
Du, D (1.7.224)
Ω Ω

In a similar fashion,
∫ ∫
   
II = − u, L  w ϕ dL n + ' w ϕ dL n .
Du, D (1.7.225)
Ω Ω

At this stage, from (1.7.222) and (1.7.224)-(1.7.225) we conclude that

228 1 Integral Representations and Integral Identities
   
divF = Lu, w − u, L  w in D (Ω). (1.7.226)

In particular, the last line in (1.7.212) ensures that

divF ∈ L 1 (Ω, L n ). (1.7.227)

 
Also, if we choose an aperture parameter κ∗ ∈ 0, min{κ, κ } , from (1.7.220) and
[112, (8.2.25)] we obtain
 
Nκ∗ F ≤ C Nκ (Du)Nκ w + Nκ u Nκ ( D ' w) on ∂Ω, (1.7.228)

for some constant C ∈ (0, ∞) which depends only on the coefficient tensor A. In
turn, from (1.7.102), (1.7.217), [112, (8.2.26)], the nontangential maximal function
memberships in the second line of (1.7.212), and (1.7.71) we conclude that

Nκ∗ F ∈ L 1 (∂Ω, σ). (1.7.229)

Going further, observe that (1.7.220) and (1.7.218) imply that the nontangential
boundary trace
κ −n.t.

F exists at σ-a.e. point on ∂nta Ω. (1.7.230)
∂Ω
Moreover, with the dependence on κ dropped, from (1.7.220) and [112, Proposi-
tion 8.8.6] we obtain
 n.t.    n.t.  n.t. 
' ν) (Du) , w 
ν · F ∂Ω = (−i) Sym( D; ∂Ω ∂Ω
 n.t.   n.t.  
' w)
− u∂Ω, (−i) Sym(D ; ν) ( D ∂Ω
(1.7.231)

at σ-a.e. point on ∂∗ Ω. Finally, (1.7.214) guarantees that the vector field F satisfies the
growth condition [112, (1.2.3)]. Granted these properties of F,  [112, Theorem 1.2.1]
applies and the Divergence Formula [112, (1.2.2)] presently yields (1.7.213) on
account of (1.7.226) and (1.7.231). 

It is of interest to specialize Theorem 1.7.16 to the case when the factorization

in (1.7.210) pertains to the zero operator, in which case we obtain an integration by
parts formula on the boundary without demanding that the functions involved satisfy
a certain PDE (as was the case in (1.7.215)-(1.7.216)). More general formulas of
this flavor are established in the context of Riemannian manifolds in [112, §1.12].

Corollary 1.7.17 Let Ω ⊆ Rn be an open set with a lower Ahlfors regular boundary
and such that σ := H n−1 ∂Ω is a doubling measure. Denote by ν the geometric
measure theoretic outward unit normal to Ω Fix M, N, N ∈ N and consider a
first-order M × N system D with constant complex coefficients in Rn , together with
a first-order N × M system D' with constant complex coefficients in Rn , with the
property that
' = 0.
DD (1.7.232)
1.7 Green-Type Formulas for Second-Order Systems 229

Finally, fix some κ, κ > 0, assume (X0, Y0 ) and (X1, Y1 ) are two regular dual
systems on (∂Ω, σ) (in the sense of Definition 1.7.9), and consider two complex
vector-valued functions
N N
u ∈ Lloc
1
(Ω, L n ) , w ∈ Lloc
1
(Ω, L n ) , (1.7.233)

satisfying (with all derivatives taken in the sense of distributions in Ω):

Du ∈ Lloc
1 (Ω, L n ) M
, ' w ∈ L 1 (Ω, L n )
D
M
,
loc
' w) ∈ Y0,
Nκ u ∈ X0, Nκ (Du) ∈ X1, Nκ w ∈ Y1, Nκ ( D (1.7.234)
κ−n.t. κ−n.t. κ −n.t. κ −n.t.
u∂Ω , (Du)∂Ω , w ∂Ω , ( D' w)
∂Ω
exist σ-a.e. on ∂nta Ω.

Then for any other specified aperture parameter κ ∈ (0, ∞) the nontangential
κ −n.t. κ −n.t. κ −n.t. κ −n.t.
boundary traces u∂Ω , (Du)∂Ω , w ∂Ω , ( D ' w)
∂Ω
exist at σ-a.e. point on
∂nta Ω and are independent of κ . In addition, with the dependence on the parameter
κ dropped, the following formula holds
∫ 
 n.t.  n.t. 
' ν) (Du) , w 
(−i) Sym( D; dσ
∂Ω ∂Ω
∂∗ Ω
∫  n.t.   n.t.  
= u∂Ω, (−i) Sym(D ; ν) ( D
' w)
∂Ω
dσ (1.7.235)
∂∗ Ω

when either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when Ω

is unbounded and ∂Ω is bounded, formula (1.7.213) continues to be true under the
additional assumption that there exists λ ∈ (1, ∞) such that
∫  
' w| + |w||Du| dL n = o(R) as R → ∞.
|u|| D (1.7.236)
[B(0,λ R)\B(0,R)]∩Ω

In addition, a similar result is valid if (X0, Y0 ) and (X1, Y1 ) are two dual systems
on (∂Ω, σ) which are not necessarily regular, with the understanding that, in such
a case, all nontangential boundary traces are taken with respect to an aperture
parameter which is < min{κ, κ }.

We wish to remark that the cancelation condition demanded in (1.7.232) naturally

' = curl and D = ∇ or, more
occurs in applications. For instance, this is the case for D
'
generally, taking D and D to be the exterior derivative operator d. Another basic
example is obtained by fixing two indices j, k ∈ {1, . . . , n} and then taking

Du := (∂j u)e j + (∂k u)ek , 'w := ∂j wk − ∂k w j ,

D (1.7.237)

for all scalar functions u and vector-valued functions w = (wi )1≤i ≤n . Formula
(1.7.235) corresponding to such a scenario is then related to the result proved in
[113, Proposition 11.3.12].
230 1 Integral Representations and Integral Identities

Proof of Corollary 1.7.17 This is an immediate consequence of Theorem 1.7.16. 

Our next result offers a more general perspective on the Green formula established
in Theorem 1.7.12.

Theorem 1.7.18 Let Ω ⊆ Rn be an open set with a lower Ahlfors regular boundary
and with the property that σ := H n−1 ∂Ω is a doubling measure. Hence, Ω is a set
of locally finite perimeter, and its geometric measure theoretic outward unit normal
ν is defined σ-a.e. on ∂∗ Ω.
Fix M, N, N ∈ N and consider a first-order M × N system D with constant
complex coefficients in Rn , along with a first-order N × M system D ' with constant
complex coefficients in R . Then define the constant (complex) coefficient second-
n

order N × N system in Rn given by

'
L := DD. (1.7.238)

Finally, some κ, κ > 0, assume (X, Y) is a regular dual system on (∂Ω, σ) (in the
sense of Definition 1.7.9), and consider two complex vector-valued functions
N N
u ∈ Lloc
1
(Ω, L n ) , w ∈ Lloc
1
(Ω, L n ) , (1.7.239)

satisfying

Du ∈ Lloc
1 (Ω, L n ) M
, ' w ∈ L 1 (Ω, L n )
D
M
,
loc

Nκ (Du) belongs to X, Nκ w belongs to Y,

κ−n.t. κ −n.t.
(Du)∂Ω and w ∂Ω exist σ-a.e. on ∂nta Ω, (1.7.240)

1 (Ω, L n ) N
Lu belongs to Lloc , and
   
' w + Lu, w belongs to L 1 (Ω, L n ).
Du, D

Then for any other specified aperture parameter κ ∈ (0, ∞) the nontangential
κ −n.t. κ −n.t.
boundary traces (Du)∂Ω and w ∂Ω exist at σ-a.e. point on ∂nta Ω and are
independent of κ . Moreover, with the dependence on the parameter κ dropped,
the following Green type formula (involving absolutely convergent integrals) holds
∫ 
   
Lu, w + Du, D ' w dL n
Ω
∫   n.t.  n.t. 
= ' ν) (Du) , w 
(−i)Sym( D; dσ. (1.7.241)
∂Ω ∂Ω
∂∗ Ω

when either Ω is bounded, or ∂Ω is unbounded. In the remaining case, i.e., when Ω

is unbounded and ∂Ω is bounded, formula (1.7.241) continues to be true under the
additional assumption that there exists λ ∈ (1, ∞) such that
1.7 Green-Type Formulas for Second-Order Systems 231
∫
|Du||w| dL n = o(R) as R → ∞. (1.7.242)
[B(0,λ R)\B(0,R)]∩Ω

In particular, corresponding to the case when the penultimate condition in

(1.7.240) is strengthened to

Lu = 0 in the sense of distributions in Ω, (1.7.243)

formula (1.7.241) becomes

∫ ∫ 
   n.t.  n.t. 
' w dL n =
Du, D ' ν) (Du) , w 
(−i)Sym( D; dσ. (1.7.244)
∂Ω ∂Ω
Ω ∂∗ Ω

Finally, similar results are valid if (X, Y) is a dual system on (∂Ω, σ) which is
not necessarily regular, with the understanding that, in such a case, all nontan-
gential boundary traces are taken with respect to an aperture parameter which is
< min{κ, κ }.

Proof In concert with [112, Corollary 8.9.9], the present assumptions guarantee the
existence of some p ∈ (0, ∞] with the property that for any given aperture parameter
κ > 0 we have
p
Nκ (Du), Nκ w ∈ Lloc (∂Ω, σ), (1.7.245)
and the nontangential boundary traces
κ −n.t. κ −n.t.
(Du)∂Ω and w ∂Ω , both exist at σ-a.e. point on ∂nta Ω,
(1.7.246)
and agree with the nontangential boundary traces in (1.7.240).

Next, observe that (1.7.239), the second line in (1.7.240) (or (1.7.245)), and [112,
Lemma 8.3.1] imply
∞ M ∞ N
Du ∈ Lloc (Ω, L n ) and w ∈ Lloc (Ω, L n ) . (1.7.247)

Consider now the vector field defined implicitly by the demand that
 

ξ · F(x) ' ξ)(Du)(x), w(x)
= (−i)Sym( D;
(1.7.248)
for each ξ ∈ Rn and L n -a.e. x ∈ Ω.

In particular, from this and (1.7.247) we conclude that

F ∈ Lloc
∞ n n
(Ω, L n ) ⊆ Lloc
1
(Ω, L n ) . (1.7.249)

 in the sense of distributions in Ω, fix an arbitrary

To compute the divergence of F,
∞
real-valued function ϕ ∈ 𝒞c (Ω) and write
232 1 Integral Representations and Integral Identities
∫
 
D (Ω)
 ϕ
divF, D(Ω) =− ∇ϕ · F dL n
Ω
∫  
=− ' ∇ϕ)(Du), w dL n,
(−i)Sym( D; (1.7.250)
Ω

where the second equality comes from (1.7.248), used with ξ := ∇ϕ. To proceed,

for each sufficiently small ε > 0 consider Ωε := x ∈ Ω : dist(x, ∂Ω) > ε .
Using a Friedrichs mollifier, it is possible to construct a sequence of functions
M
w ε = (wαε )1≤α ≤M ∈ 𝒞∞ (Ωε ) such that

w ε −−−−→
+
w at L n -a.e. point in Ω,
ε→0
for each compact set K ⊂ Ω we have
N
w ε −−−−→
+
w in L 1 (K, L n ) and
ε→0 (1.7.251)
'
D w ε −−−−→
 ' w in L 1 (K, L n )
D
M
,
ε→0+
and there exists some number εK > 0
so that sup0<ε<εK #w ε #[L ∞ (K, L n )] N < ∞.

On account of (1.7.250), (1.7.247), (1.7.251), (1.7.240), and Lebesgue’s Dominated

Convergence Theorem, we may then compute
1.7 Green-Type Formulas for Second-Order Systems 233
 
D (Ω)
 ϕ
divF, D(Ω)
∫  
= − lim+ ' ∇ϕ)Du, w ε dL n
(−i)Sym( D;
ε→0 Ω
∫  
= lim+ ' ; ∇ϕ)w ε dL n
Du, (−i)Sym( D
ε→0 Ω
∫ ∫
   
= lim+ ' (ϕw ε ) dL n − lim
Du, D ' w ε dL n
Du, ϕ D
ε→0 + ε→0
Ω Ω
∫
   
' (ϕw ε )
= lim+ [D (Ω)] M Du, D [D(Ω)] M − ' w ϕ dL n
Du, D
ε→0 Ω
∫
   
'
= lim+ [D (Ω)] N DDu, ϕw ε [D(Ω)] N − ' w ϕ dL n
Du, D
ε→0 Ω
∫
   
= lim+ [D (Ω)] N Lu, ϕw ε [D(Ω)] N − ' w ϕ dL n
Du, D
ε→0 Ω
∫ ∫
   
= lim+ Lu, ϕw ε dL n − ' w ϕ dL n
Du, D
ε→0 Ω Ω
∫ ∫
   
= Lu, w ϕ dL n − ' w ϕ dL n .
Du, D (1.7.252)
Ω Ω

Having established this, we conclude that

   
divF = Lu, w + Du, D ' w in D (Ω), (1.7.253)

so the last line in (1.7.240) imply

divF ∈ L 1 (Ω, L n ). (1.7.254)

 
In addition, if we choose an aperture parameter κ∗ ∈ 0, min{κ, κ } , from (1.7.248)
and [112, (8.2.25)] we see that there exists a constant C ∈ (0, ∞) such that

Nκ∗ F ≤ C Nκ (Du) Nκ w on ∂Ω. (1.7.255)

From (1.7.255), (1.7.245), [112, (8.2.26)], the nontangential maximal function mem-
berships in the second line of (1.7.240), and (1.7.71) we conclude that

Nκ∗ F ∈ L 1 (∂Ω, σ). (1.7.256)

In addition, (1.7.246) and (1.7.248) imply that the nontangential boundary trace
κ −n.t.

F exists at σ-a.e. point on ∂nta Ω. (1.7.257)
∂Ω
234 1 Integral Representations and Integral Identities

Moreover, with the dependence on κ dropped, from (1.7.248) and [112, Proposi-
tion 8.8.6] we conclude that
 n.t.    n.t.  n.t. 
 ' ν) (Du) , w 
ν · F = (−i)Sym( D; ∂Ω ∂Ω
at σ-a.e. point on ∂∗ Ω. (1.7.258)
∂Ω

Let us also note that (1.7.242) implies that the present vector field F satisfies the
growth condition [112, (1.2.3)]. Having established the aforementioned properties of
 [112, Theorem 1.2.1] applies and the Divergence Formula [112, (1.2.2)] currently
F,
yields (1.7.241), thanks to (1.7.253) and (1.7.258). 

One of the virtues of the manner in which Theorem 1.7.16 has been formulated
and proved is that virtually the same considerations continue to be valid in the context
of manifolds. Specifically, here is the counterpart of Theorem 1.7.16 for differential
operators acting on Hermitian vector bundles over a given Riemannian manifold.

Theorem 1.7.19 Suppose

M is a connected, compact, boundaryless, oriented manifold of class
𝒞1 , of real dimension
. n, equipped with a continuous Riemannian (1.7.259)
metric tensor g = 1≤ j,k ≤n g jk dx j ⊗ dxk .

Let Ω be a nonempty, open, proper subset of M with the property that ∂Ω is lower
Ahlfors regular boundary, and that σg := Hgn−1 ∂Ω is a doubling measure on ∂Ω.
In particular, Ω is a set of finite perimeter and its (geometric measure theoretic)
outward unit conormal νg : ∂∗ Ω → T ∗ M is defined σg -a.e. on ∂∗ Ω. Denote by Lgn
the measure induced by the volume element dVg on M.
Consider next three Hermitian vector bundles ℰ, ℱ, 𝒢 → M and, having fixed
two first-order differential operators D : ℰ → ℱ, and D ' : ℱ → 𝒢, introduce
' Also, fix p, q ∈ [1, ∞], denote by p , q their Hölder conjugate exponents,
L := DD.
pick some κ, κ > 0, and consider two sections

u ∈ Lloc
1
(Ω, Lgn ) ⊗ ℰ, w ∈ Lloc
1
(Ω, Lgn ) ⊗ 𝒢, (1.7.260)

satisfying (with all operators applied in the sense of distributions in Ω):

Du ∈ Lloc
1 (Ω, L n ) ⊗ ℱ,
g
' w ∈ L 1 (Ω, Lgn ) ⊗ ℱ,
D loc
Nκ u ∈ L p (∂Ω, σg ), Nκ (Du) ∈ L q (∂Ω, σg ),
Nκ w ∈ L q (∂Ω, σg ), ' w) ∈ L p (∂Ω, σg ),
Nκ ( D
κ−n.t. κ−n.t. κ −n.t. κ −n.t. (1.7.261)
u∂Ω , (Du)∂Ω , w ∂Ω , ( D' w)
∂Ω
exist σg -a.e. on ∂nta Ω,
1 (Ω, L n ) ⊗ 𝒢,
Lu ∈ Lloc g L  w ∈ Lloc
1 (Ω, L n ) ⊗ ℰ,
g
   

Lu, w 𝒢 − u, L w ℰ belongs to L (Ω, Lgn ).
1
1.7 Green-Type Formulas for Second-Order Systems 235

Then for any other specified aperture parameter κ ∈ (0, ∞) the nontangential
κ −n.t. κ −n.t. κ −n.t. κ −n.t.
boundary traces u∂Ω , (Du)∂Ω , w ∂Ω , ( D ' w)
∂Ω
exist at σg -a.e. point on
∂nta Ω and are independent of κ . Moreover, with the dependence on the parameter
κ dropped, the following Green type formula (involving absolutely convergent
integrals) holds:
∫ 
    
Lu, w 𝒢 − u, L  w ℰ dLgn
Ω
∫   n.t.  n.t. 
= ' ν) (Du) , w 
(−i) Sym( D; dσg
∂Ω ∂Ω 𝒢
∂∗ Ω
∫  n.t.   n.t.  
− u∂Ω, (−i) Sym(D ; ν) ( D
' w)
∂Ω
dσg . (1.7.262)
∂∗ Ω ℰ

Moreover, this formula continues to be valid when the nontangential maximal

function conditions stipulated in (the second and third lines of) (1.7.261) are replaced
by more general memberships to the components of any two regular dual systems on
(∂Ω, σg ) (defined in a fashion similar to Definition 1.7.9).
Proof A cursory inspection reveals that real-variable argument used in the proof
of Theorem 1.7.16 may be naturally adapted to the context of manifolds. More
specifically, the definition of the vector field F in (1.7.220) continues to make sense
in the latter context if ξ is now selected from the co-tangent bundle T ∗ M and
the dot product now interpreted as the duality pairing between vectors in T M and
co-vectors in T ∗ M. Since [112, (1.7.17)-(1.7.20)] remain presently valid (see the
discussion pertaining to [112, (1.11.21)-(1.11.25)]), virtually the same argument as
in the Euclidean case carries over and yields the versions of (1.7.226) and (1.7.231)
in the manifold setting. Granted these, [112, Corollary 1.11.5] applies and yields all
desired conclusions. 
We conclude by presenting the version of Theorem 1.7.18 in the setting of dif-
ferential operators acting on Hermitian vector bundles over a given Riemannian
manifold.
Theorem 1.7.20 Suppose the Riemannian manifold (M, g) is as in (1.7.259). Let
Ω be a nonempty, open, proper subset of M with the property that ∂Ω is lower
Ahlfors regular boundary, and that σg := Hgn−1 ∂Ω is a doubling measure on ∂Ω.
In particular, Ω is a set of finite perimeter and its (geometric measure theoretic)
outward unit conormal νg : ∂∗ Ω → T ∗ M is defined σg -a.e. on ∂∗ Ω. Denote by Lgn
the measure induced by the volume element dVg on M.
Consider next three Hermitian vector bundles ℰ, ℱ, 𝒢 → M and, having fixed
two first-order differential operators D : ℰ → ℱ, and D ' : ℱ → 𝒢, introduce
'
L := DD.
In this setting, fix p ∈ [1, ∞], denote by p its Hölder conjugate exponent, pick
some κ, κ > 0, and consider two sections

u ∈ Lloc
1
(Ω, Lgn ) ⊗ ℰ, w ∈ Lloc
1
(Ω, Lgn ) ⊗ 𝒢, (1.7.263)
236 1 Integral Representations and Integral Identities

satisfying

Du ∈ Lloc
1 (Ω, L n ) ⊗ ℱ,
g
' w ∈ L 1 (Ω, Lgn ) ⊗ ℱ,
D loc

Nκ (Du) ∈ L p (∂Ω, σg ), Nκ w ∈ L p (∂Ω, σg ),

κ−n.t. κ −n.t.
(Du)∂Ω and w ∂Ω exist σg -a.e. on ∂nta Ω, (1.7.264)

1 (Ω, L n ) ⊗ 𝒢, and
Lu belongs to Lloc g
   
' w + Lu, w belongs to L 1 (Ω, Lgn ).
Du, D ℱ 𝒢

Then for any other specified aperture parameter κ ∈ (0, ∞) the nontangential
κ −n.t. κ −n.t.
boundary traces (Du)∂Ω and w ∂Ω exist at σg -a.e. point on ∂nta Ω and are
independent of κ . Moreover, with the dependence on the parameter κ dropped,
the following Green type formula (involving absolutely convergent integrals) holds:
∫ 
    
Lu, w 𝒢 + Du, D ' w dLgn
ℱ
Ω
∫   n.t.  n.t. 
= ' νg ) (Du) , w 
(−i)Sym( D; dσg . (1.7.265)
∂Ω ∂Ω 𝒢
∂∗ Ω

Furthermore, formula (1.7.265) remains true when the nontangential maximal

function conditions stipulated in (the second and third lines of) (1.7.264) are replaced
by more general memberships to the components of any regular dual system on
(∂Ω, σg ) (defined in a fashion similar to Definition 1.7.9).

Proof All desired conclusions follow by observing that the proof of Theorem 1.7.18
naturally adapts to the context of manifolds. The most significant novel aspect is
that, this time, in place of [112, Theorem 1.2.1] or [112, Theorem 1.4.1], we rely on
[112, Corollary 1.11.5] (see the discussion in the proof of Theorem 1.7.19). 

1.8 Rellich Identities

We begin by discussing an important integral formula in potential theory, originally

proved by D. Jerison and C. Kenig in the category of Lipschitz domains in [69] (cf.
also [79, Lemma 2.1.7, p. 46]), which has been further studied in [16, Theorem 5.8,
p. 80] and [81, Lemma A.3.1, p. 389] (cf. also [80, §6]) where the identity in question
has also been shown to be valid in the class of chord arc domains with a sufficiently
small constant, as well as the category of δ-Reifenberg flat chord arc domains with
the property that the logarithm of the Poisson kernel has vanishing mean oscillations.
In all scenarios, the proof involves a suitable approximating sequence {Ω j } j ∈N of
the given set Ω which, after establishing said identity in each Ω j , allows one to
pass to the limit j → ∞ and obtain the desired formula for the original domain Ω.
1.8 Rellich Identities 237

The availability of such an approximating scheme in the class of chord-arc domains

has been established in [81] under suitable additional flatness assumptions. The
aforementioned assumptions are then inherited by any result whose proof makes
use of such an approximating scheme. Here, by relying on our earlier results on the
nature of the Green function, we are able to work directly with the original domain
Ω and, most importantly, avoid imposing any artificial flatness conditions on Ω.
Recall that the reverse Hölder classes Bq (σ), 1 < q < ∞, are defined as in item
(v) of [112, Proposition 7.7.9] (with μ := σ and X := ∂Ω).

Theorem 1.8.1 Assume Ω ⊂ Rn is a bounded NTA domain whose boundary is an

upper Ahlfors regular set. Let ν stand for the geometric measure theoretic outward
unit normal to Ω, and abbreviate σ := H n−1 ∂Ω. With ω x0 denoting the harmonic
measure with pole at a fixed point x0 ∈ Ω, suppose

ω x0 ∈ B2 (σ) (1.8.1)

and consider the Poisson kernel k x0 = dω x0 /dσ with pole at x0 .

Then the following Rellich type identity holds:
∫ ∫
k x0 (y) 1  x0  2
dσ(y) = k (y) y − x0, ν(y) dσ(y). (1.8.2)
∂Ω |y − x0 |
n−2 ωn−1 ∂Ω

Proof We begin by defining u : Ω \ {x0 } → R by setting

u(x) := G(x, x0 ) + EΔ (x − x0 ), ∀x ∈ Ω \ {x0 }, (1.8.3)

where GΩ (·, ·) is the Green function for the Laplacian in Ω, and EΔ is the standard
fundamental solution for the Laplacian in Rn defined in (1.5.56). Keeping (5.3.3) in
mind, it follows that u extends to a harmonic function in Ω. Retaining the notation u
for this extension we therefore have

u ∈ 𝒞∞ (Ω) and Δu = 0 in Ω. (1.8.4)

Moreover, (1.8.1) ensures that

k x0 ∈ L 2 (∂Ω, σ) (1.8.5)

which, in concert with (5.4.7), the boundedness of the Hardy-Littlewood maximal

operator on L 2 (∂Ω, σ) (cf. [112, Corollary 7.6.3]), and [112, Proposition 8.4.1],
implies that for each aperture parameter κ > 0 we have

Nκ (∇u) ∈ L 2 (∂Ω, σ). (1.8.6)

In addition, from (5.6.1) we deduce that, for each κ > 0,

  κ−n.t.
the nontangential trace ∇u ∂Ω exists σ-a.e. on ∂Ω
(1.8.7)
and is actually independent of the parameter κ > 0.
238 1 Integral Representations and Integral Identities

Going further, define the function w : Ω → R by setting

w(x) := (x − x0 ) · (∇u)(x), ∀x ∈ Ω. (1.8.8)

Then from (1.8.8), (1.8.4), (1.8.6), and (1.8.7) we see that, for each κ > 0,

w ∈ 𝒞∞ (Ω), Δw = 0 in Ω, Nκ w ∈ L 2 (∂Ω, σ), w(x0 ) = 0, and

κ−n.t. (1.8.9)
w ∂Ω exists σ-a.e. on ∂Ω and is actually independent of κ > 0.

In fact, as far as the latter nontangential boundary trace is concerned, relying on

(1.8.8), (1.8.3), and (5.6.18), at σ-a.e. point y ∈ ∂Ω we may compute
 κ−n.t.     κ−n.t. 
w ∂Ω (y) = (y − x0 ) · ∇u ∂Ω (y)
 &
 κ−n.t. 

= (y − x0 ) · ∇GΩ (·, x0 )  (y) + (∇EΔ )(y − x0 )
∂Ω

1 1
= −k x0 (y)y − x0, ν(y) + . (1.8.10)
ωn−1 |y − x0 | n−2
Having established (1.8.9)-(1.8.10), Corollary 5.7.8 applies to w (thanks to (1.8.1))
and this permits us to write
∫  κ−n.t. 
0 = w(x0 ) = k x0 (y) w ∂Ω (y) dσ(y)
∂Ω
∫
 2
=− k x0 (y) y − x0, ν(y) dσ(y)
∂Ω
∫
1 k x0 (y)
+ dσ(y). (1.8.11)
ωn−1 ∂Ω |y − x0 | n−2
With this in hand, (1.8.2) readily follows. 

Shifting perspectives, we shall establish an integral identity on the boundary

of an Ahlfors regular domain Ω ⊆ Rn , involving functions u ∈ 𝒞∞ (Ω) satisfying
Δu + λ u = 0 in Ω, for some λ ∈ R. Subsequently, in Corollaries 1.8.3-1.8.4 we
shall specialize this result by imposing, respectively, homogeneous Dirichlet and
Neumann boundary conditions on u in order to obtain very general formulas in the
spirit of those originally proved by Rellich in [154].

Theorem 1.8.2 Assume Ω is an open nonempty bounded subset of Rn with a lower

Ahlfors regular boundary and such that σ := H n−1 ∂Ω is a doubling measure on
∂Ω. In particular, Ω is a set of locally finite perimeter, so its geometric measure
theoretic outward unit normal ν is defined σ-a.e. on ∂∗ Ω. Fix κ ∈ (0, ∞), λ ∈ R,
and assume that u ∈ 𝒞∞ (Ω) is a real-valued function satisfying
1.8 Rellich Identities 239
κ−n.t. κ−n.t.
the nontangential traces u∂Ω , (∇u)∂Ω exist σ-a.e. on ∂nta Ω,
(1.8.12)
Nκ u, Nκ (∇u) ∈ L 2 (∂Ω, σ), and Δu + λ u = 0 in Ω.
κ −n.t. κ −n.t.
Then for any κ > 0 the nontangential traces u∂Ω , (∇u)∂Ω exist σ-a.e. on
∂nta Ω and are actually independent of κ . Moreover, suppressing the dependence on
the parameter κ and introducing
 n.t. 
∂ν u := ν, (∇u)∂Ω at σ-a.e. point on ∂∗ Ω, (1.8.13)

one has n.t. n.t.

u∂Ω, (∇u)∂Ω, ∂ν u ∈ L 2 (∂∗ Ω, σ), u ∈ L 2 (Ω, L n ), (1.8.14)
and
⎧ ⎫
⎨ ∫  n.t.  2
⎪
⎪ ∫ ⎪
⎬
⎪
λ u∂Ω ν, x dσ − 2 u2 dL n
⎪
⎪ ⎪
⎪
⎩ ∂∗ Ω Ω ⎭
∫ ∫
 n.t.   n.t. 
= (∇u) 2 ν, x dσ − 2 (∇u)∂Ω, x ∂ν u dσ
∂Ω
∂∗ Ω ∂∗ Ω
∫
 n.t. 
− (n − 2) u∂Ω ∂ν u dσ, (1.8.15)
∂∗ Ω

with all integrals absolutely convergent.

κ −n.t. κ −n.t.
Proof Given any κ > 0, [112, Proposition 8.9.8] implies that u∂Ω , (∇u)∂Ω exist
σ-a.e. on ∂nta Ω and are actually independent of κ . Consequently, suppressing the
dependence on the aperture parameter when considering said nontangential traces
creates no ambiguity. With this convention in mind, the membership of the functions
n.t. n.t.
u∂Ω , (∇u)∂Ω , ∂ν u to L 2 (∂∗ Ω, σ) is seen from assumptions, [112, (8.9.44)], [112,
Proposition 8.8.6], [112, (8.9.8)], (1.8.13), and [112, (5.6.23)]. In addition, [112,
(8.6.51) in Proposition 8.6.3] (used with p := 2 and E := Ω) implies
 
#u# 2n n
≤ C# Nκ u L 2 (∂Ω,σ) < +∞. (1.8.16)
L n−1 (Ω, L )

Since the Lebesgue scale on Ω is nested (given that Ω has finite measure), this allows
us to conclude that 2n
u ∈ L n−1 (Ω, L n ) → L 2 (Ω, L n ), (1.8.17)
which finishes the proof of (1.8.14).
To proceed, introduce the vector field F : Ω → Rn by setting
240 1 Integral Representations and Integral Identities


F(x) : = |(∇u)(x)| 2 x − 2(∇u)(x), x(∇u)(x)

− (n − 2)u(x)(∇u)(x) − λ u(x)2 x, (1.8.18)

for each x ∈ Ω. Then F ∈ 𝒞∞ (Ω) and a straightforward computation (which

n

twice uses the fact that Δu = −λ u) yields

divF = −2∇u, xΔu − (n − 2)uΔu − 2λ∇u, xu − nλ u2

= −2λ u2 in Ω. (1.8.19)

In particular, thanks to (1.8.17),

divF ∈ L 1 (Ω, L n ). (1.8.20)

Moreover, from (1.8.18) and assumptions, it follows that

κ−n.t.
the nontangential trace F∂Ω exists at σ-a.e. point on ∂nta Ω and
(1.8.21)
the nontangential maximal function Nκ F belongs to L 1 (∂Ω, σ).

Granted (1.8.19)-(1.8.21), the Divergence Formula [112, (1.2.2)] applies and readily
gives (1.8.15). This finishes the proof of Theorem 1.8.2. 

More than 75 years ago, F. Rellich has discovered a basic identity for the Dirichlet
eigenvalue problem for the Laplacian in smooth domains (cf. [154]), expressing the
eigenvalue as a weighted boundary integral of the square of the normal derivative
of the normalized eigenfunction. As evidenced by work in [12], [13], [47], [101],
[152], and the references therein, this formula remains an important tool in problems
in partial differential equations and spectral theory. The following consequence
of Theorem 1.8.2 is a geometrically sharp version of Rellich’s celebrated identity
which is valid in a much larger category of domains than previously considered in
the literature.

Corollary 1.8.3 Retain the assumptions on Ω and u made in Theorem 1.8.2 and, in
addition, suppose u does not vanish identically in Ω and satisfies
n.t.
u∂Ω = 0 at σ-a.e. point on ∂∗ Ω. (1.8.22)

Then ∫
2 
∂ν u ν, x dσ
∂Ω
λ= ∗ ∫ (1.8.23)
2 u2 dL n
Ω
with all integrals absolutely convergent.
1.8 Rellich Identities 241

Proof Identity (1.8.23) is a direct consequence of (1.8.15) and [113, (11.3.45)]

(also bearing in mind (1.8.22)). Finally, the fact that the intervening integrals are
absolutely convergent is a consequence of (1.8.14). 
Theorem 1.8.2 also yields a sharp version of Rellich’s identity for the Neumann
eigenvalue problem for the Laplacian in very general domains, of the sort described
in the corollary below.

Corollary 1.8.4 Retain the assumptions on Ω and u made in Theorem 1.8.2 and, in
addition, suppose
∂ν u = 0 at σ-a.e. point on ∂∗ Ω. (1.8.24)
Then, with all integrals absolutely convergent,
⎧ ⎫
⎨ ∫  n.t.  2
⎪
⎪ ∫ ⎬ ∫ 
⎪
⎪ n.t. 
λ u∂Ω ν, x dσ − 2 u2 dL n = (∇u) 2 ν, x dσ (1.8.25)
⎪
⎪ ⎪
⎪
∂Ω
⎩ ∂∗ Ω Ω ⎭ ∂∗ Ω
and
∫ ∫ ∫
 n.t.  2  n.t. 
λ u∂Ω ν, x dσ = (∇u) 2 ν, x dσ + 2
∂Ω
|∇u| 2 dL n . (1.8.26)
∂∗ Ω ∂∗ Ω Ω

n.t.
Moreover, if Ω is actually a UR domain, then (∇u)∂Ω may be replaced by
 n.t. 
∇tan u∂Ω in (1.8.25) and (1.8.26).

Proof Formula (1.8.25) is readily implied by (1.8.15) and (1.8.24). In turn, formula
(1.8.26) is obtained from (1.8.25) upon observing that
∫ ∫
λ u2 dL n = |∇u| 2 dL n . (1.8.27)
Ω Ω

The latter integral identity is justified by applying the Divergence Formula [112,
(1.2.2)] to the vector field F := u∇u ∈ 𝒞∞ (Ω) which, thanks to the present
n

hypotheses, satisfies Nκ F ∈ L 1 (∂Ω, σ),

divF = |∇u| 2 + uΔu = |∇u| 2 − λ u2 ∈ L 1 (Ω, L n ) (1.8.28)

κ−n.t.
as well as F∂Ω exists at σ-a.e. point on ∂nta Ω, and
 κ−n.t.   κ−n.t. 
ν · F∂Ω = u∂Ω ∂ν u = 0 at σ-a.e. point on ∂∗ Ω. (1.8.29)

Finally, the last claim in the statement of the corollary is a consequence of [113,
(11.4.41)]. 
The principle behind the proof of Theorem 1.8.2, i.e., applying our Divergence
Theorem to a vector field suitably tailored from the original function(s), is robust
242 1 Integral Representations and Integral Identities

and may be implemented in a variety of situations of interest. Here is a Rellich type

identity involving holomorphic functions.

Theorem 1.8.5 Suppose Ω is an open nonempty proper subset of R2 ≡ C with a

lower Ahlfors regular boundary and such that σ := H 1 ∂Ω is a doubling measure
on ∂Ω. Consequently, Ω is a set of locally finite perimeter, so its geometric measure
theoretic outward unit normal

ν = (ν1, ν2 ) ≡ ν1 + iν2 (1.8.30)

is defined σ-a.e. on ∂∗ Ω. Fix κ ∈ (0, ∞) and assume u, w : Ω → C are two

holomorphic functions satisfying
κ−n.t. κ−n.t.
the nontangential traces u∂Ω , w ∂Ω exist σ-a.e. on ∂nta Ω,
   2 (1.8.31)
and the product Nκ u · Nκ w belongs to L 1 (∂Ω, σ).

Then, with bar denoting complex conjugation, one has

∫
 κ−n.t.   κ−n.t. 2
Re ν u∂Ω w ∂Ω  dσ
∂∗ Ω
∫
 κ−n.t.   κ−n.t.    κ−n.t.  
= 2 Re u∂Ω w ∂Ω Re ν w ∂Ω dσ (1.8.32)
∂∗ Ω

when either Ω is bounded, or ∂Ω is unbounded. In the remaining cases, i.e., when

Ω is unbounded and ∂Ω is bounded, formula (1.8.32) continues to hold under the
additional assumption that there exists λ ∈ (1, ∞) such that
∫
|u||w| 2 dL 2 = o(R) as R → ∞. (1.8.33)
[B(0,λ R)\B(0,R)]∩Ω

Proof Define two real vector fields a, b : Ω → R2 by setting

   
a := Re u, Im u and b := Re w, −Im w , (1.8.34)

then consider the vector field

 2 a − 2a, b
F := | b|  b ∈ 𝒞∞ (Ω) 2 . (1.8.35)

A direct computation based on the Cauchy-Riemann equations shows that

divF = 0 in Ω. (1.8.36)

Also, from (1.8.34)-(1.8.35) and (1.8.31) we conclude that

κ−n.t.
F∂Ω exists σ-a.e. on ∂nta Ω, and Nκ F ∈ L 1 (∂Ω, σ). (1.8.37)
1.8 Rellich Identities 243

Moreover, at σ-a.e. point on ∂∗ Ω we have

 κ−n.t.    κ−n.t.    κ−n.t. 
ν, F∂Ω = Re ν u∂Ω w 
∂Ω
2
  κ−n.t.   κ−n.t.     κ−n.t.  
− 2 Re u∂Ω w ∂Ω Re ν w ∂Ω . (1.8.38)

Lastly, when Ω is an exterior domain, (1.8.33) implies that

∫
| F | dL 2 = o(R) as R → ∞. (1.8.39)
B(0,λ R)\B(0,R)

Having established (1.8.35)-(1.8.39), the Divergence Formula [112, (1.2.2)] may be

invoked to conclude that (1.8.32) holds. 

Another embodiment of the principle used in the proof of Theorem 1.8.2 is as

follows. Assume Ω ⊆ Rn is an open set with an Ahlfors regular boundary and
abbreviate σ := H n−1 ∂Ω. Also, let ν stand for the geometric measure theoretic
outward unit normal to Ω and fix an aperture parameter κ ∈ (0, ∞). In this setting,
we have a Rellich-type identity of the following sort:
∫ ∫ ∫
 κ−n.t. 2  κ−n.t. 
ν, h (∇u)∂Ω  dσ = 2 h, (∇u)∂Ω (∂ν u) dσ + (divh)|∇u| 2 dL n
∂∗ Ω ∂∗ Ω Ω
∫
−2 Dh, ∇u ⊗ ∇u dL n . (1.8.40)
Ω
n
Above, h is a vector field in 𝒞1c (Rn ) , the symbol Dh denotes the Jacobian matrix
of h, and u is a harmonic function in Ω which, for some fixed κ > 0, satisfies
κ−n.t.
2 (∂Ω, σ) and the nontangential trace (∇u)
Nκ (∇u) ∈ Lloc ∂Ω
exists at σ-a.e. point on
 κ−n.t. 

∂nta Ω. Also, ∂ν u := ν · (∇u) ∂Ω is the normal derivative of u on ∂∗ Ω, and ∇u ⊗ ∇u
 
is (cf. (1.3.7)) the n × n matrix ∂j u∂k u 1≤ j,k ≤n .
Identity (1.8.40) is obtained by applying the Divergence Formula [112, (1.2.2)]
to the vector field

F := |∇u| 2 h − 2h, ∇u∇u ∈ 𝒞1 (Ω) .

n
(1.8.41)

The role of the hypothesis Nκ (∇u) ∈ Lloc

2 (∂Ω, σ) is to ensure that N F
κ
 ∈ L 1 (∂Ω, σ)
which is a key assumption in [112, Theorem 1.2.1]. In the particular case when Ω is
bounded and h(x) = x for each x near Ω (in which scenario divh = n and Dh = I in
Ω), formula (1.8.40) becomes
∫
 κ−n.t. 2
ν, x (∇u)∂Ω  dσ (1.8.42)
∂∗ Ω
∫ ∫
 κ−n.t. 
=2 x, (∇u) ∂Ω
(∂ν u) dσ + (n − 2) |∇u| 2 dL n
∂∗ Ω Ω
244 1 Integral Representations and Integral Identities

which refines results of a similar nature derived under more stringent regularity
assumptions in [70, (2), p. 204], [79, Lemma 2.1.13, p. 48], [144, §3.7-§3.8], [179].
According to [112, Theorem 1.2.1], formula (1.8.40) continues to be valid when
n
Ω has an unbounded boundary and h ∈ 𝒞1 (Rn ) is bounded, with bounded first-
order partial derivatives. For example, if we take Ω to be the domain in Rn above the
graph of a continuous function ϕ : Rn−1 → R whose first-order partial derivatives
belong to BMO(Rn−1 ), and consider h ≡ e j with j ∈ {1, . . . , n}, eventually allows
us to recast (1.8.40) as
∫ ∫
 κ−n.t. 2  κ−n.t. 
ν (∇u)∂Ω  dσ = 2 (∇u)∂Ω (∂ν u) dσ. (1.8.43)
∂∗ Ω ∂∗ Ω

This refines earlier work, including [79, Lemma 2.2.15, p. 55], [105, Théorème 6,
p. 496], and [141].
Due to the influence of the book [141] by Nečas on this topic, integral formulas of
this nature are occasionally referred to as Nečas-Rellich identities. Here is a Nečas-
Rellich identity for systems of differential equations, refining work similar in spirit
from [29], [40, Lemma 2.1], [42], [38], [68], [126], [135], [141, (5.47), p. 264].
Proposition 1.8.6 Let Ω be an open nonempty proper subset of Rn (n ∈ N, n ≥ 2)
with a lower Ahlfors regular boundary and such that σ := H n−1 ∂Ω is a doubling
measure on ∂Ω. Denote by ν the geometric measure theoretic outward unit normal
to Ω, and fix some aperture
 αβ parameter
 κ ∈ (0, ∞). Next, for some M ∈ N, consider a
coefficient tensor A = ar s 1≤r,s ≤n with complex entries satisfying the symmetry
1≤α,β ≤M
condition
αβ βα
ar s = asr for each α, β ∈ {1, . . . , M } and each r, s ∈ {1, . . . , n}, (1.8.44)
 αβ 
then associate with A the M × M second-order system L = ar s ∂r ∂s 1≤α,β ≤M .
In addition, consider a C M -valued function u = (uβ )1≤β ≤M ∈ [𝒞2 (Ω)] M with
κ−n.t.
the property that Nκ (∇u) belongs to L 2 (∂Ω, σ), the nontangential trace (∇u)
loc ∂Ω
2n/(n+1) M
exists at σ-a.e. point on ∂nta Ω, and Lu belongs to Lbdd (Ω, L n ) . Finally, fix
an arbitrary vector field h = (hi )1≤i ≤n ∈ [𝒞c (R )] .
1 n n

Then for any other aperture parameter parameter κ > 0 the nontangential
κ −n.t.
trace (∇u)∂Ω exists at σ-a.e. point on ∂nta Ω and is actually independent of κ .
Moreover, with the dependence on the parameter κ dropped, one has (with the
conormal derivative ∂νAu defined as in (1.7.9))
∫  
αβ  n.t.   n.t.   n.t. 
ν, har s (∂r uα )∂Ω (∂s uβ )∂Ω − 2h j (∂j uα )∂Ω (∂νAu)α dσ
∂∗ Ω
∫ 
αβ αβ
= (∂i hi )ar s (∂r uα )(∂s uβ ) − 2(∂i h j )ais (∂j uα )(∂s uβ )
Ω

− 2h j (∂j uα )(Lu)α dL n, (1.8.45)
1.8 Rellich Identities 245

where all integrals are absolutely convergent. As a corollary,

∫ 
αβ  n.t.   n.t. 
− ν, har s (∂r uα )∂Ω (∂s uβ )∂Ω
∂∗ Ω
  n.t.   n.t.   αβ  n.t.  
+ 2h j νk (∂j uα )∂Ω − ν j (∂k uα )∂Ω ak (∂ uβ )∂Ω dσ
∫ 
αβ αβ
= (∂i hi )ar s (∂r uα )(∂s uβ ) − 2(∂i h j )ais (∂j uα )(∂s uβ )
Ω

− 2h j (∂j uα )(Lu)α dL n . (1.8.46)

κ −n.t.
Proof For starters, the fact that for each κ > 0 the nontangential trace (∇u)∂Ω
n·M
exists at σ-a.e. point on ∂nta Ω, belongs to L 2 (∂∗ Ω, σ) and is actually inde-
pendent of κ follows from assumptions and [112, Propositions 8.9.5, 8.9.8, 8.8.6].
As such, there is no ambiguity in dropping the dependence on the parameter κ
n.t.
and simply denoting this nontangential trace by (∇u)∂Ω . Next, the idea is to apply
 
[112, Theorem 1.2.1] to the vector field F = Fi 1≤i ≤n : Ω → Cn whose scalar
components are given by
αβ αβ
Fi := hi ar s (∂r uα )(∂s uβ ) − 2h j ais (∂j uα )(∂s uβ ), ∀i ∈ {1, . . . , n}. (1.8.47)

Thus, F ∈ 𝒞1 (Ω)
n
and a direct computation gives
αβ αβ
∂i Fi = (∂i hi )ar s (∂r uα )(∂s uβ ) + hi ar s (∂i ∂r uα )(∂s uβ )
αβ αβ
+ hi ar s (∂r uα )(∂i ∂s uβ ) − 2(∂i h j )ais (∂j uα )(∂s uβ )
αβ αβ
− 2h j ais (∂i ∂j uα )(∂s uβ ) − 2h j ais (∂j uα )(∂i ∂s uβ )
=: I + II + III + IV + V + VI. (1.8.48)

Since we are assuming the symmetry condition (1.8.44), in term II we may switch α
with β and r with s to conclude that II = III. Hence,
αβ αβ
II + III = 2hi ar s (∂i ∂r uα )(∂s uβ ) = 2h j ais (∂j ∂i uα )(∂s uβ ) = −V, (1.8.49)

where the second equality in (1.8.49) is obtained by re-denoting i by j, and then

re-denoting r by i. Also, using the definition of L we see that

VI = −2h j (∂j uα )(Lu)α . (1.8.50)

Combining (1.8.48)-(1.8.50) leads to

246 1 Integral Representations and Integral Identities
αβ αβ
divF = (∂i hi )ar s (∂r uα )(∂s uβ ) − 2(∂i h j )ais (∂j uα )(∂s uβ )
− 2h j (∂j uα )(Lu)α . (1.8.51)

Since Nκ (∇u) ∈ Lloc

2 (∂Ω, σ), [112, Proposition 8.6.3] implies that ∇u belongs to
2n/(n−1) n·M 2n/(n+1) M
Lbdd (Ω, L n ) . It is also assumed that Lu ∈ Lbdd (Ω, L n ) , so we
may ultimately conclude from (1.8.51) that divF ∈ L (Ω, L ). In view of (1.8.47),
1 n

the hypothesis Nκ (∇u) ∈ Lloc2 (∂Ω, σ) also entails N F

κ
 ∈ L 1 (∂Ω, σ). By design, F
κ−n.t.
vanishes outside of a bounded subset of Ω. Finally, F∂Ω exists at σ-a.e. point on
∂nta Ω and
 κ−n.t.  αβ  n.t.   n.t. 
ν · F∂Ω = ν, har s (∂r uα )∂Ω (∂s uβ )∂Ω
 n.t. 
− 2h j (∂j uα )∂Ω (∂νAu)α (1.8.52)

at σ-a.e. point on ∂∗ Ω. At this stage, (1.8.45) follows from [112, (1.2.2)] and (1.8.51)-
(1.8.52). Finally, (1.8.46) follows from (1.8.45) by expanding
 n.t. 
−2h j (∂j uα )∂Ω (∂νAu)α
 n.t.  αβ  n.t. 
= −2h j (∂j uα )∂Ω νk ak (∂ uβ )∂Ω
  n.t.   n.t.   αβ  n.t. 
= −2h j νk (∂j uα )∂Ω − ν j (∂k uα )∂Ω ak (∂ uβ )∂Ω

αβ  n.t.   n.t. 
− ν, hak (∂k uα )∂Ω (∂ uβ )∂Ω , (1.8.53)

and combining like-terms in the context of (1.8.45). 

Rellich type identities have been extended from the Euclidean space to the setting
of Lipschitz subdomains of Riemannian manifolds in [119], [129], [130], [132],
[133], [134]. Given the availability of a suitable version of the Divergence Theorem
on manifolds (cf. the discussion in [112, §1.11]), such results may be established in
considerably larger classes of domains on manifolds. Here we further contribute to
this line of work by establishing a very general Rellich-type identity40 on Riemannian
manifolds, of the sort described in the theorem below.

Theorem 1.8.7 Let (M, g) compact, connected, boundaryless, oriented Riemannian

manifold of class 𝒞2 , of real dimension n, and denote by Lgn the measure induced by
the volume element on M.  Suppose
 E, F → M are two Hermitian vector bundles of
class 𝒞1 , and denote by ·, · E and ·, · F the real (i.e., complex bilinear) pointwise
pairings in the fibers of E and F , respectively. Also, assume ∇ E and ∇ F are con-
nections on E and F , respectively. Next, let D : E → F be a first-order differential
operator whose coefficients are of class 𝒞1 . Also, pick some A ∈ Hom (F , F ) with

40 which may well be labeled “the mother of all Rellich identities”

1.8 Rellich Identities 247

coefficients of class 𝒞1 and satisfying

A∗ = A, (1.8.54)

then use it to define the second-order differential operator

L := −D∗ AD. (1.8.55)

Going further, let Ω be a nonempty, open, proper subset of M such that ∂Ω is

a lower Ahlfors regular set, and σg := Hgn−1 ∂Ω is a doubling measure on ∂Ω.
In particular, Ω is a set of finite perimeter, hence its geometric measure theoretic
outward unit conormal νg : ∂∗ Ω → T ∗ M is defined σg -a.e. on ∂∗ Ω. Finally, consider
a section
2,1
u ∈ Wloc (Ω) ⊗ E (1.8.56)
which, for some aperture parameter κ > 0, satisfies

Nκ u, Nκ (∇ E u) belong to L 2 (∂Ω, σg ),
κ−n.t. κ−n.t.
u∂Ω and (∇ E u)∂Ω exist σg -a.e. on ∂nta Ω, (1.8.57)
2n
Lu belongs to the space L n+1 (Ω, Lgn ) ⊗ E,

and define the conormal derivative

κ−n.t.
∂νL u := iSym(D∗ ; νg )A(Du)∂Ω at σg -a.e. point on ∂∗ Ω. (1.8.58)

Then for each vector field h ∈ 𝒞1 (M, T M) with real-valued components one has
∫ 
 κ−n.t.  c 
2 Re ∂νL u, (∇hE u)∂Ω dσg
∂∗ Ω E
∫  κ−n.t.  κ−n.t.  c 
− T ∗ M (νg, h)T M A(Du)∂Ω , (Du)∂Ω dσg
∂∗ Ω F
∫ ∫
   
= 2 Re Lu, (∇hE u)c E
dLgn − (divg h) ADu, (Du)c F dLgn
Ω Ω
∫ ∫
     
+ O |Du| 2 [∇hF, A] dLgn + O | A||Du| [D, ∇h ]u dLgn
Ω Ω
∫
 
+ O | A||h||Du| 2 dLgn, (1.8.59)
Ω

where the superscript ‘c’ indicates complex conjugation, and where the constants
implicit in all big “O” terms depend at most on the metrics.

Before we present the proof of this theorem we make a number of comments

aimed to shed further light on its scope and nature.
248 1 Integral Representations and Integral Identities

Comment 1. It is of significance to note that the commutator

[D, ∇h ] := D∇hE − ∇hF D is a first-order differential operator, (1.8.60)

since its principal symbol (as a second-order operator) vanishes. Indeed, for each
covector ξ ∈ T ∗ M we have
 
Sym [D, ∇h ]; ξ = Sym(D; ξ)iT ∗ M (ξ, h)T M I E

− iT ∗ M (ξ, h)T M I F Sym(D; ξ) = 0. (1.8.61)

As such, all residual terms (appearing in the last three integrals in (1.8.59)) involve
at most first-order derivatives on the section u. This is a key feature of the identity
recorded in (1.8.59).
Let us also observe that since
 
(−i)Sym ∇hF ; ξ = T ∗ M (ξ, h)T M I F for each ξ ∈ T ∗ M, (1.8.62)

(cf. [118, (9.1.45), p. 376]), it follows that for each covector ξ ∈ T ∗ M we have
 
(−i)Sym [∇hF, A]; ξ = T ∗ M (ξ, h)T M I F A − AT ∗ M (ξ, h)T M I F = 0. (1.8.63)

From this we conclude that

[∇hF, A] is a zero-order operator. (1.8.64)

Comment 2. Recall that the connection ∇ F satisfies a product rule. Specifically, for
any two given sections s, t of class 𝒞1 in F the following Leibniz formula holds:

∇h s, t F = ∇hF s, t F + s, ∇hF t F + Ch (s, t) (1.8.65)

in a pointwise sense, where Ch is a bilinear form on F . In other words, the connection

∇ F is “almost” metric (see [118, (9.1.50), p. 377]); if ∇ F is a metric connection (aka
Hermitian connection), then Ch = 0 and the very last integral in (1.8.59) (for which
Ch is responsible) may be omitted.
Comment 3. The proof of Theorem 1.8.7 given below also works in the case when
the manifold M is the entire Euclidean space Rn (equipped with the standard metric),
and E, F are trivial vector bundles, say E ≡ R N and F ≡ R N , provided either
the (real, 𝒞1 ) vector field h is compactly supported, or h is a constant vector field
and D, A have constant coefficients. In such a scenario, the set Ω is allowed to be
unbounded, with the understanding that it Ω is an exterior domain then ∇u is required
to vanish at infinity.
We now take up the task of presenting the proof of Theorem 1.8.7.
Proof of Theorem 1.8.7 To get started, fix a vector field h ∈ 𝒞1 (M, T M) with
real-valued components. Thanks to (1.8.65) and the fact that h has real-valued
components, we then have the following sequence of equalities:
1.8 Rellich Identities 249
 
ADu, (∇hF Du) c
F
 
= ADu, ∇hF (Du)c F
     
= ∇h ADu, (Du) c
F
− ∇hF ADu, (Du)c F − Ch ADu, (Du)c
     
= ∇h ADu, (Du)c F − [∇hF, A]Du, (Du)c F − Ch ADu, (Du)c
   
− A∇hF Du, (Du)c F − Ch ADu, (Du)c , (1.8.66)

pointwise Lgn -a.e. in Ω. The hypothesis adopted in (1.8.54) allows us to re-write the
penultimate term in (1.8.66) as
 F      c
A∇h Du, (Du)c F = ∇hF Du, (ADu)c F = ADu, (∇hF Du)c F . (1.8.67)

In view of (1.8.66)-(1.8.67) and bearing in mind the observation made in (1.8.64)

(which is relevant in relation to the third-from-last term in (1.8.66)), we may ulti-
mately recast (1.8.66) as
   
2 Re ADu, (∇hF Du)c F = ∇h ADu, (Du)c F
    
+ O |Du| 2 [∇hF, A] + O | A||h||Du| 2 . (1.8.68)

Going further, the current hypotheses on Ω, the first line in (1.8.57), and [112,
(8.6.51)] imply
∇ E u ∈ L n−1 (Ω, Lgn ) ⊗ E.
2n
(1.8.69)
In particular,
2n
Du ∈ L n−1 (Ω, Lgn ) ⊗ F . (1.8.70)

Next, define the vector field F : Ω → T M implicitly, via the requirement that at
each point x ∈ Ω we have
     

Tx∗ M ξ, F(x) Tx M = − Tx∗ M ξ, h(x) Tx M A(x)(Du)(x), (Du)(x)
c
Fx
 
− 2Re (−i)Sym(D∗ ; ξ)A(x)(Du)(x), (∇hE u)(x)c (1.8.71)
Ex

for each covector ξ ∈ Tx∗ M. Since the right-hand side of (1.8.71) is linear in ξ, this
definition is meaningful. Moreover, it is apparent from (1.8.71) and (1.8.56) that

F ∈ Lloc
1
(Ω, Lgn ) ⊗ T M. (1.8.72)

Also,
Nκ F ≤ C Nκ (∇ E u)2 pointwise on ∂Ω. (1.8.73)
250 1 Integral Representations and Integral Identities

As a consequence of (1.8.73), the first line in (1.8.57), and [112, (8.2.26)] we

conclude that
Nκ F ∈ L 1 (∂Ω, σg ). (1.8.74)
In addition, the second line in (1.8.57) ensures that
the pointwise nontangential boundary
κ−n.t. (1.8.75)
trace F∂Ω exists σg -a.e. on ∂nta Ω,

and at σg -a.e. point x ∈ ∂∗ Ω we have

  κ−n.t.  
T ∗ M νg (x), F
 (x) T M
x ∂Ω x

    κ−n.t.   κ−n.t.  
= − Tx∗ M νg (x), h(x) Tx M A(x) (Du)∂Ω (x), (Du)∂Ω (x)c
Fx
  κ−n.t.   κ−n.t.  
− 2Re (−i)Sym(D∗ ; νg (x))A(x) (Du)∂Ω (x), (∇hE u)∂Ω (x)c
Ex

    κ−n.t.   κ−n.t.  
= − Tx∗ M νg (x), h(x) Tx M A(x) (Du)∂Ω , (Du)∂Ω (x)c
Fx
  κ−n.t.  
+ 2Re (∂νL u)(x), (∇hE u)∂Ω (x)c . (1.8.76)
Ex

To compute the (differential geometric) divergence of F,  we find it convenient

to work in the sense of distributions. To this end, select an arbitrary real-valued
function ϕ ∈ 𝒞1c (Ω) and recall (cf. [118, (9.1.4) on p. 372, (9.2.1) on p. 381]) that

∇h ϕ = h(ϕ) = dϕ(h) = T ∗ M (dϕ, h)T M , (1.8.77)

and that
(∇h ) = −∇h − divg h (1.8.78)
(cf. [118, (9.2.16), p. 383]). Use these to compute
1.8 Rellich Identities 251
∫
   
 ϕ =−
divg F, T∗M dϕ, F TM dLgn
Ω
∫
   
= T∗M dϕ, h TM ADu, (Du)c F
dLgn
Ω
∫  
+ 2Re (−i)Sym(D∗ ; dϕ)ADu, (∇hE u)c dLgn
Ω E
∫
 
= (∇h ϕ) ADu, (Du)c F dLgn
Ω
∫  
+ 2Re (−i)Sym(D∗ ; dϕ)ADu, (∇hE u)c dLgn
Ω E

=: I + II, (1.8.79)

where the second equality is implied by (1.8.77) and by (1.8.71) written for ξ := dϕ.
As regards the first term above, use (1.8.78) and integration by parts to write (bearing
in mind that ϕ is compactly supported in Ω)
∫  ∫
   
I=− ϕ∇h ADu, (Du) F dLg −
c n
ϕ(divg h) ADu, (Du)c F dLgn .
Ω Ω
(1.8.80)

Concerning the second term in the last line of (1.8.79), in view of the commutator
identity (cf. [112, (1.11.25)])

iSym(D∗ ; dϕ) = ϕD∗ − D∗ (ϕ ·), (1.8.81)

we obtain
252 1 Integral Representations and Integral Identities
∫   ∫  
II = 2Re D∗ (ϕADu), (∇hE u)c dLgn − 2Re ϕ D∗ ADu, (∇hE u)c dLgn
Ω E Ω E
∫   ∫  
= 2Re ϕ ADu, (D∇hE u)c dLgn − 2Re ϕ D∗ ADu, (∇hE u)c dLgn
Ω E Ω E
∫   ∫   c
= 2Re ϕ ADu, (∇hE Du)c dLgn + 2Re ϕ ADu, [D, ∇hE ]u dLgn
Ω E Ω E
∫  
− 2Re ϕ D∗ ADu, (∇hE u)c dLgn
Ω E
∫  ∫
   
= ϕ∇h ADu, (Du) c
F
dLgn + 2Re ϕ Lu, (∇hE u)c E dLgn
Ω Ω
∫ ∫
     
+ ϕ · O |Du| 2 [∇hF, A] dLgn + ϕ · O | A||Du| [D, ∇h ]u dLgn
Ω Ω
∫
 
+ ϕ · O | A||h||Du| 2 dLgn, (1.8.82)
Ω

where the last equality is based on (1.8.68), the fact that ϕ is real-valued, and (1.8.55).
In concert, (1.8.79), (1.8.80), and (1.8.82) prove that
      
divg F = 2Re Lu, (∇hE u)c E − (divg h) ADu, (Du)c F + O |Du| 2 [∇hF, A]
    
+ O | A||Du| [D, ∇h ]u + O | A||h||Du| 2 . (1.8.83)

In view of the last line in (1.8.57) and (1.8.69)-(1.8.70), this implies (keeping in
mind that (2n)/(n + 1) and (2n)/(n − 1) are conjugate exponents) that

divg F ∈ L 1 (Ω, Lgn ). (1.8.84)

At this stage, having established (1.8.72), (1.8.74), (1.8.75), and (1.8.84), [112,
Corollary 1.11.5] applies and, on account of (1.8.76) and (1.8.83), the Divergence
Formula [112, (1.11.20)] presently yields (1.8.59). 
We would like to further refine the Rellich identity recorded in (1.8.59) under the
additional assumption that the second-order differential operator L is weakly elliptic.
As in the past, this amounts to having Sym(L; ξ) an invertible linear map for each
ξ ∈ T ∗ M \ {0}. Loosely speaking, this refinement is carried out by decomposing
D into its tangential and normal component on ∂Ω, analogously to the standard
decomposition of the full gradient operator in Rn into its tangential and normal
components on ∂Ω.
Let us describe a procedure which, given an arbitrary first-order differential
operator P : E → G, where G → M is a Hermitian vector bundle of class 𝒞1 ,
allows one to decompose P as the sum of a tangential differential operator on ∂Ω
and a suitable “multiple” of the conormal derivative operator ∂νL . The key observation
is that the operator
1.8 Rellich Identities 253

τP := P − iSym(P; νg )Sym(L; νg )−1 ∂νL (1.8.85)

is tangential on ∂Ω, in the sense that this may be expressed as a linear combina-
tion of tangential differential operators expressed in local coordinates as (see [113,
(11.6.23)])
∂τ j k := (νg ) j ∇∂k − (νg )k ∇∂j , 1 ≤ j, k ≤ n, (1.8.86)
eventually plus a zero-th order term. This is ensured by the fact that Sym(τP ; νg ) = 0
at σg -a.e. point on ∂∗ Ω, which follows readily from definitions:

Sym(τP ; νg )

= Sym(P; νg ) − iSym(P; νg )Sym(L; νg )−1 iSym(D∗ ; νg )A Sym(D; νg )

= Sym(P; νg ) + Sym(P; νg )Sym(L; νg )−1 Sym(−L; νg )

= Sym(P; νg ) − Sym(P; νg ) = 0. (1.8.87)

Indeed, if a first-order differential operator Q locally written locally as


Q= A j ∇∂ j + B (1.8.88)
j
.
satisfies Sym(Q; νg ) = 0 at σg -a.e. point on ∂∗ Ω, then j (νg ) j A j = 0, and since
 
(νg )k (νg )k = g jk (νg )k (νg ) j = νg, νg  = |νg | 2 = 1, (1.8.89)
k j,k

we may write
 
Q= A j ∇∂ j + B − A j (νg ) j (νg )k ∇∂k
j j,k
 
= A j (νg )k (νg )k ∇∂j + B − A j (νg ) j (νg )k ∇∂k
j,k j,k
  
= A j (νg )k (νg )k ∇∂j − (νg ) j ∇∂k + B
j,k

=− A j (νg )k ∂τ j k + B, (1.8.90)
j,k

which goes to show that the operator Q is, as claimed, tangential.

Having fixed some background aperture parameter κ > 0, the manner in which
the operator τP from (1.8.85) acts on a section
1,1
u ∈ Wloc (Ω) ⊗ E with Nκ u, Nκ (∇ E u) ∈ L 2 (∂Ω, σg ),
κ−n.t. κ−n.t. (1.8.91)
and so that u , (∇ E u)
∂Ω ∂Ω
exist σg -a.e. on ∂ Ω, nta
254 1 Integral Representations and Integral Identities

is κ−n.t.
τP u := (Pu)∂Ω − iSym(P; νg )Sym(L; νg )−1 ∂νL u. (1.8.92)
In the case when this procedure is applied to D (and G := F ), the resulting
tangential operator

τD := D − iSym(D; νg )Sym(L; νg )−1 ∂νL (1.8.93)

has the extra property that

Sym(D∗ ; νg )AτD = −i∂νL − iSym(D∗ ; νg )ASym(D; νg )Sym(L; νg )−1 ∂νL

= −i∂νL + i∂νL = 0. (1.8.94)

Let us now fix a section u satisfying the properties listed in (1.8.91). Then, as

seen from (1.8.93) and (1.8.58),
 κ−n.t. 
|∂νL u| + |τD u| ≈ (Du)∂Ω  uniformly, pointwise on ∂∗ Ω. (1.8.95)

Indeed, the design of the conormal derivative (cf. (1.8.58)) implies

 κ−n.t.   κ−n.t. 

|∂ L u| =  Sym(D∗ ; νg )A(Du)
ν  ≤ C (Du)
∂Ω ∂Ω
 (1.8.96)

at σg -a.e. point on ∂∗ Ω, while from (1.8.92)-(1.8.93) we see that

 κ−n.t.  
|τD u| ≤ C (Du)∂Ω  + |∂νL u| at σg -a.e. point on ∂∗ Ω. (1.8.97)

Together, (1.8.96) and (1.8.97) establish the left-pointing inequality in (1.8.95). The
right-pointing inequality is justified by decomposing

D = i Sym(D; νg )Sym(L; νg )−1 ∂νL + τD . (1.8.98)

Henceforth, work on ∂∗ Ω. Using (1.8.98) we may expand

 κ−n.t.  κ−n.t.  c 
A(Du)∂Ω , (Du)∂Ω
F
  κ−n.t.  c 
= iASym(D; νg )Sym(L; νg )−1 ∂νL u, (Du)∂Ω
F
  c
+ AτD u, iSym(D; νg )Sym(L; νg )−1 ∂νL u
F
 
+ AτD u, (τD u) c
F

=: I + II + III. (1.8.99)

Observe that, thanks to [112, (1.7.19)] and (1.8.58),

1.8 Rellich Identities 255
   κ−n.t.   c 
I = Sym(L; νg )−1 ∂νL u, − iSym(D∗ ; νg )A (Du)∂Ω
E
  c
= Sym(L; ν)−1 ∂νL u, − ∂νL u (1.8.100)
E

and that, by [112, (1.7.19)] and (1.8.94),

  c
II = Sym(D∗ ; νg )AτD u, iSym(L; νg )−1 ∂νL u = 0. (1.8.101)
E

All in all, we have the following identity41:

 κ−n.t.  κ−n.t.  c    c
A(Du)∂Ω , (Du)∂Ω = Sym(−L; νg )−1 ∂νL u, ∂νL u
F E
 
+ AτD u, (τD u) c
F
. (1.8.102)

Similarly, we decompose

∇hE = T ∗ M (νg, h)T M Sym(−L; νg )−1 ∂νL + τh, (1.8.103)

where τh abbreviates τ∇E , the tangential differential operator associated as in (1.8.85)

h
with the first-order differential operator P := ∇hE : E → E. Specifically, according
to (1.8.85) and (1.8.62),

τh := ∇hE − T ∗ M (νg, h)T M Sym(−L; νg )−1 ∂νL . (1.8.104)

Thus, bearing in mind that L ∗ = L, we obtain

  κ−n.t.  c 
Re ∂νL u, (∇hE u)∂Ω
E
  c  
= T ∗ M (νg, h)T M ∂νL u, Sym(−L; νg )−1 ∂νL u + Re ∂νL u, (τh u)c E
E
 
= T ∗ M (νg, h)T M Sym(−L; νg )−1 ∂νL u, (∂νL u)c
E
 
+ Re ∂νL u, (τh u)c E . (1.8.105)

Plugging (1.8.102) and (1.8.105) back in (1.8.59) and canceling like-terms (one
being the first term in the right side of (1.8.102) and one associated with the first
term in the right side of (1.8.105)), finally proves the following general Rellich-type
identity:

Theorem 1.8.8 Retain the assumptions made in Theorem 1.8.7 on the n-dimensional
Riemannian manifold (M, g), the two Hermitian vector bundles E, F → M, the first
order differential operator D : E → F , the linear map A ∈ Hom (E, F ), and
41 reminiscent of the Pythagorean Theorem, according to which the square of the norm of a vector
is the sum of the squares of the norms of its normal and tangential parts
256 1 Integral Representations and Integral Identities

assume that
L := −D∗ AD is weakly elliptic, (1.8.106)
in the sense that its principal symbol Sym(L; ξ) is an invertible linear map for each
ξ ∈ T ∗ M \ {0}.
Next, consider a nonempty, open, proper subset Ω of M such that ∂Ω is a lower
Ahlfors regular set, and σg := Hgn−1 ∂Ω is a doubling measure on ∂Ω. Denote by
νg : ∂∗ Ω → T ∗ M geometric measure theoretic outward unit conormal of the set
of locally finite perimeter Ω, and denote by Lgn the measure induced by the volume
element on M. Finally, consider a section
2,1
u ∈ Wloc (Ω) ⊗ E (1.8.107)

which, for some aperture parameter κ > 0, satisfies the properties listed in (1.8.57),
and pick a vector field h ∈ 𝒞1 (M, T M) with real-valued components.
Then, with the tangential first-order differential operators τD and τh defined as
in (1.8.93) and (1.8.104), respectively, and with the conormal derivative ∂νL defined
as in (1.8.58), the following Rellich-type identity holds:
∫  
−1 L
T ∗ M (νg, h)T M Sym(−L; νg ) ∂ν u, (∂ν u)
L c
dσg
∂∗ Ω E
∫
 
= T ∗ M (νg, h)T M AτD u, (τD u)c F
dσg
∂∗ Ω
∫
 
− 2 Re ∂νL u, (τh u)c E
dσg
∂∗ Ω
∫ ∫
    
+ 2 Re Lu, (∇hE u)c E
dLgn + O |Du| 2 [∇hF, A] dLgn
Ω Ω
∫ ∫
     
+ O |Du| 2 divg h dL n + g O | A||Du| [D, ∇h ]u dLgn
Ω Ω
∫
 
+ O | A||h||Du| 2 dLgn, (1.8.108)
Ω

where the superscript ‘c’ indicates complex conjugation, ∇ F is a connection on the

vector bundle F , and all the big O’s involve only constants depending on the metrics.
It should be noted that the same comments as for the previous theorem apply to
Theorem 1.8.8 as well.
Next we discuss a number of consequences of the Rellich-type identities (1.8.59),
(1.8.108). These are particularly useful when the vector field h is transverse to ∂Ω,
i.e., when
essinf ∂∗ Ω T ∗ M (νg, h)T M > 0. (1.8.109)
From [62] we know that the existence of a continuous transverse vector field to ∂Ω
is equivalent to Ω being a Lipschitz domain. Other crucial hypotheses have to do
1.8 Rellich Identities 257

with the (semi or strict) positive definiteness of the homomorphism A, and the strong
ellipticity of the second-order operator L.

Corollary 1.8.9 Retain the hypotheses made in Theorem 1.8.7. Strengthen the as-
sumptions on the set Ω by now assuming that this is a Lipschitz subdomain of
the manifold M, and make the assumption that the self-adjoint homomorphism
A ∈ Hom (F , F ) is also uniformly strictly positive definite on ∂Ω in the sense that
there exists some c > 0 such that

A(x)ζ, ζ c  Fx ≥ c|ζ | 2Fx for all x ∈ ∂Ω and ζ ∈ Fx . (1.8.110)

Fix some scalar function V ∈ L ∞ (M, Lgn ) along with an aperture parameter
κ > 0, and consider a section u : Ω → E which satisfies
2,1
u ∈ Wloc (Ω) ⊗ E, (L − V)u = 0 in Ω,
Nκ u ∈ L 2 (∂Ω, σg ), Nκ (∇ E u) ∈ L 2 (∂Ω, σg ), (1.8.111)
κ−n.t. κ−n.t.
u∂Ω and (∇ E u)∂Ω exist σg -a.e. on ∂Ω.

Then there exists a constant C ∈ (0, ∞), independent of u and, for each ε > 0,
there exists a constant Cε ∈ (0, ∞) which depends on ε but is independent of u, with
the property that
∫ ∫ ∫
 κ−n.t. 2   E  κ−n.t. 2
(Du)  dσg ≤ Cε |∂ L 2
u| dσ + ε  ∇ u   dσg
∂Ω ν g ∂Ω
∂Ω ∂Ω ∂Ω
∫
 
+C |u| 2 + |∇ E u| 2 dLgn . (1.8.112)
Ω

In the opposite direction, (1.8.96) implies that we also have

∫ ∫
 κ−n.t. 2
|∂ν u| dσg ≤ C
L 2 (Du)  dσg . (1.8.113)
∂Ω
∂Ω ∂Ω

Together with (1.8.112) this shows that, in a uniformly fashion for sections u satis-
fying (1.8.111), there holds
∫ ∫
 κ−n.t. 2
|∂ν u| dσg ≈
L 2 (Du)  dσg
∂Ω
∂Ω ∂Ω (1.8.114)
modulo small and lower-order terms.

Let us now give the proof of Corollary 1.8.9.

Proof of Corollary 1.8.9 The inequality claimed in (1.8.112) is a direct consequence
of the Rellich-type identity (1.8.59) written for a real vector field h ∈ 𝒞1 (M, T M)
258 1 Integral Representations and Integral Identities

which is transverse to ∂Ω (hence (1.8.109) holds), and Cauchy’s inequality with

epsilon42. 
To state another application of a similar flavor, we first make a definition. Con-
p
cretely, given any function f ∈ L1 (∂Ω, σg ) ⊗ E with 1 ≤ p ≤ ∞, we define its
tangential gradient as the section

∇tan f : ∂Ω −→ T M ⊗ E (1.8.115)

locally expressed as

∇tan f := (ν  ) j g k ∂ ⊗ (∂τ j k f ) = νs g s j g k ∂ ⊗ (∂τ j k f ), (1.8.116)

where ∂τ j k are as in (1.8.86) (when acting on smooth functions). In this regard, see
also [113, §11.4] and [118, pp. 441-442]. Under the metric identification of T M with
T ∗ M, in place of (1.8.115)-(1.8.116) we may alternatively consider

∇tan f : ∂Ω −→ T ∗ M ⊗ E (1.8.117)

expressed in local coordinates as

∇tan f = (ν  ) j dxk ⊗ (∂τ j k f ) = νs g s j dxk ⊗ (∂τ j k f ). (1.8.118)

Thus, the elementary tangential differential operators ∂τ j k serve as building blocks

for ∇tan . Remarkably, it is also possible to extract the former from the latter. Indeed,
with the interpretation given in (1.8.118), for each a, b ∈ {1, . . . , n} we have the
following pointwise equality of sections in E:
   
T ∗ M ∇tan f , νa ∂b T M − T ∗ M ∇tan f , νb ∂a T M = ∂τa b f . (1.8.119)

To justify this, via density, it suffices to consider the case when f = ϕ|∂Ω with ϕ a
smooth section of the bundle E. In such a scenario, we may compute (mindful of
(1.8.86) and the fact that (ν  ) j ν j = 1)
   
T ∗ M ∇tan f , νa ∂b T M − T ∗ M ∇tan f , νb ∂a T M

= νa (ν  ) j (∂τ j b f ) − νb (ν  ) j (∂τ j a f )
   
= νa (ν  ) j ν j ∇∂b ϕ − νb ∇∂j ϕ − νb (ν  ) j ν j ∇∂a ϕ − νa ∇∂j ϕ

= νa ∇∂b ϕ − νb ∇∂a ϕ = ∂τa b f , (1.8.120)

as wanted.
We are now ready to state the result alluded to earlier.
Corollary 1.8.10 Preserve the notation and background hypotheses employed in
Theorem 1.8.8. This time, strengthen the assumptions by now assuming that Ω is a
ε
42 to the effect that ab ≤ 2 a2 + 1 2
2ε b for each a, b ∈ R and each ε > 0
1.8 Rellich Identities 259

Lipschitz subdomain of the manifold M, and impose the condition that the second-
order operator L is strongly elliptic at points on ∂Ω in a uniform fashion, i.e., there
exists some c > 0 such that for each x ∈ ∂Ω there holds
 
Re Sym(−L; ξ)η, η c ≥ c|ξ | 2 |η| 2, for all ξ ∈ Tx∗ M and η ∈ E x . (1.8.121)
Ex

Finally, some scalar function V ∈ L ∞ (M, Lgn ) along with an aperture parameter
κ > 0, and consider a section u : Ω → E satisfying
2,1
u ∈ Wloc (Ω) ⊗ E, (L − V)u = 0 in Ω,
Nκ u ∈ Nκ (∇ E u) ∈ L 2 (∂Ω, σg ),
L 2 (∂Ω, σg ), (1.8.122)
κ−n.t. κ−n.t.
u∂Ω and (∇ E u)∂Ω exist σg -a.e. on ∂Ω.

Then for each real vector field h ∈ 𝒞1 (M, T M) which is transverse to ∂Ω there
exists a constant C ∈ (0, ∞), independent of u, with the property that
∫ ∫
 
|∂νL u| 2 dσg ≤ C |τD u| 2 + |τh u| 2 dσg
∂Ω ∂Ω
∫
 
+C |u| 2 + |∇ E u| 2 dLgn . (1.8.123)
Ω

Moreover, there exists a constant C ∈ (0, ∞), independent of u, with the property
that
∫ ∫
 E κ−n.t. 2   κ−n.t.  2
(∇ u)  dσg ≤ C ∇tan u  dσg
∂Ω ∂Ω
∂Ω ∂Ω
∫
 
+C |u| 2 + |∇ E u| 2 dLgn, (1.8.124)
Ω

where ∇tan is the tangential gradient on ∂Ω.

Proof In the current setting, (1.8.123) is an immediate consequence of the Rellich-
type identity (1.8.108), Cauchy’s inequality with epsilon, and (1.8.121). As regards
(1.8.124), we first note that, through the freedom of choosing h arbitrary, (1.8.103)
implies 
 E κ−n.t.    κ−n.t.  
(∇ u)  ≤ C |∂ L u| + ∇tan u  (1.8.125)
∂Ω ν ∂Ω

at σg -a.e. point on ∂∗ Ω (cf. also (1.7.54) and [113, (11.3.26)], in this regard). Then
(1.8.124) is implied by (1.8.123) and (1.8.125). 
We want to briefly elaborate on the significance of Corollaries 1.8.9-1.8.10 from
the perspective of singular integral operators and boundary value problems. Work
in the setting adopted in Corollary 1.8.10. Make the additional assumption that

L − V : H 1 (M) ⊗ E −→ H −1 (M) ⊗ E is an isomorphism, (1.8.126)

260 1 Integral Representations and Integral Identities

and denote by E L (x, y) the Schwartz kernel of (L − V)−1 . Hence,

∫
 
(L − V)−1 w(x) = E L (x, y), w(y) E y dLgn (y), x ∈ M. (1.8.127)
M

Then E L (x, y) is a fundamental solution for the differential operator L − V whose

main singularity (along the diagonal) makes its first-order derivatives amenable to
the treatment of singular integral operators from §2.8. As done in Theorem 2.8.4,
use this to define the boundary-to-domain single layer potential operator 𝒮 acting
on each f ∈ L 2 (∂Ω, σg ) ⊗ E according to
∫
 
𝒮 f (x) := E L (x, y), f (y) E y dσg (y), x ∈ M \ ∂Ω, (1.8.128)
∂Ω

as well as its boundary-to-boundary version S, acting on each f ∈ L 2 (∂Ω, σg ) ⊗ E

according to
∫
 
S f (x) := E L (x, y), f (y) E y dσg (y), x ∈ ∂Ω. (1.8.129)
∂Ω

To proceed, abbreviate
Ω+ := Ω, Ω− := M \ Ω, (1.8.130)
and, having fixed an arbitrary f ∈ L 2 (∂Ω, σg ) ⊗ E, introduce

u± := 𝒮 f in Ω± . (1.8.131)

Then, thanks to Theorem 2.8.4, the sections u± satisfy all conditions listed in
(1.8.122) in Ω± and, in addition,
κ−n.t. κ−n.t.
at σg -a.e. point on ∂Ω we have u+ ∂Ω = S f = u− ∂Ω
  (1.8.132)
and τD u+ = τD u−, τh u+ = τh u−, ∂νL u± = ∓ 12 I + K L# f ,

where the conormal derivatives are defined as in (1.8.58) relative to Ω± (with νg

retaining its original meaning, i.e., the outward unit conormal to Ω = Ω+ ). The last
property in (1.8.132) comes from (2.8.235) and involves the principal-value singular
integral operator K L# on ∂Ω defined as in (2.8.227) corresponding to the factorization
of L given by (compare with (1.8.106))
' where D
L = DD ' := −D∗ A. (1.8.133)

In particular, the last property in (1.8.132) implies

∂νL u+ − ∂νL u− = − f . (1.8.134)

Based on these properties and (1.8.123) we then conclude that

# f # L 2 (∂Ω,σg )⊗ E ≤ C#S f # L 2 (∂Ω,σg )⊗ E + #Comp f # (1.8.135)

1
1.8 Rellich Identities 261

where Comp is a compact operator mapping L 2 (∂Ω, σg ) ⊗ E into a Banach space.

A word of clarification regarding the latter aspect is in order. In the present scenario,
the solid integral in (1.8.123) corresponding to Ω± in place of just Ω are simply
 2
𝒮 f  1,2 , so the presence of #Comp f # in (1.8.135) is attributable to the fact
W (Ω± )⊗ E
that

𝒮 : L 2 (∂Ω, σg ) ⊗ E −→ W 1,2 (Ω± ) ⊗ E are compact operators. (1.8.136)

In turn, (1.8.136) is a consequence of the fact that, as seen from (2.8.125),

2,2
𝒮 : L 2 (∂Ω, σg ) ⊗ E −→ B3/2 (Ω± ) ⊗ E are bounded operators (1.8.137)

and that, thanks to the Rellich-Kondrachov theorem, the embeddings

2,2
B3/2 (Ω± ) ⊗ E → W 1,2 (Ω± ) ⊗ E are compact. (1.8.138)

Pressing on, (1.8.135) implies that the linear and bounded operator (cf. (2.8.128))

S : L 2 (∂Ω, σg ) ⊗ E −→ L12 (∂Ω, σg ) ⊗ E (1.8.139)

has closed range and a finite dimensional kernel. In particular, the boundary-to-
boundary single layer S is a semi-Fredholm operator in the context of (1.8.139). To
determine its index, we take a continuous family Lt of second order, strongly elliptic
operators on M, indexed by t ∈ [0, 1], such that L1 = L and L0 is scalar. This gives
a norm continuous family

St : L 2 (∂Ω, σg ) ⊗ E −→ L12 (∂Ω, σg ) ⊗ E, all semi-Fredholm. (1.8.140)

Since S0 is Fredholm of index 0 (cf. [129]), so are all the operators St in (1.8.140)
and, in particular, S = S1 . In conclusion, we have the following basic result:

S : L 2 (∂Ω, σg ) ⊗ E → L12 (∂Ω, σg ) ⊗ E is Fredholm with index zero. (1.8.141)

Via duality, we also see that

2
S : L−1 (∂Ω, σg ) ⊗ E → L 2 (∂Ω, σg ) ⊗ E is Fredholm with index zero. (1.8.142)

Ultimately, these results imply the Fredholm solvability of the Dirichlet Problem and
Regularity Problem for the operator L in Lipschitz subdomains of the Riemannian
manifold M.
Moving on, work in the setting described in Theorem 1.8.8 strengthened by now
assuming that Ω is a UR domain, and make the additional assumption formulated
in (1.8.126). Having fixed some f ∈ L 2 (∂Ω, σg ) ⊗ E, define u± as in (1.8.131) and
write the Rellich-type identity (1.8.108) for u± in Ω± . Pick some number λ ∈ R,
then multiply the formula written for Ω+ by λ − 12 , multiply the formula written for
Ω− by λ + 12 and, finally, add the resulting identities. Keeping in mind that u± satisfy
both the conditions listed in (1.8.122) (written relative to Ω± ) and the properties
262 1 Integral Representations and Integral Identities

described in (1.8.132), after some algebra (involving canceling like-terms) we arrive

at the identity
∫  
 2 1 −1
λ −4 T ∗ M (νg, h)T M Sym(−L; νg ) f , f
c
dσg
∂Ω E
∫  
   
(νg, h)T M Sym(−L; νg )−1 λI + K L# f , λI + K L# f
c
= T∗M dσg
∂Ω E
∫
  c
− T ∗ M (νg, h)T M AτD (S f ), τD (S f ) dσg
∂Ω F
∫    c
+ 2 Re λI + K L# f , τh (S f ) dσg
∂∗ Ω E
∫
  2 
+ O |𝒮 f | 2 + ∇ E (𝒮 f ) dLgn . (1.8.143)
Ω±

Suppose next that Ω is actually a Lipschitz domain and that the real vector field h
is transverse to ∂Ω (in the sense of (1.8.109)). Also, suppose L is in fact a strongly
elliptic operator (i.e., (1.8.121) holds) and assume the self-adjoint homomorphism
A ∈ Hom (F , F ) is semi-positive definite on ∂Ω, in the sense that

A(x)ζ, ζ c  Fx ≥ 0 for all x ∈ ∂Ω and ζ ∈ Fx . (1.8.144)

Granted these hypotheses, from (1.8.143), Cauchy’s inequality with epsilon, and
(1.8.136) we obtain that for each λ ∈ R with |λ| > 12 there exists a constant
Cλ ∈ (0, ∞), independent of f , with the property that
  
# f # L 2 (∂Ω,σg )⊗ E ≤ Cλ  λI + K L# f  L 2 (∂Ω,σg )⊗ E + #Comp f # (1.8.145)

where Comp is a compact operator mapping L 2 (∂Ω, σg ) ⊗ E into a Banach space

(equipped with the norm # · #). As a consequence, the operator λI + K L# has closed
range and finite dimensional kernel, in particular, λI + K L# is semi-Fredholm on
L 2 (∂Ω, σg ) ⊗ E. Since λI + K L# is also invertible when |λ| is large enough, from
the homotopic invariance of the index (see item (8) of [113, Theorem 2.1.2]) we
ultimately conclude that the operator

λI + K L# : L 2 (∂Ω, σg ) ⊗ E −→ L 2 (∂Ω, σg ) ⊗ E
(1.8.146)
is Fredholm with index zero for each λ ∈ R with |λ| > 12 .

From (2.8.238) and the fact that we presently have L = L ∗ , conclude that

the (Hermitian) adjoint of K L# acting on L 2 (∂Ω, σg ) ⊗ E is the operator

K L , defined as in (2.8.226) relative to the factorization of L given in (1.8.147)
(1.8.133), also acting on the space L 2 (∂Ω, σg ) ⊗ E.
1.8 Rellich Identities 263

Bearing this in mind, via duality we also obtain that the operator

λI + K L : L 2 (∂Ω, σg ) ⊗ E −→ L 2 (∂Ω, σg ) ⊗ E
(1.8.148)
is Fredholm with index zero for each λ ∈ R with |λ| > 12 .

To be able to treat the important end-point cases λ = ± 12 , assume that the self-
adjoint homomorphism A ∈ Hom (F , F ) is uniformly strictly positive definite on
∂Ω (in the sense that (1.8.110) holds), and suppose there exists some C ∈ (0, ∞)
such that  
|∇ E w| ≤ C |Dw| + |w| in a pointwise sense,
1,1
(1.8.149)
for each section w ∈ Wloc (M) ⊗ E.
Once again, fix an arbitrary f ∈ L 2 (∂Ω, σg ) ⊗ E and define u± as in (1.8.131). Then
(1.8.149) implies
∫ ∫
  κ−n.t.  2  κ−n.t. 2
∇tan u±   dσg ≤ C (Du± )  dσg
∂Ω ∂Ω
∂Ω ∂Ω
∫
 κ−n.t. 2
+C u±   dσg, (1.8.150)
∂Ω
∂Ω

while from (1.8.112) and (1.8.149) we deduce that

∫ ∫
 κ−n.t. 2
(Du± )  dσg ≤ C |∂νL u± | 2 dσg
∂Ω
∂Ω ∂Ω
∫
 
+C |u± | 2 + |∇ E u± | 2 dLgn . (1.8.151)
Ω±

Together with the trace formulas in (1.8.132) and the mapping properties from
(1.8.136) and (1.8.139), these show that
∫ ∫
   1  
∇tan (S f )2 dσg ≤ C  ± I + K # f 2 dσg + #Comp f #, (1.8.152)
2 L
∂Ω ∂Ω

where Comp is a compact operator mapping L 2 (∂Ω, σg ) ⊗ E into a Banach space

(equipped with the norm # · #). In the opposite direction, from (1.8.123) written for
u± and (1.8.136) we see that
∫ ∫
 1    
 ± I + K # f 2 dσg ≤ C ∇tan (S f )2 dσg + #Comp f #. (1.8.153)
2 L
∂Ω ∂Ω

Combining (1.8.152) with (1.8.153) and using the triangle inequality then proves
that, in the current scenario, (1.8.145) also holds for λ = ± 12 , i.e., we have
  
# f # L 2 (∂Ω,σg )⊗ E ≤ C  ± 12 I + K L# f  L 2 (∂Ω,σg )⊗ E + #Comp f # (1.8.154)
264 1 Integral Representations and Integral Identities

for each f ∈ L 2 (∂Ω, σg ) ⊗ E. As such, in place of (1.8.146)-(1.8.148) we now obtain

λI + K L# , λI + K L : L 2 (∂Ω, σg ) ⊗ E −→ L 2 (∂Ω, σg ) ⊗ E
(1.8.155)
are Fredholm operators with index zero for all λ ∈ R with |λ| ≥ 12 .

More can be said when (1.8.145), (1.8.154), and (1.8.141) can be used in tandem.
To be specific, (1.8.141) implies (see item (7) in [113, Theorem 2.1.2]) that there
exists a linear and bounded operator

T : L12 (∂Ω, σg ) ⊗ E −→ L 2 (∂Ω, σg ) ⊗ E (1.8.156)

such that, with I denoting the identity operator,

ST − I is compact from L12 (∂Ω, σg ) ⊗ E into itself,

(1.8.157)
T S − I is compact from L 2 (∂Ω, σg ) ⊗ E into itself.

Based on these, the fact that S intertwines K L and K L# (cf. Corollary 2.8.13), and
(1.8.145) together with (1.8.154), for each λ ∈ R with |λ| ≥ 12 we may then estimate
(with Comp denoting generic compact operators)
  
# f # L 2 (∂Ω,σg )⊗ E =  ST − Comp f  L 2 (∂Ω,σg )⊗ E
1 1

≤ #S(T f )# L 2 (∂Ω,σg )⊗ E + #Comp f #

1

≤ C#T f # L 2 (∂Ω,σg )⊗ E + #Comp f #

 
≤ C  λI + K L# (T f )# L 2 (∂Ω,σg )⊗ E + #Comp f #
  
≤ C S λI + K L# (T f )# L 2 (∂Ω,σg )⊗ E + #Comp f #
1
 
= C  λI + K L (ST f )# L12 (∂Ω,σg )⊗ E + #Comp f #
 
≤ C  λI + K L f # L 2 (∂Ω,σg )⊗ E + #Comp f #, (1.8.158)
1

for each f ∈ L12 (∂Ω, σg ) ⊗ E. Also, bearing in mind (2.8.239), this ultimately proves
that
λI + K L : L12 (∂Ω, σg ) ⊗ E −→ L12 (∂Ω, σg ) ⊗ E
(1.8.159)
is Fredholm with index zero for each λ ∈ R with |λ| ≥ 12 .

Via duality (cf. (1.8.147)) we therefore obtain

λI + K L# : L−1
2 (∂Ω, σ ) ⊗ E −→ L 2 (∂Ω, σ ) ⊗ E
g −1 g
(1.8.160)
is Fredholm with index zero for each λ ∈ R with |λ| ≥ 12 .
1.8 Rellich Identities 265

In this vein, it is also useful to note that from (1.8.141), (1.8.142), (1.8.155),
(1.8.159), (1.8.160) and item (f) of [113, Theorem 2.2.4] we have the following
regularity results:
 
Ker S : L−12
(∂Ω, σg ) ⊗ E → L 2 (∂Ω, σg ) ⊗ E
 
= Ker S : L 2 (∂Ω, σg ) ⊗ E → L12 (∂Ω, σg ) ⊗ E (1.8.161)

and, for each λ ∈ R with |λ| ≥ 12 ,

   
Ker λI + K L# ; L−1
2
(∂Ω, σg ) ⊗ E = Ker λI + K L# ; L 2 (∂Ω, σg ) ⊗ E , (1.8.162)
   
Ker λI + K L ; L 2 (∂Ω, σg ) ⊗ E = Ker λI + K L ; L12 (∂Ω, σg ) ⊗ E . (1.8.163)

Also, since the quality of being Fredholm and the value of the index are stable on
complex interpolation scales (cf. [77, Theorem 2.9]), all results recorded in (1.8.141),
(1.8.142), (1.8.155), (1.8.159), (1.8.160) have versions in which the integrability
exponent 2 has been replaced by p ∈ (2 − ε, 2 + ε) for some small ε = ε(Ω) ∈ (0, 1).
In turn, the aforementioned L p -versions of (1.8.155), (1.8.159), (1.8.160) lead to
Fredholm solvability results for the Dirichlet Regularity, Neumann, and Transmission
Problems for the differential operator L in Lipschitz subdomains of the Riemannian
manifold M, wit data in Lebesgue and Sobolev spaces involving an integrability
exponent p near 2. Results allowing arbitrary p ∈ (1, ∞) are found in [64], in the
setting of Lipschitz domains with unit normals in VMO. This topic is re-visited
in [116, Chapter 8], where we further refine this work by allowing Ahlfors regular
domains which are sufficiently flat either globally, or at infinitesimal scales.
Chapter 2
Calderón-Zygmund Theory on Uniformly
Rectifiable Sets

Here we take on the task of methodically developing a Calderón-Zygmund theory

for singular integral operators on uniformly rectifiable sets which is applicable to
boundary layer potentials associated with elliptic systems both in the Euclidean set-
ting and in the setting of Riemannian manifolds. This body of results continues work
of many predecessors, such as Calderón, Zygmund, Mikhlin, Coifman, McIntosh,
Meyer, David, Jerison, Kenig, Semmes, Stein, among many others, and refines those
in [63] by placing more economical demands on the ambient geometry. An infor-
mative historical account of the formative years in the development of the classical
Calderón-Zygmund theory of singular integrals, which also touches on the personal
perspective of the author, has been written by E.M. Stein in [168]. Among many
aspects, it is particularly interesting to learn how this area of mathematics eventually
became main-stream, though this accepted was hard-earned. In this regard, Stein
writes on [168, p. 1134]: “I do recall that we, in the small group who were interested
in singular integrals then, felt a certain separateness from the larger community of
analysts - not that this isolation was self-imposed, but more because our subject
matter was seen by our colleagues as somewhat arcane, rarefied, and possibly not
very relevant.” In the early stages, this attitude was not limited to analysts in the US.
On page 143 of his autobiographical book [98], V. Maz’ya writes: “S.G. Mikhlin’s
highest accomplishment was creation of the theory of multidimensional singular
integral equations which was present in a large article, published already in the
1936 edition of Matematicheskiy Sbornik. [...] However, some influential Leningrad
experts1 on partial differential equations did not admit that Mikhlin’s subject area
was part of the ‘main trend’ and this attitude nagged the life out of him.”
To set the stage for subsequent work in this chapter, in §2.1 we consider integral
operators on Hölder spaces on upper Ahlfors regular sets, and in §2.2 we treat singular
integrals on Lebesgue spaces on Ahlfors regular quasi-metric spaces. Principal-value
singular integral operators on uniformly rectifiable sets are dealt with in §2.3, while in
§2.4 we consider boundary-to-domain integral operators on open sets with uniformly
rectifiable boundaries. A powerful and versatile jump-formula is then proved §2.5.

1 including O.A. Ladyzhenskaya (our note)

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 267
D. Mitrea et al., Geometric Harmonic Analysis III, Developments in Mathematics 74,
https://doi.org/10.1007/978-3-031-22735-6_2
268 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

This describes the nontangential trace of a boundary-to-domain integral operator in

an open set with a uniformly rectifiable boundary in terms of a jump-term (involving
the Fourier transform of the integral kernel function) and a principal-value singular
integral operator on the boundary (of the sort alluded to earlier). A number of
applications to integral operators whose kernels are derived from the fundamental
solutions of elliptic differential operators are presented in §2.9.
In §2.8, the last section of this chapter, we extend the scope of the Calderón-
Zygmund theory developed so far by considering singular integral operators, on
uniformly rectifiable sets, with variable coefficient kernels that are Schwartz kernels
of classical pseudodifferential operators of order −1 with odd principal symbol,
acting between two vector bundles over a given Riemannian manifold.

2.1 Integral Operators Acting on Hölder Spaces on Upper

Ahlfors Regular Sets

The main result pertaining to the action of integral operators on Hölder functions in
this section is the estimate presented in Proposition 2.1.3. Its proof is divided into
two parts, dealt with separately in Lemmas 2.1.1-2.1.2 below.

Lemma 2.1.1 Suppose Ω is an open subset of Rn such that ∂∗ Ω is bounded and

satisfies an upper Ahlfors regularity condition with constant c ∈ (0, ∞). In this
setting, define σ∗ := H n−1 ∂∗ Ω and consider an integral operator
∫
𝒯f (x) := k(x, y) f (y) dσ∗ (y), x ∈ Ω, (2.1.1)
∂∗ Ω

whose kernel k : Ω × ∂∗ Ω → R is such that k(x, ·) is a σ∗ -measurable function for

each x ∈ Ω, and has the property that there exists some finite positive constant C0
such that
C0
|k(x, y)| ≤ (2.1.2)
|x − y| n−1
for each x ∈ Ω and σ∗ -a.e. y ∈ ∂∗ Ω. Also, suppose the action of 𝒯 on the constant
function 1 is bounded in Ω, i.e.,
 
M := sup 𝒯1(x) < +∞. (2.1.3)
x ∈Ω

Then for every function f ∈ 𝒞α (∂∗ Ω) with α ∈ (0, 1) one has

 22(n−1+2α) 
sup |𝒯f (x)| ≤ c C0 2n−1+α + [diam(∂∗ Ω)]α  f 𝒞. α (∂ Ω)
x ∈Ω α ln 2 ∗

 
+ max M, c C0 · sup | f |. (2.1.4)
∂∗ Ω
2.1 Integral Operators Acting on Hölder Spaces on Upper Ahlfors Regular Sets 269

Proof Pick a function f ∈ 𝒞α (∂∗ Ω) and fix an arbitrary point x ∈ Ω. Consider

first the case when dist(x, ∂∗ Ω) ≥ diam(∂∗ Ω), in which scenario we may directly
estimate

|𝒯f (x)| ≤ C0 diam(∂∗ Ω)1−n σ∗ (∂∗ Ω) sup | f | ≤ c C0 sup | f |. (2.1.5)

∂∗ Ω ∂∗ Ω

In the remaining case, i.e., when dist(x, ∂∗ Ω) < diam(∂∗ Ω) < ∞, fix ε ∈ (0, 1)
arbitrary and select a point xε ∈ ∂∗ Ω such that

(1 + ε) dist(x, ∂∗ Ω) > |x − xε | =: r ∈ 0, diam(∂∗ Ω) . (2.1.6)

Next, split 𝒯f (x) = I + I I + I I I, where

∫
I := k(x, y) f (y) − f (xε ) dσ∗ (y), (2.1.7)
∂∗ Ω∩B(x ε ,2r)
∫
I I := k(x, y) f (y) − f (xε ) dσ∗ (y), (2.1.8)
∂∗ Ω\B(x ε ,2r)

and
I I I := (𝒯1)(x) f (x∗ ). (2.1.9)
Note that
∫
|I | ≤ |k(x, y)| | f (y) − f (xε )| dσ∗ (y)
∂∗ Ω∩B(x ε ,2r)
∫
|y − xε | α
≤ C0  f 𝒞. α (∂ Ω) dσ∗ (y)
∗
∂∗ Ω∩B(x ε ,2r) |x − y| n−1
(1 + ε)n−1 (2r)α 
≤ C0  f 𝒞. α (∂ Ω) σ∗ ∂∗ Ω ∩ B(xε, 2r) , (2.1.10)
∗ r n−1
where the third inequality comes from the fact that |y − xε | α < (2r)α on the domain
of integration, and the fact that 1/|x − y| ≤ (1 + ε)/|x − xε | = (1 + ε)/r, for all
y ∈ ∂∗ Ω. Hence,

|I | ≤ 2n−1+α c C0 (1 + ε)n−1  f 𝒞. α (∂ Ω), (2.1.11)

∗

bearing in mind (2.1.6) and the upper Ahlfors regularity of ∂∗ Ω. Also,

∫
  |y − xε | α
 I I  ≤ C0  f  . dσ∗ (y). (2.1.12)
𝒞α (∂∗ Ω)
∂∗ Ω\B(x ε ,2r) |x − y|
n−1

Note that if y ∈ ∂∗ Ω \ B(xε, 2r) then

|y − xε |
|y − xε | ≤ |y − x| + r and r ≤ implies |y − xε | ≤ 2|y − x|. (2.1.13)
2
270 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

From this and [112, Lemma 7.2.1] we therefore obtain

∫
1
|I I | ≤ 2n−1 C0  f 𝒞. α (∂ Ω) dσ∗ (y) (2.1.14)
∂∗ Ω\B(x ε ,2r) |y − xε |
∗ n−1−α

α
≤ 22(n−1+2α) c C0 (α ln 2)−1 diam(∂∗ Ω)  f 𝒞. α (∂ Ω) . (2.1.15)
∗

Clearly, |I I I | ≤ M · sup∂∗ Ω | f |, so the desired conclusion follows by combining the

individual estimates for I, I I, I I I and then letting ε → 0+ . 

The following is a companion result to Lemma 2.1.1.

Lemma 2.1.2 Let Ω be an open subset of Rn such that ∂∗ Ω satisfies an upper Ahlfors
regularity condition with constant c ∈ (0, ∞). Define σ∗ := H n−1 ∂∗ Ω and consider
the integral operator
∫
𝒬f (x) := q(x, y) f (y) dσ∗ (y), x ∈ Ω, (2.1.16)
∂∗ Ω

whose kernel q : Ω × ∂∗ Ω → R is such that q(x, ·) is a σ∗ -measurable function for

each x ∈ Ω, and has the property that for each x ∈ Ω and all σ∗ -a.e. y ∈ ∂∗ Ω one
has
C1
|q(x, y)| ≤ (2.1.17)
|x − y| n

for some finite positive constant C1 . Also, suppose there exists α ∈ (0, 1) with the
property that the action of 𝒬 on the constant function 1 satisfies
 
C2 := sup dist(x, ∂∗ Ω)1−α |(𝒬1)(x)| < +∞. (2.1.18)
x ∈Ω

Then for every function f ∈ 𝒞α (∂∗ Ω) one has

   23n−2 
sup dist(x, ∂∗ Ω)1−α |𝒬f (x)| ≤ c C1 2n−1+α +  f 𝒞. α (∂ Ω)
x ∈Ω (1 − α) ln 2 ∗

+ C2 · sup | f |. (2.1.19)
∂∗ Ω

Proof Select an arbitrary function f ∈ 𝒞α (∂∗ Ω). Fix some ε > 0 arbitrary, pick a
point x ∈ Ω, and choose xε ∈ ∂∗ Ω such that r := |x − xε | < (1 + ε) dist(x, ∂∗ Ω).
Split 𝒬f (x) = I + I I + I I I, where
∫
I := q(x, y) f (y) − f (xε ) dσ∗ (y), (2.1.20)
∂∗ Ω∩B(x ε ,2r)
∫
I I := q(x, y) f (y) − f (xε ) dσ∗ (y), (2.1.21)
∂∗ Ω\B(x ε ,2r)
2.1 Integral Operators Acting on Hölder Spaces on Upper Ahlfors Regular Sets 271

and
I I I := (𝒬1)(x) f (xε ). (2.1.22)
Then since r/(1 + ε) < dist(x, ∂∗ Ω) ≤ |x − y| for each y ∈ ∂∗ Ω, and r ≥ dist(x, ∂∗ Ω),
we may estimate
∫
|I | ≤ |q(x, y)| | f (y) − f (xε )| dσ∗ (y)
∂∗ Ω∩B(x ε ,2r)
∫
|y − xε | α
≤ C1  f  . α dσ∗ (y)
𝒞 (∂∗ Ω)
∂∗ Ω∩B(x ε ,2r) |x − y| n
(1 + ε)n (2r)α 
≤ C1  f 𝒞. α (∂ Ω) σ ∂∗ Ω ∩ B(xε, 2r)
∗ rn
≤ 2n−1+α c C1 (1 + ε)n dist(x, ∂∗ Ω)α−1  f 𝒞. α (∂ Ω) . (2.1.23)
∗

Next, keeping in mind that 1/|x − y| n ≤ 2n /|y − xε | n on ∂∗ Ω\ B(xε, 2r) (cf. (2.1.13)),
with the help of [112, Lemma 7.2.1] we may estimate
∫
|y − xε | α
|I I | ≤ C1  f 𝒞. α (∂ Ω) dσ∗ (y).
∂∗ Ω\B(x ε ,2r) |x − y|
∗ n
∫
1
≤ 2n C1  f 𝒞. α (∂ Ω) dσ∗ (y)
∂∗ Ω\B(x ε ,2r) |y − xε |
∗ n−α

23n−2
≤ c C1 dist(x, ∂∗ Ω)α−1  f 𝒞. α (∂ Ω) . (2.1.24)
(1 − α) ln 2 ∗

Given that dist(x, ∂∗ Ω)1−α |I I I | ≤ C2 · sup∂∗ Ω | f |, estimate (2.1.19) is established

after combining the bounds for I, I I, I I I and passing to limit as ε → 0+ . 

Combining Lemma 2.1.1 and Lemma 2.1.2 then yields the following result which
is useful in the treatment of the harmonic double layer potential operator, considered
later on.

Proposition 2.1.3 Let Ω be an open subset of Rn such that ∂∗ Ω is bounded and

satisfies an upper Ahlfors regularity condition with constant c ∈ (0, ∞). In this
setting, define σ∗ := H n−1 ∂∗ Ω and consider the integral operator
∫
T f (x) := K(x, y) f (y) dσ∗ (y), x ∈ Ω, (2.1.25)
∂∗ Ω

whose kernel K : Ω × ∂∗ Ω → R has the property that there exists a finite constant
B > 0 such that
B
|K(x, y)| + |x − y||(∇x K)(x, y)| ≤ (2.1.26)
|x − y| n−1

for each x ∈ Ω and σ∗ -a.e. y ∈ ∂∗ Ω. Fix some α ∈ (0, 1) and suppose that
272 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
    
A := sup (T 1)(x) + sup dist(x, ∂∗ Ω)1−α ∇(T 1)(x) < +∞. (2.1.27)
x ∈Ω x ∈Ω

Then there exists a finite constant Cn,α,Ω > 0, which depends only on n, α, c, and
diam(∂∗ Ω), with the property that for every function f ∈ 𝒞α (∂∗ Ω) one has
  
sup |T f (x)| + sup dist(x, ∂∗ Ω)1−α ∇(T f )(x)
x ∈Ω x ∈Ω

≤ Cn,α,Ω (A + B) f 𝒞α (∂∗ Ω) . (2.1.28)

Proof This is an immediate consequence of Lemma 2.1.1 and Lemma 2.1.2. 

We conclude by recording a compactness criterion for singular integral operators

with weakly singular kernels which is a particular case of a slightly more general
result proved on space of homogeneous type in [63, Lemma 2.20, p. 2608].

Lemma 2.1.4 Suppose Σ ⊂ Rn is compact and Ahlfors regular, and let k(·, ·) be a
complex-valued, Borel-measurable function on {(x, y) ∈ Σ × Σ : x  y} satisfying

ω(|x − y|)
|k(x, y)| ≤ for each x, y ∈ Σ with x  y, (2.1.29)
|x − y| n−1

where ω : (0, ∞) → (0, ∞) is non-decreasing, satisfies Dini’s integrability condition

∫ 1
ω(t)
dt < ∞, (2.1.30)
0 t
and is slowly varying, i.e., there exists a finite constant c > 0 such that

ω(2t) ≤ c ω(t), ∀t ∈ (0, ∞). (2.1.31)

In this setting, define σ := H n−1 Σ and consider the integral operator given by
∫
R f (x) := k(x, y) f (y) dσ(y), x ∈ Σ. (2.1.32)
Σ

Then for each p ∈ (1, ∞), the operator

R : L p (Σ, σ) −→ L p (Σ, σ) is compact (hence also bounded). (2.1.33)

2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces

The most general environment in which a Calderón-Zygmund theory may be devel-

oped, in a manner that retains much of its original force from the Euclidean setting,
is that of spaces of homogeneous type. In this section we review a number of results
pertaining to the nature of singular integral operators of Calderón-Zygmund type on
Ahlfors regular quasi-metric spaces. We begin by making the following definition.
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 273

Definition 2.2.1 Given d ∈ (0, ∞), call a triplet (X, ρ, μ) a d-Ahlfors regular
(quasi-metric) space (or simply, a d-AR space) if the pair (X, ρ) is a quasi-
metric space, μ is a non-negative measure defined on a sigma-algebra of subsets
of X which contains all ρ-balls and such that there exists CAR ∈ [1, ∞) with the
property that

(CAR )−1 r d ≤ μ Bρ (x, r) ≤ CAR r d, for all x ∈ X
 (2.2.1)
and all finite r ∈ 0, diamρ (X) .

Clearly, if (X, ρ, μ) is a d-Ahlfors regular quasi-metric space, then (X, ρ, μ) is also

a space of homogeneous type. In what follows, given a nonempty set X, we let

diag (X) := {(x, y) ∈ X × X : x = y} (2.2.2)

denote the diagonal in the Cartesian product X × X. If (X, ρ, μ) is a d-Ahlfors regular

quasi-metric space, it follows that the diagonal diag(X) is a μ ⊗ μ-measurable subset
of X × X, of zero measure.

Lemma 2.2.2 Let (X, ρ, μ) be a d-Ahlfors regular quasi-metric space. The diagonal
diag(X) is a μ ⊗ μ-measurable subset of X × X and

(μ ⊗ μ) diag(X) = 0. (2.2.3)

Proof For each ε > 0 introduce

 
Δε := (x, y) ∈ X × X : ρ# (x, y) < ε . (2.2.4)

From [112, (7.1.18)] it follows that the set Δε is open in τρ × τρ for each ε > 0.
Since
Δ1/j = diag(X), (2.2.5)
j ∈N

we conclude that diag(X) is a μ ⊗ μ-measurable subset of X × X.

To show that the diagonal in question has measure zero, we argue as follows. Fix
a small number ε ∈ (0, 1) and consider a subset Uε of X such that

ρ(x, y) ≥ ε for every x, y ∈ Uε, (2.2.6)

and
Uε is maximal (with respect to inclusion) among
(2.2.7)
all subsets of X satisfying property (2.2.6).
The maximality of Uε implies that

X= Bρ (x, ε). (2.2.8)

x ∈Uε

Also, from (2.2.6) we deduce that

274 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

 
the family Bρ x, ε/(2Cρ ) consists of mutually disjoint sets. (2.2.9)
x ∈Uε

To proceed, fix a point z ∈ X. We claim that2

 d
  2(Cρ )2 r + Cρ
# Bρ (z, r) ∩ Uε ≤ (CAR ) 2
, ∀r > 0. (2.2.10)
ε

Indeed, for each r > 0 we have

 
x ∈ Bρ (z, r) ∩ Uε =⇒ Bρ x, ε/(2Cρ ) ⊆ Bρ z, Cρ r + 1/2 (2.2.11)

which, in concert with (2.2.1) and (2.2.9), permits us to estimate

       
d
CAR Cρ r + 1/2 ≥ μ Bρ z, Cρ r + 1/2 ≥ μ Bρ x, ε/(2Cρ )
x ∈B ρ (z,r)∩Uε
  
≥ # Bρ (z, r) ∩ Uε (CAR )−1 ε/(2Cρ ) .
d
(2.2.12)

Then (2.2.10) readily follows from this. Note that, as a consequence of (2.2.10), the
set Uε is at most countable. Moving on, for z as above define
 
diag j (X) := diag(X) ∩ Bρ (z, j) × Bρ (z, j) , ∀ j ∈ N. (2.2.13)

From what we have proved so far, it follows that

each diag j (X) is a μ ⊗ μ-measurable subset of X × X. (2.2.14)

For each fixed j ∈ N we now claim that

diag j (X) ⊆ Bρ (x, ε) × Bρ (x, ε). (2.2.15)

x ∈B ρ (z,Cρ j)∩Uε

This is a consequence of (2.2.8) and the observation that Bρ (z, j) ∩ Bρ (x, ε)  

forces x ∈ Bρ (z, Cρ j). Having established (2.2.15), we now invoke (2.2.14), (2.2.1),
and (2.2.10), in order to estimate

2 with #(E) denoting the cardinality of the set E

2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 275
  
(μ ⊗ μ) diag j (X) ≤ (μ ⊗ μ) Bρ (x, ε) × Bρ (x, ε)
x ∈B ρ (z,Cρ j)∩Uε
  2

≤ μ Bρ (x, ε) ≤ (C AR )2 ε 2d
x ∈B ρ (z,Cρ j)∩Uε x ∈B ρ (z,Cρ j)∩Uε
 
= # Bρ (z, Cρ j) ∩ Uε (C AR )2 ε 2d
 d
≤ (CAR )4 2(Cρ )3 j + Cρ ε d (2.2.16)

for each fixed j ∈ N. Since ε ∈ (0, 1) may be taken to be arbitrarily small, this forces

(μ ⊗ μ) diag j (X) = 0 for each j ∈ N. (2.2.17)

Given that, as seen from (2.2.13),

diag(X) = diag j (X), (2.2.18)

j ∈N

we ultimately conclude from (2.2.17)-(2.2.18) that (2.2.3) holds. 

The point of the proposition below is that any Ahlfors regular quasi-metric space
is a reasonable topological and measure theoretic environment.

Proposition 2.2.3 Let (X, ρ, μ) be a d-Ahlfors regular quasi-metric space. Then3

(X, τρ ) is a separable topological space, (2.2.19)

and
μ is a Borel measure on (X, τρ ). (2.2.20)
Moreover4,

(X, τρ ) is a second countable topological space. (2.2.21)

In particular5
(X, τρ ) is a strongly Lindelöf space. (2.2.22)

Proof We begin by observing that, as a consequence of [6, Theorem 6.2] (cf. also
[117, Theorem 4.21, p. 184]), the sigma-algebra generated by all ρ-balls contains all
open sets in τρ . This renders μ a Borel measure. As a consequence of (2.2.1), we
have that μ is a doubling measure. In turn, this implies that (X, ρ) is geometrically
3 recall that a topological space is called separable if contains a countable dense subset
4 a topological space is said to be second countable if there exists some countable collection U
of open sets with the property that any open set may be expressed as a union of elements of some
subfamily of U
5 a strongly Lindelöf space is a topological space such that every open cover of any of its open
subsets has a countable sub-cover
276 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

doubling (cf. [112, Definition 7.5.1]), hence (2.2.19) holds by virtue of [117, (4.49),
p. 164]. From (2.2.19) and the fact that the topology induced by any quasi-metric is
metrizable (cf. [112, Theorem 7.1.2]), we see that (2.2.21) holds. Finally, (2.2.22)
follows from this, given that any second countable space is a strongly Lindelöf
space. 
We continue by making a basic definition.
Definition 2.2.4 Let (X, ρ, μ) be a d-Ahlfors regular quasi-metric space for some
d ∈ (0, ∞), and consider a function

K : X × X \ diag(X) −→ R. (2.2.23)

(i) Define the transpose kernel of K to be the function

K  : X × X \ diag(X) −→ R given by
(2.2.24)
K  (x, y) := K(y, x) ∀x, y ∈ X, with x  y.

(ii) Call K a standard kernel in the first variable if K is μ ⊗ μ-measurable

and there exist finite constants γ > 0, C0 ≥ 0, C1 > 0 such that
C0
|K(x, y)| ≤ , ∀x, y ∈ X, with x  y, (2.2.25)
ρ(x, y)d
and
ρ(x , x)γ
|K(x, y) − K(x , y)| ≤ C0 , ∀x, x , y ∈ X,
ρ(x , y)d+γ
(2.2.26)
satisfying y  {x, x  } as well as ρ(x , y) ≥ C1 ρ(x , x).

(iii) Call K a standard kernel in the second variable if K  is a standard

kernel in the first variable.
Our next proposition contains a refined version of Cotlar’s inequality.
Proposition 2.2.5 Assume (X, ρ, μ) is a d-Ahlfors regular quasi-metric space for
some d ∈ (0, ∞) with the property that the quasi-distance ρ : X × X → [0, +∞) is
continuous in the product topology τρ × τρ . Suppose

K : X × X \ diag(X) −→ R (2.2.27)

is a standard kernel in the first variable (i.e., K is μ ⊗ μ-measurable and (2.2.25)-

(2.2.26) are satisfied). Fix p ∈ [1, ∞) and assume that

T : L p (X, μ) −→ L p,∞ (X, μ) (2.2.28)

is an operator6 which satisfies the following properties:

6 not necessarily linear
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 277

(T1) there exists CT ∈ [0, ∞) such that for every f , g ∈ L p (X, μ) one has

|T( f + g)| ≤ CT |T f | + |T g| μ-a.e. in X; (2.2.29)

(T2) with the constant CT as above, for every f ∈ L p (X, μ) there holds

T f  L p,∞ (X,μ) ≤ CT  f  L p (X,μ) ; (2.2.30)

(T3) for each f ∈ L p (X, μ) one has7

∫
(T f )(x) = K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X \ supp f . (2.2.31)
X

Next, having fixed some reference point x0 ∈ X, for each ε > 0 and each function
 
μ(x)
f ∈ L 1 X, 1+ρ(x, x ) d (2.2.32)
0

define
∫
(Tε f )(x) := K(x, y) f (y) dμ(y), ∀x ∈ X. (2.2.33)
y ∈X, ρ(x,y)>ε

Finally, for each f as in (2.2.32), set

 
(Tmax f )(x) := sup (Tε f )(x) for every x ∈ X. (2.2.34)
ε>0

Then the following statements are true.

(i) If ℳ+ (X, μ) denotes the cone of μ-measurable functions defined on X with

values in [0, +∞], then the operator
 
μ(x)
Tmax : L 1 X, 1+ρ(x, x ) d −→ ℳ+ (X, μ) (2.2.35)
0

is well defined and sub-linear, in the sense that

 
μ(x)
Tmax ( f + g) ≤ Tmax f + Tmax g, ∀ f , g ∈ L 1 X, 1+ρ(x, x0 ) d
,
  (2.2.36)
μ(x)
Tmax (λ f ) = |λ| Tmax f , ∀ f ∈ L 1 X, 1+ρ(x, x )d
, ∀λ ∈ R.
0

 
μ(x)
As a consequence, for any two given functions f , g ∈ L 1 X, 1+ρ(x, d one has
x 0)

  
(Tmax f )(x) − (Tmax g)(x) ≤ Tmax ( f − g) (x) (2.2.37)

at each point x ∈ X where (Tmax f )(x) < +∞ and (Tmax g)(x) < +∞.

7 with supp f understood in the sense of [112, Definition 3.8.3]

278 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

(ii) For each r ∈ (0, ∞) there exists a constant C ∈ (0, ∞) which depends only on
ρ , CAR , CT , d, γ, p, and r, with the property that the following
C0 , C1 , Cρ , C
Cotlar-type inequality holds for every function f ∈ L p (X, μ) and every point
x ∈ X:
   1/r    1/p
(Tmax f )(x) ≤ C M X |T f | r (x) + C M X | f | p (x)

= CM X,r (T f )(x) + CM X, p ( f )(x), (2.2.38)

where M X is the ordinary Hardy-Littlewood maximal operator on the space of

homogeneous type (X, ρ, μ) and, for each s ∈ (0, ∞), the symbol M X,s denotes
the L s -based Hardy-Littlewood maximal operator on (X, ρ, μ) (cf. (A.0.83)).
(iii) The operator
Tmax : L p (X, μ) −→ L p,∞ (X, μ) (2.2.39)
is well defined, sub-linear8, bounded9, and continuous10. In fact, there exists
some constant C ∈ (0, ∞) with the property that
 
Tmax f − Tmax g  p,∞ ≤ C f − g L p (X,μ) for all f , g ∈ L p (X, μ).
L (X,μ)
(2.2.40)

As a consequence,

the operator Tε : L p (X, μ) −→ L p,∞ (X, μ) is

well-defined linear bounded for every ε > 0, (2.2.41)
and supε>0 Tε  L p (X,μ)→L p,∞ (X,μ) < +∞.

(iv) The operator

 
μ(x) p p,∞
Tmax : L 1 X, 1+ρ(x, x )d
∩ Lloc (X, μ) −→ Lloc (X, μ) (2.2.42)
0

is well defined, sub-linear, bounded, and continuous.

Proof We proceed in a series of steps, starting with the following claim.

 
Step I: Fix x ∈ X along with ε > 0 and f ∈ L 1 X, 1+ρ(·,μ x ) d . Then
0

8 in the sense that (2.2.36) holds

9 meaning that bounded sets are mapped into bounded sets
10 with respect to the topology induced by the respective (quasi-)norms
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 279
∫
|K(x, y)|| f (y)| dμ(y) ≤ C0 · Mρ,d,x0,ε,x ·  f  L 1  X, μ ,
y ∈X, ρ(x,y)≥ε 1+ρ(·, x0 ) d
     d
where Mρ,d,x0,ε,x := 2 max ε −1, Cρ max Cρ, ρ(x, x0 )/ε .
(2.2.43)

To justify this, first note that for each y ∈ X with ρ(x, y) ≥ ε we may estimate (recall
(A.0.25) and (A.0.26))
 
ρ(y, x0 ) ≤ Cρ max ρ(y, x), ρ(x, x0 )
 
≤ Cρ max C ρ, ρ(x, x0 )/ε ρ(x, y) (2.2.44)

hence
     d
1 + ρ(y, x0 )d ≤ 2 max 1, ρ(y, x0 )d = 2 max 1, ρ(y, x0 )

≤ Mρ,d,x0,ε,x · ρ(x, y)d . (2.2.45)

On account of (2.2.45) and (2.2.25) we may then write

∫
|K(x, y)|| f (y)| dμ(y)
y ∈X, ρ(x,y)≥ε
∫
| f (y)|
≤ C0 dμ(y)
y ∈X, ρ(x,y)≥ε ρ(x, y)d
∫
| f (y)|
≤ C0 · Mρ,d,x0,ε,x dμ(y), (2.2.46)
X 1 + ρ(y, x0 )d
as wanted.
 
μ(x)
Step II: Fix an arbitrary f ∈ L 1 X, 1+ρ(x, x0 ) d . Then O f := X \ supp f is an
open set in the topological space (X, τρ ), and the following function is well defined
and continuous:
∫
Φ f : O f → R, Φ f (x) := K(x, y) f (y) dμ(y) for each x ∈ O f . (2.2.47)
X

That O f is open in (X, τρ ) is known from [112, Lemma 3.8.4]. We shall show
that Φ f is a well-defined function which actually belongs to the local Hölder space
.γ 
𝒞loc O f , ρ (cf. (A.0.30)). In particular, this justifies the claim that Φ f is continuous
on O f . To this end, pick an arbitrary point x ∈ O f and note that (A.0.124) guarantees
the existence of some rx ∈ (0, ∞) with the property that Bρ (x, rx ) ⊆ O f , hence

supp f ⊆ X \ Bρ (x, rx ). (2.2.48)

280 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Since from [112, Lemma 3.8.4] we know that f = 0 at μ-a.e. point in X \ supp f , it
follows that
∫ ∫
|K(x, y)|| f (y)| dμ(y) = |K(x, y)|| f (y)| dμ(y) < +∞, (2.2.49)
X X\B ρ (x,r x )
 
μ(x)
thanks to (2.2.48), (2.2.43), and the fact that f ∈ L 1 X, 1+ρ(x, x0 ) d
. In turn, (2.2.49)
proves that for each x ∈ O f the expression Φ f (x) in (2.2.47) is defined via an
absolutely convergent integral. Thus, Φ f : O f → R is a well-defined function.
.γ 
To show that Φ f ∈ 𝒞loc O f , ρ , fix an arbitrary point x∗ ∈ O f . Recall that, by
design, rx∗ ∈ (0, ∞) has the property that supp f ⊆ X \ Bρ x∗, rx∗ . Observe that
  
∀x  ∈ Bρ x∗, rx∗ /Cρ =⇒ Bρ x , rx∗ /Cρ ⊆ Bρ x∗, rx∗ , (2.2.50)

hence
 
∀x  ∈ Bρ x∗, rx∗ /Cρ =⇒ supp f ⊆ X \ Bρ x , rx∗ /Cρ . (2.2.51)

Define
r x∗
ε∗ := ∈ (0, ∞). (2.2.52)
Cρ · C1

Such a choice ensures that


∀x, x , y ∈ X with ρ(x , x) < ε∗
=⇒ ρ(x , y) ≥ C1 ρ(x , x). (2.2.53)
and ρ(x , y) ≥ rx∗ /Cρ

Granted this, for every x  ∈ Bρ x∗, rx∗ /Cρ and each x ∈ Bρ (x , ε∗ ) we may estimate,
by relying on (2.2.26),
∫ 
  
|Φ f (x) − Φ f (x )| =  K(x, y) − K(x , y) f (y) dμ(y)
supp f
∫
 
≤ K(x, y) − K(x , y) | f (y)| dμ(y)
X\B ρ (x ,r x∗ /Cρ )
∫
| f (y)|
≤ C0 ρ(x , x)γ dμ(y)
X\B ρ (x ,r x∗ /Cρ ) ρ(x , y)d+γ
∫
| f (y)|
≤ Cr ho(x , x)γ dμ(y), (2.2.54)
X 1 + ρ(x , y)d

for some constant C ∈ (0, ∞) which is independent of x and x . Note that taking
 ε rx 
∗
R∗ := min , ∗ ∈ (0, ∞) (2.2.55)
ρ Cρ
Cρ C
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 281

implies

Bρ (x∗, R∗ ) ⊆ Bρ x∗, rx∗ /Cρ ⊆ O f , and
(2.2.56)
ρ(x , x) < ε∗ for every x, x  ∈ Bρ (x∗, R∗ ).

Granted these, we deduce from (2.2.54) that

 
Φ f (x) − Φ f (x )
sup < +∞, (2.2.57)
x,x  ∈B ρ (x∗,R∗ ) ρ(x , x)γ
xx 
.γ 
hence Φ f ∈ 𝒞loc O f , ρ , as wanted.
Step III: If for each f ∈ L p (X, μ) we define

T f in supp f ,
 f :=
T (2.2.58)
Φ f in O f ,

 maps L p (X, μ) into L p,∞ (X, μ). In fact, for every given function
then the operator T
f ∈ L (X, μ) one has
p

 f = T f at μ-a.e. point in X,
T (2.2.59)

and (compare with (2.2.31))

∫
 f )(x) =
(T K(x, y) f (y) dμ(y) for every x ∈ X \ supp f . (2.2.60)
X

All these claims are clear from definitions, Steps I-II, and (2.2.31).
Step IV: For each fixed ε > 0, the function

X × X  (x, y) −→ 1 {y ∈X, ρ(x,y)>ε } (y) ∈ R
(2.2.61)
is lower-semicontinuous, hence μ ⊗ μ-measurable.

To justify this claim, observe that for every number λ ∈ R the set of points in X × X
where the given function is > λ may be described as
⎧
⎪  if λ ≥ 1,
⎪
⎪
⎨
⎪ 
(x, y) ∈ X × X : ρ(x, y) > ε if λ ∈ [0, 1), (2.2.62)
⎪
⎪
⎪
⎪
⎩ X × X if λ < 0.
Thanks to [112, (7.1.18)], all sets appearing in (2.2.62) are open in τρ × τρ . This
proves that the function (2.2.61) is lower-semicontinuous.
282 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Step V: If Q+ denotes the collectionof all positive rational numbers, then for each
function f belonging to the space L 1 X, 1+ρ(·,μ x ) d we have
0

 
(Tmax f )(x) = sup (Tε f )(x) for every x ∈ X. (2.2.63)
ε ∈Q+

To justify this, pick some f ∈ L 1 X, 1+ρ(·,μ x ) d . We claim that if x ∈ X is
0
arbitrary and fixed then for each ε ∈ (0, ∞) and each sequence {ε j } j ∈N ⊆ (0, ∞)
such that ε j  ε as j → ∞ we have
∫ ∫
lim K(x, y) f (y) dμ(y) = K(x, y) f (y) dμ(y). (2.2.64)
j→∞ y ∈X, ρ(x,y)>ε j y ∈X, ρ(x,y)>ε

To justify (2.2.64) note that

{y ∈ X : ρ(x, y) > ε j }  {y ∈ X : ρ(x, y) > ε} as j → ∞, (2.2.65)

in the sense that

{y ∈ X : ρ(x, y) > ε} = {y ∈ X : ρ(x, y) > ε j } and

j ∈N (2.2.66)
{y ∈ X : ρ(x, y) > ε j } ⊆ {y ∈ X : ρ(x, y) > ε j+1 } for every j ∈ N.

Then (2.2.64) follows from (2.2.65), (2.2.43), and Lebesgue’s Dominated Conver-
gence Theorem.
 What we have just proved amounts to saying that for every function
f ∈ L 1 X, 1+ρ(·,μ x ) d we have
0

lim (Tε j f )(x) = (Tε f )(x) for every x ∈ X, (2.2.67)

j→∞

whenever ε ∈ (0, ∞) and {ε j } j ∈N ⊆ (0, ∞) are such that ε j  ε as j → ∞. Having

established this, (2.2.63) readily follows on account of the density of Q+ in (0, ∞).

Step VI: Given any f ∈ L 1 X, 1+ρ(·,μ x ) d , the function Tmax f is μ-measurable.
0
Indeed, granted (2.2.63) and since the supremum of some countable family of μ-
measurable functions is itself a μ-measurable function, it suffices to show that

Tε f is a μ-measurable function, for each fixed

 (2.2.68)
ε ∈ (0, ∞) and each fixed f ∈ L 1 X, 1+ρ(·,μ x ) d .
0


With this goal in mind, fix ε ∈ (0, ∞) along with f ∈ L 1 X, 1+ρ(·,μ x ) d , and for each
0
j ∈ N define

G j : X × X −→ R given at every point (x, y) ∈ X × X by

  (2.2.69)
G j (x, y) := 1Bρ (x0, j) (x)K(x, y) f (y) 1 {y ∈X, ρ(x,y)>ε } (y).
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 283

Then, thanks to assumptions, (2.2.61), and bearing in mind that ρ-balls are open, it
follows that G j is a μ ⊗ μ-measurable function for each j ∈ N. In addition, from
(2.2.69), (2.2.43), and since ρ-balls have finite measure, we see that
∫
|G j (x, y)| dμ(x) dμ(y) < ∞. (2.2.70)
X×X

Granted these properties, Fubini’s Theorem (whose applicability is ensured by the

fact that (X, μ) is a sigma-finite measure space) then guarantees that
∫
g j : X → R, g j (x) := G j (x, y) dμ(y), ∀x ∈ X,
X (2.2.71)
is a μ-measurable function, for each integer j ∈ N.

On the other hand, from (2.2.69), (2.2.71), and (2.2.33) it is apparent that for each
j ∈ N we have
g j = 1Bρ (x0, j) Tε f everywhere in X. (2.2.72)
In particular, this implies

lim g j = Tε f pointwise everywhere in X. (2.2.73)

j→∞

At this stage, the fact that Tε f is a μ-measurable function follows from (2.2.73) and
(2.2.71).
Step VII: Fix a point x∗ ∈ X and select a constant (whose relevance will become
clear shortly) 
A := 1 + C1 Cρ Cρ (Cρ )2 . (2.2.74)
Also, pick some ε > 0, along with some f ∈ Lloc 1 (X, μ). For these choices, define a

pair of functions, f1, f2 : X → R, by setting at each point x ∈ X


− f (x) if ρ(x∗, x) ≤ A ε,
f1 (x) := (2.2.75)
0 if ρ(x∗, x) > A ε,

and 
0 if ρ(x∗, x) ≤ A ε,
f2 (x) := (2.2.76)
f (x) if ρ(x∗, x) > A ε.
Then
f1, f2 ∈ Lloc
1 (X, μ), | f1 |, | f2 | ≤ | f | and f2 = f + f1 on X,
(2.2.77)
and also Bρ (x∗, ε) ∩ supp f2 = .

The properties in the first line of (2.2.77) are clear from definitions (upon recalling
that we are presently assuming that ρ : X × X → [0, +∞) is continuous in the
product topology τρ × τρ ). To justify the last property in (2.2.77), pick an arbitrary
point z ∈ Bρ (x∗, ε) and consider some point w ∈ Bρ (z, ε). Then (A.0.25)-(A.0.26)
284 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

and the choice of A imply

 
ρ ρ(x∗, w) ≤ Cρ C
ρ(w, x∗ ) ≤ C ρ max ρ(x∗, z), ρ(z, w)
 
ρ )2 max ρ(z, x∗ ), ρ(w, z) < A ε,
≤ Cρ (C (2.2.78)

proving that w ∈ Bρ (x∗, A ε). Thus, ultimately,

Bρ (z, ε) ⊆ Bρ (x∗, A ε) (2.2.79)

which goes to show that f2 = 0 everywhere on Bρ (z, ε). Having established this, the
last property in (2.2.77) follows from [112, Lemma 3.8.4].
Step VIII: Fix a point x∗ ∈ X and pick some ε > 0. With C1, A ∈ (0, ∞) as in
(2.2.26) and (2.2.74), respectively, we have
 
x, x  ∈ Bρ (x∗, ε) and ρ(x , y) ≥ C1 ρ(x , x),
=⇒ (2.2.80)
y ∈ X, ρ(x∗, y) > A ε ρ(x∗, y) ≤ Cρ Cρ ρ(x , y).

Indeed, if x, x , y are as in the left side of (2.2.80) then

 
ρ(x , x) ≤ Cρ max ρ(x , x∗ ), ρ(x∗, x)
 
≤ Cρ Cρ max ρ(x∗, x ), ρ(x∗, x) < Cρ C
ρ ε. (2.2.81)

Since ρ(x∗, y) > A ε by assumption, whereas ρ(x∗, x ) ≤ C ρ ρ(x , x∗ ) < C

ρ ε and

A ε > Cρ Cρ ε, the quasi-subadditive estimate
 
ρ(x∗, y) ≤ Cρ max ρ(x∗, x ), ρ(x , y) (2.2.82)

reduces simply to

ρ(x∗, y) ≤ Cρ ρ(x , y). (2.2.83)

Given that
ρ ρ(x , y),
ρ(x , y) ≤ C (2.2.84)

we may then conclude from (2.2.83) and assumptions that

ρ ρ(x , y).
A ε < Cρ C (2.2.85)

At this stage, the first estimate in the right side of (2.2.80) follows from (2.2.81),
(2.2.85), and (2.2.74), while the second estimate in the right side of (2.2.80) follows
from (2.2.83) and (2.2.84).
Step IX: Fix a point x∗ ∈ X and pick some ε > 0. Also, select f ∈ L p (X, μ) and
define f1, f2 as in (2.2.75)-(2.2.76). Then there exists a constant C ∈ (0, ∞) which
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 285

depends only on C0 , C1 , Cρ , C ρ , CAR , d, and γ, with the property that

  
(T  f2 )(x ) ≤ C M X f (x∗ ) for all x, x  ∈ Bρ (x∗, ε).
 f2 )(x) − (T (2.2.86)

To prove this, we begin by noting that, in light of (2.2.60), the properties listed in
(2.2.77) ensure that we may express
∫
 f2 )(x) =
(T K(x, y) f2 (y) dμ(y) (2.2.87)
X
∫
= K(x, y) f (y) dμ(y) for all x ∈ Bρ (x∗, ε).
y ∈X, ρ(x∗,y)> A ε

Consequently, for every x, x  ∈ Bρ (x∗, ε) we may estimate based on (2.2.80), (2.2.87)

(used both for x and x ), (2.2.26), and (2.2.81),
 
(T
 f2 )(x) − (T f2 )(x )
∫ 
  
= K(x, y) − K(x , y) f (y) dμ(y)

y ∈X, ρ(x∗,y)> A ε
∫
 
≤ K(x, y) − K(x , y) | f (y)| dμ(y)
y ∈X, ρ(x∗,y)> A ε
∫
ρ )d+2γ εγ | f (y)|
≤ C0 (Cρ C dμ(y)
y ∈X, ρ(x∗,y)> A ε ρ(x∗, y)d+γ
∞ ∫

ρ )d+2γ εγ | f (y)|
= C0 (Cρ C dμ(y)
j=0 2 j+1 A ε ≥ρ(x∗,y)>2 j A ε ρ(x∗, y)d+γ


∞ ∫
ρ )d+2γ εγ 1
≤ C0 (Cρ C d+γ  | f (y)| dμ(y)
2j A ε
j=0 2 j+1 A ε ≥ρ(x∗,y)

  
∞ μ Bρ x∗, 2 j+2 A ε ⨏
ρ )d+2γ εγ
≤ C0 (Cρ C  j | f (y)| dμ(y)
d+γ
j=0 2 Aε B ρ (x∗,2 j+2 A ε)

  
∞ μ Bρ x∗, 2 j+2 (Cρ )2 A ε

ρ )d+2γ εγ
≤ C0 (Cρ C  j M X f (x∗ )
d+γ
j=0 2 Aε

 
∞
ρ )d+2γ M X f (x∗ )
≤ 4d A−γ CAR C0 (Cρ )3d+2γ (C 2−jγ
j=0

= C M X f (x∗ ), (2.2.88)
286 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

where the last equality defines the constant C ∈ (0, ∞).

Step X: There exists a constant C ∈ (0, ∞) which depends only on C0 , C1 , Cρ , C ρ ,
CAR , and d, with the property that, with all objects involved retaining their earlier
significance, there holds
  
(T
 f2 )(x) − (Tε f )(x) ≤ C M X f (x∗ ) for every x ∈ Bρ (x∗, ε). (2.2.89)

To see that this is the case, observe first that for every x ∈ Bρ (x∗, ε) we have
∫
 f2 )(x) − (Tε f )(x) =
(T K(x, y) f (y) dμ(y)
y ∈X, ρ(x∗,y)> A ε
∫
− K(x, y) f (y) dμ(y)
y ∈X, ρ(x,y)>ε
∫
=− K(x, y) f (y) dμ(y). (2.2.90)
ρ(x,y)>ε, ρ(x∗,y)≤ A ε

Above, the first equality follows from (2.2.60) and (2.2.33). The second equality is
a consequence of the fact that

x ∈ Bρ (x∗, ε) and y ∈ X
=⇒ ρ(x, y) > ε. (2.2.91)
such that ρ(x∗, y) > A ε

Indeed, if x, y are as in the left side of (2.2.91), then the estimate

 
A ε < ρ(x∗, y) ≤ Cρ max ρ(x∗, x), ρ(x, y) (2.2.92)

reduces simply to

A ε < ρ(x∗, y) ≤ Cρ ρ(x, y), (2.2.93)

given that ρ(x∗, x) ≤ C ρ ρ(x, x∗ ) < Cρ ε and A > Cρ C ρ . Bearing in mind that

Cρ ≥ 1 (cf. (A.0.26)) forces A > Cρ , we then deduce from (2.2.93) that ρ(x, y) > ε.
This justifies (2.2.91) which, in turn, concludes the proof of (2.2.90).
Combining (2.2.90) with (2.2.25) then permits us to estimate, for every point
x ∈ Bρ (x∗, ε),
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 287
∫
  | f (y)|
(T
 f2 )(x) − (Tε f )(x) ≤ C0 dμ(y)
ρ(x,y)>ε, ρ(x∗,y)≤ A ε ρ(x, y)d
∫
−d
≤ C0 · ε | f (y)| dμ(y)
ρ(x∗,y)≤ A ε
∫
≤ C0 · ε −d | f (y)| dμ(y)
ρ A ε
ρ(y,x∗ )<2C

  ⨏
ρ A ε
≤ C0 · ε −d μ Bρ x∗, 2C | f (y)| dμ(y)
ρ A ε)
B ρ (x∗,2C

 
ρ A d M X f (x∗ ),
≤ C0 · CAR 2C (2.2.94)

ρ A
hence choosing C := C0 CAR 2C
d
finishes the proof of (2.2.89).
Step XI: Given any r ∈ (0, p), there exists a constant C ∈ (0, ∞) which depends
ρ , CAR , CT , d, p, and r, with the property that, if all objects involved
only on C1 , Cρ , C
retaining their earlier significance,
∫   r/p
 
ε −d  f1 )(x)r dμ(x) ≤ C M X (| f | p )(x∗ )
(T . (2.2.95)
B ρ (x∗,ε)

To justify this estimate, based on (2.2.59), [112, (6.2.37)], and (2.2.30), we write
∫ ∫
   
(T
 f1 )(x)r dμ(x) = (T f1 )(x)r dμ(x)
B ρ (x∗,ε) B ρ (x∗,ε)
 p    1− pr
≤ μ Bρ (x∗, ε) T f1  Lr p,∞ (X,μ)
p−r
 p  r r
≤ (CT )r (CAR )1− p ε d(1− p )  f1  Lr p (X,μ)
p−r
∫  pr
 p 
1− pr d(1− pr )
= (CT )r
(CAR ) ε | f (y)| dμ(y)
p
p−r y ∈X, ρ(x∗,y)≤ A ε

 p   dr
  pr
≤ (CT )r C 2C ρ A p
ε d M X (| f | p )(x∗ ) , (2.2.96)
p − r AR
from which the desired conclusion follows.
Step XII: The end-game in the proof of Proposition 2.2.5.
Proof of claims in part (i): For starters, observe that given any number ε > 0, any

point x ∈ X, and any function f ∈ L 1 X, 1+ρ(·,μ x ) d , the integral defining (Tε f )(x)
0
in (2.2.33) is absolutely convergent thanks to (2.2.43). In concert with (2.2.68), this
shows that each truncated operator Tε : L q (X, μ) → ℳ+ (X, μ) is well defined and
288 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

linear. From this and (2.2.34) we then conclude that the maximal operator Tmax is
well defined and sub-linear in the context of (2.2.35).
Proof of claims in part (ii): First assume some integrability exponent r ∈ (0, p) has
been given. Fix a function f ∈ L p (X, μ), some point x∗ ∈ X, and some number
ε > 0. With f1, f2 defined as in (2.2.75)-(2.2.76) we then have
       
(Tε f )(x ) ≤ (Tε f )(x ) − (T
 f2 )(x ) + (T  f2 )(x) + (T
 f2 )(x ) − (T  f2 )(x)
  
≤ C M X f (x∗ ) + (T
 f2 )(x) for every x, x  ∈ Bρ (x∗, ε), (2.2.97)

where the last inequality is based on (2.2.89) (written for x  in place of x) and
(2.2.86). Based on (2.2.59), (2.2.77), and (2.2.29), we may further write
         
T f2  = T f2  = T( f + f1 ) ≤ CT T f  + T f1 

   
= CT  T f  + 
T f1  at μ-a.e. point in X. (2.2.98)

Combining (2.2.97) with (2.2.98) then proves that

for every point x  belonging to Bρ (x∗, ε), the estimate

      
(Tε f )(x ) ≤ C M X f (x∗ ) + CT (T
 f )(x) + CT (T
 f1 )(x) (2.2.99)
holds for μ-a.e. x ∈ Bρ (x∗, ε).

Raise both sides of the estimate in (2.2.99) to the power r, then average in the variable
x over the ball Bρ (x∗, ε). In concert with (2.2.1), [112, (7.1.21)], and (A.0.84), this
leads to the conclusion that for every point x  ∈ Bρ (x∗, ε) we have
   
(Tε f )(x )r ≤ C M X f (x∗ )r + CM X |T  f | r (x∗ )
∫
 
+ Cε −d  f1 )(x)r dμ(x),
(T (2.2.100)
B ρ (x∗,ε)

where C ∈ (0, ∞) depends only on C0 , C1 , Cρ , C ρ , CAR , CT , d, γ, p, and r. By

combining (2.2.100) with (2.2.95) we therefore arrive at the conclusion that
   
(Tε f )(x )r ≤ C M X f (x∗ )r + CM X |T
 f | r (x∗ )
   r/p
+ C M X | f | p (x∗ ) for every x  ∈ Bρ (x∗, ε), (2.2.101)

with the constant C ∈ (0, ∞) having the same nature as above. Since Hölder’s
inequality (recall that p ∈ [1, ∞)) gives
    1/p
M X f (x∗ ) ≤ M X | f | p (x∗ ) , (2.2.102)
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 289

by keeping (2.2.59) in mind we may re-write (2.2.101) simply as

      r/p
(Tε f )(x )r ≤ CM X |T f | r (x∗ ) + C M X | f | p (x∗ )

for every point x  ∈ Bρ (x∗, ε), (2.2.103)

where C ∈ (0, ∞) has the same nature as before. Specializing (2.2.103) to the case
when x  := x∗ , then taking the supremum over ε ∈ (0, ∞) establishes (2.2.38)
(written with x∗ in place of x) in the case when r ∈ (0, p). Finally, this version of
(2.2.38) self-improves to also allow r ∈ [p, ∞). Indeed, having fixed r ∈ [p, ∞), with
r0 ∈ (0, p) Hölder’s inequality gives
   1/r0    1/r
M X |T f | r0 (x) ≤ M X |T f | r (x) , ∀x ∈ X. (2.2.104)

Proof of claims in part (iii): Given an arbitrary f ∈ L p (X, μ), the function Tmax f is
μ-measurable and non-negative, thanks to (2.2.35). Granted this, we may then invoke
(2.2.38) for some r ∈ (0, p), [112, (6.2.16)], [112, (6.2.19)], [112, Corollary 7.6.3],
and (2.2.30), in order to estimate (bearing in mind that our present choice of r entails
p/r ∈ (1, ∞), and that | f | p belongs to L 1 (X, μ))
   1/r     1/p 
   
Tmax f  L p,∞ (X,μ) ≤ C  M X |T f | r  + C MX | f | p 
L p,∞ (X,μ) L p,∞ (X,μ)
  1/r   1/p
   
= C M X |T f | r  p/r,∞ + C M X | f | p  1,∞
L (X,μ) L (X,μ)
 1/r  1/p
≤ C  |T f | r  p/r,∞
L (X,μ)
+ C| f | p  1
L (X,μ)

= CT f  L p,∞ (X,μ) + C f  L p (X,μ)

≤ C f  L p (X,μ), (2.2.105)

for some constant C ∈ (0, ∞) which depends only on C0 , C1 , Cρ , C ρ , CAR , CT , d,

γ, p, and r. This concludes the proof of the fact that the operator (2.2.39) is well
defined, sub-linear, and bounded. Granted this, the claims in (2.2.41) readily follow
in view of (2.2.34). Next, (2.2.40) is proved observing that (2.2.37), [112, (6.2.16)],
and (2.2.105) imply the existence of a constant C ∈ (0, ∞) such that, as desired,
   
Tmax f − Tmax g  p,∞ ≤ Tmax ( f − g) L p,∞ (X,μ)
L (X,μ)

≤ C f − g L p (X,μ) (2.2.106)

for every f , g ∈ L p (X, μ). In particular, this shows that Tmax is also continuous in
the context of (2.2.39).
290 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Proof of claims in part (iv): Fix a sufficiently large constant C∗ ∈ (0, ∞), depending
only on ρ, and pick some radius R ∈ (0, ∞). For an arbitrarily chosen function
 μ(x)
f ∈ L 1 X, 1+ρ(x,x )d
, define
o

 

gR := Tmax f 1X\Bρ (xo,C∗ R)  and
  ρ (xo,R)
X∩B
(2.2.107)

hR := Tmax f 1Bρ (xo,C∗ R)  .
X∩B ρ (x o ,R)

Note that there exists CR ∈ (0, ∞) with the property that for each x ∈ Bρ (xo, R) we
have
∫
 
|K(x, y)|  f 1X\Bρ (xo,C∗ R) (y) dμ(y)
X
∫
| f (y)|
≤ CR dμ(y)
X\B ρ (x o ,C∗ R) ρ(x, y)d
∫
| f (y)|
≤ CR dμ(y) < +∞. (2.2.108)
X 1 + ρ(x, xo )d
Granted this, it follows that

gR ∈ L ∞ Bρ (xo, R), μ and
(2.2.109)
gR  L ∞ (Bρ (xo,R),μ) ≤ CR  f  1  μ(x) .
L X,
1+ρ(x,xo ) d

From [112, (6.2.27)] we know that the following continuous embeddings hold:
  
L ∞ Bρ (xo, R), μ → L p Bρ (xo, R), μ → L p,∞ Bρ (xo, R), μ . (2.2.110)

In concert with (2.2.109), this shows that


gR ∈ L p,∞ Bρ (xo, R), μ and
(2.2.111)
gR  L p,∞ (Bρ (xo,R),μ) ≤ CR  f  1  μ(x) .
L X,
1+ρ(x,xo ) d

Next, since

f 1Bρ (xo,C∗ R) ∈ L p (X, μ) with

     (2.2.112)
 f 1B  = f  p ,
ρ (x o ,C∗ R) p
L (X,μ) B ρ (x o ,C∗ R) L (B ρ (x o ,C∗ R),μ)

we conclude from (2.2.41) that


T f 1Bρ (xo,C∗ R) ∈ L p,∞ (X, μ) and we have
      (2.2.113)
T f 1B (x ,C R)  p,∞ ≤ C  f Bρ (xo,C∗ R)  L p (Bρ (xo,C∗ R),μ) .
ρ o ∗ L (X,μ)
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 291

Based on (2.2.113) and the second line in [112, (6.2.17)] we deduce that

hR ∈ L p,∞ Bρ (xo, R), μ and
   (2.2.114)
hR  L p,∞ (Bρ (xo,R),μ) ≤ C  f Bρ (xo,C∗ R)  L p (Bρ (xo,C∗ R),μ) .

In view of the fact that, thanks to what we have already proved in item (i),

Tmax f is a μ-measurable function on X satisfying

 (2.2.115)
0 ≤ (Tmax f )Bρ (xo,R) ≤ gR + hR on Bρ (xo, R),

we ultimately conclude from (2.2.111), (2.2.114), and (2.2.115) that the operator
Tmax in (2.2.42) is indeed well defined, sub-linear, bounded, and continuous. 

There are a number of other related versions of Cotlar’s inequality presented in

Proposition 2.2.5, which are of interest. One such version, valid for a larger class of
functions, is discussed in the corollary below.

Corollary 2.2.6 Assume (X, ρ, μ) is a d-Ahlfors regular quasi-metric space for some
d ∈ (0, ∞) with the property that the quasi-distance ρ : X × X → [0, +∞) is
continuous in the product topology τρ × τρ . Demote by ℳ(X, μ) the space of all μ-
measurable functions defined on X, and by ℳ+ (X, μ) the collection of non-negative
functions in ℳ(X, μ). Suppose K : X × X \ diag(X) → R is a standard kernel in the
first variable (i.e., K is μ ⊗ μ-measurable and (2.2.25)-(2.2.26) are satisfied). Fix a
reference point x0 ∈ X and assume that
 
μ(x)
T : L 1 X, 1+ρ(x, x )d
−→ ℳ(X, μ) (2.2.116)
0

is an operator11 which satisfies the following properties:

 
μ(x)
(T1) there exists CT ∈ [0, ∞) such that for every f , g ∈ L 1 X, 1+ρ(x, d one has
x 0)


|T( f + g)| ≤ CT |T f | + |T g| μ-a.e. in X; (2.2.117)
 
μ(x)
(T2) with the constant CT as above, for every f ∈ L 1 (X, μ) ⊆ L 1 X, 1+ρ(x, x0 ) d
there holds
T f  L 1,∞ (X,μ) ≤ CT  f  L 1 (X,μ) ; (2.2.118)
 
μ(x)
(T3) for each f ∈ L 1 X, 1+ρ(x, x )d
one has12
0

∫
(T f )(x) = K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X \ supp f . (2.2.119)
X

11 again, not necessarily linear

12 with supp f understood in the sense of [112, Definition 3.8.3]
292 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 
μ(x)
Finally, for each f ∈ L 1 X, 1+ρ(x, x )d
set
0

∫ 
 
(Tmax f )(x) := sup  K(x, y) f (y) dμ(y) for every x ∈ X.
ε>0 y ∈X, ρ(x,y)>ε
(2.2.120)
Then for each r ∈ (0, ∞) there exists a constant C ∈ (0, ∞) which depends
ρ , CAR , CT , d, γ, and r, with the property that the following
only on C0 , C1 , Cρ , C
Cotlar-type inequality holds:
  1/r 
(Tmax f )(x) ≤ C M X (|T f | r ) (x) + C M X f (x)
  (2.2.121)
μ(x)
for every f ∈ L 1 X, 1+ρ(x, x )d
and every x ∈ X,
0

where M X is the Hardy-Littlewood maximal operator on the space of homogeneous

type (X, ρ, μ).

Proof We retrace the steps taken in the proof of Proposition 2.2.5. Specifically,
Steps I-II and Steps IV-VIII go through unchanged, and in place of Step III we now
have
 
μ(x)
New Step III: If for each f ∈ L 1 X, 1+ρ(x, x )d
we define
0


T f in supp f ,
 f :=
T (2.2.122)
Φ f in O f ,
 
then T maps L 1 X, μ(x) d into ℳ(X, μ). In fact, for every given function
 1+ρ(x, x0 )
μ(x)
f ∈ L 1 X, 1+ρ(x, x )d
one has
0

 f = T f at μ-a.e. point in X,
T (2.2.123)

and (compare with (2.2.119))

∫
 f )(x) =
(T K(x, y) f (y) dμ(y) for every x ∈ X \ supp f . (2.2.124)
X

Indeed, all these claims are clear from definitions, Steps I-II, and (2.2.119). With
this alteration, the old Steps IX-X now go through unchanged. Here are the new
formulations of the remaining steps, namely Steps XI-XII:
New Step XI: Fix  a point x∗ ∈ X and pick some ε > 0. Also, select an arbitrary
μ(x)
function f ∈ L 1 X, 1+ρ(x, x0 ) d
and define f1, f2 as in (2.2.75)-(2.2.76). Then for
each r ∈ (0, 1) there exists a constant C ∈ (0, ∞) which depends only on C1 , Cρ ,
Cρ , CAR , CT , d, and r, with the property that
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 293
∫  r
 
ε −d (T
 f1 )(x)r dμ(x) ≤ C M X f (x∗ ) . (2.2.125)
B ρ (x∗,ε)

Indeed, thanks to (2.2.123), [112, (6.2.37)], the fact that (X, ρ, μ) is a d-Ahlfors
regular quasi-metric space, and (2.2.118), we may estimate
∫ ∫
   
ε −d (T
 f1 )(x)r dμ(x) = ε −d (T f1 )(x)r dμ(x)
B ρ (x∗,ε) B ρ (x∗,ε)
 1    1−r
≤ ε −d μ Bρ (x∗, ε) T f1  Lr 1,∞ (X,μ)
1−r
 1 
≤ (CT )r (CAR )1−r ε −dr  f1  Lr 1 (X,μ)
1−r
∫ r
 1 
= (CT )r (CAR )1−r ε −dr | f (y)| dμ(y)
1−r y ∈X, ρ(x∗,y)≤ A ε
 1    r
≤ (CT )r ρ A
CAR 2C
dr
M X f (x∗ ) , (2.2.126)
1−r
as wanted.
New Step  XII: Theend-game in the proof of (2.2.121). Pick an arbitrary function
μ(x)
f ∈ L X, 1+ρ(x,
1
x0 ) d
, some point x∗ ∈ X, some number ε > 0, and define f1, f2
as in (2.2.75)-(2.2.76). Also, recall the truncated singular integral operator Tε from
(2.2.33). First, fix some integrability exponent r ∈ (0, 1). Then the same argument
as in (2.2.97)-(2.2.100) leads to the conclusion that for every point x  ∈ Bρ (x∗, ε) we
have
   r 
(Tε f )(x )r ≤ C M X f (x∗ ) + CM X |T  f | r (x∗ )
∫
 
+ Cε −d (T
 f1 )(x)r dμ(x), (2.2.127)
B ρ (x∗,ε)

where C ∈ (0, ∞) depends only on C0 , C1 , Cρ , C ρ , CAR , CT , d, γ, and r. By

combining (2.2.127) with (2.2.125) and keeping (2.2.123) in mind we therefore
obtain
   r 
(Tε f )(x )r ≤ C M X f (x∗ ) + CM X |T f | r (x∗ )
(2.2.128)
for all x  ∈ Bρ (x∗, ε),

with the constant C ∈ (0, ∞) having the same nature as above. Specializing (2.2.128)
to the case when x  := x∗ , then taking the supremum over ε ∈ (0, ∞) establishes
(2.2.121) (written with x∗ in place of x) in the case when r ∈ (0, 1). Finally, this
version of (2.2.121) self-improves as to also allow r ∈ [1, ∞). Indeed, having fixed
r ∈ [1, ∞), choose r0 ∈ (0, 1) and apply Hölder’s inequality to write
294 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
   1/r0    1/r
M X |T f | r0 (x) ≤ M X |T f | r (x) , ∀x ∈ X, (2.2.129)

from which the desired conclusion follows. 

For operators which are actually linear, the version of Cotlar’s inequality contained
in Proposition 2.2.5 takes a more streamlined form. It is also of interest to assume
boundedness at two different levels. Here is a result of this flavor.
Proposition 2.2.7 Suppose (X, ρ, μ) is a d-Ahlfors regular quasi-metric space for
some d ∈ (0, ∞) with the property that the quasi-distance ρ : X × X → [0, +∞) is
continuous in the product topology τρ × τρ . Assume

K : X × X \ diag(X) −→ R (2.2.130)

is a standard kernel in the first variable (that is, K is μ ⊗ μ-measurable and (2.2.25)-
(2.2.26) are satisfied). Fix p ∈ [1, ∞) and assume that an operator

T : L p (X, μ) −→ L p,∞ (X, μ), linear and bounded, (2.2.131)

has been given, with the property that (2.2.31) holds for each f ∈ L p (X, μ). In
addition, assume that
there exists some q ∈ (p, ∞) with the property that the re-
striction of T to L q (X, μ) ∩ L p (X, μ) extends to a linear and (2.2.132)
bounded operator from the space L q (X, μ) into itself.
For each ε > 0 define Tε as in (2.2.32)-(2.2.33).
Then the maximal operator Tmax , defined as in (2.2.34), induces a sub-linear,
bounded, and continuous mapping in each of the following contexts:

Tmax : L p (X, μ) −→ L p,∞ (X, μ), (2.2.133)

Tmax : L r (X, μ) −→ L r (X, μ) for each r ∈ (p, q], (2.2.134)

and, for each fixed reference point x0 ∈ X,

 
μ(x)
Tmax : L 1 X, 1+ρ(x, x )d
∩ Lloc
r (X, μ) −→ L r (X, μ)
loc
0 (2.2.135)
for each integrability exponent r ∈ (p, q].

As a consequence of (2.2.133)-(2.2.134), for each ε > 0 the linear operator Tε

is bounded in the following contexts:

Tε : L p (X, μ) −→ L p,∞ (X, μ), and (2.2.136)

Tε : L r (X, μ) −→ L r (X, μ) for each r ∈ (p, q]. (2.2.137)

In addition,
sup Tε  L p (X,μ)→L p,∞ (X,μ) < ∞, (2.2.138)
ε>0
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 295

and
sup Tε  L r (X,μ)→L r (X,μ) < ∞ for each r ∈ (p, q]. (2.2.139)
ε>0

As is apparent from the proof of Proposition 2.2.7 given below, if in place of

(2.2.132) one assumes a weak-(q, q) bound, i.e.,

there exists some q ∈ (p, ∞) with the property that the re-
striction of T to L q (X, μ) ∩ L p (X, μ) extends to a linear and (2.2.140)
bounded operator from L q (X, μ) into L q,∞ (X, μ),
then in place of (2.2.133)-(2.2.134) one concludes that

Tmax : L r (X, μ) −→ L r (X, μ) for each r ∈ (p, q), as well as (2.2.141)

Tmax : L p (X, μ) → L p,∞ (X, μ) and Tmax : L q (X, μ) → L q,∞ (X, μ) (2.2.142)

are well-defined, bounded, continuous, sub-linear operators, and so is Tmax in the

context of (2.2.135) with r ∈ (p, q), i.e.,
 
μ(x)
Tmax : L 1 X, 1+ρ(x, x0 ) d
∩ Lloc
r
(X, μ) −→ Lloc
r
(X, μ)
(2.2.143)
is well-defined, bounded, continuous, for each r ∈ (p, r).

Here is the actual proof of Proposition 2.2.7.

Proof of Proposition 2.2.7 That Tmax induces a sub-linear, bounded, and continuous
mapping in the context of (2.2.133) follows directly
 from Proposition 2.2.5. Assume
now that q ∈ (p, ∞) is an exponent such that T  L q (X,μ)∩L p (X,μ) extends as a linear
and bounded operator from L q (X, μ) into itself. Such an extension is unique (cf.
[112, (3.1.13)]) and, without ambiguity, we may continue to denote it by T. Hence,
in addition to (2.2.131), we also have

T : L q (X, μ) −→ L q (X, μ), linear and bounded. (2.2.144)

We claim that the analogue of hypothesis (T3) (appearing in the statement of

Proposition 2.2.5) presently holds for functions in L q (X, μ). Specifically, we claim
that
for each function f ∈ L q (X, μ) one has
∫ (2.2.145)
(T f )(x) = K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X \ supp f .
X

To justify this claim, fix some function f ∈ L q (X, μ) along with an arbitrary point
x  belonging to X \ supp f . Since the latter set is open, there exists r0 > 0 such that

Bρ (x , r0 ) ⊆ X \ supp f . (2.2.146)
296 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

From [112, (3.1.13)] we know that there exists a sequence of functions { 

f j } j ∈N in
q p 
L (X, μ) ∩ L (X, μ) with the property that f = lim f j in L (X, μ). Introducing
q
j→∞
f j := 
f j 1X\Bρ (x,r0 ) for each j ∈ N, then produces a sequence

{ f j } j ∈N ⊂ L q (X, μ) ∩ L p (X, μ) such that (2.2.147)

f = lim f j in L q (X, μ), and (2.2.148)

j→∞

Bρ (x , r0 ) ⊆ X \ supp f j for each j ∈ N. (2.2.149)

From (2.2.144) and (2.2.147)-(2.2.148) it follows that T f = lim T f j in L q (X, μ).

j→∞
By eventually passing to a subsequence, we may then write

(T f )(x) = lim (T f j )(x) for μ-a.e. x ∈ X. (2.2.150)

j→∞

For each j ∈ N fixed, the fact that T satisfies (2.2.31) for every function in L p (X, μ)
implies, in view of (2.2.147) and (2.2.149), that there exists a nullset N j of μ with
the property that
∫
(T f j )(x) = K(x, y) f j (y) dμ(y) for all x ∈ Bρ (x , r0 ) \ N j .
X\B ρ (x ,r0 )
(2.2.151)
 
Then N := N j is a nullset of μ, and for each x ∈ Bρ x , r0 /(2Cρ ) \ N we may
j ∈N
write
∫
lim (T f j )(x) = lim K(x, y) f j (y) dμ(y)
j→∞ j→∞ X\B ρ (x ,r0 )
∫
= K(x, y) f (y) dμ(y)
X\B ρ (x ,r0 )
∫
= K(x, y) f (y) dμ(y). (2.2.152)
X

Above, the first equality is a consequence of (2.2.151) and the definition of N. To

justify the second equality, we start by observing that

∀x, y ∈ X with ρ(x , y) ≥ r0
=⇒ ρ(x, y) ≥ r0 /Cρ . (2.2.153)
and ρ(x , x) ≤ r0 /(2Cρ )

Indeed,
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 297

ρ(x , y) ≤ Cρ max{ρ(x , x), ρ(x, y)}
=⇒ ρ(x , y) ≤ Cρ ρ(x, y) (2.2.154)
ρ(x , y) ≥ r0 and ρ(x , x) ≤ r0 /(2Cρ )

from which (2.2.153) follows. Having established (2.2.153),   if q ∈ (1, ∞) denotes
the Hölder conjugate exponent of q then for each x ∈ Bρ x , r0 /(2Cρ C ρ ) we may
use (2.2.25), Hölder’s inequality, and [112, (7.2.5)] to estimate, for some C ∈ (0, ∞)
independent of j,
∫
|K(x, y)|| f j (y) − f (y)| dμ(y)
X\B ρ (x ,r0 )
∫
| f j (y) − f (y)|
≤ C0 dμ(y)
y ∈X, ρ(x,y)≥r0 /Cρ ρ(x, y)d
∫  1/q
dμ(y)
≤ C0  f j − f  L q (X,μ) 
y ∈X, ρ(x,y)≥r0 /Cρ ρ(x, y)dq

≤ C(r0 /Cρ )−d/q  f − f j  L q (X,μ) . (2.2.155)

In concert with (2.2.148), the above estimate implies

∫
lim |K(x, y)|| f j (y) − f (y)| dμ(y) = 0 (2.2.156)
j→∞ X\B ρ (x ,r0 )

which, in turn, justifies the second equality in (2.2.152). Finally, the third equality in
(2.2.152) is a consequence of (2.2.146) and item (9) in [112, Lemma 3.8.4] (while
also mindful of (2.2.22)). At this stage, from (2.2.150) and (2.2.152) we conclude
(bearing in mind that N is a nullset of μ) that

for each point x  ∈ X \ supp f there exist some rx ∈ (0, ∞),
along with some nullset N(x ) of μ, with the property that
∫ (2.2.157)
(T f )(x) = K(x, y) f (y) dμ(y) for each
X 
ρ ) \ N(x ).
x ∈ Bρ x , rx /(2Cρ C

Invoking (2.2.22), we may now refine the open cover

  
ρ )
Bρ x , rx /(2Cρ C (2.2.158)
 x ∈X\supp f

of X \ supp f to a countable cover of this set, say

  
Bρ xi, ri /(2Cρ Cρ ) . (2.2.159)
i ∈N

Then N0 := N(xi ) is a nullset of μ, and
i ∈N
298 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫

(T f )(x) = K(x, y) f (y) dμ(y) for each x ∈ X \ supp f \ N0 . (2.2.160)
X

This finishes the proof of (2.2.145).

At this stage, we may conclude that T satisfies hypotheses (T1)-(T3) from the
statement of Proposition 2.2.5 with q in place of p. Consequently, Proposition 2.2.5
applies (with q in place of p) and gives that Tmax induces a sub-linear and bounded
mapping in the context

Tmax : L q (X, μ) −→ L q,∞ (X, μ). (2.2.161)

This is weaker than the conclusion we ultimately seek but, as we shall see momentar-
ily, is nonetheless useful. For now, we note that since the properties listed in (2.2.36)
imply   
(Tmax f )(x) − (Tmax g)(x) ≤ Tmax ( f − g) (x)
(2.2.162)
for every f , g ∈ L q (X, μ) and every x ∈ X,
from the boundedness of (2.2.161) as well as [112, (6.2.16)] we conclude that
   
Tmax f − Tmax g  q,∞ ≤ Tmax ( f − g) L q,∞ (X,μ)
L (X,μ)

≤ C f − g L q (X,μ) (2.2.163)

for every f , g ∈ L q (X, μ). In particular, this shows that Tmax is also continuous in the
context of (2.2.161).
To proceed, fix an arbitrary function f ∈ L q (X, μ) ∩ L p (X, μ). Since the original
assumptions on the operator T imply that the hypotheses (T1)-(T3) from Proposi-
tion 2.2.5 are satisfied for the given exponent p, and since f belongs to the space
L p (X, μ), we may write the pointwise inequality (2.2.38). Taking L q norms, much as
in (2.2.105) we may then estimate, for some r ∈ (0, q) and some constant C ∈ (0, ∞)
independent of f ,
   1/r     1/p 
   
Tmax f  L q (X,μ) ≤ C  M X |T f | r  + C MX | f | p 
L q (X,μ) L q (X,μ)
  1/r   1/p
   
= C M X |T f | r  q/r + C M X | f | p  q/p
L (X,μ) L (X,μ)
 1/r  1/p
≤ C  |T f | r  L q/r (X,μ) + C  | f | p  L q/p (X,μ)

= CT f  L q (X,μ) + C f  L q (X,μ)

≤ C f  L q (X,μ), (2.2.164)

where the last step uses (2.2.144). Pick now an arbitrary f ∈ L q (X, μ). From [112,
(3.1.13)] we know that there exists a sequence { f j } j ∈N ⊂ L q (X, μ) ∩ L p (X, μ) such
that
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 299

lim f j = f in L q (X, μ). (2.2.165)

j→∞

Now (2.2.162) implies that |Tmax f j − Tmax fk | ≤ Tmax ( f j − fk ) pointwise on X, for

every j, k ∈ N. In concert with the estimate obtained in (2.2.164) (written for the
difference f j − fk ∈ L q (X, μ) ∩ L p (X, μ) in place of f ) this implies
   
Tmax f j − Tmax fk  q ≤ Tmax ( f j − fk ) L q (X,μ)
L (X,μ)

≤ C f j − fk  L q (X,μ), ∀ j, k ∈ N. (2.2.166)

This proves that {Tmax f j } j ∈N is a Cauchy sequence in the Banach space L q (X, μ).
As such,

g := lim Tmax f j exists in L q (X, μ). (2.2.167)

j→∞

Having established this, pass to limit k → ∞ in (2.2.166) to obtain

Tmax f j − g L q (X,μ) ≤ C f j − f  L q (X,μ), ∀ j ∈ N. (2.2.168)

Based on (2.2.168) and (2.2.164) (written for f j in place of f ), for each j ∈ N we

may estimate

g L q (X,μ) ≤ Tmax f j − g L q (X,μ) + Tmax f j  L q (X,μ)

≤ C f j − f  L q (X,μ) + C f j  L q (X,μ) . (2.2.169)

By passing to limit j → ∞ we therefore arrive at the conclusion that, for some

constant C ∈ (0, ∞) independent of f ,

g L q (X,μ) ≤ C f  L q (X,μ) . (2.2.170)

We now claim that g = Tmax f at μ-a.e. point in X. To see this, observe that, thanks
to [112, (6.2.27)] and (2.2.167), we have g = lim Tmax f j in L q,∞ (X, μ). Since from
j→∞
(2.2.165) and the continuity of Tmax in (2.2.161) we also have

lim Tmax f j = Tmax f in L q,∞ (X, μ), (2.2.171)

j→∞

we ultimately conclude that g = Tmax f at μ-a.e. point in X. Utilizing this back in

(2.2.170) proves that for every f ∈ L q (X, μ) we have

Tmax f  L q (X,μ) ≤ C f  L q (X,μ), (2.2.172)

for some constant C ∈ (0, ∞) independent of f . This proves that Tmax induces a
sub-linear and bounded mapping in the context

Tmax : L q (X, μ) −→ L q (X, μ). (2.2.173)

300 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

From this and (2.2.162), we then conclude that there exists a constant C ∈ (0, ∞)
with the property that
   
Tmax f − Tmax g  q ≤ Tmax ( f − g) L q (X,μ)
L (X,μ)

≤ C f − g L q (X,μ) (2.2.174)

for every f , g ∈ L q (X, μ). In particular, Tmax in (2.2.173) is also continuous.

Having proved the boundedness of Tmax both in the context of (2.2.133) as well
as (2.2.173), the real interpolation result for quasi-subadditive operators recorded
in [113, Proposition 1.3.7] applies and, in concert with [112, (6.2.47)], leads to the
conclusion that Tmax further induces a sub-linear and bounded mapping in the context

Tmax : L r (X, μ) −→ L r (X, μ) for each r ∈ (p, q]. (2.2.175)

Parenthetically, we note that we could have arrived at this conclusion by first inter-
polating between (2.2.131) and (2.2.144) to obtain that T maps L r (X, μ) boundedly
into itself for each r ∈ (p, q], then run the same program as above starting from this
premise in lieu of (2.2.144).
Returning to the mainstream discussion, the boundedness of (2.2.175) then yields,
much as before via (2.2.162) (and the fact that L r (X, μ) is a quasi-normed lattice of
functions), that for each r ∈ (p, q] there exists C ∈ (0, ∞) with the property that
 
Tmax f − Tmax g  r,∞ ≤ C f − g L r (X,μ) for all f , g ∈ L r (X, μ). (2.2.176)
L (X,μ)

In particular, the mapping (2.2.175) is also continuous.

The claims pertaining to Tmax in the context of (2.2.135) are justified by reasoning
as in the proof of (2.2.42), with two important changes. First, in place of (2.2.111)
we now conclude from (2.2.109) that

gR ∈ L r Bρ (xo, R), μ and
(2.2.177)
gR  L r (Bρ (xo,R),μ) ≤ CR  f  1  μ(x) .
L X,
1+ρ(x,xo ) d

Second, since

f 1Bρ (xo,C∗ R) ∈ L r (X, μ) with

     (2.2.178)
 f 1B (x ,C R)  r =  f Bρ (xo,C∗ R)  L r (Bρ (xo,C∗ R),μ),
ρ o ∗ L (X,μ)

in place of (2.2.113)-(2.2.114) we deduce from (2.2.134) that


hR ∈ L r Bρ (xo, R), μ and
   (2.2.179)
hR  L r (Bρ (xo,R),μ) ≤ C  f Bρ (xo,C∗ R)  L r (Bρ (xo,C∗ R),μ) .
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 301

With (2.2.177) and (2.2.179) in hand, we then conclude as before that Tmax is a
well-defined, sub-linear, bounded, continuous operator in the context of (2.2.135).
Finally, the claims pertaining to (2.2.136)-(2.2.139) are direct consequences of
what we have proved so far, (2.2.34), and (2.2.68). 
We proceed to discuss a number of consequences of the Calderón-Zygmund
lemma. First, as Proposition 2.2.8 below shows, for a linear operator T being weak-
(q, q) bounded for some integrability exponent q ∈ (1, ∞) and being associated with
a standard kernel in the second variable extrapolates to strong-(p, p) bounds for T
for any p ∈ (1, q).

Proposition 2.2.8 Assume (X, ρ, μ) is a d-Ahlfors regular quasi-metric space for

some d ∈ (0, ∞) with the property that the quasi-distance ρ : X × X → [0, +∞) is
continuous in the product topology τρ × τρ . Suppose

K : X × X \ diag(X) −→ R (2.2.180)

is a standard kernel in the second variable (cf. Definition 2.2.4). Fix q ∈ (1, ∞) and
assume that an operator

T : L q (X, μ) −→ L q,∞ (X, μ), linear and bounded, (2.2.181)

has been given, with the property that T is associated in L q (X, μ) with the kernel
K, in the sense that for each f ∈ L q (X, μ) one has
∫
(T f )(x) = K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X \ supp f . (2.2.182)
X

Then the following statements are true:

(i) There exists C ∈ (0, ∞) with the property that

T f  L 1,∞ (X,μ) ≤ C f  L 1 (X,μ), ∀ f ∈ L q (X, μ) ∩ L 1 (X, μ). (2.2.183)

As a consequence, the restriction of T to L q (X, μ) ∩ L 1 (X, μ) extends to a linear

and bounded operator from L 1 (X, μ) into L 1,∞ (X, μ), henceforth, still denoted
by T. Thus,

T : L 1 (X, μ) −→ L 1,∞ (X, μ) linearly and boundedly. (2.2.184)

(ii) For each p ∈ (1, q), the operator T extends to a mapping

T : L p (X, μ) −→ L p (X, μ) linear and bounded. (2.2.185)

To set the stage for presenting the proof of Proposition 2.2.8 we state a lemma
containing an estimate for a Marcinkiewicz-type integral.

Lemma 2.2.9 Assume that (X, ρ, μ) is a d-Ahlfors regular space for some d > 0
with the property that the quasi-distance ρ : X × X → [0, +∞) is continuous in the
302 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

product topology τρ × τρ . Then for each γ > 0 there exists C ∈ (0, ∞) such that
whenever F is a nonempty closed subset of (X, τρ ) one has
∫ ∫
distρ (y, F )γ 
dμ(y) dμ(x) ≤ C μ X \ F . (2.2.186)
F X ρ(x, y)d+γ

Proof Since distρ (·, F ) vanishes on F and γ > 0, we may write based on Fubini’s
Theorem
∫ ∫ 
distρ (y, F )γ
dμ(y) dμ(x)
X ρ(x, y)
d+γ
F
∫ ∫ 
distρ (y, F )γ
= dμ(y) dμ(x)
X\F ρ(x, y)
d+γ
F
∫ ∫
γ dμ(x) 
= distρ (y, F ) dμ(y)
F ρ(x, y)
d+γ
X\F
∫ ∫
γ dμ(x) 
≤ distρ (y, F ) dμ(y)
X\B ρ (y,dist ρ (y, F)) ρ(x, y)
d+γ
X\F
∫

≤C dμ = C μ X \ F , (2.2.187)
X\F

where the last inequality is based on [112, (7.2.5)]. 

We are now ready to present the proof of Proposition 2.2.8.

Proof of Proposition 2.2.8 We shall show the existence of some C ∈ (0, ∞) with
the property that for each function f ∈ L 1 (X, μ) ∩ L q (X, μ) we have
   f  L 1 (X,μ)
μ x ∈ X : |(T f )(x)| > λ ≤C , ∀λ > 0. (2.2.188)
λ
To get started, assume some arbitrary function f ∈ L 1 (X, μ) ∩ L q (X, μ) has been
fixed. In the case when 0 < λ ≤  f  L 1 (X,μ) /μ(X) (which may only happen when
X is bounded), we have
   f  L 1 (X,μ)
μ x ∈ X : |(T f )(x)| > λ ≤ μ(X) ≤ , (2.2.189)
λ
so (2.2.188) holds in this case if we choose C ≥ 1.
Consider now the case when λ >  f  L 1 (X,μ) /μ(X). There is no loss of generality
in assuming that f has bounded support, and we shall perform a Calderón-Zygmund
decomposition of f at level λ. More precisely, there exist two finite constants C > 0,
N ∈ N (depending only on geometry), along with an at most countable family of
balls (Q j ) j ∈J , say Q j := Bρ (x j , r j ) for each index j ∈ J, and two real-valued
functions g, b : X → R satisfying the following properties (cf., e.g., [22]):
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 303

f = g + b at μ-a.e. point in X, (2.2.190)

g ∈ L 1 (X, μ) ∩ L ∞ (X, μ) → L q (X, μ) and b ∈ L 1 (X, μ) ∩ L q (X, μ), (2.2.191)

g L 1 (X,μ) ≤ C f  L 1 (X,μ), |g(x)| ≤ Cλ for μ-a.e. x ∈ X, (2.2.192)
 ⨏ 
if b j := f − f 1Q j for each j ∈ J, then (2.2.193)
Qj

b= b j with convergence both in L 1 (X, μ) and L q (X, μ), (2.2.194)
j ∈J

b j  L 1 (X,μ) ≤ C f  L 1 (X,μ), (2.2.195)
j ∈J
∫ ⨏
b j ≡ 0 in X \ Q j , b j dμ = 0, and |b j | dμ ≤ Cλ, ∀ j ∈ J, (2.2.196)
X Qj
 
if O := j ∈J Q j and F := X \ O, then 1 O ≤ j ∈J 1Q j ≤ N1 O,
(2.2.197)
μ(O) ≤ Cλ−1  f  L 1 (X,μ), and distρ (Q j , F ) ≈ r j uniformly in j ∈ J.

To prove that the series j ∈J b j converges in L q (X, μ) to b, there is no loss of
generality in assuming that J = N. Suppose this is the case and pick some ε > 0.
Given that | f | q dμ is a finite measure which is absolutely continuous with respect to
μ, it follows that there exists δ > 0 with the property that
∫
| f | q dμ < ε whenever E ⊆ X is μ-measurable and μ(E) < δ. (2.2.198)
E

Next, observe that 1 O ∈ L

 (X, μ) since O is open and of finite measure. Mindful
q

of this and the fact that j ∈N 1Q j ≤ N1 O , Lebesgue’s Dominated Convergence

Theorem implies that

m 
the sequence 1Q j is convergent in L q (X, μ). (2.2.199)
m
j=1

Consequently, the aforementioned sequence is Cauchy in the space L q (X, μ). Hence,
there exists some integer M = M(δ) ∈ N such that
∫  m+k q
 
 1Q j  dμ < δ for all m, k ∈ N with m ≥ M. (2.2.200)
X j=m

Upon noting that, pointwise on X we have


m+k
1Q j ≥ 1m+k Q j for all m, k ∈ N, (2.2.201)
j=m
j=m
304 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

we conclude from (2.2.200) that

 m+k 
μ Q j < δ for all m, k ∈ N with m ≥ M. (2.2.202)
j=m

To proceed, for each j ∈ N use the triangle inequality and Hölder’s inequality to
estimate
∫  ⨏  q
 
|b j | q dμ =  f 1Q j − f 1Q j  q
X Qj L (X,μ)

 q
   ⨏  
 
≤  f 1Q j  L q (X,μ) +  f 1Q j 
Qj L q (X,μ)

∫ ⨏ q
≤ 2q | f | q dμ + 2q μ(Q j ) f
Qj Qj
∫
≤ (2q + 1) | f | q dμ. (2.2.203)
Qj

Also, since for each subset I of N and for each point x ∈ X the cardinality of the set
{ j ∈ I : x ∈ Q j } is at most N, for every m, k ∈ N we may write

 m+k
 q   q 
|b j (x)| = |b j (x)| ≤ Nq |b j (x)| q
j=m {m≤ j ≤m+k: x ∈Q j } {m≤ j ≤m+k: x ∈Q j }


m+k
≤ Nq |b j (x)| q for μ-a.e. x ∈ X. (2.2.204)
j=m

Together, (2.2.203), (2.2.204), (2.2.202), and (2.2.198) then permit us to estimate

2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 305

 m+k  ∫  m+k q ∫
  q  m+k
 bj  q ≤ |b j | dμ ≤ N q
|b j | q dμ
L (X,μ) X X
j=m j=m j=m

∫
m+k
≤ N q (2q + 1) | f | q dμ
j=m Qj

∫  m+k
 
≤ N (2 + 1)
q q
 m+k
| f |q 1Q j dμ
j=m Qj j=m
∫
≤ N q+1 (2q + 1)  m+k
| f | q dμ
j=m Qj

≤ N q+1 (2q + 1)ε, for all m, k ∈ N with m ≥ M. (2.2.205)

Since the number ε > 0 has been arbitrarily chosen, the above estimate proves that the

series j ∈N b j converges in L q (X, μ). Since the same series is known to converge

in L 1 (X, μ) to the function b, we ultimately conclude that j ∈N b j converges in
L q (X, μ) to b.
Let us now return to the main stream discussion. Since

|(T f )(x)| ≤ |(T g)(x)| + |(T b)(x)| for μ-a.e. x ∈ X, (2.2.206)

as far as (2.2.188) is concerned it suffices to prove that

   f  L 1 (X,μ)
μ x ∈ X : |(T g)(x)| > λ/2 ≤C , (2.2.207)
λ
and
   f  L 1 (X,μ)
μ x ∈ X : |(T b)(x)| > λ/2 . ≤C (2.2.208)
λ
The boundedness of T in the context of (2.2.181) permits us to estimate
 
 g L q (X,μ) q
μ x ∈ X : |(T g)(x)| > λ/2 ≤ C
λ
q−1
g L ∞ (X,μ) g L 1 (X,μ)
≤C
λq
 f  L 1 (X,μ)
≤C , (2.2.209)
λ
thus (2.2.207) is established. We are therefore left with proving (2.2.208). To justify
this, first note that (2.2.197) directly implies
306 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
   f  L 1 (X,μ)
μ x ∈ O : |(T b)(x)| > λ/2 ≤ μ(O) ≤ C . (2.2.210)
λ
Second, Chebytcheff’s inequality gives
 ∫
 1
μ x ∈ F : |(T b)(x)| > λ/2 ≤ |T b| dμ. (2.2.211)
λ F

Therefore, since X = O ∪ F , in view of (2.2.210) and (2.2.211), estimate (2.2.208)

will follow as soon as we prove that
∫
|T b| dμ ≤ C f  L 1 (X,μ) (2.2.212)
F

for some C ∈ (0, ∞) independent of f . With this goal in mind, we fix j ∈ J and

x ∈ F arbitrary and look for a pointwise estimate for
∫
(T b j )(x) = K(x, y)b j (y) dμ(y). (2.2.213)
X

With x j and r j denoting, respectively, the center and radius of Q j , based on the third
condition in (2.2.195), we may write
∫  ∫
   
(T b j )(x) =   
K(x, y)b j (y) dμ(y) = 

K(x, y) − K(x, x j ) b j (y) dμ(y)
X X
∫ 
 
= K(x, y) − K(x, x j ) b j (y) dμ(y) ≤ I1 + I2, (2.2.214)
Qj

where we have set

∫
 
I1 := K(x, y) − K(x, x j ) |b j (y)| dμ(y), (2.2.215)
y∈Q j
ρ(y,x j )<C1 ρ(x,x j )

∫
 
I2 := K(x, y) − K(x, x j ) |b j (y)| dμ(y). (2.2.216)
y∈Q j
ρ(y,x j )≥C1 ρ(x,x j )

Note that ρ(y, x j ) < C1 ρ(x, x j ) on the domain of integration in I1 . Based on this,
the fact that the kernel K is standard in the second variable, and (2.2.196), we may
then estimate this term as follows
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 307
∫
ρ(y, x j )γ
I1 ≤ C0 |b j (y)| dμ(y)
Qj ρ(x, x j )d+γ

rj
γ ∫ rj
d+γ
≤C |b j (y)| dμ(y) ≤ Cλ
ρ(x, x j )d+γ Q j ρ(x, x j )d+γ
∫
distρ (y, F )γ
≤ Cλ dμ(y), (2.2.217)
Q j ρ(x, y)
d+γ

since distρ (y, F ) ≈ r j and ρ(x, x j ) ≈ ρ(x, y) uniformly for y ∈ Q j and x ∈ F .

Estimating I2 requires a few geometrical preliminaries. Recall that x ∈ F and
consider an arbitrary point y ∈ Q j such that ρ(y, x j ) ≥ C1 ρ(x, x j ). Observe that, on
the one hand,
ρ ρ(x, x j ),
r j ≈ distρ (Q j , F ) ≤ ρ(x j , x) ≤ C (2.2.218)
ρ ρ(x, y).
r j ≈ distρ (Q j , F ) ≤ ρ(y, x) ≤ C (2.2.219)

On the other hand, the fact that y ∈ Q j forces ρ(x j , y) < r j which in turn allows
ρ ρ(x j , y) < C −1 C
us to estimate ρ(x, x j ) ≤ C1−1 ρ(y, x j ) ≤ C1−1 C ρ r j . Together with
1
(2.2.218), this ultimately implies

r j ≈ ρ(x, x j ), uniformly in j ∈ J and x ∈ F . (2.2.220)

ρ, 1}r j which, in light

Also, ρ(x, y) ≤ Cρ max{ρ(x, x j ), ρ(x j , y)} ≤ Cρ max{C1−1 C
of (2.2.219), entails

r j ≈ ρ(x, y), uniformly in y ∈ Q j and x ∈ F . (2.2.221)

Consequently, on the domain of integration in I2 we have thanks to (2.2.220) and

(2.2.221)
     
K(x, y) − K(x, x j ) ≤ K(x, y) + K(x, x j )
γ
C0 C0 Cr j
≤ + ≤ . (2.2.222)
ρ(x, y)d ρ(x, x j )d ρ(x, x j )d+γ

Together with (2.2.216), this allows us to estimate

rj
γ ∫ rj
d+γ
I2 ≤ C |b j (y)| dμ(y) ≤ Cλ
ρ(x, x j )d+γ Qj ρ(x, x j )d+γ
∫
distρ (y, F )γ
≤ Cλ dμ(y). (2.2.223)
Qj ρ(x, y)d+γ
308 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

From (2.2.214), (2.2.217), and (2.2.223) we conclude that there exists C ∈ (0, ∞)
with the property that for every j ∈ J we have
∫
 
(T b j )(x) ≤ Cλ distρ (y, F )γ
dμ(y), ∀x ∈ F . (2.2.224)
Q j ρ(x, y)
d+γ

Summing up inequalities of the form (2.2.224) over j ∈ N and using the linearity of
the operator T as well as the finite-overlap property in (2.2.197), we obtain
∫
 
(T b)(x) ≤ Cλ distρ (y, F )γ
dμ(y), ∀x ∈ F . (2.2.225)
O ρ(x, y)
d+γ

Consequently, from (2.2.225), Lemma 2.2.9, and (2.2.197), we deduce that

∫ ∫ ∫
distρ (y, F )γ
|T b| dμ ≤ Cλ dμ(y) dμ(x)
F O ρ(x, y)
d+γ
F

∫ ∫
distρ (y, F )γ
≤ Cλ dμ(y) dμ(x)
F X ρ(x, y)d+γ

≤ Cλ μ X \ F = C λ μ(O) ≤ C f  L 1 (X,μ) . (2.2.226)

This proves (2.2.212), thus completing the proof of (2.2.188). From this, the claims
in item (i) follow.
Interpolating between (2.2.184) and (2.2.181) gives (in light of [112, (6.2.47)])

T : L r (X, μ) −→ L r (X, μ)
(2.2.227)
linearly and boundedly for each r ∈ (1, q).

This takes care of item (ii), and finishes the proof of the proposition. 

The reader is reminded that the space of Hölder functions with bounded support,
in the quasi-metric setting, has been defined in (A.0.33). The following result can be
thought of as a refinement of [112, Proposition 3.7.2]. Before stating it, the reader is
reminded that distρ (·, ·) is defined in (A.0.42).

Lemma  2.2.10 Suppose (X, ρ) is a quasi-metric space, and fix a finite exponent

β ∈ 0, (log2 Cρ )−1 . Assume μ is a locally finite Borel-semiregular measure on X
(cf. [112, Definition 3.4.3]). Also, let F ⊆ X be a closed set and consider a function

h : X \ F −→ R which is absolutely integrable with respect to μ on

(2.2.228)
any bounded closed subset E of X \ F satisfying distρ (F, E) > 0,

and with the property that

∫
β 
h φ dμ = 0 for each φ ∈ 𝒞c (X, ρ) with distρ F, supp φ > 0. (2.2.229)
X
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 309

Then h = 0 at μ-a.e. point in X \ F.

A particular case of interest is F = , a scenario in which we set distρ (, ·) := +∞.

Proof of Lemma 2.2.10 Fix an arbitrary point x in the open set X \ F. Then there
exists some small radius r > 0 such that Bρ# (x, r) ⊆ X \ F. Consider a closed
set E ⊆ Bρ# (x, r/4). From [117, Lemma 4.14, p. 166] we know that there exists a
β
sequence {φ j } j ∈N ⊆ 𝒞c (X, ρ) with the following properties:

supp φ j ⊆ Bρ# (x, r/2) ⊆ X \ F, 0 ≤ φ j ≤ 1 for all j ∈ N,

(2.2.230)
and φ j  1E pointwise everywhere on X, as j → ∞.

Note that 
distρ X \ Bρ# (x, r), Bρ# (x, r/2) > 0. (2.2.231)
Indeed, having y ∈ X \ Bρ# (x, r) and z ∈ Bρ# (x, r/2) forces ρ# (y, x) ≥ r as well as
ρ# (z, x) ≤ r/2. Granted these properties, we may then use [112, (7.1.19)-(7.1.20)]
to estimate
β β β β
r β ≤ ρ# (y, x) ≤ ρ# (y, z) + ρ# (z, x) ≤ ρ# (y, z) + (r/2)β (2.2.232)

which, in light of [112, (7.1.21)], further implies

ρ · ρ(y, z).
r(1 − 2−β )1/β ≤ ρ# (y, z) ≤ C (2.2.233)

As such,

distρ X \ Bρ# (x, r),Bρ# (x, r/2)
 
= inf ρ(y, z) : y ∈ X \ Bρ# (x, r) and z ∈ Bρ# (x, r/2)

ρ )−1 > 0,
≥ r(1 − 2−β )1/β (C (2.2.234)

justifying (2.2.231). In turn, this implies that


distρ F, supp φ j > 0 for each j ∈ N. (2.2.235)
∫
Based on (2.2.235), (2.2.228), and (2.2.229) we conclude that X h φ j dμ = 0 for
each j ∈ N. Since for each j ∈ N we have |h φ j | ≤ 1Bρ# (x,r/2) ∈ L 1 (X, μ) we may
invoke Lebesgue’s Dominated Convergence Theorem to write
∫ ∫
h dμ = lim h φ j dμ = 0. (2.2.236)
E j→∞ X

The argument so far proves that

∫
h dμ = 0 for each closed set E ⊆ Bρ# (x, r/4). (2.2.237)
E
310 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

To proceed, introduce
 
A± := x ∈ Bρ# (x, r/4) : ±h(x) ≥ 0 . (2.2.238)

Then A± are μ-measurable subsets of Bρ# (x, r/4) and μ(A± ) ≤ μ Bρ# (x, r/4) < ∞.
Since the measure μ is Borel-semiregular it follows that there exist two Borel sets
B± ⊆ X such that
μ(A± B± ) = 0. (2.2.239)
Replacing B± by B± ∩ Bρ# (x, r/4) (and keeping in mind that all ρ# -balls are open)
there is no loss of generality in assuming that we actually have B± ⊆ Bρ# (x, r/4).
Observe that, having fixed any reference point x0 ∈ X,
 
X= Bρ# (x0, j) and for each j ∈ N we have μ Bρ# (x0, j) < ∞
j ∈N (2.2.240)
and Bρ# (x0, j) is open hence, in particular, it belongs to Borelτ (X).

Granted (2.2.240) and [112, Lemma 3.4.13], we may then invoke item (1) in [112,
Proposition 3.4.15] to conclude that
 
μ(B± ) = sup μ(E± ) : E± closed subset of B± . (2.2.241)

Fix ε > 0 arbitrary. Since the 1Bρ# (x,r/4) |h| dμ is a finite measure which is absolutely
continuous with respect to μ, it follows that there exists δ > 0 with the property that
∫
|h| dμ < ε for each μ-measurable set A ⊆ X with μ(A) < δ.
A∩B ρ# (x,r/4)
(2.2.242)
For this δ, we may rely on (2.2.241) to pick two closed sets E± ⊆ B± with

μ(B± \ E± ) < δ. (2.2.243)

Then, thanks to (2.2.239) and (2.2.237), we may write

∫ ∫ ∫ ∫ ∫
|h| dμ = h dμ − h dμ = h dμ − h dμ
B ρ# (x,r/4) A+ A− B+ B−
∫ ∫  ∫ ∫ 
= h dμ + h dμ − h dμ + h dμ
B+ \E+ E+ B− \E− E−
∫ ∫
= h dμ − h dμ. (2.2.244)
B+ \E+ B− \E−

We may also rely on (2.2.242) and (2.2.243) to estimate

∫  ∫
 
 
 h dμ ≤ |h| dμ < ε. (2.2.245)
 B± \E±  B± \E±
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 311

Together, (2.2.244) and (2.2.245) prove that

∫
|h| dμ < 2ε. (2.2.246)
B ρ# (x,r/4)
∫
Since ε > 0 is arbitrary, this forces B (x,r/4) |h| dμ = 0 hence, further, h = 0 at
ρ#
μ-a.e. point in Bρ# (x, r/4). Given that the topological space (X, τρ ) is Lindelöf (cf.
Proposition 2.2.3), we ultimately conclude that h = 0 at μ-a.e. point in X \ F. 

A version of Proposition 2.2.8, for kernels which are standard in both variables,
is presented next.

Corollary 2.2.11 Assume (X, ρ, μ) is a d-Ahlfors regular quasi-metric space for

some d ∈ (0, ∞) with the property that the quasi-distance ρ : X × X → [0, +∞)
is continuous in the product topology τρ × τρ . Also, assume that the measure μ is
Borel-semiregular. Next, suppose

K : X × X \ diag(X) −→ R (2.2.247)

is a kernel which is standard in both variables, i.e., K is μ ⊗ μ-measurable and

both K(x, y) and K  (x, y) := K(y, x) satisfy (2.2.25)-(2.2.26). Fix q ∈ (1, ∞) and
assume that an operator

T : L q (X, μ) −→ L q,∞ (X, μ), linear and bounded, (2.2.248)

has been given, with the property that T is associated in L q (X, μ) with the kernel
K, in the sense that for each f ∈ L q (X, μ) one has
∫
(T f )(x) = K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X \ supp f . (2.2.249)
X

For each ε > 0 define Tε as in (2.2.32)-(2.2.33), and introduce the maximal operator
Tmax as in (2.2.34). Then

T : L p (X, μ) −→ L p (X, μ) linearly and boundedly for each p ∈ (1, ∞),

(2.2.250)
and the maximal operator Tmax induces a sub-linear, bounded, and continuous
mapping

Tmax : L p (X, μ) −→ L p (X, μ) for each p ∈ (1, ∞). (2.2.251)

Also, for each fixed reference point x0 ∈ X,

 
μ(x) p p
Tmax : L 1 X, 1+ρ(x, x ) d ∩ Lloc (X, μ) −→ Lloc (X, μ)
0
(2.2.252)
is well-defined, bounded, continuous, for each p ∈ (1, ∞).

Consequently, for each ε > 0 the linear operator Tε is bounded in the context
312 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Tε : L p (X, μ) −→ L p (X, μ) for each p ∈ (1, ∞), (2.2.253)

and
sup Tε  L p (X,μ)→L p (X,μ) < ∞ for each p ∈ (1, ∞). (2.2.254)
ε>0

Proof From (2.2.227) we know that

T : L r (X, μ) −→ L r (X, μ),

(2.2.255)
linearly and boundedly for each r ∈ (1, q).

Fix some arbitrary ro ∈ (1, q) and denote by ro its Hölder conjugate exponent.
Then (2.2.255) ensures that T maps L ro (X, μ) linearly and boundedly into itself. Via
duality, we then conclude that

T  maps L ro (X, μ) linearly and boundedly into itself. (2.2.256)

We also claim that


for each function f ∈ L ro (X, μ) we have
∫ (2.2.257)

(T f )(x) = K  (x, y) f (y) dμ(y) for μ-a.e. x ∈ X \ supp f .
X

To justify the claim in (2.2.257), pick some f ∈ L ro (X, μ) and consider an arbitrary
β
test function φ ∈ 𝒞c (X, μ), for some finite exponent β ∈ 0, (log2 Cρ )−1 , satisfying

distρ supp f , supp φ > 0. (2.2.258)

The latter property then permits us to use Fubini’s Theorem which, in concert with
(2.2.249), gives
∫ ∫

(T f )(x)φ(x) dμ(x) = f (y)(T φ)(y) dμ(y)
X X
∫ ∫ 
= f (y) K(y, x)φ(x) dμ(x) dμ(y)
X X
∫ ∫ 
= K  (x, y) f (y) dμ(y) φ(x) dμ(x). (2.2.259)
X X

Note that T  f ∈ L ro (X, μ) ⊆ Lloc 1 (X, μ), thanks to (2.2.256). In addition,

(2.2.25) implies
∫ that the function defined at μ-a.e. x ∈ X \ supp f according to
g(x) := X K  (x, y) f (y) dμ(y) is absolutely integrable with respect to μ on any

  E of X \ supp f satisfying distρ E, supp f > 0. As a consequence,
closed subset
h := g − T f X\supp f satisfies (2.2.228)-(2.2.229) (with F := supp f ). As such,
Lemma 2.2.10 applies and proves that (2.2.257) holds. Granted this, and given that
K  has the same qualities as K, we may then run the program that has produced
(2.2.227) in the proof of Proposition 2.2.8 for the transpose operator T  and arrive
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 313

at the conclusion that

T  : L r (X, μ) −→ L r (X, μ),
(2.2.260)
linearly and boundedly for each r ∈ (1, ro ).

Via duality we therefore obtain

T : L r (X, μ) −→ L r (X, μ),

(2.2.261)
linearly and boundedly for each r ∈ (ro, ∞).

Since ro ∈ (1, q) has been arbitrarily chosen, this implies

T : L r (X, μ) −→ L r (X, μ),

(2.2.262)
linearly and boundedly for each r ∈ (1, ∞),

proving (2.2.250). All other claims now follow from this and Proposition 2.2.7. 
The point of our next result is that smoothly truncating a given kernel retains
some of its essential properties, in a quantitatively uniform fashion with respect to
the truncation parameter.
Lemma 2.2.12 Suppose (X, ρ, μ) is a d-Ahlfors regular quasi-metric space for some
d ∈ (0, ∞) and assume K : X × X \ diag(X) → R is a standard kernel in the first
variable (i.e., the function K is μ ⊗ μ-measurable and (2.2.25)-(2.2.26) are satisfied).
Pick a real-valued function ψ ∈ 𝒞∞ (R) with the property that ψ ≡ 1 on R \ [−1, 1]
and ψ ≡ 0 on [−1/2, 1/2]. Having chosen a finite exponent β ∈ 0, (log2 Cρ )−1 , for
each ε > 0 define
 ρβ (x, y) 
#
K(ε) (x, y) := K(x, y)ψ , ∀(x, y) ∈ X × X \ diag(X). (2.2.263)
εβ
Then each K(ε) is also a standard kernel in the first variable, with constants
independent of ε. Moreover, if actually K : X × X \ diag(X) → R is a standard
kernel in both variables, then each K(ε) is also a standard kernel in both variables,
with constants independent of ε.
Proof Fix an arbitrary ε > 0. First, from (2.2.263) and (2.2.25) we see that for each
x, y ∈ X with x  y we have
    C0
K(ε) (x, y) ≤ sup |ψ| . (2.2.264)
R ρ(x, y)d

Going further, pick x1, x2, y ∈ X with y  {x1, x2 } and ρ(x2, y) ≥ C1 ρ(x2, x1 ). The
goal is to estimate the difference K(ε) (x1, y) − K(ε) (x2, y), much as in (2.2.26). To
this end, select a real number
C > (Cρ )3 . (2.2.265)
We shall analyze several cases. First, if ρ(x2, y) > Cε then
314 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 
ρ(x2, y) ≤ Cρ · max ρ(x2, x1 ), ρ(x1, y)
 
≤ Cρ · max ρ(x2, y)/C1, ρ(x1, y) (2.2.266)

which, in view of the fact that C > Cρ (cf. (2.2.265)), forces ρ(x2, y) ≤ Cρ · ρ(x1, y).
Hence,
Cε < ρ(x2, y) ≤ Cρ · ρ(x1, y). (2.2.267)
From this and [112, (7.1.21)] we then conclude that

ε < (Cρ )−3 Cε ≤ (Cρ )−2 · ρ(x1, y) ≤ ρ# (x1, y). (2.2.268)

Since we also have

ε < (Cρ )−2 Cε < (Cρ )−2 ρ(x2, y) ≤ ρ# (x2, y), (2.2.269)

it follows that
 ρβ (x1, y)   ρβ (x2, y) 
#
ψ = 0 and ψ # β = 0, (2.2.270)
εβ ε
hence K(ε) (x1, y) = 0 and K(ε) (x2, y) = 0. Hence, in this case we have

K(ε) (x1, y) − K(ε) (x2, y) = 0. (2.2.271)

Assume next that ρ(x2, y) ≤ Cε. In this case, write

     ρβ (x , y)  
K(ε) (x1, y) − K(ε) (x2, y) ≤ K(ε) (x1, y) − K(ε) (x2, y)ψ # 1 
εβ
 
   ρ#β (x1, y)   ρβ (x2, y)  

+ K(ε) (x2, y)ψ −ψ #

 εβ εβ 

=: I + II. (2.2.272)

Then (2.2.26) gives

  ρ(x , x )γ
2 1
I ≤ C0 sup |ψ| . (2.2.273)
R ρ(x2, y)
d+γ

Also, based on (2.2.25), the Mean Value Theorem, [112, (7.1.19)-(7.1.20)], and the
fact that we are presently assuming ρ(x2, y) ≤ Cε we may estimate
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 315
 β 
  C0  ρ (x1, y) − ρβ (x2, y)
II ≤ sup |ψ  | # #
R ρ(x2, y)d εβ
  C0
β
ρ# (x2, x1 )
≤ sup |ψ  |
R ρ(x2, y) d εβ
  β
ρ )β sup |ψ  | ρ(x2, x1 ) .
≤ CC0 (C (2.2.274)
R ρ(x2, y)d+β

If γ ≤ (log2 Cρ )−1 we may choose β := γ to begin with, then deduce from (2.2.264),
(2.2.273), and (2.2.274) that K(ε) is indeed a standard kernel. In the case when
γ > (log2 Cρ )−1 , the fact that we have ρ(x2, y) ≥ C1 ρ(x2, x1 ) entails

ρ(x2, x1 )γ β−γ ρ(x2, x1 )

β
≤ (C1 ) . (2.2.275)
ρ(x2, y)d+γ ρ(x2, y)d+β
In concert with (2.2.264), (2.2.273), and (2.2.274) this once again permits us to
conclude that K(ε) is a standard kernel in this case. Ultimately, we conclude that K(ε)
is a standard kernel in the first variable, with constants independent of ε.
Finally, since
 ρβ (y, x) 
 #
K(ε) (x, y) = K(ε) (y, x) = K(y, x)ψ
εβ
 ρβ (x, y) 
#
= K(y, x)ψ for all (x, y) ∈ X × X \ diag(X), (2.2.276)
εβ
the same type of reasoning as above shows that if K is a standard kernel in both
 is also a standard kernel in both variables with constants
variables then each K(ε)
independent of ε. 

Our next proposition concludes boundedness results for the maximal operator
starting from the assumption that the truncated operators are uniformly bounded.

Proposition 2.2.13 Suppose (X, ρ, μ) is a d-Ahlfors regular quasi-metric space for

some d ∈ (0, ∞) with the property that the quasi-distance ρ : X × X → [0, +∞) is
continuous in the product topology τρ × τρ . Assume

K : X × X \ diag(X) −→ R (2.2.277)

is a standard kernel in both variables (cf. Definition 2.2.4). For each ε > 0 introduce
the truncated operator Tε as in (2.2.32)-(2.2.33), and assume that there exists an
exponent q ∈ (1, ∞) with the property that

sup Tε  L q (X,μ)→L q,∞ (X,μ) < ∞. (2.2.278)

ε>0
316 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

The maximal operator Tmax , defined as in (2.2.34), induces a sub-linear, bounded,

and continuous mapping in the following contexts:

Tmax : L 1 (X, μ) −→ L 1,∞ (X, μ), and (2.2.279)

Tmax : L p (X, μ) −→ L p (X, μ) for each p ∈ (1, ∞). (2.2.280)

Proof Recall that we are assuming that the kernel K is standard in both variables,
i.e., both K and K  satisfy (2.2.25)-(2.2.26) for some finite constants γ > 0 and
C0 ≥ 0, C1 > 0. For each ε > 0 define K(ε) as in (2.2.263). From Lemma 2.2.12
we know that each K(ε) is also a standard kernel in both variables, with constants
independent of ε. To proceed, fix some reference point x0 ∈ X. For each ε > 0 and
each function  
μ(x)
f ∈ L 1 X, (2.2.281)
1 + ρ(x, x0 ) d

define ∫
(T(ε) f )(x) := K(ε) (x, y) f (y) dμ(y), ∀x ∈ X. (2.2.282)
X
Let M X be the Hardy-Littlewood maximal function on X (cf. (A.0.84)). We claim
that there exists some C ∈ (0, ∞) with the property that for each function f as in
(2.2.281) we have
 
Tε f − T(ε) f  ≤ CM X f on X, for each ε > 0. (2.2.283)

To justify this, fix ε > 0 and pick a function f as in (2.2.281). For each x ∈ X write
 
(Tε f )(x) − (T(ε) f )(x) (2.2.284)
∫  
  ρβ (x, y)  
 
≤ |K(x, y)| ψ # β − 1 {y ∈X:ρ(x,y)>ε } (y) | f (y)| dμ(y).
 ε 
X
 ρ β (x,y) 
Recall (2.2.25) and observe that ψ # ε β  1 {y ∈X:ρ(x,y)>ε } (y) forces ρ(x, y)
to be comparable to ε, i.e., there exists θ ∈ (0, 1) independent of x, y such that
θ < ρ(x, y)/ε < θ −1 . Consequently, there exists a constant C ∈ (0, ∞) with the
property that
∫
 
(Tε f )(x) − (T(ε) f )(x) ≤ C | f (y)| dμ(y) for each x ∈ X. (2.2.285)
ε d Bρ (x,Cε)

Since X is d-Ahlfors regular, this readily implies (2.2.283).

Next, from (2.2.278), (2.2.283), and [112, (7.6.18)] we see that
 
sup T(ε)  L q (X,μ)→L q,∞ (X,μ) < ∞. (2.2.286)
ε>0
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 317

Granted this, (2.2.282), and the fact that each K(ε) is a standard kernel in both vari-
ables with constants independent of ε, we may invoke Corollary 2.2.11 to conclude
that

(T(ε) )max : L 1 (X, μ) −→ L 1,∞ (X, μ), and (2.2.287)

(T(ε) )max : L p (X, μ) −→ L p (X, μ) for p ∈ (1, ∞), (2.2.288)

are all bounded operators uniformly in ε > 0.

Recall next from (2.2.33)-(2.2.34) and (2.2.282) that for each ε > 0 and each
function f as in (2.2.281) we have
∫ 
  
 
(T(ε) )max f (x) = sup  K(ε) (x, y) f (y) dμ(y) (2.2.289)
ε >0  y ∈X, ρ(x,y)>ε 

for all points x ∈ X. Observe that whenever ε  > (Cρ )2 ε and ρ(x, y) > ε  then we
have ρ(x, y) > (Cρ )2 ε which, in view of [112, (7.1.21)], further implies ρ# (x, y) > ε.
 ρ β (x,y) 
Consequently, ψ # ε β = 1 so K(ε) (x, y) = K(x, y) in this case. Consequently,
∫
K(ε) (x, y) f (y) dμ(y) = (Tε f )(x),
y ∈X, ρ(x,y)>ε (2.2.290)
for f as in (2.2.281), x ∈ X, and ε > (Cρ )2 ε.

From (2.2.289) and (2.2.290) we then conclude that for each f as in (2.2.281) and
each ε > 0 we have
  
(T(ε) )max f (x) ≥ sup (Tε f )(x), ∀x ∈ X. (2.2.291)
ε >(Cρ )2 ε

Since for each x ∈ X we have

    
sup (Tε f )(x)  sup (Tε f )(x) = Tmax f (x) as ε  0, (2.2.292)
ε >(Cρ )2 ε ε >0

Lebesgue’s Monotone Convergence Theorem applies and gives that for each f as in
(2.2.281) and each p ∈ (1, ∞) we have
∫ ∫
 p
sup (Tε f ) p dμ  Tmax f dμ as ε  0. (2.2.293)
ε >(Cρ )2 ε
X X

Collectively, (2.2.291)-(2.2.293) and (2.2.288) then permit us to conclude that for

each exponent p ∈ (1, ∞) there exists some constant C = C(X, K, p) ∈ (0, ∞) such
that
318 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫  1/p
 
Tmax f  p = lim sup (Tε f ) dμ
p
L (X,μ) ε0 ε >(Cρ )2 ε
X
∫  1/p
 p
≤ lim sup (T(ε) )max f dμ
ε0
X
 
≤ sup (T(ε) )max f  L p (X,μ) ≤ C f  L p (X,μ), (2.2.294)
ε>0

for each function f ∈ L p (X, μ). From this, the claims about (2.2.280) readily follow.
Finally, the claims about (2.2.279) are dealt with similarly, now using (2.2.287). 

Next we elaborate on how the functional analytic properties of the maximal

operator and the principal-value singular integral operator associated with a given
kernel are related to one another. Essentially, Proposition 2.2.14 shows that “TPV is
bounded if and only if Tmax is bounded.”

Proposition 2.2.14 Assume (X, ρ, μ) is a d-Ahlfors regular quasi-metric space for

some d ∈ (0, ∞) with the property that the quasi-distance ρ : X × X → [0, +∞)
is continuous in the product topology τρ × τρ , andsuch that μ is a complete Borel-
semiregular measure. Fix a finite exponent β ∈ 0, (log2 Cρ )−1 along with some
p ∈ (1, ∞). Also, suppose

K : X × X \ diag(X) −→ R (2.2.295)

is a kernel which is standard in the first variable (cf. Definition 2.2.4). Finally, for
each ε > 0 associate Tε with the kernel K as in (2.2.32)-(2.2.33), and introduce the
maximal operator Tmax as in (2.2.34). Then the following statements are true.
β
(1) For each function f ∈ 𝒞c (X, ρ) the limit
∫
lim+ K(x, y) f (y) dμ(y) exists for μ-a.e. x ∈ X (2.2.296)
ε→0 y ∈X, ρ(x,y)>ε

if and only if the limit

∫
𝒦(x) := lim+ K(x, y) dμ(y) exists for μ-a.e. x ∈ X. (2.2.297)
ε→0
y ∈X
1>ρ(x,y)>ε

β
(2) If (2.2.297) holds then the mapping assigning to each f ∈ 𝒞c (X, ρ) the function
∫

TPV f (x) := lim+ K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X
ε→0 y ∈X, ρ(x,y)>ε
(2.2.298)
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 319

extends to a linear and bounded operator from L p (X, μ) into L p,∞ (X, μ) if and
only if

Tmax : L p (X, μ) −→ L p,∞ (X, μ) is a bounded mapping. (2.2.299)

Moreover, if either (hence both) of these conditions happen, then (2.2.296) holds
β
for each f ∈ L p (X, μ), the extension of 𝒞c (X, ρ)  f → TPV f to a linear and
bounded operator from L (X, μ) into L (X, μ) is
p p,∞

TPV : L p (X, μ) −→ L p,∞ (X, μ), acting on each f ∈ L p (X, μ) by

∫

TPV f (x) = lim+ K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X.
ε→0 y ∈X, ρ(x,y)>ε
(2.2.300)
(3) Strengthen the original hypotheses by now assuming that K is actually a standard
kernel in both variables. Once again, suppose (2.2.297) holds. Then
β
the mapping assigning to each f ∈ 𝒞c (X, ρ) the function TPV f
defined as in (2.2.298) extends to a linear and bounded operator (2.2.301)
from L p (X, μ) into L p,∞ (X, μ)
if and only if

Tmax : L q (X, μ) −→ L q (X, μ) is a bounded mapping

(2.2.302)
for some (or each) index q ∈ (1, ∞).

Moreover, the validity of either (hence both) of these conditions implies that for
each exponent q ∈ (1, ∞) the operator

TPV : L q (X, μ) −→ L q (X, μ), acting on each f ∈ L q (X, μ) by

∫

TPV f (x) = lim+ K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X,
ε→0 y ∈X, ρ(x,y)>ε
(2.2.303)
is well defined, linear, bounded,

for each function f ∈ L q (X, μ) one has

(2.2.304)
Tε f −→ TPV f in L q (X, μ) as ε → 0+,

and, corresponding to q = 1, the operator

TPV : L 1 (X, μ) −→ L 1,∞ (X, μ), acting on each f ∈ L 1 (X, μ) by

∫

TPV f (x) = lim+ K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X,
ε→0 y ∈X, ρ(x,y)>ε
(2.2.305)
is also well defined, linear, bounded.
320 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

(4) As above, strengthen the original hypotheses by assuming that K is in fact a

standard kernel in both variables and, once more, suppose (2.2.297) holds. In
addition, assume the limit
∫
lim+ K  (x, y) dμ(y) exists for μ-a.e. x ∈ X, (2.2.306)
ε→0
y ∈X
1>ρ(y,x)>ε

and suppose there exists some integrability exponent q ∈ (1, ∞) such that

Tmax : L q (X, μ) −→ L q (X, μ) is a bounded mapping. (2.2.307)

Then, if q  ∈ (1, ∞) is the Hölder conjugate exponent of q, the (real) transpose

of (2.2.303) is the operator
 
 : L q (X, μ) −→ L q (X, μ), acting on each f ∈ L q (X, μ) as 
TPV
∫
 
TPV f (x) = lim+ K  (x, y) f (y) dμ(y) for μ-a.e. x ∈ X.
ε→0 y ∈X, ρ(y,x)>ε
(2.2.308)
β
Proof Given an arbitrary function f ∈ 𝒞c (X, ρ) along with some ε ∈ (0, 1), for
each x ∈ X decompose
∫ ∫
K(x, y) f (y) dμ(y) = K(x, y) f (y) dμ(y)
y ∈X, ρ(x,y)>ε y ∈X, ρ(x,y)≥1
∫
+ K(x, y)[ f (y) − f (x)] dμ(y)
y ∈X
1>ρ(x,y)>ε
 ∫ 
+ K(x, y) dμ(y) f (x).
y ∈X
1>ρ(x,y)>ε
(2.2.309)

The fact that f is Hölder continuous and K satisfies (2.2.25) then ensures (cf. [112,
(7.2.5)]) that
∫
lim+ K(x, y)[ f (y) − f (x)] dμ(y)
ε→0
y ∈X
1>ρ(x,y)>ε
∫
= K(x, y)[ f (y) − f (x)] dμ(y). (2.2.310)
y ∈X, 1>ρ(x,y)
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 321

Bearing this in mind it follows that (2.2.296) holds if and only if (2.2.297) holds, in
which scenario we actually have
∫
lim+ K(x, y) f (y) dμ(y) (2.2.311)
ε→0 y ∈X, ρ(x,y)>ε
∫
= K(x, y) f (y) dμ(y)
y ∈X, ρ(x,y)≥1
∫
+ K(x, y)[ f (y) − f (x)] dμ(y) + 𝒦(x) f (x)
y ∈X, 1>ρ(x,y)

for μ-a.e. x ∈ X. This proves the equivalence claimed in item (1).

Turning attention to item (2), work under the assumption that (2.2.297) holds.
Recall from [112, (7.4.13)] that
β
𝒞c (X, ρ) → L p (X, μ) densely. (2.2.312)

Suppose first that the mapping (2.2.298) extends to a linear and bounded operator
on L p (X, μ) (which is a meaningful assumption in view of (2.2.297) and (2.2.312)).
β
Denote by T" PV the (unique) extension by density of the operator TPV acting on 𝒞c (X, ρ)
as in (2.2.298) as a linear and bounded mapping

T"
PV : L (X, μ) −→ L
p p,∞
(X, μ). (2.2.313)

We claim that
for each function f ∈ L p (X, μ) we have
∫
 (2.2.314)
T"
PV f (x) = K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X \ supp f .
X

To prove this, fix an arbitrary f ∈ L p (X, μ) and pick some xo ∈ X \ supp f . From

β
(2.2.312) we know that there exists a sequence { f j } j ∈N ⊆ 𝒞c (X, ρ) convergent to
f in L p (X, μ). After eventually multiplying by a suitable cutoff function, matters
may be arranged so that, for some r > 0, we have

Bρ (xo, r) ∩ supp f j =  for each j ∈ N. (2.2.315)

Then, on the one hand, the continuity of (2.2.313) implies

T" "
PV f = lim TPV f j = lim TPV f j in L
p,∞
(X, μ). (2.2.316)
j→∞ j→∞

By eventually passing to a subsequence we may therefore assume that

T"
PV f = lim TPV f j at μ-a.e. point in X. (2.2.317)
j→∞
322 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Consequently, for μ-a.e. x ∈ Bρ (xo, r) we have

∫
 
T"
PV f (x) = lim TPV f j (x) = lim K(x, y) f j (y) dμ(y)
j→∞ j→∞ X
∫
= K(x, y) f (y) dμ(y), (2.2.318)
X

where the last equality is justified as in (2.2.156). Reasoning as in (2.2.157)-(2.2.160)

then finishes the proof of (2.2.314). Granted (2.2.313)-(2.2.314), we may now invoke
(2.2.39) to conclude that (2.2.299) is true.
β
Conversely, assume (2.2.299) is true. Given that for each f ∈ 𝒞c (X, ρ) we have
 
T f  ≤ Tmax f for μ-a.e. point on X, (2.2.319)
PV

and TPV f happens to be a μ-measurable function, as the pointwise limit of a sequence

of μ-measurable functions (cf. (2.2.298) and [112, Remark 3.1.2], where the latter
uses the fact that the measure μ is assumed to be complete). In view of these
properties, it follows from [112, (6.2.16)] that the operator (2.2.298) extends to a
linear and bounded operator from L p (X, μ) into L p,∞ (X, μ).
Assume next that (2.2.299) holds, with the goal of showing that (2.2.296) holds
β
for each function f ∈ L p (X, μ), and that the extension of 𝒞c (X, ρ)  f → TPV f to
a linear and bounded operator on L (X, μ) is as described in (2.2.300). To this end,
p

we shall use [112, Proposition 6.2.11] for the ambient 𝒳 := X × [0, ∞) equipped
with the canonical product topology, L p (X, μ) playing the role of the normed space
Y , and X × {0} playing the role of X in the statement of [112, Proposition 6.2.11].
Also, take Γ (x, 0) := {x} × (0, ∞) for each (x, 0) ∈ X × {0}, a choice ensuring that

(x, 0) ∈ Γ (x, 0) for every (x, 0) ∈ X × {0}.
Next, consider the linear operator T , mapping functions defined on X into func-
tions defined on 𝒳 \ (X × {0}) = X × (0, ∞), of the following sort. For every given
f ∈ Y = L p (X, μ) define
∫
(T f )(x, ε) := K(x, y) f (y) dμ(y) for all (x, ε) ∈ X × (0, ∞), (2.2.320)
y ∈X
ρ(x,y)>ε

and note that this implies

 ∫ 
 
 
(T f )(x, 0) := sup |(T f )(z, ε)| = sup  K(x, y) f (y) dμ(y)
(z,ε)∈Γ((x,0)) ε>0  
y ∈X
ρ(x,y)>ε

= (Tmax f )(x) for every x ∈ X. (2.2.321)

In turn, from (2.2.321) and (2.2.299) we conclude that condition [112, (6.2.74)]
is verified in the present setting. To finish the implementation of [112, Proposi-
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 323
β
tion 6.2.11] we finally choose V := 𝒞c (X, ρ), which is a dense subset of L p (X, μ)
(cf. (2.2.312)). Moreover, condition [112, (6.2.75)] is also satisfied since for every
g ∈ V the limit
∫
lim (T g)(x, ε) = lim+ K(x, y)g(y) dμ(y) = lim+ (Tε g)(x)
Γ((x,0))(x,ε)→(x,0) ε→0 ε→0
y ∈Y
ρ(x,y)>ε
(2.2.322)

exists for μ-a.e. x ∈ X, thanks to the assumption that (2.2.297) holds and item
(1). Hence, all hypotheses of [112, Proposition 6.2.11] are satisfied for the present
choices, so we may conclude from it that for every function f ∈ L p (X, μ) the limit

TPV f (x) = lim+ (Tε f )(x) = lim (T f )(x, ε) (2.2.323)
ε→0 Γ((x,0))(x,ε)→(x,0)

exists for μ-a.e. x ∈ X. From this, the desired conclusions follow. This concludes
the treatment of item (2).
As far as item (3) is concerned, Corollary 2.2.11 applied to T := TPV proves
the equivalence of (2.2.301) with (2.2.302). If either of these conditions holds then
items (1)-(2) imply that for each exponent q ∈ (1, ∞) the operator (2.2.303) is well
defined, linear, and bounded. In concert with (the strongest version of) (2.2.301)
and Lebesgue’s Dominated Convergence Theorem, this also ensures that (2.2.304) is
valid. Consider the very last claim, pertaining to (2.2.305). Assuming (2.2.302), item
(i) in Proposition 2.2.8 gives that TPV extends to a bounded mapping of L 1 (X, μ) into
L 1,∞ (X, μ). With this in hand, what we have proved in the current item (2) applies
(with p := 1) and yields all desired conclusions about (2.2.305).
Finally, consider the duality result claimed in item (4). For each ε > 0 define
 
Q ε : L q (X, μ) −→ L q (X, μ) given by
∫ (2.2.324)
(Q ε f )(x) := K  (x, y) f (y) dμ(y) for all x ∈ X,
y ∈X, ρ(y,x)>ε


for each function f ∈ L q (X, μ). Observe that, for each ε > 0, the assumption made
in (2.2.307) implies that Tε : L q (X, μ) → L q (X, μ) is a well-defined, linear, and
bounded operator, whose (real) transpose is

Tε = Q ε on L q (X, μ). (2.2.325)

Consequently, for each ε > 0 and each f ∈ L q (X, μ) we have
324 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫  ∫ 
   
Q ε f  L q (X,μ) = sup  (Q ε f )g dμ = sup  f (Tε g) dμ
g  L q (X, μ) =1 X g  L q (X, μ) =1 X

≤  f  L q (X,μ) Tε g L q (X,μ) ≤  f  L q (X,μ) Tmax g L q (X,μ)

≤  f  L q (X,μ) Tmax  L q (X,μ)→L q (X,μ) (2.2.326)

hence
sup Q ε  L q (X,μ)→L q (X,μ) ≤ Tmax  L q (X,μ)→L q (X,μ) . (2.2.327)
ε>0

From (2.2.327), the fact that K is a standard kernel in both variables, and Proposi-
tion 2.2.13 we conclude that
Qmax : L p (X, μ) −→ L p (X, μ) is a bounded mapping
(2.2.328)
for each integrability exponent p ∈ (1, ∞),

where

(Qmax f )(x) := sup |(Q ε f )(x)| for each f ∈ L p (X, μ). (2.2.329)
ε>0

Granted (2.2.328), and bearing in mind (2.2.306), what we have proved in the current
item (3) guarantees that the operator
  
Q : L q (X, μ) −→ L q (X, μ), acting on each f ∈ L q (X, μ) by
∫ (2.2.330)
(Q f )(x) := lim+ K  (x, y) f (y) dμ(y) for μ-a.e. x ∈ X,
ε→0 y ∈X, ρ(y,x)>ε

is well defined, linear, bounded, and


Q ε f −→ Q f in L q (X, μ) as ε → 0+
 (2.2.331)
for each function f ∈ L q (X, μ).

Then the claim in relation to (2.2.308) is seen from this, (2.2.325), (2.2.330), and
(2.2.304). 

It is also useful to note that the principal-value singular integral operator associ-
ated with a given kernel is bounded whenever the corresponding truncated singular
integral operators are bounded in a uniform fashion with respect to the truncation
parameter.

Corollary 2.2.15 Suppose (X, ρ, μ) is a d-Ahlfors regular quasi-metric space for

some d ∈ (0, ∞) with the property that the quasi-distance ρ : X × X → [0, +∞)
is continuous in the product topology τρ × τρ , and such that μ is a complete Borel-
semiregular measure. Assume K : X × X \ diag(X) → R is a standard kernel in
both variables (cf. Definition 2.2.4) such that
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 325
∫
the limit lim+ K(x, y) dμ(y) exists for μ-a.e. x ∈ X. (2.2.332)
ε→0
y ∈X
1>ρ(x,y)>ε

For each ε > 0 introduce the truncated operator Tε as in (2.2.32)-(2.2.33), and

assume that there exists some q ∈ (1, ∞) with the property that

sup Tε  L q (X,μ)→L q,∞ (X,μ) < ∞. (2.2.333)

ε>0

Finally, fix a finite exponent β ∈ 0, (log2 Cρ )−1 .
β
Then for each p ∈ [1, ∞) the mapping assigning to each f ∈ 𝒞c (X, ρ) the
function
∫

TPV f (x) := lim+ K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X (2.2.334)
ε→0 y ∈X, ρ(x,y)>ε

extends to a linear and bounded operator

TPV : L p (X, μ) −→ L p (X, μ) if p ∈ (1, ∞),

(2.2.335)
TPV : L 1 (X, μ) −→ L 1,∞ (X, μ) if p = 1,

acting on each f ∈ L p (X, μ) according to

∫

TPV f (x) = lim+ K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X. (2.2.336)
ε→0 y ∈X, ρ(x,y)>ε

Of course, (2.2.333) is satisfied if

Tmax : L q (X, μ) −→ L q (X, μ) is bounded. (2.2.337)

See Corollary 2.2.11 and Proposition 2.2.7 for more on this.

In the converse direction, if for some fixed p ∈ (1, ∞) the operator TPV is meaning-
fully defined as in (2.2.336) for each f ∈ L p (X, μ) and induces a bounded mapping
TPV : L p (X, μ) → L p,∞ (X, μ), then Corollary 2.2.11 and Proposition 2.2.8 applied
(to T := TPV ) plus Proposition 2.2.14 give that TPV continues to be meaningfully
defined by the formula in (2.2.336) for any function f as in (2.2.281), and

TPV : L 1 (X, μ) −→ L 1,∞ (X, μ) boundedly, and

(2.2.338)
TPV : L q (X, μ) −→ L q (X, μ) boundedly for each q ∈ (1, ∞).

Here is the proof of Corollary 2.2.15.

Proof of Corollary 2.2.15 All desired properties are obtained by combining Propo-
sition 2.2.13 with Proposition 2.2.14. 
326 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Moving on, the goal is to prove the basic weak convergence result stated a little
later, in Theorem 2.2.20. This requires some preparations, and a number of auxiliary
tools are developed in Lemmas 2.2.16, 2.2.17, 2.2.19.
Lemma 2.2.16 Let X, Y be two normed spaces and suppose a : X × Y → C is a
bilinear map that is also continuous, i.e., there exists C ∈ [0, ∞) such that

|a(x, y)| ≤ C xX  yY for all (x, y) ∈ X × Y . (2.2.339)

Then there exists a unique map Ra ∈ L(X → Y ∗ ) such that

a(x, y) = Y ∗ Ra x, yY for all (x, y) ∈ X × Y . (2.2.340)

Proof Define the map Ra : X → Y ∗ by setting Ra (x) := a(x, ·) for each x ∈ X.

Then Ra ∈ L(X → Y ∗ ) and for each (x, y) ∈ X × Y we have Y ∗ Ra x, yY = a(x, y)
as wanted. Also, uniqueness follows from identity (2.2.340). 
Let X be a Banach space and recall the notation L(X) := L(X → X) and the
operator norm
 
T  L(X) := sup T x : x ∈ X,  xX = 1 , for each T ∈ L(X). (2.2.341)

Then the normed space L(X),  ·  L(X) is a Banach space. In this a setting, fix
(x, ξ) ∈ X × X ∗ and define the linear map

ωx,ξ : L(X) −→ C, ωx,ξ (T) := X T x, ξX ∗ for each T ∈ L(X). (2.2.342)

Since for each T ∈ L(X) we may estimate

   
ωx,ξ (T) ≤ X T x, ξX ∗  ≤ ξ X ∗ T xX ≤ ξ X ∗  xX T  L(X), (2.2.343)
 
it follows that ωx,ξ ∈ L(X)∗ and ωx,ξ  L(X) ∗ ≤  x X ξ  X ∗ .
∗
 discussion, we now∗ define the linear space generated in L(X) by
Based on this
the collection ωx,ξ : x ∈ X, ξ ∈ X , i.e.,


N 
Lω (X) := ωx j ,ξ j : N ∈ N, x j ∈ X, ξ j ∈ X ∗ , (2.2.344)
j=1

and set
L(X)∗
L∗ (X) := Lω (X) . (2.2.345)
Then L∗ (X) is a close subspace of L(X)∗ and is a Banach space when endowed
with the norm inherited from L(X)∗ .
Lemma 2.2.17 Let X be a reflexive Banach space. Consider the bilinear form

a : L(X) × L∗ (X) −→ C
(2.2.346)
a(T, Λ) := Λ(T) for all (T, Λ) ∈ L(X) × L∗ (X),
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 327

and let Ra ∈ L L(X) → (L∗ (X))∗ be the operator associated with a as in
Lemma 2.2.16. Then Ra : L(X) −→ (L∗ (X))∗ is an isometric isomorphism.

Proof Pick T ∈ L(X) arbitrary. Then (2.2.342), (2.2.346), and (2.2.340) allow us
to write
 
T  L(X) = sup T xX : x ∈ X,  xX = 1
  
= sup X T x, ξX ∗  : x ∈ X,  xX = 1, ξ ∈ X ∗, ξ X ∗ = 1
  
= sup ωx,ξ (T) : x ∈ X,  xX = 1, ξ ∈ X ∗, ξ X ∗ = 1
  
= sup a(T, ωx,ξ ) : x ∈ X,  xX = 1, ξ ∈ X ∗, ξ X ∗ = 1
  
≤ sup a(T, Λ) : Λ ∈ L∗ (X), Λ L(X)∗ ≤ 1
  
= sup (L∗ (X))∗ Ra (T), Λ L∗ (X)  : Λ ∈ L∗ (X), Λ L(X)∗ ≤ 1
 
≤  Ra (T)(L∗ (X))∗ (2.2.347)

Also,
    
 Ra (T) = sup (L∗ (X))∗ Ra (T), Λ L∗ (X)  : Λ ∈ L∗ (X), Λ L(X)∗ ≤ 1
(L∗ (X))∗
  
= sup a(T, Λ) : Λ ∈ L∗ (X), Λ L(X)∗ ≤ 1
  
= sup Λ(T) : Λ ∈ L∗ (X), Λ L(X)∗ ≤ 1

≤ T  L(X) . (2.2.348)

Thus,
 
 Ra (T) = T  L(X) for all T ∈ L(X). (2.2.349)
(L∗ (X))∗

In particular, Ra is injective. To finish the proof of the lemma it remains to show that
Ra is surjective. To this end, pick Θ ∈ (L∗ (X))∗ and define b : X × X ∗ −→ C by

b(x, ξ) := (L∗ (X))∗ Θ, ωx,ξ  L∗ (X) for all (x, ξ) ∈ X × X ∗ . (2.2.350)

This is a bilinear and continuous map and we may apply

 Lemma 2.2.16 to conclude
that there exists a unique map T := Rb with T ∈ L X → (X ∗ )∗ = L(X) (recall
that X is reflexive) such that

(L∗ (X))∗ Θ, ω x,ξ  L∗ (X) = X T x, ξX ∗ for all (x, ξ) ∈ X × X ∗ . (2.2.351)

This, in concert with (2.2.342) and (2.2.346) further implies

328 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

(L∗ (X))∗ Θ, ω x,ξ  L∗ (X)
= ωx,ξ (T) = a T, ωx,ξ (2.2.352)
 
for all (x, ξ) ∈ X × X ∗ . Since ωx,ξ : x ∈ X, ξ ∈ X ∗ spans a dense subspace of
L∗ (X) and since Θ and a(T, ·) are continuous, it follows that

(L∗ (X))∗ Θ, Λ L∗ (X) = a T, Λ = (L∗ (X))∗ Ra (T), Λ L∗ (X) (2.2.353)

for all Λ ∈ L∗ (X). Hence, Ra (T) = Θ as functionals in (L∗ (X))∗ . This completes
the proof of the fact that Ra is onto. 

Next we consider two topologies defined on L(X). One is the weak operator
topology (WOT) which is the topology induced by the family of seminorms
 
L(X)  T → X T x, ξX ∗  for all x ∈ X, ξ ∈ X ∗ . (2.2.354)

The other is the ultraweak operator topology (UWOT) which is defined as the
topology induced by the family of seminorms

L(X)  T → |Λ(T)| for all Λ ∈ L∗ (X). (2.2.355)

The relationship between these topologies on the closed unit ball in L(X) in the case
when X is Banach is discussed in Lemma 2.2.19 below. As a preamble, we record
the following result which appears in [173, Lemma, §1.2, p. 14].
 
Proposition 2.2.18 Let E,  ·  E be a Banach space and let E ∗,  ·  E ∗ be its dual.
Consider a linear space F ⊆ E ∗ and denote by σ(E, F ) the topology induced on
E by the family of seminorms {p f : f ∈ F } defined by p f (e) := | f (e)| for every
e ∈ E and each f ∈ F . Similarly, consider the topology σ(E, F ), where F denotes
the closure of F in E ∗,  ·  E ∗ .
Then the topologies σ(E, F ) and σ(E, F ) coincide on the closed unit ball in E.

Here is the lemma alluded to earlier.

Lemma
 2.2.19 Let X be a Banach space and consider the closed unit ball in
L(X),  ·  L(X) denoted by B L(X) := {T ∈ L(X) : T  L(X) ≤ 1}. Then WOT
and UWOT coincide on B L(X) .

Proof We apply Proposition 2.2.18 with E := L(X) and F := Lω (X). For this
selection, we have (recall (2.2.345))
(L(X))∗
F = Lω (X) = L∗ (X) (2.2.356)

 notation in Proposition 2.2.18) σ L(X), L∗ (X) is the UWOT
and since (recall the
on L(X) and σ L(X), Lω (X) is the WOT on L(X), the desired conclusion
follows. 

We are now prepared to prove the following basic weak convergence result.
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 329

Theorem 2.2.20 Let X be a reflexive Banach space. Then the closed unit ball B L(X)
in L(X) is compact in WOT. Consequently, for every sequence {Tj } j ∈N of operators
in L(X) with the property that sup j ∈N Tj  L(X) < ∞, there exists a subsequence
{Tjk }k ∈N and an operator T ∈ L(X) such that, for all x ∈ X and all ξ ∈ X ∗ ,

X T jk x, ξ X ∗ −→ X T x, ξX ∗ as k → ∞. (2.2.357)

Proof Using the isometric isomorphism from Lemma 2.2.17, we may identify L(X)
∗
with L∗ (X) . From Alaoglu’s theorem we know that B L(X) is compact in UWOT.
Invoking Lemma 2.2.19, it follows that B L(X) is also compact in WOT, as wanted.
One particularly useful consequence of the above theorem is singled out in the
corollary below.

Corollary 2.2.21 Let (X, μ) be a measure space, fix an exponent p ∈ (1, ∞), and
suppose {Tj } j ∈N is a sequence of linear and bounded operators from L p (X, μ) into
itself with the property that sup j ∈N Tj  L(L p (X,μ)) < ∞.

Then there exists a subsequence {Tjk }k ∈N along with some T ∈ L L p (X, μ)
such that ∫ ∫
lim Tjk f · g dμ = T f · g dμ
k→∞ X X (2.2.358)

for all f ∈ L p (X, μ), g ∈ L p (X, μ),
where p := (1 − 1/p)−1 .

Proof This follows from Theorem 2.2.20 applied with L p (X, μ) in place of X
and Riesz’s duality theorem to the effect that (L p (X, μ))∗ may be identified with

L p (X, μ). 
Pressing on, in the next proposition we shall show that the only linear, bounded,
and local operators are those of pointwise multiplication by an essentially bounded
function.

Proposition 2.2.22 Let d ∈ (0, ∞) and suppose (X, ρ, μ) is a d-Ahlfors regular

quasi-metric space with the property that μ is a Borel-semiregular measure. Fix
p ∈ (0, ∞). Then for each given operator

T : L p (X, μ) −→ L p (X, μ) (2.2.359)

the following are equivalent:

(1) There exists b ∈ L ∞ (X, μ) such that T f = b · f for all f ∈ L p (X, μ), i.e., T is
the operator of pointwise multiplication by an essentially bounded function.
(2) The operator T is linear, bounded, and local in the sense that

supp(T f ) ⊆ supp f for each f ∈ L p (X, μ). (2.2.360)

Moreover, if either (1) or (2) holds, then b as in (1) is uniquely determined up to

a set of μ-measure zero and
330 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

b L ∞ (X,μ) = T  L p (X,μ)→L p (X,μ) . (2.2.361)

Before presenting the proof of this result, we make two comments. First, the
property described in (2.2.360) may be rephrased, equivalently, as saying that

for each function f ∈ L p (X, μ) we have

(2.2.362)
(T f )(x) = 0 for μ-a.e. x ∈ X \ supp f .

Second, as a consequence of Proposition 2.2.22, whenever two operators are

associated in L p (X, μ) with the same kernel (as in item (T3) of Proposition 2.2.5)
it follows that their difference is the operator of pointwise multiplication by an
essentially bounded function.
Here is the proof of Proposition 2.2.22.
Proof of Proposition 2.2.22 The implication (1) ⇒ (2) is immediate so we shall
focus on proving (2) ⇒ (1). Suppose T as in (2.2.359) is linear, bounded, and local.
The first claim we make is that whenever O1, O2 are given open subsets of X with
μ(O1 ) < ∞ and μ(O2 ) < ∞, it follows that
 
T1 O1 (x) = T1 O2 (x) for μ-a.e. x ∈ O1 ∩ O2 . (2.2.363)

To see why this is true, note that 1 O1 , 1 O2 ∈ L p (X, μ), the set
 O1 ∩ O2 is open, and
1 O1 − 1 O2 ≡ 0 on O  1 ∩ O 2 . Hence, O 1 ∩ O 2 ⊆ X \ supp 1 O1 − 1 O2 and since T
is local, we have T 1 O1 − 1 O2 (x) = 0 for μ-a.e. x ∈ X \ supp 1 O1 − 1 O2 . Thus,
T 1 O1 − 1 O2 (x) = 0 for μ-a.e. x ∈ O1 ∩ O2 . This and the linearity of T now imply
(2.2.363).  
To proceed, recall the dyadic grid D(X) = Q αk : k ∈ Z, k ≥ κX , α ∈ Ik from
[112, Proposition 7.5.4]. In particular, for each k ∈ Z with k ≥ κX there exists a
nullset Nk of μ such that 
X = α∈Ik Q αk  Nk . (2.2.364)
Next, fix k0 ∈ Z with k0 ≥ κX and define the function b : X → R by setting

T1Q k0 on Q αk0 for each α ∈ Ik0 ,
b := α (2.2.365)
0 on Nk0 .

Then b is μ-measurable and, due to (2.2.363) and the structure of the dyadic grid,
we actually have

b = T1Qαk at μ-a.e. point in Q αk ,

(2.2.366)
for each k ∈ Z with k ≥ κX and each α ∈ Ik .

In turn, this implies

2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 331

T1Qαk = b · 1Qαk at μ-a.e. point in X,

(2.2.367)
for each k ∈ Z with k ≥ κX and each α ∈ Ik .

To justify this, observe that on X \ supp(1Qαk ) both b · 1Qαk and T1Qαk are zero μ-a.e.
(for the latter we use the fact that T is a local operator). Having proved (2.2.366),
and since

supp(1Qαk ) = Q αk and 0 = μ(∂Q αk ) = μ Q αk \ Q αk , (2.2.368)

(cf. [112, (3.8.9)] and [112, (7.5.20)]) it follows that (2.2.367) is indeed valid. As a
consequence of (2.2.367) and the mapping properties of T we have
p
b ∈ Lloc (X, μ). (2.2.369)

Consequently, |b| p ∈ Lloc

1 (X, μ), and we may invoke [112, Proposition 7.4.4] (recall

that μ is Borel-semiregular) to conclude

⨏
|b(x)| = lim+
p
|b| p dμ at μ-a.e. x ∈ X. (2.2.370)
r→0 B ρ# (x,r)

Fix now an arbitrary open set O ⊆ X with μ(O) < ∞. From item (11) in [112,

Proposition  some μ-nullset N and some countable
 7.5.4] we know that there exist
set I O ⊆ (α, k) : k ∈ Z, k ≥ κX , α ∈ Ik , such that
 

O =  j ∈IO Q j  N (2.2.371)

where Q j := Q αk if j ∈ I O is such that j = (α, k). By invoking Lebesgue’s Dominated

Convergence Theorem (with pointwise convergence guaranteed by (2.2.371) and
domination by 1 O ) we further obtain that

1 O j = 1 O in L p (X, μ). (2.2.372)
j ∈IO

Since T is linear and continuous as an operator from L p (X, μ) into L p (X, μ), from
(2.2.372) and (2.2.367) we may conclude that, pointwise μ-a.e. on X we have
  
T1 O = T1 O j = b · 1Oj = b · T1 O j = b · 1 O . (2.2.373)
j ∈IO j ∈IO j ∈IO

In summary, the argument so far proves that

T1 O = b · 1 O at μ-a.e. point in X,
(2.2.374)
for every open set O ⊆ X with μ(O) < ∞.

Next, for each x ∈ X and each r ∈ (0, ∞), with ρ# denoting the regularized
version of the quasi-distance ρ (cf. [112, (7.1.18)]), we may write
332 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫ ∫ ∫
   
|b| p dμ = T1B (x,r)  p dμ ≤ T1B (x,r)  p dμ
ρ# ρ#
B ρ# (x,r) B ρ# (x,r) X
∫
 p
p
≤ T  L p (X,μ)→L p (X,μ) 1 B  dμ
ρ# (x,r)
X
p 
= T  L p (X,μ)→L p (X,μ) · μ Bρ# (x, r) , (2.2.375)

hence ⨏
p
|b| p dμ ≤ T  L p (X,μ)→L p (X,μ) . (2.2.376)
B ρ# (x,r)

Together, (2.2.376) and (2.2.370) imply

b ∈ L ∞ (X, μ) and b L ∞ (X,μ) ≤ T  L p (X,μ)→L p (X,μ) . (2.2.377)

Going further, the goal is to show that (2.2.374) continues to hold with O replaced
by any μ-measurable set E ⊆ X satisfying μ(E) < ∞. Fix such a set E. Since μ
is Borel-semiregular there exists some Borel set B ⊆ X such that μ(EB) = 0. In
turn, this implies

μ(B) < ∞ and 1E = 1B at μ-a.e. point in X. (2.2.378)

Consequently,
T1E = T1B at μ-a.e. point in X. (2.2.379)
In addition, by relying on [112, (3.4.44)] (whose applicability in the present setting
is ensured by [112, Lemma 3.4.13] and the fact that quasi-metric space (X, ρ, μ) is
d-Ahlfors regular), it is possible to find a sequence {O j } j ∈N of open sets in X with

B ⊆ O j for all j ∈ N and μ(O j \ B) −→ 0 as j → ∞. (2.2.380)

Fix some reference point x∗ ∈ X and introduce the sets

Bk := B ∩ Bρ# (x∗, k) for each k ∈ N, (2.2.381)

and
O jk := O j ∩ Bρ# (x∗, k) for each j, k ∈ N. (2.2.382)
Based on these definition and (2.2.380), for each j, k ∈ N we see that

O jk is open, Bk ⊆ O jk , and μ(O jk ) ≤ μ Bρ# (x∗, k) < ∞. (2.2.383)

In addition, since O jk \ Bk = (O j \ B) ∩ Bρ# (x∗, k) for each j, k ∈ N, it follows that

μ(O jk \ Bk ) ≤ μ(O j \ B) −→ 0 as j → ∞, for each fixed k ∈ N. (2.2.384)

As a consequence of (2.2.384) and the inclusion in (2.2.383), we conclude that

2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 333
p
1 O k − 1Bk  L p (X,μ) = μ(O jk \ Bk ) −→ 0 as j → ∞, (2.2.385)
j

for each fixed k ∈ N. Since also 1Bk → 1B in L p (X, μ) as k → ∞ (by Lebesgue’s

Dominated Convergence Theorem) and T is assumed to be bounded in the context
of (2.2.359), we may further infer that
 
T1B = lim T1Bk = lim lim T1 O k = lim lim b · 1 O k
k→∞ k→∞ j→∞ j k→∞ j→∞ j

= lim b · 1Bk = b · 1B in L p (X, μ), (2.2.386)

k→∞

where the third equality in (2.2.386) relies on (2.2.374). The leftmost and right-
most terms in (2.2.386) being equal in L p (X, μ) are equal μ-a.e. on X which, in
combination with (2.2.378) and (2.2.379), implies

T1E = T1B = b · 1B = b · 1E at μ-a.e. point in X. (2.2.387)

Let us record our progress: the argument so far proves that

T1E = b · 1B for every μ-measurable set E ⊆ X with μ(E) < ∞. (2.2.388)

Since T is linear, (2.2.388) further extends to simple functions, i.e.,


N
T s = b · s for every simple function s = λ j 1E j , (2.2.389)
j=1

where N ∈ N, the set E j ⊆ X is μ-measurable with μ(E j ) < ∞, and λ j ∈ C, for all
j ∈ {1, . . . , N }. By the density of simple functions in L p (X, μ) (cf. [112, (3.1.11)])
and the boundedness of T from L p (X, μ) to L p (X, μ), it follows that T f = b · f for
every f ∈ L p (X, μ). This completes the proof of the implication (2) ⇒ (1).
There remains to prove the final claim in the statement. If (1) holds, it is immediate
that a function b as in (1) is uniquely determined up to a set of μ-measure zero. In
addition,

T  L p (X,μ)→L p (X,μ) = sup T f  L p (X,μ)

 f  L p (X, μ) =1

= sup b · f  L p (X,μ) ≤ b L ∞ (X,μ), (2.2.390)

 f  L p (X, μ) =1

which when combined with (2.2.377) and the fact that b is unique up to a set of
μ-measure zero, yields the equality in (2.2.361). 

Our next proposition amounts to structure result to the effect that if T is a linear
operator satisfying both a weak-(p, p) bound as well as a strong (q, q) bound for
1 ≤ p < q < ∞ and which is associated with a kernel K in L p which is standard
kernel in the first variable and satisfies (2.2.332) then necessarily T is of the form
TPV + Mb for some essentially bounded function b.
334 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Proposition 2.2.23 Let d ∈ (0, ∞) and suppose (X, ρ, μ) is a d-Ahlfors regular

quasi-metric space with the property the quasi-distance ρ : X × X → [0, +∞) is
continuous in the product topology τρ × τρ , and such that μ is a complete Borel-
semiregular measure. Fix p ∈ [1, ∞) and let

T : L p (X, μ) −→ L p,∞ (X, μ) (2.2.391)

be a linear and bounded operator satisfying

∫
T f (x) = K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X \ supp f , (2.2.392)
X

where K is a standard kernel in the first variable

 (cf. Definition 2.2.4). Also, assume
that there exists q ∈ (p, ∞) such that T  L p (X,μ)∩L q (X,μ) extends to a linear and
bounded operator from L q (X, μ) into L q (X, μ).
Then for each integrability exponent r ∈ (p, q] there exist a sequence
 of num-
bers {ε j } j ∈N ⊆ (0, ∞) convergent to zero, an operator T(r) ∈ L L r (X, μ) and a
function b ∈ L ∞ (X, μ) such that
∫ ∫
lim Tε j f · g dμ = T(r) f · g dμ
j→∞ X X (2.2.393)

for all f ∈ L r (X, μ) and g ∈ L r (X, μ),

where r  ∈ (1, ∞) is such that 1/r + 1/r  = 1, and

T = T(r) + Mb in L r (X, μ), (2.2.394)

where Mb is the operator of pointwise multiplication by the function b.

Moreover, if in addition to the original hypotheses one also assumes that
∫
the limit lim+ K(x, y) dμ(y) exists for μ-a.e. x ∈ X, (2.2.395)
ε→0 1>ρ(x,y)>ε

then for each r ∈ (p, q] it follows that

TPV : L r (X, μ) −→ L r (X, μ), acting on each f ∈ L r (X, μ) by

∫
 (2.2.396)
TPV f (x) = lim+ K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X,
ε→0 y ∈X, ρ(x,y)>ε

is a well-defined, linear, bounded operator, and T(r) = TPV on L r (X, μ). In particular,
in this case, (2.2.394) becomes

T = TPV + Mb in L r (X, μ) for each r ∈ (p, q]. (2.2.397)

We remark that formula (2.2.397) yields the pointwise inequality

2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 335

|(T f )(x)| ≤ (Tmax f )(x) + C| f (x)| for μ-a.e. x ∈ X,

(2.2.398)
whenever f ∈ L r (X, μ) with r ∈ (p, q],

where C := b L ∞ (X,μ) ; compare with (2.2.38). We next present the proof of Propo-
sition 2.2.23.
Proof of Proposition 2.2.23 Proposition 2.2.7 ensures that the maximal operator
Tmax is bounded in the following contexts:

Tmax : L p (X, μ) −→ L p,∞ (X, μ), and (2.2.399)

Tmax : L r (X, μ) −→ L r (X, μ) for each r ∈ (p, q]. (2.2.400)


In particular, {Tε }ε>0 is a bounded sequence in L L r (X, μ) for each fixed exponent
r ∈ (p, q]. To proceed, pick a sequence {ε j } j ∈N such that ε j  0 as j → ∞ and
apply Corollary 2.2.21 to the sequence of operators {Tε j } j ∈N acting on L r (X, μ),
where r  ∈ (p, q] is arbitrary, fixed. This guarantees the existence of an operator
T(r) ∈ L L r (X, μ) and a subsequence {ε jk }k ∈N of {ε j } j ∈N such that
∫ ∫
lim Tε j k f · g dμ = T(r) f · g dμ
k→∞ X X (2.2.401)

for all f ∈ L r (X, μ) and g ∈ L r (X, μ),

where r  ∈ (1, ∞) is such that 1/r + 1/r  = 1. After relabeling, this proves (2.2.393).

Now pick f ∈ L r (X, μ) and g ∈ L r (X, μ) with δ := distρ (supp f , supp g) > 0.
Then, for each ε ∈ (0, δ/2), we have
∫ ∫ ∫ 
(Tε f )(x)g(x) dμ(x) = ρ(x,y)>ε
K(x, y) f (y) dμ(y) g(x) dμ(x)
X x ∈supp g
y ∈supp f
∫ ∫ 
= K(x, y) f (y) dμ(y) g(x) dμ(x)
x ∈supp g y ∈supp f
∫ ∫
= (T f )(x)g(x) dμ(x) = T f · g dμ. (2.2.402)
supp g X

In concert, (2.2.402) and (2.2.401) imply

∫ ∫
T f · g dμ = T(r) f · g dμ for all f ∈ L r (X, μ)
X X (2.2.403)

and g ∈ L r (X, μ) with distρ (supp f , supp g) > 0.

From (2.2.403) we further deduce (based on Lemma 2.2.10) that, for every function
f ∈ L r (X, μ),
T f = T(r) f pointwise μ-a.e. on X \ supp f . (2.2.404)
336 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Thus, T − T(r) ∈ L L r (X, μ) is a local operator. Proposition 2.2.22 then guarantees
the existence of some b ∈ L ∞ (X, μ) such that T −T(r) = Mb on L r (X, μ). This proves
(2.2.394).
As regards the claims in the last part of the statement, assume now that (2.2.395)
is also satisfied. Then, thanks to (2.2.400), we may invoke the last claim in part (2) of
Proposition 2.2.14 to conclude that for each r ∈ (p, q] the principal-value operator

TPV : L r (X, μ) −→ L r,∞ (X, μ), acting on each f ∈ L r (X, μ) by

∫
 (2.2.405)
TPV f (x) = lim+ K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X,
ε→0 y ∈X, ρ(x,y)>ε

is well defined, linear, and bounded. In view of the fact that |TPV f | ≤ Tmax f for
each f ∈ L r (X, μ), another appeal to (2.2.400) shows that (2.2.396) is indeed a
well-defined, linear, and bounded operator for each r ∈ (p, q].
In addition, Lebesgue’s Dominated Convergence Theorem plus the fact that
|Tε f | ≤ Tmax f for each f ∈ L r (X, μ) and each ε > 0 show that for every f ∈ L r (X, μ)
with r ∈ (p, q] we have

lim Tε f = TPV f in L r (X, μ). (2.2.406)

ε→0+

From (2.2.406) and (2.2.401) we then obtain that necessarily TPV = T(r) as operators
on L r (X, μ) for each r ∈ (p, q]. 

The weak boundedness property is introduced next.

Definition 2.2.24 Assume (X, ρ, μ) is a d-Ahlfors

 regular quasi-metric space for
some d ∈ (0, ∞), and fix a finite exponent β ∈ 0, (log2 Cρ )−1 . A linear operator
β  β ∗
T : 𝒞c (X, ρ) −→ 𝒞c (X, ρ) (2.2.407)
 β ∗ β
where 𝒞c (X, ρ) denotes the algebraic dual of 𝒞c (X, ρ), is said to satisfies the
weak boundedness property (WBP for short) provided there exists some constant
C ∈ [0, ∞) such that for each point xo ∈ X, each radius r > 0, and each functions
β
f , g ∈ 𝒞c (X, ρ) with

supp f ⊆ Bρ (xo, r), sup | f | + r β  f 𝒞. β (X,ρ) ≤ 1 (2.2.408)

c

and
supp g ⊆ Bρ (xo, r), sup |g| + r β g𝒞. β (X,ρ) ≤ 1, (2.2.409)
c

it follows that  
 # $  
 
(𝒞βc (X,ρ))∗ T f , g β  ≤ C μ Bρ (xo, r) . (2.2.410)
 𝒞 c (X,ρ) 
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 337

With each given antisymmetric kernel is possible to associate an operator mapping

“test functions” into “distributions” of the sort described below.

Definition 2.2.25 Let (X, ρ, μ) be a d-Ahlfors regular quasi-metric space for some
d ∈ (0, ∞), and assume K : X × X \ diag(X) → R is a μ ⊗ μ-measurable function
which is antisymmetric, in the sense that

K(y, x) = −K(x, y) for all x, y ∈ X with x  y, (2.2.411)

and with the property that there exists some constant C ∈ [0, ∞) such that

|K(x, y)| ≤ C ρ(x, y)−d for all x, y ∈ X with x  y. (2.2.412)


Finally, fix a finite exponent β ∈ 0, (log2 Cρ )−1 .
In this setting, consider the operator TK canonically induced by (or, associated
with) the kernel K in the distributional sense as the mapping
β  β ∗
TK : 𝒞c (X, ρ) −→ 𝒞c (X, ρ) (2.2.413)

defined as
# $
β
(𝒞 c (X,ρ))∗
TK f , g β
𝒞 c (X,ρ)
∫ ∫
1
:= K(x, y) f (y)g(x) − f (x)g(y) dμ(y) dμ(x), (2.2.414)
2 X X

β
for all f , g ∈ 𝒞c (X, ρ).

That the above definition is indeed meaningful is clarified by the next proposition.

Proposition 2.2.26 Let (X, ρ, μ) be a d-Ahlfors regular quasi-metric space for some
d ∈ (0, ∞), and assume K : X × X \ diag(X) → R is a μ ⊗ μ-measurable function
which is antisymmetric,
 in the sense of (2.2.411), and satisfies (2.2.412). Also, fix a
finite exponent β ∈ 0, (log2 Cρ )−1 .
Then the operator TK canonically induced by (or, associated with) the kernel K
in the distributional sense as in (2.2.413)-(2.2.414) is a well-defined linear mapping
β  β ∗
TK : 𝒞c (X, ρ) −→ 𝒞c (X, ρ) (2.2.415)

which satisfies the weak boundedness property (cf. Definition 2.2.24).

β
Proof Fix a point xo ∈ X, a radius r > 0, and two functions f , g ∈ 𝒞c (X, ρ)
satisfying (2.2.408)-(2.2.409). Then we may rely on [112, (7.2.5)] to estimate
338 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫ ∫
 
|K(x, y)|  f (y)g(x) − f (x)g(y) dμ(y) dμ(x)
X X
∫ ∫  ρ (x, y)  β
C #
≤ dμ(y) dμ(x)
B ρ# (x o ,r) B ρ# (x o ,r) ρ# (x, y)d r
 
≤ C μ Bρ# (xo, r) ≤ C μ Bρ (xo, r) , (2.2.416)

from which all desired conclusions follow. 

It turns out that, in an appropriate setting, the operator TK canonically induced by

an antisymmetric kernel K in the distributional sense is associated with this kernel
in L p in the sense described below.

Lemma 2.2.27 Let (X, ρ, μ) be a d-Ahlfors regular quasi-metric space for some
d ∈ (0, ∞), with the property that μ is a locally finite Borel-semiregular
 measure on
X (cf. [112, Definition 3.4.3]). Also, fix a finite exponent β ∈ 0, (log2 Cρ )−1 . Assume
K : X × X \ diag(X) → R is a μ ⊗ μ-measurable function which is antisymmetric, in
the sense of (2.2.411), and satisfies (2.2.412). Suppose that, for some p ∈ (1, ∞), the
operator TK canonically induced by the kernel K in the distributional sense (as in
(2.2.413)-(2.2.414)) extends to a linear operator mapping L p (X, μ) into Lloc1 (X, μ).

Then for each f ∈ L (X, μ) one has13

p

∫
(TK f )(x) = K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X \ supp f . (2.2.417)
X

Proof Thanks to Lemma 2.2.10, the assignment


1 (X, μ)  φ → Λ ∈ 𝒞β (X, ρ) ∗ defined by
Lloc φ c
∫
% &
β
(𝒞 (X,ρ))∗
Λ φ , ψ β
𝒞 (X,ρ)
:= φψ dμ (2.2.418)
c c
X
β
1 (X, μ) and ψ ∈ 𝒞 (X, ρ),
for all φ ∈ Lloc c


1 (X, μ) into 𝒞β (X, ρ) ∗ .
is injective. This amounts to embedding Lloc c
β
Fix now an arbitrary f ∈ 𝒞c (X, ρ) ⊆ L p (X, μ). Then for each
β 
g ∈ 𝒞c (X, ρ) with distρ supp f , supp g > 0, (2.2.419)

in view of the embedding (2.2.418) and (2.2.414) we may write

13 with supp f understood in the sense of [112, Definition 3.8.3]

2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 339
∫ # $
(TK f )g dμ = (𝒞β (X,ρ))∗ TK f , g β
X c 𝒞 c (X,ρ)
∫ ∫
1
= K(x, y) f (y)g(x) − f (x)g(y) dμ(y) dμ(x)
2 X X
∫ ∫
= K(x, y) f (y)g(x) dμ(y) dμ(x), (2.2.420)
X X

where the last equality is obtained by naturally breaking up the previous integral,
interchanging the roles of x, y, and using the fact that K is antisymmetric. In turn,
(2.2.420) implies
∫  ∫ 
(TK f )(x) − K(x, y) f (y) dμ(y) g(x) dμ(x) = 0 (2.2.421)
X X

for each function g as in (2.2.419). Granted this, Lemma 2.2.10 applies and gives
β
that (2.2.417) holds whenever f ∈ 𝒞c (X, ρ). With this in hand, the same type
of argument which has established (2.2.145) (see (2.2.146)-(2.2.160)), now using
β
the fact that 𝒞c (X, ρ) is dense in L p (X, μ) (cf. [112, Proposition 7.4.4] with ρ
replaced by ρ# from [112, Theorem 7.1.2]) then proves that (2.2.417) holds for each
f ∈ L p (X, μ). 

One way to summarize the essence of our next result is to say that “TK is nice if
and only if TPV is nice.”

Proposition 2.2.28 Assume (X, ρ, μ) is a d-Ahlfors regular quasi-metric space for

some d ∈ (0, ∞) such that the quasi-distance ρ : X × X → [0, +∞) is continuous
in the product topology τρ × τρ , and with the property
 that μ is a complete Borel-
semiregular measure. Fix a finite exponent β ∈ 0, (log2 Cρ )−1 along with some
p ∈ (1, ∞). Also, suppose

K : X × X \ diag(X) −→ R (2.2.422)

is a kernel which is standard in the first variable (cf. Definition 2.2.4), is antisym-
metric (in the sense of (2.2.411)), and such that
∫
the limit lim+ K(x, y) dμ(y) exists for μ-a.e. x ∈ X. (2.2.423)
ε→0
y ∈X
1>ρ(x,y)>ε

Finally, recall the operator TK canonically induced by (or, associated with) the
kernel K in the distributional sense as in (2.2.413)-(2.2.414).
Then the following statements are equivalent:
(1) The operator TK extends to a linear and bounded operator mapping L p (X, μ)
into L p,∞ (X, μ) for some p ∈ (1, ∞).
340 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

(2) The operator TK induces well-defined, linear, and bounded mappings in the
following contexts:

TK : L q (X, μ) −→ L q (X, μ) for each q ∈ (1, ∞),

(2.2.424)
TK : L 1 (X, μ) −→ L 1,∞ (X, μ).

(3) There exists p ∈ (1, ∞) such that principal-value operator assigning to each
β
f ∈ 𝒞c (X, ρ) the function
∫

TPV f (x) := lim+ K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X
ε→0 y ∈X, ρ(x,y)>ε
(2.2.425)
extends to a linear and bounded mapping from L p (X, μ) into L p,∞ (X, μ).
(4) There exists p ∈ (1, ∞) such the that principal-value operator

TPV : L p (X, μ) −→ L p,∞ (X, μ), acting on each f ∈ L p (X, μ) by

∫

TPV f (x) = lim+ K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X,
ε→0 y ∈X, ρ(x,y)>ε
(2.2.426)
is well defined, linear, and bounded.
(5) The operator TPV induces well-defined, linear, and bounded mappings in the
following contexts:

TPV : L q (X, μ) −→ L q (X, μ) for each q ∈ (1, ∞),

(2.2.427)
TPV : L 1 (X, μ) −→ L 1,∞ (X, μ).

Moreover, the veracity of either (hence all) of the above conditions implies that

TK = TPV on L q (X, μ) for each q ∈ [1, ∞). (2.2.428)

Proof Assume (1). Then Lemma 2.2.27 implies that (2.2.417) holds for each function
f ∈ L p (X, μ). In turn, this permits us to invoke item (i) of Proposition 2.2.8 and
Corollary 2.2.11. The former gives that

TK : L 1 (X, μ) −→ L 1,∞ (X, μ) linearly and boundedly, (2.2.429)

while the latter presently gives two things. First,

TK : L q (X, μ) −→ L q (X, μ)
(2.2.430)
linearly and boundedly for each q ∈ (1, ∞),

and, second, the maximal operator Tmax induces a sub-linear, bounded, and contin-
uous mapping in the context
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 341

Tmax : L q (X, μ) −→ L q (X, μ) for each q ∈ (1, ∞). (2.2.431)

Collectively, (2.2.429)-(2.2.430) prove the claim made in the current item (2). Since,
obviously, (2) ⇒ (1), we ultimately conclude that (1) ⇔ (2). Granted (2.2.431), from
Proposition 2.2.14 we conclude that (1) implies either of the items (3), (4), (5).
Proposition 2.2.14 also shows that the items (3), (4), (5) are equivalent to one
another.
β
Assume next that (3)-(5) hold. For each f , g ∈ 𝒞c (X, ρ) write
# $
β
(𝒞 (X,ρ))∗ TK f , g β
c 𝒞 c (X,ρ)
∫ ∫
1
= K(x, y) f (y)g(x) − f (x)g(y) dμ(y) dμ(x)
2 X X
∫ ∫
1
= lim+ K(x, y) f (y)g(x) − f (x)g(y) dμ(y) dμ(x)
ε→0 2
X y ∈X, ρ(x,y)>ε
∫ ∫
= lim+ K(x, y) f (y)g(x) dμ(y) dμ(x)
ε→0
X y ∈X, ρ(x,y)>ε
∫
= lim+ (Tε f (x)g(x) dμ
ε→0 X
∫

= TPV f (x)g(x) dμ
X
# $
= (𝒞β (X,ρ))∗ TPV f , g β
. (2.2.432)
c 𝒞 c (X,ρ)

Above, the first equality comes from (2.2.414), the second equality is a consequence
of Lebesgue’s Dominated Convergence Theorem and (2.2.416), the third equality is
obtained by naturally breaking up the previous integral, interchanging the roles of
x, y, and using the fact that K is antisymmetric, the fourth equality uses (2.2.33), the
fifth equality is implied by (2.2.304), and the last equality is seen from (2.2.418).
β
In turn, (2.2.432) shows that for each f ∈ 𝒞c (X, ρ) we have TK f = TPV f as
 β ∗
functionals in 𝒞c (X, ρ) . In view of this, the density result in item (5) of [112,
Proposition 7.4.4], and the fact that TPV induces linear, bounded, and continuous
mappings in the contexts described in (2.2.427), we deduce that the operator TK
induces a well-defined, linear, and bounded mapping from L q (X, μ) into itself for
each q ∈ (1, ∞), and that said extension agrees with TPV . This proves that (3)-
(5) imply (1) and that TK = TPV on L q (X, μ) for each q ∈ (1, ∞). Finally, that
TK f = TPV f for each f ∈ L 1 (X, μ) follows by density and the continuity of the
operators involved from the space L 1 (X, μ) into L 1,∞ (X, μ). 
As a preamble to the statement of the T(1) theorem, given a little further below,
we make some definitions.
342 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Definition 2.2.29 Let (X, ρ, μ) be a d-Ahlfors regular quasi-metric space for some
d ∈ (0, ∞), and fix a β ∈ 0, (log2 Cρ )−1 . Assume K : X × X \ diag(X) → R is a
standard kernel in the second variable (cf. Definition 2.2.4). In this context, a linear
operator  β
β ∗
T : 𝒞c (X, ρ) −→ 𝒞c (X, ρ) (2.2.433)
β
is said to be associated with the kernel K in 𝒞c (X, ρ) (or, simply, in a weak sense)
provided
# $ ∫ ∫
β
(𝒞 (X,ρ))∗ T f , g β
= K(x, y) f (y)g(x) dμ(y) dμ(x) (2.2.434)
c 𝒞 c (X,ρ) X X

for all 
β
f , g ∈ 𝒞c (X, ρ) with distρ supp f , supp g > 0. (2.2.435)
Also, define the weak transpose T  of T in (2.2.433) as the mapping
β  β ∗
T  : 𝒞c (X, ρ) −→ 𝒞c (X, ρ) defined as
# $ # $
T  f, g = T g, f (2.2.436)
β ∗ β ∗
(𝒞 (X,ρ))
c β (𝒞 (X,ρ))
𝒞 c (X,ρ) c β
𝒞 c (X,ρ)
β
for all functions f , g ∈ 𝒞c (X, ρ).

It follows from the above considerations that:

in the context of Definition 2.2.29, if T is weakly associated with
K then T  is weakly associated with the kernel K  , defined as (2.2.437)
K  (x, y) := K(y, x) for all points x, y ∈ X with x  y.
Let us also observe that, in the context of Definition 2.2.29, for each
∫
β
f ∈ 𝒞c (X, ρ) with f dμ = 0 (2.2.438)
X
 β ∗
it is meaningful to extend the action of the functional T f ∈ 𝒞c (X, ρ) to any
function
g ∈ 𝒞β (X, ρ) such that there exists ε > 0 with
 (2.2.439)
g ≡ 0 on {x ∈ X : distρ x, supp f ) < ε}
by setting, for some fixed reference point y0 ∈ X,
∫ ∫
% &
T f , g := K(x, y) − K(x, y0 ) f (y)g(x) dμ(y) dμ(x)
X X (2.2.440)
for each f as in (2.2.438) and each g as in (2.2.439).

Indeed, the fact that the kernel K is standard in the second variable plus [112, (7.2.5)]
ensure that the integral above is absolutely convergent, while the vanishing moment
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 343

condition of f guarantees that the definition in (2.2.440) is independent of the choice

of the reference point y0 ∈ X.
Definition 2.2.30 Let (X, ρ, μ) be a d-Ahlfors
 regular quasi-metric space for some
d ∈ (0, ∞), and fix a finite exponent β ∈ 0, (log2 Cρ )−1 . Consider a linear operator
β  β ∗
T : 𝒞c (X, ρ) −→ 𝒞c (X, ρ) (2.2.441)

which is weakly associated with a kernel K : X × X \ diag(X) → R which is

standard in the first variable (cf. Definition 2.2.29). In this context, define T(1) as
the functional acting on each function
∫
β
f ∈ 𝒞c (X, ρ) with f dμ = 0 (2.2.442)
X

according to14
 # $ % &
T(1), f := (𝒞β (X,ρ))∗ T φ, f β
+ T  f , (1 − φ) , (2.2.443)
c 𝒞 c (X,ρ)

where
β
φ ∈ 𝒞c (X, ρ) is such that φ ≡ 1 near supp f . (2.2.444)
In definition (2.2.443), the last pairing is understood in the sense of (2.2.440),
with T  now playing the role of the operator T (note that T  is weakly associated
with K  (x, y) := K(y, x), which is a standard kernel in the second variable; cf.
(2.2.437)), and 1 − φ playing the role of g. It turns out that the above definition of is
independent of the auxiliary function φ.
Continue to work in the setting of Definition 2.2.30, and let us temporarily assume
β β
that X is ρ-bounded. Then 𝒞c (X, ρ) = 𝒞β (X, ρ). In particular, 1 ∈ 𝒞c (X, ρ), so
β ∗
T1 ∈ 𝒞c (X, ρ) . Also, for each f as in (2.2.442) and each φ as in (2.2.444) we
may write
 # $ # $
T(1), f = (𝒞β (X,ρ))∗ T φ, f β + (𝒞β (X,ρ))∗ T  f , (1 − φ) β
𝒞 (X,ρ) 𝒞 (X,ρ)
# $ # $
= (𝒞β (X,ρ))∗ T φ, f + (𝒞β (X,ρ))∗ T(1 − φ), f
𝒞β (X,ρ) 𝒞β (X,ρ)
# $
= (𝒞β (X,ρ))∗ T1, f (2.2.445)
𝒞β (X,ρ)

thanks to% (2.2.443),

& the consistency of the pairing in (2.2.440) with the pairing
β
(𝒞 (X,ρ))∗ ·, · 𝒞β (X,ρ)
, and the definition of the transpose given in (2.2.436). We can
summarize this observation as follows:
% &
if X is ρ-bounded, then T(1) = (𝒞β (X,ρ))∗ T1, · 𝒞β (X,ρ) as
 ∫  (2.2.446)
β
functionals on f ∈ 𝒞c (X, ρ) : X f dμ = 0 .

14 here, T (1), f simply means the action of the functional T (1) on the function f
344 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Let us assign a precise meaning to the symbol “T(1) ∈ BMO(X, μ).”

Definition 2.2.31 In the context of Definition 2.2.30, say that “T(1) ∈ BMO(X, μ)”
provided there exists a function b ∈ BMO(X, μ) with the property that
∫ # $ % &
b f dμ = (𝒞β (X,ρ))∗ T φ, f β + T  f , (1 − φ) (2.2.447)
X c 𝒞 c (X,ρ)

whenever ∫
β
f ∈ 𝒞c (X, ρ) satisfies f dμ = 0 and
X (2.2.448)
β
φ ∈ 𝒞c (X, ρ) is such that φ ≡ 1 near supp f .
β
∫
Owing to the existence of a function fo ∈ 𝒞c (X, ρ) with X fo dμ  0 (cf., e.g.,
[117]), it follows that
 β ∗ % &
if Λ ∈ 𝒞c (X, ρ) is such that (𝒞β (X,ρ))∗ Λ, f 𝒞β (X,ρ) = 0
c c
∫
β
for each function f ∈ 𝒞c (X, ρ) with f dμ = 0 then (2.2.449)
X
∫
β
there exists c ∈ R so that Λ f = c f for all f ∈ 𝒞c (X, ρ).
X

In view of this property, it follows that

the function b ∈ BMO(X, μ) satisfying (2.2.447)-(2.2.448)

(2.2.450)
is in fact uniquely determined up to additive constants.
In particular, from this, (2.2.446), and (2.2.418) we see that

if X is ρ-bounded and “T(1) ∈ BMO(X, μ)” then the functional

T1 agrees (up to additive constants, in the sense of (2.2.418)) with (2.2.451)
the function b ∈ Lloc
1 (X, μ) doing the job in (2.2.447).

Conversely, from (2.2.446) and (2.2.447) with φ ≡ 1 we see that

whenever X is ρ-bounded and T1 ∈ BMO(X, μ),

(2.2.452)
the function b := T1 does the job in (2.2.447).
Collectively, (2.2.451) and (2.2.452) prove that

if X is ρ-bounded then “T(1) ∈ BMO(X, μ)” (cf. the sense given in

(2.2.453)
Definition 2.2.31) if and only if T1 ∈ BMO(X, μ) (as in (2.2.418)).

The stage is set to state the following version of the T(1) theorem of G. David
and J.-L. Journé (cf. the original [33], or [104] for the Euclidean setting, and [8,
Theorem 12.2, p. 291], [18], [34] for the setting of spaces of homogeneous type).
2.2 Singular Integrals on Ahlfors Regular Quasi-Metric Spaces 345

Theorem 2.2.32 Let (X, ρ, μ) be a d-Ahlfors regular quasi-metric space for some
d ∈ (0, ∞) with
 the property that μ is a Borel-semiregular measure. Fix a finite
exponent β ∈ 0, (log2 Cρ )−1 and suppose
β  β ∗
T : 𝒞c (X, ρ) −→ 𝒞c (X, ρ) (2.2.454)

is a linear operator associated in a weak sense with a kernel

K : X × X \ diag(X) −→ R (2.2.455)

which is standard in both variables (cf. Definition 2.2.29 and Definition 2.2.4).
Then T extends to a linear and bounded operator from L 2 (X, μ) into itself if and
only if the following three conditions are satisfied:
(i) T satisfies the weak boundedness property (aka WBP; cf. Definition 2.2.24);
(ii) T(1) ∈ BMO(X, μ) (in the sense of Definition 2.2.31);
(iii) T  (1) ∈ BMO(X, μ) (cf. (2.2.436) and Definition 2.2.31).

Remark 2.2.33 According to Proposition 2.2.26 and (2.2.437), in the case when
K is antisymmetric (cf. (2.2.411)), conditions (i)-(iii) above simply reduce to the
demand that T(1) ∈ BMO(X, μ) (in the sense of Definition 2.2.31).

The final result elaborates on the functional analytic properties of TK , the operator
canonically induced by an antisymmetric kernel K in the distributional sense, in light
of the T(1) theorem and our earlier results in this section.

Proposition 2.2.34 Assume (X, ρ, μ) is a d-Ahlfors regular quasi-metric space for

some value d ∈ (0, ∞) with the property that X is ρ-bounded, the quasi-distance
ρ : X × X → [0, +∞) is continuous in the product topology τρ × τρ , and μ is a
complete Borel-semiregular measure. Fix an exponent β ∈ 0, (log2 Cρ )−1 along
with some p ∈ (1, ∞). Also, suppose

K : X × X \ diag(X) −→ R (2.2.456)

is a kernel which is standard in the first variable (cf. Definition 2.2.4) as well as
antisymmetric (in the sense of (2.2.411)). Bring in the operator TK canonically
induced by the kernel K in the distributional sense, i.e.,
 ∗
TK : 𝒞β (X, ρ) −→ 𝒞β (X, ρ) (2.2.457)

defined for all f , g ∈ 𝒞β (X, ρ) as

# $
β
(𝒞 (X,ρ))∗ TK f , g β𝒞 (X,ρ)
∫ ∫
1
:= K(x, y) f (y)g(x) − f (x)g(y) dμ(y)dμ(x). (2.2.458)
2 X X
346 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Then TK extends to a linear and bounded operator from the space L 2 (X, μ) into
itself if and only if T1 ∈ BMO(X, μ) (in the sense of Definition 2.2.31), i.e., if and
only if there exists b ∈ BMO(X, μ) such that
∫ ∫ ∫
1
bg dμ = K(x, y) g(x) − g(y) dμ(y) dμ(x) (2.2.459)
X 2 X X

for all g ∈ 𝒞β (X, ρ).

Moreover, if either of the aforementioned conditions holds then:

(a) The operator TK induces well-defined, linear, and bounded mappings in the
following contexts:

TK : L p (X, μ) −→ L p (X, μ) for each p ∈ (1, ∞), (2.2.460)

TK : L 1 (X, μ) −→ L 1,∞ (X, μ). (2.2.461)

(b) The maximal operator Tmax is sub-linear and bounded in the following contexts:

Tmax : L p (X, μ) −→ L p (X, μ) for each p ∈ (1, ∞), (2.2.462)

Tmax : L 1 (X, μ) −→ L 1,∞ (X, μ). (2.2.463)

(c) For each ε > 0 the truncated operator Tε is linear and bounded in the following
contexts:

Tε : L p (X, μ) −→ L p (X, μ) for each p ∈ (1, ∞), (2.2.464)

Tε : L 1 (X, μ) −→ L 1,∞ (X, μ), (2.2.465)

with operator norm bounds uniform with respect to ε ∈ (0, ∞).

(d) Under the additional assumption that
∫
the limit lim+ K(x, y) dμ(y) exists for μ-a.e. x ∈ X, (2.2.466)
ε→0
y ∈X
1>ρ(x,y)>ε

it follows that the principal-value operator

∫

TPV f (x) := lim+ K(x, y) f (y) dμ(y) for μ-a.e. x ∈ X,
ε→0 y ∈X, ρ(x,y)>ε
(2.2.467)
is well defined, linear, and bounded in the following contexts:

TPV : L p (X, μ) −→ L p (X, μ) for each p ∈ (1, ∞), (2.2.468)

TPV : L 1 (X, μ) −→ L 1,∞ (X, μ). (2.2.469)

2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 347

In addition, TK from (2.2.460)-(2.2.461) coincides with TPV from (2.2.468)-

(2.2.469).

Proof The first claim in the statement is a consequence Theorem 2.2.32, Re-
mark 2.2.33, (2.2.453), and (2.2.452). Assume next that TK extends to a linear
and bounded operator from L 2 (X, μ) into itself. Much as in the proof of Propo-
sition 2.2.28 this implies that the operators (2.2.460)-(2.2.462) are well defined
and bounded. In addition from (2.2.462) and Proposition 2.2.13 we also see that
(2.2.463) is well defined and bounded. This takes care of items (a)-(b). In turn,
(2.2.462)-(2.2.463) readily imply the claims in item (c). Finally, the claims in item
(d) are consequences of what we have proved so far and Proposition 2.2.28. 

2.3 Principal Value Singular Integral Operators on Uniformly

Rectifiable Sets

The results in this section build on the work of many predecessors, including
Calderón, Zygmund, Mikhlin, Coifman, McIntosh, Meyer, David, Jerison, Kenig,
Semmes, Stein, and a multitude of others. Among other things, here we expand,
refine, and sharpen work in [63] by presently placing more economical demands on
the ambient geometry.
We begin by recording a useful result, pertaining to the manner in which Euclidean
singular integrals are truncated, which appears in [64, Proposition B.2, p. 163].

Proposition 2.3.1 Let A : Rn → Rm be some vector-valued Lipschitz function, and

assume that F : Rm → R is an odd function of class 𝒞 N , for some sufficiently large

integer N = N(m). Also, suppose B : Rn → Rm is a bi-Lipschitz function15, and
pick p ∈ (1, ∞). Then for each given function f ∈ L p (Rn, L n ) the limit
∫  A(x) − A(y) 
1
lim+ F f (y) dy (2.3.1)
{y ∈R n : |B(x)−B(y) |>ε } |x − y| |x − y|
ε→0 n

exists at L n -a.e. point x ∈ Rn . Moreover, this limit is independent of the choice of

the function B which, in particular, implies that for each given f ∈ L p (Rn, L n ) the
limit (2.3.1) is equal to
∫  A(x) − A(y) 
1
lim+ F f (y) dy (2.3.2)
{y ∈R n : |x−y |>ε } |x − y| |x − y|
ε→0 n

at L n -a.e. x ∈ Rn .

Our first main result in this section is Theorem 2.3.2, which builds on the work of
G. David in [31] and refines [63, Propositions 3.18-3.19, pp. 2639-3240] as well as

15 in the sense that |B(x) − B(y)| ≈ |x − y | uniformly in x, y ∈ R n (which, in the present context,
entails m ≥ n)
348 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

the portion of Theorem 3.33 on p. 2669 in [63] dealing with the pointwise existence
of principal-value singular integral operators (by allowing the underlying UR set
Σ ⊆ Rn to be more general than the topological boundary of an open set Ω ⊆ Rn
whose boundary is a UR set satisfying H n−1 (∂Ω \ ∂∗ Ω) = 0, as considered in the
aforementioned results in [63], and by allowing a larger class of functions).
Theorem 2.3.2 For each n ∈ N with n ≥ 2 there exists N = N(n) ∈ N with the
following significance. Suppose Σ ⊆ Rn is a UR set, abbreviate σ := H n−1 Σ, and
consider a kernel
k ∈ 𝒞 N (Rn \ {0}) satisfying k(−x) = −k(x) as well as
(2.3.3)
k(λ x) = λ1−n k(x) for all x ∈ Rn \ {0} and λ ∈ (0, ∞).

In this setting, define the maximal singular integral operator Tmax acting on functions
 σ(x) 
f ∈ L 1 Σ, (2.3.4)
1 + |x| n−1
according to  
(Tmax f )(x) := sup (Tε f )(x), ∀x ∈ Σ, (2.3.5)
ε>0

where, for each ε > 0, the truncated singular integral operator Tε is given by16
∫
(Tε f )(x) := k(x − y) f (y) dσ(y), ∀x ∈ Σ. (2.3.6)
y ∈Σ
|x−y |>ε

Then the following statements are true.

(1) The maximal operator is a well-defined, bounded17 and continuous mapping
 σ(x)  1,∞
Tmax : L 1 Σ, −→ Lloc (Σ, σ), (2.3.7)
1 + |x| n−1

which for each p ∈ (1, ∞) induces a well-defined, bounded and continuous

mapping
 σ(x) 
p p
Tmax : Lloc (Σ, σ) ∩ L 1 Σ, −→ Lloc (Σ, σ). (2.3.8)
1 + |x| n−1
Moreover, Tmax is sub-linear, in the sense that
 
Tmax ( f + g) ≤ Tmax f + Tmax g, ∀ f , g ∈ L 1 Σ, 1+σ(x)
|x | n−1 ,
  (2.3.9)
Tmax (λ f ) = |λ| Tmax f , ∀ f ∈ L 1 Σ, 1+σ(x)
|x | n−1
, ∀λ ∈ R.

16 the fact that the integral in the right-hand side of (2.3.6) is absolutely convergent is a consequence
of (2.3.4)

17 i.e., it maps bounded subsets of L 1 Σ, 1+|σ(x)
x | n−1
1,∞
into bounded subsets of Lloc (Σ, σ)
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 349
 
As a consequence, given any f , g ∈ L 1 Σ, 1+σ(x)
|x | n−1
one has
  
(Tmax f )(x) − (Tmax g)(x) ≤ Tmax ( f − g) (x) at each point
(2.3.10)
x ∈ Σ where (Tmax f )(x) < +∞ and (Tmax g)(x) < +∞.

(2) For each p ∈ [1, ∞) there exists a constant C ∈ (0, ∞) depending only on n, p,
and the UR character of Σ, such that for every function f ∈ L p (Σ, σ) one has
  
Tmax f  L p (Σ,σ) ≤ C  k S n−1 𝒞 N (S n−1 )  f  L p (Σ,σ) if 1 < p < ∞, (2.3.11)
  
Tmax f  L 1,∞ (Σ,σ) ≤ C  k S n−1 𝒞 N (S n−1 )  f  L 1 (Σ,σ) if p = 1. (2.3.12)

As a consequence of this and (2.3.9)-(2.3.10), it follows that

Tmax : L p (Σ, σ) −→ L p (Σ, σ) if 1 < p < ∞, (2.3.13)

Tmax : L 1,∞ (Σ, σ) −→ L 1 (Σ, σ) if p = 1, (2.3.14)

are Lipschitz (hence, continuous) sub-linear mappings.

(3) One has that
 
for each function f ∈ L 1 Σ, 1+σ(x)
|x | n−1
the limit
(T f )(x) := lim+ (Tε f )(x) exists for σ-a.e. x ∈ Σ, (2.3.15)
ε→0
and the function T f thus defined is σ-measurable.

In fact, T is a well-defined, linear, and continuous18 operator in the context

 σ(x)  1,∞
T : L 1 Σ, −→ Lloc (Σ, σ), (2.3.16)
1 + |x| n−1

which for each p ∈ (1, ∞) induces a well-defined, linear and continuous mapping
 σ(x) 
p p
T : Lloc (Σ, σ) ∩ L 1 Σ, −→ Lloc (Σ, σ). (2.3.17)
1 + |x| n−1
Moreover, the operators induced by the action of T from (2.3.16) on ordinary
Lebesgue spaces (cf. [112, (7.7.106)]), i.e.,

18 This means that for each

 point xo ∈ Σ  and each radius r ∈ (0, ∞) there exists C x o , r ∈ (0, ∞)
  
with the property that (T f )Σ∩B(x o , r )  1,∞ ≤ C x o , r  f  1  σ(x) for every
L (Σ∩B(x o , r ), σ) L Σ,
1+| x | n−1
 σ(x)
function f ∈ L 1 Σ, 1+| x | n−1
.
350 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

T : L p (Σ, σ) −→ L p (Σ, σ), p ∈ (1, ∞), (2.3.18)

T : L 1 (Σ, σ) −→ L 1,∞ (Σ, σ), (2.3.19)

are well-defined, linear, bounded, and compatible with one another, with oper-
ator norms19
  
T  L p (Σ,σ)→L p (Σ,σ) ≤ Cp  k S n−1 𝒞 N (S n−1 ) if 1 < p < ∞, (2.3.20)
  
T  L 1 (Σ,σ)→L 1,∞ (Σ,σ) ≤ C  k S n−1 𝒞 N (S n−1 ) . (2.3.21)

As a consequence of (2.3.21) and [112, Lemma 6.2.3], for each θ ∈ (0, 1)

there exists a constant C = C(Σ, θ, n, k) ∈ (0, ∞) with the property that for any
σ-measurable set E ⊆ Σ and any function f ∈ L 1 (Σ, σ) one has
∫
|T f | θ dσ ≤ Cσ(E)1−θ  f  Lθ 1 (Σ,σ) . (2.3.22)
E

Finally, for each r ∈ (0, ∞) there exists C ∈ (0, ∞), which depends only on n,
the UR constants of Σ, and r, with the property that the following Cotlar-type
inequality holds:
  1/r   
(Tmax f )(x) ≤ C M Σ (|T f | r ) (x) + Ck S n−1 𝒞 N (S n−1 ) M Σ f (x)
 
for every function f ∈ L 1 Σ, 1+σ(x)|x | n−1 and every point x ∈ Σ,
(2.3.23)
where M Σ is the Hardy-Littlewood maximal operator on Σ.
(4) For each fixed function f ∈ L p (Σ, σ) with 1 < p < ∞ one has

Tε f −→ T f in L p (Σ, σ) as ε → 0+ . (2.3.24)

(5) One has that

the (real) transpose of the operator (2.3.18) is
 
−T : L p (Σ, σ) → L p (Σ, σ), where the expo-
 (2.3.25)
nent p ∈ (1, ∞) is the Hölder conjugate of the
exponent p ∈ (1, ∞).
In addition, given any two functions

19 The dependence of the constant C p appearing in (2.3.20) on the exponent p follows from
interpolating between p = 1 (cf. (2.3.21)), p = 2, and duality: ultimately, C p = O(p/(p − 1)) as
p → ∞ and as p → 1.
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 351
 
p
f ∈ Lloc (Σ, σ) ∩ L 1 Σ, 1+σ(x)
p
|x | n−1
and g ∈ Lcomp (Σ, σ)

with p, p ∈ (1, ∞) such that 1/p + 1/p = 1,

∫ ∫ (2.3.26)
one has |T f ||g| dσ < +∞, | f ||T g| dσ < +∞
Σ ∫ ∫ Σ
and (T f )g dσ = − f (T g) dσ.
Σ Σ

(6) If H p (Σ, σ)with (n − 1)/n < p < ∞ denotes the scale of Hardy spaces on Σ,
introduced in [113, Definition 4.2.1], then the operator (2.3.18) extends uniquely
to a linear and bounded mapping

T : H p (Σ, σ) −→ L p (Σ, σ), n−1

n < p ≤ 1, (2.3.27)

according to

H p (Σ, σ)  f → lim T f j in L p (Σ, σ) where

j→∞
p,q
{ f j } j ∈N ⊆ Hfin (Σ, σ) with q ∈ (1, ∞) is any (2.3.28)
sequence satisfying f = lim f j in H p (Σ, σ),
j→∞

with operator quasi-norm

  
T  H p (Σ,σ)→L p (Σ,σ) ≤ CΣ, p  k S n−1 𝒞 N (S n−1 ) for n−1
n < p ≤ 1. (2.3.29)

In particular, with P.V. used to indicate that an integral is taken in the principal-
value sense,
∫
∞ 
∞
Tf = λjT aj = λ j P.V. k(· − y)a j (y) dσ(y) in L p (Σ, σ),
j=1 j=1 Σ
when n−1 < p ≤ 1 and when the distribution f ∈ H p (Σ, σ) (2.3.30)
n 
is expressed as f = ∞ j=1 λ j a j in H (Σ, σ) for some sequence
p

{λ j } j ∈N ∈  (N) and some sequence {a j } j ∈N of (p, q)-atoms on Σ,

p

where q ∈ [1, ∞] with q > p is some fixed background index.

Moreover, the operators in (2.3.27) corresponding to various values of the
exponent p ∈ n−1n , 1 are compatible with one another. In addition, the operator
T defined on Hardy spaces as in (2.3.27) acts in a coherent fashion with the
operator T defined on Lebesgue spaces as in (2.3.18), and the operator T defined
as in (2.3.27) with p = 1 acts in a coherent fashion with the operator T defined
in (2.3.19).
(7) For the end-point p = ∞, consider the following modified version of the
principal-value operator in (2.3.15):
352 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
 
(Tmod f )(x) := lim+ k ε (x − y) − k1 (−y) f (y) dσ(y) for σ-a.e. x ∈ Σ,
ε→0 Σ  σ(x) 
where k ε := k · 1Rn \B(0,ε) for each ε > 0, and f ∈ L 1 Σ, .
1 + |x| n
(2.3.31)
Then Tmod is meaningfully defined in the above context. In particular, since
Hölder’s inequality and [112, Lemma 7.2.1] guarantee a continuous embedding
 σ(x) 
L p (Σ, σ) → L 1 Σ, for each p ∈ [1, ∞], (2.3.32)
1 + |x| n

it follows that Tmod acts in a well-defined manner on L p (Σ, σ) for each exponent
p ∈ [1, ∞].
Most generally, Tmod from (2.3.31) induces a well-defined, linear and continuous
operator20
 σ(x)  1,∞
Tmod : L 1 Σ, −→ Lloc (Σ, σ). (2.3.33)
1 + |x| n
Defined as such, Tmod is compatible with T from (2.3.16) (hence also with T from
(2.3.18)-(2.3.19)), in the sense that

for each function f belonging to the space L 1 Σ, 1+σ(x) |x | n−1
(hence,
in particular, for each f ∈ L (Σ, σ) with p ∈ [1, ∞)) the diff-
p
(2.3.34)
erence C f := Tmod f − T f is a constant function on Σ with
|C f | ≤ CΣ,k ·  f  L 1 (Σ, σ(x) ) for some finite CΣ,k > 0.
1+| x | n−1

Also, in a quantitative fashion,

p
Tmod f belongs to the space Lloc (Σ, σ) whenever
 σ(x)  (2.3.35)
p
f ∈ L 1 Σ, ∩ Lloc (Σ, σ) for some p ∈ (1, ∞),
1 + |x| n
while (compare with (2.3.26))

if p, p ∈ (1, ∞) are such that 1/p + 1/p = 1 then given any functions
  ∫
p
f ∈ Lloc (Σ, σ) ∩ L 1 Σ, 1+σ(x)
p
|x | n and g ∈ Lcomp (Σ, σ) with g dσ = 0,
Σ
∫ ∫
it follows that |Tmod f ||g| dσ < +∞, | f ||T g| dσ < +∞,
Σ Σ
∫ ∫
and (Tmod f )g dσ = − f (T g) dσ.
Σ Σ
(2.3.36)

20 with a similar interpretation as in the case of (2.3.16)

2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 353

Furthermore, there exists a finite constant C = C(Σ, k) > 0 with the property
that
 
T f  ≤ C f  L ∞ (Σ,σ) for every f ∈ L ∞ (Σ, σ), (2.3.37)
mod BMO(Σ,σ)

In particular,
Tmod : L ∞ (Σ, σ) −→ BMO(Σ, σ)
(2.3.38)
is well defined, linear and bounded.
Also,
if the set Σ is bounded then T from (2.3.18) induces a linear
and bounded mapping T : L ∞ (Σ, σ) −→ BMO(Σ, σ). (2.3.39)

 angled ∗brackets ·, · denoting the duality pairing between

Finally, with the
H 1 (Σ, σ) and H 1 (Σ, σ) (cf. [113, Theorem 4.6.1]), for any two given func-
tions f ∈ L ∞ (Σ, σ) and g ∈ H 1 (Σ, σ) one has
∫  % &
H 1 (Σ,σ) g, [Tmod f ] BMO(Σ,σ)/∼ if Σ is unbounded,
f T g dσ = % & (2.3.40)
Σ H 1 (Σ,σ) g, T f BMO(Σ,σ) if Σ is bounded.

(8) If p ∈ (1, ∞) and q ∈ (0, ∞] then (cf. [112, (7.7.107)])

for each given function f ∈ L p,q (Σ, σ) the limit

(T f )(x) = lim+ (Tε f )(x) exists for σ-a.e. x ∈ Σ (2.3.41)
ε→0
and the function T f thus defined is σ-measurable.

Also,

Tmax : L p,q (Σ, σ) −→ L p,q (Σ, σ), (2.3.42)

T:L p,q
(Σ, σ) −→ L p,q
(Σ, σ), (2.3.43)

are well-defined, bounded (sub-linear and linear, respectively) operators. Fi-

nally,

if p, q ∈ (1, ∞) then the (real) transpose of the operator (2.3.43)

   
is the operator −T : L p ,q (Σ, σ) → L p ,q (Σ, σ), where the
(2.3.44)
exponents p, q  ∈ (1, ∞) are such that 1/p + 1/p = 1 and
1/q + 1/q  = 1.

(9) For each exponent p ∈ n−1 n , ∞ denote by Tp the operator (2.3.27)-(2.3.28) if
n < p ≤ 1 and, respectively, the operator (2.3.18) if 1 < p < ∞. Then the
n−1
 
family of operators Tp n−1 are mutually compatible with one another, in
n <p<∞
the sense that
354 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 n−1
if p0, p1 ∈ n , ∞ and f ∈ H p0 (Σ, σ) ∩ H p1 (Σ, σ), it
(2.3.45)
follows that Tp0 f = Tp1 f at σ-a.e. point on Σ.
 n−1
As such, given any p0, p1 ∈ n , ∞ , one may define, in a coherent fashion, a
mapping

p0, p1 : H p0 (Σ, σ) + H p1 (Σ, σ) −→ L p0 (Σ, σ) + L p1 (Σ, σ),

T
p0, p1 f := Tp0 f0 + Tp1 f1 whenever f ∈ H p0 (Σ, σ) + H p1 (Σ, σ)
T (2.3.46)
splits as f = f0 + f1 with f0 ∈ H p0 (Σ, σ) and f1 ∈ H p1 (Σ, σ).

Moreover, whenever n−1

n < p0 < p1 < ∞,

p0, p1 is a linear and bounded operator, which satisfies

T
 (2.3.47)
p0, p1 
T = Tp for each p ∈ [p0, p1 ].
p
H (Σ,σ)

Also,
 
the operators in the family Tp0, p1 n−1
n <p0, p1 <∞ (2.3.48)
are mutually compatible with one another,
thus giving rise to a common linear mapping. Specifically,

HSum (Σ, σ) := H p0 (Σ, σ) + H p1 (Σ, σ) (2.3.49)
n−1
n <p0, p1 <∞

and

LSum (Σ, σ) := L p0 (Σ, σ) + L p1 (Σ, σ) (2.3.50)
n−1
n <p0, p1 <∞

are linear spaces and there exists a unique linear operator

T : HSum (Σ, σ) −→ LSum (Σ, σ) with the property that

  (2.3.51)

T =Tp0, p1 for each p0, p1 ∈ n−1 , ∞ .
H p p
0 (Σ,σ)+H 1 (Σ,σ) n

In addition, the restriction of the operator  from (2.3.51) to any Lorentz-

T
 n−1
based Hardy space H (Σ, σ) with p ∈ n , ∞ and q ∈ (0, ∞] maps into
p,q

Lorentz space L p,q (Σ, σ) in a linear and bounded fashion, and the operator
thus induced is compatible with T from (2.3.18) and (2.3.27). As a consequence,
one may employ the same notation for said operator, namely
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 355

T : H p,q (Σ, σ) −→ L p,q (Σ, σ)

 (2.3.52)
for p ∈ n−1 n , ∞ and q ∈ (0, ∞].

This satisfies the operator quasi-norm estimate

  
T  H p, q (Σ,σ)→L p, q (Σ,σ) ≤ CΣ, p,q  k  
S n−1 𝒞 N (S n−1 )
(2.3.53)

and enjoys the following property:

T f = T f0 + T f1 at σ-a.e. point on Σ, whenever f ∈ H p,q (Σ, σ),

f0 ∈ H p0,q0 (Σ, σ), f1 ∈ H p1,q1 (Σ, σ), with p, p0, p1 ∈ n−1
n , ∞ and
 
q, q0, q1 ∈ (0, ∞], are such that f = f0 + f1 in Lipc (Σ) .
(2.3.54)

Finally,

given p ∈ n−1n , ∞ , if q ∈ (0, ∞) then the operator (2.3.52) acts
according to H p,q (Σ, σ)  f −→ lim T f j in L p,q (Σ, σ), provided
j→∞
{ f j } j ∈N ⊆ H p0 (Σ, σ) ∩ H p1 (Σ, σ) with n−1
n < p0 < p < p1 < ∞ is any
given sequence satisfying f = lim f j in the space H p,q (Σ, σ).
j→∞
(2.3.55)
(10) Denote by Ap (Σ, σ), 1 ≤ p < ∞, the classes of Muckenhoupt
 weights associated
as in (A.0.3) with the space of homogeneous type Σ, | · − · |, σ (cf. [112,
Example 7.4.1]). Then the principal-value singular integral operator T from
(2.3.15) induces, via restriction (cf. [112, (7.7.104)]), well-defined linear and
bounded mappings on weighted versions of the Lebesgue spaces in (2.3.18),
for weights in the corresponding Muckenhoupt class Ap (Σ, σ), 1 < p < ∞.
Specifically, the mapping

T : L p (Σ, wσ) −→ L p (Σ, wσ), p ∈ (1, ∞), w ∈ Ap (Σ, σ), (2.3.56)

induced (via restriction; cf. [112, (7.7.104)]) by the operator T from (2.3.15) is
well defined, linear, and bounded. Moreover, for each exponent p ∈ (1, ∞) and
w ∈ Ap (Σ, σ), one has
  
T  L p (Σ,wσ)→L p (Σ,wσ) ≤ C(Σ, p, [w] A p ) k S n−1 𝒞 N (S n−1 ) . (2.3.57)

In addition,

Tmax : L p (Σ, wσ) −→ L p (Σ, wσ), p ∈ (1, ∞), w ∈ Ap (Σ, σ), (2.3.58)

is a well-defined, sub-linear, bounded, and continuous operator, whose operator

norm may be estimated as in (2.3.57), i.e.,
  
Tmax  L p (Σ,wσ)→L p (Σ,wσ) ≤ C(Σ, p, [w] A p ) k S n−1 𝒞 N (S n−1 ) . (2.3.59)
356 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Also, bearing in mind the natural identification

 p ∗  
L (Σ, wσ) = L p (Σ, w 1−p σ)
(2.3.60)
for p, p ∈ (1, ∞) with 1/p + 1/p = 1,
∫
under the canonical integral pairing ( f , g) → Σ f g dσ it follows that21

the (real) transpose of the operator (2.3.56) is

   
−T : L p (Σ, w 1−p σ) → L p (Σ, w 1−p σ), where (2.3.61)
p ∈ (1, ∞) is chosen so that 1/p + 1/p = 1.

Furthermore, for each p ∈ (1, ∞) and w ∈ Ap (Σ, σ), the following mapping is
well defined, linear, and continuous:
 σ(x) 
p p
T : Lloc (Σ, wσ) ∩ L 1 Σ, −→ Lloc (Σ, wσ). (2.3.62)
1 + |x| n−1

Finally, for each p, p ∈ (1, ∞) with 1/p + 1/p = 1 and each w ∈ Ap (Σ, σ) there
exists a constant C ∈ (0, ∞) with the property that
 
if f ∈ L p (Σ, wσ) and g ∈ .L p (Σ, w 1−p σ) then (T f )g + f (T g)
 space H (Σ, σ) (cf. [113, (4.2.12)]) and one (2.3.63)
belongs 1
 to the Hardy
 
has (T f )g + f (T g) H 1 (Σ,σ) ≤ C f  L p (Σ,wσ) · g L p (Σ,w 1−p σ) .

(11) As before, denote by M Σ the Hardy-Littlewood

 maximal operator associated
with the space of homogeneous type Σ, | · − · |, σ . Consider a Generalized
Banach Function Space X on (Σ, σ) and denote by X its associated space (cf.
[113, Definitions 5.1.4, 5.1.11]). Then if

M Σ : X → X and M Σ : X → X
(2.3.64)
are well-defined bounded mappings

it follows that the maximal operator Tmax , originally considered as in (2.3.4)-

(2.3.5), induces well-defined, bounded and continuous mappings

Tmax : X −→ X and Tmax : X −→ X

     (2.3.65)
with max Tmax X→X, Tmax X →X ≤ C  k S n−1 𝒞 N (S n−1 ),

where C ∈ (0, ∞) depends only on Σ and the operator norms of M Σ on X, X.

Likewise, under the assumptions made in (2.3.64), the principal-value singu-
lar integral operator T, originally defined as in (2.3.15), induces well-defined,
bounded, mappings


21 In relation to (2.3.61), it is reassuring to note that if w ∈ A p (Σ, σ) then w 1−p ∈ A p  (Σ, σ)
(cf. item (2) in [112, Lemma 7.7.1])
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 357

T : X −→ X and T : X −→ X
     (2.3.66)
with max T X→X, T X →X ≤ C  k S n−1 𝒞 N (S n−1 ),

where C ∈ (0, ∞) depends only on Σ and the operator norms of M Σ on X and

X. Next, let X̊ be defined in relation to X as in [113, Definition 5.2.6], i.e., set

∞ (Σ, σ)  · X
X̊ := Lcomp . (2.3.67)

Then, under the assumptions made in (2.3.64), one also has

Tmax : X̊ −→ X̊ and T : X̊ −→ X̊
     (2.3.68)
with max Tmax X̊→X̊, T X̊→X̊ ≤ C  k S n−1 𝒞 N (S n−1 ),

where, once again, C ∈ (0, ∞) depends only on Σ and the operator norms of
M Σ on X and X. Finally, granted the assumptions in (2.3.64), the following
transposition formula holds:
∫ ∫
(T f )g dσ = − f (T g) dσ for any f ∈ X̊ and g ∈ X . (2.3.69)
Σ Σ

Proof We begin by noting that the fundamental operator norm estimate recorded in
(2.3.11) has been established in [31, Proposition 4 bis]. The claims in (2.3.15) follow
directly from [112, Corollary 5.3.6] (bearing in mind [112, Proposition 6.10.5] and
the fact that UR sets are closed and upper Ahlfors regular). In concert with (2.3.11),
this further implies the fact that T is well defined, linear, and bounded in the context of
(2.3.18). Granted this, all claims in item (1) are directly implied by Proposition 2.2.5
and (2.2.252) in Corollary 2.2.11.
Before going any further, we wish to present an alternative proof of the claims
made in (2.3.15) in a series of steps, presented below.
Step I: The proof of the first claim in (2.3.15) when Σ is a Lipschitz graph in Rn
and f ∈ L p (Σ, σ) with p ∈ (1, ∞). Concretely, assume that for some Lipschitz
function φ : Rn−1 → R we have
 
Σ = x , φ(x ) : x  ∈ Rn−1 , (2.3.70)

and fix some function f ∈ L p (Σ, σ) with p ∈ (1, ∞). In this scenario, [112, (2.8.69)]
gives
'
σ(x , φ(x ) = 1 + |(∇  φ)(x )| 2 L n−1 (x ), x  ∈ Rn−1 . (2.3.71)

Given that the above square-root is bounded away from zero and infinity, if we now
define the function f : Rn−1 → R by setting
' 

f (x ) := 1 + |(∇  φ)(x )| 2 f x , φ(x ) for each x  ∈ Rn−1, (2.3.72)
358 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

it follows that  f ∈ L p (Rn−1, L n−1 ). Moreover, changing variables in (2.3.6) gives

that for each ε > 0 and each x  ∈ Rn−1 we have
∫
  

(Tε f ) x , φ(x ) = k x  − y , φ(x ) − φ(y ) 
f (y ) dy  . (2.3.73)
y  ∈R n−1
|x  −y  | 2 + |φ(x  )−φ(y  ) | 2 >ε 2

To study the behavior of the integral in the right side of (2.3.73), the idea is to rely
on Proposition 2.3.1 used for a suitable choice
 of the functions A, B, F. Concretely,
we consider A : Rn−1 → Rn by A(z ) := z , φ(z ) for each z  ∈ Rn−1 , and take
B := A. Clearly, A is Lipschitz
' and B is bi-Lipschitz. To define the function F, first
pick a finite number R > 1 + ∇  φ L2 ∞ (Rn−1, L n−1 ) and consider a cutoff function

ψ ∈ 𝒞∞
c (R ) even, such that ψ ≡ 0 on B(0, 1/4),
n
(2.3.74)
and ψ ≡ 1 on B(0, R) \ B(0, 1/2).

Having done that, define the function F := ψk in Rn and note that this choice entails
F ∈ 𝒞 N (Rn ) and F is odd. In addition, for each x , y  ∈ Rn−1 with x   y , the
positive homogeneity of k permits us to express
 1  x  − y  φ(x ) − φ(y ) 
k x  − y , φ(x ) − φ(y ) = k ,
|x  − y  | n−1 |x  − y  | |x  − y  |
1  x  − y  φ(x ) − φ(y ) 
=  (ψk) ,
|x − y  | n−1 |x  − y  | |x  − y  |
1  A(x ) − A(y ) 
= F . (2.3.75)
|x  
−y | n−1 |x  − y  |

Granted this, it follows from (2.3.73) and Proposition 2.3.1 that the first claim in
(2.3.15) holds in this case.
Step II: The proof of the first claim in (2.3.15) when Σ is a rotated Lipschitz graph
in Rn and f ∈ L p (Σ, σ) with p ∈ (1, ∞). Given that the original assumptions on the
kernel k are rotation invariant, and that the (n − 1)-dimensional Hausdorff measure
is also rotation invariant, this is a consequence of Step I and a change of variables
reducing matters to the case when Σ is a genuine Lipschitz graph in Rn .
First Intermission: Having dealt at this point with the first claim in (2.3.15)
when Σ is a rotated Lipschitz graph in Rn and f ∈ L p (Σ, σ) with p ∈ (1, ∞), we now
consider the case when Σ ⊆ Rn is a UR set and f ∈ L p (Σ, σ) with p ∈ (1, ∞). In
particular, from [112, Proposition 6.10.5] it follows that Σ is a countably rectifiable
set. Using [112, Proposition 5.3.3] (while keeping in mind that Σ is a closed set) we
may therefore decompose
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 359
 
Σ= S j ∪ A, where H n−1 (A) = 0 and, for each j ∈ N,
j ∈N (2.3.76)
S j is a compact subset of a rotated Lipschitz graph Σ j in Rn .

For further use, in the above context let us also define

j−1
S1 := S1 then, inductively, set Sj := S j \ Si for j ≥ 2. (2.3.77)
i=1

Step III: The proof of the fact that the limit in (2.3.15) exists at σ-a.e. point on
Σ in the case when Σ is as in (2.3.76), and f ∈ L p (Σ, σ) with p ∈ (1, ∞) has the
additional property that supp f ⊆ S j0 for some j0 ∈ N. In this situation, Step II
applies and gives that, on the one hand, lim+ (Tε f )(x) exists for σ-a.e. x ∈ S j0 . On
ε→0
the other hand, for each x ∈ Σ \ supp f the aforementioned limit trivially exists, as
in this latter case we have
∫
lim+ (Tε f )(x) = lim+ k(x − y) f (y) dσ(y)
ε→0 ε→0
Σ\B(x,ε)
∫
= lim+ k(x − y) f (y) dσ(y)
ε→0
(supp f )\B(x,ε)
∫
= k(x − y) f (y) dσ(y), (2.3.78)
supp f

thanks to item (1) in [112, Lemma 3.8.4].

Step IV: The proof of the existence of the limit in (2.3.15) at σ-a.e. point on Σ
in the case when Σ is as in (2.3.76), and f ∈ L p (Σ, σ) with p ∈ (1, ∞) has the
additional property that
m
supp f ⊆ S j for some m ∈ N. (2.3.79)
j=1

In this scenario, recall the notation introduced in (2.3.77) and define f j := f 1Sj for
1 ≤ j ≤ m. Then, given that each f j belongs to L p (Σ, σ) and is supported on S j ,
Step III applies and gives that, for each j ∈ {1, . . . , m}, the limit lim+ (Tε f j )(x)
ε→0

m
exists for σ-a.e. x ∈ Σ. Since, by design, f = f j on Σ, and since Tε is linear, we
j=1
ultimately conclude that, indeed, lim+ (Tε f )(x) exists for σ-a.e. x ∈ Σ.
ε→0

Step V: The proof of the fact that the limit in (2.3.15) exists at σ-a.e. point on Σ in
the case when Σ is as in (2.3.76), and f ∈ L p (Σ, σ) with p ∈ (1, ∞) is arbitrary.
360 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

The idea now is to use [112, Proposition 6.2.11] for the ambient 𝒳 := Σ × [0, ∞)
equipped with the canonical product topology, the normed space Y := L p (Σ, σ),
X := Σ × {0},
 μ := σ which is a complete measure on X by [112, Lemma 3.6.4],
and with Γ (x, 0) := {x} × (0, ∞) for each (x, 0) ∈ X = Σ × {0}. The latter choice

ensures that (x, 0) ∈ Γ (x, 0) for every (x, 0) ∈ X = Σ × {0}. Also, consider a linear
operator T mapping functions on X into functions on 𝒳 \ X = Σ × (0, ∞) of the
following sort. For every f ∈ Y = L p (Σ, σ) define
∫
(T f )(x, ε) := k(x − y) f (y) dσ(y)
y ∈Σ (2.3.80)
|x−y |>ε
for all (x, ε) ∈ 𝒳 \ X = Σ × (0, ∞),

and note that this entails

(T f )(x, 0) : = sup |(T f )(z, ε)|

(z,ε)∈Γ((x,0))
 ∫ 
 
 
= sup  k(x − y) f (y) dσ(y)
ε>0  
y ∈Σ
|x−y |>ε

= (Tmax f )(x) for every x ∈ Σ. (2.3.81)

In turn, from (2.3.81) and (2.3.11) we conclude that condition [112, (6.2.74)] is ver-
ified in the present setting. To finish the implementation of [112, Proposition 6.2.11]
we finally choose
 m 
V := g ∈ L p (Σ, σ) : supp g ⊆ S j for some m ∈ N . (2.3.82)
j=1

Then V is a dense subset of L p (Σ, σ) since for every f ∈ L p (Σ, σ) Lebesgue’s

Dominated Convergence Theorem gives

m
V  gm := f 1Sj −→ f in L p (Σ, σ) as m → ∞. (2.3.83)
j=1

Moreover, condition [112, (6.2.75)] is also satisfied since for every g ∈ V the limit
∫
lim (T g)(x, ε) = lim+ k(x − y)g(y) dσ(y)
Γ((x,0))(x,ε)→(x,0) ε→0
y ∈Σ
|x−y |>ε

= lim+ (Tε g)(x) (2.3.84)

ε→0
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 361

exists for μ-a.e. (x, 0) ∈ X = Σ × {0}, thanks to Step IV. At this stage, all hypotheses
of [112, Proposition 6.2.11] have been verified for the present choices, so we may
conclude from it that for every function f ∈ L p (Σ, σ) ≡ L p (X, μ) the limit

lim (Tε f )(x) = lim (T f )(x, ε) (2.3.85)

ε→0+ Γ((x,0))(x,ε)→(x,0)

exists for σ-a.e. x ∈ Σ.

Second Intermission: Let us record our progress. At this point we have
proved the first claim in (2.3.15) in the case when f ∈ L p (Σ, σ) with p ∈ (1, ∞).
In particular, we may consider the operator T acting on an arbitrary f ∈ L p (Σ, σ)
with p ∈ (1, ∞) according to

(T f )(x) := lim+ (Tε f )(x) for σ-a.e. x ∈ Σ. (2.3.86)

ε→0

Thanks to this, (2.2.68), the equivalence (i) ⇔ (iii) in [112, Remark 3.1.2], and [112,
(3.6.26)] it follows that
T f is a σ-measurable function on Σ for
(2.3.87)
each f ∈ L p (Σ, σ) with p ∈ (1, ∞).
Combining (2.3.5), (2.3.11), (2.3.86), and (2.3.87) we arrive at the conclusion that

the operator T : L p (Σ, σ) −→ L p (Σ, σ)

(2.3.88)
is well defined, linear, and bounded, if 1 < p < ∞,

with operator norm

  
T  L p (Σ,σ)→L p (Σ,σ) ≤ CΣ, p  k S n−1 𝒞 N (S n−1 ) . (2.3.89)

These justify (2.3.18) and (2.3.20) in the statement of the theorem. Next, item (iii)
of Proposition 2.2.8 applied to the operator T from (2.3.86) yields, on account of
(2.3.88)-(2.3.89), that

Tmax : L 1 (Σ, σ) −→ L 1,∞ (Σ, σ) and there exists C ∈ (0, ∞) such that
  
Tmax f  L 1,∞ (Σ,σ) ≤ C  k S n−1 𝒞 N (S n−1 )  f  L 1 (Σ,σ) for each f ∈ L 1 (Σ, σ).
(2.3.90)

This establishes (2.3.12) in the statement of the theorem.

Step VI: The proof of the first claim in (2.3.15) when Σ ⊆ Rn is an arbitrary UR
set and when f ∈ L 1 (Σ, σ). The idea is to once again implement [112, Proposi-
tion 6.2.11], this time with p = 1. We make the same choices for all the relevant
entities as in treatment of Step V, with the exception of (2.3.82), in place of which
we now take

Y := L 1 (Σ, σ) and V := L 1 (Σ, σ) ∩ L p (Σ, σ) for a fixed p ∈ (1, ∞). (2.3.91)

362 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Then V is a dense subset of Y (cf. [112, (3.1.13)]) and, as seen from Step V (cf.
(2.3.86)), condition [112, (6.2.75)] is presently verified. Moreover, [112, (6.2.74)]
currently holds thanks to (2.3.90). Granted these, [112, Proposition 6.2.11] applies
(with p = 1) and proves that the limit in (2.3.15) exists, for each fixed function
f ∈ L 1 (Σ, σ), at σ-a.e. point on Σ.
Step VII: The proof of the claims in (2.3.15) when Σ ⊆ Rn is an arbitrary  UR
σ(x)
set and f is an arbitrary function belonging to the space L Σ, 1+ |x | n−1 . In such
1

a scenario, fix x0 ∈ Σ and, for each j ∈ N, abbreviate Δr := B(x0, j) ∩ Σ. Having

fixed a function f ∈ L Σ, 1+σ(x)
1
|x | n−1
, and given ε ∈ (0, 1) arbitrary, for each j ∈ N
and each x ∈ Δ j split
∫
(Tε f )(x) = Iε (x) + k(x − y) f (y) dσ(y) (2.3.92)
y ∈Σ
|x−y | ≥1

where, with f j := f · 1Δ j+1 ,

∫ ∫
Iε (x) := k(x − y) f (y) dσ(y) = k(x − y) f j (y) dσ(y)
y ∈Σ y ∈Σ
1> |x−y |>ε 1> |x−y |>ε
∫
= (Tε f j )(x) − k(x − y) f j (y) dσ(y). (2.3.93)
y ∈Σ
|x−y | ≥1

Since the function f j = f · 1Δ j+1 belongs to L 1 (Σ, σ), the result proved in Step VI
ensures the existence of some σ-measurable set N j ⊆ Σ such that σ(N j ) = 0
and lim+ (Tε f j )(x) exists for each x ∈ Σ \ N j . Granted this, it follows from
ε→0
(2.3.92)-(2.3.93) that lim+ (Tε f )(x) exists for each x ∈ Δ j \ N j . Given that
ε→0  

Σ = Δ j , this ultimately shows that for each function f ∈ L 1 Σ, 1+σ(x)
|x | n−1 the
j ∈N
limit (T f )(x) := lim+ (Tε f )(x) exists for σ-a.e. x ∈ Σ. With this in hand, from
ε→0
(2.2.68), the equivalence (i) ⇔ (iii) in [112, Remark 3.1.2], and [112, (3.6.26)] we
then conclude that the function T f thus defined is σ-measurable. Hence, the claims
in (2.3.15) hold as stated.
At this stage, the claims in (2.3.15) have been fully justified. Having accomplished
this task, arguing as in (2.3.86)-(2.3.87) and availing ourselves of (2.3.90) then proves
(2.3.19) and (2.3.21). As regards the claim made in (2.3.16), fix a point xo ∈ Σ and

pick some radius r ∈ (0, ∞). For an arbitrarily chosen function f ∈ L 1 Σ, 1+σ(x) |x | n−1
,
define
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 363
 

gr := T f 1Σ\B(xo,2r)  and
Σ∩B(x o ,r)
  (2.3.94)

hr := T f 1Σ∩B(xo,2r)  .
Σ∩B(x o ,r)

Since there exists Ck,r ∈ (0, ∞) with the property that for each x ∈ Σ ∩ B(xo, r) we
have
∫
 
|k(x − y)|  f 1Σ\B(xo,2r) (y) dσ(y)
Σ
∫
| f (y)|
≤ Ck,r dσ(y)
Σ\B(x o ,2r) |x − y| n−1
∫
| f (y)|
≤ Ck,r dσ(y) < +∞, (2.3.95)
Σ 1 + |y| n−1
it follows that 
gr ∈ L ∞ Σ ∩ B(xo, r), σ and
gr  L ∞ (Σ∩B(xo,r),σ) ≤ Ck,r  f  1  . (2.3.96)
σ(x)
L Σ,
1+ |x | n−1

Given that, according to [112, (6.2.27)], we have the continuous embeddings

  
L ∞ Σ ∩ B(xo, r), σ → L 1 Σ ∩ B(xo, r), σ → L 1,∞ Σ ∩ B(xo, r), σ , (2.3.97)

we deduce from (2.3.96) that


gr ∈ L 1,∞ Σ ∩ B(xo, r), σ and
gr  L 1,∞ (Σ∩B(xo,r),σ) ≤ Ck,xo,r  f  1  . (2.3.98)
σ(x)
L Σ,
1+ |x | n−1

Also, since there exists Cxo,r ∈ (0, ∞) such that

f 1Σ∩B(xo,2r) ∈ L 1 (Σ, σ) with the property that

 
 f 1Σ∩B(x ,2r)  1 ≤ Cxo,r  f  1  σ(x) , (2.3.99)
o L (Σ,σ) L Σ,
1+ |x | n−1

we conclude from (2.3.19) (which has already been established at this point) that

T f 1Σ∩B(xo,2r) ∈ L 1,∞ (Σ, σ) and we have
  
T f 1 
Σ∩B(x o ,2r) ≤C f  k,x o ,r . (2.3.100)
σ(x)
L 1,∞ (Σ,σ) L 1 Σ,
1+ |x | n−1

In turn, from (2.3.100) and the second line in [112, (6.2.17)] we see that
364 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

hr ∈ L 1,∞ Σ ∩ B(xo, r), σ and
hr  L 1,∞ (Σ∩B(xo,r),σ) ≤ Ck,xo,r  f  1  . (2.3.101)
σ(x)
L Σ,
1+ |x | n−1

In view of the fact that, by design, (T f )B(xo,r)∩Σ = gr + hr on Σ ∩ B(xo, r), we
ultimately conclude from (2.3.98) and (2.3.101) that the operator T in (2.3.16) is
indeed well defined, linear, and continuous.
Note that if the above argument is carried out for  a function  f which is now
p σ(x)
assumed to belong to the intersection Lloc (Σ, σ) ∩ L Σ, 1+ |x | n−1 for some exponent
1

p ∈ (1, ∞), then in place of (2.3.99) we now have

f 1Σ∩B(xo,2r) ∈ L p (Σ, σ) with

  (2.3.102)
 f 1Σ∩B(x ,2r)  p =  f  L p (Σ∩B(xo,2r),σ) .
o L (Σ,σ)

Thanks to this and (2.3.18), in place of (2.3.100) we now obtain


T f 1Σ∩B(xo,2r) ∈ L p (Σ, σ) with
   (2.3.103)
T f 1Σ∩B(x ,2r)  p ≤ Ck,xo,r  f  L p (Σ∩B(xo,2r),σ) .
o L (Σ,σ)

Hence, (2.3.101) presently improves to


hr ∈ L p Σ ∩ B(xo, r), σ with
(2.3.104)
hr  L p (Σ∩B(xo,r),σ) ≤ Ck,xo,r  f  L p (Σ∩B(xo,2r),σ) .

Together with (2.3.96), this ultimately shows that the mapping in (2.3.17) is, as
claimed, well defined, linear, and continuous.
Next, the Cotlar-type inequality claimed in (2.3.23) is a direct consequence of
Corollary 2.2.6 and (2.3.21). Going further, (2.3.24) is a consequence of (2.3.11),
the fact that the limit in (2.3.15) exists at σ-a.e. point, and Lebesgue’s Dominated
Convergence Theorem. As regards the claim in (2.3.25), pick two arbitrary functions,

f ∈ L p (Σ, σ) and g ∈ L p (Σ, σ), with p, p ∈ (1, ∞) Hölder conjugate exponents.
Then for every ε > 0 Fubini’s Theorem gives
∫ ∫ ∫
(Tε f )(x)g(x) dσ(x) = k(x − y) f (y)g(x)1 |x−y |>ε dσ(y) dσ(x)
Σ Σ Σ
∫
=− f (x)(Tε g)(x) dσ(x). (2.3.105)
Σ

Keeping in mind that Tε f → T f in L p (Σ, σ) and Tε g → T g in L p (Σ, σ) as
ε → 0+ (cf. (2.3.24)), we deduce from (2.3.105) that
∫ ∫
(T f )g dσ = − f (T g) dσ, (2.3.106)
Σ Σ
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 365

proving (2.3.25). As regards theclaims in (2.3.26), pick some arbitrary functions

p
f ∈ Lloc (Σ, σ) ∩ L 1 Σ, 1+σ(x) and g ∈ Lcomp (Σ, σ) with p, p ∈ (1, ∞) such
p
|x | n−1
that 1/p + 1/p = 1, and select xo ∈ Σ along with R > 0 with the property that
supp g ⊆ Σ ∩ B(xo, R). Then since f 1B(xo,2R)∩Σ ∈ L p (Σ, σ) we conclude from
(2.3.18) that T f 1Σ∩B(xo,2R) ∈ L p (Σ, σ), hence
∫
  
T f 1Σ∩B(x ,2R)  |g| dσ < +∞, (2.3.107)
o
Σ

by Hölder’s inequality. Also, there exists a constant C = C(k, xo, R) ∈ (0, ∞) for
which
∫
 
T f 1Σ\B(x ,2R) (x)||g(x)| dσ(x) (2.3.108)
o
Σ
∫ ∫
≤ |k(x − y)|| f (y)||g(x)| dσ(y) dσ(x)
x ∈Σ∩B(x o ,R) y ∈Σ\B(xo ,2R)
∫ | f (y)|  ∫ 
≤C dσ(y) |g(x)| dσ(x) < +∞,
Σ\B(x o ,2R) 1 + |y| n−1 Σ∩B(x o ,R)

given the integrability∫ properties assumed of f and g. Combining

∫ (2.3.107) and
(2.3.108) then proves Σ |T f ||g| dσ < +∞. That we also have Σ | f ||T g| dσ < +∞
is then established in a very similar fashion. In turn, the finiteness properties just
proved permit us to write
∫ ∫ ∫
 
(T f )g dσ = T f 1Σ∩B(xo,2R) g dσ + T f 1Σ\B(xo,2R) g dσ
Σ Σ Σ
∫
=− f 1Σ∩B(xo,2R) (T g) dσ
Σ
∫ ∫ 
+ k(x − y) f (y) dσ(y) g(x) dσ(x)
Σ∩B(x o ,R) Σ\B(x o ,2R)
∫
=− f 1Σ∩B(xo,2R) (T g) dσ
Σ
∫ ∫ 
− f (y) k(y − x)g(x) dσ(x) dσ(y)
Σ\B(x o ,2R) Σ∩B(x o ,R)
∫ ∫
=− f 1Σ∩B(xo,2R) (T g) dσ − f 1Σ\B(xo,2R) (T g) dσ
Σ Σ
∫
=− f (T g) dσ, (2.3.109)
Σ
366 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

where we have also used (2.3.25) in the second equality above (bearing in mind that

the function f 1B(xo,2R)∩Σ belongs to L p (Σ, σ) and that g ∈ L p (Σ, σ)), as well as
Fubini’s Theorem and the fact that k is odd in the third equality. This establishes the
final formula in (2.3.25). Hence, at this point in the proof we have justified all claims
in items (1)-(5).
To deal with the claims in item (6), fix some q ∈ (1, ∞) and let a be an arbitrary
(p, q)-atom on Σ. Hence, there exist xo ∈ Σ and r ∈ 0, diam Σ such that
∫
supp a ⊆ B(xo, r) ∩ Σ, a dσ = 0,
Σ (2.3.110)
 1/q−1/p
and a L q (Σ,σ) ≤ σ B(xo, r) ∩ Σ .

If σ(Σ) < +∞ then, by definition, the constant function a ≡ σ(Σ)−1/p is also a

(p, q)-atom. We may then estimate
∫
p  1−p/q
|T a| p dσ ≤ T a L q (Σ,σ) · σ B(xo, 2r) ∩ Σ
B(x o ,2r)∩Σ
  p 
≤ C  k S n−1 𝒞 N (S n−1 ) a L q (Σ,σ) · σ B(xo, 2r) ∩ Σ
p 1−p/q

  p
≤ C(Σ, p, q) k S n−1 𝒞 N (S n−1 ) . (2.3.111)

Above, the first estimate is based on Hölder’s inequality, the second estimate is a
consequence of (2.3.18) (with p := q), and the third estimate relies on (2.3.110) and
the Ahlfors regularity of Σ. Next, if x ∈ Σ is such that |x − xo | ≥ 2r then based on
(2.3.110), the Mean Value Theorem, and Hölder’s inequality we may estimate
∫ 
 
|(T a)(x)| =  k(x − y)a(y) dσ(y)
Σ
∫
≤ |k(x − y) − k(x − xo )||a(y)| dσ(y)
B(x o ,r)∩Σ
∫
   r
≤ C  k S n−1 𝒞1 (S n−1 ) |a(y)| dσ(y)
|x − xo | n B(x o ,r)∩Σ
   r 
≤ C  k S n−1 𝒞1 (S n−1 )
1−1/p
σ B(xo, r) ∩ Σ . (2.3.112)
|x − xo | n
With this in hand, [112, Lemma 7.2.1] applies and, since p n > n − 1, gives
∫
  p
|(T a)(x)| p dσ(x) ≤ C(Σ, p) k S n−1 𝒞1 (S n−1 ) . (2.3.113)
Σ\B(x o ,2r)

Collectively, (2.3.111) and (2.3.113) imply that, whenever the atom a is as in

(2.3.110), then
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 367
  
T a L p (Σ,σ) ≤ Ck  S n−1
 ,
𝒞1 (S n−1 )
(2.3.114)

for some finite constant C = C(Σ, p, q) > 0 independent of the atom a. In the
scenario when σ(Σ) < +∞, it follows from (2.3.18) (with p = 2) and Hölder’s
inequality that the same estimate holds for the constant function a ≡ σ(Σ)−1/p .
Having established this, we may now invoke the general boundedness criterion
described in [113, Theorem 4.4.7] (presently used with q := 2, X := L 2 (Σ, σ), and
Y := L p (Σ, σ)) in order to conclude that the operator (2.3.18) extends uniquely (as
indicated in [113, (4.4.144)] with Y := L p (Σ, σ)) to a linear and bounded mapping
as in (2.3.27), with operator quasi-norm satisfying  (2.3.29). That the operators in
(2.3.27) corresponding to various values of p ∈ n−1 n , 1 are compatible with one
another is a consequence of the simultaneous approximation result established in
[113, Theorem 4.4.3]. In fact, [113, Theorem 4.4.3] also implies that the operator T
defined on Hardy spaces as in (2.3.27) acts in a coherent fashion with the operator
T defined on Lebesgue spaces as in (2.3.18).
To also see that the operator T defined as in (2.3.27) with p = 1 acts in a
coherent fashion with the operator T defined in (2.3.19), pick an arbitrary function
1,q
f ∈ H 1 (Σ, σ) and consider a sequence { f j } j ∈N ⊆ Hfin (Σ, σ) for some fixed
q ∈ (1, ∞) such that f = lim f j in H 1 (Σ, σ). Then, on the one hand, (2.3.28)
j→∞
implies that T f j converges as j → ∞ in L 1 (Σ, σ) to T f defined as in (2.3.27).
On the other hand, since H 1 (Σ, σ) → L 1 (Σ, σ) continuously, the continuity of
the operator in (2.3.19) ensures that T f j converges as j → ∞ in L 1,∞ (Σ, σ) to
T f defined as in (2.3.15). The desired compatibility property now readily follows,
finishing the treatment of item (6).
Let us now deal with the claims in item (7) in the statement of the theorem. For
starters, recall that
k ε := k · 1Rn \B(0,ε) for each ε > 0, (2.3.115)
and observe that, thanks to the smoothness and the homogeneity of the kernel k, for
each multi-index α ∈ N0n we have

 α  α
(∂ k)(z) ≤ supS n−1 |(∂ k)| , ∀z ∈ Rn \ {0}. (2.3.116)
|z| n−1+ |α |
In relation to (2.3.115) we claim that for each fixed x ∈ Rn and ε > 0 there exists
a constant Cx,ε ∈ (0, ∞) such that
  Cx,ε
 k ε (x − y) − k1 (−y) ≤ , ∀y ∈ Rn . (2.3.117)
1 + |y| n

This is seen by analyzing two cases. First, if either |y| ≤ max{2|x|, 1} or |x − y| ≤ ε

then (2.3.117) readily follows from (2.3.115) and (2.3.116) (with |α| = 0), bearing
in mind that y is bounded. Second, consider the case when both |y| > max{2|x|, 1}
and |x − y| > ε. In particular, the former inequality implies that for each t ∈ [0, 1]
368 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

we have |t x − y| ≥ |y| − |x| ≥ 12 |y|. As such, F(t) := k(t x − y) for each t ∈ [0, 1] is
a well-defined function which is continuously differentiable on [0, 1]. Granted this,
the Mean Value Theorem together with (2.3.116) permit us to estimate
 
 k ε (x − y) − k1 (−y) = |F(1) − F(0)| ≤ sup |F (t)|
0<t<1
2n |x|
≤ |x| sup |(∇k)(t x − y)| ≤
0<t<1 |y| n
22n |x|
≤ , ∀y ∈ Rn . (2.3.118)
(1 + |y|)n
This completes the proof of (2.3.117). In concert with [112, Lemma 7.2.1], estimate
(2.3.117) implies that
∫
 
 k ε (x − y) − k1 (−y) | f (y)| dσ(y) < +∞
Σ  σ  (2.3.119)
for each ε > 0, each x ∈ Σ, and each f ∈ L 1 Σ, .
1+|·| n

Hence, the integrals appearing in the definition of (T  mod f )(x) in (2.3.31) are all
absolutely convergent for each x ∈ Σ and each f ∈ L 1 Σ, 1+σ| · | n .
To proceed, fix an arbitrary number R ∈ (1, ∞) and select an arbitrary  point
x ∈ Σ ∩ B(0, R). Also, pick some ε ∈ (0, 1). Having selected some f ∈ L 1 Σ, 1+σ| · | n ,
we then proceed to split
∫
 
k ε (x − y) − k1 (−y) f (y) dσ(y) = Iε + IIε (2.3.120)
Σ

where the term

∫
 
Iε := k ε (x − y) − k1 (−y) f (y) dσ(y)
y ∈Σ
|x−y |>1
∫
 
= k(x − y) − k1 (−y) f (y) dσ(y) (2.3.121)
y ∈Σ
|x−y | ≤1

turns out to be independent of ε > 0, and

∫
 
IIε := k ε (x − y) − k1 (−y) f (y) dσ(y). (2.3.122)
y ∈Σ
|x−y | ≤1

Given that |y| ≤ |x − y| + |x| ≤ 1 + R < 2R on the domain of integration in (2.3.122),

we may replace f by the function
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 369

fR := 1B(0,2R) · f which belongs to L 1 (Σ, σ) and has bounded support,

(2.3.123)

and write
∫
 
IIε = k ε (x − y) − k1 (−y) fR (y) dσ(y). (2.3.124)
y ∈Σ
|x−y | ≤1

Pressing on, observe that since k1 is bounded in Rn (cf. (2.3.116)), we have

k1 (−·) fR ∈ L 1 (Σ, σ). In light of this and (2.3.119), we may further decompose
(bearing (2.3.6) in mind)
∫ ∫
IIε = k ε (x − y) fR (y) dσ(y) − k1 (−y) fR (y) dσ(y)
y ∈Σ y ∈Σ
|x−y | ≤1 |x−y | ≤1
∫ ∫
= k(x − y) fR (y) dσ(y) − k(x − y) fR (y) dσ(y)
y ∈Σ y ∈Σ
|x−y |>ε |x−y |>1
∫
− k1 (−y) fR (y) dσ(y)
y ∈Σ
|x−y | ≤1
∫
= (Tε fR )(x) − k(x − y) fR (y) dσ(y)
y ∈Σ
|x−y |>1
∫
− k1 (−y) fR (y) dσ(y). (2.3.125)
y ∈Σ
|x−y | ≤1

Note that the last two integrals above are independent of ε (as well as absolutely
convergent). Also, thanks to (2.3.15) and (2.3.123), the limit lim+ (Tε fR )(x) exists
ε→0
for σ-a.e. x ∈ Σ ∩ B(0, R). Let us record our progress. In light of the arbitrariness of
R ∈ (1, ∞), the argument so far shows that

for each given function f in the space L 1 Σ, 1+σ| · | n , the limit
∫ (2.3.126)
 
lim+ k ε (x − y) − k1 (−y) f (y) dσ(y) exists for σ-a.e. x ∈ Σ.
ε→0 Σ

Upon also invoking (2.2.68), the equivalence (i) ⇔ (iii) in [112, Remark 3.1.2], and
[112, (3.6.26)], ultimately this proves that
370 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

the operator Tmod introduced in (2.3.31) is well defined on the

weighted Lebesgue space L 1 Σ, 1+σ(x)
|x | n which it maps linearly
(2.3.127)
into the space of σ-measurable functions on Σ.
Next, the claims made in relation to the operator (2.3.33) may be proved in an
analogous manner to the earlier justification of similar claims for the operator in
(2.3.16).  
Observe now that if f ∈ L 1 Σ, 1+σ(x)|x | n−1 then based on (2.3.115) and (2.3.116)
(with |α| = 0) we may estimate
∫ ∫
| f (y)|
|k1 (−y) f (y)| dσ(y) ≤ C dσ(y) < +∞. (2.3.128)
Σ 1 + |y|
n−1
Σ
∫
Having proved that Σ k1 (−y) f (y) dσ(y) is absolutely convergent, from (2.3.126),
(2.3.31), (2.3.6), and (2.3.15) we conclude that
∫ ∫
(Tmod f )(x) = lim+ k ε (x − y) f (y) dσ(y) − k1 (−y) f (y) dσ(y)
ε→0 Σ Σ

= (T f )(x) − C f at σ-a.e. x ∈ Σ, (2.3.129)

where the constant C f is given by

∫
C f := k1 (−y) f (y) dσ(y) ∈ C. (2.3.130)
Σ

In particular, from this and (2.3.128) we have |C f | ≤ CΣ,k ·  f  L 1 (Σ, σ(x)

)
for
1+| x | n−1
some finite constant CΣ,k > 0. This finishes the proof of (2.3.34).
Suppose next that
 σ(x)  p
f ∈ L 1 Σ, ∩ Lloc (Σ, σ) for some p ∈ (1, ∞), (2.3.131)
1 + |x| n
and pick an arbitrary surface ball Δ ⊆ Σ. For each x ∈ Δ decompose
    
Tmod f (x) = Tmod 12Δ f (x) + Tmod 1Σ\2Δ f (x). (2.3.132)

 that since 12Δ f ∈ L (Σ, σ) it follows from (2.3.34) that Tmod 12Δ f and
Note p

T 12Δ f differ by a constant. On account of this and (2.3.18) we then conclude that
 p
Tmod 12Δ f ∈ Lloc (Σ, σ). (2.3.133)

There remains to observe that, thanks to the nature of the constant in (2.3.117), for
each x ∈ Δ we may estimate
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 371
∫
    
T 1Σ\2Δ f (x) ≤  k(x − y) − k1 (−y) | f (y)| dσ(y)
mod
Σ\2Δ
∫
| f (y)|
≤ cΔ dσ(y) < +∞, (2.3.134)
1 + |y| n
Σ

for some constant cΔ ∈ (0, ∞) which depends on k and Δ but not on x. Since the last
quantity above is a finite constant independent of x, we conclude that

Tmod 1Σ\2Δ f is bounded on Δ. (2.3.135)

At this point, (2.3.35) is clear from (2.3.133)-(2.3.135).

To deal with the claim in (2.3.36) we refine the argument which has produced
(2.3.26). To get started, suppose p, p ∈ (1, ∞) satisfy 1/p + 1/p = 1 and pick two
functions,  
f ∈ Lloc (Σ, σ) ∩ L 1 Σ, 1+σ(x)
p
|x | n (2.3.136)

and ∫
p
g ∈ Lcomp (Σ, σ) with g dσ = 0. (2.3.137)
Σ

To fix ideas, choose xo ∈ Σ and r ∈ 0, 2diamΣ such that

supp g ⊆ Δ(xo, r) := B(xo, r) ∩ Σ. (2.3.138)

Granted (2.3.136), from (2.3.35) we conclude that

p
Tmod f ∈ Lloc (Σ, σ). (2.3.139)

Together
∫ with the membership stipulated (2.3.137), this property shows that we do
have Σ |Tmod f ||g| dσ < +∞, which is one of the claims in (2.3.36). Next, much as
in (2.3.112), for each x ∈ Σ such that |x − xo | ≥ 2r we obtain (based on (2.3.137)-
(2.3.138), (2.3.3), the Mean Value Theorem, and Hölder’s inequality)
∫ 
 
|(T g)(x)| =  k(x − y)g(y) dσ(y)
Σ
∫
≤ |k(x − y) − k(x − xo )||g(y)| dσ(y)
B(x o ,r)∩Σ
⨏
   r · σ(Δ(xo, r))
≤ Ck  n−1

𝒞1 (S n−1 )
|g(y)| dσ(y)
S |x − xo | n Δ(x o ,r)
⨏
   r · σ(Δ(xo, r))   1/p
≤ Ck   
n−1 𝒞 (S n−1 )
|g| p dσ
|x − xo | n
S 1
Δ(x o ,r)

C(Σ, n, p, k, xo, r, g)
≤ , (2.3.140)
1 + |x| n
372 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

for some finite constant C(Σ, n, p, k,∫xo, r, g) > 0. In concert with (2.3.136),
this pointwise inequality proves that Σ\B(x ,2r) | f ||T g| dσ < +∞. Since we also
∫ o
have Σ∩B(x ,2r) | f ||T g| dσ < +∞, thanks to (2.3.137), (2.3.18), and the fact that
o

∫f · 1Σ∩B(xo,2r) belongs to L (Σ, σ) (cf. (2.3.136)), we ultimately deduce that

p

Σ
| f ||T g| dσ < +∞, which is another claim in (2.3.36). Bearing this in mind,
we may then write
∫ ∫ ∫
f (T g) dσ = f (T g) dσ + f (T g) dσ =: I + II, (2.3.141)
Σ Σ∩B(x o ,2r) Σ\B(x o ,2r)

with the last equality defining I and II. In relation to these, note that on account of
(2.3.25) we may recast I as
∫ ∫  
 
I= f · 1Σ∩B(xo,2r) (T g) dσ = − T f · 1Σ∩B(xo,2r) g dσ
Σ Σ
∫  

=− Tmod f · 1Σ∩B(xo,2r) g dσ (2.3.142)
Σ
 
since T f · 1Σ∩B(xo,2r) differs from Tmod f · 1Σ∩B(xo,2r) by a constant on Σ (thanks
to (2.3.34), bearing in mind that f · 1Σ∩B(xo,2r) ∈ L p (Σ, σ)) and the fact that g has
integral zero (cf. the last property in (2.3.137)). Also,
∫ ∫ 
II = f (x) k(x − y)g(y) dσ(y) dσ(x)
Σ\B(x o ,2r) Σ

∫ ∫ 
= f (x) k ε (x − y)g(y) dσ(y) dσ(x)
Σ\B(x o ,2r) Σ

∫ ∫ 
 
= f (x) k ε (x − y) + k1 (−x) g(y) dσ(y) dσ(x)
Σ\B(x o ,2r) Σ

∫ ∫ 
 
= f (x) k ε (x − y) + k1 (−x) g(y) dσ(y) dσ(x)
Σ\B(x o ,2r) Σ∩B(x o ,r)

∫ ∫ 
 
= k ε (x − y) + k1 (−x) f (x) dσ(x) g(y) dσ(y)
Σ∩B(x o ,r) Σ\B(x o ,2r)

∫ ∫ 
 
= lim k ε (x − y) + k1 (−x) f (x) dσ(x) g(y) dσ(y)
ε→0+ Σ\B(x o ,2r)
Σ
∫  

=− Tmod f · 1Σ\B(xo,2r) (y) g(y) dσ(y). (2.3.143)
Σ
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 373

The first equality in (2.3.143) is implied by (2.3.15) bearing in mind that, thanks
to (2.3.138), the variables x, y are uniformly separated. The second equality in
(2.3.143) uses the definition of k ε in the second line of (2.3.31) and is valid for
each choice ε ∈ (0, r). The third equality in (2.3.143) is a consequence of the can-
celation property of the function g (cf. the last property in (2.3.137)), while the
fourth equality in (2.3.143) is seen from (2.3.138). The fifth equality in (2.3.143)
follows from Fubini’s Theorem whose applicability is presently ensured by the fact
that the double integral is absolutely convergent, thanks to the properties listed in
(2.3.137)-(2.3.138), the estimate in (2.3.117) (with the roles of x, y reversed, and
a constant which stays bounded for y in a compact subset of Σ; applicability here
uses the fact that k is odd), and (2.3.136). The sixth equality in (2.3.143) uses the
fact that the inner integral is actually independent of ε ∈ (0, r), and also the sup-
port property of g (cf. (2.3.138)). Finally, the last equality in (2.3.143) is seen from
(2.3.31) (bearing in mind that k is odd). Collectively, from (2.3.141)-(2.3.143) we
conclude that ∫ ∫

f (T g) dσ = − Tmod f g dσ, (2.3.144)
Σ Σ
finishing the proof of (2.3.36).
Moving on, fix a function f ∈ L ∞ (Σ, σ) and pick an arbitrary surface ball Δ ⊆ Σ,
of center xΔ ∈ Σ and radius rΔ ∈ (0, ∞). Introduce
∫
C f , Δ := k1 (−y) f (y) dσ(y) ∈ C, (2.3.145)
2Δ

and for each x ∈ Δ define

   
I(x) := Tmod 12Δ f (x) = T 12Δ f (x) + C f ,Δ (2.3.146)

where the second equality is a consequence of (2.3.129)-(2.3.145), bearing in mind

that 12Δ f belongs to L 1 (Σ, σ). Then, on account of (2.3.146), (2.3.20) (with p := 2),
and the fact that σ is doubling, we may estimate
⨏  ⨏    1/2
 
I(x) − C f ,Δ | dσ(x) ≤  T 12Δ f (x)2 dσ(x)
Δ Δ
 
≤ σ(Δ)−1/2 T(12Δ f ) L 2 (Σ,σ)
 
≤ Cσ(Δ)−1/2 12Δ f  L 2 (Σ,σ) ≤ C f  L ∞ (Σ,σ) . (2.3.147)

Next, introduce
∫
 
f ,Δ :=
C k(xΔ − y) − k1 (−y) f (y) ∈ C, (2.3.148)
Σ\2Δ

(with the membership guaranteed by (2.3.119) and [112, (7.4.118)]) and for each
x ∈ Δ define
374 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
   
II(x) := Tmod 1Σ\2Δ f (x) = k(x − y) − k1 (−y) f (y) dσ(y)
Σ\2Δ
∫
 
= f ,Δ
k(x − y) − k(xΔ − y) f (y) dσ(y) + C (2.3.149)
Σ\2Δ

where the second equality is a consequence of the fact that the variables x ∈ Δ and
y ∈ Σ \ 2Δ are separated. Based on (2.3.149), the Mean Value Theorem, (2.3.116),
and [112, Lemma 7.2.1], for each x ∈ Δ we may then estimate
∫
  
II(x) − C
f ,Δ | ≤  k(x − y) − k(xΔ − y) | f (y)| dσ(y)
Σ\2Δ
∫
rΔ
≤ C f  L ∞ (Σ,σ) dσ(y)
Σ\2Δ |y − xΔ | n
≤ C f  L ∞ (Σ,σ), (2.3.150)

for some C ∈ (0, ∞) which depends only on Σ. Consequently,

⨏

II(x) − C
f ,Δ | dσ(x) ≤ C f  L ∞ (Σ,σ) . (2.3.151)
Δ

If we now set
f ,Δ ∈ C
c f ,Δ := C f ,Δ + C (2.3.152)

then from (2.3.146), (2.3.149), (2.3.147), and (2.3.151) we conclude that

⨏ ⨏ ⨏
   
T f − c f ,Δ  dσ ≤ I(x) − C f ,Δ | dσ + II(x) − C
f ,Δ | dσ
mod
Δ Δ Δ

≤ C f  L ∞ (Σ,σ) . (2.3.153)

In view of [112, (7.4.57)] this ultimately establishes (2.3.37).

We shall next consider the claim formulated in (2.3.39). Work under the assump-
tion that Σ is bounded, a scenario guaranteeing that the space L ∞ (Σ, σ) is contained
(
in L p (Σ, σ). Granted this, we see from (2.3.128) that if C f is associated with
1≤p ≤∞
f as in (2.3.130) then there exists a constant c ∈ (0, ∞), which depends only on Σ
and k, such

|C f | ≤ c f  L ∞ (Σ,σ) for every f ∈ L ∞ (Σ, σ). (2.3.154)

With this in hand, (2.3.39) follows with the help of (2.3.34) and (2.3.37).
To deal with the claim in (2.3.40), recall from (2.3.32) that
 σ(x)   
L ∞ (Σ, σ) ⊆ L 1 Σ,
p
∩ L (Σ, σ) . (2.3.155)
1 + |x| n 1<p<∞
loc
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 375

Pick an arbitrary function f ∈ L ∞ (Σ, σ). Let us also fix a background exponent

q ∈ (1, ∞) and consider an arbitrary (1, q)-atom a on Σ with vanishing moment22.
Specifically, this a σ-measurable
 function defined on Σ, with the property that there
exist xo ∈ Σ and r ∈ 0, 2diamΣ such that
 1/q−1
supp a ⊆ B(xo, r) ∩ Σ, a L q (Σ,σ) ≤ σ B(xo, r) ∩ Σ ,
∫ (2.3.156)
and a dσ = 0.
Σ

From (2.3.36), (2.3.155), (2.3.156) we then conclude that

∫ ∫
|Tmod f ||a| dσ < +∞, | f ||T a| dσ < +∞, (2.3.157)
Σ Σ

and ∫ ∫

f (T a)dσ = − Tmod f a dσ. (2.3.158)
Σ Σ
Having proved (2.3.157)-(2.3.158), the claim in (2.3.40) now follows with the help
of [113, Theorem 4.4.1], the duality result from [113, Theorem 4.6.1, (4.6.8)] (cf.
[113, (4.6.8)], plus the very last claim in its statement), (2.3.27) with p = 1, and the
current item (5) (in the case when Σ is bounded). This completes the treatment of
item (7).
Consider next the claims made in item (8). For starters, (2.3.41) is a direct
corollary of [112, (7.7.107)] and (2.3.15). Next, thanks to [112, (7.7.106)] and
(2.3.18), it follows that

T : L p0 (Σ, σ) + L p1 (Σ, σ) −→ L p0 (Σ, σ) + L p1 (Σ, σ)

(2.3.159)
is well defined, linear, and bounded for all p0, p1 ∈ (1, ∞).

Based on (2.3.159), [112, (6.2.48)], and [113, Proposition 1.3.7] we then conclude
that the claims about the operator (2.3.43) are valid. With this in hand, the Cotlar-
type inequality recorded in (2.2.38) (presently used with p := 1 and r := 1) together
with the boundedness of the Hardy-Littlewood maximal operator on Lorentz spaces
(cf. [112, (6.2.20)]) justify the claims made about the operator (2.3.42). Finally, from
(2.3.25) and [112, (6.2.51), (6.2.62)-(6.2.63)] we see that (2.3.44) holds.
Regarding item (9), since the operators from (2.3.27) and (2.3.18) are compatible
with one another (cf. the compatibility claim in item (3) of Theorem 2.3.2 and the
very last part of
 item (6) of Theorem 2.3.2), the claim in (2.3.45) follows. In turn, for
each p0, p1 ∈ n−1 n , ∞ , this permits us to define, in a coherent fashion, the mapping
p0, p1 as in (2.3.46).
T
Going further, fix two exponents p0, p1 with n−1 n < p0 < p1 < ∞. As a byproduct

of the lack of ambiguity in the definition of Tp0, p1 above, plus the fact that Tp0 , Tp1
are linear and bounded, it follows that

22 this is always the case if Σ is unbounded

376 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

p0, p1 is a linear and bounded operator, which satisfies

T
  (2.3.160)
p0, p1 
T = Tp0 as well as p0, p1 
T = Tp1 .
H p0 (Σ,σ) H p 1 (Σ,σ)

Granted (2.3.46) and (2.3.160), we may use real interpolation (cf. [113, Proposi-
tion 1.3.7, (1.3.66)]), together with [113, (4.3.3)] and [112, (6.2.48)], to conclude
p0, p1 from (2.3.46) induces a linear and bounded mapping
that the operator T

p0, p1 : H p (Σ, σ) −→ L p (Σ, σ) for each p ∈ (p0, p1 ).

T (2.3.161)

Moreover, given any p ∈ (p0, p1 ), since H p0 (Σ, σ) ∩ H p1 (Σ, σ) is a dense subspace

of H p (Σ, σ) (cf. [113, (4.3.145)]) and since

p0, p1 f = Tp0 f = Tp f
T
(2.3.162)
for each f ∈ H p0 (Σ, σ) ∩ H p1 (Σ, σ),

where the first equality comes from (2.3.160) and the second equality is implied by
(2.3.45), we ultimately conclude that the properties claimed in (2.3.47) are all valid.
Consider now the case when we have n−1 n < q0 < p0 < p1 < q1 < ∞ and pick
some arbitrary f ∈ H p0 (Σ, σ) + H p1 (Σ, σ). In particular, there exist f0 ∈ H p0 (Σ, σ)
and f1 ∈ H p1 (Σ, σ) such that f = f0 + f1 . Then repeated applications of property
recorded in the second line of (2.3.47) allows us to compute
q0,q1 f = T
T q0,q1 f0 + T
q0,q1 f1 = Tp0 f0 + Tp1 f1 = T
p0, p1 f . (2.3.163)

This goes to show that Tq0,q1 is an extension of Tp0, p1 , from which the claim in (2.3.48)
follows. Next, since [113, (1.3.41)] and the real interpolation result established in
[113, (4.3.4)] ensure that

H p (Σ, σ) ⊆ H p0 (Σ, σ) + H p1 (Σ, σ)

(2.3.164)
whenever n−1
n < p0 < p < p1 < ∞,

we conclude that HSum (Σ, σ) defined in (2.3.49) is indeed a linear space. Likewise,
LSum (Σ, σ) defined in (2.3.50) is a linear space, thanks to [112, (6.2.25), (6.2.52)].
Thus, introducing

T : HSum (Σ, σ) −→ LSum (Σ, σ) by setting T f := T

p0, p1 f
 (2.3.165)
whenever f ∈ H p0 (Σ, σ) + H p1 (Σ, σ) for some p0, p1 ∈ n−1 n ,∞ ,

yields an unambiguously defined linear mapping which satisfies

 

T =Tp0, p1 for each p0, p1 ∈ n−1 , ∞ , (2.3.166)
p
H p
0 (Σ,σ)+H 1 (Σ,σ) n

and which is unique with this property. If, much as in (2.3.161), we now use real
interpolation (again, relying on [113, Proposition 1.3.7, (1.3.66), (4.3.3)] and [112,
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 377

(6.2.48)]), we conclude that the operator T from (2.3.51) induces a linear and bounded
mapping

 : H p,q (Σ, σ) −→ L p,q (Σ, σ)

T
 (2.3.167)
for each p ∈ n−1 n , ∞ and q ∈ (0, ∞],

obeying (cf. [113, (1.3.64)]) the operator quasi-norm estimate

  
 H p, q (Σ,σ)→L p, q (Σ,σ) ≤ CΣ, p,q  k  n−1  N n−1 .
T (2.3.168)
S 𝒞 (S )
 n−1
Given any p, p∗ ∈ n , ∞ and q ∈ (0, ∞], choose p0, p1 such that
n−1
n < p0 < min{p, p∗ } ≤ max{p, p∗ } < p1 < ∞, (2.3.169)

then for each f ∈ H p∗ (Σ, σ) ∩ H p,q (Σ, σ) write

f = T
T p0, p1 f = Tp∗ f , (2.3.170)

thanks to the definition of T  and the second line in (2.3.47). This shows that T 
from (2.3.167) is compatible with T from (2.3.18) and (2.3.27). As such, there is
no ambiguity if we may retain the same notation, i.e., T, for the operator (2.3.167).
In such a scenario, (2.3.52)-(2.3.53) become consequences of (2.3.167)-(2.3.168).
Also, the property claimed in  (2.3.54) is a manifestation of the fact that T acting
on any H p,q (Σ, σ) with p ∈ n−1 n , ∞ and q ∈ (0, ∞] is merely the restriction of the

linear operator T : HSum (Σ, σ) → LSum (Σ, σ) to H p,q (Σ, σ). Finally, from [113,
(4.3.145)] we see that (2.3.55) holds.
Let us now deal with item (10). As regards (2.3.56)-(2.3.57), see [19], [104,
Corollaire 2, p. 254] for the Euclidean setting, and [93] for the setting of spaces
of homogeneous type. In turn, from (2.3.56), Cotlar’s inequality from item (ii) in
Proposition 2.2.5 (which, thanks to (2.3.19), is presently applicable with p := 1
and r := 1), and item (1) in [112, Lemma 7.7.1] we also see that (2.3.58) holds.
Parenthetically, we note that we could have also obtained this conclusion based on
good-λ inequalities for Tmax and M Σ , in the spirit of [104, Théorème 9, pp. 252-253];
cf. also [104, Corollaire 1, Corollaire 2, p. 254]. Next, (2.3.61) is a consequence of
(2.3.25) and a standard density argument. Also, the claim pertaining to (2.3.62) may
be justified by reasoning much as in the proof of (2.3.17) (now keeping in mind
(2.3.56)). Lastly, (2.3.63) is a consequence of (2.7.115), (2.3.57), and (2.3.61).
As regards item (11), the claims made in (2.3.65)-(2.3.66) are consequences of
what we have proved in the current item (10) and [113, Corollary 5.2.3]. Let us turn
our attention to (2.3.68). To set the stage, bring in an exponent q  ∈ (1, ∞) associated
with X as in [113, Proposition 5.2.7, (5.2.90)], and recall from the current item (3)

that T is a well-defined linear and bounded operator from the space L q (Σ, σ) into
itself. In particular,
 q 
T Lcomp (Σ, σ) ⊆ L q (Σ, σ). (2.3.171)
378 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

q
To proceed, fix a point x0 ∈ Σ, and select an arbitrary function f ∈ Lcomp (Σ, σ).
From (2.3.41) and the nature of the kernel k (cf. (2.3.3)), we see that there exists a
constant C f ∈ (0, ∞), which is allowed to depend on f , with the property that for
each sufficiently large j ∈ N we have

1Σ\Δ(x0, j)
1Σ\Δ(x0, j) · |T f | ≤ C f · at σ-a.e. point in Σ. (2.3.172)
1 + | · −x0 | n−1
In addition, (2.3.171) implies that
q
φ j := 1Δ(x0, j) · T f ∈ Lcomp (Σ, σ) for each j ∈ N. (2.3.173)

For each large j ∈ N now estimate

   1Σ\Δ(x0, j) 

T f − φ j X = 1Σ\Δ(x0, j) · |T f | X ≤ C f  
1 + | · −x0 | n−1 X
 1 1 + log+ | · −x0 | 
 Σ\Δ(x0, j)
= Cf  · 
1 + log+ | · −x0 | 1 + | · −x0 | n−1 X
C f  1 + log+ | · −x0 | 
≤  
ln j 1 + | · −x0 | n−1 X
C f  

≤ (M Σ ◦ M Σ )(1Δ(x0,1) )
ln j X

C f  2  
≤ M Σ X→X 1Δ(x0,1) X (2.3.174)
ln j
where the first equality is seen from (2.3.173), the subsequent inequality is a con-
sequence of (2.3.172) (keeping in mind property (P2) in [113, Definition 5.1.1]),
the second equality is clear, the penultimate inequality is based on [112, (7.6.69)]
and once again the monotonicity of the norm in X, while the final inequality follows
from (2.3.64).
 Given
 that (2.3.64) entails M Σ X→X < ∞, and that [113, (5.2.75)]
implies 1Δ(x0,1) X < ∞, this proves that

φ j → T f in X as j → ∞. (2.3.175)

Together with (2.3.173) and (2.3.67), this ultimately proves that T f ∈ X̊. Thus,
 q
T Lcomp (Σ, σ) ⊆ X̊. (2.3.176)

From (2.3.176), (2.3.67), and item (1) in [113, Lemma 1.2.20] we then deduce
that T maps X̊ linearly and boundedly into itself. Thanks to the fact that X̊ is a
closed linear subspace of X, it follows that T X̊→X̊ is dominated
  by T X→X (cf.
[113, (1.2.20)]), and the latter has been shown to be ≤ C  k S n−1 𝒞 N (S n−1 ) where
the constant C ∈ (0, ∞) depends only on Σ and the operator norms of M Σ on X
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 379

and X. Once this has been established, the corresponding properties for Tmax listed
in (2.3.68) become consequences of the Cotlar-type inequality recorded in (2.3.23)
(presently used with r := 1), whose applicability is ensured by the embedding
X → L 1 Σ, 1+σ(x)
|x | n−1
(see [113, (5.2.88)]), the assumption made in (2.3.64), and the
monotonicity of the norm in X (see (P2) in [113, Definition 5.1.1]).
Finally, to justify the transposition formula claimed in (2.3.69), fix f ∈ X̊ and
g ∈ X, both arbitrary. From [113, (5.2.90)] we know that there exists some p ∈ (1, ∞)
such that  
g ∈ X ⊆ Lloc (Σ, σ) ∩ L 1 Σ , 1+σ(x)
p
|x | n−1 . (2.3.177)

In addition, since Lcomp∞ (Σ, σ) is dense in X̊ (cf. (2.3.67)), there exists a sequence

{ f j } j ∈N ⊆ Lcomp (Σ, σ) which converges to f in X. In particular, if p ∈ (1, ∞)

∞
p
is such that 1/p + 1/p = 1 then each f j belongs to Lcomp (Σ, σ). Granted these
properties we may invoke (2.3.26) and conclude that
∫ ∫
(T f j )g dσ = − f j (T g) dσ for each j ∈ N. (2.3.178)
Σ Σ

Passing to limit j → ∞ in (2.3.178) then yields (2.3.69), on account of the fact that
T is continuous both on X and on X (cf. (2.3.66) and (2.3.66)) plus the generalized
Hölder inequality recorded in [113, Proposition 5.1.12]. All claims in item (11) are
therefore justified. This finishes the proof of Theorem 2.3.2. 
In the last part of this section we wish to further elaborate on Tmod , the modified
version of the principal-value singular integral operator T, originally introduced in
item (7) of Theorem 2.3.2.
Proposition 2.3.3 Suppose Σ ⊆ Rn , n ∈ N with n ≥ 2, is a UR set, and abbreviate
σ := H n−1 Σ. Having picked a sufficiently large integer N = N(n) ∈ N, consider a
kernel k ∈ 𝒞 N (Rn \ {0}) which is odd and positive homogeneous of degree 1 − n.
In this setting, associate with Σ and k the principal-value singular integral operator
T as in (2.3.15), together with its modified version Tmod , defined as in (2.3.31). Then
the following statements are true:
 
(i) For each function f ∈ L 1 Σ, 1+σ(x)
|x | n and each surface ball Δ ⊆ Σ there exists a
constant C f ,Δ ∈ R with the property that for σ-a.e. point x ∈ Δ one has
∫
  
(Tmod f )(x) = T f 1Δ (x) + k(x − y) − k(xΔ − y) f (y) dσ(y) + C f ,Δ,
Σ\Δ
(2.3.179)
where xΔ ∈ Σ is the center of the surface ball Δ.
(ii) Fix a reference point x0 ∈ Σ and for each
 j ∈ N set Δ j := B(x0, j) ∩ Σ. Given
σ(x)
an arbitrary function f ∈ L Σ, 1+ |x | n , for each j ∈ N define
1

∫

g j (x) := T f 1Δ j (x) − k(x0 − y) f (y) dσ(y) for σ-a.e. x ∈ Σ. (2.3.180)
Δ j \Δ1
380 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Then the sequence {g j } j ∈N converges pointwise σ-a.e. on Σ. In particular,

  ∫ 
mod f )(x) := lim T f 1Δ j (x) −
(T k(x0 − y) f (y) dσ(y) (2.3.181)
j→∞ Δ j \Δ1

f ,x0 ∈ R
is well defined at σ-a.e. point x ∈ Σ. Moreover, there exists a constant C
with the property that
mod f )(x) = (Tmod f )(x) + C
(T f ,x0 for σ-a.e. x ∈ Σ. (2.3.182)
 
Proof Pick a function f ∈ L 1 Σ, 1+σ(x) |x | n and a surface ball Δ ⊆ Σ, with center
xΔ ∈ Σ and radius R. Define the constant
∫
 
C f ,Δ := k R (xΔ − y) − k1 (−y) f (y) dσ(y) ∈ R. (2.3.183)
Σ

At σ-a.e. point x ∈ Δ we may then use (2.3.31), (2.3.183), and (2.3.15) to write
∫
 
(Tmod f )(x) = lim+ k ε (x − y) − k1 (−y) f (y) dσ(y)
ε→0 Σ
∫
 
= lim+ k ε (x − y) − k R (xΔ − y) f (y) dσ(y) + C f ,Δ
ε→0 Σ
∫
 
= lim+ k ε (x − y) − k R (xΔ − y) f (y) dσ(y)
ε→0 Δ
∫
 
+ lim+ k ε (x − y) − k R (xΔ − y) f (y) dσ(y) + C f ,Δ
ε→0 Σ\Δ
∫
  
= T f 1Δ (x) + k(x − y) − k(xΔ − y) f (y) dσ(y)
Σ\Δ

+ C f ,Δ . (2.3.184)

The last equality is a consequence of two observations. First, k R (xΔ − y) = 0 for

each y ∈ Δ. Second, since x ∈ Δ it follows that if ε > 0 is small enough then
k ε (x − y) = k(x − y) for each y ∈ Σ \ Δ. This finishes the proof of (2.3.179).
Turning attention to item (ii), fix an arbitrary compact set F ⊆ Σ and choose
jF ∈ N such that F ⊆ Δ jF . Then for each j ∈ N with j ≥ jF and σ-a.e. point x ∈ F
we have
2.3 Principal Value Singular Integral Operators on Uniformly Rectifiable Sets 381
∫

g j (x) − g jF (x) = T f 1Δ j \Δ j F (x) − k(x0 − y) f (y) dσ(y)
Δ j \Δ j F
∫ ∫
= k(x − y) f (y) dσ(y) − k(x0 − y) f (y) dσ(y)
Δ j \Δ j F Δ j \Δ j F
∫
 
= k(x − y) − k(x0 − y) f (y) dσ(y). (2.3.185)
Δ j \Δ j F

Granted this, Lebesgue’s Dominated Convergence Theorem applies and shows that
at σ-a.e. point x ∈ F we have
∫
 
lim g j (x) = g jF (x) + k(x − y) − k(x0 − y) f (y) dσ(y). (2.3.186)
j→∞ Σ\Δ j F

Hence, at σ-a.e. point x ∈ F we may write

∫
 

(Tmod f )(x) = lim g j (x) = g jF (x) + k(x − y) − k(x0 − y) f (y) dσ(y)
j→∞ Σ\Δ j F
∫

= T f 1Δ j F (x) − k(x0 − y) f (y) dσ(y)
Δ j F \Δ1
∫
 
+ k(x − y) − k(x0 − y) f (y) dσ(y)
Σ\Δ j F

= (Tmod f )(x) + CF, f ,x0 , (2.3.187)

where CF, f ,x0 ∈ R is a constant, defined as

∫
CF, f ,x0 := − k(x0 − y) f (y) dσ(y) − C f ,Δ j F , (2.3.188)
Δ j F \Δ1

with C f ,Δ j F as in (2.3.183) (with jF in place of R, and x0 in place of xΔ ). In

(2.3.187), the first equality comes from (2.3.180)-(2.3.181), the second equality is
(2.3.186), the third equality uses (2.3.180) (with jF in place of j), and the final
equality is implied by (2.3.179) (presently employed with Δ := B(x0, jF ) ∩ Σ) plus
the definition made in (2.3.188).
In summary, (2.3.187) proves that T mod f − Tmod f is a constant function on each
compact subset of Σ. In turn, this shows that the constant CF, f ,x0 is actually inde-
pendent of F, which then establishes (2.3.182). 
382 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

2.4 Boundary-to-Domain Integral Operators on Open sets with

Uniformly Rectifiable Boundaries

Theorem 2.3.2 is one of the main ingredients in the proof of our main result in this
section, contained in Theorem 2.4.1 below. Items (1)-(5) deal with nontangential
maximal function estimates for integral operators mapping functions defined on a
UR set Σ ⊆ Rn into functions defined in Rn \ Σ (refining [63, Propositions 3.20-3.21,
pp. 2641-3243], originally proved in a more restrictive geometric setting; see the
preamble to the statement of Theorem 2.3.2). Items (6)-(8) in this theorem, containing
mixed-norm, area-function, square-function, and Carleson estimates, make essential
use of work in [61]. Items (9)-(10) in Theorem 2.4.1 deal with off-diagonal Carleson
measure estimates for singular integral operators on UR sets. Recall that given an
open set Ω ⊆ Rn with an Ahlfors regular boundary, a Borel measure μ in Ω is said
to be a Carleson measure provided its Carleson constant is finite, i.e.,

μ B(x, r) ∩ Ω
sup  < +∞. (2.4.1)
r ∈(0,2diam(∂Ω)) H
n−1 B(x, r) ∩ ∂Ω
and x ∈∂Ω

For example, a simple application of Hölder’s inequality shows that

if Ω ⊆ Rn is an open set with an Ahlfors regular boundary
and if F ∈ L n (Ω, L n ) is some non-negative function, then (2.4.2)
μ := FL n is a Carleson measure in Ω.
Finally, item (11) contains regularity results, with smoothness measured on the scale
of Triebel-Lizorkin spaces.

Theorem 2.4.1 Fix n ∈ N with n ≥ 2 and suppose Σ ⊆ Rn is a UR set. Abbreviate

Σ c := Rn \ Σ, which entails Σ = ∂(Σ c ), and pick an arbitrary aperture parameter
c
κ ∈ (0, ∞). Let ΓΣκ (x), with x ∈ Σ, stand for the nontangential approach regions with
c
aperture parameter κ in the open set Σ c . Also, denote by NκΣ the corresponding
nontangential maximal operator with aperture parameter κ in the open set Σ c .
Finally, define σ := H n−1 Σ, and let H p (Σ, σ) with n−1 n < p < ∞ denote the scale
of Hardy spaces on Σ, as introduced in [113, Definition 4.2.1].
Consider next a kernel k ∈ 𝒞 N (Rn \ {0}), with N ≥ 2, which is odd and positive
homogeneous of degree 1 − n (cf. (2.3.3)), then define the integral operator T acting

on L 1 Σ, 1+σ(y)
|y | n−1
, i.e., on σ-measurable functions f : Σ → C satisfying
∫
| f (y)|
dσ(y) < +∞, (2.4.3)
Σ 1 + |y| n−1
according to
∫
(T f )(x) := k(x − y) f (y) dσ(y), ∀x ∈ Rn \ Σ. (2.4.4)
Σ
2.4 Boundary-to-Domain Integral Operators on Open sets with UR Boundaries 383

In relation to this, the following statements are true.

(1) There exists some constant C = C(Σ, κ, n) ∈ (0, ∞) such that

   
NκΣ (T f )(x) ≤ (Tmax f )(x) + C  k S n−1 𝒞1 (S n−1 ) M Σ f (x),
c

  (2.4.5)
for each function f ∈ L 1 Σ, 1+σ(y)
|y | n−1
and each point x ∈ Σ,

where Tmax is the maximal singular integral operator defined as in (2.3.4)-(2.3.6),

and M Σ is the Hardy-Littlewood maximal operator on the set23 Σ.
As a consequence of (2.4.5) and the Cotlar-type inequality from (2.3.23), in
the case when the smoothness exponent N of the kernel k is sufficiently large
(depending only on the dimension n ∈ N of the ambient Euclidean space) it
follows that for each r ∈ (0, ∞) one has
  1/r    
+ C  k S n−1 𝒞 N (S n−1 ) M Σ f (x),
c
NκΣ (T f )(x) ≤ C M Σ (|T f | r ) (x)
 
for each function f ∈ L 1 Σ, 1+σ(y)
|y | n−1
and each point x ∈ Σ,
(2.4.6)
where T is the principal-value singular integral operator associated with the
kernel k and the set Σ as in (2.3.15), and C ∈ (0, ∞) depends only on n, the UR
constants of Σ, and r.
(2) Assume the smoothness exponent N of the kernel k is sufficiently large, depend-
ing only on the dimension n ∈ N of the ambient Euclidean space. Then the
assignment
 σ(x)  c
L 1 Σ,  f −→ NκΣ (T f ) ∈ Lloc
1,∞
(Σ, σ) (2.4.7)
1 + |x| n−1

is well defined, sub-linear, bounded24 and continuous, and for each p ∈ (1, ∞)
it induces a well-defined, sub-linear, bounded, and continuous mapping
 σ(x)  c
 f −→ NκΣ (T f ) ∈ Lloc (Σ, σ).
p p
Lloc (Σ, σ) ∩ L 1 Σ, (2.4.8)
1 + |x| n−1
(3) Assume the smoothness exponent N of the kernel k is sufficiently large, depending
only on the dimension n ∈ N of the ambient Euclidean space. Then for each
p ∈ [1, ∞) there exists a constant C = C(Σ, p, κ, n) ∈ [0, ∞) such that for every
f ∈ L p (Σ, σ) one has

23 regarded as a space of homogeneous type when equipped with the Euclidean distance and the
measure σ; cf. the discussion in [112, Chapter 7]

24 i.e., it maps bounded subsets of L 1 Σ, 1+|σ(x)
x | n−1
1,∞
into bounded subsets of Lloc (Σ, σ)
384 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 Σc    
N (T f ) p ≤ C  k S n−1 𝒞 N (S n−1 )  f  L p (Σ,σ) if 1 < p < ∞, (2.4.9)
κ L (Σ,σ)

 Σc    
N (T f ) 1,∞ ≤ C  k S n−1 𝒞 N (S n−1 )  f  L 1 (Σ,σ) if p = 1. (2.4.10)
κ L (Σ,σ)

In addition, in view of work in [112, §8.5], similar estimates to (2.4.9)-(2.4.10)

are valid with the nontangential maximal operator replaced by the tangential
maximal operator (associated as in [112, Definition 8.5.1] with Ω := Σ c and
a sufficiently large power M; cf. (A.0.85)). Also, if for each p ∈ (0, ∞] one
considers (in agreement with (A.0.98))
 c

∞
(Σ c, L n ) : u Nκp (Σ c ;σ) := NκΣ u L p (Σ,σ) < +∞ ,
p
Nκ (Σ c ; σ) := u ∈ Lloc
(2.4.11)
then
p
T : L p (Σ, σ) −→ Nκ (Σ c ; σ) is a well-defined, linear
(2.4.12)
and bounded operator for each exponent p ∈ (1, ∞).
Moreover, the operator T as originally defined in (2.4.3)-(2.4.4) further extends
in a unique and coherent fashion to a linear mapping
p 
T : H p (Σ, σ) −→ Nκ (Σ c ; σ) for each p ∈ n−1
n ,1 , (2.4.13)

which, for each f ∈ H p (Σ, σ) with n−1

n < p ≤ 1, satisfies (for some Cp ∈ (0, ∞)
independent of f )
 Σc    
N (T f ) p ≤ Cp  k S n−1 𝒞 N (S n−1 )  f  H p (Σ,σ), (2.4.14)
κ L (Σ,σ)

plus a similar estimate in which the nontangential maximal operator is replaced

by the tangential maximal operator (associated as in [112, Definition 8.5.1] with
Ω := Σ c and a sufficiently large power M; cf. (A.0.85)).
An actual formula for said extension is given by

∞ ∫
(T f )(x) = λj k(x − y)a j (y) dσ(y), ∀x ∈ Rn \ Σ, (2.4.15)
j=1 Σ

with uniform
∞ convergence of the series on compact subsets of Rn \ Σ, whenever
f = j=1 λ j a j in H p (Σ, σ) for some sequence {λ j } j ∈N ∈  p (N) and family
{a j } j ∈N of H p -atoms on Σ. In fact, the same formula remains valid if the
a j ’s happen to be H p -molecules on Σ (in the sense of [113, Definition 4.5.1]).
 n−1 the aforementioned extension acts on each given f ∈ H (Σ, σ)
Alternatively, p

with p ∈ n , 1 according to
%
k(x − ·), f  if Σ is bounded,
(T f )(x) = % ∀x ∈ Rn \ Σ. (2.4.16)
[k(x − ·)], f  if Σ is unbounded,
2.4 Boundary-to-Domain Integral Operators on Open sets with UR Boundaries 385

Above, ·, · stands for the duality bracket described in [113, Theorem 4.6.1].
In addition, if Σ is bounded,
 k(x − ·) is regarded as a function in 𝒞α (Σ) with
α := (n − 1) p1 − 1 if p ∈ n−1 , 1 , and as a function in L ∞ (Σ, σ) ⊆ BMO(Σ, σ)
n
if p = 1. Also, if Σ is unbounded,  [k(x − ·)] is the equivalence class. (modulo)
constants) of the function k(x − ·)Σ , considered in the quotient space 𝒞α (Σ) ∼
 n−1
if p ∈ n , 1 , and in the quotient space BMO(Σ,* σ) in the case when p = 1.
Finally, whenever f ∈ H p (Σ, σ) has compact support (as a distribution, which
is automatically the case if Σ is bounded) and ψ ∈ Lipc (Σ) is identically one
near supp f one has
#  &
(T f )(x) = Lipc (Σ) ψk(x − ·)∂Ω, f (Lipc (Σ)) for each x ∈ Rn \ Σ,
(2.4.17)
where ψ ∈ Lipc (Σ) is identically one near supp f .

(4) Assume the smoothness exponent N of the kernel k is large enough (depending
on n ∈ N), and recall that Ap (Σ, σ), 1 ≤ p < ∞, are the classes of Muckenhoupt

weights associated as in (A.0.3) with the space of homogeneous type Σ, |·−· |, σ
(cf. [112, Example 7.4.1]). Then

for each p ∈ (1, ∞) and w ∈ Ap (Σ, σ) there exists

C = C(Σ, p, [w] A p ) ∈ (0, ∞) with the property that
 Σc     (2.4.18)
N (T f ) p ≤ C  k S n−1 𝒞 N (S n−1 )  f  L p (Σ,wσ)
κ L (Σ,wσ)
for each function f ∈ L p (Σ, wσ).

In fact, a more general result is valid. To state it, bring back the Hardy-Littlewood
maximal operator M Σ on Σ, and assume X is a Generalized Banach Function
Space on (Σ, σ) such that

M Σ : X → X and M Σ : X → X
(2.4.19)
are well-defined bounded mappings,

where X is the associated space of X (cf. [113, Definitions 5.1.4, 5.1.11]). Then
there exists a constant C ∈ (0, ∞), which depends only on Σ and the operator
norms of M Σ on X and X, with the property that
 Σc    
N (T f ) ≤ C  k  n−1  N n−1  f X for each f ∈ X,
κ X S 𝒞 (S )
 Σc     (2.4.20)
N (T f )  ≤ C  k  n−1  N n−1  f X for each f ∈ X .
κ X S 𝒞 (S )

Also, if X̊ is defined in relation to X as in [113, Definition 5.2.6], i.e.,

∞ (Σ, σ)  · X
X̊ := Lcomp , (2.4.21)

then, under the assumptions made in (2.4.19), one also has

386 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
c
NκΣ (T f ) ∈ X̊ for each f ∈ X̊. (2.4.22)

Finally, in light of work in [112, §8.5], estimates similar to (2.4.18) and (2.4.20)
are valid with the nontangential maximal operator replaced by the tangential
maximal operator (associated as in [112, Definition 8.5.1] with Ω := Σ c and a
sufficiently large power M; cf. (A.0.85)).
(5) Assume the smoothness exponent N of the kernel k is sufficiently large, depending
on the dimension n of the ambient Euclidean space. Then for each p ∈ (1, ∞)
and q ∈ (0, ∞] there exists a constant C ∈ (0, ∞) such that
 Σc 
N (T f ) p, q ≤ C f  L p, q (Σ,σ) for all f ∈ L p,q (Σ, σ). (2.4.23)
κ L (Σ,σ)

More generally, the operator T originally defined as in (2.4.3)-(2.4.4) further ex-

tends, in a coherent fashion,  to a well-defined mapping acting on each arbitrary
f ∈ H p,q (Σ, σ) with p ∈ n−1 n , ∞ and q ∈ (0, ∞] according to
% 
(T f )(x) = (H p, q (Σ,σ))∗ k(x − ·)Σ, f  H p, q (Σ,σ) for all x ∈ Rn \ Σ. (2.4.24)

Then
the identity recorded in (2.4.17) continues to be valid in the case
(2.4.25)
when f ∈ H p,q (Σ, σ) has compact support (as a distribution).
In particular, (2.4.25) and [113, (4.2.34)] imply that

T δxo (x) = k(x − xo ) for each xo ∈ Σ and each x ∈ Rn \ Σ. (2.4.26)
 n−1
Also, interpreting T as in (2.4.24), it follows that for each p ∈ n , ∞ and
q ∈ (0, ∞] there exists a constant C ∈ (0, ∞) such that
 Σc 
N (T f ) p, q ≤ C f  H p, q (Σ,σ)
κ L (Σ,σ)
(2.4.27)
for each f ∈ H p,q (Σ, σ).

As a consequence of (2.4.27) and [112, (8.6.50)], for each Lebesgue-measurable

set E ⊆ Rn \ Σ which is also assumed
 n−1 to satisfy L (E) < +∞ in the case when
n

(2.4.43) holds, and for each p ∈ n , ∞ and q ∈ (0, ∞] there exists a constant
C ∈ (0, ∞), which is actually independent of E in the case when H n−1 (Σ) = +∞,
with the property that the operator

T : H p,q (Σ, σ) −→ L np/(n−1),q (E, L n )

(2.4.28)
is well defined, linear, and bounded.

Finally, thanks to work in [112, §8.5], similar estimates to (2.4.23) and (2.4.27)
hold with the nontangential maximal operator replaced by the tangential max-
imal operator (associated as in [112, Definition 8.5.1] with Ω := Σ c and a
sufficiently large power M; cf. (A.0.85)).
2.4 Boundary-to-Domain Integral Operators on Open sets with UR Boundaries 387

(6) Given any integrability expoents p ∈ n−1 n , ∞ and q ∈ (1, ∞), there exists some
constant C = C(Σ, k, p, q, κ, n) ∈ [0, ∞) such that for every f ∈ H p (Σ, σ) the
following mixed-norm estimate holds:
∫  1/p
∫  p/q
c
|(∇T f )(z)| q dist(z, Σ)q−n dz dσ(x)
Σ ΓΣκ (x)

≤ C f  H p (Σ,σ), (2.4.29)

where the function T f is defined as in (2.4.4) if p ∈ (1, ∞), and as in (2.4.15)

or (2.4.16) if p ∈ n−1n ,1 .
In addition, corresponding to the case when p = 1, for every fixed exponent
q ∈ (1, ∞) there exists a constant C = C(Σ, k, q, κ, n) ∈ [0, ∞) with the
property that for each f ∈ L 1 (Σ, σ) the following weak estimate holds:
+  ∫  ,
sup λ · σ x ∈ Σ : c
|∇(T f )| dist(·, Σ)
q q−n
dL > λ
n q
λ>0 ΓΣκ (x)

≤ C f  L 1 (Σ,σ) . (2.4.30)

In the language of the L q -based area-function (0 < q < ∞), whose action on a
1,1 n
generic function u ∈ Wloc (R \ Σ) is defined as
∫  1/q
(A q,κ u)(x) := c
|(∇u)(y)| q |x − y| q−n dy , ∀x ∈ Σ, (2.4.31)
ΓΣκ (x)

(compare with [113, (9.2.206)]), the estimate in (2.4.29) becomes the statement
that for each integrability exponents p ∈ n−1 n , ∞ and q ∈ (1, ∞) there exists a
constant C ∈ [0, ∞) such that for every f ∈ H p (Σ, σ) one has
 
A q,κ (T f ) p ≤ C f  H p (Σ,σ), (2.4.32)
L (Σ,σ)

while the estimate in (2.4.30) translates into saying that for each q ∈ (1, ∞)
there exists some constant C ∈ [0, ∞) such that for each f ∈ L 1 (Σ, σ) one has
 
A q,κ (T f ) 1,∞ ≤ C f  L 1 (Σ,σ) . (2.4.33)
L (Σ,σ)

(7) For each p ∈ (1, ∞) there exists a constant C = C(Σ, p, n) ∈ [0, ∞) with the
property that for each function f ∈ L p (Σ, σ) one has the L p -square function
estimate
∫  1/p
|(∇T f )(x)| p dist(x, Σ) p−1 dx
R n \Σ
  
≤ C  k S n−1 𝒞2 (S n−1 )  f  L p (Σ,σ) . (2.4.34)
388 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

In particular, corresponding to p = 2, the following L 2 -square function estimate

holds: there exists a constant C = C(Σ, k, n) ∈ [0, ∞) with the property that for
each f ∈ L 2 (Σ, σ) one has
∫ ∫
|(∇T f )(x)| 2 dist(x, Σ) dx ≤ C | f (x)| 2 dσ(x). (2.4.35)
R n \Σ Σ

Also, for each p ∈ (1, ∞) there exists some C = C(Σ, k, p, n) ∈ [0, ∞) such that
for each function f ∈ L p (Σ, σ) there holds
  n−1   n−1

sup (T f )(x) dist(x, Σ) p + ∇(T f )(x) dist(x, Σ)1+ p
x ∈R n \Σ

≤ C f  L p (Σ,σ) . (2.4.36)

Finally, whenever 1 < q < ∞ and q < p < ∞ there exists a finite constant
C = C(Σ, k, p, q) > 0 such that the following type of square function estimate
  ∫  1/q 
  
 sup  1 ∇T f q dist(·, Σ)q−1 dL n 
 σ Σ∩B(z,r)  p
r >0 B(z,r)\Σ Lz (Σ,σ)

≤ C f  L p (Σ,σ) (2.4.37)

holds for every f ∈ L p (Σ, σ). In particular, corresponding to q = 2 and the

end-point p = ∞, for each function f ∈ L ∞ (Σ, σ) one has the following type of
Carleson measure estimate:
 ∫ 
 
sup  1 ∇T f 2 dist(·, Σ) dL n
z ∈Σ, r >0 σ Σ∩B(z,r) B(z,r)\Σ

≤ C f  L2 ∞ (Σ,σ) . (2.4.38)

(8) Define
∫
(∇T ) f (x) := (∇k)(x − y) f (y) dσ(y) for all x ∈ Rn \ Σ,
Σ (2.4.39)
 σ(y)
for each given function f ∈ L1 Σ, 1+ |y | n .

Then for each q ∈ (1, ∞) there exists some constant C = C(Σ, k, n, q) ∈ [0, ∞)
with the property that for each function f ∈ L ∞ (Σ, σ) one has the Carleson
measure type estimate
 ∫
  1/q
sup  1 (∇T ) f q dist(·, Σ) dL n
x ∈Σ, r >0 σ Σ∩B(x,r) B(x,r)\Σ

≤ C f  L ∞ (Σ,σ), (2.4.40)
2.4 Boundary-to-Domain Integral Operators on Open sets with UR Boundaries 389

as well as the uniform estimate

  
sup (∇T ) f (x) dist(x, Σ) ≤ C f  L ∞ (Σ,σ) . (2.4.41)
x ∈R n \Σ

(9) Assume the smoothness exponent N of the kernel k is sufficiently large, depending
only on the dimension n ∈ N of the ambient Euclidean space, and suppose μ is
a Borel measure on Rn \ Σ with the property that there exists a number α ≥ 1
such that 
μ B(x, r) \ Σ
Cμ := sup < +∞. (2.4.42)
r ∈(0,2diam(Σ)) r (n−1)α
and x ∈Σ

Also, let E ⊆ Rn \ Σ be a μ-measurable set which is also assumed to satisfy

μ(E) < +∞ in the case when H n−1 (Σ) < +∞. (2.4.43)

Then for each p ∈ (1, ∞) there exists C = C(Σ, n, p, α, Cμ, μ(E)) > 0 which is
actually independent of E in the case when H n−1 (Σ) = +∞, with the property
that every f ∈ L p (Σ, σ) one has
∫  α1p   
|T f | αp dμ ≤ C  k S n−1 𝒞 N (S n−1 )  f  L p (Σ,σ) . (2.4.44)
E

(10) Assume the smoothness exponent N of the kernel k is sufficiently large, depending
only on the dimension n ∈ N of the ambient Euclidean space. Then whenever

p ∈ (1, ∞), θ ∈ [0, ∞), α := n−1 ,

n+θ
0 < q ≤ αp, (2.4.45)

there exists some constant C = C(Σ, n, p, q, θ, k) ∈ (0, ∞) such that for each
point x ∈ Σ and each radius r ∈ 0, 2 diam(Σ) one has
∫
1
|T f | q dist(·, Σ)θ dL n ≤ C f  L p (Σ,σ),
q
 α−(q/p)
σ B(x, r) ∩ Σ
B(x,r)\Σ

for every function f belonging to the space L p (Σ, σ).

(2.4.46)

Corresponding to the special case when q := 1+θ n−1 p, there exists some constant
C = C(Σ, n, p, θ, k) ∈ (0, ∞) with the property that for every f ∈ L p (Σ, σ) the
expression
 1+θ
|T f | n−1 p dist(·, Σ)θ L n is a Carleson measure
 1+θin the
p
(2.4.47)
open R \ Σ, with constant bounded by C f  L pn−1
n
(Σ,σ)
.

In particular, if n ≥ 3, there exists C = C(Σ, n, k) ∈ (0, ∞) such that

390 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

given any f ∈ L n−1 (Σ, σ), it follows that |T f | 2 dist(·, Σ) L n is a

Carleson measure in Rn \ Σ with constant ≤ C f  L2 n−1 (Σ,σ) . (2.4.48)

(11) For each p ∈ (1, ∞) and each bounded Lipschitz domain Ω ⊆ Rn satisfying
Ω ∩ Σ =  there exists a constant C ∈ (0, ∞) with the property that

T f B p, p (Ω) ≤ C f  L p (Σ,σ) for every f ∈ L p (Σ, σ). (2.4.49)

1/p

Finally, suppose the kernel k is C M×M -valued and is a null-solution in Rn \ {0}

of a homogeneous, constant (complex) coefficient, M × M system L, which is
weakly elliptic (in the sense that its M × M symbol matrix is invertible; cf.
(1.3.2)). Also, suppose Ω ⊆ Rn is a bounded one-sided NTA domain satisfying
Ω ∩ Σ = , and fix

1 < p < ∞, n
n+1/p < q1 ≤ ∞, p ≤ q2 ≤ ∞. (2.4.50)

Then there exists a constant C = C(Σ, Ω, k, L, p, q1, q2 ) ∈ [0, ∞) with the property
that
 
max T f [F p, q1 (Ω)] M , T f [B p, q2 (Ω)] M ≤ C f [L p (Σ,σ)] M
1/p 1/p (2.4.51)
M
for each function f ∈ L p (Σ, σ) .

In particular, corresponding to p = q1 = q2 = 2, one has (recall (A.0.57))

M
T f [H 1/2 (Ω)] M ≤ C f [L 2 (Σ,σ)] M for each f ∈ L 2 (Σ, σ) . (2.4.52)

Before presenting the proof of this theorem we make a few comments.

Condition (2.4.3) ensures that the integral defining (T f )(x) in (2.4.4) is abso-
lutely convergent for each fixed point x ∈ Rn \ Σ. Regarding (2.4.9)-(2.4.10), it
is reassuring to recall [112, (7.7.106)-(7.7.107)]. We also want to point out that a
natural version of Theorem 2.4.1 holds when the kernel k is matrix-valued and the
functions/distributions this is paired with are vector-valued.
Two comments pertaining to item (9) of Theorem 2.4.1 are as follows. First
assume that μ is the Lebesgue measure on Rn \ Σ. In such a scenario, (2.4.42) holds
with α = n/(n − 1). Granted this, (2.4.44) then gives that for each p ∈ (1, ∞) and
each L n -measurable set E ⊆ Rn \ Σ which, in the case when Σ is bounded and
πΣ c ,κ (E) = Σ for each κ > 0, is also assumed to satisfy L n (E) < +∞, the operator

T : L p (Σ, σ) −→ L np/(n−1) (E, L n )

(2.4.53)
is well defined, linear, and bounded.

Of course, this also may be seen by specializing (2.4.28) to the case p = q ∈ (1, ∞)
(bearing in mind [112, (6.2.27)].
Second, if Σ is unbounded and μ is a Carleson measure in Rn \ Σ then for each
p ∈ (1, ∞) the operator
2.4 Boundary-to-Domain Integral Operators on Open sets with UR Boundaries 391

T : L p (Σ, σ) −→ L p (Rn \ Σ, μ)
(2.4.54)
is well defined, linear, and bounded.

We next turn to the proof of Theorem 2.4.1.

Proof of Theorem 2.4.1 A few preliminary observations are as follows. First, since
Σ is closed and has an empty interior (due to the upper Ahlfors regularity condition),
it follows that Σ c = Rn \ Σ is an open set whose topological boundary is precisely
Σ. Second, hypotheses (2.3.3) imposed on the kernel k imply
 
|k(x)| ≤ sup |k | |x| 1−n for each x ∈ Rn \ {0}, (2.4.55)
S n−1

and  
|(∇k)(x)| ≤ sup |∇k | |x| −n for each x ∈ Rn \ {0}. (2.4.56)
S n−1

Third, for each function f as in (2.4.3), the integral in (2.4.4) is meaningfully defined
and 
T f ∈ 𝒞 N Rn \ Σ . (2.4.57)

To deal with item (1), pick some function f ∈ L 1 Σ, 1+ |σ· | n−1 . Also, assume that
c
z ∈ Σ and x ∈ Σ c are two fixed points such that x ∈ ΓΣκ (z), i.e.,

|x − z| < (1 + κ) dist(x, Σ). (2.4.58)

Recall the truncated singular integral operator from (2.3.6). We presently consider
the threshold
ε := |x − z| > 0 (2.4.59)
and estimate the difference
 
(T f )(x)−(T2ε f )(z)
∫ ∫ 
 
 
=  k(x − y) f (y) dσ(y) − k(z − y) f (y) dσ(y)
 
Σ y ∈Σ
|z−y |>2ε
 ∫ 
 
 
≤ k(x − y) f (y) dσ(y)
 
y ∈Σ
|z−y |<2ε
 ∫ 
   
 
+ k(x − y) − k(z − y) f (y) dσ(y)
 
y ∈Σ
|z−y |>2ε

=: I + I I. (2.4.60)
392 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Since
|x − z| ε
|x − y| ≥ dist(x, Σ) > = , (2.4.61)
1+κ 1+κ
making use of (2.4.55), (2.4.61), and the Ahlfors regularity of Σ, we obtain
  C(κ, n) ∫
|I | ≤ sup |k | | f (y)| dσ(y)
S n−1 ε n−1
y ∈Σ
|z−y |<2ε
 
≤ C(Σ, κ, n) sup |k | M Σ f (z), (2.4.62)
S n−1

where M Σ denotes the Hardy-Littlewood maximal operator on Σ. On the other hand,

if y ∈ Σ is such that |y − z| > 2ε, then for each t ∈ [0, 1] we have

ε + 12 |z − y| ≤ 12 |z − y| + 12 |z − y| = |z − y|

≤ |z − (t x + (1 − t)z)| + |t x + (1 − t)z − y|

= t|x − z| + |t x + (1 − t)z − y|

≤ ε + |t x + (1 − t)z − y|, (2.4.63)

hence

|t x + (1 − t)z − y| ≥ 12 |z − y| for all t ∈ [0, 1]. (2.4.64)

Granted this, the Mean Value Theorem applies and, on account of (2.4.59) and
(2.4.56), gives
  
|k(x − y) − k(z − y)| ≤ |x − z| sup (∇k) t x + (1 − t)z − y 
0≤t ≤1
  ε
≤ 2n sup |∇k | . (2.4.65)
S n−1 |z − y| n

Consequently, a familiar argument gives

2.4 Boundary-to-Domain Integral Operators on Open sets with UR Boundaries 393
  ∫
ε
|I I | ≤ 2n sup |∇k | | f (y)| dσ(y)
S n−1 |z − y| n
y ∈Σ
|z−y |>2ε

 
∞ ∫
ε
=2 n
sup |∇k | | f (y)| dσ(y)
S n−1 j=1
|z − y| n
y ∈Σ
2 j ε< |z−y | ≤2 j+1 ε

 
∞ ∫
−j
≤2 n
sup |∇k | 2 (2 j+1
ε) 1−n
| f (y)| dσ(y)
S n−1 j=1 y ∈Σ
|z−y | ≤2 j+1 ε

 
∞

≤ C(Σ, n) sup |∇k | 2−j M Σ f (z)
S n−1 j=1
 
= C(Σ, n) sup |∇k | M Σ f (z), (2.4.66)
S n−1

where the next-to-last step utilizes the upper Ahlfors regularity of Σ.

In summary, the above analysis (consisting of (2.4.60), (2.4.62), and (2.4.66))
proves that
     
(T f )(x) − (T2ε f )(z) ≤ C(Σ, κ, n) k  n−1  1 n−1 M Σ f (z)
S 𝒞 (S )
(2.4.67)
for any two points x ∈ Σ c and z ∈ Σ as in (2.4.58)-(2.4.59).

In view of (2.3.5), this further entails

       
(T f )(x) ≤ (Tmax f )(z) + C(Σ, κ, n) k  n−1  1 n−1 M Σ f (z)
S 𝒞 (S )
c
(2.4.68)
whenever z ∈ Σ and x ∈ ΓΣκ (z).
c
For each fixed point z ∈ Σ, with the property that ΓΣκ (z)  , take the supremum
c
in x ∈ ΓΣκ (z) in (2.4.68). Bearing (2.4.57) in mind, we arrive at the conclusion that
there exists some constant C = C(Σ, κ, n) ∈ (0, ∞) such that
   
NκΣ (T f )(z) ≤ (Tmax f )(z) + C  k S n−1 𝒞1 (S n−1 ) M Σ f (z).
c
(2.4.69)
c
Since the above estimate is trivially valid when ΓΣκ (z) = , it follows that (2.4.5)
holds for all points in Σ.
Next, we shall deal with items (2)-(3) simultaneously (in an intertwined fashion).
Make the assumption that the smoothness exponent N of the kernel k is large enough.
Then (2.4.9)-(2.4.10) in item (3) become consequences of (2.4.5), (2.3.11)-(2.3.12),
and [112, Corollary 7.6.3].
Returning now to item (2), consider first the claim made in (2.4.7). To set the
stage, fix a reference point xo ∈ Σ and choose some radius r > 0. For an arbitrary
394 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

function f ∈ L 1 Σ, 1+ |σ· | n−1 , define

c
  
gr := NκΣ T f 1Σ\B(xo,4r) and
c
   (2.4.70)
hr := NκΣ T f 1Σ∩B(xo,4r) on Σ ∩ B(xo, r).

Fix an arbitrary point x ∈ Σ ∩ B(xo, r). Note that if Γκ (x)   then [112, (8.1.12)]
permits us to estimate

|y − xo | ≈ |y − x| ≤ (2 + κ) dist y, Γκ (x) ≤ (2 + κ)|y − z|,
(2.4.71)
uniformly for y ∈ Σ \ B(xo, 4r) and z ∈ Γκ (x).

Consequently, there exists some C = C(k, r, xo ) ∈ (0, ∞) with the property that if
x ∈ Σ ∩ B(xo, r) and z ∈ Γκ (x) then
∫
  
T f 1Σ\B(xo,4r) (z) ≤ |k(z − y)|  f 1Σ\B(xo,4r) (y) dσ(y)
Σ
∫
| f (y)|
≤C dσ(y)
Σ\B(x o ,4r) |z − y| n−1
∫
| f (y)|
≤C dσ(y)
Σ\B(x o ,4r) |xo − y| n−1
∫
| f (y)|
≤C dσ(y) < +∞. (2.4.72)
Σ 1 + |y| n−1

This proves that there exists C = C(k, r, xo ) ∈ (0, ∞) such that


gr ∈ L ∞ Σ ∩ B(xo, r), σ and
(2.4.73)
gr  L ∞ (Σ∩B(xo,r),σ) ≤ C f  L 1 Σ, σ(x) .
1+| x | n−1

In turn, from this and (2.3.97) we deduce that


gr ∈ L 1,∞ Σ ∩ B(xo, r), σ and
gr  L 1,∞ (Σ∩B(xo,r),σ) ≤ C f  1  . (2.4.74)
σ(x)
L Σ,
1+ |x | n−1

Also, since there exists C = C(r, xo ) ∈ (0, ∞) such that

f 1Σ∩B(xo,4r) ∈ L 1 (Σ, σ) with

  (2.4.75)
 f 1Σ∩B(x ,4r)  1
o L (Σ,σ)
≤ C f  L 1 Σ, σ(x) ,
1+| x | n−1

we conclude from (2.4.10) (which has already been established at this point) and
the second line in [112, (6.2.17)] that there exists C = C(k, r, xo ) ∈ (0, ∞) with the
2.4 Boundary-to-Domain Integral Operators on Open sets with UR Boundaries 395

property that 
hr ∈ L 1,∞ Σ ∩ B(xo, r), σ and
(2.4.76)
hr  L 1,∞ (Σ∩B(xo,r),σ) ≤ C f  L 1 Σ, σ(x) .
1+| x | n−1

Altogether, from (2.4.74), (2.4.76), [112, (8.2.28)], and the fact that, at each point
c
on Σ ∩ B(xo, r), we have 0 ≤ NκΣ T f ≤ gr + hr , we ultimately conclude that the
assignment in (2.4.7) is indeed well defined, sub-linear, bounded, and continuous.
Next, the claims about the mapping (2.4.8) are dealt with in a similar fashion,

taking into account that if now f ∈ Lloc (Σ, σ) ∩ L 1 Σ, 1+σ(x)
p
|x | n−1
then, thanks to
(2.4.9), in place of (2.4.76) we now have

hr ∈ L p Σ ∩ B(xo, r), σ and
(2.4.77)
hr  L p (Σ∩B(xo,r),σ) ≤ C f  L p (Σ∩B(xo,4r),σ),

and that (2.4.73) implies


gr ∈ L p Σ ∩ B(xo, r), σ and
gr  L p (Σ∩B(xo,r),σ) ≤ C f  1  . (2.4.78)
σ(x)
L Σ,
1+ |x | n−1

This finishes the justification of the claims in item (2).

Going back to item (3), consider the Hardy space estimate in (2.4.14). To justify
this, fix some q ∈ (1, ∞) with the Σc
 goal of estimating Nκ (T a) when the function
a is an arbitrary (p, q)-atom on Σ, | · − · |, σ , viewed as a space of homogeneous
type (cf. [112, Example 7.4.1]). Recall that this means that there exist xo ∈ Σ along
with some finite r ∈ 0, diam Σ with the property that
∫
supp a ⊆ B(xo, r) ∩ Σ, a dσ = 0,
 Σ (2.4.79)
1/q−1/p
and a L q (Σ,σ) ≤ σ B(xo, r) ∩ Σ .

When σ(Σ) < +∞, the constant function a ≡ σ(Σ)−1/p is also considered to be a
(p, q)-atom.
First, estimate (2.4.9) used with p := q and f := a as in (2.4.79) gives that, on
the one hand,
∫
 Σc 
N (T a) p dσ
κ
B(x o ,2(2+κ)r)∩Σ
 c p 
≤ NκΣ (T a) L q (Σ,σ) · σ B(xo, 2(2 + κ)r) ∩ Σ
1−p/q

  p 
≤ C  k S n−1 𝒞 N (S n−1 ) a L q (Σ,σ) · σ B(xo, 2(2 + κ)r) ∩ Σ
p 1−p/q

  p
≤ C(Σ, κ, p, q) k S n−1 𝒞 N (S n−1 ), (2.4.80)
396 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

where we have also used Hölder’s inequality, (2.4.79), and the Ahlfors regularity of
Σ. On the other hand, whenever z ∈ Σ c is such that |z − xo | ≥ 2r, and whenever a
is as in (2.4.79), based on the vanishing moment condition for the atom, the Mean
Value Theorem, and Hölder’s inequality we may estimate
∫
|(T a)(z)| ≤ |k(z − y) − k(z − xo )||a(y)| dσ(y)
B(x o ,r)∩Σ
∫
   r
≤ C  k S n−1 𝒞1 (S n−1 ) |a(y)| dσ(y)
|z − xo | n B(x o ,r)∩Σ
   r 
≤ C  k S n−1 𝒞1 (S n−1 )
1−1/p
σ B(xo, r) ∩ Σ . (2.4.81)
|z − xo | n

Now, if x ∈ Σ and z ∈ Σ c are such that

|x − xo | ≥ 2(2 + κ)r and |z − x| < (1 + κ) dist(z, Σ) (2.4.82)

then

2(2 + κ)r ≤ |x − xo | ≤ |x − z| + |z − xo | ≤ (2 + κ)|z − xo |. (2.4.83)

In particular,
1
|z − xo | ≥ 2r and |z − xo | ≥
|x − xo |. (2.4.84)
2+κ
In light of (2.4.84), after taking the supremum in (2.4.81) over z ∈ Σ c satisfying
|z − x| < (1 + κ) dist(z, Σ), we obtain
 Σc     r 
N (T a)(x) ≤ C  k  n−1  1 n−1 σ B(xo, r) ∩ Σ
1−1/p
,
κ 𝒞 (S )
S |x − xo | n
(2.4.85)

for every point x ∈ Σ \ B xo, 2(2 + κ)r .

Keeping in mind that p n > n − 1, on account of (2.4.85) and [112, Lemma 7.2.1]
we may then estimate
∫
 Σc    
N (T a)(x) p dσ(x) ≤ C(Σ, κ, p) k  n−1  p 1 n−1 , (2.4.86)
κ S 𝒞 (S )
Σ\B(x o ,2(2+κ)r)

for some finite constant. From (2.4.80) and (2.4.86) it follows that
 Σc    
N (T a) p ≤ C  k S n−1 𝒞 N (S n−1 ), (2.4.87)
κ L (Σ,σ)

for some finite constant C = C(Σ, κ, n, p, q) > 0, independent of the atom a as in

(2.4.79). When σ(Σ) < +∞, it follows from (2.4.9) (with p = 2) and Hölder’s
inequality that the same estimate holds for the constant function a ≡ σ(Σ)−1/p .
Having established this, we may now invoke the general boundedness criterion from
[113, Theorem 4.4.7] (see also [7, Proposition 8.32, pp. 421-422] for a general result
of this flavor), used here with
2.4 Boundary-to-Domain Integral Operators on Open sets with UR Boundaries 397
p
q := 2, X := Nκ2 (Σ c ; σ), and Y := Nκ (Σ c ; σ), (2.4.88)

(bearing in mind [112, Proposition 8.3.5]), in order to conclude that (2.4.14) holds.
Next, formula (2.4.15) is implied by [113, (4.4.144)] (keeping in mind [112,
(8.3.33)]). Formula (2.4.16) follows from (2.4.15) and the duality result from [113,
Theorem 4.6.1], upon noting that since for each fixed point x ∈ Rn \ Σ the function
Σ  y → k(x − y) ∈ C is both bounded and Lipschitz (cf. (2.4.55)-(2.4.56)), we
infer from [112, (7.3.25)] that

k(x − ·)Σ ∈ 𝒞α (Σ), ∀x ∈ Rn \ Σ. (2.4.89)
0<α<1

That formula (2.4.15) remains valid if the a j ’s happen to be H p -molecules on Σ is

seen from (2.4.16), (2.4.89), and [113, (4.8.68), (4.8.70)]. Also, that the operator T
in (2.4.16)
  n−1 T in (2.4.4) considered on functions from
acts in a coherent fashion with
L 1 Σ, 1+σ(y)
|y | n−1
is implied when p ∈ n , 1 by the compatibility result established
in [113, Proposition 4.8.8] (bearing in mind [113, (4.8.57)] and (2.4.55)-(2.4.56)),
and when p = 1 by the compatibility result proved in [113, Proposition 4.8.6].
To deal with the claim made in (2.4.17), assume f ∈ H p (Σ, σ) is a compactly
supported distribution, and pick some function ψ ∈ Lipc (Σ) which is identically one
near supp f . Also, choose a function θ ∈ 𝒞∞ c (R ) satisfying θ ≡ 1 near the origin in
n

Rn , and for each R > 0 define θ R (x) := θ(x/R) for every x ∈ Rn . Finally, fix a point
x ∈ Rn \ Σ. We may then rely on [112, (7.3.17)] (with α := 1) to conclude that
 
lim θ R k(x − ·)Σ = k(x − ·)Σ in 𝒞α (Σ) for each α ∈ (0, 1). (2.4.90)
R→∞

In concert with (2.4.16) and [113, Lemma 4.6.4], this permits us to write
#  $
(T f )(x) = lim Lipc (Σ) θ R k(x − ·)Σ, f (Lipc (Σ))
R→∞
#  $
= lim Lip c (Σ) θ R ψk(x − ·)Σ, f (Lipc (Σ))
R→∞
#  $
= Lipc (Σ) ψk(x − ·)Σ, f (Lipc (Σ)) (2.4.91)

which proves (2.4.17). At this point, the justification of all claims made in item (3)
of the statement of the theorem is complete.
As regards item (4), that T f in (2.4.4) is well defined to being with, for each
f ∈ L p (Σ, wσ), is seen from [112, (7.7.104)], while the claim in (2.4.18) is imply
by (2.3.58)-(2.3.57), (2.4.5), and item (1) in [112, Lemma 7.7.1]. Then (2.4.20)
becomes a consequence of (2.4.18) and [113, Corollary 5.2.3]. Moreover, the claim
made in (2.4.22) is seen from (2.4.5), (2.3.68), and [113, (5.2.86), (5.2.102)].
Let us now turn to item (5). That, to being with, T f in (2.4.4) is well defined
for each given function f ∈ L p,q (Σ, σ) is a consequence of [112, (7.7.107)]. The
estimate claimed in (2.4.23) follows from [113, Proposition 1.3.7], [112, (6.2.48)],
and (2.4.9). Since for each fixed point x ∈ Σ c = Rn \ Σ we have (cf. [112, (7.2.5),
398 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

(7.3.17)])
    
k(x − ·)Σ ∈ Lip (Σ) ∩ L p (Σ, σ) ∩ 𝒞α (Σ) , (2.4.92)
1<p ≤∞ 0<α<1

From [113, Proposition 4.8.9] we see that the operator T , initially defined as in
(2.4.3)-(2.4.4) may be further extended, in a coherent  fashion, to a well-defined
mapping acting on each f ∈ H p,q (Σ, σ) with p ∈ n−1 n , ∞ and q ∈ (0, ∞] as
in (2.4.24). To justify (2.4.25), consider a compactly supported distribution f in
H p,q (Σ, σ) along with ψ ∈ Lipc (Σ) satisfying ψ ≡ 1 near supp f . As before, pick
some θ ∈ 𝒞∞ c (R ) with θ ≡ 1 near the origin in R , and for each R > 0 set
n n

θ R (x) := θ(x/R) for every x ∈ Rn . Let us also choose p0, p1 ∈ n−1 n , ∞ such that
(0) (1) (i)
p0 < p < p1 and decompose f = f + f where f ∈ H (Σ, σ) for i ∈ {0, 1}.
p i

Fix an arbitrary point x ∈ Rn \ Σ. Combining (2.4.90) with [113, Lemma 4.6.4] and
results from [115, §2.2] allows us to write
 (i) #  $
T f (x) = lim Lipc (Σ) θ R k(x − ·)Σ, f (i) (Lipc (Σ)) for i ∈ {0, 1}. (2.4.93)
R→∞

Summing up the two formulas in (2.4.93) then yields

#  $
(T f )(x) = lim Lipc (Σ) θ R k(x − ·)Σ, f (Lipc (Σ))
R→∞
#  $
= lim Lip c (Σ) θ R ψk(x − ·)Σ, f (Lipc (Σ))
R→∞
#  $
= Lipc (Σ) ψk(x − ·)Σ, f (Lipc (Σ)) (2.4.94)

which proves (2.4.25). Finally, the estimate in (2.4.27) is a consequence of (2.4.9),

(2.4.14), [113, (4.3.3)], and the real interpolation result from [113, Proposition 1.3.7].
Consider next the claim made in item (6). According to [61, Theorem 1.3(1),
pp. 9-10] the L 2 -Square Function Estimate stated in [61, Theorem 1.1(1), p. 6] holds
for the choices
𝒳 := Rn, E := Σ, m := n, d := n − 1, υ := 1, α := 1,
(2.4.95)
and θ(x, y) := (∇k)(x − y) (which entails Θ = ∇T ).

Granted this, [61, Theorem 1.1(13), pp. 6-8] then applies and readily gives the
estimate recorded in (2.4.29) (after unraveling notation, and keeping (2.4.95) in
mind). In turn, once (2.4.29) has been established, the weak estimate in (2.4.30)
follows by invoking [61, Theorem 6.17, p. 94].
To deal with the first claim in item (7), fix some f ∈ L p (Σ, σ) with p ∈ (1, ∞).
Then, on the one hand, (2.4.29) used with q := p reads
∫  p1
∫ 
c
|∇T f | p dist(·, Σ) p−n dL n dσ(x) ≤ C f  L p (Σ,σ) . (2.4.96)
Σ ΓΣκ (x)
2.4 Boundary-to-Domain Integral Operators on Open sets with UR Boundaries 399

On the other hand, using Fubini’s Theorem as in [112, (8.1.43)], we may recast the
left-hand side above as
∫ ∫  1/p
 
c
|(∇T f )(z)| dist(z, Σ)
p p−n
dz dσ(x)
Σ ΓΣκ (x)
∫  1/p

= |(∇T f )(z)| dist(z, Σ)
p p−n
σ πκ ({z}) dz (2.4.97)
R n \Σ

where for each point z ∈ Rn \ Σ, the “(reverse) conical projection” πκ ({z}) of z onto
Σ is defined as  
c
πκ ({z}) := x ∈ Σ : z ∈ ΓΣκ (x) . (2.4.98)
In view of the Ahlfors regularity of Σ, from [112, (8.1.19)] it follows that

σ πκ ({z}) ≈ dist(z, Σ)n−1, uniformly in z ∈ Rn \ Σ. (2.4.99)

Combining (2.4.96), (2.4.97), and (2.4.99) then yields (2.4.34). In turn, (2.4.35)
follows by specializing (2.4.34) to the case when p = 2. Next, the estimate claimed in
(2.4.36) is a consequence of [112, Proposition 8.7.13] used twice (with Ω := Rn \ Σ).
Finally, consider (2.4.37). To this end, pick q ∈ (1, ∞) and p ∈ (q, ∞), and select
an arbitrary function f ∈ L p (Σ, σ). Also, let z ∈ Σ and r > 0 be arbitrary. The goal
is to establish the pointwise estimate
∫
  
r 1−n ∇T f (x)q dist(x, Σ)q−1 dx ≤ C M Σ | f | q (z), (2.4.100)
B(z,r)\Σ

where M Σ stands for the Hardy-Littlewood maximal operator on Σ, i.e.,

⨏

M Σ g (x) := sup |g| dσ, ∀x ∈ Σ. (2.4.101)
ρ>0 Δ(x,ρ)

Here and elsewhere, Δ(x, ρ) := B(x, ρ) ∩ Σ is the surface ball centered at x ∈ Σ and
having radius ρ > 0. Once (2.4.100) has been justified, (2.4.37) is readily obtained
from this and the fact that M Σ is bounded on L p/q (Σ, σ) since p/q ∈ (1, ∞) (cf.
[112, (7.6.18)]).
As regards (2.4.100), decompose

f = 1Δ2r f + 1Σ\Δ2r f (2.4.102)

where, for each R > 0, we let ΔR abbreviate Δ(z, R). Making use of the square
function estimate (2.4.34) (written for q in place of p) and the doubling property of
the surface measure σ, we may then estimate
400 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫  q
  
r 1−n ∇T 1Δ2r f (x) dist(x, Σ)q−1 dx
B(z,r)\Σ
∫  q
  
≤ r 1−n ∇T 1Δ2r f (x) dist(x, Σ)q−1 dx
R n \Σ
∫ ⨏
≤ Cr 1−n |1Δ2r f | q dσ ≤ C | f | q dσ
Σ Δ(z,2r)

≤ C M Σ | f | q (z), (2.4.103)

which suits our purposes. Next, observe that if x ∈ B(z, r) \ Σ is fixed then

|x − y| ≈ |z − y| uniformly for y ∈ Σ \ Δ2r . (2.4.104)

Also, by homogeneity,
C
|(∇k)(x − y)| ≤ . (2.4.105)
|x − y| n
In turn, (2.4.104), (2.4.105), the Ahlfors regularity of ∂Ω, Hölder’s inequality, and
(2.4.101) can be used to justify the following sequence of inequalities:
  ∫
   1
∇T 1Σ\Δ2r f (x) ≤ C | f (y)| dσ(y)
Σ\Δ2r |z − y|
n

 ∞ ∫
1
≤C | f (y)| dσ(y)
j=1
(2 j r)n Δ2 j+1 r \Δ2 j r
⨏ ⨏
  1   1/q
∞ ∞
1
≤C | f | dσ ≤ C | f | q
dσ
j=1
2 j r Δ2 j+1 r j=1
2jr Δ2 j+1 r

  1/q
≤ Cr −1 M Σ | f | q (z) . (2.4.106)

Hence, after raising the most extreme sides to the power q, we obtain the pointwise
inequality    
∇T 1Σ\Δ f (x)q ≤ Cr −q M Σ | f | q (z) (2.4.107)
2r

for each x ∈ B(z, r) \ Σ. In turn, this implies (keeping in mind that dist(x, Σ) ≤ r on
the domain of integration) that
∫  q
   
r 1−n
∇T 1Σ\Δ2r f (x) dist(x, Σ)q−1 dx ≤ C M Σ | f | q (z), (2.4.108)
B(z,r)\Σ

which is of the right order. Now, (2.4.37) becomes an immediate consequence of

(2.4.103) and (2.4.108).
Moving on, at least when q = 2, the Carleson measure estimate (2.4.40) in item
(8) is contained in [61, Theorem 1.3(3), pp. 9-10]. We may directly justify (2.4.40)
2.4 Boundary-to-Domain Integral Operators on Open sets with UR Boundaries 401

for any q ∈ (1, ∞) by re-running the proof of (2.4.37) with p := ∞. The uniform
estimate (2.4.41) is implied by (the very last claim in) [112, Proposition 8.7.13] used
with p := ∞, α := 1, and Ω := Rn \ Σ.
Turning to the claims in part (9), assume the smoothness exponent N of the kernel
k is large enough, and pick a set E as specified in the statement of the theorem. Also,
select an integrability exponent p ∈ (1, ∞) and fix some arbitrary aperture parameter
κ > 0. In the current setting, condition (2.4.42) amounts to saying that (μ, σ) is
an (α, 0)-Carleson pair for the ambient (Σ c, Σ) (cf. [112, Definition 8.6.1])). Having
observed this, we may invoke [112, Theorem 8.6.2, (8.6.13)] and (2.4.9) in order to
estimate
∫  α1p  c 
|T f | αp dμ ≤ NκΣ (T f ) L p (Σ,σ)
E
  
≤ C  k S n−1 𝒞 N (S n−1 )  f  L p (Σ,σ), (2.4.109)

for each f ∈ L p (Σ, σ), where the constant C = C(Σ, n, p, α, Cμ, κ) is a finite positive
number. This establishes (2.4.44).
Consider
 next the claims made in item (10). First we observe that whenever
r ∈ 0, diam(Σ) the estimate in (2.4.46) holds under the assumptions made on
p, θ, α, q in (2.4.45). Indeed, this is a consequence of (2.4.44) (applied for the set
E := B(x, r) \ Σ and the measure μ := dist(·, Σ)θ L n in Rn \ Σ which satisfies
(2.4.42) with α := n+θ n−1 ≥ 1, and which also satisfies μ(E) < +∞ in the case
when Σ is compact, thus rendering (2.4.43) valid), followed by an application of
Hölder’s inequality (used to increase the integrability exponent from q to αp).
If diam(Σ) = +∞ this proves (2.4.46) as stated. In turn, (2.4.47) is obtained by
specializing (2.4.46), and (2.4.48) is a particular case of (2.4.47).
As far as item (11) is concerned, the first claim is a direct consequence of [113,
Lemma 9.2.40], the L p -square function estimate (2.4.34), and the nontangential
maximal function estimate (2.4.9) together with [112, (8.6.51) in Proposition 8.6.3].
To treat the second claim in item (11), we shall work under the additional as-
sumptions on the kernel specified there. In particular, the fact that k is a null-solution
M
of a weakly elliptic M × M system L implies that k ∈ 𝒞∞ (Rn \ {0}) (cf. [112,
M
Theorem 6.5.7]). Next, fix an arbitrary function f ∈ L p (Σ, σ) with p ∈ (1, ∞)
and consider

u := T f Ω ∈ 𝒞∞ (Ω) .
M
(2.4.110)

Since Lu = 0 in Ω, [112, Theorem 6.5.7] implies that u is subaveraging in Ω. To

proceed, observe that

dist(x, ∂Ω) ≤ dist(x, Σ) for each x ∈ Ω. (2.4.111)

Indeed, given x ∈ Ω arbitrary, if r := dist(x, ∂Ω) then a simple argument (using

the fact that Ω is open and B(x, r) is connected) implies that B(x, r) ⊆ Ω. Hence,
B(x, r) ⊆ Rn \ Σ which goes to show that dist(x, Σ) ≥ r = dist(x, ∂Ω). Having
402 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

justified (2.4.111), we may then estimate

∫  1/p  ∫  1/p
|∇u| p dist(·, ∂Ω) p−1 dL n ≤ |∇(T f )| p dist(·, Σ) p−1 dL n
Ω R n \Σ

≤ C f [L p (Σ,σ)] M , (2.4.112)

where the last step comes from (2.4.35). Also, [112, (8.6.51) in Proposition 8.6.3]
presently implies  Ω 
u p n M ≤ C Nκ u p (2.4.113)
n
L n−1 (Ω, L ) L (Σ,σ)

for some constant C ∈ [0, ∞) independent of u. In concert with Hölder’s inequality,

(2.4.110), and (2.4.9), this implies
 Ω 
u[L p (Ω, L n )] M ≤ Cu p n M ≤ C Nκ u p
n
L n−1 (Ω, L ) L (Σ,σ)

 c 
≤ C NκΣ (T f ) L p (Σ,σ) ≤ C f [L p (Σ,σ)] M . (2.4.114)

Hence, if n/(n + 1/p) < q1 ≤ ∞, p ≤ q2 ≤ ∞, and k := 0, α := 1/p, we may invoke

[113, Theorem 9.2.30] (with q ∈ {q1, q2 }) which, in concert with (2.4.112) and
(2.4.114), yields (2.4.51). Finally, (2.4.52) is a consequence of (2.4.51) and [113,
(9.2.22)]. The proof of Theorem 2.4.1 is therefore complete. 

In the next corollary we establish a Carleson measure estimate which brings into
play the role of the Divergence Theorems from [112, §1.2].

Corollary 2.4.2 Assume Ω ⊆ Rn (where n ∈ N with n ≥ 2) is an open set with the

property that ∂Ω is a UR set; in particular, Ω is a set of locally finite perimeter.
Denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal
to Ω and abbreviate σ := H n−1 ∂Ω. Assume k ∈ 𝒞2 (Rn \ {0}) is an odd function
which is positive homogeneous of degree 1 − n and, for each i, j ∈ {1, . . . , n},
consider the integral operator acting on functions f ∈ BMO(∂Ω, σ) according to
∫  
Qi j f (x) := νi (y)∂y j k(x − y) − ν j (y)∂yi k(x − y) f (y) dσ(y), (2.4.115)
∂∗ Ω

for each x ∈ Ω.
Then for each f ∈ BMO(∂Ω, σ) the function Qi j f is well defined in Ω and, given
any exponent p ∈ (1, ∞), there exists some C = C(Ω, k, p) ∈ (0, ∞), independent of
f , with the property that
  
Qi j f  p dist(·, ∂Ω) p−1 dL n (2.4.116)
1≤i, j ≤n

is a Carleson measure in Ω, in the quantitative sense that

2.4 Boundary-to-Domain Integral Operators on Open sets with UR Boundaries 403
∫ 
1
n
 
sup  Qi j f  p dist(·, ∂Ω) p−1 dL n
x ∈∂Ω, r >0 σ B(x, r) ∩ ∂Ω B(x,r)∩Ω i, j=1

≤ C f  .
p
, (2.4.117)
BMO(∂Ω,σ)

with the piece of notation introduced in (A.0.14). In particular, corresponding to

p = 2 it follows that, in a quantitative fashion,
  
Qi j f 2 dist(·, ∂Ω) dL n is a Carleson measure in Ω,
1≤i, j ≤n (2.4.118)
for each function f ∈ BMO(∂Ω, σ).

Proof Pick i, j ∈ {1, . . . , n} and fix an arbitrary function f ∈ BMO(∂Ω, σ). Then

for each fixed x ∈ Ω we may estimate
∫  
 
νi (y)∂y j k(x − y) − ν j (y)∂yi k(x − y) | f (y)| dσ(y)
∂∗ Ω
∫
  | f (y)|
≤ 2 sup ∇k  · dσ(y) < +∞, (2.4.119)
S n−1 ∂Ω |x − y| n

thanks to the homogeneity of the kernel k, the fact that ∂Ω is an Ahlfors regular set,
and [112, (7.4.116)]. This goes to show that the function Qi j f is well defined in Ω,
via an absolutely convergent integral.
Next, we claim that
Qi j 1 ≡ 0 in Ω. (2.4.120)
To prove this, fix an arbitrary point x ∈ Ω and note that k(x − ·) ∈ Lloc
1 (Ω, L n ). As

such, it is meaningful to consider the vector field

F! := ∂j k(x − ·) ei − ∂i k(x − ·) e j (2.4.121)

where the partial derivatives are taken in the sense of distributions in Ω (in the “dot”
variable). This places F! in D (Ω) . Since its singular support is the singleton {x},
n

we see that actually

F! ∈ Lloc (Ω, L n ) + ℰ(Ω) .
1 n
(2.4.122)
Clearly,

divF! = ∂i ∂j k(x − ·) − ∂j ∂i k(x − ·) = 0 in D (Ω). (2.4.123)

Moreover, with d := dist(x, ∂Ω) > 0 and κ > 0 arbitrary, [112, Lemma 8.3.7]
ensures that there exists a constant C ∈ (0, ∞) such that
 
NκΩ\B(x,d/2) F! (y) ≤ C|y − x| −n for each y ∈ ∂Ω. (2.4.124)

From this, [112, (8.2.26)], and [112, Lemma 7.2.1], we may therefore conclude that
404 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

NκΩ\B(x,d/2) F! ∈ L 1 (∂Ω, σ). (2.4.125)

Furthermore, since F! is continuous in a neighborhood of ∂Ω it follows that the

κ−n.t.
pointwise nontangential boundary trace F!∂Ω exists at every point in ∂nta Ω and, in
fact,
 κ−n.t. 
F!∂Ω (y) = ∂y j k(x − y) ei − ∂yi k(x − y) e j at every y ∈ ∂nta Ω. (2.4.126)

In concert with [112, Proposition 8.8.6] this also shows that

 κ−n.t. 
F!∂Ω (y) = ∂y j k(x − y) ei − ∂yi k(x − y) e j for σ-a.e. y ∈ ∂∗ Ω. (2.4.127)

Our final observation is that in the case when the set Ω is unbounded we have
!
| F(y)| = O(|y| −n ) for y ∈ Ω with |y| → ∞. Thus, in such a scenario,
∫
!
|y · F(y)| dL n (y) = O(R) as R → ∞. (2.4.128)
[B(0,2R)\B(0,R)]∩Ω

Granted (2.4.122), (2.4.123), (2.4.125), (2.4.127), and (2.4.128), [112, Theo-

rem 1.4.1] applies and formula [112, (1.4.6)] presently yields
∫  κ−n.t. 

ν(y) · F! (y) dσ(y) = 0. (2.4.129)
∂∗ Ω ∂Ω

In view of (2.4.127), the identity claimed in (2.4.120) now readily follows from
(2.4.129). For further reference we wish to note that, as an inspection of the above
argument shows,
the cancelation property recorded in (2.4.120) holds even when
we merely assume that ∂Ω is an Ahlfors regular set and only
(2.4.130)
requiring that the kernel k ∈ 𝒞2 (Rn \ {0}) is positive homoge-
neous of degree 1 − N for some N ≥ n.

Going forward, select an integrability exponent p ∈ (1, ∞). Also, fix a point

xo ∈ ∂Ω and for each R > 0 abbreviate
⨏
SR := B(xo, R) ∩ ∂Ω, T(SR ) := B(xo, R) ∩ Ω, and fS R := f dσ. (2.4.131)
SR

In addition, pick a scale r ∈ 0, 2 diam(∂Ω) and consider a cutoff function η in Rn
satisfying

η ∈ 𝒞∞
c (R ),
n 0 ≤ η ≤ 1,

η ≡ 1 on B(xo, 2r), η ≡ 0 outside B(xo, 4r), (2.4.132)

|∂ α η(x)| ≤ Cα r − |α |, ∀x ∈ Rn and ∀α ∈ N0n .

2.4 Boundary-to-Domain Integral Operators on Open sets with UR Boundaries 405

We then proceed to split

f = η( f − fS4r ) + (1 − η)( f − fS4r ) + fS4r . (2.4.133)

From (2.4.133) and the cancelation property (2.4.120) we see that

 
Qi j f = Qi j η( f − fS4r ) + Qi j (1 − η)( f − fS4r ) . (2.4.134)

Thus,
∫  p
 
Qi j f (x) dist(x, ∂Ω) p−1 dx
T (Sr )
∫   p
 
≤C Qi j η( f − fS4r ) (x) dist(x, ∂Ω) p−1 dx
T (Sr )
∫   p
 
+C Qi j (1 − η)( f − fS4r ))(x) dist(x, ∂Ω) p−1 dx
T (Sr )

=: I + I I. (2.4.135)

We shall show that

I ≤ Cσ(Sr ) fp# (xo ) p and I I ≤ Cσ(Sr ) f1# (xo ) p (2.4.136)

where the L q -based Fefferman-Stein sharp maximal function fq# , with q ∈ [1, ∞),
has been defined in (A.0.123). To justify the first inequality in (2.4.136), we write
∫  p
  
I≤C Qi j η( f − fS4r ) (x) dist(x, ∂Ω) p−1 dx
Ω
∫ ∫
   
≤C η( f − fS ) p dσ ≤ C  f − fS  p dσ
4r 4r
∂Ω S4r
⨏
 
= Cσ(S4r )  f − fS  p dσ ≤ Cσ(Sr ) f # (xo ) p . (2.4.137)
4r p
S4r

Above, the first inequality follows from the definition of I in (2.4.135), the second
inequality follows from (2.4.34), the third inequality is clear from the support prop-
erties of the function η introduced in (2.4.132), the subsequent equality is obvious,
and the fourth inequality is due to the fact that σ is doubling (itself, a consequence
of the Ahlfors regularity of ∂Ω), and the definition of fp# in (A.0.123).
To justify the second inequality in (2.4.136), we first observe that for each point
x ∈ T(Sr ) we have
   ∫
  | f (y) − fS4r |
Qi j (1 − η)( f − fS4r ))(x) ≤ C dσ(y)
∂Ω\S2r |xo − y| n

≤ Cr −1 f1# (xo ). (2.4.138)

406 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Indeed, the first inequality above follows from the definition of Qi j (1−η)( f − fS4r ) ,
the properties of the function η from (2.4.132), and the fact that

|x − y| ≈ |xo − y| uniformly for x ∈ T(Sr ) and y ∈ ∂Ω \ S2r . (2.4.139)

The second inequality in (2.4.138) is a direct consequence of [112, (7.4.115)]. In

turn, from (2.4.135) and (2.4.138) we obtain
∫
I I ≤ Cr −p f1# (xo ) p dist(x, ∂Ω) p−1 dx
T (Sr )

≤ Cr n−1 f1# (xo ) p ≤ Cσ(Sr ) f1# (xo ) p, (2.4.140)

where the last equality

 is a consequence of the lower Ahlfors regularity of ∂Ω and
the fact that r ∈ 0, 2 diam(∂Ω) . This establishes the second inequality in (2.4.136).
At this stage, from (2.4.135), (2.4.136), and [112, (7.4.111)] we obtain
∫ 
1
n
 
sup  Qi j f  p dist(·, ∂Ω) p−1 dL n
x ∈∂Ω and σ B(x, r) ∩ ∂Ω B(x,r)∩Ω i, j=1
0<r <2 diam(∂Ω)
p
≤ C f  . . (2.4.141)
BMO(∂Ω,σ)

From this, (2.4.117) immediately follows in the case when ∂Ω is unbounded, or Ω is

bounded. To complete the proof of the corollary there remains to establish a similar
estimate in the case when Ω is an exterior domain and when the supremum is taken
in the regime r ∈ 2 diam(∂Ω), ∞ .
In such a scenario, the inequality we seek to establish reads
∫ 
1
n
 
sup Qi j f  p dist(·, ∂Ω) p−1 dL n ≤ C f  p . .
x ∈∂Ω σ(∂Ω) B(x,r)∩Ω i, j=1 BMO(∂Ω,σ)

(2.4.142)
⨏
Using the abbreviation f∂Ω := ∂Ω
f dσ we can write
∫ 
1
n
 
sup Qi j f  p dist(·, ∂Ω) p−1 dL n
x ∈∂Ω σ(∂Ω) B(x,r)∩Ω i, j=1

∫ 
1
n
  
≤ Qi j f − f∂Ω  p dist(·, ∂Ω) p−1 dL n
σ(∂Ω) Ω i, j=1
⨏
 
≤C  f − f∂Ω  p dσ
∂Ω
p
≤ C f  . , (2.4.143)
BMO(∂Ω,σ)
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 407

where the first inequality is based on (2.4.120), and the second one comes from
(2.4.34). This justifies (2.4.142) and finishes the proof of (2.4.117). In turn, (2.4.118)
follows by specializing (2.4.117) to the case p = 2. 

2.5 The Jump-Formula for Boundary-to-Domain Integral

Operators in Open Sets with Uniformly Rectifiable
Boundaries

Having proved operator bounds on UR sets in Theorem 2.3.2 and Theorem 2.4.1, we
shall now use these results in order to establish jump-formulas for integral operators
of the sort described in Theorem 2.5.1 below25. Once again, this refines work in the
first part of [63, Theorem 3.33, p. 2669] carried out for a smaller class of functions (we
presently identify the optimal class of functions for which said jump-formulas hold),
and in the more restrictive setting of UR domains Ω ⊆ Rn (in the terminology of
[63, Definition 3.7, p. 2631]; in particular, the condition that the geometric measure
theoretic boundary of Ω has full H n−1 -measure in its topological boundary – as
stipulated in [63, (3.1.22), p. 2631] – is no longer required here).

Theorem 2.5.1 For each n ∈ N with n ≥ 2 there exists an integer N = N(n) ≥ 4

with the following significance. Suppose Ω ⊆ Rn is a nonempty open set with the
property that ∂Ω is a UR set; in particular, Ω is a set of locally finite perimeter.
Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic outward
unit normal to Ω.
Next, consider a kernel k ∈ 𝒞 N (Rn \ {0}) which is odd and positive homogeneous
of degree 1 − n (cf. (2.3.3)) and define the singular integral operator acting on
arbitrary functions
 σ(x) 
f ∈ L 1 ∂Ω, (2.5.1)
1 + |x| n−1
according to
∫
(T f )(x) := k(x − y) f (y) dσ(y) for each x ∈ Ω, (2.5.2)
∂Ω

along with its principal-value version

∫
(T f )(x) := lim+ k(x − y) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω. (2.5.3)
ε→0
y ∈∂Ω
|x−y |>ε

25 the roots of this results go back in time to the work of Julian Karol Sokhotski who first proved
a special version in 1868, and Josip Plemelj who rediscovered it and used it as the main ingredient
in his solution of a Riemann-Hilbert Problem in 1908
408 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 
Then for each aperture parameter κ > 0 and each function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1

the following jump-formula26 holds:

 κ−n.t. 
 1
Tf  (x) = √ - k ν(x) f (x) + (T f )(x) for σ-a.e. x ∈ ∂∗ Ω, (2.5.4)
∂Ω 2 −1
√
where −1 = i ∈ C, and ‘hat’ denotes the Fourier transform in Rn (cf. (1.4.14)).

A few comments to help clarify the nature of the statement of Theorem 2.5.1 are
in order.
Comment 1. For each function f as in (2.5.1), the integral in (2.5.2) is absolutely
convergent and
T f ∈ 𝒞∞ (Ω). (2.5.5)
As such, for each given κ > 0, the nontangential boundary trace in (2.5.4) is defined
as κ−n.t.
 
Tf  (x) := lim (T f )(z) for σ-a.e. x ∈ ∂∗ Ω. (2.5.6)
∂Ω z ∈Γ κ (x)
z→x

That for σ-a.e. x ∈ ∂∗ Ω it is actually meaningful to consider the limit in the right-
hand side of (2.5.6) follows from [112, (8.8.52)] in [112, Proposition 8.8.6] and the
current assumptions on Ω.
Comment 2. The fact that the principal-value singular integral operator in (2.5.3) is
well defined when acting on functions as in (2.5.1) is guaranteed by Theorem 2.3.2
(presently used with Σ := ∂Ω).
Comment 3. It is natural to restrict attention to points on the geometric measure
theoretic boundary ∂∗ Ω (as opposed to, say, points on the topological boundary
∂Ω) when considering the jump-formula (2.5.4) since this involves the geometric
measure theoretic outward unit normal ν(x) which is well defined for σ-a.e. x ∈ ∂∗ Ω
(but not necessarily in a larger context).
Comment 4. The Fourier transform in Rn employed here is the one defined earlier in
(1.4.14). The reader is alerted to the fact that this is different from the normalizations
used in [175] and [166].
Comment 5. Estimate (2.4.55) guarantees that k defines a tempered distribution in
Rn (via integration against Schwartz functions). Moreover, -
k originally considered
in the class of tempered distributions in Rn satisfies

-
k ∈ 𝒞m (Rn \ {0}) if N ∈ N is even
(2.5.7)
and m ∈ N0 is such that m < N − 1.

Indeed, according to [109, Exercise 4.62, p. 144],

26 or saltus-condition, using an older piece of terminology; see, e.g., [4, p. 275]

2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 409

if u is a tempered distribution in Rn with uRn \{0} ∈ 𝒞 N (Rn \ {0}) for

some even N ∈ N, and uRn \{0} is positive homogeneous of degree
 (2.5.8)
θ < N − n, then - uRn \{0} ∈ 𝒞m (Rn \ {0}) whenever m ∈ N0 is such
that m < N − n − θ.
In particular, (2.5.7) ensures that

if N ≥ 2 then the function - k ν(·) is meaningfully defined σ-a.e. (2.5.9)
on ∂∗ Ω and, in fact, said function belongs to L ∞ (∂∗ Ω, σ).
Moreover, given that k is odd, it follows that
-
k : Rn \ {0} −→ C is an odd function. (2.5.10)

Comment 6. If for each given ω ∈ S n−1 we consider the hyperplane in Rn consisting


⊥
of vectors which are orthogonal to ω, i.e., ω := ξ ∈ R : ξ, ω = 0 , then
n

according to [109, Proposition 4.80, p. 172] we have

∫
-
k(ω) = −2i lim k(ξ + ω) dH n−1 (ξ), ∀ω ∈ S n−1 . (2.5.11)
r→∞
ξ ∈ ω ⊥, |ξ |<r

Since for each r ∈ (0, ∞) and each ω ∈ S n−1 the parity of k implies that
∫
k(ξ + ω) dH n−1 (ξ) (2.5.12)
ξ ∈ ω ⊥, |ξ |<r
∫
 
= 1
2 k(ξ + ω) − k(ξ − ω) dH n−1 (ξ),
ξ ∈ ω ⊥, |ξ |<r

and since the integrand in (2.5.12) exhibits better decay than the integrand in (2.5.11)
(specifically, we now have k(ξ + ω) − k(ξ − ω) = O(|ξ | −n ) as |ξ | → ∞), we may
recast (2.5.11) as
∫
 
-k(ω) = −i k(ξ + ω) − k(ξ − ω) dH n−1 (ξ), ∀ω ∈ S n−1 . (2.5.13)
ξ ∈ ω ⊥

In addition, if we introduce

x
Θ(x) := k for each x ∈ Rn \ {0}, (2.5.14)
|x|

then
410 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Θ ∈ 𝒞 N (Rn \ {0}) is odd, positive homogeneous of degree zero,

Θ(x) (2.5.15)
and k(x) = for each x ∈ Rn \ {0}.
|x| n−1
In turn, the latter formula implies that

Θ(ξ + ω) − Θ(ξ − ω)
k(ξ + ω) − k(ξ − ω) =
(1 + |ξ | 2 )(n−1)/2 (2.5.16)
whenever ω ∈ S n−1 and ξ ∈ ω ⊥ .

As a consequence, for σ-a.e. x ∈ ∂∗ Ω the quantity -
k ν(x) appearing in (2.5.4) may
be alternatively expressed as
∫
 
-
k ν(x) = −2i lim k ξ + ν(x) dH n−1 (ξ)
r→∞
ξ ∈ ν(x) ⊥, |ξ |<r
∫
 
= −i k(ξ + ν(x)) − k(ξ − ν(x)) dH n−1 (ξ)
ξ ∈ ν(x) ⊥
∫
Θ(ξ + ν(x)) − Θ(ξ − ν(x))
= −i dH n−1 (ξ). (2.5.17)
(1 + |ξ | 2 )(n−1)/2
ξ ∈ ν(x) ⊥

Comment 7. In relation to (2.5.8), for further reference it is also useful to make two
elementary observations. First, based on (2.5.8) and [109, Exercise 4.63] we have
that
if N ≥ 4 is even and b ∈ 𝒞 N (Rn \ {0}) is positive homogeneous
of degree θ ∈ (1 − n, N − 1 − n), then for each j ∈ {1, . . . , n} the
Fourier transforms of the tempered distributions induced by ∂j b and
(2.5.18)
b in Rn (via integration against Schwartz functions) are continuous
when restricted to Rn \ {0} and satisfy (∂. -
j b)(ξ) = iξ j b(ξ) for all
ξ ∈ R \ {0}.
n

Second, from (2.5.8) and [109, Theorem 4.26(b), p. 132] we may also conclude that

if u is a tempered distribution in Rn such that uRn \{0} ∈ 𝒞 N (Rn \ {0})

for some N > 1 odd, and (∇u)Rn \{0} is positive homogeneous of
 (2.5.19)
degree θ < N − 1 − n, then ∇u . n is continuous in Rn \ {0} and
R \{0}
ξk (∂. /
j u)(ξ) = ξ j (∂k u)(ξ) for each j, k ∈ {1, . . . , n} and ξ ∈ R \ {0}.
n

Comment 8. Recall (cf., e.g., [109, Corollary 4.65, p. 147]) that whenever n ∈ N
with n ≥ 2 and j ∈ {1, . . . , n} then
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 411
ξ
k-j (ξ) = −iωn−1 |ξ j|2 in 𝒮(Rn )
xj
(2.5.20)
if k j (x) := |x | n for all x ∈ Rn \ {0}.

More generally, we recall from [166, p. 73] that if P j is a homogeneous harmonic

polynomial of degree j ∈ N in Rn (where n ∈ N satisfies n ≥ 2), then

for each α ∈ (0, n) it follows that the Fourier transform (considered

with the normalization given in (1.4.14)) of the tempered distribu-
tion induced by the locally integrable function P j (x)/|x| n+j−α which
decays at infinity in Rn (via integration against functions from the
Schwartz class) is the tempered distribution induced by the locally in-
tegrable function which decays at infinity in Rn given by (with Γ the (2.5.21)
Gamma function)

Γ( j/2 + α/2) P j (x)

2α π n/2 (−i) j , ∀x ∈ Rn \ {0}.
Γ( j/2 + n/2 − α/2) |x| j+α

In the case when n ∈ N satisfies n ≥ 2, the exponent j ∈ N is odd, and

k(x) := P j (x)/|x| n−1+j for each x ∈ Rn \ {0} where

(2.5.22)
P j is a homogeneous harmonic polynomial of degree j in Rn

we conclude from (2.5.21) used with α := 1 that the Fourier transform (considered
with the normalization given in (1.4.14)) of k is the tempered distribution induced,
via integration against Schwartz functions, by the locally integrable function

Γ( j/2 + 1/2) P j (ξ)

-
k(ξ) = 2π n/2 (−i) j (2.5.23)
Γ( j/2 + n/2 − 1/2) |ξ | j+1

Γ( j/2 + 1/2)Γ(n/2) P j (ξ)

= (−i) j ωn−1 , ∀ξ ∈ Rn \ {0}.
Γ( j/2 + n/2 − 1/2) |ξ | j+1

Hence, in such a scenario, for σ-a.e. x ∈ ∂∗ Ω the quantity -
k ν(x) appearing in
(2.5.4) may be computed as
 Γ( j/2 + 1/2)Γ(n/2) 
-
k ν(x) = (−i) j ωn−1 P j ν(x) . (2.5.24)
Γ( j/2 + n/2 − 1/2)
For example, in the special situation when j = 1 and, for some fixed vector a ∈ Rn ,
we now take P1 (x) := x, a for each x ∈ Rn then (2.5.24) becomes
 % &
-
k ν(x) = −iωn−1 ν(x), a for σ-a.e. x ∈ ∂∗ Ω. (2.5.25)

Also, in the particular case when n = 2 (so that R2 ≡ C) and k is as in (2.5.22) with
j ∈ N odd number and where, for some a, b ∈ C, we now take P j (z) := az j + bz j
412 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

for each z ∈ C, then (2.5.24) reduces to

  j
-
k ν(x) = 2π(−i) j aν(x) j + bν(x) for σ-a.e. x ∈ ∂∗ Ω. (2.5.26)

Finally, we wish to note that whenever n ∈ N with n ≥ 3 and a, b, c ∈ {1, . . . , n}

then the Fourier transform of the tempered distribution induced by the locally inte-
grable function in Rn defined as
x a xb xc
k abc (x) := for each x ∈ Rn \ {0} (2.5.27)
|x| n+2

is the tempered distribution 0 k abc ∈ 𝒮(Rn ) which is of function type outside the
origin, where it is explicitly given by
 
0 ω n−1 ξ c ξ b ξ a ξ a ξ b ξ c
k abc (ξ) = −i δab 2 + δac 2 + δbc 2 − 2 . (2.5.28)
n |ξ | |ξ | |ξ | |ξ | 4

(Indeed, this may be seen from (2.5.20), [109, (7.14.9), p. 320], and a straightforward
computation based of properties of the Fourier transform.) Note that the kernel
(2.5.27) falls under the scope of Theorem 2.5.1, so (2.5.28) gives an explicit formula
for the jump-term in (2.5.4) in this significant speacial case.
Comment 9. The jump-formula (2.5.4) readily extends to the case when the kernel k
is a matrix-valued function and the -
 function f is vector-valued. In such a scenario, k
-
is computed entry-by-entry and k ν(x) f (x) is considered in the sense of the natural
action of a matrix on a vector.
Comment 10. In the case when ∂Ω is bounded we have 1+ |x1| n−1 ≈ 1 uniformly for
 
x ∈ ∂Ω, hence the space L 1 ∂Ω, 1+σ(x)
|x | n−1 simply reduces to L 1 (∂Ω, σ) in such a
situation.
Comment 11. It is relevant to contrast the hypotheses guaranteeing the validity of
our jump-formula (2.5.4) to the most accommodating geometric setting in which
its very formulation is meaningful. Two key aspects in this regard are as follows.
First, the very presence of ν(x) in the right-hand side limits x to ∂∗ Ω since, up
to a σ-nullset, this is the largest ambient where the geometric measure theoretic
outward unit normal may be considered. Second, on the analytic side, assuming as

we presently do that f belongs to the weighted Lebesgue space L 1 ∂Ω, 1+σ(x) |x | n−1
is
optimal since this is the largest class of functions on which the integral operator T
is meaningfully defined as in (2.5.2) via an absolutely convergent integral.
Before turning to the proof Theorem 2.5.1, in the next several results (Corollar-
ies 2.5.2-2.5.7) we shall discuss a variety of its special cases and natural adaptations,
which are of particular interest in applications. We begin by presenting a version of
Theorem 2.5.1 pertaining to the scales of ordinary Lebesgue and Lorentz spaces.
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 413

Corollary 2.5.2 In the context of Theorem 2.5.1 the jump-formula (2.5.4) holds for
each function f in the Lebesgue space L p (∂Ω, σ) with 1 ≤ p < ∞, as well as
each function f belonging to the Lorentz space L p,q (∂Ω, σ) with 1 < p < ∞ and
0 < q ≤ ∞.
Proof The embedding in [112, (7.7.106)] (used with Σ := ∂Ω) ensures that the
jump-formula (2.5.4) holds, in particular, for every function f ∈ L p (∂Ω, σ) with
1 ≤ p < ∞, while [112, (7.7.107)] (again, used with Σ := ∂Ω) guarantees that
the jump-formula (2.5.4) is also valid for every function f ∈ L p,q (∂Ω, σ) with
p ∈ (1, ∞) and q ∈ (0, ∞]. 
Next we present a result which, in particular, gives a version of the jump-formula
(2.5.4) for bounded functions.
Corollary 2.5.3 Let Ω ⊆ Rn (where n ∈ N with n ≥ 2) be a nonempty open set
with the property that ∂Ω is an Ahlfors regular set, and abbreviate σ := H n−1 ∂Ω.
Given a kernel k ∈ 𝒞1 (Rn \ {0}) which is odd and positive homogeneous of degree
1 − n, consider the following modified version of the operator in (2.5.1)-(2.5.2):
∫
 
(Tmod f )(x) := k(x − y) − k1 (−y) f (y) dσ(y) for each x ∈ Ω,
∂Ω  σ(x)  (2.5.29)
where k1 := k · 1Rn \B(1,0) and f ∈ L 1 ∂Ω, is arbitrary.
1 + |x| n

Then Tmod is meaningfully defined in the above context and is compatible with T
from (2.5.1)-(2.5.2), in the sense that for each function f belonging to the smaller

space L 1 ∂Ω, 1+σ(x)
|x | n−1
(hence, in particular, for each function f ∈ L p (∂Ω, σ) with
p ∈ [1, ∞)) the difference

C f := Tmod f − T f is a constant in Ω. (2.5.30)

As a consequence, ∇Tmod f = ∇T f in Ω for each such function f .

Next, strengthen the original hypotheses by assuming now that k ∈ 𝒞 N (Rn \ {0})
for some integer N = N(n) ≥ 4 large enough and that ∂Ω is a UR set. In this setting,
denote by ν the geometric measure theoretic outward unit normal to Ω, and bring
in the operator Tmod from (2.3.31) with Σ := ∂Ω. Then, with the same notational
conventions as in (2.5.4), the following jump-formula holds:

 κ−n.t. 
 1
Tmod f  (x) = √ - k ν(x) f (x) + (Tmod f )(x) for σ-a.e. x ∈ ∂∗ Ω,
∂Ω 2 −1  σ(x) 
(2.5.31)
for each given κ > 0 and each given function f ∈ L 1 ∂Ω, .
1 + |x| n

Also, for each aperture parameter κ ∈ (0, ∞) and each truncation parameter
ε ∈ (0, ∞), in a quantitative way one has

Nκε (Tmod f ) ∈ Lloc (∂Ω, σ) for each function

p

 (2.5.32)
f ∈ L 1 ∂Ω, 1+σ(x)
p
|x | n ∩ Lloc (∂Ω, σ) with p ∈ (1, ∞).
414 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Proof Note that if x ∈ Ω, y ∈ ∂Ω, and 0 < ε < |x − y|, then k(x − y) = k ε (x − y),
where k ε is as in (2.3.115). Keeping this in mind, the same argument used to prove
(2.3.119) presently gives that
∫
 
 k(x − y) − k1 (−y) | f (y)| dσ(y) < +∞
∂Ω  σ(x)  (2.5.33)
for each x ∈ Ω and each f ∈ L 1 ∂Ω, .
1 + |x| n

Thus, the integral in the definition of (Tmod f )(x) in (2.5.29) is absolutely convergent

for each x ∈ Ω and each f ∈ L 1 ∂Ω, 1+σ(x)|x | n . From (2.3.116) (used with |α| = 0) we
also see that
 σ(x) 
if actually f belongs to the space L 1 ∂Ω, then
1 + |x| n−1
∫ ∫ (2.5.34)
| f (y)|
|k1 (−y) f (y)| dσ(y) ≤ dσ(y) < +∞.
∂Ω 1 + |y|
n−1
∂Ω

In concert, (2.5.33)-(2.5.33) prove that

∫
C f := Tmod f − T f = k1 (−y) f (y) dσ(y) ∈ C,
∂Ω
 σ(x) 
(2.5.35)
is a constant in Ω, whenever f ∈ L 1 ∂Ω, .
1 + |x| n−1
Assuming the stronger hypotheses made in the second part of the statement, there
remains to prove the jump-formula recorded in (2.5.31). To this end, fix a point
x0 ∈ ∂Ω and pick an arbitrary number r ∈ (0, ∞). Next, having selected some

function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n , split

f = f1 + f2 on ∂Ω, where
(2.5.36)
f1 := 1∂Ω∩B(x0,4r) · f and f2 := 1∂Ω\B(x0,4r) · f ,

then decompose
Tmod f = Tmod f1 + Tmod f2 in Ω. (2.5.37)
Note that since Tmod f2 has a continuous extension to B(x0, r) we trivially have that
κ−n.t.

the nontangential trace Tmod f2  exists at every point in ∂nta Ω ∩ B(x0, r). More
∂Ω
precisely,
κ−n.t. ∫
  
Tmod f2  (x) = k(x − y) − k1 (−y) f2 (y) dσ(y)
∂Ω ∂Ω

= Tmod f2 (x) for each x ∈ ∂nta Ω ∩ B(x0, r). (2.5.38)
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 415

In particular, from (2.5.38) and item (iii) in [112, Proposition 8.8.6] we conclude
that, on the one hand,
κ−n.t. 

Tmod f2  (x) = Tmod f2 (x) at σ-a.e. x ∈ ∂∗ Ω ∩ B(x0, r). (2.5.39)
∂Ω

On the other hand, since f1 ∈ L 1 (∂Ω, σ), at σ-a.e. point x ∈ ∂∗ Ω ∩ B(x0, r) we may
write
κ−n.t. κ−n.t.
 
Tmod f1  (x) = T f1  (x) + C f1
∂Ω ∂Ω

1 
= √ - k ν(x) f1 (x) + (T f1 )(x) + C f1
2 −1
1  
= √ - k ν(x) f (x) + Tmod f1 (x). (2.5.40)
2 −1
Above, the first equality uses (2.5.35) (with f replaced by f1 ), the second equality
is provided by Theorem 2.5.1, and the final equality is seen from (2.3.34) (with
Σ := ∂Ω, a scenario in which C f from (2.3.34) coincides with C f from (2.5.35)).
In turn, from (2.5.39)-(2.5.40) and (2.5.36) we conclude that
 κ−n.t. 
 √1 -
Tmod f  (x) = k ν(x) f (x) + (Tmod f )(x)
∂Ω 2 −1 (2.5.41)
for σ-a.e. x ∈ ∂∗ Ω ∩ B(x0, r).

Given that r > 0 has been arbitrarily chosen, this ultimately proves (2.5.31).
Finally, to justify the claim made in (2.5.32), fix some κ, ε ∈ (0, ∞) and pick an

arbitrary function f ∈ L 1 ∂Ω, 1+σ(x)
p
|x | n ∩ Lloc (∂Ω, σ) with

p ∈ (1, ∞). Having picked

a reference point x0 ∈ ∂Ω along with a radius r > max (1 + κ)ε, 14 (1 + |x0 |) , split
f as in (2.5.36) then decompose Tmod f as in (2.5.37). For each x ∈ ∂Ω ∩ B(x0, r),
each z ∈ Γκ (x) with dist(z , ∂Ω) < ε, and each ξ ∈ [z, x] we have

|z − x| < (1 + κ) dist(z , ∂Ω) ≤ (1 + κ)ε < r, (2.5.42)

and

|x − y| ≤ |x − ξ | + |ξ − y| ≤ |x − z| + |ξ − y|

≤ r + |ξ − y| ≤ 13 |x − y| + |ξ − y|, (2.5.43)

hence, |x − y| ≤ 32 |ξ − y|. Also, |z| ≤ |z − x| + |x − x0 | + |x0 | ≤ 2r + |x0 |, and the fact

that we have 4r ≤ |x0 − y| ≤ |x0 | + |y| forces |y| ≥ 1. On account of these estimates
and the Mean Value Theorem we may then write
416 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
  
Tmod f2 (z) ≤ |k(z − y) − k1 (−y)|  f 1∂Ω\B(x0,4r) (y) dσ(y)
∂Ω
∫
 
= |k(z − y) − k(−y)|  f 1∂Ω\B(x0,4r) (y) dσ(y)
∂Ω
∫  |z| 
≤ C(k, n) sup | f (y)| dσ(y)
∂Ω\B(x0,4r) ξ ∈[z,x] |ξ − y| n
∫
| f (y)|
≤ C(k, n, r, x0 ) dσ(y)
∂Ω\B(x0,4r) |x − y| n
∫
| f (y)|
≤ C(k, n, r, x0 ) dσ(y) < +∞. (2.5.44)
∂Ω 1 + |y| n
In concert with [112, (8.2.28)] this proves that, in a quantitative fashion,
 
Nκε Tmod f2 ∈ L ∞ ∂Ω ∩ B(x0, r), σ . (2.5.45)

Since f1 ∈ L p (∂Ω, σ) with p ∈ (1, ∞), we conclude from (2.5.30) and (2.4.9) that
(once again in a quantitative manner)

Nκε Tmod f1 ∈ L p (∂Ω, σ) + Lloc
∞
(∂Ω, σ). (2.5.46)

At this stage, the claim made in (2.5.32) follows from (2.5.45) and (2.5.46) (keeping
in mind (2.5.37) and [112, (8.2.9), (8.2.28)]). 

There is also a suitable version of Theorem 2.5.1 on the scale of Hardy spaces,
as described next.

Corollary 2.5.4 Let Ω ⊆ Rn (where n ∈ N with n ≥ 2) be a nonempty open set with

the property that ∂Ω is a UR set. Abbreviate σ := H n−1 ∂Ω and denote by ν the
geometric measure theoretic outward unit normal to Ω.
Next, fix a kernel k ∈ 𝒞 N (Rn \{0}) (for some integer N = N(n) ≥ 4 large enough)
which is odd and positive homogeneous of degree 1 − n, and pick p ∈ n−1 n , 1 along
with κ ∈ (0, ∞). Recall from item (3) in Theorem 2.4.1 (with Σ := ∂Ω) that if for
each f ∈ H p (∂Ω, σ) one defines
%
k(x − ·), f  if ∂Ω is bounded,
(T f )(x) := % ∀x ∈ Ω, (2.5.47)
[k(x − ·)], f  if ∂Ω is unbounded,

where ·, · stands for the duality bracket described in [113, Theorem 4.6.1] (with
Σ := ∂Ω), then T is a well-defined linear and bounded operator
p
T : H p (∂Ω, σ) −→ Nκ (Ω; σ). (2.5.48)
p
(recall that Nκ (Ω; σ) has been introduced in (A.0.98)) which acts in a coherent
fashion with T as originally defined in (2.4.3)-(2.4.4) (with Σ := ∂Ω).
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 417

Also, recall from item (6) of Theorem 2.3.2 that the principal-value singular
integral of formal convolution type with the kernel k on ∂Ω (cf. (2.5.3)) induces a
well-defined linear and bounded operator

T : H p (∂Ω, σ) −→ L p (∂Ω, σ). (2.5.49)

Finally, recall the L p -filtering operator H : H p (∂Ω, σ) → L p (∂Ω, σ) from [113,

Theorem 4.9.1] (used here with Σ := ∂Ω).
Then for each f ∈ H p (∂Ω, σ) the following jump-formula holds:

 κ−n.t. 
 1
Tf  (x) = √ - k ν(x) (H f )(x) + (T f )(x) for σ-a.e. x ∈ ∂∗ Ω. (2.5.50)
∂Ω 2 −1
One significant consequence of the above result is that, in the context of Corol-
lary 2.5.4, the operator T from (2.5.49) acts in a compatible fashion with its principal-
value version (defined as in (2.3.15) with Σ := ∂Ω), when the outputs are compared
on ∂∗ Ω, i.e.,
  
if f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
∩ H p (∂Ω, σ) for some p ∈ n−1
n ,1
then T f ∈ L p (∂Ω, σ) defined in the sense of (2.5.49) satisfies
∫ (2.5.51)
(T f )(x) = lim+ k(x − y) f (y) dσ(y) at σ-a.e. x ∈ ∂∗ Ω.
ε→0
y ∈∂Ω
|x−y |>ε

Indeed, this is seen from (2.5.50), [113, (4.9.3)], and the fact that, as noted earlier,
the operator T from (2.5.48) acts in a coherent fashion with its version originally
defined as in (2.4.3)-(2.4.4) with Σ := ∂Ω.
In the case when Ω is actually a UR domain, the identity recorded in (2.5.50) also
provides an alternative formula for the action of the operator (2.5.49), namely

if f ∈ H p (∂Ω, σ) with p ∈ n−1n , 1 , then T f ∈ L (∂Ω, σ) is given by
p

  κ−n.t. (2.5.52)
1 
(T f )(x) = − √ - k ν(x) (H f )(x) + T f  (x) for σ-a.e. x ∈ ∂Ω.
2 −1 ∂Ω

For example, upon recalling from [113, (4.2.17)]

 that for each given pair of points

x0, x1 ∈ ∂Ω the distribution δx0 − δx1 ∈ Lipc (∂Ω) belongs to the Hardy space
H p (∂Ω, σ) for every p ∈ n−1 n , 1 , from formula (2.5.52) and [113, (4.9.9)] we
conclude that

T(δx0 − δx1 ) (x) = k(x − x0 ) − k(x − x1 ) for σ-a.e. x ∈ ∂Ω. (2.5.53)

Parenthetically, given the properties of the kernel k (cf. (2.4.55)-(2.4.56)), [112,

Lemma 7.2.1] (cf. [112,  (7.2.5)]) provides independent confirmation of the fact that
   n−1
k(· − x0 ) − k(· − x1 )  belongs to L (∂Ω, σ) for each p ∈ n , 1 .
p
∂Ω
418 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Proof of Corollary 2.5.4 Fix some integrability exponent q ∈ (1, ∞). Then for each
function f ∈ H p (∂Ω, σ) ∩ L q (∂Ω, σ) the function in left-hand side of (2.5.50)
becomes (thanks to [113, (4.9.3)] and the very last claim in item (6) of Theorem 2.3.2)
∫
1 
√ - k ν(x) f (x) + lim+ k(x − y) f (y) dσ(y) for σ-a.e. x ∈ ∂∗ Ω.
2 −1 ε→0
y ∈∂Ω
|x−y |>ε
(2.5.54)
On the other hand, on account of (2.5.47) and the compatibility results from [113,
Lemmas 4.6.5-4.6.6] we have
∫
(T f )(x) = k(x − y) f (y) dσ(y) for each x ∈ Ω. (2.5.55)
∂Ω

In turn, from (2.5.55) and (2.5.54) we see that the jump-formula (2.5.50) holds for
each function f ∈ H p (∂Ω, σ) ∩ L q (∂Ω, σ). Consequently (cf. [113, (4.4.114)]),
the jump-formula recorded in (2.5.50) holds for any
p,q
function f belonging to the space Hfin (∂Ω, σ). (2.5.56)

Granted this, the idea now is to apply [112, Proposition 6.2.11] with 𝒳 := Ω∪∂∗ Ω
equipped with the topology inherited from the ambient Euclidean space, X := ∂∗ Ω,
μ := H n−1 ∂∗ Ω which is a locally finite complete Borel-regular measure on X
by virtue of [112, Lemma 3.6.4] used with s := n − 1 (that the hypotheses in
[112, (3.6.25)] are presently satisfied is seen from [112, (5.6.33), (5.6.35)]), the sets
Γ(x) := Γκ (x) for each x ∈ ∂∗ Ω which satisfy [112, (6.2.71)] thanks to [112, (8.8.45)],
the quasi-Banach space Y := H p (∂Ω, σ), the linear operator T as in (2.5.48), and
p,q
V := Hfin (∂Ω, σ), which is a dense subset of Y , as indicated in [113, (4.4.114)]. In
view of these choices, (2.5.56) implies that [112, (6.2.75)] holds. Also, the maximal
operator associated with the present choices as in [112, (6.2.73)] is (bearing in mind
that T maps distributions from H p (∂Ω, σ) into continuous functions on Ω)

T f (x) := supy ∈Γκ (x) |(T f )(y)| = Nκ (T f ) (x)
(2.5.57)
at σ-a.e. x ∈ ∂∗ Ω, for each f ∈ H p (∂Ω, σ).

Since (2.4.14) implies that

Nκ (T ·) : H p (∂Ω, σ) −→ L p (∂Ω, σ) (2.5.58)

is a well-defined, sub-linear and bounded operator, and since the restriction operator

L p (∂Ω, σ)  g −→ g ∂∗ Ω ∈ L p,∞ (∂∗ Ω, σ) (2.5.59)

is well defined, linear, and bounded (cf. [112, (6.2.17)] and [112, (6.2.27)]), it follows
that [112, (6.2.74)] holds. As such, all hypotheses of [112, Proposition 6.2.11] are
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 419

verified, which then permits us to conclude that

for each f ∈ H p (∂Ω, σ), the nontangential limit

 κ−n.t.
 (2.5.60)
Tf  (x) exists at σ-a.e. point x ∈ ∂∗ Ω.
∂Ω

Next, thanks to (2.5.60), (2.5.58), [112, Corollary 8.9.6], and [112, (8.9.8)], we
see that the operator

 κ−n.t.

H p (∂Ω, σ)  f −→ T f  ∈ L p (∂∗ Ω, σ) (2.5.61)
∂Ω

is well defined, linear and bounded. Moreover, from [113, (4.9.2)], (2.5.9), and
(2.5.49) we also have that
1 
H p (∂Ω, σ)  f −→ √ - k ν(·) H f + T f ∈ L p (∂∗ Ω, σ) (2.5.62)
2 −1
is a well-defined, linear and bounded operator. Combining (2.5.61), (2.5.62), (2.5.56),
and [113, (4.4.114)] then establishes (2.5.50) for each f ∈ H p (∂Ω, σ). 

Corollary 2.5.4 also extends naturally to the more inclusive scale of Lorentz-based
Hardy spaces, as indicated below.

Corollary 2.5.5 Suppose Ω ⊆ Rn (where n ∈ N with n ≥ 2) is a nonempty open

set with the property that ∂Ω is a UR set. Abbreviate σ := H n−1 ∂Ω and denote
by ν the geometric measure theoretic outward unit normal to Ω. Next, fix a kernel
k ∈ 𝒞 N (Rn \ {0}) (for some integer N = N(n) ≥ 4 large enough) which is odd and
n ,1
positive homogeneous of degree 1 − n, and pick integrability exponents p ∈ n−1
and q ∈ (0, ∞] along with an aperture parameter κ ∈ (0, ∞). Recall from (2.3.52)-
(2.3.55) that the principal-value singular integral of formal convolution type with
the kernel k on ∂Ω (cf. (2.5.3)) induces a well-defined linear and bounded operator

T : H p,q (∂Ω, σ) −→ L p,q (∂Ω, σ). (2.5.63)

Also, as in (2.4.24) (with Σ := ∂Ω), for each f ∈ H p,q (∂Ω, σ) define

% 
(T f )(x) := (H p, q (∂Ω,σ))∗ k(x − ·)∂Ω, f  H p, q (∂Ω,σ) for all x ∈ Ω (2.5.64)

which, according to [113, Proposition 4.8.9], acts in a coherent fashion with the
original version of T defined as in (2.4.3)-(2.4.4) with Σ := ∂Ω. Finally, recall from
[113, (4.9.5) in Theorem 4.9.1] (used here with Σ := ∂Ω) the L p,q -filtering operator
H : H p,q (∂Ω, σ) → L p,q (∂Ω, σ) .
Then for each f ∈ H p,q (∂Ω, σ) the following jump-formula holds:

 κ−n.t. 
 1
Tf  (x) = √ - k ν(x) (H f )(x) + (T f )(x) for σ-a.e. x ∈ ∂∗ Ω. (2.5.65)
∂Ω 2 −1
420 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

A byproduct of Corollary 2.5.5 is that, in the above context, the operator T from
(2.5.63) acts in a coherent manner with its principal-value counterpart (defined as in
(2.3.15) with Σ := ∂Ω), when the outputs are compared on ∂∗ Ω, i.e.,
  
if f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 ∩ H p,q (∂Ω, σ) for p ∈ n−1
n , 1 and q ∈ (0, ∞],
then with T f ∈ L p,q (∂Ω, σ) considered in the sense of (2.5.63) we have
∫
(T f )(x) = lim+ k(x − y) f (y) dσ(y) at σ-a.e. point x ∈ ∂∗ Ω.
ε→0
y ∈∂Ω
|x−y |>ε
(2.5.66)
This follows from (2.5.65), [113, (4.9.7)], and the fact that, as remarked above, the
operator T from (2.5.64) acts in a coherent fashion with its version originally defined
as in (2.4.3)-(2.4.4) with Σ := ∂Ω.
When Ω is in fact a UR domain, identity (2.5.65) may also be construed as an
alternative formula for the action of the operator (2.5.63), namely

if f ∈ H p,q (∂Ω, σ) with p ∈ n−1
n , 1 and q ∈ (0, ∞]
then T f ∈ L p,q (∂Ω, σ) is given at σ-a.e. x ∈ ∂Ω by
(2.5.67)
  κ−n.t.
1 
(T f )(x) = − √ - k ν(x) (H f )(x) + T f  (x).
2 −1 ∂Ω

To provide a concrete example, recall

 from [113, (4.2.34)] that for each fixed point

x0 ∈ ∂Ω the distribution δx0 ∈ Lipc (∂Ω) belongs to the weak Hardy space
H 1,∞ (∂Ω, σ). Then from (2.5.67), [113, (4.9.8)], and (2.4.26) we see that

the operator T, considered as in (2.5.63) (with p := 1 and q := ∞),

(2.5.68)
satisfies (T δx0 )(x) = k(x − x0 ) for σ-a.e. point x ∈ ∂Ω.

Incidentally, this yields independent confirmation of the fact that k(·− x0 )∂Ω belongs
to the weak Lebesgue space L 1,∞ (∂Ω, σ) (see [112, Example 6.2.2] in this regard).
Proof of Corollary 2.5.5 When q ∈ (0, ∞), we may follow in broad outline the
same argument used earlier in the proof of Corollary 2.5.4. Once again, the idea is
to apply [112, Proposition 6.2.11] for the same choices of the parameters 𝒳, X, and
μ as before, except that we now take Y := H p,q (∂Ω, σ), consider the linear operator
T as in (2.5.64) and, this time, define V := H q0 (∂Ω, σ) ∩ L q1 (∂Ω, σ) for some
q0 ∈ n−1n , p and q1 ∈ (1, ∞). According to [113, (4.3.145)], V is a dense subset
of the present choice of Y , since q < ∞. Thanks to (2.4.27) and Corollary 2.5.4,
all hypotheses of [112, Proposition 6.2.11] are then satisfied. This permits us to
conclude that
for each f ∈ H p,q (∂Ω, σ), the nontangential limit
 κ−n.t.
 (2.5.69)
Tf  (x) exists at σ-a.e. point x ∈ ∂∗ Ω.
∂Ω
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 421

Granted this, (2.4.27), [112, Corollary 8.9.6], [112, (8.9.8)], [112, (6.2.16)], and
[112, (6.2.17)] then imply that we have a well-defined, linear, and bounded operator

 κ−n.t.

H p,q (∂Ω, σ)  f −→ T f  ∈ L p,q (∂∗ Ω, σ). (2.5.70)
∂Ω

In addition, from [113, (4.9.5) in Theorem 4.9.1], (2.5.9), (2.5.63), and [112, (6.2.17)]
we also know that
1 
H p,q (∂Ω, σ)  f −→ √ - k ν(·) H f + T f ∈ L p,q (∂∗ Ω, σ) (2.5.71)
2 −1
is a well-defined, linear and bounded operator. From (2.5.70), (2.5.71), (2.5.56),
(2.5.50), and [113, (4.3.145)] we may now conclude that the jump-formula (2.5.65)
is indeed valid for each f ∈ H p,q (∂Ω, σ) when q < ∞.
An argument which works for the full range  q ∈ (0, ∞] is based on the ob-
servation that once we have fixed p0 ∈ n−1 n , p and p1 ∈ (1, ∞) then [113,
(1.3.41)], [113, (4.3.3)], and [112, (3.6.27)] ensure that the space H p,q (∂Ω, σ) em-
beds (continuously) into H p0 (∂Ω, σ) + L p1 (∂Ω, σ). Hence, any given distribution
f ∈ H p,q (∂Ω, σ) may be decomposed as f = f0 + f1 (in the sense of distributions
on ∂Ω) with f0 ∈ H p0 (∂Ω, σ) and f1 ∈ L p1 (∂Ω, σ). Then [113, Lemma 4.7.1]
implies that   
T f (x) = T f0 (x) + T f1 (x) for each x ∈ Ω, (2.5.72)
where T f0 is considered as in (2.5.47) (with p replaced by p0 ) and T f1 is considered
as in (2.5.1)-(2.5.2) (cf. also [112, (7.7.106)]). In turn, from (2.5.72), (2.5.50), and
(2.5.4) we conclude that at σ-a.e. point x ∈ ∂∗ Ω we have

 κ−n.t. 
 1
Tf  (x) = √ - k ν(x) (H f0 )(x) + (T f0 )(x)
∂Ω 2 −1
1 
+ √ - k ν(x) f1 (x) + (T f1 )(x). (2.5.73)
2 −1
From [113, (4.9.3), (4.9.6)] we also know that

(H f0 ) + f1 = H f0 + H f1 = H( f0 + f1 ) = H f at σ-a.e. point on ∂Ω, (2.5.74)

where as (2.3.54) implies

T f = T f0 + T f1 at σ-a.e. point on ∂Ω. (2.5.75)

At this stage, (2.5.65) follows from (2.5.73), (2.5.74), and (2.5.75). 

We continue by presenting a version of Theorem 2.5.1 adapted to the scale of

Muckenhoupt weighted Lebesgue spaces, as well as to the scale of Morrey spaces.
422 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Corollary 2.5.6 Suppose Ω ⊆ Rn (where n ∈ N with n ≥ 2) is a nonempty open set

with the property that ∂Ω is a UR set. Abbreviate σ := H n−1 ∂Ω and denote by
ν the geometric measure theoretic outward unit normal to Ω. Also, fix a sufficiently
large integer N = N(n) and consider a kernel k ∈ 𝒞 N (Rn \ {0}) as in (2.3.3).
Finally, define the integral operator T , along with its principal version T, in relation
to the domain Ω and the kernel k as in (2.5.2)-(2.5.3), and fix an aperture parameter
κ ∈ (0, ∞).
Then for each integrability exponent p ∈ (1, ∞), each Muckenhoupt weight
w ∈ Ap (∂Ω, σ), and each function f ∈ L p (∂Ω, wσ), the following jump-formula
holds:
 κ−n.t. 
 1
Tf  (x) = √ - k ν(x) f (x) + (T f )(x) for σ-a.e. x ∈ ∂∗ Ω. (2.5.76)
∂Ω 2 −1
Moreover, the jump-formula (2.5.76) also holds for each function f belonging to
the Morrey space M p,λ (∂Ω, σ) with p ∈ (1, ∞) and λ ∈ (0, n − 1).

Proof Thanks to [112, (7.7.104)] and [113, (6.2.25)], we may invoke Theorem 2.5.1
to conclude that the jump-formula (2.5.76) holds both when f ∈ L p (∂Ω, wσ) with
p ∈ (1, ∞) and w ∈ Ap (∂Ω, σ), as well as when f ∈ M p,λ (∂Ω, σ) with p ∈ (1, ∞)
and λ ∈ (0, n − 1).
An alternative proof of (2.5.76) in the former scenario, which avoids [112,
(7.7.104)] and, instead, involves approximating functions in weighted Lebesgue
spaces by functions in ordinary Lebesgue spaces, goes as follows. The fact that
simple functions are dense in Lebesgue space associated with any given generic
measure space (cf. [112, (3.1.11)]) implies that L p (∂Ω, σ) ∩ L p (∂Ω, wσ) is dense
in L p (∂Ω, wσ). As such, having fixed an arbitrary function f ∈ L p (∂Ω, wσ) we
can find a sequence

{ f j } j ∈N ⊆ L p (∂Ω, σ) ∩ L p (∂Ω, wσ) such that

(2.5.77)
f j −→ f in L p (∂Ω, wσ) as j → ∞.

In concert with item (10) in Theorem 2.3.2 and item (5) in Theorem 2.4.1, the
convergence result in (2.5.77) implies that

Nκ T ( f − f j ) → 0 in L p (∂Ω, wσ) and
(2.5.78)
T f j → T f in L p (∂Ω, wσ) as j → ∞.

As such, by passing to a subsequence if necessary, matters may be arranged so that,

as j → ∞,
f j (x) → f (x), (T f j )(x) → (T f )(x), and
 (2.5.79)
Nκ T ( f − f j ) (x) → 0, for σ-a.e. x ∈ ∂Ω.
Fix now j ∈ N arbitrary. Then for σ-a.e. x ∈ ∂∗ Ω we may rely on Corollary 2.5.2 to
write
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 423
   
 
lim sup(T f )(z) − √1 - k ν(x) f (x) + (T f )(x)  (2.5.80)
2 −1
z ∈Γκ (x)
z→x
 
 
≤ lim sup  T ( f − f j ) (z)
z ∈Γκ (x)
z→x
   
 √1 -

+ lim sup  T f j (z) − k ν(x) f j (x) + (T f j )(x) 
2 −1
z ∈Γκ (x)
z→x
     
 √1 - √1 -

+ k ν(x) f j (x) + (T f j )(x) − k ν(x) f (x) + (T f )(x) 
2 −1 2 −1
    
≤ Nκ T ( f − f j ) (x) + Ck | f j (x) − f (x)| + (T f j )(x) − (T f )(x),

where Ck ∈ (0, ∞) appearing above is a constant which depends only on the kernel
k. In light of (2.5.79) we may now conclude that for each f ∈ L p (∂Ω, wσ) the
jump-formula (2.5.76) is valid at σ-a.e. point x ∈ ∂∗ Ω. 

Here is a two-sided jump-formula, across the boundary of a UR domain. Results

similar in spirit to Corollary 2.5.7 below are also valid in relation to the jump-
formulas from Corollary 2.5.4, Corollary 2.5.5, and Corollary 2.5.6.

Corollary 2.5.7 For each n ∈ N with n ≥ 2 there exists an integer N = N(n) ≥ 4

with the following significance. Let k ∈ 𝒞 N (Rn \ {0}) be an odd, positive ho-
mogeneous kernel of degree 1 − n (cf. (2.3.3)), and let Ω  Rn be a UR domain.
Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic outward
unit normal to Ω. Next, define

Ω+ := Ω and Ω− := Rn \ Ω, (2.5.81)
 
then consider the integral operators acting on functions f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
ac-
cording to
∫
(T± f )(x) := k(x − y) f (y) dσ(y) for all x ∈ Ω± and (2.5.82)
∂Ω
∫
(T f )(x) := lim+ k(x − y) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω. (2.5.83)
ε→0
y ∈∂Ω
|x−y |>ε

Then Ω− is also a UR domain, whose topological boundary coincides with that

of Ω, and whose geometric measure theoretic boundary agrees with that of Ω (i.e.,
∂(Ω− ) = ∂Ω and ∂∗ (Ω− ) = ∂∗ Ω). In addition, the geometric measure theoretic
 to Ω− is −ν
outward unit normal  at σ-a.e. point on ∂Ω. Furthermore, for each κ > 0
σ(x)
and each f ∈ L ∂Ω, 1+ |x | n−1 the following jump-formulas hold:
1
424 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

 κ−n.t. 
 1
T± f  (x) = ± √ - k ν(x) f (x) + (T f )(x) for σ-a.e. x ∈ ∂Ω. (2.5.84)
∂Ω± 2 −1
Proof All claims about the nature of the set Ω− have been already established in
[112, Lemma 5.10.9]. Granted these, Theorem 2.5.1 applies both to Ω+ and Ω− ,
a scenario in which the jump-formula (2.5.4) yields (2.5.84), thanks to (2.5.10)
(keeping in mind that the geometric measure theoretic outward unit normal to Ω− is
−ν at σ-a.e. point on ∂Ω). 

The proof of Theorem 2.5.1 will be presented later on, as a corollary of the more
general result recorded in Theorem 2.5.38 where we deal with variable coefficient
kernels. In turn, Theorem 2.5.38 is proved by first treating in Proposition 2.5.37
kernels of a special algebraic structure, i.e., functions of the form

P(x)
k(x) = , for each x ∈ Rn \ {0},
|x| n−1+deg P (2.5.85)
with P homogeneous, odd, harmonic polynomial in Rn,

which then permits us to deal with arbitrary kernels as in (2.3.3) via a spherical
harmonic expansion. Next, treating the case of kernels as in (2.5.85) is done via
induction on deg P, the degree of the homogeneous, odd, harmonic polynomial P in
Rn . The initial step in this induction involves polynomials of the form P(x) = x j , with
j ∈ {1, . . . , n}. Up to normalization, these produce Riesz transforms in the context
of (2.5.85) which are treated in Proposition 2.5.35. As a preamble to the proof of this
latter result, in Propositions 2.5.21-2.5.27 we deal with the harmonic double layer
and other related operators (whose integral kernels are obtained by taking certain
“tangential” derivatives on the fundamental solution for the Laplacian). Finally, the
proofs of Propositions 2.5.21-2.5.27 require certain auxiliary results, presented in
Lemmas 2.5.8-2.5.20.
The result in our lemma next refines and strengthens [63, Proposition 3.3, p. 2628]
and [120, Lemma 2.5]. The main conclusion (see (2.5.88) below) may be interpreted
as saying that, for a given set of locally finite perimeter E, the surface measure of
the portion contained in E of any sphere with center on ∂ ∗ E is roughly the area of a
half-sphere, and the degree of accuracy improves indefinitely as the radius shrinks
to zero trough a suitable set of admissible values. For this reason we shall refer to a
result of this type as a “half-sphere lemma.”

Lemma 2.5.8 Let E ⊆ Rn be a set of locally finite perimeter. Then for each x ∈ ∂ ∗ E
there exists a Lebesgue measurable set Ox ⊂ (0, 1) of density 1 at 0, i.e., satisfying

L 1 Ox ∩ (0, ε)
lim = 1, (2.5.86)
ε→0+ ε
with the property that
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 425

the sets E ∩ ∂B(x, r) and ∂B(x, r) \ E

(2.5.87)
are H n−1 -measurable for L 1 -a.e. r ∈ Ox

and, with ωn−1 denoting the surface area of the unit sphere in Rn ,

H n−1 E ∩ ∂B(x, r) 1
lim = , (2.5.88)
O x r→0 ωn−1 r n−1 2

H n−1 ∂B(x, r) \ E 1
lim = . (2.5.89)
O x r→0 ωn−1 r n−1 2
Proof Having fixed some x ∈ ∂ ∗ E and R ∈ (0, ∞) arbitrary, invoke [112, (5.3.7)
in Proposition 5.3.2] with m := 1, A := B(x, R), g := 1B(x,R)∩E , and the Lipschitz
function

f : Rn −→ R given by f (z) := |z − x| for each z ∈ Rn . (2.5.90)

Bearing in mind that R > 0 is arbitrary, this eventually proves that

E ∩ ∂B(x, r) is H n−1 -measurable for L 1 -a.e. r ∈ (0, ∞). (2.5.91)

Moving on, fix x ∈ ∂ ∗ E and define the half-spaces

Hx± := {y ∈ Rn : ±ν(x) · (y − x) > 0}. (2.5.92)

For each number r > 0, set ∂ ± B(x, r) := ∂B(x, r) ∩ Hx± and introduce

W(x, r) := ∂ − B(x, r) E ∩ ∂B(x, r)

= ∂B(x, r) ∩ EHx− (2.5.93)

where, generally speaking, UV := (U\V)∪(V \U) denotes the symmetric difference
of any two given sets U, V. For an arbitrary number R ∈ (0, ∞), the idea is now to
invoke formula [112, (5.3.7) in Proposition 5.3.2] with m := 1, A := B(x, R),
the Lipschitz function f from  (2.5.90), and g := 1B(x,R)∩(E Hx ) . In particular,
−

these choices entail A ∩ f −1 {r } =  whenever the number r ∈ R \ [0, R], and

(D f )(z) = (z − x)/|z − x| which implies J f = 1 at every point in Rn \ {x}. From
[112, Proposition 5.3.2] we then conclude that

the set W(x, r) is H n−1 -measurable for L 1 -a.e. r ∈ (0, R) (2.5.94)

and
∫ R  
H n−1 W(x, r) dr = L n B(x, R) ∩ (EHx− ) (2.5.95)
0
   
= L n B(x, R) ∩ E ∩ Hx+ + L n B(x, R) \ E ∩ Hx− .
426 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

In turn, from (2.5.95) and [112, (5.6.57)-(5.6.58)] we conclude that

∫ R

H n−1 W(x, r) dr = o(Rn ) as R → 0+ . (2.5.96)
0

To proceed, consider the function

φ : (0, 1) −→ [0, ∞) given by

 (2.5.97)
φ(r) := r 1−n H n−1 W(x, r) for L 1 -a.e. r ∈ (0, 1).

It follows from (2.5.96) that

∫ R  R  1−n ∫ R 
φ(r) dr ≤ H n−1 W(x, r) dr = o(R) as R → 0+ . (2.5.98)
R/2 2 0

Bring in the dyadic intervals Ik := 2−(k+1), 2−k for k ∈ N0 and note that (2.5.98)
entails ⨏
δk := φ(r) dr −→ 0+ as k → ∞. (2.5.99)
Ik

For each k ∈ N0 split

 1 
Ik = Ak ∪ Bk , with Bk := r ∈ Ik : φ(r) > δk and Ak := Ik \ Bk . (2.5.100)

Then Chebytcheff’s inequality permits us to estimate

⨏ 1
L 1 (Bk ) 1
≤ √ φ(r) dr = δk , ∀k ∈ N0 . (2.5.101)
L (Ik )
1
δk Ik
In light of (2.5.99), this implies that if we now define

Ox := Ak ⊂ (0, 1), (2.5.102)

k ∈N0

then
lim φ(r) = 0, (2.5.103)
O x r→0

hence 
H n−1 W(x, r)
lim = 0. (2.5.104)
O x r→0 r n−1
We claim that (2.5.86) also holds for this choice of Ox . To see that this is the case,
assume some arbitrary θ > 0 has been fixed. For each ε ∈ (0, 1), let Nε ∈ N0 be
such that 2−Nε −1 < ε ≤ 2−Nε . Since Nε → ∞ as ε → 0+ , it follows from (2.5.99)
that there exists εθ > 0 with the property that

δk ≤ θ 2 whenever 0 < ε < εθ and k ≥ Nε . (2.5.105)

2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 427

Assuming that 0 < ε < εθ we may then estimate


ε − L 1 Ox ∩ (0, ε) 
0≤ = ε −1 L 1 (0, ε) \ Ox
ε
  ∞
≤ ε −1 L 1 (0, 2−Nε ) \ Ox = ε −1 L 1 (Bk )
k=Nε


∞ 1
≤ ε −1 L 1 (Ik ) δk ≤ ε −1 θ 2−Nε ≤ θ/2. (2.5.106)
k=Nε

This finishes the proof of (2.5.86).

At this stage there remains to observe that since, generally speaking,
 n−1 
H (U) − H n−1 (V) ≤ H n−1 (UΔV) (2.5.107)

for any H n−1 -measurable sets U, V ⊆ Rn of finite H n−1 -measure, from (2.5.91),
(2.5.107), and (2.5.93) we obtain
     
 H n−1  E ∩ ∂B(x, r) 1  H n−1 E ∩ ∂B(x, r) − H n−1 ∂ − B(x, r) 

 − =
 ωn−1 r n−1 2 ωn−1 r n−1

H n−1 W(x, r)
≤ for L 1 -a.e. r ∈ (0, 1). (2.5.108)
ωn−1 r n−1
Then formula (2.5.88) is a consequence of this and (2.5.104). Moreover, (2.5.89) is
readily implied by (2.5.88) and (2.5.91). Finally, taking (2.5.91) into account and
eventually adjusting Ox by an L 1 -nullset, we may also ensure that (2.5.87) holds. 

It turns out that the “half-sphere lemma” proved earlier for arbitrary sets of locally
finite perimeter (cf. Lemma 2.5.8) has a version in which the radius approaches zero
unrestrictedly, except for an L 1 -nullset. En route to Lemma 2.5.12, where such a
version is formulated, we first establish several pointwise convergence results.

Lemma 2.5.9 Let Ω ⊆ Rn be a set of locally finite perimeter whose topological

boundary, ∂Ω, is compact. Abbreviate σ∗ := H n−1 ∂∗ Ω and denote by ν the geo-
metric measure theoretic outward unit normal to Ω. Make the assumption that either
Ω is bounded or n ≥ 2. Also, consider a vector-valued function

k! = (k j )1≤ j ≤n ∈ 𝒞2 (Rn \ {0})

n
which is
odd, positive homogeneous of degree 1 − n, (2.5.109)
and which satisfies div k! = 0 in Rn \ {0}.

Finally, set
428 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
% &
ϑ := !
ω, k(ω) dH n−1 (ω) ∈ C. (2.5.110)
S n−1

Then for σ∗ -a.e. point x ∈ ∂∗ Ω there holds

⎧
⎪ ϑ
⎪
∫
% & ⎨ − 2 if Ω is bounded,
⎪
⎪
lim ! − y) dσ∗ (y) =
ν(y), k(x (2.5.111)
ε→0+ ⎪
⎪ ϑ
y ∈∂∗ Ω ⎪
⎪ + if Ω is unbounded.
|x−y |>ε ⎩ 2

In particular, for σ∗ -a.e. point x ∈ ∂∗ Ω one has

∫ 
1 ν(y), y − x + 12 if Ω is bounded,
lim dσ∗ (y) = (2.5.112)
ε→0+ ωn−1 |x − y| n − 12 if Ω is unbounded.
y ∈∂∗ Ω
|x−y |>ε

Proof Let us first work under the assumption that Ω ⊆ Rn is a bounded set of
locally finite perimeter. Pick some x ∈ Rn and, for each  ∈ N, select a scalar
function ψ ∈ 𝒞∞ (Rn ) satisfying
 
ψ ≡ 0 in B x, 1/(4) and ψ ≡ 1 in Rn \ B x, 1/(2) . (2.5.113)

The idea is to use [112, Lemma 5.7.2] for the bounded set E := Ω and the vector
field

F! ∈ 𝒞2 (Rn ) ,
n ! − y) for all y ∈ Rn .
F! (y) := ψ (y) k(x (2.5.114)

Given that the vector field F! is divergence-free near Ω \ B(x, 1/) (cf. the last line
in (2.5.109)), this lemma implies that there exists an L 1 -nullset Nx, ⊂ (0, ∞) such
that for each ε ∈ (1/, ∞) \ Nx, the set Ω ∩ ∂B(x, ε) is H n−1 -measurable and we
have
∫ ∫
% &
! − y) dσ∗ (y) =
ν(y), k(x ν(y) · F! (y) dσ∗ (y)
(∂∗ Ω)\B(x,ε)
|x−y |>ε
y ∈∂∗ Ω
∫ y − x
= · F! (y) dH n−1 (y)
Ω∩∂B(x,ε) ε
∫

= (y − x) j /ε k j (x − y) dH n−1 (y).
|x−y |=ε
y ∈Ω
(2.5.115)

For further use, introduce

2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 429

Nx := Nx, (2.5.116)
 ∈N

and note that Nx ⊂ (0, ∞) is an L 1 -nullset. In particular, [112, Lemma 5.7.2] also
implies (via the same type of argument as in (2.5.115)) that for each ε ∈ (0, ∞) \ Nx
we have
∫ ∫
% & 
!
ν(y), k(x − y) dσ∗ (y) = (y − x) j /ε k j (x − y) dH n−1 (y).
|x−y | ≥ε |x−y |=ε
y ∈∂∗ Ω y ∈Ω
(2.5.117)

Assume next that in fact x ∈ ∂ ∗ Ω and, for each r ∈ (0, ∞), abbreviate

∂ − B(x, r) := ∂B(x, r) ∩ Hx− where

(2.5.118)
Hx− := {y ∈ Rn : ν(x) · (y − x) < 0}.

For each r ∈ (0, ∞) let us also introduce

W(x, r) := ∂ − B(x, r) Ω ∩ ∂B(x, r) = ∂B(x, r) ∩ ΩHx− (2.5.119)

where, as in the past,  denotes the set-theoretic symmetric difference operator.

Since, generally speaking,
∫ ∫  ∫
 
 f dH n−1 − f dH n−1  ≤ | f | dH n−1 (2.5.120)
A B AB

for any two H n−1 -measurable sets A, B ⊆ Rn and any function f ∈ L 1 (A∪ B, H n−1 ),
on the one hand we obtain, for each ε ∈ (0, ∞) \ Nx ,
 ∫
 

 (y − x) j /ε k j (x − y) dH n−1 (y)

Ω∩∂B(x,ε)
∫ 
 

− (y − x) j /ε k j (x − y) dH n−1 (y)

∂− B(x,ε)

H n−1 W(x, ε)
≤ · sup | k! |. (2.5.121)
ε n−1 S n−1

Recall that k! is odd and positive homogeneous of degree 1 − n (cf. (2.5.109)). Fix
ε ∈ (0, ∞) \ Nx . Making the changes of variables ω := ±(y − x)/ε ∈ S n−1 yields,
respectively,
430 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫ ∫
 % &
(y − x) j /ε k j (x − y) dH n−1 (y) = − !
ω, k(ω) dH n−1 (y)
∂− B(x,ε) ω ∈S n−1
ν(x)·ω<0
(2.5.122)

and
∫ ∫
 % &
(y − x) j /ε k j (x − y) dH n−1
(y) = − !
ω, k(ω) dH n−1 (y).
∂− B(x,ε) ω ∈S n−1
ν(x)·ω>0
(2.5.123)
From (2.5.122)-(2.5.123) and (2.5.110) we then conclude that
∫
 ϑ
(y − x) j /ε k j (x − y) dH n−1 (y) = − (2.5.124)
2
∂− B(x,ε)

for each ε ∈ (0, ∞) \ Nx .

In addition, from (2.5.104) we know that there exists a Lebesgue measurable set
Ox ⊂ (0, 1) of density 1 at 0, i.e., satisfying (2.5.86), with the property that the set
Ω ∩ ∂B(x, ε) happens to be H n−1 -measurable for L 1 -a.e. ε ∈ Ox and

H n−1 W(x, ε)
lim = 0. (2.5.125)
O x ε→0 ε n−1
From (2.5.121), (2.5.124), and (2.5.125) we then obtain
∫
 ϑ
lim (y − x) j /ε k j (x − y) dH n−1 (y) = − . (2.5.126)
O x \N x ε→0 2
|x−y |=ε
y ∈Ω

In turn, from (2.5.115) and (2.5.126) we conclude that, on the one hand,
∫
% &
lim ! − y) dσ∗ (y) = − ϑ for each x ∈ ∂ ∗ Ω. (2.5.127)
ν(y), k(x
O x \N x ε→0 2
y ∈∂∗ Ω
|x−y |>ε

On the other hand, [112, Proposition 5.6.7] is presently applicable (with f taken to be
the scalar components of the geometric measure theoretic outward unit normal to Ω;
indeed, given that the set ∂Ω is bounded, the finite perimeter assumption ensures that
σ∗ (∂∗ Ω) = H n−1 (∂∗ Ω) < +∞). As such, [112, (5.6.45)] may be used to conclude
that the limit
∫
% &
lim+ ! − y) dσ∗ (y) exists for H n−1 -a.e. x ∈ ∂∗ Ω.
ν(y), k(x (2.5.128)
ε→0
y ∈∂∗ Ω
|x−y |>ε
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 431

Having established (2.5.127) and (2.5.128), the version of (2.5.111) corresponding

to the case when Ω is bounded follows, on account of [112, (5.6.21)].
Next, the version of (2.5.111) corresponding to the case when the set Ω is un-
bounded and n ≥ 2 becomes a consequence of what we have proved so far, applied to
Ωc := Rn \ Ω. Indeed, since n ≥ 2 and ∂Ω is compact, from [112, Lemma 5.10.10],
[112, (5.2.3)], [112, (5.6.16)], and assumptions, we conclude that Ωc := Rn \ Ω is a
bounded set of locally finite perimeter whose geometric measure theoretic outward
unit normal is −ν at each point on ∂ ∗ (Ωc ) = ∂ ∗ Ω. This completes the proof of
(2.5.111).
Finally, specializing (2.5.111) to the case when

! 1 x
k(x) := − for all x ∈ Rn \ {0}, (2.5.129)
ωn−1 |x| n
yields (2.5.112). 
For further use, we isolate a technical result in the following lemma.
Lemma 2.5.10 Let Ω ⊆ Rn be a set of locally finite perimeter and consider a vector-
valued function k! ∈ 𝒞2 (Rn \ {0}) which is odd, positive homogeneous of degree
n

1 − n, and satisfies div k! = 0 in Rn \ {0}. Also, abbreviate

∫
% &
ϑ := !
ω, k(ω) dH n−1 (ω) ∈ C. (2.5.130)
S n−1

Then for H n−1 -a.e. point x ∈ ∂∗ Ω there exists some L 1 -nullset Nx ⊂ (0, ∞) with
the property that for each ε ∈ (0, ∞) \ Nx the set Ω ∩ ∂B(x, ε) is H n−1 -measurable
and
∫ #y − x $
lim ! − y) dH n−1 (y) = − ϑ .
, k(x (2.5.131)
ε ∈(0,∞)\N x Ω∩∂B(x,ε) ε 2
ε→0

Proof Assume first that Ω is bounded. As in the past, abbreviate σ∗ := H n−1 ∂∗ Ω.
From (2.5.115) and (2.5.117) we know that if for each x ∈ Rn we define Nx as
in (2.5.116) then Nx ⊂ (0, ∞) is an L 1 -nullset with the property that for each
ε ∈ (0, ∞) \ Nx the set Ω ∩ ∂B(x, ε) is H n−1 -measurable and
∫ #y − x $ ∫
% &
! − y) dH n−1 (y) =
, k(x ! − y) dσ∗ (y)
ν(y), k(x
ε
Ω∩∂B(x,ε) |x−y |>ε
y ∈∂∗ Ω
∫
% &
= ! − y) dσ∗ (y).
ν(y), k(x
|x−y | ≥ε
y ∈∂∗ Ω
(2.5.132)

Then (2.5.131) in the case when Ω is bounded is seen from (2.5.132) and (2.5.111).
432 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

To treat the case when Ω is not necessarily bounded we reason as follows. From
[112, Lemma 5.7.3] we know that there exists some L 1 -nullset N0 ⊂ (0, ∞) such that
for each r ∈ (0, ∞) \ N0 the set Ωr := Ω ∩ B(0, r) is bounded, has finite perimeter,
and
 
∂∗ Ωr ≡ Ω ∩ ∂B(0, r) ∪ ∂∗ Ω ∩ B(0, r) modulo H n−1 . (2.5.133)

Select a sequence {r j } j ∈N ⊆ (2, ∞) \ N0 with r j  ∞ as j → ∞. From what we

have established in the first part of the current proof and (2.5.133) we conclude that,
given any integer j ∈ N, there exists an H n−1 -nullset A j such that for each point
x ∈ ∂∗ Ω ∩ B(0, r j ) \ A j we may find some L 1 -nullset Nx, j ⊂ (0, ∞) with the
property that for each ε ∈ (0, ∞) \ Nx, j the set Ωr j ∩ ∂B(x, ε) is H n−1 -measurable
and
∫ #y − x $
lim ! − y) dH n−1 (y) = − ϑ .
, k(x (2.5.134)
ε ∈(0,∞)\N x, j Ωr j ∩∂B(x,ε) ε 2
ε→0

Moving on, introduce

A := Aj (2.5.135)
j ∈N

which continues to be an H n−1 -nullset. In addition, for each x ∈ ∂∗ Ω \ A define

Nx := Nx, j (2.5.136)
j ∈N: |x |<r j

so Nx is a meaningfully define L 1 -nullset subset of (0, ∞). Let us also observe that

Ωr j ∩ ∂B(x, ε) = Ω ∩ ∂B(x, ε)
(2.5.137)
for each x ∈ B(0, r j /2) and ε ∈ (0, r j /2).

Combining (2.5.134)-(2.5.137) then proves the following result: Given any j ∈ N

and any x ∈ ∂∗ Ω ∩ B(0, r j /2) \ A, whenever ε ∈ (0, 1) \ Nx the set Ω ∩ ∂B(x, ε)
is H n−1 -measurable and
∫ #y − x $
lim ! − y) dH n−1 (y) = − ϑ .
, k(x (2.5.138)
ε ∈(0,1)\N x Ω∩∂B(x,ε) ε 2
ε→0

Since r j  ∞ as j → ∞, this ultimately shows that for any x ∈ ∂∗ Ω \ A and any

ε ∈ (0, 1) \ Nx the set Ω ∩ ∂B(x, ε) is H n−1 -measurable and (2.5.138) holds. From
this, the claim made in (2.5.131) readily follows. 

Particularizing (2.5.132) yields a number of significant results.

2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 433

Remark 2.5.11 Throughout, assume Ω ⊆ Rn is a bounded set of locally finite

perimeter, abbreviate σ∗ := H n−1 ∂∗ Ω, and denote by ν = (ν1, . . . , νn ) the geometric
measure theoretic outward unit normal to Ω.
(a) When k! is as in (2.5.129) we obtain the following result: For each x ∈ Rn
there exists an L 1 -nullset Nx ⊂ (0, ∞) such that, for each r ∈ (0, ∞) \ Nx ,
∫ ∫
ν(y), y − x ν(y), y − x
dσ∗ (y) = dσ∗ (y)
|x − y| n |x − y| n
y ∈∂∗ Ω y ∈∂∗ Ω
|x−y |>r |x−y | ≥r

H n−1 Ω ∩ ∂B(x, r)
= . (2.5.139)
r n−1
(b) Fix j, k ∈ {1, . . . , n} and specialize (2.5.132) to the case when

k! := (∂k EΔ )e j − (∂j EΔ )ek , (2.5.140)

where EΔ is the standard fundamental solution for the Laplacian in Rn defined in

(1.5.56). This allows us to conclude that for each x ∈ Rn there exists an L 1 -nullset
Nx ⊂ (0, ∞) such that, for each r ∈ (0, ∞) \ Nx ,
∫  
ν j (y)(∂k EΔ )(y − x) − νk (y)(∂j EΔ )(y − x) dσ∗ (y) = 0 and
|x−y |>r
∫
y ∈∂∗ Ω
  (2.5.141)
ν j (y)(∂k EΔ )(y − x) − νk (y)(∂j EΔ )(y − x) dσ∗ (y) = 0.
|x−y | ≥r
y ∈∂∗ Ω

Here is the version of the “half-sphere lemma” mentioned earlier, valid in the
class of sets of locally finite perimeter, refining Lemma 2.5.8.
Lemma 2.5.12 Assume Ω ⊆ Rn is a set of locally finite perimeter and abbreviate
σ := H n−1 ∂Ω. Then for σ-a.e. x ∈ ∂∗ Ω there exists some L 1 -nullset Nx ⊂ (0, ∞)
with the property that

the sets Ω ∩ ∂B(x, ε) and ∂B(x, ε) \ Ω

(2.5.142)
are H n−1 -measurable for each ε ∈ (0, ∞) \ Nx

and such that (with ωn−1 denoting the surface area of the unit sphere in Rn )

H n−1 Ω ∩ ∂B(x, ε) 1
lim = , (2.5.143)
ε ∈(0,∞)\N x ωn−1 ε n−1 2
ε→0

H n−1 ∂B(x, ε) \ Ω 1
lim = . (2.5.144)
ε ∈(0,∞)\N x ωn−1 ε n−1 2
ε→0
434 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Proof Much as in the proof of Lemma 2.5.8, the claim culminating in (2.5.142) is
a consequence of the co-area theorem. Next, specializing (2.5.131) to the case when
k! is as in (2.5.129) yields (2.5.143). Finally, (2.5.144) is an immediate consequence
of (2.5.143). 

In the lemma below we prove that a certain maximal operator on the boundary of a
Lebesgue measurable set Ω maps Hölder functions into bounded functions provided
∂Ω is compact and upper Ahlfors regular.

Lemma 2.5.13 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a Lebesgue measurable set such

that ∂Ω is bounded and ∂∗ Ω is upper Ahlfors regular. Abbreviate σ∗ := H n−1 ∂∗ Ω
and denote by ν the geometric measure theoretic outward unit normal to Ω. Also,
consider a divergence-free vector-valued function k! ∈ 𝒞1 (Rn \ {0}) with the
n

property that there exists some C ∈ (0, ∞) such that

!
| k(x)| ≤ C|x| 1−n for all x ∈ Rn \ {0}. (2.5.145)

In this setting, define the action of the maximal operator Zmax on an arbitrary
function f ∈ L 1 (∂∗ Ω, σ∗ ) according to
 
 
 ∫ 
 % & 
 ! 
(Zmax f )(x) := sup  ν(y), k(x − y) f (y) dσ∗ (y) (2.5.146)
ε>0  
 y ∈∂∗ Ω 
 |x−y |>ε 

for all x ∈ ∂Ω.

Then for each given exponent α > 0 there exists some number N ∈ (0, ∞),
! n, α, and the upper Ahlfors regularity constant of ∂∗ Ω, with
depending only on k,
the property that

·  f 𝒞. α (∂ Ω) + N · sup | f |,
α
sup (Zmax f )(x) ≤ N diam(∂∗ Ω) (2.5.147)
∗
x ∈∂∗ Ω ∂∗ Ω

for each function f ∈ 𝒞α (∂∗ Ω).

Proof The present assumptions imply that Ω is a set of locally finite perimeter (cf.
[112, (5.9.16)]), so the geometric measure theoretic outward unit normal ν to Ω is
well defined at σ∗ -a.e. point on ∂∗ Ω. In a first stage, make the additional assumption
that Ω is bounded. Fix an arbitrary point x ∈ Rn and observe that (2.5.132) implies
that for L 1 -a.e. ε > 0 we have
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 435
∫ % & 
 ! − y) dσ∗ (y)
 ν(y), k(x
∂∗ Ω\B(x, ε)
 
 ∫ 
 #y − x $ 
 ! − y) dH n−1 (y)
= , k(x
 ε 
 Ω∩∂B(x,ε) 

H n−1 ∂B(x, ε)
≤C·
ε n−1
= C · ωn−1 . (2.5.148)

Let us now pick an arbitrary r ∈ (0, ∞). Based on (2.5.148) we conclude that there
exists ε ∈ (r/2, r) such that
∫ % & 
 ! − y) dσ∗ (y) ≤ C · ωn−1 .
 ν(y), k(x (2.5.149)
∂∗ Ω\B(x, ε)

For this choice of ε we then proceed to estimate

∫ % & 
 ! − y) dσ∗ (y)
 ν(y), k(x
∂∗ Ω\B(x, r)
∫ % & 
 ! − y) dσ∗ (y)
≤ ν(y), k(x
∂∗ Ω\B(x, ε)
∫ % & 
 ! − y) dσ∗ (y)
+ ν(y), k(x
[B(x, r)\B(x, ε) ]∩∂∗ Ω
 ∫ 
1
≤ C ωn−1 + dH n−1 (y)
[B(x, r)\B(x, ε) ]∩∂∗ Ω |x − y| n−1
 ∫ 
1
≤ C ωn−1 + dH n−1 (y)
[B(x, 2ε)\B(x, ε) ]∩∂∗ Ω |x − y| n−1
  
≤ C ωn−1 + ε −(n−1) H n−1 B(x, 2ε) ∩ ∂∗ Ω

! n) < +∞.
≤ C = C(∂∗ Ω, k, (2.5.150)

The argument so far shows that

∫ % & 
 ! − y) dσ∗ (y) ≤ C,
sup sup  ν(y), k(x (2.5.151)
x ∈R n r ∈(0,∞) ∂∗ Ω\B(x, r)

! the dimension n, and the upper Ahlfors

where C ∈ (0, ∞) depends only on k,
regularity constant of ∂∗ Ω.
436 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

To proceed, select f ∈ 𝒞α (∂∗ Ω) with α > 0, and fix x ∈ ∂∗ Ω along with r > 0

arbitrary. Split
∫
% &
! − y) f (y) dσ∗ (y) = I + II,
ν(y), k(x (2.5.152)
∂∗ Ω\B(x, r)

where
∫
% &
I := ! − y) [ f (y) − f (x)] dσ∗ (y)
ν(y), k(x (2.5.153)
∂∗ Ω\B(x, r)

and
∫
% &
II := f (x) ! − y) dσ∗ (y).
ν(y), k(x (2.5.154)
∂∗ Ω\B(x, r)

To estimate I, we use (2.5.145), (2.5.153), the fact that σ∗ = H n−1 ∂∗ Ω, and [112,
Lemma 7.2.1] to write
∫
1
|I| ≤ C · | f (y) − f (x)| dσ∗ (y)
∂∗ Ω |x − y| n−1
∫
1
≤ C ·  f 𝒞. α (∂ Ω) dH n−1 (y)
∗ |x − y| n−1−α
∂∗ Ω
α
≤ C · diam(∂∗ Ω) ·  f 𝒞. α (∂ Ω), (2.5.155)
∗

! n, α) ∈ (0, ∞) independent of f . Also, (2.5.154) and

for some C = C(∂∗ Ω, k,
(2.5.151) imply

|II| ≤ C · sup | f |. (2.5.156)

∂∗ Ω

Together, (2.5.146), (2.5.152), (2.5.155), and (2.5.156) establish (2.5.147) in the case
when Ω is bounded.
Finally, the case when the set Ω unbounded is a consequence of what we have
proved so far, applied to Ωc := Rn \ Ω. Indeed, since n ≥ 2 and ∂Ω is compact, from
[112, Lemma 5.10.10], [112, (5.2.3)], [112, (5.6.16)], and assumptions, we conclude
that Ωc is a bounded set of locally finite perimeter whose geometric measure theoretic
outward unit normal is −ν at H n−1 -a.e. point on ∂∗ (Ωc ) = ∂∗ Ω. 

As a preamble to Proposition 2.5.16, in the next couple of lemmas we study the

boundary-to-domain version of the integral operator with a kernel as in (2.5.146).

Lemma 2.5.14 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set of locally finite

perimeter with the property that ∂Ω is bounded. Denote by ν the geometric measure
theoretic outward unit normal to Ω and abbreviate σ∗ := H n−1 ∂∗ Ω. Next, bring
in a divergence-free vector-valued function k! ∈ 𝒞1 (Rn \ {0}) which is positive
n
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 437

homogeneous of degree 1 − n. Use this to define the integral operator acting on each
function f ∈ L 1 (∂∗ Ω, σ∗ ) according to
∫
% &
Z f (x) := ! − y) f (y) dσ∗ (y) for all x ∈ Ω.
ν(y), k(x (2.5.157)
∂∗ Ω
∫ % &
Then, with ϑ := !
ω, k(ω) dH n−1 (ω) ∈ C, one has
S n−1

−ϑ in Ω, if Ω is bounded,
Z1 ≡ (2.5.158)
0 in Ω, if Ω is unbounded.

Proof In a first stage, assume that Ω is bounded and fix an arbitrary point x ∈ Ω.

The idea is to apply [112, Corollary 2.8.8] to a vector field F! ∈ ℰ(Rn ) whose
n

𝒞1 -singular support is a compact subset of Ω and which coincides near the compact
set Ω with
! − y) for L n -a.e. y ∈ Ω.
k(x (2.5.159)
 
Since from [112, (4.5.46)] we know that divF! Ω = −ϑδx where δx is the Dirac
distribution with mass at x in Ω, the Divergence Formula [112, (2.8.57)] reduces
precisely to (Z1)(x) = 1. As such, the version of (2.5.158) when Ω is bounded
follows on account of the arbitrariness of x ∈ Ω.
In the case when the set Ω is unbounded and ∂Ω is bounded, [112, Lemma 5.10.10]
and [112, (5.2.3), (5.6.16)] imply that Ωc := Rn \ Ω is a compact set of locally finite
perimeter, satisfying ∂∗ (Ωc ) = ∂∗ Ω, ∂ ∗ (Ωc ) = ∂ ∗ Ω, and whose geometric measure
theoretic outward unit normal is −ν. Having fixed an arbitrary x ∈ Ω, apply the De
Giorgi-Federer version of the Divergence Theorem (see [112, Theorem 1.1.1]) to the
set Ωc and a vector field F! ∈ 𝒞1c (Rn ) which coincides in an open neighborhood
n

O ⊆ R \ {x} of R \ Ω with
n n

! − y) ∈ Cn .
O  y −→ k(x (2.5.160)
 
In view of the fact that divF! Ωc = 0, the Divergence Formula [112, (1.1.8)] now
simply reads (Z1)(x) = 0, finishing the proof of (2.5.158). 

Here is the second auxiliary lemma alluded to earlier.

Lemma 2.5.15 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set such that ∂Ω

is bounded and ∂∗ Ω is upper Ahlfors regular (in particular Ω has locally finite
perimeter; cf. [112, (5.9.16)]). Consider a divergence-free vector-valued function
k! ∈ 𝒞1 (Rn \ {0}) which is positive homogeneous of degree 1 − n and use it to
n

define the integral operator Z as in (2.5.157). Finally, fix an arbitrary exponent

α ∈ (0, 1).
Then there exists some C ∈ (0, ∞), depending only on k,! n, α, diam(∂∗ Ω), and the
upper Ahlfors regularity constant of ∂∗ Ω, such that for every function f ∈ 𝒞α (∂∗ Ω)
one has
438 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
    
sup (Z f )(x) + sup dist(x, ∂∗ Ω)1−α ∇(Z f )(x) ≤ C f 𝒞α (∂∗ Ω) . (2.5.161)
x ∈Ω x ∈Ω

Moreover, under the additional assumption that Ω is a uniform domain, the

operator 
Z : 𝒞α (∂∗ Ω) −→ 𝒞α Ω (2.5.162)
is well defined, linear, and bounded; hence so is

Z : 𝒞α (∂Ω) −→ 𝒞α Ω . (2.5.163)

Proof The estimate in (2.5.161) is a consequence of (2.5.158) and Proposition 2.1.3.

In particular, since ∂∗ Ω ⊆ ∂Ω, this further entails
    
sup (Z f )(x) + sup dist(x, ∂Ω)1−α ∇(Z f )(x) ≤ C f 𝒞α (∂∗ Ω), (2.5.164)
x ∈Ω x ∈Ω

for each f ∈ 𝒞α (∂∗ Ω).

In the case when the set Ω ⊆ Rn is also assumed to
be a uniform domain, the fact that the operator Z in the context of (2.5.162) is
well defined, linear, and bounded, becomes a consequence of (2.5.164) and [112,
(5.11.78)] (keeping in mind that Z f ∈ 𝒞1 (Ω) for each function f ∈ 𝒞α (∂∗ Ω)). 
We are now in a position to prove the following basic result for singular integral
operators whose kernel is the “dot product” between the geometric measure theoretic
outward unit normal to a given Lebesgue measurable set with a compact upper
Ahlfors boundary boundary and a smooth, divergence-free, vector-valued function
which is odd and positive homogeneous of degree 1 − n in Rn \ {0}. One way to
think of this class of singular integral operators is as a natural generalization of the
standard harmonic double layer operator.

Proposition 2.5.16 Suppose Ω ⊆ Rn (where n ∈ N with n ≥ 2) is a Lebesgue

measurable set such that ∂Ω is compact and ∂∗ Ω is upper Ahlfors boundary. Denote
by ν the geometric measure theoretic outward unit normal to Ω and abbreviate
σ∗ := H n−1 ∂∗ Ω. Also, consider k! ∈ 𝒞2 (Rn \ {0}) a divergence-free vector-
n

valued function which is odd and positive homogeneous of degree 1 − n. Finally, fix

an exponent α ∈ (0, 1) and define
∫
% &
ϑ := !
ω, k(ω) dH n−1 (ω) ∈ C. (2.5.165)
S n−1

Then the following statements are true.

(i) For each function f ∈ 𝒞α (∂∗ Ω) the limit
∫
% &
Z f (x) := lim+ ! − y) f (y) dσ∗ (y)
ν(y), k(x (2.5.166)
ε→0
y ∈∂∗ Ω
|x−y |>ε

exists for σ∗ -a.e. point x ∈ ∂∗ Ω.

2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 439

(ii) After possibly redefining Z f on a σ∗ -nullset contained in ∂∗ Ω, the assignment

f → Z f thus defined induces a well-defined, linear, and bounded operator

Z : 𝒞α (∂∗ Ω) −→ 𝒞α (∂∗ Ω) (2.5.167)

with the property that


− ϑ2 on ∂∗ Ω, if Ω is bounded,
Z1 ≡ (2.5.168)
+ ϑ2 on ∂∗ Ω, if Ω is unbounded.

(iii) Under the additional assumption that the set Ω is open and ∂Ω is Ahlfors regular,
the following jump-formula (where Z is the integral operator introduced in
(2.5.157), I denotes the identity operator, and κ > 0 is an arbitrary fixed
aperture parameter) is valid:

for every given function f ∈ 𝒞α (∂∗ Ω) one has

 κ−n.t. (2.5.169)

Zf  = (− ϑ2 I + Z) f at σ-a.e. point on ∂∗ Ω.
∂Ω

(iv) In the case when it is also assumed that H n−1 (∂Ω \ ∂∗ Ω) = 0, one may replace
∂∗ Ω by ∂Ω in the formulation of (2.5.167), (2.5.168), and (2.5.169).

Proof To deal with the claim in part (i), pick an arbitrary function f ∈ 𝒞α (∂∗ Ω).
Then (2.5.111) gives that for σ∗ -a.e. point x ∈ ∂∗ Ω we have
∫
% &
lim+ ! − y) f (y) dσ∗ (y)
ν(y), k(x
ε→0
y ∈∂∗ Ω
|x−y |>ε
∫
% & ϑ
= lim+ ! − y)
ν(y), k(x f (y) − f (x) dσ∗ (y) ∓ f (x)
ε→0 2
y ∈∂∗ Ω
|x−y |>ε
∫
% & ϑ
= ! − y)
ν(y), k(x f (y) − f (x) dσ∗ (y) ∓ f (x), (2.5.170)
2
∂∗ Ω

where the sign of ϑ2 f (x) is minus if Ω is bounded and plus if Ω is unbounded.

For the last equality in (2.5.170) we have used Lebesgue’s Dominated Convergence
Theorem. More concretely, given that f (y)− f (x) = O(|x − y| α ), [112, Lemma 7.2.1]
shows that the last integrand above is absolutely integrable for each fixed x ∈ ∂∗ Ω.
In turn, (2.5.170) allows us to conclude that the limit defining (Z f )(x) in (2.5.166)
exists for σ∗ -a.e. x ∈ ∂∗ Ω. This takes care of the claim made in part (i).
As regards the claims in part (ii), we first observe from (2.5.166) and (2.5.170)
that for every given f ∈ 𝒞α (∂∗ Ω) the function Z f may be redefined on a σ∗ -nullset
so that
440 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
ϑ % &
(Z f )(x) = ∓ f (x) + ! − y)
ν(y), k(x f (y) − f (x) dσ∗ (y) (2.5.171)
2
∂∗ Ω

for all x ∈ ∂∗ Ω, with the sign dictated by whether Ω is bounded (minus), or Ω is

unbounded (plus). Then (2.5.171) readily gives (2.5.168). We next proceed to show
that, in the context of (2.5.167), the operator f → Z f defined as in (2.5.171) is well
defined and bounded. To this end, fix two arbitrary distinct points x1, x2 ∈ ∂∗ Ω and,
bearing (2.5.171) in mind, write

(Z f )(x1 ) − (Z f )(x2 ) = I + II (2.5.172)

where
ϑ
I := ∓ f (x1 ) − f (x2 ) (2.5.173)
2
and
∫ 
% &
II := ! 1 − y)
ν(y), k(x f (y) − f (x1 )
∂∗ Ω
% & 
! 2 − y) f (y) − f (x2 ) dσ∗ (y).
− ν(y), k(x (2.5.174)

Next, introduce r := |x1 − x2 | > 0 and estimate

 
II ≤ II1 + II2 + II3, (2.5.175)

where
 ∫
 % &
 ! 1 − y)
II1 :=  ν(y), k(x f (y) − f (x1 )

y ∈∂∗ Ω
|x1 −y |>2r

% & 
! 2 − y) 
− ν(y), k(x f (y) − f (x2 ) dσ∗ (y), (2.5.176)


while
∫ % 
 &
II2 := ! 1 − y) f (y) − f (x1 )  dσ∗ (y),
 ν(y), k(x (2.5.177)
y ∈∂∗ Ω
|x1 −y | ≤2r
∫ % 
 &
II3 := ! 2 − y) f (y) − f (x2 )  dσ∗ (y).
 ν(y), k(x (2.5.178)
y ∈∂∗ Ω
|x1 −y | ≤2r

Note that
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 441
  ∫
dH n−1 (y)
II2 ≤ sup | k! |  f 𝒞. α (∂ Ω) , (2.5.179)
S n−1
∗ |x1 − y| n−1−α
y ∈∂∗ Ω
|x1 −y | ≤2r

and, given that |x1 − y| ≤ 2r forces |x2 − y| ≤ |x1 − x2 | + |x1 − y| ≤ 3r,

∫ % 
 &
II3 ≤ ! 2 − y) f (y) − f (x2 )  dσ∗ (y)
 ν(y), k(x
y ∈∂∗ Ω
|x2 −y | ≤3r

  ∫
dH n−1 (y)
≤ sup | k! |  f 𝒞. α (∂ Ω) . (2.5.180)
S n−1
∗ |x2 − y| n−1−α
y ∈∂∗ Ω
|x2 −y | ≤3r

Invoking [112, Lemma 7.2.1] (cf. [112, (7.2.5)]), we obtain from (2.5.179) and
(2.5.180) (keeping in mind the significance of the number r) that there exists some
constant M = M(∂Ω, k, ! n, α) ∈ (0, ∞) with the property that

II2 + II3 ≤ M  f 𝒞. α (∂ Ω) |x1 − x2 | α . (2.5.181)

∗

Going further, bound

II1 ≤ II1a + II1b, (2.5.182)

with
 ∫ 
 % & 
 ! 1 − y) 
II1a :=  ν(y), k(x f (x2 ) − f (x1 ) dσ∗ (y)
 
y ∈∂∗ Ω
|x1 −y |>2r
 ∫ 
 % &  
 ! 1 − y) dσ∗ (y) f (x2 ) − f (x1 )
= ν(y), k(x
 
y ∈∂∗ Ω
|x1 −y |>2r
 
≤ (Zmax 1)(x1 ) f (x2 ) − f (x1 ) ≤ M r α  f 𝒞. α (∂ Ω), (2.5.183)
∗

where the last inequality is based on (2.5.147) (with f := 1), and

442 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 ∫ 
 % & 
 ! 1 − y) − k(x
! 2 − y) 
b
II1 :=  ν(y), k(x f (y) − f (x2 ) dσ∗ (y)
 
y ∈∂∗ Ω
|x1 −y |>2r
∫
 
≤ !k(x1 − y) − k(x
! 2 − y) | f (y) − f (x2 )| dσ∗ (y)
y ∈∂∗ Ω
|x1 −y |>2r
∫
dH n−1 (y)
≤ C r  f 𝒞. α (∂ Ω) , (2.5.184)
∗ |x1 − y| n−α
y ∈∂∗ Ω
|x1 −y |>2r

using the Mean Value Theorem and the fact that the function f is Hölder of order
α on ∂∗ Ω. Here it helps to note that if |x1 − y| > 2r then |ξ − y| ≈ |x1 − y|
for all ξ ∈ [x1, x2 ], and |y − x2 | < 3|y − x1 |/2. Combining (2.5.182), (2.5.183),
(2.5.184), and [112, (7.2.5)] we may then conclude that there exists some constant
M = M(∂Ω, k, ! n, α) ∈ (0, ∞) with the property that

II1 ≤ M  f 𝒞. α (∂ Ω) |x1 − x2 | α . (2.5.185)

∗

From (2.5.172), (2.5.173), (2.5.175), (2.5.181), and (2.5.185) we may then conclude
that
 
(Z f )(x1 ) − (Z f )(x2 ) ≤ M  f  . |x − x2 | α, ∀x1, x2 ∈ ∂∗ Ω, (2.5.186)
𝒞α (∂ Ω) 1 ∗

for some constant M = M(∂Ω, k, ! n, α) ∈ (0, ∞). The argument so far gives that
the principal-value singular integral operator Z is well defined and bounded in the
context . .
Z : 𝒞α (∂∗ Ω) −→ 𝒞α (∂∗ Ω). (2.5.187)
Let us also note that (2.5.171) and [112, Lemma 7.2.1] imply that for every function
f ∈ 𝒞α (∂∗ Ω) we have
∫
dσ∗ (y)
sup |(Z f )(x)| ≤ C sup | f (x)| + C f 𝒞. α (∂ Ω) sup
x ∈∂∗ Ω x ∈∂∗ Ω ∗
x ∈∂∗ Ω |x − y| n−1−α
∂∗ Ω

α
≤ C sup | f | + C f 𝒞. α (∂ Ω) · diam(∂∗ Ω) , (2.5.188)
∗
∂∗ Ω

for some C ∈ (0, ∞) which depends only on k,! n, α, and the upper Ahlfors regularity
constant of ∂∗ Ω. Together, (2.5.187) and (2.5.188) complete the proof of the claim
made about (2.5.167).
Turning our attention to the claim in part (iii), make the additional assumption
that the set Ω is open and that ∂Ω is Ahlfors regular. As far as the jump-formula
(2.5.169) is concerned, it has been already noted in Lemma 2.5.15 that the action
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 443

of the integral operator Z (cf. (2.5.157)) is presently meaningful on Hölder func-

tions. Also, given that ∂Ω is presently assumed to be Ahlfors regular, item (iii) in
[112, Proposition 8.8.6] ensures that it is meaningful to consider the nontangential
boundary trace in the left-hand side of the equality in (2.5.169) in the present setting.
Assume now that some f ∈ 𝒞α (∂∗ Ω) with α ∈ (0, 1) has been given and observe
from (2.5.157) that Z f is continuous in Ω.
Also, recall from the current item (i) that there exists a σ∗ -measurable set A ⊆ ∂∗ Ω
satisfying σ∗ (A) = 0 and with the property that the limit
∫
% &
lim+ ! − y) f (y) dσ∗ (y) exists for each x ∈ ∂∗ Ω \ A. (2.5.189)
ν(y), k(x
ε→0
|x−y |>ε
y ∈∂∗ Ω

Fix an arbitrary point


x ∈ ∂∗ Ω \ A ∩ ∂nta Ω (2.5.190)

and let Nx ⊂ (0, ∞) be the union between the L 1 -nullset associated with x as in
Lemma 2.5.10 (used with Ω replaced by Rn \ Ω) and the L 1 -nullset associated with
x as in [112, Lemma 5.7.2, (5.7.25)] (used with E := Rn \ Ω). In particular, for each
ε ∈ (0, ∞) \ Nx the set ∂B(x, ε) \ Ω is H n−1 -measurable and formula (2.5.131)
holds with Ω replaced by Rn \ Ω. For some arbitrary κ > 0 fixed, write

lim Z f (z)
z ∈Γκ (x)
z→x
∫
% &
= lim lim ! − y) f (y) dσ∗ (y)
ν(y), k(z
ε ∈(0,∞)\N x z ∈Γκ (x)
ε→0 z→x |x−y |>ε
y ∈∂∗ Ω
∫
% &
+ lim lim ! − y) ( f (y) − f (x)) dσ∗ (y)
ν(y), k(z
ε ∈(0,∞)\N x z ∈Γκ (x)
ε→0 z→x |x−y | ≤ε
y ∈∂∗ Ω

2 ∫ 5
3 % & 6
+ 33 lim lim ν(y), k(z − y) dσ∗ (y)66 f (x)
!
3ε ∈(0,∞)\N x z ∈Γκ (x) 6
ε→0 z→x |x−y | ≤ε
4 y ∈∂∗ Ω 7
=: I1 + I2 + I3 . (2.5.191)

For each fixed threshold ε > 0, Lebesgue’s Dominated Convergence Theorem applies
to the limit as Γκ (x)  z → x in I1 and yields
444 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
% &
I1 = lim ! − y) f (y) dσ∗ (y)
ν(y), k(x
ε ∈(0,∞)\N x
ε→0 |x−y |>ε
y ∈∂∗ Ω
∫
% &
= lim+ ! − y) f (y) dσ∗ (y) = Z f (x),
ν(y), k(x (2.5.192)
ε→0
|x−y |>ε
y ∈∂∗ Ω

where the second equality is a consequence of (2.5.189), and the last equality comes
from (2.5.166).
To handle I2 , we first observe that for every y ∈ ∂∗ Ω ⊆ ∂Ω and z ∈ Γκ (x) we may
estimate

|x − y| ≤ |z − y| + |z − x| ≤ |z − y| + (1 + κ) dist(z, ∂Ω)
≤ |z − y| + (1 + κ)|z − y| = (2 + κ)|z − y|. (2.5.193)

Hence, since f is Hölder of order α on ∂∗ Ω,

% &   (2 + κ)n−1
 ! − y)  | f (y) − f (x)| ≤ sup | k! |  f  . α
 ν(y), k(z 𝒞 (∂ Ω)
. (2.5.194)
S n−1
∗ |x − y| n−1−α

Based on this, [112, (7.2.5)], the upper Ahlfors regularity of ∂∗ Ω, and once again
Lebesgue’s Dominated Convergence Theorem, we obtain that
 ∫ 
   % & 

I2  =  lim !
ν(y), k(x − y) ( f (y) − f (x)) dσ∗ (y)
 ε ∈(0,∞)\N x 
ε→0 |x−y | ≤ε
y ∈∂∗ Ω

  ∫
dσ∗ (y)
≤ sup | k! | ·  f 𝒞. α (∂ Ω) lim sup
S n−1
∗
ε→0+ |x − y| n−1−α
|x−y | ≤ε
y ∈∂∗ Ω

≤ C ·  f 𝒞. α (∂ Ω) lim sup ε α = 0. (2.5.195)

∗
ε→0+

Thus,
I2 = 0. (2.5.196)

As for I3 in (2.5.191), for each fixed z ∈ Γκ (x) pick r ∈ 0, dist(z, ∂Ω) and
consider a scalar function ψ ∈ 𝒞∞ (Rn ) satisfying

ψ ≡ 0 in B(z, r/2) and ψ ≡ 1 in Rn \ B(z, r). (2.5.197)

The idea is now to use [112, Lemma 5.7.2] with E := Rn \ Ω and the vector field

F! ∈ 𝒞2 (Rn ) ,
n
F(y) ! − y) for all y ∈ Rn .
! := ψ(y) k(z (2.5.198)
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 445

Specifically, bearing in mind that F! is divergence-free near E (given that ψ ≡ 1 near

E), [112, Lemma 5.7.2] applied to this vector field implies (upon recalling that, by
design, Nx contains the L 1 -nullset associated as in [112, Lemma 5.7.2, (5.7.25)]
with the point x from (2.5.190) and the set E := Rn \ Ω) that for each ε ∈ (0, ∞) \ Nx
we have
∫
% &
! − y) dσ∗ (y)
ν(y), k(z
|x−y | ≤ε
y ∈∂∗ Ω
∫
% &
= ! − y) dH n−1 (y),
(y − x)/ε, k(z (2.5.199)
|x−y |=ε
y ∈R n \Ω

bearing in mind that the geometric measure theoretic outward unit normal to the set
E := Rn \ Ω is −ν. From (2.5.199) and (2.5.131) (used with Ω replaced by Rn \ Ω)
we therefore conclude that
∫
% &
lim lim ! − y) dσ∗ (y)
ν(y), k(z (2.5.200)
ε ∈(0,∞)\N x z ∈Γκ (x)
ε→0 z→x |x−y | ≤ε
y ∈∂∗ Ω
∫
% &
= lim ! − y) dH n−1 (y) = − ϑ .
(y − x)/ε, k(x
ε ∈(0,∞)\N x 2
ε→0 |x−y |=ε
y ∈R n \Ω

A combination of (2.5.191), (2.5.192), (2.5.196), and (2.5.200) shows that the limit
in the left-hand side of (2.5.191) exists and matches (− ϑ2 I + Z) f (x). In view of
(2.5.190) and [112, (8.8.52)], this proves that the formula in the second line of
(2.5.169) holds for each function f ∈ 𝒞α (∂∗ Ω) at σ∗ -a.e. point x ∈ ∂∗ Ω.
Finally, under the additional assumption that σ(∂Ω \ ∂∗ Ω) = 0, the claim in part
(iv) is seen from what we have proved so far together with [112, (7.3.24)] and [112,
(5.9.18)] which, in concert, show that 𝒞α (∂∗ Ω) may be canonically identified with
𝒞α (∂Ω) in the present setting. 

Next, we turn our attention to the study the harmonic double layer. In the most
general setting, we introduce the latter operator as follows.

Definition 2.5.17 Given a set Ω ⊆ Rn (where n ∈ N, n ≥ 2) of locally finite

perimeter, denote by ν its geometric measure theoretic outward unit normal, and
abbreviate σ∗ := H n−1 ∂∗ Ω. In this setting, define the action of the harmonic double
layer D associated with Ω on an arbitrary σ∗ -measurable function f : ∂∗ Ω → C
satisfying ∫
| f (y)|
dσ∗ (y) < +∞ (2.5.201)
1 + |y| n−1
∂∗ Ω

according to
446 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
1 ν(y), y − x
D f (x) := f (y) dσ∗ (y), ∀x ∈ Ω̊, (2.5.202)
ωn−1 ∂∗ Ω |x − y| n

where ωn−1 is the surface area of the unit sphere in Rn .

In the same geometric context, [112, Proposition 5.6.7] guarantees that the
principal-value limit
∫
1 ν(y), y − x
K f (x) := lim+ f (y) dσ∗ (y) (2.5.203)
ε→0 ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε

exists for any given function f ∈ L 1 ∂∗ Ω, 1+ |σ· |∗n−1 at σ∗ -a.e. point x ∈ ∂∗ Ω. The

assignment f → K f , mapping functions from the space L 1 ∂∗ Ω, 1+ |σ· |∗n−1 into σ∗ -
measurable functions on ∂∗ Ω (again, see [112, Proposition 5.6.7]), is then referred to
as the principal-value, or boundary-to-boundary, harmonic double layer
potential operator.
Finally, if one strengthens the original hypotheses on the underlying domain by
assuming that Ω is a Lebesgue measurable set whose topological boundary ∂Ω
is countably rectifiable (of dimension n − 1) and has locally finite H n−1 measure,
then the last claim in [112, Proposition 5.6.7] guarantees that for each function

f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
the limit in (2.5.203) actually exists for H n−1 -a.e. x ∈ ∂Ω
and gives rise to a H n−1 -measurable function on ∂Ω. This is how K f , the action
of the principal-value (or boundary-to-boundary) harmonic double layer potential
operator on the function f , is to be understood in this context.

A few comments pertaining to Definition 2.5.17 are as follows. First, it is worth

noting that the conditions imposed on f ensure that the integral in the right-hand side
of (2.5.202) is absolutely convergent for each x ∈ Ω̊. Second, if ∂Ω is upper Ahlfors
regular then any function f ∈ L p (∂∗ Ω, σ∗ ) with p ∈ [1, ∞) is σ∗ -measurable and
satisfies (2.5.201) (as maybe seen using [112, Lemma 7.2.1]). Third, recall from
(1.5.56) that EΔ denotes the standard fundamental solution for the Laplacian in Rn ,
where n ∈ N with n ≥ 2. If Ω is as in Definition 2.5.17, then the integral kernel
of the harmonic double layer D associated with Ω may be expressed in terms of
(1.5.56) as
1 ν(y), y − x
= ν(y) · (∇EΔ )(y − x) = ∂ν(y) EΔ (x − y)
ωn−1 |x − y| n (2.5.204)
for all x ∈ Ω and σ∗ -a.e. y ∈ ∂∗ Ω,

where ∂ν := ν·∇ denotes the normal derivative on ∂∗ Ω (i.e., the directional derivative
along the unit vector ν). Lastly, we wish to remark that whenever Ω is a Lebesgue
measurable set whose topological boundary is a UR set then automatically ∂Ω is
countably rectifiable (of dimension n − 1) and has locally finite H n−1 measure (cf.
[112, Definition 5.10.1] and [112, Proposition 6.10.5]).
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 447

Some trade-mark properties of the integral operator (2.5.202) are discussed be-
low. Specifically, the harmonic double layer is a mechanism for producing lots of
harmonic functions in Ω, using σ∗ -measurable functions satisfying (2.5.201) (called
“densities”) as inputs, and the harmonic double layer reproduces constants if Ω is
bounded.
Lemma 2.5.18 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a set of locally finite perimeter.
Denote by ν its geometric measure theoretic outward unit normal, and abbreviate
σ∗ := H n−1 ∂∗ Ω. Consider the harmonic double layer associated with Ω as in
(2.5.202).
Then for each σ∗ -measurable function f : ∂∗ Ω → C satisfying the integrability
condition in (2.5.201) one has

D f ∈ 𝒞∞ Ω̊ and Δ(D f ) = 0 in Ω̊, (2.5.205)

where Δ := ∂12 + · · · + ∂n2 is the Laplacian in Rn . Moreover,

whenever Ω is open and ∂Ω is bounded then


1 in Ω, if Ω is bounded, (2.5.206)
D1 ≡
0 in Ω, if Ω is unbounded.

Proof The claims in (2.5.205) are seen straight from definitions. The claim in
(2.5.206) is a direct consequence of Lemma 2.5.14 used with k! as in (2.5.129). 
As seen from our next lemma, the harmonic double layer is reasonably behaved
on Hölder spaces, especially when considered on uniform domains (cf. [112, Defi-
nition 5.11.10]) with compact Ahlfors regular boundaries.
Lemma 2.5.19 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set such that ∂Ω is
bounded and ∂∗ Ω is upper Ahlfors regular. In particular (cf. [112, (5.9.16)]), Ω has
locally finite perimeter, so one may associate the harmonic double layer D with Ω
as in (2.5.202).
Then for each α ∈ (0, 1) there exists some C ∈ (0, ∞), depending only on n, α,
diam(∂∗ Ω), and the upper Ahlfors regularity constant of ∂∗ Ω, such that for every
function f ∈ 𝒞α (∂∗ Ω) one has
    
sup (D f )(x) + sup dist(x, ∂∗ Ω)1−α ∇(D f )(x) ≤ C f 𝒞α (∂∗ Ω) . (2.5.207)
x ∈Ω x ∈Ω

In particular, if Ω ⊆ Rn is a uniform domain whose boundary is compact and

upper Ahlfors regular then, for each α ∈ (0, 1),

D : 𝒞α (∂∗ Ω) −→ 𝒞α Ω (2.5.208)

is a well-defined, linear, and bounded operator; hence so is


D : 𝒞α (∂Ω) −→ 𝒞α Ω . (2.5.209)
448 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Proof All claims are direct consequence of Lemma 2.5.15 specialized to the case
when the vector-valued function k! is defined as in (2.5.129). 
Given a set Ω ⊆ Rn (where n ∈ N, n ≥ 2) of locally finite perimeter, abbreviate
σ∗ := H n−1 ∂∗ Ω and denote by ν the geometric measure theoretic outward unit
normal to Ω. In this
 setting, we let the maximal harmonic double layer Kmax act on
functions f ∈ L 1 ∂∗ Ω, 1+ |σ· |∗n−1 according to
 
 
 ∫ 
 1 ν(y), y − x 
 
(Kmax f )(x) := sup  f (y) dσ∗ (y) , (2.5.210)

ε>0 ωn−1 |x − y| n 
 y ∈∂∗ Ω 
 |x−y |>ε


for all x ∈ ∂Ω.

In the lemma below we prove that Kmax maps Hölder functions into bounded
functions provided Ω is a Lebesgue measurable set whose boundary is compact and
upper Ahlfors regular.
Lemma 2.5.20 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be a Lebesgue measurable set
such that ∂Ω is bounded and ∂∗ Ω is upper Ahlfors regular. Then for each given
exponent α > 0 there exists some number N ∈ (0, ∞), depending only on n, α, and
the upper Ahlfors regularity constant of ∂∗ Ω, with the property that

·  f 𝒞. α (∂ Ω) + N · sup | f |,
α
sup (Kmax f )(x) ≤ N diam(∂∗ Ω) (2.5.211)
∗
x ∈∂∗ Ω ∂∗ Ω

for each function f ∈ 𝒞α (∂∗ Ω).

Proof This is a direct consequence of Lemma 2.5.13 used with k! as in (2.5.129). 
The next order of business is to study the boundary-to-boundary harmonic double
layer potential operator from (2.5.203). In this vein, we shall first prove the following
result, dealing with its action on Hölder functions.
Proposition 2.5.21 Let Ω ⊆ Rn (where n ∈ N with n ≥ 2) be a Lebesgue measurable
set with a compact upper Ahlfors regular boundary and abbreviate σ := H n−1 ∂Ω.
Then for each α ∈ (0, 1) the following statements are true.
(i) For each f ∈ 𝒞α (∂∗ Ω) the limit defining the principal-value harmonic double
layer potential (K f )(x) in (2.5.203) exists for σ-a.e. point x ∈ ∂∗ Ω.
(ii) After possibly redefining K f on a σ-nullset contained in ∂∗ Ω, the assignment
f → K f thus defined induces a well-defined, linear, and bounded operator

K : 𝒞α (∂∗ Ω) −→ 𝒞α (∂∗ Ω) (2.5.212)

with the property that


+ 12 on ∂∗ Ω, if Ω is bounded,
K1 ≡ (2.5.213)
− 12 on ∂∗ Ω, if Ω is unbounded.
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 449

(iii) Under the additional assumption that the set Ω is open and ∂Ω is also lower
Ahlfors regular, the following jump-formula (where I denotes the identity oper-
ator and κ > 0 is an arbitrary fixed aperture parameter) is valid:

for every given function f ∈ 𝒞α (∂∗ Ω) one has

 κ−n.t. (2.5.214)

Df  = ( 12 I + K) f at σ-a.e. point on ∂∗ Ω.
∂Ω

(iv) In the case when it is also assumed that H n−1 (∂Ω \ ∂∗ Ω) = 0, one may replace
∂∗ Ω by ∂Ω in the formulation of (2.5.212), (2.5.213), and (2.5.214).

Proof All claims are direct consequences of Proposition 2.5.16 used for the vector-
valued function k! defined as in (2.5.129). 

As the proposition below illustrates, the most natural environment in which the
principal-value harmonic double layer (2.5.203) is reasonably behaved on Lebesgue
spaces is that of uniformly rectifiable sets.

Proposition 2.5.22 Consider a Lebesgue measurable set Ω ⊆ Rn (where n ∈ N,

n ≥ 2) whose boundary is a UR set. Abbreviate σ := H n−1 ∂Ω, σ∗ := H n−1 ∂∗ Ω,
and denote by ν the geometric measure theoretic outward unit normal to Ω. In this
setting, consider the harmonic double layer potential operator D and the maximal
operator Kmax associated with Ω as in (2.5.202) and (2.5.210), respectively.
Then for each p ∈ [1, ∞) one can select a constant C ∈ (0, ∞) depending only
on n, p, and the UR character of ∂Ω, such that for every f ∈ L p (∂∗ Ω, σ∗ ) one has

Kmax f  L p (∂Ω,σ) ≤ C f  L p (∂∗ Ω,σ∗ ) if 1 < p < ∞, (2.5.215)

Kmax f  L 1,∞ (∂Ω,σ) ≤ C f  L 1 (∂∗ Ω,σ∗ ) if p = 1, (2.5.216)

Furthermore, for each f ∈ L p (∂∗ Ω, σ∗ ) with p ∈ [1, ∞) the limit

∫
1 ν(y), y − x
(K f )(x) = lim+ f (y) dσ∗ (y) exists for σ-a.e. x ∈ ∂Ω,
ε→0 ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε
(2.5.217)
and the operator K is well defined, linear, and bounded in the following settings:

K : L p (∂∗ Ω, σ∗ ) −→ L p (∂Ω, σ), p ∈ (1, ∞), (2.5.218)

K : L 1 (∂∗ Ω, σ∗ ) −→ L 1,∞ (∂Ω, σ). (2.5.219)

Finally, under the additional assumption that Ω is open, for each aperture pa-
rameter κ ∈ (0, ∞) and integrability exponent p ∈ [1, ∞), there exists a constant
C ∈ (0, ∞) depending only on n, p, κ, and the UR character of ∂Ω, such that for
every f ∈ L p (∂∗ Ω, σ∗ ) one has
450 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Nκ (D f ) L p (∂Ω,σ) ≤ C f  L p (∂∗ Ω,σ∗ ) if 1 < p < ∞, (2.5.220)

Nκ (D f ) L 1,∞ (∂Ω,σ) ≤ C f  L 1 (∂∗ Ω,σ∗ ) if p = 1. (2.5.221)

Proof The claims about Kmax and K are seen from Theorem 2.3.2 applied with
Σ := ∂Ω and a natural choice of the kernel, keeping in mind that any given
f ∈ L p (∂∗ Ω, σ∗ ) may be canonically viewed as a function in L p (∂Ω, σ) sim-
ply extending it by zero on ∂Ω \ ∂∗ Ω. In a similar fashion, the nontangential maximal
function estimates from (2.5.220)-(2.5.221) are consequences of Theorem 2.4.1. 

Remark 2.5.23 In the setting of Proposition 2.5.22, it is worth noting that we may
also view K as a well-defined, linear, and bounded operator

K : L p (∂Ω, σ) −→ L p (∂Ω, σ), p ∈ (1, ∞), (2.5.222)

K : L 1 (∂Ω, σ) −→ L 1,∞ (∂Ω, σ), (2.5.223)

simply by composing (2.5.218)-(2.5.219) to the left with the canonical restriction


L p (∂Ω, σ)  f −→ f ∂∗ Ω ∈ L p (∂∗ Ω, σ∗ ), p ∈ [1, ∞). (2.5.224)

In other words, for f ∈ L p (∂Ω, σ) with p ∈ [1, ∞) we let K f retain the definition
(2.5.217), simply by now regarding f as being restricted to ∂∗ Ω in the integrand in
the right-hand side of (2.5.217).

To further shed some light on the nature of the principal-value limit in (2.5.217),
consider the simple case of the harmonic double layer Kpoly associated with an open
planar polygon region Ωpoly ⊂ R2 , with vertexes x j , 1 ≤ j ≤ N, and corresponding
angles α j ∈ (0, 2π), 1 ≤ j ≤ N. In particular, Ωpoly is a bounded open subset of R2 ,
whose topological boundary is a compact UR set, satisfying ∂∗ Ωpoly = ∂Ωpoly and
∂ ∗ Ωpoly = ∂Ωpoly \ {x1, . . . , x N }. As such, the general result from (2.5.217) presently
ensures that
∫
1 ν(y), y − x
(Kpoly 1)(x) = lim+ dH 1 (y) exists for H 1 -a.e. x ∈ ∂Ω.
ε→0 2π |x − y| 2
y ∈∂Ωpoly
|x−y |>ε
(2.5.225)
This being said, from (2.5.139) we see that actually for every point x ∈ ∂Ωpoly the
limit defining (Kpoly 1)(x) exists and, in fact,

H 1 Ωpoly ∩ ∂B(x, r)
(Kpoly 1)(x) = lim+
ε→0 2πr

1/2 if x ∈ ∂Ωpoly \ {x1, . . . , x N },
= (2.5.226)
α j /(2π) if x = x j for some 1 ≤ j ≤ N.
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 451

Moreover, writing (Kpoly f )(x) = Kpoly ( f − f (x)) (x) + f (x)(Kpoly 1)(x) we see that,
given any function f ∈ 𝒞α (∂Ωpoly ) with α ∈ (0, 1), the limit defining (Kpoly 1)(x)
exists for every x ∈ ∂Ωpoly . In fact, with an absolutely convergent integral we have
∫
1 ν(y), y − x 
(Kpoly f )(x) = f (y) − f (x) dH 1 (y)
2π |x − y| 2
∂Ωpoly

f (x)/2 if x ∈ ∂Ωpoly \ {x1, . . . , x N }
+ (2.5.227)
f (x)α j /(2π) if x = x j for some 1 ≤ j ≤ N.

In our next proposition we discuss Carleson measure estimates for the harmonic
double layer acting on functions in L n−1 as well as for its gradient acting on functions
of bounded mean oscillations.

Proposition 2.5.24 Let Ω ⊆ Rn (where n ∈ N, n ≥ 2) be an open set with the

property that ∂Ω is a UR set; in particular, Ω is a set of locally finite perimeter.
Denote by ν the geometric measure theoretic outward unit normal to Ω and abbreviate
σ := H n−1 ∂Ω.
Then there exists a constant C = C(∂Ω) ∈ (0, ∞) with the property that

given any f ∈ L n−1 (∂Ω, σ), it follows that |D f | 2 dist(·, ∂Ω) dL n is

a Carleson measure in Ω, with constant ≤ C f  L2 n−1 (∂Ω,σ) . (2.5.228)

Furthermore, if in the same setting one defines the action of the gradient of the
harmonic double layer ∇D on functions f ∈ BMO(∂Ω, σ) according to
 1 ∫ #  y − x $ 
(∇D) f (x) := ν(y), ∂x j f (y) dσ(y) (2.5.229)
ωn−1 ∂∗ Ω |x − y| n 1≤ j ≤n

for all x ∈ Ω, then for each f ∈ BMO(∂Ω, σ) the function (∇D) f is well defined
in Ω and, given any p ∈ (1, ∞), it follows that
 
(∇D) f  p dist(·, ∂Ω) p−1 dL n (2.5.230)

is a Carleson measure in Ω, in the quantitative sense that there exists a constant

C ∈ (0, ∞) independent of f with the property that
∫
1  
sup  (∇D) f  p dist(·, ∂Ω) p−1 dL n
x ∈∂Ω, r >0 σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
p
≤ C f BMO(∂Ω,σ) . (2.5.231)

In particular, corresponding to p = 2,
 
(∇D) f 2 dist(·, ∂Ω) dL n is a Carleson measure in Ω,
(2.5.232)
for each function f ∈ BMO(∂Ω, σ).
452 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Proof The claim in (2.5.228) is clear from (2.4.48) and Definition 2.5.17. To proceed,
fix an arbitrary function f ∈ BMO(∂Ω, σ). Since for each x ∈ Ω we have
∫ #  y − x  $
 
 ν(y), ∂x j | f (y)| dσ(y)
∂∗ Ω |x − y| n
∫
| f (y)|
≤ cn dσ(y) < +∞, (2.5.233)
∂Ω |x − y| n

thanks to the fact that ∂Ω is an Ahlfors regular set, and [112, (7.4.116)], we conclude
that the function (∇D) f is well defined in Ω, via absolutely convergent integrals.
We next claim that
(∇D)1 ≡ 0 in Ω. (2.5.234)
To justify this claim, fix some index j ∈ {1, . . . , n} and pick an arbitrary point x ∈ Ω.
Then the function (∂j E)(x − ·) is locally absolutely integrable in Ω and this permits
us to meaningfully define the vector field

F!j := ∇ (∂j EΔ )(x − ·) (2.5.235)

by taking the gradient in the sense of distributions in Ω (in the “dot” variable). Hence,
F!j belongs to D (Ω) . In fact, since the singular support of F!j is the singleton
n

{x}, we see that actually

F!j ∈ Lloc (Ω, L n ) + ℰ(Ω) .

1 n
(2.5.236)

With the divergence taken in the sense of distributions in Ω we may compute (bearing
in mind that EΔ is a fundamental solution for Δ in Rn )

divF!j = Δ (∂j EΔ )(x − ·) = −∂j Δ EΔ (x − ·) = −∂j δx in D (Ω), (2.5.237)

where δx is the Dirac distribution in Ω with mass at x ∈ Ω. Also, if we introduce

d := dist(x, ∂Ω) > 0 and select some arbitrary κ > 0, [112, Lemma 8.3.7] shows
that there exists a constant C ∈ (0, ∞) such that
 
NκΩ\B(x,d/2) F!j (y) ≤ C|y − x| −n for each y ∈ ∂Ω. (2.5.238)

In turn, from (2.5.238), [112, (8.2.26)], and [112, Lemma 7.2.1], we conclude that

NκΩ\B(x,d/2) F!j ∈ L 1 (∂Ω, σ). (2.5.239)

In addition, since F!j is continuous in a neighborhood of ∂Ω it follows that the

κ−n.t.
pointwise nontangential boundary trace F!j ∂Ω exists at every point in ∂nta Ω and, in
fact,  κ−n.t. 
F!j ∂Ω (y) = ∇y (∂j EΔ )(x − y) at every y ∈ ∂nta Ω. (2.5.240)
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 453

In concert with (1.5.56) and [112, Proposition 8.8.6] this ultimately proves that
 κ−n.t.  1  y−x 
F!j  ∂Ω
(y) = ∂x j for σ-a.e. y ∈ ∂∗ Ω. (2.5.241)
ωn−1 |x − y| n

Finally, we note that in the case when Ω is unbounded we have | F!j (y)| = O(|y| −n )
for y ∈ Ω with |y| → ∞. Hence, in such a scenario,
∫
|y · F!j (y)| dL n (y) = O(R) as R → ∞. (2.5.242)
[B(0,2R)\B(0,R)]∩Ω

From (2.5.236), (2.5.237), (2.5.239), (2.5.241), and (2.4.128), we see that the hy-
potheses of [112, Theorem 1.4.1] are satisfied by F!j . As such formula [112, (1.4.6)]
applies and presently gives
∫  κ−n.t. 
 
!j , 1 ∞ =
∗ div F ν(y) · F!j  (y) dσ(y). (2.5.243)
(𝒞 (Ω))
∞
b 𝒞 (Ω) b ∂∗ Ω ∂Ω

Now, on the one hand, thanks to (2.5.237) and the compatibility property recorded
in [112, (4.6.21)], we have

!
(𝒞∞b (Ω))∗ divFj , 1 𝒞∞b (Ω) = − ℰ (Ω) ∂j δx, 1 ℰ(Ω)

= ℰ (Ω) δx, ∂j 1 ℰ(Ω) = 0. (2.5.244)

On the other hand, (2.5.241) implies that

∫  κ−n.t. 

ν(y) · F!j  (y) dσ(y)
∂∗ Ω ∂Ω
∫ #  y − x $
1
= ν(y), ∂x j dσ(y). (2.5.245)
ωn−1 ∂∗ Ω |x − y| n

Collectively, (2.5.243), (2.5.244), (2.5.245), and (2.5.229) imply the cancelation

property claimed in (2.5.234).
With (2.5.234) in hand, the remaining argument proceeds very much as in the
case of the proof of Corollary 2.4.2. Specifically, having fixed a point xo ∈ ∂Ω and
a scale r > 0, the same arguments which have produced (2.4.136) (based on the
estimate recorded in (2.4.34) and the decay of the integral kernel of ∇D) now give
∫
 
(∇D f )(x) p dist(x, ∂Ω) p−1 dx
B(x o ,r)∩Ω

≤ Cr n−1 fp# (xo ) p + Cr n−1 f1# (xo ) p, (2.5.246)

454 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

where the L q -based Fefferman-Stein maximal function fq# has been defined in
(A.0.123). From this, (2.5.231) now follows on account of [112, (7.4.112)]. The
final claim is obtained by specializing the earlier result to the case when p = 2. 

We desire to augment Proposition 2.5.22 and Proposition 2.5.24 with a jump-

formula relating the harmonic double layers D and K much as in (2.5.214) but when
the operators in question now act on functions from Lebesgue spaces. This is done
in the proposition below where we consider the nontangential trace of D acting on
L p functions at almost all points in the geometric measure theoretic boundary of the
underlying domain.

Proposition 2.5.25 Suppose Ω ⊆ Rn (where n ∈ N with n ≥ 2) is a nonempty open

set with the property that ∂Ω is a UR set. In particular, Ω is a set of locally finite
perimeter. Abbreviate σ∗ := H n−1 ∂∗ Ω and denote by ν the geometric measure
theoretic outward unit normal to Ω. In this context, define the (harmonic) double
layer potential operator associated with Ω as in (2.5.202), and also consider its
principal-value version acting on functions f ∈ L p (∂∗ Ω, σ∗ ) with p ∈ [1, ∞) as in
(2.5.217) (for σ∗ -a.e. x ∈ ∂∗ Ω). Finally, fix an arbitrary κ > 0.
Then the double layer potential (2.5.202) has the property that for every function
f ∈ L p (∂∗ Ω, σ∗ ) with 1 ≤ p < ∞ the following jump-formula holds (where I
denotes the identity operator):

lim (D f )(z) = ( 12 I + K) f (x) for σ∗ -a.e. x ∈ ∂∗ Ω. (2.5.247)

z ∈Γκ (x)
z→x

In the setting of Proposition 2.5.25, it has been noted in Remark 2.5.23 that
the principal-value harmonic double layer K may be naturally considered as an
operator acting on functions belonging to L p (∂Ω, σ) with p ∈ [1, ∞) and where
σ := H n−1 ∂Ω. Since this is also the case for the boundary-to-domain harmonic
double layer D (originally defined in (2.5.202)), one may wonder whether the jump-
formula (2.5.247) written for generic functions f ∈ L p (∂Ω, σ) with p ∈ [1, ∞) is
actually valid σ-a.e. on E for a possibly larger set E ⊆ ∂Ω than ∂∗ Ω. However, this
is not the case since such an eventuality would imply that
κ−n.t.
f = 2(D f )∂Ω − 2K f at σ-a.e. point on E. (2.5.248)

Given that both D f and K f depend only on f ∂∗ Ω , this would ultimately imply that

f ∂∗ Ω always determines f on the bigger set E, which is obviously not the case.
Hence, from this point of view, the jump-formula (2.5.247) is optimally formulated.
Proof of Proposition 2.5.25 First note that [112, Proposition 8.8.6] ensures that it
is meaningful to consider the limit in (2.5.247). Second, as seen from (2.5.220)-
(2.5.221), the nontangential maximal operator associated with the type of limit
implicit in (2.5.247), i.e.,
 
f −→ Nκ (D f )(x) := sup (D f )(z), x ∈ ∂∗ Ω, (2.5.249)
z ∈Γκ (x)
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 455

is bounded from L p (∂∗ Ω, σ∗ ) into itself if 1 < p < ∞, and from L 1 (∂∗ Ω, σ∗ ) into
L 1,∞ (∂∗ Ω, σ∗ ).
Third, [112, Proposition 6.2.11] may be presently applied with 𝒳 := Ω ∪ ∂∗ Ω
endowed with the topology inherited from the ambient Euclidean space, X := ∂∗ Ω,
μ := σ∗ which is a locally finite complete Borel-regular measure on X by virtue of
[112, Lemma 3.6.4] (the fact that the hypotheses in [112, (3.6.25)] are presently sat-
isfied is seen from [112, (5.6.35)] and [112, (5.6.33)]), the sets Γ(x) := Γκ (x) for each
x ∈ ∂∗ Ω which satisfy [112, (6.2.71)] thanks to item (iii) in [112, Proposition 8.8.6],
Y := L p (∂∗ Ω, σ∗ ), the linear operator T := D, and
 
V := f |∂∗ Ω : f ∈ Lipc (∂Ω) (2.5.250)

which is a dense subset of Y by [112, (3.7.3)] and [112, Proposition 3.7.1]. These
choices imply that the associated maximal operator (cf. [112, (6.2.73)]) is (2.5.249)
which, according to earlier remarks, satisfies [112, (6.2.74)]. As such, the only
hypothesis in [112, Proposition 6.2.11] left to check at this point is [112, (6.2.75)].
This follows once we show that
for each function f belonging to Lipc (∂Ω),
lim (D f )(z) = ( 12 I + K) f (x) for σ∗ -a.e. x ∈ ∂∗ Ω. (2.5.251)
z ∈Γκ (x)
z→x

However, this may be seen from a cursory inspection of the argument used in the proof
of (2.5.169) in Proposition 2.5.16 (with k! as in (2.5.129)) which continues to work
in the present setting since the compactness of ∂Ω stipulated in Proposition 2.5.16
is not currently needed. The only significant difference is that now, having fixed an
arbitrary function f ∈ Lipc (∂Ω), the existence of a σ∗ -measurable set A ⊆ ∂∗ Ω
with the property that (2.5.189) holds is implied by (2.5.217) (in lieu of item (i) of
Proposition 2.5.16, as was the case in the past).
The application of [112, Proposition 6.2.11] in the manner just described then
guarantees that

for each given function f ∈ L p (∂∗ Ω, σ∗ ) with 1 ≤ p < ∞ the

nontangential limit lim (D f )(z) exists for σ∗ -a.e. x ∈ ∂∗ Ω, and
z ∈Γκ (x)
  z→x (2.5.252)
 
 lim (D f )(z) ≤ Nκ (D f )(x) at σ∗ -a.e. point x ∈ ∂∗ Ω.
z ∈Γκ (x)
z→x

With this in hand, [112, Corollary 8.9.6] additionally gives that

for each f ∈ L p (∂∗ Ω, σ∗ ) with 1 ≤ p < ∞ the function defined

σ∗ -a.e. as ∂∗ Ω  x → lim (D f )(z) is σ∗ -measurable on ∂∗ Ω. (2.5.253)
z ∈Γκ (x)
z→x

In concert, (2.5.252), (2.5.253), and the boundedness of the mapping in (2.5.249)

ensure that the assignment
456 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

f −→ lim (D f )(z), x ∈ ∂∗ Ω, (2.5.254)

z ∈Γκ (x)
z→x

is linear and bounded from L p (∂∗ Ω, σ∗ ) into itself if 1 < p < ∞, and from
L 1 (∂∗ Ω, σ∗ ) into the Lorentz space L 1,∞ (∂∗ Ω, σ∗ ). Thanks to (2.3.18)-(2.3.19), the
operator 12 I +K is also bounded in the very same context. Granted these boundedness
properties, the identity recorded in (2.5.251), and the density of V in L p (∂∗ Ω, σ∗ )
for each p ∈ [1, ∞), we ultimately conclude that (2.5.247) holds. 

Fix an integer n ∈ N with n ≥ 2 and recall the standard fundamental solution EΔ

for the Laplace operator in Rn introduced in (1.5.56). Given an open set of locally
finite perimeter Ω ⊆ Rn , abbreviate σ∗ := H n−1 ∂∗ Ω and denote by ν = (ν1, . . . , νn )
the geometric measure theoretic outward unit normal to Ω. In this setting, we are next
interested in considering integral operators of a similar nature to the harmonic double
layer D but whose integral kernels are obtained by applying certain “tangential”
derivatives to EΔ (x − y) (instead of taking the normal derivative as done in the
case of D in (2.5.204)). Specifically, with Ω as above, for each j, k ∈ {1, . . . , n} we
presently consider the operator R jk acting on σ∗ -measurable functions f : ∂∗ Ω → C
satisfying ∫
| f (y)|
dσ∗ (y) < +∞ (2.5.255)
1 + |y| n−1
∂∗ Ω

according to
∫  
R jk f (x) := ν j (y)∂yk EΔ (x − y) − νk (y)∂y j EΔ (x − y) f (y) dσ∗ (y)
∂∗ Ω
∫
−1 ν j (y)(xk − yk ) − νk (y)(x j − y j )
= f (y) dσ∗ (y) (2.5.256)
ωn−1 |x − y| n
∂∗ Ω

for each x ∈ Ω. In particular,

R j j = 0 and R k j = −R jk for every j, k ∈ {1, . . . , n}. (2.5.257)

Proposition 2.5.26 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an open set of locally

finite perimeter. Denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward
unit normal to Ω, and abbreviate σ∗ := H n−1 ∂∗ Ω, σ := H n−1 ∂Ω. Then for each
j, k ∈ {1, . . . , n} the following properties hold.

(i) For every σ∗ -measurable function f : ∂∗ Ω → C satisfying the integrability

condition demanded in (2.5.255) one has

R jk f ∈ 𝒞∞ (Ω) and Δ(R jk f ) = 0 in Ω. (2.5.258)

(ii) Under the additional assumption that ∂Ω is bounded one has

2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 457

1 in Ω, if Ω is bounded,
R jk 1 ≡ (2.5.259)
0 in Ω, if Ω is unbounded.

(iii) Under the additional assumption that ∂Ω is bounded and ∂∗ Ω is upper Ahlfors
regular, it follows that for each α ∈ (0, 1) there exists some C ∈ (0, ∞),
depending only on n, α, diam(∂∗ Ω), and the upper Ahlfors regularity constant
of ∂∗ Ω, such that for every function f ∈ 𝒞α (∂∗ Ω) one has
    
sup (R jk f )(x) + sup dist(x, ∂∗ Ω)1−α ∇(R jk f )(x) ≤ C f 𝒞α (∂∗ Ω) .
x ∈Ω x ∈Ω
(2.5.260)

In particular, if Ω ⊆ Rn is a uniform domain whose boundary is compact and

upper Ahlfors regular then R jk induces a well-defined, linear, and bounded
operator in the context

R jk : 𝒞α (∂∗ Ω) −→ 𝒞α Ω , ∀α ∈ (0, 1). (2.5.261)

As a corollary, if Ω ⊆ Rn is simultaneously a uniform domain, and an Ahlfors

regular domain with compact boundary, then for each α ∈ (0, 1) the operator
R jk is well defined, linear, and bounded in the context

R jk : 𝒞α (∂Ω) −→ 𝒞α Ω . (2.5.262)

(iv) If ∂Ω is a UR set then for each aperture parameter κ ∈ (0, ∞) and integrability
exponent p ∈ [1, ∞) there exists a constant C = C(Ω, n, κ, p) ∈ (0, ∞) with the
property that for each function f ∈ L p (∂∗ Ω, σ∗ ) one has

Nκ (R jk f ) L p (∂Ω,σ) ≤ C f  L p (∂∗ Ω,σ∗ ) if 1 < p < ∞, (2.5.263)

Nκ (R jk f ) L 1,∞ (∂Ω,σ) ≤ C f  L 1 (∂∗ Ω,σ∗ ) if p = 1. (2.5.264)

(v) Suppose ∂Ω is a UR set. Then there exists C = C(∂Ω) ∈ (0, ∞) such that
 2
given any f ∈ L n−1 (∂Ω, σ), it follows that R jk f  dist(·, ∂Ω) dL n is a
Carleson measure in Ω, with constant ≤ C f  L2 n−1 (∂Ω,σ) .
(2.5.265)
In addition, if one defines the action of the operator ∇R jk on arbitrary functions
f belonging to the space BMO(∂Ω, σ) according to
∫  
(∇R jk ) f (x) := νk (y)(∇∂j EΔ )(x − y) − ν j (y)(∇∂k EΔ )(x − y) f (y) dσ∗ (y)
∂∗ Ω
(2.5.266)
at each x ∈  Ω, then for
p every f ∈ BMO(∂Ω, σ) the function (∇R )
jk f is well
defined and (∇R jk ) f  dist(·, ∂Ω) p−1 dL n is a Carleson measure in Ω for each
p ∈ (1, ∞), in the quantitative sense that there exists some constant C ∈ (0, ∞),
458 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

independent of f , with the property that

∫
1  
sup  (∇R jk ) f  p dist(·, ∂Ω) p−1 dL n
x ∈∂Ω, r >0 σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
p
≤ C f BMO(∂Ω,σ) . (2.5.267)
 2
In particular, for p = 2, one has that (∇R jk ) f  dist(·, ∂Ω) dL n is a Carleson
measure in Ω, for each function f ∈ BMO(∂Ω, σ).

Proof The claims in items (i)-(iii) are consequences of (2.5.256) and Proposi-
tion 2.5.16 used with k! as in (2.5.140). Next, item (iv) is a consequence of Theo-
rem 2.4.1. Finally, the first claim in item (v) is a direct consequence of (2.4.48) and
(2.5.256), while the second claim is implied by Corollary 2.4.2. 

Let us now bring in the principal-value versions of the integral operators con-
sidered in (2.5.256). Specifically, fix n ∈ N with n ≥ 2 and suppose Ω ⊆ Rn
is a set of locally finite perimeter. Abbreviate σ∗ := H n−1 ∂∗ Ω and denote by
ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal to Ω. Pick
j, k ∈ {1, . . . , n} arbitrary. Then [112, Proposition 5.6.7] ensures that, given any
function f ∈ L 1 ∂∗ Ω, 1+ |σ· |∗n−1 , the principal-value limit
∫  
R jk f (x) := lim+ ν j (y)(∂k EΔ )(y − x) − νk (y)(∂j EΔ )(y − x) f (y) dσ∗ (y)
ε→0
y ∈∂∗ Ω
|x−y |>ε
(2.5.268)
exists at σ∗ -a.e. point x ∈ ∂∗ Ω and gives rise to a σ∗ -measurable function on ∂∗ Ω.
Note that, by design,

R j j = 0 and Rk j = −R jk for every j, k ∈ {1, . . . , n}. (2.5.269)

From [112, Proposition 5.6.7] we also know that, if Ω is a Lebesgue measurable set
whose topological boundary ∂Ω is countably rectifiable (of dimension n − 1) and has

locally finite H n−1 measure, then for each function f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
the limit in
(2.5.268) actually exists for H n−1 -a.e. x ∈ ∂Ω and gives rise to a H n−1 -measurable
function on ∂Ω.
Finally,
 we define the corresponding maximal operator acting on each function
f ∈ L 1 ∂∗ Ω, 1+ |σ· |∗n−1 according to

(R jk )max f (x) (2.5.270)

 
 
 ∫ 
   
 
:= sup  ν j (y)(∂k EΔ )(y − x) − νk (y)(∂j EΔ )(y − x) f (y) dσ∗ (y)
ε>0  
 y ∈∂∗ Ω 
|x−y |>ε 
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 459

at each x ∈ ∂Ω. In relation to the operators (2.5.268), (2.5.270), we have the following
result.

Proposition 2.5.27 Assume Ω ⊆ Rn (where n ∈ N with n ≥ 2) is a set of locally

finite perimeter, and abbreviate σ := H n−1 ∂Ω, σ∗ := H n−1 ∂∗ Ω. Then for each
j, k ∈ {1, . . . , n} the following statements are true.

(i) If ∂Ω is bounded and ∂∗ Ω is upper Ahlfors regular then for each α > 0 there
exists a number N ∈ (0, ∞), depending only on the dimension n, exponent α, and
the upper Ahlfors regularity constant of ∂∗ Ω, with the property that for every
function f ∈ 𝒞α (∂∗ Ω) one has

sup (R jk )max f (x) ≤ N diam(∂∗ Ω) ·  f 𝒞. α (∂ Ω) + N · sup | f |. (2.5.271)
α
∗
x ∈∂∗ Ω ∂∗ Ω

(ii) If ∂Ω is a bounded upper Ahlfors regular set then for each f ∈ 𝒞α (∂∗ Ω) with
α ∈ (0, 1) the limit defining R jk f (x) in (2.5.268) exists for σ-a.e. point x ∈ ∂∗ Ω.
Moreover, after possibly redefining R jk f on a σ-nullset contained in ∂∗ Ω, the
assignment f → R jk f thus defined induces a well-defined, linear, and bounded
operator

R jk : 𝒞α (∂∗ Ω) −→ 𝒞α (∂∗ Ω), ∀α ∈ (0, 1), (2.5.272)

with the property that

R jk 1 ≡ 0 on ∂∗ Ω. (2.5.273)
(iii) Suppose Ω is actually open and ∂Ω is a compact Ahlfors regular set. Then for
each arbitrary fixed number κ > 0, the following nontangential boundary trace
formula hods:

for each f ∈ 𝒞α (∂∗ Ω) with α ∈ (0, 1) one has

 κ−n.t. (2.5.274)

R jk f  = R jk f at σ-a.e. point on ∂∗ Ω.
∂Ω

(iv) In the case when it is also assumed that σ(∂Ω \ ∂∗ Ω) = 0, one may replace ∂∗ Ω
by ∂Ω in the formulation of (2.5.272), (2.5.273), and (2.5.274).
(v) Strengthen the assumptions by requiring that ∂Ω is a UR set. Then for each
p ∈ [1, ∞) there exists a constant C = C(Ω, n, p) ∈ (0, ∞) with the property that
for each f ∈ L p (∂∗ Ω, σ∗ ) one has

(R jk )max f  L p (∂Ω,σ) ≤ C f  L p (∂∗ Ω,σ∗ ) if 1 < p < ∞, (2.5.275)

(R jk )max f  L 1,∞ (∂Ω,σ) ≤ C f  L 1 (∂∗ Ω,σ∗ ) if p = 1. (2.5.276)

Furthermore, given any function f ∈ L p (∂∗ Ω, σ∗ ) with p ∈ [1, ∞) the limit

defining R jk f (x) in (2.5.268) exists for σ-a.e. x ∈ ∂Ω and the operator R jk is
well defined, linear, and bounded in the following settings:
460 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

R jk : L p (∂∗ Ω, σ∗ ) −→ L p (∂Ω, σ), p ∈ (1, ∞), (2.5.277)

R jk : L 1 (∂∗ Ω, σ∗ ) −→ L 1,∞ (∂Ω, σ). (2.5.278)

(vi) Assume that actually Ω is open and that ∂Ω is a UR set. Then for each fixed
arbitrary number κ > 0, then following nontangential boundary trace formula
holds:
for each f ∈ L p (∂∗ Ω, σ∗ ) with p ∈ [1, ∞) one has
  κ−n.t. (2.5.279)

R jk f  = R jk f at σ∗ -a.e. point on ∂∗ Ω.
∂Ω

Proof The claim made in item (i) is a direct consequence of Lemma 2.5.13 special-
ized to the case when k! is as in (2.5.140). The claims in items (ii)-(iv) follows from
Proposition 2.5.16 once again with k! as in (2.5.140) (bearing in mind that ϑ = 0
in this case). Next, all claims in item (v) are direct consequences of Theorem 2.3.2.
Finally, the nontangential boundary trace formula in item (vi) may be proved by
arguing as in Proposition 2.5.25 (making use of (2.5.274) the same way (2.5.214)
was used in the end-game of the proof of (2.5.247)). 

Let Ω ⊂ Rn be a set of locally finite perimeter and denote by ν = (ν1, . . . , νn ) its

geometric measure theoretic outward unit normal and abbreviate σ∗ := H n−1  ∂∗ Ω.
Embed Rn into the Clifford algebra Cn as in [112, (6.4.3)]. In particular, we agree
to identify ν with the Cn -valued function defined σ∗ -a.e. on ∂∗ Ω as

ν = ν1 e1 + · · · + νn en . (2.5.280)

The action of the (boundary-to-domain) Cauchy-Clifford integral operator on

any given σ∗ -measurable Clifford algebra-valued function f : ∂∗ Ω → Cn satisfying
∫
| f (y)|
dσ∗ (y) < +∞ (2.5.281)
∂∗ Ω 1 + |y|
n−1

is defined as
∫
1 x−y
C f (x) := % ν(y) % f (y) dσ∗ (y), ∀x ∈ Ω̊, (2.5.282)
ωn−1 ∂∗ Ω |x − y| n

while the action of the (boundary-to-boundary) maximal operator on f is defined as

 
 
 ∫ 
 1 x−y 
 
Cmax f (x) := sup  % ν(y) % f (y) dσ∗ (y) , (2.5.283)
ε>0  ω |x − y| n 
 n−1 y ∈∂∗ Ω 
 |x−y |>ε


for all x ∈ ∂Ω. From [112, Proposition 5.6.7] we also see that for any function
f ∈ L 1 ∂∗ Ω, 1+ |σ· |∗n−1 ⊗ Cn the principal-value limit
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 461
∫
1 x−y
C f (x) := lim+ % ν(y) % f (y) dσ∗ (y) (2.5.284)
ε→0 ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε

exists at σ∗ -a.e. point x ∈ ∂∗ Ω and gives rise to a σ∗ -measurable Clifford algebra-

 function on ∂∗ Ω. Then C is a well-defined operator, sending functions from
valued
L 1 ∂∗ Ω, 1+ |σ· |∗n−1 ⊗ Cn into σ∗ -measurable Clifford algebra-valued function on ∂∗ Ω.
In fact, the last claim in [112, Proposition 5.6.7] implies that if Ω is a Lebesgue
measurable set whose topological boundary ∂Ω is countably rectifiable (of di-
mension n − 1) and has locally finite H n−1 measure, then for each function

f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)| n−1
⊗ Cn the limit in (2.5.284) actually exists for H n−1 -a.e.
x ∈ ∂Ω and gives rise to a H n−1 -measurable Clifford algebra-valued function on
∂Ω.

Remark 2.5.28 Note that in (2.5.282), (2.5.284), the factor (x − y)/|x − y| n is

located to the left. For this reason, we shall occasionally, use notation designed
to emphasize this specific feature, i.e., write CL and C L in place of C and C,
respectively. This becomes particularly relevant upon noting that there are also
“right-handed” versions of the operators (2.5.282),
 (2.5.284). Specifically, in the
same geometric context as before, for each f ∈ L 1 ∂∗ Ω, 1+ |σ· |∗n−1 ⊗ Cn we define
∫
1 x−y
CR f (x) := f (y) % ν(y) % dσ∗ (y), ∀x ∈ Ω̊, (2.5.285)
ωn−1 ∂∗ Ω |x − y| n

and, for σ∗ -a.e. point x ∈ ∂∗ Ω,

∫
1 x−y
C R f (x) := lim+ f (y) % ν(y) % dσ∗ (y). (2.5.286)
ε→0 ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε

It is then apparent from the above definitions and [112, (6.4.29), (6.4.32)] that for
each given function f ∈ L 1 ∂∗ Ω, 1+ |σ· |∗n−1 ⊗ Cn we have (with ‘bar’ denoting Clifford
conjugation)
CR f = CL ( f ) in Ω̊, (2.5.287)
and
C R f = C L ( f ) at σ∗ -a.e. point on ∂∗ Ω. (2.5.288)
Consequently, the operators CR , C R enjoy very similar properties of their standard,
left-handed counterparts C, C. for this reason, we shall primarily
 just focus on the
latter. Here we only wish to note that for each function f ∈ L 1 ∂∗ Ω, 1+ |σ· |∗n−1 ⊗ Cn
we have (cf. (2.5.290) below)

DR CR f = 0 and D L CL f = 0 in Ω̊, (2.5.289)

462 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

where D L and DR denote the Dirac operator acting from the right and from the left,
respectively (cf. (A.0.40)-(A.0.41)).
The operators (2.5.282)-(2.5.284) are studied in the proposition below.
Proposition 2.5.29 Assume Ω ⊆ Rn (where n ∈ N, n ≥ 2) is a given set of locally fi-
nite perimeter. Denote by ν its geometric measure theoretic outward unit normal, and
abbreviate σ∗ := H n−1 ∂∗ Ω, σ := H n−1 ∂Ω. Then, for the Cauchy-Clifford integral
operators associated with Ω as in (2.5.282)-(2.5.284), the following statements are
true.
(i) For each σ∗ -measurable function f : ∂∗ Ω → Cn satisfying the integrability
condition in (2.5.281) one has

C f ∈ 𝒞∞ Ω̊ and D(C f ) = 0 in Ω̊, (2.5.290)

where D := e1 ∂1 + · · · + en ∂n is the Dirac operator in Rn .

(ii) Whenever Ω is open and ∂Ω is bounded, one has

1 in Ω, if Ω is bounded,
C1 ≡ (2.5.291)
0 in Ω, if Ω is unbounded.

(iii) Under the additional assumption that ∂Ω is bounded and ∂∗ Ω is upper Ahlfors
regular, it follows that for each α ∈ (0, 1) there exists some C ∈ (0, ∞), depending
only on n, α, diam(∂∗ Ω), and the upper Ahlfors regularity constant of ∂∗ Ω, such
that for every function f ∈ 𝒞α (∂∗ Ω) one has
    
sup (C f )(x) + sup dist(x, ∂∗ Ω)1−α ∇(C f )(x) ≤ C f 𝒞α (∂∗ Ω) . (2.5.292)
x ∈Ω x ∈Ω

In particular, if Ω ⊆ Rn is a uniform domain whose boundary is compact

and upper Ahlfors regular then C induces a well-defined, linear, and bounded
operator in the context

C : 𝒞α (∂∗ Ω) ⊗ Cn −→ 𝒞α Ω ⊗ Cn, ∀α ∈ (0, 1). (2.5.293)

As a corollary, if Ω ⊆ Rn is simultaneously a uniform domain, and an Ahlfors

regular domain with compact boundary, then for each α ∈ (0, 1) the operator
C is well defined, linear, and bounded in the context

C : 𝒞α (∂Ω) ⊗ Cn −→ 𝒞α Ω ⊗ Cn . (2.5.294)

(iv) If ∂Ω is a UR set then for each aperture parameter κ ∈ (0, ∞) and integrability
exponent p ∈ [1, ∞) there exists a constant C = C(Ω, n, κ, p) ∈ (0, ∞) with the
property that for each function f ∈ L p (∂∗ Ω, σ∗ ) ⊗ Cn one has

Nκ (C f ) L p (∂Ω,σ) ≤ C f  L p (∂∗ Ω,σ∗ )⊗ Cn if 1 < p < ∞, (2.5.295)

Nκ (C f ) L 1,∞ (∂Ω,σ) ≤ C f  L 1 (∂∗ Ω,σ∗ )⊗ Cn if p = 1. (2.5.296)
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 463

(v) Suppose ∂Ω is a UR set. Then there exists C = C(∂Ω) ∈ (0, ∞) such that

given any f ∈ L n−1 (∂Ω, σ) ⊗ Cn it follows that

|C f | 2 dist(·, ∂Ω) dL n is a Carleson measure in Ω, (2.5.297)
with constant ≤ C f  L2 n−1 (∂Ω,σ)⊗ C .
n

Moreover, if one defines the action of the operator ∇C on arbitrary functions f

belonging to the space BMO(∂Ω, σ) ⊗ Cn according to
 ∫ 
1  x−y 
(∇C) f (x) := ∂x j % ν(y) % f (y) dσ∗ (y) (2.5.298)
ωn−1 |x − y| n
∂∗ Ω 1≤ j ≤n

at each point x ∈ Ω, then for each f ∈ BMO(∂Ω, σ) ⊗ Cn the function (∇C) f
is well defined and for each p ∈ (1, ∞) it follows that
 
(∇C) f  p dist(·, ∂Ω) p−1 dL n (2.5.299)

is a Carleson measure in Ω, in the quantitative sense that there exists some

constant C ∈ (0, ∞), independent of f , such that
∫
1  
sup  (∇C) f  p dist(·, ∂Ω) p−1 dL n
x ∈∂Ω, r >0 σ B(x, r) ∩ ∂Ω B(x,r)∩Ω
p
≤ C f BMO(∂Ω,σ)⊗ Cn . (2.5.300)
 2
Thus, corresponding to p = 2, it follows that (∇C) f  dist(·, ∂Ω) dL n is a
Carleson measure in Ω, for each function f ∈ BMO(∂Ω, σ) ⊗ Cn .
(vi) If ∂Ω is bounded and ∂∗ Ω is upper Ahlfors regular then for each α > 0 there
exists a number N ∈ (0, ∞), depending only on the dimension n, exponent α,
and the upper Ahlfors regularity constant of ∂∗ Ω, with the property that for every
function f ∈ 𝒞α (∂∗ Ω) ⊗ Cn one has

sup Cmax f (x) ≤ N diam(∂∗ Ω) ·  f 𝒞. α (∂ Ω) + N · sup | f |. (2.5.301)
α
∗
x ∈∂∗ Ω ∂∗ Ω

(vii) Assume that ∂Ω is a bounded upper Ahlfors regular set. Then for each function
f ∈ 𝒞α (∂∗ Ω) ⊗ Cn with α ∈ (0, 1) the limit defining C f (x) in (2.5.284)
exists for σ-a.e. point x ∈ ∂∗ Ω. Moreover, after possibly redefining C f on a
σ∗ -nullset, the assignment f → C f thus defined induces a well-defined, linear,
and bounded operator

C : 𝒞α (∂∗ Ω) ⊗ Cn −→ 𝒞α (∂∗ Ω) ⊗ Cn, ∀α ∈ (0, 1), (2.5.302)

with the property that

 464 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

+ 12 on ∂∗ Ω, if Ω is bounded,
C1 ≡ (2.5.303)
− 12 on ∂∗ Ω, if Ω is unbounded.

In particular, all these properties hold whenever Ω is Lebesgue measurable with

a compact Ahlfors regular boundary.
(viii) Suppose Ω is actually open and ∂Ω is a compact Ahlfors regular set. Then for
each arbitrary fixed number κ > 0, the following nontangential boundary trace
formula hods:

if f ∈ 𝒞α (∂∗ Ω) ⊗ Cn with α ∈ (0, 1) then

 κ−n.t.  (2.5.304)
C f ∂Ω = 12 I + C f at σ∗ -a.e. point in ∂∗ Ω.

(ix) In the case when it is also assumed that σ(∂Ω \ ∂∗ Ω) = 0, one may replace ∂∗ Ω
by ∂Ω in the formulation of (2.5.302), (2.5.303), and (2.5.304).
(x) Strengthen the assumptions by requiring that ∂Ω is a UR set. Then for each
p ∈ [1, ∞) there exists a constant C = C(Ω, n, p) ∈ (0, ∞) with the property
that for each f ∈ L p (∂∗ Ω, σ∗ ) ⊗ Cn one has

Cmax f  L p (∂Ω,σ) ≤ C f  L p (∂∗ Ω,σ∗ )⊗ Cn if 1 < p < ∞, (2.5.305)

Cmax f  L 1,∞ (∂Ω,σ) ≤ C f  L 1 (∂∗ Ω,σ∗ )⊗ Cn if p = 1. (2.5.306)

Furthermore, given any function f ∈ L p (∂∗ Ω, σ∗ ) ⊗ Cn with p ∈ [1, ∞) the

limit defining C f (x) in (2.5.284) exists for σ-a.e. x ∈ ∂Ω and the operator C is
well defined, linear, and bounded in the following settings:

C : L p (∂∗ Ω, σ∗ ) ⊗ Cn −→ L p (∂Ω, σ) ⊗ Cn, p ∈ (1, ∞), (2.5.307)

C : L 1 (∂∗ Ω, σ∗ ) ⊗ Cn −→ L 1,∞ (∂Ω, σ) ⊗ Cn . (2.5.308)

(xi) Assume that actually Ω is open and that ∂Ω is a UR set. Then for each fixed
arbitrary number κ > 0, then following nontangential boundary trace formula
holds:
for each f ∈ L p (∂∗ Ω, σ∗ ) ⊗ Cn with p ∈ [1, ∞) one has
 κ−n.t.  (2.5.309)
C f ∂Ω = 12 I + C f at σ∗ -a.e. point on ∂∗ Ω.

Proof The claims in item (i) are seen directly from definitions. To proceed, denote
by (ν1, . . . , νn ) the components of ν. Then from [112, (6.4.2)-(6.4.3)] it follows that
for each y = (y1, . . . , yn ) ∈ ∂ ∗ Ω and x = (x1, . . . xn ) ∈ Rn \ {y} we have
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 465

x−y ν(y), y − x
% ν(y) = (2.5.310)
|x − y| n |x − y| n

1  ν j (y)(xk − yk ) − νk (y)(x j − y j )
n
− · e j % ek .
2 j,k=1 |x − y| n

From (2.5.310), (2.5.282), (2.5.202), and (2.5.256) we then conclude that, for each
σ∗ -measurable function f : ∂∗ Ω → Cn satisfying the integrability condition in
(2.5.281),

1 
n
C f (x) = D f (x) + · e j % ek % R jk f (x), ∀x ∈ Ω̊. (2.5.311)
2 j,k=1

Also, whenever the individual entities are meaningful,

1 
n
C f (x) = K f (x) + · e j % ek % R jk f (x), x ∈ ∂Ω. (2.5.312)
2 j,k=1

Granted these observations, all other claims become consequences of Lemma 2.5.20,
Proposition 2.5.21, Proposition 2.5.22, Proposition 2.5.24, Proposition 2.5.25,
Proposition 2.5.26, and Proposition 2.5.27. 

Incidentally, it is possible to first prove Proposition 2.5.29 (by suitably specializing

Proposition 2.5.16) and then read off the corresponding properties for D, R jk , K,
R jk directly from those of the Cauchy-Clifford operators via (2.5.311)-(2.5.312).
The thrust of the next result is that, up to normalization, the principal-value
Cauchy-Clifford operator is its own inverse when acting on Hölder space.

Proposition 2.5.30 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an open set, with the

property that ∂Ω is a compact Ahlfors regular set, and such that

H n−1 ∂nta Ω \ ∂∗ Ω = 0. (2.5.313)

Then for each α ∈ (0, 1) the principal-value Cauchy-Clifford integral operator C

associated with Ω as in (2.5.284) satisfies

C 2 = 14 I on 𝒞α (∂∗ Ω) ⊗ Cn . (2.5.314)

As a corollary, whenever Ω ⊆ Rn , with n ≥ 2, is an Ahlfors regular domain with

compact boundary for each α ∈ (0, 1) one has

C 2 = 14 I on 𝒞α (∂Ω) ⊗ Cn . (2.5.315)

Proof Introduce σ := H n−1 ∂Ω and select some apperture parameter κ > 0. Also,
fix some α ∈ (0, 1) and pick an arbitrary function f ∈ 𝒞α (∂∗ Ω) ⊗ Cn . In concert
with (2.5.313), Proposition 2.5.29 ensures that if we now define u := C f in Ω then
466 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

u ∈ 𝒞∞ (Ω) ⊗ Cn, Du = 0 in Ω,
κ−n.t.
u∂Ω = ( 12 I + C) f at σ-a.e. point on ∂nta Ω, (2.5.316)
 
and sup u(x) ≤ C f 𝒞α (∂∗ Ω) < +∞.
x ∈Ω

Thanks to [112, (8.2.26)], the last property above implies (bearing in mind that ∂Ω
is compact)
∫
(Nκ u)(x)
Nκ u ∈ L ∞ (∂Ω, σ), hence dσ(x) < +∞. (2.5.317)
∂Ω + |x|
1 n−1

Granted (2.5.325)-(2.5.326) and the current assumptions on Ω, Theorem 1.2.2 ap-

plies and gives that
∫  κ−n.t. 
1 x−y
u(x) = % ν(y) % u∂Ω (y) dσ(y)
ωn−1 ∂∗ Ω |x − y| n

= C( 12 I + C) f (x), ∀x ∈ Ω. (2.5.318)

Taking the nontangential boundary traces of the most extreme sides in (2.5.318) then
yields (on account of (2.5.316) and (2.5.304))

( 12 I + C) f = ( 12 I + C) f ( 12 I + C) f at σ-a.e. point on ∂∗ Ω. (2.5.319)

From this, (2.5.314) readily follows. Finally, for the very last claim in the state-
ment, pertaining to (2.5.315), is justified similarly, with the help of item (ix) in
Proposition 2.5.29. 

To state our next result, the reader is reminded the notation employed for the
commutator [A; B] := AB − BA and anti-commutator { A; B} := AB + BA of two
operators A, B (cf. [112, (6.4.60)-(6.4.61)]).

Proposition 2.5.31 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an open set, with

the property that its topological boundary ∂Ω is a compact Ahlfors regular set, and
such that H n−1 ∂nta Ω \ ∂∗ Ω = 0. Then for each α ∈ (0, 1) the following operator
identities are valid on 𝒞α (∂∗ Ω):

1 
n
K2 − (R jk )2 = 14 I, (2.5.320)
2 j,k=1


n
[R j ; Rk ] = {K; R jk } for each j, k ∈ {1, . . . , n}, (2.5.321)
=1

and
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 467

{Ri j ; Rk } + {Ri ; R jk } = {Rik ; R j } for each i, j, k,  ∈ {1, . . . , n}. (2.5.322)

As a corollary, whenever Ω ⊆ Rn , with n ≥ 2, is an Ahlfors regular domain with

compact boundary the operator identities (2.5.320)-(2.5.322) hold on 𝒞α (∂Ω) for
each α ∈ (0, 1).

Proof This is a direct consequence of Proposition 2.5.30 in the sense that (2.5.320)-
(2.5.322) are seen by untangling the Clifford algebra identity C 2 = 14 I component-
wise, as in [112, Lemma 6.4.2]. 
We next present a version of Proposition 2.5.30 on Lebesgue space.

Proposition 2.5.32 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 2) is an open set such that

∂Ω is a UR set and 
H n−1 ∂nta Ω \ ∂∗ Ω = 0. (2.5.323)
Abbreviate σ∗ := H n−1 ∂∗ Ω and fix p ∈ (1, ∞) arbitrary. Then the principal-value
Cauchy-Clifford integral operator C associated with Ω as in (2.5.284) satisfies

C 2 = 14 I on L p (∂∗ Ω, σ∗ ) ⊗ Cn . (2.5.324)

Equivalently, the operator identities (2.5.320)-(2.5.322) are valid on the space

L p (∂∗ Ω, σ∗ ) ⊗ Cn .

Proof Introduce σ := H n−1 ∂Ω and pick an arbitrary κ > 0. Also, fix p ∈ (1, ∞)
and select an arbitrary function f ∈ L p (∂∗ Ω, σ∗ ) ⊗ Cn . If we now define u := C f
in Ω then, in light of (2.5.323), Proposition 2.5.29 guarantees that

u ∈ 𝒞∞ (Ω) ⊗ Cn, Du = 0 in Ω,
κ−n.t.
u∂Ω = ( 12 I + C) f at σ-a.e. point on ∂nta Ω,
(2.5.325)
Nκ u belongs to the space L p (∂Ω, σ), and
u(x) = O(|x| 1−n ) if Ω is an exterior domain.

In particular, the next-to-last property recorded above implies (thanks to Hölder’s

inequality and [112, Lemma 7.2.1])
∫
(Nκ u)(x)
dσ(x) < +∞, (2.5.326)
∂Ω 1 + |x|
n−1

while the last property in (2.5.325) implies that, in the case when Ω is an exterior
domain,
∫
|u(x)| dL n = o(Rn ) as R → ∞. (2.5.327)
[B(0,2R)\B(0,R)]∩Ω

Having established (2.5.325)-(2.5.327), we may now invoke Theorem 1.2.2 in order

to conclude that
468 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫  κ−n.t. 
1 x−y
u(x) = % ν(y) % u∂Ω (y) dσ(y)
ωn−1 ∂∗ Ω |x − y| n


= C( 12 I + C) f (x), ∀x ∈ Ω. (2.5.328)

Taking the nontangential boundary traces of the most extreme sides in (2.5.328) then
yields (on account of (2.5.325) and (2.5.309))

( 12 I + C) f = ( 12 I + C) f ( 12 I + C) f at σ∗ -a.e. point on ∂∗ Ω. (2.5.329)

This establishes (2.5.324). Finally, the operator identities (2.5.320)-(2.5.322) are

seen by considering the components of the Clifford algebra expressions involved in
the formula (2.5.324), as in [112, Lemma 6.4.2]. 

In the class of UR domains, we have the following version of Proposition 2.5.32

on Muckenhoupt weighted Lebesgue spaces.

Corollary 2.5.33 Let Ω ⊆ Rn , where n ≥ 2, be a UR domain. Set σ := H n−1 ∂Ω

and fix p ∈ (1, ∞) along with w ∈ Ap (∂Ω, σ). Then

C : L p (∂Ω, w) ⊗ Cn −→ L p (∂Ω, w) ⊗ Cn (2.5.330)

is a well-defined, linear, and bounded operator, which satisfies

C 2 = 14 I on L p (∂Ω, w) ⊗ Cn . (2.5.331)

Equivalently, the operator identities (2.5.320)-(2.5.322) are valid on the Mucken-

houpt weighted Lebesgue space L p (∂Ω, w) for each p ∈ (1, ∞) and w ∈ Ap (∂Ω, σ).

Proof The claims regarding (2.5.330) are seen from (2.5.284) and item (10) in
Theorem 2.3.2. Also, since we are currently assuming that Ω is a UR domain,
(2.5.324) becomes
C 2 = 14 I on L p (∂Ω, σ) ⊗ Cn . (2.5.332)
In turn, formula (2.5.331) follows from this, the density of L p (∂Ω, σ) ∩ L p (∂Ω, w)
in the space L p (∂Ω, w), and (2.5.330). 

In the two-dimensional setting, we may recast results on the Cauchy-Clifford

operator in the manner described below.

Corollary 2.5.34 Given a UR domain Ω ⊆ R2 , set σ := H 1 ∂Ω and denote by

ν = (ν1, ν2 ) its geometric measure theoretic outward unit normal. Also, for each
 σ(x)
function f ∈ L 1 ∂Ω, 1+ |x | define
∫  
R f (x) := lim+ ν1 (y)(∂2 EΔ )(y − x) − ν2 (y)(∂1 EΔ )(y − x) f (y) dσ(y),
ε→0
y ∈∂Ω
|x−y |>ε
(2.5.333)
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 469

for σ-a.e. x ∈ ∂Ω, where EΔ is the standard fundamental solution for the two-
dimensional Laplacian (cf. (1.5.56)). Then for each p ∈ (1, ∞) one has
 
K R = −RK and 12 I + K − 12 I + K = R2 on L p (∂Ω, σ). (2.5.334)

Proof These identities are readily implied by the versions of (2.5.320)-(2.5.321)

on Lebesgue spaces (cf. the last claim in Proposition 2.5.32) bearing in mind that,
as seen from (2.5.269), (2.5.268), and (2.5.333), we have R11 = R22 = 0 and
R12 = −R21 = R. 

Let Σ ⊆ Rn (where n ∈ N with n ≥ 2) be a closed set which is countably

rectifiable (of dimension n − 1) and satisfies H n−1 (K ∩ Σ) < +∞ for each compact
set K ⊆ Rn . Abbreviate σ := H n−1 Σ. In this setting, for j ∈ {1, . . . , n} introduce
the boundary-to-domain Riesz transforms acting on functions f ∈ L 1 Σ, 1+ |σ· | n−1
according to
∫
2 xj − yj
R j f (x) := f (y) dσ(y), ∀x ∈ Rn \ Σ, (2.5.335)
ωn−1 Σ |x − y| n
along with their boundary-to-boundary versions
∫
2 xj − yj
R j f (x) := lim+ f (y) dσ(y) for σ-a.e. x ∈ Σ. (2.5.336)
ε→0 ωn−1 |x − y| n
|x−y |>ε
y ∈Σ

Granted the current assumptions, [112, Proposition 5.6.7] guarantees that the latter
principal-value limit exists for σ-a.e. point x ∈ Σ and gives rise to a σ-measurable
function on Σ.
Suppose next that Ω ⊆ Rn (where n ∈ N with n ≥ 2) is an open set such that
∂Ω is a UR set. All previous considerations work with Σ := ∂Ω,  and for each
integrability exponent p ∈ [1, ∞) we now have L p (∂Ω, σ) ⊆ L 1 ∂Ω, 1+ |σ· | n−1 . In
such a setting, the Riesz transforms (2.5.336) (with Σ := ∂Ω) fall directly under the
scope of Theorem 2.3.2 which gives that the operators

R j : L p (∂Ω, σ) −→ L p (∂Ω, σ), p ∈ (1, ∞), (2.5.337)

R j : L 1 (∂Ω, σ) −→ L 1,∞ (∂Ω, σ). (2.5.338)

are well defined, linear, and bounded. Moreover, Theorem 2.4.1 implies that for each
κ > 0 we have the estimates

n
Nκ (R j f ) L p (∂Ω,σ) ≤ C f  L p (∂Ω,σ), 1 < p < ∞, (2.5.339)
j=1


n
Nκ (R j f ) L 1,∞ (∂Ω,σ) ≤ C f  L 1 (∂Ω,σ), (2.5.340)
j=1
470 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

for some finite constant C > 0 depending only on p, κ as well as the Ahlfors
regularity and UR constants of ∂Ω.
In addition, for these Riesz transforms we have the jump-formulas stated in the
proposition below.

Proposition 2.5.35 Assume Ω ⊆ Rn (where n ∈ N with n ≥ 2) is an open set with

the property that ∂Ω is a UR set (in particular, Ω is a set of locally finite perimeter).
Abbreviate σ := H n−1 ∂Ω, σ∗ := H n−1 ∂∗ Ω and denote by ν = (ν1, . . . , νn ) the
geometric measure theoretic outward unit normal to Ω.
Then, for every f ∈ L p (∂∗ Ω, σ∗ ) with 1 ≤ p < ∞ (canonically regarded as a
function in the space L p (∂Ω, σ) by extending it by zero outside ∂∗ Ω to the entire
∂Ω), and every j ∈ {1, . . . , n}, the limit in (2.5.336) exists at σ∗ -almost every
x ∈ ∂∗ Ω, and

lim (R j f )(z) = −ν j (x) f (x) + R j f (x) at σ∗ -a.e. x ∈ ∂∗ Ω. (2.5.341)

z ∈Γκ (x)
z→x

Proof Fix some arbitrary function f ∈ L p (∂∗ Ω, σ∗ ) where 1 ≤ p < ∞, along

with j ∈ {1, . . . , n}. That the limit in (2.5.336) exists σ∗ -a.e. on ∂∗ Ω is a direct
consequence of Theorem 2.3.2. Turning our attention to the jump-formula (2.5.341),
let ∂ν := ν · ∇ denote the normal derivative. Then for each x ∈ Ω and σ∗ -a.e. y ∈ ∂∗ Ω
we may write
1 xj − yj
= (∂j EΔ )(x − y) (2.5.342)
ωn−1 |x − y| n
= −ν j (y)∂ν(y) [EΔ (x − y)]

n  
+ νk (y) νk (y)(∂j EΔ )(x − y) − ν j (y)(∂k EΔ )(x − y) .
k=1

In view of the fact that (cf. (2.5.204))

1 ν(y), y − x
∂ν(y) [EΔ (x − y)] = , (2.5.343)
ωn−1 |x − y| n
the decomposition in (2.5.342) entails

n
R j f = −2D(ν j f ) + 2 R jk (νk f ) in Ω. (2.5.344)
k=1

Granted (2.5.344), Proposition 2.5.25 and Proposition 2.5.27 may be used (with f
replaced by −ν j f and νk f , respectively) in order to compute


n
lim (R j f )(z) = −ν j (x) f (x) − 2K(ν j f )(x) + 2 R jk (νk f )(x) (2.5.345)
z ∈Γκ (x)
z→x k=1
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 471

at σ∗ -a.e. x ∈ ∂∗ Ω. Since the decomposition (2.5.342) also holds for σ∗ -a.e. points
x, y ∈ ∂∗ Ω, from (2.5.342), (2.5.343), (2.5.217), (2.5.268), and (2.5.336) we conclude
that

n
−2K(ν j f ) + 2 R jk (νk f ) = R j f at σ∗ -a.e. point on ∂∗ Ω. (2.5.346)
k=1

At this stage, the jump-formula in (2.5.341) follows from (2.5.345)-(2.5.346). 

Changing course, recall the Clifford algebra ambient (Cn, +, %) from [112, §6.4],
which serves as an enlargement of the Euclidean space Rn to a non-commutative
unitary real algebra. In this regard, we shall require one additional piece of notation.
Specifically, for each s ∈ {1, . . . , n} we agree to let [·]s denote the projection onto
the s-th Euclidean coordinate, i.e.,

[x]s := xs if x = (x1, . . . , xn ) ∈ Rn . (2.5.347)

The lemma below, originating in [159], will play an important role for us shortly.
Here we shall follow the exposition in [120] where the following more refined version
has been established.

Lemma 2.5.36 For any odd, harmonic, homogeneous polynomial P(x), x ∈ Rn

(with n ≥ 2), of degree  ≥ 3, there exist a family Pr s (x), 1 ≤ r, s ≤ n, of harmonic,
homogeneous polynomials, which are either zero or have degree  − 2, as well as a
family of odd, 𝒞∞ functions

kr s : Rn \ {0} −→ Rn → Cn, with r, s ∈ {1, . . . , n}, (2.5.348)

which are homogeneous of degree 1 − n, and for each x ∈ Rn \ {0} satisfy (with the
piece of notation introduced in (2.5.347))

P(x) n
= [kr s (x)]s (2.5.349)
|x| n−1+ r,s=1

as well as (with DR denoting the Dirac operator acting from the right, as in (A.0.41))

−1 ∂ Pr s (x)
(DR kr s )(x) = for all r, s ∈ {1, . . . , n}. (2.5.350)
n +  − 3 ∂ xr |x| n+−3

Moreover, there exists a finite purely dimensional constant cn > 0 such that
   
max sup |kr s | + max sup |∇kr s | ≤ cn 2 P L 1 (S n−1, H n−1 ) . (2.5.351)
1≤r,s ≤n S n−1 1≤r,s ≤n S n−1

Proof Given now an odd, harmonic, homogeneous polynomial P(x) of degree  ≥ 3

in Rn , for each r, s ∈ {1, . . . , n} introduce
472 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

1
Pr s (x) := (∂r ∂s P)(x), ∀x ∈ Rn . (2.5.352)
( − 1)
Then each Pr s is an odd, harmonic, homogeneous polynomial of degree  − 2 in Rn ,
and Euler’s formula for homogeneous functions gives

n
P(x) = xr xs Pr s (x), ∀x ∈ Rn, (2.5.353)
r,s=1

and, for each r, s ∈ {1, . . . , n},

(∇Pr s )(x), x = ( − 2)Pr s (x), ∀x ∈ Rn . (2.5.354)

To proceed, assume first that n ≥ 3 and, for each pair of indices r, s ∈ {1, . . . , n},
define the function kr s : Rn \ {0} −→ Rn → Cn by setting

1  n  P (x) 
rs
kr s (x) := ∂r ∂j ej ∀x ∈ Rn \ {0}. (2.5.355)
(n +  − 3)(n +  − 5) j=1 |x| n+−5

The fact that n,  ≥ 3 ensure that both n +  − 3  0 and n +  − 5  0 so each kr s is

well defined, odd, 𝒞∞ and homogeneous of degree −(n − 1) in Rn \ {0}. In addition,

1   P (x)  
rs
kr s (x) = DR ∂r , ∀x ∈ Rn \ {0}, (2.5.356)
(n +  − 3)(n +  − 5) |x| n+−5

hence for all x ∈ Rn \ {0} we may write

1   P (x)  
rs
(DR kr s )(x) = D2R ∂r
(n +  − 3)(n +  − 5) |x| n+−5
−1   P (x)  
rs
= Δ ∂r
(n +  − 3)(n +  − 5) |x| n+−5
=: I + I I + I I I, (2.5.357)

where
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 473

−1  (ΔP )(x) 
rs
I := ∂r = 0,
(n +  − 3)(n +  − 5) |x| n+−5
−1  % &
I I := ∂r 2 (∇Pr s )(x), ∇ |x| −(n+−5)
(n +  − 3)(n +  − 5)
2  %(∇Pr s )(x), x & 
= ∂r
n+−3 |x| n+−3
2( − 2)  Pr s (x) 
= ∂r ,
n+−3 |x| n+−3
−1  
I I I := ∂r Pr s (x)Δ |x| −(n+−5)
(n +  − 3)(n +  − 5)
− + 3  P (x) 
rs
= ∂r , (2.5.358)
n+−3 |x| n+−3
by the harmonicity of P, (2.5.354), and straightforward algebra. This proves that
(2.5.349) holds when n ≥ 3. Going further, from (2.5.355) and the fact that

n 
n 
n
(∂r Pr s )(x) = (∂s Pr s )(x) = 0 and Prr (x) = 0 (2.5.359)
r=1 s=1 r=1

(as seen from formula (2.5.352) and the harmonicity of P), we deduce that for each
point x ∈ Rn \ {0}

n
1 n  P (x) 
rs
[kr s (x)]s = ∂r ∂s
r,s=1
(n +  − 3)(n +  − 5) r,s=1 |x| n+−5

1 n
= Pr s (x)∂r ∂s |x| −(n+−5)
(n +  − 3)(n +  − 5) r,s=1

−1 n  δ xr xs 
rs
= Pr s (x) − (n +  − 3)
n +  − 3 r,s=1 |x| n+−3 |x| n+−1

P(x)
= . (2.5.360)
|x| n−1+
This establishes (2.5.350) for n ≥ 3. Moving on, for each γ ∈ N0n , interior estimates
for the harmonic function P give
474 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫ 
∂γ P L ∞ (S n−1, H n−1 ) ≤ cn,γ |P(x)| dxr n−1+ dr dH n−1 (ω)
B(0,2)
∫ ∫ 2 
= cn,γ |P(ω)| r n−1+ dr dH n−1 (ω)
S n−1 0

2
= cn,γ P L 1 (S n−1, H n−1 ), (2.5.361)
n+
where we have also used the fact that P is homogeneous of degree . The estimates
in (2.5.351) now readily follow on account of (2.5.355), (2.5.352), and (2.5.361).
To treat the two-dimensional case, first we observe that if Q m (x) is an arbitrary
homogeneous polynomial of degree m ∈ N0 in Rn with n ≥ 2 and λ > 0 then

Q m (x)
is a tempered distribution in Rn . (2.5.362)
|x| n+m−λ

If, in addition, Q m (x) is harmonic and λ < n then (cf. [166, p. 73]) also
 Q (x)  Q m (ξ)
m
Fx→ξ = γn,m,λ m+λ as tempered distributions in Rn, (2.5.363)
|x| n+m−λ |ξ |

where Fx→ξ is an alternative notation for the Fourier transform in Rn from (1.4.14)
and
Γ(m/2 + λ/2)
γn,m,λ := (−1)3m/2 π n/2 2λ . (2.5.364)
Γ(m/2 + n/2 − λ/2)
Pick now an odd, harmonic, homogeneous polynomial P(x) of degree  ≥ 3 in R2
and define Pr s for r, s ∈ {1, . . . , n} as in (2.5.352). Hence, once again each Pr s is an
odd, harmonic, homogeneous polynomial of degree  − 2 in R2 , and (2.5.353) holds.
Moreover, (2.5.363) used for n = 2, m =  − 2, λ = 1, and Q m = Pr s yields

Pr s (x)  P (ξ) 
−1 rs
= −(−1) 3/2
2π Fξ→x . (2.5.365)
|x| −1 |ξ | −1

Now, for each r, s ∈ {1, 2} define the function kr s : R2 \ {0} −→ R2 → C2 by

setting


2  Pr s (ξ) 
−1
kr s (x) := (−1)3/2 2π Fξ→x ξr ξ j +1 e j , ∀x ∈ R2 \ {0}. (2.5.366)
j=1
|ξ |

By (2.5.362) used with n = 2, m = , λ = 1, and Q m (ξ) = ξr ξ j Pr s (ξ), it follows

(ξ)
that ξr ξ j P|ξr s| +1 is a tempered distribution in R2 . Consequently, kr s in (2.5.366)
is meaningfully defined and, from [109, Proposition 4.60, p. 143], we deduce that
kr s ∈ 𝒞∞ (R2 \ {0}). Also, based on standard properties of the Fourier transform
(cf., e.g., [109, Chapter 4, p. 117]) it follows that kr s is odd and homogeneous of
degree −1 in R2 \ {0}. In addition,
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 475


2  Pr s (ξ) 
−1
(DR kr s )(x) = (−1)3/2 2π ∂xi Fξ→x ξr ξ j +1 e j % ei
i, j=1
|ξ |

√  2  Pr s (ξ) 
−1
= −1(−1)3/2 2π Fξ→x ξr ξ j ξi +1 e j % ei
i, j=1
|ξ |

=: I + I I, (2.5.367)

where I and I I are the pieces produced by summing up over j = i and j  i,

respectively. Since in the latter scenario ξi ξ j = ξ j ξi while e j % ei = −ei % e j it
follows that I I = 0. Given that e j % e j = −1 for each j ∈ {1, 2}, we conclude that

√ 2  Pr s (ξ) 
−1
(DR kr s )(x) = − −1(−1)3/2 2π Fξ→x ξr ξ j2 +1
j=1
|ξ |
√  P (ξ) 
−1 rs
= − −1(−1)3/2 2π Fξ→x ξr −1
|ξ |
  P (ξ)   P (x) 
−1 rs rs
= −(−1)3/2 2π ∂xr Fξ→x −1
= ∂xr , (2.5.368)
|ξ | |x| −1
where the last step uses (2.5.365). Hence, (2.5.349) holds when n = 2. Finally, from
(2.5.355), (2.5.353), and (2.5.363) (used for P) we deduce that for each x ∈ R2 \ {0}
we have

2 
2  Pr s (ξ) 
−1
[kr s (x)]s = (−1)3/2 2π Fξ→x ξr ξs +1
r,s=1 r,s=1
|ξ |
 P(ξ)  P(x)
−1
= (−1)3/2 2π Fξ→x = +1 . (2.5.369)
|ξ | +1 |x|
This establishes (2.5.350) when n = 2.
At this stage, there remains to justify (2.5.351) in the case n = 2. To this end,
pick ψ ∈ 𝒞∞ c (R ) with 0 ≤ ψ ≤ 1, ψ = 1 on B(0, 1) and ψ = 0 on R \ B(0, 2). Fix
2 2

r, s, j ∈ {1, 2} and abbreviate u(ξ) := ξr ξ j Pr s (ξ)/|ξ | +1 for ξ ∈ R \ {0}. Then u is

2

locally integrable and defines a tempered distribution in R2 . Hence, for each α ∈ N20
with |α| = 2 and ξ ∈ B(0, 1) we may write
   % √ & %  √ &
Fx→ξ ψ(x)∂ α u(x)  =  ψ∂ α u, e− −1 ξ, ·  =  u, ∂ α ψe− −1 ξ, · 
∫ ∫
≤C |u(x)| dx ≤ C |Pr s (ω)| dH 1 (ω)
B(0,2) S1

≤ C2 P L 1 (S 1, H 1 ), (2.5.370)

and
476 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
  
Fx→ξ (1 − ψ(x))∂ α u(x)  ≤ (1 − ψ)∂ α u L 1 (R2, L 2 ) ≤ |∂ α u(x)| dx
R2 \B(0,1)
∫
≤C |∂ α u(ω)| dH 1 (ω)
S1

≤ C2 P L 1 (S 1, H 1 ) . (2.5.371)

Collectively, (2.5.370) and (2.5.371) give that, for each α ∈ N20 with |α| = 2 and
ξ ∈ B(0, 1),
        
Fx→ξ ∂ α u(x)  ≤ Fx→ξ ψ(x)∂ α u(x)  + Fx→ξ (1 − ψ(x))∂ α u(x) 

≤ C2 P L 1 (S 1, H 1 ), (2.5.372)

hence for each ξ ∈ B(0, 1) we have

   2
 2  2
  
|ξ | 2 -
u(ξ) = ξ - 
i u(ξ) =
Fx→ξ ∂ 2 u(x)  ≤ C2 P L 1 (S 1, H 1 ) .
i (2.5.373)
i=1 i=1
 
In particular, kr s  L ∞ (S 1, H 1 ) ≤ C sup |ξ |=1 -
u(ξ) ≤ C2 P L 1 (S 1, H 1 ) . A similar circ-
le of ideas also yields ∇kr s  L ∞ (S 1, H 1 ) ≤ C2 P L 1 (S 1, H 1 ) . This proves (2.5.351)
in the case n = 2 and completes the proof of Lemma 2.5.36. 

We are now ready to establish jump-formulas for integral operators associated

with kernels which have the special algebraic structure stipulated in (2.5.85).

Proposition 2.5.37 Suppose Ω ⊆ Rn (where n ∈ N with n ≥ 2) is a nonempty

open set with the property that ∂Ω is a UR set. In particular, Ω is a set of locally
finite perimeter. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure
theoretic outward unit normal to Ω. Also, consider an odd, harmonic, homogeneous
polynomial P(x) of degree  ∈ N in Rn and define

P(x)
k(x) := for each x ∈ Rn \ {0}. (2.5.374)
|x| n−1+
In relation to the kernel k and the domain Ω, associate the integral operators
∫
(T f )(x) := k(x − y) f (y) dσ(y), ∀ x ∈ Ω, (2.5.375)
∂Ω

and ∫
T f (x) := lim+ k(x − y) f (y) dσ(y), ∀ x ∈ ∂Ω. (2.5.376)
ε→0
y ∈∂Ω
|x−y |>ε
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 477

Then for each fixed κ > 0, and each f ∈ L p (∂Ω, σ) with 1 ≤ p < ∞, the
jump-formula 
lim (T f )(z) = √1 -
k ν(x) f (x) + T f (x) (2.5.377)
z ∈Γκ (x) 2 −1
z→x

is valid at σ-a.e. x ∈ ∂∗ Ω.

Proof That for each f ∈ L p (∂Ω, σ) the limit in (2.5.376) exists for σ-a.e. x ∈ ∂Ω,
and the induced operators

T : L p (∂Ω, σ) −→ L p (∂Ω, σ), p ∈ (1, ∞), (2.5.378)

T : L 1 (∂Ω, σ) −→ L 1,∞ (∂Ω, σ), (2.5.379)

are well-defined, linear, and bounded, follows from Theorem 2.3.2. Moreover, for
each κ > 0 and f ∈ L p (∂Ω, σ) with p ∈ [1, ∞), Theorem 2.4.1 also ensures that

Nκ (T f ) L p (∂Ω,σ) ≤ C f  L p (∂Ω,σ), 1 < p < ∞, (2.5.380)

Nκ (T f ) L 1,∞ (∂Ω,σ) ≤ C f  L 1 (∂Ω,σ) . (2.5.381)

Pressing on, fix some κ > 0 and an exponent p ∈ [1, ∞) arbitrary. Also, abbreviate
σ∗ := H n−1 ∂∗ Ω. In a first stage, we propose to prove by induction on  ∈ N that
given an integral kernel k as in (2.5.374) for an odd, harmonic,
homogeneous polynomial P(x) of degree  in Rn , and given any
f ∈ L p (∂∗ Ω, σ∗ ), canonically regarded as a function in L p (∂Ω, σ) (2.5.382)
by extending it by zero outside ∂∗ Ω, the jump-formula (2.5.377)
holds at σ∗ -a.e. point x ∈ ∂∗ Ω.
The case  = 1 has been already dealt with in Proposition 2.5.35, so we may assume
that  ≥ 3 and that the corresponding jump-formula is true for any kernel k associated
with an odd harmonic homogeneous polynomial P of degree  − 2 as in (2.5.374).
Now,
  let P, k be as in the statement of the current proposition and recall the kernels
kr s 1≤r,s ≤n introduced in relation to these as in Lemma 2.5.36. For future purposes
it will be convenient to set, for each r, s ∈ {1, . . . , n},

 − 1 Pr s (x)
k r s (x) := for each x ∈ Rn \ {0}, (2.5.383)
n +  − 3 |x| n+−3

where Pr s are the harmonic homogeneous polynomials in Rn which are either zero
or have degree  − 2, associated with the given function k as in Lemma 2.5.36.
For each r, s ∈ {1, . . . , n} then define
∫
(Tr s f )(x) := kr s (x − y) f (y) dσ∗ (y), ∀ x ∈ Ω. (2.5.384)
∂∗ Ω

On the one hand, identity (2.5.349) gives

478 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets


n
Tf = [Tr s f ]s for every f ∈ L p (∂∗ Ω, σ∗ ). (2.5.385)
r,s=1

On the other hand, for every f ∈ L p (∂∗ Ω, σ∗ ) ⊗ Cn with 1 ≤ p < ∞ from (2.3.15)
in Theorem 2.3.2 we know that there exists a set A f ⊆ ∂Ω satisfying H n−1 (A f ) = 0
and such that for each r, s ∈ {1, . . . , n} the limits
∫ ∫
lim+ kr s (x−y) f (y) dσ∗ (y) and lim+ k r s (x − y) f (y) dσ∗ (y)
ε→0 ε→0
y ∈∂∗ Ω y ∈∂∗ Ω
|x−y |>ε |x−y |>ε
(2.5.386)
exist for each point x ∈ ∂∗ Ω \ A f .

From (2.5.385)-(2.5.386) it is then clear that it suffices to show that, having fixed
r, s ∈ {1, . . . , n} and f ∈ L p (∂∗ Ω, σ∗ ) with 1 ≤ p < ∞, at σ∗ -a.e. x ∈ ∂∗ Ω we have

lim (Tr s f )(z) = √1 - kr s ν(x) f (x)
z ∈Γκ (x) 2 −1
z→x
∫
+ lim+ kr s (x − y) f (y) dσ∗ (y). (2.5.387)
ε→0
y ∈∂∗ Ω
|x−y |>ε

In fact, by arguing as in the first part of the proof of Proposition 2.5.25 (a process
in which the properties recorded in (2.5.378)-(2.5.381) are relevant) we see that it
suffices to prove (2.5.387) in the case when

f = ν % g with g ∈ Lipc (∂∗ Ω) ⊗ Cn arbitrary. (2.5.388)

To this end, fix an arbitrary point


x ∈ ∂ ∗ Ω \ A f ∩ ∂nta Ω (2.5.389)

and let Ox ⊂ (0, 1) be associated with the point x and the set E := Ω (a choice
which places the fixed point x in ∂ ∗ E) as in Lemma 2.5.8. Also, let Nx ⊂ (0, ∞) be
an L 1 -measurable set such that L 1 (Nx ) = 0 associated with the point x and the set
E := Rn \ Ω as in [112, Lemma 5.7.2]. In this scenario, paralleling the treatment in
(2.5.191) we decompose
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 479

lim (Tr s f )(z)

z ∈Γκ (x)
z→x
∫
= lim lim kr s (z − y) f (y) dσ∗ (y)
ε ∈(O x \N x ) z ∈Γκ (x)
ε→0 z→x |x−y |>ε
y ∈∂∗ Ω
∫
1
+ lim lim kr s (z − y) % ν(y) % [g(y) − g(x)] dσ∗ (y)
ε ∈(O x \N x ) z ∈Γκ (x) ωn−1
ε→0 z→x |x−y | ≤ε
y ∈∂∗ Ω

2 ∫ 5
3 6
3
+ 3 lim lim kr s (z − y) % ν(y) dσ∗ (y)66 % g(x)
3 ε ∈(O x \N x ) z ∈Γ κ (x) 6
ε→0 z→x |x−y | ≤ε
4 y ∈∂∗ Ω 7
=: I1 + I2 + I3 . (2.5.390)

Bearing in mind (2.5.386), as in the case of (2.5.191) (with k! given by (2.5.129)) we

obtain
∫
I1 = lim+ kr s (x − y) f (y) dσ∗ (y) and I2 = 0. (2.5.391)
ε→0
y ∈∂∗ Ω
|x−y |>ε

There remains to treat I3 and we begin by rewriting the last integral in (2.5.390) in
a more convenient fashion. To accomplish this, To accomplish this, fix an arbitrary
point z ∈ Γκ (x) then consider the vector field

F! = Fj 1≤ j ≤n with Cn -valued components given by
(2.5.392)
Fj (y) := kr s (z − y) % e j − k r s (z − y)δr j for each y ∈ Rn \ {z},

(no sum over r) where the kernels k r s are as in (2.5.383). Then, thanks to (2.5.392),
(2.5.350), and (2.5.383), for each y ∈ Rn \ {z} we have

n

! =
divF(y) ∂j Fj (y)
j=1

= −(DR kr s )(z − y) + (∂r k r s )(z − y) = 0. (2.5.393)

On account of this, [112, Lemma 5.7.2] used with the set E := Rn \ Ω (which
ensures that the vector field F! is divergence-free near E, and that the geometric
measure theoretic outward unit normal to E is −ν), and the point x as in (2.5.389),
guarantees, in view of the current choice of the L 1 -nullset Nx ⊂ (0, ∞), that for
each ε ∈ (0, ∞) \ Nx we have
480 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
kr s (zi − y) % ν(y) dσ∗ (y)
|x−y | ≤ε
y ∈∂∗ Ω
∫
= k r s (zi − y)νr (y) dσ∗ (y)
|x−y | ≤ε
y ∈∂∗ Ω
∫ y − x
+ kr s (zi − y) % dH n−1 (y)
ε
(∂B(x,ε))\Ω
∫ y − x 
r r
− k r s (zi − y) dH n−1 (y). (2.5.394)
ε
(∂B(x,ε))\Ω

To proceed, for each ε > 0 define

∂ − B(x, ε) := {y ∈ ∂B(x, ε) : ν(x) · (y − x) > 0} (2.5.395)

and

W(x, ε) := ∂ − B(x, ε) (∂B(x, ε)) \ Ω . (2.5.396)

Then (2.5.104) in the proof of Lemma 2.5.8 presently gives


H n−1 W(x, ε)
lim = 0. (2.5.397)
O x ε→0 ε n−1
Also, for each ε > 0 fixed, [112, (6.4.35)] and the positive homogeneity of the kernel
kr s permit us to estimate
 ∫
 y − x

 kr s (x − y) % dH n−1 (y)
 ε
(∂B(x,ε))\Ω

∫ 
y − x 

− kr s (x − y) % dH n−1
(y)
ε 
∂− B(x,ε)
∫
≤ |kr s (x − y)| dH n−1 (y)
W (x,ε)

  H n−1 W(x, ε)
≤ sup |kr s | . (2.5.398)
S n−1 ε n−1

Based on (2.5.398) and (2.5.397) we may therefore write

2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 481
∫ y − x
lim lim kr s (z − y) % dH n−1 (y)
ε ∈(O x \N x ) z ∈Γκ (x) ε
ε→0 z→x (∂B(x,ε))\Ω

∫ y − x
= lim kr s (x − y) % dH n−1 (y)
ε ∈(O x \N x ) ε
ε→0 (∂B(x,ε))\Ω
∫ y − x
= lim kr s (x − y) % dH n−1 (y)
ε ∈(O x \N x ) ε
ε→0 ∂− B(x,ε)
∫
= kr s (x − 
y ) % (
y − x) dH n−1 (
y) (2.5.399)
∂− B(x,1)

where, the last equality, is seen by making the change of variables

y − x
∂ − B(x, ε)  y −→ y := x + ∈ ∂ − B(x, 1), (2.5.400)
ε
while bearing in mind the positive homogeneity of the kernel kr s . Moreover, in a
very similar fashion, we have
∫ y − x 
r r
lim lim k r s (z − y) dH n−1 (y)
ε ∈(O x \N x ) z ∈Γκ (x) ε
ε→0 z→x (∂B(x,ε))\Ω

∫
= k r s (x − 
y )(
yr − xr ) dH n−1 (
y ). (2.5.401)
∂− B(x,1)

In summary, the above argument (which, starting with the decomposition in

(2.5.390), has produced (2.5.391), (2.5.394), (2.5.399), and (2.5.401)) gives (keeping
in mind (2.5.389) and [112, (8.8.52)]) that for σ∗ -a.e. x ∈ ∂∗ Ω we have (after a slight
adjustment in notation)
482 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

lim (Tr s f )(z)

z ∈Γκ (x)
z→x
∫
= lim+ kr s (x − y) f (y) dσ∗ (y)
ε→0
y ∈∂∗ Ω
|x−y |>ε

2 ∫ 5
3 6
3
+ 3 lim lim k (z − y)νr (y) dσ∗ (y)66 % g(x)
rs
3ε ∈(O x \N x ) z ∈Γ κ (x) 6
ε→0 z→x |x−y | ≤ε
4 y ∈∂∗ Ω 7

2 ∫ 5
+ 33 kr s (x − y) % (y − x) dH n−1 (y)66 % g(x)
4∂− B(x,1) 7

2 ∫ 5
+ 33 k r s (x − y)(yr − xr ) dH n−1 (y)66 % g(x). (2.5.402)
4∂− B(x,1) 7
To proceed, for each h ∈ L p (∂∗ Ω, σ∗ )
introduce
∫
(T r s h)(x) := k r s (x − y)h(y) dσ∗ (y), ∀x ∈ Ω. (2.5.403)
∂∗ Ω

Two notable things about this family of integral operators are as follows. On the one
hand, given the nature of the kernels k r s from (2.5.383) (in particular, recall that
Pr s is a harmonic homogeneous polynomial in Rn which is either zero or has degree
 −2), the current induction hypothesis pertaining to the validity of the jump-formula
(2.5.377) allows us to write that, for each function h ∈ L p (∂∗ Ω, σ∗ ),

lim T r s (νr h)(z) = √1 k.r s ν(x) ν (x)h(x)
r
z ∈Γκ (x) 2 −1
z→x
∫
+ lim+ k r s (x − y)νr (y)h(y) dσ∗ (y) (2.5.404)
ε→0
y ∈∂∗ Ω
|x−y |>ε

at σ∗ -almost every x ∈ ∂∗ Ω. On the other hand, by breaking up the domain of

integration in (2.5.403) into ∂∗ Ω \ B(x, ε) and (∂∗ Ω) ∩ B(x, ε), then passing to the
limit Γκ (x)  z → x and ε → 0+ , we obtain that for every function h ∈ L p (∂∗ Ω, σ∗ )
we have
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 483
∫
lim T r s (νr h)(z) = lim+ k r s (x − y)νr (y)h(y) dσ∗ (y)
z ∈Γκ (x) ε→0
z→x y ∈∂∗ Ω
|x−y |>ε
∫
+ lim+ lim k r s (z − y)νr (y)h(y) dσ∗ (y) (2.5.405)
ε→0 z ∈Γκ (x)
z→x |x−y | ≤ε
y ∈∂∗ Ω

for σ∗ -a.e. x ∈ ∂∗ Ω. By comparing (2.5.404) with (2.5.405) (and choosing h to be

identically 1 near the point x) we eventually arrive at the conclusion that
∫

lim+ lim k r s (z − y)νr (y) dσ∗ (y) = √1 k.
r s ν(x) ν (x)
r (2.5.406)
ε→0 z ∈Γκ (x) 2 −1
z→x |x−y | ≤ε
y ∈∂∗ Ω

for σ∗ -a.e. x ∈ ∂∗ Ω. In turn, from (2.5.406), (2.5.402), and the fact that g = −ν % f at
σ∗ -a.e. point on ∂∗ Ω (as seen from (2.5.388) and [112, (6.4.59)]), we finally conclude
that for σ∗ -a.e. x ∈ ∂∗ Ω we have
∫
lim (Tr s f )(z) = lim+ kr s (x − y) f (y) dσ∗ (y)
z ∈Γκ (x) ε→0
z→x y ∈∂∗ Ω
|x−y |>ε

− √1 k.
rs ν(x) νr (x) % ν(x) % f (x)
2 −1

2 ∫ 5
− 33 kr s (x − y) % (y − x) dH n−1 (y)66 % ν(x) % f (x)
4∂− B(x,1) 7

2 ∫ 5
− 33 k r s (x − y)(yr − xr ) dH n−1 (y)66 % ν(x) % f (x).
4∂− B(x,1) 7
(2.5.407)

In short, for every f ∈ L p (∂∗ Ω, σ∗ ) with 1 ≤ p < ∞, at σ∗ -a.e. x ∈ ∂∗ Ω we have

∫
lim (Tr s f )(z) = αr s (x) % f (x) + lim+ kr s (x − y) f (y) dσ∗ (y) (2.5.408)
z ∈Γκ (x) ε→0
z→x y ∈∂∗ Ω
|x−y |>ε

where, at σ∗ -a.e. x ∈ ∂∗ Ω, the coefficient αr s (x) is given by

484 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

αr s (x) = − √1 k.
rs ν(x) νr (x) % ν(x)
2 −1

2 ∫ 5
− 33 kr s (x − y) % (y − x) dH n−1 (y)66 % ν(x)
4∂− B(x,1) 7

2 ∫ 5
− 33 k r s (x − y)(yr − xr ) dH n−1 (y)66 % ν(x). (2.5.409)
4∂− B(x,1) 7
Hence, in order to fully justify (2.5.387), it remains to show that

αr s (x) = √1 - kr s ν(x) at σ∗ -a.e. x ∈ ∂∗ Ω. (2.5.410)
2 −1

A key characteristic of αr s (x), visible from (2.5.409), is that, as far as the geometry
is concerned, this quantity depends only on the vector ν(x) and not on the set Ω
itself. Consequently, in the process of computing the actual value of αr s (x0 ) for a
given point x0 ∈ ∂ ∗ Ω, using (2.5.408), we may replace Ω by any other open set in Rn
with a UR boundary and which has the same geometric measure theoretic outward
unit normal ν(x0 ) at x0 as Ω does. Of course, the simplest such replacement is a
suitably rotated and translated half-space, namely,
 := {x ∈ Rn : ν(x0 ) · (x − x0 ) < 0}.
Ω (2.5.411)

In this latter scenario, we agree to decorate with tilde objects associated with Ω 
in the same manner as their plain counterparts have been introduced in relation to

the original set Ω. With this convention in mind, for every f ∈ L p (∂∗ Ω, σ∗ ) where
1 ≤ p < ∞, the jump-formula
r s f )(z)
lim (T (2.5.412)
z ∈
Γκ (x)
z→x
∫

= √1 -kr s ν(x0 ) f (x) + lim+ kr s (x − y) f (y) d
σ∗ (y),
2 −1 ε→0

y ∈∂∗ Ω
|x−y |>ε

at   , is well known (see, e.g., [109, Corollary 4.81, p. 174] and the
σ∗ -a.e. x ∈ ∂∗ Ω
discussion in [129]). Altogether, from (2.5.412) and (2.5.408) we then conclude that
(2.5.410) holds. This justifies (2.5.387) and finishes the proof of (2.5.382).
Having established (2.5.382), the goal now is to prove the validity of the jump-
formula (2.5.377) at σ∗ -a.e. point in ∂∗ Ω in the case when f is an arbitrary function
in L p (∂Ω, σ). With this goal in mind, pick some arbitrary f ∈ L p (∂Ω, σ), fix a
point x0 ∈ ∂Ω and, for each m ∈ N, define

fm := f · 1B(x0,m)∩∂Ω ∈ L p (∂Ω, σ). (2.5.413)

2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 485

Lebesgue’s Dominated Convergence Theorem then ensures that

fm −→ f in L p (∂Ω, σ) as m → ∞. (2.5.414)

In turn, from (2.5.414), Theorem 2.3.2, and Theorem 2.4.1 we conclude that, as
m → ∞,

L p (∂Ω, σ) if 1 < p < ∞,
T fm −→ T f in (2.5.415)
L 1,∞ (∂Ω, σ) if p = 1,

and 
 L p (∂Ω, σ) if 1 < p < ∞,
Nκ T ( f − fm ) → 0 in (2.5.416)
L 1,∞ (∂Ω, σ) if p = 1.
In view of [112, Proposition 6.2.7], by eventually passing to a subsequence there is
no loss of generality in assuming that there exists A ⊆ ∂Ω, which is σ-measurable
and satisfies σ(A) = 0, with the property that, as m → ∞, we also have

fm (x) → f (x), (T fm )(x) → (T f )(x), and

 (2.5.417)
Nκ T ( f − fm ) (x) → 0, for each x ∈ ∂Ω \ A.

To proceed, make the working assumption that

for each m ∈ N there exists some σ∗ -nullset Am ⊆ ∂∗ Ω

with the property that for each x ∈ ∂∗ Ω \ Am we have
 (2.5.418)
lim (T fm )(z) = √1 - k ν(x) fm (x) + (T fm )(x).
z ∈Γκ (x) 2 −1
z→x

Introduce   
 := ∂∗ Ω ∩ A ∪
A Am (2.5.419)
m∈N

so that
A  = 0.
 ⊆ ∂∗ Ω is σ∗ -measurable and satisfies σ∗ ( A) (2.5.420)
 and each m ∈ N we may write
Then, for each fixed x ∈ ∂∗ Ω \ A
486 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
   
 
lim sup (T f )(z) − √1 - k ν(x) f (x) + (T f )(x) 
2 −1
z ∈Γκ (x)
z→x
 
 
≤ lim sup  T ( f − fm ) (z)
z ∈Γκ (x)
z→x
   
 √1 -

+ lim sup  T fm (z) − k ν(x) fm (x) + (T fm )(x) 
2 −1
z ∈Γκ (x)
z→x
     
 √1 - √1 -

+ k ν(x) fm (x) + (T fm )(x) − k ν(x) f (x) + (T f )(x) 
2 −1 2 −1

=: Im (x) + IIm (x) + IIIm (x). (2.5.421)

Note that
  
0 ≤ Im (x) ≤ Nκ T ( f − fm ) (x) (2.5.422)

which, in concert with the last property in (2.5.417), implies

lim Im (x) = 0. (2.5.423)

m→∞

Moreover, thanks to the working assumption (2.5.418), we have

IIm (x) = 0 for each m ∈ N, (2.5.424)

whereas the first line in (2.5.417) entails

lim IIIm (x) = 0. (2.5.425)

m→∞

From (2.5.420)-(2.5.425) we then conclude that, for any given f ∈ L p (∂Ω, σ) with
1 ≤ p < ∞, the jump-formula (2.5.377) is valid at σ∗ -a.e. point x ∈ ∂∗ Ω.
At this stage, there remains to dispense with the extra working assumption made in
(2.5.418) where, as before, f is an arbitrary function in L p (∂Ω, σ) with 1 ≤ p < ∞.
To this end, fix m ∈ N arbitrary and introduce

gm := fm ∂∗ Ω ∈ L p (∂∗ Ω, σ∗ ). (2.5.426)

gm the extension of gm by zero outside ∂∗ Ω to the entire ∂Ω,

Let us also denote by 
and define
hm := fm − 
gm ∈ L p (∂Ω, σ). (2.5.427)
From (2.5.382) we know that for σ∗ -a.e. x ∈ ∂∗ Ω we have
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 487
  
lim T
gm (z) = √1 -k gm (x) + T
ν(x)  gm (x)
z ∈Γκ (x) 2 −1
z→x
 
= √1 -k ν(x) fm (x) + Tgm (x). (2.5.428)
2 −1

As regards the function defined in (2.5.427), note that


hm = 0 on ∂∗ Ω ∪ ∂Ω \ B(x0, m) = ∂Ω \ E (2.5.429)

where 
E := ∂Ω \ ∂∗ Ω ∩ B(x0, m). (2.5.430)
Observe that the set E is σ-measurable and, thanks to the fact that ∂Ω is upper
Ahlfors regular, satisfies

σ(E) ≤ σ ∂Ω ∩ B(x0, m) < +∞. (2.5.431)

In view of [112, Lemma 3.4.13] and the upper Ahlfors regularity of ∂Ω, the hypothe-
ses in item (3) of [112, Proposition 3.4.15] are satisfied by the ambient X := ∂Ω
(a closed set regarded as a topological space in its own right when equipped with
the topology inherited from Rn ), and the measure σ which is Borel-regular on X
(thanks to [112, Lemma 3.6.4]). Since E ⊆ ∂Ω is σ-measurable, the inner-regularity
property [112, (3.4.47)] presently implies
 
σ(E) = sup σ(C) : C closed subset of ∂Ω contained in E . (2.5.432)

Bearing in mind that σ(E) < +∞, we conclude from (2.5.432) that there exists a
nested sequence {C j } j ∈N of closed subsets of ∂Ω with the property that

C j ⊆ E for each j ∈ N, and

(2.5.433)
1C j  1E as j → ∞, for σ-a.e. point in ∂Ω.

Hence, if we define

hm, j := hm · 1C j ∈ L p (∂Ω, σ) for each j ∈ N, (2.5.434)

then from (2.5.433), Lebesgue’s Dominated Convergence Theorem, and (2.5.429)

we conclude that

hm, j −→ hm · 1E = hm in L p (∂Ω, σ) as j → ∞. (2.5.435)

In turn, with the help of Theorem 2.3.2 and Theorem 2.4.1, from this we conclude
that, as j → ∞,

L p (∂Ω, σ) if 1 < p < ∞,
T hm, j → T hm in (2.5.436)
L 1,∞ (∂Ω, σ) if p = 1,
488 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

and

 L p (∂Ω, σ) if 1 < p < ∞,
Nκ T (hm − hm, j ) → 0 in (2.5.437)
L 1,∞ (∂Ω, σ) if p = 1.

As in the past, by eventually passing to a subsequence there is no loss of generality

in assuming that

(T hm, j )(x) → (T hm )(x) and Nκ T (hm − hm, j ) (x) → 0
(2.5.438)
as j → ∞, for σ-a.e. point x ∈ ∂Ω.

Going further, recall the set Aκ (∂Ω) defined in (A.0.2). We claim that

lim (T hm, j )(z) = (T hm, j )(x)

z ∈Γκ (x)
z→x (2.5.439)
for each j ∈ N and each x ∈ ∂∗ Ω ∩ Aκ (∂Ω).

To prove this claim, fix j ∈ N and x ∈ ∂∗ Ω ∩ Aκ (∂Ω) arbitrary. In particular, x  C j .

Since C j is a closed set, this implies that there exists some ε j (x) > 0 with the
property that 
B x, ε j (x) ∩ C j = . (2.5.440)
Let us also note that for any z ∈ Γκ (x) and any y ∈ ∂Ω we have

|x − y| ≤ |x − z| + |z − y| < (1 + κ) dist(z, ∂Ω) + |z − y|

≤ (1 + κ)|z − y| + |z − y| = (2 + κ)|z − y|, (2.5.441)

hence (bearing in mind the homogeneity property of the kernel k),

  (2 + κ)n−1
|k(z − y)| ≤ sup |k | · =: Fx (y). (2.5.442)
S n−1 |x − y| n−1

On the one hand, from (2.5.440) and the definition of Fx (y) just given above we see
that
  (2 + κ)n−1
sup |Fx (y)| ≤ sup |k | · < +∞, (2.5.443)
y ∈C j S n−1 ε j (x)n−1

while on the other hand for each integrability exponent q ∈ (1, ∞) we may rely on
the definition of Fx from (2.5.442) (which among other things, shows that Fx is
continuous), (2.5.440), and [112, Lemma 7.2.1] to estimate
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 489
∫
|Fx (y)| q dσ(y) (2.5.444)
Cj
 q ∫
1
≤ sup |k | (2 + κ)q(n−1) dσ(y) < +∞.
S n−1 ∂Ω\B(x,ε j (x)) |x − y| q(n−1)

Collectively (2.5.443)-(2.5.444) imply

Fx ∈ L q (C j , σ). (2.5.445)
1<q ≤∞

From (2.5.442), (2.5.445), and (2.5.434) (keep in mind that 1 ≤ p < ∞), we conclude
that for each point z ∈ Γκ (x) we have
    
 k(z − y)hm, j (y) ≤ Fx (y) hm, j (y) for every y ∈ C j ,
   (2.5.446)
and the function Fx  hm, j  belongs to L 1 (C j , σ).

For each point x ∈ ∂∗ Ω ∩ Aκ (∂Ω) we are now prepared to compute

∫

lim T hm, j (z) = lim k(z − y)hm, j (y) dσ(y)
z ∈Γκ (x) z ∈Γκ (x) ∂Ω
z→x z→x
∫
= lim k(z − y)hm, j (y) dσ(y)
z ∈Γκ (x) C j
z→x
∫
= k(x − y)hm, j (y) dσ(y)
Cj
∫
= lim+ k(x − y)hm, j (y) dσ(y)
ε→0 ∂Ω\B(x,ε)

= (T hm, j )(x). (2.5.447)

Above, the first equality uses the definition of the operator T (cf. (2.5.375)), while
the second equality is clear from (2.5.434). The third equality in (2.5.447) is a
consequence of Lebesgue’s Dominated Convergence Theorem, whose applicability
is ensured by the continuity of k on Rn \ {0}, the fact that x ∈ Γκ (x), and (2.5.446).
The fourth equality in (2.5.447) is implied by the fact that the set C j (outside of which
hm, j vanishes identically) is contained in ∂Ω \ B(x, ε) whenever 0 < ε < ε j (x) (as
seen from (2.5.440)). Finally, the fifth equality in (2.5.447) relies on the definition of
the operator T (cf. (2.5.376)). In summary, (2.5.447) finishes the proof of (2.5.439).
With (2.5.439) in hand, for each x ∈ ∂∗ Ω ∩ Aκ (∂Ω) and each j ∈ N we may write
490 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
   
   
lim sup (T hm )(z) − (T hm )(x) ≤ lim sup  T (hm − hm, j ) (z)
z ∈Γκ (x) z ∈Γκ (x)
z→x z→x
 
 
+ lim sup  T hm, j (z) − (T hm, j )(x)
z ∈Γκ (x)
z→x
 
+ (T hm, j )(x) − (T hm )(x)
  
≤ Nκ T (hm − hm, j ) (x)
 
+ (T hm, j )(x) − (T hm )(x). (2.5.448)

In concert with (2.5.438) and [112, (8.8.45)], this ultimately implies that

lim (T hm )(z) = (T hm )(x) for σ∗ -a.e. x ∈ ∂∗ Ω. (2.5.449)

z ∈Γκ (x)
z→x

The end-game in the proof is as follows. First, (2.5.427) implies that

fm = 
gm + hm on ∂Ω. (2.5.450)

Second, based on (2.5.450), (2.5.428), and (2.5.449), for σ∗ -a.e. x ∈ ∂∗ Ω we may

write
  
lim T fm (z) = lim T  gm (z) + lim T hm (z)
z ∈Γκ (x) z ∈Γκ (x) z ∈Γκ (x)
z→x z→x z→x
  
= √1 -k ν(x) fm (x) + Tgm (x) + T hm (x)
2 −1
 
= √1 -k ν(x) fm (x) + T fm (x). (2.5.451)
2 −1

This proves that (2.5.418) holds, thus the proof of Proposition 2.5.37 is complete. 
Having established Proposition 2.5.37, the general case described in Theo-
rem 2.5.1 may be dealt via a spherical harmonics expansion. We execute this in the
more general context described in Theorem 2.5.38 below, which may be regarded as
a variable coefficient version of Theorems 2.3.2, 2.4.1, 2.5.1.
Theorem 2.5.38 For each n ∈ N with n ≥ 2 there exists a positive integer M = M(n)
with the following significance. Let b(x, z) be a function which is odd and positive
homogeneous of degree 1 − n in the variable z ∈ Rn \ {0}, and such that ∂zα b(x, z)
is continuous and bounded on Rn × S n−1 for each multi-index α ∈ N0n satisfying
|α| ≤ M. Also, let Ω ⊆ Rn be a nonempty open set with the property that its
topological boundary ∂Ω is a UR set; in particular, Ω is a set of locally finite
perimeter. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure
theoretic outward unit normal to Ω.
Then the following statements are true.
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 491
 
(a) For every f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
the limit

(B f )(x) := lim+ Bε f (x) where, for each ε > 0, (2.5.452)
ε→0
∫

Bε f (x) := b(x, x − y) f (y) dσ(y), ∀ x ∈ ∂Ω, (2.5.453)
y ∈∂Ω
|x−y |>ε

exists at σ-a.e. x ∈ ∂Ω, and the principal-value singular integral operator B

thus defined is bounded from L p (∂Ω, σ) into itself for every p ∈ (1, ∞), and
from L (∂Ω, σ) into L (∂Ω, σ). In fact, if for each f ∈ L ∂Ω, 1+σ(x)
1 1,∞ 1
|x | n−1
one
defines   
Bmax f (x) := sup (Bε f )(x), ∀ x ∈ ∂Ω, (2.5.454)
ε>0

then, for every p ∈ (1, ∞) there exists CΩ, p ∈ (0, ∞) such that
 α 
Bmax f  L p (∂Ω,σ) ≤ CΩ, p · sup  ∂ b (x, z)  f  L p (∂Ω,σ) . (2.5.455)
z
(x,z)∈R n ×S n−1
|α | ≤M

Corresponding to p = 1, estimate (2.5.455) holds if the weak-L 1 norm is used

in the left-hand side.
 
(b) Given any f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
define
∫
(B f )(x) := b(x, x − y) f (y) dσ(y) for each x ∈ Ω. (2.5.456)
∂Ω

Further extend the action of the above boundary-to-domain

 integral operator to
distributions f ∈ H p (∂Ω, σ) with p ∈ n−1
n , 1 by setting, at each x ∈ Ω,
%
b(x, x − ·) , f  if ∂Ω is bounded,
(B f )(x) := % if n−1
n < p ≤ 1,
[b(x, x − ·)] , f  if ∂Ω is unbounded,
(2.5.457)
where ·, · stands for the duality bracket described in [113, Theorem 4.6.1].
Then for each aperture parameter κ ∈ (0, ∞) and each integrability exponent
p ∈ n−1n , ∞ there exists some constant C = C(Ω, κ, p) ∈ (0, ∞) such that
 α 
Nκ (B f ) L p (∂Ω,σ) ≤ C · sup  ∂ b (x, z)  f  H p (∂Ω,σ) (2.5.458)
z
(x,z)∈R n ×S n−1
|α | ≤M

for each f ∈ H p (∂Ω, σ). Also, there exists a constant C = C(Ω, κ) ∈ (0, ∞)
for which
492 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 α 
Nκ (B f ) L 1,∞ (∂Ω,σ) ≤ C · sup  ∂ b (x, z)  f  L 1 (∂Ω,σ) (2.5.459)
z
(x,z)∈R n ×S n−1
|α | ≤M

whenever f ∈ L 1 (∂Ω, σ).

 
(c) For every f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 and every κ ∈ (0, ∞), the following jump-
formula holds:
1 
lim (B f )(z) = √ - b x, ν(x) f (x) + (B f )(x) (2.5.460)
z ∈Γκ (x)
z→x
2 −1

at σ-a.e. x ∈ ∂∗ Ω, where ‘hat’ stands for the Fourier transform in the second
variable. 
n , 1 the operator B induces a linear and bounded
Moreover, in the range p ∈ n−1
mapping
B : H p (∂Ω, σ) −→ L p (∂Ω, σ) (2.5.461)
according to

∞
B f := λ j Ba j in L p (∂Ω, σ), (2.5.462)
j=1

whenever the given distribution f ∈ H p (∂Ω, σ) is expressed as f = ∞ j=1 λ j a j
in H p (∂Ω, σ) for some sequence {λ j } j ∈N ∈  p (N) and some sequence {a j } j ∈N
of (p, q)-atoms on ∂Ω (where q ∈ [1, ∞] with q > p is some fixed background
integrability exponent). Also, having fixed an aperture parameter κ ∈ (0, ∞), if
H : H p (∂Ω, σ) → L p (∂Ω, σ) stands for the L p -filtering operator from [113,
Theorem 4.9.1] (currently used with Σ := ∂Ω) then for each f ∈ H p (∂Ω, σ)
with p ∈ n−1 n , 1 the following jump-formula holds:

 κ−n.t. 
 1
Bf  (x) = √ - b x, ν(x) (H f )(x) + (B f )(x) (2.5.463)
∂Ω 2 −1
for σ-a.e. x ∈ ∂∗ Ω.
(d) Similar results to those stated
 in items (a)-(c) are valid for the operators defined
for each function f ∈ L ∂Ω, 1+σ(x)
1
|x | n−1
as
  
#
Bmax f (x) := sup (Bε# f )(x) at each x ∈ ∂Ω, and (2.5.464)
ε>0

(B# f )(x) := lim+ Bε# f (x) at σ-a.e. x ∈ ∂Ω, where (2.5.465)
ε→0
∫

Bε# f (x) := b(y, x − y) f (y) dσ(y), ∀ x ∈ ∂Ω, ∀ ε > 0, (2.5.466)
y ∈∂Ω
|x−y |>ε
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 493

as well as
∫
(B f )(x) :=
#
b(y, x − y) f (y) dσ(y) for each x ∈ Ω. (2.5.467)
∂Ω

In particular, the limit in (2.5.465) exists at σ-a.e. point in ∂Ω and, given

integrability exponent p ∈ [1, ∞) along with any aperture parameter κ ∈ (0, ∞),
there exists a constant C = C(Ω, κ, p) ∈ (0, ∞) such that
 α 
Bmax
#
f  L p (∂Ω,σ) ≤ C · sup  ∂ b (y, z)  f  L p (∂Ω,σ), (2.5.468)
z
(y,z)∈R n ×S n−1
|α | ≤M
   α 
Nκ (B # f ) ≤C· sup  ∂ b (y, z)  f  L p (∂Ω,σ), (2.5.469)
L p (∂Ω,σ) z
(y,z)∈R n ×S n−1
|α | ≤M

if p ∈ (1, ∞) plus similar estimates (involving the space L 1,∞ (∂Ω, σ) in the
left-hand side) when p = 1. Also,

the (real) transpose of B acting on L p (∂Ω, σ) with p ∈ (1, ∞)


is −B# acting on L p (∂Ω, σ), provided p, p ∈ (1, ∞) satisfy (2.5.470)
1/p + 1/p = 1.
 
Moreover, for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 and each κ > 0, the following jump-
formula holds:
1 
lim (B # f )(z) = √ - b x, ν(x) f (x) + (B# f )(x) (2.5.471)
z ∈Γκ (x)
z→x
2 −1

at σ-a.e. x ∈ ∂∗ Ω, again, with ‘hat’ denoting the Fourier transform in the second
variable.
(e) In all results stated in items (a)-(d) above which involve Lebesgue spaces
L p (∂Ω, σ) with p ∈ (1, ∞) one may use instead the more inclusive scale
of Muckenhoupt weighted Lebesgue spaces L p (∂Ω, wσ) with p ∈ (1, ∞) and
w ∈ Ap (∂Ω, σ) arbitrary.
(f) In a slightly more restrictive setting than originally assumed, there are also
natural Hardy space estimates for the operators
1 B # , B# . Specifically, suppose
∂Ω is compact, p ∈ n , 1 , r > (n − 1) p − 1 , and whenever α ∈ N0n has
n−1

|α| ≤ M the function ∂zα b(·, z) belongs to 𝒞r (Rn ), uniformly for z ∈ S n−1 . Then
the operator B # , now interpreted as
%
(B # f )(x) := b(·, x − ·) , f  for each f ∈ H p (∂Ω, σ) and x ∈ Ω, (2.5.472)
494 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

where ·, · stands for the duality bracket described in [113, Theorem 4.6.1],
satisfies
 
Nκ (B # f ) p (2.5.473)
L (∂Ω,σ)
 α 
≤ C · max sup  ∂z b (·, z)𝒞r (Rn )  f  H p (∂Ω,σ),
|α | ≤M z ∈S n−1

for each f ∈ H p (∂Ω, σ), where C ∈ (0, ∞) is independent of f . Also, B# may

be extended to a linear and bounded operator

B# : H p (∂Ω, σ) −→ L p (∂Ω, σ) (2.5.474)

by setting

∞
B# f := λ j B# a j in L p (∂Ω, σ) (2.5.475)
j=1

whenever the given distribution f ∈ H p (∂Ω, σ) is expressed as f = ∞ j=1 λ j a j
in H p (∂Ω, σ) for some sequence {λ j } j ∈N ∈  p (N) and some sequence {a j } j ∈N
of (p, q)-atoms on ∂Ω (where q ∈ [1, ∞] with q > p is some fixed background
integrability exponent).
Moreover, with H : H p (∂Ω, σ) → L p (∂Ω, σ) denoting the L p -filtering operator
from [113, Theorem 4.9.1] (presently used with Σ := ∂Ω) and having fixed
an aperture parameter κ ∈ (0, ∞), it follows that for each f ∈ H p (∂Ω, σ) the
following jump-formula holds:

 κ−n.t. 
 1
B# f  (x) = √ - b x, ν(x) (H f )(x) + (B# f )(x) (2.5.476)
∂Ω 2 −1
for σ-a.e. x ∈ ∂∗ Ω.
(g) There is a natural L p -square function estimate for the operator B # . Concretely,
for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) depending only on n, p,
and the UR character of ∂Ω with the property that for each f ∈ L p (∂Ω, σ) one
has
∫
 p−1
|∇B # f (x)| p dist x, ∂Ω dx (2.5.477)
Ω
 p ∫
 α 
≤C· sup  ∂ b (y, z) | f | p dσ.
z
(y,z)∈R n ×S n−1 ∂Ω
|α | ≤M

Also, under the stronger assumption that Ω is actually a bounded Lipschitz

domain,
p, p
B # : L p (∂Ω, σ) −→ B1/p (Ω) (2.5.478)
is a well-defined, linear, and bounded operator, for each p ∈ (1, ∞).
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 495

(h) A similar L p -square function estimate also holds for the variable coefficient
operator B defined in (2.5.456), with some caveats. Specifically, under the
additional assumptions that Ω is bounded and whenever α, β ∈ N0n satisfy
β
|α| ≤ M and | β| ≤ 1 the function ∂x ∂zα b(x, z) is continuous and bounded on
R × S , it follows that for each p ∈ (1, ∞) there exists a constant C ∈ (0, ∞)
n n−1

depending only on n, p, the UR character of ∂Ω, and the diameter of Ω, with the
property that for each f ∈ L p (∂Ω, σ) one has the following L p -square function
estimate:
∫
 p−1
|∇B f (x)| p dist x, ∂Ω dx
Ω
∫
 β 
≤ C · max sup  ∂x ∂zα b (x, z) | f | p dσ. (2.5.479)
|α | ≤M x ∈R n ∂Ω
|β | ≤1 z ∈S n−1

Furthermore, if Ω is actually a bounded Lipschitz domain then for each exponent

p ∈ (1, ∞) the operator
p, p
B : L p (∂Ω, σ) −→ B1/p (Ω) (2.5.480)

is well defined, linear, and bounded.

Proof Define
   
n−1+ n+−3
H0 := 1, H1 := n, and H :=  −  − 2 if  ≥ 2, (2.5.481)
 
and, for each  ∈ N0 , let Ψi 1≤i ≤H be an orthonormal basis for the space of
spherical harmonics of degree  on the (n − 1)-dimensional sphere S n−1 equipped
with the measure H n−1 . In particular,

H ≤ ( + 1) · ( + 2) · · · (n +  − 2) · (n +  − 1) ≤ Cn  n−1 for  ≥ 1 (2.5.482)

and, if ΔS n−1 denotes the Laplace-Beltrami operator on S n−1 , then for each  ∈ N0
and 1 ≤ i ≤ H ,

ΔS n−1 Ψi = −(n +  − 2)Ψi on S n−1, and

 x  P (x) (2.5.483)
i
Ψi = for every x ∈ Rn \ {0},
|x| |x| 
for some homogeneous harmonic polynomial Pi of degree  in Rn , thus

Ψi : S n−1 → R is an even function whenever  is even. (2.5.484)

Also,
 
Ψi  ∈N0, 1≤i ≤H
is an orthonormal basis for L 2 (S n−1, H n−1 ), (2.5.485)
496 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

hence,

Ψi  L 2 (S n−1, H n−1 ) = 1 for each  ∈ N0 and 1 ≤ i ≤ H . (2.5.486)

More details on these matters may be found in, e.g., [171, pp. 137–152] and [167,
pp. 68–75].
Next, recall the integer N = N(n) ∈ N from Theorems 2.3.2-2.5.1 and fix

an even integer d ∈ N with d > [(n − 1)/2] + N. (2.5.487)

Sobolev’s embedding theorem then gives that for each  ∈ N0 and 1 ≤ i ≤ H we

have (with I standing for the identity operator on S n−1 )
 
Ψi 𝒞 N (S n−1 ) ≤ Cn (I − ΔS n−1 )d/2 Ψi  L 2 (S n−1, H n−1 ) ≤ Cn max{1, } d, (2.5.488)

where the last inequality is a consequence of (2.5.483)-(2.5.486).

Going further, pick a number

M = M(n) ∈ N even, such that M ≥ 2(n + 1) + d, (2.5.489)

and consider a function b(·, ·) : Rn × S n−1 → R which is odd and positive homoge-
neous of degree 1 − n in the second variable and such that ∂zα b(x, z) is continuous
and bounded on Rn × S n−1 for each α ∈ N0n with |α| ≤ M. Consider now  ∈ N0 and
1 ≤ i ≤ H arbitrary. If we define
∫
ai (x) := b(x, ω)Ψi (ω) dH n−1 (ω), for each x ∈ Rn, (2.5.490)
S n−1

it follows from (2.5.484) and the assumptions on b(x, z) that

the function ai is continuous in Rn and
(2.5.491)
is identically zero whenever  is even.
In addition, for each number m ∈ N with 2m ≤ M we have
∫

[−(n +  − 2)]m ai (x) = b(x, ω) ΔSmn−1 Ψi (ω) dH n−1 (ω)
S n−1
∫

= ΔSmn−1 b (x, ω)Ψi (ω) dH n−1 (ω), (2.5.492)
S n−1

from which we conclude that

2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 497
   
sup [−(n +  − 2)]m ai (x) ≤ sup  ΔSmn−1 b (x, ·) L 2 (S n−1, H n−1 )
x ∈R n x ∈R n
 α 
≤ Cn sup  ∂ b (x, z)
z
(x,z)∈R n ×S n−1
|α | ≤M

=: Cb ∈ (0, ∞). (2.5.493)

Hence, the coefficients ai are rapidly decreasing, in a uniform manner, in the sense
that for each number m ∈ N with 2m ≤ M there exists a constant Cn,m ∈ (0, ∞) such
that
 
sup ai (x) ≤ Cn,m · Cb · max{1, }−2m,  ∈ N0, 1 ≤ i ≤ H . (2.5.494)
x ∈R n

For each fixed x ∈ Rn , expand the function b(x, ·) ∈ L 2 (S n−1, H n−1 ) with respect to
the orthonormal basis (2.5.485) to obtain that (in the sense of L 2 (S n−1, H n−1 ) in the
variable z/|z| ∈ S n−1 )

 z   H  z 
b(x, z) = b x, |z| 1−n = ai (x)Ψi |z| 1−n
|z|  ∈N i=1
|z|
0

 
H  z 
= ai (x)Ψi |z| 1−n, (2.5.495)
 ∈2N0 +1 i=1
|z|

where the last equality is a consequence of (2.5.491). For each  ∈ N0 and 1 ≤ i ≤ H

let us now set
 z  Pi (z)
ki (z) := Ψi |z| 1−n = n−1+ for z ∈ Rn \ {0}. (2.5.496)
|z| |z|
In this notation, (2.5.495) streamlines to

 
H

b(x, z) = ai (x)ki (z), ∀(x, z) ∈ Rn × Rn \ {0} . (2.5.497)
 ∈2N0 +1 i=1

Moreover, from definition (2.5.496) and estimate (2.5.488) we conclude that if d is

as in (2.5.487) then there exists some constant Cn, N ∈ (0, ∞) with the property that
  
 ki  n−1  N n−1 ≤ Cn, N · max{1, } d if  ∈ N0 and 1 ≤ i ≤ H . (2.5.498)
S 𝒞 (S )

For future use, let us also note here that, as seen from the expansion (2.5.497) and
the estimates in (2.5.494) and (2.5.498), we have
498 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

 
H
-
b(x, ξ) = ai (x)-
ki (ξ) with uniform convergence
 ∈2N0 +1 i=1 (2.5.499)

for (x, ξ) in compact subsets of Rn × Rn \ {0} .
 
Moving on, given any f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
, for each  ∈ N0 and 1 ≤ i ≤ H
define
∫
Bε f (x) :=
i ki (x − y) f (y) dσ(y), x ∈ ∂Ω,
y ∈∂Ω
|x−y |>ε
 
i f (x) := sup  Bi f (x),
(2.5.500)
Bmax ε x ∈ ∂Ω,
ε>0

Bi f (x) := lim+ Bεi f (x), x ∈ ∂Ω.

ε→0

Then for each p ∈ [1, ∞) Theorem 2.3.2 guarantees the existence of a constant
CΩ, p ∈ (0, ∞) such that whenever  ∈ N0 and 1 ≤ i ≤ H it follows that for each
f ∈ L p (∂Ω, σ) we have
  
Bmax
i
f  L p (∂Ω,σ) ≤ CΩ, p  ki S n−1 𝒞 N (S n−1 )  f  L p (∂Ω,σ)

≤ CΩ, p · Cn, N · max{1, } d  f  L p (∂Ω,σ) (2.5.501)

if 1 < p < ∞ and, corresponding to p = 1, a constant CΩ ∈ (0, ∞) such that for

every f ∈ L 1 (∂Ω, σ) we have
  
Bmax
i
f  L 1,∞ (∂Ω,σ) ≤ CΩ  ki S n−1 𝒞 N (S n−1 )  f  L 1 (∂Ω,σ)

≤ CΩ · Cn, N · max{1, } d  f  L 1 (∂Ω,σ) . (2.5.502)

Next, fix f ∈ L p (∂Ω, σ) with 1 ≤ p < ∞ arbitrary along with ε > 0 and, for
each threshold μ ∈ N, split
   
(Bε f )(x) = ai (x)Bεi f (x) + ai (x)Bεi f (x),
 ∈2N0 +1 1≤i ≤H  ∈2N0 +1 1≤i ≤H
 ≤μ >μ
(2.5.503)
for each x ∈ ∂Ω. In relation to this splitting, observe that if 1 < p < ∞, thanks
to (2.5.500), Minkowski’s inequality, (2.5.494), (2.5.501), and (2.5.482), for each
m ∈ N such that
n + 1 + d ≤ 2m ≤ M (2.5.504)
(a viable choice given (2.5.489)) we have, with all multiplicative constants positive
and finite,
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 499
 
  
    

sup  
ai Bεi f 
  
 ε>0  ∈2N0 +1 1≤i ≤H
 >μ 
L p (∂Ω,σ)
   i 
≤ sup |ai (x)|  Bmax f  L p (∂Ω,σ)
n
 ∈2N0 +1 1≤i ≤H x ∈R
>μ

≤ C(Ω, b, p, n)  n−1+d−2m  f  L p (∂Ω,σ)
>μ
 
≤ C(Ω, b, p, n)  −2  f  L p (∂Ω,σ)
>μ

C(Ω, b, p, n)
≤  f  L p (∂Ω,σ) . (2.5.505)
μ

Corresponding to p = 1, there is a similar estimate, involving the weak-L 1 space

L 1,∞ (∂Ω, σ).Since the latter is only quasi-Banach, more care is required in the
step when Minkowski’s inequality has been used in the past. Specifically, from the
equivalence in [112, (6.2.21)] and [112, Proposition 6.2.10] (used with p = 1, which
forces α1 = 1/2) we conclude that if we now take m ∈ N such that

2(n + 1) + d ≤ 2m ≤ M (2.5.506)

(once again, a viable choice thanks to (2.5.489)) then, with all multiplicative constants
positive and finite,
 
  
    

sup  i 
 ai Bε 
f
 ε>0   ∈2N0 +1 1≤i ≤H 
 >μ  1,∞
L (∂Ω,σ)
 2
   1/2
≤C sup |ai (x)| 1/2  Bmax
i
f  1,∞
L (∂Ω,σ)
n
 ∈2N0 +1 1≤i ≤H x ∈R
>μ
 2
≤ C(Ω, b, n)  n−1+(d/2)−m  f  L 1 (∂Ω,σ)
>μ
 2
≤ C(Ω, b, n)  −2  f  L 1 (∂Ω,σ)
>μ

C(Ω, b, n)
≤  f  L 1 (∂Ω,σ) . (2.5.507)
μ2
500 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

On the other hand, Theorem 2.3.2 gives that, for each fixed μ ∈ N, the limit as
ε → 0+ of the first sum in (2.5.503), i.e.,
   
lim+ ai (x)Bεi f (x) = ai (x) lim+ Bεi f (x)
ε→0 ε→0
 ∈2N0 +1 1≤i ≤H  ∈2N0 +1 1≤i ≤H
 ≤μ  ≤μ
 
= ai (x)Bi f (x), (2.5.508)
 ∈2N0 +1 1≤i ≤H
 ≤μ

exists at σ-a.e. x ∈ ∂Ω. Thus, if we consider the disagreement function

D f (x) := lim sup Bε f (x) − lim inf

+
Bε f (x), x ∈ ∂Ω, (2.5.509)
ε→0+ ε→0

then for every μ ∈ N we have

   
D f (x) = lim sup ai (x)Bεi f (x)
ε→0+  ∈2N0 +1 1≤i ≤H
>μ
   
− lim inf
+
ai (x)Bεi f (x) (2.5.510)
ε→0
 ∈2N0 +1 1≤i ≤H
>μ

at σ-a.e. x ∈ ∂Ω. Granted this, in the case when 1 < p < ∞, from Chebytcheff’s
inequality and (2.5.505) we conclude that for every μ ∈ N we have
   C(Ω, b, p, n)  p
p
σ {x ∈ ∂Ω : D f (x) > λ} ≤  f  L p (∂Ω,σ), ∀λ > 0. (2.5.511)
μ·λ
Similarly, corresponding to p = 1, from (2.5.510) and (2.5.507) we deduce that
  C(Ω, b, n)
σ {x ∈ ∂Ω : D f (x) > λ} ≤  f  L 1 (∂Ω,σ), ∀λ > 0. (2.5.512)
μ2 · λ

Passing to limit μ → ∞ in (2.5.511)-(2.5.512) then shows that D f (x) = 0 for σ-a.e.

x ∈ ∂Ω and this shows that
the limit in (2.5.452) exists at σ-a.e. point on ∂Ω
(2.5.513)
for any function f ∈ L p (∂Ω, σ) with 1 ≤ p < ∞.
Let us now turn our attention to the jump-formula (2.5.460) in the case when
f ∈ L p (∂Ω, σ) with 1 ≤ p < ∞. To set the stage we note that, as seen from
(2.5.497), for each μ ∈ N the difference
 
B− ai Bi (2.5.514)
 ∈2N0 +1 1≤i ≤H
 ≤μ
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 501

is a principal-value integral operator on ∂Ω with kernel

 
ai (x)ki (x − y), x, y ∈ ∂Ω. (2.5.515)
 ∈2N0 +1 1≤i ≤H
>μ

Arguing as in (2.5.505) and (2.5.507), based on (2.3.20)-(2.3.21), this permits us to

conclude that
   
 
lim B − ai Bi  p = 0 if 1 < p < ∞ (2.5.516)
μ→∞ p
L (∂Ω,σ)→L (∂Ω,σ)
 ∈2N0 +1 1≤i ≤H
 ≤μ

and, corresponding to p = 1,
   
 
lim B − ai Bi  = 0. (2.5.517)
μ→∞ L 1 (∂Ω,σ)→L 1,∞ (∂Ω,σ)
 ∈2N0 +1 1≤i ≤H
 ≤μ

Hence,  
B= ai Bi (2.5.518)
 ∈2N0 +1 1≤i ≤H
 
with convergence in B L p (∂Ω, σ) → L p (∂Ω, σ) if 1 < p < ∞ and, correspond-
 
ing to p = 1, in B L 1 (∂Ω, σ) → L 1,∞ (∂Ω, σ) .
 
To proceed, given any function f ∈ L 1 ∂Ω, 1+σ(x) |x | n−1 , for each  ∈ N0 and
1 ≤ i ≤ H we now introduce
∫
(Bi f )(x) := ki (x − y) f (y) dσ(y), ∀x ∈ Ω. (2.5.519)
∂Ω

Having fixed an arbitrary threshold μ ∈ N then split

   
B f (z) = ai (z)Bi f (z) + ai (z)Bi f (z)
 ∈2N0 +1 1≤i ≤H  ∈2N0 +1 1≤i ≤H
 ≤μ >μ

=: Iμ (z) + I Iμ (z), ∀z ∈ Ω. (2.5.520)

Then whenever f ∈ L p (∂Ω, σ) with 1 ≤ p < ∞, the jump-formula (2.5.377) applies

(given (2.5.496)) and yields
    
lim Iμ (z) = ai (z) √1 -ki ν(x) f (x) + Bi f (x) (2.5.521)
z ∈Γκ (x) 2 −1
z→x  ∈2N0 +1 1≤i ≤H
 ≤μ

at σ-a.e. x ∈ ∂∗ Ω. From this, (2.5.499), and (2.5.516)-(2.5.517) we then conclude

that there exists a strictly increasing sequence {μ j } j ∈N ⊆ N with the property that
502 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

lim Iμ j (z) converges at σ-a.e. x ∈ ∂∗ Ω to

z ∈Γκ (x)
z→x (2.5.522)

√1 -b x, ν(x) f (x) + B f (x) as j → ∞.
2 −1

On the other hand, for each x ∈ ∂Ω we have

    
sup  I Iμ (z) ≤ sup |ai (x)| Nκ Bi f (x) (2.5.523)
n
z ∈Γκ (x)  ∈2N0 +1 1≤i ≤H x ∈R
>μ

which, in concert with Theorem 2.4.1, (2.5.494), and (2.5.498), proves that
  

lim  sup  I Iμ (z) p = 0. (2.5.524)
μ→∞ z ∈Γκ (x) L (∂Ω,σ)

Hence, after possibly refining our earlier strictly increasing sequence {μ j } j ∈N ⊆ N,

matters may be arranged so that

lim I Iμ j (z) converges to 0 at σ-a.e. x ∈ ∂Ω as j → ∞. (2.5.525)

z ∈Γκ (x)
z→x

Ultimately, from (2.5.520), (2.5.522), and (2.5.525), we conclude that

the jump-formula (2.5.460) holds whenever the func-
(2.5.526)
tion f belongs to L p (∂Ω, σ) for some 1 ≤ p < ∞.
We shall now show that both the σ-a.e. existence of the limit in (2.5.452) and
the jump-formula (2.5.460) are actually valid for functions f belonging to the larger

space L 1 ∂Ω, 1+σ(x)
|x | n−1
(which contains any L p (∂Ω, σ) with 1 ≤ p < ∞). In prepa-
ration to dealing with the former claim, given any surface ball Δ ⊆ ∂Ω, for each

truncation parameter ε > 0 and each function f ∈ L 1 ∂Ω, 1+σ(x)|x | n−1
define
∫

Bε,Δ f (x) := b(x, x − y) f (y) dσ(y), ∀ x ∈ Δ. (2.5.527)
y ∈∂Ω
|x−y |>ε

As far as the σ-a.e. existence on

 ∂Ω of the limit in (2.5.452) is concerned, it suffices
to show that whenever f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
the limit

lim+ Bε,Δ f (x) exists for σ-a.e. x ∈ Δ. (2.5.528)
ε→0

With this goal in mind, pick an arbitrary f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
and for each fixed
ε > 0 decompose
 
Bε,Δ f = Bε,Δ 12Δ · f + Bε,Δ 1∂Ω\2Δ · f . (2.5.529)
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 503
 

Note that the first term in the right side of (2.5.529) is simply Bε 12Δ · f  and
Δ
12Δ · f ∈ L 1 (∂Ω, σ). In view of (2.5.513) we may then conclude that, on the one
hand, 
lim+ Bε,Δ 12Δ · f (x) exists for σ-a.e. x ∈ Δ. (2.5.530)
ε→0

On the other hand, for each x ∈ Δ the number Bε,Δ 1∂Ω\2Δ · f (x) is actually
independent of ε > 0 whenever this truncation parameter is smaller than the radius
of Δ. In concert with (2.5.530) and (2.5.529), this establishes (2.5.528).
 Hence, the
proof of the σ-a.e. existence of the limit in (2.5.452) for any f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
is
complete.
Let us now show that the jump-formula stated in (2.5.460) is in fact true for

any given function f ∈ L 1 ∂Ω, 1+σ(x)|x | n−1
. To get started, fix a point x0 ∈ ∂Ω and
pick an arbitrary number r ∈ (0, ∞). Next, having selected some arbitrary function

f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
, decompose

f = f1 + f2 on ∂Ω, where
(2.5.531)
f1 := 1∂Ω∩B(x0,2r) · f and f2 := 1∂Ω\B(x0,2r) · f ,

then split
B f = B f1 + B f2 in Ω. (2.5.532)
Fix an aperture parameter κ > 0. Since the function B f2 has a continuous extension
κ−n.t.

to B(x0, r), we trivially have that the nontangential trace B f2  exists at every
∂Ω
point in ∂nta Ω ∩ B(x0, r). More precisely,
κ−n.t. ∫
 
B f2  (x) = b(x, x − y) f2 (y) dσ(y)
∂Ω ∂Ω

= B f2 (x) for each x ∈ ∂nta Ω ∩ B(x0, r). (2.5.533)

In particular, from (2.5.533) and item (iii) in [112, Proposition 8.8.6] we conclude
that, on the one hand,
 κ−n.t. 

B f2  (x) = B f2 (x) at σ-a.e. x ∈ ∂∗ Ω ∩ B(x0, r). (2.5.534)
∂Ω

On the other hand, since f1 ∈ L 1 (∂Ω, σ), we may invoke (2.5.526) to obtain that, at
σ-a.e. point x ∈ ∂∗ Ω ∩ B(x0, r), we have

 κ−n.t. 
 1
B f1  (x) = √ - b x, ν(x) f1 (x) + (B f1 )(x). (2.5.535)
∂Ω 2 −1
In turn, from (2.5.534), (2.5.535), and (2.5.531) we conclude that
504 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

 κ−n.t. 
 1
Bf  (x) = √ - b x, ν(x) f (x) + (B f )(x) (2.5.536)
∂Ω 2 −1
for σ-a.e. x ∈ ∂∗ Ω ∩ B(x0, r). In view of the fact that r > 0 has been arbitrarily
chosen, this ultimately proves that the jump-formula (2.5.460) is in fact true for any

function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
.
Save for the L p -square function estimate (2.5.479), all remaining claims in the
statement of the theorem pertaining to the operators (2.5.452), (2.5.456), (2.5.457),
are either implicit in what we have done so far, or may be proved in a similar fashion,
using Theorems 2.3.2, 2.4.1, 2.5.1, Corollary 2.5.4, and the expansion in spherical
harmonics given in (2.5.497). Also, in the case of the operators (2.5.464), (2.5.465),
(2.5.467), the same sort of analysis works with x replaced by y in the spherical
harmonic expansion (2.5.495).
The claims from item  (f), when the operators in question act from the Hardy space
H p (∂Ω, σ) with p ∈ n−1 n , 1 , also make use of [113, Proposition 4.4.8]. In this
regard, it is of relevance to note that starting from (2.5.492), in place of (2.5.493)
we now conclude that for each m ∈ N with 2m ≤ M we have
 
[(n +  − 2)]m ai 𝒞r (Rn ) ≤ Cn · max sup  ∂zα b (·, z)𝒞r (Rn ) . (2.5.537)
|α | ≤M z ∈S n−1

Thus, the size of the coefficients ai measured on the Hölder space 𝒞r (Rn ) in the
variable x, in a uniform manner in the variable z ∈ S n−1 , is rapidly decreasing.
Specifically, (2.5.537) implies that for each number m ∈ N with 2m ≤ M there exists
a constant Cn,m ∈ (0, ∞) such that
 
ai 𝒞r (Rn ) ≤ Cn,m · max{1, }−2m · max sup  ∂zα b (·, z)𝒞r (Rn )
|α | ≤M z ∈S n−1
(2.5.538)
whenever  ∈ N0 and 1 ≤ i ≤ H .

In view of [113, (4.4.159)], this tells that each Mai , the operator of multiplication
by ai , has the property that for each number m ∈ N with 2m ≤ M there exists a
constant Cn,m ∈ (0, ∞) such that
 
 Ma  p p
i H (∂Ω,σ)→H (∂Ω,σ)
 
≤ Cn,m · max{1, }−2m · max sup  ∂zα b (·, z)𝒞r (Rn )
|α | ≤M z ∈S n−1

whenever  ∈ N0 and 1 ≤ i ≤ H . (2.5.539)

Granted this, the same type of argument based on the spherical harmonic expansion
(2.5.497) with x replaced by y goes through and, on account of Theorems 2.3.2,
2.4.1, and Corollary 2.5.4, all claims in item (f) follow.
Moreover, the L p -square function estimate (2.5.477) becomes a direct conse-
quence of (2.4.34), keeping in mind that the coefficients ai are rapidly decreasing
(in the sense described in (2.5.494)) and the “plain” convolution kernels ki satisfy
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 505

the growth condition (2.5.498). Having proved (2.5.477), the claim made in (2.5.478)
follows, when Ω is actually a bounded Lipschitz domain, from [113, (9.2.215)] (used
with θ := 1/p) plus (2.5.469) and [112, (8.6.51)].
In contrast to (2.5.477), when dealing with the L p -square function estimate
(2.5.479) via a similar approach, based on the expansion in spherical harmonics
given in (2.5.497), there is one additional series to consider, in which the gradient
∇x now falls on the coefficients ai (x) rather than the kernels ki (x − y). This leads
to the issue of estimating
 p
  H ∫  ∫ 
 
 (∇ai )(x)ki (x − y) f (y) dσ(y) dist(x, ∂Ω) p−1 dx (2.5.540)
 ∈2N0 +1 i=1 Ω  ∂Ω 
∫ p
H ∫
    p  

= (∇ai )(x)  ki (x − y) f (y) dσ(y) dist(x, ∂Ω) p−1 dx
Ω  ∂Ω 
 ∈2N0 +1 i=1
∫
by a finite multiple of ∂Ω
| f | p dσ. In this regard, under the assumption that
β
∂x ∂zα b(x, z)is continuous and bounded on Rn × S n−1 whenever α, β ∈ N0n sat-
isfy |α| ≤ M and | β| ≤ 1, we may differentiate (2.5.490) to obtain
∫
(∇ai )(x) = (∇x b)(x, ω)Ψi (ω) dH n−1 (ω) for all x ∈ Rn, (2.5.541)
S n−1

and then reason as in (2.5.493) to conclude that for each number m ∈ N with 2m ≤ M
we may find a constant Cn,m ∈ (0, ∞) for which
    β 
sup (∇ai )(x) ≤ Cn,m · sup sup  ∂x ∂zα b (x, z) · max{1, }−2m
x ∈R n |α | ≤M x ∈R n
|β | ≤1 z ∈S n−1

whenever  ∈ N0 and 1 ≤ i ≤ H .
(2.5.542)
In view of the format of (2.5.540), this serves our current purpose well. To deal
with the last boundary integral appearing in (2.5.540), let p ∈ (1, ∞) be such that
1/p + 1/p = 1. For any two distinct points x, y ∈ Rn if r := |x − y|/2 then the
homogeneity of ki together with (2.5.498) permit us to estimate
  
ki (· − y) L p (B(x,r), L n ) ≤ Cn, p sup |ki | |x − y| 1−n/p
S n−1

≤ Cn, p  d · |x − y| 1−n/p , (2.5.543)

for all  ∈ 2N0 +1 and 1 ≤ i ≤ H . In particular, if we make the additional assumption

that Ω is bounded, it follows that [112, (8.7.67)] is satisfied, for any choice ε ∈ (0, p1 ),
by the function b(x, y) := ki (x − y). As such, we may rely on [112, Lemma 8.7.11]
which, in concert with (2.5.542), ultimately shows that in the current scenario the
506 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
series (2.5.540) is dominated by C ∂Ω | f | p dσ, with C ∈ (0, ∞) independent of f .
This finishes the proof of the L p -square function estimate (2.5.479).
Subsequently, when Ω is actually a bounded Lipschitz domain, the claims made
in relation to (2.5.480) follow from (2.5.479), [113, (9.2.215)] (used with θ := 1/p),
(2.5.458) (bearing in mind [112, (3.6.27)]), and [112, (8.6.51)]. This completes the
proof of Theorem 2.5.38. 

At last, we are ready to present the proof of Theorem 2.5.1.

Proof of Theorem 2.5.1 By invoking the jump-formula in item (c) of Theorem 2.5.38
with 
b(·, ·) : Rn × Rn \ {0} −→ R given by
(2.5.544)
b(x, z) := k(z) for all x ∈ Rn and z ∈ Rn \ {0},
the desired result follows. 

We augment the results in this section with a proposition pertaining to the non-
tangential behavior of integral operators with weakly singular kernels.

Proposition 2.5.39 Assume Ω ⊆ Rn (where n ∈ N with n ≥ 2) is an open set with

an upper Ahlfors regular boundary and abbreviate σ := H n−1 ∂Ω. Also, suppose
b(·, ·) is a Borel measurable function on Ω × ∂Ω with the property that there exist
some exponent α ∈ (0, n − 1) and some constant C ∈ (0, ∞) such that

for L n -a.e. point x ∈ Ω one has |b(x, y)| ≤ C |x − y| −(n−1−α)

(2.5.545)
at σ-a.e. point y belonging to ∂Ω.

In this context, define the integral operator

∫
ℬ f (x) := b(x, y) f (y) dσ(y) at L n -a.e. x ∈ Ω,
∂Ω
 σ(x)  (2.5.546)
for each function f ∈ L 1 ∂Ω, .
1 + |x| n−1−α
Finally, denote by Iα the fractional integral operator of order α on ∂Ω, acting on
each non-negative σ-measurable function g on ∂Ω according to
∫
g(y)
Iα g(z) := dσ(y), ∀z ∈ ∂Ω. (2.5.547)
∂Ω |z − y| n−1−α
 
Then for each function f ∈ L 1 ∂Ω, 1+ |xσ(x) | n−1−α
the integral in (2.5.546) is abso-
lutely convergent and for each aperture parameter κ > 0 there exists some constant
C = C(Ω, n, α, κ) ∈ (0, ∞) with the property that

Nκ (ℬ f ) (z) ≤ C Iα (| f |)(z), ∀z ∈ ∂Ω. (2.5.548)

Also, one has the continuous embedding

2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 507
 σ(x) 
L p (∂Ω, σ) → L 1 ∂Ω, for each p ∈ 1, (n − 1)/α , (2.5.549)
1 + |x| n−1−α

and for each integrability exponent p ∈ 1, (n − 1)/α and each aperture parameter
κ > 0 there exists C ∈ (0, ∞) such that for every function f ∈ L p (∂Ω, σ) one has
 
Nκ (ℬ f ) p ≤ C f  L p (∂Ω,σ)
L ∗ (∂Ω,σ)
(2.5.550)
with 1/p∗ = 1/p − α/(n − 1) if p > 1,

and, corresponding to the end-point p = 1,

 
Nκ (ℬ f ) (n−1)/(n−1−α), ∞ ≤ C f  L 1 (∂Ω,σ) . (2.5.551)
L (∂Ω,σ)

In addition, for each p ∈ 1, (n − 1)/α there exists C ∈ (0, ∞) with the property that
 n−1 −α 
 p 
δ∂Ω ℬ f  ≤ C f  L p (∂Ω,σ) for each f ∈ L p (∂Ω, σ). (2.5.552)
L ∞ (Ω, L n )

Moreover, under the additional assumption that

for σ-a.e. point y ∈ ∂Ω the function
(2.5.553)
b(·, y) extends continuously to Ω \ {y}

it follows that for each function f ∈ L 1 ∂Ω, 1+ |xσ(x)
| n−1−α
one has
∫
|b(x, y)|| f (y)| dσ(y) < +∞ at σ-a.e. x ∈ ∂Ω, (2.5.554)
∂Ω
κ−n.t.
and for each aperture parameter κ > 0 the nontangential boundary limit ℬ f ∂Ω
exists at σ-a.e. point on Aκ (∂Ω); in fact, for each aperture parameter κ > 0,
 ∫
κ−n.t. 

ℬ f ∂Ω (x) = b(x, y) f (y) dσ(y) at σ-a.e. point x ∈ Aκ (∂Ω). (2.5.555)
∂Ω

We wish to note that if ∂Ω is bounded then estimates in the spirit of (2.5.550)-

(2.5.551)
 continue to hold for the range p ∈ (n − 1)/α, ∞ . Specifically, for each
p ∈ (n − 1)/α, ∞ Hölder’s inequality and the first estimate in [112, (7.2.5)] imply
that there exists Cp ∈ (0, ∞), which is now also allowed to depend on diam(∂Ω),
such that
 
ℬ f  ∞ ≤ Cp  f  L p (∂Ω,σ), ∀ f ∈ L p (∂Ω, σ). (2.5.556)
L (Ω, L n )

Also, corresponding to the critical value p = (n − 1)/α, from (2.5.551) and Hölder’s
inequality we obtain
508 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 
Nκ (ℬ f ) ≤ Cq  f  L (n−1)/α (∂Ω,σ),
L q (∂Ω,σ)
(2.5.557)
∀ f ∈ L (n−1)/α (∂Ω, σ), ∀q ∈ (1, ∞).

Proof of Proposition 2.5.39 That the integral in (2.5.546) is absolutely convergent

for each x ∈ Ω is clear from (2.5.545). Also, given any p ∈ 1, (n − 1)/α , if p
denotes its Hölder conjugate exponent then n−1−α n−1
< p ≤ ∞ so for each function
f ∈ L (∂Ω, σ) we may estimate
p

∫ ∫  1/p
| f (x)| dσ(x)
dσ(x) ≤ C f  L p (∂Ω,σ) (n−1−α)p  < +∞,
∂Ω 1 + |x| ∂Ω 1 + |x|
n−1−α
(2.5.558)
thanks to Hölder’s inequality and [112, Lemma 7.2.1]. This takes care of (2.5.549).
To proceed, recall from [112, (8.1.8)] that

|z − y| < (2 + κ)|x − y|, ∀z ∈ ∂Ω,

∀x ∈ Γκ (z), ∀y ∈ ∂Ω. (2.5.559)
 
Consequently, for each f ∈ L 1 ∂Ω, 1+ |xσ(x)
| n−1−α
we have
∫
   | f (y)|
Nκ (ℬ f ) (z) = ℬ f  L ∞ (Γκ (z), L n ) ≤ C dσ(y)
∂Ω |z − y| n−1−α
= C Iα (| f |)(z), ∀z ∈ ∂Ω, (2.5.560)

proving (2.5.548). With this in hand, the estimates claimed in (2.5.550)-(2.5.551)

now follow with the help of the Fractional Integration Theorem (cf. [112, (7.8.7)-
(7.8.9)]).
As regards (2.5.552), given f ∈ L p (∂Ω, σ) with p ∈ 1, (n − 1)/α , for L n -a.e.
point x ∈ Ω we may estimate
  ∫ dσ(x)  1/p
(ℬ f )(x) ≤ C f  L p (∂Ω,σ)
(n−1−α)p 
∂Ω |x − y|
n−1
−(n−1−α)
≤ C f  L p (∂Ω,σ) δ∂Ω (x) p , (2.5.561)

thanks to (2.5.546), Hölder’s inequality, (2.5.545), and [112, Corollary 8.7.12].

The estimate claimed in (2.5.552) now follows from (2.5.561) upon observing that
p − (n − 1 − α) = − p + α. Parenthetically, we wish to note that, under the
n−1 n−1

additional assumptions that ∂Ω is lower Ahlfors regular and p > 1, the estimate in
(2.5.552) may be alternatively deduced from (2.5.550) and [112, (8.6.62)].
Henceforth, work under the additional assumption made in (2.5.553). We claim
that
for each x ∈ ∂Ω we have |b(x, y)| ≤ C |x − y| −(n−1−α)
(2.5.562)
at σ-a.e. point y belonging to ∂Ω.
To see that this is the case, denote by N the exceptional subset of Ω satisfying
L n (N) = 0 and such that for each x ∈ Ω \ N we have |b(x, y)| ≤ C |x − y| −(n−1−α)
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 509

at σ-a.e. point y ∈ ∂Ω. Next, fix an arbitrary point x ∈ ∂Ω, and observe that
there exists a sequence {x j } j ∈N ⊆ Ω \ N which converges to x. For each j ∈ N
use (2.5.562) to find a σ-measurable set N j ⊆ ∂Ω such that σ(N j ) = 0 and
|b(x j , y)| ≤ C |x j − y| −(n−1−α) for each y ∈ ∂Ω \ N j . The hypothesis listed in
(2.5.553) also guarantees that there exists a σ-measurable set N0 ⊆ ∂Ω such that
σ(N0 ) = 0 and for each y ∈ ∂Ω \ N0 the function b(·, y) has a continuous extension
to Ω \ {y}. Consider now an arbitrary point
 
y ∈ ∂Ω \ ∪ j ∈N N j ∪ N0 ∪ {x} . (2.5.563)

Then b(·, y) has a continuous extension to Ω \ {y} and |b(x j , y)| ≤ C |x j − y| −(n−1−α)
for each j ∈ N. Passing to limit j → ∞ yields |b(x, y)| ≤ C |x − y| −(n−1−α), from

which the conclusion in (2.5.562) follows given that that σ ∪ j ∈N N j ∪ N0 = 0.

Having proved (2.5.562), that (2.5.554) holds for any f ∈ L 1 ∂Ω, 1+ |xσ(x) | n−1−α
be-
comes a consequence of [112, (7.8.5)]. Finally, consider the claim made in (2.5.555).

To this end, fix some arbitrary f ∈ L 1 ∂Ω, 1+ |xσ(x)
| n−1−α
. Then [112, (7.8.5)] implies
that there exists some σ-measurable set N ⊆ ∂Ω satisfying

Iα (| f |)(z) < ∞ for every z ∈ ∂Ω \ N, and σ(N) = 0. (2.5.564)

For each z ∈ ∂Ω bring in the function

| f (y)|
G z : ∂Ω → [0, ∞], G z (y) := for every y ∈ ∂Ω. (2.5.565)
|z − y| n−1−α
∫
In particular, since ∂Ω
G z (y) dσ(y) = Iα (| f |)(z) for each z ∈ ∂Ω, it follows from
(2.5.564) that

G z ∈ L 1 (∂Ω, σ) for every z ∈ ∂Ω \ N. (2.5.566)

Fix now z ∈ Aκ (∂Ω) \ N and consider an arbitrary sequence {x j } j ∈N ⊂ Γκ (z) which

converges to z. For each j ∈ N introduce the function Fj : ∂Ω → R by setting
Fj (y) := b(x j , y) f (y) for σ-a.e. y ∈ ∂Ω. Finally, define F : ∂Ω → R by setting
F(y) := b(z, y) f (y) for σ-a.e. y ∈ ∂Ω. The fact that for σ-a.e. y ∈ ∂Ω the function
b(·, y) extends continuously to Ω \ {y} then implies

lim Fj (y) = F(y) for σ-a.e. y ∈ ∂Ω. (2.5.567)

j→∞

In addition, from (2.5.545), (2.5.559) and definitions we deduce that for each j ∈ N
we have

|Fj (y)| ≤ C · G z (y) for σ-a.e. y ∈ ∂Ω. (2.5.568)

In concert with Lebesgue’s Dominated Convergence Theorem, (2.5.566), (2.5.567),

(2.5.568) then prove that for every z ∈ ∂Ω \ N we have
510 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Fj ∈ L 1 (∂Ω, σ) for every j ∈ N, F ∈ L 1 (∂Ω, σ),

∫ ∫ (2.5.569)
and lim Fj (y) dσ(y) = F(y) dσ(y).
j→∞ ∂Ω ∂Ω

Unraveling definitions this ultimately yields

 ∫
κ−n.t. 
ℬ f ∂Ω (z) = k(z, y) f (y) dσ(y) for every z ∈ Aκ (∂Ω) \ N. (2.5.570)
∂Ω

Upon recalling from (2.5.564) that σ(N) = 0, we conclude that (2.5.555) holds. 

A significant consequence of Proposition 2.5.39, pertaining to global integrability

properties of functions expressed as integrals involving weakly singular kernels, is
presented below.

Proposition 2.5.40 Assume Ω ⊆ Rn , where n ∈ N with n ≥ 2, is an arbitrary

UR domain and abbreviate σ := H n−1 ∂Ω. Also, for a sufficiently large number
N = N(n) ∈ N, consider a complex-valued function b ∈ 𝒞 N (Rn \ {0}) with the
property that there exist C ∈ (0, ∞) and α ∈ (0, n − 1) such that

|b(x)| ≤ C|x| −(n−1−α) for each x ∈ Rn \ {0} (2.5.571)

and such that ∇b is odd and positive homogeneous of degree 1 − n in Rn \ {0}. Fix
a function 
f ∈ L p (∂Ω, σ) with p ∈ 1, (n − 1)/α . (2.5.572)
Finally, define
∫
u(x) := b(x − y) f (y) dσ(y) for each x ∈ Rn \ ∂Ω, (2.5.573)
∂Ω

and regard u as a function defined L n -a.e. in Rn .

Then there exists some constant C ∈ (0, ∞) independent of f such that with
1 α  −1
pα := − ∈ (1, ∞) (2.5.574)
p n−1
one has

u ∈ L npα /(n−1) (Rn, L n ) and u L n pα /(n−1) (Rn, L n ) ≤ C f  L p (∂Ω,σ) . (2.5.575)

Also, for each j ∈ {1, . . . , n}, the distribution ∂j u ∈ D (Rn ) actually belongs the
Lebesgue space L np/(n−1) (Rn, L n ) and

n
∂j u L n p/(n−1) (Rn, L n ) ≤ C f  L p (∂Ω,σ) (2.5.576)
j=1

for some constant C ∈ (0, ∞) independent of f . In particular,

2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 511

1, np/(n−1)
u ∈ Wloc (Rn ). (2.5.577)

Proof Denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit

normal to Ω. If we introduce Ω+ := Ω and Ω− := Rn \ Ω then [112, Lemma 5.10.9]
ensures that Ω− is also a UR domain, whose topological boundary coincides with
that of Ω, whose geometric measure theoretic boundary agrees with that of Ω, and
whose geometric measure theoretic  outward unit normal is −ν at σ-a.e. point on
∂Ω. To proceed, define u± := uΩ± . Also, fix an aperture parameter κ > 0. From
(2.5.550) we know that
 
Nκ u±  p ≤ C f  L p (∂Ω,σ) . (2.5.578)
L α (∂Ω,σ)

In turn, from (2.5.578) and [112, (8.6.51) in Proposition 8.6.3] we conclude that, in
a quantitative fashion,
 np /(n−1)
L α (Ω±, L n ) if Ω± is bounded, or ∂Ω is unbounded,
u± ∈ np /(n−1)
(2.5.579)
Lbddα (Ω±, L n ) if Ω± is an exterior domain.

Also, it is apparent from (2.5.573) that

u± (x) = O(|x| −(n−1−α) ) as |x| → ∞ if Ω± is an exterior domain. (2.5.580)

At this stage, the claims in (2.5.575) become consequences of (2.5.578), (2.5.579),

and (2.5.580).
Moving on, fix an arbitrary j ∈ {1, . . . , n}. Then Theorem 2.4.1 applies and gives
that
 
Nκ (∂j u± ) p ≤ C f  L p (∂Ω,σ) (2.5.581)
L (∂Ω,σ)

for some constant C ∈ (0, ∞) independent of f . Much as before, as a consequence of

(2.5.581), the decay property (∂j u± )(x) = O(|x| 1−n ) as |x| → ∞ if Ω± is an exterior
domain, and [112, Proposition 8.6.3], we may conclude that the function defined as

∂j u+ in Ω+,
w j := (2.5.582)
∂j u− in Ω−,

belongs to the Lebesgue space L np/(n−1) (Rn, L n ) (again, in a quantitative fashion).

We now claim that ∂j u ∈ D (Rn ) coincides with the distribution induced by the
locally integrable function w j in Rn (via integration against test functions). To see
that this is the case, pick an arbitrary ϕ ∈ 𝒞∞
c (R ) and write
n
512 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
% & % &
D  (R n ) ∂j u, ϕ D(R n ) = − D  (Rn ) u, ∂j ϕ D(Rn ) = − u ∂j ϕ dL n
Rn
∫ ∫
=− u+ ∂j ϕ dL n − u− ∂j ϕ dL n
Ω+ Ω−
∫ ∫
 κ−n.t.
= (∂j u+ )ϕ dL n + ν j u+ ∂Ω ϕ dσ
Ω+ ∂Ω
∫ ∫
 κ−n.t.
+ (∂j u− )ϕ dL n − ν j u− ∂Ω ϕ dσ
Ω− ∂Ω
∫
= w j ϕ dL n . (2.5.583)
Rn

The fourth equality above is implied by the integration by parts formula recorded in
[112, Theorem 1.7.1] (whose applicability in the present circumstances is guaranteed
by Proposition 2.5.39 and [112, Proposition 8.8.4]), while the fifth equality above
makes use of (2.5.582) and (2.5.555). From (2.5.583) we then conclude that ∂j u = w j
in D (Rn ), from which all desired conclusions follow. 
Here is a version of Proposition 2.5.40 for UR domains with compact boundaries.
This makes less demanding assumptions on the function b, albeit the given function
f now has vanishing moment.
Proposition 2.5.41 Assume Ω ⊆ Rn , where n ∈ N with n ≥ 2, is a UR domain with
compact boundary and abbreviate σ := H n−1 ∂Ω. Also, for a sufficiently large
number N = N(n) ∈ N, consider a complex-valued function b ∈ 𝒞 N (Rn \ {0}) with
the property that there exist C ∈ (0, ∞) and α ∈ (0, n − 1) such that

|b(x)| ≤ C|x| −(n−1−α) for each x ∈ B(0, 1) \ {0} (2.5.584)

and such that ∇b is odd and positive homogeneous of degree 1 − n in Rn \ {0}. Fix
a function
∫
f ∈ L p (∂Ω, σ) with p ∈ (1, ∞) and f dσ = 0. (2.5.585)
∂Ω

Finally, define
∫
u(x) := b(x − y) f (y) dσ(y) for each x ∈ Rn \ ∂Ω, (2.5.586)
∂Ω

and regard u as a function defined L n -a.e. in Rn .

Then there exists some constant C ∈ (0, ∞) independent of f such that with
1 α  −1
pα := − ∈ (1, ∞) (2.5.587)
p n−1
one has, in a quantitative sense,
2.5 The Jump-Formula for Integral Operators in Open Sets with UR Boundaries 513

u ∈ L npα /(n−1) (Rn, L n ) if p ∈ 1, (n − 1)/α , (2.5.588)

and

u∈ L q (Rn, L n ) if p ∈ (n − 1)/α, ∞ . (2.5.589)

n
n−1 <q<∞

Moreover, for each j ∈ {1, . . . , n}, the distribution ∂j u ∈ D (Rn ) actually belongs
any Lebesgue space L q (Rn, L n ) with 1 < q ≤ np/(n − 1).

Proof The argument proceeds along the lines of the proof of Proposition 2.5.40, so
we will elaborate only on the main differences. To get started, there is no loss of
generality in assuming that Ω+ is bounded (and Ω− is an exterior domain). Fix a
cutoff function ψ ∈ 𝒞∞ c (R ) with ψ ≡ 1 near Ω+ and define b(x, y) := ψ(x)b(x − y)
n

for each distinct points x, y ∈ Rn . Next, for f as in (2.5.585), define

∫
ℬ± f (x) := b(x, y) f (y) dσ(y) for each x ∈ Ω± . (2.5.590)
∂Ω

As in the past, set u± := uΩ± where u is as in (2.5.586). From definitions, we have
ℬ± f = ψu± in Ω± . Since ∂Ω is compact and ψ is compactly supported, b(x, y)
satisfies (2.5.545) both relative to Ω+ and Ω− . As a consequence of (2.5.550) we
then have
 
Nκ (ψu± ) p ≤ C f  L p (∂Ω,σ) if 1 < p < (n − 1)/α. (2.5.591)
L α (∂Ω,σ)

In concert with [112, (8.6.51)], this eventually leads to the conclusion that
np /(n−1)
u± ∈ Lbddα (Ω±, L n ) if 1 < p < (n − 1)/α. (2.5.592)

Upon taking into account the cancelation property of f and the fact that ∇b is positive
homogeneous of degree 1 − n in Rn \ {0}, we conclude that

(∇k u− )(x) = O(|x| −(n−1+k) ) as |x| → ∞, for all k ∈ {0, 1, · · · , N }. (2.5.593)

At this stage, (2.5.589) follows from (2.5.592) and (2.5.593) (used with k := 0). In
the case when p ∈ (n − 1)/α, ∞ , in place of (2.5.591) we now obtain
 
Nκ (ψu± ) q ≤ Cq  f  L p (∂Ω,σ) for each q ∈ (1, ∞), (2.5.594)
L (∂Ω,σ)

thanks to (2.5.556)-(2.5.557). By once again using [112, (8.6.51)] and (2.5.593), we

thus arrive at the conclusion that (2.5.589) holds.
Moving on, the same argument as in the proof of Proposition 2.5.40 gives
that the distributional derivatives ∂j u ∈ D (Rn ) with j ∈ {1, . . . , n} belong to
L np/(n−1) (Rn, L n ). Then the last claim in the statement follows from this and
(2.5.593) used with k := 1. 
514 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

2.6 Singular Integrals on Morrey Spaces and Their Pre-Duals

This section contains our main results regarding singular integral operators on Mor-
rey spaces, as well as their pre-duals, considered on uniformly rectifiable sets. We
begin treating singular integral operators of a general nature in the theorem below,
which we subsequently apply to the study of boundary layer potentials in [115, §3.3].

Theorem 2.6.1 Suppose Σ ⊆ Rn (where n ∈ N with n ≥ 2) is a UR set and abbreviate

σ := H n−1 Σ. Consider a kernel k ∈ 𝒞 N (Rn \ {0}), with N = N(n) ∈ N sufficiently
large, which is odd and positive homogeneous of degree 1−n (cf. (2.3.3)). Associated
with Σ and k, define the principal-value singular integral operator T as in (2.3.15)
as well as its boundary-to-domain version, T defined as in (2.4.3)-(2.4.4). Finally,
fix two integrability exponents p, q ∈ (1, ∞) such that 1/p + 1/q = 1 along with some
parameter λ ∈ (0, n − 1). Then the following claims are true:

(1) The restriction (cf. [113, (6.2.25)]) of the principal-value singular integral op-
erator T to the Morrey space M p,λ (Σ, σ) induces a well-defined, linear, and
bounded mapping

T : M p,λ (Σ, σ) −→ M p,λ (Σ, σ), (2.6.1)

whose operator norm satisfies, for some C = C(Σ, n, p, λ) ∈ (0, ∞),

  
T  M p, λ (Σ,σ)→M p, λ (Σ,σ) ≤ C  k S n−1 𝒞 N (S n−1 ) . (2.6.2)

Also, the maximal operator Tmax (cf. (2.3.5), (2.3.7)) induces a well-defined,
sub-linear, bounded, and continuous mapping in the context

Tmax : M p,λ (Σ, σ) −→ M p,λ (Σ, σ), (2.6.3)

and its norm may be estimated as in (2.6.2). Furthermore, for each given aperture
parameter κ ∈ (0, ∞) there exists some constant C = C(Σ, n, q, λ, κ) ∈ (0, ∞)
with the property that boundary-to-domain integral operator T satisfies
 Σc    
N (T f ) p, λ ≤ C  k S n−1 𝒞 N (S n−1 )  f  M p, λ (Σ,σ)
κ M (Σ,σ)
(2.6.4)
for each function f belonging to the Morrey space M p,λ (Σ, σ).

Also, in view of work in [112, §8.5], an estimate analogous to (2.6.4) holds if the
nontangential maximal operator is replaced by the tangential maximal operator
(associated as in [112, Definition 8.5.1] with Ω := Σ c and a sufficiently large
power M; cf. (A.0.85)).
In addition, there exists some C ∈ (0, ∞), depending only on n, p, λ, k, and
the Ahlfors regularity constants of Σ, with the property that for each multi-index
α ∈ N0n with |α| ≤ N one has
2.6 Singular Integrals on Morrey Spaces and Their Pre-Duals 515
 n−1−λ      
sup dist (x, Σ) |α |+ p ∂ α T f (x) ≤ C  k S n−1 𝒞 N (S n−1 )  f  M p, λ (Σ,σ)
x ∈R n \Σ
for each function f ∈ M p,λ (Σ, σ).
(2.6.5)
Finally, similar results are valid with the Morrey space M p,λ (Σ, σ) replaced by
its version M̊ p,λ (Σ, σ) introduced in (A.0.88).
(2) More generally, let b(x, z) be a function which is odd and positive homogeneous
of degree 1− n in the variable z ∈ Rn \ {0}, and such that ∂zα b(x, z) is continuous
and bounded on Rn × S n−1 for each multi-index α ∈ N0n satisfying |α| ≤ N.
If one then defines the variable-coefficient
 kernel integral operators acting on
σ(x)
functions f ∈ L Σ, 1+ |x | n−1 according to
1

∫
(B f )(x) := lim+ b(x, x − y) f (y) dσ(y) for σ-a.e. x ∈ Σ, (2.6.6)
ε→0
y ∈Σ
|x−y |>ε
 ∫ 
  
 
Bmax f (x) := sup  b(x, x − y) f (y) dσ(y) for each x ∈ Σ, (2.6.7)
ε>0  
y ∈Σ
|x−y |>ε

and
∫
(B f )(x) := b(x, x − y) f (y) dσ(y) for each x ∈ Rn \ Σ, (2.6.8)
Σ

it follows that there exists a constant C = C(Σ, p, λ) ∈ (0, ∞) such that for each
function f ∈ M p,λ (Σ, σ) one has
 
max B f  M p, λ (Σ,σ) , Bmax f  M p, λ (Σ,σ) (2.6.9)
 α 
≤C· sup  ∂ b (x, z)  f  M p, λ (Σ,σ) .
z
(x,z)∈R n ×S n−1
|α | ≤ N

Furthermore, for any given aperture parameter κ belonging to (0, ∞) one can
find some constant C = C(Σ, n, p, λ, κ) ∈ (0, ∞) with the property that for each
function f ∈ M p,λ (Σ, σ) one has
 Σc   α 
N (B f ) p, λ ≤C· sup  ∂ b (x, z)  f  M p, λ (Σ,σ) . (2.6.10)
κ M (Σ,σ) z
(x,z)∈R n ×S n−1
|α | ≤ N

Once again, thanks to work in [112, §8.5], a similar estimate to (2.6.10) holds
with the nontangential maximal operator replaced by the tangential maximal
operator (associated as in [112, Definition 8.5.1] with Ω := Σ c and a sufficiently
large power M; cf. (A.0.85)).
516 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Lastly, analogous results are true with the Morrey space M p,λ (Σ, σ) replaced
by its version M̊ p,λ (Σ, σ) introduced in (A.0.88).
(3) With the function b as in item (2), similar results to those described item (2) are
valid for
 the variable-coefficient
 kernel integral operators acting on functions
σ(x)
f ∈ L Σ, 1+ |x | n−1 according to
1

∫
(B# f )(x) := lim+ b(y, x − y) f (y) dσ(y) for σ-a.e. x ∈ Σ, (2.6.11)
ε→0
y ∈Σ
|x−y |>ε
 ∫ 
  
 
#
Bmax f (x) := sup  b(y, x − y) f (y) dσ(y) for each x ∈ Σ, (2.6.12)
ε>0  
y ∈Σ
|x−y |>ε

and
∫
(B # f )(x) := b(y, x − y) f (y) dσ(y) for each x ∈ Rn \ Σ. (2.6.13)
Σ

(4) The principal-value singular integral operator T from (2.3.18) induces (in view
of the embedding in [113, (6.2.71)]) a well-defined linear and bounded mapping

T : B q,λ (Σ, σ) −→ B q,λ (Σ, σ), (2.6.14)

and
the real transpose of (2.6.14) is the opposite of the operator
(2.6.15)
T acting on Morrey spaces as in (2.6.1) (that is, T  = −T),
while

the real transpose of the operator T acting on M̊ p,λ (Σ, σ) is

(2.6.16)
the opposite of the operator T from (2.6.14) (i.e., T  = −T).
Also,

if f ∈ M̊ p,λ (Σ, σ) and g ∈ B q,λ

. (Σ, σ) then (T f )g + f (T g) belongs to the
homogeneous Hardy  space H 1 (Σ, σ) (cf. [113, (4.2.12)]) and one has
(T f )g + f (T g) 1 ≤ C f  M p, λ (Σ,σ) · g B q, λ (Σ,σ) .
H (Σ,σ)
   (2.6.17)
Moreover, the operator norm of T in (2.6.14) is ≤ C  k S n−1 𝒞 N (S n−1 ) where the
constant C ∈ (0, ∞) depends only on Σ, n, q, λ. Also, the maximal operator Tmax
(cf. (2.3.5), (2.3.7)) induces a well-defined, sub-linear, bounded, and continuous
mapping in the context

Tmax : B q,λ (Σ, σ) −→ B q,λ (Σ, σ) (2.6.18)

2.6 Singular Integrals on Morrey Spaces and Their Pre-Duals 517

and its norm may be estimated in a similar fashion as before.

In addition, if one also considers the boundary-to-domain integral operator T
with kernel k, as in (2.4.3)-(2.4.4), then for each given aperture parameter κ > 0
there exists some constant C = C(Σ, n, q, λ, κ) ∈ (0, ∞) with the property that
 Σc    
N (T f ) q, λ ≤ C  k S n−1 𝒞 N (S n−1 )  f  B q, λ (Σ,σ)
κ B (Σ,σ)
(2.6.19)
for each function f belonging to the space B q,λ (Σ, σ).

In light of work in [112, §8.5], a similar estimate to (2.6.19) holds with the
nontangential maximal operator replaced by the tangential maximal operator
(associated as in [112, Definition 8.5.1] with Ω := Σ c and a sufficiently large
power M; cf. (A.0.85)).
More generally, if the function b(x, z) is as described in item (2) above, then for
each function f ∈ B q,λ (Σ, σ) the operators (2.6.6)-(2.6.8) presently satisfy
  c  
max B f  B q, λ (Σ,σ) , Bmax f  B q, λ (Σ,σ), N Σ (B f ) q, λ
κ B (Σ,σ)
 α 
≤C· sup  ∂ b (x, z)  f  B q, λ (Σ,σ), (2.6.20)
z
(x,z)∈R n ×S n−1
|α | ≤ N

where the constant C ∈ (0, ∞) depends only on the ambient. Finally, an estimate
similar to (2.6.20) is valid for the variable-coefficient kernel integral operators
(2.6.11)-(2.6.13).
(5) Fix q ∈ (1, ∞) along with λ ∈ (0, n − 1) and recall the (q, λ)-midway space
ℳq,λ (Σ, σ) introduced in (A.0.89)-(A.0.90). Also, recall that ℋq,λ (Σ, σ), the
pre-dual of a Morrey-Campanato space, has been introduced in (A.0.55)-
(A.0.56). Finally, fix an aperture parameter κ > 0. Then there exists some
constant C = C(Σ, k, p, λ, κ, n) ∈ [0, ∞) with the property that
 Σc    
N (T f ) q, λ ≤ C  k S n−1 𝒞 N (S n−1 )  f ℋq, λ (Σ,σ)
κ ℳ (Σ,σ)
(2.6.21)
for each function f belonging to the space ℋq,λ (Σ, σ).

Finally, a similar estimate holds for the variable-coefficient kernel integral

operator (2.6.13).

The results pertaining to the integral operators with variable coefficient ker-
nels considered in Theorem 2.6.1 apply to the Schwartz kernels of certain pseudo-
differential operators. In particular, this yields results in the spirit of Theorem 2.8.2
on Morrey (and related) spaces on manifolds.
We also wish to note that, thanks to [112, (6.2.23)], given any open set Ω ⊆ Rn
whose topological boundary ∂Ω is a UR set the jump-formula (2.5.4) continues
to hold for each function f belonging to the Morrey space M p,λ (∂Ω, σ) with
p ∈ (1, ∞) and λ ∈ (0, n − 1). More generally, the jump-formulas (2.5.460), (2.5.471)
remain true for each f ∈ M p,λ (∂Ω, σ).
518 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Proof of Theorem 2.6.1 The fact that the operator T is well defined, linear, and
bounded in the context of (2.6.1) is a consequence of [113, Proposition 6.2.12] and
(2.3.56) (also bearing in mind the monotonicity of the Muckenhoupt classes; cf. item
(5) in [112, Lemma 7.7.1]). Alternatively, we could have used (2.3.56) in combination
with the extrapolation result from [113, Corollary 5.2.3], whose applicability in the
present setting with X := M p,λ (∂Ω, σ) is ensured by [113, Proposition 6.2.17]
together with [113, Corollaries 6.2.11, 6.2.13].
Next, the estimate in (2.6.2) is then implied by [113, (6.2.121)-(6.2.122)] and
(2.3.57). In fact, thanks to (2.3.58), the same argument works in the case of Tmax in
(2.6.3). As another justification for (2.6.3) we may rely on Cotlar’s inequality from
(2.3.23) (written for r := 1), (2.6.1), [113, (6.2.3), (6.2.25)], [113, Corollary 6.2.13],
and (2.3.7).
Likewise, the estimate in (2.6.4) is a consequence of [113, Proposition 6.2.12]
and (2.4.18). An alternative proof of (2.6.4) is obtained by relying on (2.4.5), [113,
(6.2.25)], (2.6.3), [113, Corollary 6.2.13], [113, (6.2.3)], and [112, (8.2.28)]. Yet
another justification is seen from (2.4.18), [113, Corollary 5.2.3], [113, Proposi-
tion 6.2.17], and [113, Corollaries 6.2.11, 6.2.13].
To justify (2.6.5), fix some α ∈ N0n with |α| ≤ N and note that
  
|(∂ α k)(x − y)| ≤  k S n−1 𝒞 N (S n−1 ) · |x − y| −(n−1+ |α |)
(2.6.22)
for all points x ∈ Rn \ Σ and y ∈ Σ.

Consequently, given a function f ∈ M p,λ (Σ, σ), for any x ∈ Rn \ Σ we may estimate
∫
 α 
∂ T f (x) ≤ |(∂ α k)(x − y)|| f (y)| dσ(y)
Σ
∫
  | f (y)|
≤ k S n−1 𝒞 N (S n−1 ) dσ(y). (2.6.23)
Σ |x − y| n−1+ |α |

To proceed, pick x∗ ∈ Σ such that |x − x∗ | = dist (x, Σ) := δΣ (x), and write

2.6 Singular Integrals on Morrey Spaces and Their Pre-Duals 519
∫ ∫
| f (y)| | f (y)|
dσ(y) = dσ(y)
Σ |x − y| n−1+ |α | Δ(x∗,2δΣ (x)) |x − y| n−1+ |α |
∞ ∫
 | f (y)|
+ dσ(y)
j=1 Δ(x∗,2 j+1 δΣ (x))\Δ(x∗,2 j δΣ (x)) |x − y| n−1+ |α |
⨏
≤ CδΣ (x)− |α | | f (y)| dσ(y)
Δ(x∗,2δΣ (x))


∞ ⨏
 − |α |
+C 2 j δΣ (x) | f (y)| dσ(y)
j=1 Δ(x∗,2 j+1 δΣ (x))

⨏  1/p
− |α |− n−1−λ n−1−λ
≤ CδΣ (x) p · δΣ (x) p | f | p dσ
Δ(x∗,2δΣ (x))


∞
 − |α |− n−1−λ
+C 2 j δΣ (x) p
×
j=1

 n−1−λ
⨏  1/p
× 2 j δΣ (x) p
| f | p dσ
Δ(x∗,2 j+1 δΣ (x))


∞ 
− |α |− n−1−λ n−1−λ
≤ CδΣ (x) p (2 j )− |α |− p  f  M p, λ (Σ,σ)
j=0

n−1−λ
= CδΣ (x)− |α |− p  f  M p, λ (Σ,σ) (2.6.24)

for some constant C ∈ (0, ∞) independent of x and f , based on (2.6.22), the Ahlfors
regularity of Σ, Hölder’s inequality, the definition of the Morrey norm from [113,
(6.2.2)], and bearing in mind that |α| + (n − 1 − λ)/p > 0. With (2.6.24) in hand, the
claim made in (2.6.5) readily follows.
The very last claim in item (1) is clear from what we have proved so far, the defini-
tion of M̊ p,λ (Σ, σ) in (A.0.88), and the fact that the operators involved are continuous
on L s (Σ, σ) with s := p(n−1)
n−1−λ ∈ (1, ∞) (cf. Theorem 2.3.2 and Theorem 2.4.1).
The claims in item (2) about the integral operators (2.6.6)-(2.6.8), recorded in
(2.6.9)-(2.6.10), may be justified based on what we have proved so far, making
use of the same spherical harmonics expansion technique used in the proof of
Theorem 2.5.38 and taking into account [113, (6.2.5)]. Alternatively, we may rely
on the very last property in the statement of Theorem 2.5.38 together with [113,
Proposition 6.2.12], as before. The very last claim in item (2) is handled as before.
Next, the claims in item (3) are dealt with in a very analogous fashion.
On to item (4), let us deal with the claims made in relation to the operator in
(2.6.14). These follow by combining (2.3.56) with the extrapolation result from [113,
Corollary 5.2.3] whose applicability in the present setting with X := B q,λ (∂Ω, σ)
520 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

is guaranteed by [113, Proposition 6.2.17] together with [113, Corollaries 6.2.11,

6.2.13].
Another, more direct, proof of the claims made in relation to the operator in
(2.6.14) is as follows. Fix an arbitrary
 B q,λ -block b on Σ. Then there exist some
point xo ∈ Σ and some radius R ∈ 0, 2 diam(Σ) such that
1

λ −1
supp b ⊆ B(xo, R) ∩ Σ and b L q (Σ,σ) ≤ R q . (2.6.25)

We may then use (2.3.18) (with p := q) to estimate

∫  1/q   
|(T b)(x)| q dσ(x) ≤ C  k S n−1 𝒞 N (S n−1 ) b L q (Σ,σ)
Σ

   1
λ −1
≤ Ck  S n−1

𝒞 N (S n−1 )
R q . (2.6.26)

Next, if x ∈ Σ is such that |x − xo | ≥ 2R then based on (2.6.25), the homogeneity of

the kernel, and the Ahlfors regularity of Σ we may estimate
∫ 
 
|(T b)(x)| =  k(x − y)b(y) dσ(y)
B(x o ,R)∩Σ
∫

≤ C sup |k | |x − xo | −(n−1) |b(y)| dσ(y)
S n−1 B(x o ,R)∩Σ
⨏

≤ C sup |k | |x − xo | −(n−1) Rn−1 |b| dσ
S n−1 B(x o ,R)∩Σ
 ⨏  1/q

≤ C sup |k | |x − xo | −(n−1) Rn−1 |b| q dσ
S n−1 B(x o ,R)∩Σ

 −
n−1
≤ C sup |k | |x − xo | −(n−1) Rn−1 R q b L q (Σ,σ)
S n−1
1
 −
n−1
λ
≤ C sup |k | |x − xo | −(n−1) Rn−1 R q R q −1
S n−1
1
−(n−1−λ) q −1
 R
≤ C sup |k | . (2.6.27)
S n−1 |x − xo | n−1

Collectively, (2.6.26), (2.6.27), and [113, Lemma 6.2.10] (used with θ := 0, which
is a permissible value in [113, (6.2.108)]) imply that there exists some finite constant
C = C(Σ, n, q, λ) > 0 with the property that
  
T b ∈ B q,λ (Σ, σ) and T b B q, λ (Σ,σ) ≤ C  k S n−1 𝒞 N (S n−1 )
(2.6.28)
for each given B q,λ -block b on Σ.
2.6 Singular Integrals on Morrey Spaces and Their Pre-Duals 521

Suppose now that some function f ∈ B q,λ (Σ, σ) has been given. Then there
exist some numerical sequence {λ j } j ∈N ∈  1 (N) along with a sequence {b j } j ∈N of

B q,λ -blocks on Σ such that f = ∞  λ j b j in the sense of distributions onr Σ. As
j=1
noted in [113, (6.2.73)], the series ∞ j=1 λ j b j actually converges to f in L (Σ, σ)
q(n−1)
with r := n−1+λ(q−1) ∈ (1, q). Also, (2.6.28) readily implies that the sequence
  J 
j=1 λ j b j J ∈N is Cauchy, hence convergent, in the Banach space B
T q,λ (Σ, σ).

 inL (Σ, σ) as well.

In light of [113, (6.2.71)], this latter convergence takes place r

However, since T is continuous on L r (Σ, σ), it follows that T Jj=1 λ j b j J ∈N is,

in fact, convergent to T f in L r (Σ, σ). Bearing (2.6.28) and (A.0.22) in mind, this
argument ultimately proves the following: given any f ∈ B q,λ (Σ, σ), regarding f
q(n−1)
as a function in L r (Σ, σ) with r := n−1+λ(q−1) ∈ (1, q) and considering T f in the
principal-value sense (cf. (2.3.18)) yields a function which satisfies

the function T f belongs to B q,λ (Σ, σ) and

   (2.6.29)
T f  B q, λ (Σ,σ) ≤ C  k S n−1 𝒞 N (S n−1 )  f  B q, λ (Σ,σ),

for some constant C = C(Σ, n, q, λ) ∈ (0, ∞) independent of f . Hence, the operator

T in (2.6.14) is well defined, linear, continuous, and satisfied the desired operator
norm bound.
To proceed, observe that for each function f ∈ M p,λ (Σ, σ) and each B q,λ -block
b on Σ we may write
∫ ∫
(T f )b dσ = − f (T b) dσ, (2.6.30)
Σ Σ

thanks to (2.3.26) and [113, (6.2.25)]. On account of [113, Proposition 6.2.8],

(A.0.21), [113, (6.2.67)], the fact that the operator in (2.6.1) is well defined and
the operator in (2.6.14) is continuous, this further implies that
∫ ∫
(T f )g dσ = − f (T g) dσ for each
Σ Σ (2.6.31)
functions f ∈ M p,λ (Σ, σ) and g ∈ B q,λ (Σ, σ).

We then conclude from (2.6.31) and [113, Proposition 6.2.8] that the real transpose
of the operator T in (2.6.14) is the opposite of the operator T in (2.6.1). The claim
in (2.6.15) is therefore established. In light of [113, (6.2.15)], we also deduce from
(2.6.31) that
∫ ∫
(T f )g dσ = − f (T g) dσ for each
Σ Σ (2.6.32)
functions f ∈ M̊ (Σ, σ) and g ∈ B (Σ, σ).
p,λ q,λ

In concert with [113, Proposition 6.2.16], this proves (2.6.16). Having established
this, (2.6.17) follows from (2.6.16) and the abstract (commutator type) estimate
(2.7.72), established independently of the current considerations.
522 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

There is an alternative approach to the transposition formulas just proved, which

is more resourceful. Specifically, recall from [113, (6.2.203)] that
   
M p,λ (Σ, σ) = B q,λ (Σ, σ) and B q,λ (Σ, σ) = M p,λ (Σ, σ), (2.6.33)

and note from [113, (5.2.100), (6.2.20), (6.2.70)] that

 ◦
∞ (Σ, σ)  ·  M p, λ (Σ, σ) coincides with M̊ p,λ (Σ, σ),
M p,λ (Σ, σ) := Lcomp
 ◦ (2.6.34)
∞ (Σ, σ)  · B q, λ (Σ, σ) coincides with B q,λ (Σ, σ).
B q,λ (Σ, σ) := Lcomp

Then (2.6.15) and (2.6.16) follow from (2.6.33)-(2.6.34) and the general transposition
formula obtained in (2.3.69).
Next, Cotlar’s inequality recorded in (2.3.23) (presently employed with r := 1)
shows that  a constant C ∈ (0, ∞) with the property that for each function
 there exists
σ(x)
g ∈ L Σ, 1+ |x | n−1 we have
1


(Tmax g)(x) ≤ CM Σ (T g)(x) + C M Σ g (x), ∀x ∈ Σ, (2.6.35)

where M Σ is the Hardy-Littlewood maximal operator on Σ. From this, [112, (6.2.72)],

(2.6.14), [113, Corollary 6.2.11], [113, (6.2.75)], and (2.3.7) we conclude that Tmax
in (2.6.18) is a well-defined, sub-linear, bounded, and continuous mapping.
As regards the nontangential maximal function estimate in (2.6.19), one route is
to rely on (2.4.5), [112, (6.2.72)], (2.6.18), [113, Corollary 6.2.11], [113, (6.2.75)],
and [112, (8.2.28)]. Another route is to combine (2.4.18), [113, Corollary 5.2.3],
[113, Proposition 6.2.17], and [113, Corollaries 6.2.11, 6.2.13].
Yest another argument for (2.6.19) is as follows. Fix some aperture  parameter
κ > 0 and pick an arbitrary B q,λ -block b on Σ. If xo ∈ Σ and R ∈ 0, 2 diam(Σ) are
such that the properties in (2.6.25) hold, we may employ (2.4.9) (with f := a and
p := q) to write
∫  c   1/q   
N Σ (T b)(x)q dσ(x) ≤ C  k S n−1 𝒞 N (S n−1 ) b L q (Σ,σ)
κ
Σ
1
   λ
≤ C  k S n−1 𝒞 N (S n−1 ) R q −1 . (2.6.36)

Also, whenever z ∈ Σ c is such that |z− xo | ≥ 2R, based on support and normalization
properties of the atom, the homogeneity of the kernel, and the Ahlfors regularity of
Σ we may estimate
2.6 Singular Integrals on Morrey Spaces and Their Pre-Duals 523
∫
|(T b)(z)| ≤ |k(z − y)||b(y)| dσ(y)
Σ∩B(x o ,R)
∫

≤ C sup |k | |z − xo | −(n−1) |b(y)| dσ(y)
S n−1 Σ∩B(x o ,R)
⨏

≤ C sup |k | |z − xo | −(n−1) Rn−1 |b| dσ
S n−1 Σ∩B(x o ,R)

 ⨏  1/q
−(n−1) n−1
≤ C sup |k | |z − xo | R |b| q dσ
S n−1 Σ∩B(x o ,R)

 −
n−1
≤ C sup |k | |z − xo | −(n−1) Rn−1 R q b L q (Σ,σ)
S n−1
1
 −
n−1
λ
≤ C sup |k | |z − xo | −(n−1) Rn−1 R q R q −1
S n−1
1
−(n−1−λ) q −1
 R
≤ C sup |k | . (2.6.37)
S n−1 |z − xo | n−1
 c
Now, if x ∈ Σ \ B xo, 2(2 + κ)R and z ∈ ΓΣκ (x), then

2(2 + κ)R ≤ |x − xo | ≤ |x − z| + |z − xo | ≤ (1 + κ) dist(z, Σ) + |z − xo |

≤ (1 + κ)|z − xo | + |z − xo | = (2 + κ)|z − xo |. (2.6.38)

In particular,
1
|z − xo | ≥ 2R and |z − xo | ≥ |x − xo |, (2.6.39)
2+κ
c
so (2.6.37) gives, after taking the supremum over z ∈ ΓΣκ (x),
1
−(n−1−λ) q −1
 Σc  
N (T b)(x) ≤ C sup |k | R
κ
S n−1 |x − xo | n−1 (2.6.40)

at each point x ∈ Σ \ B xo, 2(2 + κ)R .

From (2.6.36), (2.6.40), and [113, Lemma 6.2.10] used with θ := 0 (which
is a permissible value in [113, (6.2.108)]) we obtain that there exists some
C = C(Σ, n, κ, q, λ) ∈ (0, ∞) for which
  
NκΣ (T b) ∈ B q,λ (Σ, σ) and NκΣ (T b) B q, λ (Σ,σ) ≤ C  k S n−1 𝒞 N (S n−1 ) .
c c

for each given B q,λ -block b on Σ.

(2.6.41)
The end-game in the (second) proof of (2.6.19) is as follows. Having fixed a
function f ∈ B q,λ (Σ, σ), there exist a numerical sequence {λ j } j ∈N ∈  1 (N) along
524 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

with a sequence {b j } j ∈N of B q,λ -blocks on Σ such that f = ∞ λ b in the
 j j
j=1
sense of distributions on Σ. Thanks to [113, (6.2.73)], the series ∞ j=1 λ j b j ac-
q(n−1)
tually converges to f in L r (Σ, σ) with r := n−1+λ(q−1) ∈ (1, q). Collectively,
[112, (8.2.11)],
 c  [112, (8.2.9)], [113, (6.2.3)], and (2.6.41) also imply that the se-
quence Nκ T Σ
j=1 λ j b j
J
is Cauchy, hence convergent, in the Banach
J ∈N
space B q,λ (Σ, σ). In light of [113, (6.2.71)], this latter convergence takes place in
L r (Σ, σ) as well. However,
 thanks to (2.4.9) (with
    p := r), [112, (8.2.11)], andc [113,
Σ c
(6.2.3)], it follows that Nκ T j=1 λ j b j
J
actually converges to NκΣ (T f )
J ∈N
in L r (Σ, σ). By also taking (2.6.41) into account, this argument ultimately gives that
c
NκΣ (T f ) ∈ B q,λ (Σ, σ) and
 Σc     (2.6.42)
N (T f ) q, λ ≤ C  k S n−1 𝒞 N (S n−1 )  f  B q, λ (Σ,σ)
κ B (Σ,σ)

for some constant C = C(Σ, n, κ, q, λ) ∈ (0, ∞). This finishes the proof of (2.6.19).
Going further, the estimate recorded in (2.6.20) may be justified based on what we
have proved so far, making use of the same spherical harmonics expansion technique
used in the proof of Theorem 2.5.38 (as well as the boundedness results established
in Theorem 2.5.38). Finally, on account of [113, (6.2.74)], the same type of argument
may be used to justify the very last claim made in item (4).
Let us now deal with item (5). To justify the estimate claimed in (2.6.21), consider
first the case when Σ is unbounded. Pick an arbitrary ℋq,λ -atom a on Σ. Then there
exist some point xo ∈ Σ and some radius R ∈ (0, ∞) such that
1 ∫
λ −1
supp a ⊆ B(xo, R) ∩ Σ, a L q (Σ,σ) ≤ R q , a dσ = 0. (2.6.43)
Σ

In particular, b := a satisfies the properties in (2.6.25), so according to (2.6.36) we

have
∫  c  1/q 1
   
N Σ (T a)q dσ ≤ C  k  n−1  N n−1 Rλ q −1 . (2.6.44)
κ S 𝒞 (S )
Σ

In addition, whenever z ∈ Σ c is such that |z−xo | ≥ 2R, based on support, cancelation,

and normalization properties of the atom, the homogeneity of the kernel, the Mean
Value Theorem, and the Ahlfors regularity of Σ we may estimate
2.6 Singular Integrals on Morrey Spaces and Their Pre-Duals 525
∫ 
   
 
|(T a)(z)| =  k(z − y) − k(z − xo ) a(y) dσ(y)
 Σ∩B(xo,R) 
∫
≤ |k(z − y) − k(z − xo )||a(y)| dσ(y)
Σ∩B(x o ,R)
∫
 −n
≤ CR sup |∇k | |z − xo | |a(y)| dσ(y)
S n−1 Σ∩B(x o ,R)
⨏
n −n
≤ CR sup |∇k | |z − xo | |a| dσ
S n−1 Σ∩B(x o ,R)
 ⨏  1/q

≤ CRn sup |∇k | |z − xo | −n |a| q dσ(y)
S n−1 Σ∩B(x o ,R)

 −
n−1
≤ CRn sup |∇k | |z − xo | −n R q a L q (Σ,σ)
S n−1

n
n−1 1
− λ −1
≤ CR sup |∇k | |z − xo | −n R q R q
S n−1

1
−(n−1−λ) q −1
 R
≤ CR sup |∇k | . (2.6.45)
S n−1 |z − xo | n
 c
Next, if x ∈ Σ \ B xo, 2(2 + κ)R and z ∈ ΓΣκ (x) then (2.6.39) holds. Combining this
c
with (2.6.45) and taking the supremum over z ∈ ΓΣκ (x) we arrive at the conclusion
that 1
1−(n−1−λ) q −1
 Σc   R
N (T a)(x) ≤ C sup |∇k |
κ
S n−1 |x − xo | n (2.6.46)

at each point x ∈ Σ \ B xo, 2(2 + κ)R .
In turn, from (2.6.44), (2.6.46), and [113, Definition 6.2.23] we conclude that
c
NκΣ (T a) is a fixed multiple of what we call a (q, λ)-dull molecule concen-
trated near Δ(xo, R) := Σ ∩ B(xo, R). In view of this and the nature of the
constants involved, from [113, (6.2.251)] we see that there exists some constant
C = C(Σ, n, κ, q, λ, λ) ∈ (0, ∞) for which
c
NκΣ (T a) ∈ ℳq,λ (Σ, σ) and
  
N Σ (T a)ℳq, λ (Σ,σ) ≤ C  k  n−1  N
c
κ S 𝒞 (S n−1 )
, (2.6.47)
for each ℋq,λ -atom a on Σ.

When Σ is compact, the constant 1 is also considered to be an ℋq,λ -atom on Σ. That,

in such a scenario, the properties listed in (2.6.47) are also valid for the function a = 1
may be seen directly from (2.4.9), [113, Definition 6.2.23], and [113, (6.2.251)].
526 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Having established (2.6.47) in all cases, the end-game in the proof of (2.6.21)
is very similar to the argument that, starting with (2.6.41), has produced (2.6.42).
Finally, the variable coefficient version of (2.6.21) may be established much as
before.
We close with some comments on the particular case when Σ is compact. In
such a scenario, we know from [113, (6.2.77)] that B q,λ (Σ, σ) = ℋq,λ (Σ, σ). As
such, the results in item (4) may be formulated in terms of the latter space. In fact,
it is possible to give an alternative proof for the boundedness of T on ℋq,λ (Σ, σ)
when Σ is compact, based on the atomic description of this space given in (A.0.55).
Specifically, having fixed a parameter θ ∈ 0, qn − n + 1 and with a ∈ L q (Σ, σ) an
ℋq,λ -atom on Σ as in [113, (6.1.15)], we may use (2.3.18) (with p := q) to estimate
∫
|(T a)(x)| q |x − xo | θ dσ(x)
B(x o ,2R)∩Σ
∫
≤ CRθ · |T a| q dσ ≤ CRθ T a L q (Σ,σ)
q

B(x o ,2R)∩Σ
  q
≤ Ck   Rθ a L q (Σ,σ)
q
S n−1 𝒞 N (S n−1 )
  q
≤ C  k S n−1 𝒞 N (S n−1 ) Rλ(1−q)+θ . (2.6.48)

Next, if x ∈ Σ is such that |x − xo | ≥ 2R then based on [113, (6.1.15)], the Mean

Value Theorem, and the Ahlfors regularity of Σ we may estimate
∫ 
 
|(T a)(x)| =  k(x − y)a(y) dσ(y)
Σ
∫
≤ |k(x − y) − k(x − xo )||a(y)| dσ(y)
B(x o ,R)∩Σ
∫
   R
≤ Ck  S n−1

𝒞1 (S n−1 ) |x − x | n
|a(y)| dσ(y)
o B(x o ,R)∩Σ
   R 
≤ C  k S n−1 𝒞1 (S n−1 )
1−1/q
a L q (Σ,σ) σ B(xo, r) ∩ Σ
|x − xo | n
   R1+(n−1−λ)(1−1/q)
≤ C  k S n−1 𝒞1 (S n−1 ) . (2.6.49)
|x − xo | n
With this in hand, [112, Lemma 7.2.1] applies since qn − θ > n − 1 and gives
2.6 Singular Integrals on Morrey Spaces and Their Pre-Duals 527
∫
|(T a)(x)| q |x − xo | θ dσ(x)
Σ\B(x o ,2R)
∫
  q dσ(x)
≤ C  k S n−1 𝒞1 (S n−1 ) R1+(n−1−λ)(q−1)
Σ\B(x o ,2R) |x − xo | qn−θ
  q
≤ Ck  S n−1
 1
𝒞 (S n−1 )
R 1+(n−1−λ)(q−1)
· Rn−1−nq+θ
  q
= C  k S n−1 𝒞1 (S n−1 ) Rλ(1−q)+θ . (2.6.50)

Collectively, (2.6.48) and (2.6.50) imply that whenever a is an in [113, (6.1.15)] we

have
∫
  q
|(T a)(x)| q |x − xo | θ dσ(x) ≤ C  k S n−1 𝒞 N (S n−1 ) Rλ(1−q)+θ , (2.6.51)
Σ

for some finite constant C = C(Σ, n, q, λ, θ) > 0 independent

 of a. Specializing
(2.6.51) first to the case when θ := 0 and then fixing θ ∈ (n − 1)(q − 1), qn − n + 1
proves that the function T a is a fixed multiple of some ℋq,λ,θ -dome on Σ (cf. [113,
(6.1.33)]). Since (2.3.18) (with p := q) implies that T1 ∈ L q (Σ, σ), we conclude
from this and [113, (6.2.54)] that the same conclusion is valid when a ≡ constant.
With θ ∈ (n − 1)(q − 1), qn − n + 1 fixed, the argument so far shows that

for each ℋq,λ -atom a on Σ, the function T a is

(2.6.52)
a fixed multiple of a ℋq,λ,θ -dome on Σ.
Ultimately, from (2.6.52) and [113, (6.2.59)] we conclude that there exists some
C = C(Σ, n, q, λ) ∈ (0, ∞) with the property that
  
T a ∈ ℋq,λ (Σ, σ) and T aℋq, λ (Σ,σ) ≤ C  k S n−1 𝒞 N (S n−1 ) .
(2.6.53)
for each given ℋq,λ -atom a on Σ.

Granted this, the same type of reasoning which starting with (2.6.28) has produced
(2.6.29) now leads to the conclusion that T is a bounded endomorphism of ℋq,λ (Σ, σ)
when Σ is compact.
In fact, considerations of a similar nature (cf. also (2.4.79)-(2.4.87) for relevant
estimates in fairly analogous circumstances)
 may be used to show that if Σ is compact
then, having picked a parameter θ ∈ (n − 1)(q − 1), qn − n + 1 , we have that
c
for each ℋq,λ -atom a on Σ, the function NκΣ (T a) is
(2.6.54)
a fixed multiple of some ℋq,λ,θ -dome on Σ.
In turn, from (2.6.54) and [113, (6.2.59)] we deduce that there exists some constant
C = C(Σ, n, κ, q, λ) ∈ (0, ∞) for which
  
NκΣ (T a) ∈ ℋq,λ (Σ, σ) and NκΣ (T a)ℋq, λ (Σ,σ) ≤ C  k S n−1 𝒞 N (S n−1 ) .
c c

for each given ℋq,λ -atom a on Σ.

(2.6.55)
528 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Ultimately, this proves that for each function f ∈ ℋq,λ (Σ, σ) we have
c
NκΣ (T f ) ∈ ℋq,λ (Σ, σ) and
 Σc     (2.6.56)
N (T f ) q, λ ≤ C  k S n−1 𝒞 N (S n−1 )  f ℋq, λ (Σ,σ)
κ ℋ (Σ,σ)

for some constant C = C(Σ, n, κ, q, λ) ∈ (0, ∞). This provides an alternative proof
of (2.6.19) in the case when Σ is compact. 

For a multitude of reasons it is important to augment Theorem 2.6.1 with similar

results for weakly singular integral operators. We do so in the next two propositions.
First, we deal with boundary-to-boundary weakly singular integral operators, as
indicated below.

Proposition 2.6.2 Suppose Σ ⊆ Rn (where n ∈ N with n ≥ 2) is a closed Ahlfors

regular set and abbreviate σ := H n−1 Σ. Fix a parameter λ ∈ (0, n − 1) along
with an exponent α ∈ (0, n − 1 − λ), and consider some σ ⊗ σ-measurable function
θ : Σ × Σ → C with the property that there exists a constant C ∈ (0, ∞) such that

|θ(x, y)| ≤ C|x − y| −(n−1−α) for a.e. x, y ∈ Σ. (2.6.57)

Use this to define the integral operator


Θ : L 1 Σ, 1+ |xσ(x)
| n−1−α
−→ Lloc 1
(Σ, σ)

acting on each f ∈ L 1 Σ, 1+ |xσ(x)
| n−1−α
according to (2.6.58)
∫

Θ f (x) := θ(x, y) f (y) dσ(y) for σ-a.e. x ∈ Σ.
Σ

Finally, assume that

n−1−λ 1 α  −1
1<p< and p∗ := − (2.6.59)
α p n−1−λ
and recall from [113, (6.2.140), (6.2.230)] that there are well-defined, linear, and
continuous embeddings
 σ(x) 
M p,λ (Σ, σ) → L 1 Σ, , (2.6.60)
1 + |x| n−1−α

 σ(x) 
B p,λ (Σ, σ) → L 1 Σ, . (2.6.61)
1 + |x| n−1−α
Then the operators induced via (2.6.60)-(2.6.61) by Θ from (2.6.58) in the contexts
2.6 Singular Integrals on Morrey Spaces and Their Pre-Duals 529

Θ : M p,λ (Σ, σ) −→ M p∗,λ (Σ, σ), (2.6.62)

Θ : B p,λ (Σ, σ) −→ B p∗,λ (Σ, σ), (2.6.63)

Θ : M̊ p,λ (Σ, σ) −→ M̊ p∗,λ (Σ, σ), (2.6.64)

are well-defined, linear, and bounded.

Proof For starters, under the assumptions made in (2.6.59), the embedding recorded
in [112, (6.2.23)] guarantees that (2.6.60) holds. Going further, let IΣ,α denote the
fractional integral operator of order α on Σ. Since there exists a constant C ∈ (0, ∞)
with the property that

for each function f ∈ L 1 Σ, 1+ |xσ(x)
| n−1−α
we have
(2.6.65)
|(Θ f )(x)| ≤ C IΣ,α (| f |)(x) at σ-a.e. point x ∈ Σ,

the fact that the operator Θ in (2.6.62) is well defined, linear, and bounded is implied
by [113, Proposition 6.2.14] and [113, (6.2.3)], while the fact that the operator Θ in
(2.6.63) is a consequence of [113, Proposition 6.2.20] and [113, (6.2.75)].
As regards the claims made in relation to the operator Θ in (2.6.64), recall from
(A.0.88) that M̊ p,λ (Σ, σ) is the closure of L s (Σ, σ) with s := p(n−1)
n−1−λ in M
p,λ (Σ, σ),
,λ p (n−1) ,λ
and that M̊ ∗ (Σ, σ) is the closure of L ∗ (Σ, σ) with s∗ := n−1−λ in M ∗ (Σ, σ). In
p s ∗ p

particular, from definitions and (2.6.59) we see that

1 1 1
s ∈ (1, n − 1) and = − . (2.6.66)
s∗ s n−1
Granted this, we conclude from (2.6.65) and [112, (7.8.7)] that Θ maps L s (∂Ω, σ)
linearly and boundedly into L s∗ (∂Ω, σ). In concert with (2.6.62) and the definitions
of the spaces M̊ p,λ (Σ, σ), M̊ p∗,λ (Σ, σ) just recalled above, this further implies that
Θ maps the space M̊ s,λ (∂Ω, σ) in a linear and bounded fashion into the space
M̊ s∗,λ (∂Ω, σ). 

The second proposition alluded to above, further complementing the body of

results from Theorem 2.6.1 and Proposition 2.6.2, deals with boundary-to-domain
weakly singular integral operators on Morrey spaces and their pre-duals.

Proposition 2.6.3 Let Ω ⊆ Rn (where n ∈ N with n ≥ 2) be an open set with an

Ahlfors regular boundary and abbreviate σ := H n−1 ∂Ω. Also, pick λ ∈ (0, n − 1)
and α ∈ (0, n − 1 − λ). In this context, suppose b(·, ·) is a Borel-measurable function
on Ω × ∂Ω with the property that there exists some constant C ∈ (0, ∞) such that

for L n -a.e. point x ∈ Ω one has |b(x, y)| ≤ C |x − y| −(n−1−α)

(2.6.67)
at σ-a.e. point y belonging to ∂Ω.

Use this function as a kernel to define the integral operator

530 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
ℬ f (x) := b(x, y) f (y) dσ(y) at L n -a.e. x ∈ Ω,
∂Ω
 σ(x)  (2.6.68)
for each function f ∈ L 1 ∂Ω, .
1 + |x| n−1−α
Finally, fix an aperture parameter κ > 0, assume that

n−1−λ 1 α  −1
1<p< and p∗ := − . (2.6.69)
α p n−1−λ
Then the integral operator ℬ acts in a meaningful manner on any of the spaces
M p,λ (∂Ω, σ), B p,λ (∂Ω, σ), M̊ p,λ (∂Ω, σ), and there exists a constant C ∈ (0, ∞)
such that
 
Nκ (ℬ f ) p , λ ≤ C f  M p, λ (∂Ω,σ) for each f ∈ M p,λ (∂Ω, σ),
M ∗ (∂Ω,σ)

hence, in particular, for each function f ∈ M̊ p,λ (∂Ω, σ),

(2.6.70)
and
 
Nκ (ℬ f ) p , λ ≤ C f  B p, λ (∂Ω,σ) for each f ∈ B p,λ (∂Ω, σ). (2.6.71)
B ∗ (∂Ω,σ)

Moreover, under the additional assumption that

for σ-a.e. point y ∈ ∂Ω the function
(2.6.72)
b(·, y) extends continuously to Ω \ {y}
κ−n.t.
it follows that the nontangential boundary limit ℬ f ∂Ω exists at σ-a.e. point on
Aκ (∂Ω) and, in fact,
 ∫
κ−n.t. 
ℬ f ∂Ω (x) = b(x, y) f (y) dσ(y) at σ-a.e. point x ∈ Aκ (∂Ω), (2.6.73)
∂Ω

for each function f belonging to any of the spaces M p,λ (∂Ω, σ), B p,λ (∂Ω, σ),
M̊ p,λ (∂Ω, σ).

Proof For starters, the fact that the integral operator ℬ acts in a meaningful manner
on any of the spaces M p,λ (∂Ω, σ), B p,λ (∂Ω, σ), M̊ p,λ (∂Ω, σ) follows from ob-
serving that the embeddings from [113, (6.2.140), (6.2.230)] hold with Σ := ∂Ω. In
concert with (2.5.555), this observation also guarantees the validity of the nontan-
gential boundary trace formula stated in (2.6.73), under the additional assumption
made in (2.6.72). As such, there remains to justify the estimates claimed in (2.6.70)-
(2.6.71). To this end, let I∂Ω,α denote the fractional integral operator of order α on
∂Ω and recall from (2.5.560) that there exists C ∈ (0, ∞) with the property that for

each f ∈ L 1 ∂Ω, 1+ |xσ(x)
| n−1−α
we have

Nκ (ℬ f ) (x) ≤ C I∂Ω,α (| f |)(x), ∀x ∈ ∂Ω. (2.6.74)
2.7 Commutator Estimates 531

Granted this, (2.6.70) follows on account of [113, Proposition 6.2.14], [113, (6.2.3)],
and [112, (8.2.26)]. Likewise, (2.6.71) is implied by (2.6.74), [113, Proposi-
tion 6.2.20], [113, (6.2.75)], and [112, (8.2.26)]. 

2.7 Commutator Estimates

Our first result, which refines [63, Theorem 2.16, p. 2603], is a combination of the
extrapolation theorem of Rubio de Francia with the commutator theorem of Coifman
et al., [21], suitably adapted to setting of spaces of homogeneous type. The reader is
reminded that the semi-norm  · BMO . has been introduced in (A.0.14).

Theorem 2.7.1 Let Σ ⊆ Rn be a closed Ahlfors regular set, and set σ := H n−1 Σ .
Fix p0 ∈ (1, ∞) along with some non-decreasing function Φ : (0, ∞) → (0, ∞) and
let T be a linear operator which is bounded on L p0 (Σ, ω) for every ω ∈ Ap0 (Σ, σ),
with operator norm ≤ Φ [ω] A p0 .
Then for each integrability exponent p ∈ (1, ∞) there exist C1, C2 ∈ (0, ∞) which
depend exclusively on the dimension n, the exponents p0, p, and the Ahlfors regularity
constants of Σ, with the property that for any Muckenhoupt weight w ∈ Ap (Σ, σ) the
operator

T : L p (Σ, w) −→ L p (Σ, w) (2.7.1)

is well defined, linear, and bounded, with operator norm27

 
1+(p −1)/(p−1)
T  L p (Σ,w)→L p (Σ,w) ≤ C1 · Φ C2 · [w] A p 0 . (2.7.2)

In addition, given any exponent p ∈ (1, ∞) along with some weight w ∈ Ap (Σ, σ),
there exists a constant C = C(Σ, n, p0, p, [w] A p ) ∈ (0, ∞), which stays bounded if
[w] A p stays bounded, and with the property that for every complex-valued function
b ∈ L ∞ (Σ, σ) one has (with C1 as before)
 
[Mb, T] p
L (Σ,w)→L p (Σ,w)
.
≤ C1 · Φ(C)bBMO(Σ,σ) , (2.7.3)

where [Mb, T] is the commutator of Mb , the operator of pointwise multiplication on

L p (Σ, w) by the function b, with T considered as in (2.7.1), i.e.,

[Mb, T] f := b(T f ) − T(b f ) for each f ∈ L p (Σ, w). (2.7.4)

Of course, ordinary Lebesgue spaces are included in (2.7.3). Indeed, specializing

(2.7.3) to the case when w ≡ 1 yields

27 by adapting to the setting of measure metric spaces the Euclidean argument employed in [36, The-
orem 3.2], [24, Theorem 3.22, p.40], and also making use of the estimates
 recorded in [112, (7.7.14)],

max{1,(p0 −1)/(p−1)}
it is possible to improve (2.7.2) to T  L p (Σ, w)→L p (Σ, w) ≤ C1 · Φ C2 · [w] A p
532 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 
[Mb, T] p .
≤ C1 · Φ(C)bBMO(Σ,σ) . (2.7.5)
L (Σ,σ)→L p (Σ,σ)

In particular, from (2.7.5) and real interpolation we obtain the following commutator
estimate on Lorentz spaces: For each p ∈ (1, ∞) and q ∈ (0, ∞] there exists
C = C(Σ, n, p, q) ∈ (0, ∞) with the property that for every complex-valued function
b ∈ L ∞ (Σ, σ) one has
 
[Mb, T] p, q .
≤ C1 · Φ(C)bBMO(Σ,σ) . (2.7.6)
L (Σ,σ)→L p, q (Σ,σ)

We also wish to note that the formulation of Theorem 2.7.1 is such that this
result may be iterated. Specifically, in view of (2.7.3), the commutator [Mb, T] has
the same nature as the original operator T, so another application of Theorem 2.7.1
gives that there exists a constant C ∈ (0, ∞) with the property that for any two
functions b1, b2 ∈ L ∞ (Σ, σ) we have
 
  . .
 Mb1 , [Mb2 , T]  p p
≤ Cb1 BMO(Σ,σ) b2 BMO(Σ,σ) . (2.7.7)
L (Σ,w)→L (Σ,w)

Inductively, it follows that for each m ∈ N there exists a constant C ∈ (0, ∞) with
the property that for any family of functions b1, . . . , bm ∈ L ∞ (Σ, σ) we have
  8
m
  .
 Mb1 , · · · Mbm−1 , [Mbm , T] · · ·  ≤C b j BMO(Σ,σ) .
L p (Σ,w)→L p (Σ,w)
j=1
(2.7.8)

Parenthetically, we wish to note that if T is an integral operator on Σ with kernel

k(x, y), then the iterated commutator Mb1 , · · · Mbm−1 , [Mbm , T] · · · is also an
9 
integral operator on Σ, whose kernel is k(x, y) m j=1 b j (x) − b j (y) .

The proof of Theorem 2.7.1 is given next.

Proof of Theorem 2.7.1 From [112, (7.7.82)] we know that

L p0 (Σ, ω) = L p (Σ, w), (2.7.9)

ω ∈ A p0 (Σ,σ) 1<p<∞
w ∈ A p (Σ,σ)

and the current assumptions ensure that the operator T is a well-defined mapping in
the context
T: L p0 (Σ, ω) −→ L p0 (Σ, ω). (2.7.10)
ω ∈ A p0 (Σ,σ) ω ∈ A p0 (Σ,σ)

For an arbitrary
f ∈ L p0 (Σ, ω), (2.7.11)
ω ∈ A p0 (Σ,σ)
2.7 Commutator Estimates 533

it follows that T f is a well-defined σ-measurable function on Σ and, as hypothesized

in the statement of the theorem, we have
∫  1/p0  ∫  1/p0
|T f | p0 dω ≤ Φ [ω] A p0 | f | p0 dω for each ω ∈ Ap0 (Σ, σ).
Σ Σ
(2.7.12)
Granted this, Rubio de Francia’s extrapolation theorem (recalled in [112, Proposi-
tion 7.7.6]) applies and gives that for each p ∈ (1, ∞) there exists C ∈ (0, ∞), which
depends only on p, p0 , and the Ahlfors regularity constants of Σ, with the property
that for each w ∈ Ap (Σ, σ) we have
∫  1/p  ∫  1/p
1+(p −1)/(p−1)
|T f | p dw ≤ 4 · Φ C · [w] A p 0 | f | p dw . (2.7.13)
Σ Σ

From (2.7.9), (2.7.11), and (2.7.13) we then conclude that the linear operator T
induces a bounded linear mapping on L p (Σ, w), whose operator norm is controlled
as in (2.7.2). This takes care of first claim in the statement of the theorem.
To proceed, fix p ∈ (1, ∞) and w ∈ Ap (Σ, σ). We propose to show that
 
[Mb, T] p ≤ C1 · Φ(C)bBMO(Σ,σ) . (2.7.14)
L (Σ,w)→L p (Σ,w)

To justify this estimate, we shall adapt arguments from [21], [75], [63]. First, from
simple linearity and homogeneity considerations, there is no loss of generality in
assuming that b ∈ L ∞ (Σ, σ) is actually real-valued and satisfies bBMO(Σ,σ) = 1
(the case when b is constant28 is trivial). From item (12) of [112, Lemma 7.7.1] we
know that there exists some small ε = ε(Σ, p, [w] A p ) > 0 with the property that for
each complex number z with |z| ≤ ε we have

w · e(Re z)b ∈ Ap (Σ, σ) with w · e(Re z)b Ap

≤ C, (2.7.15)

where the constant C = C(Σ, p, [w] A p ) ∈ (0, ∞) is independent of z. The idea is

now to observe that
  
Φ : z ∈ C : |z| < ε/2 −→ Bd L p (Σ, w) defined as
(2.7.16)
Φ(z) := Mez b T Me−z b for each z ∈ C with |z| < ε/2,

is an analytic map which, for each z ∈ C with |z| < ε/2 and each f ∈ L p (Σ, w),
satisfies

28 the only scenario which would prevent dividing by b BMO(Σ, σ) = 0

534 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
 
Φ(z) f (x) p w(x) dσ(x)
Σ
∫
 −zb p
= T(e f )(x) w(x) · e(Re z)b(x) dσ(x)
Σ
p
≤ T  L p (Σ,w ·e(Re z)b )→L p (Σ,w ·e(Re z)b ) ×
∫
 p
× e−zb(x) f (x) w(x) · e(Re z)b(x) dσ(x)
Σ
 p
≤ C1 · Φ C2 · C 1+(p0 −1)/(p−1)  f  L p (Σ,w),
p p
(2.7.17)

thanks to (2.7.16), (2.7.15), and (2.7.2). In addition, from (2.7.16) and Cauchy’s
reproducing formula for analytic functions we see that
∫
1 Φ(z)
[Mb, T] = Φj (0) = dz. (2.7.18)
2πi |z |=ε/4 z 2

Consequently, for each f ∈ L p (Σ, w) and x ∈ Σ, we have

∫
   
[Mb, T] f (x) ≤ 8 Φ(z) f (x) dH 1 (z), (2.7.19)
πε |z |=ε/4
2

hence
   p ∫  
[Mb, T] f (x) p ≤ 8 Φ(z) f (x) p dH 1 (z). (2.7.20)
πε 2
|z |=ε/4

From (2.7.1) and (2.7.4) we see that for each f ∈ L p (Σ, w) the function [Mb, T] f
is σ-measurable. In concert with (2.7.20) and (2.7.17), this permits us to estimate
∫
 
[Mb, T] f (x) p dw(x)
Σ
 
 8 p ∫ ∫
 
≤ Φ(z) f (x) p dH 1 (z) dw(x)
πε 2
Σ |z |=ε/4
∫ ∫ 
 8 p  
= Φ(z) f (x) p dw(x) dH 1 (z)
πε 2
|z |=ε/4 Σ

 23p−1  p  p
1+(p0 −1)/(p−1) p
≤ C · Φ C2 · C  f  L p (Σ,w) . (2.7.21)
π p−1 ε 2p−1 1
In turn, (2.7.21) implies (2.7.14).
The end-game in the proof is as follows. If Σ is unbounded, then (2.7.3) is a
direct consequence of (2.7.14) and (A.0.15). If Σ is bounded, then (2.7.3) is obtained
2.7 Commutator Estimates 535
⨏
by applying (2.7.14) to the function  b := b − Σ b dσ. Indeed,  b ∈ L ∞ (Σ, σ)
and, as seen from (A.0.15) and (2.7.4), we have  .
bBMO(Σ,σ) = bBMO(Σ,σ) and
[Mb, T] = [Mb, T]. Then (2.7.3) readily follows from these observations. 

In turn, Theorem 2.7.1 is a basic ingredient in the proof of the following com-
mutator estimate on Muckenhoupt weighted Lebesgue spaces for convolution type
singular integral operators.

Theorem 2.7.2 Suppose Σ ⊆ Rn is a UR set, and abbreviate σ := H n−1 Σ. Given

a sufficiently large integer N = N(n) ∈ N, fix a function k ∈ 𝒞 N (Rn \ {0}) which
is complex-valued, odd, and positive homogeneous of degree 1 − n. Also, pick an
exponent p ∈ (1, ∞) along with a Muckenhoupt weight w ∈ Ap (Σ, σ) and recall from
item (10) in Theorem 2.3.2 (cf. (2.3.56)) that the singular integral operator defined

originally on functions f ∈ L 1 Σ, 1+σ(x)
|x | n−1
as
∫
T f (x) := lim+ k(x − y) f (y) dσ(y) for σ-a.e. x ∈ Σ, (2.7.22)
ε→0
y ∈Σ
|x−y |>ε

induces a well-defined, linear, and bounded mapping

T : L p (Σ, w) −→ L p (Σ, w). (2.7.23)

Then there exists a constant C ∈ (0, ∞), which depends only on n, k, p, [w] A p ,
and the UR character of Σ, such that for each function b ∈ L ∞ (Σ, σ) one has
 
[T, Mb ] p
L (Σ,w)→L p (Σ,w)
.
≤ CbBMO(Σ,σ) . (2.7.24)

More generally, consider a function b(x, z) which is odd and positive homo-
geneous of degree 1 − n in the variable z ∈ Rn \ {0}, and such that ∂zα b(x, z)
is continuous and bounded on Rn × S n−1 for each multi-index α ∈ N0n satisfying
|α| ≤ M where M = M(n) is a sufficiently large positive integer. Then the singular

integral operators acting on functions f ∈ L 1 Σ , 1+σ(x)
|x | n−1
via
∫
B f (x) := lim+ b(x, x − y) f (y) dσ(y) for σ-a.e. x ∈ Σ, (2.7.25)
ε→0
y ∈Σ
|x−y |>ε

and, respectively,
∫
B f (x) := lim+
#
b(y, x − y) f (y) dσ(y) for σ-a.e. x ∈ Σ, (2.7.26)
ε→0
y ∈Σ
|x−y |>ε

induce well-defined, linear, and bounded mappings

536 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

B, B# : L p (Σ, w) −→ L p (Σ, w) (2.7.27)

and there exists a constant C ∈ (0, ∞), which depends only on n, b, p, [w] A p , and
the UR character of Σ, such that for each function ψ ∈ L ∞ (Σ, σ) one has
    
max [B, Mψ ] L p (Σ,w)→L p (Σ,w) , [B#, Mψ ] L p (Σ,w)→L p (Σ,w)

.
≤ CψBMO(Σ,σ) . (2.7.28)

Proof The claims pertaining to (2.7.23)-(2.7.24) are consequences of Theorem 2.7.1

and item (10) in Theorem 2.3.2, while the claims made in (2.7.27)-(2.7.28) are seen
from Theorem 2.7.1 and item (e) in Theorem 2.5.38. 

In the following companion result to Theorem 2.7.2 we estimate the essential

norm of the commutator (2.7.4) in the case when the underlying set is compact.

Theorem 2.7.3 Let Σ ⊆ Rn be a compact UR set, and abbreviate σ := H n−1 Σ.

Having fixed a sufficiently large integer N = N(n) ∈ N, consider a complex-valued
function k ∈ 𝒞 N (Rn \ {0}) is which is odd and positive homogeneous of degree
1 − n. Pick an exponent p ∈ (1, ∞) along with a Muckenhoupt weight w ∈ Ap (Σ, σ)
and recall from item (10) in Theorem 2.3.2 (cf. (2.3.56)) that the singular integral
operator defined originally on functions f ∈ L 1 (Σ, σ) as
∫
T f (x) := lim+ k(x − y) f (y) dσ(y) for σ-a.e. x ∈ Σ, (2.7.29)
ε→0
y ∈Σ
|x−y |>ε

induces a well-defined, linear, and bounded mapping

T : L p (Σ, w) −→ L p (Σ, w). (2.7.30)

Then there exists a constant C ∈ (0, ∞), which depends only on n, p, [w] A p , and
the UR character of Σ, such that for each function b ∈ L ∞ (Σ, σ) one has
      
dist [T, Mb ], Cp L p (Σ, w) ≤ C sup |∂ α k | · dist b, VMO(Σ, σ)
n−1
|α | ≤ N S
 (2.7.31)
where the distance in the left-hand side is measured in Bd L p (Σ, w) and the distance
in the right-hand side is measured in BMO(Σ, σ).
As a consequence of (2.7.31) with w ≡ 1, real interpolation, and [113, Propo-
sition 1.4.24] (whose applicability for Lebesgue spaces is ensured by item (3) in
[113, Proposition 7.3.6] together with the identification in [113, (7.1.55)]), for each
q ∈ (0, ∞] there exists a constant C ∈ (0, ∞), which depends only on n, p, q, and
the UR character of Σ, such that for each function b ∈ L ∞ (Σ, σ) one has
2.7 Commutator Estimates 537
      
dist [T, Mb ], Cp L p,q (Σ, σ) ≤C sup |∂ α k | dist b, VMO(Σ, σ)
n−1
|α | ≤ N S
 (2.7.32)
where the distance in the left-hand side is measured in Bd L p,q (Σ, σ) and the
distance in the right-hand side is measured in BMO(Σ, σ).
In particular,

[T, Mb ] ∈ Cp L p (Σ, w) whenever b ∈ L ∞ (Σ, σ) ∩ VMO(Σ, σ), (2.7.33)

and, for each q ∈ (0, ∞],


[T, Mb ] ∈ Cp L p,q (Σ, σ) whenever b ∈ L ∞ (Σ, σ) ∩ VMO(Σ, σ). (2.7.34)

Utilizing the piece of notation introduced in (A.0.102), we may rephrase (2.7.31)

and (2.7.32) as
 ess    
[T, Mb ] p ≤C sup |∂ α k | · dist b, VMO(Σ, σ) (2.7.35)
L (Σ,w)→L p (Σ,w)
n−1
|α | ≤ N S

and, respectively,
 ess    
[T, Mb ] p, q ≤C sup |∂ α k | · dist b, VMO(Σ, σ) .
L (Σ,σ)→L p, q (Σ,σ)
n−1
|α | ≤ N S
(2.7.36)
We can also prove a version of Theorem 2.7.3 for commutators involving singular
integral operators with variable coefficient kernels. To elaborate, assume that

b(x, z) is a function which is odd and positive homogeneous of degree

1−n in the variable z ∈ Rn \{0}, and such that ∂zα b(x, z) is continuous
(2.7.37)
and bounded on Rn × S n−1 for each multi-index α ∈ N0n satisfying
|α| ≤ M, where M = M(n) is a sufficiently large positive integer.

Given a compact UR set Σ ⊆ Rn , abbreviate σ := H n−1 Σ. For each function


f ∈ L 1 Σ, 1+σ(x)
|x | n−1
define
∫
B f (x) := lim+ b(x, x − y) f (y) dσ(y) for σ-a.e. x ∈ Σ. (2.7.38)
ε→0
y ∈Σ
|x−y |>ε

Then for each p ∈ (1, ∞) and w ∈ Ap (Σ, σ) there exists some C ∈ (0, ∞), depending
only on n, p, [w] A p , and Σ, with the property that for any function ψ ∈ L ∞ (Σ, σ)
we have
 ess 
[B, Mψ ] p = dist [B, Mψ ] , Cp(L p (Σ, w))
L (Σ,w)→L p (∂Ω,w)

≤ C · Cb · dist ψ , VMO(Σ, σ) , (2.7.39)
538 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

where  α 
Cb := sup  ∂ b (x, z) ∈ (0, ∞). (2.7.40)
z
(x,z)∈R n ×S n−1
|α | ≤M

To justify (2.7.39), we shall avail ourselves of notation and results from the proof of
Theorem 2.5.38. Much as with the decomposition (2.5.518), we presently obtain
 
B= ai Bi (2.7.41)
 ∈2N0 +1 1≤i ≤H
 
with convergence in B L p (Σ, w) → L p (Σ, w) . For an arbitrary fixed threshold
μ ∈ N, split
   
B= ai Bi + ai Bi . (2.7.42)
 ∈2N0 +1 1≤i ≤H  ∈2N0 +1 1≤i ≤H
 ≤μ >μ

Similarly to (2.5.505), we may estimate

 
 
   
  C(Ω, b, p, w, n)
 i 
ai B  ≤ . (2.7.43)
 μ
 ∈2+1N0 1≤i ≤H 
 >μ  p
L (Σ,w)→L p (Σ,w)

∞
 p function ψ ∈ L (Σ, σ). For each  ∈ 2N0 + 1 and 1 ≤ i ≤ H , let
Fix an arbitrary
C ∈ Cp L (Σ, w) be such that
i

 i  
[B , Mψ ] − C i  p ≤ 2 · dist [Bi, Mψ ] , Cp(L p (Σ, w)) . (2.7.44)
L (Σ,w)→L p (Σ,w)

By (2.7.31), for each  ∈ 2N0 + 1 and 1 ≤ i ≤ H we have


dist [Bi, Mψ ] , Cp(L p (Σ, w))
   
≤C sup |∂ α ki | dist ψ , VMO(Σ, σ) . (2.7.45)
n−1
|α | ≤ N S

Also, (2.5.494), (2.5.498), and (2.5.482) ensure that

     
sup |ai | sup |∂ α ki | ≤ Cn · Cb . (2.7.46)
 ∈N0 1≤i ≤H Rn |α | ≤ N S
n−1

If we now define
  
Cμ := ai C i ∈ Cp L p (Σ, w) , (2.7.47)
 ∈2N0 +1 1≤i ≤H
 ≤μ

then (2.7.42)-(2.7.47) imply

2.7 Commutator Estimates 539
  
dist [B, Mψ ] , Cp(L p (Σ, w)) ≤ [B, Mψ ] − Cμ  L p (Σ,w)→L p (Σ,w) (2.7.48)

 C(Ω, b, p, w, n)
≤ C · Cb · dist ψ , VMO(Σ, σ) + .
μ
Passing to limit μ → ∞ now yields (2.7.39).
Lastly, we note that an estimate similar to (2.7.39) is also valid for commutators
involving the variable coefficient kernel singular integral operator acting on functions

f ∈ L 1 Σ, 1+σ(x)
|x | n−1
according to
∫
B# f (x) := lim+ b(y, y − x) f (y) dσ(y) for σ-a.e. x ∈ Σ. (2.7.49)
ε→0
y ∈Σ
|x−y |>ε

We now turn to the proof of Theorem 2.7.3.

Proof of Theorem 2.7.3 Fix some α ∈ (0, 1) and consider an arbitrary function
φ ∈ VMO(Σ, σ). Based on [113, (3.1.50)] we may then find a sequence of functions
{b j } j ∈N ⊆ 𝒞α (Σ) with the property that

lim φ − b j BMO(Σ,σ) = 0. (2.7.50)

j→∞

From (2.7.3) we conclude that there exists constant C = C(Σ, n, p, [w] A p ) ∈ (0, ∞)
with the property that for each j ∈ N we have
 
[Mb, T]−[Mb , T] p (2.7.51)
j L (Σ,w)→L p (Σ,w)
 
= [Mb−b j , T] L p (Σ,w)→L p (Σ,w)
  
≤C sup |∂ α k | b − b j BMO(Σ,σ)
n−1
|α | ≤ N S
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ) .
n−1
|α | ≤ N S

To proceed, we make the claim


[Mb j , T] ∈ Cp L p (Σ, w) for each j ∈ N. (2.7.52)

Postponing the proof of (2.7.52), we conclude from (2.7.51) that

  
dist [T, Mb ] , Cp L p (Σ, w) (2.7.53)
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ)
n−1
|α | ≤ N S
540 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

for each j ∈ N. Passing to the limit j → ∞ in (2.7.53) then yields

     
dist [T, Mb ], Cp L p (Σ, w) ≤ C sup |∂ α k | b − φBMO(Σ,σ) (2.7.54)
n−1
|α | ≤ N S

on account of (2.7.50). Finally, taking the infimum over all φ ∈ VMO(Σ, σ) estab-
lishes (2.7.31).
There remains to justify (2.7.52). Fix j ∈ N and observe that for each f ∈ L p (Σ, w)
we have
∫

[Mb j , T] f (x) = lim+ b j (x) − b j (y) k(x − y) f (y) dσ(y)
ε→0
y ∈Σ
|x−y |>ε
∫

= b j (x) − b j (y) k(x − y) f (y) dσ(y) for σ-a.e. x ∈ Σ.
Σ
(2.7.55)

Above, the last equality is a consequence of Lebesgue’s Dominated Convergence

Theorem, whose applicability in the presence setting is ensured by the membership
of b j to 𝒞α (Σ), the properties of the kernel k, [112, (7.8.5)], and the fact that (cf.
[112, (7.7.103)])
L p (Σ, w) → L 1 (Σ, σ). (2.7.56)
To proceed, fix some ψ ∈ 𝒞∞ (R) with ψ ≡ 0 on (−1, 1) and ψ ≡ 1 on R \ (−2, 2).
For every ε > 0, consider the truncated singular integral operator acting on each
function f ∈ L p (Σ, w) according to
∫
Q ε f (x) := qε (x, y) f (y) dσ(y) for σ-a.e. x ∈ Σ, (2.7.57)
Σ

where
 |x − y|  
qε (x, y) := ψ b j (x) − b j (y) k(x − y) for x, y ∈ Σ. (2.7.58)
ε
Then [112, Lemma 7.7.16] ensures that there exists some C ∈ (0, ∞) with the
property that for each ε > 0 and each f ∈ L p (Σ, w) we have
 
[Mb , T] f − Q ε f  p ≤ Cε α  f  L p (Σ,w) . (2.7.59)
j L (Σ,w)

In particular,
 
[Mb , T] − Q ε  p ≤ Cε α for each ε > 0, (2.7.60)
j L (Σ,w)→L p (Σ,w)

hence, on the one hand,


[Mb j , T] = lim+ Q ε in Bd L p (Σ, w) → L p (Σ, w) . (2.7.61)
ε→0
2.7 Commutator Estimates 541

On the other hand, for each fixed ε > 0 the kernel qε (x, y) of the integral operator
Q ε is bounded and there exists some C ∈ (0, ∞) such that
 
qε (x0, y) − qε (x1, y) ≤ C|x0 − x1 | α for all x0, x1, y ∈ Σ. (2.7.62)

Indeed, (2.7.62) is a simple consequence of the fact that F(z) := ψ |zε | k(z) for
z ∈ Rn is bounded and Lipschitz (hence, Hölder continuous of order α), and that
b j ∈ 𝒞α (Σ). In turn, from (2.7.62) and the Arzelà-Ascoli Theorem we conclude that
Q ε maps L 1 (Σ, σ) compactly into the space of continuous functions on Σ, equipped
with the supremum norm. As a consequence,

Q ε ∈ Cp L 1 (Σ, σ) → L ∞ (Σ, σ) for each ε > 0. (2.7.63)

From this, (2.7.56), and the fact that L ∞ (Σ, σ) embeds continuously into L p (Σ, w),
we ultimately see that

Q ε ∈ Cp L p (Σ, w) → L p (Σ, w) for each ε > 0. (2.7.64)

At this stage, the claim made in (2.7.52) becomes a consequence of (2.7.61), (2.7.64),
and [113, (1.2.53)]. 
We next turn to the task of establishing commutator estimates in Morrey and
block spaces. Here is a version of the result from Theorem 2.7.1.
Theorem 2.7.4 Suppose Σ ⊆ Rn is a closed Ahlfors regular set and abbreviate
σ := H n−1 Σ. Fix an exponent p0 ∈ (1, ∞) along with some non-decreasing
function Φ : (0, ∞) → (0, ∞) and let T be a linear operator  which is bounded on
L p0 (Σ, w) for every w ∈ Ap0 (Σ, σ), with operator norm ≤ Φ [w] A p0 .
Then for each integrability exponent p ∈ (1, ∞) and each parameter λ ∈ (0, n − 1)
the operator T induces well-defined, linear, and bounded mappings in the contexts

T : M p,λ (Σ, σ) −→ M p,λ (Σ, σ), (2.7.65)

T : M̊ p,λ (Σ, σ) −→ M̊ p,λ (Σ, σ). (2.7.66)

Also, given any exponent p ∈ (1, ∞) along with some λ ∈ (0, n − 1), there exist
two constants, C1 = C1 (Σ, n, p0, p, λ) ∈ (0, ∞) and C2 = C2 (Σ, n, p0, p) ∈ (0, ∞), with
the property that for every complex-valued function b ∈ L ∞ (Σ, σ) one has
   
[Mb, T] p, λ p, λ ≤ [Mb, T] p, λ
M̊ (Σ,σ)→ M̊ (Σ,σ) M (Σ,σ)→M p, λ (Σ,σ)

.
≤ C1 Φ(C2 )bBMO(Σ,σ), (2.7.67)

where [Mb, T] := bT(·) − T(b ·) is the commutator of the operator Mb of pointwise

multiplication (either on M p,λ (Σ, σ), or on M̊ p,λ (Σ, σ)) by the function b and the
operator T (considered either as in (2.7.65), or as in (2.7.66)).
Moreover, as a consequence of the duality result from [113, Proposition 6.2.16],
the (real) transpose of (2.7.66), i.e.,
542 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

T  : B q,λ (Σ, σ) −→ B q,λ (Σ, σ), (2.7.68)

is also well defined, linear, and bounded if q := (1 − 1/p)−1 ∈ (1, ∞) is the Hölder
conjugate exponent of p. In addition, there exist two positive finite constants, say
C1 = C1 (Σ, n, p0, q, λ) and C2 = C2 (Σ, n, p0, q), with the property that for every
complex-valued function b ∈ L ∞ (Σ, σ) one has
 
[Mb, T  ] q, λ
B (Σ,σ)→B q, λ (Σ,σ)
.
≤ C1 Φ(C2 )bBMO(Σ,σ) . (2.7.69)

Finally, the (real) transpose of (2.7.68), i.e.,


 := T   : M p,λ (Σ, σ) −→ M p,λ (Σ, σ),
T (2.7.70)

is an extension of (2.7.66), and there exists a constant C ∈ (0, ∞) with the property
that

if f ∈ M p,λ (Σ, σ) and g ∈ B q,λ (Σ, .σ) then (T f )g − f (T  g) belongs

 space H (Σ, σ) (cf. [113, (4.2.12)]) and
to the homogeneous Hardy 1
 (2.7.71)
 f )g − f (T  g) 1
one has (T H (Σ,σ)
≤ C f  M p, λ (Σ,σ) · g B q, λ (Σ,σ) .

In particular,

if f ∈ M̊ p,λ (Σ, σ) and g ∈ B q,λ (Σ, .σ) then (T f )g − f (T  g) belongs

 space H (Σ, σ) (cf. [113, (4.2.12)]) and
to the homogeneous Hardy 1
 (2.7.72)
  
one has (T f )g − f (T g) H 1 (Σ,σ) ≤ C f  M p, λ (Σ,σ) · g B q, λ (Σ,σ) .

When Σ is the boundary of a UR domain Ω ⊆ Rn and T is a Riesz transform

on Σ (which makes T skew-symmetric), the results in (2.7.71)-(2.7.72) are already
contained in the version of [113, (10.1.137)] corresponding to the scenario described
in item (2) of [113, Corollary 10.1.12].
Proof of Theorem 2.7.4 The present hypotheses ensure that it makes sense to define
the following family of pairs of σ-measurable functions on Σ:
 

F := T f , f : f ∈ L p0 (Σ, σ) . (2.7.73)
w ∈ A p0 (Σ,σ)

In view of the monotonicity property of the classes of Muckenhoupt weights on Σ

(cf. item (5) in [112, Lemma 7.7.1]), the current assumptions also guarantee that

T f  L p (Σ,w) ≤ Φ [w] A1  f  L p (Σ,w)
(2.7.74)
for each pair (T f , f ) ∈ F and weight w ∈ A1 (Σ, σ).

Granted this, once some p ∈ (1, ∞) and λ ∈ (0, n − 1) have been fixed, the extrap-
olation result established in [113, Proposition 6.2.12] applies and gives a constant
C ∈ (0, ∞) with the property that
2.7 Commutator Estimates 543

T f  M p, λ (Σ,σ) ≤ C f  M p, λ (Σ,σ) for each pair (T f , f ) ∈ F . (2.7.75)

In view of [113, (6.2.246) in Proposition 6.2.22], we conclude from (2.7.75) that we

actually have T f  M p, λ (Σ,σ) ≤ C f  M p, λ (Σ,σ) for each function f ∈ M p,λ (Σ, σ).
Thus, T induces a well-defined, linear, and bounded operator in the context of
(2.7.65). Since from (2.7.1) we also know that
p(n−1)
T : L s (Σ, σ) −→ L s (Σ, σ) with s := n−1−λ (2.7.76)

is a well-defined, linear, and bounded operator, and since M̊ p,λ (Σ, σ) is the closure
of L s (Σ, σ) in M p,λ (Σ, σ) (cf. (A.0.88)), we additionally conclude that T induces a
well-defined, linear, and bounded operator in the context of (2.7.66).
In a similar fashion, the second estimate in (2.7.67) is implied by Theorem 2.7.1,
and [113, Propositions 6.2.12, 6.2.22]. The first estimate in (2.7.67) is a simple
consequence of [113, (6.2.15)].
At this stage, the claims regarding (2.7.68) and (2.7.69) follow from what we have
proved already and the duality result from [113, Proposition 6.2.8]. Alternatively,
we could have used the extrapolation result from [113, Corollary 5.2.3], whose
applicability in the present setting with X := B q,λ (∂Ω, σ) is ensured by [113,
Proposition  6.2.17] together with [113, Corollaries 6.2.11, 6.2.13]. Next, the fact
that T := T   from (2.7.70) is an extension of (2.7.66) is seen from definitions,
(A.0.88), [113, (6.2.80)] and [113, (6.2.155)].
There remains to justify the claims made in (2.7.71)-(2.7.72). To this end, fix
arbitrary functions, f ∈ M p,λ (Σ, σ) and g ∈ B q,λ (Σ, σ). Then from [113, Proposi-
tion 6.2.8, (6.2.78)] we conclude that the function h := (T  f )g − f (T  g) belongs to
L (Σ, σ) and satisfies
1

∫ ∫  ∫
   
h dσ = T f g dσ − f (T  g) dσ
Σ Σ Σ
∫ ∫
= f (T  g) dσ − f (T  g) dσ = 0. (2.7.77)
Σ Σ

To proceed, pick an arbitrary φ ∈ Lipc (Σ) and let Mφ denote the operator of
pointwise multiplication by the function φ. Then [113, (6.2.74)] ensures that φg
belongs to B q,λ (Σ, σ). Keeping this in mind and using [113, Proposition 6.2.8,
(6.2.78)] we may then compute
∫  ∫  
    f )g − f (T  g) dσ 
 φ h dσ  =  φ (T
Σ Σ
∫   
 
= f T , Mφ g dσ  ≤  f  M p, λ (Σ,σ) ·  T , Mφ g  B q, λ (Σ,σ)
Σ

.
≤ C f  M p, λ (Σ,σ) · g B q, λ (Σ,σ) · φBMO(Σ,σ), (2.7.78)
544 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

with the last inequality provided by (2.7.69). Granted this, we conclude from [113,
(4.4.10), (4.6.26)] and (2.7.77) that all conclusions in (2.7.71) are valid. Finally,
 is an extension of T in (2.7.66).
(2.7.72) follows from this and the fact that T 

In turn, Theorem 2.7.4 is one of the key ingredients in the proof of the following
commutator estimates on Morrey and block spaces for convolution type singular
integral operators.

Theorem 2.7.5 Assume Σ ⊆ Rn is a UR set, and abbreviate σ := H n−1 Σ. Having

fixed a sufficiently large integer N = N(n) ∈ N, consider a complex-valued function
k ∈ 𝒞 N (Rn \ {0}) which is odd and positive homogeneous of degree 1 − n. Also, pick
two integrability exponents p, q ∈ (1, ∞) along with some parameter λ ∈ (0, n − 1).
Recall from Theorem 2.6.1 that, in this setting, the singular integral operator defined

originally on functions f ∈ L 1 Σ, 1+σ(x)
|x | n−1
as
∫
T f (x) := lim+ k(x − y) f (y) dσ(y) for σ-a.e. x ∈ Σ, (2.7.79)
ε→0
y ∈Σ
|x−y |>ε

induces well-defined, linear, and bounded mappings

T : M p,λ (Σ, σ) −→ M p,λ (Σ, σ), (2.7.80)

T : M̊ p,λ (Σ, σ) −→ M̊ p,λ (Σ, σ), (2.7.81)

T : B q,λ (Σ, σ) −→ B q,λ (Σ, σ). (2.7.82)

Then there exists a constant C ∈ (0, ∞), which depends only on n, p, q, λ, and the
UR character of Σ, such that for each function b ∈ L ∞ (Σ, σ) one has
 
[T, Mb ] p, λ
M (Σ,σ)→M p, λ (Σ,σ)
.
≤ CbBMO(Σ,σ), (2.7.83)
 
[T, Mb ]
M̊ p, λ (Σ,σ)→ M̊ p, λ (Σ,σ)
.
≤ CbBMO(Σ,σ), (2.7.84)
 
[T, Mb ]
B q, λ (Σ,σ)→B q, λ (Σ,σ)
.
≤ CbBMO(Σ,σ) . (2.7.85)

More generally, similar norm estimates are valid for commutators involving the
variable coefficient kernel singular integral operators B from (2.7.25) and B# from
(2.7.26).

Proof All estimates in (2.7.83)-(2.7.85) are consequence of Theorem 2.7.4 and

Theorem 2.6.1. The very last claim is justified in a similar fashion, using the last part
in Theorem 2.7.2. 

In the case when the underlying set is compact, we may use Theorem 2.7.5 to
estimate the essential norm of commutators on Morrey and block spaces.
2.7 Commutator Estimates 545

Theorem 2.7.6 Let Σ ⊆ Rn be a compact UR set, and abbreviate σ := H n−1 Σ. Fix
a sufficiently large integer N = N(n) ∈ N and consider a complex-valued function
k ∈ 𝒞 N (Rn \ {0}) which is odd and positive homogeneous of degree 1 − n. Pick two
integrability exponents p, q ∈ (1, ∞) along with some parameter λ ∈ (0, n − 1), and
recall from Theorem 2.6.1 that the singular integral operator defined originally on
functions f ∈ L 1 (Σ, σ) as
∫
T f (x) := lim+ k(x − y) f (y) dσ(y) for σ-a.e. x ∈ Σ, (2.7.86)
ε→0
y ∈Σ
|x−y |>ε

induces well-defined, linear, and bounded mappings

T : M p,λ (Σ, σ) −→ M p,λ (Σ, σ), (2.7.87)

T : M̊ p,λ (Σ, σ) −→ M̊ p,λ (Σ, σ), (2.7.88)

T : B q,λ (Σ, σ) −→ B q,λ (Σ, σ). (2.7.89)

Then there exists a constant C ∈ (0, ∞), which depends only on n, p, λ, and the
UR character of Σ, such that for each function b ∈ L ∞ (Σ, σ) one has
      
dist [T, Mb ], Cp M p,λ (Σ, σ) ≤ C sup |∂ α k | · dist b, VMO(Σ, σ)
n−1
|α | ≤ N S
(2.7.90)
and
      
dist [T, Mb ], Cp M̊ p,λ (Σ, σ) ≤ C sup |∂ α k | · dist b, VMO(Σ, σ) ,
n−1
|α | ≤ N S
 p,λ (2.7.91)
where
 the distances in the left-hand side are measured in Bd M (Σ, σ) and
Bd M̊ p,λ (Σ, σ) , respectively, and the distances in the right-hand side are measured
in BMO(Σ, σ).
Moreover, there exists a constant C ∈ (0, ∞), which depends only on n, q, λ, and
the UR character of Σ, such that for each function b ∈ L ∞ (Σ, σ) one has
      
dist [T, Mb ], Cp B q,λ (Σ, σ) ≤ C sup |∂ α k | · dist b, VMO(Σ, σ) ,
n−1
|α | ≤ N S
 (2.7.92)
where the distance in the left-hand side is measured in Bd B q,λ (Σ, σ) and the
distance in the right-hand side is measured in BMO(Σ, σ).
In particular,

[T, Mb ] is compact on M p,λ (Σ, σ), M̊ p,λ (Σ, σ), B q,λ (Σ, σ)
(2.7.93)
whenever b ∈ L ∞ (Σ, σ) ∩ VMO(Σ, σ).
546 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Finally, similar results are valid for commutators involving the variable coefficient
kernel singular integral operators B from (2.7.38) and B# from (2.7.49), in relation
to a kernel function as in (2.7.37).

In view of the piece of notation introduced in (A.0.102), we may recast (2.7.90)-

(2.7.92) as
 ess    
[T, Mb ] p, λ p, λ ≤ C sup |∂ α
k | · dist b, VMO(Σ, σ) ,
M (Σ,σ)→M (Σ,σ)
n−1
|α | ≤ N S
(2.7.94)
 ess    
[T, Mb ] p, λ ≤C sup |∂ α k | · dist b, VMO(Σ, σ) ,
M̊ (Σ,σ)→ M̊ p, λ (Σ,σ)
n−1
|α | ≤ N S
(2.7.95)
and
 ess    
[T, Mb ] q, λ ≤C sup |∂ α k | · dist b, VMO(Σ, σ) ,
B (Σ,σ)→B q, λ (Σ,σ)
n−1
|α | ≤ N S
(2.7.96)
respectively.
Proof of Theorem 2.7.6 Fix an arbitrary function φ ∈ VMO(Σ, σ). As in the
proof of Theorem 2.7.3, fix some exponent α ∈ (0, 1) and consider a sequence
{b j } j ∈N ⊆ 𝒞α (Σ) such that lim φ − b j BMO(Σ,σ) = 0. From (2.7.51) and [113,
j→∞
Proposition 6.2.12] (cf. Comment 2 following its statement) we conclude that there
exists some C = C(Σ, p, λ) ∈ (0, ∞) with the property that for each j ∈ N we have
 
[Mb, T]−[Mb , T] p, λ (2.7.97)
j M (Σ,σ)→M p, λ (Σ,σ)
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ) .
n−1
|α | ≤ N S

Likewise, from (2.7.60) and [113, Proposition 6.2.12] we see that, if Q ε is as in

(2.7.57)-(2.7.58), there exists some C ∈ (0, ∞) with the property that for each j ∈ N
we have
 
[Mb , T] − Q ε  p, λ ≤ Cε α for each ε > 0. (2.7.98)
j M (Σ,σ)→M p, λ (Σ,σ)

Consequently, for each j ∈ N we obtain


[Mb j , T] = lim+ Q ε in Bd M p,λ (Σ, σ) → M p,λ (Σ, σ) . (2.7.99)
ε→0

Since (2.7.63) and [113, (6.2.7)] imply that


Q ε ∈ Cp M p,λ (Σ, σ) → M p,λ (Σ, σ) for each ε > 0, (2.7.100)

we conclude from (2.7.99) and [113, (1.2.53)] that

2.7 Commutator Estimates 547

[Mb j , T] ∈ Cp M p,λ (Σ, σ) for each j ∈ N. (2.7.101)

Together with (2.7.97), this proves that for each j ∈ N we have

  
dist [T, Mb ] , Cp M p,λ (Σ, σ) (2.7.102)
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ) .
n−1
|α | ≤ N S

Passing to the limit j → ∞ in (2.7.102) then yields (keeping in mind (2.7.50))

     
dist [T, Mb ], Cp M p,λ (Σ, σ) ≤ C sup |∂ α k | b − φBMO(Σ,σ) .
n−1
|α | ≤ N S
(2.7.103)
At this stage, taking the infimum over all φ ∈ VMO(Σ, σ) establishes (2.7.90).
In view of (A.0.88)-[113, (6.2.15)] and [113, (1.2.20)], the same proof as above
works, and gives (2.7.91), as soon as we show that

Q ε ∈ Cp M̊ p,λ (Σ, σ) → M̊ p,λ (Σ, σ) for each ε > 0. (2.7.104)
p(n−1)
To justify this, observe first that since with s := n−1−λ ∈ (p, ∞) we have
 
Q ε ∈ Bd M p,λ (Σ, σ) → M p,λ (Σ, σ) ∩ Bd L s (Σ, σ) → L s (Σ, σ) (2.7.105)

(cf. (2.7.100), (2.7.63), and [113, (1.2.52)]), it follows from (A.0.88) and [113,
(6.2.15)] that 
Q ε ∈ Bd M̊ p,λ (Σ, σ) → M̊ p,λ (Σ, σ) . (2.7.106)
From (2.7.100), (2.7.106), and [113, (1.1.17), (6.2.15)] we then conclude that the
claim made in (2.7.104) is true. This completes the proof of (2.7.91).
Let us now show (2.7.92). To this end, assume that 1/p + 1/q = 1. Via duality,
from (2.7.104), Schauder’s theorem (cf. [113, (1.2.57)]), and [113, (6.2.155)] we
conclude that the transpose of the operator Q ε satisfies
 q,λ
Qε ∈ Cp B (Σ, σ) → B q,λ (Σ, σ) for each ε > 0. (2.7.107)

Next, based on duality, (2.7.99), (2.7.107), [113, (6.2.155)], [113, (1.2.53)], and
(2.6.15) we see that

[Mb j , T] ∈ Cp B q,λ (Σ, σ) for each j ∈ N. (2.7.108)

Note that (2.7.97), [113, (6.2.15)], and [113, (1.2.20)] imply that
548 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 
[Mb, T]−[Mb , T] p, λ (2.7.109)
j M̊ (Σ,σ)→ M̊ p, λ (Σ,σ)
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ) .
n−1
|α | ≤ N S

Using duality, (2.7.109), [113, (1.2.68), (6.2.155)], and (2.6.15) we next deduce that
 
[Mb, T]−[Mb , T] q, λ (2.7.110)
j B (Σ,σ)→B q, λ (Σ,σ)
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ) .
n−1
|α | ≤ N S

In particular, (2.7.110) and (2.7.108) prove that for each j ∈ N we have

  
dist [T, Mb ] , Cp B q,λ (Σ, σ) (2.7.111)
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ) .
n−1
|α | ≤ N S

Upon sending j → ∞ in (2.7.111) and taking into account (2.7.50) we obtain

  
dist [T, Mb ],Cp B q,λ (Σ, σ) (2.7.112)
  
≤C sup |∂ α k | b − φBMO(Σ,σ) .
n−1
|α | ≤ N S

At this stage, taking the infimum over all φ ∈ VMO(Σ, σ) yields (2.7.92).
Finally, the very last claim in the statement of the theorem is justified in a similar
fashion, relying on the discussion just prior to the proof of Theorem 2.7.3. 
The following theorem refines some of the work in [21].
Theorem 2.7.7 Suppose Σ ⊆ Rn is a closed Ahlfors regular set, and abbreviate
σ := H n−1 Σ. Fix an exponent p0 ∈ (1, ∞) along with some non-decreasing
function Φ : (0, ∞) → (0, ∞) and let T be a linear operator  which is bounded on
L p0 (Σ, ω) for every ω ∈ Ap0 (Σ, σ), with operator norm ≤ Φ [ω] A p0 .
Then for each exponent p ∈ (1, ∞) and Muckenhoupt weight w ∈ Ap (Σ, σ) the
operator

T : L p (Σ, w) −→ L p (Σ, w) (2.7.113)

is well defined, linear, and bounded. In particular, its transpose

 
T  : L p (Σ, w ) −→ L p (Σ, w ) (2.7.114)

is also well defined, linear, and bounded, where p := (1 − 1/p)−1 ∈ (1, ∞) is the

conjugate exponent of p and w  := w 1−p ∈ Ap (Σ, σ) is the conjugate weight of w
2.7 Commutator Estimates 549

(cf. item (2) in [112, Lemma 7.7.1]). Finally, there exists a constant C ∈ (0, ∞) with
the property that

if f ∈ L p (Σ, w) and g ∈ L p (Σ, w .) then (T f )g − f (T  g) belongs
 space H (Σ, σ) (cf. [113, (4.2.12)]) and
to the homogeneous Hardy 1
 (2.7.115)
  
one has (T f )g − f (T g) H 1 (Σ,σ) ≤ C f  L p (Σ,w) · g L p (Σ,w ) .

In the case when Σ is the boundary of a UR domain Ω ⊆ Rn , (2.7.115) contains

as a particular case [113, (10.1.137)] in the scenario described in item (1) of [113,
Corollary 10.1.12] (since each Riesz transform R j is skew-symmetric). See also item
(10) in Theorem 2.3.2.
Proof of Theorem 2.7.7 That T is well defined, linear, and bounded in the context
of (2.7.113) has been seen in Theorem 2.7.1. Via duality (cf. (2.3.60)), this further
implies that T  is also well defined, linear, and bounded in the context of (2.7.114).
To justify the claims made in (2.7.115), fix two arbitrary functions f ∈ L p (Σ, w)

and g ∈ L p (Σ, w ). In particular, the difference h := (T f )g − f (T  g) belongs to
L 1 (Σ, σ) and satisfies ∫
h dσ = 0. (2.7.116)
Σ
Next, pick an arbitrary φ ∈ Lipc (Σ) and denote by Mφ the operator of pointwise
multiplication by the function φ. Then φ f ∈ L p (Σ, w) and we may then write
∫  ∫  
   
 φ h dσ  =  φ (T f )g − f (T  g) dσ 
Σ Σ
∫   
 
=  g[Mφ, T] f dσ  ≤ g L p (Σ,w ) · [Mφ, T] f  L p (Σ,w)
Σ

.
≤ Cg L p (Σ,w ) ·  f  L p (Σ,w) · φBMO(Σ,σ), (2.7.117)

using Hölder’s inequality and (2.7.3). With this in hand, we conclude from [113,
(4.4.10)], (2.7.116), and [113, (4.6.26)] that all conclusions in (2.7.115) hold. 
In the last part of this section we shall extend our earlier commutator estimates
to abstract Generalized Banach Function Spaces. Here is the first result of this
flavor, generalizing Theorem 2.7.4 (here [113, Proposition 6.2.17] is relevant) and
Theorem 2.7.7.
Theorem 2.7.8 Suppose Σ ⊆ Rn is a closed Ahlfors regular set, and abbreviate
σ := H n−1 Σ. With M Σ denoting the Hardy-Littlewood maximal operator on Σ,
assume X is a Generalized Banach Function Space on (Σ, σ) such that

M Σ : X → X and M Σ : X → X
(2.7.118)
are well-defined bounded mappings,

where X is the associated space of X (cf. [113, Definitions 5.1.4, 5.1.11]). Finally,
fix an integrability exponent p0 ∈ (1, ∞) along with some non-decreasing function
550 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Φ : (0, ∞) → (0, ∞) and let T be a linear operator

 which is bounded on L (Σ, w)
p0

for every w ∈ Ap0 (Σ, σ), with operator norm ≤ Φ [w] A p0 .

Then the operator T induces well-defined, linear, and bounded mappings in the
contexts

T : X −→ X, (2.7.119)

T : X −→ X . (2.7.120)

Also, there exist two constants C1, C2 ∈ (0, ∞), depending only on Σ, p0 , and the
operator norms of M Σ on X and X, with the property that for every complex-valued
function b ∈ L ∞ (Σ, σ) one has
 
[Mb, T]
X→X
.
≤ C1 Φ(C2 )bBMO(Σ,σ), (2.7.121)
 
[Mb, T]
X →X
.
≤ C1 Φ(C2 )bBMO(Σ,σ), (2.7.122)

where [Mb, T] := bT(·) − T(b ·) is the commutator of the operator Mb of pointwise

multiplication by the function b and the operator T (considered in either of the
scenarios described in (2.7.119)-(2.7.120)). Moreover,
similar results to (2.7.119), (2.7.120) and (2.7.121), (2.7.122)
are also valid for T  , the (real) transpose of the original (2.7.123)
operator T.

Finally, there exists a constant C ∈ (0, ∞) with the property that

whenever f ∈ X̊ and g ∈ X it follows 

. 1 that (T f )g − f (T g) belongs
to the homogeneous Hardy space H (Σ,  σ) (cf. [113, (4.2.12)]) and (2.7.124)
one has the estimate (T f )g − f (T  g) H 1 (Σ,σ) ≤ C f X · gX ,

and
whenever f ∈ X and g ∈ (X)◦ it follows 
. 1 that (T f )g − f (T g) belongs
to the homogeneous Hardy space H (Σ,  σ) (cf. [113, (4.2.12)]) and (2.7.125)
one has the estimate (T f )g − f (T  g) H 1 (Σ,σ) ≤ C f X · gX .

In relation to Theorem 2.7.8, we make two comments. First, thanks to item (10)
in Theorem 2.3.2, the conclusions in Theorem 2.7.8 apply to all principal-value
singular integral operators of the sort considered in (2.3.15). Given the significance
of this special case, we shall state it as a stand-alone result a little later, subsumed in
Corollary 2.7.9.
Our second comment regarding Theorem 2.7.8 is that, as noted in [113, Proposi-
tion 5.3.14],
2.7 Commutator Estimates 551

the conditions demanded in (2.7.118) automatically hold in the case

when X is a rearrangement invariant Banach function space on the
(2.7.126)
(non-atomic, sigma-finite measure space (Σ, σ)) whose Boyd indices
satisfy 1 < pX ≤ qX < ∞.
Consequently, all conclusions in Theorem 2.7.8 are valid in the class of re-
arrangement invariant Banach function spaces X whose Boyd indices satisfy
1 < pX ≤ qX < ∞. In particular, this is the case when the Generalized Banach
Function Space X is actually an Orlicz space, X := L Φ (Σ, σ), associated with the
(non-atomic, sigma-finite) measure space (Σ, σ) and a Young function Φ satisfying
1 < i(Φ) ≤ I(Φ) < ∞; see [113, Proposition 5.3.15] and [113, (5.3.62)]. As a
concrete example, one may take Φ(t) ≈ t p [ln(e + t)]α for some fixed p ∈ (1, ∞) and
α ∈ R, a choice for which the Orlicz space L Φ actually becomes Zygmund’s space
L p (log L)α (see [113, Example 5.3.10]).
Here is the proof of Theorem 2.7.8.
Proof of Theorem 2.7.8 All claims in (2.7.119)-(2.7.120) and (2.7.121)-(2.7.122)
are consequences of Theorem 2.7.1 and [113, Corollary 5.2.3]. Also, (2.7.123) is
justified in a similar manner, keeping in mind that T  satisfies the same properties as
the original operator T, with p0 replaced by q0 := (1 − 1/p0 )−1 ∈ (1, ∞), its Hölder
conjugate exponent.
As a preamble to the proof of (2.7.124), we first establish that
∫ ∫
(T f )g dσ = f (T  g) dσ for each f ∈ X̊ and g ∈ X . (2.7.127)
Σ Σ

With this aim in mind, fix f ∈ X̊ and g ∈ X, arbitrary. Since Lcomp ∞ (Σ, σ) is dense

in X̊ (cf. [113, (5.2.82)]), there exists a sequence { f j } j ∈N ⊆ Lcomp ∞ (Σ, σ) which

converges to f in X. Also, [113, Corollary 5.2.3] implies that there exists a weight
ω ∈ Aq0 (Σ, σ) such that g ∈ L q0 (Σ, ωσ). Then item (2) in [112, Lemma 7.7.1]
ensures that w := ω1−p0 ∈ Ap0 (Σ, σ) so, bearing in mind (2.7.119) as well as [113,
Proposition 5.1.12], we may write
∫ ∫ ∫
(T f )g dσ = lim (T f j )g dσ = lim f j (T  g) dσ
Σ j→∞ Σ j→∞ Σ
∫
= f (T  g) dσ, (2.7.128)
Σ

where the second equality in (2.7.128) uses the fact that the transposed of the
operator T is considered at the level of Muckenhoupt weighted Lebesgue spaces
∞ (Σ, σ) ⊆ L p0 (Σ, wσ) = L q0 (Σ, ωσ) ∗ . This finishes the proof of
and that Lcomp
(2.7.127).
To proceed, select two arbitrary functions, f ∈ X̊ and g ∈ X. Since X̊ ⊆ X, from
(2.7.119)-(2.7.120) and [113, Proposition 5.1.12] we conclude that the function
h := (T f )g − f (T  g) is absolutely integrable on Σ (with respect to the measure σ).
In addition, (2.7.127) permits us to write
552 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫ ∫ ∫
h dσ = (T f )g dσ − f (T  g) dσ
Σ Σ Σ
∫ ∫
= (T f )g dσ − (T f )g dσ = 0. (2.7.129)
Σ Σ

Fix now an arbitrary function φ ∈ Lipc (Σ) and denote by Mφ the operator of
pointwise multiplication by φ. From [113, (5.2.85)] we know that φ f ∈ X̊. Granted
this and using (2.7.127) we may then compute
∫  ∫   ∫ 
      
 φ h dσ  =  φ (T f )g − f (T g) dσ  =  φ(T f )g − T(φ f )g dσ 
Σ Σ Σ
∫     
 
= Mφ, T f g dσ  ≤  T, Mφ f X · gX
Σ

.
≤ C f X · gX · φBMO(Σ,σ), (2.7.130)

with the penultimate inequality coming from [113, (5.1.37)], and last inequality
provided by (2.7.121). At this stage, we conclude from [113, (4.4.10)], (2.7.129),
(2.7.130), [113, (4.6.26)] that all conclusions in (2.7.124) are valid. With this in hand,
(2.7.125) follows from (2.7.124) by replacing X with X (a permissible choice in
light of [113, Proposition 5.1.14]), interchanging the roles of f and g, and replacing
T by T  (which enjoys the same properties as the original operator). 

It is of interest to specialize Theorem 2.7.8 to convolution-type singular integral

operators on UR sets, a scenario in which we have the result described in the following
corollary, which also encompasses X̊ is defined in relation to a given Generalized
Banach Function space X as in [113, Definition 5.2.6].

Corollary 2.7.9 Let Σ ⊆ Rn be a UR set, and abbreviate σ := H n−1 Σ. With M Σ

denoting the Hardy-Littlewood maximal operator on Σ, assume X is a Generalized
Banach Function Space on (Σ, σ) such that

M Σ : X → X and M Σ : X → X
(2.7.131)
are well-defined bounded mappings,

where X is the associated space of X (cf. [113, Definitions 5.1.4, 5.1.11]). Finally,
fix a sufficiently large integer N = N(n) ∈ N and consider a complex-valued function
k ∈ 𝒞 N (Rn \ {0}) which is odd and positive homogeneous of degree 1 − n, and

consider the singular integral operator acting on functions f ∈ L 1 Σ, 1+σ(x) |x | n−1
according to
∫
T f (x) := lim+ k(x − y) f (y) dσ(y) for σ-a.e. x ∈ Σ. (2.7.132)
ε→0
y ∈Σ
|x−y |>ε
2.7 Commutator Estimates 553

Then the operator T induces well-defined, linear, and bounded mappings in the
contexts

T : X −→ X, T : X̊ −→ X̊, (2.7.133)

T : X −→ X, T : (X)◦ −→ (X)◦, (2.7.134)

where
 · X  · X
∞ (Σ, σ)
X̊ := Lcomp and (X)◦ := Lcomp
∞ (Σ, σ) . (2.7.135)

In addition, there exists a constant C ∈ (0, ∞), depending only on Σ, k, plus the
operator norms of M Σ on X and X, with the property that for every complex-valued
function b ∈ L ∞ (Σ, σ) one has
   
[Mb, T]
X̊→X̊
.
≤ [Mb, T]X→X ≤ CbBMO(Σ,σ) , (2.7.136)
   
[Mb, T]
(X )◦ →(X )◦
.
≤ [Mb, T]X →X ≤ CbBMO(Σ,σ), (2.7.137)

where [Mb, T] := bT(·) − T(b ·) is the commutator of the operator Mb of pointwise

multiplication by the function b and the operator T (considered in either of the
scenarios described in (2.7.133)-(2.7.134)).
Furthermore, there exists a constant C ∈ (0, ∞) with the property that

whenever either f ∈ X̊ and g ∈ X, or f ∈ X and g ∈ (X)◦ , it follows

that the. function (T f )g + f (T g) belongs to the homogeneous Hardy
(2.7.138)
space
 H 1 (Σ, σ) (defined in [113, (4.2.12)]) and one has the estimate
(T f )g + f (T g) 1 ≤ C f X · gX .
H (Σ,σ)

More generally, similar mapping properties and norm estimates are valid for
commutators involving the variable coefficient kernel singular integral operators B
from (2.7.25) and B# from (2.7.26).

We wish to remark that, in light of [113, Proposition 5.3.14], all conclusions in

Corollary 2.7.9 are valid in the class of rearrangement invariant Banach function
spaces X whose Boyd indices satisfy 1 < pX ≤ qX < ∞. For example, this is the
case when the Generalized Banach Function Space X is actually an Orlicz
space, X := L Φ (Σ, σ), associated with the (non-atomic, sigma-finite) measure space
(Σ, σ) and a Young function Φ satisfying 1 < i(Φ) ≤ I(Φ) < ∞; see [113,
Proposition 5.3.15] and [113, (5.3.62)].
Proof of Corollary 2.7.9 All claims pertaining to constant coefficient kernel singu-
lar integral operators are consequences of item (11) in Theorem 2.3.2, and The-
orem 2.7.8, bearing in mind [113, (1.2.20)] and (2.3.25) The more general case,
corresponding to singular integral operators with variable coefficient kernels, are
dealt with virtually verbatim, while keeping in mind item (e) in Theorem 2.5.38 and
(2.5.470). 
554 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Finally, in the case when the underlying set is compact, we have the following
version of Theorem 2.7.6 with abstract Generalized Banach Function Spaces playing
the role of Morrey and block spaces.

Theorem 2.7.10 Let Σ ⊆ Rn be a compact UR set, and abbreviate σ := H n−1 Σ.

With M Σ denoting the Hardy-Littlewood maximal operator on Σ, assume X is a
Generalized Banach Function Space on (Σ, σ) such that

M Σ : X → X and M Σ : X → X
(2.7.139)
are well-defined bounded mappings,

where X is the associated space of X (cf. [113, Definitions 5.1.4, 5.1.11]). Lastly, fix
a sufficiently large integer N = N(n) ∈ N and consider a complex-valued function
k ∈ 𝒞 N (Rn \ {0}) which is odd and positive homogeneous of degree 1 − n, and
recall from Corollary 2.7.9 that the singular integral operator defined originally on
functions f ∈ L 1 (Σ, σ) as
∫
T f (x) := lim+ k(x − y) f (y) dσ(y) for σ-a.e. x ∈ Σ, (2.7.140)
ε→0
y ∈Σ
|x−y |>ε

has X, X, as well as

 · X  · X
X̊ := L ∞ (Σ, σ) and (X)◦ := L ∞ (Σ, σ) (2.7.141)

as invariant subspaces (cf. [113, (5.2.76)]), hence

T : X −→ X, T : X̊ −→ X̊, (2.7.142)

T : X −→ X, T : (X)◦ −→ (X)◦, (2.7.143)

are well-defined, linear, bounded operators.

Then there exists a constant C ∈ (0, ∞), which depends only on the operator
norms of M Σ on X and X and the UR character of Σ, with the property that for
each function b ∈ L ∞ (Σ, σ) one has
  
dist [T, Mb ] , Cp X̊
   
≤C sup |∂ α k | · dist b, VMO(Σ, σ) , (2.7.144)
n−1
|α | ≤ N S

and
2.7 Commutator Estimates 555
  
dist [T, Mb ] , Cp X
   
≤C sup |∂ α k | · dist b, VMO(Σ, σ) , (2.7.145)
n−1
|α | ≤ N S

 
where the distances in the left are measured in Bd X̊ and Bd X , respectively, while
the distances in the right are measured in BMO(Σ, σ), as well as
  
dist [T, Mb ] , Cp (X)◦
   
≤C sup |∂ α k | · dist b, VMO(Σ, σ) , (2.7.146)
n−1
|α | ≤ N S

and
  
dist [T, Mb ] , Cp X
   
≤C sup |∂ α k | · dist b, VMO(Σ, σ) , (2.7.147)
n−1
|α | ≤ N S
 
where the distances in the left are measured in Bd (X)◦ and Bd X , respectively,
and the distances in the right are measured in BMO(Σ, σ).
In particular,
[T, Mb ] is compact on X, X, X̊, (X)◦
(2.7.148)
whenever b ∈ L ∞ (Σ, σ) ∩ VMO(Σ, σ).
Finally, analogous estimates are true for commutators involving the more general
variable coefficient kernel singular integral operators B from (2.7.38) and B# from
(2.7.49).

Upon recalling the piece of notation introduced in (A.0.102), we may reformulate

(2.7.144)-(2.7.145) and (2.7.146)-(2.7.147) simply as
 ess    
[T, Mb ] ≤C sup |∂ α k | · dist b, VMO(Σ, σ) , (2.7.149)
X̊→X̊
n−1
|α | ≤ N S

 ess    
[T, Mb ] ≤C sup |∂ α k | · dist b, VMO(Σ, σ) , (2.7.150)
X→X
n−1
|α | ≤ N S

and
556 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 ess    
[T, Mb ]  ◦ ≤C sup |∂ α k | · dist b, VMO(Σ, σ) , (2.7.151)
(X ) →(X )◦
n−1
|α | ≤ N S

 ess    
[T, Mb ]   ≤ C sup |∂ α
k | · dist b, VMO(Σ, σ) , (2.7.152)
X →X
n−1
|α | ≤ N S

respectively.
It is also of significance to point out that, thanks to [113, Proposition 5.3.14],
all conclusions in Theorem 2.7.10 remain true for rearrangement invariant Banach
function spaces X whose Boyd indices satisfy 1 < pX ≤ qX < ∞. In particular,
said conclusions hold for X := L Φ (Σ, σ), the Orlicz space associated with the
(non-atomic, sigma-finite) measure space (Σ, σ) and a Young function Φ satisfying
1 < i(Φ) ≤ I(Φ) < ∞ (cf. [113, Proposition 5.3.15] and [113, (5.3.62)]).
Proof of Theorem 2.7.10 Fix an arbitrary function φ ∈ VMO(Σ, σ). Having picked
an exponent α ∈ (0, 1), the characterization of the Sarason space given in
[113, (3.1.50)] guarantees the existence of a sequence {b j } j ∈N ⊆ 𝒞α (Σ) such
that lim φ − b j BMO(Σ,σ) = 0. Also, from (2.7.133), [113, (5.2.85)], [113,
j→∞
(1.2.20)], (2.7.51), and [113, Corollary 5.2.3] we see that there exists a constant
C = C(Σ, X) ∈ (0, ∞) with the property that for each j ∈ N we have
   
[Mb, T]−[Mb , T] ≤ [Mb, T] − [Mb j , T]X→X (2.7.153)
j X̊→X̊
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ) ,
n−1
|α | ≤ N S

and
   
[Mb, T]−[Mb , T]  ◦ ≤ [Mb, T] − [Mb j , T]X →X (2.7.154)
j (X ) →(X )◦
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ) .
n−1
|α | ≤ N S

Next, if Q ε is as in (2.7.57)-(2.7.58), then from [113, (1.2.20)], (2.7.60), and [113,

Corollary 5.2.3] we conclude (also bearing in mind (2.7.63) and [113, (5.2.76)]) that
there exists some constant C ∈ (0, ∞) with the property that for each j ∈ N we have
   
[Mb , T] − Q ε  ≤ [Mb j , T] − Q ε X→X ≤ Cε α for each ε > 0, (2.7.155)
j X̊→X̊

and
   
[Mb , T] − Q ε   ◦ ≤ [Mb j , T] − Q ε X →X ≤ Cε α (2.7.156)
j (X ) →(X )◦

for each ε > 0.

Consequently, for each j ∈ N we obtain
2.7 Commutator Estimates 557
 
[Mb j , T] = lim+ Q ε in Bd X , Bd X ,
ε→0
  (2.7.157)
Bd X̊ , and in Bd (X)◦ .

In view of the fact that (2.7.63) and [113, (5.2.76), (5.2.84)] imply
   
Q ε ∈ Cp X ∩ Cp X ∩ Cp X̊ ∩ Cp (X)◦ for each ε > 0, (2.7.158)

we conclude from (2.7.157) and [113, (1.2.53)] (bearing in mind the final conclusion
in [113, Proposition 5.1.8]) that for each j ∈ N we have
   
[Mb j , T] ∈ Cp X ∩ Cp X ∩ Cp X̊ ∩ Cp (X)◦ . (2.7.159)

Together with (2.7.153), this proves that for each j ∈ N we have

  
dist [T, Mb ] , Cp X̊ (2.7.160)
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ) ,
n−1
|α | ≤ N S

and
  
dist [T, Mb ] , Cp X (2.7.161)
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ) .
n−1
|α | ≤ N S

Also, from (2.7.154) and (2.7.159) we conclude that for each j ∈ N we have
  
dist [T, Mb ] , Cp (X)◦ (2.7.162)
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ) ,
n−1
|α | ≤ N S

and
  
dist [T, Mb ] , Cp X (2.7.163)
   
≤C sup |∂ α k | b − φBMO(Σ,σ) + φ − b j BMO(Σ,σ) .
n−1
|α | ≤ N S

Keeping in mind (2.7.50) and sending j → ∞ in (2.7.160)-(2.7.163) then yields

558 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
  
dist [T, Mb ] , Cp X̊
  
≤C sup |∂ α k | b − φBMO(Σ,σ), (2.7.164)
n−1
|α | ≤ N S

and
  
dist [T, Mb ] , Cp X
  
≤C sup |∂ α k | b − φBMO(Σ,σ), (2.7.165)
n−1
|α | ≤ N S

as well as
  
dist [T, Mb ] , Cp (X)◦
  
≤C sup |∂ α k | b − φBMO(Σ,σ), (2.7.166)
n−1
|α | ≤ N S

and
  
dist [T, Mb ] , Cp X
  
≤C sup |∂ α k | b − φBMO(Σ,σ) . (2.7.167)
n−1
|α | ≤ N S

After taking the infimum over all φ ∈ VMO(Σ, σ), we now arrive at (2.7.144)-
(2.7.147). Let us also note that (2.7.148) is a consequence of [113, (1.2.53)] (again,
bearing in mind the final conclusion in [113, Proposition 5.1.8]) and (2.7.144)-
(2.7.147).
Lastly, the final claim in the statement of the theorem is seen from what we
have proved so far and the spherical harmonics expansion employed in the proof of
Theorem 2.5.38, by reasoning much as in (2.7.37)-(2.7.49) (with L p (Σ, w) replaced
by X, X, X̊, (X)◦ ), while bearing in mind [113, (5.1.13)]. 

2.8 Calderón-Zygmund Theory for Singular Integrals on

Riemannian Manifolds

The Calderón-Zygmund machinery developed so far may be naturally adapted to

integral operators, on uniformly rectifiable sets, with variable coefficient kernels
that are actually the Schwartz kernels of classical pseudodifferential operators of
order −1 with odd principal symbol, acting between vector bundles over a manifold.
Introducing this very general class of integral operators requires some preparations.
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 559

To set the stage, recall that there are several symbol classes of importance. One
m , is defined by
class, denoted S1,0
 β 
q(x, ξ) ∈ S1,0
m
⇐⇒  Dx Dαξ q(x, ξ) ≤ Cα,β (1 + |ξ |)m− |α | . (2.8.1)

With each symbol q(x, ξ) in Hörmander’s class S1,0

m associate a pseudodifferential

operator
q(x, D) := Q(x, D) (2.8.2)
acting29 on each Schwartz function30 u as
∫
Q(x, D)u(x) := (2π)−n u(ξ)ei x, ξ  dξ
q(x, ξ)-
∬
−n
= (2π) q(x, ξ)ei x−y, ξ  u(y) dy dξ. (2.8.3)

The latter is an oscillatory integral, in the sense, e.g., of [65]. It is also of interest to
note that (see [176, (2.1)-(2.2), p. 5])

the Schwartz kernel of q(x, D) is b(x, x − y) where

∫ (2.8.4)
b(x, z) := (2π)−n q(x, ξ)ei z,ξ  dξ,

with the integral once again interpreted in an oscillatory sense. It turns out that b is
of class 𝒞 ∞ in Rn × Rn \ {0} (cf. [176, Proposition 2.1, p. 5]), and satisfies (see
[176, Proposition 2.2, p. 6]) the estimate
 α β 
∂ ∂z b(x, z) ≤ Cα,β |z| −m−n− |α |− |β | if |α| + | β| > −m − n. (2.8.5)
x

After taking a partial Fourier transform in the second variable, (2.8.4) also gives
-
b(x, ξ) = q(x, ξ). (2.8.6)

Here, we are concerned with a smaller class of symbols, Sclm with m ∈ Z, defined
by requiring that the (matrix-valued) function q(x, ξ) has an asymptotic expansion
of the form
q(x, ξ) ∼ qm (x, ξ) + qm−1 (x, ξ) + · · · , (2.8.7)
with q j smooth in x and ξ and positive homogeneous of degree j in ξ for |ξ | ≥ 1 (in
the sense that q j (x, λξ) = λ j q j (x, ξ) if λ ≥ 1 and |ξ | ≥ 1). The meaning of (2.8.7)
is that, for each k ∈ N, the difference between the left side and the sum of the first k
terms on the right belong to S1,0 m−k . In particular, the “error”

29 with “hat” denoting the Fourier transform in R n , with the normalization given in (1.4.14)
30 this subsequently extends to tempered distributions in a natural fashion; see [176, Lemma 1.1,
p. 3]
560 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

rm−1 (x, ξ) := q(x, ξ) − qm (x, ξ) belongs to S1,0

m−1
. (2.8.8)

Define  
OPSclm := q(x, D) : q(x, ξ) ∈ Sclm . (2.8.9)
Also, given any q(x, D) ∈ OPSclm , call qm (x, ξ), i.e., the leading term in (2.8.7), the
principal symbol of q(x, D) = Q(x, D). In the sequel, we shall use the notation

Sym(Q; ξ) := qm (x, ξ). (2.8.10)

Let us illustrate how this information ties up with earlier work. Start with a
classical pseudodifferential operator q(x, D) of order m, and consider its Schwartz
kernel b(x, x − y) as well as its principal symbol qm (x, ξ). The latter is known
to be homogeneous of degree m in the ξ variable only for |ξ | ≥ 1 (in the sense
that qm (x, λξ) = λ m qm (x, ξ) if λ ≥ 1 and |ξ | ≥ 1). To correct this, consider
qm (x, ξ) := |ξ | m qm x, |ξξ | for ξ ∈ Rn \ {0}. Then 
 qm (x, ξ) is smooth and (genuinely)
positive homogeneous of degree m in R \ {0}, and also matches qm (x, ξ) when
n

|ξ | ≥ 1. Moreover, if qm (x, ξ) is odd in ξ then so is  qm (x, ξ). Hence, 

qm (x, ξ)
induces a unique tempered distribution in Rn , which is positive homogeneous of
degree m. If we now define
∫
btop (x, z) := (2π)−n qm (x, ξ)ei z,ξ  dξ,
 (2.8.11)

then (see [175, Proposition 8.1, p. 239 in Chapter 3])

btop (x, z) is positive homogeneous of degree −m − n

as well as smooth in the variable z ∈ Rn \ {0}, (2.8.12)
and it is also odd in z if qm (x, ξ) is odd in ξ.

In addition, if
∫
−n
b (x, z) := (2π)
err
qm (x, ξ) ei z,ξ  dξ
qm (x, ξ) − 
∫
−n
+ (2π) rm−1 (x, ξ)ei z,ξ  dξ, (2.8.13)

then the first term in the right-hand side belongs to 𝒞∞ (Rn ) (as the Fourier transform
of a compactly supported distribution; cf., e.g., [109, Theorem 4.35, p. 135], [175,
Proposition 4.3, p. 208]), while the second term in the right-hand side of (2.8.13)
satisfies estimates of the sort described in (2.8.5) with m replaced by m−1. Altogether
we obtain
 α β 
∂ ∂z berr (x, z) ≤ Cα,β |z| 1−m−n− |α |− |β | for |α| + | β| > 1 − m − n. (2.8.14)
x

This argument shows that we have the decomposition

2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 561

b(x, z) = btop (x, z) + berr (x, z) (2.8.15)

of the Schwartz kernel of q(x, D) into a top part btop as in (2.8.12) and the error term
berr exhibiting a milder singularity at the origin, manifested by the estimate recorded
in (2.8.14). We also wish to make the important observation that

if m = −1 and the principal symbol q−1 (x, ξ) of q(x, D) ∈ OPScl−1 is

odd in ξ then Theorem 2.5.38 applies to btop (x, z) since this is odd and (2.8.16)
positive homogeneous of degree 1 − n in z.
From (2.8.6) and (2.8.10) we also see that

b.
top (x, ξ) = Sym(Q; ξ) for |ξ | ≥ 1, (2.8.17)

where once again ‘hat’ stands for the partial Fourier transform in the second variable.
In fact, we shall find it convenient to work with classes of symbols that only
exhibit a limited amount of regularity in the spatial variable (while still 𝒞∞ in the
Fourier variable). Specifically, for each r ≥ 0 define
 
C r S1,0
m
:= q(x, ξ) : Dαξ q(·, ξ)𝒞r ≤ Cα (1 + |ξ |)m− |α |, ∀α ∈ N0n . (2.8.18)

Denote by OPC r S1,0

m the class of pseudodifferential operators associated with

such symbols. As before, we write OPC r Sclm for the subclass of classical
pseudodifferential operators in OPC r S1,0 m whose symbols can be expanded
m−j
as in (2.8.7), where q j (x, ξ) ∈ C r S1,0 is homogeneous of degree j in ξ for |ξ | ≥ 1,
j = m, m − 1, . . . . Basic material on these symbol classes, as well as other symbol
classes with limited regularity in x, can be found in [174].
Another type of pseudodifferential operator is defined by
∬
−n
Q(D, x)u(x) := (2π) q(ξ, y)ei x−y,ξ  u(y) dy dξ (2.8.19)

(often, we simply write q(D, x) in place of Q(D, x)). The Schwartz kernel of Q(D, x)
is 
b(y, x − y) where
∫

b(y, z) := (2π) −n
q(ξ, y)ei z,ξ  dξ, (2.8.20)

also interpreted as an oscillatory integral. Any operator of the form (2.8.19) can be
rewritten in the form (2.8.3) if q(ξ, x) belongs to Sclm , but for symbols in C r Sclm with
r < ∞, these operator classes do not coincide. If q(ξ, y) belongs to such a symbol
class, we write q(D, x) ∈ ØPC r Sclm . Hence, ØPC r Sclm consists of formal adjoints of
elements of OPC r Sclm . There is also a natural formula for composition of operators
in these two classes, namely
∬
p(x, D)q(D, x)u = (2π)−n p(x, ξ)q(ξ, y)ei x−y,ξ  u(y) dy dξ. (2.8.21)
562 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Let us also remark that similar considerations are valid for operators in ØPC 0 Scl−1
with an odd principal symbol in the cotangent variable. We therefore have the
following basic result:

operators in OPC 0 Scl−1 and ØPC 0 Scl−1 with odd principal symbols in the
cotangent variable have Schwartz kernels that differ from the variable
(2.8.22)
coefficient kernels treated in Theorem 2.5.38 by error terms with weak
singularities along the diagonal.
This is going to be a guiding principle in extending our Euclidean results to the
setting of manifolds in this section.
Consider next an n-dimensional smooth oriented manifold M equipped with a
Riemannian metric tensor g. Let dVg be the volume element on M induced by the
metric g, and denote by Lgn the measure canonically associated with dVg . Also,
let Hgn−1 stand for the (n − 1)-dimensional Hausdorff measure associated with the
metric g on M. By TM and T ∗ M we denote, respectively, the tangent and cotangent
bundles on M.
Given two Hermitian vector bundles E, F → M, denote by OPC r S1,0 m (E, F )

the collection of pseudodifferential operators mapping sections of the vector bundle

E into sections of the vector bundle F which, in local coordinates of M and over
local trivializations of E, F , can be represented as matrices with entries from
m . Finally, let ØPC r S m (E, F ) consist of all formal adjoints of operators in
OPC r S1,0 cl
OPC Sclm (F , E).
r

Next, given a classical pseudodifferential operator Q(x, D) ∈ OPC 0 Scl−1 (E, F ),

denote its principal symbol by

Sym(Q; ·) : T ∗ M \ 0 → Hom (E, F ), Sym(Q; ·) ∈ C 0 Scl−1 (E, F ). (2.8.23)

Also, let
 
k Q ∈ D  M × M, F ⊗ E ∩ 𝒞0 M × M \ diag, F ⊗ E (2.8.24)

stand for the Schwartz kernel of Q(x, D). Hence, for all test functions φ, ψ on M,

D  (M) Qφ, ψ D(M) = D  (M×M) k Q, ψ ⊗ φ D(M×M), (2.8.25)

and
∫
% &
Q(x, D)φ(x) = k Q (x, y), φ(y) Ey
dLgn (y), ∀ x ∈ M \ supp φ. (2.8.26)
M
√
In view of (2.8.26) and the fact that dLgn = g dL n locally, where L n is the ordinary
n-dimensional Lebesgue measure in Rn , we conclude that

in local coordinates, the Schwartz kernel of Q(x, D) with1respect to

(2.8.27)
the ordinary Euclidean Lebesgue measure L n is k Q (x, y) g(y).
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 563

Going further, fix an open set Ω ⊆ M satisfying Hgn−1 (∂Ω) < ∞ abbreviate

σg := Hgn−1 ∂Ω. In this context, introduce the boundary-to-domain integral operator
acting on f ∈ L 1 (∂Ω, σg ) ⊗ E according to
∫
% &
𝒦Q f (x) := k Q (x, y), f (y) E y dσg (y), ∀ x ∈ M \ ∂Ω. (2.8.28)
∂Ω

With dg (x, y) standing for the geodesic distance between x and y in M, with respect
to the Riemannian metric g (cf. [112, (1.11.5)]), let us also introduce the maximal
operator
 ∫ 
 % & 
 
KQ,max f (x) := sup  k Q (x, y), f (y) E y dσg (y), ∀ x ∈ ∂Ω, (2.8.29)
ε>0  
y ∈∂Ω
dg (x,y)>ε

for each f ∈ L 1 (∂Ω, σg ) ⊗ E. Similar considerations apply to pseudodifferential

operators in the class ØPC 0 Scl−1 (E, F ).
Going further, we make the following definition which basically stipulates that
for singular integral operators on a Riemannian manifold the principal-value limit is
considered in local Euclidean coordinates.

Definition 2.8.1 In the same geometric setting as above, given a kernel


k ∈ 𝒞0 M × M \ diag, F ⊗ E , (2.8.30)

for each f ∈ L 1 (∂Ω, σg ) ⊗ E the meaning assigned to the formal expression

∫
% &
P.V. k(x, y), f (y) E y dσg (y), ∀ x ∈ ∂Ω, (2.8.31)
∂Ω

is as follows. First, for each x ∈ ∂Ω \ supp f define (2.8.31) as being the absolutely
convergent integral ∫
% &
k(x, y), f (y) E y dσg (y). (2.8.32)
∂Ω
Second, suppose supp f is contained in a local coordinate patch over which the
vector-bundles E, F trivialize.
 Using local orthonormal frames, identify the kernel
k(x, y) with a matrix k αβ (x, y) α,β and f with a vector-valued function with
components { fβ }β . Work in local Euclidean coordinates. Hence, in addition to the
original surface measure σg induced by the Riemannian metric, one can also speak
of the Euclidean surface measure σ E on ∂Ω. Locally, these measures are related
to one another as in [112, (1.11.8)], (so ρ := dσg /dσ E is meaningfully defined).
Then, in local Euclidean coordinates, (2.8.31) is identified with the vector whose
α-component is given by
564 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫
lim k αβ (x, y) fβ (y)ρ(y) dσ E (y), (2.8.33)
ε→0+
y ∈∂Ω
|x−y |>ε

at points x ∈ supp f for which the limit exists. Finally, in the remaining case, (2.8.31)
is defined using a partition of unity which reduces matters to the previous situations.
In this vein, we want to point out that, as seen from [112, (1.11.10)], the density
ρ appearing in (2.8.33) satisfies

√  rs E E
1/2
ρ= g g νr νs on ∂∗ Ω, (2.8.34)
r,s

where ν E is the Euclidean geometric measure theoretic outward unit normal to Ω.

Lastly, we observe that the same blueprint for fashioning smoothness spaces in
open sets out of their global versions in the entire Euclidean space used in [113,
§9.2] works in setting of a given manifold31 M as well. For example, first the global
p
Bessel potential space Ls (M) is transported to the manifold M from the Euclidean
setting by means of local coordinate charts after localization via a sufficiently smooth
partition of unity. Then the Bessel potential spaces in an open set Ω ⊆ M is defined
via restriction, as  
p p
Ls (Ω) := u|Ω : u ∈ Ls (M) . (2.8.35)
Similar considerations apply to other scales of interest, including Besov spaces and
Triebel-Lizorkin spaces.
Theorem 2.8.2 Suppose M is an n-dimensional compact boundaryless oriented
manifold of class 𝒞2 equipped with a continuous Riemannian metric tensor g, and
let Lgn , Hgn−1 denote, respectively, the n-dimensional Lebesgue measure and the
(n − 1)-dimensional Hausdorff measure associated with the metric g on M.
Next, consider two vector bundles E, F → M endowed with continuous Hermi-
tian metrics, and let Q(x, D) ∈ OPC 0 Scl−1 (E, F ) be a classical pseudodifferential
operator with the property that its principal symbol Sym(Q; ξ) is odd in the cotangent
variable ξ ∈ T ∗ M.
Having fixed an open set Ω ⊂ M such that ∂Ω is a UR set, denote the geometric
measure theoretic outward unit conormal to Ω by νg ∈ T ∗ M and let σg := Hgn−1 ∂Ω.
Finally, recall the integral operators from (2.8.28)-(2.8.29).
Then the following statements are true:
(1) For each function f ∈ L 1 (∂Ω, σg ) ⊗ E the principal-value integral
∫
% &
KQ f (x) := P.V. k Q (x, y), f (y) E y dσg (y), (2.8.36)
∂Ω

considered in the sense of Definition 2.8.1, is well defined at σ-a.e. point x ∈ ∂Ω,
and
31 assumed to be sufficiently regular, relative to the smoothness indices one has in mind
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 565

KQ : L p (∂Ω, σg ) ⊗ E −→ L p (∂Ω, σg ) ⊗ F (2.8.37)

is a well defined, linear, and bounded operator whenever p ∈ (1, ∞). Also,
corresponding to p = 1,

KQ : L 1 (∂Ω, σg ) ⊗ E −→ L 1,∞ (∂Ω, σg ) ⊗ F is bounded. (2.8.38)


Finally, for each p ∈ n−1 n , 1 the operator KQ induces a well-defined, linear,
and bounded mapping

KQ : H p (∂Ω, σg ) ⊗ E −→ L p (∂Ω, σg ) ⊗ F . (2.8.39)

(2) The maximal operators

KQ,max : L p (∂Ω, σg ) ⊗ E −→ L p (∂Ω, σg ) ⊗ F if 1 < p < ∞,

(2.8.40)
KQ,max : L 1 (∂Ω, σg ) ⊗ E −→ L 1,∞ (∂Ω, σg ) ⊗ F ,

are well defined, sub-linear, and bounded.

(3) With KQ denoting the singular integral operator associated with Q in the
same manner KQ has been associated with Q in (2.8.36), for each p ∈ (1, ∞)
one has  
KQ = KQ , as operators from
(2.8.41)
L p (∂Ω, σg ) ⊗ F into L p (∂Ω, σg ) ⊗ E.
(4) For every κ > 0 and p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) such that

Nκ (𝒦Q f ) L p (∂Ω,σg ) ≤ C f  L p (∂Ω,σg )⊗ E , (2.8.42)

for all f ∈ L p (∂Ω, σg ) ⊗ E, and, corresponding to p = 1, there exists C ∈ (0, ∞)

such that
Nκ (𝒦Q f ) L 1,∞ (∂Ω,σg ) ≤ C f  L 1 (∂Ω,σg )⊗ E (2.8.43)

for all f ∈ L 1 (∂Ω, σg ) ⊗ E. Moreover, if p ∈ n−1 n , 1 then

Nκ (𝒦Q f ) L p (∂Ω,σg ) ≤ C f  H p (∂Ω,σg )⊗ E , (2.8.44)

for all f ∈ H p (∂Ω, σg ) ⊗ E, where the boundary-to-domain integral operator

𝒦Q from (2.8.28) now acts on any given f ∈ H p (∂Ω, σg ) ⊗ E according to
(with ·, · denoting the natural duality bracket)
% &
𝒦Q f (x) := k Q (x, ·), f , ∀ x ∈ M \ ∂Ω. (2.8.45)

(5) Given any f ∈ L 1 (∂Ω, σg ) ⊗ E, the function 𝒦Q f possesses a nontangential

boundary trace at σg -a.e. point on ∂∗ Ω, in the precise sense that for any aperture
parameter κ > 0 one has
566 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
κ−n.t.
 1
𝒦Q f  = √ Sym(Q; νg ) f + KQ f at σg -a.e. point on ∂∗ Ω, (2.8.46)
∂Ω 2 −1
with the singular integral operator KQ considered
 as in (2.8.38). Furthermore,
for any given f ∈ H p (∂Ω, σg ) ⊗ E with p ∈ n−1n , 1 , the function 𝒦Q f , now
defined as in (2.8.45), also possesses a nontangential boundary trace at σg -a.e.
point on ∂∗ Ω, in the precise sense that for any aperture parameter κ > 0 one
has
κ−n.t.
 1
𝒦Q f  = √ Sym(Q; νg )H f + KQ f at σg -a.e. point on ∂∗ Ω, (2.8.47)
∂Ω 2 −1
where H : H p (∂Ω, σg ) ⊗ E → L p (∂Ω, σg ) ⊗ E is the L p -filtering operator on
∂Ω (cf. the discussion in [113, §4.9]), and the operator KQ is now considered
as in (2.8.39).
(6) On the scale of Lebesgue spaces with integrability exponents p ∈ [1, ∞), similar
results to those in items (1)-(6) above are valid for integral operators associ-
ated with pseudodifferential operators Q(D, x) from the class ØPC 0 Scl−1 (E, F ),

while for Hardy spaces corresponding to p ∈ n−1 n , 1 similar results to those
in items (1)-(6) above are valid for integral operators associated with pseudod-
ifferential operators Q(D, x) ∈ ØPC r Scl−1 (E, F ) where it is now assumed that

the smoothness index satisfies r > (n − 1) p1 − 1 .
In addition, given any Q(D, x) ∈ ØPC 0 Scl−1 (E, F ) it follows that for each
p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) with the property that for each
f ∈ L p (∂Ω, σg ) ⊗ E one has the L p -square function estimate
∫ ∫
 
∇(𝒦Q f )(x) p distg (x, ∂Ω) p−1 dL n (x) ≤ C | f | p dσg, (2.8.48)
g
Ω ∂Ω

where ∇ is a connection on the vector bundle F , and distg (·, ∂Ω) stands for the
distance to ∂Ω naturally associated with the metric g (or any other sufficiently
smooth background Riemannian metric on M).
Finally, if Ω is a Lipschitz domain, then for each Q(D, x) ∈ ØPC 0 Scl−1 (E, F ) the
operator
p, p
𝒦Q : L p (∂Ω, σg ) ⊗ E −→ B1/p (Ω) ⊗ F (2.8.49)
is well defined, linear, and bounded; in particular, corresponding to the case
when p = 2, the following is a well defined, linear, and bounded operator:

𝒦Q : L 2 (∂Ω, σg ) ⊗ E −→ H 1/2 (Ω) ⊗ F . (2.8.50)

(7) The L p -square function estimate (2.8.48) is also valid if we actually have
Q(x, D) ∈ OPC 1 Scl−1 (E, F ). Also, if Ω is a Lipschitz domain then the map-
ping properties recorded in (2.8.49)-(2.8.50) remain true when 𝒦Q is now
associated with Q(x, D) ∈ OPC 1 Scl−1 (E, F ).
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 567

The jump-formula (2.8.46) is particularly elegant and beautiful, as it brings to-

gether in an invariant, coordinate-free formalism, the algebraic and analytic structures
underpinning the current setting, through the involvement of the principal symbol
and, respectively, the principal-value singular integral operator associated with the
given pseudodifferential operator (we wish to emphasize that it is only for pseudod-
ifferential operators of order −1 that a genuine jump-formula like the one recorded in
(2.8.46) is to be expected). The fact that the pseudodifferential operator in question
acts between generic Hermitian vector bundles32 makes this result especially general
and versatile.
We now turn to the task of providing the proof of Theorem 2.8.2.
Proof of Theorem 2.8.2 The problem localizes and, given the invariance of the class
of pseudodifferential operators and symbols under discussion, it can be transported
in to the Euclidean setting via coordinate mappings. By fixing local frames in E
and F , the operator Q(x, D) can be locally identified, in a canonical manner, with
a matrix Q jk (x, D) j,k whose entries Q jk (x, D) are classical pseudodifferential
operators having principal symbols odd in the cotangent variable. Also, k Q (x, y),
the
 Schwartz kernel of Q(x, D), can be identified with a matrix-valued distribution
θ jk (x, y) j,k . Thanks to (2.8.27), the latter is such that

the Schwartz kernel b jk (x, x − y) of Q jk (x, D) with respect to the

1 (2.8.51)
ordinary Euclidean Lebesgue measure is θ jk (x, y) g(y).

As in (2.8.15), decompose each b jk (x, z) as

b jk (x, z) = btop
jk
(x, z) + berr
jk
(x, z) (2.8.52)

with
the top term btop
jk
(x, z) odd and positive homogeneous of degree 1 − n
(2.8.53)
in z, and the error term berr
jk
(x, z) only weakly singular (cf. (2.8.14)).

Also, with ‘hat’ denoting the partial Fourier transform in the second variable, from
(2.8.17) we see that each top term satisfies

b.
top
jk
(x, ξ) = Sym(Q jk ; ξ) for |ξ | ≥ 1. (2.8.54)

Going further, Ω may be locally identified with an open set having a UR boundary
in Rn . While we retain the same notation, Ω, for the latter set, there are now two
measures operating on ∂Ω, namely the original σg induced by the Riemannian
metric, and the Euclidean surface measure which we denote by σ E . They are related
to one another as in [112, (1.11.10)]. Moreover, the original differential geometric
outward unit conormal νg and the Euclidean outward unit normal, denoted by ν E ,

32 as the old adage goes, working on a vector bundle is like doing algebra vertically (along fibers,
which are vector spaces), and doing analysis horizontally (relative to the base point, in the underlying
manifold)
568 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

are related to one another as in [112, (1.11.10)]. Consequently, 𝒦Q can be identified

with the operator-valued matrix (ℬ jk ) j,k where, for x ∈ Ω,
∫ ;  <
1 1/2
ℬ jk f (x) := θ jk (x, y) f (y) g(y) g (y)νr (y)νs (y)
rs E E
dσ E (y)
∂Ω r,s

∫  1/2
= b jk (x, x − y) gr s (y)νrE (y)νsE (y) f (y) dσ E (y), (2.8.55)
∂Ω r,s

with the last equality coming from (2.8.51). To justify the jump-formula claimed in
(2.8.46), at σ E -a.e. point x ∈ ∂Ω write
 n.t. 

ℬ jk f  (x)
∂Ω

1 top  E
1/2
= b. x, ν (x) gr s (x)νrE (x)νsE (x) f (x)
2i jk r,s
∫  1/2
+ lim+ b jk (x, x − y) gr s (y)νrE (y)νsE (y) f (y) dσ E (y)
ε→0
y∈∂Ω
r,s
|x−y |>ε

 1/2
= − 12 iSym Q jk ; ν (x)
E
gr s (x)νrE (x)νsE (x) f (x)
r,s
∫ 
1 1/2
+ lim+ θ jk (x, y) g(y) gr s (y)νrE (y)νsE (y) f (y) dσ E (y)
ε→0
y∈∂Ω
r,s
|x−y |>ε
∫

= − 12 iSym Q jk ; νg (x) f (x) + lim+ θ jk (x, y) f (y)ρ(y) dσ E (y)
ε→0
y∈∂Ω
|x−y |>ε
∫

= − 12 iSym Q jk ; νg (x) f (x) + P.V. θ jk (x, y) f (y) dσg (y)
∂Ω

= − 12 iSym Q jk ; νg (x) f (x) + K jk f (x). (2.8.56)

Above, the first equality is a consequence of (2.8.55), (2.8.52)-(2.8.53), and the jump-
formulas proved in Theorem 2.5.38 in the class of variable coefficient Calderón-
Zygmund operators in the Euclidean setting; here, ‘hat’ stands for partial Fourier
transform in the second variable. The second equality in (2.8.56) is implied by
(2.8.54) and (2.8.51). In the third equality in (2.8.56) we have used the fact that
Sym(Q jk ; ξ) is homogeneous of degree −1 in ξ, plus the formula relating νg to ν E in
[112, (1.11.10)], and (2.8.34) (keeping in mind that currently ∂∗ Ω has full measure in
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 569

∂Ω). The fourth equality in (2.8.56) is seen from Definition 2.8.1. The final equality
in (2.8.56) actually defines the operator K jk as the principal-value integral in the
penultimate line of (2.8.56). Under the identification KQ ≡ (K jk ) j,k , the analysis
above amounts to
 n.t.  

𝒦Q f  (x) = − 12 iSym Q; νg (x) f (x) (2.8.57)
∂Ω
∫
+ P.V. k Q (x, y), f (y) E y dσg (y) at σg -a.e. x ∈ ∂Ω,
∂Ω

which is precisely (2.8.46).

Next, the claims pertaining to (2.8.36)-(2.8.40), (2.8.42)-(2.8.44) and (2.8.47)
follow in a similar fashion from the corresponding theory for singular integrals with
variable kernel in UR domains in the Euclidean setting from Theorem 2.5.38. Also,
the L p -square function estimates claimed in items (6)-(7) may be established in an
analogous manner by relying on the Euclidean result recorded in Theorem 2.5.38 for
the top terms and using [112, Lemma 8.7.11] to control less singular terms. Once
these have been established, then we may invoke [113, Lemma 9.2.40] together with
the nontangential maximal function estimates from the current item (4) and [112,
(8.6.51)] to conclude that, if Ω is actually a Lipschitz domain, then the global Besov
regularity result claimed in (2.8.49) holds. With this in hand, (2.8.50) is implied by
[113, (9.2.22)].
Finally, consider the claim made in (2.8.41). With this in mind, first note that
 
k Q (y, x) = k Q (x, y) in D (M × M, E ⊗ F ). (2.8.58)

Indeed, for two arbitrary test functions φ ∈ 𝒞2 (M, E) and ψ ∈ 𝒞2 (M, F ) we have
# $ # $

k Q (y, x), φ(x) ⊗ ψ(y) = k Q (x, y), ψ(x) ⊗ φ(y)
E ⊗F F⊗E

= Qφ, ψ F = Q ψ, φ E
% &
= k Q (x, y), φ(x) ⊗ ψ(y) E ⊗ F (2.8.59)

from which (2.8.58) follows. To start the proof of (2.8.41) in earnest, observe that
using a partition of unity and the linearity of the operators involved, matters are
reduced to showing that
 
KQ h = KQ  h (2.8.60)

for each h ∈ L p (∂Ω, σg ) ⊗ F such that supp h is contained in a local coordinate patch
O (which we will freely identify with an open set in Rn , still denoted by O) over
which the vector-bundles E, F trivialize. Using local orthonormal frames, identify
h with a vector-valued function with components {hα }α , and identify the kernel
αβ
k Q (x, y) with a matrix k Q (x, y) α,β . If the kernel k Q (x, y) is also identified in
 αβ
O with the matrix k Q (x, y) α,β , then (2.8.58) implies
570 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
αβ βα
k Q (x, y) = k Q (y, x). (2.8.61)

Pick an exponent p ∈ (1, ∞) satisfying 1/p + 1/p = 1, and bring in another


function, f ∈ L p (∂Ω, σg ) ⊗ E supported in the local coordinate patch O. Use
the local orthonormal frame in E to identify f with a vector-valued function with
components { fβ }β . Working in local Euclidean coordinates, for each ε > 0 define
 ∫ 
αβ
KQ,ε f (x) := k Q (x, y) fβ (y)ρ(y) dσ E (y) (2.8.62)
y ∈∂Ω α
|x−y |>ε

for each x ∈ ∂Ω ∩ O, and also

 ∫  
 
 αβ 
KQ,∗ f (x) := sup  k Q (x, y) fβ (y)ρ(y) dσ (y) 
E
(2.8.63)
ε>0  
y ∈∂Ω α
|x−y |>ε

for each x ∈ ∂Ω ∩ O. Then (2.8.36) and Definition 2.8.1 ensure that

lim KQ,ε f (x) = KQ f (x) at σ E -a.e. point x ∈ ∂Ω ∩ O, (2.8.64)

ε→0+

while the Euclidean Calderón-Zygmund theory for variable coefficient kernels (cf.
Theorem 2.5.38) guarantees that

KQ,∗ f ∈ L p (∂Ω ∩ O, σ E ). (2.8.65)

Granted (2.8.64)-(2.8.65), Lebesgue’s Dominated Convergence Theorem applies and

gives
lim+ KQ,ε f = KQ f in L p (∂Ω ∩ O, σ E ). (2.8.66)
ε→0

If, much as in (2.8.62)-(2.8.63), we also define, for each y ∈ ∂Ω ∩ O,

 ∫ 
βα
KQ,ε h(y) := k Q (y, x)hα (x)ρ(x) dσ E (x) (2.8.67)
x ∈∂Ω β
|y−x |>ε

then the same type of argument based on Lebesgue’s Dominated Convergence The-
orem implies that

lim+ KQ,ε h = KQ h in L p (∂Ω ∩ O, σ E ). (2.8.68)
ε→0

Thanks to the aforementioned identifications, for each ε > 0 we may now write
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 571
∫

hα (x) KQ,ε f α (x)ρ(x) dσ
E
(x)
∂Ω∩O
∫  ∫ 
αβ
= hα (x) k Q (x, y) fβ (y)ρ(y) dσ (y) ρ(x) dσ E (x)
E
x ∈∂Ω∩O
y ∈∂Ω
|x−y |>ε
∫  ∫ 
αβ
= fβ (y) k Q (x, y)hα (x)ρ(x) dσ E (x) ρ(y) dσ E (y)
y ∈∂Ω∩O
x ∈∂Ω
|x−y |>ε
∫  ∫ 
βα
= fβ (y) k Q (y, x)hα (x)ρ(x) dσ (x) ρ(y) dσ E (y)
E
y ∈∂Ω∩O
x ∈∂Ω
|y−x |>ε
∫

= fβ (y) KQ,ε h β (y)ρ(y) dσ E (y), (2.8.69)
∂Ω∩O

where we have used Fubini’s Theorem together with (2.8.61). Use (2.8.66) and
(2.8.68) to pass to limit ε → 0+ in (2.8.69), then move the integrals back to the
manifold M. This gives
∫ ∫ ∫
% 
& % & % &
f , (KQ ) h E dσg = h, KQ f F dσg = f , KQ h E dσg, (2.8.70)
∂Ω ∂Ω ∂Ω

with the first equality based on the actual definition of (KQ ) . In view of the arbi-
trariness of f , this proves that

(KQ ) h = KQ h on ∂Ω ∩ O. (2.8.71)

Recall that the support of h is contained in ∂Ω ∩ O. Bearing this in mind, it follows

from (2.8.32) in Definition 2.8.1, (2.8.58), and an argument based on Fubini’s The-
orem in the spirit of (2.8.69) (in which we now take the function f to be supported
in ∂Ω \ O) that we also have

(KQ ) h = KQ h on ∂Ω \ O. (2.8.72)

Altogether, this shows that (KQ ) h = KQ h on ∂Ω, finishing the proof of (2.8.60).

It is interesting to contrast the results in items (6)-(7) of Theorem 2.8.2 with the
following global regularity result in NTA domains on manifolds for null-solutions
of weakly elliptic systems. Recall that a differential operator L : E → F acting
between two vector bundles E, F → M is called weakly elliptic if the symbol map
Sym(L; ξ) : E x → Fx is invertible for each x ∈ M and ξ ∈ Tx∗ M \ {0}. This should
be contrasted with the Legendre-Hadamard strong ellipticity condition, demanding
the existence of a constant c ∈ (0, ∞) such that
572 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
% &
Re Sym(−L; ξ)η, η c Fx
≥ c|η| 2Fx
(2.8.73)
for all x ∈ M, η ∈ Fx and ξ ∈ Tx∗ M with |ξ | = 1.

Proposition 2.8.3 Let M be an n-dimensional compact boundaryless oriented man-

ifold of class 𝒞2 equipped with a Lipschitz Riemannian metric tensor g. Denote by
Lgn the n-dimensional Lebesgue measure associated with the metric g on M, and let
dg (·, ·) stand for the geodesic distance between points in M. Also, let E, F → M be
two Hermitian vector bundles with Lipschitz metrics, and denote by ∇ a connection
on F . Consider a weakly elliptic, second-order, differential operator L mapping
sections of E into sections of F such that, when written in local coordinates as
   
αβ αβ
Lu = ∂j a jk ∂k uβ + b j ∂ j uβ + d αβ uβ (2.8.74)
α
j,k β j β β

where u = (uβ )β , its coefficients exhibit the following type of regularity:

αβ αβ
a jk ∈ Lip, bj ∈ L1r , d αβ ∈ L r for some r > n. (2.8.75)

Then, given any one-sided NTA domain Ω ⊂ M along with p ∈ (r/(r − 1), r)
and s ∈ (0, 1), there exists some constant C = C(Ω, L, p, s) ∈ (0, ∞) such that for
each section u ∈ L p (Ω, Lgn ) ⊗ E satisfying Lu = 0 in Ω one has

uBsp, p (Ω)⊗ E ≤ Cu L p (Ω, L gn )⊗ E

∫  1/p
+C |(∇u)(x)| p distg (x, ∂Ω) p(1−s) dLgn (x) . (2.8.76)
Ω

As a corollary, corresponding to the particular choice p := 2 and s := 1/2, one

has

u H 1/2 (Ω)⊗ E ≤ Cu L 2 (Ω, L gn )⊗ E

∫  1/2
+C |(∇u)(x)| 2 distg (x, ∂Ω) dLgn (x) . (2.8.77)
Ω

Proof Work in local coordinates in a small neighborhood of a boundary point

x∗ ∈ ∂Ω, over which the vector bundles trivialize, and drop the dependence of
various pieces of notations on E, F in what follows. We shall need an interior
elliptic estimate that appears in [131, Proposition 3.4] to the effect that, under the
regularity assumptions on the coefficients imposed in (2.8.75), if p ∈ (r/(r − 1), r)
then whenever x ∈ Ω and 0 < t < 2 dist(x, ∂Ω)/3 one has
⨏  1/p
sup |∇u| ≤ Ct −1 |u(y) − u(x)| p dy + Ct 1−n/r |u(x)|, (2.8.78)
B(x,t) B(x,3t/2)

where the constant C ∈ (0, ∞) is independent of u, x and t. For each point x, let us
abbreviate
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 573

ρ(x) := dist(x, ∂Ω) (2.8.79)

and, whenever x ∈ Ω and 0 < t < ρ(x)/2, define the oscillation
⨏
osc (u, x, t) :=
0
|u(y) − u(x)| dy. (2.8.80)
B(x,t)

Then [158, Corollary 1, p. 398] implies that whenever 1 < p < ∞ and s ∈ (0, 1) we
have

uBsp, p (Ω) ≈ u L p (Ω, L n )

  1/p 
 ∫ 1 ρ(·)
 2 dt 
+  
p
osc0 (u, ·, t) 1+sp  . (2.8.81)
 0 t 
L p (Ω, L n )

Next, for every x ∈ Ω and t ∈ 0, ρ(x)/2 we estimate
⨏ ⨏
osc0 (u, x, t) = |u(y) − u(x)| dy ≤ Ct |(∇u)(y)| dy
B(x,t) B(x,t)

≤ Ct · sup |∇u| ≤ Ct · sup |∇u|

B(x,t) B(x,ρ(x)/2)
⨏  1/p
≤ Ct ρ(x)−1 |u(y) − u(x)| p dy
B(x,3ρ(x)/4)

+ Ct ρ(x)1−n/r |u(x)|
⨏  1/p
≤ Ct |(∇u)(y)| p dy + Ct ρ(x)1−n/r |u(x)|, (2.8.82)
B(x,3ρ(x)/4)

by Poincaré’s inequality (used twice) and the

 interior estimate recorded in (2.8.78).
This implies that whenever x ∈ Ω and t ∈ 0, ρ(x)/2 we have
⨏
p 1
osc0 (u, x, t) ≤ C |(∇u)(y)| p dy
tp B(x,3ρ(x)/4)

+ C ρ(x) p−(np)/r |u(x)| p . (2.8.83)

Consequently,
 ∫ 
 ρ(·)/2 1/p 
 p dt 
 osc0 (u, ·, t)  ≤ I + II, (2.8.84)
 0 t 1+sp 
L p (Ω, L n )

where
574 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫ ∫ ρ(x)/2 ⨏ 1/p
I := C t p−1−sp
|(∇u)(y)| dy dt dx
p
(2.8.85)
Ω 0 B(x,3ρ(x)/4)

and
∫ ∫ ρ(x)/2   1/p
II := C t p−1−sp ρ(x) p−(np)/r |u(x)| p dt dx . (2.8.86)
Ω 0

Note that
∫ ⨏ 1/p
I≤C ρ(x) p(1−s)
|(∇u)(y)| p dy dx
Ω B(x,3ρ(x)/4)
∫ ∫ 1/p
≤C ρ(x) p(1−s)−n |(∇u)(y)| p 1dist(x,y)≤3ρ(x)/4 dy dx . (2.8.87)
Ω Ω

Since whenever dist(x, y) ≤ 3ρ(x)/4 we necessarily have ρ(x) ≈ ρ(y), we may

further estimate
∫ ∫ 1/p
I≤C ρ(y) p(1−s)−n |(∇u)(y)| p 1dist(x,y)≤cρ(y) dy dx
Ω Ω
∫ ∫ 1/p
≤C ρ(y) p(1−s)−n
|(∇u)(y)| p
1dist(x,y)≤cρ(y) dx dy
Ω Ω
∫ 1/p
≤C ρ(y) p(1−s) |(∇u)(y)| p dy , (2.8.88)
Ω

which suits our purposes. There remains to observe that, since r > n and s < 1,
∫ 1/p
II ≤ C ρ(x) p(2−s−n/r) |u(x)| p dx ≤ Cu L p (Ω, L n ) . (2.8.89)
Ω

At this stage, (2.8.76) follows from [113, (9.2.149)], (2.8.84), (2.8.88), and (2.8.89).

The stage is set to fashion out of Theorem 2.8.2 a similar result for boundary
layer potentials associated with a given elliptic operator with coefficients of limited
smoothness. Before stating this result we make the following convention. Suppose
E, F , G, H → M are four vector bundles. Given some arbitrary double distribution
E ∈ D (M × M, E ⊗ F ) along with two differential operators P  : E → G and
P : F → H , writing
x ⊗ Py )E(x, y) or, simply (P
(P  ⊗ P)E(x, y), (2.8.90)

 acts in the variable x on the columns of E(x, y) while P acts in

indicates that P
the variable y on the rows of E(x, y). More specifically, work in local coordinates
over which E, F trivialize, and let {eα }α and { fβ }β be arbitrary local orthonormal
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 575

frames for E and F , respectively. Use these to identify the distribution E(x, y) with
a matrix Eαβ (x, y)eα ⊗ fβ , in the sense that for each ϕ ∈ 𝒟(M, F ) locally expressed
as ϕ = ϕβ fβ we have
% & % &
E(x, y), ϕ(y) = Eαβ (x, y), ϕβ (y) eα . (2.8.91)

With I denoting the identity operator then define

 
x E(x, y) := (P
P x ⊗ Iy )E(x, y) := Px Eαβ (x, y) eα ⊗ fβ, (2.8.92)
 
Py E(x, y) := (Ix ⊗ Py )E(x, y) := eα ⊗ Py Eαβ (x, y) fβ . (2.8.93)

Otherwise said, we are declaring that

x = (P
P x ⊗ Iy ) acts on the columns of E(x, y), and
(2.8.94)
Py = (Ix ⊗ Py ) acts on the rows of E(x, y).

For further reference let us also point out here that

% & % &
x ⊗ Iy )E(x, y), ϕ(x) ⊗ ψ(y) = E(x, y), (P
(P  ϕ)(x) ⊗ ψ(y)
(2.8.95)
for any ϕ ∈ D(M, G) and ψ ∈ D(M, F ),

and
% & % &
(Ix ⊗ Py )E(x, y), ϕ(x) ⊗ ψ(y) = E(x, y), ϕ(x) ⊗ (P ψ)(y)
(2.8.96)
for any ϕ ∈ D(M, E) and ψ ∈ D(M, H ).

Theorem 2.8.4 Let M be an n-dimensional compact boundaryless oriented mani-

fold of class 𝒞2 equipped with a 𝒞1 Riemannian metric tensor g. Let Lgn and Hgn−1
be, respectively, the n-dimensional Lebesgue measure and the (n − 1)-dimensional
Hausdorff measure associated with the metric g on M. Also, let E, F → M be two
Hermitian vector bundles equipped with 𝒞1 metrics, and use ∇ to denote connections
on E and F .
In this context, consider a weakly elliptic, second-order, differential operator L
mapping sections of E into sections of F such that, when written in local coordinates
as
 
  
αβ αβ αβ
Lu = ∂j a jk ∂k uβ + b j ∂ j uβ + d uβ , (2.8.97)
j,k β j β β α

its coefficients satisfy

αβ αβ
a jk ∈ 𝒞1, bj ∈ L ∞, d αβ ∈ L r for some r > n. (2.8.98)
576 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

In addition, assume that in local coordinates the coefficients of L ∗ , the Hermitian

adjoint of L, also satisfy33 (2.8.98). Also, suppose that

L : H 1 (M) ⊗ E −→ H −1 (M) ⊗ F is invertible (2.8.99)

and let

E ∈ D (M × M, E ⊗ F ) denote the Schwartz kernel of L −1 . (2.8.100)

Next, let Ω be an arbitrary UR domain in M. Denote its geometric measure

theoretic outward unit conormal by νg ∈ T ∗ M and abbreviate σg := Hgn−1 ∂Ω.
Also, let dg (·, ·) stand for the geodesic distance between points in M. In this setting,
define action of the single layer operator 𝒮 and of its boundary version S on an
arbitrary function f ∈ L 1 (∂Ω, σg ) ⊗ F as
∫
% &
𝒮 f (x) := E(x, y), f (y) Fy dσg (y), x ∈ Ω, (2.8.101)
∂Ω

and, respectively,
∫
% &
S f (x) := E(x, y), f (y) Fy
dσg (y), x ∈ ∂Ω. (2.8.102)
∂Ω

In addition, for a first-order differential operator P : F → F with continuous

coefficients, consider the integral operator with kernel (Ix ⊗ Py )E(x, y), i.e., for
each function f ∈ L 1 (∂Ω, σg ) ⊗ F set
∫
% &
A f (x) := (Ix ⊗ Py )E(x, y), f (y) Fy dσg (y), x ∈ Ω. (2.8.103)
∂Ω

Define the maximal integral operator Amax acting on each f ∈ L 1 (∂Ω, σg ) ⊗ F

according to
 ∫ 
 % & 
 
Amax f (x) := sup  (Ix ⊗ Py )E(x, y), f (y) Fy dσg (y), (2.8.104)
ε>0  
y ∈∂Ω
dg (x,y)>ε

for all x ∈ ∂Ω, and consider

∫
% &
A f (x) := P.V. (Ix ⊗ Py )E(x, y), f (y) Fy dσg (y) (2.8.105)
∂Ω

for σg -a.e. x ∈ ∂Ω, with the principal-value integral understood in the sense of
Definition 2.8.1.
Then the following claims are true.
αβ αβ
33 this is the case if in place of (2.8.98) we start by assuming that a j k ∈ 𝒞1 , b j ∈ L1r , and
d αβ ∈ L r for some r > n
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 577

(1) The boundary-to-domain single layer, originally acting on L 1 (∂Ω, σg ) ⊗ F as

p
in (2.8.101) extends to a linear mapping acting on each f ∈ L−1 (∂Ω, σg ) ⊗ F
with p ∈ (1, ∞) as
#  $
𝒮 f (x) := L p (∂Ω,σ )⊗ F E(x, ·)∂Ω, f p , ∀x ∈ Ω, (2.8.106)
1 g L−1 (∂Ω,σg )⊗ F

where p ∈ (1, ∞) is such that 1/p + 1/p = 1. Defined as such, for each
p ∈ (1, ∞) one then has
p
L(𝒮 f ) = 0 in Ω, ∀ f ∈ L−1 (∂Ω, σg ) ⊗ F , (2.8.107)

and
p
𝒮 f ∈ 𝒞0loc (Ω) ⊗ E, ∀ f ∈ L−1 (∂Ω, σg ) ⊗ F , (2.8.108)
while
𝒮 f ∈ 𝒞1loc (Ω) ⊗ E, ∀ f ∈ L 1 (∂Ω, σg ) ⊗ F . (2.8.109)
Also, for each aperture parameter κ > 0 and any p ∈ [1, ∞) there exists a
constant C ∈ (0, ∞) with the property that

Nκ (𝒮 f ) L p (∂Ω,σg ) ≤ C f  L p (∂Ω,σg )⊗ F

(2.8.110)
for each f ∈ L p (∂Ω, σg ) ⊗ F , if p ∈ [1, ∞],

Nκ (𝒮 f ) L p (∂Ω,σg ) ≤ C f  L p (∂Ω,σg )⊗ F

−1
p
(2.8.111)
for each f ∈ L−1 (∂Ω, σg ) ⊗ F , if p ∈ (1, ∞),

Nκ (∇𝒮 f ) L p (∂Ω,σg ) ≤ C f  L p (∂Ω,σg )⊗ F

(2.8.112)
for each f ∈ L p (∂Ω, σg ) ⊗ F , if p ∈ (1, ∞),
and, corresponding to p = 1,

Nκ (∇𝒮 f ) L 1,∞ (∂Ω,σg ) ≤ C f  L 1 (∂Ω,σg )⊗ F

(2.8.113)
for each function f ∈ L 1 (∂Ω, σg ) ⊗ F .

In fact, a more nuanced version of (2.8.110) holds. Namely, if n = dim M ≥ 3

one has:
578 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

if p ∈ (1, n − 1) and 1/q = 1/p − 1/(n − 1) there exists C ∈ (0, ∞)

so that Nκ (𝒮 f ) L q (∂Ω,σg ) ≤ C f  L p (∂Ω,σg )⊗ F, ∀ f ∈ L p (∂Ω, σg ) ⊗ F ,

while for each p ∈ (n − 1, ∞) there exists C ∈ (0, ∞) such that

 
Nκ (𝒮 f ) ∞ ≤ C f  L p (∂Ω,σg )⊗ F, ∀ f ∈ L p (∂Ω, σg ) ⊗ F ,
L (∂Ω,σg )

and, corresponding to p = n − 1, for every q ∈ (1, ∞) there holds

 
Nκ (𝒮 f ) q ≤ Cq  f  L n−1 (∂Ω,σg )⊗ F, ∀ f ∈ L n−1 (∂Ω, σg ) ⊗ F ,
L (∂Ω,σg )

while corresponding to the critical value p = 1 one has the estimate

 
Nκ (𝒮 f ) (n−1)/(n−2),∞ ≤ C f  L 1 (∂Ω,σg )⊗ F, ∀ f ∈ L 1 (∂Ω, σg ) ⊗ F .
L (∂Ω,σg )
(2.8.114)
In addition, in the case when n = 2 one has
 
Nκ (𝒮 f ) ∞ ≤ C f  L p (∂Ω,σg )⊗ F
L (∂Ω,σg )
(2.8.115)
for each f ∈ L (∂Ω, σg ) ⊗ F with p ∈ (1, ∞),
p

and
 
Nκ (𝒮 f ) ≤ C  f  L 1 (∂Ω,σg )⊗ F
q
L q (∂Ω,σg )
(2.8.116)
for each f ∈ L (∂Ω, σg ) ⊗
1 F and each q ∈ (1, ∞).

(2) The single layer operator satisfies a number of square function styled estimates.
First, if hypothesis (2.8.98) regarding the regularity of the coefficients of the
differential operator L is strengthened to
αβ αβ
a jk ∈ 𝒞1+γ, bj ∈ L1r , d αβ ∈ L r ,
(2.8.117)
for some γ > 0 and r > n,

then for each integrability exponent p ∈ (1, ∞) there exists some constant
p
C ∈ (0, ∞) with the property that for each f ∈ L−1 (∂Ω, σg ) ⊗ F one has
∫    1/p
∇(𝒮 f )(x) p distg (x, ∂Ω) p−1 dL n (x) ≤ C f  L p (∂Ω,σg )⊗ F, (2.8.118)
g −1
Ω
while for each q ∈ (1, ∞) and p ∈ (q, ∞] there exists a finite constant C > 0
p
such that for every f ∈ L−1 (∂Ω, σg ) ⊗ F one has
  ∫  1/q 
  
sup r 1−n ∇𝒮 f q distg (·, ∂Ω)q−1 dL n  (2.8.119)
 g  p
r >0 B(z,r)∩Ω Lz (∂Ω,σg )

≤ C f  L p (∂Ω,σg )⊗ F .
−1

Second, if hypothesis (2.8.98) is strengthened to

2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 579
αβ αβ
a jk ∈ 𝒞1+γ, bj ∈ L ∞, d αβ ∈ L r ,
(2.8.120)
for some γ > 0 and r > n,

then for each integrability exponent p ∈ (1, r) there exists some C ∈ (0, ∞) with
p
the property that for each f ∈ L−1 (∂Ω, σg ) ⊗ F one has
∫    1/p
∇2 (𝒮 f )(x) p distg (x, ∂Ω) p−1 dL n (x) ≤ C f  L p (∂Ω,σg )⊗ F .
g
Ω
(2.8.121)
Third, if in place of (2.8.98) one now assumes
αβ αβ
a jk ∈ 𝒞1+γ, bj ∈ L1r , d αβ ∈ 𝒞γ,
(2.8.122)
for some γ > 0 and r > n,

then for each q ∈ (1, r) and p ∈ (q, ∞] there exists a finite constant C > 0 such
that for every function f ∈ L p (∂Ω, σg ) ⊗ F one has
  ∫  1/q 
  2 q
sup r 1−n ∇ 𝒮 f  distg (·, ∂Ω)q−1 dL n  (2.8.123)
 g  p
r >0 B(z,r)∩Ω Lz (∂Ω,σg )

≤ C f  L p (∂Ω,σg )⊗ F .

Finally, if Ω is actually a Lipschitz domain, then for each p ∈ (1, ∞) the single
layer operator induces well-defined, linear, and bounded mappings
p p, p
𝒮 : L−1 (∂Ω, σg ) ⊗ F −→ B1/p (Ω) ⊗ E for each p ∈ (1, ∞)
(2.8.124)
provided the regularity conditions in (2.8.117) hold,

and
p, p
𝒮 : L p (∂Ω, σg ) ⊗ F −→ B1+1/p (Ω) ⊗ E for each p ∈ (1, r)
(2.8.125)
provided the regularity conditions in (2.8.120) hold.

(3) For each function f ∈ L 1 (∂Ω, σg ) ⊗ F and aperture parameter κ > 0, one has
κ−n.t.
𝒮 f ∂Ω = S f at σg -a.e. point on ∂Ω, (2.8.126)

 κ−n.t.

and ∇𝒮 f  exists at σg -a.e. point on ∂Ω. (2.8.127)
∂Ω

Moreover, for each p ∈ (1, ∞), the operators

p
S : L p (∂Ω, σg ) ⊗ F −→ L1 (∂Ω, σg ) ⊗ E, (2.8.128)
p
S : L−1 (∂Ω, σg ) ⊗ F −→ L p (∂Ω, σg ) ⊗ E, (2.8.129)
580 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

are well defined, linear, and bounded. In fact, the nontangential trace formula
p
in (2.8.126) continues to hold for each f ∈ L−1 (∂Ω, σg ) ⊗ F with p ∈ (1, ∞).
Also, for all continuous vector fields X, Y , and every f ∈ L p (∂Ω, σg ) ⊗ F with
p ∈ (1, ∞), one has
∫
% & % & 
∂τXY (S f )(x) = νg (x), X  (x) P.V. ∇Y(x) ⊗ Iy E(x, y), f (y) dσg (y)
∂Ω
∫
% & % & 
− νg (x), Y  (x) P.V. ∇X(x) ⊗ Iy E(x, y), f (y) dσg (y)
∂Ω

at σg -almost every point x belonging to ∂Ω, (2.8.130)

where ∂τXY is understood in the sense of [112, (1.12.89)] (while keeping

(2.8.128) in mind).
In addition, for each p ∈ (1, ∞) the mapping

S : L p (∂Ω, σg ) ⊗ F −→ L p (∂Ω, σg ) ⊗ E (2.8.131)

is compact, and its (real ) transpose is the operator

 
S  : L p (∂Ω, σg ) ⊗ E −→ L p (∂Ω, σg ) ⊗ F , 1/p + 1/p = 1, (2.8.132)

acting on each g ∈ L p (∂Ω, σg ) ⊗ E according to
∫

% &
S g(x) = E L  (x, y), g(y) E y dσg (y), x ∈ ∂Ω, (2.8.133)
∂Ω

where E L  ∈ D (M × M, F ⊗ E) denotes the Schwartz kernel of

 −1
L : H 1 (M) ⊗ E −→ H −1 (M) ⊗ F . (2.8.134)

(4) For each aperture parameter κ > 0 and each p ∈ [1, ∞) there exists a constant
C ∈ (0, ∞) with the property that

Nκ (A f ) L p (∂Ω,σg ) ≤ C f  L p (∂Ω,σg )⊗ F

(2.8.135)
for each f ∈ L p (∂Ω, σg ) ⊗ F , if p ∈ (1, ∞),

and, corresponding to p = 1,

Nκ (A f ) L 1,∞ (∂Ω,σg ) ≤ C f  L 1 (∂Ω,σg )⊗ F

(2.8.136)
for each f ∈ L 1 (∂Ω, σg ) ⊗ F .

Moreover,

L(A f ) = 0 in Ω, ∀ f ∈ L 1 (∂Ω, σg ) ⊗ F , (2.8.137)

2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 581

and
A f ∈ 𝒞0loc (Ω) ⊗ E, ∀ f ∈ L 1 (∂Ω, σg ) ⊗ F . (2.8.138)
(5) For each f ∈ L 1 (∂Ω, σg ) ⊗ F , the principal-value integral operator
∫
% &
A f (x) := P.V. (Ix ⊗ Py )E(x, y), f (y) E y dσg (y), (2.8.139)
∂Ω

considered in the sense of Definition 2.8.1, is well defined at σg -a.e. x ∈ ∂Ω.

Moreover,

A : L p (∂Ω, σg ) ⊗ F −→ L p (∂Ω, σg ) ⊗ E if 1 < p < ∞, (2.8.140)

A : L 1 (∂Ω, σg ) ⊗ F −→ L 1,∞ (∂Ω, σg ) ⊗ E, (2.8.141)

are linear and bounded mappings. In fact, the maximal operator (defined much
as in (2.8.104)) induces sub-linear bounded mappings

Amax : L p (∂Ω, σg ) ⊗ F −→ L p (∂Ω, σg ) ⊗ E if 1 < p < ∞, (2.8.142)

Amax : L 1 (∂Ω, σg ) ⊗ F −→ L 1,∞ (∂Ω, σg ) ⊗ E. (2.8.143)

(6) For every f ∈ L 1 (∂Ω, σg )⊗ F and every aperture parameter κ > 0 the following
jump-formula holds:
κ−n.t.
A f ∂Ω = 12 iSym(L; νg )−1 Sym(P; νg ) f + A f (2.8.144)

at σg -a.e. point on ∂Ω, where the superscript  denotes (real ) transposition.

(7) If hypothesis (2.8.98) regarding the regularity of the coefficients of the differential
operator L is strengthened to
αβ αβ
a jk ∈ 𝒞1+γ, bj ∈ L1r , d αβ ∈ L r ,
(2.8.145)
for some γ > 0 and r > n,

then
A f ∈ 𝒞1loc (Ω) ⊗ E, ∀ f ∈ L 1 (∂Ω, σg ) ⊗ F , (2.8.146)
and for each p ∈ (1, ∞) there exists C ∈ (0, ∞) such that for every function
f ∈ L p (∂Ω, σg ) ⊗ F one has
∫ ∫
|∇(A f )(x)| p distg (x, ∂Ω) p−1 dLgn (x) ≤ C | f | p dσg . (2.8.147)
Ω ∂Ω

In addition, under the stronger regularity assumptions imposed in (2.8.145),

whenever one has 1 < q < ∞ and q < p ≤ ∞ it follows that there exists some
finite constant C > 0 for which the following type of square function estimate
582 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
  ∫  1/q 
  
 sup r 1−n ∇A f q distg (·, ∂Ω)q−1 dL n  (2.8.148)
 g  p
r >0 B(z,r)∩Ω Lz (∂Ω,σg )

≤ C f  L p (∂Ω,σg )⊗ F

holds for every f ∈ L p (∂Ω, σg ) ⊗ F ; in particular, corresponding to q := 2

and the end-point p = ∞, for each function f ∈ L ∞ (∂Ω, σg ) ⊗ F one has the
following Carleson measure estimate:
 ∫ 
 
∇A f 2 distg (·, ∂Ω) dL n ≤ C f  2 ∞
sup r 1−n g L (∂Ω,σg )⊗ F .
z ∈∂Ω, r >0 B(z,r)∩Ω
(2.8.149)
Finally, if in addition to (2.8.145) one also assumes that Ω is a Lipschitz domain,
then for each p ∈ (1, ∞) the operator
p, p
A : L p (∂Ω, σg ) ⊗ F −→ B1/p (Ω) ⊗ E (2.8.150)

is well defined, linear, and bounded.

(8) Similar results to those described in items (4)-(6) above (with the exception of
(2.8.137)) are valid for the integral operators
∫
% &
 f (x) :=
A x ⊗ Iy )E(x, y), f (y)
(P dσg (y), x ∈ Ω, (2.8.151)
Fy
∂Ω

and
∫
% &
 f (x) := P.V.
A x ⊗ Iy )E(x, y), f (y)
(P dσg (y), x ∈ ∂Ω, (2.8.152)
Fy
∂Ω

where P  : E → E is a first-order differential operator with continuous coeffi-

cients. In this scenario, in place of the jump-formula (2.8.144), for each function
f ∈ L 1 (∂Ω, σg ) ⊗ F and at σg -a.e. point on ∂Ω one now has
κ−n.t.
f 
A  νg ) Sym(L; νg )−1 f + A
= − 12 iSym(P; f . (2.8.153)
∂Ω

 have 𝒞1 coefficients while the lower-order ones

Moreover, if the top terms in P
are continuous, and if hypothesis (2.8.98) is strengthened to
αβ αβ
a jk ∈ 𝒞1+γ, bj ∈ L ∞, d αβ ∈ L r ,
(2.8.154)
for some γ > 0 and r > n,

then the analogues of (2.8.147) and (2.8.148), albeit in a restricted range, are
also valid in this case: first, for each p ∈ (1, r) there exists some C ∈ (0, ∞) such
that for every function f ∈ L p (∂Ω, σg ) ⊗ F one has
∫ ∫
  
∇ A f (x) p distg (x, ∂Ω) p−1 dLgn (x) ≤ C | f | p dσg (2.8.155)
Ω ∂Ω
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 583

and, second, for each q ∈ (1, r) and p ∈ (q, ∞] there exists C ∈ (0, ∞) such that
for every f ∈ L p (∂Ω, σg ) ⊗ F one has
  ∫  1/q 
  
 sup r 1−n ∇ A f q distg (·, ∂Ω)q−1 dLgn  (2.8.156)
  p
r >0 B(z,r)∩Ω Lz (∂Ω,σg )

≤ C f  L p (∂Ω,σg )⊗ F .

Finally, if in addition to (2.8.154) and the latest requirements on the coefficients

 one also assumes that Ω is a Lipschitz domain then for each p ∈ (1, r) the
of P
operator
A : L p (∂Ω, σg ) ⊗ F −→ B p, p (Ω) ⊗ E (2.8.157)
1/p

is well defined, linear, and bounded.

(9) The (real ) transpose A of A in (2.8.140) is the operator

A : L p (∂Ω, σg ) ⊗ E −→ L p (∂Ω, σg ) ⊗ F , 1 < p < ∞, (2.8.158)

acting on each h ∈ L p (∂Ω, σg ) ⊗ E at σg -a.e. point x ∈ ∂Ω according to

∫
% &
A h(x) = P.V. (Px ⊗ Iy )E L  (x, y), h(y) E y dσg (y) (2.8.159)
∂Ω

where, as before, E L  ∈ D (M × M, F ⊗ E) denotes the Schwartz kernel of

  −1
L .
 
Similarly, the (real ) transpose A  of A  from (2.8.152) is the operator sending
each function h ∈ L (∂Ω, σg ) ⊗ E with 1 < p < ∞ into
p

∫
  % &

A h(x) = P.V. y )E L  (x, y), h(y)
(Ix ⊗ P dσg (y), (2.8.160)
Ey
∂Ω

for σg -a.e. x ∈ ∂Ω.

(10) If Ω is also a one-sided NTA domain and hypothesis (2.8.98) for L is strength-
ened to
αβ αβ
a jk ∈ 𝒞1, bj ∈ L1r , d αβ ∈ L r ,
(2.8.161)
for some r > n,

then for each p ∈ r/(r − 1), r the following operator is well defined, linear,
and bounded:
p p, p
𝒮 : L−1 (∂Ω, σg ) ⊗ F −→ B1/p (Ω) ⊗ E. (2.8.162)
Moreover, if hypothesis (2.8.98) for L is further strengthened to
584 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
αβ αβ
a jk ∈ 𝒞1+γ, bj ∈ L1r , d αβ ∈ L r ,
(2.8.163)
for some γ > 0 and r > n,

while Ω is once again assumed to also be a one-sided NTA domain, then the
operator
p, p
A : L p (∂Ω, σg ) ⊗ F −→ B1/p (Ω) ⊗ E (2.8.164)

is well defined, linear, and bounded for each p ∈ r/(r − 1), r .

If the all the underlying differentiability assumptions for the Riemannian manifold
M, the Hermitian vector bundles E, F , and the coefficients of the differential
operators L, P, P  are upgraded to 𝒞∞ , then Q := L −1 P and Q  := PL
 −1 are both
−1
classical pseudodifferential operators in OPScl , with Schwartz kernels and principal
symbols given by

k Q (x, y) = (Ix ⊗ Py )E(x, y),

(2.8.165)
Sym(Q; ξ) = Sym(L; ξ)−1 Sym(P ; ξ)

and, respectively,
x ⊗ Iy )E(x, y),
k Q (x, y) = (P
(2.8.166)
 ξ) = Sym(P;
Sym(Q;  ξ)Sym(L; ξ)−1 .
As such, in the current scenario, the “mother” jump-formula for integral operators
associated with Schwartz kernels of classical pseudodifferential operators in OPScl−1
possessing odd principal symbols in the cotangent variable, recorded in (2.8.46),
directly gives the jump-formulas in (2.8.153) and (2.8.144) (also keeping in mind
[112, (1.7.17)] in the case of the latter).
In this regard, we also wish to mention that, in principle, P, P  are allowed to be
zero-th order differential operators (hence, plain homomorphisms). In such a case,
their principal symbols vanish identically. As a result, the jump-terms in (2.8.144),
(2.8.153) disappear, rendering these jump-formulas in line with the boundary trace
result for the single layer, recorded in (2.8.126). This is to be expected, and logically
consistent, since now the integral kernels of A, A  exhibit the same type of main
singularity as for the single layer operator.
We conclude this preamble with several useful observations regarding the alge-
braic nature of the double distribution E L . For one thing, the fact that E L is the
Schwartz kernel of L −1 means that

E L ∈ D  M × M, E ⊗ F (2.8.167)

has the property that for arbitrary test sections ϕ ∈ D(M, E) and ψ ∈ D(M, F ) we
have % &
E L (x, y), ϕ(x) ⊗ ψ(y) = L −1 ψ, ϕ. (2.8.168)
Based on this it may be checked that
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 585

E L (x, y) = E L (x, y), (2.8.169)


E L  (x, y) = E L (y, x) , (2.8.170)
 
E L ∗ (x, y) = E L (y, x) . (2.8.171)

We also wish to bring in the diagonal Dirac distribution


δ  = δ E ∈ D  M × M, E ⊗ E , (2.8.172)

defined by setting
∫

δ , Θ := Trace Θ(x, x) dLgn (x) for each Θ ∈ D M × M, E ⊗ E , (2.8.173)
M

where Trace Θ(x, x) := α θ αα (x, x) if the test section Θ is expressed in an arbitrary
local orthonormal frame {eα }α of E as Θ(x, y) = θ αβ (x, y) eα ⊗ eβ . In particular,
∫
% &
δ  (x, y), ϕ(x) ⊗ ψ(y) = ϕ(x), ψ(x) dLgn (x) (2.8.174)
M

for any ϕ, ψ ∈ D(M, E). Otherwise said,

% &
δ  (x, y), ψ(y) = ψ(x) for each ψ ∈ D(M, E). (2.8.175)

Let us also note that δ  is supported on the diagonal, hence

δ  (x, y) = 0 for x  y. (2.8.176)

Lastly, with the help of (2.8.95)-(2.8.96), (2.8.168), (2.8.174) it may be readily

verified that

(Lx ⊗ Iy )E L (x, y) = δ F (x, y) in D  M × M, F ⊗ F , (2.8.177)

(Ix ⊗ Ly )E L (x, y) = δ E (x, y) in D  M × M, E ⊗ E , (2.8.178)

and, more generally, for any differential operators P : E → G and Q ∈ F → H we

have

(Lx ⊗ Q y )E L (x, y) = (Ix ⊗ Q y )δ F (x, y) in D  M × M, F ⊗ H , (2.8.179)

(Px ⊗ Ly )E L (x, y) = (Px ⊗ Iy )δ E (x, y) in D  M × M, G ⊗ E . (2.8.180)

Here is the proof of Theorem 2.8.4.

Proof of Theorem 2.8.4 We begin with a review of the nature of the singularity the
Schwartz kernel of L −1 along the diagonal, which is of independent interest. Denote
the Schwartz kernel of L −1 by E(x, y). Then, in local coordinates,
586 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
1
Lx E(x, y) g(y) = δy (x), (2.8.181)

where the operator L acts in the variable x on columns, and

1
Ly E(x, y) g(x) = δx (y), (2.8.182)

where the transposed operator L  acts in the variable y on rows. Granted the regu-
larity conditions from (2.8.98), the analysis34 of the Schwartz kernel E(x, y) of L −1
provided in [119, Propositions 2.2-2.6] (cf. also the discussion in [124, Appendix])
applies and yields the following Hölder and Sobolev regularity properties:
α 
E(·, y) ∈ 𝒞loc M \ {y}, E ⊗ Fy , ∀y ∈ M, ∀α < 2 − n/r, (2.8.183)

loc M × M \ diag, E ⊗ F for some θ = θ(n, r) > 0,
E(·, ·) ∈ 𝒞1+θ (2.8.184)
p 
E(·, y) ∈ Ls,loc M \ {y}, E ⊗ Fy , ∀y ∈ M, ∀s < 2, ∀p < r, (2.8.185)

p
E(·, y) ∈ L1 M, E ⊗ Fy , ∀p < n/(n − 1), uniformly in y ∈ M. (2.8.186)

The same analysis also yields the decomposition

1
E(x, y) = 1 E0 (y, x − y) + E1 (x, y) (2.8.187)
g(y)

where the residual piece E1 (x, y) obeys, for each ε > 0,

|E1 (x, y)| ≤ Cε |x − y| −(n−3+ε), (2.8.188)

|∇x E1 (x, y)| ≤ Cε |x − y| −(n−2+ε), (2.8.189)

|∇y E1 (x, y)| ≤ Cε |x − y| −(n−2+ε), (2.8.190)

while the top singularity depends exclusively on the top coefficients of L and is
explicitly given by
∫   −1
−n
E0 (y, z) = −(2π) ξ j ξk A jk (y) ei z,ξ  dξ, (2.8.191)
j,k

 αβ
where we have set A jk := a jk α,β . As such, if n ≥ 3 it follows that

E0 (y, z) is even and smooth in z ∈ Rn \ {0}, and is

homogeneous of degree 2 − n in the variable z, (2.8.192)
with 𝒞1 dependence in the variable y.

34 the argument in [119] is based on microlocal analysis: symbol decomposition techniques, local
elliptic regularity, pseudodifferential operators
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 587

In particular, this implies that for n ≥ 3 we have

E0 (y, ρz) = | ρ| −(n−2) E0 (z, y) for any ρ ∈ R and z ∈ Rn \ {0}, (2.8.193)

and

(∇zm ∇y E0 )(y, z) = O |z| −(n−2+m) as z −→ 0,
(2.8.194)
uniformly in y, for every m ∈ N0 and  ∈ {0, 1}.

The singularity in the top term E0 (y, z) when n = 2 is at most of logarithmic type,
i.e., |E0 (y, z)| ≤ C| ln |z||, with 𝒞1 dependence in the variable y.
Let us also remark here that if ℱz→ξ denotes the Fourier transform turning a
function in the variable z into a function in the variable ξ, then (2.8.191) gives
  −1
ℱz→ξ E0 (y, ·) (ξ) = − ξ j ξk A jk (y) (2.8.195)
j,k

which, in turn, shows (cf. item (6) of Theorem 1.4.2) that

for each y fixed, the matrix-valued function E0 (y, ·) is a
n
(2.8.196)
fundamental solution for the system A jk (y)∂j ∂k .
j,k=1

Based on (2.8.195) we may also check that each row E0α• (y, ·) of E0 (y, ·) satisfies the
PDE
n
Ajk (y)∂z j ∂zk E0α• (y, z) = 0 for z ∈ Rn \ {0}. (2.8.197)
j,k=1

In this connection, it has been noted in [119] that (with the operator L acting on
columns)

Lx E0 (y, x − y) = δy (x) + R(x, y) where, if V := (d αβ )α,β,

(2.8.198)
|R(x, y)| ≤ C|x − y| −(n−1) + C|V(x)||x − y| −(n−2),

assuming n ≥ 3. Also, from (2.8.187), (2.8.188), and the above discussion pertaining
to the nature of the singularity in E0 (y, z) it follows that for each x, y ∈ M with
x  y we have

C dist(x, y)−(n−2) if n ≥ 3,
|E(x, y)| ≤ (2.8.199)
Cα dist(x, y)−(1−α) ∀α ∈ (0, 1) if n = 2.

More can be said if the leading coefficients of L exhibit more smoothness. Specif-
ically, [119, Proposition 2.6, p. 19] gives that for any two distinct points x, y ∈ M
we have
588 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
⨏ 1/p
|∇x ∇x E1 (w, y)| p dw ≤ Cp, ε |x − y| −(n−1+ε)
B(x, |x−y |/2)
(2.8.200)
for each exponent p ∈ (1, r) and each ε > 0, provided
αβ αβ
a jk ∈ 𝒞1+γ, b j ∈ L ∞, dαβ ∈ L r with γ > 0 and r > n.

Let us also note here that under the stronger regularity assumptions we can further
augment the list of properties in (2.8.183)-(2.8.188). Specifically,

|∇x ∇y E1 (x, y)| ≤ Cε |x − y| −(n−1+ε) for each ε > 0, and

p 
E(·, y) ∈ L2,loc M \ {y}, E ⊗ Fy , ∀y ∈ M, ∀p ∈ (1, r), (2.8.201)
αβ αβ
if a jk ∈ 𝒞1+γ, bj ∈ L1r , d αβ ∈ Lr with γ > 0 and r > n.

These follow from [119, Proposition 2.8, p. 21] and, respectively, (2.8.181) together
with [119, Proposition 2.7, p. 20]. Finally, it has been remarked in [119, (2.53)-
(2.54), p. 19] that the following pointwise estimate (which should be compared with
(2.8.200)) holds:

|∇x ∇x E1 (x, y)| ≤ Cε |x − y| −(n−1+ε) for each ε > 0 whenever

αβ αβ
(2.8.202)
a jk ∈ 𝒞1+γ, bj ∈ L1r , d αβ ∈ 𝒞γ with γ > 0 and r > n.

After this preamble, we are ready to turn to the proof of Theorem 2.8.4 in earnest.
First, in view of (2.8.184), the extension of the single layer operator proposed in
(2.8.106) is meaningful, and the operator so defined is linear and bounded in the
context
p
𝒮 : L−1 (∂Ω, σg ) ⊗ F −→ 𝒞0loc (Ω) ⊗ E. (2.8.203)
In particular, (2.8.108) holds. The claim in (2.8.109) also follows from (2.8.184).
p
Since the embedding of L 1 (∂Ω, σg ) ⊗ F into L−1 (∂Ω, σg ) ⊗ F is dense (cf. [113,
(11.8.4)]), and since (2.8.101) and the fact that E(x, y) is the Schwartz kernel of
L −1 imply L(𝒮 f ) = 0 in Ω for each f ∈ L 1 (∂Ω, σg ) ⊗ F , we then conclude from
(2.8.203) that (2.8.107) holds as well.
Next, (2.8.199) shows that the single layer (2.8.102) behaves like a fractional
integral operator of order 1 if n ≥ 3, and of any order α ∈ (0, 1) if n = 2.
Bearing this in mind, an argument based on the Fractional Integration Theorem (see
Proposition 2.5.39) then establishes (2.8.114)-(2.8.116). As a byproduct, we also
have (2.8.110). The trace formula in (2.8.126) is also a consequence of the above
observation and Proposition 2.5.39.
Moving on, by working in local coordinates and then reassembling the pieces by
using a smooth partition of unity, we can construct

Q0 ∈ ØPC 1 Scl−2 (F , E) (2.8.204)

2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 589

such that Sym(Q0 ; ξ) = Sym(L; ξ)−1 and the remainder R := Q0 − L −1 is relatively

tame, i.e., its Schwartz kernel E1 (x, y) satisfies (2.8.188)-(2.8.190). This happens
under the assumptions made in (2.8.98). Under stronger regularity assumptions,
E1 (x, y) satisfies additional properties, as described in (2.8.200)-(2.8.202).
The idea is now that if we employ the notation k Q0 for the Schwartz kernel of
Q0 , then (Ix ⊗ Py )k Q0 (x, y) becomes the Schwartz kernel of Q := Q0 P . Since this
satisfies
Q ∈ ØPC 0 Scl−1 (F , E) and Q has an odd principal
(2.8.205)
symbol in the cotangent variable ξ ∈ T ∗ M,

the desired conclusions for the contribution of (Ix ⊗ Py )k Q0 (x, y) follow directly
from Theorem 2.8.2, while the contribution due to (Ix ⊗ Py )E1 (x, y) is, because of
the aforementioned estimates, amenable to a much more straightforward analysis.
x ⊗ Iy )E(x, y).
Essentially, a similar analysis applies to the kernel (P
This principle applies to a sizable portion of the statement of the theorem, in-
cluding all claims in item (5)-(6), the claims in item (8) up to, and including, the
jump-formula (2.8.153), and (2.8.135)-(2.8.136). In turn, these also cover the claims
made in (2.8.127), (2.8.112)-(2.8.113), and also (2.8.111) (bearing in mind the struc-
ture of the Sobolev spaces of negative order on ∂Ω; cf. [113, (11.8.7)-(11.8.8)]).
Let us next exemplify the manner in which the aforementioned principle is im-
plemented in relation to the various L p -square function estimates recorded in the
statement of the theorem. First, having fixed some p ∈ (1, ∞), consider the task
of proving (2.8.147), working under the stronger regularity assumptions made in
(2.8.145). These imply that the error term E1 (x, y) satisfies the estimate recorded in
the first line of (2.8.201). Denote by p ∈ (1, ∞) the Hölder conjugate exponent of
p and choose some ε ∈ (0, p1 ). Also, pick a pair of arbitrary distinct points x, y, and
abbreviate ρ := |x − y|/2. Since |w − y| ≈ |x − y|, uniformly for w ∈ B(x, ρ), based
on (2.8.201) we may estimate
∫ 1/p ∫ 1/p
|∇x ∇y E1 (w, y)| dw
p
≤C |x − y| −p(n−1+ε) dw
B(x, ρ) B(x, ρ)

= C|x − y| −(n−1+ε) ρn/p


= C|x − y| 1−ε−n/p . (2.8.206)

Thanks to (2.8.206), [112, Lemma 8.7.11] applies (with a := p−1) to the contribution
to the left-hand side of (2.8.147) coming from ∇x ∇y E1 (x, y), whereas the contri-
bution to the left-hand side of (2.8.147) coming from ∇y E0 (y, x − y) g(y)−1/2 is,
thanks to (2.8.192), directly amenable to the L p -square function estimate from The-
orem 2.5.38 (alternatively, we may invoke Theorem 2.8.2). Granted these and keep-
ing in mind (2.8.187), the L p -square function estimate (2.8.147) follows. A virtually
identical argument works in the case of (2.8.118). The L p -square function estimate in
(2.8.121) is proved using the same decomposition of the fundamental solution, upon
observing that Theorem 2.5.38 applies to the expression ∇x E0 (y, x − y) g(y)−1/2
590 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

while [112, Lemma 8.7.11] can now be used directly, thanks to (2.8.200), to control
the contribution from ∇x ∇x E1 (x, y).
Once (2.8.147) has been established, the L p -square function estimate in (2.8.118)
follows as a special case, keeping in mind the structure of the Sobolev spaces
of negative order on ∂Ω (cf. [113, (11.8.7)-(11.8.8)]). In turn, (2.8.119) may be
deduced from (2.8.118) by the same type of argument used in the proof of (2.4.37).
This time, in place of (2.4.105), we use the fact that

|∇x ∇y E(x, y)| ≤ C dist(x, y)−n, (2.8.207)

which follows from (2.8.187), (2.8.192), and (2.8.201).

Likewise, (2.8.121) directly implies the L p -square function estimate claimed in
(2.8.121) which, in turn, further implies (2.8.123) by reasoning as in the proof of
(2.4.37) and keeping in mind that under the assumptions in (2.8.122) we have

|∇x ∇x E(x, y)| ≤ C dist(x, y)−n, (2.8.208)

as may be seen from (2.8.187), (2.8.192), and (2.8.202).

Moving on, if Ω is actually a Lipschitz domain then the mapping properties
claimed in (2.8.125) become consequences of the L p -square function estimates in
(2.8.118)-(2.8.121), [113, Lemma 9.2.40], and known lifting results on the Besov
scale in Lipschitz domains (cf. [73, Proposition 2.18(b), p. 173]). Similarly, if Ω is
Lipschitz, then the claims pertaining to the operator in (2.8.124) are direct conse-
quences of the L p -square function estimate (2.8.147) and [113, Lemma 9.2.40].
Next, the first claims in item (8) may be established by reasoning as in the proof
of item (3) in Theorem 2.8.2, bearing in mind that

(I ⊗ P)E (y, x) = (P ⊗ I)E L  (x, y), (2.8.209)

which may be checked directly from definitions. A similar argument also allows us
to identify the transpose of the single layer as claimed in (2.8.131).
Regarding the single layer on Sobolev spaces, from (2.8.112), (2.8.110), (2.8.127),
the existence of the pointwise nontangential trace in (2.8.126), and [112, Proposi-
tion 1.12.9] it follows that the operator (2.8.128) is well-defined, linear, and bounded.
Via duality, the same conclusions are valid for (2.8.129).
Going further, the claims pertaining to the operator (2.8.162) (under the as-
sumption (2.8.161)) follow from Proposition 2.8.3, (2.8.107), (2.8.111), and [112,
(8.6.51)], while the claims pertaining to the operator (2.8.164) (under the assumption
(2.8.163)) follow from Proposition 2.8.3, (2.8.137), (2.8.135), and [112, (8.6.51)].
Consider now (2.8.148). Thanks to the L p -square function estimate recorded in
(2.8.147), the same type of argument as in the proof of (2.4.37) works virtually
verbatim in the current context; the most notable novelty is that in place of (2.4.105)
we presently use (2.8.207).
The Carleson measure estimate (2.8.149) is, of course, a particular case of
(2.8.148). Also, the L p -square function estimate (2.8.119) for the single layer op-
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 591

erator is a consequence of (2.8.148), bearing in mind the structure of the Sobolev

spaces of negative order on ∂Ω (cf. [113, (11.8.7)-(11.8.8)]).
As regards formula (2.8.130), fix a pair of continuous vector fields X, Y , along
with a function f ∈ L p (∂Ω, σg ) ⊗ F for some p ∈ (1, ∞). From the jump-formula
(2.8.153) and [112, (1.12.52)] we know that for every continuous vector field Z we
have
κ−n.t. ∫
  −1
% &
∇Z 𝒮 f  = 2 νg (Z) Sym(L; νg ) f +P.V.
1
(∇ Z ⊗ I)E, f F dσg (2.8.210)
∂Ω ∂Ω

at σ-a.e. point on ∂Ω. Then (2.8.130) follows from this (used with Z = X and Z = Y ),
and [112, (1.12.101)], keeping in mind [112, (1.12.140)].
The first claim in item (10), pertaining to the operator (2.8.162), is a consequence
of Proposition 2.8.3 whose applicability in the current setting is ensured by (2.8.107)
and the L p -square function estimate (2.8.118). Similarly, the second claim in item
(10), regarding the operator (2.8.164), is once again implied by Proposition 2.8.3,
now bearing in mind (2.8.137) and the L p -square function estimate (2.8.147). 

Finally, we elaborate on the definition and nature of double layer potential oper-
ators associated with various (quasi-)factorizations of weakly elliptic second order
operators acting between sections of a vector bundle over a given Riemannian man-
p
ifold. The reader is reminded that Ls stands for the L p -based Sobolev space of
p
(smoothness) order s ∈ R, and that Ls,loc denotes the local version of this scale. As
in the past, we shall also abbreviate H s := Ls2 .

Hypothesis 2.8.5 (Analytic Assumptions and Notation) Let M denote a com-

pact, oriented, boundaryless manifold of class 𝒞2 and real dimension n ∈ N (where
n ≥ 2), equipped with a Riemannian metric tensor of class 𝒞1 . Consider a second
order differential operator L : E → E acting between sections of a given Hermitian
vector bundle E → M, satisfying the following properties:
(i) One has the quasi-factorization
 + W,
L = DD (2.8.211)
 D are first-order differential operators
where D,

D : E −→ G,  : G −→ E
D (2.8.212)

for some Hermitian vector bundle G → M which, in any local coordinate chart
U on M and with respect to local trivializations of E, G, may be represented as

Du(x) = A j (x)∂j u(x) + B(x)u(x) where, for some r > n,
j
  (2.8.213)
A j ∈ L2,loc U, CrankG×rankE , B ∈ L1,loc
r r U, CrankG×rankE ,

and
592 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

  
Dw(x) = A j (x)∂ j w(x) + B(x)w(x) where, for some r > n,
j
  (2.8.214)
j ∈ L r  ∈ Lr
2,loc U, C
rankE×rankG ,
1,loc U, C
A B rankE×rankG ,

while

W ∈ Hom E, E has coefficients in L r for some r > n. (2.8.215)

(ii) The operator L is weakly elliptic, in the sense that its principal symbol has the
property that

Sym(L; ξ) : E → E is invertible for each ξ ∈ T ∗ M \ 0. (2.8.216)

(iii) The operator L is invertible as a mapping

L : H 1 (M) ⊗ E −→ H −1 (M) ⊗ E. (2.8.217)

In view of the quasi-factorization (2.8.211) and the subsequent assumptions on

 D, it follows that L may be locally written as
the operators D,
  
Lu(x) = ∂j A jk (x)∂k u(x) + B j (x)∂j u(x) + V(x)u(x), (2.8.218)
j,k j

with coefficients
j Ak ∈ 𝒞1+γ, for some γ > 0,
A jk := A loc
  j B + BA
 j ∈ Lr ,
B j := − (∂k Ak )A j + A 1,loc (2.8.219)
k
  + W ∈ Lr .
V := A j ∂ j B + BB loc
j

If E L denotes the Schwartz kernel of the inverse L −1 of L in (2.8.217) then


E L ∈ D (M × M, E ⊗ E) ∩ 𝒞1+θ loc M × M \ diag, E ⊗ E (2.8.220)

for some θ > 0; see (2.8.184) in this regard.

Associated with the quasi-factorization (2.8.211), introduce the family of first
order differential operators indexed by sections in the cotangent bundle

∂ξL := (−i) Sym D;  ξ D, ξ ∈ T ∗ M. (2.8.221)

Whenever L is as in Hypothesis 2.8.5, it follows that L  , the real transposed of

L, also satisfies all conditions in Hypothesis 2.8.5. In particular, we now have the
quasi-factorization
 + W  .
L  = D D (2.8.222)
  −1
We shall denote by E L  the Schwartz kernel of the inverse L of the operator
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 593

L  : H 1 (M) ⊗ E −→ H −1 (M) ⊗ E (2.8.223)

which continues to enjoy a regularity property analogous to (2.8.220). Also, asso-

ciated with the quasi-factorization (2.8.222), we introduce the family of first order
differential operators indexed by sections in the cotangent bundle
 
,
∂ξL := (−i) Sym D ; ξ D ξ ∈ T ∗ M. (2.8.224)

We now turn to assumptions of a geometric nature.

Hypothesis 2.8.6 (Geometric Assumptions and Notation) Let M denote a com-
pact, oriented, boundaryless manifold of class 𝒞2 and real dimension n ∈ N (where
n ≥ 2), equipped with a Riemannian metric tensor g of class 𝒞1 . Assume Ω ⊂ M is
an Ahlfors regular domain. With Hgn−1 denoting the (n − 1)-dimensional Hausdorff
measure associated with the metric g on M, abbreviate σg := Hgn−1 ∂Ω. Also,
denote by νg ∈ T ∗ M the geometric measure theoretic outward unit conormal to Ω
(which is well defined at σg -a.e. point on ∂Ω). Finally, given any Hermitian vector
bundle E → M and any exponent p ∈ (0, ∞], the space of all σg -measurable
sections f : ∂Ω → E which are p-th power integrable with respect to the surface
measure σg is denoted by L p (∂Ω, σg ) ⊗ E.
We are now prepared to introduce the boundary-to-domain double layer potential
operator and its principal value version, as indicated below.
Definition 2.8.7 (Double Layers) Assume Hypotheses 2.8.5-2.8.6. In this context,
define the double layer associated with the quasi-factorization of L in (2.8.211) as
the integral operator sending any f ∈ L 1 (∂Ω, σg ) ⊗ E into the function defined at
each x ∈ M \ ∂Ω by
∫ # $
 
D L f (x) := y E L (x, y), f (y)
Ix ⊗ (−i) Sym D ; νg (y) D dσg (y)
Ey
∂Ω
∫ # $
 
= Ix ⊗ ∂νLg (y) E L (x, y), f (y) dσg (y), (2.8.225)
Ey
∂Ω

(where I denotes the identity, here acting in the variable x, etc.). In addition, consider
its principal value version on ∂Ω (in the sense of Definition 2.8.1) acting on each
f ∈ L 1 (∂Ω, σg ) ⊗ E according to
∫ # $
 
K L f (x) := P.V. Ix ⊗ (−i) Sym D ; νg (y) Dy E L (x, y), f (y) dσg (y)
Ey
∂Ω
∫ # $
 
= P.V. Ix ⊗ ∂νLg (y) E L (x, y), f (y) dσg (y), (2.8.226)
Ey
∂Ω

at σg -a.e. point x ∈ ∂Ω. Finally, define the singular integral operator acting on each
function f ∈ L 1 (∂Ω, σg ) ⊗ E according to
594 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫ # $
 
K L# f (x) := P.V.  νg (x) Dx ⊗ Iy E L (x, y), f (y)
(−i)Sym D; dσg (y)
Ey
∂Ω
∫ # $

= P.V. ∂νLg (x) ⊗ Ix E L (x, y), f (y) dσg (y), (2.8.227)
Ey
∂Ω

at σg -a.e. point x ∈ ∂Ω.

The remark below further clarifies some nuances in the above definition.

Remark 2.8.8 In the context of Hypothesis 2.8.5, the given second-order operator
L does not determine uniquely the first-order operators D,  D. Indeed, L can have
multiple (infinitely many, actually) quasi-factorizations as in (2.8.211). This being
said, the formulas of the double layer operators (both boundary-to-boundary and
boundary-to-domain) in Definition 2.8.7 strongly depend on the choice of D  and D.
As such, in place of D L , K L , K L (which seems to suggest that L alone determines
#

said operators) it would be perhaps more appropriate to write DD,D  , KD,D

#
 , KD,D

,
respectively. Same comment applies in relation to the conormal derivative from
(2.8.221). The reader may want to keep in mind this aspect throughout this section.

In the subsequent discussion, adopt the stronger hypothesis that Ω is actually a

UR domain. Then the boundary-to-domain double layer D L falls under the general
template of the operator A discussed in (2.8.103), corresponding to the special case
when the first-order differential operator P is given by35

P := (−i) Sym D ; νg D  . (2.8.228)

Also, the principal-value double layer K L is a particular embodiment of (2.8.105),

corresponding to the choice of P given in (2.8.228), while the singular integral
operator K L# is a particular case of (2.8.152) corresponding to the choice

 := (−i)Sym( D;
P  νg )D. (2.8.229)

Granted this interpretation, Theorem 2.8.4 then yields a wealth of properties of

our brand of double layer operators. For example, under Hypotheses 2.8.5-2.8.6 and
assuming that Ω is actually a UR domain, it follows that

L(D L f ) = 0 in Ω, for each f ∈ L 1 (∂Ω, σg ) ⊗ E. (2.8.230)

and for each integrability exponent p ∈ (1, ∞) and each aperture parameter κ > 0
there exists a constant C ∈ (0, ∞) with the property that

35 Strictly speaking, P given in (2.8.228) does not have continuous coefficients due to the
presence
 of νg but, working in local coordinates, we may expand (−i) Sym D ; νg =
(ν
j g j) (−i) Sym D  ; dx j and focus on the first-order differential operators Sym D  ; dx j
as playing the role of P, while the components (νg ) j are regarded as pointwise multipliers on
the density f .
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 595
 
Nκ (D L f ) ≤ C f  L p (∂Ω,σg )⊗ E , ∀ f ∈ L p (∂Ω, σg ) ⊗ E,
L p (∂Ω,σg )
(2.8.231)
Also, for each p ∈ (1, ∞), the operators

K L : L p (∂Ω, σg ) ⊗ E −→ L p (∂Ω, σg ) ⊗ E,
(2.8.232)
K L# : L p (∂Ω, σg ) ⊗ E −→ L p (∂Ω, σg ) ⊗ E,

are well defined, linear, and bounded. In addition, for each f ∈ L 1 (∂Ω, σg ) ⊗ E the
following jump-formula holds:
κ−n.t. 

(D L f ) = 12 I + K L f at σg -a.e. point on ∂Ω. (2.8.233)
∂Ω

Indeed, (2.8.233) follows from the jump-formula (2.8.144) written for P as in

(2.8.228), upon noting that
−1 
2 iSym(L; νg ) Sym(P; νg )
1

    
 ; νg
= 12 iSym(L; νg )−1 (−i) Sym D ; νg Sym D
 
 νg Sym D; νg
= 12 iSym(L; νg )−1 (−i) Sym D;

= 1  νg
Sym(L; νg )−1 Sym DD;
2

= 1
2 Sym(L; νg )−1 Sym L; νg = 12 I, (2.8.234)

thanks to [112, (1.7.17)] and (2.8.211). In summary, the boundary-to-domain double

layer is a mechanism for producing lots of null-solutions of the operator L, all of
which have quantitative control of their nontangential maximal functions, and have
nontangential pointwise boundary traces given explicitly (in terms of the principal
value double layer and a constant jump term).
It is also of significance to remark that, in the same setting as above, for each
 as in (2.8.229) presently
f ∈ L 1 (∂Ω, σg ) ⊗ E the jump-formula (2.8.153) used with P
gives

   κ−n.t.
 νg ) D𝒮 L f 
∂νLg 𝒮 L f := (−i)Sym( D;
∂Ω

 νg ) Sym(D; νg ) Sym(L; νg )−1 f + K L# f

= − 12 i(−i)Sym( D;
 νg ) Sym(L; νg )−1 f + K L# f
= − 12 Sym( DD;

= − 12 Sym(L; νg ) Sym(L; νg )−1 f + K L# f


= − 12 I + K L# f at σg -a.e. point on ∂Ω. (2.8.235)
596 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Our next comment has to do with K L#  , the principal value singular integral oper-
ator associated with L  much as K L# has been associated with L in Definition 2.8.7,
this time making use of the quasi-factorization (2.8.222) for L  . Specifically, for
each f ∈ L 1 (∂Ω, σg ) ⊗ E we have
∫ # $
 
K L  f (x) = P.V.
# 
(−i)Sym D ; νg (x) D x ⊗ Iy E L  (x, y), f (y) dσg (y)
Ey
∂Ω
∫ # $
 
= P.V. ∂νLg (x) ⊗ Ix E L  (x, y), f (y) dσg (y), (2.8.236)
Ey
∂Ω

at σg -a.e. point x ∈ ∂Ω. Then, granted the same analytic and geometric conditions
as before, item (9) of Theorem 2.8.4 presently implies
 
K L = K L#  (2.8.237)

in the precise sense that, for each p, p ∈ (1, ∞) satisfying 1/p + 1/p = 1,

the (real) transpose of K L acting on L p (∂Ω, σg ) ⊗ E is the

 (2.8.238)
operator K L#  acting on L p (∂Ω, σg ) ⊗ E (see (2.8.232)).

We also want to comment that, in the same setting as above, for any p ∈ (1, ∞)
the operator K L originally acting from L p (∂Ω, σg ) ⊗ E into itself leaves the Sobolev
p
space L1 (∂Ω, σg ) ⊗ E invariant, and the following map is well defined, linear, and
bounded:
p p
K L : L1 (∂Ω, σg ) ⊗ E −→ L1 (∂Ω, σg ) ⊗ E. (2.8.239)
See [118, Theorem 4.9, p. 155], as well as [118, Theorem 4.10, p. 169] and [128],
for related results.
Here we give a proof of the aforementioned mapping property of the principal
value double layer on boundary Sobolev spaces in the case when all differentiability
structures related to the Riemannian manifold, the Hermitian vector bundles, and
the coefficients of the differential operator L are 𝒞∞ . Specifically, (2.8.239) is a
consequence of the jump-formula (2.8.233), [112, Proposition 1.12.9], and Theo-
rem 2.8.10 stated a little further below. In the latter theorem we consider the double
p
layer D L acting from L p -based Sobolev spaces of order one, i.e., L1 (∂Ω, σg ) ⊗ E
with 1 < p < ∞, where Ω is a UR domain of a smooth Riemannian manifold,
and show that the covariant derivative of the double layer is well-behaved near the
boundary. We begin with a few preliminary results, setting the stage for the proof of
Theorem 2.8.10.
First, we record a useful fact regarding the nature of the principal symbol of
a commutator. Work on a smooth manifold M, of real dimension n. Consider a
classical pseudodifferential operator Ψ ∈ OPSclm , with m arbitrary, acting between
two vector bundles over M, and denote the principal symbol of Ψ by

q(x, ξ) := Sym(Ψ; ξ), ∀ξ ∈ Tx∗ M. (2.8.240)

2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 597

Also, having fixed a smooth vector field X = j X j ∂j ∈ TM, denote the principal
symbol of the commutator [∇X , Ψ] := ∇X Ψ − Ψ∇X by

p(x, ξ) := Sym [∇X , Ψ]; ξ , ∀ξ ∈ Tx∗ M. (2.8.241)

Then the discussion in [176, p. 13] shows that


p(x, ξ) = i X j {ξ j , q(x, ξ)}, ∀ξ ∈ Tx∗ M, (2.8.242)
j

where, in general, the Poisson bracket of p1 (x, ξ) and p2 (x, ξ) is defined (cf., e.g.,
[176, (3.26), p. 13]) by

n

{p1 (x, ξ), p2 (x, ξ)} := ∂ξ p1 ∂x p2 − ∂x p1 ∂ξ p2 , ∀ξ ∈ Tx∗ M. (2.8.243)
=1

Next, let E, F , and G be Hermitian vector bundles of class 𝒞∞ over a given

Riemannian manifold M of class 𝒞∞ . We make the following convention:

given T : D(M, E) → D (M, F ) linear and continuous,

(2.8.244)
we agree to let kT (x, y) denote the Schwartz kernel of T .

Then, if T is as in (2.8.244), for any differential operators P : F → G and Q : G → E

we have (bearing in mind the conventions made in (2.8.90)-(2.8.94))

k PT (x, y) = (Px ⊗ Iy )kT (x, y) in D  M × M, G ⊗ E , (2.8.245)

kT Q (x, y) = (Ix ⊗ Q
y )kT (x, y) in D M × M, F ⊗ G , (2.8.246)


k PT Q (x, y) = (Px ⊗ Q
y )kT (x, y) in D M × M, G ⊗ G . (2.8.247)

The following lemma plays a key role in the proof of Theorem 2.8.10.

Lemma 2.8.9 Let (M, g) be a compact Riemannian manifold of class 𝒞∞ , having

(real) dimension n. Suppose E, G → M are two Hermitian vector bundles of class
𝒞∞ equipped with connections

∇ E : TM × E −→ E, ∇ G : TM × G −→ G. (2.8.248)

Assume L : E → E is a weakly elliptic second-order partial differential operator

with smooth coefficients such that

L : H 1 (M) ⊗ E −→ H −1 (M) ⊗ E is an isomorphism. (2.8.249)

In addition, suppose L has the quasi-factorization

 +W
L = DD (2.8.250)
598 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

for some first-order differential operators with smooth coefficients

D : E −→ G,  : G −→ E,
D (2.8.251)

and a homomorphism W ∈ Hom (E, E). Let E L (x, y) denote the Schwartz kernel of
L −1 .
Next, let Ω ⊂ M be UR domain with outward unit conormal νg : ∂Ω → T ∗ M
and surface measure σg := Hgn−1 ∂Ω. Choose an arbitrary vector field X ∈ TM,
and consider the kernel function
  
y E L (x, y)
ωX (x, y) := ∇XE x ⊗ (−i)Sym D ; νg (y) D
  
− Ix ⊗ (−i)Sym D ; νg (y) (∇XG ) 
y Dy E L (x, y). (2.8.252)

Use this to define an integral operator 𝒦X mapping any integrable section f in

L 1 (∂Ω, σg ) ⊗ E into the section 𝒦X f : Ω → E defined by
∫
% &
(𝒦X f )(x) := ωX (x, y), f (y) E dσg (y), ∀x ∈ Ω. (2.8.253)
∂Ω

Finally, fix an aperture parameter κ > 0.

Then for each f ∈ L 1 (∂Ω, σg ) ⊗ E the nontangential boundary trace
κ−n.t.

(𝒦X f ) exists σg -a.e. on ∂Ω (2.8.254)
∂Ω

and, in fact, using the abbreviation [∇X , D] for the first-order differential operator
∇XG D − D∇XE , at σg -a.e. point x ∈ ∂Ω one has
 κ−n.t.  1 
(𝒦X f )∂Ω (x) = − Sym(L; νg )−1 Sym( D;
 νg )Sym [∇X , D]; νg f (x)
2
∫
% &
+ P.V. ωX (x, y), f (y) E dσg (y). (2.8.255)
∂Ω

Also, for each p ∈ (1, ∞) there exists C = C(Ω, L, X, κ, p) ∈ (0, ∞) such that

Nκ (𝒦X f ) L p (∂Ω,σg ) ≤ C f  L p (∂Ω,σg )⊗ E , ∀ f ∈ L p (∂Ω, σg ) ⊗ E. (2.8.256)

Proof First we introduce some notation by rewriting ωX (x, y) as

  
ωX (x, y) = Ix ⊗ (−i)Sym D ; νg (y) ω X (x, y), (2.8.257)

where
 
y E L (x, y) − Ix ⊗ (∇ G )
X (x, y) := ∇XE x ⊗ D
ω 
X y Dy E L (x, y). (2.8.258)
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 599

Further defining 
y E L (x, y),
Θ(x, y) := Ix ⊗ D (2.8.259)
allows us to recast ω
X (x, y) as
 
X (x, y) = ∇XE x ⊗ Iy Θ(x, y) − Ix ⊗ (∇XG )
ω y Θ(x, y). (2.8.260)

The idea is to now interpret the expressions in (2.8.259) and (2.8.260) as Schwartz
kernels. First recall that
E L (x, y) = k L −1 (x, y), (2.8.261)
 in (2.8.246)
where k L −1 (x, y) denotes the Schwartz kernel of L −1 . Taking Q := D
then yields
Θ(x, y) = k L −1 D (x, y). (2.8.262)
This suggests considering the composition

Ψ := L −1 D. (2.8.263)

Since the differential geometric structures of M, E, G and the coefficients of L are

all 𝒞∞ , we have
L −1 ∈ OPScl−2 and D ∈ OPS 1 , (2.8.264)
cl

which implies
 ∈ OPS −1 with Θ(x, y) = k Ψ (x, y).
Ψ = L −1 D (2.8.265)
cl

Moreover,
 ξ) = Sym(L; ξ)−1 Sym( D;
Sym(Ψ; ξ) = Sym(L −1 ; ξ)Sym( D;  ξ). (2.8.266)

Since L is a second-order operator and D is first-order, their principal symbols are

even and odd in ξ, respectively. Hence,

Sym(Ψ; ξ) is odd in ξ. (2.8.267)

Moving on, by (2.8.245) it follows that

 E
∇Xx ⊗ Iy Θ(x, y) = k ∇E Ψ (x, y). (2.8.268)
X

Also,

Ix ⊗ (∇XG )
y Θ(x, y) = k Ψ∇G (x, y), (2.8.269)
X

which follows from (2.8.246) with Q := (∇XG ) . Then by (2.8.260), (2.8.268), and
(2.8.269) it follows that
600 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

ω
X (x, y) = k ∇E Ψ (x, y) − k Ψ∇G (x, y) = k ∇E Ψ−Ψ∇G (x, y)
X X X X

= k[∇ X ,Ψ] (x, y). (2.8.270)

Going further, if q(x, ξ) := Sym(Ψ; ξ), then based on (2.8.242) and (2.8.243) we
have
 
Sym [∇X , Ψ]; ξ = i X j {ξ j , q(x, ξ)}
j
 
n

=i Xj (∂ξ ξ j )(∂x q)(x, ξ) − (∂x ξ j )(∂ξ q)(x, ξ)
j =1
 n 
=i Xj δ j (∂x q)(x, ξ) = i X j (∂x j q)(x, ξ)
j =1 j

= i∇Xx q(x, ξ) , ∀ξ ∈ Tx∗ M. (2.8.271)

Hence, comparing (2.8.240), (2.8.241), and (2.8.271) yields

 
Sym [∇X ; Ψ], ξ = i∇X Sym(Ψ; ξ) , ∀ξ ∈ T ∗ M. (2.8.272)

Recalling from (2.8.267) that Sym(Ψ; ξ) is odd in ξ, and also observing that the
covariant derivative along X which appears in (2.8.272) does not affect the parity of
Sym(Ψ; ξ) in the variable ξ, we then conclude that

Sym [∇X , Ψ]; ξ is odd in ξ. (2.8.273)

To proceed, recall (see, e.g., the discussion in [176, pp. 12-13]) that, in general,
given two classical pseudodifferential operators, P1, P2 of order r, such
that Sym(P1 ; ξ) = Sym(P2 ; ξ) for every cotangent variable ξ, it follows (2.8.274)
that P1 − P2 is a classical pseudodifferential operator of order ≤ r − 1.

We also note that since Sym(∇XE ; ξ) = iξ(X) I E and Sym(∇XG ; ξ) = iξ(X) I G by

[112, (1.12.52)], where I E is the identity on E and I G is the identity on G, we have
 
Sym(∇XE ; ξ)Sym(Ψ; ξ) = iξ(X)I E Sym(Ψ; ξ) = Sym(Ψ; ξ) iξ(X) I G

= Sym(Ψ; ξ)Sym(∇XG ; ξ), (2.8.275)

i.e., the symbols of Ψ intertwines the symbols of ∇XE and ∇XG . It then follows that

[∇X , Ψ] := ∇XE Ψ − Ψ∇XG ∈ OPScl−1, (2.8.276)

from (2.8.274), (2.8.275), plus the fact that both ∇XE Ψ and Ψ∇XG belong to OPScl0 (as
seen from (2.8.265)).
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 601

Next, fix p ∈ (1, ∞) and let f ∈ L p (∂Ω, σg ) ⊗ E be arbitrary. By (2.8.273) and

(2.8.276), we may apply (2.8.42) with Q := [∇X , Ψ] to conclude that the singular
integral operator defined by
∫
% &
(𝒦[∇ X ,Ψ] f )(x) := k[∇ X ,Ψ] (x, y), f (y) E y dσg (y), ∀x ∈ Ω, (2.8.277)
∂Ω

satisfies the nontangential maximal function estimate

Nκ (𝒦[∇ X ,Ψ] f ) L p (∂Ω,σg ) ≤ C f  L p (∂Ω,σg )⊗ E , (2.8.278)

for all f ∈ L p (∂Ω, σg ) ⊗ E, where the constant C ∈ (0, ∞) is independent of f .

Recalling (2.8.270), it follows from (2.8.278) that for each p ∈ (1, ∞) there exists a
constant C ∈ (0, ∞) such that
  
Nκ 𝒦 "X f  p ≤ C f  L p (∂Ω,σg )⊗ E , ∀ f ∈ L p (∂Ω, σg ) ⊗ E, (2.8.279)
L (∂Ω,σg )

where
∫
 % &
"X f (x) :=
𝒦 ω
X (x, y), f (y) dσg (y), ∀x ∈ Ω. (2.8.280)
Ey
∂Ω

In turn, one may deduce that the same type of nontangential maximal function esti-
mate, namely (2.8.256), for the singular integral operator 𝒦X defined in (2.8.253).
Indeed, since ωX (x, y) is of the form (2.8.257), we see that
∫ #  $

(𝒦X f )(x) = Ix ⊗ (−i)Sym D ; νg (y) ω X (x, y), f (y) dσg (y)
∂Ω Ey
∫ # $

= ω
X (x, y), iSym D; νg (y) f (y) dσg (y)
∂Ω Ey
  
= 𝒦"X iSym(D; νg ) f (x), ∀ x ∈ Ω. (2.8.281)

From this and (2.8.279) it follows that there exists C = C(Ω, X, L, κ, p) ∈ (0, ∞) with
the property that
   
 "X iSym(D; νg ) f 
Nκ (𝒦X f ) L p (∂Ω,σg ) = Nκ 𝒦 p L (∂Ω,σg )
 
≤ C iSym(D; νg ) f  L p (∂Ω,σg )⊗ E

≤ C f  L p (∂Ω,σg )⊗ E , (2.8.282)

for all f ∈ L p (∂Ω, σg ) ⊗ E, as desired. Finally, the analysis above combined with
(2.8.46) also implies that for each section f ∈ L 1 (∂Ω, σg ) ⊗ E the nontangential
κ−n.t.
trace (𝒦X f ) ∂Ω
exists, and at σg -a.e. point x ∈ ∂Ω we have
602 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 κ−n.t.  1 
(𝒦X f )∂Ω (x) = Sym [∇X , Ψ]; νg i Sym(D; νg ) f (x)
2i
∫
% &
+ P.V. ωX (x, y), f (y) E dσg (y)
∂Ω

1 
 νg )Sym [∇X , D]; νg f (x)
= − Sym(L; νg )−1 Sym( D;
2
∫
% &
+ P.V. ωX (x, y), f (y) E dσg (y), (2.8.283)
∂Ω

since
 
Sym [∇X , Ψ]; νg Sym(D; νg ) = Sym [∇X , Ψ]D; νg (2.8.284)
 
= Sym ∇X ΨD − Ψ∇X D; νg
 
 − L −1 D∇
= Sym ∇X L −1 DD  X D; νg
 
 X , D] + L −1 DD∇
= Sym ∇X − L −1 D[∇  X ; νg
 
 X , D]; νg
= Sym − L −1 D[∇

 νg )Sym [∇X , D]; νg ,
= −Sym(L; νg )−1 Sym( D;

 differ by a lower-order term. This

thanks to the fact that the operators L and DD
concludes the proof of Lemma 2.8.9. 

We are now ready to state and prove the result alluded to earlier, regarding size
estimates and boundary behavior for the boundary-to-domain double layer potential
operator acting from Sobolev spaces defined on the boundary of a UR domain in a
smooth Riemannian manifold.

Theorem 2.8.10 Let (M, g) be a compact Riemannian manifold of class 𝒞∞ . Sup-

pose E, G → M are Hermitian vector bundles of class 𝒞∞ equipped with connec-
tions
∇ E : TM × E −→ E, ∇ G : TM × G −→ G. (2.8.285)
Assume L : E → E is a weakly elliptic second-order partial differential operator
with smooth coefficients such that

L : H 1 (M) ⊗ E −→ H −1 (M) ⊗ E is an isomorphism. (2.8.286)

In addition, suppose L has the quasi-factorization

 +W
L = DD (2.8.287)
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 603

for some first-order differential operators with smooth coefficients

D : E −→ G,  : G −→ E,
D (2.8.288)

and some homomorphism W ∈ Hom(E, E). Let E L (x, y) denote the Schwartz kernel
of L −1 .
Next, let Ω ⊂ M be UR domain with outward unit conormal νg : ∂Ω → T ∗M and
surface measure σg := Hgn−1 ∂Ω. Finally, fix an integrability exponent p ∈ (1, ∞)
and an aperture parameter κ > 0.
p
Then for each f ∈ L1 (∂Ω, σg ) ⊗ E, the nontangential pointwise trace
κ−n.t.

(∇ E D L f ) exist at σg -a.e. point on ∂Ω. (2.8.289)
∂Ω

Moreover, there exists a constant C = C(Ω, L, p, κ) ∈ (0, ∞) such that

Nκ (∇ E D L f ) L p (∂Ω,σg ) ≤ C f  L p (∂Ω,σg )⊗ E , (2.8.290)

1

p
for all f ∈ L1 (∂Ω, σg ) ⊗ E.
p
Proof Let f ∈ L1 (∂Ω, σg ) ⊗ E be arbitrary. Also fix an arbitrary vector field
X ∈ TM. For any point x ∈ Ω, we have
∫ =  >
 
E
(∇X D L f )(x) = E 
∇Xx ⊗ (−i)Sym D ; νg (y) Dy E L (x, y), f (y) dσg (y)
∂Ω Ey

= (𝒦X f )(x) + (𝒦X f )(x), (2.8.291)

where 𝒦X is the singular integral operator defined by (2.8.253) in Lemma 2.8.9 and
𝒦X is the singular integral operator defined as
∫ =  >

(𝒦X f )(x) := 
Ix ⊗ (−i)Sym D ; νg (y) (∇XG ) 
y Dy E L (x, y), f (y) dσg (y)
Ey
∂Ω
(2.8.292)
at each point x ∈ Ω. Observing that

0 ≤ Nκ (∇XE D L f ) ≤ Nκ (𝒦X f ) + Nκ (𝒦X f ), (2.8.293)

one may infer

Nκ (∇XE D L f ) L p (∂Ω,σg ) ≤ Nκ (𝒦X f ) L p (∂Ω,σg )

+ Nκ (𝒦X f ) L p (∂Ω,σg ) . (2.8.294)

From Lemma 2.8.9, there exists a constant C = C(Ω, L, X, p, κ) ∈ (0, ∞) such that

Nκ (𝒦X f ) L p (∂Ω,σg ) ≤ C f  L p (∂Ω,σg )⊗ E ≤ C f  L p (∂Ω,σg )⊗ E , (2.8.295)

1
604 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
p
where the second inequality is due to L1 (∂Ω, σg ) ⊗ E → L p (∂Ω, σg ) ⊗ E. Fur-
thermore, (2.8.254) shows that at σg -a.e. point x ∈ ∂Ω we have
 κ−n.t.  1 
(𝒦X f )∂Ω (x) = − Sym(L; νg )−1 Sym( D;
 νg )Sym [∇X , D]; νg f (x)
2
∫
% &
+ P.V. ωX (x, y), f (y) E dσg (y), (2.8.296)
∂Ω

where the kernel ωX (x, y) is as in (2.8.252). Thus, it remains to prove

Nκ (𝒦X f ) L p (∂Ω,σg ) ≤ C f  L p (∂Ω,σg )⊗ E , (2.8.297)

1

for some constant C = C(Ω, L, X, p, κ) ∈ (0, ∞), and find a suitable formula for
κ−n.t.
(𝒦X f )∂Ω at σg -a.e. point on ∂Ω.
To achieve these goals, the integral kernel of 𝒦X requires a closer examination.
The delicate issue with the integral kernel of 𝒦X is that there are two derivatives on
E L (x, y), increasing its singularity from order n − 2 to order n, and thus making the
integral kernel of 𝒦X hypersingular (due to the dimension of ∂Ω being n − 1). Our
strategy involves using the quasi-factorization L  = D D  + W  and then applying
integration by parts on the boundary to transfer the burden of one of the derivatives
on E L (x, y) to the function f in the form of a tangential derivative. This strategy is
p
only effective because f ∈ L1 (∂Ω, σg ) ⊗ E, so it can take a tangential derivative
within the context of the nontangential maximal function estimate we are trying
to prove in (2.8.297) (as well as the subsequent boundary trace). With only one
remaining derivative on E L (x, y), the Calderón-Zygmund theory in Theorem 2.8.4
can be invoked to finish the job.
To handle 𝒦X f , for ease of writing abbreviate
    
Υ(x, y) := Ix ⊗ (−i)Sym D ; νg (y) (∇XG )
y +νg (X)(y)D  
y Dy E L (x, y) (2.8.298)

for each x ∈ Ω and y ∈ ∂Ω. To proceed in earnest, fix x ∈ Ω, add and subtract

∫ =  >
y E L (x, y), f (y)
Ix ⊗ νg (X)(y)Dy D dσg (y) (2.8.299)
∂Ω Ey

from (𝒦X f )(x), and then use the fact that D D

 = L  − W  to obtain
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 605
∫ =  >

(𝒦X f )(x) = Ix ⊗ (−i)Sym D ; νg (y) (∇XG ) 
y Dy E L (x, y), f (y) dσg (y)
∂Ω Ey
∫ = >
= Υ(x, y), f (y) dσg (y)
∂Ω Ey
∫ =  >
− y E L (x, y), f (y)
Ix ⊗ νg (X)(y)Dy D dσg (y)
∂Ω Ey
∫ = >
= Υ(x, y), f (y) dσg (y)
∂Ω Ey
∫ =  >
− Ix ⊗ νg (X)(y)Ly E L (x, y), f (y) dσg (y)
∂Ω Ey
∫ =  >
+ Ix ⊗ νg (X)(y)Wy E L (x, y), f (y) dσg (y)
∂Ω Ey

=: I1 (x) + I2 (x) + I3 (x). (2.8.300)

Next, observe that

∫ =  >
I2 (x) = − Ix ⊗ νg (X)(y)Ly E L (x, y), f (y) dσg (y)
∂Ω Ey
∫
% &
=− Ix ⊗ Ly E L (x, y), νg (X)(y) f (y) Ey
dσg (y)
∂Ω

= 0, (2.8.301)

where the third equality is a consequence of the fact that Ix ⊗ Ly E L (x, y) = 0 for
x ∈ Ω and y ∈ ∂Ω. Moving on, the term I3 (x) can be handled by rewriting
∫ =  >
I3 (x) = Ix ⊗ νg (X)(y)Wy E L (x, y), f (y) dσg (y)
∂Ω Ey
∫
% &
= E L (x, y), νg (X)(y)Wy f (y) Ey
dσg (y). (2.8.302)
∂Ω

Since the integral kernel of I3 (x) contains no derivatives on E L (x, y), it is weakly
singular and the term I3 (x) falls under the scope of Proposition 2.5.39 with
(ℬ f )(x) := I3 (x). As such,
 κ−n.t.  ∫
 % &
I3  (x) = E L (x, y) , νg (X)(y)Wy f (y) E y dσg (y) (2.8.303)
∂Ω ∂Ω

at σg -a.e. point x ∈ ∂Ω, and

606 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Nκ (I3 ) L p (∂Ω,σg ) ≤ C f  L p (∂Ω,σg )⊗ E ≤ C f  L p (∂Ω,σg )⊗ E , (2.8.304)

1

p
where the fact that L1 (∂Ω, σg ) ⊗ E → L p (∂Ω, σg ) ⊗ E is a continuous embedding
is used the last inequality.
To treat the term I1 (x), in a first stage we shall use integration by parts on
the boundary, as described in [112, Corollary 1.12.12], in which we apply [112,
(1.12.132)] with P := D , the vector bundle G in place of F , and with the mapping
y → (Ix ⊗ D y )E L (x, ·) playing the role of the test function36 ϕ. In this regard,
bring in the tangential derivative operator ∂τQ,P defined as in [112, (1.12.131)] for
P := D (and with the vector bundle G in place of F ). The point of performing the
aforementioned integration by parts on the boundary is that the Calderón-Zygmund
theory from Theorem 2.8.4 may then be brought to bear, keeping in mind that, much
as in [112, (1.12.131)], the mapping
p
∂τQ,P : L1 (∂Ω, σg ) ⊗ E −→ L p (∂Ω, σg ) ⊗ E is continuous. (2.8.305)

Turning to specifics, we use [112, (1.12.132)] as explained earlier (recall the notation
in (2.8.298)) to write
∫ = >
I1 (x) = Υ(x, y), f (y) dσg (y)
∂Ω Ey
∫ # $

=− y E L (x, y), ∂τQ,P f (y)
Ix ⊗ D dσg (y)
∂Ω Ey
∫ =  >
  
+ Ix ⊗ iSym ∇XG D − D∇XE ; νg (y) D y E L (x, y), f (y) dσg (y)
∂Ω Ey

=: I1(a) (x) + I(b)

1 (x). (2.8.306)

Having decomposed I1 (x) as in (2.8.306), the Calderón-Zygmund theory (pertain-

ing to the operator A) from Theorem 2.8.4 can be invoked for the piece I1(a) (x),
since ∂τQ,P f ∈ L p (∂Ω, σg ) ⊗ E. This guarantees that there exists a constant
C = C(Ω, L, X, p, κ) ∈ (0, ∞) with the property that
  (a)   
Nκ I  p ≤ C ∂τQ,P f  L p (∂Ω,σg )⊗ E ≤ C f  L p (∂Ω,σg )⊗ E , (2.8.307)
1 L (∂Ω,σg ) 1

where the last inequality in (2.8.307) follows from (2.8.305), and (2.8.144) gives
 κ−n.t.  
 i
I1(a)   νg ) ∂τQ,P f (x)
(x) = Sym(L; νg )−1 Sym( D; (2.8.308)
∂Ω 2
∫ # $
 
− P.V. y E L (x, y), ∂τQ,P f (y)
Ix ⊗ D dσg (y)
∂Ω Ey

36 the singularity at x is away from ∂Ω, so it may be safely ignored

2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 607

at σg -a.e. point x ∈ ∂Ω.

Next, we observe that from [112, (1.12.52)] that Sym(∇XG ; ξ) = iξ(X) I G and
Sym(∇XE ; ξ) = iξ(X) I E , where I G is the identity on G and I E is the identity on E,
for each ξ ∈ T ∗ M. Consequently, for each ξ ∈ T ∗ M we have
 
Sym(∇XG ; ξ)Sym(D; ξ) = iξ(X)I G Sym(D; ξ) = Sym(D; ξ) iξ(X) I E

= Sym(D; ξ)Sym(∇XE ; ξ) (2.8.309)

which further implies that

[∇X , D] := ∇XG D − D∇XE is a first-order differential operator. (2.8.310)

Upon expressing

I(b)
1 (x) (2.8.311)
∫ =  >
  
= Ix ⊗ iSym ∇XG D − D∇XE ; νg (y) D y E L (x, y), f (y) dσg (y)
∂Ω Ey
∫ = >
 
= y E L (x, y), iSym ∇ G D − D∇ E ; νg (y) f (y)
Ix ⊗ D dσg (y)
X X
∂Ω Ey

and bearing in mind (2.8.310), we may once again call upon Theorem 2.8.4 which
guarantees the existence of a constant C = C(Ω, L, X, p, κ) ∈ (0, ∞) with
  (b) 
Nκ I  p ≤ C f  L p (∂Ω,σg )⊗ E ≤ C f  L p (∂Ω,σg )⊗ E , (2.8.312)
1 L (∂Ω,σg ) 1

p
where the fact that L1 (∂Ω, σg ) ⊗ E
→ L p (∂Ω, σg ) ⊗ E is a continuous embedding is
used in the second inequality. Furthermore, granted the format of I(b)
1 (x) in (2.8.312),
the jump-formula (2.8.144) from Theorem 2.8.4 also ensures that
 κ−n.t. 

I(b)
1  (x) (2.8.313)
∂Ω

1 
=  νg )Sym [∇X , D]; νg f (x)
Sym(L; νg )−1 Sym( D;
2
∫ = >
 
+ P.V. 
Ix ⊗ Dy E L (x, y) , i Sym [∇X , D]; νg (y) f (y) dσg (y)
∂Ω Ey

at σg -a.e. point x ∈ ∂Ω. Note that the jump term in (2.8.313) cancels the one in
(2.8.255). Based on (2.8.301), (2.8.303), (2.8.308), (2.8.300), (2.8.255), (2.8.306),
and (2.8.291) we therefore conclude that at σg -a.e. point x ∈ ∂Ω we have
608 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
 κ−n.t. 
(∇ E D L f )
X ∂Ω
(x) (2.8.314)

i 
=  νg ) ∂τQ,P f (x)
Sym(L; νg )−1 Sym( D;
2
∫ # $
 
− P.V. y E L (x, y), ∂τQ,P f (y)
Ix ⊗ D dσg (y)
∂Ω Gy
∫ = >
 
+ P.V. y E L (x, y) , i Sym [∇X , D]; νg (y) f (y)
Ix ⊗ D dσg (y)
∂Ω Gy
∫
% &
+ P.V. E L (x, y) , νg (X)(y)Wy f (y) Ey
dσg (y)
∂Ω
∫
% &
+ P.V. ωX (x, y), f (y) Ey
dσg (y), (2.8.315)
∂Ω

where the tangential derivative operator ∂τQ,P is defined as in [112, (1.12.131)] for
P := D (and with the vector bundle G in place of F ), and where the kernel
ωX (x, y) is as in (2.8.252). This establishes (2.8.289). From (2.8.301), (2.8.304),
(2.8.307), (2.8.312), (2.8.300), and (2.8.306) we also see that (2.8.297) does in fact
hold. Hence, the result claimed in (2.8.290) follows from (2.8.294), (2.8.295), and
(2.8.297). The proof of Theorem 2.8.10 is therefore complete. 

Continue to work in the context of Theorem 2.8.10. To set the stage, let us agree
that

∂τ(∇ X ;P) stands for tangential derivative operator ∂τQ,P defined as in

[112, (1.12.131)] for a given vector field X on the manifold M and (2.8.316)
a given first-order differential operator P : G → E.
Introduce
Ω+ := Ω and Ω− := M \ Ω. (2.8.317)
p
Given f ∈ L1 (∂Ω, σg ) ⊗ E with 1 < p < ∞, define

D L± f := (D L f )Ω± . (2.8.318)

Then Theorem 2.8.10 together with (2.8.231) and (2.8.233) imply that if u± := D L± f
we have
Nκ u± ∈ L p (∂Ω, σg ), Nκ (∇u± ) ∈ L p (∂Ω, σg ),
κ−n.t. κ−n.t. (2.8.319)
both u± ∂Ω , (∇u± )∂Ω exist σg -a.e. on ∂Ω,

and, in fact,
κ−n.t. 

u±  = ± 12 I + K L f at σg -a.e. point on ∂Ω. (2.8.320)
∂Ω
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 609

To proceed, observe from (2.8.314) that for each vector field X we have
κ−n.t. κ−n.t.
(∇XE D L+ f )∂Ω − (∇XE D L− f )∂Ω

 νg )∂τ(∇ X ;D ) f .
= i Sym(L; νg )−1 Sym( D; (2.8.321)

Use (2.8.320) to decompose the section f as

κ−n.t. κ−n.t.
f = u+ ∂Ω − u− ∂Ω . (2.8.322)

In this regard, we observe from (2.8.316) and [112, (1.12.137) in Theorem 1.12.13]
(whose applicability is ensured by (2.8.319)) employed with P, Q in place of P, Q,
where 
P : E → G ⊕ E, Pu := Du, −∇XE u ,
(2.8.323)
Q : G ⊕ E → G, Q(v, w) := ∇XG v + Dw,
that for any vector field X we have
)  κ−n.t. κ−n.t.
∂τ(∇ X ;D u+ ∂Ω = i Sym(Q; νg )(Pu+ )∂Ω
κ−n.t.
= −(−i)Sym(∇X ; νg )(Du+ )∂Ω
κ−n.t.
+ (−i)Sym(D; νg )(∇X u+ )∂Ω . (2.8.324)

Thus, if we locally express

D = A j ∇∂ j + B (2.8.325)
where the A j ’s and B are in Hom(E, G), then (2.8.324) implies that for each j we
have
(∇ ∂ ;D  )  κ−n.t.  κ−n.t.
∂τ j u+ ∂Ω = −(−i)Sym(∇∂j ; νg )(Du+ )∂Ω
κ−n.t.
+ (−i)Sym(D; νg )(∇∂j u+ )∂Ω (2.8.326)

hence at σg -a.e. point on ∂Ω we may write

610 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

(∇ ∂ j ;D  )  κ−n.t. 
 νg )A j i Sym(L; νg )−1 Sym( D;
Sym( D;  νg )∂τ u+ ∂Ω
κ−n.t.
 νg )Sym(∇∂j ; νg )(Du+ )
 νg ) A j Sym(L; νg )−1 Sym( D;
= −Sym( D; ∂Ω
κ−n.t.
 νg )Sym(D; νg )(∇∂j u+ )
 νg ) A j Sym(L; νg )−1 Sym( D;
+ Sym( D; ∂Ω
κ−n.t.
 νg )(Du+ )
 νg ) A j Sym(∇∂j ; νg )Sym(L; νg )−1 Sym( D;
= −Sym( D; ∂Ω
κ−n.t.
 νg )(∇∂j u+ )
 νg ) A j Sym(L; νg )−1 Sym( DD;
+ Sym( D; ∂Ω
κ−n.t.
 νg )(Du+ )
 νg ) Sym(D; νg )Sym(L; νg )−1 Sym( D;
= −Sym( D; ∂Ω
κ−n.t.
 νg ) A j (∇∂j u+ )
+ Sym( D; ∂Ω
κ−n.t.
 νg )(Du+ )
 νg )Sym(L; νg )−1 Sym( D;
= −Sym( DD; ∂Ω
κ−n.t.  κ−n.t.
 νg )(Du+ )
+ Sym( D;  νg ) B u+ 
− Sym( D;
∂Ω ∂Ω
κ−n.t.
 νg )(Du+ )
= −Sym( D; ∂Ω
κ−n.t.  κ−n.t.
 νg )(Du+ )
+ Sym( D;  νg ) B u+ 
− Sym( D;
∂Ω ∂Ω
 κ−n.t.
 νg ) B u+ 
= −Sym( D; (2.8.327)
∂Ω

plus a similar formula when u− is used in place of u+ . In concert, these two formulas
imply
(∇ ∂ j ;D  )  κ−n.t. 
 νg )A j i Sym(L; νg )−1 Sym( D;
Sym( D;  νg )∂τ u+ ∂Ω

(∇ ∂ j ;D  )  κ−n.t. 
 νg ) A j i Sym(L; νg )−1 Sym( D;
− Sym( D;  νg )∂τ u− ∂Ω
 κ−n.t. κ−n.t. 
 νg ) B u+ 
= −Sym( D; − u− ∂Ω
∂Ω

 νg ) B f ,
= −Sym( D; (2.8.328)

where the last equality comes from (2.8.322). Finally, from (2.8.325), (2.8.321),
(2.8.328), and (2.8.322) we conclude that
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 611
κ−n.t. κ−n.t.
 νg )(DD L+ f )
Sym( D;  νg )(DD L− f )
− Sym( D;
∂Ω ∂Ω
κ−n.t. κ−n.t.
 νg ) A j (∇∂j D L+ f )
= Sym( D;  νg ) A j (∇∂j D L− f )
− Sym( D;
∂Ω ∂Ω
κ−n.t. κ−n.t.
+ B(D L+ f )∂Ω − B(D L− f )∂Ω
(∇ ∂ j ;D  )  κ−n.t. 
 νg ) A j i Sym(L; νg )−1 Sym( D;
= Sym( D;  νg )∂τ u+ ∂Ω

(∇ ∂ j ;D  )  κ−n.t. 
 νg ) A j i Sym(L; νg )−1 Sym( D;
− Sym( D;  νg )∂τ u− ∂Ω

 κ−n.t.  κ−n.t.
+ B u+ ∂Ω − B u− ∂Ω

= −Sym( D;  νg ) B f = 0.
 νg ) B f + Sym( D; (2.8.329)

Then Theorem 2.8.10 together with the above argument prove the following result,
which may be regarded as a far-reaching generalization of the classical Lyapunov-
Tauber Theorem regarding the lack of jump for the normal derivative of the standard
harmonic double layer potential operator in the flat Euclidean setting (cf. [56, The-
orem 1,pp. 72-73], [90, p. 58]).

Theorem 2.8.11 Work in the context of Theorem 2.8.10 and retain notation in-
p
troduced above in (2.8.317)-(2.8.318). Then for each f ∈ L1 (∂Ω, σg ) ⊗ E with
1 < p < ∞ one has
κ−n.t. κ−n.t.
 νg )(DD L+ f )
Sym( D;  νg )(DD L− f )
= Sym( D; (2.8.330)
∂Ω ∂Ω

at σg -a.e. point on ∂Ω. In particular, the conormal derivatives

κ−n.t.
 νg )(DD L± f )
∂νLg D L± f := (−i)Sym( D; (2.8.331)
∂Ω

agree with one another, so it makes sense to simply write

κ−n.t.
 νg )(DD L± f )
∂νLg D L f := (−i)Sym( D; . (2.8.332)
∂Ω

Consider as such, the conormal derivative operator

p
∂νLg D L : L1 (∂Ω, σg ) ⊗ E −→ L p (∂Ω, σg ) ⊗ E (2.8.333)

is then well-defined, linear, and bounded.

The single and double layer potential operators play a basic role in the context
of the following Green type integral representation formula for null-solutions of a
weakly elliptic second-order differential operator L acting between vector bundles
over a Riemannian manifold.
612 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Theorem 2.8.12 Assume Hypotheses 2.8.5-2.8.6. Bring in the double layer operator
D L associated as in Definition 2.8.7 with the quasi-factorization of L in (2.8.211),
and recall the single layer operator 𝒮 associated with L as in (2.8.101). Fix an
aperture parameter κ > 0 and suppose
1,1
u ∈ Wloc (Ω) ⊗ E, Lu = 0 in Ω,
Nκ u and Nκ (Du) belong to L 1 (∂Ω, σg ), (2.8.334)
κ−n.t. κ−n.t.
u∂Ω and (Du)∂Ω exist at σg -a.e. point on ∂Ω.

Then, with the conormal derivative defined as

κ−n.t.
 νg )(Du)
∂νLg u := (−i)Sym( D; , (2.8.335)
∂Ω

the following integral representation formula holds:

 κ−n.t.  
u = D L u∂Ω − 𝒮 ∂νLg u in Ω. (2.8.336)

Proof We begin by fixing an arbitrary point x ∈ Ω, and considering the finite

dimensional vector space 𝒱 := E x . Next, define the mapping F! : Ω → 𝒱 ⊗ TM
via the requirement that for each y ∈ Ω and each ξ ∈ Ty∗ M we have
 #  $
Ty∗ M
!
ξ, F(y) = y E L (x, y), u(y)
Ix ⊗ (−i) Sym D ; ξ D
Ty M Ey
# $
 ξ)(Du)(y)
− E L (x, y), (−i)Sym( D; (2.8.337)
Ey

as vectors in 𝒱. Thanks to the first

 property in (2.8.334) and (2.8.186), this implies
that F! belongs to Lloc ! = Fj ∂j
1 (Ω, L n ) ⊗ 𝒱 ⊗ TM . In fact, we may locally express F
g
with
#  $
Fj (y) = Ix ⊗ (−i) Sym D ; dx j D y E L (x, y), u(y)
Ey
# $
 dx j )(Du)(y)
− E L (x, y), (−i)Sym( D; . (2.8.338)
Ey

Moreover, for any scalar-valued function ψ ∈ 𝒞1c (Ω) we may write

2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 613

divg F,
D  (Ω)
!ψ
D(Ω)

! grad ψ
= − D  (Ω) F, D(Ω)
∫
% &
=− Ty M
!
(grad ψ)(y), F(y) dLgn (y)
Ty M
Ω
∫

=− Ty∗ M
!
(grad ψ) (y), F(y) dLgn (y)
Ty M
Ω
∫

=− Ty∗ M
!
dψ(y), F(y) dLgn (y)
Ty M
Ω
∫ # $
 
=− y E L (x, y), u(y)
Ix ⊗ (−i) Sym D ; dψ(y) D dLgn (y)
Ω Ey
∫ # $

+  dψ(y) (Du)(y)
E L (x, y), (−i)Sym D; dLgn (y)
Ω Ey

=: I + II. (2.8.339)

Based on [112, (1.7.17), (1.7.20)], and integration by parts we may compute

∫ # $
 
I=− I ⊗ (−i) Sym D ; dψ D  E L (x, ·), u dLgn
Ω E
∫ # $
   
= I ⊗ (−i) Sym D; dψ D E L (x, ·), u dLgn
Ω E
∫ # $
 
=  E L (x, ·), (−i) Sym D; dψ u dLgn
I⊗D
Ω E
∫
% &
=  E L (x, ·), D(ψu)
I⊗D dLgn
E
Ω
∫
% &
−  E L (x, ·), ψ(Du)
I⊗D dLgn
E
Ω
∫
% &
= (ψu)(x) − ψ I ⊗ W  E L (x, ·), u E dLgn
Ω
∫
% &
−  E L (x, ·), Du
ψ I⊗D dLgn, (2.8.340)
E
Ω

where we have also used (2.8.222) and the fact that E L (x, y) is the Schwartz kernel
of L −1 . Likewise, from [112, (1.7.17), (1.7.20)], integration by parts, (2.8.211), and
the fact that Lu = 0 in the sense of distributions in Ω, we obtain
614 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
∫ # $

II =  dψ Du dLgn
E L (x, ·), (−i)Sym D;
Ω E
∫ # $
  
=−  ; dψ E L (x, ·), Du dLgn
I ⊗ (−i)Sym D
Ω E
∫ # $
 
=−  ψE L (x, ·) , Du
I⊗D dLgn
Ω E
∫ # $
+  E L (x, ·), Du dLgn
ψ I⊗D
Ω E
∫ ∫
% & % &
= ψ E L (x, ·), Wu E dLgn +  E L (x, ·), Du dLgn
ψ I⊗D E
Ω Ω
∫
% &
= ψ (I ⊗ W  )E L (x, ·), u E
dLgn
Ω
∫
% &
+  E L (x, ·), Du dLgn .
ψ I⊗D (2.8.341)
E
Ω

Altogether, (2.8.339)-(2.8.341) imply

 
!
D  (Ω) divg F, ψ D(Ω) = (ψu)(x) = u(x)δx, ψ
D  (Ω) D(Ω), (2.8.342)

where δx denotes the Dirac distribution with mass at x. Hence,

divg F! = u(x)δx ∈ V ⊗ ℰ(Ω). (2.8.343)

In addition, it follows from (2.8.337) and the third line in (2.8.334) that the nontan-
 κ−n.t.
gential trace F!∂Ω (y) exists at σg -a.e. point y ∈ ∂Ω and
  κ−n.t. 
Ty∗ M νg (y), F! ∂Ω (y)
Ty M
#   κ−n.t. $
= y E L (x, y), u
Ix ⊗ (−i) Sym D ; νg (y) D (y)
∂Ω Ey
#  κ−n.t.  $ 
 νg (y) (Du)
− E L (x, y), (−i)Sym D; (y) . (2.8.344)
∂Ω Ey

ρ
Also, if Nκ denotes the nontangential maximal operator truncated at some height
ρ ∈ 0, dist(x, ∂Ω) , then from (2.8.337), (2.8.334), and (2.8.220) we see that
ρ 
Nκ F! ≤ C Nκ u + Nκ (Du) pointwise on ∂Ω. (2.8.345)

In light of [112, (8.2.28)] and the second line in (2.8.334), this gives the mem-
ρ
bership Nκ F! ∈ L 1 (∂Ω, σg ). At this point the version of [112, Theorem 1.11.3]
mentioned in [112, Remark 1.11.4] applies and, account of (2.8.343)-(2.8.344),
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 615

(2.8.225), (2.8.335), (2.8.101), ultimately yields the integral representation formula

(2.8.336). 

In turn, the Green type integral representation formula established in Theo-

rem 2.8.12 has a wealth of useful consequences. Here is one such relevant application,
which plays a role in the discussion in the last part of §1.8.

Corollary 2.8.13 Work under Hypotheses 2.8.5-2.8.6 and strengthen the demand on
Ω by assuming that this is actually a UR domain. Associate with the differential
operator L the principal value singular integral operator K L# as in (2.8.236), and
the principal value singular integral operator K L as in (2.8.226). Also, let S be the
boundary-to-boundary single layer potential operator, associated with L much as in
(2.8.102). Then the following intertwining formula is true:

K L S = SK L# (2.8.346)
p
as operators from L p (∂Ω, σg ) ⊗ E into L1 (∂Ω, σg ) ⊗ E, for each p ∈ (1, ∞).

Proof Given any f ∈ L p (∂Ω, σg ) ⊗ E with p ∈ (1, ∞), Theorem 2.8.4 ensures that
the function u := 𝒮 f satisfies all properties listed in (2.8.334). As such, The-
orem 2.8.12 guarantees that integral representation formula (2.8.336) holds for
this u. Going nontangentially to the boundary in (2.8.336) then ultimately yields
K L (S f ) = S(K L# f ), thanks to (2.8.126), (2.8.233), and (2.8.235). That both K L S
p
and SK L# map L p (∂Ω, σg ) ⊗ E into L1 (∂Ω, σg ) ⊗ E is seen from (2.8.128), (2.8.232),
and (2.8.239). 

Work in the setting described in Hypotheses 2.8.5-2.8.6 and consider an additional

Hermitian vector bundle F → M. We shall describe a procedure which allows us
to decompose a generic first-order differential operator P : E → F as the sum of
a tangential differential operator on ∂Ω and a suitable “multiple” of the conormal
derivative operator ∂νLg defined in (2.8.335). The key observation is that the operator

τP := P − i Sym(P; νg )Sym(L; νg )−1 ∂νLg (2.8.347)

is tangential on ∂Ω, in the sense that this may be expressed as a linear combina-
tion of tangential differential operators expressed in local coordinates as (see [113,
(11.6.23)])

∂τ j k := (νg ) j ∇∂k − (νg )k ∇∂j , 1 ≤ j, k ≤ n, (2.8.348)

eventually plus a zero-th order term. This is ensured by the fact that Sym(τP ; νg ) = 0
at σg -a.e. point on ∂Ω, which follows readily from definitions:
616 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Sym(τP ; νg )

 νg )Sym(D; νg )
= Sym(P; νg ) − i Sym(P; νg )Sym(L; νg )−1 (−i)Sym( D;

= Sym(P; νg ) − Sym(P; νg )Sym(L; νg )−1 Sym(L; νg )

= Sym(P; νg ) − Sym(P; νg ) = 0. (2.8.349)

In turn, this fits into a more general narrative to the ffect that if a first-order differential
operator Q written locally as

Q= A j ∇∂ j + B (2.8.350)
j

satisfies
Sym(Q; νg ) = 0 at σg -a.e. point on ∂Ω, (2.8.351)
then 
(νg ) j A j = 0, (2.8.352)
j

and since
 
(νg )k (νg )k = g jk (νg )k (νg ) j = νg, νg  = |νg | 2 = 1, (2.8.353)
k j,k

we may write
 
Q= A j ∇∂ j + B − A j (νg ) j (νg )k ∇∂k
j j,k
 
= A j (νg )k (νg )k ∇∂j + B − A j (νg ) j (νg )k ∇∂k
j,k j,k
  
= A j (νg )k (νg )k ∇∂j − (νg ) j ∇∂k + B
j,k

=− A j (νg )k ∂τ j k + B, (2.8.354)
j,k

which goes to show that the operator Q is tangential. This observation then permits
to think of τP as a linear and continuous operator
p
τP : L1 (∂Ω, σg ) ⊗ E −→ L p (∂Ω, σg ) ⊗ E for p ∈ [1, ∞] (2.8.355)

defined as

A j (νg )k (∂τ j k f ) + B f for each f ∈ L1 (∂Ω, σg ) ⊗ E,
p
τP f := − (2.8.356)
j,k
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 617

where the homomorphisms A j ’s and B are as in (2.8.351) when Q := τP (and where

each ∂τ j k f is considered as in [112, Definition 1.12.7] with X := ∂j and Y := ∂k ).
On the other hand, we can also consider the action of τP from (1.8.85) on any section
1,1
u ∈ Wloc (Ω) ⊗ E with Nκ u, Nκ (∇ E u) ∈ L 1 (∂Ω, σg ),
κ−n.t. κ−n.t. (2.8.357)
and such that u∂Ω , (∇ E u)∂Ω exist σg -a.e. on ∂Ω,

(where κ > 0 is a some fixed background aperture parameter) according to

κ−n.t.
τP u := (Pu)∂Ω − i Sym(P; νg )Sym(L; νg )−1 ∂νLg u, (2.8.358)

with the conormal derivative understood in the sense of (2.8.335). It is significant to

note that, as seen from (2.8.358) and (2.8.335), we have
 νg )τD u = 0 for each u as in (2.8.357).
Sym( D; (2.8.359)

In regard to the two definitions of τP given above we wish to note that:

given any section u as in (2.8.357), it follows that the nontangential
κ−n.t.
boundary trace u∂Ω belongs to the Sobolev space L11 (∂Ω, σg ) ⊗ E
 κ−n.t. (2.8.360)
and τP u∂Ω (considered as in (2.8.355)-(2.8.356)) is equal to τP u
(considered as in (2.8.358)).
κ−n.t.
To justify the claims made above, suppose u is as in (2.8.357). That u∂Ω belongs to
the Sobolev space L11 (∂Ω, σg )⊗ E is then a consequence of [112, Proposition 1.12.9].
Furthermore, based on (2.8.355), (2.8.356), and [112, (1.12.102)] we may write
 κ−n.t.   κ−n.t.  κ−n.t.
τP u∂Ω = − A j (νg )k ∂τ j k u∂Ω + B u∂Ω
j,k
   κ−n.t.  κ−n.t.   κ−n.t.
=− A j (νg )k (νg ) j ∇∂k u ∂Ω − (νg )k ∇∂j u ∂Ω + B u∂Ω
j,k
  κ−n.t.  κ−n.t.
= A j ∇∂j u ∂Ω + B u∂Ω = τP u, (2.8.361)
j

thanks to (2.8.352), (2.8.353), and (2.8.358). This establishes (2.8.360).

Additional basic properties of τP and other related tangential differential operators
are contained in the lemma below.

Lemma 2.8.14 Assume Hypotheses 2.8.5-2.8.6. Then, having fixed an aperture pa-
rameter κ > 0, given any section
1,1
u ∈ Wloc (Ω) ⊗ E with Nκ u, Nκ (∇ E u) ∈ L 1 (∂Ω, σg ),
κ−n.t. κ−n.t. (2.8.362)
and such that u∂Ω , (∇ E u)∂Ω exist σg -a.e. on ∂Ω,
618 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
κ−n.t.
it follows that the nontangential boundary trace u∂Ω belongs to the Sobolev space
L11 (∂Ω, σg ) ⊗ E and for any two vector fields X, Y ∈ TM one has
 κ−n.t.
νg (X)τ∇Y u − νg (Y )τ∇ X u = ∂τXY u∂Ω (2.8.363)

where τ∇ X , τ∇Y are considered as in (2.8.358), and where ∂τXY acts on the space
L11 (∂Ω, σg ) ⊗ E as in [112, Definition 1.12.7]. In particular, corresponding to
X := ∂j and Y := ∂k for some j, k ∈ {1, . . . , n}, one obtains
 κ−n.t.
(νg ) j τ∇∂k u − (νg )k τ∇∂ j u = ∂τ j k u∂Ω (2.8.364)

for any section u as in (2.8.362).

Henceforth strengthen the demand on Ω by assuming that this is actually a UR
domain. Then any two vector fields X, Y ∈ TM one has

νg (X)τ∇Y f − νg (Y )τ∇ X f = ∂τXY f

p
(2.8.365)
for each f ∈ L1 (∂Ω, σg ) ⊗ E with p ∈ (1, ∞),

where the actions of τ∇ X , τ∇Y on f are considered as in (2.8.355)-(2.8.356), and

where ∂τXY f is considered as in [112, Definition 1.12.7]. As a consequence, corre-
sponding to X := ∂j and Y := ∂k for some j, k ∈ {1, . . . , n}, one has

(νg ) j τ∇∂k f − (νg )k τ∇∂ j f = ∂τ j k f

p
(2.8.366)
for each f ∈ L1 (∂Ω, σg ) ⊗ E with p ∈ (1, ∞).

Furthermore, the tangential differential operator τD defined as in (2.8.355)-(2.8.356)

with P := D satisfies
 νg )τD f = 0 for each f ∈ L (∂Ω, σg ) ⊗ E with p ∈ (1, ∞).
Sym( D;
p
(2.8.367)
1

Also, for any vector field X ∈ TM one has


 νg )∂τ(∇ X ;D ) f = τ∇ X f
i Sym(L; νg )−1 Sym( D;
p
(2.8.368)
for each f ∈ L1 (∂Ω, σg ) ⊗ E with p ∈ (1, ∞),

where ∂τ(∇ X ;D ) is defined as in (2.8.316) with P := D . Finally, for any two vector
fields X, Y ∈ TM one has
 
νg (X)∂τ(∇Y ;D ) f − νg (Y )∂τ(∇ X ;D ) f = (−i)Sym(D; νg )∂τXY f
p
(2.8.369)
for each f ∈ L1 (∂Ω, σg ) ⊗ E with p ∈ (1, ∞)

so, in particular,
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 619

(∇ ∂k ;D  ) (∇ ∂ j ;D  )
(νg ) j ∂τ f − (νg )k ∂τ f = (−i)Sym(D; νg )∂τ j k f
(2.8.370)
p
for each f ∈ L1 (∂Ω, σg ) ⊗ E with p ∈ (1, ∞).

Proof Select an arbitrary section u satisfying the properties listed in (2.8.362). Then
κ−n.t.
[112, (1.12.100)] ensures that u∂Ω belongs to L11 (∂Ω, σg ) ⊗ E, while (2.8.358) and
[112, (1.12.101)] permit us to compute

νg (X)τ∇Y u − νg (Y )τ∇ X u
κ−n.t.
= νg (X)(∇Y u)∂Ω − νg (X) i Sym(∇Y ; νg )Sym(L; νg )−1 ∂νLg u
κ−n.t.
− νg (Y )(∇X u)∂Ω + νg (Y ) i Sym(∇X ; νg )Sym(L; νg )−1 ∂νLg u
κ−n.t.
= νg (X)(∇Y u)∂Ω + νg (X)νg (Y )Sym(L; νg )−1 ∂νLg u
κ−n.t.
− νg (Y )(∇X u)∂Ω − νg (Y )νg (X)Sym(L; νg )−1 ∂νLg u
κ−n.t. κ−n.t.
= νg (X)(∇Y u)∂Ω − νg (Y )(∇X u)∂Ω
 κ−n.t.
= ∂τXY u∂Ω . (2.8.371)

This proves (2.8.363).

For the remainder of the proof work under the stronger assumption that Ω is
actually a UR domain, and pick an integrability exponent p ∈ (1, ∞). In such
a scenario, (2.8.365) becomes a consequence of the decomposition in (2.8.322),
κ−n.t.
(2.8.363), and (2.8.360). We next observe that whenever f = u∂Ω for some
1,1
u ∈ Wloc (Ω) ⊗ E with Nκ u, Nκ (∇ E u) ∈ L p (∂Ω, σg ),
κ−n.t. κ−n.t. (2.8.372)
and such that u∂Ω , (∇ E u)∂Ω exist σg -a.e. on ∂Ω,

it follows that
 κ−n.t.
Sym( D;  νg )τD u
 νg )τD f = Sym( D;  νg )τD u = 0,
= Sym( D; (2.8.373)
∂Ω

thanks to (2.8.360) and (2.8.359). The claim made in (2.8.367) then becomes a
consequence of this special case and the decomposition (2.8.322).
As regards the claim made in (2.8.368), we once again first treat the case when
κ−n.t.
the section f ∈ L1 (∂Ω, σg ) ⊗ E is of the form f = u∂Ω for some u as in (2.8.372).
p

In such a scenario we may employ (2.8.324), the fact that ∇X has a scalar principal
symbol, and (2.8.358) with P := ∇X to write
620 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

 νg )∂τ(∇ X ;D ) f
i Sym(L; νg )−1 Sym( D;
κ−n.t.
 νg )Sym(∇X ; νg )(Du)
= −Sym(L; νg )−1 Sym( D; ∂Ω
κ−n.t.
 νg )Sym(D; νg )(∇X u)
+ Sym(L; νg )−1 Sym( D; ∂Ω
κ−n.t.
 νg )(Du)
= −Sym(∇X ; νg )Sym(L; νg )−1 Sym( D; ∂Ω
κ−n.t.
 νg )(∇X u)
+ Sym(L; νg )−1 Sym( DD; ∂Ω
κ−n.t.
= −i Sym(∇X ; νg )Sym(L; νg )−1 ∂νLg u + (∇X u)∂Ω

= τ∇ X f , (2.8.374)

as wanted. The general case of (2.8.368) then follows from this and the decomposition
(2.8.322).
In the proof (2.8.369) we shall adopt the same strategy of first checking the case
κ−n.t.
when the section f ∈ L1 (∂Ω, σg ) ⊗ E has the particular structure f = u∂Ω where
p

u as in (2.8.372). In this special situation we make use of (2.8.324), (2.8.358) with

P := ∇X , the formulas for the principal symbols of ∇X and ∇Y , as well as [112,
(1.12.101)] to compute
 
νg (X)∂τ(∇Y ;D ) f − νg (Y )∂τ(∇ X ;D ) f
)  κ−n.t.    κ−n.t.
= νg (X)∂τ(∇Y ;D u∂Ω − νg (Y )∂τ(∇ X ;D ) u∂Ω
κ−n.t.
= −νg (X)(−i)Sym(∇Y ; νg )(Du)∂Ω
κ−n.t.
+ νg (X)(−i)Sym(D; νg )(∇Y u)∂Ω
κ−n.t.
+ νg (Y )(−i)Sym(∇X ; νg )(Du)∂Ω
κ−n.t.
− νg (Y )(−i)Sym(D; νg )(∇X u)∂Ω
 κ−n.t. κ−n.t. 
= (−i)Sym(D; νg ) νg (X)(∇Y u)∂Ω − νg (Y )(∇X u)∂Ω

 κ−n.t.
= (−i)Sym(D; νg )∂τXY u∂Ω

= (−i)Sym(D; νg )∂τXY f , (2.8.375)

as desired. Having established this, then (2.8.369) follows in full generality by

availing ourselves of the decomposition in (2.8.322). 
2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 621

We are now in a position to give global, coordinate-free formulas for nontangential

traces of generic first-order derivatives of the boundary-to-domain double layer
introduced in Definition 2.8.7.

Theorem 2.8.15 Work in the context of Theorem 2.8.10. Consider an additional

Hermitian vector bundle F → M and suppose P : E → F is a generic first-order
p
differential operator. Then for each f ∈ L1 (∂Ω, σg ) ⊗ E with 1 < p < ∞ one has
κ−n.t.
(PD L f )∂Ω = τP ( 12 I + K L ) f + i Sym(P; νg )Sym(L; νg )−1 ∂νLg (D L f ) (2.8.376)

at σg -a.e. point on ∂Ω, where the tangential derivative operator τP is as in (2.8.355)-

(2.8.356). In particular, using notation introduced in (2.8.317)-(2.8.318), from
(2.8.376) and Theorem 2.8.11 one obtains
κ−n.t. κ−n.t.
(PD L+ f )∂Ω − (PD L− f )∂Ω = τP f . (2.8.377)
p
Furthermore, given any vector field X ∈ TM and any f ∈ L1 (∂Ω, σg ) ⊗ E with
1 < p < ∞, at σg -a.e. point x ∈ ∂Ω one has
 κ−n.t. 
(∇XE D L f )∂Ω (x)

1
= τ∇ X f (x)
2
∫ # $
  
− P.V. y E L (x, y), ∂τ(∇ X ;D ) f (y)
Ix ⊗ D dσg (y)
∂Ω Gy
∫ = >
 
+ P.V. y E L (x, y) , i Sym [∇X , D]; νg (y) f (y)
Ix ⊗ D dσg (y)
∂Ω Gy
∫
% &
+ P.V. E L (x, y) , νg (X)(y)Wy f (y) Ey
dσg (y)
∂Ω
∫
% &
+ P.V. ωX (x, y), f (y) Ey
dσg (y), (2.8.378)
∂Ω

where τ∇ X is considered as in (2.8.355)-(2.8.356), the tangential derivative oper-


ator ∂τ(∇ X ;D ) is defined as in (2.8.316), and where the kernel ωX (x, y) has been
introduced in (2.8.252).
p
Proof Pick f ∈ L1 (∂Ω, σg ) ⊗ E with 1 < p < ∞, arbitrary. From Theorem 2.8.10,
(2.8.231), (2.8.233) we see that we may employ (2.8.358) with u := D L f . In concert
with (2.8.360) this permits us to write
622 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets
κ−n.t.
(PD L f )∂Ω = τP (D L f ) + (−i)Sym(P; νg )Sym(−L; νg )−1 ∂νLg (D L f )
 κ−n.t. 
= τP (D L f )∂Ω + (−i)Sym(P; νg )Sym(−L; νg )−1 ∂νLg (D L f )

= τP ( 12 I + K L ) f + (−i)Sym(P; νg )Sym(−L; νg )−1 ∂νLg (D L f )

(2.8.379)

as wanted. Also, (2.8.378) is a direct consequence of (2.8.314), (2.8.368), and

(2.8.316). 
Parenthetically, we wish to note that, thanks to (2.8.367), the jump-formula
recorded in (2.8.377) written for P := D is consistent with the lack of jump for
the conormal derivative of the double layer (see (2.8.330)).
We continue by presenting an important commutator identity to the effect that
one can pass tangential derivatives through the principal value double layer all the
way to the section on which this acts (itself assumed to be in a boundary Sobolev
space), at the expense of operators which may be regarded (at least in a favorable
geometric setting) as being residual.

Theorem 2.8.16 Retain the setting of Theorem 2.8.10 and consider two vector fields
X, Y ∈ TM. Then, for each p ∈ (1, ∞),
p
the commutator ∂τXY , K L : L1 (∂Ω, σg ) ⊗ E −→ L p (∂Ω, σg ) ⊗ E
may be written as a linear combination of operators of the form
[Mνg , T] ◦ ∂τ and a bounded map B : L p (∂Ω, σg ) ⊗ E → L p (∂Ω, σg ) ⊗ E,
(2.8.380)
where [Mνg , T] is the commutator between the pointwise multiplication with either
νg (X) or νg (Y ) and a Calderón-Zygmund singular integral operator T on ∂Ω
(whose kernel involves at most first-order derivatives of E L (x, y)), and where the
p
map ∂τ : L1 (∂Ω, σg ) ⊗ E → L p (∂Ω, σg ) ⊗ G is bounded.
p
Proof Fix an arbitrary f ∈ L1 (∂Ω, σg ) ⊗ E. Based on [112, (1.12.101) in Propo-
sition 1.12.9] (whose applicability in the present setting is guaranteed by Theo-
rem 2.8.10, (2.8.231), and (2.8.233)) and the jump-formula (2.8.233) we may write
 κ−n.t.
∂τXY (K L f ) = ∂τXY ( 12 I + K L ) f − 12 ∂τXY f = ∂τXY D L f ∂Ω − 12 ∂τXY f
 κ−n.t.  κ−n.t.
= νg (X) ∇Y D L f ∂Ω − νg (Y ) ∇X D L f ∂Ω − 12 ∂τXY f . (2.8.381)

At this stage, we shall employ (2.8.378) to compute, at σg -a.e. point x ∈ ∂Ω,

2.8 Calderón-Zygmund Theory for Singular Integrals on Riemannian Manifolds 623
 κ−n.t.  κ−n.t.
νg (X)(x) ∇Y D L f ∂Ω (x) − νg (Y )(x) ∇X D L f ∂Ω (x)

1   
= νg (X)(x) τ∇Y f (x) − νg (Y )(x) τ∇ X f (x)
2
∫ # $
  
− νg (X)(x) P.V. y E L (x, y), ∂τ(∇Y ;D ) f (y)
Ix ⊗ D dσg (y)
∂Ω Gy
∫ # $
 
+ νg (Y )(x) P.V. y E L (x, y), ∂τ(∇ X ;D ) f (y)
Ix ⊗ D dσg (y)
∂Ω Gy

+ (B f )(x), (2.8.382)

where B is a linear and bounded operator from L p (∂Ω, σg ) ⊗ E into itself, originating
from the last three lines in (2.8.378). In relation to this formula, we make two key
observations. First, thanks to (2.8.365),
 the expression in the first line in the right-
hand side of (2.8.382) is simply 12 ∂τXY f )(x) (which ultimately is going to cancel
the last term in (2.8.381)). Second, up to operators of the form [Mνg , T] ◦ ∂τ where
[Mνg , T] is the commutator between the pointwise multiplication with νg (X) or
νg (Y ) and the Calderón-Zygmund singular integral operator T on ∂Ω, acting on each
φ ∈ L p (∂Ω, σg ) ⊗ G according to
∫
% &
(T φ)(x) := −P.V. y E L (x, y), φ(y)
Ix ⊗ D dσg (y), (2.8.383)
Gy
∂Ω

(note that the integral kernel of T involves at most first-order derivatives on E L (x, y))
 
and ∂τ is either ∂τ(∇ X ;D ) or ∂τ(∇Y ;D ) , the expressions in the second and third lines
in the right-hand side of (2.8.382) combine to:
∫ #
  
P.V. Ix ⊗ Dy E L (x, y), νg (Y )(y) ∂τ(∇ X ;D ) f (y)
∂Ω
 
$
− νg (X)(y) ∂τ(∇Y ;D ) f (y) dσg (y)
Gy
∫ # $

= P.V. y E L (x, y), i Sym(D; νg (y)) ∂τXY f (y)
Ix ⊗ D dσg (y)
∂Ω Gy
∫ # $
 
= P.V. y E L (x, y) , f (y)
Ix ⊗ (−i) Sym D ; νg (y) D dσg (y)
Ey
∂Ω
  
= K L ∂τXY f (x). (2.8.384)

Above, the second equality comes from (2.8.369) and the last equality is just the
definition of K L from (2.8.226).
Altogether, this argument establishes the claim made in (2.8.380). 
We conclude with the following remark, pointing to further extensions.
624 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

Remark 2.8.17 In view of the fact that integral operators with variable coefficient
kernels are allowed in the formulation of Theorem 2.6.1, the same type localization
argument as in the proofs of Theorem 2.8.2 and Theorem 2.8.4 shows that similar
results to those described in Theorem 2.8.2, Theorem 2.8.4, Theorem 2.8.10, The-
orem 2.8.11, Theorem 2.8.15, and Theorem 2.8.16 are also valid in the context of
Morrey and block spaces, naturally defined on the boundaries of UR domains in
Riemannian manifolds.

2.9 Some Applications to Singular Integrals Associated with

Elliptic Differential Operators

The powerful Calderón-Zygmund theory developed as the collection of results cen-

tered around Theorem 2.3.2, Theorem 2.4.1, and Theorem 2.5.1 (as well as their
proofs and various extensions) has a multitude of applications of special interest. In
this section we shall elaborate on some of these applications, with the understanding
that much more will follow in Volumes III-V.
To set the stage for our first application, presented in Theorem 2.9.1 below,
consider next a second-order, homogeneous, constant (complex) coefficient, elliptic
differential operator
n
L= ar s ∂r ∂s (2.9.1)
r,s=1

in Rn , where n ∈ N with n ≥ 2. The aforementioned ellipticity condition means that

the coefficient matrix A := (ar s )1≤r,s ≤n ∈ Cn×n has the property that there exists
some constant c > 0 such that
 n 

Re ar s ξr ξs ≥ c|ξ | 2, ∀ξ = (ξ1, . . . , ξn ) ∈ Rn . (2.9.2)
r,s=1

Then the locally integrable function E defined in (1.4.171) is a fundamental solution

for the differential operator L in Rn .
Theorem 2.9.1 below provides natural examples of vector fields satisfying the
hypotheses of [112, Theorem 1.2.1]. Theorem 2.9.1 is also very much relevant in
the treatment of the Neumann Problem for elliptic differential operators in rough
domains, via boundary layer potentials, as done in [28] in the case of the Laplace
operator in Lipschitz domains.

Theorem 2.9.1 Suppose Ω ⊆ Rn (where n ∈ N with n ≥ 2) is a nonempty open

set with the property that ∂Ω is a UR set; in particular, Ω is a set of locally
finite perimeter. Abbreviate σ := H n−1 ∂Ω, and denote by ν = (ν1, . . . , νn ) the
geometric measure theoretic outward unit normal to Ω. Also, consider a second-
order, homogeneous, constant (complex) coefficient, elliptic differential operator L
as in (2.9.1), and recall its fundamental solution E defined in (1.4.171).
2.9 Some Applications to Singular Integrals Associated with Elliptic Differential Operators 625

In relation to the differential operator L and the domain Ω, introduce the vector
field
∫ n 

!
F(x) := ar s (∂s E)(x − y) f (y) dσ(y) for x ∈ Ω, (2.9.3)
∂Ω s=1 1≤r ≤n

where (ar s )1≤r,s ≤n ∈ Cn×n is the coefficient matrix of L, and f ∈ H 1 (∂Ω, σ) is an

arbitrary function.
Then F! ∈ 𝒞∞ (Ω) is a divergence-free vector field in Ω, with the property that
n

for each fixed parameter κ ∈ (0, ∞) there exists a constant CΩ,κ ∈ (0, ∞) such that
 
Nκ F! ∈ L 1 (∂Ω, σ) and Nκ F! L 1 (∂Ω,σ) ≤ CΩ,κ  f  H 1 (∂Ω,σ), (2.9.4)
κ−n.t.
and such that the nontangential boundary trace F!∂Ω exists σ-a.e. on ∂∗ Ω and is
independent of κ. In fact, as regards the latter property, at σ-a.e. point x ∈ ∂∗ Ω one
actually has (with the dependence on κ dropped)
 n 
 n.t.  ar s νs (x) n
 1
F! (x) = − n s=1
f (x) + ar s (Ts f )(x) (2.9.5)
∂Ω 2 s,=1 as νs (x)ν (x) s=1 1≤r ≤n

where, for each s ∈ {1, . . . , n},

∫
Ts f (x) := lim+ (∂s E)(x − y) f (y) dσ(y) at σ-a.e. x ∈ ∂Ω. (2.9.6)
ε→0
y ∈∂Ω
|x−y |>ε

Moreover, if the set Ω is actually a UR domain (i.e., if the additional condition

that H n−1 (∂Ω \ ∂∗ Ω) = 0 also holds), then there exists a constant CΩ ∈ (0, ∞) such
that
 n.t.    n.t.  
   
ν · F! ∈ H 1 (∂Ω, σ) and ν · F!  1 ≤ CΩ  f  H 1 (∂Ω,σ) . (2.9.7)
∂Ω ∂Ω H (∂Ω,σ)

Proof By design, F! is a 𝒞∞ smooth, divergence-free vector field in Ω. The properties

in (2.9.4) are direct consequences of (2.4.14) (used with p = 1, Σ := ∂Ω, and
k := ∂s E) in Theorem 2.4.1, whereas the nontangential boundary trace formula
(2.9.5) is implied by (2.5.4) (with k := ∂s E), bearing in mind that the tempered
- is of function type in Rn \ {0} and, in fact (see, e.g., [109, §7.12,
distribution E
p. 305]),
  −1

n
- =−
E(ξ) ar s ξr ξs , ∀ξ = (ξ1, . . . , ξn ) ∈ Rn \ {0}. (2.9.8)
r,s=1
626 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

There remains to deal with the claims made in (2.9.7) under the assumption that Ω
is actually a UR domain. To this end, consider the singular integral operator defined
for each f ∈ L p (∂Ω, σ) with 1 ≤ p < ∞ according to
∫ 
n
T f (x) := lim+ ar s νr (x)(∂s E)(x − y) f (y) dσ(y) (2.9.9)
ε→0
y ∈∂Ω r,s=1
|x−y |>ε

at σ-a.e. x ∈ ∂Ω. In particular, Theorem 2.3.2 guarantees that

T : L p (∂Ω, σ) −→ L p (∂Ω, σ) with p ∈ (1, ∞), as well as

T : L 1 (∂Ω, σ) −→ L 1,∞ (∂Ω, σ), are well-defined, linear, and (2.9.10)
bounded operators.

To proceed, observe that if F! is the vector field associated with a given function
f ∈ L p (∂Ω, σ) with 1 ≤ p < ∞ as in (2.9.3), then much as in the case of (2.9.5)
we have
 n 
 n.t.  n
ar s νs (x) n
 1
ν(x) · F! (x) = νr (x) − n s=1
f (x) + ar s (Ts f )(x)
∂Ω
r=1
2 s,=1 as νs (x)ν (x) s=1

= − 12 f (x) + (T f )(x) for σ-a.e. x ∈ ∂Ω. (2.9.11)

Suppose now that we fix some function

∫
p
f ∈ Lcomp (∂Ω, σ), with p ∈ (1, ∞), satisfying f dσ = 0 (2.9.12)
∂Ω

and, once again, associate the vector field F! with this f as in (2.9.3). Then the
function f ∈ H 1 (∂Ω, σ) (cf. [7, Lemma 5.8, p. 183]) hence, according to what we
have established in the first part of the present proof, F! ∈ 𝒞∞ (Ω) is a divergence-
n

free vector field in Ω, satisfying (2.9.4), and such that (2.9.5) holds for each κ > 0.
We may therefore compute
∫ ∫ ∫  n.t. 

T f dσ = (− 2 f + T f ) dσ =
1
ν · F! dσ = 0, (2.9.13)
∂Ω ∂Ω ∂Ω ∂Ω

on account of (2.9.11), the cancelation condition in (2.9.12), and the Divergence

Formula [112, (1.2.2)] from [112, Theorem 1.2.1] (whose current applicability in the
present setting is ensured by the hypotheses on Ω and the aforementioned properties
! In turn, (2.9.9), (2.9.10), and (2.9.13) allow us to invoke the Hardy space
of F).
boundedness criterion from [7, Theorem 8.40, p. 438] according to which
the operator T from (2.9.10) induces a unique linear
(2.9.14)
bounded mapping T : H 1 (∂Ω, σ) −→ H 1 (∂Ω, σ).
2.9 Some Applications to Singular Integrals Associated with Elliptic Differential Operators 627

Then the claims in (2.9.7) are seen from (2.9.11), the fact that we have the inclusion
H 1 (∂Ω, σ) ⊂ L 1 (∂Ω, σ), and (2.9.14). 

It turns out that if Ω is a UR domain in Rn then the entire Hardy space H 1 (∂Ω, σ)
 n.t.
is actually spanned by functions of the form ν · F!∂Ω where F! is a divergence-free
vector field whose nontangential maximal operator is integrable on the boundary,
along with their counterparts defined relative to Rn \ Ω. Specifically, we have the
following result.

Corollary 2.9.2 Assume Ω ⊆ Rn (where n ∈ N with n ≥ 2) is UR domain, with

geometric measure theoretic outward unit normal ν, and abbreviate σ := H n−1 ∂Ω.
Also, fix κ ∈ (0, ∞) and define

Ω+ := Ω and Ω− := Rn \ Ω. (2.9.15)

Then there exists a constant CΩ,κ ∈ (0, ∞) with the property that for every
f ∈ H 1 (∂Ω, σ) one can find two divergence-free vector fields F!± ∈ 𝒞∞ (Ω± )
n

satisfying:
 
Nκ F!±  1 ≤ CΩ,κ  f  H 1 (∂Ω,σ), (2.9.16)
L (∂Ω,σ)
κ−n.t.
F!± ∂Ω± exist σ-a.e. on ∂Ω, (2.9.17)
 κ−n.t.
ν · F!± ∂Ω± ∈ H 1 (∂Ω, σ), (2.9.18)
  κ−n.t. 
ν · F!±   1 ≤ CΩ,κ  f  H 1 (∂Ω,σ), (2.9.19)
∂Ω± H (∂Ω,σ)

 κ−n.t.  κ−n.t.
ν · F!+ ∂Ω+ − ν · F!− ∂Ω− = f σ-a.e. on ∂Ω. (2.9.20)

Proof Select some second-order, homogeneous, constant coefficient, elliptic dif-

ferential operator L as in (2.9.1), and let E be its fundamental solution defined in
(1.4.171). Also, fix an arbitrary f ∈ H 1 (∂Ω, σ). If we now define F!± as in (2.9.3)
relative to Ω± and the function − f , Theorem 2.9.1 and [112, Lemma 5.10.9] give
that the vector fields F!± are 𝒞∞ smooth and divergence-free in Ω± , and satisfy
(2.9.16)-(2.9.19). Next, observe that (2.9.5) presently gives (bearing in mind that
f is currently replaced by − f , and taking [112, Lemma 5.10.9] once again into
account) that at σ-a.e. x ∈ ∂Ω we have
 n 
 κ−n.t.  ar s νs (x) n
 1
F!±  (x) = ± n s=1
f (x) − ar s (Ts f )(x) . (2.9.21)
∂Ω± 2 s,=1 as νs (x)ν (x) s=1 1≤r ≤n

Then (2.9.20) readily follows from (2.9.21) and simple algebra. 

We proceed by recalling the fundamental solution associated with higher-order ho-

mogeneous constant (complex) coefficient elliptic systems from [109, Theorem 11.1,
628 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

pp. 393-395]. This is going to be relevant in the formulation and proof of Corol-
lary 2.9.4, discussed shortly.
Theorem 2.9.3 Fix n, m, M ∈ N with n ≥ 2, and assume that L is an M × M system
in Rn of order 2m of the form

L= ∂ α Aαβ ∂ β (2.9.22)
|α |= |β |=m

where each Aαβ ∈ C M×M . Assume that the characteristic matrix of L, i.e., the
M × M matrix-valued function37

L(ξ) := (−1)m ξ α+β Aαβ, ∀ξ ∈ Rn, (2.9.23)
|α |= |β |=m

satisfies
det [L(ξ)]  0, ∀ξ ∈ Rn \ {0}. (2.9.24)
Then the M × M matrix E defined at each x ∈ \ {0} byRn
∫
(−1)m −1
E(x) := Δ(n−1)/2
x (x · ξ)2m−1 sgn (x · ξ) L(ξ) dH n−1 (ξ)
4(2πi)n−1 (2m − 1)! S n−1
(2.9.25)
if n is odd, and
∫
(−1)m+1 −1
E(x) := Δn/2 (x · ξ)2m ln |x · ξ | L(ξ) dH n−1 (ξ) (2.9.26)
(2πi)n (2m)! x S n−1

if n is even, and satisfies the following properties.

(1) The entries in E are real-analytic functions in Rn \ {0} (hence, in particular,
functions in 𝒞∞ (Rn \ {0})) and

E(−x) = E(x) for each x ∈ Rn \ {0}. (2.9.27)

(2) Define the M × M matrix-valued function

∫
(−1)m+1 −1
P(x) := (x · ξ)2m−n L(ξ) dH n−1 (ξ) (2.9.28)
(2πi)n (2m − n)! S n−1
at each x ∈ Rn . Then the entries of P are identically zero when either n is odd
or n > 2m, and are homogeneous polynomials of degree 2m − n when n ≤ 2m.
Moreover, there exists a matrix-valued function Φ : Rn \ {0} → C M×M , whose
entries belong to 𝒞∞ (Rn \ {0}) and are positive homogeneous of degree 2m − n,
such that

E(x) = Φ(x) + ln |x| P(x), ∀x ∈ Rn \ {0}. (2.9.29)
37 The reader is alerted to the fact that our present definition of L(ξ) differs by a factor of (−1) m
from the one in [109, (11.3.2), p. 391]
2.9 Some Applications to Singular Integrals Associated with Elliptic Differential Operators 629

(3) For each multi-index γ ∈ N0n there exists Cγ ∈ (0, ∞) such that for every
x ∈ Rn \ {0} one has

⎧
⎪ Cγ
⎪
⎪ if either n is odd, or n > 2m, or if |γ| > 2m − n,
 γ  ⎪
⎨ |x|
⎪ n−2m+ |γ |
(∂ E)(x) ≤
⎪
⎪ Cγ (1 + | ln |x||)
⎪
⎪ if 0 ≤ |γ| ≤ 2m − n.
⎪
⎩ |x| n−2m+ |γ |
(2.9.30)
In particular,
1 (Rn, L n ), hence they may
the entries of E belong to Lloc
be canonically regarded as distributions in Rn ; in the (2.9.31)
latter scenario, the entries of E are actually tempered
distributions in Rn .
Also,
for each multi-index γ ∈ N0n with |γ| ≥ 2m − 1, the func-
tion ∂γ E is of class 𝒞∞ , as well as positive homogeneous (2.9.32)
of degree 2m − n − |γ|, in Rn \ {0}.
(4) For each fixed y ∈ Rn one has (in the sense of tempered distributions in Rn )

Lx E(x − y) = δy (x) I M×M . (2.9.33)

Above, the subscript x denotes the fact that the operator L in (2.9.33) is applied
to each column of E in the variable x. Also, δy stands for the Dirac distribution
with mass at y in Rn , and I M×M is the M × M identity matrix.
(5) When restricted to Rn \ {0}, the entries of -E (with “hat” denoting the Fourier
transform in Rn ; cf. (1.4.14)) are 𝒞∞ functions and, in fact,
−1
-
E(ξ) = L(ξ) for each ξ ∈ Rn \ {0}. (2.9.34)

(6) Writing E L in place of E to emphasize the dependence on L, the fundamental

solution E L with entries as in (2.9.25)-(2.9.26) satisfies
   ∗
EL = EL , EL = EL , EL = EL∗,
(2.9.35)
and EλL = λ−1 E L for each λ ∈ C \ {0}.

(7) Any M × M matrix U whose entries are tempered distributions in Rn and which is
a fundamental solution of the system L in Rn is of the form U = E + Q where E is
as in (2.9.25)-(2.9.26) and Q is an M × M matrix whose entries are polynomials
in Rn and whose columns, Qk , k ∈ {1, . . . , M }, satisfy the pointwise equations
LQk = 0 ∈ C M in Rn for each k ∈ {1, . . . , M }.
Collectively, Theorem 2.3.2, Theorem 2.4.1, and Theorem 2.5.1 permit us to
deal with integral operators on uniformly rectifiable sets whose kernels consists
630 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

of derivatives of the fundamental solution of a higher-order system of differential

operators, of the sort described in the corollary below.

Corollary 2.9.4 Fix n, m, M ∈ N with n ≥ 2, and suppose L is an M × M system in

Rn of order 2m of the form

L= ∂ α Aαβ ∂ β, where Aαβ ∈ C M×M , (2.9.36)
|α |= |β |=m

satisfying the weak ellipticity condition

⎡  ⎤
⎢ ⎥
⎢
det ⎢ ξ α+β
Aαβ ⎥⎥  0, ∀ξ ∈ Rn \ {0}. (2.9.37)
⎢ |α |= |β |=m ⎥
⎣ ⎦
Denote by E L the fundamental solution of L defined as in (2.9.25)-(2.9.26). Next,
assume Ω ⊆ Rn is a nonempty open set with the property that ∂Ω is a UR set; in
particular, Ω is a set of locally finite perimeter. Abbreviate σ := H n−1 ∂Ω and
denote by ν the geometric measure theoretic outward unit normal to Ω.
Finally, fix a multi-index γ ∈ N0n with |γ| = 2m−1 and define the singular integral
operator acting on C M -valued functions
  σ(x)   M
f ∈ L 1 ∂Ω, (2.9.38)
1 + |x| n−1
according to
∫
(TL,γ f )(x) := (∂γ E L )(x − y) f (y) dσ(y), ∀x ∈ Ω, (2.9.39)
∂Ω

along with its principal-value version

∫
(TL,γ f )(x) := lim+ (∂γ E L )(x − y) f (y) dσ(y) for σ-a.e. x ∈ ∂Ω. (2.9.40)
ε→0
y ∈∂Ω
|x−y |>ε

Then
for each function f as in (2.9.38), the limit defining
(TL,γ f )(x) in (2.9.40) exists at σ-a.e. point x ∈ ∂Ω (2.9.41)

and the induced operators

M M
TL,γ : L p (∂Ω, σ) −→ L p (∂Ω, σ) , p ∈ (1, ∞), (2.9.42)
M M
TL,γ : L 1 (∂Ω, σ) −→ L 1,∞ (∂Ω, σ) , (2.9.43)

are well-defined, linear, and bounded. The operator (2.9.42) further extends to a
linear and bounded mapping
2.9 Some Applications to Singular Integrals Associated with Elliptic Differential Operators 631
M M
TL,γ : H p (∂Ω, σ) −→ L p (∂Ω, σ) for n−1
n < p ≤ 1, (2.9.44)

and, corresponding to the end-point p = ∞,

TL,γ : L ∞ (∂Ω, σ)
M M
−→ BMO(∂Ω, σ) . (2.9.45)

Moreover, for each p ∈ (1, ∞) there exists C = C(Ω, L, p, n) ∈ [0, ∞) with the
M
property that for each function f ∈ L p (∂Ω, σ) one has
∫  1/p
|(∇TL,γ f )(x)| p dist(x, ∂Ω) p−1 dx ≤ C f [L p (∂Ω,σ)] M . (2.9.46)
Ω

Also, for each κ > 0 and p ∈ n−1 n , ∞ there exists C = C(Ω, L, p, κ, n) ∈ [0, ∞)
such that
 
Nκ (TL,γ f ) p ≤ C f [L p (∂Ω,σ)] M if 1 < p < ∞, (2.9.47)
L (∂Ω,σ)

 
Nκ (TL,γ f ) 1,∞ ≤ C f [L 1 (∂Ω,σ)] M , (2.9.48)
L (∂Ω,σ)

 
Nκ (TL,γ f ) ≤ C f [H p (∂Ω,σ)] M if n−1
< p ≤ 1. (2.9.49)
L p (∂Ω,σ) n


Finally, for each κ > 0 and each f ∈ L 1 ∂Ω, 1+σ(x)
M
|x | n−1
the following jump-
formula holds at σ-a.e. point x ∈ ∂∗ Ω:

 κ−n.t. ν(x)γ    −1

TL,γ f  (x) = − ν(x)α+β Aαβ f (x) + (TL,γ f )(x). (2.9.50)
∂Ω 2
|α |= |β |=m

Proof Let k be a generic entry in the matrix ∂γ E L . From Theorem 2.9.3 we know that
each such function belongs to 𝒞∞ (Rn \ {0}), and is odd and positive homogeneous
of degree 1 − n in Rn \ {0}. In turn, this permits us to invoke Theorem 2.3.2 and
Theorem 2.4.1 which yield all conclusions in (2.9.41)-(2.9.49). Formula (2.9.50)
is going to be a consequence of Theorem 2.5.1. Concretely, thanks to (2.9.34) and
standard Fourier analysis, for each x ∈ ∂ ∗ Ω we have
 √ 
∂0 .L ν(x)
γ E ν(x) = ( −1) |γ | ν(x)γ E
L

√    −1
= − −1ν(x)γ ν(x)α+β Aαβ . (2.9.51)
|α |= |β |=m

Granted this, the jump-formula (2.9.50) now follows on account of (2.5.4). 

It is worth singling out the formula giving the nontangential traces of generic
derivatives of the single layer potential operator. We do so in the corollary below.
Corollary 2.9.5 Assume Ω ⊆ Rn is a nonempty open set with the property that ∂Ω is
a UR set; in particular, Ω is a set of locally finite perimeter. Let σ := H n−1 ∂Ω and
632 2 Calderón-Zygmund Theory on Uniformly Rectifiable Sets

denote by ν = (ν1, . . . , νn ) the geometric measure theoretic outward unit normal

to Ω. Also, let L be a homogeneous, second-order, constant complex coefficient,
weakly elliptic M × M system in Rn and consider the matrix-valued fundamental
solution E = (Eαβ )1≤α,β  ≤M associated with L as in Theorem 1.4.2. Finally, pick a
αβ
coefficient tensor A = ar s 1≤α,β ≤M for which L A = L, and consider the matrix-
1≤r,s ≤n
valued function

B = (bαβ )1≤α,β ≤M : ∂∗ Ω → C M×M defined σ-a.e. on ∂∗ Ω as

αβ −1
(2.9.52)
B := [L(ν)]−1 = − (ar s νr νs )1≤α,β ≤M .

For each α, β ∈ {1, . . . , M } and r ∈ {1, . . . , n} introduce an operator, denoted


by (∂r 𝒮αβ ), which acts on each function f ∈ L 1 ∂Ω, 1+σ(x) |x | n−1
according to38
∫
(∂r 𝒮αβ ) f (x) := (∂r Eαβ )(x − y) f (y) dσ(y) for all x ∈ Ω. (2.9.53)
∂Ω

Then for each indexes α, β ∈ {1, . . . , M } and r ∈ {1, . . . , n}, each function

f ∈ L 1 ∂Ω, 1+σ(x) |x | n−1
, and each aperture parameter κ > 0, at σ-a.e. point x ∈ ∂∗ Ω
one has
κ−n.t. ∫
 1
(∂r 𝒮αβ ) f  (x) = νr (x)bαβ (x) f (x) + lim+ (∂r Eαβ )(x − y) f (y) dσ(y).
∂Ω 2 ε→0
y ∈∂Ω
|x−y |>ε
(2.9.54)

Proof This is an immediate consequence of (2.9.50). 

38 we emphasize that (∂r 𝒮αβ ) f is simply a convenient piece of notation, and should not be
interpreted as ∂r (𝒮αβ f )
Chapter 3
Quantitative Fatou-Type Theorems in Arbitrary
UR Domains

The trademark blueprint of a Fatou-type theorem is that size/integrability properties

of the nontangential maximal operator for a null-solution of an elliptic equation in
a certain domain implies the a.e. existence of the pointwise nontangential boundary
trace of said function. It is natural to call such a theorem quantitative if the boundary
trace does not just simply exist, but also encodes significant information regarding
the size and regularity of the original function.
One of the main results in this chapter is Theorem 3.1.6, where we establish a
quantitative Fatou-type theorem for any null-solution u of an injectively elliptic first-
order (homogeneous, constant complex coefficient) system of differential operators in
an arbitrary UR domain Ω ⊆ Rn , assuming that the nontangential maximal operator
of u is p-th power integrable on ∂Ω (with   to the Hausdorff measure H )
respect n−1

for some integrability exponent p ∈ n−1 n , ∞ . Such a result has a wide range of
applications, including the theory of Hardy spaces associated with injectively elliptic
first-order systems in UR domains. We also prove here a quantitative Fatou-type
theorem for the gradient of null-solutions of second-order systems in UR domains
(cf. Theorem 3.3.4).
Quantitative Fatou-type result for null-solutions u of weakly elliptic second-order
system in arbitrary UR domains Ω ⊆ Rn , in which integrability properties of the
nontangential maximal functions of u and ∇u together ensure the existence of non-
tangential pointwise traces of both u and ∇u on ∂Ω, are presented in Theorem 3.3.9.
In addition we show that, for a null-solution of a weakly elliptic system, having suf-
ficient regularity on the Sobolev/Besov/Triebel-Lizorkin scales in a given bounded
two-sided NTA domain with an Ahlfors regular boundary guarantees the existence
of its nontangential pointwise trace on the boundary of said domain. In the final
section in this chapter we establish Fatou-type theorems for first and second order
differential operators on Riemannian manifolds.

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 633
D. Mitrea et al., Geometric Harmonic Analysis III, Developments in Mathematics 74,
https://doi.org/10.1007/978-3-031-22735-6_3
634 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

3.1 Quantitative Fatou-Type Theorems in UR Domains for

First-Order Systems

Recall the class of injectively elliptic first-order systems from Definition 1.3.4. The
integral representation formula in the theorem below is central to the considerations
in this section.

Theorem 3.1.1 Fix n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an Ahlfors regular

domain. Denote by ν the geometric measure theoretic outward unit normal to Ω and
abbreviate σ := H n−1 ∂Ω. Next, consider a homogeneous, first-order N × M system
D with constant complex coefficients in Rn (where N, M ∈ N) which is injectively
 is a homogeneous first-order M × N system
elliptic (cf. (1.3.18)), and suppose D
with constant complex coefficients in Rn which complements D (i.e., (1.3.21) holds).
In particular, since L := DD is a weakly elliptic second-order M × M system in Rn
it is meaningful to consider the matrix-valued fundamental solution E L  associated
with L  as in Theorem 1.4.2. Also, with D acting on the columns of E L  define

 := D
E  E L  . (3.1.1)

Next, suppose u : Ω → C M is a vector-valued function with Lebesgue measurable

components satisfying, for an aperture parameter κ ∈ (0, ∞),
 
Nκ u ∈ L p (∂Ω, σ) for some p ∈ n−1n ,∞ . (3.1.2)

In addition, with Du considered in the sense of distributions in Ω, assume that

 1 N
Du ∈ Lloc (Ω, L n ) and P(Du) ∈ L p (∂Ω, σ). (3.1.3)
   N
Then Sym D; ν • u belongs to the Hardy space H p (∂Ω, σ) and there exists
some constant C = C(Ω, D, p, κ) ∈ (0, ∞), independent of u, such that
       
Sym D; ν • u p ≤ C Nκ u L p (∂Ω,σ) + C P(Du) L p (∂Ω,σ) . (3.1.4)
[H (∂Ω,σ)] N
 n−1 
Moreover, if p ∈ n , 1 and ∂Ω is unbounded then for L n -a.e. x ∈ Rn \ ∂Ω with
the property that
∫
|(Du)(y)|
dy < +∞ (3.1.5)
Ω |x − y| n−1
which is also assumed to be a Lebesgue point for the function u in the case when
x ∈ Ω, one has
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 635
   
 (x − ·)
E , (−i)Sym D; ν • u
∂Ω

∫ 
u(x) if x ∈ Ω,
− 
E (x − y), (Du)(y) dy = (3.1.6)
Ω 0 if x ∈ Rn \ Ω.

The bracket ·, · in the first line of (3.1.6) is viewed as the duality pairing (in the
sense of [113, Theorem
  4.6.1], presently used with Σ := ∂Ω) between the rows of the
M × N matrix E  (x − ·) , each of which belonging to1
∂Ω

⎧  . (n−1)( 1 −1)  N
  N ∗ ⎪
⎨
⎪ 𝒞 p (∂Ω) ∼ if p < 1,
H p (∂Ω, σ) =  N   ∞  N
⎪
⎪ BMO(∂Ω,

⎩ σ) ⊃ L (∂Ω, σ) ∼ if p = 1,
(3.1.7)
   N
and (−i)Sym D; ν • u ∈ H p (∂Ω, σ) . The bracket inside the solid integral in the
second line of (3.1.6) is simply the pointwise (real inner product)
 pairing between
the columns of the M × N matrix E  (x − y) and the vector Du)(y) ∈ C N .
 
A similar integral representation formula holds if p ∈ n−1n , 1 and ∂Ω is bounded,
this time without having to take the equivalence class of E  (x − ·) modulo
∂Ω
constants in the first line of (3.1.6), provided one also assumes that there exists
λ ∈ (1, ∞) such that
⨏
|u| dL n = o(1) as R → ∞ (3.1.8)
B(0,λ R)\B(0,R)

in the case when Ω happens to be an exterior domain.

Finally, an analogous integral representation formula holds when p ∈ (1, ∞),
this time interpreting the duality pairings as integration on ∂Ω with respect to σ,
i.e.,
∫
   
E (x − y),(−i) Sym D; ν • u (y) dσ(y)
∂Ω
∫ 
u(x) if x ∈ Ω,
−  (x − y), (Du)(y) dy =
E (3.1.9)
Ω 0 if x ∈ Rn \ Ω,

with the same caveat as before when Ω is an exterior domain.

Proof The fact that we have the membership

   N
Sym D; ν • u ∈ H p (∂Ω, σ) (3.1.10)

in the quantitative manner, indicated in (3.1.4), is a direct consequence of [113,

Corollary 10.2.13] (used with q := p), [112, (6.2.25)], and [113, (4.2.25)]. Let us
also remark here that (3.1.2) and [112, Lemma 8.3.1] ensure that

1 membership implied by the estimates in Theorem 1.4.2

636 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
 ∞ M
u ∈ Lloc (Ω, L n ) . (3.1.11)

To proceed, we distinguish two cases.

 
Case I: Suppose p ∈ n−1 n , 1 . To fix ideas, assume ∂Ω is unbounded, and pick a
Lebesgue point x ∈ Ω for the function u, with the additional property that (3.1.5)
is satisfied. Also, consider a real-valued function η ∈ 𝒞∞ (R ) satisfying η ≡ 0 on
n

B(0, 1) and η ≡ 1 on Rn \ B(0, 2). For each number ε ∈ 0, 12 dist(x, ∂Ω) define
ηε : Rn → R by setting
y − x
ηε (y) := η for every y ∈ Rn . (3.1.12)
ε
Then
ηε ∈ 𝒞∞ (Rn ) is a bounded function,
lim ηε (y) = 1 for every y ∈ Rn \ {x},
ε→0+
1 − ηε ∈ 𝒞∞
c (Ω), ηε ≡ 0 on B(x, ε), (3.1.13)
supp (∇ηε ) ⊆ B(x, 2ε) \ B(x, ε), and also
supy ∈Rn |(∇ j ηε )(y)| ≤ Cε −j for all j ∈ N0,

for some constant C ∈ (0, ∞) independent of the parameter ε. In particular, the

 
 − ·)ηε belongs to 𝒞∞ (Rn ) N ×M and coincides with E(x
function E(x  − ·) near
∂Ω. As such,
   
 (x − ·)
E ,(−i)Sym D; ν • u
∂Ω
   
=  (x − ·)
ηε E , (−i)Sym D; ν • u . (3.1.14)
∂Ω

To proceed, fix a dilation factor λ ∈ (1, ∞) and choose a real-valued function

θ ∈ 𝒞∞ c (R ) satisfying 0 ≤ θ ≤ 1 as well as θ ≡ 1 on B(0, 1), and θ ≡ 0 on
n

Rn \ B(0, λ). For each given R > 0, define θ R (y) := θ(y/R) at every point y ∈ Rn .
Fix now some number ε ∈ 0, 12 dist(x, ∂Ω) and observe that
     
lim  (x − ·)
θ R ηε E =  (x − ·)
ηε E
R→∞ ∂Ω ∂Ω

⎧ .   M×N
⎪
⎨ 𝒞(n−1)(1/p−1) (∂Ω) ∼
⎪ if p < 1, (3.1.15)
weak-∗ in
⎪ BMO(∂Ω,
⎪  M×N
⎩  σ) if p = 1.

This is implied by the general weak-∗ convergence results established in [113,

Lemma 4.8.4] and, respectively, [113, Lemma 4.8.1] (in the latter case also bearing
in mind the trivial bounded embedding L ∞ (∂Ω, σ) → BMO(∂Ω, σ)). At this stage,
from (3.1.14), (3.1.15), (3.1.10), the duality result from [113, Theorem 4.6.1], and
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 637

[113, Lemma 4.6.4] we conclude that for each index α ∈ {1, . . . , M } we have
   
 (x − ·)
E , (−i)Sym D; ν • u (3.1.16)
∂Ω α
   
=  )α . (x − ·)
(E , (−i)Sym D; ν • u
∂Ω

   
= lim [(Lip c (∂Ω)) ] N
. α (x − ·)
(−i)Sym D; ν • u, θ R ηε E N.
R→∞ ∂Ω [Lip c (∂Ω)]

Based on (3.1.16) and [113, Proposition 10.2.11] (presently used with F := u) we

obtain (cf. [113, (10.2.94)-(10.2.95)])
   
 (x − ·)
E , (−i)Sym D; ν • u
∂Ω
∫
= lim  − ·) dL n
Du , θ R ηε E(x
R→∞ Ω
∫
 
− lim  − ·) dL n,
u, D θ R ηε E(x (3.1.17)
R→∞ Ω

 − ·). Thanks to (3.1.5),

where D acts on the columns of the matrix θ R ηε E(x
(1.4.24), and Lebesgue’s Dominated Convergence Theorem,
∫ ∫
lim+ lim 
Du , θ R ηε E(x − ·) dL =
n  − ·) dL n
Du , E(x
ε→0 R→∞ Ω Ω
∫
=  (x − ·), Du dL n .
E (3.1.18)
Ω

In order to be specific, suppose D is expressed as in (1.3.19). Upon recalling (1.3.22),

from (3.1.17), (3.1.18), and (1.3.22) we then deduce that
∫
   

E(x − ·) ∂Ω , (−i)Sym D; ν • u −  (x − ·), Du dL n
E
Ω
 ∫ 
αβ  
αγ (x − ·) dL n
= lim+ lim uβ a j ∂ j θ R ηε E . (3.1.19)
ε→0 R→∞ Ω 1≤γ ≤M

Going further, observe that (1.6.7) entails

αβ  αγ (x − ·) = 0 in Rn \ {x},
a j (∂j E) (3.1.20)

for each β, γ ∈ {1, . . . , M }. Make use of (3.1.20) to split the last expression in
(3.1.19) as
 ∫ 
αβ  
αγ (x − ·) dL n
lim+ lim uβ a j ∂ j θ R ηε E = I + II, (3.1.21)
ε→0 R→∞ Ω 1≤γ ≤M
638 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

where
 ∫ 
αβ αγ (x − ·) dL n
I := lim+ lim uβ a j (∂j θ R )ηε E
ε→0 R→∞ Ω 1≤γ ≤M
 ∫ 
αβ αγ (x − ·) dL n
= lim uβ a j (∂j θ R )E , (3.1.22)
R→∞ Ω 1≤γ ≤M

and
 ∫ 
αβ αγ (x − ·) dL n
II := lim+ lim uβ a j θ R (∂j ηε )E
ε→0 R→∞ Ω 1≤γ ≤M
 ∫ 
αβ αγ (x − ·) dL n
= lim+ uβ a j (∂j ηε )E . (3.1.23)
ε→0 Ω 1≤γ ≤M

Note that since we are presently assuming

 ∂Ω to be unbounded,
 estimate [112,
(8.6.51)] is applicable with E := Ω ∩ B(0, λ R) \ B(0, R) and p := (n − 1)/n.
Together with Hölder’s inequality and (1.4.24), this gives
 ∫ 
−n
I ≤ C lim sup R |u| dL n
R→∞ Ω∩[B(0,λ R)\B(0,R)]

 ∫ np
 n−1
np
∫  1− n−1
np

−n
≤ C lim sup R |u| n−1 dL n
dL n
R→∞ Ω∩[B(0,λ R)\B(0,R)] B(0,λ R)
 
≤ C lim sup R−(n−1)/p Nκ u L p (∂Ω,σ) = 0. (3.1.24)
R→∞

Hence,

I = 0. (3.1.25)

Next, we decompose

II = II(a) + II(b) (3.1.26)

where
 ∫ 
  αβ
(a)
II := lim+ αγ (x − y) dy
uβ (y) − uβ (x) a j (∂j ηε )(y)E ,
ε→0 Ω 1≤γ ≤M
 ∫ 
(b) αβ αγ (x − y) dy
II := uβ (x) lim+ a j (∂j ηε )(y)E . (3.1.27)
ε→0 Ω 1≤γ ≤M

In relation to II(a) observe that since x ∈ Ω is a Lebesgue point for the function u
we may estimate (using (3.1.13) and (1.4.24))
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 639
⨏
II(a) ≤ C · lim sup u(y) − u(x) dy = 0. (3.1.28)
ε→0+ B(x,2ε)

Also, for each β, γ ∈ {1, . . . , M } fixed, keeping (3.1.13) in mind permits us to

compute
∫
αβ 
lim+ aj E αγ (x − y)(∂ j ηε )(y) dy
ε→0 Ω
∫
αβ 
= lim+ aj E αγ (x − y)∂ j (ηε − 1)(y) dy
ε→0 Ω

αβ 
= lim+ D  (Ω) a j Eαγ (x − ·), ∂ j (ηε − 1) D(Ω)
ε→0

αβ   
= − lim+ D  (Ω) a j ∂j Eαγ (x − ·) , ηε − 1 D(Ω)
ε→0

  
 − ·)
= lim+ D  (Ω) D E(x
ε→0 βγ, ηε − 1 D(Ω)

= − lim+ δβγ (ηε − 1)(x) = δβγ . (3.1.29)

ε→0

where the penultimate step makes use of (1.6.7). From (3.1.26)-(3.1.29) we then
conclude that

II = u(x). (3.1.30)

Combining (3.1.19)-(3.1.21) with (3.1.25) and (3.1.30) proves (3.1.6) in the case
when ∂Ω is unbounded. The case when Ω is bounded is dealt with in an absolutely
similar fashion. Finally, in the case when Ω is an exterior domain, the same type of
reasoning applies, this time with (3.1.25) justified based on the additional assumption
made in (3.1.8) and the first line of (3.1.24).
Case II: Suppose p ∈ (1, ∞). By and large, we reason as before, bearing in mind
that (3.1.10) now becomes (cf. [112, (3.6.27)])
   N
Sym D; ν • u ∈ L p (∂Ω, σ) , (3.1.31)

that in the current case all duality pairings used in the past are now ordinary inte-
gration on ∂Ω with respect to σ, and that ∫if p is the Hölder conjugate exponent
 − ·)| p dσ < +∞. The integral
of p then (1.4.24) and [112, (7.2.5)] imply ∂Ω | E(x
representation formula (3.1.9) eventually follows. 
There is also a natural version of Theorem 3.1.1 which employs the more inclusive
scale of Lorentz spaces in place of Lebesgue spaces in the hypotheses made on the
given function u, of the sort described below.
Theorem 3.1.2 Retain the assumptions from Theorem 3.1.1 on the set Ω, as well as
 and continue to define E
the first-order systems D, D,  as in (3.1.1). This time, assume
640 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

u : Ω → C M is a vector-valued function with Lebesgue measurable components

satisfying, for an aperture parameter κ ∈ (0, ∞),
 
Nκ u ∈ L p,q (∂Ω, σ) for some p ∈ n−1n , ∞ and q ∈ (0, ∞),
 1 N (3.1.32)
Du ∈ Lloc (Ω, L n ) and P(Du) ∈ L p,q (∂Ω, σ).
   N
Then Sym D; ν • u belongs to the Lorentz-based Hardy space H p,q (∂Ω, σ)
and there exists some constant C = C(Ω, D, p, q, κ) ∈ (0, ∞), independent of u,
such that
   
Sym D; ν •u p, q N
[H (∂Ω,σ)]
   
≤ C Nκ u L p, q (∂Ω,σ) + C P(Du) L p, q (∂Ω,σ) . (3.1.33)

In addition,
  N ×M
 − ·)
E(x ∈ H p,q (∂Ω, σ)∗ for each x ∈ Rn \ ∂Ω, (3.1.34)
∂Ω

and for L n -a.e. x ∈ Rn \ ∂Ω with the property that

∫
|(Du)(y)|
dy < +∞ (3.1.35)
Ω |x − y| n−1
which is also assumed to be a Lebesgue point for the function u in the case when
x ∈ Ω one has
 

([H p, q (∂Ω,σ)] N )∗ E (x − ·) ∂Ω, (−i)Sym D; ν • u [H p, q (∂Ω,σ)] N (3.1.36)

∫ 
u(x) if x ∈ Ω,
− 
E (x − y), (Du)(y) dy =
Ω 0 if x ∈ Rn \ Ω

provided u also vanishes at infinity (in the sense of Definition 1.6.3) in the case when
Ω happens to be an exterior domain.
Proof We follow in broad outline the argument used in the proof of Theorem  3.1.1,

so we shall only discuss the main differences. First, that actually Sym D; ν • u
 p,q N
belongs to the Lorentz-based Hardy space H (∂Ω, σ) and the estimate in
(3.1.33) holds is seen from [113, Corollary 10.2.13]. Second, the membership in
(3.1.34) is a consequence of [113, (4.7.2)], whose applicability in the present setting
is ensured by (3.1.1), Theorem 1.4.2, and (2.4.92). Third, the integral representation
formula (3.1.36) is established much as (3.1.6). In this regard, there are two important
distinctions. Specifically, in place of (3.1.15) we now have
   
lim θ R ηε E (x − ·) = ηε E (x − ·)
R→∞ ∂Ω ∂Ω
(3.1.37)
 p,q  M×N
weak-∗ in H (∂Ω, σ)∗ .
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 641

This is implied by the general weak-∗ convergence result established in [113,

Lemma 4.8.13]. Also, in place of (3.1.16) we now have
 
 

([H p, q (∂Ω,σ)] N )∗ E (x − ·) ∂Ω, (−i)Sym D; ν • u [H p, q (∂Ω,σ)] N (3.1.38)
α
 
 )α . (x − ·) , (−i)Sym D; ν • u
= ([H p, q (∂Ω,σ)] N )∗ (E [H p, q (∂Ω,σ)] N
∂Ω

   
= lim [(Lip c (∂Ω)) ] N
. α (x − ·)
(−i)Sym D; ν • u, θ R ηε E N
R→∞ ∂Ω [Lip c (∂Ω)]

for each α ∈ {1, . . . , M }. This is justified as before, with [113, Lemma 4.8.14]
presently employed in lieu of [113, Lemma 4.6.4]. Finally, to see that (3.1.25) holds
in the case when ∂Ω is unbounded, in place of (3.1.24) we now estimate, with ΩR
abbreviating
 Ω ∩ [B(0,
 λ R) \ B(0, R)] for each radius R > 0, some fixed exponent
r ∈ max{1, q}, ∞ , and with prime indicating Hölder conjugation,
 ∫ 
I ≤ C lim sup R−n |u| dL n
R→∞ Ω∩[B(0,λ R)\B(0,R)]
 
= C lim sup R−n u L 1 (Ω R , L n )
R→∞
 
≤ C lim sup R−n u np
,r
n−1 (Ω, L n )
1Ω R  (
np  
) ,r
R→∞ L L n−1 (Ω, L n )

   1− n−1 
≤ C lim sup R−n u np
,q
n−1 (Ω, L n )
L n (ΩR ) np

R→∞ L

 
≤ C lim sup R−(n−1)/p Nκ u L p, q (∂Ω,σ) = 0, (3.1.39)
R→∞

thanks to [112, (6.2.26), (6.2.40), (6.2.61), (8.6.50)] and (3.1.32). Modulo these
alterations, the argument in the proof of Theorem 3.1.1 goes through, and presently
yields (3.1.36). 

There is actually a more general2 phenomenon at play here, valid for arbitrary (not
necessarily injectively elliptic) first-order systems, which we would like to single out
in the next two propositions.

Proposition 3.1.3 Fix n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an Ahlfors regular

domain. Denote by ν the geometric measure theoretic outward unit normal to Ω and
abbreviate σ := H n−1 ∂Ω. Next, consider a homogeneous, first-order N × M system
D with constant complex coefficients in Rn (where N, M ∈ N) and suppose u is a
vector-valued satisfying, for an aperture parameter κ ∈ (0, ∞),

2 at least in the class of smooth functions

642 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
 M
u ∈ 𝒞∞ (Ω) , Du = 0 in Ω, and
  (3.1.40)
Nκ u ∈ L p (∂Ω, σ) for some p ∈ n−1
n ,∞ ,

as well as ⨏
|u| dL n = o(1) as R → ∞ (3.1.41)
B(0,λ R)\B(0,R)

for some λ ∈ (1, ∞) in the case when Ω happens to be an exterior domain. In

addition, assume that
 N  M
w ∈ D (Ω) has D w ∈ ℰ(Ω) as well as
 N (3.1.42)
w Ω\K ∈ Lip(Ω \ K) for a compact set K ⊆ Ω,

and
w(x) = O(|x| 1−n ) as |x| → ∞
(3.1.43)
in the case when Ω is unbounded.
   N
Then (−i)Sym D; ν • u belongs to the Hardy space H p (∂Ω, σ) . Also, the
κ−n.t.
nontangential trace w ∂Ω
exists at σ-a.e. point on ∂Ω, as a function this trace
belongs to the space

⎧  (n−1)( 1 −1) N
⎪
⎪ 𝒞 p (∂Ω) if ∂Ω compact and p < 1,
  ∗ ⎪⎪
⎨
⎪  
N N
H p (∂Ω, σ) = BMO(∂Ω, σ) if ∂Ω compact and p = 1,
⎪
⎪
⎪
⎪     −1
⎪ L p (∂Ω, σ) N if p > 1 and p := 1 − p1 ,
⎩
(3.1.44)
and, in all cases described above,
  κ−n.t.
(−i)Sym D; ν • u, w ∂Ω
= −[ℰ(Ω)] M u, D w [ℰ (Ω)] M (3.1.45)

where the bracket ·, · in the left-hand side of (3.1.45) is viewed as the duality
pairing (in the sense of [113, Theorem 4.6.1], presently
   N   used with Σ := ∂Ω) between
κ−n.t. N ∗
(−i)Sym D; ν • u ∈ H p (∂Ω, σ) and w ∂Ω ∈ H p (∂Ω, σ) .
 κ−n.t.  κ−n.t.
Finally, if ∂Ω is unbounded then w ∂Ω , the equivalence class of w ∂Ω modulo
constant vectors (in C N ), belongs to
⎧
⎪ .  N
  N ∗ ⎨ 𝒞(n−1)( p −1) (∂Ω) ∼
1
⎪ if p < 1,
H (∂Ω, σ)
p
=  N (3.1.46)
⎪ 
⎪ BMO(∂Ω,
⎩ σ) if p = 1,

and in all these cases one has

   κ−n.t. 
(−i)Sym D; ν • u, w ∂Ω
= −[ℰ(Ω)] M u, D w [ℰ (Ω)] M (3.1.47)
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 643

with a similar interpretation of the duality bracket in the left-hand side of (3.1.47).
The version of Theorem 3.1.1 in which the function u is a smooth null-solution
of the system D becomes a consequence of Proposition 3.1.3 applied with w taken
 − ·).
to be an arbitrary column of the matrix E(x
There is also a natural variant of Proposition 3.1.3 which employs the more
inclusive scale of Lorentz spaces in place of Lebesgue spaces in the hypotheses
made on the given function u. As indicated earlier, this variant yields a direct proof
of the special case of Theorem 3.1.2 in which u is a smooth null-solution of the
system D.
We now present the proof of Proposition 3.1.3.
   N
Proof of Proposition 3.1.3 That (−i)Sym D; ν • u belongs to H p (∂Ω, σ) is
seen from [113, Corollary 10.2.13] (presently employed with q := p), [112, (6.2.25)],
and [113, (4.2.25)]. From (3.1.42)-(3.1.43) and a rudimentary extension theorem3
we see that there exists
 N
a bounded function W ∈ Lip( Ω)
(3.1.48)
such that W Ω\K = w Ω\K .

Based on this and [112, Proposition 8.8.4] we then conclude that the nontangential
κ−n.t.
trace w ∂Ω exists at σ-a.e. point on ∂Ω. In fact,
κ−n.t.  N
w ∂Ω
=W ∂Ω
∈ L ∞ (∂Ω, σ) ∩ Lip(∂Ω) (3.1.49)

and (here (3.1.43) is also relevant) there exists C ∈ (0, ∞) with the property that for
each x ∈ ∂Ω we have
 κ−n.t.  C
w ∂Ω (x) ≤ . (3.1.50)
1 + |x| n−1
Since the function in the right side of (3.1.50) belongs to any L q (∂Ω, σ) with
q ∈ (1, ∞), and since
  
L ∞ (∂Ω, σ) ∩ Lip(∂Ω) ⊆ BMO(∂Ω, σ) ∩ 𝒞α (∂Ω) , (3.1.51)
0<α<1

κ−n.t.  κ−n.t. 
all claims made about the memberships of w ∂Ω and w ∂Ω
in the statement of
the proposition follow.
To proceed, pick a scalar function

η ∈ 𝒞∞ 
c (Ω) such that η ≡ 1 near K ∪ supp (D w). (3.1.52)

Then  N
(1 − η)w = (1 − η)W as distributions in D (Ω) , (3.1.53)

3 for, say, bounded uniformly continuous functions

644 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

since the distributions in questions agree on Ω \ K (by (3.1.48)), and they both vanish
near K (by (3.1.52)). Moreover,
 N   κ−n.t.
(1 − η)W ∈ Lip( Ω) and (1 − η)W ∂Ω
=w ∂Ω (3.1.54)
at σ-a.e. point on ∂Ω,

given that 1 − η ≡ 1 near ∂Ω. Let us choose a scalar function θ ∈ 𝒞∞ c (R ) satisfying

n

0 ≤ θ ≤ 1, θ ≡ 1 on B(0, 1), θ ≡ 0 on R \ B(0, λ) and, for each R > 0, define

n

θ R (x) := θ(x/R) for every x ∈ Rn . Then for each R > 0 it follows that
 N
the function θ R (1 − η)W belongs to Lipc ( Ω) ,
vanishes identically outside a bounded subset of Ω, (3.1.55)
 κ−n.t.    N
and θ R w ∂Ω ) = θ R (1 − η)W ∂Ω ∈ Lipc (∂Ω) .

Also, for each α ∈ (0, 1) we may write

      
  κ−n.t.
sup  θ R (1 − η)W ∂Ω  α N
= sup θ R w ∂Ω  < ∞, (3.1.56)
R>0 [𝒞 (∂Ω)] R>0 [𝒞α (∂Ω)] N

thanks to (3.1.49) and (3.1.51). If ∂Ω is unbounded we therefore have

    κ−n.t. 
lim θ R (1 − η)W ∂Ω = w ∂Ω
R→∞
⎧ .  N
⎪
⎨ 𝒞(n−1)(1/p−1) (∂Ω) ∼
⎪ if p < 1, (3.1.57)
weak-∗ in  
⎪
⎪ BMO(∂Ω,
 N
⎩ σ) if p = 1,

while if ∂Ω is bounded we have

  κ−n.t.
lim θ R (1 − η)W ∂Ω
=w ∂Ω
R→∞
⎧  N
⎪
⎨ 𝒞(n−1)(1/p−1) (∂Ω) if p < 1,
⎪ (3.1.58)
weak-∗ in  
⎪
⎪ BMO(∂Ω, σ) N if p = 1.
⎩
Indeed, all these weak convergence properties are implied by (3.1.57) plus the general
weak-∗ convergence results established in [113, Lemma 4.8.4] and, respectively,
[113, Lemma 4.8.1] (in the latter case also bearing in mind the trivial bounded
embedding L ∞ (∂Ω, σ) → BMO(∂Ω, σ)). Furthermore, in the case when p > 1,

since the function in the right side of (3.1.50) belongs to the space L p (∂Ω, σ) for
  −1
p := 1 − p1 ∈ (1, ∞), Lebesgue’s Dominated Convergence Theorem ensures
that   κ−n.t.   N
lim θ R (1 − η)W ∂Ω = w ∂Ω in L p (∂Ω, σ) . (3.1.59)
R→∞
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 645

For clarity of exposition consider first the case when ∂Ω is unbounded (which
forces Ω to be unbounded). In this scenario, we may write
   κ−n.t. 
[H p (∂Ω,σ)] N (−i)Sym D; ν • u, w ∂Ω ([H p (∂Ω,σ)] N )∗

    
= lim [H p (∂Ω,σ)] N (−i)Sym D; ν • u, θ R (1 − η)W ∂Ω ([H p (∂Ω,σ)] N )∗
R→∞

    
= lim [(Lip c (∂Ω)) ] N (−i)Sym D; ν • u, θ R (1 − η)W ∂Ω [Lip c (∂Ω)] N
R→∞
∫
 
= − lim u, D θ R (1 − η)W dL n, (3.1.60)
R→∞ Ω

where the first equality is a consequence of (3.1.57) and (3.1.59) (depending on how
p compares with 1), the second equality in (3.1.60) is implied by [113, Lemma 4.6.4],
and the third equality in (3.1.60) follows from (3.1.55) and the definition in [113,
Proposition 10.2.11] (cf. [113, (10.2.94)]).
To continue, for each fixed R > 0 we compute
 
D θ R (1 − η)W
   
= [D, θ R ] (1 − η)W + θ R D (1 − η)W
   
= (1 − η)(−i) Sym D ; ∇θ R W + θ R D (1 − η)w
 
= (1 − η)(−i) Sym D ; ∇θ R W + θ R [D, 1 − η]w + θ R (1 − η)D w
   
= (1 − η)(−i) Sym D ; ∇θ R W + θ R (−i) Sym D ; ∇(1 − η) w
   
= (1 − η)(−i) Sym D ; ∇θ R W − θ R (−i) Sym D ; ∇η w
   
= (1 − η)(−i) Sym D ; ∇θ R W − θ R (−i) Sym D ; ∇η W, (3.1.61)

making repeated applications of formula [112, (1.7.20)], the observation that we

currently have (1 − η)D w = 0 since 1 − η ≡ 0 on the support of the distribution
D w, and the fact that ∇η is supported in Ω \ K, an open set where W and w coincide
(cf. (3.1.48)). In turn, we may use (3.1.61) to decompose
∫
 
u, D θ R (1 − η)W dL n = IR + IIR (3.1.62)
Ω

where ∫
 
IR := (−i) u, (1 − η) Sym D ; ∇θ R W dL n, (3.1.63)
Ω
and ∫
 
IIR := i u, θ R Sym D ; ∇η W dL n . (3.1.64)
Ω
646 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

In view of the fact that we are presently assuming the set ∂Ω to be unbounded,
the estimate recorded in [112, (8.6.51)  in Proposition 8.6.3]
 is applicable for the
choice p := (n − 1)/n and E := Ω ∩ B(0, λ R) \ B(0, R) . In concert with Hölder’s
inequality, (3.1.48), and (3.1.43), this gives
∫
IR ≤ CR−n |u| dL n
Ω∩[B(0,λ R)\B(0,R)]

∫ np
 n−1
np
∫  1− n−1
np
≤ CR−n |u| n−1 dL n dL n
Ω∩[B(0,λ R)\B(0,R)] B(0,λ R)

≤ CR−(n−1)/p Nκ u L p (∂Ω,σ) → 0 as R → ∞. (3.1.65)

Thus, on the one hand,

lim IR = 0. (3.1.66)
R→∞

On the other hand, another appeal to [112, (1.7.20)] permits us to compute

 
D w = D (1 − η)w + D (ηw)

= [D, 1 − η]w + (1 − η)D w + D (ηw)

 
= i Sym D ; ∇η w + D (ηw) (3.1.67)

since, as noted earlier, (1 − η)D w = 0 in Ω. Next, we make use of (3.1.67) to write

 
[ℰ(Ω)] M u, D w [ℰ (Ω)] M = [ℰ(Ω)] M u, i Sym D ; ∇η w [ℰ (Ω)] M

+ [ℰ(Ω)] M u, D (ηw) [ℰ (Ω)] M

 
= [ℰ(Ω)] M u, i Sym D ; ∇η W [ℰ (Ω)] M
∫
 
= u, i Sym D ; ∇η W dL n
Ω

= lim IIR, (3.1.68)

R→∞

where we have used the fact that

[ℰ(Ω)] M u, D (ηw) [ℰ (Ω)] M = [ℰ(Ω)] M Du, ηw [ℰ (Ω)] M =0 (3.1.69)

 M
since ηw ∈ ℰ(Ω) and Du = 0 (cf. (3.1.40)), that w = W on the support of ∇η,
the definition of IIR from (3.1.64), and the observation that θ R becomes identically
1 on the support of ∇η when R is large.
Hence, in the case when ∂Ω is unbounded, gathering (3.1.60), (3.1.62)-(3.1.64),
(3.1.66), and (3.1.68) proves (3.1.46). Finally, the case when ∂Ω is bounded is dealt
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 647

with in a similar fashion, with one important difference. Specifically, when Ω is an

exterior domain then (3.1.66) is proved by using (3.1.63), (3.1.43), and (3.1.41). The
proof of Proposition 3.1.3 is therefore complete. 

Here is the second proposition alluded to earlier, a close companion to Propo-

sition 3.1.3 which we shall use later in Corollary 3.1.5 to further expand on Theo-
rem 3.1.1.

Proposition 3.1.4 Pick n ∈ N with n ≥ 2, and assume Ω ⊆ Rn is an Ahlfors regular

domain. Denote by ν the geometric measure theoretic outward unit normal to Ω
and abbreviate σ := H n−1 ∂Ω. Next, suppose X is a Generalized Banach Function
Space on (∂Ω, σ) with the property that

M ∂Ω : X → X and M ∂Ω : X → X
(3.1.70)
are well-defined bounded mappings,

where M ∂Ω is the Hardy-Littlewood maximal operator on ∂Ω, and X is the as-

sociated space of X (cf. [113, Definitions 5.1.4, 5.1.11]). In addition, consider a
homogeneous first-order N × M system D with constant complex coefficients in Rn
(where N, M ∈ N) and assume u is a vector-valued satisfying, for some aperture
parameter κ ∈ (0, ∞),
 M
u ∈ 𝒞∞ (Ω) , Du = 0 in Ω, and Nκ u ∈ X, (3.1.71)

as well as ⨏
|u| dL n = o(1) as R → ∞ (3.1.72)
B(0,λR)\B(0,R)

for some λ ∈ (1, ∞) in the case when Ω happens to be an exterior domain. Finally,
assume that
 N  M
w ∈ D (Ω) has D w ∈ ℰ(Ω) as well as
 N (3.1.73)
w Ω\K ∈ Lip(Ω \ K) for a compact set K ⊆ Ω,

and also
w(x) = O(|x| 1−n ) as |x| → ∞
(3.1.74)
in the case when Ω is unbounded.
 
Then the (vector) distribution (−i)Sym D; ν • u actually belongs to the space
 N
X and satisfies the estimate
     
Sym D; ν • u N ≤ C Nκ u , (3.1.75)
[X] X

κ−n.t.
the nontangential trace w ∂Ω exists at σ-a.e. point on ∂Ω, as a function this trace
 N
belongs to the space X , and one has (with an absolutely convergent integral in
the left side)
648 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
∫
  κ−n.t.
(−i)Sym D; ν • u , w ∂Ω
dσ = −[ℰ(Ω)] M u , D w [ℰ (Ω)] M . (3.1.76)
∂Ω

Proof Combining (3.1.70) with [113, Proposition 5.2.7] shows that

there exist q ∈ (1, ∞) and ε ∈ (0, 1) such that

 σ(x)  (3.1.77)
X → L q ∂Ω , continuously.
1 + |x| n−1−ε
In light of the current assumptions, [113, Proposition 5.2.7] also guarantees that if
q  := (1 − 1/q)−1 ∈ (1, ∞) denotes the Hölder conjugate exponent of q, then

 

L q ∂Ω , (1 + |x| n−1−ε )q −1 σ(x) → X . (3.1.78)

As notes in (3.1.48), there exists

 N
a bounded function W ∈ Lip(Ω)
(3.1.79)
such that W Ω\K
= w Ω\K .

From this, [112, Proposition 8.8.4], and (3.1.74) we then conclude that the nontan-
κ−n.t.
gential trace w ∂Ω exists at σ-a.e. point on ∂Ω, satisfies
κ−n.t.  N
w ∂Ω
=W ∂Ω
∈ Lip(∂Ω) , (3.1.80)

and there exists C ∈ (0, ∞) such that for each x ∈ ∂Ω we have

 κ−n.t.  C
w ∂Ω
(x) ≤ . (3.1.81)
1 + |x| n−1
κ−n.t.
In particular, (3.1.80) implies that w ∂Ω is σ-measurable which, in concert with
(3.1.81), may be used to conclude that
∫
 κ−n.t.  q 
w ∂Ω (x) (1+|x| n−1−ε )q −1 dσ(x)
∂Ω
∫
dσ(x)
≤C <∞ (3.1.82)
∂Ω 1 + |x| n−1+ε(q −1)
since ∂Ω is Ahlfors regular. Together with (3.1.78), this ultimately shows that
κ−n.t.
w ∂Ω
belongs to X . (3.1.83)

To proceed, suppose the homogeneous first-order N × M system D with constant

complex coefficients is expressed as
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 649


n 
αβ
D= a j ∂j 1≤α ≤ N (3.1.84)
j=1 1≤β ≤M

and that the vector-valued function u has scalar components (uβ )1≤β ≤M in Ω. For
each fixed index α ∈ {1, . . . , N }, consider the vector field

Fα = (Fjα )1≤ j ≤n : Ω −→ Cn (3.1.85)

with components given by


M
αβ
Fjα := a j uβ in Ω, for j ∈ {1, . . . , n}. (3.1.86)
β=1

Then
Nκ Fα ∈ X ⊆ Lloc
1
(∂Ω, σ), (3.1.87)
and

n 
M 
n
αβ
div Fα = ∂j Fjα = a j ∂j uβ = (Du)α = 0 in D (Ω). (3.1.88)
j=1 β=1 j=1

 N
Given any ψ = (ψα )1≤α ≤ N ∈ Lipc (∂Ω) , if we let Ψ = (Ψα )1≤α ≤ N be any
C N -valued function satisfying
 N
Ψ ∈ Lip(Ω) , Ψ ∂Ω = ψ, and Ψ ≡ 0
(3.1.89)
outside of some compact subset of Ω,
then according to the definitions given in [113, (10.2.94)] and (A.0.95) we have
(bearing in mind that we are currently assuming Du = 0 and div Fα = 0 for each α)
∫ ∫
αβ
(−i)Sym(D; ν) • u, ψ = − u, D Ψ dL n = a j uβ (∂j Ψα ) dL n
Ω Ω
∫
= Fα · ∇Ψα dL n = ν • Fα, ψα , (3.1.90)
Ω

where the brackets stand for (vector) distributional pairing on ∂Ω. In view of the
arbitrariness of ψ, this proves that
   N
(−i)Sym(D; ν) • u = ν • Fα 1≤α ≤ N ∈ Lipc (∂Ω) . (3.1.91)

Having established this, from [113, Proposition 10.2.6] and (3.1.97) we then conclude
that    N
Sym D; ν • u belongs to X (3.1.92)
and the estimate claimed in (3.1.75) holds. As in the proof of Proposition 3.1.3, pick
a scalar function
650 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

η ∈ 𝒞∞ 
c (Ω) such that η ≡ 1 near K ∪ supp (D w). (3.1.93)

and also choose a scalar function θ ∈ 𝒞∞ c (R ) satisfying 0 ≤ θ ≤ 1, as well as θ ≡ 1

n

on B(0, 1), and θ ≡ 0 on R \ B(0, λ). For each given dilation factor R > 0, define
n

θ R (x) := θ(x/R) at every point x ∈ Rn . Then

∫
   κ−n.t. 
(−i)Sym D; ν • u , w ∂Ω dσ
∂Ω
∫  
  
= lim (−i)Sym D; ν • u , θ R (1 − η)W ∂Ω
dσ
R→∞ ∂Ω
    
= lim [(Lip c (∂Ω)) ] N (−i)Sym D; ν • u , θ R (1 − η)W ∂Ω [Lip c (∂Ω)] N
R→∞
∫
 
= − lim u , D θ R (1 − η)W dL n . (3.1.94)
R→∞ Ω

Above, the first equality is a consequence of (3.1.83), (3.1.92), the generalized Hölder
inequality from [113, Proposition 5.1.12], and Lebesgue’s Dominated Convergence
Theorem. The second equality in (3.1.94) is implied by (3.1.55) and [112, Proposi-
tion 4.1.4]. The third equality in (3.1.94) follows from (3.1.55) and the definition in
[113, Proposition 10.2.11, (10.2.94)].
Formula (3.1.94) should be compared with (3.1.60). From this point on we follow
the proof of Proposition 3.1.3 with one key alteration. Specifically, in the case when
∂Ω is unbounded, in place of (3.1.65) we now make use of Hölder’s inequality, [112,
Proposition 8.6.3], [112, (8.1.17)], the last membership in (3.1.71), and (3.1.77) to
estimate (with q and ε as in (3.1.77)):
∫
IR ≤ CR−n |u| dL n
Ω∩[B(0,λR)\B(0,R)]

∫ nq
 n−1
nq
∫  1− n−1
nq
≤ CR−n |u| n−1 dL n dL n
Ω∩[B(0,λR)\B(0,R)] B(0,λR)

n−1
∫  1/q
≤ CR− q |Nκ u| q dσ
∂Ω∩B(0,λ(2+κ)R)

n−1
∫ |(Nκ u)(y)| q  1/q
≤ CR− q Rn−1−ε dσ(y)
∂Ω∩B(0,λ(2+κ)R) 1 + |y| n−1−ε
ε  
≤ CR− q Nκ u L q  ∂Ω, σ(y)
 ≤ CR− qε Nκ uX
1+|y | n−1−ε

= o(1) as R → ∞. (3.1.95)
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 651

Consequently, (3.1.66) continues to be true in the current setting. Having established

this, the rest of the argument in the proof of Proposition 3.1.3 goes through and gives
(3.1.76). 
As promised earlier, we shall now employ Proposition 3.1.4 to produce the fol-
lowing version of Theorem 3.1.1 with the role of the Lebesgue spaces played by
Generalized Banach Function Spaces.
Corollary 3.1.5 Pick n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an Ahlfors regular
domain. Denote by ν the geometric measure theoretic outward unit normal to Ω
and abbreviate σ := H n−1 ∂Ω. Consider a homogeneous, first-order N × M system
D with constant complex coefficients in Rn (where N, M ∈ N) which is injectively
 is a homogeneous first-order M × N system
elliptic (cf. (1.3.18)), and suppose D
with constant complex coefficients in Rn which complements D (i.e., (1.3.21) holds).
In particular, since L := DD is a weakly elliptic second-order M × M system in Rn
it is meaningful to consider the matrix-valued fundamental solution E L  associated
with L  as in Theorem 1.4.2. With D  acting on the columns of E L  , define

 := D
E  E L  . (3.1.96)

Next, suppose X is a Generalized Banach Function Space on (∂Ω, σ) with the

property that
M ∂Ω : X → X and M ∂Ω : X → X
(3.1.97)
are well-defined bounded mappings,
where M ∂Ω is the Hardy-Littlewood maximal operator on ∂Ω, and X is the asso-
ciated space of X (cf. [113, Definitions 5.1.4, 5.1.11]). Finally, let u : Ω → C M be
a vector-valued function with Lebesgue measurable components satisfying, for an
aperture parameter κ ∈ (0, ∞),

Nκ u ∈ X and Du = 0 in [D (Ω)] N . (3.1.98)

   N
Then the distribution Sym D; ν • u actually belongs to the space X and there
exists some constant C ∈ (0, ∞), which depends only on Ω, D, X, and κ, such that
     
Sym D; ν • u N ≤ C Nκ u . (3.1.99)
[X] X

Moreover, with an absolutely convergent integral in the right side one has
∫
   
u(x) =   (x − y) , (−i) Sym D; ν • u (y) dσ(y) ∀ x ∈ Ω, (3.1.100)
E
∂Ω

if one also assumes that there exists λ ∈ (1, ∞) such that

⨏
|u| dL n = o(1) as R → ∞ (3.1.101)
B(0,λR)\B(0,R)

in the case when Ω happens to be an exterior domain.

652 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

Proof All claims follow from Proposition 3.1.4 and Theorem 1.4.2 (also bearing
in mind the elliptic regularity result recorded in [112, (6.5.40)]) taking the vector
 − ·).
distribution w to be an arbitrary column of the matrix E(x 

We are now in a position to discuss a rather powerful and elegant quantitative

Fatou-type theorem, for null-solutions of injectively elliptic first-order systems in
arbitrary UR domains in Rn . At this level of generality, such a result is the first of its
kind. In particular, we wish to stress that the underlying domain may fail to satisfy a
corkscrew condition (which has typically been “de rigueur” even in the scalar case),
and may even fail to be n-thick (in the sense of [112, Definition 5.1.1]). The “bullet
product” technology, initiated in [112, §4.2] (cf. [112, Proposition 4.2.3]) and subse-
quently developed more fully in [113, §10.2] (starting with [113, Theorem 10.2.1]),
is key to the proof of our result, even though the qualitative aspect of the theorem
(i.e., the a.e. existence of the nontangential pointwise boundary trace; see item (i)
below) does not involve this notion in any way. Finally, we note that for the member-
ship of the nontangential
 maximal operator to Lebesgue spaces arbitrary exponents
p ∈ n−1 n , ∞ are allowed, and that said membership may also be prescribed in large
classes of Generalized Banach Function Spaces.

Theorem 3.1.6 Fix n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an arbitrary UR

domain. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic
outward unit normal to Ω. Next, consider an injectively elliptic homogeneous first-
order N × M system D with constant complex coefficients in Rn (where N, M ∈ N),
and suppose u : Ω → C M is a vector-valued function with Lebesgue measurable
components satisfying, for some aperture parameter κ ∈ (0, ∞),
 
Nκ u ∈ L p (∂Ω, σ) with p ∈ n−1
n ,∞ ,
(3.1.102)
and Du = 0 in [D (Ω)] N .

Then the following statements are true.

(i) The nontangential boundary trace

κ−n.t.
u ∂Ω exists (in C M ) at σ-a.e. point on ∂Ω and
κ−n.t.
the function u ∂Ω is σ-measurable on ∂Ω;
κ−n.t.  M (3.1.103)
also, u ∂Ω belongs to the space L p (∂Ω, σ)
 κ−n.t. 
and one has u ∂Ω [L p (∂Ω,σ)] M ≤ Nκ u L p (∂Ω,σ) .

Moreover, for any other given aperture parameter κ  ∈ (0, ∞) the nontangential
κ  −n.t. κ−n.t.
boundary trace u ∂Ω also exists, and agrees with u ∂Ω , at σ-a.e. point on ∂Ω.
κ−n.t.
Henceforth, it is therefore meaningful to drop the dependence of u ∂Ω on the
n.t.
aperture parameter κ, and simply denote this function by u ∂Ω .
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 653

(ii) Suppose p ∈ (1, ∞) and, in the case when Ω is an exterior domain, make the
additional assumption that u vanishes at infinity (in the sense of Definition 1.6.3).
Then one has
      n.t.  
Nκ u p ≈ Sym D; ν u ∂Ω [L p (∂Ω,σ)] N
L (∂Ω,σ)
 n.t. 
≈ u ∂Ω [L p (∂Ω,σ)] M . (3.1.104)

 is a homogeneous first-order M × N system with constant complex

Moreover, if D
coefficients in Rn which complements D (i.e., (1.3.21) holds) and D, K are the
boundary double layer potential operators associated with the weakly elliptic
second-order M × M system L := DD  and the set Ω as in [115, §1.4] then
 n.t.     n.t. 
u = D u ∂Ω in Ω and − 12 I + K u ∂Ω = 0 on ∂Ω. (3.1.105)

In addition,
 ξ) = I M×M for each ξ ∈ Rn \ {0} (in
if Sym(D; ξ)[L(ξ)]−1 Sym( D;

particular, if D and D commute), then the double layer D has the
   M
property that D(D f ) = 0 in Ω for each f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
.
(3.1.106)
Furthermore, for each q ∈ (1, ∞) one has
n.t.  M
u ∂Ω ∈ L q (∂Ω, σ) ⇐⇒ Nκ u ∈ L q (∂Ω, σ) (3.1.107)

and
n.t.  p,q M
u ∂Ω ∈ L1 (∂Ω, σ) ⇐⇒ Nκ (∇u) ∈ L q (∂Ω, σ), (3.1.108)
n.t.
in a quantitative fashion. In particular, the nontangential pointwise trace u ∂Ω
 p M
belongs to the boundary Sobolev space L1 (∂Ω, σ) if and only if Nκ (∇u)
belongs to L p (∂Ω, σ), once again in a quantitative fashion, namely
 n.t.     
u  p ≈ Nκ u L p (∂Ω,σ) + Nκ (∇u) L p (∂Ω,σ), (3.1.109)
∂Ω [L (∂Ω,σ)] M
1

where the proportionality constants are independent of u.

(iii) In the case p = 1, one has
   n.t.   N
Sym D; ν u ∂Ω ∈ H 1 (∂Ω, σ) (3.1.110)

and, for some constant C = C(Ω, D, κ) ∈ (0, ∞) independent of u,

    n.t.    
Sym D; ν u  ≤ C Nκ u L 1 (∂Ω,σ) . (3.1.111)
∂Ω [H 1 (∂Ω,σ)] N
654 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

In fact, one also has

 n.t.        n.t.  
u  1 M ≤ Nκ u 1 ≤ C Sym D; ν u 
∂Ω [L (∂Ω,σ)] L (∂Ω,σ) ∂Ω [H 1 (∂Ω,σ)] N
 n.t. 
≤ C u ∂Ω [L 1 (∂Ω,σ)] M (3.1.112)

provided, in the case when Ω is an exterior domain, it is assumed that u vanishes

at infinity (in the sense of Definition 1.6.3).
 
(iv) In the case p ∈ n−1 n , 1 , one has
   N
Sym D; ν • u ∈ H p (∂Ω, σ) and
      (3.1.113)
Sym D; ν • u p ≤ C Nκ u L p (∂Ω,σ)
[H (∂Ω,σ)] N

for some constant C = C(Ω, D, κ, p) ∈ (0, ∞) independent of u.

Henceforth, in the case when Ω is an exterior domain, make the additional
assumption that u vanishes at infinity (in the sense of Definition 1.6.3). Then one
also has
     
Nκ u p ≤ C Sym D; ν • u[H p (∂Ω,σ)] N . (3.1.114)
L (∂Ω,σ)

Furthermore, if D is a homogeneous first-order M × N system with constant

complex coefficients in Rn which complements D (i.e., (1.3.21) holds) then4
        
 ν −1 Sym D;
n.t.
u ∂Ω = Sym DD;  ν H Sym D; ν • u
(3.1.115)
at σ-a.e. point on ∂Ω,

where H is the L p -filtering operator from [113, (4.9.2) in Theorem 4.9.1]

(presently used with Σ := ∂Ω). Consequently, with D∗ denoting the Hermi-
tian adjoint of D, one has5
n.t.     −1      
u ∂Ω = Sym D∗ D; ν Sym D∗ ; ν H Sym D; ν • u
(3.1.116)
at σ-a.e. point on ∂Ω.

In particular, if the operator D is of Dirac type (i.e., if D∗ D has scalar principal

symbol) then
       n.t. 
H Sym D; ν • u = Sym D; ν u ∂Ω
(3.1.117)
at σ-a.e. point on ∂Ω.

 
4 with Sym DD; ξ denoting the principal symbol of DD,
 viewed as a second-order differential
operator
 
5 again, with Sym D ∗ D; ξ denoting the principal symbol of D ∗ D, regarded as a second-order
differential operator
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 655

(v) For each q ∈ (1, ∞) there exists some C = C(Ω, D, p, q, κ) ∈ (0, ∞), indepen-
dent of u, such that the following mixed-norm estimate holds:
∫  p1
∫  qp  
|∇u| dist(·, ∂Ω)
q q−n
dL n
dσ(x) ≤ C Nκ u L p (∂Ω,σ) .
∂Ω Γκ (x)
(3.1.118)
Alternatively, with the L q -based area-function defined as in [113, (9.2.206)],
i.e.,
∫  1/q
(A q,κ u)(x) := |(∇u)(y)| q |x − y| q−n dy , ∀x ∈ ∂Ω, (3.1.119)
Γκ (x)

the estimate in (3.1.118) may be recast as

   
A q,κ u p ≤ C Nκ u L p (∂Ω,σ) . (3.1.120)
L (∂Ω,σ)

In addition, when p ∈ (1, ∞) there exists a constant C = C(Ω, D, p, κ) ∈ (0, ∞)

with the property that
∫  1/p  
|(∇u)(x)| p dist(x, ∂Ω) p−1 dx ≤ C Nκ u L p (∂Ω,σ) . (3.1.121)
Ω

(vi) In the case when Ω is actually a bounded NTA  n domain  with Ahlfors regular
boundary in Rn and p ∈ (1, ∞), for each q ∈ n+1/p , ∞ there exists a constant
C = C(Ω, D, p, q) ∈ (0, ∞) with the property that
   n.t. 
max u[F p, q (Ω)] M , u[B p, p (Ω)] M ≤ C u ∂Ω [L p (∂Ω,σ)] M . (3.1.122)
1/p 1/p

In particular, corresponding to the case when p = q = 2, one has

 κ−n.t. 
u[H 1/2 (Ω)] M ≤ C u ∂Ω [L 2 (∂Ω,σ)] M . (3.1.123)

(vii) Suppose in place of the first line in (3.1.102) one now assumes
 
Nκ u ∈ L p,q (∂Ω, σ) for some p ∈ n−1 n , ∞ and q ∈ (0, ∞). (3.1.124)
κ−n.t.
Then the nontangential boundary trace u ∂Ω exists at σ-a.e. point on ∂Ω, is
independent of the given aperture parameter κ, belongs to the Lorentz space
 p,q M
L (∂Ω, σ) , and satisfies the estimate
 κ−n.t. 
u  p, q ≤ Nκ u L p, q (∂Ω,σ) . (3.1.125)
∂Ω [L (∂Ω,σ)] M

Henceforth, make the additional assumption that u vanishes at infinity (in the
sense of Definition 1.6.3) in the case when Ω is an exterior domain. Then one
has
 656 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
   N
Sym D; ν • u ∈ H p,q (∂Ω, σ) and
     (3.1.126)
Sym D; ν • u p, q 
≈ Nκ u L p, q (∂Ω,σ),
[H (∂Ω,σ)] N

and formulas (3.1.115)-(3.1.117) remain valid, where H is now interpreted as

the L p,q -filtering operator from [113, (4.9.5) in Theorem 4.9.1] (again, used
with Σ := ∂Ω). Also, in the particular case when p ∈ (1, ∞), the formulas in
(3.1.105) continue to hold and one has
      κ−n.t.  
Nκ u ≈ Sym D; ν u ∂Ω [L p, q (∂Ω,σ)] N
L p, q (∂Ω,σ)
 κ−n.t. 
≈ u ∂Ω [L p, q (∂Ω,σ)] M . (3.1.127)

Finally, with the L r -based area-function defined similarly to (3.1.119) for some
r ∈ (1, ∞), one has
   
Ar,κ u p, q ≤ C Nκ u L p, q (∂Ω,σ), (3.1.128)
L (∂Ω,σ)

for some constant C ∈ (0, ∞) independent of u.

(viii) Suppose in place of the first line in (3.1.102) one now assumes

Nκ u ∈ X, (3.1.129)

where X is a Generalized Banach Function Space on (∂Ω, σ) with the property

that
M ∂Ω : X → X and M ∂Ω : X → X
(3.1.130)
are well-defined bounded mappings
(with M ∂Ω denoting the Hardy-Littlewood maximal operator on ∂Ω, and with
X denoting the associated space of X; cf. [113, Definitions 5.1.4, 5.1.11]).
κ−n.t.
Then the nontangential boundary trace u ∂Ω exists at σ-a.e. point on ∂Ω
κ−n.t.
and is independent of the given aperture parameter κ. Also, as a function, u ∂Ω
 M  κ−n.t. 
belongs to the space X , and satisfies u  M ≤ Nκ uX . Furthermore,
∂Ω [X]

     κ−n.t.    N
Sym D; ν • u = Sym D; ν u ∂Ω ∈ X (3.1.131)

and, under the additional assumption that u vanishes at infinity (in the sense of
Definition 1.6.3) in the case when Ω is an exterior domain, one has
      κ−n.t.    κ−n.t. 
Nκ u ≈ Sym D; ν u  N ≈ u  M. (3.1.132)
X ∂Ω [X] ∂Ω [X]

We continue by including several comments which further clarify the context and
scope of the above theorem.
Comment 1. In light of [112, (8.3.6) in Lemma 8.3.2], for each given integrability
exponent p ∈ (1, ∞), the equivalence (cf. (3.1.104))
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 657
   n.t. 
Nκ u ≈ u ∂Ω [L p (∂Ω,σ)] M (3.1.133)
L p (∂Ω,σ)

may be naturally interpreted as an L p -Maximum Principle for null-solutions u of

D in Ω (which also vanish at infinity if Ω is an exterior domain). Likewise, (3.1.109)
p
may be thought of as an L1 -Maximum Principle, while (3.1.127) may be construed
as an L p,q -Maximum Principle. Finally, (3.1.132) contains what may be called the
X-Maximum Principle for null-solutions u of D in Ω, namely
   κ−n.t. 
Nκ u ≈ u  M. (3.1.134)
X ∂Ω [X]

Comment 2. The first identity in (3.1.105) may be regarded as Cauchy’s reproducing

formula. To add substance to this perspective, let D := ∂x + i∂y be the Cauchy-
Riemann operator in the plane, and pick D  := D∗ . Since D∗ = −∂x + i∂y , we
∗
have L := D D = −Δ in R . The fundamental solution associated with L as in
2

Theorem 1.4.2 then becomes E L (x, y) := −(2π)−1 ln x 2 + y 2 . In particular,

1 x 1 y
(∂x E L )(x, y) := − and (∂y E L )(x, y) := − . (3.1.135)
2π x 2 + y 2 2π x 2 + y 2

Expressing D = ∂¯ as a j ∂j with complex coefficients a1 := 1 and a2 := i, it follows

that L = D∗ D = A jk ∂j ∂k , where A jk = −a j ak , for 1 ≤ j, k ≤ 2, i.e.,

A11 = −1, A12 = i, A21 = −i, A22 = −1. (3.1.136)

Then the boundary-to-domain double layer D associated with the coefficient tensor
A = A jk 1≤ j,k ≤2 acts on functions f ∈ L p (∂Ω, σ) with 1 < p < ∞ according to
∫ 
  1 −ν1 (ζ)(z1 − ζ1 ) + iν1 (ζ)(z2 − ζ2 )
D f (z) =
2π |z − ζ | 2
∂Ω
!
−iν2 (ζ)(z1 − ζ1 ) − ν2 (ζ)(z2 − ζ2 )
+ f (ζ) dσ(ζ)
|z − ζ | 2
∫   
1 − ν1 (ζ) + iν2 (ζ) (z1 − ζ1 ) − i(z2 − ζ2 )
= f (ζ) dσ(ζ)
2π |z − ζ | 2
∂Ω
∫   ∫
−1 ν(ζ) z − ζ −1 f (ζ)
=   f (ζ) dσ(ζ) = iν(ζ) dσ(ζ)
2π (z − ζ) z − ζ 2πi ∂Ω z − ζ
∂Ω
∫
1 f (ζ)
= dζ for each z ∈ Ω, (3.1.137)
2πi ∂Ω ζ−z

which is the familiar format of the Cauchy operator on ∂Ω. In this setting, the first
identity in (3.1.105) simply asserts that the holomorphic function u may be recovered
658 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

from its (nontangential) boundary trace via the action of Cauchy operator on the latter
function.
However, as opposed to the case of the Cauchy operator which maps L p -functions
on ∂Ω into holomorphic functions (hence, null-solutions of the Cauchy-Riemann
operator) in Ω, in general it is not true that D(D f ) = 0 in Ω for arbitrary functions
   M
f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 . This being said, we see from (3.1.106) that

if D is an injectively elliptic constant complex coefficient first-order

system in Rn which commutes with its Hermitian adjoint (i.e., one has
D∗ D = DD∗ ) then the double layer D associated with the M × M (3.1.138)
weakly elliptic system L := D∗ D satisfies D(D f ) = 0 in Ω for each
   M
function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1 .

See [67, §4.5] for a multitude of examples of first-order elliptic systems which
commute with their Hermitian adjoints.
Comment 3. Formula (3.1.116),  relating
 the nontangential pointwise trace of u on
∂Ω to the bullet product Sym D; ν • u (itself a distribution in the Hardy space
H p on the boundary, whose definition is inspired by a variational approach) via the
L p -filtering operator H, is remarkable in view of the nature of the objects involved.
  ∗   −1  
In this vein, it is worth pointing out that Sym  D D; ν Sym D∗ ; ν is actually a
left-inverse for the (injective) mapping Sym D; ν .
Comment 4. Specializing Theorem 3.1.6 to the case when n = 2 and D := ∂, the
Cauchy-Riemann operator (hence, M = N = 1), yields the following corollary:

any
∫ holomorphic
p function u in a UR domain Ω ⊆ C satisfying

∂Ω
Nκ u dH 1 < +∞, for some κ ∈ (0, ∞) and p ∈ 1 , ∞ , has
2 (3.1.139)
κ−n.t.
the property that u ∂Ω exists in C at H 1 -a.e. point on ∂Ω.

A similar property is valid if the integrability condition imposed above on the

nontangential maximal function is replaced by membership to a Lorentz space (cf.
(3.1.124)).
A large category of UR domains in the complex plane has been singled out
in [112, Proposition 5.10.7]. While this covers many mundane cases, which arise
routinely in applications, the reader is reminded that the topology of a UR domain
can be quite wild. See [123, §A.4, p. 755] where a class of UR domains in Rn , with
n ≥ 2 arbitrary, of infinite topological type has been produced.
Combining Theorem 1.1.1 with (3.1.139) then ∫ proves
 that if u is a holomorphic
p
function in a UR domain Ω ⊆ C satisfying ∂Ω Nκ u dH 1 < +∞ for some
κ ∈ (0, ∞) and p ∈ (1, ∞), and which also vanishes at infinity in the case when Ω
κ−n.t.
is an exterior domain, then u ∂Ω exists at H 1 -a.e. point on ∂Ω and the following
Cauchy integral representation formula holds:
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 659

∫  κ−n.t. 
1 u ∂Ω (ζ)
u(z) = dζ, ∀z ∈ Ω. (3.1.140)
2πi ∂Ω ζ−z
Let us also point out that a version of the Fatou-type result for holomorphic func-
tions recorded in (3.1.139) continues to hold even in the presence of singularities.
Concretely, we have the following result (which, in particular, allows the considera-
tion of meromorphic functions with singularities contained in a compact subset of
their domain):

if Ω ⊆ C is a UR domain and u ∈ D (Ω)∫ verifies ∂u = 0 in Ω\K, where

 p
K is a compact subset of Ω, as well as ∂Ω NκΩ\K u dH 1 < +∞, for (3.1.141)
  κ−n.t.
some κ > 0 and p ∈ 12 , ∞ , then u ∂Ω exists at H 1 -a.e. point on ∂Ω.

This is seen by applying (3.1.139) to the restriction of u to Ω \ O, where O is

a bounded Lipschitz domain satisfying K ⊆ O and O ⊆ Ω. When used together
with Theorem 1.1.3, the Fatou-type result recorded in (3.1.141) yields a version
of the residue formula (1.1.66) in which the existence of nontangential pointwise
boundary traces need not be assumed a priori, but rather becomes a consequence of
the hypotheses.
Finally, we wish to note that a result in the spirit of (3.1.139), in which we now
allow any integrability exponent in the interval (0, ∞), holds as described below:

given any simply connected UR domain Ω ⊆∫ C along

  with any non-
p
vanishing holomorphic function u in Ω with ∂Ω Nκ u dH 1 < +∞,
κ−n.t. (3.1.142)
for some κ ∈ (0, ∞) and p ∈ (0, ∞), it follows that u ∂Ω exists in C
at H 1 -a.e. point on ∂Ω.

Indeed, this follows by applying (3.1.139) to u1/m , which is a well-defined holomor-

phic function in Ω (cf., e.g., [156, Theorem 13.11, p.274]), with m ∈ N sufficiently
large so that mp > 1/2.
Comment 5. Another natural candidate for the (homogeneous, constant coefficient)
first-order operator with injective principal symbol intervening in the statement of
"
Theorem 3.1.6 is the Dirac operator D := nj=1 e j  ∂j (cf. the discussion in [112,
§6.4]). Upon recalling that Cn -valued functions which are null-solutions of D are
called monogenic, Theorem 3.1.6 implies the following Fatou-type result:

∫any monogenic
p function u in a UR domain Ω ⊆ Rn satisfying
 
∂Ω
Nκ u dH n−1 < +∞, for some κ ∈ (0, ∞) and p ∈ n−1
n ,∞ , (3.1.143)
κ−n.t.
has the property that u ∂Ω exists in Cn at H n−1 -a.e. point on ∂Ω.

A similar result is valid when the above integrability condition imposed on the
nontangential maximal function is replaced by membership to a Lorentz space (cf.
(3.1.124)).
660 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

Comment 6. Holomorphic functions of several complex variables also fit into  the
framework of Theorem 3.1.6 by considering the first-order operator D := ∂z̄ j 1≤ j ≤n .
Since from Example 1.3.10 we know that this is injectively elliptic, Theorem 3.1.6
implies the following Fatou-type result:

∫any holomorphic
p function u in a UR domain Ω ⊆ Cn satisfying 
∂Ω
N κ u dH 2n−1 < +∞, for some κ ∈ (0, ∞) and p ∈ 2n−1 , ∞ ,
2n (3.1.144)
κ−n.t.
has the property that u ∂Ω exists in C at H 2n−1 -a.e. point on ∂Ω.

A similar Fatou-type result is valid if the integrability condition imposed above on

the nontangential maximal function is replaced by membership to a Lorentz space
(cf. (3.1.124)), and also for separately monogenic functions of m Clifford variables
in UR domains in (Rn )m , where m ∈ N is arbitrary (see [92, §6.8] for a brief
introduction to Clifford analysis for functions of several variables).
Comment 7. Specializing Theorem 3.1.6 to the case when D is the Hodge-Dirac
operator d + δ (where d, δ are, respectively, the exterior derivative operator and its
formal adjoint; cf. [112, (6.6.141)]) acting on differential forms yields the following
Fatou-type result:
any (mixed degree) differential form u defined∫in aUR domain
p Ω ⊆ Rn
and satisfying (d + δ)u = 0 in Ω, as well as ∂Ω Nκ u dH n−1 <∞
 n−1 
for some κ ∈ (0, ∞) and some p ∈ n , ∞ , has the property that the (3.1.145)
κ−n.t.
pointwise nontangential trace u ∂Ω exists at H n−1 -a.e. point on ∂Ω.

In turn, this further implies that

any (mixed degree) differential form u defined
∫  in a UR p domain Ω ⊆ Rn
and satisfying Δu = 0 in Ω, as well as ∂Ω Nκ (du) dH n−1 < ∞ and
∫  p  n−1  (3.1.146)
∂Ω
Nκ (δu) dH n−1 < ∞ for some κ ∈ (0, ∞) and p ∈ n , ∞ ,
κ−n.t. κ−n.t.
has the property that (du) ∂Ω and (δu) ∂Ω exist H n−1 -a.e. on ∂Ω.

Indeed, thanks to the above hypotheses and the conclusion we now seek, there is
no loss of generality in assuming that u is a homogeneous differential form, having
a (well-defined) degree  ∈ {0, 1, . . . , n}. Since (d + δ)2 = −Δ, we may invoke
(3.1.145) for the (mixed degree) differential form w := du + δu ∈ Λ +1 ⊕ Λ −1 and
κ−n.t.
conclude that w ∂Ω exists at H n−1 -a.e. point on ∂Ω. In terms of the individual
components of w of degree  + 1 and  − 1, this translates precisely as saying that
κ−n.t. κ−n.t.
(du) ∂Ω and (δu) ∂Ω exist at H n−1 -a.e. point on ∂Ω, proving (3.1.146).
Again, similar properties are valid if the integrability conditions imposed above
on the nontangential maximal function are replaced by memberships to Lorentz
spaces (cf. (3.1.124)).
Comment 8. Recall the d-bar operator ∂ in the several complex variable theory, and
its Hermitian adjoint ϑ (cf. [115, Chapter 7]). Since the so-called complex Laplacian
in Cn , defined as  := ∂ϑ ¯ + ϑ ∂¯ = (∂¯ + ϑ)2 turns out a scalar multiple of the
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 661

real Laplacian Δ, the same type of argument as in Comment 7 above produces the
following Fatou-type results:

 p Ω ⊆ C and
any complex differential form u defined in∫a UR n
 domain
satisfying (∂¯ + ϑ)u = 0 in Ω, as well as ∂Ω Nκ u dH 2n−1 < +∞
  (3.1.147)
2n , ∞ , has the property that the
for some κ ∈ (0, ∞) and some p ∈ 2n−1
κ−n.t.
pointwise nontangential trace u ∂Ω exists at H 2n−1 -a.e. point on ∂Ω,

which further implies that

any complex differential form u defined ∫ in a UR domain Ω ⊆ Cn and

p
satisfying Δu = 0 in Ω, as well as ∂Ω Nκ (∂u) ¯ dH 2n−1 < ∞ and
∫  p  2n−1  (3.1.148)
∂Ω
Nκ (ϑu) dH 2n−1 < ∞ for some κ ∈ (0, ∞) and p ∈ 2n , ∞ ,
κ−n.t. κ−n.t.
has the property that (∂u)
¯
∂Ω
and (ϑu) ∂Ω
exist H 2n−1 -a.e. on ∂Ω.

Once more, similar Fatou-type results hold when the integrability condition imposed
above on the nontangential maximal function is replaced by membership to a Lorentz
space (cf. (3.1.124)).
Comment 9. It turns out that family of functions satisfying the Moisil-Teodorescu
system (aka the generalized
" Cauchy-Riemann equations), namely ∂j uk = ∂k u j
for 1 ≤ j, k ≤ n and nj=1 ∂j u j = 0 in an arbitrary UR domain Ω ⊆ Rn with
∫ "  p
n ≥ 2, and satisfying ∂Ω nj=1 Nκ u j dH n−1 < ∞ for some κ ∈ (0, ∞) and
  κ−n.t.
some p ∈ n−1 n , ∞ , have nontangential pointwise traces u j ∂Ω at H
n−1 -a.e. point

on ∂Ω, for 1 ≤ j ≤ n. Indeed, this follows by applying Theorem 3.1.6 to the

Moisil-Teodorescu system D from Example 1.3.8.
A similar Fatou-type result holds in connection with the modified Moisil-
Teodorescu system associated with a matrix A = (a jk )1≤ j,k ≤n ∈ Cn×n which is
weakly elliptic, namely for vector-valued functions u = (u j )1≤ j ≤n defined in an
arbitrary UR domain Ω ⊆ Rn with n ≥ 2 and satisfying in Ω:

n
∂j uk = ∂k u j for 1 ≤ j, k ≤ n, and a jk ∂j uk = 0. (3.1.149)
j,k=1

∫Then, based on Example 1.3.9 and Theorem 3.1.6, we conclude that whenever
"n  p 
∂Ω j=1
Nκ u j dH n−1 < ∞ for some κ ∈ (0, ∞) and some p ∈ n−1 n , ∞ it
κ−n.t.
follows that the nontangential pointwise traces u j ∂Ω
, with 1 ≤ j ≤ n, exist at
H n−1 -a.e. point on ∂Ω.
Comment 10. Recall from [112, Definition 5.10.6] that for a nonempty open subset
Ω of Rn to be a UR domain means that ∂Ω is a UR set  (cf. [112,
 Definition 5.10.1])
and that ∂∗ Ω has full H n−1 measure in ∂Ω, i.e., H n−1 ∂Ω\∂∗ Ω = 0. It turns out that
the failure of the latter condition may lead to the failure of the conclusion recorded
in the first line of (3.1.103). Indeed, this is seen by taking n = 2 and D := ∂ the
Cauchy-Riemann operator in the plane, take Ω to be the slit disk defined in (1.1.24)
662 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

(which a bounded open set and ∂Ω is a UR set, but Ω is not a UR domain), and
consider the function u as in (1.1.25). Then (1.1.29) implies the property in the first
line of (3.1.103) fails in such a setting.
It is easy to see that having D injectively elliptic is also a necessary condition for
the validity of the Fatou-type property in the first line of (3.1.103). For example, if
n := 2, M := 1, N := 1, D := ∂x , Ω := (0, 1) × (0, 1), and

u(x, y) := sin(1/y) for each (x, y) ∈ Ω, (3.1.150)

then Ω is a bounded UR domain in R2 , D is a homogeneous, constant coefficient,

first-order operator, and u is a bounded function satisfying Du = 0 in Ω. Hence, for
each aperture parameter κ > 0 we have

Nκ u ∈ L p (∂Ω, H 1 ), (3.1.151)
0<p ≤∞

κ−n.t.
but u ∂Ω does not exist at any point on A := (0, 1) × {0} ⊂ ∂Ω. Thus, the Fatou-
type property in the first line of (3.1.103) presently fails, since H 1 (A) > 0. The
source of this failure is precisely the lack of injectivity for the principal symbol of
D = ∂x . Indeed, for each ξ = (ξ1, ξ2 ) ∈ R2 we have Sym(∂x ; ξ) = iξ1 which vanishes
identically on {0} × R.
Comment 11. We wish to stress that in the statement of Theorem 3.1.6 no (interior,
or exterior) corkscrew condition is imposed on the underlying set Ω. In particular,
Ω may fail to be an NTA domain, which stands in sharp contrast to the Fatou-
type result from Proposition 5.5.2. Examples of UR domains which are not NTA
include heart-shaped regions, since the presence of cusps invalidates the corkscrew
condition.

Fig. 3.1 A UR domain which is not NTA for which Theorem 3.1.6 holds

 
αβ
Comment 12. Suppose L = ar s ∂r ∂s 1≤α ≤M (with the summation convention
1≤β ≤M
over repeated indices in effect) is a weakly elliptic, constant (complex) coefficient,
second-order, M × M system in Rn . To this, associate the first-order, homogeneous,
constant coefficient, (n · M) × M system
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 663
 αβ 
Du := A∇u := ar s ∂s uβ 1≤α ≤M (3.1.152)
1≤r ≤n

which, from Example 1.3.11, is known to be injectively elliptic. Specializing Theo-

rem 3.1.6 to this system then yields the following result: Let Ω ⊆ Rn (where n ≥ 2)
be a UR domain and set σ := H n−1 ∂Ω. Also, consider a coefficient tensor A
with the property that L A is a weakly elliptic M × M system. Then each function
 M
u ∈ 𝒞∞ (Ω) satisfying A∇u = 0 in Ω and Nκ u ∈ L p,q (∂Ω, σ) for some κ > 0,
 n−1  κ−n.t.
p ∈ n , ∞ , and q ∈ (0, ∞], has a nontangential boundary trace u ∂Ω at σ-a.e.
 κ−n.t. 
point on ∂Ω. Moreover, if actually p ∈ (1, ∞), then u = D u ∂Ω in Ω, where D
is the double layer potential operator associated with the coefficient tensor A as in
[115, §1.3].
Comment 13. It turns out that the proof of Theorem 3.1.6 may be extended, by
reasoning largely along the lines of the flat case discussed below, to the setting of
UR subdomains of compact boundaryless manifolds of class 𝒞2 , and for injectively
elliptic first-order differential operators acting between vector bundles over said
manifold. On example of such an operator is the so-called deformation tensor,
a choice which leads to Fatou-type theorems for Killing fields on UR domains
in Riemannian manifolds. Such body of results complements work in [118, §6]
which contains Fatou theorems for the Hodge-Laplacian on regular SKT domains
on Riemannian manifolds.
We now turn to the proof of Theorem 3.1.6.
Proof of Theorem 3.1.6 For starters, from [113, Corollary 10.2.13] (presently used
with q := p), [112, (6.2.25)], [113, (4.2.25)], and (3.1.102) it follows that
   N
Sym D; ν • u belongs to the space H p (∂Ω, σ)
     
and Sym D; ν • u[H p (∂Ω,σ)] N ≤ C Nκ u L p (∂Ω,σ) (3.1.153)
for some C = C(Ω, D, κ, p) ∈ (0, ∞) independent of u.

To deal with the claims in item (i),

 assume  first that either ∂Ω is unbounded, or Ω
is bounded. Also, assume that p ∈ n−1 n , 1 . Granted these assumptions, and bearing
in mind that Du = 0 in Ω and that u is continuous in Ω, formula (3.1.6) currently
implies that for each x ∈ Ω we have

⎧     
⎪ 
⎨ E (x − ·) ∂Ω , (−i)Sym D; ν • u
⎪
⎪ if ∂Ω is unbounded,
u(x) = (3.1.154)
⎪
⎪  
 (x − ·) , (−i)Sym D; ν • u
⎪ E if ∂Ω is bounded.
⎩ ∂Ω

In turn, from (3.1.153), (3.1.154), (3.1.1), Theorem 1.4.2, Corollary 2.5.4, and [112,
κ−n.t.
Corollary 8.9.6] we conclude that u ∂Ω exists (in C M ) at σ-a.e. point on ∂Ω and
κ−n.t.
the function u ∂Ω is σ-measurable on ∂Ω. Moreover, on account of [112, (8.9.8)]
 κ−n.t.  κ−n.t.
we also have u ∂Ω [L p (∂Ω,σ)] M ≤ Nκ u L p (∂Ω,σ) < +∞, hence u ∂Ω belongs
664 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
 M
to L p (∂Ω, σ) . This establishes (3.1.103) in the case when either Ω is bounded,
 
or ∂Ω is unbounded, under the assumption that p ∈ n−1 n ,1 .
When either Ω is bounded, or ∂Ω is unbounded, but p ∈ (1, ∞), we rea-
son as above with natural alterations. First, (3.1.153) and [112, (3.6.27)] we have
   N
Sym D; ν • u ∈ L p (∂Ω, σ) . Thus, if we now employ the integral representation
formula (3.1.9), in place of (3.1.154), which presently reads
∫
   
u(x) =  (x − y), (−i) Sym D; ν • u (y) dσ(y) for all x ∈ Ω,
E
∂Ω
(3.1.155)

from Theorem 2.5.1, Theorem 1.4.2, and [112, Corollary 8.9.6] we again conclude
κ−n.t. κ−n.t.
that u ∂Ω exists σ-a.e. on ∂Ω and the function u ∂Ω is σ-measurable on ∂Ω.
κ−n.t.  M
Together with [112, (8.9.8)], these also imply that u ∂Ω belongs to L p (∂Ω, σ)
 κ−n.t. 
and satisfies u ∂Ω [L p (∂Ω,σ)] M ≤ Nκ u L p (∂Ω,σ) .
Thus, as far as (3.1.103) is concerned, there remains to consider the case when
Ω is unbounded and ∂Ω is bounded. Assuming this is the case, invoke [112,
Lemma 5.10.10] to find R ∈ (0, ∞) such that Rn \ Ω ⊂ B(0, R/2), then run the
first part of the current proof with Ω replaced by ΩR := Ω ∩ B(0, R) which is
a bounded UR domain with ∂ΩR = (∂Ω) ∪ ∂B(0, R). Specifically, observe first
 M
that since Lu = 0 in D (Ω) , with L := D∗ D weakly elliptic M × M system,
from [112, (6.5.40) in Theorem 6.5.7] we see that actually u ∈ [𝒞∞ (Ω)] M . As a
consequence,

u is continuous and bounded near ∂B(0, R). (3.1.156)

Next, if for each x ∈ ∂ΩR we denote by

# $
ΓκR (x) := y ∈ ΩR : |x − y| < (1 + κ) dist(y, ∂ΩR ) (3.1.157)

the nontangential approach region with vertex at x in the set ΩR , then the fact
that ∂Ω ⊂ ∂ΩR forces ΓκR (x) ⊆ Γκ (x) for each x ∈ ∂Ω, where Γκ (x) is the
nontangential approach region with vertex at x in the original set Ω. As such, if NκR
denotes the nontangential maximal operator associated with the domain ΩR , we have
NκR u ≤ Nκ u pointwise on ∂Ω. In view of [112, (8.2.26)], (3.1.102), and (3.1.156),
this implies

NκR u ∈ L p (∂ΩR, σR ) where σR := H n−1 ∂ΩR . (3.1.158)

Since clearly Du = 0 in ΩR , the first part of the proof gives that the nontangential
boundary trace of u, considered from within nontangential approach regions ΓκR (x)
defined as in (3.1.157), exists at σ-a.e. point x ∈ ∂Ω and the resulting function is
σ-measurable. In concert with [112, (8.9.8)], these properties further imply that said
 M
function belongs to L p (∂Ω, σ) and its quasi-norm in this space is dominated by
Nκ u L p (∂Ω,σ) . At this stage, there remains to observe that there exists ε > 0 with
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 665

the property that

ΓκR (x) ∩ B(x, ε) = Γκ (x) ∩ B(x, ε) for each x ∈ ∂Ω, (3.1.159)

so the nontangential boundary trace of u at points on ∂Ω, taken from within ΩR ,

κ−n.t.
agrees, whenever meaningful, with u ∂Ω . This completes the proof of (3.1.103).
κ−n.t.
The last claim in item (i), i.e., that u ∂Ω is actually independent of the aperture
parameter κ, then becomes a consequence of what we have just proved, the first
property in (3.1.102), [112, Proposition 8.9.8], and [112, Proposition 8.8.6].
Turning to item (ii), assume that p ∈ (1, ∞). Then from [113, (10.2.96)-(10.2.97)],
(3.1.102)-(3.1.103), (3.1.153), [112, (3.6.27)], and [112, Corollary 3.7.3] we con-
clude that
 κ−n.t. 
Sym(D; ν) • u = Sym(D; ν) u ∂Ω
(3.1.160)
at σ-a.e. point on ∂Ω.
Since we are presently assuming that u also vanishes at infinity in the case when
Ω is an exterior domain, the integral representation formula (3.1.9) together with
(3.1.160) then give (bearing in mind that u is a null-solution of D)
∫   κ−n.t.  
u(x) =  (x − y), (−i) Sym(D; ν) u
E (y) dσ(y) for all x ∈ Ω.
∂Ω
∂Ω
(3.1.161)

In concert with item (3) in Theorem 2.4.1 (whose applicability is ensured by (3.1.1)
and Theorem 1.4.2), this further implies
      κ−n.t.  
Nκ u ≤ C Sym D; ν u ∂Ω [L p (∂Ω,σ)] N (3.1.162)
L p (∂Ω,σ)

for some constant C = C(Ω, D, κ, p) ∈ (0, ∞) independent of u. Together with the

estimate in (3.1.103), this proves the first equivalence in (3.1.104). The second
equivalence in (3.1.104) is a direct consequence of the fact that, since

Sym(D; ξ) : C M −→ C N is injective for each ξ ∈ Rn \ {0}, (3.1.163)

there exists λ > 0 such that

Sym(D; ξ)w ≥ λ|ξ ||u| for all ξ ∈ Rn and w ∈ C M . (3.1.164)

Next, the first formula in (3.1.105) is seen from the Cauchy-type reproducing
formula for null-solutions of first-order injectively elliptic systems from [115, §1.4].
The second identity in (3.1.105) then becomes a consequence of the first, and the
jump-formula for the boundary-to-domain double layer proved in [115, §1.5]. The
claim in (3.1.106) is implied by the definition of the boundary-to-domain double
layer operator (cf. [115, §1.4]) and Proposition 1.6.4.
Going further, if Nκ u ∈ L q (∂Ω, σ) for some q ∈ (1, ∞) then item (i) implies that
κ−n.t.  M
u ∂Ω belongs to L q (∂Ω, σ) with control of the norm. Conversely, assume that
666 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
κ−n.t.  M
u ∂Ω ∈ L q (∂Ω, σ) for some q ∈ (1, ∞). Then from the integral representation
formula (3.1.161) and item (3) in Theorem 2.4.1 we conclude that there exists some
constant C = C(Ω, D, κ, q) ∈ (0, ∞) independent of u such that
      κ−n.t.  
Nκ u q ≤ C Sym D; ν u ∂Ω [L q (∂Ω,σ)] N
L (∂Ω,σ)
 κ−n.t. 
≤ C u ∂Ω [L q (∂Ω,σ)] M . (3.1.165)

This finishes the proof of the equivalence in (3.1.107). Next, in the case when we
κ−n.t.  p,q M
actually have u ∂Ω ∈ L1 (∂Ω, σ) for some q ∈ (1, ∞) the nontangential max-
imal function estimates for the boundary-to-domain double layer potential operator
established in [115, §1.5] imply Nκ (∇u) ∈ L q (∂Ω, σ) in a quantitative fashion. This
justifies the right-pointing implication in (3.1.108). The left-pointing implication in
(3.1.108) is a consequence of [113, Proposition 11.3.4].
Regarding the claims in item (iii), assume that p = 1. First, from [113, (10.2.96)-
(10.2.97)], (3.1.102)-(3.1.103), and (3.1.153), we see that
 κ−n.t.   N
Sym(D; ν) u ∂Ω = Sym(D; ν) • u ∈ H 1 (∂Ω, σ)
  κ−n.t.     (3.1.166)
and Sym(D; ν) u ∂Ω [H 1 (∂Ω,σ)] N ≤ C Nκ u L 1 (∂Ω,σ)

for some C = C(Ω, D, κ) ∈ (0, ∞) independent of u. This takes care of (3.1.110)-

(3.1.111). Assuming that u also vanishes at infinity in the case when Ω is an exterior
domain, we may invoke the integral representation formula (3.1.154), together with
the nontangential maximal function estimates from (2.4.14) (with p := 1, keeping in
mind (3.1.1) and Theorem 1.4.2), as well as the first line in (3.1.166), the estimate
in (3.1.103) (with p = 1), and the first embedding in [113, (4.2.28)] to conclude that
(3.1.112) holds.  
Moving on, to deal with item (iv) suppose p ∈ n−1 n , 1 . The claims in (3.1.113)
have been already justified in (3.1.153). For the remainder of the proof make the
additional assumption that u vanishes at infinity in the case when Ω is an exterior
domain. Granted this, and taking into account that u is a null-solution of D in Ω,
formula (3.1.6) presently becomes

     u(x) if x ∈ Ω,

E (x − ·) ∂Ω , (−i)Sym D; ν • u = (3.1.167)
0 if x ∈ Rn \ Ω.

In concert with the jump-formula (2.5.50), at σ-a.e. point x ∈ ∂Ω this allows us to

write
1      
∓ % k ν(x) H Sym D; ν • u (x)
2
  κ−n.t. 
    u ∂Ω (x)
+ T (−i)Sym D; ν • u (x) = (3.1.168)
0
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 667

where
 
 ∈ 𝒞∞ (Rn \ {0}) M×N
k := E (3.1.169)

and
 N  M
T : H p (∂Ω, σ) −→ L p (∂Ω, σ) (3.1.170)

is the mapping induced as in item (6) of Theorem 2.3.2 by the principal-value

singular integral of formal convolution type with the matrix kernel k on ∂Ω (cf.
(2.5.3)). Subtracting the two versions of (3.1.168) then yields
       κ−n.t. 
−%k ν(x) H Sym D; ν • u (x) = u ∂Ω (x) (3.1.171)

at σ-a.e. point x ∈ ∂Ω. Observe that since E= D E L  , we may express the Fourier
transform of E at each ξ ∈ R \ {0} as
 n

          
Sym D ;ξ E
& L  (ξ) = − Sym D; ξ  E'L (ξ) 

         
 ξ −1
 ξ  Sym DD;
= − Sym D;
    
 ξ −1 Sym D;
= − Sym DD;  ξ , (3.1.172)

where we have also used the properties of E L recorded in Theorem 1.4.2. Hence,
    
%  ξ −1 Sym D;
k(ξ) = − Sym DD;  ξ for each ξ ∈ Rn \ {0}. (3.1.173)

Together with (3.1.171), this proves (3.1.115). Specializing (3.1.115) to the case
when D  := D∗ (which, according to (1.3.28), is an admissible choice) yields
(3.1.116). In the case when D∗ D has scalar principal symbol it follows that
    −1    
Sym D∗ D; ν commutes with Sym D; ν . As such, applying Sym D; ν to
both sides in (3.1.171) produces (3.1.117), bearing in mind that
    −1    
Sym D∗ D; ν Sym D; ν Sym D∗ ; ν = I, (3.1.174)

the identity operator. The claims in the current items (v)-(vi) are consequences
of the integral representation formulas (3.1.154), (3.1.161), items (4), (6), (7) in
Theorem 2.4.1, as well as (3.1.104) and (3.1.113).
Going further, the claims in item (vii) may be justified much as their counterparts
in items (i)-(vi), making use of Theorem 3.1.2 in place of Theorem 3.1.1.
Turning our attention to item (viii), work under the assumptions made in (3.1.129)-
(3.1.130). All claims with the exception of (3.1.131)-(3.1.132) then follow from
Corollary 3.1.5, [113, Proposition 5.2.7], Theorem 2.5.1, Theorem 1.4.2, [112,
Corollary 8.9.9], [113, (5.1.12)], [112, (8.9.8)], and [112, Corollary 8.9.6]. To justify
(3.1.131), we use what we have proved up to this point, [113, Proposition 10.2.9],
668 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

and (3.1.91) to write, with the family of vector fields

Fα = (Fjα )1≤ j ≤n : Ω −→ Cn (3.1.175)

indexed by α ∈ {1, . . . , N } as in (3.1.86),

   κ−n.t.   αβ  κ−n.t. 
(−i)Sym D; ν u ∂Ω = a j ν j uβ ∂Ω
(3.1.176)
1≤α ≤ N
  κ−n.t.    κ−n.t. 
= ν j Fjα ∂Ω
= ν · Fα ∂Ω
1≤α ≤ N 1≤α ≤ N
  κ−n.t.    
= ν • Fα ∂Ω
= (−i)Sym D; ν • u,
1≤α ≤ N

as wanted. Finally, (3.1.132) is a consequence of what we have proved so far, the

integral representation formula from Corollary 3.1.5, Theorem 1.4.2, the estimate
recorded in (2.4.20), and [113, (5.1.13)]. The proof of Theorem 3.1.6 is therefore
complete. 

Theorem 3.1.6 is surprisingly robust, and a version of its item (i), in which in
place of u being a null-solution for the system D in Ω we now ask that Du has
components in L q (Ω, L n ) with q > n, is presented below, in the class of bounded
UR domains Ω ⊆ Rn .

Theorem 3.1.7 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be a bounded UR domain, and

abbreviate σ := H n−1 ∂Ω. Also, consider an injectively elliptic homogeneous first-
order N × M system D with constant complex coefficients in Rn (where N, M ∈ N),
and suppose u : Ω → C M is a vector-valued function with Lebesgue measurable
components satisfying, for some aperture parameter κ ∈ (0, ∞),
 
Nκ u ∈ L p (∂Ω, σ) for some p ∈ n−1n ,∞ ,
 N (3.1.177)
and Du ∈ L q (Ω, L n ) for some q ∈ (n, ∞].
κ−n.t.
Then the nontangential boundary trace u ∂Ω exists (in C M ) at σ-a.e. point on
∂Ω, and is in fact independent of the aperture parameter κ. Moreover, the function
κ−n.t. κ−n.t.  M
u ∂Ω is σ-measurable on ∂Ω and, in fact, u ∂Ω ∈ L p (∂Ω, σ) .

In light of [112, (8.3.6) in Lemma 8.3.2], the case p = ∞ of Theorem 3.1.7 asserts
that:
if D is an injectively elliptic homogeneous first-order N × M system
D with constant complex coefficients in Rn , and u is an essentially
bounded function in a bounded UR domain Ω ⊆ Rn such that Du
has components in L q (Ω, L n ) for some q > n then the nontangential (3.1.178)
n.t.
boundary trace u ∂Ω exists at H n−1 -a.e. point on ∂Ω and, in fact,
n.t.  M
u ∂Ω ∈ L ∞ (∂Ω, H n−1 ) .
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 669

It is interesting to compare this Fatou-type theorem with an earlier result of A. Nagel

and W. Rudin (cf. [155, Theorem 11.1.2, p. 235]) to the effect that
 
if a, b, c ∈ R satisfy a < b and c > 0, and if u ∈ 𝒞1 (a, b) × (0, c)
is a bounded function satisfying ∂u/∂ z̄ ∈ L p (a, b) × (0, c), L 2 for
some p > 1, then the normal limit lim+ u(x + iy) exists for H 1 -a.e. (3.1.179)
y→0
number x ∈ (a, b).
By further specializing (3.1.178) to scalar functions and D := ∇, the gradient
operator, we obtain the following result, about nontangential pointwise boundary
traces of bounded functions belonging to Sobolev spaces in bounded UR domains:
Corollary 3.1.8 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be an arbitrary bounded UR
domain. Abbreviate σ := H n−1 ∂Ω and fix an aperture parameter κ ∈ (0, ∞). Then
any function
u ∈ W 1,q (Ω) ∩ L ∞ (Ω, L n ) for some q > n (3.1.180)
κ−n.t.
has a nontangential boundary trace u ∂Ω at σ-a.e. point on ∂Ω. Moreover, this
trace is in fact independent of the aperture parameter κ and belongs to L ∞ (∂Ω, σ).
In connection to Corollary 3.1.8 we wish to note if Ω ⊆ Rn is a Sobolev extension
domain6 then any function u ∈ W 1,q (Ω) with q > n necessarily belongs to 𝒞0 (Ω).
Hence, in the case when Ω is also bounded, any such function u is globally bounded.
However, in the class of bounded UR domains, while any function u ∈ W 1,q (Ω)
with q > n is locally continuous (hence, locally bounded), it is not true in general
that u belongs to L ∞ (Ω, L n ). For example, given n ∈ N with n ≥ 2 along with any
q ∈ (1, ∞), if a > q − 1 and if b > 0 is small enough so that a > q(1 + b) − 1, then
taking
# $
Ω := (x1, . . . , xn ) ∈ Rn : 0 < x1, . . . , xn−1 < 1 and 0 < xn < xn−1
a ,
(3.1.181)
and u(x) := xn−1 −b for each x = (x , . . . , x ) ∈ Ω,
1 n

yields a bounded UR domain Ω ⊆ Rn and an unbounded function u ∈ W 1,q (Ω). In

summary, in the class of bounded UR domains Ω ⊆ Rn , membership to W 1,q (Ω) does
not automatically imply global essential boundedness in Ω, even when q ∈ (0, ∞) is
large. Hence, the hypothesis in (3.1.180) is not formulated in a redundant manner.
Another point worth making is that without the hypothesis that u is essentially
bounded, one cannot expect the conclusion in Corollary 3.1.8 to retain its full
strength. For example, if Ω and u are as in (3.1.181) then for each fixed κ > 0 the
κ−n.t.
nontangential boundary trace u ∂Ω exists at σ-a.e. point on ∂Ω but since
 κ−n.t.  −b
u ∂Ω (x1, . . . , xn−1, 0) = xn−1
(3.1.182)
for each (x1, . . . , xn−1, 0) ∈ (0, 1)n−1 × {0} ⊆ ∂Ω,

6 i.e., there exists an operator E : W 1, q (Ω) → W 1, q (R n ) with the property that (Eu) Ω = u for
each u ∈ W 1, q (Ω)
670 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

this nontangential trace no longer belongs to the space L ∞ (∂Ω, σ), even in the case
when the integrability exponent q is large.
We now turn to the task of presenting the proof of Theorem 3.1.7.
Proof of Theorem 3.1.7 From the second property in (3.1.177), [113, Lemma 10.1.1],
and the fact that σ(∂Ω) < +∞, we see that

P(Du) ∈ L ∞ (∂Ω, σ) ⊆ L p (∂Ω, σ). (3.1.183)

 n 
Denote by q  ∈ 1, n−1 the conjugate exponent of q ∈ (n, ∞]. Then the second
property in (3.1.177) also implies, thanks to Hölder’s inequality and the boundedness
of Ω, that for each x ∈ Rn we have
∫ ∫  1/q
|(Du)(y)| dy
dy ≤ Du[L q (Ω, L n )] N < +∞. (3.1.184)
|x − y| n−1 |x − y| (n−1)q
Ω Ω

 from (3.1.1) and observe that Theorem 1.4.2 implies

To proceed, recall E

 (x − y) ≤ C
E for each x, y ∈ Rn with x  y. (3.1.185)
|x − y| n−1
Granted (3.1.184)-(3.1.185), the second property in (3.1.177) then permits us to
conclude (also bearing in mind that composition with translations is continuous on
the Lebesgue scale) that
∫
the mapping Rn  x −→  (x − y), (Du)(y) dy ∈ C M
E
Ω (3.1.186)
is a well-defined C M -valued function which is continuous in Rn .

With (3.1.183) and (3.1.184) in hand, and granted the first property in (3.1.177),
   N
Theorem 3.1.1 applies and gives that Sym D; ν • u belongs to H p (∂Ω, σ) and
that the representation formula (3.1.6) holds at each Lebesgue point (hence, L n -a.e.,
since u is locally integrable) in Ω. In turn, from (3.1.6), (3.1.186), Theorem 2.5.1,
Theorem 1.4.2, Corollary 2.5.4, the first property in (3.1.177), and [112, Corol-
κ−n.t.
lary 8.9.6] we conclude that u ∂Ω exists at σ-a.e. point on ∂Ω and the function
κ−n.t. κ−n.t.
u ∂Ω is σ-measurable on ∂Ω. Finally, that u ∂Ω is actually independent of the
aperture parameter is a consequence of what we have just proved, the first property
in (3.1.177), [112, Proposition 8.9.8], and [112, Proposition 8.8.6]. 

Here is a local version of Theorem 3.1.6 which, among other things, allows the
end-point p = ∞ to be considered in (3.1.102). In particular, this implies that any
bounded null-solution of an injectively elliptic first-order (homogeneous, constant
coefficient) system in an arbitrary UR domain has a nontangential trace at H n−1 -a.e.
point on the boundary.
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 671

Theorem 3.1.9 Suppose Ω ⊆ Rn , where n ∈ N with n ≥ 2, is an arbitrary UR

domain and abbreviate σ := H n−1 ∂Ω. Also, consider an injectively elliptic homo-
geneous first-order N × M system D with constant complex coefficients in Rn (where
 M
N, M ∈ N), and suppose u ∈ 𝒞∞ (Ω) is a vector-valued function satisfying, for
some aperture parameter κ ∈ (0, ∞) and some truncation parameter ε > 0,

Nκε u ∈ Lloc (∂Ω, σ) for some p ∈ (1, ∞], and Du = 0 in Ω.

p
(3.1.187)
κ−n.t.
Then the nontangential boundary trace u ∂Ω exists (in C M ) at σ-a.e. point
κ−n.t.
on ∂Ω, is actually independent of the aperture parameter κ, the function u ∂Ω is
 p M
σ-measurable on ∂Ω and belongs to the space Lloc (∂Ω, σ) .
As a corollary, the same conclusions stated above remain valid for each given
 ∞ M
function u ∈ Lbdd (Ω, L n ) satisfying Du = 0 in [D (Ω)] N hence, in particular,
 M
for any bounded null-solution u ∈ 𝒞∞ (Ω) of the system D in Ω.
p
If ∂Ω is bounded then, of course, Lloc (∂Ω, σ) = L p (∂Ω, σ), so all conclusions
are readily implied Theorem 3.1.6. Hence, the crux of the matter here is allowing
∂Ω to be an unbounded set.
To offer an example, let Ω be a UR domain in #the complex plane, $ and consider a
holomorphic mapping ϕ : Ω → D, where D := z ∈ C : |z| < 1 is the unit disk
in the plane. In particular, ϕ is bounded and satisfies ∂z ϕ = 0 in Ω. Then the last
part in the statement of Theorem 3.1.9 (used with n = 2 and D the Cauchy-Riemann
operator) guarantees that, for each aperture parameter  ∈ (0, ∞),
κ−n.t.
the nontangential boundary trace ϕ ∂Ω exists (in C) at H 1 -a.e. point
on ∂Ω, is actually independent of the aperture parameter κ, and the (3.1.188)
κ−n.t.
function ϕ ∂Ω is H 1 -measurable and (essentially) bounded on ∂Ω.

In particular, this applies to the conformal mapping of Ω onto D, from the clas-
sical Riemann Mapping Theorem. The latter result should be compared with
Carathéodory’s theorem (cf., e.g., [49, Theorem 3.1, p 13]), which states that if
ψ : D → Ω is a conformal mapping of the unit disk onto a Jordan domain
Ω ⊆ R2 ≡ C then ψ may be extended, as a continuous one-to-one mapping, from D
onto Ω.
We now present the proof of Theorem 3.1.9.
Proof of Theorem 3.1.9 Fix a point x0 ∈ ∂Ω along with an arbitrary radius R > 0,
and consider a scalar function ϕ ∈ 𝒞∞ c (R ) such that supp ϕ ⊆ B(x0, 2R) and ϕ ≡ 1
n

on B(x0, R). Since from (A.0.92) and [112, (8.1.17)] (also keeping in mind the second
line in [112, (8.2.26)]) we have
   
Nκ (ϕu) ≤ sup |ϕ| · max Nκε u, sup |u| · 1∂Ω∩B(x0,(4+2κ)R) on ∂Ω,
Rn B(x0,2R)\Oε
(3.1.189)
we conclude from [112, Proposition 8.2.3] and the first property in (3.1.187) that
672 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
p
Nκ (ϕu) ∈ Lcomp (∂Ω, σ) ⊆ L r (∂Ω, σ) for each r ∈ (0, p]. (3.1.190)

In concert with [112, (8.6.51) from Proposition 8.6.3], the first property in (3.1.187)
 M  ∞ M
also implies (bearing in mind that u ∈ 𝒞∞ (Ω) ⊆ Lloc (Ω, L n ) ) that
 np/(n−1) M
u ∈ Lbdd (Ω, L n ) . (3.1.191)

In addition, thanks to [112, (1.7.20)] and the second property in (3.1.187), we have

D(ϕu) = (−i) Sym(D; ∇ϕ)u in Ω. (3.1.192)

By combining (3.1.191) and (3.1.192) we arrive at the conclusion that

 N
D(ϕu) belongs to L 1 (Ω, L n ) and
  (3.1.193)
D(ϕu) = 0 on Ω \ B(x0, 2R) \ B(x0, R) .

In particular, for each point x ∈ Ω ∩ B(x0, R/2) we have

∫   ∫  
D(ϕu) (y) D(ϕu) (y)
dy = dy
Ω |x − y| Ω∩[B(x0,2R)\B(x0,R)] |x − y|
n−1 n−1

∫
≤ (R/2) 1−n
D(ϕu) dL n < ∞. (3.1.194)
Ω

Going further, keeping (3.1.192) in mind and reasoning as in (3.1.189)-(3.1.190)

gives
  p
Nκ D(ϕu) ∈ Lcomp (∂Ω, σ) ⊆ L q (∂Ω, σ) for each q ∈ (0, p]. (3.1.195)
 n−1 
Suppose now that q ∈ n , 1 is fixed and invoke [113, (10.1.8)-(10.1.9)] to conclude
from (3.1.195) that
  ∗  
1 −1
P D(ϕu) ∈ L q (∂Ω, σ) where q∗ := q1 − n−1 . (3.1.196)
  ∗
 
Given that the assignment n−1n , 1  q → q ∈ 1,
n−1
n−2 is bijective, we deduce from
(3.1.190) and (3.1.196) (since p > 1) that

there exists some r ∈ (1, ∞) with the property that

  (3.1.197)
Nκ (ϕu) ∈ L r (∂Ω, σ) and P D(ϕu) ∈ L r (∂Ω, σ).

Collectively, (3.1.197), (3.1.193), and (3.1.194) ensure that Theorem 3.1.1 is

applicable to the function ϕu, with the old integrability exponent p currently replaced
by some r ∈ (1, ∞). With ν denoting the geometric measure theoretic outward unit
normal to Ω, this theorem implies that
   N
Sym D; ν • (ϕu) belongs to the Lebesgue space L r (∂Ω, σ) . (3.1.198)
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 673

Also, keeping in mind that the decay condition (3.1.8) is trivially satisfied by the
function ϕu, and that each point x ∈ Ω ∩ B(x0, R/2) is, trivially, a Lebesgue point
for the (continuous) function ϕu, the integral representation formula (3.1.9) gives
∫
   
u(x) =  (x − y), (−i) Sym D; ν • (ϕu) (y) dσ(y)
E
∂Ω
∫
 
−  (x − y), D(ϕu) (y) dy
E (3.1.199)
Ω∩[B(x0,2R)\B(x0,R)]

for each point x ∈ Ω ∩ B(x0, R/2). In addition, from (3.1.193), Theorem 1.4.2, and
Lebesgue’s Dominated Convergence Theorem it follows that the assignment
∫
 
B(x0, R/2)  x −→  (x − y), D(ϕu) (y) dy
E
Ω∩[B(x0,2R)\B(x0,R)] (3.1.200)
is a well-defined, C M -valued, continuous function.

At this stage, from (3.1.198), (3.1.199), (3.1.200), and Corollary 2.5.2 we deduce
κ−n.t.
that u ∂Ω exists at σ-a.e. point on ∂Ω ∩ B(x0, R/2). Given that R > 0 has been taken
to be an arbitrary number this, in turn, implies that said nontangential boundary
trace actually exists at σ-a.e. point on ∂Ω. Having proved this, we may invoke
[112, Corollary 8.9.6] to conclude (also mindful of the fact that we currently have
κ−n.t.
σ(∂Ω \ ∂∗ Ω) = 0) that the function u ∂Ω is σ-measurable on ∂Ω, and we may rely
κ−n.t.
on [112, Corollary 8.9.9] (also bearing [112, (8.8.52)] in mind) to justify that u ∂Ω
is actually independent of the aperture parameter κ. Finally, [112, (8.9.8)] and the
κ−n.t.  p M
first property in (3.1.187) guarantee that, in fact, u ∂Ω ∈ Lloc (∂Ω, σ) . 

We continue by presenting a version of Theorem 3.1.6 which amounts to a

quantitative Fatou theorem emphasizing the scale of Morrey spaces, in place of
Lebesgue spaces as done in the past.

Theorem 3.1.10 Suppose Ω ⊆ Rn , where n ∈ N with n ≥ 2, is an arbitrary

UR domain and abbreviate σ := H n−1 ∂Ω. Also, consider an injectively elliptic
homogeneous first-order N × M system D with constant complex coefficients in
 M
Rn (where N, M ∈ N), and suppose u ∈ 𝒞∞ (Ω) is a vector-valued function
satisfying, for some κ ∈ (0, ∞), p ∈ (1, ∞), and λ ∈ (0, n − 1),

Nκ u ∈ M p,λ (∂Ω, σ) and Du = 0 in Ω. (3.1.201)

κ−n.t.
Then the nontangential boundary trace u ∂Ω exists (in C M ) at σ-a.e. point on
κ−n.t.
∂Ω, is actually independent of the aperture parameter κ, the function u ∂Ω happens
 M
to be σ-measurable on ∂Ω, belongs to the Morrey space M p,λ (∂Ω, σ) , and
satisfies
 κ−n.t.   
u  p, λ M ≤ Nκ u p, λ . (3.1.202)
∂Ω [M (∂Ω,σ)] M (∂Ω,σ)
674 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

In addition, under the additional assumption that u vanishes at infinity (in the sense
of Definition 1.6.3) in the case when Ω is an exterior domain, one actually has
 κ−n.t.   
u  p, λ M ≈ Nκ u p, λ . (3.1.203)
∂Ω [M (∂Ω,σ)] M (∂Ω,σ)

Proof One way to justify these claims is to invoke item (viii) of Theorem 3.1.6 with
X := M p,λ (∂Ω, σ), a permissible choice in light of [113, Proposition 6.2.17], and
[113, Corollaries 6.2.11, 6.2.13].
An alternative, more self-contained, proof goes as follows. Since we have the
p
inclusion M p,λ (∂Ω, σ) ⊆ Lloc (∂Ω, σ), Theorem 3.1.9 applies and gives that the
κ−n.t.
nontangential boundary trace u ∂Ω exists at σ-a.e. point on ∂Ω, is independent of
κ−n.t.
the aperture parameter κ, and the function u ∂Ω is σ-measurable on ∂Ω. Granted
these properties, on account of [113, (6.2.3)], [112, (8.9.8)], and the first property in
κ−n.t.  M
(3.1.201) we further conclude that u ∂Ω belongs to M p,λ (∂Ω, σ) and (3.1.202)
holds. From what we have proved up to this stage, [113, (10.2.96)-(10.2.97)] (used
with F := u), [113, (6.2.5)], and [112, (6.2.23)] we also see that
 κ−n.t.   N
Sym(D; ν) • u = Sym(D; ν) u ∂Ω ∈ M p,λ (∂Ω, σ) ,
  σ(x)   N
(3.1.204)
hence Sym(D; ν) • u ∈ L 1 ∂Ω, .
1 + |x| n−1
At this stage, there remains to prove (3.1.203) (in fact, only the right-pointing
inequality requires justification), working under the additional assumption that u
vanishes at infinity in the case when Ω is an exterior domain. We make the observation
that there is a natural version of Theorem 3.1.1 which employs Morrey spaces in
place of Lebesgue spaces. Specifically, with E  defined as in (3.1.1), we claim that in
the current setting we have
∫
   
u(x) =  (x − y), (−i) Sym D; ν • u (y) dσ(y) for each x ∈ Ω.
E
∂Ω
(3.1.205)

To justify (3.1.205), we reason as in the proof of Theorem 3.1.1 (and retain notation
introduced on that occasion). For starters, having fixed an arbitrary point x ∈ Ω, in
place of (3.1.16) we now have, for each α ∈ {1, . . . , M },
∫
   
 (x − y), (−i) Sym D; ν • u (y) dσ(y)
E (3.1.206)
∂Ω α

   
= lim [(Lip c (∂Ω)) ] N
. α (x − ·)
(−i)Sym D; ν • u, θ R ηε E N
R→∞ ∂Ω [Lip c (∂Ω)]

thanks to the second line in (3.1.204), the fact that there exists a constant C ∈ (0, ∞)
 − y) ≤ C(1 + |y|)−(n−1) for each y ∈ ∂Ω (cf. (3.1.1) and
with the property that E(x
(1.4.24)), Lebesgue’s Dominated Convergence Theorem, and [112, (4.1.47)]. We
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 675

then proceed along the lines of (3.1.17)-(3.1.30), bearing in mind that we are now
assuming Du = 0. The only aspect we need to reconsider is the fact that the term
 ∫ 
αβ αγ (x − ·) dL n
I = lim uβ a j (∂j θ R )E (3.1.207)
R→∞ Ω 1≤γ ≤M

continues to vanish (much as it did in (3.1.25)) in the present case as well. To see
that this is indeed the case, assume first that ∂Ω is unbounded and, without loss
 suppose 0 ∈ ∂Ω. Then we may invoke [112, (8.6.51)] for the choice
of generality
E := Ω ∩ B(0, CR) \ B(0, R) (with C ∈ (1, ∞) standing for what used to be λ in
the proof of Theorem 3.1.1) and p := (n − 1)/n. In concert with (1.4.24), Hölder’s
inequality, [112, (8.1.17)], and (A.0.87), this permits us to estimate
 ∫ 
I ≤ C lim sup R−n |u| dL n
R→∞ Ω∩[B(0,C R)\B(0,R)]

 ∫ np
 n−1
np
∫  1− n−1
np

≤ C lim sup R−n |u| n−1 dL n dL n
R→∞ Ω∩[B(0,C R)\B(0,R)] B(0,C R)
 !
∫  p1
− n−1
≤ C lim sup R p |Nκ u| p dσ
R→∞ ∂Ω∩B(0,C(2+κ)R)
 !
−
n−1−λ n−1−λ ⨏  p1
≤ C lim sup R p R p |Nκ u| p dσ
R→∞ ∂Ω∩B(0,C(2+κ)R)

n−1−λ
−
≤ C lim sup R p Nκ u M p, λ (∂Ω,σ) = 0. (3.1.208)
R→∞

This proves that I = 0, as wanted. That the same conclusion remains valid when Ω
is bounded is clear since Ω ∩ [B(0, CR) \ B(0, R)] =  if R is sufficiently large.
Finally, in the case when Ω is an exterior domain, we once again have I = 0 thanks to
the assumption that u vanishes at infinity and the first line of (3.1.208). This finishes
the proof of (3.1.205).
At this stage, from (3.1.205), (3.1.1), Theorem 1.4.2, (2.6.4), (3.1.204), and [113,
(6.2.5)] we see that there exists C = C(Ω, D, n, κ, p, λ) ∈ (0, ∞) with the property
that
     
Nκ u p, λ ≤ C Sym D; ν • u p, λ N
M (∂Ω,σ) [M (∂Ω,σ)]
 κ−n.t. 
≤ C u ∂Ω [M p, λ (∂Ω,σ)] M . (3.1.209)

Together with (3.1.202), this completes the alternative proof of Theorem 3.1.10. 

We next study the issue whether the L p -Maximum Principle for null-solutions
u of an injectively elliptic first-order system D in a UR domain Ω singled out in
(3.1.133) has a suitable pointwise version. In Proposition 3.1.11 below we show
676 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

that this is indeed the case provided the underlying domain also satisfies an exterior
corkscrew condition. Specifically, in such a geometric setting, we prove that the
nontangential maximal function of u lies in between (the absolute value of) the
nontangential boundary trace of u and (a fixed multiple of) the Hardy-Littlewood
maximal operator acting on the nontangential boundary trace of u; see (3.1.212)
below (which, in concert with [112, Proposition 8.2.3] and [112, (7.6.18)], actually
implies (3.1.133)).

Proposition 3.1.11 Having fixed n ∈ N with n ≥ 2, assume Ω ⊆ Rn is a UR domain

satisfying an exterior corkscrew condition (cf. [112, Definition 5.1.3]). Abbreviate
σ := H n−1 ∂Ω and pick an arbitrary aperture parameter κ ∈ (0, ∞). Also, suppose
D is an injectively elliptic homogeneous first-order N × M system with constant
complex coefficients in Rn (where N, M ∈ N), and consider a vector-valued function
u : Ω → C M with Lebesgue measurable components satisfying
 σ(x)  p
Nκ u ∈ L 1 ∂Ω , ∩ L loc (∂Ω, σ) for some p ∈ (1, ∞),
1 + |x| n−1
Du = 0 in [D (Ω)] N , and u vanishes at infinity (3.1.210)

(cf. Definition 1.6.3) if Ω is an exterior domain.

κ−n.t.
Then the nontangential boundary trace u ∂Ω exists (in C M ) at σ-a.e. point on
∂Ω and is independent of the aperture parameter, the function
κ−n.t.   σ(x)  M
p
u ∂Ω belongs to L 1 ∂Ω , ∩ L (∂Ω, σ) (3.1.211)
1 + |x| n−1 loc

and, with M ∂Ω denoting the Hardy-Littlewood maximal operator on ∂Ω, one has
κ−n.t.  κ−n.t. 
u ∂Ω ≤ Nκ u ≤ C · M ∂Ω u ∂Ω at σ-a.e. point on ∂Ω, (3.1.212)

for some constant C = C(Ω, D, κ) ∈ (0, ∞) independent of u.

κ−n.t.
Proof From Theorem 3.1.9 and [112, (8.9.8)] we see that u ∂Ω is a well-defined
function, independent of the aperture parameter, and (3.1.211) holds. Granted this
property and the current assumptions, the integral representation formula from The-
orem 1.6.1 gives (with ν denoting the geometric measure theoretic outward unit
normal to Ω)
∫ 
  κ−n.t.   u(x) if x ∈ Ω,
 (x − y), (−i) Sym(D; ν) u
E 
(y) dσ(y) =
∂Ω
∂Ω 0 if x ∈ Rn \ Ω.
(3.1.213)

To proceed, fix an arbitrary point x ∈ Ω and consider some z ∈ ∂Ω such that

r := dist(x, ∂Ω) = |x − z|. Denote by x∗ ∈ Rn \ Ω a corkscrew point relative to z at
scale r. That is, for some θ ∈ (0, 1) independent of z and r, we have
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 677
 
B x∗, θr ⊆ B(z, r) \ Ω. (3.1.214)

Subtracting the two versions of (3.1.213), written for the current point x and for x∗
then permits us to estimate
∫
 κ−n.t. 
|u(x)| ≤ C  − y) − E(x
E(x  ∗ − y) u (y) dσ(y), (3.1.215)
∂Ω
∂Ω

for some C = C(D) ∈ (0, ∞) independent of u, z, x, x∗ . We claim that

 − y) − E(x
 ∗ − y) ≤  Cr
E(x  n for each y ∈ ∂Ω, (3.1.216)
r + |z − y|

where C = C(Ω, D, κ) ∈ (0, ∞) is independent of z, x, x∗, y, r. To justify this, we

study two cases.
Case I: Assume y ∈ ∂Ω is such that |z − y| ≥ 4r. Then, for each ξ ∈ [x, x∗ ], we
have

|ξ − z| ≤ |x − ξ | + |x − z| ≤ |x − x∗ | + |x − z| ≤ |z − x∗ | + 2|x − z|

< r + 2r = 3r < 34 |z − y|, (3.1.217)

hence
 
|ξ − y| ≥ |z − y| − |ξ − z| ≥ 1 − 34 |z − y| = 14 |z − y|. (3.1.218)

In addition, |x − x∗ | ≤ |x − z| + |z − x∗ | < r +r = 2r. Granted these, we may invoke the

Mean Value Theorem to estimate (bearing in mind (3.1.1), (1.4.24), and the working
hypotheses)
 − y) − E(x
E(x  ∗ − y) ≤ |x − x∗ | · sup  − y)
(∇ E)(ξ
ξ ∈[x,x∗ ]

 C  Cr
≤ 2r · sup ≤
ξ ∈[x,x∗ ] |ξ − y| n |z − y| n

Cr
≤  n, (3.1.219)
r + |z − y|

as wanted.
Case II: Assume y ∈ ∂Ω is such that |z − y| < 4r. In this scenario, the right-hand
side in (3.1.216) is ≥ C/r n−1 , and thanks to (3.1.1), (1.4.24), and assumptions we
have
678 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

 − y) − E(x
E(x  ∗ − y) ≤ E(x
 − y) + E(x
 ∗ − y)

C C
≤ +
|x − y| n−1 |x∗ − y| n−1
C C
≤ +
dist(x, ∂Ω)n−1 dist(x∗, ∂Ω)n−1
C
≤ , (3.1.220)
r n−1
from which (3.1.216) follows.
Having established (3.1.216), we make the observation that since for each point
y ∈ ∂Ω we have |x − y| ≤ |x − z| + |z − y| ≤ r + |z − y|, it follows that

E(x  ∗ − y) ≤ C · dist(x, ∂Ω) for each y ∈ ∂Ω.

 − y) − E(x (3.1.221)
|x − y| n
When used in concert with (3.1.215) and [112, (8.4.152)] (with α := 1) from [112,
Proposition 8.4.12], this permits us to conclude that the second estimate in (3.1.212)
holds. Since the first estimate in (3.1.212) is also true (by virtue of [112, (8.9.8)] and
[112, Proposition 8.8.4]), the proof of the proposition is complete. 
We conclude this section by including the purely real variable Fatou type result
from Proposition 3.1.13, As a preamble, in the lemma below we isolate a useful
geometric result employed in its proof.

Lemma 3.1.12 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be an open set of locally finite

perimeter with a lower Ahlfors regular boundary ∂Ω, and such that7

H n−1 (∂Ω \ ∂∗ Ω) = 0. (3.1.222)

In particular, the geometric measure theoretic outward unit normal ν is well defined
H n−1 -a.e. on ∂Ω.
Then for any κ ∈ (0, ∞) there exist an angle θ ∈ (0, π) and an aperture parameter

κ ∈ (0, ∞), both depending only on κ and the lower Ahlfors regularity constant of
∂Ω, with the property that for H n−1 -a.e. point x ∈ ∂Ω one can find a radius
r(x) ∈ (0, ∞), depending only on x, κ, and the lower Ahlfors regularity constant of
∂Ω, such that
     
Γκ (x) ∩ B x, r(x) ⊆ Cθ x, ν(x), r(x) ⊆ Γκ (x) ∩ B x, r(x) (3.1.223)

where, generally speaking,

# $
Cθ (x, h, r) := y ∈ Rn ∩ B(x, r) : (y − x) · h > cos(θ/2) |y − x| (3.1.224)

7 hence, any Ahlfors regular domain will do

3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 679

denotes the one-component circular (roundly) truncated cone with vertex at x ∈ Rn ,

symmetry axis along the vector h ∈ S n−1 , full aperture angle θ ∈ (0, π), and
truncation radius r ∈ (0, ∞).

Proof From (3.1.222) and [112, Proposition 5.6.11] we see that at each point x ∈ ∂ ∗ Ω
there exists an approximate tangent (n − 1)-plane to the set ∂Ω (cf. [112, (5.4.1),
(5.4.2)]). Moreover, the vector ν(x) is perpendicular to the approximate tangent
plane in question, call it π(x).
To proceed, denote by c ∈ (0, ∞) the lower Ahlfors regularity constant of ∂Ω,
and fix an arbitrary number s ∈ (0, 1). Also, choose some s  ∈ (0, s). Finally, pick
an arbitrary point x ∈ ∂ ∗ Ω. Then [112, (5.4.2)] implies that there exists a radius
r(x) ∈ (0, ∞), depending only on x, s , and c, such that
     
H n−1 Δ(x, r) ∩ Cx,s < c/2n−1 r n−1 ∀ r ∈ 0, r(x) , (3.1.225)

where
#   $
Cx,s := y ∈ Rn : dist y, π(x) > s  |y − x|
# $
= y ∈ Rn : (y − x) · ν(x) > s  |y − x| (3.1.226)

is the double cone with vertex at x, axis along ν(x), and full aperture angle
2 arccos s  ∈ (0, π). Suppose there exists a point
 
z ∈ ∂Ω ∩ Cx,s ∩ B x, r(x)/2 , (3.1.227)

where the double cone Cx,s is defined as in (3.1.226) with s  replaced by s. Then
r := |z − x| > 0 (since
 x belongs to ∂Ω and the set Ω is open) hence, for one
thing, r ∈ 0, r(x)/2 . For another thing, there exists some small number λ  ∈ (0, 1),
depending only on n, s, and s , with the property that B(z, λ r) ⊆ Cx,s . In particular,
 
Δ(z, λ r) ⊆ B(z, λ r) ⊆ B x, (1 + λ )r ∩ Cx,s, (3.1.228)
 
Given that (1 + λ )r ∈ 0, r(x) , from (3.1.228) and (3.1.225) we obtain
     
cr n−1 ≤ H n−1 Δ(z, λr) ≤ H n−1 B x, (1 + λ )r ∩ Cx,s
   n−1  n−1 
< c/2n−1 (1 + λ )r ≤ c/2 (2r)n−1 = cr n−1 . (3.1.229)

This is a contradiction which invalidates the happenstance assumed in (3.1.227). As

a result, the above argument shows that
 
∂Ω ∩ Cx,s ∩ B x, r(x)/2 = . (3.1.230)

Going further, denote by C±x,s the two connected components of the open double
cone Cx,s . Then (3.1.230) implies
 thatΩ+ := Ω together with Ω− := Rn \Ω
 constitute
an open cover of Cx,s ∩ B x, r(x)/2 as well as of C−x,s ∩ B x, r(x)/2 . Since the
+
680 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

latter sets are connected, it follows that each of them is solely contained in either
Ω+ or Ω− . We claim that they cannot be both simultaneously contained in Ω+ , or in
Ω− for that
 matter. To justify this, we shall analyze the case when both intersections
C±x,s ∩ B x, r(x)/2 are contained in Ω+ (the other case, when said intersections are
both in Ω− , is similar8). Then Cx,s ∩ B x, r(x)/2 ⊆ Ω, a scenario in which [112,
Lemma 5.6.12] together with the first equality in [112, (5.6.69)] (both used with
E := Ω) lead to the conclusion that
 
L n {y ∈ B(x, r) ∩ Ω : (y − x) · ν(x) > 0}
0 = lim+  
r→0 L n B(x, r)
 
L n {y ∈ B(x, r) ∩ Cx,s : (y − x) · ν(x) > 0}
≥ lim+   > 0, (3.1.231)
r→0 L n B(x, r)
 
which is obviously a contradiction.
 Hence,
 C+x,s ∩ B x, r(x)/2 is contained in one
of the sets Ω± , while C−x,s ∩ B x, r(x)/2 is contained in the other one. To make a
choice, suppose the components of Cx,s are labeled so that
   
C+x,s ∩ B x, r(x)/2 ⊆ Ω and C−x,s ∩ B x, r(x)/2 ⊆ Rn \ Ω. (3.1.232)

Form (3.1.226), the first inclusion in (3.1.232), and [112, Lemma 5.6.15] we also
conclude that
# $
C+x,s = y ∈ Rn : (x − y) · ν(x) > s  |y − x| ,
# $ (3.1.233)
C−x,s = y ∈ Rn : (y − x) · ν(x) > s  |y − x| ,

hence the simple cones C+x,s , C−x,s open up in the direction of −ν(x) and ν(x),
respectively.
Next, pick some number s  ∈ (s, 1). Then Cx,s ⊆ Cx,s and, in fact, there
exists some small number λ  ∈ (0, 1), depending only on n, s, and s , with the
property that B y, λ |y − x| ⊆ C+x,s for each y ∈ C+x,s . Thus, given any point

   
y ∈ C+x,s ∩ B x, r(x)/4 we have ρ := |x − y| ∈ 0, r(x)/4 and
 
B(y, λ  ρ) ⊆ B x, (1 + λ )ρ ∩ C+x,s ⊆ B(x, 2ρ) ∩ C+x,s
 
⊆ B x, r(x)/2 ∩ C+x,s ⊆ Ω. (3.1.234)

Hence, y ∈ Ω and dist (y, ∂Ω) ≥ λ  |x − y|, two properties placing y in Γκ (x) if
κ > 0 is such that λ  = (1 + 
 κ )−1 . This proves that
 
C+x,s ∩ B x, r(x)/2 ⊆ Γκ (x) if  κ := (1/λ ) − 1 > 0. (3.1.235)

In the opposite direction,

  κ > 0. For
start now by fixing an aperture parameter
any point y ∈ Γκ (x) ∩ B x, r(x)/4 we therefore have
 ρ := |x − y| ∈ 0, r(x)/4 and
dist (y, ∂Ω) ≥ ρ(1 + κ)−1 . As such, B y, ρ(1 + κ)−1 ⊆ Ω and also

8 now using the second equality in [112, (5.6.69)] and keeping in mind that L n (∂Ω) = 0
3.1 Quantitative Fatou-Type Theorems in UR Domains for First-Order Systems 681
   
B y, ρ(1 + κ)−1 ⊆ B(x, 2ρ) ⊆ B x, r(x)/2 (3.1.236)

so that, ultimately,
   
B y, ρ(1 + κ)−1 ⊆ Ω ∩ B x, r(x)/2 . (3.1.237)

In view of (3.1.232), this goes to show that

    
B y, ρ(1 + κ)−1 ∩ C−x,s ∩ B x, r(x)/2 =  (3.1.238)

which, thanks to (3.1.236) and the definition of ρ, self-improves to

 
B y, |x − y|(1 + κ)−1 ∩ C−x,s = . (3.1.239)

Based on elementary Euclidean geometry, it transpires that if s ∈ (0, 1) is small

enough to beging with such that
 
arccos s > π2 − arcsin (1 + κ)−1 (3.1.240)

then the collection of all points y ∈ Rn satisfying (3.1.239) is

  # $
Cθ x, −ν(x) := y ∈ Rn : (x − y) · ν(x) > cos(θ/2) |x − y| (3.1.241)

i.e., the infinite circular one-component cone with vertex at x ∈ Rn , symmetry axis
along the vector −ν(x), and full aperture
 
θ := 2π − 2 arcsin (1 + κ)−1 − 2 arccos s ∈ (0, π). (3.1.242)
 
Keeping in mind the fact that the point y ∈ Γκ (x) ∩ B x, r(x)/4 has been arbitrarily
chosen, we arrive at the conclusion that

if θ ∈ (0, π) is as in (3.1.242) (with s ∈ (0, 1) as in (3.1.240))

    (3.1.243)
then Γκ (x) ∩ B x, r(x)/4 ⊆ Cθ x, −ν(x) , the cone from (3.1.241).

Keeping in mind the fact that we presently have H n−1 (∂Ω \ ∂ ∗ Ω) = 0, the first
inclusion in (3.1.223) is seen from (3.1.243), while the second inclusion in (3.1.223)
is a consequence of (the more general result established in) (3.1.235). 
Here is the Fatou type result of a purely real variable nature, alluded to earlier.

Proposition 3.1.13 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be an Ahlfors regular

domain, and abbreviate σ := H n−1 ∂Ω. Also, assume u ∈ 𝒞1 (Ω) is a function
with the property that, for some aperture parameter κ ∈ (0, ∞), some integrability
exponent p ∈ (0, ∞], and some truncation parameter ε > 0 satisfies

Nκε (∇u) ∈ Lloc (∂Ω, σ).

p
(3.1.244)

Then
682 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
κ−n.t.
the nontangential boundary trace u ∂Ω
(3.1.245)
exists at σ-a.e. point on ∂Ω.
Moreover, for any other given aperture parameter κ  ∈ (0, ∞) the nontangential
κ  −n.t. κ−n.t.
boundary trace u ∂Ω also exists, and agrees with u ∂Ω , at σ-a.e. point on ∂Ω.
n.t.
Finally, with the dependence on the aperture parameter dropped, the function u ∂Ω
is σ-measurable on ∂Ω.

Proof From Lemma 3.1.12 we know that there exist an angle θ ∈ (0, π) and an
aperture parameter  κ ∈ (0, ∞) with the property that for σ-a.e. point x ∈ ∂Ω one
can find a radius r(x) ∈ (0, ∞) such that the inclusions
 in (3.1.223)
 hold. Observe
that this implies that for any given radius r ∈ 0, min{ε, r(x)} and any given pair
of points y1, y2 ∈ Γκ (x) ∩ B(x, r) the line segment [y1, y2 ] is contained in the
intersection Γκ (x) ∩ B(x, r), and the Man Value Theorem allows us to estimate
 
|u(y1 ) − u(y2 )| ≤ |y1 − y2 | · sup |(∇u)(ξ)| ≤ 2r · Nκε (∇u) (x). (3.1.246)
ξ ∈(y1,y2 )

Based on this, Cauchy’s criterion for

 convergence,
 and Definition A.0.99, whenever
we also have x ∈ Aκ (∂Ω) and Nκε (∇u) (x) < ∞ we see that the nontangen-
 κ−n.t. 
tial boundary trace u ∂Ω (x). Since [112, Proposition 8.8.4] guarantees that we
 
presently have σ ∂Ω \ Aκ (∂Ω) = 0, while [112, Corollary 8.4.2] together with
(3.1.244) ensure that Nκε (∇u) (x) < ∞ at σ-a.e. point x ∈ ∂Ω, we ultimately
conclude that the claim made in (3.1.245) is true. Granted this, the fact that Ω is an
Ahlfors regular domain together with [112, Corollary 8.9.6] and [112, Lemma 3.6.4]
κ−n.t.
then imply that u ∂Ω is a σ-measurable function on ∂Ω.
Finally, the remaining claims in the statement follow from what we have proved
so far and the fact that, thanks to [112, Corollary 8.4.2], the assumption made in
(3.1.244) is unaffected by changing the aperture parameter. 

3.2 Brief Look at Hardy Spaces Associated with Injectively

Elliptic First-Order Systems

Let D be first-order homogeneous N × M system in Rn (where n ∈ N satisfies n ≥ 2)

with constant (complex) coefficients, which is injectively elliptic. Also, consider an
open set Ω ⊆ Rn such that σ := H n−1 ∂Ω has the property that
 
σ B(x, r) ∩ ∂Ω > 0 for each x ∈ ∂Ω and r > 0. (3.2.1)

In this setting, having fixed a background parameter κ ∈ (0, ∞), for each integrability
exponent p ∈ (0, ∞] define the Hardy space associated with the operator D in
Ω as follows:
3.2 Brief Look at Hardy Spaces Associated with Injectively Elliptic First-Order Systems 683
 M
H p (Ω; D) is the collection of all functions u ∈ 𝒞∞ (Ω) satisfying
Nκ u ∈ L p (∂Ω, σ) and Du = 0 in Ω, and which also vanish at infinity (3.2.2)
(in the sense described in Definition 1.6.3) when Ω is an exterior
domain.
Equip the space H p (Ω; D) introduced in (3.2.2) with the quasi-norm
 
u H p (Ω;D) := Nκ u L p (∂Ω,σ), ∀u ∈ H p (Ω; D). (3.2.3)
p
Hence, H p (Ω; D) is a subspace of Nκ (Ω; σ) (defined as in (A.0.98)). Thanks to
(3.2.1), we may then rely on [112, Proposition 8.3.5] to conclude that

H p (Ω; D) is a quasi-Banach space and  ·  H p (Ω;D) is a p-norm for

(3.2.4)
p ∈ (0, 1); also, H p (Ω; D) is an actual Banach space if p ∈ [1, ∞].

It is comforting to note that, whenever σ is a doubling measure on ∂Ω, [112,

Proposition 8.4.1] guarantees that the definition of the Hardy space H p (Ω; D) in
(3.2.2) is actually independent of the aperture parameter κ ∈ (0, ∞).
Under the stronger hypothesis that the underlying set Ω is actually a UR domain,
assumed henceforth, Theorem 3.1.6 ensures that the Hardy spaces defined in (3.2.2)
enjoy a wealth of remarkable properties. Among other things, in such a scenario the
nontangential pointwise trace
n.t.  M
H p (Ω; D)  u −→ u ∂Ω ∈ L p (∂Ω, σ) (3.2.5)

defines a well-defined, linear, and bounded operator. In analogy with more classical
settings, we define the boundary Hardy space associated with the operator D as the
image of this mapping, i.e.,
 n.t. 
H p (∂Ω; D) := u ∂Ω : u ∈ H p (Ω; D) , (3.2.6)
 M
regarded as a subspace of L p (∂Ω, σ) and equipped with the quasi-norm
 n.t. 
u H p (∂Ω;D) := u ∂Ω [L p (∂Ω,σ)] M , ∀u ∈ H p (∂Ω; D). (3.2.7)

In particular,
n.t.
the mapping H p (Ω; D)  u −→ u ∂Ω ∈ H p (∂Ω; D)
is well defined, linear, bounded, surjective, and (3.2.8)
actually an isomorphism whenever p ∈ (1, ∞),

with the very last claim implied by item (ii) of Theorem 3.1.6. Let us additionally
note here that if D and K are the boundary double layer potential operators associated
as in [115, §1.4] with the set Ω and the weakly elliptic second-order M × M system
684 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

 where D
L := DD,  is a M × N homogeneous first-order system with
constant complex coefficients in Rn which complements D (in the sense (3.2.9)
that (1.3.21) holds)
then (3.1.105) implies that, for each p ∈ (1, ∞),
  M
D : Ker − 12 I + K; L p (∂Ω, σ) −→ H p (Ω; D) (3.2.10)

is a well-defined, linear, bounded, and surjective operator, with the nontangential

pointwise trace
n.t.   M
H p (Ω; D)  u −→ u ∂Ω ∈ Ker − 12 I + K; L p (∂Ω, σ) (3.2.11)

serving as a right-inverse. Since the jump-formula for the double layer (cf. [115,
§1.5]) shows that the operator (3.2.10) is also injective, we ultimately conclude that,
for each p ∈ (1, ∞), the operator
  M
D : Ker − 12 I + K; L p (∂Ω, σ) → H p (Ω; D)
is an isomorphism with (3.2.11) as inverse, and (3.2.12)
  M
H p (∂Ω; D) = Ker − 12 I + K; L p (∂Ω, σ) .

We also wish to remark that [112, Corollary 1.7.3] used with p = 1 and w arbi-
trary constant in C N establishes, keeping in mind the decay property (1.6.46), the
cancelation property to the effect that
∫
 n.t. 
(−i)Sym(D; ν) u ∂Ω dσ = 0 for each u ∈ H 1 (Ω; D). (3.2.13)
∂Ω

In fact, if D is actually elliptic, in the sense that its principal symbol Sym(D; ξ)
is invertible for each ξ ∈ Rn \ {0}, then formula [112, (1.7.35) in Corollary 1.7.3]
ensures (in view of (3.2.2) and Definition 1.6.3) that
∫
 n.t.  n.t.
(−i)Sym(D; ν) u ∂Ω , w ∂Ω dσ = 0 for each
∂Ω (3.2.14)

u ∈ H p (Ω; D) and w ∈ H p (Ω; D ) with 1/p + 1/p = 1.

In light of (3.2.6), we may simply rephrase this cancelation property as

∫
(−i)Sym(D; ν) f , g dσ = 0 for any two functions
∂Ω (3.2.15)

f ∈ H p (∂Ω; D) and g ∈ H p (∂Ω; D ) with 1/p + 1/p = 1.

Here is one last comment pertaining to our brand of Hardy spaces, when con-
sidered in the category of UR domains satisfying an exterior corkscrew condition
(which, in particular, includes all two-sided NTA domains with Ahlfors regular
3.2 Brief Look at Hardy Spaces Associated with Injectively Elliptic First-Order Systems 685

boundaries). Specifically, Proposition 3.1.11 implies that

if Ω ⊆ Rn (with n ≥ 2) is a UR domain satisfying an exterior corkscrew
condition, D is an injectively elliptic homogeneous first-order system
with constant complex coefficients in Rn , and κ ∈ (0, ∞), then there
exists some C = C(Ω, D, κ) ∈ (0, ∞) such that for each u ∈ H p (Ω; D) (3.2.16)
κ−n.t.  κ−n.t. 
with p ∈ (1, ∞) we have u ∂Ω ≤ Nκ u ≤ C · M ∂Ω u ∂Ω at almost
every point on ∂Ω (with respect to H n−1 ).
The history of Hardy spaces is particularly rich and influential, going back to its
origins involving holomorphic functions of one complex variable in the planar unit
disk (and the upper half-plane; cf. [48], [49], [59], [84], [156] and the references
therein) to the modern theory of Hardy spaces associated with the Moisil-Teodorescu
system (aka generalized Cauchy-Riemann equations) (as originally developed in [45],
[166], [167], [169]-[171]) and, further, to more general domains and differential
operators (cf. [78], [72], [125]), including extensions to null-solutions for Dirac-
type operators on manifolds (see [118], [124], [123], [127]). Here we only briefly
elaborate on the impact of our current results on the theory of Hardy spaces for the
standard Dirac operator in arbitrary UR domains. To state a significant result in this
regard, the reader is advised to recall the Clifford bullet product defined in (A.0.97).
The reader is also alerted to the fact that we shall occasionally use notation and
results discussed later (independently of the current considerations), such as those
pertaining to the Clifford-Riesz transform from [115, §2.1].

Theorem 3.2.1 Fix n ∈ N with n ≥ 2, and suppose Ω ⊆ Rn is an arbitrary UR

domain. Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic
outward unit normal
  Ω. Pick
to a background aperture parameter κ ∈ (0, ∞) and,
for each p ∈ n−1n , ∞ , let H p (Ω; D) be the Hardy space associated as in (3.2.2)

with the Dirac operator D from (A.0.37).

κ−n.t.
(1) For each u ∈ H p (Ω; D), the nontangential boundary trace u ∂Ω exists (in Cn )
at σ-a.e. point on ∂Ω, and is actually independent of the aperture parameter κ.
n.t.
Simplifying notation by dropping the dependence on κ, the function u ∂Ω belongs
 n.t. 
to the space L p (∂Ω, σ) ⊗ Cn and u  p ≤ Nκ u L p (∂Ω,σ) . In
∂Ω L (∂Ω,σ)⊗ C n
addition, if p ∈ (1, ∞) one actually has
   n.t. 
Nκ u p ≈ u ∂Ω  L p (∂Ω,σ)⊗ C n (3.2.17)
L (∂Ω,σ)

as well as
n.t.
p
u ∂Ω ∈ L1 (∂Ω, σ) ⊗ Cn ⇐⇒ Nκ (∇u) ∈ L p (∂Ω, σ), and
 n.t.      (3.2.18)
u  p ≈ Nκ u L p (∂Ω,σ) + Nκ (∇u) L p (∂Ω,σ)
∂Ω L (∂Ω,σ)⊗ C n
1

with proportionality constants independent of the function u in both cases.

686 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

(2) With the Clifford-Riesz transform operator R C originally defined as in (A.0.116),

and subsequently extended to Hardy spaces, for each function u ∈ H p (Ω; D)
the following version of Cauchy’s Reproducing Formula holds

u = 12 R C (ν 
• u) in Ω, (3.2.19)

and one has

• u ∈ H p (∂Ω, σ) ⊗ Cn and
ν
(3.2.20)
ν 
• u H p (∂Ω,σ)⊗ C n ≈ Nκ u L p (∂Ω,σ),
with proportionality constants independent of the function u.
(3) If one defines the “bullet” boundary Hardy space
p # $
ℋ• (∂Ω; D) := ν  • u : u ∈ H p (Ω; D) (3.2.21)

and equip it with the quasi-norm inherited from H p (∂Ω, σ) ⊗ Cn , then
p
ℋ• (∂Ω; D) is a closed linear subspace of H p (∂Ω, σ) ⊗ Cn . Also, if C # is the
“transpose” Cauchy-Clifford operator (originally defined as in (A.0.36) and
then extended to Hardy spaces) then 12 I + C # is a projection9 of H p (∂Ω, σ) ⊗ Cn
p
onto ℋ• (∂Ω; D). In addition,
 
p
ℋ• (∂Ω; D) = Im 12 I − C # : H p (∂Ω, σ) ⊗ Cn → H p (∂Ω, σ) ⊗ Cn
 
= Ker 1
2I + C # : H p (∂Ω, σ) ⊗ Cn → H p (∂Ω, σ) ⊗ Cn
(3.2.22)

and
p
R C : ℋ• (∂Ω; D) −→ H p (Ω; D) isomorphically,
p
(3.2.23)
with inverse H p (Ω; D)  u → 12 ν 
• u ∈ ℋ• (∂Ω; D).

In particular, a monogenic10 function u belongs to the Hardy space H p (Ω; D)

if and only if u = R C f for some f ∈ H p (∂Ω, σ) ⊗ Cn , hence

R C : H p (∂Ω, σ) ⊗ Cn −→ H p (Ω; D)

(3.2.24)
is a well-defined, linear, bounded, surjective operator.

(4) With H denoting the L p -filtering operator from [113, Theorem 4.9.1] (used here
with Σ := ∂Ω), for each u ∈ H p (Ω; D) one has
   n.t. 
• u = ν  u ∂Ω at σ-a.e. point on ∂Ω.
H ν (3.2.25)

9 i.e., a linear, bounded, and idempotent operator

10 i.e., a null-solution of the Dirac operator (A.0.37)
3.2 Brief Look at Hardy Spaces Associated with Injectively Elliptic First-Order Systems 687

As a consequence, with the boundary Hardy space H p (∂Ω; D) defined as in

(3.2.6) (where now D stands for the Dirac operator from (A.0.37)), the mapping
p
H : ℋ• (∂Ω; D) −→ ν  H p (∂Ω; D) (3.2.26)

is well defined, linear, bounded, and surjective.

(5) Introduce Ω+ := Ω and Ω− := Rn \ Ω. Then for each f ∈ H p (∂Ω, σ) ⊗ Cn
there exist unique functions u± ∈ H p (Ω± ; D) with the property that

f =ν
• u+ + ν 
• u− in H p (∂Ω, σ) ⊗ Cn . (3.2.27)

Consequently, with the direct sum topological, the following Calderón-type

decomposition formula holds:
p p
H p (∂Ω, σ) ⊗ Cn = ℋ• (∂Ω+ ; D) ⊕ ℋ• (∂Ω− ; D). (3.2.28)

(6) Suppose ∂Ω is bounded and recall the boundary-to-boundary “transpose”

Cauchy-Clifford operator C # (originally defined as in (A.0.36) and then ex-
tended to Hardy spaces). Then, whenever n−1 < p < p∗ ≤ 1, the p∗ -envelope
p n
of the “bullet” boundary Hardy space ℋ• (∂Ω; D) may be described as
 p  
p ∗, p ∗
E p∗ ℋ• (∂Ω; D) = Im 12 I − C # : B 1 (∂Ω, σ) ⊗ Cn
1
−(n−1)( p − p ∗ )

p ∗, p ∗
→B (∂Ω, σ) ⊗ Cn
−(n−1)( p1 − p1∗ )

p ∗, p ∗
= Ker 1
2I + C# : B (∂Ω, σ) ⊗ Cn
−(n−1)( p1 − p1∗ )

p ∗, p ∗
→B (∂Ω, σ) ⊗ Cn .
−(n−1)( p1 − p1∗ )
(3.2.29)

Finally, strengthen the original hypotheses by assuming that Ω is a bounded

NTA domain with an Ahlfors regular boundary, and recall the Clifford-Riesz
transform R C from (A.0.116). Then, whenever n−1 ∗
n < p < p ≤ 1, it follows
∗
that p -envelope of the Hardy space H (Ω; D) may be described as
p

 
E p∗ H p (Ω; D) (3.2.30)
 
p ∗, p ∗ p ∗, p ∗
= Im R C : B 1 (∂Ω, σ) ⊗ Cn → B 1 1 (Ω)
−(n−1)( p − p ∗ )
1 1
p∗
−(n−1)( p − p ∗ )
 1  
p ∗, p ∗
= RC f : f ∈ B (∂Ω, σ) ⊗ Cn and 2I + C# f = 0 .
−(n−1)( p − p∗
1 1
)

Before presenting the proof of this theorem, we make some comments pointing to
further research on this topic. Specifically, it is natural to consider Hardy spaces in UR
688 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

domains Ω ⊆ Rn associated with a given injectively elliptic first-order homogeneous

constant (complex) coefficient system D in Rn , by demanding the membership of
the nontangential maximal operator to a variety of other function spaces on ∂Ω.
For example, as apparent from the proof of Theorem 3.2.1 given below, similar
results to those described in items (1)-(5) above are valid for Muckenhoupt weighted
Lebesgue-styled Hardy spaces, and Morrey-styled Hardy spaces (involving super-
unital integrability exponents). The same is true for Lorentz-based Hardy spaces
associated with the Dirac operator D in Ω, i.e.,
 
H p,q (Ω; D) defined for each p ∈ n−1 , ∞ and q ∈ (0, ∞) as the
 M n
space of functions u ∈ 𝒞∞ (Ω) satisfying Nκ u ∈ L p,q (∂Ω, σ) (3.2.31)
and Du = 0 in Ω, and which also vanish at infinity (in the sense of
Definition 1.6.3) in the case when Ω is an exterior domain.
It is also of interest to consider more general sets and larger ranges of indices. To
offer a concrete example, work in the two-dimensional setting
 under the canonical
identification R2 ≡ C, and take D to be ∂z̄ := 12 ∂x + i∂y , the Cauchy-Riemann
operator. Then, if Ω ⊆ C is an open set with an Ahlfors regular boundary and
zo ∈ ∂Ω is a fixed point, then for each m ∈ N the function
1
u(z) := , ∀z ∈ Ω, (3.2.32)
(z − zo )m

belongs to the Lorentz-based Hardy space H 1/m,∞ (Ω; ∂z̄ ). This is seen from [112,
Lemma 8.3.7] and [112, Example 6.2.2].
We now turn to the task of providing the proof of Theorem 3.2.1.
Proof of Theorem 3.2.1 Since the Dirac operator D is injectively elliptic, the claims
in item (1) are direct consequences of items (i)-(ii) in Theorem 3.1.6. Next, the
discussion in [113, Example 10.2.14] implies that

ν
• u ∈ H p (∂Ω, σ) ⊗ Cn and
(3.2.33)
ν 
• u H p (∂Ω,σ)⊗ C n ≤ CNκ u L p (∂Ω,σ),

for some finite constant C > 0 independent of u. Also, in view of the properties
of the Clifford-Riesz transform from [115, §2.1] and (A.0.97), formula (3.1.154)
reduces precisely to the integral representation formula (3.2.19). Having established
this, the nontangential maximal function estimates for the Clifford-Riesz transform
established in [115, §2.1] then permit us to write (for some C ∈ (0, ∞) independent
of u)
    
Nκ u p • u)  L p (∂Ω,σ)
= 12 Nκ R C (ν 
L (∂Ω,σ)

≤ Cν 
• u H p (∂Ω,σ)⊗ C n . (3.2.34)
3.2 Brief Look at Hardy Spaces Associated with Injectively Elliptic First-Order Systems 689

Together with (3.2.33), this finishes the proof of the claims made in (3.2.20). Then
p
the fact that ℋ• (∂Ω; D) is a closed subspace of H p (∂Ω, σ) ⊗ Cn becomes a con-
sequence of (3.2.20) and the completeness of H p (Ω; D) (cf. (3.2.4)). Next, by once
again relying on the properties of the Clifford-Riesz transform from [115, §2.1] we
conclude that
R C : H p (∂Ω, σ) ⊗ Cn −→ H p (Ω; D)
(3.2.35)
is well defined, linear, bounded, and surjective.

In fact, it is shown in [115, §2.1] that

   
ν • R C f = I − 2C # f for each f ∈ H p (∂Ω, σ) ⊗ Cn . (3.2.36)

In concert, (3.2.35), (3.2.36), (3.2.21), and (3.2.19) imply

  
p
ℋ• (∂Ω; D) = 12 I − C # f : f ∈ H p (∂Ω, σ) ⊗ Cn , (3.2.37)

which proves the first equality in (3.2.22). The second equality in (3.2.22) is a
consequence of the this and the fact that, as readily seen from the fact that 2C # is
idempotent when acting on Hardy spaces, 12 I −C # is a projection on H p (∂Ω, σ)⊗ Cn .
Moving on, (3.2.23) is implied by (3.2.35),
 (3.2.19),
 and (3.2.21). Also, (3.2.25)
is a consequence of (3.1.117) when p ∈ n−1 n , 1 , and of [113, (4.9.3)] together with
[113, (10.2.104)] in the case when 1 ≤ p < ∞.
Consider next the claim made in item (5). Given a function f ∈ H p (∂Ω, σ) ⊗ Cn ,
the jump-formula (3.2.36) written both in Ω+ and Ω− implies that (3.2.27) holds
for u± := 12 R C f in Ω± . To prove the uniqueness of the functions u± ∈ H p (Ω± ; D)
doing the job in (3.2.27), assume f = 0. Then the function

u+ in Ω+,
u= (3.2.38)
u− in Ω−,

is locally integrable in Rn and, thanks to [113, (10.2.101)] and the present working
hypotheses, satisfies Du = 0 in the sense of distributions in Rn . If ∂Ω is bounded,
then u also vanishes at infinity, so Liouville’s theorem ultimately implies that u
vanishes in Rn . Hence, in this case we have u± = 0 in Ω± , as wanted. An alternative
approach, which works regardless of whether ∂Ω is bounded or not, is to observe
that if

u± ∈ H p (Ω± ; D) are such that ν 

• u+ + ν 
• u− = 0, (3.2.39)
then representation formula (3.1.167) presently implies that for each x ∈ Ω− we have
   
u− (x) = 12 R C (ν 
• u− ) (x) = − 12 R C (ν 
• u+ ) (x) = 0. (3.2.40)

Thus, u− = 0 in Ω− and, analogously, u+ = 0 in Ω+ .

690 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

Finally, consider the claims in item (6). From (3.2.22) we know that the “trans-
pose” Cauchy-Clifford operator C # (originally defined as in (A.0.36) and then ex-
tended to Hardy spaces) induces is a well-defined, linear, bounded, and surjective
mapping
p
1
2I − C # : H p (∂Ω, σ) ⊗ Cn −→ ℋ• (∂Ω; D). (3.2.41)

Assume that ∂Ω is bounded and n−1 ∗ ∗

n < p < p ≤ 1. By taking p -envelopes in
(3.2.41) we then conclude, based on item (2) in [113, Proposition 7.8.15], [113,
(7.8.153)], [113, Proposition 7.8.13], and the mapping properties of C # on the Besov
scale (established in [115, §4.1]), that the mapping
p ∗, p ∗  p 
1
2I − C# : B ⊗ Cn −→ E p∗ ℋ• (∂Ω; D) (3.2.42)
−(n−1)( p1 − p1∗ )

continues to be surjective, where C # is now the “transpose” Cauchy-Clifford operator

in the context of boundary Besov spaces (with negative smoothness). From this, the
first equality in (3.2.29) readily follows. Also, the second equality in (3.2.29) is a
consequence of the first and the fact that 2C # is idempotent when acting on Besov
spaces.
Finally, work under the assumption that Ω is a bounded NTA domain with an
∗ ∗
Ahlfors regular boundary. Also, suppose n−1 n < p < p ≤ 1. Taking p -envelopes in
(3.2.24) yields, on account of based on item (2) in [113, Proposition 7.8.15], [113,
(7.8.153)], and (A.0.116) that
p ∗, p ∗  
RC : B ⊗ Cn −→ E p∗ H p (Ω; D)
−(n−1)( p1 − p1∗ ) (3.2.43)
is a well-defined, linear, bounded, surjective mapping,

where the Clifford-Riesz transform R C now acts from boundary Besov spaces with
a negative amount of smoothness.
 Granted
 this, the first equality in (3.2.30) follows.
One way to see that E p∗ H p (Ω; D) may also be described as in the second line of
(3.2.30) is to take p∗ -envelopes in the first line of (3.2.23), then invoke (3.2.29) and
item (1) in [113, Proposition 7.8.15]. 

3.3 Quantitative Fatou-Type Theorems in UR Domains for

Second-Order Systems

Here we establish certain Fatou-type theorems in arbitrary UR domains, involving

weakly elliptic second-order systems. To put matters in the proper perspective we be-
gin by recalling that T. Wolff has proved (cf. [181, Theorem 3, p. 321] and subsequent
comments) that, given n ∈ N with n ≥ 3,
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 691

there exists some harmonic function u in R+n with the property that
∫  1/p
sup |(∇u)(x , t)| p dx  < +∞ for some p > 0, and yet (3.3.1)
t>0 R n−1
∇u does not have nontangential limits H n−1 -a.e. on Rn−1 ≡ ∂R+n .

The actual value of critical exponent p in (3.3.1) is unknown, but E. Stein and
G. Weiss proved (cf. [169, Theorem A, p. 28]) that if u is harmonic in some open
subset of Rn then |∇u| p is subharmonic if p ≥ n−2 n−1 . This, in turn, implies that
harmonic gradients have limits almost everywhere if the uniform L p -integrability
condition in (3.3.1) is satisfied with p ≥ n−2
n−1 (cf. [169, Theorem B, p. 46]). Hence,
one necessarily has 0 < p < n−1 in (3.3.1). In this vein, let us also note that
n−2

C. Fefferman and E. Stein have shown in [45, Theorem 9, p. 168] that, for a harmonic
gradient ∇u, the uniform L p -integrability condition in (3.3.1) is equivalent, in the
n−1 , to actually having Nκ (∇u) ∈ L (R
regime p > n−2 p n−1, H n−1 ) (for some, or any,

aperture parameter κ > 0). Our goal is to generalize the positive Fatou-type result of
Stein-Weiss, in the Fefferman-Stein formulation, by showing that
if u is a (vector-valued) null-solution in a UR domain Ω ⊆ Rn (with
n ≥ 2) of a weakly elliptic homogeneous constant coefficient second-
order system in Rn , having the property that the nontangential maximal (3.3.2)
 Jacobian belongs to the space L (∂Ω, H n−1 ) for some
function of its p

p ∈ n−1 n , ∞ , then ∇u has nontangential limits H n−1 -a.e. on ∂Ω.

In light of Wolff’s counterexample in (3.3.1), this goal seems reasonably sharp, and
in Theorem 3.3.4 we shall accomplish just that. This theorem is stated later, and
is given two proofs. A key ingredient in one of these proofs is the representation
formula contained in the theorem below.

Theorem 3.3.1 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be an Ahlfors regular

domain, and abbreviate σ := H n−1 ∂Ω. For some M ∈ N, consider a coefficient
αβ 
tensor A = ar s 1≤r,s ≤n with complex entries, with the property that the M × M
1≤α,β ≤M
homogeneous second-order system L = L A associated with A in Rn as in (1.3.15)
is weakly elliptic (in the sense of (1.3.3)). Also, denote by E = (Eγβ )1≤γ,β ≤M the
matrix-valued fundamental solution associated with L as in Theorem 1.4.2.
 1,1  M
In this setting, assume u = (uβ )1≤β ≤M ∈ Wloc (Ω) is a vector-valued function
which, for some aperture parameter κ > 0, satisfies
 
Nκ (∇u) ∈ L p (∂Ω, σ) for some p ∈ n−1 n ,∞ . (3.3.3)

With Lu considered in the sense of distributions in Ω, suppose also that

 1 M
Lu ∈ Lloc (Ω, L n ) and P(Lu) ∈ L p (∂Ω, σ). (3.3.4)

Then for any , s ∈ {1, . . . , n} one has

692 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
.  M .  M
∂τ s u ∈ H p (∂Ω, σ) and ∂νAu ∈ H p (∂Ω, σ) . (3.3.5)

Moreover, given any  ∈ {1, . . . , n} and γ ∈ {1, . . . , M }, it follows that for

L n -a.e. Lebesgue point x ∈ Ω of ∇u with the property that
∫
|(Lu)(y)|
dy < +∞ (3.3.6)
Ω |x − y|
n−1

one has (with the duality pairings understood in the sense of [113, Theorem 4.6.1]
with Σ := ∂Ω, and the summation convention over repeated indices in effect)
βα   .
(∂ uγ )(x) = ar s (∂r Eγβ )(x − ·) ∂Ω , ∂τ s uα
  . 
− (∂ Eγβ )(x − ·) ∂Ω , ∂νAu β
∫
 
+ (∂ Eγβ )(x − y) Lu β (y) dy (3.3.7)
Ω
 
assuming that p ∈ n−1 , 1 and ∂Ω is unbounded. A similar formula holds when
  n
p ∈ n−1 n , 1 and Ω is bounded, this time omitting taking equivalence classes of
functions
 modulo
 constants in the duality pairings in (3.3.7). In the case when
p ∈ n−1 n , 1 and Ω is an exterior domain, the latter formula remains true under the
additional assumption that there exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (3.3.8)
B(0,λ R)\B(0,R)

In fact, a similar integral representation formula holds when p ∈ (1, ∞) (with

the same caveat as before when Ω is an exterior domain), this time interpreting the
duality pairings as integration on ∂Ω with respect to σ, i.e.,
∫
βα . 
(∂ uγ )(x) = ar s (∂r Eγβ )(x − y) ∂τ s uα (y) dσ(y)
∂Ω
∫
. 
− (∂ Eγβ )(x − y) ∂νAu β (y) dσ(y)
∂Ω
∫
 
+ (∂ Eγβ )(x − y) Lu β (y) dy. (3.3.9)
Ω

Finally, if in place of (3.3.3)-(3.3.4) one now assumes

 
Nκ (∇u) ∈ L p,q (∂Ω, σ) for some p ∈ n−1 n , ∞ and q ∈ (0, ∞),
 1 M (3.3.10)
as well as Lu ∈ Lloc (Ω, L n ) and P(Lu) ∈ L p,q (∂Ω, σ),

then
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 693
.  M
∂τ s u ∈ H p,q (∂Ω, σ) for all , s ∈ {1, . . . , n}
. A  p,q M (3.3.11)
and ∂ν u ∈ H (∂Ω, σ) ,

and given any  ∈ {1, . . . , n}, γ ∈ {1, . . . , M }, it follows that for L n -a.e. Lebesgue
point x ∈ Ω of ∇u with the property that (3.3.6) holds one has
βα .
(∂ uγ )(x) = (H p, q (∂Ω,σ))∗ ar s (∂r Eγβ )(x − ·) ∂Ω, ∂τ s uα H p, q (∂Ω,σ)

. 
− (H p, q (∂Ω,σ))∗ (∂ Eγβ )(x − ·) ∂Ω, ∂νAu β H p, q (∂Ω,σ)
∫
 
+ (∂ Eγβ )(x − y) Lu β (y) dy (3.3.12)
Ω

with the duality pairings understood in the sense of [113, Lemma 4.7.1, (4.7.2)]
(with Σ := ∂Ω), the summation convention over repeated indices in effect, and also
assuming the decay condition (3.3.8) in the case when Ω is an exterior domain.

A significant consequence of Theorem 3.3.1, which may be justified along the

lines of the proof of Corollary 1.6.2, is that if u is a null-solution of a weakly elliptic
homogeneous constant coefficient second-order system in an exterior domain, then
having ∇u vanish at infinity in the sense of (3.3.8) is equivalent to having
 
(∂ α u)(x) = O |x| 2−n− |α | as |x| → ∞,
(3.3.13)
for each multi-index α ∈ N0n with |α| ≥ 1.

Proof of Theorem 3.3.1 First we shall work under the assumptions made in (3.3.3)-
(3.3.4). From [113, (10.2.14) in Example 10.2.2], [113, (10.2.182)], and (3.3.3)-
(3.3.4) we see that
.  M
∂τ s u ∈ H p (∂Ω, σ) for each , s ∈ {1, . . . , n},
.A  p M (3.3.14)
and ∂ν u ∈ H (∂Ω, σ) .

The remainder of the proof of (3.3.7) and (3.3.9) is divided into two cases,
depending on the size of p.
 
Case I: Suppose p ∈ n−1 n , 1 . To fix ideas, suppose for now that ∂Ω is unbounded,
and choose a Lebesgue point x ∈ Ω for ∇u with the property that (3.3.6) holds. Also,
select two arbitrary indexes, γ ∈ {1, . . . , M } and  ∈ {1, . . . , n}. Bring in a scalar-
∞ n
 η1 ∈ 𝒞 (R ) satisfying η = 0 onB(0, 1), η = 1 on R \ B(0, 2),
valued function n then
for each ε ∈ 0, 2 dist(x, ∂Ω) define ηε (y) := η (y − x)/ε for every y ∈ Rn . Then
 
there exists a constant C ∈ (0, ∞) such that for each ε ∈ 0, 12 dist(x, ∂Ω) we have
694 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

ηε ∈ 𝒞∞ (Rn ), bounded function,

lim ηε (y) = 1 for every y ∈ Rn \ {x},
ε→0+
1 − ηε ∈ 𝒞∞
c (Ω), ηε ≡ 0 on B(x, ε), (3.3.15)
supp (∇ηε ) ⊆ B(x, 2ε) \ B(x, ε), and
|(∇ j η ε )(y)| ≤ Cε −j for all j ∈ N0 and y ∈ Rn .

In particular, for each β ∈ {1, . . . , M } the function Eγβ (x − ·)ηε belongs to 𝒞∞ (Rn )
and coincides with Eγβ (x − ·) near ∂Ω.
Let us also fix some number λ ∈ (1, ∞) and select a function θ ∈ 𝒞∞ c (R )
n

satisfying 0 ≤ θ ≤ 1, θ ≡ 1 on B(0, 1), θ ≡ 0 on R \ B(0, λ) and, for each R > 0,

n

define θ R (y) := θ(y/R) for every y ∈ Rn . Then the general weak-∗ convergence
results established in [113, Lemma 4.8.4] and [113, Lemma 4.8.1] (also bearing
in mind the trivial bounded
 1 embedding L ∞ (∂Ω, σ) → BMO(∂Ω, σ)) imply that
for each fixed ε ∈ 0, 2 dist(x, ∂Ω) , and any two indexes β ∈ {1, . . . , M } and
j ∈ {1, . . . , n}, we have
    
(∂j Eγβ )(x − ·) ∂Ω = (∂j Eγβ )(x − ·)ηε ∂Ω
  
= lim θ R (∂j Eγβ )(x − ·)ηε ∂Ω (3.3.16)
R→∞

⎧ . 
 ∗ ⎪
⎨ 𝒞(n−1)(1/p−1) (∂Ω) ∼ if p < 1,
⎪
weak-∗ in H (∂Ω, σ)
p
= (3.3.17)
⎪ 
⎪ BMO(∂Ω,
⎩ σ) if p = 1.

For future use, let us also observe that, thanks to (1.4.24), Hölder’s inequality, and
[112, Proposition 8.6.3] we have
∫
lim sup |∇θ R ||(∇E)(x − ·)||ηε ||∇u| dL n
R→∞ Ω
∫
1
≤ lim sup n |∇u| dL n
R→∞ R Ω∩[B(0,λR)\B(0,R)]
∫
Cλ,n  np
 n−1
np
≤ lim sup n |∇u| n−1 dL n · Rn(1−(n−1)/(np))
R→∞ R Ω∩[B(0,λR)\B(0,R)]
 
≤ C# Cλ,n Nκ (∇u) L p (∂Ω,σ) · lim sup R−(n−1)/p = 0, (3.3.18)
R→∞

with Cλ,n ∈ (0, ∞) depending only on λ, n, and with C# as in [112, (8.6.49)] (hence
independent of R, since σ(∂Ω) = +∞ in the current case).
To proceed, fix ε ∈ 0, 12 dist(x, ∂Ω) and use (3.3.16) with j := r, along with
the first membership in (3.3.5), the compatibility condition in [113, Lemma 4.6.4],
formula [113, (10.2.12], and (3.3.18) to express the first term in the right-hand side
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 695

of (3.3.7) as
βα   .
ar s (∂r Eγβ )(x − ·) ∂Ω , ∂τ s uα (3.3.19)
   .
βα
= lim ar s θ R (∂r Eγβ )(x − ·)ηε ∂Ω , ∂τ s uα
R→∞
. βα
= lim (Lip c (∂Ω)) ∂τ s uα, ar s θ R (∂r Eγβ )(x − ·)ηε Lip c (∂Ω)
R→∞
∫  
βα 
= lim ar s ∂ θ R (∂r Eγβ )(x − ·)ηε (∂s uα )
R→∞ Ω
  
− ∂s θ R (∂r Eγβ )(x − ·)ηε (∂ uα ) dL n
∫ 
βα
= lim ar s − θ R (∂ ∂r Eγβ )(x − ·)ηε (∂s uα )
R→∞ Ω

+ (∂r Eγβ )(x − ·)(∂ ηε )(∂s uα )

+ θ R (∂s ∂r Eγβ )(x − ·)ηε (∂ uα )

− (∂r Eγβ )(x − ·)(∂s ηε )(∂ uα ) dL n .

In relation to the last integrand above observe that, thanks to (1.4.33), we have
βα
ar s θ R (∂s ∂r Eγβ )(x − ·)ηε (∂ uα ) = 0. (3.3.20)

Moving on, we wish to treat the second term appearing in the right-hand side of
(3.3.7). To do so, recall from (A.0.113)-(A.0.114) that the weak conormal derivative
of u associated with the coefficient tensor A is defined as the distribution
.      M
∂νAu := ν • Fα 1≤α ≤M in Lipc (∂Ω) where
  (3.3.21)
αβ
Fα := (A∇u)α = ar s ∂s uβ 1≤r ≤n for each α ∈ {1, . . . , M }.

Combining (3.3.16) (presently used with j := ), [113, Lemma  4.6.4], and [112,
Proposition 4.2.3], for each fixed number ε ∈ 0, 12 dist(x, ∂Ω) we obtain
696 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
  . 
− (∂ Eγβ )(x − ·) ∂Ω , ∂νAu β
   . 
= − lim θ R (∂ Eγα )(x − ·)ηε ∂Ω
, ∂νAu α
R→∞

 .A 
= − lim (Lip c (∂Ω)) ∂ν u α, θ R (∂ Eγα )(x − ·)ηε Lip c (∂Ω)
R→∞
∫
= − lim (divFα )θ R (∂ Eγα )(x − ·)ηε dL n
R→∞ Ω
∫
 
− lim Fα, ∇ θ R (∂ Eγα )(x − ·)ηε dL n
R→∞ Ω
∫
=− (Lu)α (∂ Eγα )(x − ·) dL n
Ω
∫
αβ  
− lim ar s (∂s uβ )∂r θ R (∂ Eγα )(x − ·)ηε dL n, (3.3.22)
R→∞ Ω

where the last equality is based on Lebesgue’s Dominated Convergence Theorem

(whose applicability is ensured by (3.3.6) and (1.4.24)). By combining (3.3.22) with
(3.3.18) we therefore arrive at
  . 
− (∂ Eγβ )(x − ·) ∂Ω , ∂νAu β (3.3.23)
∫
=− (∂ Eγβ )(x − ·)(Lu)β dL n
Ω
∫ 
αβ
− lim ar s − θ R (∂r ∂ Eγα )(x − ·)ηε
R→∞ Ω

+ (∂ Eγα )(x − ·)(∂r ηε ) (∂s uβ ) dL n
∫
=− (∂ Eγβ )(x − ·)(Lu)β dL n
Ω
∫ 
βα
− lim ar s − θ R (∂r ∂ Eγβ )(x − ·)ηε
R→∞ Ω

+ (∂ Eγβ )(x − ·)(∂r ηε ) (∂s uα ) dL n,

where in the last step we have simply interchanged the roles of α and β in the very
last integral.
Adding (3.3.19) with (3.3.23) then yields, after taking (3.3.20) into account and
canceling the terms involving ∂r ∂ Eγβ ,
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 697

βα   .   . 
ar s (∂r Eγβ )(x − ·) ∂Ω , ∂τ s uα − (∂ Eγβ )(x − ·) ∂Ω , ∂νAu β
∫ 
βα
= ar s (∂r Eγβ )(x − ·)(∂ ηε )(∂s uα ) − (∂r Eγβ )(x − ·)(∂s ηε )(∂ uα )
Ω

− (∂ Eγβ )(x − ·)(∂r ηε )(∂s uα ) dL n
∫
− (∂ Eγβ )(x − ·)(Lu)β dL n . (3.3.24)
Ω

Note that
∫  
βα
ar s (∂r Eγβ )(x − ·)(∂ ηε )(∂s uα ) − (∂ Eγβ )(x − ·)(∂r ηε )(∂s uα ) dL n
Ω
∫  
βα
= ar s (∂r Eγβ )(x − y)(∂ ηε )(y) − (∂ Eγβ )(x − y)(∂r ηε ) (∂s uα )(y) dy
Ω

= Iε + IIε (3.3.25)

where
∫  
βα
Iε := ar s (∂r Eγβ )(x − y)(∂ ηε )(y) − (∂ Eγβ )(x − y)(∂r ηε ) ×
Ω
 
× (∂s uα )(y) − (∂s uα )(x) dy (3.3.26)

and
∫

βα
IIε := ar s (∂r Eγβ )(x − y)(∂ ηε )(y)
Ω
!

− (∂ Eγβ )(x − y)(∂r ηε ) dy (∂s uα )(x). (3.3.27)

Since
∫  
(∂r Eγβ )(x − y)(∂ ηε )(y) − (∂ Eγβ )(x − y)(∂r ηε ) dy
Ω

= D  (Ω) (∂r Eγβ )(x − ·), ∂ (ηε − 1) D(Ω)

− D  (Ω) (∂ Eγβ )(x − ·), ∂r (ηε − 1) D(Ω)

= D  (Ω) Eγβ (x − ·), ∂r ∂ (ηε − 1) D(Ω)

− D  (Ω) Eγβ (x − ·), ∂ ∂r (ηε − 1) D(Ω) = 0, (3.3.28)

698 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
 
it follows that for each ε ∈ 0, 12 dist(x, ∂Ω) we have

IIε = 0. (3.3.29)

Also, making use of (3.3.15) and the fact that x is a Lebesgue point for ∇u, we may
write
⨏
lim sup Iε ≤ C · lim sup (∇u)(y) − (∇u)(x) dy = 0. (3.3.30)
ε→0+ ε→0+ B(x,2ε)

Finally, decompose
∫
βα
− ar s (∂r Eγβ )(x − ·)(∂s ηε )(∂ uα ) dL n = IIIε + IVε (3.3.31)
Ω

where
∫
βα # $
IIIε := − ar s (∂r Eγβ )(x − y)(∂s ηε )(y) (∂ uα )(y) − (∂ uα )(x) dy (3.3.32)
Ω

and
∫ !
βα
IVε := − ar s (∂r Eγβ )(x − y)(∂s ηε )(y) dy (∂ uα )(x). (3.3.33)
Ω

Since x is a Lebesgue point for ∇u,

⨏
lim sup IIIε ≤ C · lim sup (∇u)(y) − (∇u)(x) dy = 0. (3.3.34)
ε→0+ ε→0+ B(x,2ε)

In addition, (1.4.33) eventually permits us to conclude that for each β ∈ {1, . . . , M }

we have
∫
βα
lim+ ar s (∂r Eγβ )(x − y)(∂s ηε )(y) dy
ε→0 Ω
∫
βα
= lim+ ar s (∂r Eγβ )(x − y)∂s (ηε − 1)(y) dy
ε→0 Ω

βα  
= − lim+ D  (Ω) ar s ∂r Eγβ (x − ·) , ∂s (ηε − 1) D(Ω)
ε→0

βα  
= lim+ D  (Ω) ar s ∂s ∂r Eγβ (x − ·) , ηε − 1 D(Ω)
ε→0

= lim+ δαγ (ηε − 1)(x) = −δαγ . (3.3.35)

ε→0

Thus,
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 699

lim IVε = (∂ uγ )(x). (3.3.36)

ε→0+

Collectively, (3.3.24), (3.3.25), (3.3.29), (3.3.30), (3.3.31), (3.3.34), and (3.3.36)

prove (3.3.7) in the case when ∂Ω is unbounded.
The case when Ω is bounded is treated very similarly, since (3.3.18) continues to
be valid in such a scenario. Finally, when Ω is an exterior domain, estimate [112,
(8.6.51)] (on which (3.3.18) is based) may fail. Nonetheless, the decay condition
(3.3.8) turns out to be an appropriate substitute, as this implies
∫
lim sup |∇θ R ||(∇E)(x − ·)||ηε ||∇u| dL n
R→∞ Ω
⨏
≤ lim sup |∇u| dL n = 0. (3.3.37)
R→∞ B(0,λ R)\B(0,R)

Case II: Suppose p ∈ (1, ∞). We reason largely as before, with natural alterations.
Two key aspects to bear in mind are as follows. First, [112, (3.6.27)] implies that the
memberships in (3.3.14) currently read
.  M
∂τ s u ∈ L p (∂Ω, σ) for each , s ∈ {1, . . . , n},
.A  p M (3.3.38)
and ∂ν u ∈ L (∂Ω, σ) .

Second, all duality pairings used in the past now become ordinary integral pairings
on ∂Ω, with respect to the measure∫ σ. Upon also observing that if p is the Hölder

conjugate exponent of p, we have ∂Ω |(∇E)(x −·)| dσ < +∞ thanks to (1.4.24) and
p

[112, (7.2.5)], the proof proceeds as before, and the integral representation formula
(3.3.9) follows.
Finally, consider the case when in place of (3.3.3)-(3.3.4) we now assume the
conditions imposed in (3.3.10). In such a scenario, the memberships claimed in
(3.3.11) make use of the full force of [113, (10.2.13), (10.2.182)]. Granted these,
formula (3.3.12) may be justified much as (3.3.7), making similar adjustments as in
the proof of Theorem 3.1.2. 

We momentarily digress for the purpose of making the integral representation

formulas from Theorem 3.3.1 corresponding to null-solutions of the given weakly
elliptic system subject to the technique developed earlier in Proposition 3.1.3.

Remark 3.3.2 It is of interest to point out that the version of Theorem 3.3.1 corre-
sponding to the case when the given function u is a smooth null-solution of the system
L may be directly deduced from Proposition 3.1.3 for a suitable choice of u, w, and
D. To be specific, work in the context of Theorem 3.3.1 in which we now assume
 M
that u ∈ 𝒞∞ (Ω) and Lu = 0 in Ω. Fix x ∈ Ω, along with γ ∈ {1, . . . , M } and
 ∈ {1, . . . , n}, all arbitrary. With the summation convention over repeated indices
in effect, define the Cn·M+M -valued function
700 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
 βα 
( ar s (∂r Eγβ )(x − ·) 1≤α1≤s ≤n +
w := )   ≤M , , (3.3.39)
* − (∂ E γβ )(x − ·) 1≤β ≤M -

the Cn·M -valued function  

U := ∂j uα 1≤ j ≤n , (3.3.40)
1≤α ≤M
and the first-order (n · M + M) × (n · M) system
 
D1
D := (3.3.41)
D2

where, with πs,α denoting the canonical coordinate projection selecting the (s, α)-
entry of a matrix-valued function,
   αβ 
D1 := ∂ ◦ πs,α − ∂s ◦ π ,α 1≤s ≤n and D2 := ar s ∂r ◦ πs,β 1≤α ≤M . (3.3.42)
1≤α ≤M

We then have  
∂ ∂s uα − ∂s ∂ uα 1≤s ≤n
( 1≤α ≤M +
DU = )  αβ  ,=0 (3.3.43)
* ar s ∂r ∂s uβ 1≤α ≤M -
 αβ 
since ar s ∂r ∂s uβ 1≤α ≤M = Lu = 0, thanks to the current assumptions. Next, a
direct computation based on (3.3.41)-(3.3.42) shows that D , the real transpose of
D, is given by
  
φsα 1≤s ≤n  
μα
D   1≤α ≤M = − ∂ φsα + δ s ∂j φ jα − ar s ∂r ψμ 1≤s ≤n (3.3.44)
ψβ 1≤β ≤M 1≤α ≤M

so if w is as in (3.3.39) we obtain
 
D w = − δ s δαγ δx 1≤s ≤n . (3.3.45)
1≤α ≤M
  n·M
In particular, D w ∈ ℰ(Ω) and, for U as in (3.3.40),

[ℰ(Ω)] n·M U, D w [ℰ (Ω)] n·M = −(∂ uγ )(x). (3.3.46)

In addition,
   . 
  (−i)Sym D1 ; ν ∂ u 1≤s ≤n
( τ s α 1≤α ≤M +
(−i)Sym D; ν • U =   •U = )  .  ,. (3.3.47)
(−i)Sym D2 ; ν ∂ A u
* ν 1≤β ≤M -

Hence, in the case when ∂Ω is unbounded and p ≤ 1, we have

3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 701
   κ−n.t. 
(−i)Sym D; ν • U, w ∂Ω

βα   .
= ar s (∂r Eγβ )(x − ·) ∂Ω , ∂τ s uα
  . 
− (∂ Eγβ )(x − ·) ∂Ω , ∂νAu β (3.3.48)

which in concert with (3.3.46) proves, thanks to (3.1.46) in Proposition 3.1.3 (written
with U in place of u), the versions of (3.3.7) and (3.3.9) corresponding to situation
when Lu = 0 in Ω. The case when ∂Ω is bounded, or p > 1, is deal with similarly,
now employing (3.1.45).

It is also possible to prove a version of Theorem 3.3.1 in which the membership

of the nontangential maximal function to Lebesgue spaces is now replaced by a
membership to a suitable Generalized Banach Function Space.

Proposition 3.3.3 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be an Ahlfors regular

domain, and abbreviate σ := H n−1 ∂Ω. For some M ∈ N, consider a coefficient
αβ 
tensor A = ar s 1≤r,s ≤n with complex entries, with the property that the M × M
1≤α,β ≤M
homogeneous second-order system L = L A associated with A in Rn as in (1.3.15)
is weakly elliptic (in the sense of (1.3.3)). Also, denote by E = (Eγβ )1≤γ,β ≤M the
matrix-valued fundamental solution associated with L as in Theorem 1.4.2.
In addition, suppose X is a Generalized Banach Function Space on (∂Ω, σ) such
that
M ∂Ω : X → X and M ∂Ω : X → X
(3.3.49)
are well-defined bounded mappings,
where M ∂Ω is the Hardy-Littlewood maximal operator on ∂Ω, and X denotes the
associated space of X (cf. [113, Definitions 5.1.4, 5.1.11]).
 M
In this setting, assume u = (uβ )1≤β ≤M ∈ 𝒞∞ (Ω) is a vector-valued function
which, for some aperture parameter κ > 0, satisfies

Nκ (∇u) ∈ X and Lu = 0 in Ω. (3.3.50)

In the case when Ω is an exterior domain, make the additional assumption that there
exists λ ∈ (1, ∞) such that
⨏
|∇u| dL n = o(1) as R → ∞. (3.3.51)
B(0,λR)\B(0,R)

Then
.  M .  M
∂νAu ∈ X and ∂τ s u ∈ X for any , s ∈ {1, . . . , n}, (3.3.52)

and, given any  ∈ {1, . . . , n} and γ ∈ {1, . . . , M }, at each point x ∈ Ω one has
702 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
∫
βα . 
(∂ uγ )(x) = ar s (∂r Eγβ )(x − y) ∂τ s uα (y) dσ(y)
∂Ω
∫
. 
− (∂ Eγβ )(x − y) ∂νAu β (y) dσ(y). (3.3.53)
∂Ω

Proof The argument presented in Remark 3.3.2 (for the same choice of D, U, and
w) allows us to reduce matters to the case of first-order systems, treated earlier in
Proposition 3.1.4. 

We now turn to the task of presenting the Fatou-type theorem elaborating on the
result advertised earlier, in (3.3.2).

Theorem 3.3.4 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be an arbitrary UR domain,

and abbreviate σ := H n−1 ∂Ω. Consider a homogeneous constant (complex) co-
efficient second-order M × M system L in Rn (for some M ∈ N) which is weakly
 1,1  M
elliptic (in the sense of (1.3.3)), and assume u ∈ Wloc (Ω) is a vector-valued
function which, for some aperture parameter κ > 0, satisfies
 
Nκ (∇u) ∈ L p (∂Ω, σ) with p ∈ n−1
n ,∞
 M (3.3.54)
and Lu = 0 in D (Ω) .

Then the following statements are true.

(a) The nontangential boundary trace

κ−n.t.
(∇u) ∂Ω exists (in Cn·M ) at σ-a.e. point on ∂Ω and
κ−n.t.
the function (∇u) ∂Ω is σ-measurable on ∂Ω;
κ−n.t.   n·M (3.3.55)
also, (∇u) ∂Ω belongs to the space L p (∂Ω, σ)
 κ−n.t. 
and one has (∇u) ∂Ω [L p (∂Ω,σ)]n·M ≤ Nκ (∇u) L p (∂Ω,σ) .

In addition, for any other given aperture parameter κ  ∈ (0, ∞) the nontangential
κ  −n.t. κ−n.t.
boundary trace (∇u) ∂Ω also exists, and agrees with (∇u) ∂Ω , at σ-a.e. point
on ∂Ω.
Furthermore,
κ−n.t.
the nontangential boundary trace u ∂Ω exists
(3.3.56)
at σ-a.e. point on ∂Ω and is a σ-measurable function,

and for any other given aperture parameter κ  ∈ (0, ∞) the nontangential
κ  −n.t. κ−n.t.
boundary trace u ∂Ω exists and agrees with u ∂Ω at σ-a.e. point on ∂Ω.
(b) Given any coefficient tensor A which allows representing L = L A, one has:
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 703
 n−1  .  M
if p ∈ , 1 then ∂νAu ∈ H p (∂Ω, σ)
n and
.  p M (3.3.57)
∂τ j k u ∈ H (∂Ω, σ) for each j, k ∈ {1, . . . , n}.

(c) The following H p /L p -styled boundary maximum principle holds:

  . 
Nκ (∇u) p ≈ ∂νAu[H p (∂Ω,σ)] M
L (∂Ω,σ)


n
.   
+ ∂τ u p if p ∈ n−1
jk [H (∂Ω,σ)] M n ,1 , (3.3.58)
j,k=1

and
   κ−n.t. 
Nκ (∇u) ≈ (∇u) ∂Ω [L p (∂Ω,σ)]n·M
L p (∂Ω,σ)

n 
 
 κ−n.t. κ−n.t.

≈ ν j (∂k u) ∂Ω − νk (∂j u) ∂Ω 
[L p (∂Ω,σ)] M
j,k=1
 
+ ∂νAu[L p (∂Ω,σ)] M if p ∈ (1, ∞), (3.3.59)

provided, in the case when Ω is an exterior domain, it is also assumed (for both
equivalences) that ∇u vanishes at infinity (in the sense of (3.3.8)).
Moreover, if Ω ⊆ Rn is actually
 an NTA domain with an Ahlfors regular
boundary, then if p ∈ n−1
n , 1 it follows that
κ−n.t.
the nontangential boundary trace u ∂Ω exists (in C M ) at σ-a.e. point
on ∂Ω, as a vector-valued function said trace belongs to the homoge-
 .p M
neous Hardy-based Sobolev space H1 (∂Ω, σ) ,
(3.3.60)
and, if ∇u also vanishes at infinity (in the sense of (3.3.8)) when Ω is an exterior
domain, one has (compare with (3.3.58))
  .   κ−n.t. 
Nκ (∇u) ≈ ∂νAu[H p (∂Ω,σ)] M + u ∂Ω [ H. p (∂Ω,σ)] M , (3.3.61)
L p (∂Ω,σ) 1

while if p ∈ (1, ∞) it follows that

κ−n.t.
the nontangential boundary trace u ∂Ω exists (in C M ) at σ-a.e. point
on ∂Ω, as a vector-valued function said trace belongs to the homoge-
.p M
neous Sobolev space L1 (∂Ω, σ) ,
(3.3.62)
and if ∇u also vanishes at infinity (in the sense of (3.3.8)) when Ω is an exterior
domain, one has (compare with (3.3.59))
     κ−n.t. 
Nκ (∇u) ≈ ∂νAu[L p (∂Ω,σ)] M + u ∂Ω [ L. p (∂Ω,σ)] M . (3.3.63)
L p (∂Ω,σ) 1
704 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

(d) Given any q ∈ (1, ∞), there exists some C = C(Ω, L, p, q, κ) ∈ (0, ∞), inde-
pendent of u, such that the following mixed-norm estimate holds:
∫  p1
∫  qp  
|∇ u| dist(·, ∂Ω)
2 q q−n
dL n
dσ(x) ≤ C Nκ (∇u) L p (∂Ω,σ) .
∂Ω Γκ (x)
(3.3.64)
In particular, with the L q -based area-function defined as in (3.1.119), the esti-
mate in (3.3.64) may be recast as
   
A q,κ (∇u) p ≤ C Nκ (∇u) L p (∂Ω,σ) . (3.3.65)
L (∂Ω,σ)

Furthermore, when the integrability exponent p belongs to (1, ∞) there exists

some constant C = C(Ω, L, p, κ) ∈ (0, ∞) with the property that
∫  1/p  
|(∇2 u)(x)| p dist(x, ∂Ω) p−1 dx ≤ C Nκ (∇u) L p (∂Ω,σ) . (3.3.66)
Ω

(e) In the scenario when the set Ω is actually a bounded NTA domain with an Ahlfors
 in R and when p ∈ (1, ∞), it follows that for each exponent
regular n
 nboundary
q ∈ n+1/p , ∞ there exists some constant C = C(Ω, L, κ, p, q) ∈ (0, ∞) with the
property that
   κ−n.t. 
max ∇u[F p, q (Ω)]n·M , ∇u[B p, p (Ω)]n·M ≤ C (∇u) ∂Ω [L p (∂Ω,σ)]n·M .
1/p 1/p
(3.3.67)
In particular, corresponding to the case when p = q = 2, one has
 κ−n.t. 
∇u[H 1/2 (Ω)]n·M ≤ C (∇u) ∂Ω [L 2 (∂Ω,σ)]n·M . (3.3.68)

(f) Suppose the first line in (3.3.54) is replaced by

 
Nκ (∇u) ∈ L p,q (∂Ω, σ) for some p ∈ n−1 n , ∞ and q ∈ (0, ∞). (3.3.69)
κ−n.t.
Then nontangential boundary trace (∇u) ∂Ω exists at σ-a.e. point on ∂Ω,
  n·M
belongs to the Lorentz space L p,q (∂Ω, σ) , is independent of the aperture
parameter, and satisfies the estimate
 κ−n.t. 
(∇u)  p, q ≤ Nκ (∇u) L p, q (∂Ω,σ) . (3.3.70)
∂Ω [L (∂Ω,σ)] n·M

Moreover, under the additional assumption that ∇u vanishes at infinity (in the
sense of (3.3.8)) in the case when Ω is an exterior domain, one has

   .A  
n
. 
Nκ (∇u) p, q ≈ ∂ u p, q + ∂τ u p, q ,
L (∂Ω,σ) ν [H (∂Ω,σ)] M jk [H (∂Ω,σ)] M
j,k=1
(3.3.71)
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 705
   κ−n.t. 
Nκ (∇u) ≈ (∇u) ∂Ω [L p, q (∂Ω,σ)]n·M if p ∈ (1, ∞), (3.3.72)
L p, q (∂Ω,σ)

and    
A κ (∇u) ≤ C Nκ (∇u) L p, q (∂Ω,σ) . (3.3.73)
L p, q (∂Ω,σ)

(g) Suppose the membership in the first line in (3.3.54) is now replaced by

Nκ (∇u) ∈ X, (3.3.74)

where X is a Generalized Banach Function Space on (∂Ω, σ) with the property

that
M ∂Ω : X → X and M ∂Ω : X → X
(3.3.75)
are well-defined bounded mappings
(with M ∂Ω denoting the Hardy-Littlewood maximal operator on ∂Ω, and with
X denoting the associated space of X; cf. [113, Definitions 5.1.4, 5.1.11]). Then,
in place of (3.3.55), one has
κ−n.t.
(∇u) ∂Ω exists (in Cn·M ) at σ-a.e. point on ∂Ω and
κ−n.t.
the function (∇u) ∂Ω is σ-measurable on ∂Ω;
κ−n.t.   n·M (3.3.76)
also, (∇u) ∂Ω belongs to the space X
 κ−n.t. 
and one has (∇u) ∂Ω [X]n·M ≤ Nκ (∇u)X .

Moreover, for any other given aperture parameter κ  ∈ (0, ∞) the nontangential
κ  −n.t. κ−n.t.
boundary trace (∇u) ∂Ω also exists, and agrees with (∇u) ∂Ω , at σ-a.e. point
on ∂Ω. Finally, in place of (3.3.59) one presentaly has
   κ−n.t. 
Nκ (∇u) ≈ (∇u)  n·M (3.3.77)
X ∂Ω [X]

n 
 
 κ−n.t. κ−n.t.
  
≈ ν j (∂k u) ∂Ω − νk (∂j u) ∂Ω  + ∂νAu[X] M
[X] M
j,k=1

if, in the case when Ω is an exterior domain, one also assumes (for both equiv-
alences) that ∇u vanishes at infinity (in the sense of (3.3.8)).
In the regime p ∈ [1, ∞), part (a) of the above theorem together with [113,
 αβ imply that if the M × M system L is associated with the coefficient
Theorem 10.2.24]
tensor A = ar s 1≤r,s ≤n as in (1.3.15), then for Ω and u as in the statement
1≤α,β ≤M
of Theorem 3.3.4 we have (with ν = (νr )1≤r ≤n denoting the geometric measure
theoretic outward unit normal to Ω)
.  M
∂νAu belongs to the Lebesgue space L p (∂Ω, σ) and
.  κ−n.t.  (3.3.78)
αβ
∂νAu = νr ar s (∂s uβ ) ∂Ω at H n−1 -a.e. point on ∂Ω.
1≤α ≤M
706 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

As mentioned earlier, we shall give Theorem 3.3.4 two proofs.

First proof of Theorem 3.3.4 The claims in (3.3.55) are proved in the same manner
as Theorem 3.1.6, making use of the integral representation formula from Theo-
rem 3.3.1 (in lieu of Theorem 3.1.1). The integrability properties and independence
of the nontangential boundary trace on the aperture parameter follow from assump-
tions, [112, Proposition 8.9.8], [112, Corollary 8.9.6], [112, (8.9.8)], and [112,
Proposition 8.8.6]. The claims made in (3.3.56) (and just below) are direct conse-
quences of Proposition 3.1.13. Also, the claims in (3.3.57) are implied by item (1)
in [113, Theorem 10.2.24] and [113, (10.2.14) in Example 10.2.2].
Next, (3.3.58) is a consequence of (3.3.7), (2.4.14), and the estimates that ac-
company the memberships in (3.3.57) (cf. [113, (10.2.14), (10.2.182)]). Moving on,
(3.3.59) is seen from (3.3.9), (2.4.9), and (1.7.54), keeping in mind (3.3.55) and [113,
(10.2.92), (10.2.186)]. Also, the claims made in (3.3.60)-(3.3.61) are implied by the
trace result established in the last theorem of [115, §2.2] and (3.3.58), while the
claims made in (3.3.62)-(3.3.63) follow from (3.3.59) and [113, Propositions 11.3.2,
11.5.12] (bearing in mind (3.3.54), (3.3.55), and [113, (11.5.129)]).
Going further, the claims in items (d)-(e) are implied by the integral representation
formula from Theorem 3.3.1, items (4), (6), (7) in Theorem 2.4.1, together with
(3.3.58) and (3.3.59). The claims in item (f) may be justified along the lines of the
proofs of their earlier counterparts, making use of (3.3.11)-(3.3.12) in Theorem 3.3.1.
Finally, the claims in item (g) are established reasoning as before, now making use
of the integral representation formula from Proposition 3.3.3, and bearing in mind
[113, (5.1.12), (5.1.13)], [113, Proposition 5.2.7], and item (4) in Theorem 2.4.1. 

In the second proof of Theorem 3.3.4, given next, the strategy is to reduce matters
to a suitable application of the Fatou-type theorem for first-order injectively elliptic
systems from Theorem 3.1.6.

 αβ  suppose the M × M system L

Second proof of Theorem 3.3.4 To get things started,
is associated with some coefficient tensor A = ar s 1≤r,s ≤n as in (1.3.15). Bring
1≤α,β ≤M
in the first-order system D A defined as in (1.3.39). From Example 1.3.12 we know
that D A is injectively elliptic. Going further, assume u = (uβ )1≤β ≤M is as in the
statement of Theorem 3.3.4 and set w := ∇u, i.e.,
 β β
w = ws 1≤s ≤n with ws := ∂s uβ for each
1≤β ≤M (3.3.79)
s ∈ {1, . . . , n} and β ∈ {1, . . . , M }.
  n×M
Then w ∈ L 1loc (Ω, L n ) is a vector-valued function which, thanks to (1.3.41),
satisfies
Nκ w ≈ Nκ (∇u) and
(3.3.80)
D A w = 0 in the sense of distributions in Ω.
Moreover, from (1.3.39), (A.0.107), and [113, Proposition 10.2.20] we see that
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 707
. 
  ( ∂τsr uβ 1≤r,s ≤n +
(−i)Sym D A; ν • w = ) . A1≤β ≤M , . (3.3.81)
* ∂ν u -
Granted (1.3.40), (3.3.80), and (3.3.81), Theorem 3.1.6 applies (with u replaced by
w) and all desired conclusions follow directly from it (with the exception of (3.3.56),
(3.3.60)-(3.3.61) and (3.3.62)-(3.3.63) which are established as before, based on
Proposition 3.1.13, (3.3.58), (3.3.59), the trace result established in the last theorem
of [115, §2.2], and [113, Propositions 11.3.2, 11.5.12]). 
There is also a local version of Theorem 3.3.4, of the following sort.

Theorem 3.3.5 Suppose Ω ⊆ Rn , where n ∈ N with n ≥ 2, is an arbitrary UR

domain and abbreviate σ := H n−1 ∂Ω. Also, consider a homogeneous constant
(complex) coefficient second-order M × M system L in Rn (for some M ∈ N) which
 1,1  M
is weakly elliptic (in the sense of (1.3.3)), and assume u ∈ Wloc (Ω) is a vector-
valued function which, for some aperture parameter κ > 0 and some truncation
parameter ε > 0, satisfies

Nκε (∇u) ∈ Lloc (∂Ω, σ) for some p ∈ (1, ∞]

p

 M (3.3.82)
and Lu = 0 in D (Ω) .
κ−n.t.
Then the nontangential boundary trace (∇u) ∂Ω exists (in C M×n ) at σ-a.e. point
κ−n.t.
on ∂Ω, is actually independent of the aperture parameter κ, the function (∇u) ∂Ω
 p  M×n
is σ-measurable on ∂Ω and belongs to the space Lloc (∂Ω, σ) .
κ−n.t.
Furthermore, the nontangential boundary trace u ∂Ω also exists at σ-a.e. point
on ∂Ω and is a σ-measurable function which is actually independent of the aperture
parameter.

It is worth noting that the membership condition in the first line of (3.3.82)
is satisfied if, for example, Nκε (∇u) ∈ L p (∂Ω, wσ) for some p ∈ (1, ∞) and
w ∈ Ap (∂Ω, σ) (see [112, (7.7.105)] in this regard), or if Nκε (∇u) ∈ M p,λ (∂Ω, σ)
for some p ∈ (1, ∞) and λ ∈ (0, n − 1) (cf. [113, (6.2.25)]).
Proof of Theorem 3.3.5 All claims pertaining to the nontangential boundary trace
κ−n.t.
(∇u) ∂Ω may be deduced directly from Theorem 3.1.9 by reasoning as in the second
κ−n.t.
proof of Theorem 3.3.4 presented above. The claims regarding u ∂Ω are seen from
Proposition 3.1.13. 
We continue by making yet another remark pertaining to Theorem 3.3.4. Specifi-
cally, in the case when Ω ⊆ Rn is actually a bounded UR domain, we may relax the
hypotheses in (3.3.54) to
 
Nκ (∇u) ∈ L p (∂Ω, σ) with p ∈ n−1n ,∞
 M (3.3.83)
and Lu ∈ L q (Ω, L n ) for some q ∈ (n, ∞]
708 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
κ−n.t.
and still conclude11 that the nontangential boundary trace (∇u) ∂Ω exists at σ-a.e.
point on ∂Ω and is independent of the aperture parameter κ.
Moving on, by combining the Fatou-type result from Theorem 3.3.5 with the
integral representation formulas from Theorem 1.5.10 we obtain the following result:

Theorem 3.3.6 Suppose Ω ⊆ Rn , where n ∈ N with n ≥ 2, is an arbitrary UR

domain. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn ) the geometric
measure theoretic outward unit normal to Ω. With the summation convention over
repeated indices understood throughout, let
 αβ 
L = ar s ∂r ∂s 1≤α,β ≤M (3.3.84)

be a homogeneous, weakly elliptic, second-order M × M system in Rn , with comp-

lex constant coefficients, and denote by E = (Eγβ )1≤γ,β ≤M the matrix-valued
fundamental solution associated with L as in Theorem 1.4.2. In this setting, assume
 1,1  M
u = (uβ )1≤β ≤M ∈ Wloc (Ω) is a vector-valued function which, for some aperture
parameter κ > 0 and some truncation parameter ε > 0, satisfies

Nκε (∇u) ∈ Lloc (∂Ω, σ) for some p ∈ (1, ∞],

p

∫   (3.3.85)
Nκ (∇u) (y)
and dσ(y) < +∞.
∂Ω 1 + |y|
n−1

With Lu considered in the sense of distributions in Ω, suppose also that

Lu = 0 in Ω. (3.3.86)

Then the function u is of class 𝒞∞ in Ω and for any aperture parameter κ  > 0
κ  −n.t.
the nontangential trace (∇u) ∂Ω exists σ-a.e. on ∂Ω and is actually independent
of κ . Moreover, with the dependence on the parameter κ  dropped and having fixed
 ∈ {1, . . . , n} along with γ ∈ {1, . . . , M } arbitrary, it follows that for each point
x ∈ Ω one has (with absolutely convergent integrals)
∫
βα
(∂ uγ )(x) = ar s (∂r Eγβ )(x − y)×
∂Ω
  n.t.   n.t.  
× ν (y) (∂s uα ) ∂Ω (y) − νs (y) (∂ uα ) ∂Ω (y) dσ(y)
∫
αβ  n.t. 
− (∂ Eγα )(x − y)νr (y)ar s (∂s uβ ) ∂Ω (y) dσ(y) (3.3.87)
∂Ω

if either Ω is bounded, or ∂Ω is unbounded. In the case when Ω is an exterior

domain, the same conclusion holds true under the additional assumption that there
exists λ ∈ (1, ∞) such that

11 based on Theorem 3.3.1, the same way Theorem 3.1.7 has been proved with the help of Theo-
rem 3.1.6
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 709
⨏
|∇u| dL n = o(1) as R → ∞. (3.3.88)
B(0,λ R)\B(0,R)

In addition, for each x ∈ Rn \ Ω one has

∫  
βα  n.t.   n.t. 
0= ar s (∂r Eγβ )(x − y) ν (y) (∂s uα ) ∂Ω (y) − νs (y) (∂ uα ) ∂Ω (y) dσ(y)
∂Ω
∫
αβ  n.t. 
− (∂ Eγα )(x − y)νr (y)ar s (∂s uβ ) ∂Ω (y) dσ(y) (3.3.89)
∂Ω

with the same convention as before in the case when Ω is an exterior domain.

Proof All claims are consequence of elliptic regularity (cf. [109]), the Fatou-type
result from Theorem 3.3.5, the integral formulas from Theorem 1.5.10, [112, Propo-
sition 8.8.6], and [112, Corollary 8.9.9]. 

We continue by presenting an adaptation of Theorem 3.3.4 to the scale of Morrey

spaces.

Theorem 3.3.7 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be an arbitrary UR domain,

and abbreviate σ := H n−1 ∂Ω. Consider a homogeneous constant (complex) co-
efficient second-order M × M system L in Rn (for some M ∈ N) which is weakly
 1,1  M
elliptic (in the sense of (1.3.3)), and assume u ∈ Wloc (Ω) is a vector-valued
function which, for some aperture parameter κ > 0, satisfies

Nκ (∇u) ∈ M p,λ (∂Ω, σ) with p ∈ (1, ∞), λ ∈ (0, n − 1),

 M (3.3.90)
and Lu = 0 in D (Ω) .

Then the nontangential boundary trace

κ−n.t.
(∇u) ∂Ω exists (in Cn·M ) at σ-a.e. point on ∂Ω and
κ−n.t. (3.3.91)
the function (∇u) ∂Ω is σ-measurable on ∂Ω.

In addition, for any other given aperture parameter κ  ∈ (0, ∞) the nontangential
κ  −n.t. κ−n.t.
boundary trace (∇u) ∂Ω also exists, and agrees with (∇u) ∂Ω , at σ-a.e. point on
∂Ω. Also,
κ−n.t.   n·M
(∇u) ∂Ω belongs to the Morrey space M p,λ (∂Ω, σ)
 κ−n.t.  (3.3.92)
and one has (∇u) ∂Ω [M p, λ (∂Ω,σ)]n·M ≤ Nκ (∇u) M p, λ (∂Ω,σ) .

In fact, the following Morrey-styled boundary maximum principle holds:

   κ−n.t. 
Nκ (∇u) ≈ (∇u) ∂Ω [M p, λ (∂Ω,σ)]n·M (3.3.93)
M p, λ (∂Ω,σ)
710 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

provided, in the case when Ω is an exterior domain, it is also assumed that ∇u

vanishes at infinity (in the sense of (3.3.8)).
Finally, if (3.3.90) is strengthened to

Nκ (∇u) ∈ M̊ p,λ (∂Ω, σ) with p ∈ (1, ∞), λ ∈ (0, n − 1),

 M (3.3.94)
and Lu = 0 in D (Ω) ,

then (3.3.91) continues to hold and one now has

κ−n.t.   n·M
(∇u) ∂Ω ∈ M̊ p,λ (∂Ω, σ) . (3.3.95)

Proof All claims, except (3.3.95), may be justified by reasoning as in the second
proof of Theorem 3.3.4 (in order to reduce matters to working with a null-solution
of an injectively elliptic first-order system) and then invoking Theorem 3.1.10.
Alternatively, we invoke item (g) in Theorem 3.3.4 with X := M p,λ (∂Ω, σ), a
permissible choice in light of [113, Proposition 6.2.17], and [113, Corollaries 6.2.11,
6.2.13]. Finally, (3.3.95) follows from (3.3.91), [112, (8.9.8)], and [113, (6.2.18)].
Let us also remark that a natural version of Theorem 3.3.1 holds in the context of
Morrey spaces, and such a result may be used to provide an alternative proof of
Theorem 3.3.7. 

Finally, here is a version of Theorem 3.3.4 for the scale of block spaces, which
states as follows.

Theorem 3.3.8 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be an arbitrary UR domain,

and abbreviate σ := H n−1 ∂Ω. Consider a homogeneous constant (complex) co-
efficient second-order M × M system L in Rn (for some M ∈ N) which is weakly
 1,1  M
elliptic (in the sense of (1.3.3)), and assume u ∈ Wloc (Ω) is a vector-valued
function which, for some aperture parameter κ > 0, satisfies

Nκ (∇u) ∈ B q,λ (∂Ω, σ) with q ∈ (1, ∞), λ ∈ (0, n − 1),

 M (3.3.96)
and Lu = 0 in D (Ω) .

Then the nontangential boundary trace

κ−n.t.
(∇u) ∂Ω exists (in Cn·M ) at σ-a.e. point on ∂Ω and
κ−n.t. (3.3.97)
the function (∇u) ∂Ω is σ-measurable on ∂Ω.

Moreover, for any other given aperture parameter κ  ∈ (0, ∞) the nontangential
κ  −n.t. κ−n.t.
boundary trace (∇u) ∂Ω also exists, and agrees with (∇u) ∂Ω , at σ-a.e. point on
∂Ω. In addition,
κ−n.t.   n·M
(∇u) ∂Ω belongs to the block space B q,λ (∂Ω, σ)
 κ−n.t.  (3.3.98)
and one has (∇u) ∂Ω [B q, λ (∂Ω,σ)]n·M ≤ Nκ (∇u) B q, λ (∂Ω,σ) .
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 711

In fact, the following block-styled boundary maximum principle holds:

   κ−n.t. 
Nκ (∇u) ≈ (∇u) ∂Ω [B q, λ (∂Ω,σ)]n·M (3.3.99)
B q, λ (∂Ω,σ)

provided, in the case when Ω is an exterior domain, it is also assumed that ∇u

vanishes at infinity (in the sense of (3.3.8)).

Proof One possible way to see this is to rely on item (g) in Theorem 3.3.4 with
X := B q,λ (∂Ω, σ), a valid choice thanks to [113, Proposition 6.2.17], and [113,
Corollaries 6.2.11, 6.2.13]. Alternatively, we may reason as follows. From (3.3.96)
and [113, (6.2.71)] we see that
q(n − 1)
Nκ (∇u) ∈ L r (∂Ω, σ) with r := ∈ (1, q). (3.3.100)
n − 1 + λ(q − 1)
Granted this, Theorem 3.3.4 applies and proves that (3.3.97) holds. The independence
of the boundary trace on the aperture parameter is then provided by [112, Propo-
sition 8.9.8]. Next, the claims in (3.3.98) are implied by [113, (6.2.75)] and [112,
(8.9.8)]. Finally, the right-pointing inequality in (3.3.99) is contained in (3.3.98),
while the left-pointing inequality in (3.3.99) is implied by (3.3.9) and (2.6.19). 

The remaining material in this section concerns quantitative Fatou-type results for
null-solutions u of a weakly elliptic second-order system L in arbitrary UR domains
Ω ⊆ Rn , describing scenarios in which integrability properties of the nontangen-
tial maximal functions of u and ∇u together ensure the existence of nontangential
pointwise traces of both u and ∇u on ∂Ω with quantitative control. Here is our first
theorem of this flavor.

Theorem 3.3.9 Let Ω ⊆ Rn , where n ∈ N with n ≥ 3, be an arbitrary UR domain,

 αβσ := H ∂Ω. For some integer M ∈ N, consider a coefficient
and abbreviate n−1

tensor A = ar s 1≤r,s ≤n with complex entries, with the property that the M × M
1≤α,β ≤M
homogeneous second-order system L = L A associated with A in Rn as in (1.3.15) is
weakly elliptic (in the sense of (1.3.3)). Finally, fix some aperture parameter κ > 0.
In this setting, consider a vector-valued function satisfying
 M
u = (uβ )1≤β ≤M ∈ 𝒞 ∞ (Ω) , Lu = 0 in Ω,
 
Nκ (∇u) ∈ L p (∂Ω, σ) for some p ∈ n−1
n , n−1 (3.3.101)
∗  
1 −1
and Nκ u ∈ L p (∂Ω, σ) with p∗ := p1 − n−1 .

Then
712 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
κ−n.t. κ−n.t.
the nontangential traces u ∂Ω , (∇u) ∂Ω exist σ-a.e. on ∂Ω,
κ−n.t.  ∗ M κ−n.t.   n·M
u ∂Ω ∈ L p (∂Ω, σ) and (∇u) ∂Ω ∈ L p (∂Ω, σ) ,
 κ−n.t.  (3.3.102)
u  p∗ ≤ Nκ u L p ∗ (∂Ω,σ) and
∂Ω [L (∂Ω,σ)] M
 κ−n.t. 
(∇u)  p ≤ Nκ (∇u) L p (∂Ω,σ) .
∂Ω [L (∂Ω,σ)] n·M

In addition, for any other given aperture parameter κ  ∈ (0, ∞) the nontangential
κ  −n.t. κ  −n.t. κ−n.t. κ−n.t.
boundary traces u ∂Ω , (∇u) ∂Ω also exist and agree with u ∂Ω and (∇u) ∂Ω ,
respectively, at σ-a.e. point on ∂Ω. Also, for some C = C(Ω, κ, p) ∈ (0, ∞), one
has (recall the scale of Hardy-based inhomogeneous Sobolev spaces from [113,
Definition 11.10.6]) that
κ−n.t.  p∗, p M
u ∂Ω belongs to the Hardy-based Sobolev space H1 (∂Ω, σ) ,
 κ−n.t. 
and u ∂Ω [H p ∗, p (∂Ω,σ)] M ≤ CNκ u L p ∗ (∂Ω,σ) + CNκ (∇u) L p (∂Ω,σ),
1
(3.3.103)
while the weak conormal derivative of u associated with the coefficient tensor A as
in [113, Definition 10.2.18] satisfies
.  M
∂νAu ∈ H p (∂Ω, σ) and
 .A    (3.3.104)
 ∂ u p ≤ C Nκ (∇u) L p (∂Ω,σ)
ν [H (∂Ω,σ)] M

for some constant C ∈ (0, ∞) which depends only on Ω, n, A, κ, p.

Moreover, with the single and double layer potential operators 𝒮, D, associated
with A and Ω as in [115, §1.3], one has the integral representation formula
 κ−n.t.  .
u = D u ∂Ω − 𝒮(∂νAu) in Ω, (3.3.105)

with the understanding that if Ω is an exterior domain one also assumes that there
exists λ ∈ (1, ∞) such that
⨏
|u| dL n = o(1) as R → ∞. (3.3.106)
B(0,λ R)\B(0,R)

Furthermore,
if actually ∂Ω is bounded and in place of (3.3.101) one now assumes
 M
that u ∈ 𝒞 ∞ (Ω) with Lu = 0 in Ω, Nκ (∇u) ∈ L p (∂Ω, σ) for
 n−1    (3.3.107)
some p ∈ n , ∞ , and Nκ u ∈ L q (∂Ω, σ) for some q ∈ n−1 n , ∞ , it
follows that the conclusions in (3.3.102) hold with q replacing p∗ .
Finally, the same results are valid in the case n = 2 under the additional assump-
tion that Ω is bounded.
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 713

Before presenting the proof of this result we make a couple of comments. First,
if p ≥ 1 then item (3) in [113, Theorem 10.2.24] guarantees that the weak conormal
.  M
derivative ∂νAu actually belongs to the Lebesgue space L p (∂Ω, σ) and has the
pointwise formula
.  κ−n.t. 
αβ
∂νAu = νr ar s (∂s uβ ) ∂Ω at σ-a.e. point on ∂Ω. (3.3.108)
1≤α ≤M
 n−1
Second, it is only the range q ∈ n , 1 that is genuinely new in (3.3.107). Indeed,
if one assumes q ∈ (1, ∞] in (3.3.107) then it is possible to choose
 
r ∈ n−1
n , min{p, n − 1} such that q + n−1 < r < n−1 .
1 1 1 n
(3.3.109)
 
1 −1
As a consequence, if we define r ∗ := r1 − n−1 then r ∗ < q. Given that the
Lebesgue scale on ∂Ω is nested (since presently we have σ(∂Ω) < +∞), this choice
implies Nκ (∇u) ∈ L r (∂Ω, σ) and r∗
 n−1 Nκ u ∈  L (∂Ω, σ). Hence, the hypotheses in
(3.3.101) are satisfied with r ∈ n , n − 1 replacing the integrability exponent p.
κ−n.t. κ−n.t.
Granted this, (3.3.102) ensures that the nontangential traces u ∂Ω , (∇u) ∂Ω exist at
σ-a.e. on point on ∂Ω, and the corresponding functions are σ-measurable. Having
established this, the conclusions in (3.3.107) become consequences of [112, (8.9.8)]
and assumptions.
Let us now turn to the task of giving the proof of Theorem 3.3.9.
Proof of Theorem 3.3.9 The claims in (3.3.104) are implied by item (1) in [113,
κ−n.t.
Theorem 10.2.24] and the first two lines in (3.3.101). For the existence of u ∂Ω we
may invoke Proposition 3.1.13. An alternative approach which also yields the integral
representation formula (3.3.105) goes as follows. To set the stage, fix β ∈ {1, . . . , M }
together with j ∈ {1, . . . , n}. From (3.3.101) and [112, Proposition 8.6.3] we con-
clude that the vector field uβ e j belongs to the space of “admissible” vector fields,
  1 n 
𝒜 := F ∈ Lbdd (Ω, L n ) : div F ∈ Lbdd
1
(Ω, L n ) . (3.3.110)

As such, [112, Proposition   4.2.3] guarantees that ν • (uβ e j ) makes sense as a

distribution, in Lipc (∂Ω) . In relation to this, we next claim that for each two
indices, β ∈ {1, . . . , M } and j ∈ {1, . . . , n}, we have
∗
the distribution ν • (uβ e j ) belongs to the space L p (∂Ω, σ) and
  (3.3.111)
ν • (uβ e j ) p ∗ ≤ CNκ u L p ∗ (∂Ω,σ) + CNκ (∇u) L p (∂Ω,σ)
L (∂Ω,σ)

for some constant C ∈ (0, ∞) depending only on Ω, n, κ, p. Indeed, the vector field
uβ e j satisfies Nκ (uβ e j ) ≤ Nκ u which, in view of [112, Proposition 8.2.3], implies
∗
Nκ (uβ e j ) ∈ L p (∂Ω, σ) and
    (3.3.112)
Nκ (uβ e j ) p ∗ ≤ Nκ u L p ∗ (∂Ω,σ) .
L (∂Ω,σ)
714 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

Also, div (uβ e j ) = ∂j uβ , so on the one hand we have

 
P div (uβ e j ) ≤ P(∇u). (3.3.113)

On the other hand, from the second line in (3.3.101), [113, (10.1.8)-(10.1.9)] (used
with ∇u in place of u), and [112, (6.2.26)] we know that
∗ ∗
P(∇u) ∈ L p , p (∂Ω, σ) → L p (∂Ω, σ) and
    (3.3.114)
P(∇u) p ∗ ≤ C Nκ (∇u) L p (∂Ω,σ) .
L (∂Ω,σ)

Combining (3.3.113), (3.3.114), and [113, (10.1.3)] proves that

  ∗
P div (uβ e j ) ∈ L p (∂Ω, σ) and
     (3.3.115)
P div (uβ e j )  p ∗ ≤ C Nκ (∇u) L p (∂Ω,σ) .
L (∂Ω,σ)

From (3.3.112), (3.3.115), and [113, Theorem 10.2.1] we conclude that the claims
in (3.3.111) are true.
Moving on, let us temporarily work under the additional assumption that if Ω is
an exterior domain then there exists λ ∈ (1, ∞) such that (3.3.106) holds. Denote
by E = (Eγ β )1≤γ ,β ≤M the matrix-valued fundamental solution associated with L
as in Theorem 1.4.2. Fix an arbitrary index γ ∈ {1, . . . , M }, pick an arbitrary point
x ∈ Ω, and consider the vector field Fx = Fs 1≤s ≤n with components (throughout,
the summation convention over repeated indices is assumed)
βα αβ
Fs := −ar s (∂r Eγβ )(x − ·)uα − Eγα (x − ·)asr (∂r uβ )
(3.3.116)
at L n -a.e. point in Ω.

The goal is to apply [112, Theorem 1.9.4] to the vector field Fx . Checking that
this vector field satisfies the hypotheses of [112, Theorem 1.9.4] takes some work.
For starters, observe that (3.3.116) and Theorem 1.4.2 imply
 n  n
Fx ∈ L 1bdd (Ω, L n ) ⊆ L 1bdd (Ω, L n ) + ℰ(Ω) . (3.3.117)

Moreover, from (3.3.116) and (1.4.21) we see that the divergence of Fx , computed
in the sense of distributions in Ω, is given by
βα
div Fx = ∂s Fs = ar s ∂s ∂r [Eγβ (· − x)]uα
βα αβ
− ar s (∂r Eγβ )(x − ·)(∂s uα ) + (∂s Eγα )(x − ·)asr (∂r uβ )
αβ
− Eγα (x − ·)asr (∂s ∂r uβ )

= uα δγα δx = uγ (x)δx . (3.3.118)

Hence,
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 715

div Fx ∈ ℰ(Ω) ⊆ L 1 (Ω, L n ) + ℰ(Ω). (3.3.119)

Note that (3.3.117) and (3.3.119) are in agreement with [112, (1.9.29)].
Pressing on, with λ as in (3.3.106)
 if Ω is an exterior domain and λ := 2 otherwise,
pick a function φ ∈ 𝒞c∞ B(0, λ) with the property that φ ≡ 1 on B(0, 1). Then the
family

ℱ := {φ R }R>0, where φ R := φ(·/R) for each R ∈ (0, ∞), (3.3.120)

becomes a system of auxiliary functions (in the sense of [112, (1.3.3)]). Recall from
[112, (1.9.30)] that
∫
[Fx ]ℱ = − lim R−1 (∇φ)(y/R) · Fx (y) dL n (y) (3.3.121)
R→∞ Ω

Under the assumption that Ω is bounded, or ∂Ω is unbounded, we may rely on

[112, Proposition 8.6.3] (while also mindful of (3.3.101)), and Hölder’s inequality
to estimate, for R large,
∫
−1
R |(∇φ)(y/R)||(∇E)(x − y)||u(y)| dL n (y)
Ω
∫
−n
≤ CR |u| dL n
Ω∩[B(0,λR)\B(0,R)]

∫ n p∗
 n−1∗   1− n−1∗
−n np
≤ CR |u| n−1 dL n Rn np

− n−1
≤ CR p∗ Nκ u L p ∗ (∂Ω,σ)

= o(1) as R → ∞, (3.3.122)

and
∫
−1
R |(∇φ)(y/R)||E(x − y)||(∇u)(y)| dL n (y)
Ω
∫
−(n−1)
≤ CR |∇u| dL n
Ω∩[B(0,λR)\B(0,R)]

∫ np
 n−1
np   1− n−1
≤ CR−(n−1) |∇u| n−1 dL n Rn np

1− n−1
≤ CR p Nκ (∇u) L p (∂Ω,σ)

= o(1) as R → ∞, (3.3.123)
716 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

since 1 − n−1
p < 0 given that p < n − 1. Collectively, (3.3.121), (3.3.116), (3.3.122),
(3.3.123) prove that

if either Ω is bounded, or ∂Ω is unbounded, then [Fx ]ℱ = 0. (3.3.124)

In the case when Ω is an exterior domain, Lemma 1.5.6 and (3.3.106) imply
 
(∇u)(x) = o |x| −1 as |x| → ∞. (3.3.125)

Hence,
∫ ⨏
# $
R −1 
|(∇φ)(y/R)|| F(y)| dL n (y) ≤ |u| + R|∇u| dL n
Ω B(0,λ R)\B(0,R)

= o(1) as R → ∞, (3.3.126)

by (3.3.106) and (3.3.125). Thus,

if Ω is an exterior domain then [Fx ]ℱ = 0. (3.3.127)

Together, (3.3.124) and (3.3.127) show that in all cases the limit in [112, (1.9.30)]
exists, and is actually zero.
To proceed, choose a function η ∈ 𝒞c∞ (Ω) satisfying η ≡ 1 near x. Granted
(3.3.118), (3.3.119), [112, Definition 4.2.6] permits us to define ν • Fx as
   
ν • Fx = ν • (1 − η)Fx in Lipc (∂Ω) (3.3.128)
 
where ν • (1 − η)Fx is now interpreted in the sense of [112, Proposition 4.2.3]
(bearing in mind that (1 − η)Fx belongs to 𝒜, the space of vector fields defined in
(3.3.110)). Hence, with the piece of notation introduced in [112, (1.9.31)], we have
     
ν • Fx , 1 ℱ = ν • (1 − η)Fx , 1
ℱ
 
= lim (Lip c (∂Ω)) ν • (1 − η)Fx , φ R ∂Ω Lip c (∂Ω)
R→∞

 βα 
= − lim (Lip c (∂Ω)) ν • (1 − η)ar s (∂r Eγβ )(x − ·)uα es , φ R ∂Ω Lip c (∂Ω)
R→∞

 αβ 
− lim (Lip c (∂Ω)) ν • (1 − η)Eγα (x − ·)asr (∂r uβ )es , φ R ∂Ω Lip c (∂Ω)
R→∞

=: I + II. (3.3.129)

We may further compute

3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 717
   
βα
I = − lim (Lip c (∂Ω)) ar s (1 − η)(∂r Eγβ )(x − ·) ν • uα e s , φ R ∂Ω Lip c (∂Ω)
R→∞ ∂Ω

βα  
= − lim (Lip c (∂Ω)) ar s (∂r Eγβ )(x − ·) ∂Ω ν • uα es , φ R ∂Ω Lip c (∂Ω)
R→∞

  βα  
= − lim (Lip c (∂Ω)) ν • uα es , ar s (∂r Eγβ )(x − ·)φ R ∂Ω Lip c (∂Ω)
R→∞
∫
  βα
= − lim ν • uα es ar s (∂r Eγβ )(x − ·)φ R dσ
R→∞ ∂Ω
∫
  βα
=− ν • uα es ar s (∂r Eγβ )(x − ·) dσ. (3.3.130)
∂Ω

Above, the first equality comes from [112, (4.2.14)], whose applicability in the
present setting is ensured by the fact that uβ e j belongs to the space 𝒜 of “admissi-
ble” vector fields, defined in (3.3.110). The second equality is simply the observation
that 1−η = 1 on ∂Ω, while the third equality is implied by [112, (4.1.43)]. The fourth
equality is a consequence of (3.3.111) and [112, Proposition 4.1.4]. Finally, the fifth
equality is justified by relying on Lebesgue’s Dominated Convergence Theorem
(keeping in mind the estimate for ∇E from Theorem 1.4.2, and [112, Lemma 7.2.1]).
To treat the last term in (3.3.129), let us first consider the case when ∂Ω is
bounded. We then have
   αβ 
II = − lim (Lipc (∂Ω)) (1 − η)Eγα (x − ·) ∂Ω ν • asr (∂r uβ )es , φ R ∂Ω Lipc (∂Ω)
R→∞

   αβ 
= − lim (Lip c (∂Ω)) Eγα (x − ·) ∂Ω
ν • asr (∂r uβ )es , φ R ∂Ω Lip c (∂Ω)
R→∞

 αβ   
= − lim (Lip c (∂Ω)) ν • asr (∂r uβ )es , Eγα (x − ·)φ R ∂Ω Lip c (∂Ω)
R→∞

 .A   
= − lim (Lip c (∂Ω)) ∂ν u α , Eγα (x − ·)φ R ∂Ω Lip c (∂Ω)
R→∞

 .A    
= − lim H p (∂Ω,σ) ∂ν u α , Eγα (x − ·)φ R ∂Ω (H p (∂Ω,σ))∗ . (3.3.131)
R→∞

Here, the first equality uses [112, (4.2.14)], whose present applicability is guaranteed
αβ
by the fact that, for each fixed α ∈ {1, . . . , M }, the vector field asr (∂r uβ )es belongs
to the space 𝒜 from (3.3.110). Indeed, the first two lines in (3.3.101) together with
[112, Proposition 8.6.3] ensure that this is a divergence-free vector field whose
components are absolutely integrable on bounded measurable subsets of Ω. The
second equality relies on the observation that 1 − η = 1 on ∂Ω, and the third
equality is a consequence of [112, (4.1.43)]. The fourth equality is provided by [113,
Definition 10.2.18], while the fifth equality follows on account of (3.3.104) and [113,
Lemma 4.6.4]. Going further, we also claim that for each α ∈ {1, . . . , M } we have
718 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
    
lim Eγα (x − ·)φ R ∂Ω
= Eγα (x − ·) ∂Ω weak-∗ in
R→∞

⎧ . 
⎪
⎪ 𝒞(n−1)(1/p−1) (∂Ω) ∼ if n−1
n < p < 1,
⎪
 p ∗ ⎪
⎨
⎪

H (∂Ω, σ) = BMO(∂Ω, σ) if p = 1,
⎪
⎪
⎪
⎪
⎪ L p (∂Ω, σ)

if p > 1 and p := (1 − 1/p)−1 .
⎩
(3.3.132)

This claim is implied by the general weak-∗ convergence results established in [113,
Lemma 4.8.4] if p < 1, [113, Lemma 4.8.1] (in the latter case also bearing in mind the
trivial bounded embedding L ∞ (∂Ω, σ) → BMO(∂Ω, σ)) if p = 1, and Lebesgue’s
Dominated Convergence Theorem (plus Theorem 1.4.2) if 1 < p < n − 1. In turn,
from (3.3.131), (3.3.132), the manner in which the boundary-to-domain single layer
n < p ≤ 1, plus the definition of the
acts from Hardy spaces (cf. [115, §2.2]) if n−1
boundary-to-domain single layer (cf. [115, §1.3]), and (2.5.549) (with α := 1) if
1 < p < n − 1, we conclude that
 .    
II = − H p (∂Ω,σ) ∂νAu α , Eγα (x − ·) ∂Ω (H p (∂Ω,σ))∗
 . 
= − 𝒮(∂νAu)(x) . (3.3.133)
γ

The same type of argument proves (3.3.133) in the case when ∂Ω is bounded (a
scenario in which the dual of H p and the corresponding duality bracket no longer
involve classes of functions modulo constants). Thus, in all cases, from (3.3.129),
(3.3.130), and (3.3.133) we obtain
∫  . 
    βα
ν • Fx , 1 ℱ = − ν • uα es ar s (∂r Eγβ )(x − ·) dσ − 𝒮(∂νAu)(x) .
∂Ω γ
(3.3.134)

Granted this, [112, Theorem 1.9.4] applies and, on account of (3.3.118), (3.3.124),
(3.3.127), the Divergence Formula [112, (1.9.32)] written for Fx presently gives
∫  . 
  βα
uγ (x) = − ν • uα es ar s (∂r Eγβ )(x − ·) dσ − 𝒮(∂νAu)(x) . (3.3.135)
∂Ω γ

In turn, from (3.3.135), (3.3.111), Theorem 2.5.1, the trace formula (lack of jump)
for the single layer (cf. [115, §2.2]), and (3.3.104), we conclude that
κ−n.t.
the nontangential trace u ∂Ω exists at σ-a.e. point on ∂Ω if either Ω is
bounded, or ∂Ω is unbounded, or Ω is an exterior domain and the decay (3.3.136)
condition (3.3.106) holds.
In fact, this Fatou-type property further self-improves. Specifically, in the case when
Ω is an exterior domain, working with ΩR := B(0, R) ∩ Ω in place of Ω (where
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 719

R ∈ (0, ∞) is a sufficiently large number), much as in the proof of Theorem 3.1.6,

eliminates the need of asking that the decay condition (3.3.106) holds. Simply put,
at this stage we have shown that, in all cases,
κ−n.t.
the nontangential trace u ∂Ω exists σ-a.e. on ∂Ω. (3.3.137)

With this in hand, we may now apply [113, Proposition 10.2.9] to each vector field
uα es to conclude that if ν = (ν1, . . . , νn ) is the geometric measure theoretic outward
unit normal to Ω then
   κ−n.t. 
ν • uα es = νs uα ∂Ω for each s ∈ {1, . . . , n} and α ∈ {1, . . . , M }. (3.3.138)

Returning with (3.3.138) in (3.3.135) ultimately produces (3.3.105). Having proved

(3.3.137), we may rely on [113, Definition 11.10.6] and [113, Corollary 11.10.11] to
conclude that the claims in (3.3.103) hold as well. The claims about the existence and
κ−n.t.
integrability properties of the nontangential trace (∇u) ∂Ω made in (3.3.102) are
κ−n.t.
consequences of Theorem 3.3.4. The independence of the nontangential traces u ∂Ω ,
κ−n.t.
(∇u) ∂Ω on the aperture parameter is a consequence of [112, Propositions 8.8.6,
8.9.8].
The next goal is to justify the claim made in (3.3.107). The fact that ∂Ω is
currently assumed to be compact makes the Lebesgue scale on ∂Ω nested. Given
the aims
 we have in mind, there is no loss of generality in assuming that actually
p ∈ n−1 , n − 1 . Define p ∗ as before. The idea is now to run the same proof as
n
above with a few key adjustments. First, in place of (3.3.111) we now get that, for
each β ∈ {1, . . . , M } and j ∈ {1, . . . , n},

the distribution ν • (uβ e j ) belongs

∗
(3.3.139)
to the Hardy space H min{q, p } (∂Ω, σ),

thanks to [113, Theorem 10.2.4], (3.3.114), and the definition of the scale of Hardy
spaces given in [113, Definition 4.2.1]. Second, the same conclusion in (3.3.122)
holds (with an identical proof, now using q in place of p∗ ). Third, in place of
(3.3.130) we now have
  βα  
I = − lim (Lip c (∂Ω)) ν • uα es , ar s (∂r Eγβ )(x − ·)φ R ∂Ω Lip c (∂Ω)
R→∞

  βα
= − H min{q, p ∗ } (∂Ω,σ) ν • uα es , ar s (∂r Eγβ )(x − ·) ∂Ω (H min{q, p } (∂Ω,σ))∗ .
∗

(3.3.140)

The last equality in (3.3.140) is a consequence of the membership in (3.3.139) and

the fact that for each indexes β ∈ {1, . . . , M } and r ∈ {1, . . . , n}, we have (reasoning
as in (3.3.16), and also employing Lebesgue’s Dominated Convergence Theorem
together with Theorem 1.4.2)
720 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
 
(∂r Eγβ )(x − ·) ∂Ω = lim (∂r Eγβ )(x − ·)φ R ∂Ω
R→∞
 ∗ (3.3.141)
∗
weak-∗ in H min{q, p } (∂Ω, σ) .

Hence, in place of (3.3.135) we now obtain

  βα
uγ (x) = − H min{q, p ∗ } (∂Ω,σ) ν • uα es , ar s (∂r Eγβ )(x − ·) ∂Ω ∗
(H min{q, p } (∂Ω,σ))∗
 . 
− 𝒮(∂νAu)(x) . (3.3.142)
γ

Granted this representation formula, the conclusion in (3.3.136) follows on account of

(3.3.139), Corollary 2.5.4 (if q ≤ 1) and Theorem 2.5.1 (if q > 1), the trace formula
(lack of jump) for the single layer (cf. [115, §2.2]), and (3.3.104). Everything else is
as before.
Finally, a cursory inspection reveals that the proof given above also works in the
case n = 2 under the additional assumption that Ω is bounded. 
In the corollary below we present a Fatou-type result for a null-solution u of a
weakly elliptic system L in which no a priori assumptions are made on the nontan-
gential maximal function of u. Compared with Theorem 3.3.9, we shall strengthen
the geometric assumptions by demanding that the given UR domain is bounded and
also happens to be a one-sided NTA domain. A related, more nuanced, result is
proved later, in Theorem 3.4.4, in the class of bounded two-sided NTA domains with
an Ahlfors regular boundary.
Corollary 3.3.10 Let Ω ⊆ Rn , where n ∈ N with n ≥ 2, be a UR domain with com-
pact boundary which is also a one-sided NTA domain (for example, any NTA domain
with a compact Ahlfors regular boundary will do). Abbreviate σ := H n−1 ∂Ω and
fix some aperture parameter κ > 0. Also, for some M ∈ N, consider a weakly elliptic,
homogeneous, constant complex coefficient, second-order M × M system L in Rn .
Finally, fix a truncation parameter ε > 0 and assume u is a vector-valued function
satisfying
 M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω, and
  (3.3.143)
Nκε (∇u) ∈ L p (∂Ω, σ) for some p ∈ n−1
n , ∞ .
κ−n.t.
Then Nκε u also belongs to L p (∂Ω, σ) and the nontangential traces u ∂Ω and
κ−n.t.
(∇u) ∂Ω exist at σ-a.e. point on ∂Ω.
Proof Suppose first that Ω is bounded. Since u and ∇u are locally bounded, from
assumptions and [112, (8.4.109)] we conclude that Nκ u and Nκ (∇u) belong to
L p (∂Ω, σ). This permits us to invoke (3.3.107) in Theorem 3.3.9 (with q := p) to
κ−n.t. κ−n.t.
conclude that the nontangential traces u ∂Ω and (∇u) ∂Ω exist at σ-a.e. point on
∂Ω. Finally, in the situation when Ω is unbounded, apply what we have just proved
to the bounded set Ω ∩ B(0, R) with R > 2 max{|x| : x ∈ ∂Ω}. 
3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 721

We close this section by discussing a version of Theorem 3.3.9 adapted to the

scale of Morrey spaces.

Theorem 3.3.11 Suppose Ω ⊆ Rn , where n ∈ N with n ≥ 3, is an arbitrary UR

domain, and abbreviate σ := H n−1 ∂Ω. For some M ∈ N, consider an M × M
weakly elliptic, homogeneous, second-order system L = L A in Rn , associated with
some coefficient tensor A as in (1.3.15). Also, fix some aperture parameter κ > 0
and consider a vector-valued function u = (uβ )1≤β ≤M satisfying
 M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω, Nκ (∇u) ∈ M p,λ (∂Ω, σ)
 2  
for some λ ∈ 0, (n−1)
n n ,n−1−λ ,
and p ∈ n−1 (3.3.144)
  −1
and Nκ u ∈ M p,λ (∂Ω, σ) with p := p1 − n−1−λ
1
.

Then
κ−n.t. κ−n.t.
the nontangential traces u ∂Ω , (∇u) ∂Ω exist σ-a.e. on ∂Ω,
κ−n.t.  M κ−n.t.   n·M
u ∂Ω ∈ M p,λ (∂Ω, σ) and (∇u) ∂Ω ∈ M p,λ (∂Ω, σ) ,
 κ−n.t.  (3.3.145)
u  p ,λ ≤ Nκ u M p, λ (∂Ω,σ) and
∂Ω [M  (∂Ω,σ)] M
 κ−n.t. 
(∇u)  p, λ ≤ Nκ (∇u) M p, λ (∂Ω,σ) .
∂Ω [M (∂Ω,σ)] n·M

Moreover, for any other aperture parameter κ  ∈ (0, ∞) the nontangential boundary
κ  −n.t. κ  −n.t. κ−n.t. κ−n.t.
traces u ∂Ω , (∇u) ∂Ω also exist and agree with u ∂Ω and (∇u) ∂Ω , respectively,
at σ-a.e. point on ∂Ω. Also, for some C = C(Ω, κ, p, λ) ∈ (0, ∞), it follows that
κ−n.t.  p , p,λ M
u ∂Ω lies in the off-diagonal Morrey-based Sobolev space M1  (∂Ω, σ)
 κ−n.t. 
and u ∂Ω [M p, p, λ (∂Ω,σ)] M ≤ CNκ u M p, λ (∂Ω,σ) + CNκ (∇u) M p, λ (∂Ω,σ)
1
(3.3.146)
(cf. [113, (11.7.14)-(11.7.15)]), while the weak conormal derivative of u associated
with the coefficient tensor A as in [113, Definition 10.2.18] satisfies
.  M
∂νAu belongs to M p,λ (∂Ω, σ) and
 .A    (3.3.147)
∂ u p, λ ≤ C Nκ (∇u) M p, λ (∂Ω,σ),
ν [M (∂Ω,σ)] M

for some constant C ∈ (0, ∞) which depends only on Ω, n, A, κ, p, λ.

Finally, with the single and double layer potential operators 𝒮, D, associated
with A and Ω as in [115, §1.3], one has the integral representation formula
 κ−n.t.  .
u = D u ∂Ω − 𝒮(∂νAu) in Ω, (3.3.148)
722 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

with the agreement that if Ω is an exterior domain one also assumes that also assumes
that there exists C ∈ (1, ∞) such that
⨏
|u| dL n = o(1) as R → ∞. (3.3.149)
B(0,C R)\B(0,R)

Proof We largely proceed as in the proof of Theorem 3.3.9, so we will only in-
dicate the main changes needed to carry out the same game-plan. First, Theo-
κ−n.t.
rem 3.3.7 applies and gives that the nontangential boundary trace (∇u) ∂Ω ex-
κ−n.t.
ists σ-a.e. on ∂Ω, and the function (∇u) ∂Ω is independent of the aperture pa-
  n·M
rameter, belongs to the Morrey space M p,λ (∂Ω, σ) , and satisfy the esti-
 κ−n.t. 
 
mate (∇u) ∂Ω [M p, λ (∂Ω,σ)]n·M ≤ Nκ (∇u) M p, λ (∂Ω,σ) . Second, the claims in
(3.3.147) are consequences of item (5) in [113, Theorem 10.2.24] and the first two
lines in (3.3.144). For further reference, from (3.3.147) and [112, (6.2.23)] it is useful
to note that
.  σ(y) 
∂νAu ∈ M p,λ (∂Ω, σ) → L 1 ∂Ω, . (3.3.150)
1 + |y| n−2

To proceed, fix β ∈ {1, . . . , M } and

  j ∈ {1, . . . , n}. As before, ν • (uβ e j ) makes
sense as a distribution in Lipc (∂Ω) and we claim that

ν • (uβ e j ) belongs to the Morrey space M p,λ (∂Ω, σ) and satisfies

 
ν • (uβ e j ) p , λ ≤ CNκ u M p, λ (∂Ω,σ) + CNκ (∇u) M p, λ (∂Ω,σ)
M  (∂Ω,σ)
(3.3.151)

for some constant C ∈ (0, ∞) depending only on Ω, n, κ, p, λ. To justify this, observe

that in a pointwise sense we have Nκ (uβ e j ) ≤ Nκ u which, in view of (3.3.144),
[113, (6.2.3)], and [112, Proposition 8.2.3], implies

Nκ (uβ e j ) ∈ M p,λ (∂Ω, σ) and

    (3.3.152)
Nκ (uβ e j ) p , λ ≤ Nκ u M p, λ (∂Ω,σ) .
M  (∂Ω,σ)

Next, observe that div(uβ e j ) = ∂j uβ hence

 
P div(uβ e j ) ≤ P(∇u). (3.3.153)

In addition, from the last property in the top line of (3.3.144) and [113, (10.1.11)-
(10.1.12)] (used with ∇u in place of u) we see that

P(∇u) belongs to M p,λ (∂Ω, σ) and

    (3.3.154)
P(∇u) p , λ ≤ C Nκ (∇u) M p, λ (∂Ω,σ) .
M  (∂Ω,σ)

Collectively, (3.3.153), (3.3.154), and [113, (6.2.3), (10.1.3)] establish that

3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 723
 
P div(uβ e j ) belongs to M p,λ (∂Ω, σ) and
     (3.3.155)
P div(uβ e j )  p , λ ≤ C Nκ (∇u) p, λ .
M  (∂Ω,σ) M (∂Ω,σ)

At this stage, from (3.3.152), (3.3.155), and [113, Corollary 10.2.7] we conclude
that the claims in (3.3.151) hold.
Pressing on, fix an arbitrary index γ ∈ {1, . . . , M }, pick some arbitrary point
x ∈ Ω, and define the vector field Fx = Fs 1≤s ≤n with components as in (3.3.116).
This continues to satisfy (3.3.117)-(3.3.119). With the family ℱ := {φ R }R>0 as in
(3.3.120) and with [Fx ]ℱ defined as in (3.3.121), we claim that

[Fx ]ℱ = 0. (3.3.156)

To see that this is the case, assume first that either Ω is bounded, or ∂Ω is unbounded.
Also, without any loss of generality assume that 0 ∈ ∂Ω. In such a scenario, we
may rely on [112, Proposition 8.6.3], (3.3.144), Hölder’s inequality, [112, (8.2.26)],
[112, (8.1.17)], and (A.0.87) to estimate (for R ∈ (0, ∞) large and with C ∈ (0, ∞)
presently denoting what used to be λ in the proof of Theorem 3.3.9)
∫
−1
R |(∇φ)(y/R)||(∇E)(x − y)||u(y)| dL n (y)
Ω
∫
≤ CR−n |u| dL n
Ω∩[B(0,C R)\B(0,R)]

∫ n p
 nn−1
p   1− nn−1
−n
≤ CR |u| n−1 dL n Rn p

Ω∩B(0,C R)

n−1
∫  1/p
≤ CR− p |Nκ u| p dσ
∂Ω∩B(0,C(2+κ)R)
⨏  1/p
≤C |Nκ u| p dσ
∂Ω∩B(0,C(2+κ)R)

n−1−λ  
≤ CR− p Nκ u p , λ
M  (∂Ω,σ)

= o(1) as R → ∞, (3.3.157)

p < 0 given that λ < n − 1, and

since − n−1−λ
724 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
∫
R−1 |(∇φ)(y/R)||E(x − y)||(∇u)(y)| dL n (y)
Ω
∫
≤ CR−(n−1) |∇u| dL n
Ω∩[B(0,C R)\B(0,R)]

∫ np
 n−1
np   1− n−1
≤ CR−(n−1) |∇u| n−1 dL n Rn np

Ω∩B(0,C R)

n−1
∫  1/p
≤ CR1− p |Nκ (∇u)| p dσ
∂Ω∩B(0,C(2+κ)R)
⨏  1/p
≤ CR |Nκ (∇u)| p dσ
∂Ω∩B(0,C(2+κ)R)

1− n−1−λ
 
≤ CR p Nκ (∇u)
M p, λ (∂Ω,σ)

= o(1) as R → ∞, (3.3.158)

since 1 − n−1−λ
p < 0 given that p < n − 1 − λ. Together, (3.3.157)-(3.3.158) estab-
lish (3.3.156) when either Ω is bounded, or ∂Ω is unbounded. The justification of
(3.3.156) in the case when Ω is an exterior domain is similar to (3.3.125)-(3.3.126),
based on interior estimates and (3.3.149). This finishes the proof of (3.3.156).
We then follow the computations in (3.3.128)-(3.3.131) almost verbatim, the only
two significant differences being as follows. First, the last equality in (3.3.130) now
uses the fact that, as seen from (3.3.151), [112, (6.2.23)] and the current assumptions
on p, λ, p,
 σ(y) 
ν • (uβ e j ) ∈ M p,λ (∂Ω, σ) → L 1 ∂Ω, , (3.3.159)
1 + |y| n−1

that there exists C = Cx ∈ (0, ∞) such that (cf. Theorem 1.4.2)

C
|(∇E)(x − y)| ≤ for each y ∈ ∂Ω, (3.3.160)
1 + |y| n−1
and Lebesgue’s Dominated Convergence Theorem. Hence, we once again have
∫
  βα
I=− ν • uα es ar s (∂r Eγβ )(x − ·) dσ. (3.3.161)
∂Ω

Second, in place of the last equality in (3.3.131) we now have

3.3 Quantitative Fatou-Type Theorems in UR Domains for Second-Order Systems 725
 .A   
II = − lim (Lip c (∂Ω)) ∂ν u α, Eγα (x − ·)φ R ∂Ω Lip c (∂Ω)
R→∞
∫
 .A 
= − lim ∂ν u α, Eγα (x − ·)φ R dσ
R→∞ ∂Ω
∫
 .A  .
=− ∂ν u α, Eγα (x − ·) dσ = 𝒮(∂νAu)(x), (3.3.162)
∂Ω

thanks to (3.3.150), [112, (4.1.47)], the fact that there exists C = Cx ∈ (0, ∞) such
that (cf. Theorem 1.4.2)
C
|E(x − y)| ≤ for each y ∈ ∂Ω, (3.3.163)
1 + |y| n−2
Lebesgue’s Dominated Convergence Theorem, and the definition of the boundary-
to-domain single layer (cf. [115, §1.3]).
Granted (3.3.161)-(3.3.162), we may rely on [112, Theorem 1.9.4] to conclude,
as in (3.3.135), that
∫  . 
  βα
uγ (x) = − ν • uα es ar s (∂r Eγβ )(x − ·) dσ − 𝒮(∂νAu)(x) . (3.3.164)
∂Ω γ

In turn, from (3.3.164), (3.3.159), Theorem 2.5.1, (3.3.150), and the existence of the
nontangential boundary trace for the single layer (cf. [115, §2.2]) we conclude that
κ−n.t.
the nontangential trace u ∂Ω exists at σ-a.e. point on ∂Ω if either Ω is bounded,
or ∂Ω is unbounded, or Ω is an exterior domain and the decay condition (3.3.149)
holds. With this in hand, the same argument that has led to (3.3.137) once again
shows that, in all cases,
κ−n.t.
the nontangential trace u ∂Ω exists σ-a.e. on ∂Ω. (3.3.165)

In view of this property, we may invoke [113, Proposition 10.2.9] to conclude that if
ν = (ν1, . . . , νn ) is the geometric measure theoretic outward unit normal to Ω then
   κ−n.t. 
ν • uα es = νs uα ∂Ω
for each
(3.3.166)
s ∈ {1, . . . , n} and α ∈ {1, . . . , M }.

With (3.3.166) in hand, the integral representation formula (3.3.164) reduces pre-
cisely to (3.3.148). Let us also note that, on account of (3.3.165), the last line in
(3.3.144), [113, (6.2.3)], and [112, Corollary 8.9.6], we have
κ−n.t.  M
u ∂Ω ∈ M p,λ (∂Ω, σ) and
 κ−n.t.  (3.3.167)
u  p ,λ M ≤ Nκ u M p , λ (∂Ω,σ) .
∂Ω [M  (∂Ω,σ)]
726 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

Let us now fix j, k ∈ {1, . . . , n}, arbitrary. On the one hand, from the first part of the
current proof we know that the functions
κ−n.t. κ−n.t.  M
(∂j u) ∂Ω , (∂k u) ∂Ω belong to M p,λ (∂Ω, σ)
(3.3.168)
and each has norm ≤ Nκ (∇u) M p, λ (∂Ω,σ) .

On the other hand, from (3.3.144), the second inclusion in [113, (6.2.7)], and [113,
κ−n.t.  p, p M
Proposition 11.3.2] we see that u ∂Ω belongs to L1,loc (∂Ω, σ) and
 κ−n.t.   κ−n.t.   κ−n.t. 
∂τ j k u ∂Ω = ν j (∂k u) ∂Ω − νk (∂j u) ∂Ω at σ-a.e. point on ∂Ω. (3.3.169)

Combining (3.3.168), (3.3.169), and [113, (6.2.3)] proves that

 κ−n.t.   M
∂τ j k u ∂Ω ∈ M p,λ (∂Ω, σ) and
  κ−n.t.   (3.3.170)
 
∂τ j k u ∂Ω  ≤ CNκ (∇u) M p, λ (∂Ω,σ) .
[M p, λ (∂Ω,σ)] M

Together with (3.3.167), this establishes the claims in (3.3.146). The proof of Theo-
rem 3.3.11 is therefore complete. 

3.4 Fatou-Type Theorems in Two-Sided NTA Domains for

Second-Order Systems

Our first result in this section reveals that, for a null-solution of a weakly elliptic
system, having sufficient regularity on the Sobolev/Besov/Triebel-Lizorkin scales in
a given two-sided NTA domain with a compact Ahlfors regular boundary guarantees
the existence of its nontangential pointwise trace on the boundary of said domain.

Theorem 3.4.1 Let Ω ⊆ Rn be a two-sided NTA domain with a compact Ahlfors

regular boundary. Abbreviate σ := H n−1 ∂Ω and fix an aperture parameter κ > 0.
Also, let L be a homogeneous, constant (complex) second-order M × M weakly
 M
elliptic system in Rn , and consider a null-solution u ∈ 𝒞∞ (Ω) of the system L
in Ω. Then
κ−n.t.
the nontangential boundary trace u ∂Ω exists (in C M ) at σ-a.e. point
κ−n.t. (3.4.1)
on ∂Ω, and the function u ∂Ω is σ-measurable on ∂Ω,

provided either
 p,q M
u belongs to Aα (Ω)bdd with A ∈ {B, F} and
  (3.4.2)
0 < p < ∞, 0 < q ≤ ∞, (n − 1) p1 − 1 + + p1 < α,
3.4 Fatou-Type Theorems in Two-Sided NTA Domains for Second-Order Systems 727

or  1, p M  
u ∈ Wa (Ω)bdd with p ∈ (1, ∞) and a ∈ − p1 , 1 − p1 . (3.4.3)
Moreover, in either of the scenarios described in (3.4.2), (3.4.3) the nontangential
κ−n.t.
boundary trace u ∂Ω actually agrees with the trace considered in the sense of
Sobolev/Besov/Triebel-Lizorkin spaces, i.e.,
⨏
 κ−n.t. 
u ∂Ω (x) = lim+ u dL n at σ-a.e. x ∈ ∂Ω, (3.4.4)
r→0 B(x,r)∩Ω

and for each truncation parameter ε > 0 one has

Nκε u ∈ L po (∂Ω, σ) for some po ∈ (1, ∞). (3.4.5)

Before presenting the proof of Theorem 3.4.1 we make a few remarks further
elaborating on the nature and scope of this result.

Remark 3.4.2 A careful inspection of the argument that yields (3.4.5) gives a more
precise version of this result, namely

in the scenario described in (3.4.2) we have Nκε u ∈ L po (∂Ω, σ) for any

po <  nn−1  if p ∈ (0, 1], and any po <  n−1 #
n−1 $ if p ∈ (1, ∞),
p −α p −min 1,α− p
1
+ +
(3.4.6)
and
in the scenario described in (3.4.3) we have
(3.4.7)
Nκε u ∈ L po (∂Ω, σ) for any po <  n n−1  .
p −1+a +

Remark 3.4.3 It suffices to assume Ω is an NTA domain satisfying an exterior local

John condition and whose boundary is a compact Ahlfors regular set.

The proof of Theorem 3.4.1, given below, makes essential use of Green’s integral
representation formula from [115, §4.4] and the boundary behavior of layer potential
operators.
Proof of Theorem 3.4.1 By eventually replacing Ω by Ω ∩ B(0, R) for some large
R ∈ (0, ∞), there is no loss of generality in assuming that Ω is actually a bounded
two-sided NTA domain with an Ahlfors regular boundary. Thanks to [112, (5.11.66)],
we then see that Ω ⊆ Rn is a bounded (ε, δ)-domain. Consider the scenario described
in (3.4.2) with A := B. Also, from [113, Corollary 9.2.1], [113, Theorems 9.1.1,
9.1.3, 9.1.4], and the present assumptions on p, q, α we conclude that there exist
po, qo ∈ (1, ∞) and so ∈ (0, 1) with the property that
p,q po ,qo
Bα (Ω) → B (Ω). (3.4.8)
so + p1o

As such, work in [115, §4.4] guarantees that we have the integral representation
formula
728 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
   
u = D TrΩ→∂Ω u − 𝒮 ∂νA(u, 0) in Ω, (3.4.9)
where A is a coefficient tensor used to write the given system as L = L A, where D,
𝒮 are the boundary-to-domain double and single layer potential operators associated
with A and Ω, where
 p ,q M  M
TrΩ→∂Ω u ∈ Bsoo o (∂Ω, σ) → L po (∂Ω, σ) , (3.4.10)

with the last inclusion provided by [113, (7.9.10)], and where

 p ,q M  p M
∂νA(u, 0) ∈ Bsoo−1o (∂Ω, σ) → L−1o (∂Ω, σ) , (3.4.11)

with the last inclusion provided by [113, Proposition 11.11.3] (bearing in mind that
the two-sided NTA domain Ω satisfies a two-sided local John condition; cf. [112,
(5.11.28)]).
In turn, from (3.4.9), (3.4.10), (3.4.11), as well as the Calderón-Zygmund theory
developed in relation to the boundary-to-domain double and single layer potential
operators established in [115, §1.5] we conclude that the nontangential boundary
κ−n.t.
trace u ∂Ω exists at σ-a.e. point on ∂Ω and, in fact,
κ−n.t. 1      M
u ∂Ω = 2I + K TrΩ→∂Ω u − S ∂νA(u, 0) ∈ L po (∂Ω, σ) . (3.4.12)
κ−n.t.
In particular, the function u ∂Ω is σ-measurable on ∂Ω. This concludes the proof
of (3.4.1) under the hypotheses stipulated in (3.4.2) with A := B. All other scenarios
described in (3.4.2), (3.4.3) are dealt with similarly, once again relying on Green’s
integral representation formulas derived in [115, §4.4].
Next, based on [113, Theorems 8.3.6, 9.4.5] and [113, (8.3.64), (9.4.99)] we
conclude that (3.4.4) holds. Finally, the claim made in (3.4.5) is seen from (3.4.9)-
(3.4.11), and the nontangential maximal function estimates for the double and single
layers proved in [115, §1.5]. 

In turn, we may use Theorem 3.4.1 to establish the following companion result
for Theorem 3.3.9.

Theorem 3.4.4 Suppose Ω ⊆ Rn (where n ∈ N, n ≥ 3) is a bounded two-sided NTA

domain with an Ahlfors regular boundary. Abbreviate σ := H n−1 ∂Ω and fix an
aperture parameter κ > 0. In addition, let L be a homogeneous, constant (complex)
second-order M × M weakly elliptic system in Rn , and consider a null-solution
 M
u ∈ 𝒞∞ (Ω) of the system L in Ω with the property that Nκ (∇u) ∈ L p (∂Ω, σ) for
 n−1 
some p ∈ n , ∞ . Then the nontangential boundary traces
κ−n.t. κ−n.t.
u ∂Ω and (∇u) ∂Ω exist at σ-a.e. point on ∂Ω, (3.4.13)

and
3.4 Fatou-Type Theorems in Two-Sided NTA Domains for Second-Order Systems 729

⎧ ∗   ∗
1 
1 −1
⎪
⎪ L p (∂Ω, σ) if p ∈ n−1
n , n − 1 and p := p − ,
⎪
⎨
⎪ n−1
Nκ u ∈ L q (∂Ω, σ) if p = n − 1 and q ∈ (0, ∞), (3.4.14)
⎪
⎪
⎪
⎪ L ∞ (∂Ω, σ) if p ∈ (n − 1, ∞].
⎩
An inspection of proof given below reveals that it suffices to assume Ω is a bounded
NTA domain satisfying an exterior local John condition and whose boundary is
Ahlfors regular.
Proof of Theorem 3.4.4 Since the current hypotheses imply that Ω is a UR domain
(cf. [112, (5.2.4), (5.10.24), Definition 5.10.6]), we may invoke Theorem 3.3.4 to
κ−n.t.
conclude that the nontangential boundary trace (∇u) ∂Ω exists at σ-a.e. point on
∂Ω. To proceed, from assumptions, [112, Proposition 8.4.9], and [112, (8.2.28)]
we see that Nκ u ∈ L p (∂Ω, σ). On account of [112, (8.6.51)] we therefore have
 M   M ·n
u ∈ L np/(n−1) (Ω, L n ) and ∇u ∈ L np/(n−1) (Ω, L n ) , hence
 M
u ∈ W 1,np/(n−1) (Ω) . (3.4.15)

As such, the hypotheses in (3.4.3) are satisfied (with p replaced by np/(n − 1) > 1
and a := 0) so we may rely on Theorem 3.4.1 to conclude that the nontangential
κ−n.t.
boundary trace u ∂Ω also exists at σ-a.e. point on ∂Ω. This completes the proof of
(3.4.13).
Turning our attention to (3.4.14), first work under the assumption that the inte-
grability exponent p belongs to (1, n − 1). Pick a coefficient tensor which permits to
express the given system as L = L A, and bring in D and 𝒮, the boundary-to-domain
double and single layer potential operators associated with A and Ω. Granted what
we have proved up to this point, we may invoke the Green-type integral representa-
tion formula for null-solutions of L in terms of double and single layers from [115,
§1.3] in order to express the given function u as
 κ−n.t. 
u = D u ∂Ω − 𝒮(∂νAu) in Ω, (3.4.16)

where the conormal derivative ∂νAu is considered in a pointwise sense, as in Defini-

tion 1.7.1. In particular, Lemma 1.7.3 implies that
 M
∂νAu ∈ L p (∂Ω, σ) . (3.4.17)
1 
1 −1
In addition, if p∗ := p − n−1 we claim that
κ−n.t.  p M  ∗ M
u ∂Ω ∈ L1 (∂Ω, σ) → L p (∂Ω, σ) . (3.4.18)

Indeed, the membership above is provided by [113, Proposition 11.3.2]. As regards

the embedding in (3.4.18), recall from [112, (5.11.28)] that the present hypotheses
ensure that Ω satisfies a two-sided local John condition, so we may rely on item
(ii) in [113, Theorem 11.5.18] to see that said embedding presently holds. Having
730 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

established (3.4.16)-(3.4.18), we may use the nontangential maximal function esti-

mates for the double and single layers proved in [115, Chapter 1] to conclude that
∗
Nκ u ∈ L p (∂Ω, σ).
Consider
  now the task of establishing the membership claimed in (3.4.13) when
p ∈ n−1n , 1 . For starters, the argument so far guarantees (3.4.13) and Theorem 3.4.1
further implies (3.4.5). Granted these, we may invoke the Green-type integral repre-
sentation formula for null-solutions of L in terms of double and single layers from
[115, §2.3] to presently write
 κ−n.t.  .
u = D u ∂Ω − 𝒮(∂νAu) in Ω, (3.4.19)
.
where the weak conormal derivative ∂νAu is now considered in the sense of [113,
Definition 10.2.18]. In particular, [113, Theorem 10.2.24] ensures that
.  M
∂νAu ∈ H p (∂Ω, σ) . (3.4.20)
  1 
1 −1
 
Also, with q := pn/(n − 1) ∈ 1, n
n−1 and p∗ := p − n−1 ∈ 1, n−1
n−2 ⊆ (1, 2],
we may write
κ−n.t.  q,q M  ∗ M
u ∂Ω = TrΩ→∂Ω u ∈ B 1 (∂Ω, σ) → L p (∂Ω, σ) . (3.4.21)
1− q

The equality above is a consequence of (3.4.15), (3.4.4), and (A.0.126), while the
first membership in (3.4.21) is implied by [113, (8.3.38)]. The second membership
in (3.4.21) may be justified by writing
q,q p ∗,2 p ∗,2 ∗
B (∂Ω, σ) → B0 (∂Ω, σ) → F0 (∂Ω, σ) = L p (∂Ω, σ), (3.4.22)
1− q1

where the first embedding is provided by [113, Theorem 7.7.4] upon observing that
(1−1/q)  
1
q − n−1 = n−1
np − 1
n−1 1− n−1
np = 1
p − 1
n−1 = p∗ ,
1
(3.4.23)

the second embedding is a consequence of [113, (7.7.5)] (keeping in mind that

p∗ ≤ 2), and the final equality comes from [113, (7.1.55)]. This concludes the proof
of (3.4.21). In turn, from (3.4.19), (3.4.20), and (3.4.21) we conclude, on account of
the nontangential maximal function estimates for the double and single layers proved
∗
in [115, Chapters 1 and 2], that Nκ u ∈ L p (∂Ω, σ).
Next, consider the membership in (3.4.13) corresponding to p = n − 1. Since ∂Ω
 n−1having Nκ (∇u) ∈ L (∂Ω, σ) implies that Nκ (∇u) ∈ L (∂Ω, σ) for
is bounded, n−1 q

any q ∈ n , n − 1 . Based on what we have proved already, this entails

∗
Nκ u ∈ L q (∂Ω, σ) whenever
 
1 −1
  (3.4.24)
q∗ := q1 − n−1 with q ∈ n−1n , n − 1 arbitrary.
3.5 Fatou-Type Theorems on Riemannian Manifolds 731

Thus Nκ u ∈ L r (∂Ω, σ) for each r ∈ (1, ∞) hence, since ∂Ω has finite measure,
Nκ u ∈ L r (∂Ω, σ) for each r ∈ (0, ∞).
To complete the proof of the theorem there remains to consider the membership
in (3.4.13) in the case when p ∈ (n − 1, ∞]. In such a scenario, from (3.4.15) and
[113, Theorem 8.3.2] (also keeping in mind [112, (5.11.66)]) we see that u = U Ω
 M
where U ∈ W 1,q (Rn ) with q := pn/(n − 1) > n. Since classical embedding
 M
results for Sobolev spaces in the Euclidean setting imply that U ∈ 𝒞α (Rn ) with
α := 1 − n/q ∈ (0, 1), we ultimately conclude that
 M
u ∈ 𝒞α (Ω) with α := 1 − (n − 1)/p ∈ (0, 1) (3.4.25)

in this case. In particular, u is bounded in Ω so Nκ u ∈ L ∞ (∂Ω, σ) (cf. [112,

Lemma 8.3.2]). 

3.5 Fatou-Type Theorems on Riemannian Manifolds

The main goal in this section is to prove Fatou-type theorems on Riemannian mani-
folds in the spirit of our earlier work from §3.1 and §3.3 from the Euclidean setting.
As in the past, we shall treat both first-order and second-order operators. We begin
by treating the former class, with our main result in this regard being Theorem 3.5.4,
stated a little later.
To set the stage, we first extend the notion of “bullet product” from the Euclidean
ambient (cf. [112, Proposition 4.2.3]) to the setting of Riemannian manifolds. Specif-
ically, the integration by parts formula [112, (1.11.29)] suggests that we make the
following definition:

Definition 3.5.1 Let M denote a compact, oriented, boundaryless manifold of class

𝒞2 and real dimension n ∈ N (where n ≥ 2), equipped with a Riemannian metric
tensor g of class 𝒞1 . As in the past, denote by dLgn the Lebesgue measure induced
√
by the metric tensor g (expressed in local coordinates by g dL n ). Next, suppose
E → M, G → M are two Hermitian vector bundles of class 𝒞1 and consider a
first-order differential operator D : E → G which, in local coordinates, has top
coefficients of class 𝒞1 and lower-order coefficients of class 𝒞0 . Finally, pick an
arbitrary open set Ω ⊆ M.
Then for each given section

u ∈ L 1 (Ω, Lgn ) ⊗ E with Du ∈ L 1 (Ω, Lgn ) ⊗ G (3.5.1)

define the G-valued distribution (−i) Sym(D; νg ) • u on ∂Ω by setting (dropping the

dependence on the vector bundle in all pairings)
∫ ∫
Lip(∂Ω) ϕ , (−i) Sym(D; νg ) • u Lip (∂Ω) := Du, Φ dLgn − u, D Φ dLgn
Ω Ω
(3.5.2)
732 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

for each ϕ ∈ Lip(∂Ω) ⊗ G and Φ ∈ Lip(M) ⊗ G with Φ ∂Ω = ϕ.

Here and elsewhere, D is the (real) transpose of D, the symbol ·, · in the right-
hand side denotes the (real) pointwise inner product in the fibers of the vector bundles
considered, and Lip(∂Ω) denotes the space of vector bundle-valued distributions
on the set ∂Ω.
Regarding Definition 3.5.1, we wish to emphasize that Ω ⊆ M is a completely
arbitrary open set. In particular, Ω may not be a set of locally finite perimeter and, as
such, it may not have a geometric measure theoretic outward unit normal νg . Even
though the latter symbol appears in the notation (−i) Sym(D; νg ) • u, its presence
here is purely formal, as the object in question is simply the name of the distribution
defined in (3.5.2).
Reasoning as in the proof of [112, Lemma 4.2.1] it may be shown that the
definition in (3.5.2) is unambiguous, i.e., the right side of (3.5.2) does not depend of
the particular choice of an extension Φ ∈ Lip(M) ⊗ G of the given “test function”
ϕ ∈ Lip(∂Ω) ⊗ G.
Finally, we not that the integrals in the right side of (3.5.2) are absolutely con-
vergent, since u and Du are absolutely integrable while Φ and D Φ are essentially
bounded.
Moving on, let (M, g) be a Riemannian manifold as above. Fix an arbitrary
Ahlfors regular domain Ω ⊆ M and abbreviate σg := Hgn−1 ∂Ω, where Hgn−1 is
the (n − 1)-dimensional Hausdorff measure induced by the metric g on M. Also,
consider a Hermitian vector bundle E → M. In this geometric setting, bring back the
maximal operator acting on any given Lgn -measurable section u : Ω → E according
to
 ∫ !
1
(Pg u)(x) := sup   |u| dLg ,
n
x ∈ ∂Ω, (3.5.3)
r >0 σg ∂Ω ∩ Bg (x, r) Ω∩Bg (x,r)

where Bg (x, r) denotes the geodesic ball of radius r centered at x. We are now ready
to state the following basic weak (“bullet product”) trace result, extending to the
manifold setting work done in [113, §10.2] in the Euclidean space.
Theorem 3.5.2 Let M denote a compact, oriented, boundaryless manifold of class
𝒞2 and real dimension n ∈ N (where n ≥ 2), equipped with a Riemannian metric
tensor g of class 𝒞1 . As in the past, denote by Lgn the Lebesgue measure induced
√
by the metric tensor g (expressed in local coordinates as g dL n ). Next, suppose
E → M, G → M are two Hermitian vector bundles of class 𝒞1 and consider a
first-order differential operator D : E → G which, in local coordinates, has top
coefficients of class 𝒞1 and lower-order coefficients of class 𝒞0 .
In addition, let Ω ⊆ M be an Ahlfors regular domain and set σg := Hgn−1 ∂Ω,
where Hgn−1 is the (n − 1)-dimensional Hausdorff measure induced by the metric
g on M. Also, denote by νg ∈ T ∗ M the geometric measure theoretic outward unit
conormal (with respect to the metric tensor g) to the set of locally finite perimeter
Ω. Finally, assume u : Ω → E is a section having L n -measurable components and
with the property that for some κ ∈ (0, ∞) one has
3.5 Fatou-Type Theorems on Riemannian Manifolds 733
 n−1 
Nκ u ∈ L p (∂Ω, σg ) for some p ∈ n ,∞ . (3.5.4)

In particular, u ∈ L 1 (Ω, Lgn ) ⊗ E (cf. [112, Proposition 8.6.3]), and one also assumes
that Du, computed in the sense of distributions in Ω, belongs to L 1 (Ω, Lgn ) ⊗ G and
satisfies
Pg (Du) ∈ L p (∂Ω, σg ). (3.5.5)
Then with the bullet-product sysmbol interpreted as in Definition 3.5.1

Sym(D; νg ) • u ∈ H p (∂Ω, σg ) ⊗ G (3.5.6)

and there exists a constant C ∈ (0, ∞) independent of u such that

 
 
Sym(D; νg ) • u p
H (∂Ω,σg )⊗ G
 
≤ C Nκ u L p (∂Ω,σg ) + CPg (Du) L p (∂Ω,σg ) . (3.5.7)

κ−n.t.
Finally, under the additional assumptions that Nκ u ∈ L 1 (∂Ω, σg ) and u ∂Ω
exists at σg -a.e. point on ∂Ω it follows that
 κ−n.t. 
Sym(D; νg ) • u = Sym(D; νg ) u ∂Ω (3.5.8)

as distributions on ∂Ω.

Proof Fix two parameters

 n−1+γ
γ ∈ (0, 1) and α ∈ 0 , . (3.5.9)
n−1
The crux of the matter is establishing the existence of a finite constant C > 0,
independent of the section u, with the property that the following pointwise maximal
inequality holds:
    1 
α  
Sym(D; νg ) • u γ (x) ≤ C M ∂Ω |Nκ u| α (x) + C Pg Du (x)
(3.5.10)
at every point x ∈ ∂Ω,

where ( · )γ is the Fefferman-Stein grand maximal operator (defined much as in

[113, (4.1.6)], now in the manifold setting), M ∂Ω is the Hardy-Littlewood maximal
operator on ∂Ω, and Pg is maximal operator defined in (3.5.3). Granted this, the
conclusions in (3.5.6)-(3.5.7) are then implied by the definition of the scale of Hardy
spaces given in [113, Definition 4.2.1] and the boundedness of the Hardy-Littlewood
maximal operator on Lebesgue spaces. To justify (3.5.10), we reason along the lines
of the proof of [113, Theorem 10.2.4]. The only significant difference is that in place
of [113, (10.2.31)] we now have, thanks to Definition 3.5.1,
734 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
∫ ∫
Lip(∂Ω) ψ , (−i) Sym(D; νg ) • u Lip (∂Ω) =  dLgn −
Du, ψ  dLgn
u, D ψ
Ω Ω
(3.5.11)
where ψ ∈ Lip(∂Ω) ⊗ G is arbitrary and ψ  ∈ Lip(M) ⊗ G is a suitable quantitative
extension of ψ constructed using [113, Lemma 10.1.3], much as ψ and ψ  are related
to each other in [113, (10.2.29)-(10.2.30)]. The rest of the argument in the proof of
(3.5.10) may then be adapted from that of [113, Theorem 10.2.4] to the manifold
setting in a straightforward fashion. Finally, the claim made in relation to (3.5.8) may
be justified as in the proof of [113, Proposition 10.2.11]. 
Recall that a differential operator D acting between two vector bundles over
a manifold M is said to have the unique continuation property (UCP) if
whenever u ∈ H 1 (M) (the L 2 -based Sobolev space of order 1 on M is such that
Du = 0 on M and M and u| O = 0 for some open subset O of M then necessarily
u = 0 in any connected component of M overlapping with O. Observe that if D is
an injectively elliptic first-order homogeneous constant complex coefficient system
in Rn then [109, Theorem 11.7, p. 409] (applied to the weakly elliptic second order
operator L := D∗ D) implies that any nullsolution u of D is real-analytic. In view of
[109, Theorem 6.25, pp. 229-230], this implies that
any injectively elliptic first-order homogeneous constant complex
coefficient system in Rn has UCP (i.e., the unique continuation (3.5.12)
property).
Our next proposition contains a basic representation formula for nullsolutions of
nullsolutions of an injectively elliptic first-order differential operator in an arbitrary
open subset of a given Riemannian manifold. The reader is advised to recall the
conventions made in (2.8.90)-(2.8.94).
Proposition 3.5.3 Let M denote a compact, oriented, boundaryless manifold of
class 𝒞3 and real dimension n ∈ N (where n ≥ 2), equipped with a Riemannian
metric tensor g of class 𝒞3 . Suppose E → M, G → M are two Hermitian vector
bundles of class 𝒞3 and let D : E → G be a first-order differential operator which,
in any local coordinate chart U on M and with respect to any local trivializations
of E, G, may be represented as
"
D = A j (x)∂j + B(x) where the coefficients satisfy
j
    (3.5.13)
A j ∈ 𝒞loc U, CrankG×rankE , B ∈ 𝒞2loc U, CrankG×rankE .
3

In addition suppose D is injectively elliptic and has the unique continuation property.
Next, consider
 
0 ≤ V ∈ Hom E, E with coefficients in 𝒞1,
with the property that supp V ∩ Ω =  and V > 0 (3.5.14)
in a nonempty open subset of each component of M,

and introduce the second-order differential operator

3.5 Fatou-Type Theorems on Riemannian Manifolds 735

L := D∗ D + V : E −→ E. (3.5.15)

where D∗ denotes the Hermitian adjoint (i.e., the complex conjugate transposed) of
D. Finally, having fixed an arbitrary open set Ω ⊆ M, select a section
 ∞ 
u ∈ Lloc (Ω, Lgn ) ∩ L 1 (Ω, Lgn ) ⊗ E with Du = 0 in Ω. (3.5.16)

Then12

L : H 1 (M) ⊗ E −→ H −1 (M) ⊗ E is an isomorphism (3.5.17)

and, with E L (·, ·) denoting the Schwartz kernel of L −1 (the inverse of L in (3.5.17)),
at each point x ∈ Ω one has (dropping the dependence on the vector bundle in the
distributional pairing)
 
u(x) = −Lip(∂Ω) I ⊗ (D∗ ) E L (x, ·) , (−i) Sym(D; νg ) • u Lip (∂Ω) (3.5.18)

where I is the identity and the distribution involved in the pairing is considered in
the sense of Definition 3.5.1.
Proof The fact that D is injectively elliptic makes L in (3.5.15) weakly elliptic.
In turn, this ensures that L has a two-sided parametrix E # ∈ OPS1,δ −2 (for some

δ ∈ (0, 1)) which inverts L in (3.5.17) up to smoothing operators (see the proof of
[119, Proposition 2.1, p. 13]). Since the latter are compact, by the Rellich-Kondrachov
theorem, we conclude that L in (3.5.17) is invertible modulo compact operators,
hence ultimately Fredholm. The same is true for L ∗ , and since L − L ∗ is lower-order
we conclude that L has the same index as L ∗ . However, functional analysis tells
that these numbers are opposite to each other, hence the index of L in (3.5.17) is
zero. Since L is also injective in this context (an easy consequence of (3.5.15), the
fact that the operator D has the unique continuation property, and the potential V
is nonnegative as well as strictly positive in some nonempty open subset of each
component of M), we finally conclude that L is actually an isomorphism in the
context of (3.5.17).
The regularity properties of the Schwartz kernel E L (·, ·) of L −1 (including the
nature of the singularity along the diagonal) have been discussed in the proof of
Theorem 2.8.4. The present stronger smoothness assumptions adopted for the metric
tensor g and the coefficients in (3.5.13) ensure similar regularity properties one
unit higher. Also, granted the weak ellipticity of L and the conditions imposed on
the coefficients in (3.5.13), the (local) elliptic regularity result established in [119,
Proposition 2.7, p. 20] may be used to conclude that

u ∈ 𝒞2 (Ω) ⊗ E. (3.5.19)

We continue by making two observations. First, we recall that the principal symbol
of D satisfies the properties
12 throughout this section H ±1 (M) denote L 2 -based Sobolev spaces on M; alternatively,
H 1 (M) = W 1,2 (M) and H −1 (M) = W −1,2 (M)
736 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
 
Sym D ; ξ = −Sym(D; ξ), (3.5.20)
(−i) Sym(D; dϕ) = [D, ϕ], (3.5.21)

for every cotangent vector ξ and, respectively, for every scalar-valued function ϕ of
class 𝒞1 (also identified with the operator of pointwise multiplication by ϕ) where,
generally speaking,
[A, B] := AB − BA (3.5.22)
stands for the commutator of the operators A, B. Second, we note that in the current
setting Definition 3.5.1 gives
∫
Lip(∂Ω) ϕ , (−i) Sym(D; νg ) • u Lip (∂Ω) = − u, D Φ dLgn (3.5.23)
Ω

for each ϕ ∈ Lip(∂Ω) ⊗ G and Φ ∈ Lip(M) ⊗ G with Φ ∂Ω = ϕ.

To justify (3.5.18), fix an arbitrary point x ∈ Ω and pick some scalar function
η ∈ 𝒞2c (Ω) such that η ≡ 1 near x. Then we may write
 
−Lip(∂Ω) I ⊗ (D∗ ) E L (x, ·) , (−i) Sym(D; νg ) • u Lip (∂Ω)
 
= −Lip(∂Ω) I ⊗ (1 − η)(D∗ ) E L (x, ·) , (−i) Sym(D; νg ) • u Lip (∂Ω)
∫
 
= I ⊗ D (1 − η)(D∗ ) E L (x, ·) , u dLgn
Ω
∫
 
= I ⊗ [D, 1 − η](D∗ ) E L (x, ·) , u dLgn
Ω
∫
 
+ I ⊗ (1 − η)D (D∗ ) E L (x, ·) , u dLgn
Ω
∫
 
= I ⊗ (D∗ ) E L (x, ·) , (−i) Sym(D; dη)u dLgn
Ω
∫
 
= I ⊗ (D∗ ) E L (x, ·) , D(η u) dLgn
Ω
 
= D  (Ω) I ⊗ (D∗ ) E L (x, ·) , D(η u) D(Ω)
 
= D  (Ω) I ⊗ D (D∗ ) E L (x, ·) , η u D(Ω)
 
= D  (Ω) I ⊗ L  E L (x, ·) , η u D(Ω) = (η u)(x) = u(x). (3.5.24)

The first equality in (3.5.24) is based on the fact that 1 − η ≡ 1 on ∂Ω, while
the second
 equality in (3.5.24)
 is a consequence of formula (3.5.23) applied with
Φ := I ⊗ (1 − η)(D∗ ) E L (x, ·). The third equality in (3.5.24) simply follows
from (3.5.22). The fourth equality in (3.5.24) is based on two facts. First, relying on
(3.5.21) and (3.5.20) we may write

[D, 1 − η] = −(−i) Sym(D ; dη) = (−i) Sym(D; dη) . (3.5.25)

3.5 Fatou-Type Theorems on Riemannian Manifolds 737

Second, since V = 0 on Ω, we have

   
I ⊗ (1 − η)D (D∗ ) E L (x, ·) = I ⊗ (1 − η)L  E L (x, ·)
= I ⊗ (1 − η)δΔ (x, ·) = 0 in D (Ω), (3.5.26)

by (2.8.178) and (2.8.176). For the fifth equality in (3.5.24) we used again (3.5.21)
to the effect that

(−i) Sym(D; dη)u = [D, η]u = D(ηu) − η(Du) = D(ηu) in Ω, (3.5.27)

given that, by assumption, Du = 0 in Ω. The sixth equality in (3.5.24) is simply the

reinterpretation of the integral as a distributional pairing on Ω (suitably understood,
for test functions of limited regularity). The seventh equality in (3.5.24) is the result
of “integration by parts” in the sense of distributions as in (2.8.96), also relying on
the fact that ηu ∈ 𝒞2c (Ω) ⊗ E (cf. (3.5.19)). Finally, the last equality in (3.5.24) is a
consequence of I ⊗ L  E L (x, ·) = δΔ (x, ·) (cf. (2.8.178)) and the formula recorded
in (2.8.175). 

We are now in a position to state and prove the following remarkable quantitative
Fatou-type result for nullsolutions of injectively elliptic first-order differential oper-
ators in UR subdomains of Riemannian manifolds. In view of (3.5.12), this is a most
natural generalization of Theorem 3.1.6 from the Euclidean setting.

Theorem 3.5.4 Let M denote a compact, oriented, boundaryless manifold of class

𝒞3 and real dimension n ∈ N (where n ≥ 2), equipped with a Riemannian metric
tensor g of class 𝒞3 . Suppose E → M, G → M are two Hermitian vector bundles
of class 𝒞3 and

D : E → G is a first-order differential operator which is

(3.5.28)
injectively elliptic and has the unique continuation property.

Assume that, in any local coordinate chart U on M and with respect to any local
trivializations of E, G, the operator D may be represented as
"
D = A j (x)∂j + B(x) where the coefficients satisfy
j
    (3.5.29)
A j ∈ 𝒞3loc U, CrankG×rankE , B ∈ 𝒞2loc U, CrankG×rankE .

Next, suppose Ω ⊆ M is a UR domain. Denote the (n − 1)-dimensional Hausdorff

measure associated with the metric g on M by Hgn−1 , abbreviate σg := Hgn−1 ∂Ω,
and denote by νg ∈ T ∗ M the geometric measure theoretic outward unit conormal
(with respect to the metric tensor g) to Ω. Finally, select a section u : Ω → E

Nκ u ∈ L p (∂Ω, σg ) and Du = 0 in Ω, (3.5.30)

 
for an aperture parameter κ > 0 and an integrability exponent p ∈ n−1
n
,∞ .
738 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

Then the nontangential boundary trace

κ−n.t.
u ∂Ω exists (in E) at σg -a.e. point on ∂Ω and
κ−n.t.
the function u ∂Ω is σg -measurable on ∂Ω;
κ−n.t. (3.5.31)
also, u ∂Ω belongs to the space L p (∂Ω, σg ) ⊗ E
 κ−n.t. 
and one has u ∂Ω  L p (∂Ω,σg )⊗ E ≤ Nκ u L p (∂Ω,σg ) .

Moreover, for any other given aperture parameter κ  ∈ (0, ∞) the nontangential
κ  −n.t. κ−n.t.
boundary trace u ∂Ω also exists, and agrees with u ∂Ω , at σg -a.e. point on ∂Ω.
Finally, if p ∈ (1, ∞) then
   κ−n.t. 
Nκ u p ≈ u ∂Ω  L p (∂Ω,σg )⊗ E . (3.5.32)
L (∂Ω,σg )

In particular, Theorem 3.5.4 is applicable to Dirac operators with sufficiently

regular coefficients, such as the Hodge-Dirac operator d + δ acting on differential
forms on a sufficiently regular Riemannian manifold. For D taken to be Def, the
deformation tensor acting on vector fields on a sufficiently regular Riemannian
manifold, we obtain a Fatou-type theorem for Killing fields in arbitrary UR domains.
Here is the proof of Theorem 3.5.4.
Proof of Theorem 3.5.4 Having chosen a real-valued scalar function

0 ≤ V ∈ 𝒞1 (M) such that supp V ∩ Ω =  and V > 0

(3.5.33)
in a nonempty open subset of each component of M,

define the second-order differential operator L := D∗ D + V : E → E, as in (3.5.15).

Then Proposition 3.5.3 ensures that

L : H 1 (M) ⊗ E −→ H −1 (M) ⊗ E is an isomorphism, (3.5.34)

and we shall denote by E L the Schwartz kernel of L −1 , the inverse of L in (3.5.34).

Next, from the membership in (3.5.30), [112, Lemma 8.3.1], and the last estimate
in [112, Proposition 8.6.3] we deduce (bearing in mind that Ω is relatively compact)
that  ∞ 
u ∈ Lloc (Ω, Lgn ) ∩ L 1 (Ω, Lgn ) ⊗ E. (3.5.35)
In view of the properties in (3.5.30), Theorem 3.5.2 also guarantees that
 p  
H (∂Ω, σg ) ⊗ G if p ∈ n−1 n , 1 ,
Sym(D; νg ) • u belongs to (3.5.36)
L p (∂Ω, σg ) ⊗ G if p ∈ (1, ∞),

and if p ∈ (1, ∞) there exists a constant C ∈ (0, ∞) independent of u such that

   
 
Sym(D; νg ) • u p ≤ C Nκ u L p (∂Ω,σg ) . (3.5.37)
L (∂Ω,σg )⊗ G
3.5 Fatou-Type Theorems on Riemannian Manifolds 739

Thanks to (3.5.35), (3.5.36), and the compatibility between the duality pairing from
[113, Theorem 4.6.1] and the distributional
 pairing (cf. [113, Lemma 4.6.4]), Propo-
sition 3.5.3 applies and gives that if p ∈ n−1
n , 1 then at each point x ∈ Ω we have
 
u(x) = − I ⊗ (D∗ ) E L (x, ·) , (−i) Sym(D; νg ) • u (3.5.38)
with the brackets denoting the pairing between the Hardy space H p and its dual
(H p )∗ (cf. [113, Theorem 4.6.1]), while for p ∈ (1, ∞) we have
∫
 
u(x) = − I ⊗ (D∗ ) E L (x, ·) , (−i) Sym(D; νg ) • u G dσg (3.5.39)
∂Ω

at each point x ∈ Ω. From (3.5.36), (3.5.38), (3.5.39), and item (5) in Theorem 2.8.2
κ−n.t.
we are now able to see that u ∂Ω exists at σg -a.e. point on ∂Ω. All other properties
in (3.5.31) (as well as the subsequent claim in the statement) are then seen from this,
the membership in (3.5.30), [112, (8.9.8)], and [112, Proposition 8.9.8].
Furthermore, from (3.5.36), (3.5.39) item (4) in Theorem 2.8.2, and the very last
property in the statement of Theorem 3.5.2 we conclude that if p ∈ (1, ∞) there
  κ−n.t.    

Sym(D; νg ) u ∂Ω  p ≈ Nκ u L p (∂Ω,σg ) . (3.5.40)
L (∂Ω,σg )⊗ G

Granted this, the equivalence claimed in (3.5.32) follows with the help of the estimate
in the last line of (3.5.31) and the fact that
    κ−n.t.    κ−n.t. 
Sym D; νg u  ≤ C u ∂Ω  L p (∂Ω,σg )⊗ E . (3.5.41)
∂Ω L p (∂Ω,σg )⊗ G

The proof of Theorem 3.5.4 is now complete. 

In the remaining portion of this section we shall concern ourselves with Fatou-type
theorems for second-order differential operators on UR subdomains of Riemannian
manifolds. To set the stage, we make a couple of definitions.

Definition 3.5.5 Let M be a compact, boundaryless, oriented 𝒞2 manifold equipped

with a Riemannian metric g, and consider two 𝒞2 Hermitian vector bundles E, G over
M. Suppose all metrics involved are of class 𝒞1 , and equip E, G with connections

∇ E : TM × E −→ E, ∇ G : TM × G −→ G. (3.5.42)

Consider a first-order differential operator D : E → G with top coefficients of class

𝒞1 and zero-order coefficients of class 𝒞0 . Finally, fix an arbitrary 𝒞1 vector field
X ∈ TM and introduce the first-order differential operators
 
P : 𝒞1 (M) ⊗ E → 𝒞0 (M) ⊗ G × 𝒞0 (M) ⊗ E, P := D, −∇XE , (3.5.43)
740 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

and
Q : 𝒞1 (M) ⊗ G × 𝒞1 (M) ⊗ E −→ 𝒞0 (M) ⊗ G,
(3.5.44)
Q(a, b) := ∇XG a + Db, ∀ a ∈ 𝒞1 (M) ⊗ G, ∀ b ∈ 𝒞1 (M) ⊗ E.

In addition, having picked a set of locally finite perimeter Ω ⊆ M, introduce

σg := Hgn−1 ∂Ω and denote by νg the geometric measure theoretic outward unit
conormal to Ω.. In this setting, given any section u ∈ W (Ω) ⊗ E define the G-valued
1,1

distribution ∂τP,Q u on ∂Ω as
.
∂τP,Q u := i Sym(Q; νg ) • (Pu) ∈ Lip(∂Ω). (3.5.45)

A few comments are in order here. First, the composition

QP : 𝒞2 (M) ⊗ E −→ 𝒞0 (M) ⊗ G
is a first-order differential operator, (3.5.46)
namely QP = ∇XG D − D∇XE =: [∇X , D].

Granted this and the current assumptions, Definition 3.5.1 then guarantees that the
.
“bullet tangential derivative” ∂τP,Q u is indeed a well-defined G-valued distribution
on the set ∂Ω. To better understand its action on arbitrary test functions, we need to
take a closer look at the operators P, Q from (3.5.43)-(3.5.44). Observe that for any

a ∈ 𝒞1 (M) ⊗ G, b ∈ 𝒞1 (M) ⊗ E, w ∈ 𝒞1 (M) ⊗ G (3.5.47)

we may compute, based on definitions,

Q w, (a, b) = w, Q(a, b) = w, ∇X a + w, Db

= ∇   
X w, a + D w, b = (∇ X , D )w, (a, b) , (3.5.48)

thus  
Q = ∇X, D . (3.5.49)
As such, for each u ∈ W 1,1 (Ω) ⊗ E and Φ ∈ Lip(M) ⊗ G, we see from (3.5.43) and
(3.5.49) that

Pu, Q Φ = (Du, −∇X u), (∇ 

X , D )Φ

= Du, ∇ 
X Φ − ∇ X u, D Φ . (3.5.50)
.
At this point, we may explicitly write the action of the G-valued distribution ∂τP,Q u
on an any given test function ϕ ∈ Lip(∂Ω) ⊗ G as
3.5 Fatou-Type Theorems on Riemannian Manifolds 741
.P,Q
Lip(∂Ω) ϕ , ∂τ u Lip (∂Ω) = Lip(∂Ω) ϕ , i Sym(Q; νg ) • (Pu) Lip (∂Ω)
∫ ∫

= Pu , Q Φ dLg −n
QPu , Φ dLgn
Ω Ω
∫ ∫
= Du, ∇
X Φ dL g −
n
∇X u, D Φ dLgn
Ω Ω
∫
− [∇X , D]u , Φ dLgn, (3.5.51)
Ω

where Φ ∈ Lip(M) ⊗ G is any global Lipschitz section of E extending the given ϕ,

i.e., such that Φ ∂Ω = ϕ. Indeed, the first equality in (3.5.51) comes from (3.5.45),
the second equality in (3.5.51) is implied by the definition of the bullet product
introduced in Definition 3.5.1, while the final equality in (3.5.51) is a consequence
of (3.5.50) and (3.5.46).
Going further, the second definition we wish to make pertains to the notion of weak
(“bullet product”) conormal derivative associated with a any (quasi-)factorization of
a given second-order differentential operator.

Definition 3.5.6 Retain the above hypotheses on the Riemannian manifold (M, g)
and the Hermitian vector bundles E, G → M. Also, assume L : E → E is a
second-order differential operator which has the quasi-factorization
 +W
L = DD (3.5.52)

for some first-order differential operators

D : E −→ G,  : G −→ E,
D (3.5.53)

with top coefficients of class 𝒞1 and lower-order coefficients of class 𝒞0 , and a

zero-order operator W ∈ Hom (E, E) with bounded coefficients. Finally, let Ω ⊂ M
be an Ahlfors regular domain with outward unit conormal νg : ∂Ω → T ∗ M and
surface measure σg := Hgn−1 ∂Ω.
Then, for each section u ∈ W 1,1 (Ω). ⊗ E such that Lu = 0 in Ω, define the weak
(bullet product) conormal derivative ∂νLg u as the E-valued distribution on ∂Ω given
by .
 νg ) • (Du) ∈ Lip(∂Ω),
∂νLg u := (−i)Sym( D; (3.5.54)
 νg ) • (Du) is considered as in Definition 3.5.1.
where (−i)Sym( D;

Hence, in the context of the above definition, from (3.5.54) and (3.5.2) (written
 we see that for every test function ϕ ∈ Lip(∂Ω) ⊗ E we have
with D replaced by D)
.
Lip(∂Ω) ϕ , ∂νLg u Lip (∂Ω)
 νg ) • (Du)
= Lip(∂Ω) ϕ , (−i)Sym( D; Lip (∂Ω)

∫ ∫
=−  Φ dLgn −
Du , D Wu , Φ dLgn (3.5.55)
Ω Ω
742 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

for every Φ ∈ Lip(M) ⊗ E such that Φ ∂Ω = ϕ.

We are now prepared to state and prove the following basic representation formula
for nullsolutions of second-order differential operators acting between two vector-
bundles over a Riemannian manifold. Once again, reader is advised to recall the
conventions made in (2.8.90)-(2.8.94).

Theorem 3.5.7 Let M be a compact, boundaryless, oriented 𝒞3 manifold of (real)

dimension n ≥ 2, equipped with a Riemannian metric g of class 𝒞3 . Consider two
Hermitian vector bundles, E, G over M, of class 𝒞3 and equip them with connections

∇ = ∇ E : TM × E → E, ∇ = ∇ G : TM × G → G. (3.5.56)

Next, assume L : E → E is a weakly elliptic second-order partial differential

operator which has the quasi-factorization
 +W
L = DD (3.5.57)

for some first-order differential operators

D : E −→ G,  : G −→ E,
D (3.5.58)

with top coefficients of class 𝒞3 and lower-order coefficients of class 𝒞2 , and a

zero-order operator W ∈ Hom (E, E) with coefficients of class 𝒞1 . In addition,
suppose13

L : H 1 (M) ⊗ E −→ H −1 (M) ⊗ E is an isomorphism, (3.5.59)

and denote by E L (x, y) the Schwartz kernel of L −1 , the inverse of L in (3.5.59).

Suppose Ω ⊂ M is an arbitrary Ahlfors regular domain, with outward unit
conormal νg : ∂Ω → T ∗ M and surface measure σg := Hgn−1 ∂Ω. Finally, fix an
aperture parameter κ ∈ (0, ∞) and assume the section u ∈ W 1,1 (Ω) ⊗ E is such that

Nκ u belongs to the space L 1 (∂Ω, σg ),

κ−n.t.
u ∂Ω exists σg -a.e. on ∂Ω, (3.5.60)
and Lu = 0 in Ω.

Then for each 𝒞1 vector field X ∈ TM and each point x ∈ Ω one has (with
notation introduced in Definition 3.5.5 and Definition 3.5.6)

13 recall that H ±1 (M) denote L 2 -based Sobolev spaces on M; alternatively, H 1 (M) = W 1,2 (M)
and H −1 (M) = W −1,2 (M)
3.5 Fatou-Type Theorems on Riemannian Manifolds 743
    .
 E L (x, ·) , ∂τP,Q u
∇X u (x) = −Lip(∂Ω) I ⊗ D Lip (∂Ω)

  .
− Lip(∂Ω) ∇X ⊗ I E L (x, ·) , ∂νLg u Lip (∂Ω)
∫
 κ−n.t. 
+ ωX (x, y), u ∂Ω (y) dσg (y)
∂Ω
∫
     κ−n.t. 
+ I ⊗ i Sym [∇X , D]; νg D E L (x, y), u (y) dσg (y)
∂Ω
∂Ω
∫
     κ−n.t. 
− I ⊗ (−i)Sym ∇
X ; νg W E L (x, y), u ∂Ω (y) dσg (y)
∂Ω
(3.5.61)

where ωX is the kernel function from (2.8.252).

Proof Recall that the regularity properties of the Schwartz kernel E L (·, ·) of L −1
(including the nature of the singularity along the diagonal) have been discussed in
the proof of Theorem 2.8.4. The current stronger smoothness assumptions adopted for
the metric tensor g and the coefficients of L ensure similar regularity properties one
unit higher. Also, granted the weak ellipticity of L and the conditions imposed on its
coefficients, the (local) elliptic regularity result established in [119, Proposition 2.7,
p. 20] may be used to conclude that

u ∈ 𝒞2 (Ω) ⊗ E. (3.5.62)

To proceed, fix a point x ∈ Ω along with a 𝒞1 vector field X ∈ TM, and

pick a scalar function η ∈ 𝒞2c (Ω) such that η ≡ 1 near x. Then the properties
.
of η and
 the definition of ∂τP,Q u from (3.5.51) combined with (3.5.50) written for
 E L (x, ·) imply
Φ := I ⊗ (1 − η) D
744 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
  .
 E L (x, ·) , ∂τP,Q u
−Lip(∂Ω) I ⊗ D Lip (∂Ω)

  .
= −Lip(∂Ω) I ⊗ (1 − η) D  E L (x, ·) , ∂τP,Q u Lip (∂Ω)
∫
 
=− I ⊗ [∇ 
X , 1 − η] D E L (x, ·) , Du dL g
n
Ω
∫
 
− I ⊗ (1 − η)∇ 
X D E L (x, ·) , Du dL g
n
Ω
∫
 
+  E L (x, ·) , ∇X u dLgn
I ⊗ [D, 1 − η] D
Ω
∫
 
+  E L (x, ·) , ∇X u dLgn
I ⊗ (1 − η)D D
Ω
∫
 
+  E L (x, ·) , (1 − η)[∇X , D]u dLgn .
I⊗D (3.5.63)
Ω

Next, apply (3.5.21) and (3.5.20) to rewrite the commutators involving 1 − η in

(3.5.63) in the form
    
[∇X , 1 − η] = (−i)Sym ∇ X ; d(1 − η) = (−i) Sym ∇ X ; dη
(3.5.64)
and [D, 1 − η] = i Sym(D ; dη) = (−i) Sym(D; dη) .
 
Also, since D D = L  − W  and I ⊗ (1 − η)L  E L (x, ·) = 0 in Ω thanks to
(2.8.178) and (2.8.176), we have
∫
 
 E L (x, ·) , ∇X u dLgn
I ⊗ (1 − η)D D
Ω
∫
 
=− I ⊗ (1 − η)W  E L (x, ·) , ∇X u dLgn . (3.5.65)
Ω
 
Moving on, in light of (3.5.55) applied with Φ := ∇X ⊗ I E L (x, ·) we may write
  .
−Lip(∂Ω) ∇X ⊗ I E L (x, ·) , ∂νLg u Lip (∂Ω)

  .
= −Lip(∂Ω) ∇X ⊗ (1 − η)I E L (x, ·) , ∂νLg u Lip (∂Ω)
∫
 
= , 1 − η] E L (x, ·), Du dLgn
∇X ⊗ [ D
Ω
∫
 
+  E L (x, ·), Du dLgn
∇X ⊗ (1 − η) D
Ω
∫
 
+ ∇X ⊗ I E L (x, ·), (1 − η)Wu dLgn . (3.5.66)
Ω
3.5 Fatou-Type Theorems on Riemannian Manifolds 745

Using again (3.5.21) and (3.5.20), we have

, 1 − η] = i Sym( D
[D  ; dη) = (−i) Sym( D;
 dη) . (3.5.67)

A combination of (3.5.63)-(3.5.67) and subsequent use to replace the first two terms
in the right-hand side of (3.5.61), allow us to conclude that (3.5.61) will follow once
we show the following new identity:
∫
     
∇X u (x) = − I⊗D  E L (x, ·) , (−i) Sym ∇X ; dη Du dLgn
Ω
∫
 
− I ⊗ ∇ 
X D E L (x, ·) , (1 − η)Du dL g
n
Ω
∫
 
+  E L (x, ·) , (−i) Sym(D; dη)∇X u dLgn
I⊗D
Ω
∫
 
− I ⊗ (1 − η)W  E L (x, ·) , ∇X u dLgn
Ω
∫
 
+  E L (x, ·) , (1 − η)[∇X , D]u dLgn
I⊗D
Ω
∫
 
+  dη)Du dLgn
∇X ⊗ I E L (x, ·), (−i) Sym( D;
Ω
∫
 
+  E L (x, ·), (1 − η)Du dLgn
∇X ⊗ D
Ω
∫
 
+ ∇X ⊗ I E L (x, ·), (1 − η)Wu dLgn
Ω
∫
κ−n.t.
+ ωX (x, ·), u ∂Ω dσg
∂Ω
∫
    κ−n.t.
+ I ⊗ i Sym [∇X , D]; νg D E L (x, ·), u dσg
∂Ω
∂Ω
∫
    κ−n.t.
− I ⊗ (−i)Sym ∇
X ; νg W E L (x, ·), u ∂Ω dσg
∂Ω

=: I1Ω + I2Ω + I3Ω + I4Ω + I5Ω + I6Ω + I7Ω + I8Ω + I1∂Ω + I2∂Ω + I3∂Ω . (3.5.68)

Above, for each j ∈ {1, . . . , 8} the symbol I jΩ denotes the j-th solid integral over Ω,
including the sign in front, as listed in the right-hand side of (3.5.68), while for each
k ∈ {1, 2, 3} the symbol Ik∂Ω denotes the k-th boundary integral over ∂Ω, including
the sign in front, as listed in the right-hand side of (3.5.68).
The crux of the matter now is to invoke the Divergence Formula established in
[112, Theorem 1.11.1] in the context of Riemannian manifolds in order to convert
the boundary integrals into solid integrals. We shall define the vector field F (to
which [112, Theorem 1.11.1] applies) implicitly, asking that for each point y ∈ Ω
746 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains


the inner product of F(y) ∈ Ty M with an arbitrary cotangent vector ξ ∈ Ty∗ M is
given by
     

F(y), ξ := ∇X ⊗ (−i)Sym D ; ξ D  E L (x, y) , (1 − η(y))u(y)
   
− I ⊗ (−i)Sym D ; ξ ∇
X

D 
E L (x, y) , (1 − η(y))u(y)
   
− I ⊗ (−i)Sym [∇X , D]; ξ D E L (x, y) , (1 − η(y))u(y)
   
− I ⊗ (−i)Sym ∇
X ; ξ W E L (x, y) , (1 − η(y))u(y) . (3.5.69)

From this writing, assumptions, and (3.5.62) it transpires that F ∈ Lloc

1 (Ω, L n )⊗TM,
g
κ−n.t.
that the nontangential trace that F exists at σg -a.e. point on ∂Ω, and that
∂Ω
κ−n.t.
Nκ F ∈ L 1 (∂Ω, σg ). that F ∂Ω exists at σg -a.e. point on ∂Ω. In relation to the third
term in the right side of (3.5.69) we wish to note that, based on (3.5.20) and the fact
that
[A, B] = −[A, B ] for generic operators A, B, (3.5.70)
we have    
Sym [∇X ; D]; ξ = Sym [∇ 
X ; D ]; ξ (3.5.71)
for every ξ ∈ Ty∗ M.This is going to be relevant shortly.
The applicability of the Divergence Formula from [112, Theorem 1.11.1] requires
 the differential geometric divergence of F computed in the sense of
that divg F,
distributions in Ω, belongs to L 1 (Ω, Lgn ). To see why this is indeed the case, we
shall identify divg F explicitly. To this end, pick arbitrary scalar-valued function
ϕ ∈ 𝒞2c (Ω) and use (3.5.69) and (3.5.71) to write
 
 ϕ = − F,
divg F,  dϕ (3.5.72)
∫ 
  
=−  E L (x, ·) , (1 − η)u dLgn
∇X ⊗ (−i)Sym D ; dϕ D
Ω
∫  
 
+ I ⊗ (−i)Sym D ; dϕ ∇ 
X D E L (x, ·) , (1 − η)u dL g
n
Ω
∫    
+ I ⊗ (−i)Sym [∇
X , D 
]; dϕ  E L (x, ·) , (1 − η)u dLgn
D
Ω
∫    
+ I ⊗ (−i)Sym ∇
X ; dϕ W E L (x, ·) , (1 − η)u dLgn .
Ω

Invoking (3.5.21), each one of the principal symbols in (3.5.72) may be written as
commutators in the form
3.5 Fatou-Type Theorems on Riemannian Manifolds 747
   
(−i)Sym D ; dϕ = D, ϕ ,
     
(−i)Sym [∇ 
X , D ]; dϕ = [∇ X , D ], ϕ , (3.5.73)
    
(−i)Sym ∇ X ; dϕ = ∇ X , ϕ .

Using (3.5.73) back in (3.5.72) and then integrating by parts (note that the integration
by parts does not yield any boundary terms due to the fact that ϕ is compactly
supported in Ω) further gives
∫    
 ϕ =−
divg F, ∇X ⊗ ϕ D  E L (x, ·) , D (1 − η)u dLgn
Ω
∫  
+  E L (x, ·) , (1 − η)u dLgn
∇X ⊗ ϕ D D
Ω
∫    
+ I ⊗ ϕ∇ 
X D E L (x, ·) , D (1 − η)u dL g
n
Ω
∫  
− I ⊗ ϕ D  ∇
X

D 
E L (x, ·) , (1 − η)u dLgn
Ω
∫    
−  E L (x, ·) , [∇X , D] (1 − η)u dLgn
I ⊗ ϕD
Ω
∫  
 
− I ⊗ ϕ[∇
X , D ] D E L (x, ·) , (1 − η)u dLgn
Ω
∫    
+ I ⊗ ϕW  E L (x, ·) , ∇X (1 − η)u dLgn
Ω
∫  
− I ⊗ ϕ∇
X W 
E L (x, ·) , (1 − η)u dLgn, (3.5.74)
Ω

keeping in mind for the fifth integral in (3.5.74) that [∇  

X , D ] = −[∇ X , D] (cf.
(3.5.70)).
  Invoking the quasi-factorization assumed in (3.5.57), the fact that we have
I ⊗ L  E L (x, ·) = δΔ (x, ·) (cf. (2.8.178)), and that η ≡ 1 near x, we also obtain
∫  
∇X ⊗ ϕ D D  E L (x, ·) , (1 − η)u dLgn
Ω
∫  
=− ∇X ⊗ ϕ W  E L (x, ·) , (1 − η)u dLgn . (3.5.75)
Ω

Similarly, since

D  ∇      
X D + [∇ X , D ] D + ∇ X W

= D  ∇         
X D + ∇X D D − D ∇X D + ∇X W

= ∇
XL

(3.5.76)
748 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
 
and again making use of the fact that I ⊗ L  E L (x, ·) = δΔ (x, ·) (cf. (2.8.178)) and
η ≡ 1 near x, we obtain
∫  
− I ⊗ ϕ D  ∇
X

D 
E L (x, ·) , (1 − η)u dLgn
Ω
∫  
 
− I ⊗ ϕ[∇
X , D ] D E L (x, ·) , (1 − η)u dL g
n
Ω
∫  
− I ⊗ ϕ∇
XW

E L (x, ·) , (1 − η)u dLgn
Ω
∫  
=− ϕ I ⊗ ∇
X L 
E L (x, ·) , (1 − η)u dLgn = 0. (3.5.77)
Ω

Together, (3.5.74), (3.5.75), and (3.5.77) imply

∫    
 ϕ =−
divg F, ϕ ∇X ⊗ D  E L (x, ·) , D (1 − η)u dLgn
Ω
∫  
− ϕ ∇X ⊗ W  E L (x, ·) , (1 − η)u dLgn
Ω
∫    
+ ϕ I ⊗ ∇ 
X D E L (x, ·) , D (1 − η)u dL g
n
Ω
∫    
− ϕ  E L (x, ·) , [∇X , D] (1 − η)u dLgn
I⊗D
Ω
∫    
+ ϕ I ⊗ W  E L (x, ·) , ∇X (1 − η)u dLgn . (3.5.78)
Ω

Since ϕ ∈ 𝒞2c (Ω) is arbitrary, from (3.5.78) we ultimately conclude that

   
divg F = − ∇X ⊗ D  E L (x, ·) , D (1 − η)u
 
− ∇X ⊗ W  E L (x, ·) , (1 − η)u
   
+ I ⊗ ∇ 
X D E L (x, ·) , D (1 − η)u
   
−  E L (x, ·) , [∇X , D] (1 − η)u
I⊗D
   
+ I ⊗ W  E L (x, ·) , ∇X (1 − η)u . (3.5.79)

An inspection of (3.5.79) reveals that since η ≡ 1 near x, there is no blow up term in

divg F and, in the final analysis, we obtain

div F ∈ L 1 (Ω, Lgn ). (3.5.80)

3.5 Fatou-Type Theorems on Riemannian Manifolds 749

To continue, write
∫
divg F dLgn = I I1 + I I2 + I I3 + I I4 + I I5, (3.5.81)
Ω

where
∫    
I I1 := −  E L (x, ·) , D (1 − η)u dLgn
∇X ⊗ D
Ω
∫  
I I2 := − ∇X ⊗ W  E L (x, ·) , (1 − η)u dLgn
Ω
∫    
I I3 := I ⊗ ∇
X

D 
E L (x, ·) , D (1 − η)u dLgn
Ω
∫    
I I4 := −  E L (x, ·) , [∇X , D] (1 − η)u dLgn
I⊗D
Ω
∫    
I I5 := I ⊗ W  E L (x, ·) , ∇X (1 − η)u dLgn . (3.5.82)
Ω

To proceed further, observe that

[D, 1 − η]u = −[D, η]u is a compactly supported function in Ω. (3.5.83)

Using (3.5.83) and integrating by parts, we obtain

∫   
I I1 + I7Ω = − ∇X ⊗ D  E L (x, ·) , [D, 1 − η]u dLgn
Ω
∫   
=  E L (x, ·) , [D, η]u dLgn
∇X ⊗ D
Ω
∫   
=  η]u dLgn .
∇X ⊗ I E L (x, ·) , D[D, (3.5.84)
Ω

In addition, since in general A[B, C] = [AB, C] − [A, C]B, we write

   
 η]u = DD,
D[D,  η u − D,  η Du
 
= [L, η]u − [W, η]u − (−i)Sym D; dη Du
 
= L(ηu) − (−i)Sym D;  dη Du, (3.5.85)

making use of the fact that Lu = 0 in Ω, the operator W commutes with multiplication
by η, and also relying on (3.5.21). Together, (3.5.84), (3.5.85), and integration by
parts (with no boundary terms since ηu is compactly supported in Ω) yield
750 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
∫  
I I1 + I7Ω = ∇X ⊗ I E L (x, ·) , L(ηu) dLgn
Ω
∫    
−  dη Du dLgn
∇X ⊗ I E L (x, ·) , (−i)Sym D;
Ω
∫  
= ∇X ⊗ L  E L (x, ·) , ηu dLgn − I6Ω
Ω
 
= ∇X u (x) − I6Ω . (3.5.86)

Next, (3.5.83) and integration by parts allows us to write (in light of the fact that
[D, η]u is compactly supported in Ω)
∫  
I I3 + I2Ω = I ⊗ ∇
X

D 
E L (x, ·) , [D, 1 − η]u dLgn
Ω
∫
 
=− I ⊗ ∇ 
X D E L (x, ·) , [D, η]u dL g
n
Ω
∫
 
=−  E L (x, ·) , ∇X [D, η]u dLgn .
I⊗D (3.5.87)
Ω

We may also combine I I4 and I5Ω in the form

∫    
I I4 + I5Ω = − I⊗D  E L (x, ·) , [∇X , D], 1 − η u dLgn
Ω
∫
 
=  E L (x, ·) , [∇X , D]ηu − η[∇X , D]u dLgn .
I⊗D (3.5.88)
Ω

On account of (3.5.21) we may trade the principal symbols in I1Ω and I3Ω for com-
mutators and obtain
∫
   
I1Ω + I3Ω = − I⊗D E L (x, ·) , ∇X , η Du dLgn
Ω
∫
 
+  E L (x, ·) , [D, η]∇X u dLgn .
I⊗D (3.5.89)
Ω

A close inspection of I I2 and I8Ω reveals that they cancel, thus

∫  
Ω
I I2 + I8 = − ∇X ⊗ W  E L (x, ·) , (1 − η)u dLgn
Ω
∫
  
+ ∇X ⊗ I E L (x, ·), (1 − η)Wu dLgn = 0, (3.5.90)
Ω

while
3.5 Fatou-Type Theorems on Riemannian Manifolds 751
∫    
I I5 + I4Ω = I ⊗ W  E L (x, ·) , ∇X , 1 − η u dLgn
Ω
∫
   
= I ⊗ (1 − η)W  E L (x, ·) , η, ∇X u dLgn . (3.5.91)
Ω

Thus combined, (3.5.86), (3.5.90), (3.5.87), (3.5.88), (3.5.91), and (3.5.89) imply
∫
divg F dLgn + I1Ω + I2Ω + I3Ω + I4Ω + I5Ω + I6Ω + I7Ω + I8Ω
Ω

= (I I1 + I7Ω + I6Ω ) + (I I2 + I8Ω ) + (I I3 + I2Ω )

+ (I I4 + I5Ω ) + (I I5 + I4Ω ) + (I1Ω + I3Ω )

∫
   
= ∇X u (x) − I⊗D E L (x, ·) , ∇X [D, η]u dLgn
Ω
∫
 
+  E L (x, ·) , [∇X , D]ηu − η[∇X , D]u dLgn
I⊗D
Ω
∫    
+ I ⊗ W  E L (x, ·) , η, ∇X u dLgn
Ω
∫
   
−  E L (x, ·) , ∇X , η Du dLgn
I⊗D
Ω
∫
 
+  E L (x, ·) , [D, η]∇X u dLgn
I⊗D
Ω
∫

  
= ∇X u (x) +  E L (x, ·) , D[η, ∇X ]u dLgn
I⊗D
Ω
∫    
+ I ⊗ W  E L (x, ·) , η, ∇X u dLgn, (3.5.92)
Ω

where the last equality is a consequence of

 
−∇X [D, η] + [∇X , D]η − η[∇X , D] − ∇X ; η D + [D; η]∇X
= −∇X D η + ∇X η D + ∇X D η − D∇X η − η∇X D + η D∇X
− ∇X ηD + η∇X D + D η∇X − η D∇X
= −D∇X η + D η∇X = D[η, ∇X ]. (3.5.93)

Given that [η, ∇X ]u is of class 𝒞1 and compactly supported in Ω, we may again

integrate by parts
752 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
∫
 
 E L (x, ·) , D[η, ∇X ]u dLgn
I⊗D
Ω
 
= D  (Ω)
 E L (x, ·) , D[η, ∇X ]u
I⊗D D(Ω)

 
= D  (Ω)
 E L (x, ·) , [η, ∇X ]u
I ⊗ D D D(Ω)


 
= I ⊗ L  − W  E L (x, ·) , [η, ∇X ]u D(Ω)
D  (Ω)
∫
 
= δx, [η, ∇X ]u − I ⊗ W  E L (x, ·) , [η, ∇X ]u dLgn
Ω
∫
 
=− I ⊗W E L (x, ·) , [η, ∇X ]u dLgn (3.5.94)
Ω

again recalling that [η, ∇X ]u is identically zero near x. Finally, (3.5.92), (3.5.94),
and the Divergence Theorem imply
∫ ∫
κ−n.t.
I1∂Ω + I2∂Ω + I3∂Ω = νg, F ∂Ω dσg = divg F dLgn
∂Ω Ω
 
= ∇X u (x) − I1Ω − I2Ω − I3Ω − I4Ω − I5Ω − I6Ω − I7Ω − I8Ω (3.5.95)

which proves (3.5.68), as wanted. 

As a corollary of Theorem 3.5.7, we obtain the following Fatou-type result for
nullsolutions of weakly elliptic second-order differential operators, in UR subdo-
mains Riemannian manifolds.

Corollary 3.5.8 Let M be a compact, boundaryless, oriented 𝒞3 manifold of (real)

dimension n, where n ∈ N satisfies n ≥ 2, equipped with a Riemannian metric g of
class 𝒞3 . Consider a Hermitian vector bundle E → M of class 𝒞2 and denote by
∇ = ∇ E : TM × E → E a connection on E.
Next, assume L : E → E is a weakly elliptic second-order partial differential
operator with the property that, when written in local coordinates as
 
 
αβ αβ αβ
L= ∂j a jk ∂k + b j ∂j + d (3.5.96)
j,k j α,β

its coefficients satisfy

αβ αβ
a jk ∈ 𝒞3, bj ∈ 𝒞2, d αβ ∈ 𝒞1 . (3.5.97)

In addition, suppose

L : H 1 (M) ⊗ E −→ H −1 (M) ⊗ E is an isomorphism. (3.5.98)

3.5 Fatou-Type Theorems on Riemannian Manifolds 753

Let Ω ⊆ M be an arbitrary UR domain, and abbreviate σg := Hgn−1 ∂Ω. Finally,

1,1
fix an aperture parameter κ > 0 and assume the section u ∈ Wloc (Ω) ⊗ E is such
κ−n.t.
that the nontangential boundary trace u ∂Ω exists at σg -a.e. point on ∂Ω and

Lu = 0 in Ω, as well as Nκ u ∈ L 1 (∂Ω, σg )
 n  (3.5.99)
and Nκ (∇u) ∈ L p (∂Ω, σg ) for some p ∈ n−1 ,∞ .
κ−n.t.
Then the nontangential trace (∇u) ∂Ω exists at σg -a.e. point on ∂Ω. Moreover,
κ−n.t.
as a function, (∇u) ∂Ω is σg -measurable and actually independent of the aperture
parameter κ.

Proof We debut by making two observations. First, since the scale of Lebesgue
spaces on ∂Ω is nested,
 n as
 far as the conclusion we seek is concerned we may
assume that p ∈ n−1 , 1 . Second, we can find a 𝒞3 Hermitian vector bundle
G → M along with two first-order differential operators
 : G −→ E,
D : E −→ G and D (3.5.100)

with top coefficients of class 𝒞3 and lower-order coefficients of class 𝒞2 , along with
a zeroth-order operator W ∈ Hom (E, E) with coefficients of class 𝒞1 which allow
us to express L as
 + W.
L = DD (3.5.101)
To proceed, fix an arbitrary 𝒞1 vector field X ∈ TM. In relation to D and X in-
troduce the first-order differential operators P, Q as in (3.5.43)-(3.5.44). Granted the
current assumptions, we may invoke Theorem 3.5.2 which, in light of Definition 3.5.5
and Definition 3.5.6, guarantees that
. .
∂τP,Q u ∈ H p (∂Ω, σg ) ⊗ G and ∂νLg u ∈ H p (∂Ω, σg ) ⊗ E. (3.5.102)

Having established these memberships and keeping in mind the compatibility be-
tween the duality pairing from [113, Theorem 4.6.1] and the distributional pairing
(cf. [113, Lemma 4.6.4]), we may now write the representation formula (3.5.61) from
Theorem 3.5.7 with the brackets in the right-hand side now denoting the pairing be-
tween the Hardy space H p and its dual (H p )∗ (cf. [113, Theorem 4.6.1]). With this
interpretation, item (5) in Theorem 2.8.2 applies and, together with Lemma 2.8.9,
κ−n.t.
gives that (∇X u) ∂Ω exists σg -a.e. on ∂Ω. In view of the arbitrariness of X, this
κ−n.t.
ultimately shows that the nontangential boundary trace (∇u) ∂Ω exists at σg -a.e.
point on ∂Ω. The remaining properties claimed in the statement then follow from
this, the last membership in (3.5.99), [112, (8.9.8)], and [112, Proposition 8.9.8]. 
It is of interest to prove Fatou-type results in the spirit of Corollary 3.5.8 with no
assumptions on the existence of the nontangential trace of the function in question.
We succeed in accomplishing this later, in Theorem 3.5.10 which, in effect, is
our main result for second-order operators in this section. As a preamble, we first
754 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

establish two basic representation formulas with weak traces in the theorem below
(the reader is advised to recall the conventions made in (2.8.90)-(2.8.94)).

Theorem 3.5.9 Let M be a compact, boundaryless, oriented 𝒞3 manifold of (real)

dimension n ≥ 2, equipped with a Riemannian metric g of class 𝒞3 . Consider
two Hermitian vector bundles, E and G over M, of class 𝒞3 and equip them with
connections

∇ = ∇ E : TM × E → E, ∇ = ∇ G : TM × G → G. (3.5.103)

Next, assume L : E → E is a weakly elliptic second-order partial differential

operator which has the quasi-factorization
 +W
L = DD (3.5.104)

for some first-order differential operators

D : E −→ G,  : G −→ E,
D (3.5.105)

with top coefficients of class 𝒞3 and lower-order coefficients of class 𝒞2 , and a

zero-order operator W ∈ Hom (E, E) with coefficients of class 𝒞1 . In addition,
suppose

L : H 1 (M) ⊗ E −→ H −1 (M) ⊗ E is an isomorphism, (3.5.106)

and denote by E L (x, y) the Schwartz kernel of L −1 , the inverse of L in (3.5.106).

Finally, having fixed an Ahlfors regular domain Ω ⊆ M with geometric measure
theoretic outward unit conormal νg ∈ T ∗ M, pick a section satisfying

u ∈ W 1,1 (Ω) ⊗ E and Lu = 0 in Ω. (3.5.107)

Then for each 𝒞1 vector field X ∈ TM and each point x ∈ Ω one has (with
notation introduced in Definition 3.5.1, Definition 3.5.5, and Definition 3.5.6)
    .
 E L (x, ·) , ∂τP,Q u
∇X u (x) = −Lip(∂Ω) I ⊗ D Lip (∂Ω) (3.5.108)
  .
− Lip(∂Ω) ∇X ⊗ I E L (x, ·) , ∂νLg u Lip (∂Ω)

   
 E L (x, ·) , (−i)Sym D; νg • u
− Lip(∂Ω) ∇X ⊗ D Lip (∂Ω)

   
+ Lip(∂Ω) I ⊗ ∇ 
X D E L (x, ·) , (−i)Sym D; νg • u Lip (∂Ω)

    
 E L (x, ·) , (−i)Sym [∇X , D ; νg • u
− Lip(∂Ω) I ⊗ D Lip (∂Ω)

   
+ Lip(∂Ω) I ⊗ W  E L (x, ·) , (−i)Sym ∇X ; νg • u Lip (∂Ω) .
3.5 Fatou-Type Theorems on Riemannian Manifolds 755

In addition, under the same assumptions as above, at each point x ∈ Ω one has
(with notation introduced in Definition 3.5.1 and Definition 3.5.6)
   
 E L (x, ·) , (−i) Sym D; νg • u
u(x) = −Lip(∂Ω) I ⊗ D Lip (∂Ω)

.
− Lip(∂Ω) E L (x, ·) , ∂νLg u Lip (∂Ω) . (3.5.109)

Proof The regularity properties of the Schwartz kernel E L (·, ·) of L −1 (including

the nature of the singularity along the diagonal) have been reviewed during the
course of the proof of Theorem 2.8.4. The present stronger smoothness assumptions
adopted for the metric tensor g and the coefficients of L guarantee similar regularity
properties one unit higher. In addition, granted the weak ellipticity of L and the
conditions imposed on its coefficients, the (local) elliptic regularity result established
in [119, Proposition 2.7, p. 20] may be employed to conclude that

u ∈ 𝒞2 (Ω) ⊗ E. (3.5.110)

To continue, select an arbitrary point x ∈ Ω together with a 𝒞1 vector field

X ∈ TM, and pick a scalar function η ∈ 𝒞2c (Ω) such that η ≡ 1 near x. Then the
 E L (x, ·) imply
properties of η and formula (3.5.51) written for Φ := I ⊗ (1 − η) D
  .
 E L (x, ·) , ∂τP,Q u
−Lip(∂Ω) I ⊗ D Lip (∂Ω)

  .
= −Lip(∂Ω) I ⊗ (1 − η) D  E L (x, ·) , ∂τP,Q u Lip (∂Ω)
∫
 
=− I ⊗ [∇ 
X , 1 − η] D E L (x, ·) , Du dL g
n
Ω
∫
 
− I ⊗ (1 − η)∇ 
X D E L (x, ·) , Du dL g
n
Ω
∫
 
+  E L (x, ·) , ∇X u dLgn
I ⊗ [D, 1 − η] D
Ω
∫
 
+  E L (x, ·) , ∇X u dLgn
I ⊗ (1 − η)D D
Ω
∫
 
+  E L (x, ·) , (1 − η)[∇X , D]u dLgn .
I⊗D (3.5.111)
Ω

Next, we invoke (3.5.21) and (3.5.20) to rewrite the commutators involving 1 − η

in (3.5.111) in the form
    
[∇
X , 1 − η] = (−i)Sym ∇ X ; d(1 − η) = (−i) Sym ∇ X ; dη
(3.5.112)
and [D, 1 − η] = i Sym(D ; dη) = (−i) Sym(D; dη) .
756 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
 
 = L  − W  and I ⊗ (1 − η)L  E L (x, ·) = 0 in Ω (as seen from
Also, since D D
(2.8.178) and (2.8.176)), we have
∫
 
 E L (x, ·) , ∇X u dLgn
I ⊗ (1 − η)D D
Ω
∫
 
=− I ⊗ (1 − η)W  E L (x, ·) , ∇X u dLgn . (3.5.113)
Ω
 
Moving on, in light of (3.5.55) applied with Φ := ∇X ⊗ I E L (x, ·), we may write
  .
−Lip(∂Ω) ∇X ⊗ I E L (x, ·) , ∂νLg u Lip (∂Ω)

  .
= −Lip(∂Ω) ∇X ⊗ (1 − η)I E L (x, ·) , ∂νLg u Lip (∂Ω)
∫
 
= , 1 − η] E L (x, ·), Du dLgn
∇X ⊗ [ D
Ω
∫
 
+  E L (x, ·), Du dLgn
∇X ⊗ (1 − η) D
Ω
∫
 
+ ∇X ⊗ I E L (x, ·), (1 − η)Wu dLgn . (3.5.114)
Ω

Recalling (3.5.21) and (3.5.20), we have

, 1 − η] = i Sym( D
[D  ; dη) = (−i) Sym( D;
 dη) (3.5.115)

which when used in (3.5.114) yields

  .
−Lip(∂Ω) ∇X ⊗ I E L (x, ·) , ∂νLg u Lip (∂Ω)
∫
 
= ∇X ⊗ I E L (x, ·), (−i) Sym( D; dη)Du dLgn
Ω
∫
 
+  E L (x, ·), Du dLgn
∇X ⊗ (1 − η) D
Ω
∫
 
+ ∇X ⊗ I E L (x, ·), (1 − η)Wu dLgn . (3.5.116)
Ω

For the remaining four terms in the right hand-side of (3.5.108) we use the same
circle of ideas and the definition in (3.5.2) to rewrite the
 pairings as combinations

of solid integrals over Ω. First, corresponding to Φ := ∇X ⊗ (1 − η) D E L (x, ·) in
(3.5.2), we obtain
3.5 Fatou-Type Theorems on Riemannian Manifolds 757
   
 E L (x, ·) , (−i)Sym D; νg • u
−Lip(∂Ω) ∇X ⊗ D Lip (∂Ω)

   
= −Lip(∂Ω) ∇X ⊗ (1 − η) D  E L (x, ·) , (−i)Sym D; νg • u Lip (∂Ω)
∫
 
=−  E L (x, ·) , Du dLgn
∇X ⊗ (1 − η) D
Ω
∫
 
+  E L (x, ·) , u dLgn
∇X ⊗ [D, 1 − η] D
Ω
∫
 
+  E L (x, ·) , u dLgn
∇X ⊗ (1 − η)D D
Ω
∫
 
=−  E L (x, ·) , Du dLgn
∇X ⊗ (1 − η) D
Ω
∫
 
+  E L (x, ·) , (−i) Sym(D; dη)u dLgn
∇X ⊗ D
Ω
∫
 
− ∇X ⊗ (1 − η)W  E L (x, ·) , u dLgn (3.5.117)
Ω

where for the last equality

 we have also relied on (3.5.112). Second, definition (3.5.2)
applied with Φ := I ⊗ (1 − η)∇ 
X D E L (x, ·) and (3.5.112) allow us to write

   
Lip(∂Ω) I ⊗ ∇ 
X D E L (x, ·) , (−i)Sym D; νg • u Lip (∂Ω)

   
= Lip(∂Ω) I ⊗ (1 − η)∇ 
X D E L (x, ·) , (−i)Sym D; νg • u Lip (∂Ω)
∫
 
= I ⊗ (1 − η)∇ 
X D E L (x, ·) , Du dL g
n
Ω
∫
 
− I ⊗ [D, 1 − η]∇ 
X D E L (x, ·) , u dL g
n
Ω
∫
 
− I ⊗ (1 − η)D ∇ 
X D E L (x, ·) , u dL g
n
Ω
∫
 
= I ⊗ (1 − η)∇ 
X D E L (x, ·) , Du dL g
n
Ω
∫
 
− I ⊗ ∇ 
X D E L (x, ·) , (−i)Sym(D; dη)u dL g
n
Ω
∫
 
− I ⊗ (1 − η)D ∇ 
X D E L (x, ·) , u dL g .
n
(3.5.118)
Ω
 
 E L (x, ·) and with the operator
Third, (3.5.2) applied with Φ := I ⊗ (1 − η) D
[∇X , D in place of D, implies
758 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
    
 E L (x, ·) , (−i)Sym [∇X , D ; νg • u
−Lip(∂Ω) I ⊗ D Lip (∂Ω)

    
= −Lip(∂Ω) I ⊗ (1 − η) D  E L (x, ·) , (−i)Sym [∇X , D ; νg • u Lip (∂Ω)
∫
  
=−  E L (x, ·) , [∇X , D u dLgn
I ⊗ (1 − η) D
Ω
∫
    
+  E L (x, ·) , u dLgn
I ⊗ [∇X , D , 1 − η D
Ω
∫
  
+ I ⊗ (1 − η)[∇X , D  E L (x, ·) , u dLgn
D
Ω
∫
  
=−  E L (x, ·) , [∇X , D u dLgn
I ⊗ (1 − η) D
Ω
∫
  
+  E L (x, ·) , (−i)Sym([∇X , D ; dη)u dLgn
I⊗D
Ω
∫
  
+ I ⊗ (1 − η)[∇X , D  E L (x, ·) , u dLgn
D (3.5.119)
Ω

since by (3.5.21) and (3.5.20),

      
[∇X , D , 1 − η = − [∇X , D , η = (−i) Sym([∇X , D ; dη) . (3.5.120)
 
Finally, a fourth application of (3.5.2) with Φ := I ⊗ (1 − η)W  E L (x, ·) and with
the operator ∇X in place of D, permits us to write
   
Lip(∂Ω) I ⊗ W  E L (x, ·) , (−i)Sym ∇X ; νg • u Lip (∂Ω)

   
= Lip(∂Ω) I ⊗ (1 − η)W  E L (x, ·) , (−i)Sym ∇X ; νg • u Lip (∂Ω)
∫
 
= I ⊗ (1 − η)W  E L (x, ·) , ∇X u dLgn
Ω
∫
 
− I ⊗ [∇
X , 1 − η]W E L (x, ·) , u dL g
n
Ω
∫
 
− I ⊗ (1 − η)∇
X W E L (x, ·) , u dL g
n
Ω
∫
 
= I ⊗ (1 − η)W  E L (x, ·) , ∇X u dLgn
Ω
∫
   
− I ⊗ W  E L (x, ·) , (−i) Sym ∇X ; dη u dLgn
Ω
∫
 
− I ⊗ (1 − η)∇
X W E L (x, ·) , u dL g
n
(3.5.121)
Ω
3.5 Fatou-Type Theorems on Riemannian Manifolds 759

by also invoking (3.5.112) for the last equality.

Next, a combination of (3.5.111), (3.5.113), (3.5.116), (3.5.117)-(3.5.119), and
(3.5.121) allows us to conclude that (3.5.108) will follow once we show the following
new identity:
760 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
∫
     
∇X u (x) = −  E L (x, ·) , (−i) Sym ∇X ; dη Du dLgn
I⊗D
∫Ω
 
− I ⊗ ∇ 
X D E L (x, ·) , (1 − η)Du dL g
n

∫Ω
 
+  E L (x, ·) , (−i) Sym(D; dη)∇X u dLgn
I⊗D
∫Ω
 
− I ⊗ (1 − η)W  E L (x, ·) , ∇X u dLgn
∫Ω
 
+  E L (x, ·) , (1 − η)[∇X , D]u dLgn
I⊗D
∫Ω
 
+  dη)Du dLgn
∇X ⊗ I E L (x, ·), (−i) Sym( D;
∫Ω
 
+  E L (x, ·), (1 − η)Du dLgn
∇X ⊗ D
∫Ω
 
+ ∇X ⊗ I E L (x, ·), (1 − η)Wu dLgn
∫Ω
 
−  E L (x, ·) , Du dLgn
∇X ⊗ (1 − η) D
∫Ω
 
+  E L (x, ·) , (−i) Sym(D; dη)u dLgn
∇X ⊗ D
∫Ω
 
− ∇X ⊗ (1 − η)W  E L (x, ·) , u dLgn
∫Ω
 
+ I ⊗ (1 − η)∇ 
X D E L (x, ·) , Du dL g
n

∫Ω
 
− I ⊗ ∇ 
X D E L (x, ·) , (−i)Sym(D; dη)u dL g
n

∫Ω
 
− I ⊗ (1 − η)D ∇ 
X D E L (x, ·) , u dL g
n

∫Ω
  
−  E L (x, ·) , [∇X , D u dLgn
I ⊗ (1 − η) D
∫Ω
    
+  E L (x, ·) , (−i)Sym ∇X , D ; dη u dLgn
I⊗D
∫Ω
  
+ I ⊗ (1 − η)[∇X , D  E L (x, ·) , u dLgn
D
∫Ω
 
+ I ⊗ (1 − η)W  E L (x, ·) , ∇X u dLgn
∫Ω
   
− I ⊗ W  E L (x, ·) , (−i) Sym ∇X ; dη u dLgn
∫Ω
 
− I ⊗ (1 − η)∇
X W E L (x, ·) , u dL g
n
Ω

20
=: Ij . (3.5.122)
j=1
3.5 Fatou-Type Theorems on Riemannian Manifolds 761

Above, for each j ∈ {1, . . . , 20} the symbol I j stands for the j-th solid integral over
Ω, including the sign in front, in the order listed in the right-hand side of (3.5.122).
An inspection of (3.5.122) reveals some immediate cancellations:

I2 + I12 = 0, I4 + I18 = 0, I5 + I15 = 0, I7 + I9 = 0, I8 + I11 = 0. (3.5.123)

Also, recalling (3.5.21) we may combine

I1 +I3 + I16 (3.5.124)

∫
     
=  E L (x, ·) , − ∇X , η Du + [D, η]∇X u + [∇X , D], η u dLgn .
I⊗D
Ω

Observe that
   
− ∇X , η Du+[D, η]∇X u + [∇X , D], η u
= −∇X η Du + η ∇X Du + D η ∇X u − η D∇X u
+ ∇X D ηu − D∇X ηu − η∇X Du + η D∇X u
= ∇X [D, η]u − D[∇X , η]u in Ω. (3.5.125)

This identity may be used in (3.5.124) to further obtain

∫
 
I1 + I3 + I16 + I19 = I⊗D E L (x, ·) , ∇X [D, η]u dLgn
Ω
∫
 
−  E L (x, ·) , D[∇X , η]u dLgn
I⊗D
Ω
∫
 
− I ⊗ W  E L (x, ·) , [∇X , η]u dLgn
Ω
∫
 
= I ⊗ ∇ 
X D E L (x, ·) , [D, η]u dL g
n
Ω
∫
 
−  E L (x, ·) , [∇X , η]u dLgn
I ⊗ D D
Ω
∫
 
− I ⊗ W  E L (x, ·) , [∇X , η]u dLgn
Ω
∫
 
= −I13 − I ⊗ L  E L (x, ·) , [∇X , η]u dLgn
Ω

= −I13 − δx, [∇X , η]u . (3.5.126)

The second equality in (3.5.126) is based on integration by parts (keeping in mind

that [D, η]u and [∇X , η]u are compactly supported), and the last equality uses
(2.8.178). Upon observing that [∇X , η]u is equal to zero near x, from (3.5.126) and
762 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains

(2.8.176) we conclude that

I1 + I3 + I16 + I19 + I13 = 0. (3.5.127)

Next, we claim that

I14 + I17 + I20 = 0. (3.5.128)
To see why this is true, note that

−D ∇     
X D +[∇ X , D] D − ∇ X W (3.5.129)

= −D ∇         
X D + D ∇X D − ∇X D D − ∇X W
 
= −∇   
X (D D + W ) = −∇ X L (3.5.130)

which implies
∫
 
I14 + I17 + I20 = − I ⊗ ∇
X L E L (x, ·) , (1 − η)u dL g = 0,
n
(3.5.131)
Ω
   
since I ⊗ (1 − η)∇  
X L E L (x, ·) = I ⊗ (1 − η)∇ X δΔ (x, ·) = 0 thanks to (2.8.178),
(2.8.176), and the fact that 1 − η is zero near x. This justifies the claim made in
(3.5.128).
We are left with simplifying I6 + I10 . In this regard, it is useful to observe since
Lu = 0 in Ω we have
 
 η Du = Dη
D,  Du − η DDu =D  η Du − η(L − W)u

 η Du + η Wu in Ω,
=D (3.5.132)

thus
 η]u = DD
D[D,  ηu − D  ηDu = (L − W)ηu − D
 ηDu
 
 η Du in Ω.
= L(ηu) − D, (3.5.133)

Consequently, based on (3.5.133), integration by parts, (2.8.180), and (2.8.175) we

have
3.5 Fatou-Type Theorems on Riemannian Manifolds 763
∫
   
I6 = ∇X ⊗ I E L (x, ·) , D, η Du dLgn
Ω
∫
 
=−  η]u dLgn
∇X ⊗ I E L (x, ·) , D[D,
Ω
 
+ D  (Ω) ∇X ⊗ I E L (x, ·) , L(ηu) D(Ω)

 
= −I10 + D  (Ω) ∇X ⊗ L  E L (x, ·) , ηu D(Ω)

= −I10 + (∇X u)(x). (3.5.134)

Together, (3.5.123), (3.5.127), (3.5.128), and (3.5.134) imply

(I2 + I12 ) + (I4 + I18 ) + (I5 + I15 ) + (I7 + I9 ) + (I8 + I11 ) (3.5.135)
+ (I1 + I3 + I16 + I19 + I13 ) + (I14 + I17 + I20 ) + (I6 + I10 ) = (∇X u)(x),

establishing (3.5.122). This completes the proof of (3.5.108).

The justification of (3.5.109) uses the same circle of ideas as those employed in
the proof (3.5.108). First, we note that the reasoning
 that lead to (3.5.117) now leads,
after applying (3.5.2) with Φ := I ⊗ (1 − η) D  E L (x, ·), to the following sequence
of identities:
   
−Lip(∂Ω) I ⊗ D E L (x, ·) , (−i)Sym D; νg • u Lip (∂Ω)

   
= −Lip(∂Ω) I ⊗ (1 − η) D  E L (x, ·) , (−i)Sym D; νg • u Lip (∂Ω)
∫
 
=−  E L (x, ·) , Du dLgn
I ⊗ (1 − η) D
Ω
∫
 
+  E L (x, ·) , u dLgn
I ⊗ [D, 1 − η] D
Ω
∫
 
+  E L (x, ·) , u dLgn
I ⊗ (1 − η)D D
Ω
∫
 
=−  E L (x, ·) , Du dLgn
I ⊗ (1 − η) D
Ω
∫
 
+  E L (x, ·) , (−i) Sym(D; dη)u dLgn
I⊗D
Ω
∫
− E L (x, ·) , (1 − η)Wu dLgn . (3.5.136)
Ω

Second, observe that the argument that resulted in (3.5.116) now implies
764 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
.
−Lip(∂Ω) E L (x, ·) , ∂νLg u Lip (∂Ω)
∫
=  dη)Du dLgn
E L (x, ·), (−i) Sym( D;
Ω
∫
+  E L (x, ·), Du dLgn
(1 − η) D
Ω
∫
+ E L (x, ·), (1 − η)Wu dLgn . (3.5.137)
Ω

Combined, (3.5.136), (3.5.137) and (3.5.133) yield

   
 E L (x, ·) , (−i)Sym D; νg • u
−Lip(∂Ω) I ⊗ D Lip (∂Ω)

.
− Lip(∂Ω) E L (x, ·) , ∂νLg u Lip (∂Ω)
∫
 
=− I ⊗ (1 − η) D  E L (x, ·) , Du dLgn
Ω
∫
 
+  E L (x, ·) , (−i) Sym(D; dη)u dLgn
I⊗D
Ω
∫
 
− I ⊗ (1 − η)W  E L (x, ·) , u dLgn
Ω
∫
+  dη)Du dLgn
E L (x, ·), (−i) Sym( D;
Ω
∫
 
+  E L (x, ·), Du dLgn
I ⊗ (1 − η) D
Ω
∫
+ E L (x, ·), (1 − η)Wu dLgn
Ω
∫
 
=  E L (x, ·) , [D, η]u dLgn
I⊗D
Ω
∫
 
+  η Du dLgn
E L (x, ·), D,
Ω
∫
 
=  η]u + D,
E L (x, ·) , D[D,  η Du dLgn
Ω
 
= D  (Ω) I ⊗ L  E L (x, ·) , ηu D(Ω) = u(x), (3.5.138)

with the last equality provided by (2.8.178).This establishes (3.5.109) and finishes
the proof of Theorem 3.5.9. 
In turn, the representation formula from Theorem 3.5.9 is a key ingredient in
the proof of our main Fatou-type result for nullsolutions of weakly elliptic second-
3.5 Fatou-Type Theorems on Riemannian Manifolds 765

order differential operators, in arbitrary UR subdomains of a Riemannian manifold.

Theorem 3.5.10, stated next, further refines work in Corollary 3.5.8, and extends to
the realm of manifolds earlier Euclidean results from Theorem 3.3.9.
Theorem 3.5.10 Suppose M is a compact, boundaryless, oriented 𝒞3 manifold of
(real) dimension n, where n ∈ N satisfies n ≥ 2, equipped with a Riemannian metric
g of class 𝒞3 . Consider a 𝒞3 Hermitian vector bundle E → M and denote by
∇ = ∇ E : TM × E → E a connection on E.
Next, suppose L : E → E is a weakly elliptic second-order differential operator
with the property that14

L : H 1 (M) ⊗ E −→ H −1 (M) ⊗ E is an isomorphism, (3.5.139)

and which may be expressed in local coordinates as

 
 
αβ αβ αβ
L= ∂j a jk ∂k + b j ∂j + d (3.5.140)
j,k j α,β

for coefficients satisfying

αβ αβ
a jk ∈ 𝒞3, bj ∈ 𝒞2, d αβ ∈ 𝒞1 . (3.5.141)

Also, let Ω ⊆ M be a UR domain, and abbreviate σg := Hgn−1 ∂Ω. Finally, fix

1,1
an aperture parameter κ ∈ (0, ∞) and assume the section u ∈ Wloc (Ω) ⊗ E satisfies
 
Nκ u, Nκ (∇u) ∈ L p (∂Ω, σg ) for some p ∈ n−1 , ∞
n
(3.5.142)
and Lu = 0 in Ω.
κ−n.t. κ−n.t.
Then the nontangential traces u ∂Ω and (∇u) ∂Ω exist at σg -a.e. point on ∂Ω.
Also, as functions, they are σg -measurable and are actually independent of the
aperture parameter κ.
Proof As noted in the proof of Corollary 3.5.8, there exist a Hermitian vector bundle
G → M of class 𝒞3 together with two first-order differential operators
 : G −→ E,
D : E −→ G and D (3.5.143)

having top coefficients of class 𝒞3 and lower-order coefficients of class 𝒞2 , plus a

zeroth-order operator W ∈ Hom (E, E) with coefficients of class 𝒞1 allowing us to
express L as
 + W.
L = DD (3.5.144)
Given that the scale of Lebesgue spaces on the compact set ∂Ω is nested, as far as
the existence of nontangential boundary traces is concerned we may assume without
14 recall that H ±1 (M) denote L 2 -based Sobolev spaces on M; alternatively, H 1 (M) = W 1,2 (M)
and H −1 (M) = W −1,2 (M)
766 3 Quantitative Fatou-Type Theorems in Arbitrary UR Domains
 n 
loss of generality that actually p ∈ n−1 , 1 . Thanks to the memberships in the first
line of (3.5.142) and the last estimate in [112, Proposition 8.6.3] we deduce (bearing
in mind that Ω is relatively compact) that
np
u ∈ W n−1 ,1 (Ω) ⊗ E ⊆ W 1,1 (Ω) ⊗ E. (3.5.145)

In particular,

u ∈ L 1 (Ω, Lgn ) ⊗ E and Du ∈ L 1 (Ω, Lgn ) ⊗ G. (3.5.146)

Also, from the current assumptions and [113, (10.1.9) in Lemma 10.1.1] we deduce
that ∗
Pg (Du) ∈ L p , p (∂Ω, σg ) ⊆ L p (∂Ω, σg ),
 
1 −1
(3.5.147)
where p∗ := p1 − n−1 ∈ (1, ∞).
Granted (3.5.146)-(3.5.147), we then conclude from Theorem 3.5.2 that
 
(−i)Sym D; νg • u belongs to H p (∂Ω, σg ) ⊗ G. (3.5.148)

To continue, pick an arbitrary 𝒞1 vector field X ∈ TM and bring in the first-order

differential operators P, Q constructed as in (3.5.43)-(3.5.44) for the present D
and X. In light of the current assumptions, Theorem 3.5.2, Definition 3.5.5, and
Definition 3.5.6 imply that
. .
∂τP,Q u ∈ H p (∂Ω, σg ) ⊗ G and ∂νLg u ∈ H p (∂Ω, σg ) ⊗ E. (3.5.149)

Since, much as in (3.5.146)-(3.5.147), we also have

∇X u ∈ L 1 (Ω, Lgn ) ⊗ E and [∇X , D]u ∈ L 1 (Ω, Lgn ) ⊗ G, (3.5.150)

as well as
   
Pg ∇X u ∈ L p (∂Ω, σg ) and Pg [∇X , D]u ∈ L p (∂Ω, σg ), (3.5.151)

we may once again invoke Theorem 3.5.2 to conclude that

  
(−i)Sym [∇X , D ; νg • u ∈ H p (∂Ω, σg ) ⊗ G,
  (3.5.152)
and (−i)Sym ∇X ; νg • u ∈ H p (∂Ω, σg ) ⊗ E.

Once the memberships in (3.5.148), (3.5.149), (3.5.152) have been established we

may proceed to write, keeping in mind the compatibility between the duality pairing
from [113, Theorem 4.6.1] and the distributional pairing (cf. [113, Lemma 4.6.4]),
the representation formulas (3.5.108) and (3.5.109) from Theorem 3.5.9 with all
brackets in the right side now denoting the pairing between the Hardy space H p
and its dual (H p )∗ (cf. [113, Theorem 4.6.1]). With this interpretation, item (5) in
κ−n.t. κ−n.t.
Theorem 2.8.2 applies gives that both (∇X u) ∂Ω and u ∂Ω exist σg -a.e. on ∂Ω. In
3.5 Fatou-Type Theorems on Riemannian Manifolds 767

view of the arbitrariness of X, the former ultimately proves that the nontangential
κ−n.t.
boundary trace (∇u) ∂Ω exists at σg -a.e. point on ∂Ω. The remaining properties
claimed in the statement of the theorem then follow from this, the memberships in
the first line of (3.5.142), [112, (8.9.8)], and [112, Proposition 8.9.8]. 
Chapter 4
Green Functions and Uniqueness for Boundary
Problems for Second-Order Systems

The main topic in this chapter is the study of Green functions (associated, in a rather
inclusive sense, with second-order systems) with special emphasis on the role they
play in establishing uniqueness for boundary problems. Basically all results in this
chapter make essential use of the brand of Divergence Theorem developed earlier,
in Volume I ([112]).
Any object G referred to as Green function1 (in relation to a given differential
operator L and an open set Ω ⊆ Rn ) should satisfy two basic properties. First, G
should be a fundamental solution for L with pole (or singularity) at some point
x0 ∈ Ω and, second, the boundary trace of G (understood in a suitable sense) should
vanish on ∂Ω. These two properties alone typically fail to identify such an object
uniquely (thus justifying referring to it as “a Green function,” as opposed to “the
Green function”). The reason for adopting such a level of generality is to allow for
additional flexibility in adapting these basic Green functions to the specific demands
made in relation to various aims (such as proving uniqueness for boundary value
problems, or establishing integral representation formulas for null-solutions of L in
Ω involving Green functions). This is the point of view is pervasive in this chapter.
In §4.1 we examine the role of Green functions in establishing uniqueness for the
Dirichlet Problem for weakly elliptic homogeneous, constant (complex) coefficient,
second-order systems, formulated in a very broad geometric setting. In particular,
in Theorem 4.1.1 and Theorem 4.1.6 we present some sharp embodiments of the
heuristic principle asserting that uniqueness holds in the Dirichlet Problem for a given
system L and domain Ω granted the existence of Green functions which “pair well”
with the null-solutions of this BVP. By not having to actually specify any concrete
function spaces in the formulation of these results substantially increases their range
of applicability; see the discussion in Examples 4.1.2-4.1.5 and Examples 4.1.7-
4.1.10.
The topic of §4.2 is the Reciprocity Principle (aka Law of Reciprocity). This
essentially asserts that Green functions satisfy an adjoint symmetry property with
respect to its pole and its actual variable (regarded as two independent arguments),

1 in the case of the Laplacian, Green introduced the function which bears his name in [53]

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 769
D. Mitrea et al., Geometric Harmonic Analysis III, Developments in Mathematics 74,
https://doi.org/10.1007/978-3-031-22735-6_4
770 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

which also takes into account the differential operators with respect to which they are
considered. This is made precise in Theorem 4.2.1, where a geometrically general
version of this Reciprocity Principle for second-order systems, also placing minimal
analytical demands on the Green functions involved, is formally stated. In this regard,
we have discovered that, in its most general format, the Reciprocity Principle actually
involves two Green functions, whose properties are somewhat reminiscent of those
exhibited by the solutions of the Dirichlet and Regularity boundary value problems
for the system L and its transpose L  .
In §4.3 we elaborate on the actual technology of constructing Green functions
in relation to a given domain Ω and second-order system L which are effective
tools in proving uniqueness results for boundary value problems for the system L
in Ω. Specifically, the main point of Theorem 4.3.1 is that the solvability of the
Regularity Problem for L  permits us to construct a Green function which then
yields uniqueness for the Dirichlet Problem for the original system L, formulated in
a dual setting. Likewise, Theorem 4.3.2 essentially asserts that uniqueness for the
Regularity Problem for a given system L is implied by the solvability of the Dirichlet
Problem for L  , formulated in a dual setting.
The main result in §4.4 is the Poisson integral representation formula from The-
orem 4.4.1, to the effect that the value of any function, which is a null-solution
of a given second-order system L in a domain Ω, at a point x0 ∈ Ω is given by
the integral pairing between the boundary trace of said function with the conormal
derivative of the Green function in Ω with pole at x0 for the system L  . This result is
formulated under very general analytic and geometric hypotheses. In particular, no
concrete function spaces are involved in the statement (which simply requires that
the aforementioned integral is absolutely convergent), which makes this theorem
widely applicable to specific cases of interest.
Finally, in §4.5-§4.7 we introduce and study the Poisson kernel associated with a
given weakly elliptic system, in various geometric settings.

4.1 The Role of Green Functions in Uniqueness Issues

Theorem 4.1.1 is a sharp embodiment of the heuristic principle asserting that unique-
ness holds in the Dirichlet Problem for a given system L and domain Ω granted the
existence of Green functions which “pair well” with the null-solutions of the re-
spective boundary value problem. The reader is reminded that, given an arbitrary
open set Ω ⊆  R along
 with some Lebesgue measurable function u defined in Ω, by
ρ
Nκ u := Nκ 1 Oρ · u we denote the truncated version of the nontangential maximal
 
operator at height ρ > 0 (cf. (A.0.92)), and by NκE u := Nκ 1E · u we denote the
restricted version of the nontangential maximal operator to a Lebesgue measurable
subset E of Ω (cf. (A.0.93)).
4.1 The Role of Green Functions in Uniqueness Issues 771

Theorem 4.1.1 Suppose Ω ⊂ Rn , where n ∈ N with2 n ≥ 3, is a locally pathwise

nontangentially accessible set3 (in the sense of [112, Definition 8.9.14]) with a lower
Ahlfors regular boundary and such that σ := H n−1 ∂Ω is a doubling measure on
∂Ω. Also, suppose L is a weakly elliptic homogeneous, constant (complex) coefficient,
second-order M × M system in Rn . Finally, fix an aperture  parameter κ > 0, an
arbitrary point x0 ∈ Ω, and a truncation parameter ρ ∈ 0 , 14 dist(x0, ∂Ω) .
Then there exists some κ > 0, which depends only on Ω and κ, with the following
significance. Assume G is a matrix-valued function satisfying4
⎧   1 (Ω, L n ) M×M ,
⎪
⎪ G = Gαβ 1≤α,β ≤M ∈ Lloc
⎪
⎪
⎪
⎪   
⎪
⎪ L G . β α = −δx0 δαβ in D (Ω) for all α, β ∈ {1, . . . , M },
⎪
⎪
⎨
⎪ 
κ −n.t.
(4.1.1)
⎪
⎪ G = 0 at σ-a.e. point on ∂Ω,
⎪
⎪ ∂Ω
⎪
⎪
⎪
⎪ and also G(x) = o(1) as |x| −→ ∞
⎪
⎪
⎪ whenever Ω is an exterior domain,
⎩
and assume u is a vector-valued function satisfying

⎧
⎪
M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω,
⎪
⎪
⎪
⎪ κ−n.t.
⎪
⎪
⎪ u ∂Ω = 0 at σ-a.e. point on ∂Ω,
⎪
⎪
⎪
⎨∫
⎪
ρ ρ (4.1.2)
⎪
⎪ Nκ u · Nκ (∇G) dσ < +∞,
⎪
⎪
⎪ ∂Ω
⎪
⎪
⎪ and also u(x) = o(1) as |x| −→ ∞
⎪
⎪
⎪
⎪ whenever Ω is an exterior domain,
⎩

(with the convention that 0 · ∞ = ∞ · 0 = ∞ used for the integrand in the third
line of (4.1.2)). In the case when ∂Ω is unbounded, strengthen the initial geometric
hypotheses on Ω and also strengthen the integral condition in the third line of (4.1.2)
by assuming that (cf. [112, (8.9.179)])
⎧
⎪ Ω is a globally pathwise nontangentially accessible set,
⎪
⎨
⎪ ∫ (4.1.3)
⎪
⎪ Nκ u · NκΩ\K (∇G) dσ < +∞ where K := B(x0, ρ),
⎪ and
⎩ ∂Ω

(once again with the convention that 0 · ∞ = ∞ · 0 = ∞).

Then necessarily

2 see the last part in the statement for the two-dimensional case
3 in particular, a one-sided NTA domain will do; cf. [112, (8.9.180)] and the subsequent comment
4 Here and elsewhere, G. β denotes the β-th column of the matrix G. As such, the second condition
in (4.1.1) may be recast as L  G = −δ x0 I M ×M where L  acts on the columns of G, and I M ×M is
the M × M identity matrix.
772 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

u(x0 ) = 0 ∈ C M . (4.1.4)
Finally, the same result is also valid in the case n = 2 provided either Ω is
bounded, or ∂Ω is unbounded. Also, if n = 2 and Ω is an exterior domain, the same
result is true provided one assumes
∫
−1
L(ξ) dH 1 (ξ)
S 1 (4.1.5)
is an invertible M × M matrix

(which is always the case if the system L is assumed to actually satisfy the Legendre-
Hadamard strong ellipticity condition, and if M = 1 this is true if and only if
L = ∇ · A∇ for some A ∈ M20 ; cf. Lemma 1.4.19 and (1.4.186)), and with the demand
that u and G are o(1) at infinity replaced by

u(x) = O(1) and G(x) = O(1) as |x| → ∞. (4.1.6)

In relation to the above theorem we wish to make several comments.

Comment 1. It is worth pointing out that, generally speaking,

if Ω ⊆ Rn is an open set, x0 ∈ Ω is an arbitrary point, L is a weakly el-

liptic homogeneous, constant coefficient, second-order M × M system
M×M
in Rn , and U ∈ D (Ω) satisfies L U = −δx0 I M×M in the (4.1.7)
sense of distributions in Ω (where L acts on the columns of U) then
M×M
U ∈ 𝒞∞ (Ω \ {x0 }) ∩ Wloc 1,1
(Ω) .

Indeed, the elliptic regularity result recorded in [112, (6.5.40) in Theorem 6.5.7]
implies that the matrix-distribution defined as W := U + E(· − x0 ), where E is
the fundamental solution associated with the given weakly elliptic system L as in
M×M
Theorem 1.4.2, satisfies W ∈ 𝒞∞ (Ω) . Thus, as claimed, the matrix-valued
M×M
function U = W − E(· − x0 ) belongs to 𝒞∞ (Ω \ {x0 }) ∩ Wloc 1,1
(Ω) .
When (4.1.7) is used in the setting of Theorem 4.1.1, for the matrix-distribution
G and the system L  , the first two lines of (4.1.1) imply that

G belongs to 𝒞∞ (Ω \ {x0 }) ∩ Wloc

1,1 M×M
(Ω) . (4.1.8)

In particular, this membership allows us to make sense of the property formulated

in the third line of (4.1.1).
 
Comment 2. We know that H n−1 ∂Ω \ ∂nta Ω = 0 (cf. [112, Proposition 8.9.16]),
which makes it meaningful to consider nontangential boundary traces σ-a.e. on ∂Ω.
Also, from [112, (8.2.28)] we see (bearing in mind the convention 0 · ∞ = ∞ · 0 = ∞)
ρ ρ
that Nκ u · Nκ (∇G) is a well-defined non-negative σ-measurable function on ∂Ω. In
particular, the integral condition formulated in the third line of (4.1.2) is meaningful
and, since any integrable function is finite almost everywhere, it implies (again,
thanks to the aforementioned convention)
4.1 The Role of Green Functions in Uniqueness Issues 773
 ρ   ρ 
Nκ u (x) < +∞ and Nκ (∇G) (x) < +∞ at σ-a.e. x ∈ ∂Ω. (4.1.9)

Likewise, Nκ u·NκΩ\K (∇G) is a well-defined non-negative σ-measurable function on

∂Ω and, in the scenario in which ∂Ω is unbounded, the integral condition formulated
in the second line of (4.1.3) is meaningful and implies
   
Nκ u (x) < +∞ and NκΩ\K (∇G) (x) < +∞ at σ-a.e. x ∈ ∂Ω. (4.1.10)

Comment 3. The demands imposed in (4.1.1) are far from ensuring that such a
function G is unique. For example, if we work in the two-dimensional setting where
we take Ω := B(0, 1) ⊆ R2 ≡ C and L := Δ, then adding5

1 + z  1 − |z| 2
w(z) := Re = , ∀z ∈ Ω, (4.1.11)
1−z |z − 1| 2
to a function G as in (4.1.1) does not change any of these properties.
We now turn to the actual proof of Theorem 4.1.1.
Proof of Theorem 4.1.1 For starters, the reader is reminded that [112, Proposi-
tion 8.9.16] ensures that we presently have
 
σ ∂Ω \ ∂nta Ω = 0, (4.1.12)
M×M
while (4.1.8) guarantees that G belongs to 𝒞∞ (Ω \ {x0 }) ∩ Wloc
1,1
(Ω) . Also,
 
recall the family of one-sided collar neighborhoods Oε ε>0 of ∂Ω defined in
(A.0.92).
To proceed in earnest, we claim that there exists some threshold εo ∈ (0, ρ),
some magnification factor λ ∈ (1, ∞), and some aperture parameter  κ ≥ κ with the
property that for each ε ∈ (0, εo /λ) we may write
∫ ∫ ∫
 
|u||G| dL n ≤ Cε Nκε |u||G| dσ ≤ Cε Nκε u Nκε G dσ
Oε ∂Ω ∂Ω
∫
≤ Cλε 2 Nκε u Nκλε (∇G) dσ
∂Ω
∫
ρ ρ
≤ Cλε 2 Nκ u Nκ (∇G) dσ < +∞, (4.1.13)
∂Ω

for some constant C = C(∂Ω, κ) ∈ (0, ∞). Indeed, the first inequality in (4.1.13)
comes from [112, Proposition 8.6.10] (which provides an initial specification of the
threshold εo ∈ (0, ρ)), the second one follows from [112, (8.2.10)], the third one is
ensured by (4.1.12) and [112, Proposition 8.9.17] (which provides the magnification
factor λ ∈ (1, ∞), the aperture parameter 
κ ≥ κ, and eventually re-sets the threshold

5 note that the function w is closely related to the Poisson kernel for the unit disk
774 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

εo to a smaller value), the fourth one is implied by [112, (8.2.25)] since we presently
have ρ > λε > ε, and the last one is guaranteed by hypotheses.
With C ∈ (0, ∞) of the same nature as above, for each truncation parameter
ε ∈ (0, εo ) we also have
∫ ∫ ∫
 
|u||∇G| dL n ≤ Cε Nκε |u||∇G| dσ ≤ Cε Nκε u Nκε (∇G) dσ
Oε ∂Ω ∂Ω
∫
ρ ρ
≤ Cε Nκ u Nκ (∇G) dσ < +∞. (4.1.14)
∂Ω

Above, the first inequality is provided by [112, Proposition 8.6.10], the second
inequality is implied by [112, (8.2.10)], the third inequality is a consequence of
[112, (8.2.25)] given that κ ≥ κ and ρ > ε, while the last inequality is ensured by
hypotheses. Collectively, (4.1.13) and (4.1.14) prove that
∫
 
|u||G| + |u||∇G| dL n < +∞ for each ε ∈ (0, εo ). (4.1.15)
Oε

With the goal of showing that u(x0 ) = 0, bring in the family of functions {Φε }ε>0
constructed
 αβ  in [112, Lemma 6.1.2]. In addition, select a complex coefficient tensor
A = ar s 1≤r,s ≤n which permits us to express L = L A, where L A is the homoge-
1≤α,β ≤M
neous constant coefficient second-order M × M system in Rn associated with A as
in (1.7.7). Finally, denote by (uβ )1≤β ≤M the scalar components of the vector-valued
function u. Then fix γ ∈ {1, . . . , M } along with ε ∈ (0, εo ) and define the vector
field (using the summation convention over repeated indices)

αβ βα αβ
F := Φε Gαγ a jk ∂k uβ − Φε uα ak j ∂k G βγ − uβ Gαγ ak j ∂k Φε in Ω.
1≤ j ≤n
(4.1.16)
From (4.1.16), (4.1.8), and assumptions it follows that

F ∈ Lloc
n
1
(Ω, L n ) . (4.1.17)

Also, since Φε ≡ 0 in the one-sided collar neighborhood Oε/N of ∂Ω (cf. [112,

(6.1.5)]), from (4.1.16) we see that

F vanishes identically in Oε/N thus, in particular,

κ−n.t. (4.1.18)
Nκε/N F = 0 everywhere on ∂Ω, and F ∂Ω = 0 everywhere on ∂nta Ω.

Next, a direct calculation shows that divF (considered in the sense of distributions
in Ω) is given by
4.1 The Role of Green Functions in Uniqueness Issues 775
αβ αβ
divF = (∂j Φε )Gαγ a jk (∂k uβ ) + Φε (∂j Gαγ )a jk (∂k uβ )
αβ βα
+ Φε Gαγ a jk (∂j ∂k uβ ) − (∂j Φε )uα ak j (∂k G βγ )
βα βα
− Φε (∂j uα )ak j (∂k G βγ ) − Φε uα ak j (∂j ∂k G βγ )
αβ αβ
− (∂j uβ )Gαγ ak j (∂k Φε ) − uβ (∂j Gαγ )ak j (∂k Φε )
αβ
− uβ Gαγ ak j (∂j ∂k Φε )

=: I1 + I2 + I3 + I4 + I5 + I6 + I7 + I8 + I9, (4.1.19)

where the last equality defines the Ii ’s with i ∈ {1, . . . , 9}. Let us analyze some of
these terms. Changing variables j = k and k = j in I1 yields
αβ
I1 = (∂k Φε ) Gαγ ak j
(∂j uβ ) = −I7 . (4.1.20)

For I2 we change variables j = k, k = j, α = β, β = α in order to write

βα
I2 = Φε (∂k G β γ ) ak j
(∂j uα ) = −I5 . (4.1.21)

As regards I3 , we have

I3 = Φε Gαγ (L Au)α = Φε Gαγ (Lu)α = 0, (4.1.22)

by the assumptions on u. For I6 we observe that (with G . γ := (G μγ )1≤μ ≤M denoting

the γ-th column of the matrix G)

I6 = −Φε uα (L A G . γ )α = −Φε uα (L  G . γ )α

= Φε uα δαγ δx0 = Φε uγ δx0 = uγ (x0 )δx0 , (4.1.23)

thanks to the first condition in (4.1.1). Collectively, these equalities permit us to

conclude that, in the sense of distributions in Ω,
βα
divF = uγ (x0 )δx0 − (∂j Φε ) uα ak j (∂k G βγ )
αβ αβ
− uβ (∂j Gαγ )ak j (∂k Φε ) − uβ Gαγ ak j (∂j ∂k Φε ). (4.1.24)

Note that the first term in the right-hand side is a distribution supported at the
singleton {x0 }, while the remaining terms are in L 1 (Ω, L n ), as seen from (4.1.15).
As such,

divF ∈ ℰ (Ω) + L 1 (Ω, L n ). (4.1.25)

Let us now focus on the case when ∂Ω is bounded and n ≥ 3. When Ω is an

exterior domain, from assumptions and Theorem 1.5.9 (applied both to L  and L)
we conclude that as |x| → ∞ we have
776 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

G(x) = O(|x| 2−n ), (∇G)(x) = O(|x| 1−n ),

(4.1.26)
u(x) = O(|x| 2−n ), (∇u)(x) = O(|x| 1−n ).


In concert with (4.1.16) this implies that F(x) = O(|x| 3−2n ) as |x| → ∞. Hence,
since n ≥ 3, for each μ ∈ (1, ∞) we have
∫

| F(x)| dL n (x) = o(R) as R → ∞ (4.1.27)
B(0, μR)\B(0,R)

which, in turn, shows that condition [112, (1.5.22)] presently holds, if Ω is an

exterior domain. Having established (4.1.17), (4.1.18), (4.1.25), and (4.1.27), [112,
Corollary 1.5.2] in the version recorded in [112, (1.5.23)] applies and, if ν denotes
the geometric measure theoretic outward unit normal to Ω, the Divergence Formula
[112, (1.5.20)] gives, in view of (4.1.24) and [112, (4.6.19)], that for each ε > 0
small enough we have
∫
 κ−n.t.   
0= ν · F ∂Ω dσ = (𝒞∞ (Ω))∗ divF,  1 𝒞∞ (Ω)
b
∂∗ Ω b

∫
βα
= uγ (x0 ) − (∂j Φε )uα ak j (∂k G βγ ) dL n (4.1.28)
Ω
∫ ∫
αβ αβ
− uβ (∂j Gαγ )ak j (∂k Φε ) dL n − uβ Gαγ ak j (∂j ∂k Φε ) dL n .
Ω Ω

Based on (4.1.28), [112, (6.1.5)-(6.1.6)], and re-using portions of (4.1.14)-(4.1.13),

we may then estimate (also bearing in mind [112, (8.2.25)] as well as  κ ≥ κ and
ρ > λε > ε)
∫
 −2 
|uγ (x0 )| ≤ C ε |u||G| + ε −1 |u||∇G| dL n
Oε
∫ ∫
≤C Nκε u · Nκε (∇G) dσ + C Nκε u · Nκλε (∇G) dσ
∂Ω ∂Ω
∫
ρ
≤C Nκε u · Nκ (∇G) dσ for each ε ∈ (0, εo ). (4.1.29)
∂Ω

There remains to observe that, thanks to Lebesgue’s Dominated Convergence The-

orem we have
∫
ρ
lim+ Nκε u · Nκ (∇G) dσ = 0. (4.1.30)
ε→0 ∂Ω

Indeed, from [112, (8.9.13)], (A.0.111), (4.1.12), and the hypothesis made in the
second line of (4.1.2) we see that
4.1 The Role of Green Functions in Uniqueness Issues 777

lim Nκε u = 0 at σ-a.e. point on ∂Ω. (4.1.31)

ε→0+

Together with the finiteness condition in the third line of (4.1.1) this proves that
ρ
lim Nκε u · Nκ (∇G) = 0 at σ-a.e. point on ∂Ω. (4.1.32)
ε→0+

Since for each ε ∈ (0, εo ) ⊂ (0, ρ) we also have

ρ ρ ρ
0 ≤ Nκε u · Nκ (∇G) ≤ Nκ u · Nκ (∇G) ∈ L 1 (∂Ω, σ), (4.1.33)

by virtue of [112, (8.2.25)] and the integrability condition in the third line of (4.1.2),
Lebesgue’s Dominated Convergence Theorem applies and gives (4.1.30). With this
in hand, upon letting ε → 0+ in (4.1.29) we conclude that uγ (x0 ) = 0. Since
γ ∈ {1, . . . , M } has been arbitrarily chosen, this establishes (4.1.4) in the case when
∂Ω is bounded and n ≥ 3. In fact, the same argument works in the case when Ω is
bounded and n ≥ 2, since the decay at infinity is no longer a concern.
There remains to consider the case when
either ∂Ω is unbounded and n ≥ 2,
(4.1.34)
or Ω is an exterior domain and n = 2.

Consider first the case when ∂Ω is unbounded and n ≥ 2, a scenario in which the
stronger hypotheses made in (4.1.3) take effect. We reason as before, with some
important alterations. Now the idea is to use the Divergence Formula [112, (1.5.20)
in Corollary 1.5.2], and this requires that we verify the existence of an aperture
parameter κ ∈ (0, ∞) and a compact set K ⊆ Ω with the property that (cf. [112,
(1.5.21)])
NκΩ\K F ∈ L 1 (∂Ω, σ). (4.1.35)
We claim that (4.1.35) holds provided

κ ∈ (0, κ) and K := B(x0, 3ρ). (4.1.36)

To justify this claim, first observe from (4.1.16) that there exists a constant C ∈ (0, ∞)
independent of ε > 0 such that

| F | ≤ C1 {δ∂Ω ≥ε/N } |G| |∇u| + C1 {δ∂Ω ≥ε/N } |u| |∇G|

+ Cε −1 1 {ε/N ≤δ∂Ω ≤ε } |u||G| in Ω. (4.1.37)

Note that at σ-a.e. point on ∂Ω we have

778 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems
   
NκΩ\K |G| |∇u| = NκΩ\K (|G|/δ∂Ω ) · (δ∂Ω |∇u|)
   
≤ NκΩ\K G/δ∂Ω · NκΩ\K δ∂Ω |∇u|

   
≤ NκΩ\K G/δ∂Ω · Nκ δ∂Ω |∇u|
  
≤ C Nκ u NκΩ\K (∇G) , (4.1.38)

where we have used [112, (8.2.10), (8.2.25), (8.9.147)] (with |α| = 1 and λ = 0),
[112, (8.9.227)] (with u := G and with the understanding that  κ has been taken,
to begin with, greater than the aperture parameter κo appearing in the statement of
[112, Proposition 8.9.21]), and (4.1.12). Also, at σ-a.e. point on ∂Ω we have
   
NκΩ\K ε −1 1 {ε/N ≤δ∂Ω ≤ε } |u||G| ≤ NκΩ\K |u|(|G|/δ∂Ω )
  
≤ NκΩ\K u NκΩ\K (G/δ∂Ω )
  
≤ C Nκ u NκΩ\K (∇G) , (4.1.39)

thanks to [112, (8.2.10), (8.2.25)], as well as [112, (8.9.227)] (with u := G and

the same convention as above regarding the size of κ ), and (4.1.12). From (4.1.37),
(4.1.38), and (4.1.39) we then deduce that

NκΩ\K F ≤ C Nκ u · NκΩ\K (∇G) on ∂Ω. (4.1.40)

Based on (4.1.40), (4.1.3), and [112, (8.2.28)] we finally conclude that the member-
ship in (4.1.35) is indeed valid.
At last, we are left with dealing with the second case in (4.1.34), i.e., when Ω is an
exterior domain and n = 2. In such a scenario, the additional assumptions made in
(4.1.5) and (4.1.6) come in effect. The argument in the first part of the proof applies
and yields the desired conclusion as soon as we check (4.1.27) in the present setting.
To this end, note that the last part in Theorem 1.5.9 (applied both to L  and L) shows
that as |x| → ∞ we have

G(x) = O(1), (∇G)(x) = O(|x| −2 ),

(4.1.41)
u(x) = O(1), (∇u)(x) = O(|x| −2 ).


Together with (4.1.16) this implies that F(x) = O(|x| −2 ) as |x| → ∞, hence for each
μ ∈ (1, ∞) we have
∫

| F(x)| dL 2 (x) = o(R) as R → ∞, (4.1.42)
B(0, μR)\B(0,R)

as wanted. This finishes the proof of Theorem 4.1.1. 

4.1 The Role of Green Functions in Uniqueness Issues 779

Remarkably, the only assumptions of quantitative nature made6 in the statement

of Theorem 4.1.1 are the pointwise a.e. finiteness condition in the third line of
(4.1.1), the integrability condition in the third line of (4.1.2), as well as the pointwise
finiteness conditions and integrability property stipulated in (4.1.3). These demands
are naturally satisfied in many cases of practical interest, like the ones on which we
elaborate in next series of examples.

Example 4.1.2 (Ordinary Lebesgue Spaces) In Theorem 4.1.1, the uniqueness re-
sult remains valid when the third line in (4.1.2) is replaced by
ρ
Nκ u ∈ L p (∂Ω, σ) and Nκ (∇G) ∈ L p (∂Ω, σ),
(4.1.43)
where p, p ∈ [1, ∞] are such that 1/p + 1/p = 1,

and, in the case when ∂Ω is unbounded, the integrability condition in the second
line of (4.1.3) is replaced by

Nκ u ∈ L p (∂Ω, σ) and NκΩ\K (∇G) ∈ L p (∂Ω, σ),

(4.1.44)
where p, p ∈ [1, ∞] are such that 1/p + 1/p = 1.
ρ
Indeed, [112, Proposition 8.4.1] shows that Nκ (∇G) ∈ L p (∂Ω, σ) implies
ρ
Nκ (∇G) ∈ L p (∂Ω, σ) for any other 
κ > 0, so the integrability condition in the
third line of (4.1.2) becomes a consequence of Hölder’s inequality. Also, thanks to
[112, Corollary 8.9.9] and the last condition hypothesized in (4.1.43), in the third
κ−n.t.
line of (4.1.1) we may simply ask that G ∂Ω = 0 at σ-a.e. point on ∂Ω. Similar
comments apply in relation to (4.1.44).

Example 4.1.3 (Muckenhoupt Weighted Lebesgue Spaces) The uniqueness result

formulated in Theorem 4.1.1 remains valid when the third line in (4.1.2) is replaced
by
ρ
Nκ u ∈ L p (∂Ω, wσ) and Nκ (∇G) ∈ L p (∂Ω, w 1−p σ) where
(4.1.45)
p, p ∈ (1, ∞) are such that 1/p + 1/p = 1, and w ∈ Ap (∂Ω, σ),

and, in the case when ∂Ω is unbounded, the integrability condition in the second
line of (4.1.3) is replaced by

Nκ u ∈ L p (∂Ω, wσ) and NκΩ\K (∇G) ∈ L p (∂Ω, w 1−p σ) where

(4.1.46)
p, p ∈ (1, ∞) are such that 1/p + 1/p = 1, and w ∈ Ap (∂Ω, σ).
ρ
This time we rely on [112, Corollary 8.4.4], which ensures that having Nκ (∇G)
ρ
in L p (∂Ω, w 1−p σ) implies Nκ (∇G) ∈ L p (∂Ω, w 1−p σ) for any other  κ > 0,
since
 p w 1−p ∈ A (∂Ω, σ) (cf. item (2) in [112, Lemma 7.7.1]), and the fact that
∗ p
L (∂Ω, wσ) is canonically identified with L p (∂Ω, w 1−p σ) under the integral

6 in addition to specifying the behavior at infinity, in the case when Ω is an exterior domain
780 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

pairing on ∂Ω. In addition, as seen from [112, Corollary 8.9.9] and the last con-
dition hypothesized in (4.1.45), in the third line of (4.1.1) we may simply ask
κ−n.t.
that G ∂Ω = 0 at σ-a.e. point on ∂Ω. Similar comments apply in relation to
(4.1.46). It is worth noting that the only relevant property of a Muckenhoupt weight
for the uniqueness result just described is the fact that the measures w σ and w 1−p σ
are doubling on ∂Ω.

Example 4.1.4 (Lorentz Spaces) The uniqueness result in Theorem 4.1.1 remains
valid when the third line in (4.1.2) is replaced by
ρ
Nκ u ∈ L p,q (∂Ω, σ) and Nκ (∇G) ∈ L p ,q (∂Ω, σ) where
(4.1.47)
p, q, p , q ∈ [1, ∞] satisfy 1/p + 1/p = 1 = 1/q + 1/q ,

and, in the case when ∂Ω is unbounded, the integrability condition in the second
line of (4.1.3) is replaced by

Nκ u ∈ L p,q (∂Ω, σ) and NκΩ\K (∇G) ∈ L p ,q (∂Ω, σ) where

(4.1.48)
p, q, p , q ∈ [1, ∞] satisfy 1/p + 1/p = 1 = 1/q + 1/q ,
ρ
We now bring in [112, Proposition 8.4.1], which shows that having Nκ (∇G) in
ρ
L p ,q (∂Ω, σ) implies Nκ (∇G) ∈ L p ,q (∂Ω, σ) for any other  κ > 0, and O’Neil’s
inequality (cf. [112, (6.2.61)]). Once again, [112, Corollary 8.9.9] and the last con-
dition demanded in (4.1.47) show that in the third line of (4.1.1) we may simply
κ−n.t.
ask that G ∂Ω = 0 at σ-a.e. point on ∂Ω. Similar comments are valid in relation to
(4.1.48).
Here is a final example, whose justification now involves [113, Proposition 6.2.8,
(6.2.78)] and [112, Corollary 8.4.8] (see also [112, Corollary 8.4.5]).

Example 4.1.5 (Morrey Spaces and Their Pre-Duals) In the case when ∂Ω is as-
sumed to be an Ahlfors regular set, the uniqueness result in Theorem 4.1.1 remains
valid when the third line in (4.1.1) is replaced by
κ−n.t.
G ∂Ω = 0 at σ-a.e. point on ∂Ω, (4.1.49)

and the third line in (4.1.2) is replaced by

ρ
Nκ u ∈ B p,λ (∂Ω, σ) and Nκ (∇G) ∈ M p ,λ (∂Ω, σ) where
(4.1.50)
p, p ∈ (1, ∞) are such that 1/p + 1/p = 1 and λ ∈ (0, n − 1).

Also, in the case when the set ∂Ω happens to be unbounded, one may replace (4.1.3)
by

Nκ u ∈ B p,λ (∂Ω, σ) and NκΩ\K (∇G) ∈ M p ,λ (∂Ω, σ) where

(4.1.51)
p, p ∈ (1, ∞) are such that 1/p + 1/p = 1 and λ ∈ (0, n − 1).
4.1 The Role of Green Functions in Uniqueness Issues 781

Alternatively, retaining the assumption that ∂Ω is an Ahlfors regular set, the

uniqueness result in Theorem 4.1.1 continues to hold when the third line in (4.1.2)
is replaced by
ρ
Nκ u ∈ M p,λ (∂Ω, σ) and Nκ (∇G) ∈ B p ,λ (∂Ω, σ) where
(4.1.52)
p, p ∈ (1, ∞) are such that 1/p + 1/p = 1 and λ ∈ (0, n − 1),

and, in the case when the set ∂Ω is unbounded, with (4.1.3) replaced by

Nκ u ∈ M p,λ (∂Ω, σ) and NκΩ\K (∇G) ∈ B p ,λ (∂Ω, σ) where

(4.1.53)
p, p ∈ (1, ∞) are such that 1/p + 1/p = 1 and λ ∈ (0, n − 1).

Moving on, we present a companion result to Theorem 4.1.1, pertaining to unique-

ness for the Dirichlet Problem for second-order systems in which an integrability
condition is imposed on the gradient of the solution, with a weight involving a Green
function.

Theorem 4.1.6 Fix n ∈ N with7 n ≥ 3. Suppose Ω ⊂ Rn is a locally pathwise

nontangentially accessible set8 (in the sense of [112, Definition 8.9.14]) with a lower
Ahlfors regular boundary and such that σ := H n−1 ∂Ω is a doubling measure on
∂Ω. Also, suppose L is a weakly elliptic homogeneous, constant (complex) coefficient,
second-order M × M system in Rn . Finally, fix some point x0 ∈ Ω, a truncation
parameter ρ ∈ 0 , 14 dist(x0, ∂Ω) , and an aperture parameter κ ∈ (0, ∞).
Then there exists some κ > 0, which depends only on Ω and κ, with the following
significance. Suppose G is a matrix-valued function satisfying
⎧  
⎪
⎪ G = Gαβ 1≤α,β ≤M ∈ Lloc 1 (Ω, L n ) M×M ,
⎪
⎪
⎪
⎪   
⎪
⎪ L G . β α = −δx0 δαβ in D (Ω) for all α, β ∈ {1, . . . , M },
⎪
⎪
⎨
⎪ κ−n.t.

= 0 at σ-a.e. point on ∂Ω, (4.1.54)

⎪
⎪ G
⎪
⎪ ∂Ω
⎪
⎪ and also G(x) = o(1) as |x| −→ ∞
⎪
⎪
⎪
⎪
⎪ whenever Ω is an exterior domain,
⎩
and suppose u is a vector-valued function satisfying

7 see the last part in the statement for the case n = 2

8 in particular, this is the case if Ω is a one-sided NTA domain (see [112, (8.9.180)] and the
subsequent comment)
782 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

⎧
⎪
M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω,
⎪
⎪
⎪
⎪ κ −n.t.
⎪
⎪
⎪
⎪ u ∂Ω = 0 at σ-a.e. point on ∂Ω,
⎪
⎪
⎨∫
⎪
ρ ρ (4.1.55)
⎪
⎪ Nκ (∇u) · Nκ G dσ < +∞,
⎪
⎪
⎪ ∂Ω
⎪
⎪
⎪ and also u(x) = o(1) as |x| −→ ∞
⎪
⎪
⎪
⎪ whenever Ω is an exterior domain,
⎩
(with the convention that 0 · ∞ = ∞ · 0 = ∞ used for the integrand in the third line
of (4.1.55)). In the case when ∂Ω is unbounded, strengthen the initial geometric
hypotheses on Ω and also strengthen the integral condition in the third line of
(4.1.55) by assuming that (cf. [112, (8.9.179)])
⎧
⎪ Ω is a globally pathwise nontangentially accessible set,
⎪
⎨
⎪ ∫ (4.1.56)
⎪
⎪ Nκ (∇u) · NκΩ\K (∇G) dσ < +∞ where K := B(x0, ρ)
⎪ and
⎩ ∂Ω

(once again with the convention that 0 · ∞ = ∞ · 0 = ∞).

Then necessarily
u(x0 ) = 0 ∈ C M . (4.1.57)
Moreover, the same result is also valid in the case n = 2 provided either Ω is
bounded, or ∂Ω is unbounded. Also, if n = 2 and Ω is an exterior domain, the same
result is true provided one assumes
∫
−1
L(ξ) dH 1 (ξ)
S 1 (4.1.58)
is an invertible M × M matrix

(which is always the case if the system L is assumed to actually satisfy the Legendre-
Hadamard strong ellipticity condition, and if M = 1 this is true if and only if
L = ∇ · A∇ for some A ∈ M20 ; cf. Lemma 1.4.19 and (1.4.186)), and with the demand
that u and G are o(1) at infinity replaced by

u(x) = O(1) and G(x) = O(1) as |x| → ∞. (4.1.59)

Proof The main idea is to re-run the proof of Theorem 4.1.1, with the roles of u and
G reversed, and with the coefficient tensor A now replaced by A . To elaborate on
these adjustments, denote by (uβ )1≤β ≤M the scalar components of the vector-valued
function u. Then, in place of (4.1.14), for each ε > 0 small we now have
∫ ∫
|G||∇u| dL n ≤ Cε Nκε G · Nκε (∇u) dσ < +∞, (4.1.60)
Oε ∂Ω

while in place of (4.1.13) we now have

4.1 The Role of Green Functions in Uniqueness Issues 783
∫ ∫
|u||G| dL n ≤ Cλε 2 Nκε G · Nκλε (∇u) dσ < +∞. (4.1.61)
Oε ∂Ω

In particular,
∫
 
|u||G| + |G||∇u| dL n < +∞ for each ε > 0 small. (4.1.62)
Oε

Let us also observe that (4.1.7) presently implies that actually

G ∈ 𝒞∞ (Ω \ {x0 }) ∩ Wloc
1,1 M×M
(Ω) . (4.1.63)

To proceed, fix γ ∈ {1, . . . , M } along with ε > 0 sufficiently small and recall

the family of functions {Φε }ε>0 introduced in [112, Lemma 6.1.2]. Using the
summation convention over repeated indices, we now define the vector field

βα αβ βα
F := Φε uα ak j ∂k G βγ − Φε Gαγ a jk ∂k uβ − G βγ uα a jk ∂k Φε . (4.1.64)
1≤ j ≤n

From (4.1.64), (4.1.63), and assumptions we have

F belongs to Lloc and F ≡ 0 in Oε/N thus, in particular,

1 (Ω, L n ) n

κ−n.t. (4.1.65)
Nκε/N F = 0 everywhere on ∂Ω and F ∂Ω = 0 everywhere on ∂nta Ω.

Moreover, div F considered in the sense of distributions in Ω is now given by

αβ
divF = −uγ (x0 )δx0 − Gαγ (∂k uβ )a jk (∂j Φε )
βα βα
− G βγ (∂j uα )a jk (∂k Φε ) − G βγ uα a jk (∂j ∂k Φε ). (4.1.66)

Hence, once again, divF ∈ ℰ (Ω) + L 1 (Ω, L n ), thanks to (4.1.62).

In addition, when Ω is an exterior domain the current assumptions and Theo-
rem 1.5.9 (applied both to L  and L) imply the decay properties recorded in (4.1.26)
(cf. also (4.1.42)) as |x| → ∞. In turn, much as in (4.1.27), we ultimately conclude
that condition [112, (1.5.13)] presently holds, if Ω is an exterior domain. Finally, in
the case when ∂Ω is unbounded (a scenario in which (4.1.56) comes into effect), we
claim that if
κ ∈ (0, κ) and K := B(x0, 3ρ) (4.1.67)
then there exist some aperture parameter κ > 0 and some compact set K ⊆ Ω such
that
NκΩ\K F ∈ L 1 (∂Ω, σ). (4.1.68)
To prove this claim, first note from (4.1.64) and [112, Lemma 6.1.2] that there exists
a constant C ∈ (0, ∞) independent of ε > 0 such that
784 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

| F | ≤ C1 {δ∂Ω ≥ε/N } |u| |∇G| + C1 {δ∂Ω ≥ε/N } |G| |∇u|

+ Cε −1 1 {ε/N ≤δ∂Ω ≤ε } |u||G| in Ω. (4.1.69)

In relation to this, we observe that at σ-a.e. point on ∂Ω we have

   
NκΩ\K |u| |∇G| = NκΩ\K (|u|/δ∂Ω ) · (δ∂Ω |∇G|)
   
≤ NκΩ\K u/δ∂Ω · NκΩ\K δ∂Ω |∇G|

   
≤ Nκ u/δ∂Ω · NκΩ\K δ∂Ω |∇G|
  
≤ C Nκ (∇u) NκΩ\K G , (4.1.70)

thanks to [112, (8.2.10), (8.2.25)], as well as [112, (8.9.167)] (with u := G, |α| = 1,

and λ = 0), [112, (8.9.210)] (with the understanding that the current  κ has been
taken, from the beginning, to be larger than the aperture parameter
 
κ appearing
 in
the statement of [112, Proposition 8.9.18]), and the fact that σ ∂Ω \ ∂nta Ω = 0 (cf.
[112, Proposition 8.9.16]). In addition, at σ-a.e. point on ∂Ω we have
   
NκΩ\K ε −1 1 {ε/N ≤δ∂Ω ≤ε } |u||G| ≤ NκΩ\K |G|(|u|/δ∂Ω )
  
≤ NκΩ\K G Nκ (u/δ∂Ω )
  
≤ C NκΩ\K G Nκ (∇u) , (4.1.71)

where we used [112, (8.2.10)], [112, (8.2.25)], [112, (8.9.227)] (with the same
convention as above regarding the size of κ ), and the fact that σ ∂Ω \ ∂nta Ω = 0.
From (4.1.37), (4.1.38), and (4.1.39) we then conclude that

NκΩ\K F ≤ C NκΩ\K G · Nκ (∇u) on ∂Ω. (4.1.72)

Collectively, (4.1.72), (4.1.56), and [112, (8.2.28)] permit us to conclude that the
membership claimed in (4.1.68) is indeed true.
Granted the aforementioned properties of the vector field F,  [112, Corollary 1.5.2]
(in the version recorded in [112, (1.5.23)]) applies and, with ν denoting the geometric
measure theoretic outward unit normal to Ω, the Divergence Formula [112, (1.5.20)]
gives that for each ε > 0 small enough we have
4.1 The Role of Green Functions in Uniqueness Issues 785
∫
 κ−n.t.   
0= ν · F ∂Ω
 1 𝒞∞ (Ω)
dσ = (𝒞∞ (Ω))∗ divF, b
∂∗ Ω b

∫
αβ
= −uγ (x0 ) − Gαγ (∂k uβ )a jk (∂j Φε ) dL n (4.1.73)
Ω
∫ ∫
βα βα
− G βγ (∂j uα )a jk (∂k Φε ) dL n − G βγ uα a jk (∂j ∂k Φε ) dL n .
Ω Ω

Thus,
∫ ∫
αβ βα
uγ (x0 ) = − Gαγ (∂k uβ )a jk (∂j Φε ) dL n − G βγ (∂j uα )a jk (∂k Φε ) dL n
Ω Ω
∫
βα
− G βγ uα a jk (∂j ∂k Φε ) dL n (4.1.74)
Ω

which, in concert with [112, (6.1.5)-(6.1.6)] and (4.1.60)-(4.1.61), allows us to

estimate
∫
 −2 
|uγ (x0 )| ≤ C ε |u||G| + ε −1 |G||∇u| dL n
Oε
∫
ρ
≤C Nκε G · Nκ (∇u) dσ for each ε > 0 small enough. (4.1.75)
∂Ω

On the other hand, from [112, (8.9.13)], (A.0.111), (4.1.12), and the assumption
made in the third line of (4.1.54) we see that

lim Nκε G = 0 at σ-a.e. point on ∂Ω. (4.1.76)

ε→0+

In concert with the finiteness condition in the third line of (4.1.55) this gives
ρ
lim Nκε G · Nκ (∇u) = 0 at σ-a.e. point on ∂Ω. (4.1.77)
ε→0+

Given that for each ε > 0 small we also have

ρ ρ ρ
0 ≤ Nκε G · Nκ (∇u) ≤ Nκ G · Nκ (∇u) ∈ L 1 (∂Ω, σ), (4.1.78)

by virtue of [112, (8.2.25)] and the integrability condition in the fourth line of
(4.1.55), Lebesgue’s Dominated Convergence Theorem applies and proves that
∫
ρ
lim+ Nκε G · Nκ (∇u) dσ = 0. (4.1.79)
ε→0 ∂Ω

Having established (4.1.79), upon letting ε → 0+ in (4.1.75) we conclude that

uγ (x0 ) = 0. Since γ ∈ {1, . . . , M } has been arbitrarily chosen, the desired conclusion
follows. 
786 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

As in the case of Theorem 4.1.1, the only assumptions of quantitative nature made
in the formulation of Theorem 4.1.6 are the pointwise a.e. finiteness conditions and
the integrability conditions made in the third and fourth lines of (4.1.55), as well as in
(4.1.56). Once again, these demands are naturally satisfied in many cases of practical
interest. For instance, with the same type of justifications as in Examples 4.1.2-4.1.5
we now have:
Example 4.1.7 (Ordinary Lebesgue Spaces) In Theorem 4.1.6, the uniqueness re-
sult remains valid when the second and third lines in (4.1.55) are replaced by
κ−n.t.
u ∂Ω = 0 at σ-a.e. point on ∂Ω,
ρ
Nκ (∇u) ∈ L p (∂Ω, σ) and Nκ G ∈ L p (∂Ω, σ), (4.1.80)
where p, p ∈ [1, ∞] are such that 1/p + 1/p = 1,

and, in the case when ∂Ω is unbounded, the finiteness condition in (4.1.56) is

replaced by

Nκ u ∈ L p (∂Ω, σ) and NκΩ\K G ∈ L p (∂Ω, σ)

(4.1.81)
where p, p ∈ [1, ∞] are such that 1/p + 1/p = 1.

Example 4.1.8 (Muckenhoupt Weighted Lebesgue Spaces) The uniqueness result

formulated in Theorem 4.1.6 remains valid when the second and third lines in (4.1.55)
are replaced by
κ−n.t.
u ∂Ω = 0 at σ-a.e. point on ∂Ω,
ρ
Nκ (∇u) ∈ L p (∂Ω, wσ) and Nκ G ∈ L p (∂Ω, w 1−p σ) where (4.1.82)
p, p ∈ (1, ∞) are such that 1/p + 1/p = 1, and w ∈ Ap (∂Ω, σ),

and, in the case when ∂Ω is unbounded, the finiteness condition in (4.1.56) is

replaced by

Nκ (∇u) ∈ L p (∂Ω, wσ) and NκΩ\K G ∈ L p (∂Ω, w 1−p σ) where

(4.1.83)
p, p ∈ (1, ∞) are such that 1/p + 1/p = 1, and w ∈ Ap (∂Ω, σ).

Example 4.1.9 (Lorentz Spaces) The uniqueness result in Theorem 4.1.6 remains
valid when the second and third lines in (4.1.55) are replaced by
κ−n.t.
u ∂Ω = 0 at σ-a.e. point on ∂Ω,
ρ
Nκ (∇u) ∈ L p,q (∂Ω, σ) and Nκ G ∈ L p , q (∂Ω, σ) where (4.1.84)
p, q, p , q ∈ [1, ∞] satisfy 1/p + 1/p = 1 = 1/q + 1/q ,

and, in the case when ∂Ω is unbounded, the finiteness condition in (4.1.56) is

replaced by
4.2 The Reciprocity Principle 787

Nκ (∇u) ∈ L p,q (∂Ω, σ) and NκΩ\K G ∈ L p , q (∂Ω, σ) where

(4.1.85)
p, q, p , q ∈ [1, ∞] satisfy 1/p + 1/p = 1 = 1/q + 1/q .

Example 4.1.10 (Morrey Spaces and Their Pre-Duals) In the case when ∂Ω is
assumed to be an Ahlfors regular set, the uniqueness result in Theorem 4.1.6 remains
valid when the second and third lines in (4.1.55) are replaced by
κ−n.t.
u ∂Ω = 0 at σ-a.e. point on ∂Ω,
ρ
Nκ (∇u) ∈ M p,λ (∂Ω, σ) and Nκ G ∈ B p ,λ (∂Ω, σ) where (4.1.86)
p, p ∈ (1, ∞) are such that 1/p + 1/p = 1, and λ ∈ (0, n − 1),

and, in the case when ∂Ω is unbounded, the finiteness condition in (4.1.56) is

replaced by

Nκ (∇u) ∈ M p,λ (∂Ω, σ) and NκΩ\K G ∈ B p ,λ (∂Ω, σ) where

(4.1.87)
p, p ∈ (1, ∞) are such that 1/p + 1/p = 1, and λ ∈ (0, n − 1).

Alternatively, one may replace the second and third lines in (4.1.55) by
κ−n.t.
u ∂Ω = 0 at σ-a.e. point on ∂Ω,
ρ
Nκ (∇u) ∈ B p,λ (∂Ω, σ) and Nκ G ∈ M p ,λ (∂Ω, σ) where (4.1.88)
p, p ∈ (1, ∞) are such that 1/p + 1/p = 1, and λ ∈ (0, n − 1),

and, in the case when ∂Ω is unbounded, one may replace the finiteness condition in
(4.1.56) by

Nκ (∇u) ∈ B p,λ (∂Ω, σ) and NκΩ\K G ∈ M p ,λ (∂Ω, σ) where

(4.1.89)
p, p ∈ (1, ∞) are such that 1/p + 1/p = 1, and λ ∈ (0, n − 1).

4.2 The Reciprocity Principle

The Reciprocity Principle (aka Law of Reciprocity) roughly states that Green func-
tions satisfy an adjoint symmetry property with respect to the pole and the actual
variable (regarded as two independent arguments), which also takes into account the
differential operators with respect to which they are considered. Our next theorem
contains a geometrically general version of this Reciprocity Principle for second-
order systems, which also places minimal analytical demands on the Green functions
involved. Regarding the latter aspect, Theorem 4.2.1 brings to the forefront the fact
that, in its most general format, the Reciprocity Principle actually involves two Green
functions, whose properties are somewhat reminiscent of those exhibited by the so-
788 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

lutions of the Dirichlet and Regularity boundary value problems for the system L
and its transpose L  .

Theorem 4.2.1 Let Ω ⊂ Rn , where n ∈ N with9, be a locally pathwise nontangen-

tially accessible set10 (in the sense of [112, Definition 8.9.14]) with a lower Ahlfors
regular boundary and such that σ := H n−1 ∂Ω is a doubling measure on ∂Ω.
Also, suppose L is a weakly elliptic homogeneous, constant (complex) coefficient,
second-order M × M system in Rn . Finally, fix two points x0, x1 ∈ Ω with x0  x1 ,
an aperture parameter κ > 0, and a truncation parameter
 
0 < ρ < 14 min dist(x0, ∂Ω), dist(x1, ∂Ω) . (4.2.1)

Then there exists some  κ > 0, which depends only on Ω and κ, with the fol-
lowing significance. Suppose G ∈ Lloc 1 (Ω, L n ) M×M is a matrix-valued function

satisfying11
⎧
⎪ L  G = −δx0 I M×M in D (Ω)
M×M
,
⎪
⎪
⎪
⎪
⎪
⎪ 
κ −n.t.
⎨
⎪
G = 0 at σ-a.e. point on ∂Ω, (4.2.2)
⎪ ∂Ω
⎪
⎪
⎪
⎪ and also G(x) = o(1) as |x| −→ ∞
⎪
⎪
⎪ whenever Ω is an exterior domain,
⎩
M×M
and suppose H ∈ Lloc
1 (Ω, L n ) is a matrix-valued function satisfying12

⎧
⎪ L H = −δx1 I M×M in D (Ω)
M×M
,
⎪
⎪
⎪
⎪ κ−n.t.
⎪
⎪
⎪
⎪ = 0 at σ-a.e. point on ∂Ω,
⎪
⎪ H
⎪
⎨ ∫ ∂Ω
⎪
ρ ρ (4.2.3)
⎪
⎪ Nκ H · Nκ (∇G) dσ < +∞,
⎪
⎪
⎪
⎪ ∂Ω
⎪
⎪
⎪
⎪ and also H(x) = o(1) as |x| −→ ∞
⎪
⎪
⎪ whenever Ω is an exterior domain,
⎩
(with the convention that 0 · ∞ = ∞ · 0 = ∞ used for the integrand in the third
line of (4.2.3)). In the case when ∂Ω is unbounded, strengthen the initial geometric
hypotheses on Ω and also strengthen the integral condition in the third line of (4.2.3)
by assuming that (cf. [112, (8.9.179)])

9 see the last part in the statement for the case n = 2

10 in particular, this is the case if Ω is a one-sided NTA domain (cf. [112, (8.9.180)] and the
subsequent comment)
11 with L  acting on the columns of G
12 with L acting on the columns of H
4.2 The Reciprocity Principle 789

⎧
⎪ Ω is a globally pathwise nontangentially accessible set, and
⎪
⎨∫
⎪
(4.2.4)
⎪
⎪ NκΩ\K H · NκΩ\K (∇G) dσ < ∞ where K := B(x0, ρ) ∪ B(x1, ρ)
⎪
⎩ ∂Ω

(once again with the convention that 0 · ∞ = ∞ · 0 = ∞).

Then necessarily
G(x1 ) = H(x0 ) . (4.2.5)
Moreover, the same result continues to be valid in the case when n = 2 provided
either Ω is bounded, or ∂Ω is unbounded. Also, if n = 2 and Ω is an exterior domain,
the same result is true provided one assumes
∫
−1
L(ξ) dH 1 (ξ)
S1 (4.2.6)
is an invertible M × M matrix

(which is always the case if the system L is assumed to actually satisfy the Legendre-
Hadamard strong ellipticity condition, and if M = 1 this is true if and only if
L = ∇ · A∇ for some A ∈ M20 ; cf. Lemma 1.4.19 and (1.4.186)), and with the demand
that G and H are o(1) at infinity replaced by

G(x) = O(1) and H(x) = O(1) as |x| → ∞. (4.2.7)

Before presenting the proof of this theorem we make several remarks.

Remark 1. If, in the notation for Green functions G, H, we also include the depen-
dence on the pole as a second variable, as well as the dependence on the system as
a superscript, then the Reciprocity Relation recorded in (4.2.5) takes a more precise
form, namely the transposition identity

G L (x1, x0 ) = [H L (x0, x1 )] . (4.2.8)

Remark  2. It is known from [112, Proposition 8.9.16] that we presently have

H n−1 ∂Ω \ ∂nta Ω = 0, so it is meaningful to consider nontangential boundary
traces σ-a.e. on ∂Ω. In addition, in view of [112, (8.2.28)] and the convention that
ρ ρ
0 · ∞ = ∞ · 0 = ∞, it follows that Nκ H · Nκ (∇G) is a well-defined non-negative
σ-measurable function on ∂Ω. As such, the integral condition formulated in the
third line of (4.2.3) is meaningful and, since any integrable function is finite almost
everywhere, it implies (again, thanks to the aforementioned convention)
 ρ   ρ 
Nκ H (x) < +∞ and Nκ (∇G) (x) < +∞ at σ-a.e. x ∈ ∂Ω. (4.2.9)

Similarly, NκΩ\K H · NκΩ\K (∇G) is a well-defined non-negative σ-measurable func-

tion on ∂Ω and, in the case when ∂Ω is unbounded, the finiteness condition in (4.2.4)
is meaningful and implies
   
NκΩ\K H (x) < +∞ and NκΩ\K (∇G) (x) < +∞ at σ-a.e. x ∈ ∂Ω. (4.2.10)
790 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

Remark 3. The integral condition imposed in the third line of (4.2.3), and the
integral condition in (4.2.4) if ∂Ω is unbounded, are the only demands linking G
and H directly. It is remarkable that when G, H are quantitatively compatible in
this specific fashion we can conclude that G, H are algebraically compatible, in the
sense of (4.2.5). Indeed, without imposing said integral conditions there could be
infinitely many pairs of (generally completely unrelated) functions G, H satisfying
the remaining conditions in (4.2.2)-(4.2.3)13. On the other hand, insisting on the
full set of requirements listed in (4.2.2), (4.2.3), (4.2.4) guarantees that the matrices
G(x1 ) and H(x0 ) uniquely determine one another (cf. (4.2.5)).
We now turn to the task of presenting the proof of Theorem 4.2.1.
Proof of Theorem 4.2.1 The regularity result from (4.1.8) guarantees that
  M×M
G = Gαβ 1≤α,β ≤M ∈ 𝒞∞ (Ω \ {x0 }) ∩ Wloc
1,1
(Ω) ,
  M×M
(4.2.11)
H = Hαβ 1≤α,β ≤M ∈ 𝒞∞ (Ω \ {x1 }) ∩ Wloc (Ω)
1,1
.

The idea is to re-do the proof of Theorem 4.1.1 for G as in (4.2.2) and u an arbitrary
column of H. To be specific, fix γ, μ ∈ {1, . . . , M } and for each ε ∈ (0, ρ) sufficiently
small define the vector field (using the summation convention over repeated indices)
αβ βα
F := Φε Gαγ a jk ∂k Hβμ − Φε Hαμ ak j ∂k G βγ (4.2.12)

αβ
− Hβμ Gαγ ak j ∂k Φε in Ω.
1≤ j ≤n

This is the analogue of (4.1.16) in which we take u := H. μ , the μ-th column of the
matrix-valued function H. Granted (4.2.2), (4.2.3), (4.2.11), the same arguments as
in the proof of Theorem 4.1.1 give that

F ∈ Lloc and F ≡ 0 in Oε/N thus, in particular,

1 (Ω, L n ) n

κ−n.t. (4.2.13)
Nκε/N F = 0 everywhere on ∂Ω, and F ∂Ω = 0 everywhere on ∂nta Ω.

We may also compute divF,  in the sense of distributions in Ω, much as in (4.1.19)-

(4.1.23) with u := H. μ . Given the latter choice, this time we get

I3 = Φε Gαγ (L A H. μ )α = Φε Gαγ (LH. μ )α

= −Φε Gαγ δx1 δαμ = −Φε G μγ δx1 = −G μγ (x1 )δx1 (4.2.14)

by the assumptions on H, and

13 For example, assuming n := 3, M := 1, L := Δ, and Ω := R3+ , then u(x, y, z) := z/(y 2 + z 2 )

κ−n.t.
for each (x, y, z) ∈ Ω satisfies Lu = 0 in Ω, as well as u ∂Ω = 0 and Nκ (∇u) < +∞ at every
point in ∂Ω \ {(x, 0, 0) : x ∈ R}, hence σ-a.e. on ∂Ω, for each κ > 0. As such, adding multiples
of u to either G or H does not change their properties in (4.2.2), (4.2.3), (4.2.4), with the exception
of the finiteness condition in the scond line of (4.2.4).
4.2 The Reciprocity Principle 791

I6 = −Φε Hαμ (L A G . γ )α = −Φε Hαμ (L  G . γ )α

= Φε Hαμ δαγ δx0 = Φε Hγμ δx0 = Hγμ (x0 )δx0 (4.2.15)

by the assumptions on G. Ultimately, in place of (4.1.24) we now have

βα
divF = −G μγ (x1 )δx1 + Hγμ (x0 )δx0 − (∂j Φε ) Hαμ ak j (∂k G βγ )
αβ αβ
− Hβμ (∂j Gαγ )ak j (∂k Φε ) − Hβμ Gαγ ak j (∂j ∂k Φε ). (4.2.16)

The first two terms in the right-hand side of (4.2.16) belong to ℰ (Ω). Also, since
much as in (4.1.15) we now have
∫
 
|H||G| + |H||∇G| dL n < +∞ for each ε > 0 small, (4.2.17)
Oε

it follows that the last three terms in the right-hand side of (4.2.16) belong to
L 1 (Ω, L n ). Altogether, this proves that

divF ∈ ℰ (Ω) + L 1 (Ω, L n ). (4.2.18)

Let us also observe that, much as in (4.1.29)-(4.1.33), we now have

∫
 −2 
ε |H||G| + ε −1 |H||∇G| dL n = o(1) as ε → 0+ . (4.2.19)
Oε

Moreover, in the case when Ω is an exterior domain, from assumptions and Theo-
rem 1.5.9 (applied both to L  and L) we conclude that
∫

| F(x)| dL n (x) = o(R) as R → ∞. (4.2.20)
B(0, μR)\B(0,R)

This, in turn, shows that condition [112, (1.5.22)] presently holds (assuming Ω is an
exterior domain). Finally, if ∂Ω is unbounded then, much as in (4.1.35), we continue
to have the property that

NκΩ\K F belongs to the space L 1 (∂Ω, σ),

(4.2.21)
for some aperture parameter κ > 0 and some compact set K ⊆ Ω.

Having proved (4.2.13), (4.2.13), (4.2.18) (and also [112, (1.5.22)] when Ω is an
exterior domain, and (4.2.21) when ∂Ω is unbounded), [112, Corollary 1.5.2] applies
(in the version recorded in [112, (1.5.23)]) and, if ν denotes the geometric measure
theoretic outward unit normal to Ω, the Divergence Formula [112, (1.5.20)] gives,
in view of (4.2.16) and (4.2.13), that for each ε > 0 small enough we have
792 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems
∫
 κ−n.t.   
0= ν · F ∂Ω
 1 𝒞∞ (Ω)
dσ = (𝒞∞ (Ω))∗ divF, b
∂∗ Ω b

∫
βα
= −G μγ (x1 ) + Hγμ (x0 ) − (∂j Φε ) Hαμ ak j (∂k G βγ ) dL n (4.2.22)
Ω
∫ ∫
αβ αβ
− Hβμ (∂j Gαγ )ak j (∂k Φε ) dL n − Hβμ Gαγ ak j (∂j ∂k Φε ) dL n .
Ω Ω

Together with [112, (6.1.5)-(6.1.6)] and (4.2.19), this permits us to estimate

∫
 −2 
|G μγ (x1 ) − Hγμ (x0 )| ≤ C ε |H||G| + ε −1 |H||∇G| dL n
Oε

= o(1) as ε → 0+ . (4.2.23)

Granted this, upon letting ε → 0+ we conclude that G μγ (x1 ) = Hγμ (x0 ). Since
γ, μ ∈ {1, . . . , M } have been arbitrarily chosen, this finishes the proof of (4.2.5). 

4.3 How to Construct Green Functions and Use Them in

Uniqueness Issues

The main point of our first result in this section is that uniqueness for the Dirichlet
Problem for a given system L is implied by the solvability of the Regularity Problem
for L  , formulated in a dual setting.

Theorem 4.3.1 Given n ∈ N with14 n ≥ 3, let Ω ⊂ Rn be an Ahlfors regular domain,

which is locally pathwise nontangentially accessible15 (cf. [112, Definition 8.9.14]),
and abbreviate σ := H n−1 ∂Ω. Also, consider a homogeneous constant (complex)
coefficient second-order M × M system L in Rn , which is weakly elliptic (in the sense
of (1.3.3)). Finally, fix an aperture parameter κ > 0 and assume that the spaces 𝒳,
𝒴, 𝒵 are as in any of the following cases:
M
, 𝒵 := 𝒳∗ = L p (∂Ω, σ),
p
𝒳 := L p (∂Ω, σ), 𝒴 := L1 (∂Ω, σ)
(4.3.1)
where p, p ∈ (1, ∞) are such that 1/p + 1/p = 1,
or
M
, 𝒵 := 𝒳∗ = L p (∂Ω, wσ),
p
𝒳 := L p (∂Ω, w 1−p σ), 𝒴 := L1 (∂Ω, w 1−p σ)
where p, p ∈ (1, ∞) are such that 1/p + 1/p = 1 and w ∈ Ap (∂Ω, σ),
(4.3.2)

14 see the last part in the statement for the two-dimensional case
15 in particular, this is the case if Ω is a one-sided NTA domain (see [112, (8.9.180)] and the
subsequent comment)
4.3 How to Construct Green Functions and Use Them in Uniqueness Issues 793

or
𝒳 := L p , q (∂Ω, σ), 𝒵 := 𝒳∗ = L p,q (∂Ω, σ), and
 M M 
𝒴 := f ∈ L p , q (∂Ω, σ) : ∂τ j k f ∈ L p , q (∂Ω, σ) , 1 ≤ j, k ≤ n ,
where p, p , q, q ∈ (1, ∞) are such that 1/p + 1/p = 1 = 1/q + 1/q ,
(4.3.3)
or
p ,λ M
𝒳 := M p ,λ (∂Ω, σ), 𝒴 := M1 (∂Ω, σ) , 𝒵 := B p,λ (∂Ω, σ),
(4.3.4)
where p, p ∈ (1, ∞) are such that 1/p + 1/p = 1,
or
p ,λ M
𝒳 := B p ,λ (∂Ω, σ), 𝒴 := B1 (∂Ω, σ) , 𝒵 := 𝒳∗ = M p,λ (∂Ω, σ),
(4.3.5)
where p, p ∈ (1, ∞) are such that 1/p + 1/p = 1.

Then if the Regularity-type boundary value problem for L 

⎧
⎪ M
w ∈ 𝒞∞ (Ω) , L  w = 0 in Ω,
⎪
⎪
⎪
⎪
⎪
⎪ Nκ (∇w) belongs to the space 𝒳,
⎪
⎪
⎨
⎪ κ−n.t.
w ∂Ω = f at σ-a.e. point on ∂Ω, (4.3.6)
⎪
⎪
⎪
⎪
⎪
⎪ and also w(x) = o(1) as |x| → ∞
⎪
⎪
⎪
⎪ when Ω is an exterior domain,
⎩
is solvable for each boundary datum f ∈ 𝒴, the following uniqueness result for the
dual Dirichlet-type boundary value problem for L is valid:

Lu = 0 in Ω, ⎫
M
u ∈ 𝒞∞ (Ω) , ⎪
⎪
⎪
⎪
⎪
u ∂Ω = 0 at σ-a.e. point on ∂Ω, ⎪
κ−n.t.
⎪
⎪
⎬
⎪
Nκ u belongs to the space 𝒵, ⎪ =⇒ u ≡ 0 in Ω. (4.3.7)
⎪
⎪
and also u(x) = o(1) as |x| → ∞ ⎪⎪
⎪
⎪
⎪
when Ω is an exterior domain, ⎪
⎭
Furthermore, similar results are valid in the case n = 2 provided either Ω is
bounded, or ∂Ω is unbounded. Also, if n = 2 and Ω is an exterior domain, the same
result is true provided one assumes
∫
−1
L(ξ) dH 1 (ξ)
S 1 (4.3.8)
is an invertible M × M matrix
794 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

(which is always the case if the system L is assumed to actually satisfy the Legendre-
Hadamard strong ellipticity condition, and if M = 1 this is true if and only if
L = ∇ · A∇ for some A ∈ M20 ; cf. Lemma 1.4.19 and (1.4.186)), and with the demand
that u and w are o(1) at infinity replaced by

u(x) = O(1) and w(x) = O(1) as |x| → ∞. (4.3.9)



Proof Let E = Eαβ 1≤α,β ≤M be the fundamental solution associated with the
system L  as in Theorem 1.4.2. Fix x ∈ Rn \ Ω along with x0 ∈ Ω, arbitrary. Then,
for each fixed β ∈ {1, . . . , M }, consider the C M -valued function
 
fβ (x) := Eβα (x − x0 ) − Eβα (x − x ) 1≤α ≤M , ∀x ∈ ∂Ω. (4.3.10)

In relation to this, we claim that

fβ belongs to the space 𝒴, as given in any of
(4.3.11)
the scenarios described in (4.3.1)-(4.3.5).
To justify this, bring in the set O := Rn \ {x0, x }, which is an open neighborhood
of ∂Ω, then define the function ψ ∈ 𝒞∞ (O) by setting

ψ(x) := Eβα (x − x0 ) − Eβα (x − x ) for each x ∈ O. (4.3.12)

Thanks to the Mean Value Theorem and item (4) in Theorem 1.4.2, this satisfies

|ψ(x)| ≤ Cx, x0 (1 + |x|)1−n and |(∇ψ)(x)| ≤ Cx, x0 (1 + |x|)−n,

(4.3.13)
for each point x ∈ ∂Ω,

where Cx, x0 is a positive finite constant which depends only on x , x0 , and L.

Granted these, the claim in (4.3.11) becomes a consequence of [113, Lemma 11.1.5]
together with [112, Lemma 7.2.1] if 𝒴 is as in (4.3.1), a consequence of [112,
(7.7.101)] together with item (2) in [112, Lemma 7.7.1] if 𝒴 is as in (4.3.2), a
consequence of [112, (7.7.118)] if 𝒴 is as in (4.3.3), a consequence of [113, (6.2.7)]
if 𝒴 is as in (4.3.4), and a consequence of [113, Proposition 6.2.21] (or [113,
(6.2.245)]) if 𝒴 is as in (4.3.5).
Having established (4.3.11),our current
 working hypotheses then make it possible
to consider the solution wβ = wβα 1≤α ≤M of the boundary value problem (4.3.6),
formulated for this choice of the boundary datum. Subsequently, for each pair of
indices α, β ∈ {1, . . . , M } define
 
Gαβ (x) := wβα (x) − Eβα (x − x0 ) − Eβα (x − x ) , ∀x ∈ Ω \ {x0 }. (4.3.14)
 
When viewing G := Gαβ 1≤α,β ≤M as a C M×M -valued function defined L n -a.e. in
M×M
Ω, from (4.3.14) and Theorem 1.4.2 we see that G ∈ Lloc
1 (Ω, L n ) . Also, by
design,
4.3 How to Construct Green Functions and Use Them in Uniqueness Issues 795
M×M
L  G = −δx0 I M×M in D (Ω)
κ−n.t. (4.3.15)
and G = 0 at σ-a.e. point on ∂Ω,
∂Ω
while from (4.3.14), (4.3.6), (4.3.9), the Mean Value Theorem, and Theorem 1.4.2
we see that, in the case when Ω is an exterior domain, as |x| → ∞ we have

o(1) + O(|x| 1−n ) = o(1) if n ≥ 3,
G(x) = (4.3.16)
O(1) + O(|x| −1 ) = O(1) if n = 2.

Finally, from item (7) in Theorem

 1.4.2 and
 [112, Lemma 8.3.7] it follows that
for any choice of a radius ρ ∈ 0 , dist(x0, ∂Ω) we have
   
NκΩ\K G (x) ≤ max1≤β ≤M Nκ wβ (x) + Cx0,ρ (1 + |x|)1−n and
 Ω\K   
Nκ (∇G) (x) ≤ max1≤β ≤M Nκ (∇wβ ) (x) + Cx0,ρ (1 + |x|)−n, (4.3.17)

at each point x ∈ ∂Ω, where K := B(x0, ρ).

Consequently, the conditions listed in (4.1.43)-(4.1.44) are satisfied, thanks to

(4.3.15)-(4.3.17) and [112, Lemma 7.2.1] when 𝒳, 𝒴, 𝒵 are as in (4.3.1), the
conditions listed in (4.1.45)-(4.1.46) are satisfied thanks to (4.3.15)-(4.3.17) and
[112, (7.7.101)] when 𝒳, 𝒴, 𝒵 are as in (4.3.2), the conditions listed in (4.1.47)-
(4.1.48) are satisfied thanks to (4.3.15)-(4.3.17) and [112, (7.7.118)] when 𝒳, 𝒴,
𝒵 are as in (4.3.3), the conditions listed in (4.1.50)-(4.1.51) are satisfied thanks to
(4.3.15)-(4.3.17) and [113, (6.2.7)] when 𝒳, 𝒴, 𝒵 are as in (4.3.4), and the con-
ditions listed in (4.1.52)-(4.1.53) are satisfied thanks to (4.3.15)-(4.3.17) and [113,
Proposition 6.2.21] when 𝒳, 𝒴, 𝒵 are as in (4.3.5).
As indicated in Examples 4.1.2-4.1.5, these always force u(x0 ) = 0, and since the
point x0 ∈ Ω has been arbitrarily chosen the implication in (4.3.7) follows. 

Theorem 4.3.1 has a natural counterpart, essentially asserting that uniqueness for
the Regularity Problem for a given system L is implied by the solvability of the
Dirichlet Problem for L  , formulated in a dual setting.

Theorem 4.3.2 Fix n ∈ N with16 n ≥ 3. Assume Ω ⊂ Rn is an Ahlfors regular

domain which is locally pathwise nontangentially accessible17 (cf. [112, Defini-
tion 8.9.14]). Abbreviate σ := H n−1 ∂Ω and fix an aperture parameter κ > 0. Also,
consider a homogeneous constant (complex) coefficient second-order M × M system
L in Rn , which is weakly elliptic (in the sense of (1.3.3)). In addition, assume that
the spaces 𝒳, 𝒴 are as in any of the following cases:

𝒳 := L p (∂Ω, σ) and 𝒴 := 𝒳∗ = L p (∂Ω, σ),

(4.3.18)
where p, p ∈ (1, ∞) are such that 1/p + 1/p = 1,

16 see the last part in the statement for the two-dimensional case
17 in particular, this is the case if Ω is a one-sided NTA domain; see [112, (8.9.180)]
796 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

or
𝒳 := L p (∂Ω, w 1−p σ) and 𝒴 := 𝒳∗ = L p (∂Ω, wσ),
(4.3.19)
where p, p ∈ (1, ∞) satisfy 1/p + 1/p = 1 and w ∈ Ap (∂Ω, σ),
or
𝒳 := L p , q (∂Ω, σ) and 𝒴 := 𝒳∗ = L p,q (∂Ω, σ),
(4.3.20)
where p, p , q, q ∈ (1, ∞) satisfy 1/p + 1/p = 1 = 1/q + 1/q ,
or
𝒳 := B p ,λ (∂Ω, σ) and 𝒴 := 𝒳∗ = M p,λ (∂Ω, σ),
(4.3.21)
where p, p ∈ (1, ∞) are such that 1/p + 1/p = 1,
or
𝒳 := M p ,λ (∂Ω, σ) and 𝒴 := 𝒳∗ = B p,λ (∂Ω, σ),
(4.3.22)
where p, p ∈ (1, ∞) are such that 1/p + 1/p = 1.
Then if the Dirichlet-type boundary value problem for L 
⎧
⎪ M
w ∈ 𝒞∞ (Ω) , L  w = 0 in Ω,
⎪
⎪
⎪
⎪
⎪
⎪ Nκ w belongs to the space 𝒳,
⎪
⎪
⎨
⎪ κ−n.t.
w ∂Ω = f at σ-a.e. point on ∂Ω, (4.3.23)
⎪
⎪
⎪
⎪
⎪
⎪ and also w(x) = o(1) as |x| → ∞
⎪
⎪
⎪
⎪ when Ω is an exterior domain,
⎩
is solvable for each boundary datum f ∈ 𝒳, the following uniqueness result for the
dual Regularity-type boundary value problem for L is valid:

Lu = 0 in Ω, ⎫
M
u ∈ 𝒞∞ (Ω) , ⎪
⎪
⎪
⎪
⎪
⎪
u ∂Ω = 0 at σ-a.e. point on ∂Ω, ⎪
κ−n.t.
⎪
⎪
⎬
⎪
Nκ (∇u) belongs to the space 𝒴, ⎪ =⇒ u ≡ 0 in Ω. (4.3.24)
⎪
⎪
⎪
⎪
and also u(x) = o(1) as |x| → ∞ ⎪ ⎪
⎪
⎪
when Ω is an exterior domain, ⎪
⎭
Furthermore, similar results are valid in the case n = 2 provided either Ω is
bounded or ∂Ω is unbounded. Also, if n = 2 and Ω is an exterior domain, the same
result is true provided one assumes
∫
−1
L(ξ) dH 1 (ξ)
S1 (4.3.25)
is an invertible M × M matrix
4.3 How to Construct Green Functions and Use Them in Uniqueness Issues 797

(which is always the case if the system L is assumed to actually satisfy the Legendre-
Hadamard strong ellipticity condition, and if M = 1 this is true if and only if
L = ∇ · A∇ for some A ∈ M20 ; cf. Lemma 1.4.19 and (1.4.186)), and with the demand
that u and w are o(1) at infinity replaced by

u(x) = O(1) and w(x) = O(1) as |x| → ∞. (4.3.26)

Proof This is proved similarly to Theorem 4.3.1, this time relying on Theorem 4.1.6
in the version given in Examples 4.1.7-4.1.10. 
We continue by discussing how the joint solvability of the Dirichlet Problem and
Regularity Problem in bounded uniformly rectifiable domains, formulated for dual
operators and with boundary data in Lebesgue and Sobolev spaces involving Hölder
conjugate integrability exponents, entails genuine well-posedness for the Regularity
Problem in such a setting.

Proposition 4.3.3 Let Ω ⊂ Rn , with n ≥ 2, be a bounded UR domain which is

locally pathwise nontangentially accessible (in the sense of [112, Definition 8.9.14]),
and abbreviate σ := H n−1 ∂Ω. Also, consider a homogeneous constant (complex)
coefficient second-order M × M system L in Rn , which is weakly elliptic (in the sense
of (1.3.3)). Finally, pick p, p ∈ (1, ∞) with 1/p + 1/p = 1 and fix an aperture
parameter κ > 0.
In this context, consider the Dirichlet Problem for the system L with boundary
M
data from the Lebesgue space [L p (∂Ω, σ) , formulated as

⎧
⎪ M
⎪
⎪ u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω,
⎪
⎨
⎪
Nκ u ∈ L p (∂Ω, σ), (4.3.27)
⎪
⎪
⎪
⎪ κ−n.t.
⎪u M
⎩ ∂Ω = f ∈ L (∂Ω, σ) at σ-a.e. point on ∂Ω,
p

together with the Regularity Problem for the transpose system L  with boundary
p M
data from the boundary Sobolev space [L1 (∂Ω, σ) , formulated as

⎧
⎪ M
u ∈ 𝒞∞ (Ω) , L  u = 0 in Ω,
⎪
⎪
⎪
⎨
⎪
Nκ u, Nκ (∇u) ∈ L p (∂Ω, σ), (4.3.28)
⎪
⎪
⎪
⎪ κ−n.t.
⎪u p M
⎩ ∂Ω = f ∈ L1 (∂Ω, σ) at σ-a.e. point on ∂Ω.

Then the solvability of (4.3.27) and (4.3.28) implies that the Regularity Problem
(4.3.28) is actually well-posed (i.e., one has existence, uniqueness, and continuous
dependence of the solution on the boundary datum).

Proof Uniqueness for (4.3.28) is a consequence of Theorem 4.3.2. Since existence

for (4.3.28) is assumed to begin with, there remains to establish the continuous
dependence of the solution on the boundary datum in this problem. To deal with this
issue, we begin by considering the space
798 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems
 
𝒱 := u ∈ 𝒞∞ (Ω) : L  u = 0 in Ω and Nκ u, Nκ (∇u) ∈ L p (∂Ω, σ) ,
M

(4.3.29)
and set
   
u𝒱 := Nκ u L p (∂Ω,σ)
+ Nκ (∇u) L p (∂Ω,σ)
for each u ∈ 𝒱. (4.3.30)

When equipped with this norm, we claim that 𝒱 becomes complete (hence, a
p
Banach space). To justify this, recall the Banach space Nκ (Ω; σ) (defined as in
(A.0.98) with μ := σ). We may then embed 𝒱 into the Banach space
p M p M
𝒲 := Nκ (Ω; σ) ⊕ · · · ⊕ Nκ (Ω; σ) (4.3.31)

n + 1 terms

via
𝒱  u −→ (u, ∂1 u, . . . , ∂n u) ∈ 𝒲 (4.3.32)
which satisfies
 
u𝒱 ≈ (u, ∂1 u, . . . , ∂n u)𝒲 uniformly in u ∈ 𝒱. (4.3.33)

Assume now that {uμ }μ ∈N is a Cauchy sequence in 𝒱. Then (4.3.33) and the
p M
completeness of 𝒲 imply that there exist w, w1, . . . , wn ∈ Nκ (Ω; σ) such that
p M p M
uμ → w in Nκ (Ω; σ) as μ → ∞ and ∂j uμ → w j in Nκ (Ω; σ) as μ → ∞
for each j ∈ {1, . . . , n}. Thanks to [112, (8.3.33)], the above convergences also take
place in Lloc∞ (Ω, L n ) M . Consequently, given an arbitrary ϕ ∈ 𝒞∞ (Ω) M , for
c
each j ∈ {1, . . . , n} we may write
∫ ∫
w j , ϕ dL n = lim ∂j uμ, ϕ dL n
Ω μ→∞ Ω
∫ ∫
= − lim uμ, ∂j ϕ dL = − n
w, ∂j ϕ dL n . (4.3.34)
μ→∞ Ω Ω

This proves that for each j ∈ {1, . . . , n} we have ∂j w = w j in the sense of distributions
in Ω. Let us also observe that since
∫ ∫ ∫
w, Lϕ dL n = lim uμ, Lϕ dL n = lim L  uμ, ϕ dL n = 0, (4.3.35)
Ω μ→∞ Ω μ→∞ Ω

it follows that L w
= 0 in the sense of distributions in Ω. On account of this and
elliptic regularity (see the first claim in [112, Theorem 6.5.7]), we conclude that
M
w ∈ 𝒞∞ (Ω) . Altogether, this argument shows that w ∈ 𝒱 and uμ → w in 𝒱 as
μ → ∞. Hence, the space 𝒱 is complete.
To proceed, define the operator
M κ−n.t.
p
T : 𝒱 −→ L1 (∂Ω, σ) , Tu := u ∂Ω for each u ∈ 𝒱. (4.3.36)
4.4 A Sharp Poisson Integral Representation Formula 799

Thanks to [113, Proposition 11.3.4], the version of the Fatou-type theorem recorded
in (3.3.107), and the fact that, as already noted, the Regularity Problem (4.3.28) is
uniquely solvable, it follows that the operator T is well defined, linear, bounded, and
bijective. Since 𝒱 is a Banach space, the Open Mapping Theorem implies that the
inverse of T is also bounded. This, in turn, guarantees the existence of a constant
C ∈ (0, ∞) with the property that the (unique) solution u of (4.3.28) satisfies
   
Nκ u p + Nκ (∇u) L p (∂Ω,σ) ≤ C f [L p (∂Ω,σ)] M . (4.3.37)
L (∂Ω,σ) 1

This shows that the Regularity Problem (4.3.28) is indeed well-posed, finishing the
proof of the proposition. 

4.4 A Sharp Poisson Integral Representation Formula

In Theorem 4.4.1 below we derive a Poisson integral representation formula for

solutions of the Dirichlet Problem for a given weakly elliptic second-order system
L, which involves the conormal derivative of the Green function for the transpose
system L  as integral kernel.
There are several features of this result which make the case for labeling it as being
sharp. First, the underlying domain is of a very general geometric nature. Second,
the brand of Green function employed here is minimalist, requiring only basic
characteristics typically associated with such as object, like the property of being a
fundamental solution with a vanishing (nontangential pointwise) trace, in addition
to the demand of the existence of the nontangential pointwise trace for its gradient
(which is needed in the very definition of the conormal derivative). Third, our type
of Poisson integral formula is “spaceless,” in the sense that no recognizable function
space measuring size is employed in the formulation of its hypotheses. Instead,
the function u being represented via this Poisson integral formula is only required
to pair well, at the level on nontangential maximal operators, with the gradient of
the Green function (in the sense that their product is absolutely integrable on the
boundary). This ensures that the integral of the boundary trace of u coupled with the
conormal derivative of the Green function is actually integrable, which is needed
to simply make sense of our brand of Poisson integral representation formula. In
Comment 6, following the statement of Theorem 4.4.1, we give examples indicating
that this integrability condition may not be weakened much (and certainly cannot be
eliminated).

Theorem 4.4.1 Let Ω ⊂ Rn , where n ∈ N with18 n ≥ 3, be a locally pathwise

nontangentially accessible set19 (in the sense of [112, Definition 8.9.14]) with a lower
Ahlfors regular boundary and with the property that σ := H n−1 ∂Ω is a doubling

18 see the last claim in the statement for the case n = 2

19 in particular, this is the case if Ω is a one-sided NTA domain; see [112, (8.9.180)] and the
subsequent comment
800 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

measure on ∂Ω. Suppose L is a weakly elliptic, homogeneous, constant (complex)

coefficient, second-order, M × M system in Rn . Fix an aperture parameter κ ∈ (0, ∞),
along with an arbitrary point x0 ∈ Ω, and suppose 0 < ρ < 14 dist(x0, ∂Ω).
Then there exists some 
κ > 0, which depends only on Ω and κ, with the following
significance. Assume G is a matrix-valued function satisfying
⎧   1 (Ω, L n ) M×M ,
⎪
⎪ G = Gαβ 1≤α,β ≤M ∈ Lloc
⎪
⎪
⎪
⎪   
⎪
⎪ L G . β α = −δx0 δαβ in D (Ω) for all α, β ∈ {1, . . . , M },
⎪
⎪
⎪
⎪
⎪
⎪   κ −n.t.
⎨ ∇G
⎪ 2
exists (in Cn·M ) at σ-a.e. point on ∂Ω,
∂Ω (4.4.1)
⎪
⎪ 
κ −n.t.
⎪
⎪
⎪
⎪ G = 0 ∈ C M×M at σ-a.e. point on ∂Ω,
⎪
⎪ ∂Ω
⎪
⎪
⎪
⎪ and also G(x) = o(1) as |x| −→ ∞
⎪
⎪
⎪ when Ω is an exterior domain,
⎩

and assume u = (uβ )1≤β ≤M is a C M -valued function satisfying

⎧
⎪
M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω,
⎪
⎪
⎪
⎪ κ−n.t.
⎪
⎪ exists at σ-a.e. point on ∂Ω,
⎪ u
⎪ ∂Ω
⎪
⎨∫
⎪
ρ ρ (4.4.2)
⎪ ∂Ω Nκ u · Nκ (∇G) dσ < +∞,
⎪
⎪
⎪
⎪
⎪
⎪ and also u(x) = o(1) as |x| −→ ∞
⎪
⎪
⎪
⎪ when Ω is an exterior domain.
⎩

In the case when ∂Ω is unbounded, strengthen the initial geometric hypotheses on

Ω and also strengthen the integral condition in the third line of (4.4.2) by assuming
that (cf. [112, (8.9.179)])
⎧
⎪ Ω is a globally pathwise nontangentially accessible set,
⎪
⎨
⎪ ∫ (4.4.3)
⎪
⎪ Nκ u · NκΩ\K (∇G) dσ < +∞ where K := B(x0, ρ),
⎪ and
⎩ ∂Ω

(with the convention that 0 · ∞ = ∞ · 0 = ∞).  αβ 

Then for any choice of a coefficient tensor A = ar s 1≤r,s ≤n which permits
1≤α,β ≤M
writing L as L A, the system associated with A as in (1.7.7), one has the Poisson
integral representation formula
∫
 κ−n.t. A 
uβ (x0 ) = − u ∂Ω , ∂ν G . β dσ, ∀β ∈ {1, . . . , M }, (4.4.4)
∂∗ Ω
4.4 A Sharp Poisson Integral Representation Formula 801

where ν denotes the geometric measure theoretic outward unit normal to Ω and ∂νA
stands for the conormal derivative associated with A as in (1.7.9), acting on the
columns of the matrix-valued function G.
Moreover, the same result remains true in the two-dimensional setting if either Ω
is bounded, or ∂Ω is unbounded. Also, if n = 2 and Ω is an exterior domain, the
same result is true provided one assumes
∫
−1
L(ξ) dH 1 (ξ)
S1 (4.4.5)
is an invertible M × M matrix

(which is always the case if the system L is assumed to actually satisfy the Legendre-
Hadamard strong ellipticity condition, and if M = 1 this is true if and only if
L = ∇ · A∇ for some A ∈ M20 ; cf. Lemma 1.4.19 and (1.4.186)), and with the demand
that u and G are o(1) at infinity replaced by

u(x) = O(1) and G(x) = O(1) as |x| → ∞. (4.4.6)

Before presenting the proof of this theorem we make a number of comments.

Comment 1. Recall from (4.1.7)-(4.1.8) that the conditions demanded in the first
two lines of (4.4.1) imply that actually

G ∈ 𝒞∞ (Ω \ {x0 }) ∩ Wloc
1,1 M×M
(Ω) . (4.4.7)

In particular, the condition in the third line of (4.4.1) makes sense. Together with

(4.4.7), said condition permits us to define each conormal derivative ∂νA G . β as in
(1.7.9) (with the aperture parameter κ now playing the role ofthe old κ). Moreover,
[112, Proposition 8.9.16] guarantees that we presently have σ ∂Ω \ ∂nta Ω = 0, so it
makes sense to consider nontangential boundary traces (associated with any aperture
parameter) at σ-a.e. point on ∂Ω.
Bearing this in mind, from the first two lines in (4.4.2), (4.4.7), the third line of
(4.4.1), [112, Lemma 8.9.3], and [112, Proposition 8.9.16] we conclude that
 ρ   ρ 
Nκ u (x) < +∞ and Nκ (∇G) (x) < +∞ for σ-a.e. x ∈ ∂Ω. (4.4.8)
 ρ   ρ 
In particular, (4.4.8) ensures that the product Nκ u (x) · Nκ (∇G) (x) is a well-
defined number in [0, +∞) for σ-a.e. x ∈ ∂Ω. In concert with [112, (8.2.28)], this
ρ ρ
proves that Nκ u · Nκ (∇G) is a well-defined non-negative σ-measurable function
on ∂Ω. Consequently, the integral condition formulated in the third line of (4.4.2)
is meaningful. In turn, based on [112, (8.9.8)] and the hypothesis in the third line of
(4.4.2) we see that the integral in (4.4.4) is absolutely convergent, so the conclusion
in the theorem has a clear meaning.
Finally, from the second line in (4.4.3) it follows (bearing in mind the convention
0 · ∞ = ∞ · 0 = ∞) that
   
Nκ u (x) < +∞ and NκΩ\K (∇G) (x) < +∞ for σ-a.e. x ∈ ∂Ω. (4.4.9)
802 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

Comment 2. Remarkably, the only assumptions of quantitative nature made20 in

the statement of Theorem 4.4.1 are the finiteness integral condition imposed in
the third line of (4.4.2), and its version from (4.4.3) in the case when ∂Ω is un-
bounded. These demands are automatically satisfied in any of the instances de-
scribed in (4.1.43)-(4.1.44), (4.1.45)-(4.1.46), (4.1.47)-(4.1.48), (4.1.50)-(4.1.51),
and (4.1.52)-(4.1.53). Also, the aforementioned integral conditions may be recast as
memberships to weighted Lebesgue spaces of the sort
 
ρ ρ
Nκ u ∈ L 1 ∂Ω, Nκ (∇G) σ and Nκ u ∈ L 1 ∂Ω, NκΩ\K (∇G) σ . (4.4.10)

Comment 3. The integral representation formula recorded in (4.4.4) readily implies

the vanishing property u(x0 ) = 0 for any null-solution u of the Dirichlet Problem for
the system L in Ω, in a formulation which uses (4.4.10) as the demand imposed on
the size of the solution. We have already established this in Theorem 4.1.1 asking
ρ   κ −n.t.
Nκ (∇G) < +∞ at σ-a.e. point on ∂Ω in place of the σ-a.e. existence of ∇G ∂Ω
on ∂Ω which, according to [112, (8.9.22)], is a less demanding assumption on
the corresponding Green function. This being said, the latter (stronger) condition
is automatically satisfied if, for example,
 Ω is a UR domain and we assume that
ρ
Nκ (∇G) ∈ L p (∂Ω, σ) for some p ∈ n−1 n ∞ . Indeed, this is seen by applying the
,
Fatou-type result from Theorem 3.3.4 to the columns of the matrix-valued function
 
G restricted to Ω \ B x0, 12 dist(x0, ∂Ω) .
Comment 4. Different choices of the coefficient tensor A which permit us to repre-
sent the given system L as L A yield differently-looking integral representations for-
mulas (4.4.4). At least heuristically, this somewhat curious aspect may be explained
by recalling that the conormal derivatives associated with any two such coefficient
tensors differ by a linear combination of tangential derivatives which, in the case of
the Green function, should vanish since the nontangential boundary trace of G is
zero (see the condition in the fourth line of (4.4.1)). More precisely, if the condition
ρ
in the third line of (4.4.1) is strengthened to Nκ (∇G) ∈ Lloc
1 (∂Ω, σ) then on account

of (1.7.24) (whose applicability is presently ensured by [112, Proposition 8.9.17]

and the condition in the fourth line of (4.4.1)) we see that
if A, B are two coefficient tensors such that the original system may
be represented both as L = L A and as L = LB , then for each index (4.4.11)
 
β ∈ {1, . . . , M } we have ∂νA G . β = ∂νB G . β at σ-a.e. point on ∂∗ Ω.

Comment 5. In the context of Theorem 4.4.1, the integral representation formula

(4.4.4) yields the estimate
∫
κ−n.t. 
κ −n.t.
|u(x0 )| ≤ C u ∂Ω (∇G) ∂Ω dσ. (4.4.12)
∂∗ Ω

20 apart from decay at infinity in the case when Ω is an exterior domain

4.4 A Sharp Poisson Integral Representation Formula 803

Suppose now that there exist α > 0 and C ∈ (0, ∞) such that a Green function G
as in (4.4.1) exists for each pole x0 ∈ Ω and has the additional property that21

|(∇G)(x)| ≤ C · dist(x0, ∂Ω)α |x − x0 | 1−n−α, ∀x ∈ Ω. (4.4.13)

We may then rely on (4.4.12) and [112, Proposition 8.4.12] to conclude that, with
M ∂Ω denoting the Hardy-Littlewood maximal operator on ∂Ω (cf. (A.0.84)), the
following pointwise estimate holds:
 κ−n.t. 
Nκ u ≤ C · M ∂Ω u ∂Ω on ∂Ω, (4.4.14)

for each given aperture parameter κ > 0, where C ∈ (0, ∞) is independent of u.

In turn, estimate (4.4.14) is particularly versatile. For example, this may be used
to establish a Weak Maximum Principle for null-solutions u of the system L in Ω
asserting that, for some background constant C ∈ (0, ∞), we have
 κ−n.t. 
sup |u| ≤ C u ∂Ω [L ∞ (∂Ω,σ)] M , (4.4.15)
Ω

and also prove regularity results to the effect that better integrability properties of
κ−n.t.
the nontangential boundary trace u ∂Ω entail better integrability properties of the
nontangential maximal function of the null-solution u of the system L.
Comment 6. Consider Ω := D := B(0, 1) the unit disk in the plane R2 ≡ C and
introduce
the function u : D −→ R defined as
 (4.4.16)
|z | 2 −1
u(z) := Re z+1
z−1 = |z−1 | 2 for each z ∈ D.

With T := ∂D, this implies

 
u ∈ 𝒞∞ D \ {1} , Δu = 0 in D, and u T\{1} = 0. (4.4.17)

Having fixed some aperture parameter κ > 0, we also claim that

   
Nκ u ∈ L 1,∞ T , H 1 T \ L 1 T, H 1 T . (4.4.18)

To prove this, we start by observing that

(1 − |z|)(1 + |z|) 2 dist(z, T)

|u(z)| = ≤ for each z ∈ D. (4.4.19)
|z − 1| 2 |z − 1| 2

For each ζ ∈ T and each z ∈ Γκ (ζ) we may invoke [112, (8.1.23)] to estimate

21 such an estimate entails (cf. [112, Lemma 8.3.7]) that for each fixed point x ∈ Ω and any
 Ω\K x0 
given aperture parameter κ > 0 we have Nκ (∇G) (x) ≤ C x0, κ (1 + |x |)−(n−1+α) at every
 
x ∈ ∂Ω, where K x0 := B x, 12 dist(x0, ∂Ω)
804 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems
2 2
|z − 1| 2 ≈ |z − ζ | + |ζ − 1| ≈ dist(z, T) + |ζ − 1|
2
≈ dist(z, T) + |ζ − 1| 2 ≥ 2 dist(z, T) · |ζ − 1|. (4.4.20)

Together with (4.4.19) this permits us to conclude that

C
|u(z)| ≤ for each ζ ∈ T and z ∈ Γκ (ζ), (4.4.21)
|ζ − 1|

which ultimately implies that, for some universal constant C ∈ (0, ∞),
  C
Nκ u (ζ) ≤ for each ζ ∈ T. (4.4.22)
|ζ − 1|
In fact, we claim that an inequality of similar nature in the opposite direction is also
valid. Indeed, if for each ζ ∈ T ∩ B(1, 100−1 ) in the first quadrant we let zζ denote
the intersection point of the line x + y = 1 with the line joining the origin with ζ,
then zζ ∈ Γκ (ζ) hence  
Nκ u (ζ) ≥ |u(zζ )|. (4.4.23)
Regarding zζ as a point on the line x + y = 1, permits us to express zζ = (1 − t, t)
for some t ∈ (0, 1/2). Since elementary geometry gives |zζ − 1| ≈ |ζ − 1|, it follows
that t ≈ |ζ − 1|. In view of these properties and the formula for u given in (4.4.16)
we may then compute

1 − |zζ | 2 1 − |(1 − t, t)| 2 1 1

|u(zζ )| = ≈ ≈ ≈ . (4.4.24)
|zζ − 1| 2 t2 t |ζ − 1|

Together with (4.4.23) this gives

  C
Nκ u (ζ) ≥ . (4.4.25)
|ζ − 1|

By symmetry, a similar estimate holds for each ζ ∈ T ∩ B(1, 100−1 ) in the fourth
quadrant, ergo
  C
Nκ u (ζ) ≥ for each ζ ∈ T. (4.4.26)
|ζ − 1|
In concert with (4.4.22) this proves
  1
Nκ u (ζ) ≈ uniformly for ζ ∈ T, (4.4.27)
|ζ − 1|
from which the claim made in (4.4.18) follows.
Since the harmonic function u defined in (4.4.16) is nonzero in D but its non-
tangential trace vanishes H 1 -a.e. on T, it follows that u fails to be representable via
the version of the Poisson integral representation formula (4.4.4) corresponding to
L := Δ and Ω := D. The only hypothesis in the statement of Theorem 4.4.1 that is
currently violated is the absolute integrability condition demanded in the third line
4.4 A Sharp Poisson Integral Representation Formula 805

of (4.4.2). To briefly elaborate on this aspect, recall that the Green function G for
the Laplacian in the unit disk in the plane with pole at a fixed point x0 ∈ D is given
at each x ∈ D by
1 1   x0
G(x) = ln |x − x0 | − ln |x0∗ ||x − x0∗ | , where x0∗ := ∈ C \ D. (4.4.28)
2π 2π |x0 | 2
 
As such, G belongs to 𝒞∞ D \ {x0 } and its normal derivative is given by

1 1 − |x0 | 2
x · (∇G)(x) = for each x ∈ T = ∂D. (4.4.29)
2π |x − x0 | 2

In particular, at each point x ∈ T = ∂D we may estimate

1 1 − |x0 |
|(∇G)(x)| ≥ |x · (∇G)(x)| = , (4.4.30)
2π 1 + |x0 |

which goes to show that for any given

 aperture parameter  κ > 0 and any given
truncation parameter ρ ∈ 0, 1 − |x0 | there exists a constant Cρ ∈ (0, ∞) such that

 ρ  1 1 − |x0 |
Cρ ≥ Nκ (∇G) (x) ≥ > 0 for all x ∈ T = ∂D. (4.4.31)
2π 1 + |x0 |
In view of this and (4.4.18), we conclude that the absolute integrability condition
demanded in the third line of (4.4.2) fails, though just barely. This points to the fact
that Theorem 4.4.1 is in the nature of best possible.
Comment 7. Similar counterexamples to those discussed in Comment 6 above may
be constructed in higher dimensions and for unbounded domains. For instance,
corresponding to L := Δ and Ω := R+n with n ≥ 2 arbitrary, we may consider
u : R+n → R defined as
xn
u(x) := for each x = (x , xn ) ∈ Rn−1 × (0, ∞) = R+n . (4.4.32)
|x| n
This is a harmonic function which is never zero in R+n , and whose nontangential
boundary trace vanishes L n−1 -a.e. on Rn−1 ≡ ∂R+n . As such, the version of the
Poisson integral representation formula (4.4.4) for the current choices fails. The
source of this failure is the fact that the absolute integrability condition stipulated in
the third line of (4.4.2) does not materialize, and this is the only hypothesis in the
statement of Theorem 4.4.1 that is currently violated. To see that this is indeed the
case, recall that the Green function for the Laplacian in the upper half-space with
pole at an arbitrary fixed point x0 = (x0, t0 ) ∈ Rn−1 × (0, ∞) = R+n is given by

G(x) = EΔ (x − x0 ) − EΔ (x − x0, t + t0 ) for each x = (x , t) ∈ R+n \ {x0 }, (4.4.33)

 solution for the Laplacian in R (cf. (1.5.56)).

where EΔ is the standard fundamental n
∞
 n
Hence, G belongs to 𝒞 R+ \ {x0 } and its normal derivative is given by
806 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

2 t0
(∂xn G)(x , 0) = for each x ∈ Rn−1 . (4.4.34)
ωn−1 (t0 + |x − x0 | 2 )n/2

As a consequence, there exists some constant C(n, x0 ) ∈ (0, ∞) with the property
that
C(n, x0 )
(∂xn G)(x , 0) ≥ for all x ∈ Rn−1 . (4.4.35)
1 + |x | n
If we now fix some aperture parameter 
κ > 0 and abbreviate K := B(x0, t0 /2), from
(4.4.35) we then deduce that
 R n \K  C(n, x0 )
Nκ + (∇G) (x) ≥ for all x ∈ Rn−1 . (4.4.36)
1 + |x | n
In the opposite direction, using the fact that there exists a dimensional constant
Cn ∈ (0, ∞) with the property that
t0
(∇G)(x) ≤ Cn for each x ∈ R+n with |x − x0 | > 4t0 (4.4.37)
|x − x0 | n
(itself a consequence of (4.4.33) and (1.5.56)), we may employ [112, Lemma 8.3.7]
to conclude that there exists some C(n, 
κ, x0 ) ∈ (0, ∞) for which
 R n \K  κ, x0 )
C(n, 
Nκ + (∇G) (x ) ≤ for all x ∈ Rn−1 . (4.4.38)
1 + |x | n
Altogether, (4.4.36) and (4.4.38) prove that
 R n \K  1
Nκ + (∇G) (x ) ≈ uniformly in x ∈ Rn−1 . (4.4.39)
1 + |x | n
As far as the function u defined in (4.4.32) is concerned, fix an aperture parameter
κ > 0 and observe that for each x ∈ Rn−1 \ {0 } and each y = (y , yn ) ∈ Γκ (x ) we
have
1−n
|u(y)| ≤ |y| 1−n ≤ dist(0, Γκ (x )) ≤ Cκ |x | 1−n (4.4.40)
for some constant Cκ ∈ (0, ∞) which depends exclusively on κ. In turn, this implies
that
Cκ
(Nκ u)(x ) ≤ for all x ∈ Rn−1 . (4.4.41)
|x | n−1
Since for each x ∈ Rn−1 \ {0 } we may also estimate

  |x | 2−n/2
(Nκ u)(x ) ≥ u x , |x | =   n = , (4.4.42)
x , |x | |x | n−1

we deduce from (4.4.41) and (4.4.42) that the function u defined in (4.4.32) satisfies
1
(Nκ u)(x ) ≈ uniformly in x ∈ Rn−1 . (4.4.43)
|x | n−1
4.4 A Sharp Poisson Integral Representation Formula 807

Together, (4.4.39) and (4.4.43) prove that

 R n \K  1
Nκ + (∇G) (x ) · (Nκ u)(x ) ≈ uniformly in x ∈ Rn−1 .
|x | n−1 (1 + |x | n )
(4.4.44)
In light of the fact that for each x ∈ Rn−1 \ {0 } we have
1 1 1
· 1Bn−1 (0 ,1) ≤ ≤ , (4.4.45)
2|x | n−1 |x | (1 + |x | )
n−1 n |x | n−1
we conclude from (4.4.44)-(4.4.45) that
 R n \K 
Nκ + (∇G) · (Nκ u) belongs to L 1,∞ (Rn−1, L n−1 ) \ L 1 (Rn−1, L n−1 ). (4.4.46)

Hence, the absolute integrability condition demanded in the third line of (4.4.2) fails,
though very narrowly. Once again, this highlights the optimality of Theorem 4.4.1.
Comment 8. Theorem 4.4.1 naturally yields a uniqueness result for the Dirichlet
Problem, as the conclusion in Theorem 4.1.1 is implied by Theorem 4.4.1. The
reason for which Theorem 4.1.1 is not a corollary of Theorem 4.4.1 is the fact
that, as opposed to the latter, the former theorem involves an additional hypothesis,
demanding the existence of the nontangential boundary trace of ∇G.
We now turn to the task of presenting the proof of Theorem 4.4.1.
Proof of Theorem 4.4.1 Recall that (4.4.7) holds. Fix some β ∈ {1, . . . , M } and,
using the summation convention over repeated indices, define the vector field

γα αγ
F := uα ak j ∂k Gγβ − Gαβ a jk ∂k uγ at L n -a.e. point in Ω. (4.4.47)
1≤ j ≤n

M
From (4.4.47), (4.4.7), and the fact that u ∈ 𝒞∞ (Ω) it follows that

F ∈ Lloc
n
1
(Ω, L n ) (4.4.48)

and, in the sense of distributions in Ω, we have

γα γα
divF = (∂j uα )ak j (∂k Gγβ ) + uα ak j (∂j ∂k Gγβ )
αγ αγ
− (∂j Gαβ )a jk (∂k uγ ) − Gαβ a jk (∂j ∂k uγ )

=: I1 + I2 + I3 + I4, (4.4.49)

where the last equality defines the Ii ’s. Changing variables j = k, k = j, α = γ,

and γ = α in I3 yields
γα
I3 = −(∂k Gγ β ) ak j
(∂j uα ) = −I1 . (4.4.50)

As regards I4 , we have
808 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

I4 = −Gαβ (L Au)α = −Gαβ (Lu)α = 0, (4.4.51)

by the assumptions on u. Finally, thanks to the second line in (4.4.1), we have

I2 = uα (L A G . β )α = uα (L  G . β )α

= −uα δαβ δx0 = −uβ δx0 = −uβ (x0 )δx0 . (4.4.52)

Collectively, these equalities permit us to conclude that, in the sense of distributions

in Ω,

divF = −uβ (x0 )δx0 ∈ ℰ (Ω). (4.4.53)

In particular,
 ∗
divF ∈ D (Ω) induces a continuous functional in 𝒞∞b (Ω) . (4.4.54)

Moving on, recall from [112, Proposition 8.9.16] that in the present context we
have  
σ ∂Ω \ ∂nta Ω = 0. (4.4.55)
Starting with the aperture parameter κ given in the statement of the theorem, choose
a ∈ (0, 1) small enough so that κ > 2a/(1 − a), then define

κ := κ(1 − a) − 2a ∈ (0, κ). (4.4.56)

Next, let 
κ ≥ κ along with d > 0 and c ∈ [1, ∞) be parameters associated as in
[112, (8.9.178)] with κ (now playing the role of κ in that context). If necessary,
further increase 
κ as to ensure that 
κ ≥ κ. Also, set

θ := [c(1 + κ )]−1 ∈ (0, 1), (4.4.57)

define  
ρ
N1 := x ∈ ∂Ω : Nκ (∇G)(x) = +∞ , (4.4.58)

and introduce   κ−n.t.  

N2 := x ∈ ∂nta Ω : u ∂Ω (x) fails to exist . (4.4.59)

Then (4.4.8) implies σ(N1 ) = 0. Also, from the second line in (4.4.2) and [112,
(3.6.26)] we see that N2 is a σ-measurable set and σ(N2 ) = 0.
To proceed, fix x ∈ ∂nta Ω \ (N1 ∪ N2 ) and pick an arbitrary point
 
y ∈ Γκ (x) with δ∂Ω (y) < min θ · ρ, d . (4.4.60)

The property in the fourth line of (4.4.1) ensures that the condition formulated in
[112, (8.9.196)] is satisfied by (the entries of) the matrix-valued function G. Granted
this, [112, (8.9.197)] written with κ in place of κ and for any
4.4 A Sharp Poisson Integral Representation Formula 809
 
ε ∈ 0, min ρ, d/θ with ε > θ −1 δ∂Ω (y) (4.4.61)

guarantees the existence of a σ-measurable set N3 ⊆ ∂Ω satisfying σ(N3 ) = 0 such

that
 
|G(y)| ≤ Nκθε G (x) ≤ ε · Nκε (∇G)(x)
ρ
≤ ε · Nκ (∇G)(x) if x  N3, (4.4.62)

where the first inequality takes into account [112, (8.2.13)], the second inequality
relies on [112, Remark 8.9.19], and the last inequality is a consequence of (4.4.61)
together with [112, (8.2.25)]. Sending ε  θ −1 δ∂Ω (y) and keeping in mind our
choice of θ in (4.4.57) then yields
ρ
|G(y)| ≤ c(1 + κ )δ∂Ω (y) · Nκ (∇G)(x) provided x  N3 . (4.4.63)

Next, the fact that x belongs to the set Aκ (∂Ω) \ N2 ensures that the nontangential
 κ−n.t. 
boundary trace u ∂Ω (x) exists in C M . Using interior estimates (cf. [112, Theo-
 
rem 6.5.7]) in the ball B y, a · δ∂Ω (y) for the null-solution of the weakly elliptic
system L defined as
 κ−n.t. 
w(z) := u(z) − u ∂Ω (x) for each z ∈ Ω, (4.4.64)

we may estimate, for some constant C = C(L, n, a) ∈ (0, ∞),

⨏
C  κ−n.t. 
|(∇u)(y)| = |(∇w)(y)| ≤ u − u ∂Ω (x) dL n
δ∂Ω (y) B(y,a·δ∂Ω (y))
 κ−n.t. 
≤ Cδ∂Ω (y)−1 · sup u(z) − u ∂Ω (x) , (4.4.65)
z ∈Γκ (x)
|x−z |<(1+κ +a)δ ∂Ω (y)
 
since having z ∈ B y, a · δ∂Ω (y) entails (in view of the fact that the distance function
δ∂Ω is Lipschitz with constant ≤ 1),
1
δ∂Ω (y) ≤ δ∂Ω (z) + |z − y| < δ∂Ω (z) + a · δ∂Ω (y) =⇒ δ∂Ω (y) < δ∂Ω (z),
1−a
(4.4.66)

which, bearing in mind that y is as in (4.4.60) and the formula for κ from (4.4.56),
permits us to conclude that
810 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

|z − x| ≤ |z − y| + |y − x| < a · δ∂Ω (y) + (1 + κ )δ∂Ω (y)

1+κ +a
= (1 + κ + a)δ∂Ω (y) < δ∂Ω (z) = (1 + κ)δ∂Ω (z),
1−a
hence z ∈ Γκ (x) and |z − x| < (1 + κ + a)δ∂Ω (y). (4.4.67)

By combining (4.4.63) and (4.4.65) we arrive at

ρ  κ−n.t. 
|G(y)||(∇u)(y)| ≤ CNκ (∇G)(x) · sup u(z) − u ∂Ω (x) (4.4.68)
z ∈Γκ (x)
|x−z |<(1+κ +a)δ ∂Ω (y)

which, in light of [112, Definition 8.9.1] (cf. (A.0.99)) and (4.4.58), implies

lim |G(y)||(∇u)(y)| = 0 for each x ∈ ∂nta Ω \ (N1 ∪ N2 ∪ N3 ). (4.4.69)

Γκ (x)y→x

In turn, from (4.4.47), (4.4.1), (4.4.2), (4.4.69), and [112, (8.9.12)] we conclude
(upon recalling that 0 < κ < κ ≤  κ and σ(N1 ∪ N2 ∪ N3 ) = 0) that
κ −n.t.
F ∂Ω exists at σ-a.e. point on ∂nta Ω (4.4.70)

and, in fact,

  
γα  
κ −n.t.
κ −n.t. κ −n.t.
F = uα ∂Ω
ak j ∂k Gγβ ∂Ω
∂Ω 1≤ j ≤n

  
κ−n.t.
γα   
κ −n.t.
= uα ∂Ω
ak j ∂k Gγβ ∂Ω
. (4.4.71)
1≤ j ≤n

Next, fix an arbitrary point x ∈ ∂nta Ω and, for each y ∈ Γκ (x) with

 
δ∂Ω (y) < εo := min θ · ρ, d, ρ/(1 + κ + a) , (4.4.72)

in place of (4.4.65) now write, using interior estimates for the function u, [112,
(8.2.13)], and
 keeping in mind that the conclusions in (4.4.66) are valid for each
point z ∈ B y, a · δ∂Ω (y) ,
⨏
C
|(∇u)(y)| ≤ |u| dL n (4.4.73)
δ∂Ω (y) B(y,a·δ∂Ω (y))
 ρ 
≤ Cδ∂Ω (y)−1 · sup |u(z)| ≤ Cδ∂Ω (y)−1 · Nκ u (x).
z ∈Γκ (x)
|x−z |<(1+κ +a)δ ∂Ω (y)

Then combining (4.4.63) with (4.4.73) gives, on account of (4.4.55),

   ρ  ρ
Nκεo |G||∇u| ≤ C Nκ (∇G) · Nκ u at σ-a.e. point on ∂Ω. (4.4.74)
4.4 A Sharp Poisson Integral Representation Formula 811

Since thanks to [112, (8.2.10)], [112, (8.2.25)], ρ > εo , and 

κ > κ , we also have
   
Nκεo |∇G||u| ≤ Nκεo (∇G) · Nκεo u
 ρ  ρ
≤ Nκ (∇G) · Nκ u at each point on ∂Ω, (4.4.75)

we conclude from (4.4.47), (4.4.74), (4.4.75), the third line in (4.4.2), and [112,
Proposition 8.2.3] that the truncated nontangential maximal function of F (at height
εo and aperture κ ) satisfies

Nκεo F ∈ L 1 (∂Ω, σ). (4.4.76)

Let us now specialize the discussion to the case when Ω is an exterior domain.
When n ≥ 3, from assumptions and Theorem 1.5.9 (applied both to L  and L) it
follows that as |x| → ∞ we also have

G(x) = O(|x| 2−n ), (∇G)(x) = O(|x| 1−n ),

(4.4.77)
u(x) = O(|x| 2−n ), (∇u)(x) = O(|x| 1−n ).


Together with (4.4.47), this implies that F(x) = O(|x| 3−2n ) as |x| → ∞. Thus, since
n ≥ 3, for each λ ∈ (1, ∞) we have
∫

| F(x)| dL n (x) = o(R) as R → ∞. (4.4.78)
B(0,λ R)\B(0,R)

When n = 2, the additional assumptions made in (4.4.5) and (4.4.6) come in effect.
As such, the last part in Theorem 1.5.9 (applied both to L  and L) applies and gives
that as |x| → ∞ we have

G(x) = O(1), (∇G)(x) = O(|x| −2 ),

(4.4.79)
u(x) = O(1), (∇u)(x) = O(|x| −2 ).


In concert with (4.4.47) this implies that F(x) = O(|x| −2 ) as |x| → ∞, hence once
again (4.4.78) holds (with n = 2) for each μ ∈ (1, ∞). To sum up, this analysis
shows that the condition formulated in [112, (1.5.22)] presently holds, when Ω is an
exterior domain, in all dimensions n ≥ 2.
Having established (4.4.48), (4.4.54), (4.4.76), (4.4.70), and [112, (1.5.22)] (when
Ω is an exterior domain), [112, Corollary 1.5.2] applies and the Divergence Formula
[112, (1.5.20)] gives, in view of (4.4.53), the last line in (4.4.71), (1.7.9), and with
(ν j )1≤ j ≤n denoting the components of the geometric measure theoretic outward unit
normal ν to Ω,
812 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems
∫
   κ −n.t. 
 1 𝒞∞ (Ω) =
−uβ (x0 ) = (𝒞∞ (Ω))∗ divF, ν · F dσ
b b ∂Ω
∂∗ Ω
∫
 κ−n.t.  γα   
κ −n.t.
= uα ∂Ω
ν j ak j ∂k Gγβ ∂Ω
dσ
∂∗ Ω
∫
 κ−n.t.  
= u ∂Ω , ∂νA G . β dσ. (4.4.80)
∂∗ Ω

Since β ∈ {1, . . . , M } has been arbitrarily chosen, this establishes (4.4.4) in the case
when ∂Ω is bounded and n ≥ 3. In fact, the same argument works in the case when
Ω is bounded and n = 2 (since (4.4.78) is now no longer relevant).
To finish the proof of Theorem 4.4.1 there remains to treat the case when ∂Ω is
unbounded and n ≥ 2. To this end, work under the stronger assumptions made in
(4.4.3) (in particular, Ω is a globally pathwise nontangentially accessible set). We
claim that, in this setting,

NκΩ\(2K) F ∈ L 1 (∂Ω, σ). (4.4.81)

To prove this claim, observe from (4.4.47) that

| F | ≤ C|G| |∇u| + C|u| |∇G| in Ω, (4.4.82)

and that at σ-a.e. point on ∂Ω we have

   
NκΩ\(2K) |G| |∇u| = NκΩ\(2K) (|G|/δ∂Ω ) · (δ∂Ω |∇u|)
   
≤ NκΩ\(2K) G/δ∂Ω · NκΩ\(2K) δ∂Ω |∇u|

   
≤ NκΩ\(2K) G/δ∂Ω · Nκ δ∂Ω |∇u|
  
≤ C Nκ u NκΩ\K (∇G) , (4.4.83)

thanks to [112, (8.2.10), (8.2.25), (8.9.147)] (with |α| = 1 and λ = 0), [112, (8.9.227)]
(with u := G and with the understanding that  κ has been chosen, from the beginning,
to be greater than the aperture parameter  κo appearing in the statement of [112,
Proposition 8.9.21]), and (4.4.55). From (4.4.82) and (4.4.83) we then see that

NκΩ\(2K) F ≤ C Nκ u · NκΩ\K (∇G) at σ-a.e. point on ∂Ω. (4.4.84)

On account of (4.4.84), (4.4.3), and [112, (8.2.28)] we then conclude that the mem-
bership in (4.4.81) is indeed true.
At this stage, (4.4.48), (4.4.54), (4.4.81), and (4.4.70) ensure that [112, Theo-
rem 1.4.1] (in the version recorded in Note 8 following its statement) applies. Then
the Divergence Formula [112, (1.4.6)] gives, in view of (4.4.53), the last line in
(4.4.71), and (1.7.9), that
4.5 The Poisson Kernel Associated with a System: A First Look 813
∫
   κ −n.t. 
 1 𝒞∞ (Ω) =
−uβ (x0 ) = (𝒞∞ (Ω))∗ divF, ν · F dσ
b b ∂Ω
∂∗ Ω
∫
 κ−n.t.  γα   
κ −n.t.
= uα ∂Ω
ν j ak j ∂k Gγβ ∂Ω
dσ
∂∗ Ω
∫
 κ−n.t.  
= u ∂Ω , ∂νA G . β dσ. (4.4.85)
∂∗ Ω

In view of the arbitrariness of β ∈ {1, . . . , M }, formula (4.4.4) follows in the case

when ∂Ω is unbounded and n ≥ 2. This concludes the proof of Theorem 4.4.1. 

4.5 The Poisson Kernel Associated with a System: A First Look

The goal here is to introduce and study the Poisson kernel associated with a given
weakly elliptic system, in various geometric settings.
Having fixed some n ∈ N with n ≥ 2, let Ω ⊂ Rn be a UR domain which is glob-
ally pathwise nontangentially accessible (in the sense of [112, Definition 8.9.14]).
Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic out-
ward unit normal to Ω. Also, fix a sufficiently large aperture parameter κ ∈ (0, ∞).
Finally, suppose L is a homogeneous, constant (complex) coefficient, second-order,
M × M system in Rn , satisfying the Legendre-Hadamard strong ellipticity condition.
In this context, assume that for each given x ∈ Ω there exists a matrix-valued
function G L (·, x) which satisfies

⎧  L  1 (Ω, L n ) M×M ,
⎪
⎪ G L (·, x) = Gαβ (·, x) 1≤α,β ≤M ∈ Lloc
⎪
⎪
⎪
⎪  
⎪
⎪ L G .Lβ (·, x) = − δx δαβ 1≤α ≤M in D (Ω)
M
for each β ∈ {1, . . . , M },
⎪
⎪
⎪
⎪
⎪
⎨ ∇[G L (·, x)]
κ−n.t.
⎪
⎪ 2
exists (in Cn·M ) at σ-a.e. point on ∂Ω,
∂Ω
⎪
⎪ κ−n.t.
⎪
⎪ L (·, x)
⎪
⎪ G = 0 ∈ C M×M at σ-a.e. point on ∂Ω,
⎪
⎪ ∂Ω
⎪
⎪
⎪
⎪ and also G L (·, x) = O(|x| 2−n ) at infinity
⎪
⎪
⎪ when Ω is an exterior domain.
⎩
(4.5.1)
For each β ∈ {1, . . . , M } define

P.Lβ (x, y) := −∂ν(y)

A
G .Lβ (y, x) for each x ∈ Ω and σ-a.e. y ∈ ∂Ω. (4.5.2)

Assuming that a function as in (4.5.1) exists for L replaced by L  , we shall call the
C M×M -valued function
 L 
PL (x, y) := Pβα (x, y) 1≤α,β ≤M for x ∈ Ω arbitrary and σ-a.e. y ∈ ∂Ω, (4.5.3)
814 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

the Poisson kernel associated with the system L in the domain Ω. In terms of this
piece of notation, (4.4.4) becomes the integral representation formula
∫
 κ−n.t. 
u(x) = PL (x, y) u ∂Ω (y) dσ(y), ∀x ∈ Ω,
∂Ω (4.5.4)
for each C -valued function u as in (4.4.2).
M


Suppose the Green functions G L (·, ·) and G L (·, ·) both exist (as matrix-valued
functions satisfying the conditions listed in (4.5.1) for L and L  , respectively). In
addition, assume that for any two distinct points x0, x1 ∈ Ω we have
∫
  dσ(y)
NκΩ\K ∇G L (·, x0 ) (y) < +∞, (4.5.5)
∂Ω 1 + |y| n−2
∫
  
NκΩ\K G L (·, x1 ) · NκΩ\K ∇G L (·, x0 ) dσ < +∞, (4.5.6)
∂Ω
and ∫

NκΩ\K G L (·, x1 ) · NκΩ\K G L (·, x0 ) dσ < +∞, (4.5.7)
∂Ω
where K ⊆ Ω is some compact neighborhood of {x0, x1 }. If  3 we may apply the
integral representation formula (4.5.4) (written for L  in place of L) to the columns

of G L (·, x1 ) + E L  (· − x1 ), where E is the fundamental solution associated with the
system L  as in Theorem 1.4.2 which presently reads
∫

G L (x0, x1 ) + E L  (x1 − x0 ) = PL  (x0, y)E L  (x1 − y) dσ(y). (4.5.8)
∂∗ Ω

Passing to transpose and invoking the Reciprocity Relation from (4.2.8) which
currently reads

G L (x1, x0 ) = [G L (x0, x1 )], (4.5.9)
we therefore obtain
∫

G (x1, x0 ) + E(x1 − x0 ) =
L
E(x1 − y) PL  (x0, y) dσ(y). (4.5.10)
∂∗ Ω

To proceed, we shall need notation and results in connection with boundary

layer potential operators associated with the system L and the set Ω. The (formal)
convolution-type integral operator on ∂Ω whose kernel is the fundamental solution
E = E L is referred to as single layer potential operator. If we denote by 𝒮 the
boundary-to-domain single layer potential operator, then (4.5.10) simply reads


G L (·, x0 ) + E(· − x0 ) = 𝒮 PL  (x0, ·) in Ω, (4.5.11)

or, equivalently,
4.5 The Poisson Kernel Associated with a System: A First Look 815
 
G .Lβ (·, x0 ) + E. β (· − x0 ) = 𝒮 P.Lβ (x0, ·) in Ω, for 1 ≤ β ≤ M. (4.5.12)

If S = SL denotes the boundary-to-boundary single layer potential operator on

∂Ω associated with the system L, then taking the nontangential boundary traces in
(4.5.12) leads to the conclusion that
 
S P.Lβ (x0, ·) = E. β (· − x0 ) ∂Ω for 1 ≤ β ≤ M. (4.5.13)

In other words, 

S PL  (x0, ·) = E(· − x0 ) ∂Ω . (4.5.14)

To proceed, we need the notion of double layer. Such an operator may be associated
with any coefficient tensor A which may be used to write the given system L.
Once again, associated with any such coefficient tensor A, there is a boundary-to-
boundary double layer potential operator, denoted by K A, and a boundary-to-domain
double layer potential operator, denoted by D A. Finally, we shall denote by K A#
the “transpose” double layer associated with the coefficient tensor A. All these are
discussed at length in [115, Chapter 1]. What we need here is the fact that taking
conormal derivatives in (4.5.11) yields
1 #  L 
2 I + K A P. β (x0, ·) = ∂ν E . β (· − x0 ) for 1 ≤ β ≤ M,
A
(4.5.15)

or, in matrix formalism (with the operators acting on columns),

1 
#  
2 I + K A PL  (x0, ·) = ∂νA E(· − x0 ) . (4.5.16)

Going further, suppose we work in an environment where, say, the L 2 Dirichlet

boundary value problem for the system L is well-posed. For example, such a setting
is described in [116, Chapter 8] where the solution is expressed as
 κ−n.t.  
u = 𝒮 S −1 u ∂Ω in Ω. (4.5.17)

It is interesting to note that this is compatible with the Poisson integral representation
formula (4.5.4). Indeed, for each x ∈ Ω and each β ∈ {1, . . . , M } we may use (4.5.17),
  −1
the fact that the transpose of S −1 is SL  , (4.5.13) (written with L  in place of L,
(1.4.32), and (4.5.3) to write
816 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems
∫   κ−n.t.  
uβ (x) = Eβ .(x − ·) ∂Ω, S −1 u ∂Ω dσ
∂Ω
∫  
 −1 κ−n.t.
= SL  Eβ .(x − ·) ∂Ω, u ∂Ω dσ
∂Ω
∫  
 κ−n.t.
= P.Lβ (x, ·), u ∂Ω dσ
∂Ω
∫
 κ−n.t.  
= PL (x, ·) u ∂Ω dσ, (4.5.18)
∂Ω β

from which (4.5.4) follows.

M
When 12 I + K A is an invertible operator on, say, L 2 (∂Ω, σ) , then an alternative
formula for the solution u of the L 2 Dirichlet boundary value problem for the system
L in Ω is   −1  κ−n.t.  
u = D A 12 I + K A u ∂Ω in Ω. (4.5.19)

Again, we wish to check that this is compatible with the Poisson integral representa-
tion formula (4.5.4). To justify this, given any x ∈ Ω and β ∈ {1, . . . , M } we write,
thanks to the formula for the boundary-to-domain double layer (cf. [115, §1.3]), the
fact that the transpose of K A is K A# , (4.5.15) (written for A in place of A), and
(1.4.32),
∫ 
     −1  κ−n.t.  
uβ (x) = ∂νA E L  . β (x − ·) , 12 I + K A u ∂Ω dσ
∂Ω
∫   κ−n.t. 
 −1   
= 1
2I + K A# ∂νA E L  . β (x − ·) , u ∂Ω dσ
∂Ω
∫  
 κ−n.t.
= P.Lβ (x, ·), u ∂Ω dσ
∂Ω
∫
 κ−n.t.  
= PL (x, ·) u ∂Ω dσ, (4.5.20)
∂Ω β

as wanted.
Moving on, we wish to check directly the veracity of formula (4.5.15) for the
Laplacian L := Δ in the unit ball Ω := B(0, 1) ⊆ Rn with n ≥ 3. To this end, recall
first that the Green function this special setting is given by
1 1
G(y, x) := − + (4.5.21)
(2 − n)ωn−1 |x − y| n−2 (2 − n)ωn−1 |x| n−2 y − x/|x| 2
n−2

for any two distinct points x, y ∈ B(0, 1). Of course, the first term above is just
−EΔ (x − y), where EΔ is the standard fundamental solution of the Laplacian in Rn
(defined in (1.5.56)), while the second term is a harmonic correction, designed as to
ensure that G(·, x) S n−1 = 0 for each fixed x ∈ B(0, 1). Indeed, the latter hinges on
4.5 The Poisson Kernel Associated with a System: A First Look 817

the fact that, as it may be easily checked, |x| y − x/|x| 2 = |x − y| for each x ∈ B(0, 1)
and y ∈ S n−1 . Also, in the same setting, the Poisson kernel is given by

1 − |x| 2
1
P(x, y) := −∂ν(y) G(y, x) =
ωn−1 |x − y| n (4.5.22)
for each point x ∈ B(0, 1) and each point y ∈ S n−1 ,

where the vector ν is the outward unit normal to B(0, 1). The goal is to directly check
that, for each x ∈ B(0, 1),
1 #  
2 I + KΔ P(x, ·) = ∂ν EΔ (· − x) on S ,
n−1
(4.5.23)

where I is the identity, and KΔ# is the transpose of the harmonic double layer potential
operator KΔ (associated with Ω := B(0, 1) as in (2.5.203)). To this end, use the fact
that (see the discussion in the preamble of [116, Chapter 7])
 
KΔ f = KΔ# f = 2−n 2 SΔ f if n ≥ 3, (4.5.24)

where SΔ is the boundary-to-boundary harmonic single layer potential operator on

∂B(0, 1) = S n−1 , to write
1 #   1  2−n   
2 I + KΔ P(x, ·) = 2 P(x, ·) + 2 SΔ P(x, ·) . (4.5.25)

Next, if 𝒮Δ denotes the boundary-to-domain harmonic single layer potential operator

associated with the unit ball B(0, 1), then for each point z ∈ B(0, 1) we may use
the integration by parts formula from Theorem 1.7.11 and the fact that we have
G(·, x) S n−1 = 0 for each fixed x ∈ B(0, 1) to compute
∫
 
𝒮Δ P(x, ·) (z) = EΔ (z − y)P(x, y) dσ(y)
∂B(0,1)
∫
=− EΔ (z − y)∂ν(y) G(y, x) dσ(y)
∂B(0,1)
 
= ℰ (B(0,1)) Δy EΔ (z − y) , G(y, x) ℰ(B(0,1))
 
− ℰ (B(0,1)) Δy G(y, x) , EΔ (z − y) ℰ(B(0,1))

= G(z, x) + EΔ (z − x). (4.5.26)

Consequently,

    n.t.

SΔ P(x, ·) = 𝒮Δ P(x, ·) = G(·, x) S n−1 + EΔ (· − x) S n−1

∂B(0,1)

= EΔ (· − x) S n−1 . (4.5.27)
818 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

With this in hand, we return to (4.5.25) and conclude that at each point y ∈ S n−1 we
have
1 #    
2 I + K Δ P(x, ·) (y) = 12 P(x, y) + 2−n
2 EΔ (x − y)

1 1 − |x| 2 1
= +
2ωn−1 |x − y| n 2ωn−1 |x − y| n−2
1 y, y − x
= = ∂ν(y) EΔ (y − x) , (4.5.28)
ωn−1 |x − y| n
which completes the verification of (4.5.16).

4.6 The Poisson Kernel Associated with a System in the Upper

Half-Space

Theorem 4.4.1 yields nontrivial new results even in the case when the set Ω is simply
the upper half-space R+n . To describe them in detail, consider a constant (complex)
coefficient, second-order, M × M system L in Rn satisfying the Legendre-Hadamard
strong ellipticity condition. From [35] we then know that there exists a unique
C M×M -valued function G L (·, ·) defined in R+n such that for each y = (y , yn ) ∈ R+n
the following properties hold (for some aperture parameter κ > 0):
M×M
G L (· , y) ∈ Lloc
1
(R+n, L n ) , (4.6.1)
κ−n.t.
G L (· , y) ∂Rn = 0 L n−1 -a.e. in Rn−1 ≡ ∂R+n, (4.6.2)
+
∫ 
R n \B(y,yn /2) L dx
Nκ + G (· , y) (x ) < ∞, (4.6.3)
R n−1 1 + |x | n−1
M×M
L G L (· , y) = −δy I M×M in D (R+n ) , (4.6.4)

where L acts in the “dot" variable on the columns of G L (with the outcomes also
arranged as columns). In addition, for each fixed y ∈ R+n we have
 
G L (· , y) ∈ 𝒞∞ R+n \ B(y, ε)
M×M
for every ε > 0. (4.6.5)

Also, given κ > 0, for each y ∈ R+n and each compact neighborhood K of y in R+n
there exists a finite constant C = C(n, L, κ, K, y) > 0 such that
 1 + log+ |x |
R n \K
Nκ + G L (·, y) (x ) ≤ C for each x ∈ Rn−1 . (4.6.6)
1 + |x | n−1
In fact,
4.6 The Poisson Kernel Associated with a System in the Upper Half-Space 819
 αβ 
if L may be written using a coefficient tensor A = a jk 1≤ j,k ≤n which
1≤α,β ≤M
αβ αβ
has the following block-structure: a jn = anj = 0 whenever j < n, then
(4.6.7)
the logarithm in the right side of (4.6.6) may actually be omitted and,
with y ∈ R−n denoting the reflection of y ∈ R+n across ∂R+n , we have
G L (x, y) = E(x − y) − E(x − y) for all points (x, y) ∈ R+n × R+n \ diag.
β
Moreover, if for any given multi-indices α, β ∈ N0n we write ∂Xα ∂Y G L to denote the
derivative of order α in the first variable together with the derivative of order β in
the second variable of the function G L (·, ·), then for any multi-indices α, β ∈ N0n
such that |α| + | β| > 0 there exists some constant C = C(n, L, κ, α, β, K, y) ∈ (0, ∞)
for which 
R n \K β C
Nκ + (∂Xα ∂Y G L )(·, y) (x ) ≤ |α |+ |β |
. (4.6.8)
1 + |x | n−2+

β
For each α, β ∈ N0n we also have that the function ∂Xα ∂Y G L is translation invariant
in the tangential variables, in the sense that
 β    β 
∂Xα ∂Y G L x − (z , 0), y − (z , 0) = ∂Xα ∂Y G L (x, y)
(4.6.9)
for each (x, y) ∈ R+n × R+n \ diag and z ∈ Rn−1,

and is positive homogeneous, in the sense that

 β   β 
∂Xα ∂Y G L (λx, λy) = λ2−n− |α |− |β | ∂Xα ∂Y G L (x, y)
for each x, y ∈ R+n with x  y and λ ∈ (0, ∞), (4.6.10)
provided either n ≥ 3, or |α| + | β| > 0.

Finally, if G L (·, ·) denotes the (unique) Green function for the transpose system L 
in R+n , then

!
G L (x, y) = G L (y, x) , ∀(x, y) ∈ R+n × R+n \ diag. (4.6.11)

Then Theorem 4.4.1 (in the version when the conditions in (4.4.3) are assumed)
specialized to this setting gives that any function u satisfying, for some aperture
parameter κ > 0,

⎧
⎪
M
u ∈ 𝒞∞ (R+n ) , Lu = 0 in R+n,
⎪
⎪
⎪
⎪
⎨ u κ−n.t. exists at L n−1 -a.e. point in Rn−1,
⎪
∂R+n (4.6.12)
⎪
⎪ ∫
⎪
⎪   dx
⎪
⎪ Nκ u (x ) < +∞,
⎩ Rn−1 1 + |x | n−1
has the Poisson integral representation formula
820 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems
∫
 κ−n.t. 
u(x) = PtL (x − y ) u ∂Rn (y ) dy
+
R n−1
 κ−n.t.  
= PtL ∗ u ∂Rn (x ), ∀x = (x , t) ∈ R+n, (4.6.13)
+

 L 
where P L = Pβα 1≤β,α ≤M is the Agmon-Douglis-Nirenberg Poisson kernel for
the system L in R+n (cf. [2], [3], and the discussion in [95, Theorem 2.4]) and
PtL (x ) := t 1−n P L (x /t) for each x ∈ Rn−1 and t > 0.
κ−n.t.
In particular, (4.6.13) shows that if u ∂Rn vanishes at L n−1 -a.e. point in Rn−1 then
+
actually u ≡ 0 in R+n , which is a uniqueness result established in [95, Theorem 3.2,
p. 935] (via a proof which does not yield an actual integral representation formula).
We also wish to remark that the integrability condition in the last line of (4.6.12)
cannot be weakened to membership to L 1,∞ (Rn−1, L n−1 ). This may be seen by
considering the case when n := 2, L := Δ, and the nonzero harmonic function
u(x , t) := t/(t 2 + (x )2 ) for each (x
 , t)  ∈ R+ whose nontangential trace vanishes
2

L 1 -a.e. on ∂R2+ (since, in this case, Nκ u (x ) ≈ |x | −1 uniformly for x ∈ R ≡ ∂R2+ ).

This points to the sharpness of the hypotheses made in (4.6.12).  αβ 
To justify (4.6.13), having fixed a coefficient tensor A = a jk 1≤ j,k ≤n which
1≤α,β ≤M
permits writing L as L A, the system associated with A as in (1.3.15), and denoting
by (uα )1≤α ≤M the scalar components of u, we may write
" ∫ #
 κ−n.t.   A L    
u(x) = − uα ∂Rn (y ) ∂ν G . β α (y , 0), x dy
+
R n−1
1≤β ≤M
"∫ #
 κ−n.t.  γα L  
= uα ∂R+n
(y )ann (∂Xn Gγβ ) (y , 0), x dy
R n−1
1≤β ≤M
"∫ #
 κ−n.t.  γα L  
= uα ∂R+n
(y )ann (∂Yn G βγ ) x, (y , 0) dy . (4.6.14)
R n−1
1≤β ≤M

The fact that (4.6.12) implies the first equality in (4.6.14) is a consequence of (4.4.4)
in Theorem 4.4.1 (bearing in mind the versions of (4.6.1)-(4.6.2) and (4.6.4)-(4.6.5)

for G L ). The second equality in (4.6.14) is seen from the formula for the conormal
derivative (cf. Definition 1.7.1), the fact that the outward unit normal to R+n is
 
ν := (0, . . . , 0, −1), and the observation that (∂Xk G ) (y , 0), x = 0 whenever
L

k < n since G L (·, x) vanishes (in a smooth fashion) on Rn−1 × {0} (cf. (4.6.2)
and (4.6.5)). Finally, the last equality in (4.6.14) is based on (4.6.11). Once (4.6.14)
has been justified, the Poisson integral representation formula (4.6.13) follows on
account of (4.6.9)-(4.6.10) and a formula from [95, Theorem A.4] to the effect that

L (z ) = aγα ∂ G L
 
Pβα nn Yn βγ (z , 1), 0 , ∀z ∈ Rn−1,
(4.6.15)
for each β, α ∈ {1, . . . , M }.
4.6 The Poisson Kernel Associated with a System in the Upper Half-Space 821

Finally, as regards the nature L

 of the Poisson kernel P in the case when the fun-
damental solution E = Eγβ 1≤γ,β ≤M associated with the system L as in Theo-
rem 1.4.2 happens to be a radial function, we also wish to mention that for each
β, α ∈ {1, . . . , M } we have (again, see [95, Theorem A.4])
γα
L
Pβα (z ) = 2ann (∂n Eβγ )(z , 1), ∀z ∈ Rn−1 . (4.6.16)

Remarkably, the Poisson integral representation formula (4.6.13) continues to

hold even when the existence of the pointwise nontangential boundary trace is no
longer assumed to begin with, since it turns out that this is actually implied by the
other conditions imposed in (4.6.12). Specifically, having

⎧
⎪
M
u ∈ 𝒞∞ (R+n ) , Lu = 0 in R+n,
⎪
⎨
⎪
∫ (4.6.17)
⎪   dx
⎪
⎪ Nκ u (x ) < +∞,
⎩ Rn−1 1 + |x | n−1
implies that
⎧
⎪
κ−n.t.
⎪
⎪ u ∂Rn exists at L n−1 -a.e. point in Rn−1,
⎪
⎪ +
⎪
⎪ ! M
⎨ κ−n.t.
⎪ dx
u ∂Rn belongs to L 1 Rn−1, , (4.6.18)
⎪
⎪ + 1 + |x | n−1
⎪
⎪
⎪
⎪
⎪  κ−n.t.  
⎪ u(x , t) = PtL ∗ u ∂Rn (x ) for each (x , t) ∈ R+n .
⎩ +

This refines [95, Theorem 3.2, p. 935] and [95, Theorem 6.1, p. 956]. In fact, the
result just mentioned interfaces tightly with the topic of boundary value problems
for strongly elliptic systems in the upper half-space considered in [95]. Indeed, in
view of the assumptions made in (4.6.17) it is natural to consider the linear space
   M   M
𝒵 := f ∈ L 1 Rn−1, 1+ |xdx| n−1 : MRn−1 f ∈ L 1 Rn−1, 1+ |xdx| n−1
   
M
= f : Rn−1 → C M : measurable and MRn−1 f ∈ L 1 Rn−1, 1+ |xdx| n−1 ,
(4.6.19)

where MRn−1 is the Hardy-Littlewood maximal operator in Rn−1 , equipped with the
norm

 f 𝒵 :=  f [L 1 (Rn−1, dx
)] M + MRn−1 f [L 1 (Rn−1, dx
)] M
1+| x | n−1 1+| x | n−1

≈ MRn−1 f [L 1 (Rn−1, dx

)] M , ∀ f ∈ 𝒵. (4.6.20)
1+| x | n−1

  M
Then [95, Theorem 1.1, p. 915], with X := 𝒵 and Y := L 1 Rn−1, 1+ |xdx| n−1 ,
gives that the following boundary-value problem is well posed:
822 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

⎧
⎪ M
u ∈ 𝒞∞ (R+n ) , Lu = 0 in R+n,
⎪
⎪
⎪
⎪
⎨∫
⎪
⎪   dx
Nκ u (x ) < +∞, (4.6.21)
⎪
⎪ R n−1 1 + |x | n−1
⎪
⎪ κ−n.t.
⎪
⎪
⎪ u ∂Rn = f ∈ 𝒵.
⎩ +

The relevance of the fact that (4.6.17) implies (4.6.18) in the context of (4.6.21) is that
κ−n.t.
the nontangential boundary trace u ∂Rn is guaranteed to exist by the other conditions
+
imposed on the function u in the formulation of the aforementioned boundary value
problem, and that the solution may be recovered from the boundary datum via
convolution with the Poisson kernel canonically associated with the system L.
To prove that (4.6.17) implies (4.6.18), for each ε > 0 consider
M
uε ∈ 𝒞∞ (R+n ) given by uε (x) := u(x , xn + ε)
(4.6.22)
at each point x = (x , xn ) ∈ R+n,

and note that this function satisfies the conditions stipulated in (4.6.12). As such,
we may write the Poisson integral representation formula (4.6.13) for each uε and
obtain that
∫
u(x , t + ε) = PtL (x − y ) fε (y ) dy for each x = (x , t) ∈ R+n, (4.6.23)
R n−1

with fε := u(·, ε) : Rn−1 → C M for each ε > 0. If we also consider the weight
υ : Rn−1 → (0, ∞) defined as υ(x ) := 1+ |x1 | n−1 for each x ∈ Rn−1 , then the last
condition in (4.6.17) entails
 
sup | fε | ≤ Nκ u ∈ L 1 Rn−1, υ L n−1 . (4.6.24)
ε>0

To proceed, we need a suitable weak compactness result, so we momentarily digress

to elaborate on this issue.

Lemma 4.6.1 Let υ : Rn−1 → (0, ∞) be a Lebesgue measurable function and

consider a sequence { f j } j ∈N ⊂ L 1 (Rn−1 , υL n−1 ) such that

F := sup | f j | ∈ L 1 (Rn−1 , υL n−1 ). (4.6.25)

j ∈N
 
Then there exists a subsequence f jk k ∈N of { f j } j ∈N together with some function
f ∈ L 1 (Rn−1 , υL n−1 ) with the property that
∫ ∫
f jk ϕ υ dL n−1 −→ f ϕ υ dL n−1 as k → ∞,
R n−1 R n−1 (4.6.26)
for every function ϕ ∈ 𝒞0 (Rn−1 ) ∩ L ∞ (Rn−1, L n−1 ).
4.6 The Poisson Kernel Associated with a System in the Upper Half-Space 823

Proof Begin by fixing an arbitrary function ϕ ∈ 𝒞 0 (Rn−1 ) ∩ L ∞ (Rn−1, L n−1 ). Also,

having picked some function θ ∈ 𝒞∞ c (R
n−1 ) with the property that 0 ≤ θ ≤ 1 and

θ ≡ 1 near the origin, for each R > 0 define θ R (x ) := θ(x /R) for every x ∈ Rn−1 .
In particular, for each fixed R > 0 the function θ R ϕ is continuous with compact
support. Keeping
 this in mind, [95, Lemma 6.2, p. 956] implies that there exists a
subsequence f jk k ∈N of { f j } j ∈N and a function f ∈ L 1 (Rn−1 , υL n−1 ) such for each
fixed R > 0 we have
∫ ∫
θ R f jk ϕυ dL n−1
−→ θ R f ϕυ dL n−1 as k → ∞. (4.6.27)
R n−1 R n−1

Next, fix an arbitrary threshold δ > 0. Since F, f ∈ L 1 (Rn−1 , υL n−1 ) while ϕ is

continuous and bounded, Lebesgue’s Dominated Convergence Theorem guarantees
that there exists Rδ ∈ (0, ∞) with the property that
∫
(1 − θ R )|F ||ϕ|υ dL n−1 < δ whenever R > Rδ, (4.6.28)
R n−1

and ∫
(1 − θ R )| f ||ϕ|υ dL n−1 < δ whenever R > Rδ . (4.6.29)
R n−1
For each R > Rδ we may then rely on (4.6.27)-(4.6.29) to write
∫ ∫
lim sup f jk ϕ υ dL n−1
− f ϕυ dL n−1
k→∞ R n−1 R n−1
∫ ∫
≤ lim sup θ R f jk ϕυ dL n−1 − θ R f ϕυ dL n−1
k→∞ R n−1 R n−1
∫
+ lim sup (1 − θ R )| f jk ||ϕ|υ dL n−1
k→∞ R n−1
∫
+ (1 − θ R )| f ||ϕ|υ dL n−1
R n−1
∫ ∫
≤ (1 − θ R )|F ||ϕ|υ dL n−1 + (1 − θ R )| f ||ϕ|υ dL n−1
R n−1 R n−1

≤ 2δ. (4.6.30)

In view of the arbitrariness of δ > 0, this establishes the claim made in (4.6.26). 
Returning to the mainstream discussion,
 the weak-∗ convergence result from
Lemma 4.6.1 may be used for the sequence fε ε>0 ⊂ L 1 Rn−1, υ L n−1 to conclude
 
that there exist f ∈ L 1 Rn−1, υ L n−1 and a sequence {ε j } j ∈N ⊂ (0, ∞) which
converges to zero with the property that
824 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems
∫ ∫
dy dy
lim ϕ(y ) fε j (y ) = ϕ(y ) f (y ) (4.6.31)
j→∞ R n−1 1 + |y | n−1 R n−1 1 + |y | n−1

for every bounded continuous function ϕ : Rn−1 → C M×M . The fact that there exists
a constant C ∈ (0, ∞) for which
C
|P L (z )| ≤ for each z ∈ Rn−1 (4.6.32)
1 + |z | n

(cf. [95, Theorem 2.4, p. 934]) ensures that for each fixed point (x , t) ∈ R+n the
assignment

Rn−1  y → ϕ(y ) := PtL (x − y )(1 + |y | n−1 ) ∈ C M×M

(4.6.33)
is a bounded continuous matrix-valued function.

At this stage, from (4.6.23) and (4.6.31) used for the function ϕ defined in (4.6.33)
we obtain (bearing in mind that u is continuous in R+n ) that
∫
u(x , t) = PtL (x − y ) f (y ) dy for each x = (x , t) ∈ R+n . (4.6.34)
R n−1

  dx 
With this in hand, and since L 1 Rn−1, υ L n−1 ⊆ L 1 Rn−1, , we may
1 + |x | n
invoke [95, Theorem 3.1, p. 934] to conclude that22
κ−n.t.
u ∂Rn exists and equals f at L n−1 -a.e. point in Rn−1 . (4.6.35)
+

Once this has been established, all conclusions in (4.6.18) are implied by (4.6.12)-
(4.6.13).
Lastly, we wish to remark that even in the classical case when L := Δ, the
Laplacian in Rn , our Poisson integral representation formula (4.6.17)-(4.6.18) is
more general (in the sense that it allows for a larger class of functions) than the
existing results in the literature. Indeed, the latter typically assume an L p integrability
condition for the harmonic function which, in the range 1 < p < ∞, implies
our weighted L 1 integrability condition for the nontangential maximal function
demanded in (4.6.17). In this vein see, e.g., [48, Theorems 4.8-4.9, pp. 174-175],
[166, Corollary, p. 200], [167, Proposition 1, p. 119].

4.7 More on Uniqueness and Poisson Integral Representations

Here we shall consider the Poisson integral operator for a given second-order weakly
elliptic system, in the context described below.

22 convergence holds, for instance, in the set of Lebesgue points of f

4.7 More on Uniqueness and Poisson Integral Representations 825

Theorem 4.7.1 Let Ω ⊂ Rn , where n ∈ N satisfies23 n ≥ 3, be a UR domain which is

locally pathwise nontangentially accessible (in the sense of [112, Definition 8.9.14]),
and set σ := H n−1 ∂Ω. Consider a homogeneous constant (complex) coefficient
second-order M × M system L in Rn , which is weakly elliptic (in the sense of
(1.3.3)). Also, fix an aperture parameter κ > 0 and assume that the spaces 𝒳, 𝒴, 𝒵
are as in any of the following cases:
M
, 𝒵 := 𝒳∗ = L p (∂Ω, σ),
p
𝒳 := L p (∂Ω, σ), 𝒴 := L1 (∂Ω, σ)
(4.7.1)
where p, p ∈ (1, ∞) are such that 1/p + 1/p = 1,
or
M
, 𝒵 := 𝒳∗ = L p (∂Ω, wσ),
p
𝒳 := L p (∂Ω, w 1−p σ), 𝒴 := L1 (∂Ω, w 1−p σ)
where p, p ∈ (1, ∞) are such that 1/p + 1/p = 1 and w ∈ Ap (∂Ω, σ),
(4.7.2)
or
𝒳 := L p , q (∂Ω, σ), 𝒵 := 𝒳∗ = L p,q (∂Ω, σ), and
 M M 
𝒴 := f ∈ L p , q (∂Ω, σ) : ∂τ j k f ∈ L p , q (∂Ω, σ) , 1 ≤ j, k ≤ n ,
where p, p , q, q ∈ (1, ∞) are such that 1/p + 1/p = 1 = 1/q + 1/q ,
(4.7.3)
or
p ,λ M
𝒳 := M p ,λ (∂Ω, σ), 𝒴 := M1 (∂Ω, σ) , 𝒵 := 𝒳∗ = B p,λ (∂Ω, σ),
(4.7.4)
where p, p ∈ (1, ∞) are such that 1/p + 1/p = 1,
or
p ,λ M
𝒳 := B p ,λ (∂Ω, σ), 𝒴 := B1 (∂Ω, σ) , 𝒵 := M p,λ (∂Ω, σ),
(4.7.5)
where p, p ∈ (1, ∞) are such that 1/p + 1/p = 1.

Then if the Regularity Problem for the transpose system L  , formulated as

⎧
⎪ M
w ∈ 𝒞∞ (Ω) , L  w = 0 in Ω,
⎪
⎪
⎪
⎪
⎪
⎪ Nκ (∇w) belong to the space 𝒳,
⎪
⎪
⎨
⎪ κ−n.t.
w ∂Ω = g at σ-a.e. point on ∂Ω, (4.7.6)
⎪
⎪
⎪
⎪
⎪ and also w(x) = o(1) as |x| → ∞
⎪
⎪
⎪
⎪
⎪ when Ω is an exterior domain,
⎩

23 see the last part in the statement for the case n = 2

826 4 Green Functions and Uniqueness for Boundary Problems for Second-Order Systems

has a solution for each boundary datum g ∈ 𝒴, and if the Dirichlet Problem for the
original system L, formulated as
⎧
⎪ M
u ∈ 𝒞∞ (Ω) , Lu = 0 in Ω,
⎪
⎪
⎪
⎪
⎪
⎪ Nκ u belongs to the space 𝒵,
⎪
⎪
⎨
⎪ κ−n.t.
u ∂Ω = f at σ-a.e. point on ∂Ω, (4.7.7)
⎪
⎪
⎪
⎪
⎪ and also u(x) = o(1) as |x| → ∞
⎪
⎪
⎪
⎪
⎪ when Ω is an exterior domain,
⎩
M
has a solution for each boundary datum f ∈ 𝒵 , then for each x ∈ Ω there exists
a unique function G(·, x) satisfying (with all derivatives, as well as the nontangential
maximal function and the nontangential traces, taken in the “dot” variable)
M×M
G(·, x) ∈ 𝒞∞ (Ω \ {x}) ∩ Wloc
1,1
(Ω) ,
M×M
L  G(·, x) = −δx I M×M in D (Ω) ,
   
NκΩ\K x ∇G(·, x) ∈ 𝒳 where Kx := B x, 12 dist(x, ∂Ω) ,
  κ−n.t. (4.7.8)
∇G(·, x) ∂Ω exists at σ-a.e. point on ∂Ω,
κ−n.t.
G(·, x) ∂Ω = 0 at σ-a.e. point on ∂Ω,
and also G(y, x) = o(1) as |y| → ∞,
in the case when Ω is an exterior domain.

Furthermore, if in place of the solvability of (4.7.6) one now assumes that the
following type of Regularity Problem
⎧
⎪ M
w ∈ 𝒞∞ (Ω) , L  w = 0 in Ω,
⎪
⎪
⎪
⎪
⎪
⎪ Nκ w and Nκ (∇w) belong to 𝒳,
⎪
⎪
⎨
⎪ κ−n.t.
w ∂Ω = g at σ-a.e. point on ∂Ω, (4.7.9)
⎪
⎪
⎪
⎪
⎪
⎪ and also w(x) = o(1) as |x| → ∞
⎪
⎪
⎪
⎪ when Ω is an exterior domain,
⎩
is solvable for each boundary datum g ∈ 𝒴, then in addition to the properties listed
in (4.7.8) one also has  
NκΩ\K x G(·, x) ∈ 𝒳. (4.7.10)
Moreover, if P L (x, y) is the Poisson kernel associated as in (4.5.2)-(4.5.3) with
the Green function G from (4.7.8), then the Poisson integral operator for the system
M
L, defined for each function f ∈ 𝒵 as
4.7 More on Uniqueness and Poisson Integral Representations 827
∫
(PI L f )(x) := P L (x, y) f (y) dσ(y), ∀x ∈ Ω, (4.7.11)
∂Ω

is precisely the unique solution of the Dirichlet Problem (4.7.7) with boundary datum
f.
Finally, the same results are also valid in the case n = 2 provided either Ω is
bounded, or ∂Ω is unbounded. Also, if n = 2 and Ω is an exterior domain, the same
results are true provided one assumes
∫
−1
L(ξ) dH 1 (ξ)
S 1 (4.7.12)
is an invertible M × M matrix

(which is always the case if the system L is assumed to actually satisfy the Legendre-
Hadamard strong ellipticity condition, and if M = 1 this is true if and only if
L = ∇ · A∇ for some A ∈ M20 ; cf. Lemma 1.4.19 and (1.4.186)), and with the demand
that u and G(·, x) are o(1) at infinity replaced by

u = O(1) and G(·, x) = O(1) at infinity. (4.7.13)

Proof Granted the solvability of the Regularity Problem for L  formulated in (4.7.6),
we may construct a Green function for L  as in the proof of Theorem 4.3.1 (cf.
(4.3.14)). For each fixed pole x ∈ Ω, this yields a function G(·, x) satisfying all but
the fourth condition in (4.7.8). That said condition also holds in the present setting is
then seen by applying the Fatou-type result from Theorem 3.3.5 to the columns of the
 
matrix-valued function G(·, x) restricted to Ω \ B x, 12 dist(x, ∂Ω) . This ultimately
  κ−n.t.
shows that ∇G(·, x) ∂Ω exists at σ-a.e. point on ∂Ω, completing the proof of the
existence of a Green function as in (4.7.8).
 x) is another function enjoying all properties
To prove its uniqueness, suppose G(·,
 x) satisfies L  w = 0 in Ω, hence w is
listed in (4.7.8). Then w := G(·, x) − G(·,
smooth in Ω, thanks to elliptic regularity (cf. [112, (6.5.40) in Theorem 6.5.7]).
κ−n.t.
Also, w ∂Ω = 0 at σ-a.e. point on ∂Ω and Nκ (∇w) ∈ 𝒳. Given that we are
assuming the solvability of the Dirichlet Problem (4.7.7), from Theorem 4.3.2 (used
with u := w and L  in place of L) we conclude that w ≡ 0 in Ω. This finishes the
proof of the existence of a unique Green function as in (4.7.8). Next, if in place of
the solvability of (4.7.6) one now assumes the solvability of (4.7.9) for each g ∈ 𝒴,
then from (4.3.17) we see that (4.7.10) holds.
The fact that the solution of the Dirichlet Problem (4.7.7) is unique is implied by
the solvability of the Regularity Problem (4.7.6) (cf. Theorem 4.3.1). Granted this,
Theorem 4.4.1 (cf. (4.5.4)) implies that the unique solution of the Dirichlet Problem
M
(4.7.7) with boundary datum f ∈ 𝒵 is given by (4.7.11). 
Chapter 5
Green Functions and Poisson Kernels for the
Laplacian

Good references for background material pertaining to the topics in this chapter are
[71], [49], [51], [58], [79]. The developments here are motivated by questions in
potential theory for the Laplacian in Rn of the following sort:
• When is the Poisson kernel associated with a domain Ω ⊆ Rn (as the Radon-
Nikodym derivative of the harmonic measure with respect to the surface measure)
well-defined and equal to the (minus) normal derivative of the Green function?
• When is said Poisson kernel bounded from below away from zero?
• When can one represent the solution of the Dirichlet Problem with L p data as
an integral involving the Poisson kernel?
Of course, having Ω sufficiently smooth suffices, but we are interested in minimal
regularity conditions. This is the perspective through which such questions have
been raised, for example, by J. Garnett and D.E. Marshall in [49].

5.1 Upper/Lower Semicontinuous Functions and Super/Sub

Harmonic Functions

This section contains a brief review of some basic properties of upper-semicontinuous

and lower-semicontinuous functions, as well as superharmonic and subharmonic
functions, meant to facilitate the subsequent discussion.
Let (X, τ) be a topological space. For each x ∈ X denote by 𝒜x the collection of
all neighborhoods of x, and let 𝒪x ⊆ 𝒜x be the collection of all open neighborhoods
of x (in the topology τ). Given a function f : X → [−∞, +∞], for each x ∈ X define
   
lim inf f (y) := sup A∈𝒜 x inf y ∈ A f (y) = supO ∈𝒪 x inf y ∈O f (y)
y→x
    (5.1.1)
lim sup f (y) := inf A∈𝒜 x supy ∈ A f (y) = inf O ∈𝒪 x supy ∈O f (y) .
y→x

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 829
D. Mitrea et al., Geometric Harmonic Analysis III, Developments in Mathematics 74,
https://doi.org/10.1007/978-3-031-22735-6_5
830 5 Green Functions and Poisson Kernels for the Laplacian

A function f : X → (−∞, +∞] is said to be lower-semicontinuous at x ∈ X if

lim inf f (y) ≥ f (x). Then
y→x

f is lower-semicontinuous at x ∈ X if and only if

(5.1.2)
{ f > a} is a neighborhood of x whenever a < f (x).
Also,
if the function f is lower-semicontinuous
 at the point x ∈ X
then f (x) ≤ lim inf f (x j ) := sup inf f (x j ) for any sequence
j→∞ k ∈N j ≥k
(5.1.3)
{x j } j ∈N ⊆ X such that lim x j = x in τ; moreover, the converse is
j→∞
also true provided the topology τ is metrizable.

Simply call f : X → (−∞, +∞] lower-semicontinuous if it is lower-semicontinuous

at each point x ∈ X. Then

f is lower-semicontinuous if and only if { f > a} is open for any

(5.1.4)
a ∈ R or, equivalently, { f ≤ a} is closed for any a ∈ R.
In particular,

if (X, τ) is a topological space and f : X → (−∞, +∞] is lower-

semicontinuous, then for every set Y ⊆ X the restriction f Y is a (5.1.5)
 
lower-semicontinuous function on Y, τ Y .

It is also useful to note that

if the topology τ on X is metrizable and f : X → (−∞, +∞] is
lower-semicontinuous, then on each compact set f is bounded from (5.1.6)
below and its infimum (allowed to be +∞) is attained.
To justify this, fix K ⊆ X compact and abbreviate m := inf x ∈K f (x) ∈ [−∞, +∞].
Then there exists a sequence {x j } j ∈N ⊆ K such that f (x j )  m as j → ∞. Since τ
is metrizable, it follows that we can select a subsequence {x jk }k ∈N which converges
in τ to some point x∗ ∈ K. Then the property recorded in (5.1.3) permits us to write
m ≤ f (x∗ ) ≤ lim inf f (x jk ) = m. Consequently, m = f (x∗ ) > −∞, proving (5.1.6).
k→∞

Remark 5.1.1 If (X, τ) is a metrizable topological space, μ is a locally finite Borel

measure on X, and f : X → (−∞, +∞] is a lower-semicontinuous function, then for
each compact set K ⊆ X the integral
∫
f dμ is well defined in (−∞, +∞]. (5.1.7)
K

Indeed, based (5.1.4) we deduce that f is μ-measurable on K, while (5.1.6)

ensures that there exists m ∈ R such that m ≤ f (x) for each x ∈ K. Then we set
5.1 Upper/Lower Semicontinuous Functions and Super/Sub Harmonic Functions 831
∫ ∫
f dμ := ( f − m) dμ + m · μ(K) ∈ (−∞, +∞], (5.1.8)
K K
∫
where K ( f − m) dμ ∈ [0, +∞] is the integral of the non-negative, μ-measurable
function f − m on K.

Lemma 5.1.2 Assume (X, d) is a metric space, and let f : X → (−∞, +∞] be lower-
semicontinuous and bounded from below. Then there exists a sequence { f j } j ∈N of
Lipschitz functions on X with the property that for each x ∈ X one has f j (x) f (x)
as j → ∞.

Proof If f ≡ +∞ then taking f j (x) := j for each j ∈ N and each x ∈ X will do.
Consider next the case when f  +∞. In such a scenario, define
 
f j (x) := inf f (y) + j · d(x, y) : y ∈ X for each j ∈ N and each x ∈ X. (5.1.9)

Since f is bounded from below and the distance function is non-negative, it follows
that the assignment X y → f (y) + j · d(x, y) is bounded from below. Hence, each
f j (x) is well defined in R and

f1 (x) ≤ f2 (x) ≤ · · · ≤ f (x) for each x ∈ X. (5.1.10)

Also, we claim that for each j ∈ N fixed, the function f j is Lipschitz. Indeed, from
definitions, f j (z) ≤ f (y) + j · d(z, y) for each z, y ∈ X, and given any x ∈ X along
with ε > 0 there exists some yε ∈ X such that

f j (x) + ε > f (yε ) + j · d(x, yε ). (5.1.11)

Combining these estimates yields

f j (x) + ε > f (yε ) + j · d(x, yε )

   
≥ f j (z) − j · d(z, yε ) + j d(z, yε ) − d(x, z)
= f j (z) − j · d(x, z). (5.1.12)

By letting ε → 0+ we arrive at f j (x) ≥ f j (z) − j · d(x, z) which, after interchanging

z and x, ultimately shows that | f j (x) − f j (z)| ≤ j · d(x, z). This proves that, indeed,
each f j is Lipschitz.
We claim that lim f j (x) = f (x) for each x ∈ X. To prove this, fix an arbitrary
j→∞
x ∈ X along with an arbitrary ε > 0. Then for each j ∈ N there exists x j,ε ∈ X such
that
f j (x) ≥ f (x j,ε ) + j · d(x, x j,ε ) − ε. (5.1.13)
To proceed, assume lim x j,ε = x. In such a scenario, using (5.1.10) and the lower-
j→∞
semicontinuity of f (cf. (5.1.3)) we may write
832 5 Green Functions and Poisson Kernels for the Laplacian
 
f (x) ≥ lim sup f j (x) ≥ lim inf f j (x) ≥ lim inf f (x j,ε ) + j · d(x, x j,ε ) − ε
j→∞ j→∞ j→∞

≥ lim inf f (x j,ε ) − ε ≥ f (x) − ε. (5.1.14)

j→∞
 
This ultimately shows that lim f j (x) = f (x) in this case. Finally, suppose x j,ε j ∈N
j→∞
does not converge to x. By eventually passing to a subsequence, we may assume that
there exist a strictly increasing sequence { jk }k ∈N ⊆ N and some δ > 0 such that
d(x, x jk ,ε ) ≥ δ for each k ∈ N. Let m be a lower bound for f . On account of (5.1.10)
and (5.1.13) we may then write

f (x) ≥ f jk (x) ≥ f (x jk ,ε ) + jk · d(x, x jk ,ε ) − ε ≥ m + jk · δ − ε. (5.1.15)

Passing to limit as k → ∞ then forces lim f j (x) = f (x) = +∞ in this case. 

j→∞

In turn, Lemma 5.1.2 permits us to establish the following useful characterization

of lower-semicontinuity.

Proposition 5.1.3 Let (X, τ) be a metrizable locally compact topological space, and
consider a function f : X → (−∞, +∞]. Then the following are equivalent:
(1) the function f is lower-semicontinuous on X;
(2) for each compact set K ⊆ X there exists a sequence { f j } j ∈N of continuous
functions on K such that f j (x) f (x) as j → ∞ for each x ∈ K.
  
Proof To prove that (1) ⇒ (2), fix a compact set K ⊆ X. Then K, τ  is a compact,
 K

metrizable topological space and f K is lower-semicontinuous function which is
bounded from below on K, thanks to (5.1.5) and (5.1.6). Granted this, Lemma 5.1.2
applies and the desired conclusion follows. Consider the implication (2) ⇒ (1). Fix
an arbitrary point x ∈ X along with a sequence {xi }i ∈N ⊆ X which converges to x
in the topology τ. We want to show that

f (x) ≤ lim inf f (xi ). (5.1.16)

i→∞

Since (X, τ) is locally compact, there exists a compact neighborhood K of x. Given

the nature of the conclusion we presently seek, there is no loss of generality in
assuming that {xi }i ∈N ⊆ K. The working hypotheses guarantee that there exists a
sequence { f j } j ∈N of continuous functions on K such that f j f pointwise on K as
j → ∞. Then for each j ∈ N we may write

lim inf f (xi ) ≥ lim inf f j (xi ) = f j (x), (5.1.17)

i→∞ i→∞

given that f ≥ f j on K and f j is continuous at x. Since f j (x) f (x) as j → ∞,

passing to limit as j → ∞ in (5.1.17) yields (5.1.16). Having proved (5.1.16), we
conclude from (5.1.3) that the function f is indeed lower-semicontinuous on X. 
 5.1 Upper/Lower Semicontinuous Functions and Super/Sub Harmonic Functions 833

Given topological space (X, τ), call a function f : X → [−∞, +∞) upper-
semicontinuous at a point x ∈ X provided lim sup f (y) ≤ f (x). Also, f is simply
y→x
called upper-semicontinuous if it is upper-semicontinuous at each point x ∈ X.
Since f is upper-semicontinuous if and only if − f is lower-semicontinuous, similar
properties to (5.1.2)-(5.1.6), Remark 5.1.1, Lemma 5.1.2, and Proposition 5.1.3 are
valid in the class of upper-semicontinuous functions.

Definition 5.1.4 Assume Ω ⊆ Rn is an arbitrary nonempty open set. A given function

u : Ω → (−∞, +∞] is called superharmonic if the following conditions are
satisfied:
(i) u is lower-semicontinuous;
(ii) given any x ∈ Ω together with 0 < r < dist(x, ∂Ω) and a real-valued continuous
function h on B(x, r) that is harmonic in B(x, r) and satisfies h(y) ≤ u(y) for
all y ∈ ∂B(x, r), one has h(y) ≤ u(y) for all y ∈ B(x, r);
(iii) u  +∞.
Also, call u : Ω → [−∞, +∞) subharmonic if −u is superharmonic.

It turns out that there is quite a bit of flexibility in the manner in which condition
(ii) is formulated. This is made precise in the proposition below.

Proposition 5.1.5 Let Ω be an open set in Rn and consider a lower-semicontinuous

function u : Ω → (−∞, +∞]. Then condition (ii) in Definition 5.1.4 is further
equivalent with each of the following properties (all integrals are considered in the
sense of Remark 5.1.1):
(ii.a) given any compact set K ⊆ Ω together with a real-valued continuous function h
on K that is harmonic in K̊ and satisfying h(y) ≥ u(y) for all y ∈ ∂K, one has
h(y) ≥ u(y) for all y ∈ K; ⨏
(ii.b) given any x ∈ Ω and 0 < r < dist(x, ∂Ω) one has u(x) ≥ u dH n−1 ;
⨏∂B(x,r)
(ii.c) given any x ∈ Ω and 0 < r < dist(x, ∂Ω) one has u(x) ≥ u dL n .
B(x,r)

Proof The implication (ii.a) ⇒ (ii) is obvious. To justify the implication (ii) ⇒ (ii.b),
fix x0 ∈ Ω along with 0 < r < dist(x, ∂Ω). Then Proposition 5.1.3 guarantees the
existence of a sequence { f j } j ∈N of continuous functions on B(x0, r) such that
f j (x) u(x) as j → ∞ for each x ∈ B(x0, r). For each j ∈ N, denote by P[ f j ]
the Poisson integral of f j ∂B(x0,r) in the ball B(x0, r). Since each P[ f j ] is harmonic
in B(x0, r), continuous on B(x0, r), and satisfies P[ f j ](y) = f j (y) ≤ u(y) for
all y ∈ ∂B(x, r), item (ii) implies that P[ f j ] ≤ u in B(x0, r) for each j ∈ N.
Consequently,
⨏ ⨏
u(x0 ) ≥ P[ f j ](x0 ) = f j dH n−1 u dH n−1 as j → ∞,
∂B(x0,r) ∂B(x0,r)
(5.1.18)
834 5 Green Functions and Poisson Kernels for the Laplacian

by Lebesgue’s Monotone Convergence Theorem. This establishes (ii.b). Next, the

implication (ii.b) ⇒ (ii.c) is a consequence of the “spherical” Fubini formula (cf.
[109, Theorem 14.63, p. 577]).
At this stage, there remains to prove that (ii.c) ⇒ (ii.a). To this end, let K ⊆ Ω
be a compact set and consider a real-valued continuous function h on K that is
harmonic in K̊ and which satisfies h(y) ≤ u(y) for all y ∈ ∂K. Then w := u − h
is a lower-semicontinuous function on K. Also, according to (5.1.6), the function
w is bounded from below and its infimum (allowed to be +∞) is attained, say
inf K w = w(x0 ) ∈ (−∞, +∞], for some point x0 ∈ K. If x0 actually belongs to
∂K, then inf K w = w(x0 ) = u(x0 ) − h(x0 ) ≥ 0. This ultimately implies w ≥ 0
on K, hence h(y) ≤ u(y) for all y ∈ K. On the other hand, if x0  ∂K then x0
belongs to K \ ∂K = K̊, which forces r := dist(x0, ∂K) > 0. Since x0 ∈ Ω and
0 < r < dist(x0, ∂Ω), the current working hypothesis guarantees that
⨏
w(x0 ) ≥ w dL n ≥ inf w = w(x0 ). (5.1.19)
B(x0,r) K

Thus, there exists a set A ⊆ B(x0, r) which is Lebesgue measurable and satisfies
L n (A) = 0, such that w(x) = w(x0 ) for each x ∈ B(x0, r) \ A. Given the choice of
r, we have ∂K ∩ ∂B(x0, r)  , hence there exists x∗ ∈ ∂K ∩ ∂B(x0, r). Choose
a sequence {x j } j ∈N ⊆ B(x0, r) \ A such that lim x j = x∗ . Keeping in mind that
j→∞
x∗ ∈ ∂K and (5.1.3), we see that

0 ≤ w(x∗ ) ≤ lim inf w(x j ) = w(x0 ) = inf w. (5.1.20)

j→∞ K

Hence w ≥ 0 on K from which we finally conclude that u(y) − h(y) = w(y) ≥ 0 for
all y ∈ K, as wanted. 

In relation to Definition 5.1.4 we also wish to remark that if Ω ⊆ Rn is a nonempty

open set and u ∈ 𝒞0 (Ω), then u is subharmonic if Δu ≥ 0 in the sense of distributions
in Ω, i.e.,
∫
uΔϕ dL n ≥ 0 for each non-negative function ϕ ∈ 𝒞∞ c (Ω). (5.1.21)
Ω

To justify this classical fact, assume first that actually u ∈ 𝒞∞ (Ω) and Δu ≥ 0
pointwise in Ω. For some arbitrary fixed x ∈ Ω, consider the function
⨏ ⨏
f (r) := u dH n−1 = u(x + rξ) dH n−1 (ξ)
∂B(x,r) ∂B(0,1)
 
for each r ∈ 0, dist(x, ∂Ω) . (5.1.22)

Using the classical Divergence Theorem we see that this satisfies

5.1 Upper/Lower Semicontinuous Functions and Super/Sub Harmonic Functions 835
⨏
f (r) = ξ · (∇u)(x + rξ) dH n−1 (ξ)
∂B(0,1)
∫ ⨏
1 r
= Δu dL =
n
Δu dL n . (5.1.23)
ωn−1 r n−1 B(x,r) n B(x,r)

Consequently, f (r) ≥ 0 which shows that f is non-decreasing. In turn, this allows

us to estimate ⨏
u(x) = lim+ f (ρ) ≤ f (r) = u dH n−1 (5.1.24)
ρ→0 ∂B(x,r)
 
for each r ∈ 0, dist(x, ∂Ω) , proving that u is subharmonic in this case. In the
general case, when u only belongs to 𝒞0 (Ω) and satisfies (5.1.21), a standard mollifier
argument (involving a smooth non-negative compactly
 supported function) produces

a sequence {uε }ε>0 ⊆ 𝒞∞ (Ωε ), where Ωε := x ∈ Ω : dist(x, ∂Ω) > ε , which
satisfies Δuε ≥ 0 pointwise in Ωε and which converges uniformly to u on compact
subsets of Ω. Based on what we have just proved, each uε satisfies the sub-mean
property. Passing to the limit as ε → 0+ yields that u itself satisfies the sub-mean
property, hence u is subharmonic.
We conclude this review of preliminary matters by recording the following version
of the Maximum Principle.

Theorem 5.1.6 Let Ω ⊆ Rn be bounded open set, and suppose u : Ω → [−∞, +∞)
is a subharmonic function while w : Ω → (−∞, +∞] is a superharmonic function.
Then the following statements are true:
(i) If Ω is connected and u − w has a maximum in Ω then u − w is constant.
(ii) If lim sup u(y) ≤ lim inf w(y) for all x ∈ ∂Ω, then u ≤ w in Ω.
y→x y→x

Proof Working with u − w in place of u, there is no loss of generality in assuming

that w ≡ 0 throughout. To prove the claim in part (i), suppose Ω is connected and
there exists some point x0 ∈ Ω with the property that supΩ u = u(x0 ). In particular,
if we abbreviate M := supΩ u then M ∈ (−∞, +∞). Define
   
A := x ∈ Ω : u(x) < M and B := x ∈ Ω : u(x) = M . (5.1.25)

Since the function u is upper-semicontinuous, the set A is open. Also, by design,

x0 ∈ B, hence B  . Fix 0 < r < dist(x0, ∂Ω). By the sub-mean value property (as
in Proposition 5.1.5 applied to −u), we have
⨏
M = u(x0 ) ≤ u dL n ≤ M. (5.1.26)
B(x0,r)

In turn, this ensures the existence of a Lebesgue measurable set A ⊆ B(x0, r)

satisfying L n (A) = 0 and u(x) = M at each point x ∈ B(x0, r) \ A. Given any
point a ∈ A, it is possible to select a sequence {x j } j ∈N ⊆ B(x0, r) \ A which
836 5 Green Functions and Poisson Kernels for the Laplacian

converges to a. Since u is upper-semicontinuous, (5.1.3) permits us to estimate

M ≥ u(a) ≥ lim sup u(x j ) = M which forces u(a) = M. Thus, u(x) = M for all
j→∞
x ∈ B(x0, r) which goes to shows that B is open as well. In view of the fact that Ω is
connected, this ultimately implies that B = Ω, as wanted.
As regards the claim in part (ii), by working in each connected component of
Ω we may assume that Ω is connected to begin with. Suppose this is the case and
set M := supΩ u. Then there exists a sequence {x j } j ∈N ⊆ Ω with the property
that u(x j ) → M as j → ∞. Since Ω is bounded, so is the sequence {x j } j ∈N .
By Bolzano-Weierstrass’ Theorem, by eventually passing to a subsequence there
is no loss of generality in assuming that {x j } j ∈N converges to a point x∗ ∈ Ω. If
actually x∗ ∈ Ω, then the current part (i) gives that u is constant in Ω and since
lim sup u(y) ≤ 0 for all x ∈ ∂Ω we ultimately conclude that u ≤ 0 in Ω. On
y→x
the other hand, if x∗ ∈ Ω \ Ω = ∂Ω, then the working hypothesis implies that
M = lim sup u(x j ) ≤ 0. This, in turn, once again forces u ≤ 0 in Ω. 
j→∞

5.2 The Harmonic Measure

We start by briefly recalling the construction of harmonic measure associated with

some open subset Ω of Rn which is bounded (the boundedness assumption, although
pervasive in this section, is not essential). Given f : ∂Ω → R define U f , the
upper class of functions associated with f , as the collection of all superharmonic
functions u : Ω → (−∞, +∞] satisfying

inf u > −∞ and lim inf u(x) ≥ f (y) for each point y ∈ ∂Ω, (5.2.1)
Ω Ω x→y

with the agreement that the function u ≡ +∞ in Ω is also contained in U f . The

lower class of functions associated with f is then defined as

L f := {−u : u ∈ U− f }. (5.2.2)

Note that the Maximum Principle from item (ii) in Theorem 5.1.6 ensures that for
each given f ,
w ≤ u for each w ∈ L f and u ∈ U f . (5.2.3)
In particular, if at each x ∈ Ω we define the upper and lower Poisson integral
operators associated with a given function f as

(PI f )(x) := inf {u(x) : u ∈ U f } and (PI f )(x) := sup {w(x) : w ∈ L f }, (5.2.4)

i.e., if PI f and PI f are, respectively, the upper and the lower solution of the gener-
alized Dirichlet Problem for f , then
5.2 The Harmonic Measure 837

(PI f )(x) ≤ (PI f )(x) for each x ∈ Ω. (5.2.5)

Call f a resolutive boundary function if (PI f )(x) = (PI f )(x) at every point
x ∈ Ω and if PI f , PI f are harmonic functions in the set1 Ω. Whenever f is a
resolutive boundary function, its Poisson integral is defined as the harmonic
function
(PI f )(x) := (PI f )(x) = (PI f )(x) for each x ∈ Ω. (5.2.6)
In [180], N. Wiener has proved that

every continuous real-valued function on ∂Ω is resolutive. (5.2.7)

Together with the Maximum Principle and the Riesz Representation Theorem, this
result allows us to define the harmonic measure on ∂Ω. Specifically, (5.2.7) and the
Maximum Principle for harmonic functions (cf. Theorem 5.1.6) ensures that for each
fixed x ∈ Ω the assignment

𝒞0 (∂Ω) f −→ (PI f )(x) ∈ R is a positive linear functional. (5.2.8)

As such, Riesz’s Representation Theorem (cf., e.g., [46, Theorem 7.2, p. 212], [156,
Theorem 2.14, pp. 40-41]) ensures that there exists a sigma-algebra M x which con-
tains all Borel sets on ∂Ω, along with a unique measure ω x : M x → [0, +∞]
satisfying ∫
(PI f )(x) = f dω x, ∀ f ∈ 𝒞0 (∂Ω), (5.2.9)
∂Ω
and which enjoys the following additional properties:

(1) The measure ω x is a complete probability Borel measure on ∂Ω which satisfies

ω x (O) = sup (PI f )(x) : f ∈ 𝒞0 (∂Ω), 0 ≤ f ≤ 1, supp f ⊆ O

(5.2.10)
for each relatively open subset O of ∂Ω,

as well as

ω x (K) = inf (PI f )(x) : f ∈ 𝒞0 (∂Ω), f ≥ 1K

(5.2.11)
for each compact subset K of ∂Ω.

(2) The measure ω x is Radon (cf. [112, Definition 3.5.1]). In fact, ω x satisfies the
outer-regularity property

1 There is a version of this procedure that leads to the upper and the lower solution of the generalized
Dirichlet Problem which are harmonic for each given bounded function f on ∂Ω. This requires
asking that the superharmonic functions in the upper class U f (hence also the subharmonic functions
in the lower class L f ) are continuous on Ω. See, e.g., [51, Theorem 2.12, p. 24] in this regard.
838 5 Green Functions and Poisson Kernels for the Laplacian

ω x (E) = inf ω x (O), ∀E ∈ M x, (5.2.12)

E ⊆O ⊆∂Ω
O relatively open

as well as the inner-regularity property

ω x (E) = sup ω x (K), ∀E ∈ M x . (5.2.13)

K ⊆∂Ω compact
K ⊆E

Other significant properties of the family of measures {ω x }x ∈Ω are presented in

our next lemma.

Lemma 5.2.1 Let Ω be a bounded open set in Rn . Then the following statements are
true:
(i) The sigma-algebra M x is independent of x ∈ Ω. Henceforth, it is therefore
meaningful to abbreviate M := M x for some, or any, x ∈ Ω.
(ii) For each given point x ∈ Ω,

ω x : M −→ [0, 1] is a complete Borel-regular measure. (5.2.14)

(iii) For each x0, x1 ∈ Ω one has

ω x0 (E) ≈ ω x1 (E), uniformly for E ∈ M, (5.2.15)

where the intervening proportionality constants stay bounded away from zero
and infinity as long as x0, x1 remain in a compact subset of Ω.

Proof We proceed in a number of steps.

Step I: For each x ∈ Ω, the measure ω x : M x → [0, 1] is Borel-regular. To
justify this, fix x ∈ Ω and pick some arbitrary E ∈ M x . From the outer-regularity
property (5.2.12) we conclude that there exists a sequence {O j } j ∈N of relatively
open subsets of ∂Ω such that E ⊆ O j for each j ∈ N and lim ω x (O j ) = ω x (E). If
j→∞
we define B := O j then B is a Borel subset of ∂Ω which contains E and satisfies
j ∈N
ω x (E) ≤ ω x (B) ≤ ω x (O j ) → ω x (E) as j → ∞. Hence, ω x (E) = ω x (B), proving
the claim in Step I (cf. [112, Definition 3.4.1]).
∫ ∫
Step II: For each x0, x1 ∈ Ω, we have ∂Ω f dω x0 ≈ ∂Ω f dω x1 uniformly for
non-negative functions f ∈ 𝒞0 (∂Ω), with constants bounded away from zero and
infinity as long as x0, x1 remain in a compact subset of Ω. Having fixed f as above,
the desired equivalence is a consequence of Harnack’s inequality applied to the non-
negative harmonic function PI f in Ω (cf. (5.2.7)-(5.2.8)), bearing in mind (5.2.9).
Step III: For each x0, x1 ∈ Ω, we have ω x0 (K) ≈ ω x1 (K) uniformly for compacts
K ⊆ ∂Ω, with constants bounded away from zero and infinity as long as x0, x1 remain
in a compact subset of Ω. To see that this is the case, fix a compact set K ⊆ ∂Ω and
consider a sequence of functions { f j } j ∈N ⊆ 𝒞0 (∂Ω) satisfying 0 ≤ f j ≤ 1 for each
5.2 The Harmonic Measure 839

j ∈ N and f j  1K pointwise on ∂Ω as j → ∞ (cf. [117, Lemma 4.14, p. 166]).

Then we may write
∫ ∫
ω x0 (K) ≤ (PI f j )(x0 ) = f j dω x0 ≈ f j dω x1
∂Ω ∂Ω
∫
−−−−→ 1K dω x1 = ω x1 (K), (5.2.16)
j→∞ ∂Ω

thanks to (5.2.11), (5.2.9), Step II, and Lebesgue’s Dominated Convergence Theo-
rem. Changing the roles of x0 and x1 then finishes the proof of the claim made in
Step III.
Step IV: For each x0, x1 ∈ Ω, we have ω x0 (B) ≈ ω x1 (B) uniformly for Borel sets
B ⊆ ∂Ω, with constants bounded away from zero and infinity as long as x0, x1 remain
in a compact subset of Ω. This is a consequence of Step III and the inner-regularity
property (5.2.13).
Step V: Fix x0 ∈ Ω and consider E ∈ M x0 with the property that ω x0 (E) = 0. Then
for any other point x1 ∈ Ω we have E ∈ M x1 and ω x1 (E) = 0. To prove this, recall
from Step I that for each x0 ∈ Ω the measure ω x0 : M x0 → [0, 1] is Borel-regular.
Thus, given E ∈ M x0 with ω x0 (E) = 0, we can find a Borel set B ⊆ ∂Ω such
that E ⊆ B and ω x0 (B) = ω x0 (E) = 0. Since B ∈ M x0 and B = (B \ E)  E, it
follows that ω x0 (B) = ω x0 (B \ E) + ω x0 (E), which ultimately forces ω x0 (B \ E) = 0.
Pick an arbitrary x1 ∈ Ω. Given that ω x1 : M x1 → [0, 1] is a complete measure,
B ∈ M x1 , and 0 = ω x0 (B) ≈ ω x1 (B) (cf. Step IV), we conclude (cf. [112, (3.1.23)])
that B \ E ∈ M x1 and ω x1 (B \ E) = 0. Thus, E = B \ (B \ E) ∈ M x1 and

0 = ω x1 (B) = ω x1 (B \ E) + ω x1 (E) = ω x1 (E), (5.2.17)

hence ω x1 (E) = 0 as wanted.

Step VI: For each x0, x1 ∈ Ω we have M x0 = M x1 and, abbreviating M := M x
for some, or any, x ∈ Ω, it follows that ω x0 (E) ≈ ω x1 (E) uniformly for E ∈ M,
with constants bounded away from zero and infinity as long as x0, x1 remain in a
compact subset of Ω. To justify these claims, fix x0, x1 ∈ Ω along with some arbitrary
E ∈ M x0 . Since ω x0 : M x0 → [0, 1] is a Borel-regular measure (cf. Step I), we can
find a Borel set B ⊆ ∂Ω such that E ⊆ B and ω x0 (B) = ω x0 (E). This implies that
B \ E ∈ M x0 and ω x0 (B \ E) = 0. Granted this, Step V implies that we also have
B \ E ∈ M x1 and ω x1 (B \ E) = 0. As such, E = B \ (B \ E) ∈ M x1 and

ω x1 (E) = ω x1 (B) ≈ ω x0 (B) = ω x0 (E), (5.2.18)

thanks to Step IV. At this stage, all desired conclusions follow. 

At this stage, we can properly introduce the notion of harmonic measure.

Definition 5.2.2 Given a bounded open set Ω ⊆ Rn along with an arbitrary point
x ∈ Ω, call the complete, Borel-regular, probability measure ω x : M → [0, 1]
840 5 Green Functions and Poisson Kernels for the Laplacian

(where the sigma-algebra M is as in item (i) of Lemma 5.2.1) the harmonic measure
on ∂Ω with pole at x.

We are interested in the class of bounded open sets Ω ⊆ Rn for which the Poisson
integral f → PI f is the solution operator for the classical Dirichlet Problem
  
u ∈ 𝒞∞ (Ω) ∩ 𝒞0 Ω , Δu = 0 in Ω, u∂Ω = f ∈ 𝒞0 (∂Ω). (5.2.19)

In this vein, we make the following definition.

Definition 5.2.3 A bounded open set Ω in Rn is called regular (for the Dirichlet
Problem for the Laplacian) if, given any f ∈ 𝒞0 (∂Ω) one has PI f ∈ 𝒞0 (Ω) and
(PI f )(y) = f (y) for each y ∈ ∂Ω.

Regularity in the sense of the above definition turns out to be equivalent to the
well-posedness of the classical Dirichlet Problem.

Proposition 5.2.4 Let Ω be a bounded open set in Rn . Then Ω is regular (for the
Dirichlet Problem for the Laplacian) if and only if the classical Dirichlet Problem
(5.2.19) is solvable.
Hence, if Ω ⊆ Rn is a bounded open set which is regular, then the classical
Dirichlet Problem (5.2.19) is well posed and for each f ∈ 𝒞0 (∂Ω) its unique
solution is given by u := PI f .

Proof Assume the classical Dirichlet Problem is solvable. Then for each function
f ∈ 𝒞0 (∂Ω), the solution u of (5.2.19) belongs both to the upper class U f , and the
lower class L f , of functions associated with f . In concert with (5.2.4), this implies

PI f ≤ u in Ω, and u ≤ PI f in Ω. (5.2.20)

Combining (5.2.20) with (5.2.5) then proves that

PI f = PI f = u in Ω, (5.2.21)

which further goes to show that PI f = u ∈ 𝒞0 (Ω) and (PI f )(y) = u(y) = f (y) for
each y ∈ ∂Ω. Thus, Ω is regular for the Dirichlet Problem for the Laplacian (in the
sense of Definition 5.2.3).
Conversely, if Ω is regular then, tautologically, for each f ∈ 𝒞0 (∂Ω), the func-
tion u := PI f solves the classical Dirichlet Problem (5.2.19). Finally, uniqueness
for this classical Dirichlet Problem is guaranteed by the Maximum Principle (cf.
Theorem 5.1.6). 
For example, from Proposition 5.2.4 and [49, Theorem 1.1, pp. 37-38] it follows
that
any finitely connected Jordan domain in the plane is
(5.2.22)
regular for the Dirichlet Problem for the Laplacian.
To give a simple example of a domain which is not regular for the Laplacian, consider
the bounded, open, connected set Ω := {x ∈ Rn : 0 < |x| < 1} = B(0, 1) \ {0} with
5.2 The Harmonic Measure 841

n ≥ 2. If f ∈ 𝒞0 (∂Ω) is defined at each x ∈ ∂Ω = S n−1 ∪ {0} by f (x) := 0 if |x| = 1

 f (0) :=  1, then any function u ∈ 𝒞 (Ω) which is harmonic in Ω and satisfies
and 0

uS n−1 = f S n−1 = 0 necessarily vanishes everywhere in Ω. Indeed, if for each ε > 0
we consider the harmonic function
 
ε |x| 2−n − 1 if n ≥ 3,
uε (x) := ∀x ∈ Ω, (5.2.23)
−ε ln |x| if n = 2,

then Maximum Principle (cf. Theorem 5.1.6) implies that 0 ≤ u(x) ≤ uε (x) for
each x ∈ Ω. Upon letting ε → 0+ we conclude that, indeed, u ≡ 0 in Ω. Hence,
u(0) = 0  1 = f (0) so the classical Dirichlet Problem in Ω is not solvable for the
boundary datum f . A more sophisticated example due to H. Lebesgue of a bounded,
connected, open set in R3 with a sufficiently sharp inward cusp for which the classical
Dirichlet Problem (5.2.19) fails to be solvable (hence the domain in question is not
regular) is discussed in [23].
Necessary and sufficient conditions for regularity are known. To state one of them,
recall the following concept of capacity.
Definition 5.2.5 Given a compact set K ⊂ Rn , define
∫
cap (K) := inf |∇ϕ| 2 dL n : ϕ ∈ 𝒞∞
c (R ), ϕ ≥ 1 on K .
n
(5.2.24)
Rn

See, e.g., [39, § 4.7, pp. 146-158] and [51, pp. 27-28]. For example, one may check
without difficulty that

cap (K) ≤ cap (K) for each compacts K, K ⊂ Rn with K ⊆ K,

(5.2.25)
cap (rK) = r n−2 cap (K) for each compact K ⊂ Rn and r > 0.

For other basic properties see, e.g., [39, Theorem 2, p. 151]. In relation to this notion
of capacity, following Wiener criterion holds.
Theorem 5.2.6 Assume that Ω is a bounded, open, connected set in Rn . Then Ω is
regular for the Dirichlet Problem for the Laplacian if and only if
∫ 1   dr
cap B(x, r) \ Ω n−1 = +∞ for every x ∈ ∂Ω. (5.2.26)
0 r
See also [55, Theorem 2.5, p. 25] and the discussion in [51, § 2.9, pp. 27–28] in this
regard. A discrete version of (5.2.26) reads2

∞
 
2(n−2)j cap B(x, 2−j ) \ Ω = +∞ for each x ∈ ∂Ω. (5.2.27)
j=0

See [55, Corollary 2.5, p. 27]. In fact, there is a point-by-point version of Wiener’s
criterion. Specifically, if for some given point x ∈ ∂Ω we have
2 if n = 2 then 2(n−2) j is replaced by j in (5.2.27)
842 5 Green Functions and Poisson Kernels for the Laplacian


∞
 
2(n−2)j cap B(x, 2−j ) \ Ω = +∞ (5.2.28)
j=0

then for arbitrary f ∈ 𝒞0 (∂Ω) the solution to the generalized Dirichlet Problem
Δu = 0 in Ω and “u = f on ∂Ω” constructed by the Perron-Wiener-Brelot method
has the property that u(y) → f (x) whenever Ω y → x. Moreover, if for some
  
x ∈ ∂Ω we have ∞ j=0 2
(n−2)j cap B(x, 2−j ) \ Ω < +∞ then there exists f ∈ 𝒞0 (∂Ω)

for which u(y) fails to converge to f (x) as Ω y → x.

In particular, from (5.2.25) and (5.2.27) it follows that
if Ω ⊆ Rn is some bounded open set satisfying an exterior corkscrew
(5.2.29)
condition then Ω is regular for the Dirichlet Problem for the Laplacian.
Another useful criterion for regularity is based on the notion of barrier, recalled
next.

 5.2.7 Let Ω ⊆ R be a bounded open set and fix x0 ∈ ∂Ω. A function

Definition n

w ∈ 𝒞 Ω is said to be a barrier at x0 (relative to Ω) provided

0

w is superharmonic in Ω,
(5.2.30)
w > 0 in Ω \ {x0 }, and w(x0 ) = 0.

Then (cf., e.g., [51, Theorem 2.14, p. 26]) we have:

Theorem 5.2.8 Let Ω be a bounded open set in Rn . Then the classical Dirichlet
Problem (5.2.19) is solvable (i.e., Ω is regular for the Dirichlet Problem for the
Laplacian) if and only if for each x ∈ ∂Ω there exists a barrier function at x, relative
to Ω.

5.3 The Green Function for the Laplacian

The Green function for the Laplacian in arbitrary bounded open sets is defined in
the theorem below, where we also collect its most basic properties in such a setting
(for proofs, we refer to [55]). In two subsequent theorems we then indicate how some
of these properties improve as the underlying domain becomes more regular. The
reader is reminded that W k, p (Ω) is the L p -based Sobolev space of order k ∈ N in
the open set Ω ⊆ Rn , and W̊ k, p (Ω) is the closure of 𝒞∞
c (Ω) in W
k, p (Ω).

Theorem 5.3.1 Assume that Ω is an arbitrary bounded open set in Rn . Then there
exists a unique function GΩ : Ω × Ω → [0, +∞], henceforth referred to as the Green
function for the Laplacian in Ω, such that
 
GΩ (·, y) ∈ W 1,2 Ω \ B(y, r) ∩ W̊ 1,1 (Ω), ∀y ∈ Ω, ∀r > 0, (5.3.1)

and
5.3 The Green Function for the Laplacian 843
∫
∇x GΩ (x, y), ∇ϕ(x) dx = ϕ(y), ∀ϕ ∈ 𝒞∞
c (Ω). (5.3.2)
Ω
Furthermore, the Green function also satisfies the following additional properties:

(i) GΩ (x, y) = GΩ (y, x) for all x, y ∈ Ω with x  y;

(ii) 0 < GΩ (x, y) ≤ Cn |x − y| 2−n for all x, y ∈ Ω with x  y;
(iii) GΩ (x, y) ≥ Cn |x − y| 2−n for all x, y ∈ Ω with x  y and |x − y| ≤ 1
2 dist(y, ∂Ω);
(iv) GΩ (·, y) ∈ W̊ 1, p (Ω) for all 1 ≤ p < n−1
n
, uniformly in y ∈ Ω;
n
,∞
(v) GΩ (·, y) ∈ L n−2 (Ω, L ), uniformly in y ∈ Ω;
n
n
(vi) ∇GΩ (·, y) ∈ L n−1 ,∞ (Ω, L n ), uniformly in y ∈ Ω.

From (5.3.1)-(5.3.2) we see that, in the sense of distributions in Ω,

ΔGΩ (·, y) = −δy for each fixed y ∈ Ω, (5.3.3)

where δy is the Dirac distribution in Ω with mass at y. In particular,

GΩ (·, y) ∈ 𝒞∞ (Ω \ {y}) for each fixed y ∈ Ω. (5.3.4)

In fact, in [55] the authors established Theorem 5.3.1 for arbitrary scalar second-
order elliptic operators L in divergence form, with bounded measurable coefficients.
Using the same template for the construction of the Green function as in [55], this
result has been further adapted in [60] to a class containing all second-order strongly
elliptic systems L in divergence form with bounded measurable coefficients such
that weak solutions of both L and L  are locally Hölder (a property which, for
scalar operators, is automatically guaranteed by the classical De Giorgi-Nash-Moser
theory). We are interested in the results of [60] for the case L := Δ. Specialized
as such, [60, Theorem 4.1] implies, among other things, that if Ω is a bounded
connected open set in Rn then
∞
 η ∈ 𝒞c (Ω)
for any  satisfying η ≡ 1 on B(x0, r) for some (5.3.5)
r ∈ 0, dist(x0, ∂Ω) we have (1 − η)GΩ (·, x0 ) ∈ W̊ 1,2 (Ω),

and
∞
∫ f ∈ 𝒞c (Ω), the function defined for each x ∈ Ω
for each given
as u(x) := Ω GΩ (x, y) f (y) dy belongs to W̊ 1,2 (Ω) and satisfies (5.3.6)
Δu = − f in the sense of distributions in the open set Ω.
Additional geometric assumptions on the underlying set Ω ensure extra regularity
properties for the associated Green function.

Theorem 5.3.2 Let Ω ⊆ Rn be a bounded open set which is regular for the Dirichlet
Problem for the Laplacian (in the sense of Definition 5.2.3). Then for  every x0 ∈ Ω
fixed, GΩ (·, x0 ) extends uniquely to a function belonging to 𝒞  Ω \ {x0 } and,
0

retaining the same notation for this extension, one has GΩ (·, x0 )∂Ω = 0.
844 5 Green Functions and Poisson Kernels for the Laplacian

Proof Working in each connected component of Ω, there is no loss of generality in

assuming that Ω itself is a connected set to begin with. Suppose this is the case and
fix a point x0 ∈ Ω. With d := dist(x0, ∂Ω) > 0, define

M := sup |x−x0 |=d/2 GΩ (x, x0 ), Ω0 := Ω \ B(x0, d/2)

 (5.3.7)
and introduce u := M −1 · GΩ (·, x0 ) .
Ω0

Then from (5.3.1), (5.3.4), (5.3.3), and the positivity of the Green function it follows
that u satisfies

u ∈ W 1,2 (Ω0 ) ∩ 𝒞∞ (Ω0 ), u > 0, Δu = 0 in Ω0, (5.3.8)

while (5.3.5) implies that

0 ≤ u ≤ 1 on ∂Ω0 and u = 0 on ∂Ω ⊆ Ω0, (5.3.9)

in the sense of W 1,2 (Ω0 ) (cf. [79, Definition 1.1.13, p. 3], [89], [60, § 4.2]). Granted
these, [55, Lemma 2.3, p. 23] applies and gives that there exist c ∈ (0, ∞) and
r0 ∈ (0, d/2) with the property that for each x ∈ ∂Ω we have
∫  
r0 cap B(x, t) \ Ω
0≤ sup u ≤ exp − c dt , ∀r ∈ (0, r0 ). (5.3.10)
Ω∩B(x,r) r t n−1

From this, Theorem 5.2.6, and assumptions, we see that u vanishes in a continuous
fashion at each point x ∈ ∂Ω. Since u is a fixed multiple of GΩ (·, x0 ) near ∂Ω, the
desired conclusion follows. 

The result below is contained in [79, Theorem 1.2.8, p. 7] (cf. also [60, Theo-
rem 4.8]).

Theorem 5.3.3 Assume Ω ⊆ Rn is a bounded open set with the property that Rn \ Ω
is n-thick. Then one can find some exponent α = α(Ω) > 0 along with some
constant C ∈ (0, ∞) with the property that the Green function associated with Ω as
in Theorem 5.3.1 satisfies
(i) GΩ (x, y) ≤ C dist(y, ∂Ω)α |x − y| 2−n−α for all x, y ∈ Ω;
(ii) |GΩ (x, y)−GΩ (z, y)| ≤ C|x − z| α /(|x − y| 2−n−α + |z − y| 2−n−α ) for all x, y, z ∈ Ω.

In particular,
if Ω ⊆ Rn is a bounded open set whose complement is n-thick then, for
every x0 ∈ Ω fixed, GΩ (·, x0 ) extends to a Hölder continuous function (5.3.11)
in a neighborhood of the topological boundary and GΩ (·, x0 )∂Ω = 0.

Lemma 5.3.4 Suppose Ω ⊆ Rn is a bounded open set which is regular for the
Dirichlet Problem for the Laplacian. Recall the standard fundamental solution EΔ
for the Laplacian in Rn defined in (1.5.56). Then
5.3 The Green Function for the Laplacian 845
  
GΩ (·, x0 ) = PI EΔ (· − x0 )∂Ω − EΔ (· − x0 )
∫
= EΔ (· − x0 ) dω x0 − EΔ (· − x0 ) for each x0 ∈ Ω, (5.3.12)
∂Ω

and for each x0, x1 ∈ Ω one has

∫ ∫
EΔ (y − x0 ) dω (y) =
x1
EΔ (y − x1 ) dω x0 (y). (5.3.13)
∂Ω ∂Ω

Proof Fix x0 ∈ Ω and consider

H(x, x0 ) := GΩ (x, x0 ) + EΔ (x − x0 ) for L n -a.e. x ∈ Ω. (5.3.14)

1 (Ω, L n ) and ΔH(·, x ) = 0 in D  (Ω), hence H(·, x ) ∈ 𝒞∞ (Ω).
Then H(·, x0 ) ∈ Lloc 0 0
 
From assumptions
 and  Theorem 5.3.2, we also see that H(·, x0 ) ∈ 𝒞0 Ω and
H(·, x0 )∂Ω = EΔ (· − x0 )∂Ω . Consequently,
 ∫
 
H(·, x0 ) = PI EΔ (· − x0 )∂Ω = EΔ (· − x0 ) dω x0 . (5.3.15)
∂Ω

This establishes (5.3.12). Moreover, (5.3.15) permits us to write

∫
H(x, x0 ) = EΔ (y − x0 ) dω x (y), ∀x ∈ Ω. (5.3.16)
∂Ω

On the other hand, from (5.3.14), item (i) in Theorem 5.3.1, and (1.5.56) we see that

H(x, x0 ) = H(x0, x) for each x, x0 ∈ Ω with x  x0 . (5.3.17)

Then (5.3.13) readily follows from (5.3.16) and (5.3.17). 

Our next theorem elaborates on the boundary behavior of the gradient Green
function associated with an arbitrary UR domain in Rn . In particular, it refines work
in [25, Theorem 3, p. 275], [81, Lemma 3.1, p. 336], [81, Lemmas 3.1-3.2, pp. 336-
337] [81, §A.2, pp. 384-389], [82, Lemma 3.2, p. 10], where more restrictive classes
of domains have been considered.

Theorem 5.3.5 Assume Ω ⊆ Rn is an arbitrary bounded UR domain. In particular,

Ω is a set of locally finite perimeter and, if σ := H n−1 ∂Ω, its geometric measure
theoretic outward unit normal ν = (ν1, . . . , νn ) is defined σ-a.e. on ∂Ω. Also, fix a
point x0 ∈ Ω along with some aperture parameter κ ∈ (0, ∞), and assume that the
Green function GΩ (·, ·) associated with Ω as in Theorem 5.3.1 satisfies
   
Nκε ∇GΩ (·, x0 ) ∈ L 1 (∂Ω, σ) for some ε ∈ 0, dist(x0, ∂Ω) . (5.3.18)

Then for each κ  > 0, the nontangential trace

846 5 Green Functions and Poisson Kernels for the Laplacian

 κ −n.t.


∇GΩ (·, x0 )  exists at σ-a.e. point on ∂Ω
∂Ω (5.3.19)
and is actually independent on the parameter κ .

In particular, the normal derivative of GΩ (·, x0 ) may be meaningfully defined as

   κ  −n.t. 
∂ν GΩ (·, x0 ) := ν, ∇GΩ (·, x0 ) ∂Ω at σ-a.e. point on ∂Ω, (5.3.20)

and this definition is independent of the aperture parameter κ .

Moreover, with the dependence on the aperture parameter dropped, for each
j, k ∈ {1, . . . , n} one has
  n.t.    n.t. 
ν j ∂k GΩ (·, x0 ) ∂Ω = νk ∂j GΩ (·, x0 ) ∂Ω at σ-a.e. point on ∂Ω. (5.3.21)

As a consequence,
  n.t.
∇GΩ (·, x0 ) ∂Ω = ∂ν GΩ (·, x0 )ν at σ-a.e. point on ∂Ω, (5.3.22)
 
and for any vector field T : ∂Ω → Cn which is tangential, in the sense that ν, T = 0
at σ-a.e. point on ∂Ω, one has
   n.t. 
 ∇GΩ (·, x0 ) 
T, = 0 at σ-a.e. point on ∂Ω. (5.3.23)
∂Ω

Finally, for each κ  > 0, the nontangential trace

κ  −n.t.

GΩ (·, x0 ) = 0 at σ-a.e. point on ∂Ω. (5.3.24)
∂Ω

 
Proof Observe that Ω0 := Ω \ B x0, 12 dist(x0, ∂Ω) is a bounded UR domain, and
GΩ (·, x0 ) is a harmonic in Ω0 . In addition, (5.3.18) implies that
 the nontangential
maximal function of GΩ (·, x0 ) relative to Ω0 belongs to L 1 ∂Ω0, σ0 ), where we
have set σ0 := H n−1 ∂Ω0 . Granted this, the Fatou-type result from Theorem 3.3.4
applies (with L := Δ) and gives that the nontangential boundary trace of ∇GΩ (·, x0 ),
taken from within Ω0 with any aperture parameter κ  > 0, exists at σ0 -a.e. point on
∂Ω0 and is actually independent of κ . Upon observing that there exists ε > 0 with
the property that

Γ0κ  (x) ∩ B(x, ε) = Γκ  (x) ∩ B(x, ε) for each x ∈ ∂Ω ⊆ ∂Ω0,

where Γ0κ  (x), Γκ  (x) are the nontangential approach regions (5.3.25)
with apex at x, considered relative to Ω0 and Ω, respectively,

it follows that the nontangential boundary trace of ∇GΩ (·, x0 ) at points on ∂Ω, taken
  κ  −n.t.
from within Ω0 , agrees σ-a.e. with ∇GΩ (·, x0 ) ∂Ω . This completes the proof of
(5.3.19).
5.3 The Green Function for the Laplacian 847

Next, observe that item (ii) in Theorem 5.3.1 ensures that the harmonic function
GΩ (·, x0 ) is also bounded in the UR domain Ω0 . As such, we may invoke (3.3.107)
in Theorem 3.3.9 to conclude (also bearing in mind (5.3.25)) that for each κ  > 0
the nontangential trace
κ  −n.t.

GΩ (·, x0 ) exists at σ-a.e. point on ∂Ω
∂Ω (5.3.26)
and is actually independent on the parameter κ .

To prove the stronger conclusion claimed in (5.3.24), let us select an arbitrary index
j ∈ {1, . . . , n} along with any test function ϕ ∈ 𝒞∞ c (R ) having the property that
n

x0  supp ϕ. Since GΩ (·, x0 ) ∈ W̊ 1,1 (Ω) (cf. (5.3.1)), it follows that

there exists a sequence {ψμ }μ ∈N ⊆ 𝒞∞ c (Ω) with the property that

(5.3.27)
ψμ → GΩ (·, x0 ) and ∂j ψμ → ∂j GΩ (·, x0 ) in L 1 (Ω, L n ) as μ → ∞.

Then
GΩ (·, x0 )ϕ ∈ W 1,1 (Ω) , and
    (5.3.28)
∂j GΩ (·, x0 )ϕ = ∂j GΩ (·, x0 ) ϕ + GΩ (·, x0 )(∂j ϕ).
Consequently, for each κ  > 0,
∫  κ  −n.t.  ∫
  
GΩ (·, x0 ) ϕν j dσ = ∂j GΩ (·, x0 )ϕ dL n
∂Ω ∂Ω Ω
∫
 
= ∂j GΩ (·, x0 ) ϕ + GΩ (·, x0 )(∂j ϕ) dL n
Ω
∫
= lim (∂j ψμ )ϕ + ψμ (∂j ϕ) dL n
μ→∞ Ω
∫
 
= lim ∂j ψμ ϕ dL n
μ→∞ Ω

= 0. (5.3.29)

Above, the first equality is a consequence of the Divergence Formula [112, (1.2.2)]
applied to the vector field
 n
F := GΩ (·, x0 )ϕ e j ∈ 𝒞∞c (Ω) . (5.3.30)

In this regard, we note that from item (ii) in Theorem 5.3.1 and our choice of ϕ, it
follows that F is also bounded in Ω. Based on this and [112, (8.3.6) in Lemma 8.3.2]
we conclude that
Nκ  F ∈ L ∞ (∂Ω, σ) ⊆ L 1 (∂Ω, σ). (5.3.31)
Thanks to assumptions, (5.3.26), (5.3.28), and (5.3.31) the vector field F satisfies
all hypotheses in [112, Theorem 1.2.1]. This justifies the applicability of Divergence
848 5 Green Functions and Poisson Kernels for the Laplacian

Formula [112, (1.2.2)] in the present setting. Next, the second equality in (5.3.29)
comes from (5.3.28), while the third equality in (5.3.29) is implied by (5.3.27).
Finally, the fourth equality in (5.3.29) is simply Chain Rule for smooth functions,
and the last equality in (5.3.29) is justified by noting that ψμ ϕ ∈ 𝒞∞
c (Ω) for each
μ ∈ N.
Having established (5.3.29), we may invoke [112, Corollary 3.7.3, (3.7.23)] (with
X := ∂Ω, μ := σ, and O := Rn \ {x0 }) to conclude that, for each κ  > 0 and each
j ∈ {1, . . . , n} we have
 κ  −n.t. 

GΩ (·, x0 ) ν j = 0 at σ-a.e. point on ∂Ω. (5.3.32)
∂Ω

Given that |ν| = 1 at σ-a.e. point in ∂∗ Ω (cf. [112, (5.6.20), (5.6.21)]), and since
we are presently assuming (cf. the definition of a UR domain from [112, Defini-
tion 5.10.6])
σ(∂Ω \ ∂∗ Ω) = 0, (5.3.33)
we ultimately arrive at the conclusion that (5.3.24) holds.
Finally, with (5.3.19) and (5.3.24) in hand, the claims in (5.3.21)-(5.3.23) follow
from [113, Corollary 11.3.3)], bearing in mind (5.3.33) and the fact that item (ii) in
Theorem 5.3.1 together with [112, (8.3.6) in Lemma 8.3.2] imply

Nκε G(·, x0 ) ∈ L ∞ (∂Ω, σ) ⊆ L 1 (∂Ω, σ)

  (5.3.34)
for each ε ∈ 0, dist(x0, ∂Ω) and κ  > 0.

This finishes the proof of Theorem 5.3.5. 

The next result in this section provides an answer to the first question raised in the
preamble to §5 (cf. [49, Question 2, p. 49]). Essentially, if a given domain is regular
for the Dirichlet Problem for the Laplacian in the sense of Definition 5.2.3, satis-
fies a local pathwise nontangential accessibility condition, has an Ahlfors regular
boundary, and its Green function (considered as in Theorem 5.3.1) has a reasonable
boundary behavior (which basically ensures that its normal derivative is meaning-
ful), then the harmonic measure is absolutely continuous with respect to the surface
measure and the Radon-Nikodym derivative is the (minus) normal derivative of the
Green function. Our result refines work in [16, Theorem 5.7, p. 79], [81, Propo-
sition A.1.1, p. 382], [81, Proposition A.2.2, p. 385], and [82, Lemma 3.4, p. 16],
where more restrictive classes of sets have been considered.

Theorem 5.3.6 Let Ω ⊂ Rn , where n ∈ N with n ≥ 2, be a bounded open set

which is regular for the Dirichlet Problem for the Laplacian (in the sense of Defini-
tion 5.2.3), is locally pathwise nontangentially accessible (in the sense of [112, Defi-
nition 8.9.14]), has a lower Ahlfors regular boundary, and for which σ := H n−1 ∂Ω
is a doubling measure on ∂Ω. Also, fix a point x0 ∈ Ω along with some aperture
parameter κ ∈ (0, ∞), and assume that the Green function GΩ (·, ·) associated with
Ω as in Theorem 5.3.1 satisfies
5.3 The Green Function for the Laplacian 849
ε
   
Nκ ∇GΩ (·, x0 ) ∈ L 1 (∂Ω, σ) for some ε ∈ 0, dist(x0, ∂Ω) ,
  κ−n.t. (5.3.35)
and ∇GΩ (·, x0 )  exists at σ-a.e. point on ∂Ω.
∂Ω

Then
ω x0 , the harmonic measure on ∂Ω with pole at x0 , is absolutely con-
tinuous with respect to the surface measure σ = H n−1 ∂Ω, in the
(5.3.36)
sense that any σ-measurable set belongs to the sigma-algebra M, and
if E ⊆ ∂Ω is σ-measurable and σ(E) = 0 then ω x0 (E) = 0.
Moreover, if ν denotes the geometric measure theoretic outward unit normal to
Ω, then the Radon-Nikodym derivative of the harmonic measure ω x0 with respect to
the surface measure σ is given by
dω x0
= −1∂∗ Ω · ∂ν GΩ (·, x0 ) at σ-a.e. point on ∂Ω,
dσ (5.3.37)
   κ−n.t. 
where ∂ν GΩ (·, x0 ) := ν, ∇GΩ (·, x0 ) ∂Ω .

Before giving the proof of this theorem, a few comments are in order.
Comment 1. Under the additional assumption that Ω is a UR domain, Theorem 5.3.5
guarantees that the condition in the second line of (5.3.35) is automatically satisfied.
Of course, if Ω is a UR domain then, by design, ∂Ω is Ahlfors regular, so Ω has a
lower Ahlfors regular boundary and σ = H n−1 ∂Ω is a doubling measure on ∂Ω.
Comment 2. Whenever ω x0 << σ, the Radon-Nikodym derivative

dω x0
k x0 := (5.3.38)
dσ
is called the Poisson kernel for the domain Ω. Hence, whenever ω x0 << σ it
follows that
the Poisson kernel k x0 := dω x0 /dσ is a well-defined,
∫ non-negative
function, which belongs to L 1 (∂Ω, σ), and satisfies ∂Ω k x0 dσ = 1, (5.3.39)

(since ω x0 is a probability measure). Assuming that the Poisson kernel exists, Riesz’s
duality theorem also implies that given any p, p ∈ (1, ∞) satisfying 1/p + 1/p = 1
we have

L p (∂Ω, σ) → L 1 (∂Ω, ω x0 ) ⇐⇒ k x0 ∈ L p (∂Ω, σ). (5.3.40)

Comment 3. In the context of Theorem 5.3.6, the conclusion in (5.3.36) ensures that
(5.3.39) holds, while (5.3.37) gives the pointwise representation formula

k x0 = −1∂∗ Ω · ∂ν GΩ (·, x0 ) at σ-a.e. point on ∂Ω. (5.3.41)

850 5 Green Functions and Poisson Kernels for the Laplacian

Since, as noted in (5.3.39), the very existence of the Poisson kernel entails
k x0 ∈ L 1 (∂Ω, σ), it follows from (5.3.41) that the quantitative assumption made
on ∇GΩ (·, x0 ) in the first line of (5.3.35) is actually natural.
We now turn to the proof of Theorem 5.3.6.
Proof of Theorem 5.3.6 Since Ω is a bounded open set which is regular for the
Dirichlet Problem for the Laplacian, for each f ∈ 𝒞0 (∂Ω) the function u := PI f is
the unique solution of the classical Dirichlet Problem (5.2.19). In concert with [112,
Proposition 8.9.16] and [112, Lemma 8.3.2], this implies that u satisfies
⎧
⎪ u ∈ 𝒞∞ (Ω), Δu = 0 in Ω,
⎪
⎪
⎨ κ−n.t.
⎪
u∂Ω = f at σ-a.e. point on ∂Ω, (5.3.42)
⎪
⎪
⎪
⎪ Nκ u belongs to L ∞ (∂Ω, σ).
⎩
Let us also observe that, by virtue of item (ii) in Theorem 5.3.1, whenever we have
0 < ε < dist(x0, ∂Ω) it follows
 that the function GΩ (·, x0 ) is bounded in the one-sided
collar neighborhood Oε := x ∈ Ω : dist(x, ∂Ω) < ε of ∂Ω. In particular, thanks
to [112, (8.2.26)], we have Nκε GΩ (·, x0 ) ∈ L ∞ (∂Ω, σ). This property may be also
justified based on Theorem  5.3.2 which,  in view of [112, (8.9.10)] and the fact that
we presently have H n−1 ∂Ω \ ∂nta Ω = 0 (cf. [112, (8.9.193)]), additionally gives
that κ−n.t.

GΩ (·, x0 ) = 0 at σ-a.e. point on ∂Ω. (5.3.43)
∂Ω
Thanks to these properties, the original assumptions, and [112, Corollary 8.9.9], we
may invoke Theorem 4.4.1, presently used with M := 1 and L := Δ. In view of
(5.2.9) and the Poisson formula (4.4.4), we therefore obtain
∫ ∫
f dω = u(x0 ) = −
x0
f · ∂ν GΩ (·, x0 ) dσ. (5.3.44)
∂Ω ∂∗ Ω

Hence, ultimately
∫ ∫
f dω x0 = − f · ∂ν GΩ (·, x0 ) dσ, ∀ f ∈ 𝒞0 (∂Ω). (5.3.45)
∂Ω ∂∗ Ω

To proceed, fix an arbitrary compact set K ⊆ ∂Ω. From [117, Lemma 4.14,

p. 166] we know that there exists a sequence of functions { f j } j ∈N ⊆ 𝒞0 (∂Ω)
satisfying 0 ≤ f j ≤ 1 for each j ∈ N and f j  1K pointwise on ∂Ω as j → ∞.
Granted these properties, we may use Lebesgue’s Dominated Convergence Theorem
(twice) and (5.3.45) to write
∫ ∫
ω (K) =
x0
1K dω = lim
x0
f j dω x0 (5.3.46)
∂Ω j→∞ ∂Ω
∫ ∫
= − lim f j · ∂ν GΩ (·, x0 ) dσ = − 1K · ∂ν GΩ (·, x0 ) dσ.
j→∞ ∂∗ Ω ∂∗ Ω
5.3 The Green Function for the Laplacian 851

From this, (5.2.13), and item (1) in [112, Proposition 3.4.15] (also keeping in mind
[112, Lemma 3.4.13]), we then conclude that
∫
ω x0 (B) = − 1B · ∂ν GΩ (·, x0 ) dσ for each Borel set B ⊆ ∂Ω. (5.3.47)
∂∗ Ω

Consider next an arbitrary set E ⊆ ∂Ω which is σ-measurable and satisfies

σ(E) = 0. Since σ is a Borel-regular measure (cf. [112, Lemma 3.6.4]), it follows
that there exists a Borel set B ⊆ ∂Ω such that E ⊆ B and σ(B) = σ(E) = 0. Together
with (5.3.47), this forces ω x0 (B) = 0 thus, ultimately, E ∈ M and ω x0 (E) = 0, since
ω x0 is a complete measure. Let us record our progress. The argument so far shows
that
if E ⊆ ∂Ω is σ-measurable and satisfies σ(E) = 0 then
(5.3.48)
E belongs to the sigma-algebra M and ω x0 (E) = 0.
Assume now that E ⊆ ∂Ω is an arbitrary σ-measurable set. The fact that σ is a
Borel-regular measure (cf. [112, Lemma 3.6.4]) guarantees the existence of a Borel
set B ⊆ ∂Ω such that E ⊆ B and σ(B) = σ(E). In particular,

B \ E is σ-measurable, and σ(B \ E) = σ(B) − σ(E) = 0, (5.3.49)

since σ is a finite Borel measure. Together with (5.3.48), this proves that

B \ E ∈ M and ω x0 (B \ E) = 0. (5.3.50)

Since B ∈ M, this further implies that

E = B \ (B \ E) ∈ M and ω x0 (E) = ω x0 (B). (5.3.51)

From (5.3.48) and the first property in (5.3.51) the claim in (5.3.36) follows. Finally,
based on (5.3.49), (5.3.51), and (5.3.47) we may write
∫
ω x0 (E) = ω x0 (B) = − 1B · ∂ν GΩ (·, x0 ) dσ
∂∗ Ω
∫
=− 1E · ∂ν GΩ (·, x0 ) dσ. (5.3.52)
∂∗ Ω

Thus,
∫
ω x0 (E) = − 1E · ∂ν GΩ (·, x0 ) dσ for each σ-measurable set E ⊆ ∂Ω.
∂∗ Ω
(5.3.53)

In turn, this formula may be interpreted (cf. [156, Theorem 6.9(b), p. 122]) as saying
that (5.3.37) holds. 

Moving on, the basic fact that the harmonic measure in an NTA domain is
doubling has been established in [71]. The following lemma, borrowed from [71,
852 5 Green Functions and Poisson Kernels for the Laplacian

Lemma 4.8, p. 86], is a generalization of a similar result proved by B. Dahlberg in

the category of Lipschitz domains in [25, Lemma 1, p. 276]. To state it, form here
on we agree to denote by

Δ(x, r) := B(x, r) ∩ ∂Ω, with x ∈ ∂Ω and r > 0, (5.3.54)

the surface balls on ∂Ω.

Lemma 5.3.7 Let Ω be a bounded NTA domain in Rn and denote by GΩ (·, ·) the
Green function for the Laplacian in Ω. Then there exist constants R > 0, M > 1 which
depend only on Ω such that whenever 0 < r < R/2, z ∈ ∂Ω, and x ∈ Ω \ B(z, 2r),
one has  
−1 ω x Δ(z, r)
M < n−2 < M, (5.3.55)
r GΩ (Ar (z), x)
for each point Ar (z) ∈ Ω satisfying
 
| Ar (z) − z| < Mr and dist Ar (z), ∂Ω > M −1 r (5.3.56)

(making Ar (z) a corkscrew point relative to z, at scale r; cf. [112, (5.1.5)]).

A useful estimate for the truncated nontangential maximal operator of the gradient
of the Green function, in terms of the harmonic measure, is contained in the next
lemma.

Lemma 5.3.8 Suppose Ω is a bounded NTA domain in Rn and let GΩ (·, ·) be the
Green function for the Laplacian in Ω. Then there exist κ > 0 and a constant
C ∈ (0, ∞), both depending only on Ω, such that for each point x0 ∈ Ω one can find
0 < ε0 < dist(x0, ∂Ω) with the property that for each ε ∈ (0, ε0 ) one has
 
ε  ω x0 Δ(z, r)
Nκ ∇GΩ (·, x0 ) (z) ≤ C · sup for every z ∈ ∂Ω. (5.3.57)
0<r <ε r n−1

Proof Recall the constants R, M from Lemma 5.3.7. Since M > 1 we may pick
some small κ > 0 along with some large C > 1 + κ such that

C −1 + 1 < M and (1 + κ)−1 − C −1 > M −1 . (5.3.58)

 
Fix a point x0 ∈ Ω, introduce ε0 := 12 min R, dist(x0, ∂Ω) , and take ε ∈ (0, ε0 ).
Let z ∈ ∂Ω be arbitrary, select y ∈ B(z, ε) ∩ Γκ (z), and set r := |y − z| ∈ (0, ε).
Observe that the choices just made imply x0 ∈ Ω \ B(z, 2r). Also, on the one hand,
we have dist(y, ∂Ω) > |y − z|/(1 + κ) = r/(1 + κ) > r, hence B(y, r) ⊆ Ω. On the
other hand, since dist(y, ∂Ω) ≤ |y − z| < ε and r < ε, it follows that x0  B(y, r).
Consequently, GΩ (·, x0 ) is harmonic in a neighborhood of the closure of B(y, r/C).
As such, standard interior estimates give
⨏
c
|∇GΩ (y, x0 )| ≤ GΩ (x, x0 ) dx. (5.3.59)
r B(y,r/C)
5.4 The Poisson Kernel in NTA Domains 853

Note that, thanks to (5.3.58), any x ∈ B(y, r/C) satisfies

|x − z| ≤ |x − y| + |y − z| < r/C + r < Mr (5.3.60)

as well as

r/(1 + κ) = |y − z|/(1 + κ) < dist(y, ∂Ω) ≤ dist(x, ∂Ω) + |x − y|

≤ dist(x, ∂Ω) + r/C, (5.3.61)

 
hence dist(x, ∂Ω) > (1 + κ)−1 − C −1 r > M −1 r. Granted these, estimate (5.3.55)
(used with x := x0 and Ar (z) := x) yields
 
GΩ (x, x0 ) < Mr 2−n ω x0 Δ(z, r) for every x ∈ B(y, r/C). (5.3.62)

Using this back in (5.3.59) gives

 
|∇GΩ (y, x0 )| ≤ cMr 1−n ω x0 Δ(z, r) for every y ∈ B(z, ε) ∩ Γκ (z). (5.3.63)

Abbreviating C := cM and eventually taking the supremum in y ultimately yields

(5.3.57). 

5.4 The Poisson Kernel in NTA Domains

Defining the Poisson kernel for the domain Ω as the Radon-Nikodym derivative
(5.3.38), as done earlier, requires having ω x0 << σ. The result below describes a
scenario where this absolute continuity property is valid.

Proposition 5.4.1 Assume Ω ⊆ Rn is a bounded open set, satisfying a two sided

corkscrew condition, and possessing an upper Ahlfors regular boundary. Abbreviate
σ := H n−1 ∂Ω and, having fixed some x0 ∈ Ω, denote by ω x0 the harmonic measure
on ∂Ω with pole at x0 .
Then ω xo and σ are mutually absolutely continuous. Hence, the Poisson kernel
of the domain Ω, defined as the Radon-Nikodym derivative
dω x0
k x0 := , (5.4.1)
dσ
∫
exists, is non-negative, belongs to L 1 (∂Ω, σ), and satisfies ∂Ω
k x0 dσ = 1.

Proof This is a consequence of [11, Theorem 1] in concert with [112, Proposi-

tion 5.9.16]. 

We note that [112, (5.10.24)] implies that if Ω is as in Proposition 5.4.1 then ∂Ω

is actually a UR set (this observation comes in handy frequently in this section).
It turns out that both the harmonic measure and the Poisson kernel exhibit finer
854 5 Green Functions and Poisson Kernels for the Laplacian

properties under suitable connectivity assumptions. To elaborate on this aspect,

recall that a quantitative, scale invariant, version of the mutual absolute continuity
of the harmonic measure with respect to the surface measure is the membership
condition3 ω x0 ∈ A∞ (σ) (cf. [112, Definition 7.7.8]). This A∞ property has been
studied extensively and [10, Theorem 1.2] asserts (compare with [112, (5.11.6)])
that:
if Ω ⊆ Rn is an one-sided NTA domain with an Ahlfors regular
boundary, then the harmonic measure is in A∞ with respect to
(5.4.2)
the surface measure if and only if ∂Ω is a UR set, if and only if
Ω satisfies an exterior corkscrew condition.
In addition, J. Azzam has proved the following related result in [9]:
in the class of open connected sets Ω ⊆ Rn with an Ahlfors regular
boundary, the harmonic measure is in A∞ with respect to the surface
(5.4.3)
measure if and only if Ω is semi-uniform (cf. [112, (8.9.181)]) and ∂Ω
is a UR set.
As a consequence, we have the following result (originating in [160] and [32, Theo-
rem 2, p. 842]).

Corollary 5.4.2 Let Ω ⊆ Rn be a bounded NTA domain with an upper Ahlfors

regular boundary. Abbreviate σ := H n−1 ∂Ω, fix some x0 ∈ Ω. Then the Poisson
kernel for Ω satisfies

k x0 ∈ L p (∂Ω, σ) for some p > 1. (5.4.4)

Indeed, there is no loss of generality in assuming that Ω is connected. If ω x0 denotes

the harmonic measure on ∂Ω with pole at x0 , then (5.4.3) (whose applicability is
now ensured by [112, (5.10.24), (8.9.183)]) implies that

ω x0 belongs to a reverse Hölder class Bp (σ) for some p > 1 (5.4.5)

(cf. [112, Proposition 7.7.9]). With this in hand, (5.4.4) follows.

Lemma 5.4.3 Let Ω ⊆ Rn be a bounded NTA domain with an upper Ahlfors reg-
ular boundary and abbreviate σ := H n−1 ∂Ω. Also, denote by GΩ (·, ·) the Green
function for the Laplacian in Ω, and fix an arbitrary point x0 ∈ Ω. Then there exist
an aperture parameter κ > 0, a constant C ∈ (0, ∞), and a truncation parameter
0 < ε0 < dist(x0, ∂Ω) such that for every ε ∈ (0, ε0 ) and every z ∈ ∂Ω one has
⨏
ε 
Nκ ∇GΩ (·, x0 ) (z) ≤ C · sup k x0 dσ. (5.4.6)
0<r <ε Δ(z,r)

In particular,

3 Throughout, the membership of ω x0 to A∞ (σ) should be understood in a scale invariant fashion

(see, e.g., the formulation in [11])
5.5 Hardy Spaces of Harmonic Functions in NTA Domains 855
 
Nκε ∇GΩ (·, x0 ) ≤ CM ∂Ω k x0 on ∂Ω, (5.4.7)

where M ∂Ω denotes the Hardy-Littlewood maximal operator on ∂Ω (cf. (A.0.84)),

hence
 
Nκε ∇GΩ (·, x0 ) ∈ L p (∂Ω, σ) for some p > 1. (5.4.8)

Proof Since for each z ∈ ∂Ω and r > 0 we have

  ∫
ω x0 Δ(z, r) 1
= dω x0
r n−1 r n−1 Δ(z,r)
∫ ⨏
1
= n−1 k x0 dσ ≤ C k x0 dσ, (5.4.9)
r Δ(z,r) Δ(z,r)

on account of the upper Ahlfors regularity of ∂Ω, estimate (5.4.6) follows by com-
bining (5.3.57) with (5.4.9). With (5.4.6) in hand, (5.4.8) then follows from (5.4.4)
and the boundedness of the Hardy-Littlewood maximal operator on ∂Ω (cf. [112,
Corollary 7.6.3]). 
Finally, we record an estimate for the truncated nontangential maximal operators
of the Green function and its gradient, which is going to be very useful later on.
Lemma 5.4.4 Consider a bounded NTA domain Ω ⊆ Rn with an upper Ahlfors
regular boundary, and abbreviate σ := H n−1 ∂Ω. Also, let GΩ (·, ·) denote the
Green function for the Laplacian in Ω, and fix an arbitrary point x0 ∈ Ω along
with an aperture parameter κ > 0. Then there exist a constant θ ∈ (0, 1) and an
aperture
 parameter κ > 0 with the property that for each truncation parameter
ε ∈ 0, dist(x0, ∂Ω) one has
   
Nκθ ε GΩ (·, x0 ) (z) ≤ ε · Nκε ∇GΩ (·, x0 ) (z) for σ-a.e. z ∈ ∂Ω. (5.4.10)

Proof This follows from [112, Proposition 8.9.17], bearing in mind the observation
made in [112, (8.9.180)] as well as Theorem 5.3.2, or (5.3.11). 

5.5 Hardy Spaces of Harmonic Functions in NTA Domains

We say that a function u ∈ 𝒞0 (Ω) is nontangentially bounded from below at

x ∈ ∂Ω if there exists κ > 0, ε > 0, and M ≥ 0, such that u(y) ≥ −M for every
y ∈ B(x, ε) ∩ Γκ (x). Given E ⊆ ∂Ω, call u ∈ 𝒞0 (Ω) nontangentially bounded
from below on E provided u is nontangentially bounded from below at every point
in E. The following appears in [71, Theorem 6.4, p. 112].
Proposition 5.5.1 Let Ω ⊂ Rn be a bounded NTA domain and fix some point x0 ∈ Ω
along with a set E ⊆ ∂Ω. If u is a harmonic function in Ω which is nontangentially
bounded from below on E then for every κ > 0 it follows that
856 5 Green Functions and Poisson Kernels for the Laplacian
κ−n.t.
u∂Ω exists at ω x0 -a.e. point on E. (5.5.1)

A version of the Fatou-type result from Proposition 5.5.1, with the harmonic
measure replaced by the “surface measure” σ := H n−1 ∂Ω, is stated below.
Proposition 5.5.2 Let Ω ⊂ Rn be a bounded NTA domain with an upper Ahlfors
regular boundary and abbreviate σ := H n−1 ∂Ω. If u is a harmonic function in Ω
which is nontangentially bounded from below on some set E ⊆ ∂Ω then for every
κ > 0 it follows that
κ−n.t.
u∂Ω exists σ-a.e. on E, and is independent of κ. (5.5.2)

In particular, (5.5.2) holds for any κ > 0, any E ⊆ ∂Ω, and any harmonic
function u in Ω which, for some ε > 0, satisfies
 ε 
Nκ u (x) < +∞ at σ-a.e. point x ∈ E. (5.5.3)

Proof The existence part in (5.5.2) is a consequence of Proposition 5.5.1 and Propo-
sition 5.4.1. The independence on the parameter κ then follows from this and [112,
(8.1.20)]. 
Given an open set Ω ⊆ Rn , for each κ > 0 and p ∈ (0, ∞] consider the space of
harmonic functions
 
Hκ (Ω) := u ∈ 𝒞∞ (Ω) : Δu = 0 in Ω, and Nκ u ∈ L p (∂Ω, σ)
p
(5.5.4)

where σ := H n−1 ∂Ω. In terms of this, we have the following global, quantitative
Fatou-type result. Before stating it, the reader is alerted to the fact that the space
p
Nκ (Ω; σ) is defined as in (A.0.98) (corresponding to μ := σ).
Proposition 5.5.3 Let Ω ⊂ Rn be a bounded NTA domain with an upper Ahlfors
regular boundary and abbreviate σ := H n−1 ∂Ω. Then for each aperture parameter
κ > 0 and each integrability exponent p ∈ (0, ∞],
p p
Hκ (Ω) is a closed subspace of Nκ (Ω; σ) (5.5.5)
p
and, when considering Hκ (Ω) equipped with the quasi-norm from the larger ambient
p
Nκ (Ω; σ) (cf. (A.0.98)), the nontangential boundary trace induces a well-defined,
linear, and continuous operator in the context
κ−n.t.
u −→ u∂Ω ∈ L p (∂Ω, σ).
p
Hκ (Ω) (5.5.6)
p
Moreover, both the space Hκ (Ω) and the above nontangential trace operator are
actually independent of κ.
Proof First, the claim in (5.5.5) is seen by expressing
p  p 
Hκ (Ω) = u ∈ Nκ (Ω; σ) : Δu = 0 in Ω , (5.5.7)
5.6 Boundary Behavior of the Green Function in NTA Domains 857

then invoking [112, Proposition 8.3.5] (bearing in mind that harmonicity is preserved
under uniform convergence on compact sets). Next, that the operator (5.5.6) is
well defined, linear, and continuous follows from (5.5.4), Proposition 5.5.2, [112,
(8.9.44)], [112, (8.9.8)], [112, (8.8.52)], and [112, (5.2.4)]. Lastly, the independence
on the parameter κ of the objects involved in (5.5.6) may be justified with the help
of [112, Proposition 8.9.8]. 

5.6 Boundary Behavior of the Green Function in NTA Domains

The results on the nontangential boundary trace of the derivatives of the Green
function recorded in (5.6.1)-(5.6.2) below extend work in [25, Theorem 3, p. 275]
dealing with the case of bounded Lipschitz domains, and also refine [81, Lemma 3.1,
p. 336] where δ-Reifenberg flat chord-arc domains with δ > 0 sufficiently small have
been considered.

Lemma 5.6.1 Assume Ω ⊆ Rn is a bounded NTA domain with an upper Ahlfors

regular boundary. Abbreviate σ := H n−1 ∂Ω and denote by GΩ (·, ·) the Green
function for the Laplacian in Ω. Also, fix a point x0 ∈ Ω and an aperture parameter
κ > 0. Then the nontangential boundary trace
  κ−n.t.
∇GΩ (·, x0 )  exists (in Rn ) at σ-a.e. point on ∂Ω
∂Ω (5.6.1)
and is actually independent of the parameter κ > 0.

In particular, if ν denotes the geometric measure theoretic outward unit normal

to Ω, then the normal derivative of GΩ (·, x0 ) may be meaningfully defined as
   κ−n.t. 
∂ν GΩ (·, x0 ) := ν, ∇GΩ (·, x0 ) ∂Ω at σ-a.e. point on ∂Ω, (5.6.2)

and this definition is independent of the aperture parameter κ.

We emphasize that the same result holds for unbounded NTA domains, and also
for the Green function with pole at infinity.
Proof of Lemma 5.6.1 This is a consequence of Theorem 5.3.5, bearing in mind
(5.4.8) and the fact that
any bounded NTA domain with an upper
(5.6.3)
Ahlfors regular boundary is a UR domain,
as seen from [112, (5.11.4)], [112, (5.10.24)], and the definition of a UR domain
from [112, Definition 5.10.6]. 
In [49, Question 2, p. 49] Garnett and Marshall raise the issue of determining
minimal regularity conditions on an given open set ensuring that the Poisson kernel
may be expressed as the normal derivative of the Green function. Our next theorem
858 5 Green Functions and Poisson Kernels for the Laplacian

shows that merely being an NTA domain with upper Ahlfors regular boundary will
do. In particular, this sharpens [16, Theorem 5.7, p. 79], [81, Proposition A.1.1,
p. 382], [81, Proposition A.2.2, p. 385], and [82, Lemma 3.4, p. 16], where Kenig
and Toro have considered δ-Reifenberg flat chord-arc domains in Rn and/or imposed
the condition that the Poisson kernel is locally H n−1 -square-integrable. Our result
dispenses with these additional assumptions; for example, Theorem 5.6.2 applies to
arbitrary BMO1 domains in Rn .
Theorem 5.6.2 Assume Ω ⊆ Rn is a bounded NTA domain with an upper Ahlfors
regular boundary. Let ν denote the geometric measure theoretic outward unit normal
to Ω, and abbreviate σ := H n−1 ∂Ω. Also, denote by GΩ (·, ·) the Green function
for the Laplacian in Ω, and fix an arbitrary point x0 ∈ Ω. Then, with the normal
derivative of GΩ (·, x0 ) defined as in (5.6.2), one has
dω x0
−∂ν GΩ (·, x0 ) = k x0 := at σ-a.e. point on ∂Ω. (5.6.4)
dσ
We wish to stress that the same result holds for unbounded NTA domains, in which
case we may also allow the Green function and the Poisson kernel to have poles at
infinity (in the sense of [80, Lemma 3.7, p. 390]). We shall give Theorem 5.6.2 two
proofs, the first of which casts this result as a direct corollary of Theorem 5.3.6 and
Lemma 5.4.3.
First Proof of Theorem 5.6.2 From (5.2.29), [112, (8.9.180)], [112, (5.11.4)], and
[112, (5.2.4)] we see that
any bounded NTA domain Ω with an upper Ahlfors regular boundary
is regular for the Dirichlet Problem for the Laplacian (in the sense of
Definition 5.2.3), is locally pathwise nontangentially accessible (in the (5.6.5)
sense of [112, Definition 8.9.14]), has a lower Ahlfors regular boundary
(hence σ is a doubling measure on ∂Ω), and satisfies ∂∗ Ω = ∂Ω.
Granted this, and keeping in mind (5.6.3) as well as (5.4.8), we may invoke Theo-
rem 5.3.6 to conclude that (5.6.4) holds. 
The second proof of Theorem 5.6.2 relies on the version of the Divergence
Formula recorded in [112, Theorem 1.2.1] in rather transparent fashion.
Second Proof of Theorem 5.6.2 Bearing in mind that any bounded NTA domain is
a bounded open set which is regular for the Dirichlet Problem for the Laplacian and
whose complement is n-thick, from (5.3.14) and (5.3.13) we see that
∫
GΩ (x, x0 ) = H(x, x0 ) − EΔ (x − x0 ) = EΔ (x0 − z) dω x (z) − EΔ (x − x0 )
∂Ω
∫
= EΔ (x − z) dω x0 (z) − EΔ (x − x0 ), ∀x ∈ Ω \ {x0 }, (5.6.6)
∂Ω

where EΔ is the standard fundamental solution for the Laplacian in Rn defined in

(1.5.56). Fix now ϕ ∈ 𝒞∞
c (R ) arbitrary and define
n
5.6 Boundary Behavior of the Green Function in NTA Domains 859
∫
u(z) := ϕ(z) − EΔ (x − z)(Δϕ)(x) dx, ∀z ∈ Ω. (5.6.7)
Ω

Then, by design, u ∈ 𝒞0 (Ω) and Δu = 0 in Ω. Consequently,

∫ ∫
ϕ(x0 ) − EΔ (x − x0 )(Δϕ)(x) dx = u(x0 ) = u(z) dω x0 (z)
Ω ∂Ω
∫ ∫ ∫ 
= ϕ(z) dω x0 (z) − EΔ (x − z)(Δϕ)(x) dx dω x0 (z)
∂Ω ∂Ω Ω
∫ ∫ ∫
= ϕ dω x0
− GΩ (x, x0 )(Δϕ)(x) dx − EΔ (x − x0 )(Δϕ)(x) dx.
∂Ω Ω Ω
(5.6.8)

Ultimately this goes to show that for each fixed point x0 ∈ Ω the Green function
satisfies
∫ ∫
GΩ (x, x0 )(Δϕ)(x) dx = ϕ dω x0 − ϕ(x0 ), ∀ϕ ∈ 𝒞∞c (R ).
n
(5.6.9)
Ω ∂Ω

To proceed, pick now an arbitrary function ϕ ∈ 𝒞∞ c (R \ {x0 }) and consider the

n

vector field
F := GΩ (·, x0 )∇ϕ − ϕ∇GΩ (·, x0 ) in Ω. (5.6.10)
 ∞ n
Then, thanks to (5.3.4) and the fact that ϕ vanishes near x0 , we have F ∈ 𝒞 (Ω)
and
divF = GΩ (·, x0 )Δϕ ∈ L 1 (Ω, L n ), (5.6.11)
where the membership is a consequence of the estimates for the Green function
recorded in Theorem 5.3.1. Also, from Theorem 5.3.2 (or (5.3.11)) and Lemma 5.6.1
we see that, for each κ > 0 fixed,
κ−n.t.
F∂Ω exists (in Rn ) at σ-a.e. point on ∂Ω (5.6.12)

and, in fact,
κ−n.t.   κ−n.t.
F∂Ω = −ϕ ∇GΩ (·, x0 ) ∂Ω at σ-a.e. point on ∂Ω. (5.6.13)

In particular,
 κ−n.t. 
ν · F∂Ω = −ϕ ∂ν GΩ (·, x0 ) at σ-a.e. point on ∂Ω. (5.6.14)

Granted these properties, [112, Theorem 1.2.1] applies and gives that, on the one
hand,
860 5 Green Functions and Poisson Kernels for the Laplacian
∫ ∫ ∫  κ−n.t. 
GΩ (x, x0 )(Δϕ)(x) dx = divF dL n = ν · F∂Ω dσ
Ω Ω ∂Ω
∫
=− ϕ ∂ν GΩ (·, x0 ) dH n−1 . (5.6.15)
∂Ω

On the other hand, from (5.6.9), the current choice of ϕ, and Proposition 5.4.1 we
conclude that
∫ ∫ ∫
GΩ (x, x0 )(Δϕ)(x) dx = ϕ dω =
x0
k x0 ϕ dH n−1 . (5.6.16)
Ω ∂Ω ∂Ω

Comparing (5.6.16) with (5.6.15) then yields (5.6.4) on account of [112, Corol-
lary 3.7.3]. 

The following theorem generalizes results from [81, Lemmas 3.1-3.2, pp. 336-
337] [81, §A.2, pp. 384-389] (where the authors attribute them to G. David) and
[82, Lemma 3.2, p. 10] which, compared with Theorem 5.6.3 below, have been
established under certain additional flatness assumptions on the underlying domain
and/or assuming local H n−1 -square-integrability for the associated Poisson kernel.
Here we eliminate these assumptions; in particular, Theorem 5.6.3 applies to arbitrary
BMO1 domains in Rn .

Theorem 5.6.3 Assume Ω ⊆ Rn is a bounded NTA domain with an upper Ahlfors

regular boundary; in particular, Ω is a set of locally finite perimeter. Denote by
ν = (ν1, . . . , νn ) its geometric measure theoretic outward unit normal and abbreviate
σ := H n−1 ∂Ω. Also, denote by GΩ (·, ·) the Green function for the Laplacian in
Ω, and fix an arbitrary point x0 ∈ Ω. Then (with the dependence on the aperture
parameter dropped) for each j, k ∈ {1, . . . , n} one has
  n.t.    n.t. 
ν j ∂k GΩ (·, x0 ) ∂Ω = νk ∂j GΩ (·, x0 ) ∂Ω at σ-a.e. point on ∂Ω. (5.6.17)

In fact, with the Poisson kernel k x0 defined as in (5.4.1), one has

  n.t.  dω x0 
∇GΩ (·, x0 ) ∂Ω = −k x0 ν = − ν at σ-a.e. point on ∂Ω. (5.6.18)
dσ

In
 particular,
 for any vector field T : ∂Ω → Cn which is tangential, in the sense that
ν, T = 0 at σ-a.e. point on ∂Ω, one has
   n.t. 
 ∇GΩ (·, x0 ) 
T, = 0 at σ-a.e. point on ∂Ω. (5.6.19)
∂Ω

We emphasize that the same results are valid for unbounded NTA domains, in
which scenario one may also allow the Green function and the Poisson kernel to
have poles at infinity (cf. [80, Lemma 3.7, p. 390]).
Proof of Theorem 5.6.3 This follows from Theorem 5.3.5, bearing in mind (5.4.8)
and (5.6.3), as well as Theorem 5.6.2. 
5.7 The L p Dirichlet Problem for the Laplacian and Integral Representations 861

5.7 The L p Dirichlet Problem for the Laplacian and Integral

Representations

The following result appears as [71, Lemma 4.11, p. 101].

Lemma 5.7.1 If Ω is a bounded NTA domain in Rn , then there exists R > 0 with
the following significance. Assume x0, x1 ∈ ∂Ω and 0 < r0, r1 < R are such that
Δ(x1, r1 ) ⊂ Δ(x0, r0 /2). Then, with Ar (x0 ) denoting a corkscrew point for x0 at
scale r, one has
  ω x (Δ(x1, r1 ))
ω Ar (x0 ) Δ(x1, r1 ) ≈ x , (5.7.1)
ω (Δ(x0, r0 ))
uniformly for x ∈ Ω \ B(x0, 2r0 ).
From item (iii) in Lemma 5.2.1 it follows that if x1, x2 are two arbitrary points in
Ω, then the measures ω x1 and ω x2 are mutually absolutely continuous. As such, the
Radon-Nikodym derivative dω x1 /dω x2 exists and, hence, the following definition is
meaningful.
Definition 5.7.2 Let Ω be a bounded NTA domain in Rn and fix x, x0 ∈ Ω. Then the
kernel function associated with Ω is defined as the Radon-Nikodym derivative
dω x
Kx0 (x, ·) := on ∂Ω. (5.7.2)
dω x0
In particular, if x0 ∈ Ω is fixed, then for every x ∈ Ω one has
 
ω x Δ(y, r)
Kx0 (x, y) = lim x   for ω x0 -a.e. y ∈ ∂Ω. (5.7.3)
r0 ω 0 Δ(y, r)

Even though, a priori, Kx0 (x, y) is only defined for ω x0 -a.e. y ∈ ∂Ω according to
[71, Theorem 5.5, p. 104] it turns out that

if Ω ⊆ Rn is a bounded NTA domain, then Kx0 (x, y)

exists for all y ∈ ∂Ω, is a harmonic function in x ∈ Ω, and (5.7.4)
a continuous function in the variable y ∈ ∂Ω.
In fact, [71, Corollary 7.1, p. 115] implies that
if Ω ⊆ Rn is a bounded NTA domain, then there exists α > 0
which depends only on the NTA constants of Ω such that (5.7.5)
Kx0 (x, ·) ∈ 𝒞α (∂Ω) for each fixed point x ∈ Ω.

In particular, from (5.7.5), or (5.2.15), we see that

if Ω ⊆ Rn is a bounded NTA domain, then the space
(5.7.6)
L 1 (∂Ω, ω x0 ) is independent of the choice of x0 ∈ Ω.
We continue by recording several other properties of the kernel function in NTA
domains which are due to D. Jerison and C. Kenig ([71], [79]).
862 5 Green Functions and Poisson Kernels for the Laplacian

Lemma 5.7.3 Let Ω be a bounded NTA domain in Rn , and fix x0 ∈ Ω. Then there

exists r0 > 0 such that
1
Kx0 (Ar (z), y) ≈ x , (5.7.7)
ω 0 (Δ(z, r))
uniformly for z ∈ ∂Ω \ B(x0, 2r), y ∈ Δ(z, r), and 0 < r < r0 , where Ar (z) denotes a
corkscrew point for z at scale r.

Indeed, estimate (5.7.7) is a consequence of (5.7.3) and Lemma 5.7.1.

We next recall [71, Theorem 5.8, p. 105] which provides a way of extending
boundary functions (which are absolutely integrable with respect to the harmonic
measure) harmonically inside of a given an NTA domain. Before stating this, we
recall that, given a bounded NTA domain Ω ⊆ Rn and some fixed point x0 ∈ Ω, the
Hardy-Littlewood maximal operator Mω x0 associated with the harmonic measure
ω x0 on ∂Ω is (cf. [112, (7.7.87)])
  ⨏ 
Mω x0 f (x) := sup | f | dω x0 , ∀x ∈ ∂Ω, (5.7.8)
r >0 Δ(x,r)

for each f ∈ L 1 (∂Ω, ω x0 ).

Theorem 5.7.4 Suppose Ω is a bounded NTA domain in Rn , and select κ > 0

sufficiently large (depending on the interior corkscrew constants of Ω). Also, fix a
point x0 ∈ Ω and recall the kernel function Kx0 (·, ·) from (5.7.2). In this context, if
for f ∈ L 1 (∂Ω, ω x0 ) one defines
∫ ∫
u(x) := f dω x = Kx0 (x, y) f (y) dω x0 (y), ∀x ∈ Ω, (5.7.9)
∂Ω ∂Ω

then u is a well-defined function, which is harmonic in Ω and satisfies

κ−n.t.
u∂Ω = f at ω x0 -a.e. point on ∂Ω, (5.7.10)

and
(the truncated version of) Nκ u is comparable
(5.7.11)
with (the local version of) Mω x0 f .

This is essentially [71, Theorem 5.8, p. 105]. Strictly speaking, the theorem just
cited only contains the left-pointing inequality in (5.7.11), but the opposite one is
established much as in the proof of [79, Lemma 1.4.2, p. 14]. To give the main step in
this argument, let f ∈ L 1 (∂Ω,
 ω 1 ), f ≥ 0, andlet r0 be as in Lemma 5.7.3. Then for
x0

each z ∈ ∂Ω, 0 < r < min r0, 2 dist(x0, ∂Ω) , and some point x belonging to the
set [B(z, 2r) \ B(z, r)] ∩ Γκ (z) (which, thanks to the interior corkscrew condition,
is nonempty if the aperture κ is large enough), we may write
5.7 The L p Dirichlet Problem for the Laplacian and Integral Representations 863
⨏ ∫
f (y) dω x0 (y) ≤ C Kx0 (x, y) f (y) dω x0 (y)
Δ(z,r) ∂Ω
∫
=C f (y) dω x (y) = Cu(x) ≤ C(Nκ u)(z), (5.7.12)
∂Ω

by (5.7.7). A slight variant of this estimate then yields the desired conclusion.
Our next result deals with the issue of uniqueness for the Dirichlet Problem with
the boundary trace considered in the nontangential sense and with nontangential
maximal function control.

Theorem 5.7.5 Let Ω ⊂ Rn be a bounded NTA domain with an upper Ahlfors

regular boundary and abbreviate σ := H n−1 ∂Ω. Denote by ω x0 the harmonic
measure on ∂Ω with pole at the point x0 ∈ Ω. Also, select two integrability exponents
p, p ∈ (1, ∞) satisfying 1/p + 1/p = 1, and fix an aperture parameter κ > 0.
x0
Then, if the Poisson kernel k x0 = dω p
dσ belongs to L (∂Ω, σ), it follows that

u ∈ 𝒞∞ (Ω), Δu = 0 in Ω ⎫ ⎪
⎪
κ−n.t. ⎪
⎬
⎪
u∂Ω = 0 at σ-a.e. point on ∂Ω ⎪ =⇒ u ≡ 0 in Ω. (5.7.13)
⎪
⎪
and Nκ u ∈ L p (∂Ω, σ) ⎪
⎭
Proof Denote by GΩ (·, ·) the Green function associated with the Laplacian in Ω, and
recall that we are assuming

k x0 ∈ L p (∂Ω, σ). (5.7.14)
Then Lemma 5.4.3, Lemma 5.4.4, [112, Proposition 8.4.1], and [112, Corol-
lary 7.6.3] guarantee the existence of a constant C ∈ (0, ∞) with the property
that for every ε > 0 sufficiently small we have
     
 ε
Nκ ∇GΩ (·, x0 )  p ≤ C  k x0  L p (∂Ω,σ) . (5.7.15)
L (∂Ω,σ)

In particular, for ε > 0 small, from (5.7.15), (5.4.10), and [112, (8.2.26)] we conclude
that
  
Nκε GΩ (·, x0 ), Nκε ∇GΩ (·, x0 ) ∈ L p (∂Ω, σ). (5.7.16)

Thanks to (5.7.16), [112, Proposition 5.9.16], and [112, (5.2.4), (8.8.52), (8.9.180)],
all hypotheses of Theorem 4.1.1 are satisfied for L := Δ. As such, (4.1.2) presently
implies that whenever u is as in the left-hand side of (5.7.13) we necessarily have
u(x0 ) = 0. Since for any other point x1 ∈ Ω we have

k x1 = Kx0 (x1, ·)k x0 ∈ L p (∂Ω, σ), (5.7.17)

by virtue of (5.7.2) and (5.7.4) (which implies that Kx0 (x1, ·) is continuous and
bounded on ∂Ω), the same type of argument as above with x1 playing the role of
864 5 Green Functions and Poisson Kernels for the Laplacian

x0 yields u(x1 ) = 0. Given that the point x1 ∈ Ω is arbitrary, this establishes the
implication in (5.7.13), finishing the proof of the theorem. 
The next theorem in this section refines and expands the scope of [79, Theo-
rem 1.7.3, p. 29]. Its proof relies essentially on results due to G. David, D. Jerison,
and C. Kenig (cf. [32], [71]). Among other things, this result builds a bridge be-
tween the well-posedness of the L p -Dirichlet Problem (for the Laplacian) and the
availability of L p -nontangential maximal function estimates for the solution of the
classical Dirichlet Problem. Before stating it, the reader is reminded that the space
p
Hκ (Ω) has been defined in (5.5.4).
Theorem 5.7.6 Let Ω ⊂ Rn be a bounded NTA domain with the property that ∂Ω
is an upper Ahlfors regular set, and abbreviate σ := H n−1 ∂Ω. Denote by ω x0 the
harmonic measure with pole at the point x0 ∈ Ω. Also, fix some κ > 0 along with
p, p ∈ (1, ∞) such that p1 + p1 = 1. Then the following statements are equivalent:
(i) There exists a constant C ∈ (0, ∞) such that whenever f ∈ 𝒞0 (∂Ω) and u is
the unique solution of the classical Dirichlet Problem

⎧
⎪ u ∈ 𝒞0 (Ω) ∩ 𝒞∞ (Ω),
⎪
⎪
⎨
⎪
Δu = 0 in Ω, (5.7.18)
⎪
⎪ 
⎪
⎪ u = f on ∂Ω,
⎩ ∂Ω
one has
Nκ u L p (∂Ω,σ) ≤ C f  L p (∂Ω,σ) . (5.7.19)
In other words, there exists C ∈ (0, ∞) with the property that
 
Nκ (PI f ) p ≤ C f  L p (∂Ω,σ), ∀ f ∈ 𝒞0 (∂Ω). (5.7.20)
L (∂Ω,σ)

(ii) One has

ω x0 ∈ Bp (σ), (5.7.21)
 
in the sense that for each scale r ∈ 0, diam ∂Ω , each location x ∈ ∂Ω, and
each pole x0 ∈ B(x, r) ∩ Ω with dist(x0, ∂Ω) ≈ r, it follows that the harmonic
measure ω x0 is absolutely continuous with respect to the surface measure σ,
x0
p
the Poisson kernel k x0 := dω dσ belongs to L (∂Ω, σ), and there exists some
C ∈ (0, ∞) such that the following reverse Hölder condition is satisfied:
⨏ 
 1 ⨏
p
(k x0 ) p dσ ≤C k x0 dσ. (5.7.22)
Δ(x,r) Δ(x,r)

p
(iii) For every given f ∈ L p (∂Ω, σ) there exists a unique function u ∈ Hκ (Ω) with
κ−n.t.
the property that u∂Ω = f at σ-a.e. point on ∂Ω. In other words,

the nontangential boundary trace operator

(5.7.23)
is an isomorphism in the context of (5.5.6).
5.7 The L p Dirichlet Problem for the Laplacian and Integral Representations 865

(iv) One has

 κ−n.t. 
Nκ u L p (∂Ω,σ) ≈ u∂Ω  L p (∂Ω,σ), uniformly for u ∈ Hκ (Ω).
p
(5.7.24)

From Proposition 5.2.4 and (5.2.29) we see that the classical Dirichlet Problem
(cf. (5.7.18)) is uniquely solvable in the set Ω, so the statement of Theorem 5.7.6
is meaningful. Before presenting its proof, we wish to comment on terminology.
Specifically, at least two labels have been in use for the property described in item
(i). Originally, the validity of this property is indicated in [79, Definition 1.7.4] by
simply saying that “(D) p holds (for the Laplacian)” but, subsequently, the phrase
“(D p ) is solvable” has become prevalent in the literature. We find the latter prac-
tice unfortunate since it redefines the meaning of the word “solvable” in a rather
unconventional (and somewhat misleading) manner.
Here is the proof of Theorem 5.7.6.
 
Proof of Theorem 5.7.6 To show that (i) implies (ii), fix a scale r ∈ 0, diam ∂Ω
along with a location x ∈ ∂Ω and some point x0 ∈ B(x, r) ∩ Ω satisfying
dist(x0, ∂Ω) ≈ r. From Proposition 5.4.1 we know that ω xo and σ are mutually
absolutely continuous. We shall estimate
⨏ 
 1  − 1
p
(k x0 ) p dσ = σ Δ(x, r) p k x0  L p (Δ(x,r),σ) (5.7.25)
Δ(x,r)

via duality. Specifically, pick an arbitrary non-negative function f ∈ 𝒞0 (∂Ω) with

supp f ⊆ Δ(x, r) and write
∫ ∫
f k x0 dσ = f dω x0 = u(x0 ), (5.7.26)
Δ(x,r) ∂Ω

where u is the unique solution of the classical Dirichlet Problem (5.7.18) for the
boundary datum f . Since there exists some large aperture parameter κ > 0 and some
small ε > 0 such that x0 ∈ Γκ (z) for each z ∈ Δ(x, εr), we conclude from (5.7.26)
that ∫
f k x0 dσ ≤ (Nκ u)(z) for each z ∈ Δ(x, εr). (5.7.27)
Δ(x,r)

Taking integral averages, then using Hölder’s inequality, [112, Proposition 8.4.1],
the fact that σ is doubling, and (5.7.19), we therefore obtain
866 5 Green Functions and Poisson Kernels for the Laplacian
∫ ⨏ ⨏  p1
fk x0
dσ ≤ Nκ u dσ ≤ (Nκ u) p dσ
Δ(x,r) Δ(x,εr) Δ(x,εr)

 − 1
≤ σ Δ(x, εr) p Nκ u L p (∂Ω,σ)
 − 1
≤ Cσ Δ(x, r) p Nκ u L p (∂Ω,σ)
 − 1
≤ Cσ Δ(x, r) p  f  L p (∂Ω,σ)
 − 1
= Cσ Δ(x, r) p  f  L p (Δ(x,r),σ) . (5.7.28)

This ultimately proves that

 − 1   −1
σ Δ(x, r) p k x0  L p (Δ(x,r),σ) ≤ Cσ Δ(x, r) . (5.7.29)

Since the right-hand side in (5.7.22) is

⨏  
ω x0 Δ(x, r)
C k x0 dσ = C   , (5.7.30)
Δ(x,r) σ Δ(x, r)

the desired conclusions now follow from (5.7.25), (5.7.29), (5.7.30), and Bourgain’s
Lemma.
Let us now show that (ii) implies (i). Given a function f ∈ 𝒞0 (∂Ω), we know
from (5.2.29) and Definitions 5.2.2-5.2.3 that the unique solution of the classical
Dirichlet Problem (5.7.18) is given by
∫
u(x) = (PI f )(x) = f dω x, ∀x ∈ Ω. (5.7.31)
∂Ω

Granted this, and bearing in mind (5.7.11), it follows that estimate (5.7.19) is implied
by the boundedness on L p (∂Ω, σ) of the Hardy-Littlewood maximal operator Mω x0
associated with the harmonic measure ω x0 on ∂Ω (cf. (cf. [112, (7.7.87)]). In turn,
according to item (vii) in [112, Proposition 7.7.9], the latter boundedness property
holds if ω x0 ∈ Bp (σ). The argument so far shows that we have the equivalence
(i) ⇔ (ii).
We next propose to show that (i)-(ii) imply (iii). With this goal in mind, fix an
arbitrary function f ∈ L p (∂Ω, σ). Then [112, Corollary 3.7.3] implies that there
exists a sequence
{ f j } j ∈N ⊆ 𝒞0 (∂Ω) such that
(5.7.32)
f j → f in L p (∂Ω, σ) as j → ∞.
If for each j ∈ N we now define
∫
u j (x) := (PI f j )(x) = f j dω x, ∀x ∈ Ω, (5.7.33)
∂Ω
5.7 The L p Dirichlet Problem for the Laplacian and Integral Representations 867

then
u j ∈ 𝒞0 (Ω) ∩ 𝒞∞ (Ω), Δu j = 0 in Ω,
 (5.7.34)
and u j ∂Ω = f j on ∂Ω.
Since for each j, j  ∈ N we have u j − u j  = PI( f j − f j  ), from (5.7.19) and [112,
(8.9.8)] we conclude that

u j − u j   Nκp (Ω;σ) = Nκ (u j − u j  ) L p (∂Ω,σ)

≈  f j − f j   L p (∂Ω,σ), uniformly in j, j  ∈ N. (5.7.35)

In particular, from (5.7.35) and (5.7.32) we conclude that the sequence {u j } j ∈N

p
is Cauchy in Nκ (Ω; σ). Granted this, we may invoke [112, Proposition 8.3.5] to
conclude that
p p
there exists u ∈ Nκ (Ω; σ) such that u j → u in Nκ (Ω; σ) as j → ∞. (5.7.36)
p p
Given that each u j belongs to Hκ (Ω) which is a closed subspace of Nκ (Ω; σ) (cf.
(5.5.5)), it follows that actually
p
u ∈ Hκ (Ω). (5.7.37)

Granted this and recalling (5.5.6), it follows that the nontangential boundary trace
κ−n.t.
g := u∂Ω is well defined in L p (∂Ω, σ). (5.7.38)

Upon passing to the limit j  → ∞ in (5.7.35) we obtain

u j − u Nκp (Ω;σ) ≈  f j − f  L p (∂Ω,σ), uniformly in j ∈ N, (5.7.39)

on account of (5.7.36) and (5.7.32). In concert with [112, (8.9.8)], (5.7.34), [112,
(8.9.10)], and (5.7.38), this permits us to estimate

 f j − f  L p (∂Ω,σ) ≈ u j − u Nκp (Ω;σ)

 κ−n.t. κ−n.t. 
= Nκ (u j − u) L p (∂Ω,σ) ≥ u j ∂Ω − u∂Ω  L p (∂Ω,σ)

=  f j − g L p (∂Ω,σ), uniformly in j ∈ N. (5.7.40)

From (5.7.40) and (5.7.32) we then conclude that necessarily

f = g at σ-a.e. point on ∂Ω. (5.7.41)

p
At this stage, from (5.7.37) and (5.7.41) we see that u is a function in Hκ (Ω) with
κ−n.t.
the property that u∂Ω = f at σ-a.e. point on ∂Ω. That actually u is the only
function with these two qualities is, thanks to the current working assumptions, a
868 5 Green Functions and Poisson Kernels for the Laplacian

consequence of the uniqueness result established in Theorem 5.7.5. The proof of the
fact that (i)-(ii) imply (iii) is therefore complete.
Going further, note that (5.7.24) is implied by (5.7.23) and the Open Mapping
Theorem (whose applicability is ensured by the completeness result proved in [112,
Proposition 8.3.5]). Hence, (iii) ⇒ (iv).
To conclude, there remains to show that (iv) implies (i). To this end, observe that

𝒞0 (∂Ω) ⊆ L q (∂Ω, σ) (5.7.42)

0<q ≤∞

and that
 
PI : 𝒞0 (∂Ω) −→ u ∈ 𝒞0 (Ω) ∩ 𝒞∞ (Ω) : Δu = 0 in Ω ⊆
q
Hκ (Ω).
0<q ≤∞
(5.7.43)
In addition,
κ−n.t. 
(PI f )∂Ω = (PI f )∂Ω = f , ∀ f ∈ 𝒞0 (∂Ω). (5.7.44)
Granted (5.7.42)-(5.7.44), estimate (5.7.20) is then readily implied by (5.7.24). This
establishes the implication (iv) ⇒ (i), and finishes the proof of Theorem 5.7.6. 
Our next result extends [71, Theorem 10.1, p. 132], dealing with the special case
of bounded BMO1 domains in Rn , and also refines [79, Theorem 1.7.3, p. 29]. In
particular, the integral representation formula (5.7.49) involving the normal deriva-
tive of the Green function is new in the current context (compare with [81] where the
more restrictive class of δ-Reifenberg flat chord-arc domains has been considered).
Theorem 5.7.7 Suppose Ω ⊂ Rn is a bounded NTA domain whose boundary is
an upper Ahlfors regular set. Abbreviate σ := H n−1 ∂Ω and denote by ω x0 the
harmonic measure with pole at the point x0 ∈ Ω. Consider the critical exponent
 
qΩ := sup q ∈ (1, ∞) : ω x0 ∈ Bq (σ) (5.7.45)

(which, according to (5.4.5), is well defined and belongs to the interval (1, ∞]), then
let pΩ ∈ [1, ∞) be its Hölder conjugate exponent, i.e.,
qΩ
qΩ −1 if qΩ < ∞,
pΩ := (5.7.46)
1 if qΩ = ∞.

Then, for each fixed κ > 0 and p ∈ (pΩ, ∞), the Dirichlet Problem with the
boundary trace considered in the nontangential sense and with nontangential maxi-
mal function control,

⎧
⎪ u ∈ 𝒞∞ (Ω),
⎪
⎪
⎪ Δu = 0 in Ω,
⎪
⎨
⎪
(5.7.47)
⎪
⎪ Nκ u ∈ L p (∂Ω, σ),
⎪
⎪
⎪ uκ−n.t. = f σ-a.e. on ∂Ω,
⎪
f ∈ L p (∂Ω, σ),
⎩ ∂Ω
5.7 The L p Dirichlet Problem for the Laplacian and Integral Representations 869

has a unique solution. Moreover, there exists a constant C = C(Ω, κ, p) ∈ (0, ∞)

such that said solution satisfies

Nκ u L p (∂Ω,σ) ≤ C f  L p (∂Ω,σ) . (5.7.48)

Also, if ν stands for the geometric measure theoretic outward unit normal to Ω and
if GΩ (·, ·) denotes the Green associated with the Laplacian in Ω then, for each fixed
point x0 ∈ Ω, the unique solution u of (5.7.47) may be represented as
∫ ∫ ∫
u(x) = f dω x = f k x dσ = Kx0 (x, y) f (y) dσ(y)
∂Ω ∂Ω ∂Ω
∫
=− ∂ν GΩ (·, x) f dσ, ∀x ∈ Ω, (5.7.49)
∂Ω

where the normal derivative of the Green function is defined as in (5.6.2).

Conversely, wherever p ∈ (1, ∞) is such that the Dirichlet Problem (5.7.47)
is uniquely solvable and the solution satisfies (5.7.48) it follows that p > pΩ . In
particular,
pΩ may be characterized as the infimum of all integrability expo-
nents p ∈ (1, ∞) with the property that the L p Dirichlet Problem (5.7.50)
in Ω (cf. (5.7.47)) is well posed.
Proof For starters, from (5.7.45) and item (vi) in [112, Proposition 7.7.9] we con-
clude that  
q ∈ (1, ∞) : ω x0 ∈ Bq (σ) = (1, qΩ ). (5.7.51)
Let p ∈ (1, ∞) be such that 1
p + p1 = 1. The fact that p ∈ (pΩ, ∞) forces p ∈ (1, qΩ ),
hence
ω x0 ∈ Bp (σ). (5.7.52)
In particular, (cf. (5.7.17)) we have

k x = K(x, ·)k x0 ∈ L p (∂Ω, σ), for each x ∈ Ω. (5.7.53)

To proceed, pick an arbitrary f ∈ L p (∂Ω, σ). On account of (5.7.52), Theo-

p
rem 5.7.6 ensures that there is a unique function u ∈ Hκ (Ω) with the property that
κ−n.t.
u∂Ω = f at σ-a.e. point on ∂Ω, and such that
 κ−n.t. 
Nκ u L p (∂Ω,σ) ≈ u∂Ω  L p (∂Ω,σ) (5.7.54)

where the proportionality constants depend only on Ω, p, and κ. This proves that the
Dirichlet Problem (5.7.47) is uniquely solvable and the solution satisfies (5.7.48).
From the proof of Theorem 5.7.6 (cf. (5.7.36) in particular) we also know that if
a sequence { f j } j ∈N ⊆ 𝒞0 (∂Ω) has been chosen such that f j → f in L p (∂Ω, σ) as
j → ∞ then at each fixed point x ∈ Ω the solution u of (5.7.47) may be compute
as
870 5 Green Functions and Poisson Kernels for the Laplacian
∫
u(x) = lim (PI f j )(x) = lim f j dω x . (5.7.55)
j→∞ j→∞ ∂Ω
Together with (5.7.53) and Theorem 5.6.2, for each x ∈ Ω this permits us to write
∫ ∫
u(x) = lim f j dω = lim
x
f j k x dσ
j→∞ ∂Ω j→∞ ∂Ω
∫ ∫
= f k x dσ = − ∂ν GΩ (·, x) f dσ (5.7.56)
∂Ω ∂Ω

which, bearing in mind (5.4.1) and (5.7.9), proves (5.7.49).

In the converse direction, assume p ∈ (1, ∞) is such that the Dirichlet Problem
(5.7.47) is uniquely solvable and the solution satisfies (5.7.48). Since 𝒞0 (∂Ω) is a
subspace of L p (∂Ω, σ), it follows that (5.7.19) holds whenever f ∈ 𝒞0 (∂Ω) and
u is the unique solution of the classical Dirichlet Problem (5.7.18). In concert with
(i) ⇒ (ii) in Theorem 5.7.6, this implies (cf. (5.7.21)) that ω x0 ∈ Bp (σ). In light of
(5.7.45)-(5.7.46), this forces p > pΩ . 
Here is an integral representation formula augmenting the Fatou-type result from
Proposition 5.5.3.
Corollary 5.7.8 Consider a bounded NTA domain Ω ⊂ Rn whose boundary is an
upper Ahlfors regular set and let pΩ ∈ [1, ∞) be associated with Ω as in (5.7.46).
Denote by ν the geometric measure theoretic outward unit normal to Ω and let
GΩ (·, ·) stand for the Green associated with the Laplacian in Ω. Lastly, fix some
aperture parameter κ > 0 along with some integrability exponent p ∈ (pΩ, ∞). Then
p
any function u ∈ Hκ (Ω) may be expressed as
∫ ∫  κ−n.t.  ∫  κ−n.t. 
κ−n.t.
u(x) = u∂Ω dω x = k x u∂Ω dσ = Kx0 (x, y) u∂Ω (y) dσ(y)
∂Ω ∂Ω ∂Ω
∫  κ−n.t. 
=− ∂ν GΩ (·, x) u∂Ω dσ, ∀x ∈ Ω, (5.7.57)
∂Ω

where the normal derivative of the Green function is defined as in (5.6.2).

Proof This is a direct consequence of Proposition 5.5.3 (cf. (5.5.6)) and Theo-
rem 5.7.7. 

5.8 The Nature of the Critical Index pΩ and Further Results on

the Green Function

In order to further place Theorem 5.7.7 and Corollary 5.7.8 into perspective, we
wish to comment on the nature of the critical index pΩ ∈ [1, ∞) associated with a
bounded NTA domain Ω ⊆ N possessing an Ahlfors regular boundary as in (5.7.46).
First, we note that
5.8 The Nature of the Critical Index pΩ and Further Results on the Green Function 871

if Ω ⊆ Rn is a bounded Lipschitz domain then 1 ≤ pΩ < 2. (5.8.1)

This result is due to B. Dahlberg (cf. [25]) who also proved that the range 1 ≤ pΩ < 2
is sharp in the category of bounded Lipschitz domains Ω ⊆ Rn . A new approach to
(5.8.1), based on a Rellich identity involving the Poisson kernel, has been proposed
by D. Jerison and C. Kenig in [69]. For the class a bounded Lipschitz domains
with connected boundaries, alternative proofs of (5.8.1), based on boundary layer
potentials, may be found in [28] and [179]. Incidentally, the latter approach yields
the following representation of the Poisson kernel for a bounded Lipschitz domain
Ω ⊆ Rn with connected boundary:
  −1   
k x0 = 12 I + KΔ# ∂ν EΔ (· − x0 ) , x0 ∈ Ω, (5.8.2)

where KΔ# is the “transpose” of the harmonic double layer potential operator asso-
ciated with Ω as in (2.5.203), the vector ν is the outward unit normal to Ω, and
EΔ is the standard fundamental solution for the Laplacian in Rn defined in (1.5.56).
Formula (5.8.2) should be compared with (4.5.16).
Second, it has been proved by B. Dahlberg in [26] that

if Ω ⊆ Rn is a bounded 𝒞1 domain then pΩ = 1. (5.8.3)

Under the additional assumption that the bounded 𝒞1 domain in question has a
connected boundary, this has also been established by E.B. Fabes, M. Jodeit, and
N.M. Rivière in [41] (at least implicit, via (5.7.50)). Subsequently, S. Hofmann,
M. Mitrea, and M. Taylor have succeeded in extending (5.8.3) to the larger class of
regular SKT domains (cf. [112, (5.11.93)]) with compact boundaries. Specifically,
the well-posedness result established in [63, Theorem 7.2, p. 2831], in concert with
(5.7.50), implies that

if Ω ⊆ Rn is a bounded regular SKT domain then pΩ = 1. (5.8.4)

In fact, the theorems just mentioned yield a more nuanced result, in the category
of ε-regular SKT domains (introduced in [63]). The reader is reminded that the
membership of a set Ω ⊆ Rn to this class is characterized by the following geometric
measure theoretic conditions:
Ω is an open set with an Ahlfors regular boundary, satisfying a two-
sided local John condition, and whose
 geometric measure  theoretic (5.8.5)
outward unit normal ν satisfies dist ν, [VMO(∂Ω, σ)]n < ε,

where VMO(∂Ω, σ) stands for the Sarason class of functions of vanishing mean
oscillation on ∂Ω,  n surface measure σ := H ∂Ω, and the distance
 relative to the
n−1

is measured in BMO(∂Ω, σ) . It turns out that in the class of bounded ε-regular

SKT domains Ω ⊆ Rn the critical exponent pΩ can made as closed to the value 1
as wanted by taking ε > 0 small enough (relative to the desired degree of proximity
of pΩ to 1, as well as the NTA and the Ahlfors regularity character of ∂Ω). For
872 5 Green Functions and Poisson Kernels for the Laplacian

example, this allows the consideration of bounded Lipschitz domains with small
Lipschitz constant, as a particular case.
Continuing the discussion aimed at further elucidating the nature of the critical
index pΩ ∈ [1, ∞) associated as in (5.7.46) with a bounded NTA domain Ω ⊆ N
having an Ahlfors regular boundary, we note the following result.
Proposition 5.8.1 Let Ω ⊂ Rn be a bounded NTA domain, possessing an upper
Ahlfors regular boundary, and with the property that there exist α > 0 and C ∈ (0, ∞)
such that the Green function GΩ associated with the Laplacian in Ω satisfies the
estimate

|∇y GΩ (y, x)| ≤ C dist(x, ∂Ω)α |x − y| 1−n−α, ∀x, y ∈ Ω. (5.8.6)

Then pΩ = 1.
Proof Abbreviate σ := H n−1 ∂Ω and denote by ν the geometric measure theoretic
outward unit normal to Ω. From (5.8.6) and (5.6.1) we conclude that, for each fixed
x ∈ Ω,
   n.t.  

 ∇GΩ (·, x)  (y) ≤ C dist(x, ∂Ω)α |x − y| 1−n−α for σ-a.e. y ∈ ∂Ω. (5.8.7)
∂Ω

This further entails that, for each fixed x ∈ Ω,

 
∂ν(y) GΩ (y, x) ≤ C dist(x, ∂Ω)α |x − y| 1−n−α for σ-a.e. y ∈ ∂Ω. (5.8.8)

Suppose now that some f ∈ 𝒞0 (∂Ω) has been given and consider
∫
(PI f )(x) = f dω x, ∀x ∈ Ω. (5.8.9)
∂Ω

Theorem 5.6.2 then allows us to express

∫
(PI f )(x) = − ∂ν GΩ (·, x) f dσ, ∀x ∈ Ω. (5.8.10)
∂Ω

Combining (5.8.8) with (5.8.10) therefore yields

∫
dist(x, ∂Ω)α
|(PI f )(x)| ≤ C | f (y)| dσ(y), ∀x ∈ Ω. (5.8.11)
∂Ω |x − y|
n−1+α

In concert with [112, Proposition 8.4.12] this proves that for each p ∈ (1, ∞) we
have
Nκ (PI f ) L p (∂Ω,σ) ≤ C(Ω, κ, p) f  L p (∂Ω,σ) . (5.8.12)
On account of items (i)-(ii) in Theorem 5.7.6, from the fact that we have an estimate
as in (5.8.12) for each p ∈ (1, ∞) we ultimately conclude that pΩ = 1. 
To give a concrete example when the Green function estimate hypothesized in
Proposition 5.8.1 materializes, we first recall a couple of definitions. First, a set
5.8 The Nature of the Critical Index pΩ and Further Results on the Green Function 873

Ω ⊆ Rn is said to satisfy a uniform exterior ball condition (UEBC for short),

if there exists a number r > 0 such that at each boundary point x it is possible to find
an open ball Bx,r of radius r which is contained in Rn \ Ω and such that x ∈ ∂Bx,r .
Second, the concept of uniform interior ball condition (UIBC for short) is
defined analogously, but now insisting that the ball Bx,r is actually contained in Ω.
The uniform interior/exterior ball condition has been originally introduced by
Poincaré in [147] and, through the many intervening years, it has proved to be a
natural, useful condition in the study of the boundary behavior of solutions of large
classes of elliptic differential equations. While it is known (see, e.g., [5] where more
general results of this nature may be found) that for an open set Ω in Rn

Ω satisfies both a UIBC and a UEBC ⇐⇒ Ω is a 𝒞1,1 domain, (5.8.13)

a mere one-sided ball condition permits the boundary of the set in question to be
quite rough. For instance, any bounded, open, convex set in Rn is a bounded NTA
domain, with an Ahlfors regular boundary, and which satisfies a UEBC (for every
given radius r > 0). Of course, bounded, open, convex subsets of Rn can be quite
rough, though they are, nonetheless, Lipschitz domains. In this vein we wish to note
that, in the presence of certain background geometric characteristics, additionally
imposing UEBC on a given set may enhance its regularity in surprising ways. To
illustrate this phenomenon, we first recall a couple of definitions. First, given an open
set O in Rn , call a function u : O → R semiconvex provided it is continuous and
has the property that there exists C ∈ (0, ∞) such that

2u(x) − u(x + h) − u(x − h) ≤ C|h| 2

(5.8.14)
for all x, h ∈ Rn with [x − h, x + h] ⊆ O.

Second, call a nonempty, closed set E ⊆ Rn proximally regular provided there

exists some r > 0 such that for each x ∈ Rn with 0 < δE (x) < r there exists a unique
point x∗ ∈ E such that δE (x) = |x − x∗ |. This condition turns out to be equivalent
to E being a set of positive reach in the sense of H. Federer (cf. [44, Definition 4.1,
p. 432]). The following theorem is proved in [110]:

Theorem 5.8.2 Let Ω ⊆ Rn be an open set with compact boundary. Then the fol-
lowing conditions are equivalent:
(i) Ω is a Lipschitz domain satisfying a uniform exterior ball condition (UEBC);
(ii) Ω is a semiconvex domain (i.e., Ω is an open set satisfying ∂Ω = ∂(Ω) and
whose boundary may be locally described, up to a rigid transformation, by
graphs of semiconvex functions);
(iii) Ω is an NTA domain satisfying a UEBC;
(iv) Ω satisfies an interior corkscrew condition as well as a UEBC;
(v) Ω is a Lipschitz domain with the property that Ω is proximally regular (the latter
condition also being equivalent to Ω being a set of positive reach);
(vi) Ω is proximally regular (or of positive reach), ∂Ω = ∂(Ω) and the following
boundary density condition holds
 874 5 Green Functions and Poisson Kernels for the Laplacian

L n (Ω ∩ B(x, r))
lim sup > 0, ∀x ∈ ∂Ω; (5.8.15)
r→0+ rn

(vii) Ω is n-tick, ∂Ω = ∂(Ω), and the set Ω is proximally regular (or of positive
reach);
(viii) Ω is an n-thick set satisfying a UEBC;
(ix) Ω satisfies a UEBC and (cf. (A.0.2))
!
∂Ω = Aκ (∂Ω); (5.8.16)
κ>0

(x) Ω is a set of locally finite perimeter with ∂Ω = ∂(Ω), satisfying (5.8.16) and
such that there exists C ∈ (0, ∞) with the property that if ν denotes the geometric
measure theoretic outward unit normal to Ω then
(x − y) · ν(y) + dist(x, ∂Ω) ≤ C|x − y| 2
(5.8.17)
for each x ∈ Ω and each y ∈ ∂ ∗ Ω;

(xi) Ω is a set of locally finite perimeter with ∂Ω = ∂(Ω), satisfying (5.8.16) and
such that there exists C ∈ (0, ∞) with the property that if ν denotes the geometric
measure theoretic outward unit normal to Ω then

(x − y) · ν(y) ≤ C|x − y| 2 for all x ∈ ∂Ω and y ∈ ∂ ∗ Ω. (5.8.18)

Parenthetically, we wish to remark that condition (5.8.18) amounts to a pointwise

lower bound for the integral kernel of the harmonic double layer K associated with
a bounded semiconvex domain Ω ⊆ Rn as in (2.5.203) of the following sort:

1 C 1 y − x, ν(y)
− ≤ for all x ∈ ∂Ω and y ∈ ∂ ∗ Ω. (5.8.19)
ωn−1 |x − y| n−2 ωn−1 |x − y| n

As regards the nature of the Green function, the relevance of Ω satisfying a

UEBC stems from from the fact that the latter guarantees that ∇x GΩ (x, y) exhibits a
behavior akin to that of the the Poisson kernel for the upper half-space. Specifically,
we have the following theorem proved by M. Grüter and K.-O. Widman in [55,
Theorem 3.3, p. 31], expanding on the properties of the Green function discussed in
Theorem 5.3.1.

Theorem 5.8.3 If Ω ⊂ Rn is a bounded open set which satisfies a UEBC then the
Green function associated with the Laplacian in Ω satisfies the following properties
for all distinct points x, y ∈ Ω:
(i) GΩ (x, y) ≤ C|x − y| 2−n ;
(ii) GΩ (x, y) ≤ C dist(x, ∂Ω)|x − y| 1−n ;
(iii) GΩ (x, y) ≤ C dist(y, ∂Ω)|x − y| 1−n ;
(iv) GΩ (x, y) ≤ C dist(x, ∂Ω) dist(y, ∂Ω)|x − y| −n ;
(v) |∇x GΩ (x, y)| ≤ C|x − y| 1−n ;
 5.8 The Nature of the Critical Index pΩ and Further Results on the Green Function 875

(vi) |∇x GΩ (x, y)| ≤ C dist(y, ∂Ω)|x − y| −n ;

(vii) |∇y GΩ (x, y)| ≤ C|x − y| 1−n ;
(viii) |∇y GΩ (x, y)| ≤ C dist(x, ∂Ω)|x − y| −n ;
(ix) |∇x ∇y GΩ (x, y)| ≤ C|x − y| −n ,
for some constant C ∈ (0, ∞) independent of x, y.

By combining Theorem 5.8.2, Theorem 5.8.3, Proposition 5.8.1 (used here with
α = 1), and Theorem 5.7.7, we obtain the following well-posedness result for the L p
Dirichlet Problem.

Corollary 5.8.4 Let Ω ⊂ Rn be a bounded NTA domain with an upper Ahlfors

regular boundary and which satisfies a UEBC (equivalently, let Ω ⊂ Rn be a
bounded semiconvex domain). Then for each fixed κ > 0 and p ∈ (1, ∞), the Dirichlet
Problem with the boundary trace considered in the nontangential sense and with
nontangential maximal function control, formulated in (5.7.47), is uniquely solvable.
Moreover, the solution u satisfies (5.7.48) and is representable as in (5.7.49).

Being defined as a Radon-Nikodym derivative of two positive measures the

Poisson kernel is a non-negative function, but under suitable additional geometric
assumptions this turns out to actually be bounded away from zero in a uniform
fashion. To facilitate the statement of such a result, following [5] we define a pseudo-
ball in Rn with apex at the point x ∈ Rn , symmetry axis along the vector vector
h ∈ S n−1 , truncated at height b ∈ (0, ∞), and whose profile is modeled by the “shape
function" δ : [0, R] → [0, ∞) to be the set
 
𝒫δ,b (x, h) := y ∈ B(x, R) : |y − x|δ(|y − x|) < h · (y − x) < b ⊆ Rn . (5.8.20)

Theorem 5.8.5 Let Ω ⊆ Rn be a bounded NTA domain with an upper Ahlfors reg-
ular boundary. satisfying a uniform interior pseudo-ball condition of the following
sort: there exist parameters b, R ∈ (0, ∞), a unit vector field h : ∂Ω → S n−1 , and
a continuous function δ : [0, R] → [0, ∞) which is positive and non-decreasing on
(0, R] and satisfies Dini’s integrability condition
∫ R
δ(t)
dt < +∞, (5.8.21)
0 t
with the property that
 
𝒫δ,b x, h(x) ⊆ Ω for each point x ∈ ∂Ω. (5.8.22)

Set σ := H n−1 ∂Ω and let ν stand for the geometric measure theoretic outward unit
normal to Ω. Also, denote by GΩ (·, ·) the Green function for the Laplacian in Ω, and
fix an arbitrary point x0 ∈ Ω.
Then
h(x) = −ν(x) for each x ∈ ∂ ∗ Ω, (5.8.23)
hence
876 5 Green Functions and Poisson Kernels for the Laplacian

h = −ν at σ-a.e. point on ∂Ω, (5.8.24)

and there exists a constant c > 0 such that, with the normal derivative defined as in
(5.6.2), one has

k x0 = −∂ν GΩ (·, x0 ) ≥ c at σ-a.e. point on ∂Ω. (5.8.25)

The above theorem addresses the issue raised in [49, Question 2, p. 49] where
Garnett and Marshall ask what regularity conditions on ∂Ω ensure the validity of
(5.8.25). In this regard, we also wish to note that any bounded domain of class 𝒞1,α
with α ∈ (0, 1] satisfies a uniform interior pseudo-ball condition with shape function
δ(t) := Ct α in which case Dini’s integrability condition (5.8.21) holds (in particular,
if Ω satisfies a genuine uniform interior ball condition then Ω satisfies a uniform
interior pseudo-ball condition with shape function δ(t) := Ct). From this point of
view, Theorem 5.8.5 is sharp in the sense that if Ω is merely a bounded 𝒞1 domain
then k x0 (hence also −∂ν GΩ (·, x0 )) may not be bounded away from zero (this follows
from [49, Exercise 14, p. 71]).
Proof of Theorem 5.8.5 To begin with, observe that the assumptions made on δ
ensure that this function vanishes continuously at 0. This implies that for each angle
θ ∈ (0, π) there exists a small height bθ ∈ (0, ∞) with the property that
   
Cθ,bθ x, h(x) ⊆ 𝒫δ,b x, h(x) for each x ∈ ∂Ω (5.8.26)

where, generally speaking,

 
Cθ,b (x, h) := y ∈ Rn : cos(θ/2) |y − x| < (y − x) · h < b (5.8.27)

is the cone with vertex at x ∈ Rn , symmetry axis along h ∈ S n−1 , and full aperture
θ ∈ (0, π). In view of (5.8.26), the uniform interior pseudo-ball condition (5.8.22),
and [112, Lemma 5.6.15], we may conclude that (5.8.23) holds. Hence, (5.8.24)
also holds (thanks to [112, (5.2.4)] and [112, (5.6.21)]). Keeping this and (5.8.26) in
mind, the implication recorded in [112, (8.1.6)] presently gives that for each κ > 0
there exists a small ε > 0 such that
 
x − tν(x) : 0 < t < ε ⊆ Γκ (x) for σ-a.e. x ∈ ∂Ω. (5.8.28)

Having established (5.8.24), the Boundary Point Principle proved in [5] then
provides some constant c > 0 with the property that
 
GΩ x − tν(x), x0 − GΩ (x, x0 )
lim inf ≥ c at σ-a.e. x ∈ ∂Ω. (5.8.29)
t→0+ t
Then (5.8.25) follows from (5.8.29) with the help of (5.8.28) and Lemma 5.6.1. 

The notion of ε-regular SKT domain (cf. (5.8.5)) plays a significant role in the
formulation of Theorem 5.8.6 below. In light of [113, (9.2.22)], said theorem refines
a result of E. Fabes who has shown in [40] that if Ω ⊆ Rn is a bounded Lipschitz
5.8 The Nature of the Critical Index pΩ and Further Results on the Green Function 877

 σ denotes its surface measure and κ > 0 is a fixed aperture parameter, then
domain,

Nκ u 2 ≈ u 1/2 , uniformly for u harmonic function in Ω.
L (∂Ω,σ) H (Ω)

Theorem 5.8.6 Let Ω be a bounded NTA domain in Rn , with n ≥ 2, whose boundary

is upper Ahlfors regular, and abbreviate σ := H n−1 ∂Ω. Also, fix an arbitrary
aperture parameter κ > 0 along with some integrability exponent p ∈ (1, 2].
p, p
Then any harmonic function u ∈ B1/p (Ω) satisfies Nκ u ∈ L p (∂Ω, σ), has the
κ−n.t.
property that the nontangential boundary trace u∂Ω exists at σ-a.e. point on ∂Ω,
and there exists a finite constant C > 0, independent of u, such that
 
Nκ u p ≤ CuB p, p (Ω) . (5.8.30)
L (∂Ω,σ) 1/p

Moreover, as regards the opposite inequality in (5.8.30),

given any p ∈ (1, ∞) and M ∈ (0, ∞) there exists ε > 0 such that
if Ω is a bounded ε-regular SKT domain whose local John constants
and Ahlfors regularity character are controlled
 by M then one can find (5.8.31)
some C ∈ (0, ∞) for which uB p, p (Ω) ≤ C Nκ u L p (∂Ω,σ) whenever
1/p
u is harmonic in Ω.
In particular,

given any p ∈ (1, 2] and M ∈ (0, ∞) there exists ε > 0 such that if
Ω is a bounded ε-regular SKT domain whose local John constants
(5.8.32)
 Ahlfors
and  regularity character are controlled by M then actually
Nκ u p ≈ uB p, p (Ω) uniformly for u harmonic in Ω,
L (∂Ω,σ) 1/p

hence, as a corollary,

 Ω is
if  a bounded regular SKT domain and p ∈ (1, 2] then
Nκ u p ≈ uB p, p (Ω) uniformly for u harmonic in Ω. (5.8.33)
L (∂Ω,σ) 1/p

Proof For starters, observe that [112, (5.11.4)] presently ensures that ∂Ω is an
Ahlfors regular set. To proceed, recall the L 2 -based area-function defined as in
(3.1.119) with q := 2. From the version of [27, Theorem, p. 98] given in [27, item
(A), p. 106] (as well as Proposition 5.4.1) we see that there exists some finite constant
C > 0 such that for any harmonic function u in Ω we have
   
Nκ u p ≤ C A2,κ u L p (∂Ω,σ) + Cu L p (Ω, L n ) . (5.8.34)
L (∂Ω,σ)

Granted this, (5.8.30) follows by invoking [113, Theorem 9.2.39] (with q := 2).
κ−n.t.
Also, the existence of the nontangential boundary trace u∂Ω at σ-a.e. point on ∂Ω
is a consequence of (5.8.34) and Proposition 5.5.3.
To deal with (5.8.31), fix p ∈ (1, ∞) and work under the assumption that Ω is
actually a bounded ε-regular SKT domain for some sufficiently small ε = ε(p) > 0.
In this setting, suppose u is a harmonic function in Ω satisfying Nκ u ∈ L p (∂Ω, σ).
878 5 Green Functions and Poisson Kernels for the Laplacian

From the Fatou-type result from Proposition 5.5.2 and the well-posedness result for
the L p Dirichlet Problem in Ω from [63] we then conclude that there exists some
f ∈ L p (∂Ω, σ) such that
 
 f  L p (∂Ω,σ) ≤ C Nκ u L p (∂Ω,σ) and u = D f in Ω, (5.8.35)

where the constant C ∈ (0, ∞) is independent of u, and where D is the boundary-

to-domain harmonic double layer potential operator associated with Ω. Combining
(5.8.35) with the mapping properties of the boundary-to-domain double layer acting
from Lebesgue spaces into Besov spaces
  from [115, §4.3] then establishes that
u ∈ B1/p (Ω) and uB p, p (Ω) ≤ C Nκ u L p (∂Ω,σ) . This completes the proof of
p, p
1/p
(5.8.31), from which (5.8.32) follows on account of (5.8.30). 
We conclude this section with an application to the mapping properties of the
Poisson integral operator on Besov-Triebel-Lizorkin spaces, which has significant
implications for the solvability of the Dirichlet Problem for the Laplacian in this
context (for related results the interested reader is referred to [37], [43], [54], [73],
[85], [100], [99], [97], [96], [121] and the references therein).
Theorem 5.8.7 Let Ω be a bounded Lipschitz domain in Rn satisfying a UEBC.
Denote its outward unit normal ν, and abbreviate σ := H n−1 ∂Ω. Also, consider
the Green function GΩ (·, ·) for the Laplacian in Ω and for each f ∈ L 1 (∂Ω, σ)
define
∫
(PI f )(x) := − ∂ν(y) GΩ (x, y) f (y) dσ(y) for each x ∈ Ω. (5.8.36)
∂Ω

Then the operators

p,q p,q p, p p,q
PI : Bs (∂Ω, σ) → B (Ω) and PI : Bs (∂Ω, σ) → F (Ω) (5.8.37)
s+ p1 s+ p1

are bounded whenever 0 < p, q ≤ ∞ and (n − 1)( p1 − 1)+ < s < 1, with the
additional condition that p, q  ∞ in the case of Triebel-Lizorkin spaces.
It is worth recalling from Theorem 5.8.2 within the class of bounded open sets
satisfying a UEBC, being an NTA domain, or being a Lipschitz domain, or being a
semiconvex domain are all equivalent conditions.
Proof We claim that there exist a constant C ∈ (0, ∞) and some σ-measurable set
E ⊆ ∂Ω with σ(E) = 0 such that
  C
∇x ∇y GΩ (x, y) ≤ ∀x ∈ Ω, ∀y ∈ ∂Ω \ E. (5.8.38)
|x − y| n
Indeed, from Theorem 5.8.3 we know that the estimate in (5.8.38) holds for every
x, y ∈ Ω. Then, keeping x ∈ Ω, by Fatou’s Theorem proved in [25] for bounded,
harmonic functions in Lipschitz domains, it follows that there exists Ex ⊆ ∂Ω with
σ(Ex ) = 0 and such that the estimate in (5.8.38) holds for every y ∈ Ω \ Ex . Pick
5.8 The Nature of the Critical Index pΩ and Further Results on the Green Function 879
"
a countable dense subset D of Ω, and set E := Ex . Then σ(E) = 0 and the
x ∈D
estimate in (5.8.38) holds for every x ∈ D and y ∈ Ω \ E. Keeping now y ∈ Ω \ E
fixed, by density then the estimate in (5.8.38) holds for every x ∈ Ω.
Next, recall that we have

PI(1) = 1 in Ω and Δ ◦ PI = 0 in Ω. (5.8.39)

As a consequence, for each fixed θ ∈ (0, 1) we may rely on the mapping properties
for certain boundary-to-domain integral operators acting from Besov spaces with
a positive amount of smoothness into Lebesgue spaces suitably weighted in terms
of the distance to the boundary established in [115, §4.2] to the operator ∇PI and
n < p ≤ ∞ and (n − 1)( p − 1)+ < s < 1 then there exists
conclude that if n−1 1
p, p
C ∈ (0, ∞) such that for each f ∈ Bs (∂Ω, σ) we have
 1− 1 −s   
 
δ∂Ω ∇PI f 
p
,θ  L p (Ω, L n )
≤ C f Bsp, p (∂Ω,σ) . (5.8.40)

p, p
We claim that there exists C ∈ (0, ∞) such that for each f ∈ Bs (∂Ω, σ) we
also have  
PI f  p ≤ C f Bsp, p (∂Ω,σ) . (5.8.41)
L (Ω, L n )

To justify this, assume first that p ∈ (1, ∞) and s ∈ (0, 1). Since
 as is apparent
from (5.8.38), the integral kernel k(x, y) := ∇x ∂ν(y) GΩ (x, y) of the operator
∇PI satisfies [112, (8.7.52)] for each ε∗ ∈ (0, 1), we may invoke [112, (8.7.54)
in Proposition 8.7.10] (with β := 0 and ε∗ > 1 − p1 ) to conclude (also keeping
in mind [113, (7.9.10)]) that (5.8.41) holds. The case when n−1 n < p ≤ 1 and
(n − 1)( p1 − 1) < s < 1 may be reduced to this by using a suitable embedding (cf.
[113, §7.7]), and this completes the proof of (5.8.41).
From (5.8.40)-(5.8.41), the second formula in (5.8.39), and [113, (9.2.159)]
p, p
(presently used with L := Δ) we may then conclude that PI maps Bs (∂Ω, σ)
p,q
boundedly into F 1 (Ω) whenever 0 < p < ∞, (n − 1)( p − 1)+ < s < 1, and
1
s+ p
n
n+s+1/p < q ≤ ∞. The extension to q ∈ (0, ∞] then follows by invoking [76,
Theorem 1.6]. Finally, the boundedness of the first operator in (5.8.37) now follows
from this and real interpolation (cf. [122, Proposition 2.57, pp. 102-103] and [76,
Theorem 9.4]). 
Chapter 6
Scattering by Rough Obstacles

The Gauss-Green formula (aka Divergence Theorem) has proved to be a very useful
tool in scattering theory. For scattering by rough obstacles one needs such a formula
for rough integrands and rough boundaries, of the sort we have developed in Volume I
([112]). The behavior at infinity is of particular significance, and various decay
conditions (aka radiation conditions) have been used in the literature.
In this chapter the goal is to provide a unified approach to scattering theory in
rough settings which is also aimed at identifying the broadest possible spectrum of
radiation conditions for null-solutions of the vector Helmholtz operator Δ + k 2 . The
latter contains, as particular cases, the Sommerfeld, Silver-Müller, and McIntosh-
Mitrea radiation conditions corresponding to scattering by acoustic waves, electro-
magnetic waves, and null-solutions of perturbed Dirac operators, respectively. In this
regard, we succeeded in producing a family of radiation conditions, (RC A), indexed
by all possible writings of the vector Laplace operator as Δ = divA∇ involving
complex coefficient tensors A. In turn, this opens the door for proving well-posedness
results and for boundary value problems as well as integral representation formulas
for the vector Helmholtz operator in some very general types of exterior domains.

6.1 Integral Representations for Null-Solutions of the Helmholtz

Operator

To discuss certain scattering phenomena, throughout we assume that n ∈ N, n ≥ 2,

and fix a wave number k ∈ (0, ∞). We also agree to abbreviate


x := x/|x| for each x ∈ Rn \ {0}. (6.1.1)

Definition 6.1.1 Call a complex-valued smooth function u defined in an exterior

domain in Rn radiating at infinity provided u is a null-solution of the Helmholtz
operator Δ + k 2 and satisfies Sommerfeld’s radiation condition

© The Author(s), under exclusive license to Springer Nature Switzerland AG 2023 881
D. Mitrea et al., Geometric Harmonic Analysis III, Developments in Mathematics 74,
https://doi.org/10.1007/978-3-031-22735-6_6
882 6 Scattering by Rough Obstacles
∫
 2
iku(x) − 
x · (∇u)(x) dH n−1 (x) = o(1) as R → ∞. (6.1.2)
|x |=R

Furthermore, call a vector-valued function radiating at infinity provided each of its

scalar components is a null-solution of the Helmholtz operator Δ + k 2 and satisfies
Sommerfeld’s radiation condition (6.1.2).

Originally (6.1.2) was augmented by the Sommerfeld’s finiteness condition

∫
|u(x)| 2 dH n−1 (x) = O(1) as R → ∞, (6.1.3)
|x |=R

However, it is now known that (6.1.2) and (6.1.3) are not independent in the sense
that, for null-solutions of the Helmholtz operator, the former implies the latter. This
is clear from the lemma below, which is a particular case of a more general result
proved later in Proposition 6.5.1.

Lemma 6.1.2 Let Ω be an exterior domain in Rn , where n ∈ N with n ≥ 2, and fix

k ∈ (0, ∞). Suppose u ∈ 𝒞∞ (Ω) satisfies (Δ + k 2 )u = 0 in Ω and
∫
 2
iku(x) − 
x · (∇u)(x) dH n−1 (x) = O(1) as R → ∞. (6.1.4)
|x |=R

Then ∫
|u(x)| 2 dH n−1 (x) = O(1) as R → ∞ (6.1.5)
|x |=R

and ∫
 2
x · (∇u)(x) dH n−1 (x) = O(1) as R → ∞. (6.1.6)
|x |=R

For example (see the discussion in [91]), the only fundamental solution of the
Helmholtz operator Δ + k 2 in Rn which is radiating at infinity is given at each
x ∈ Rn \ {0} by
(1)
H(n−2)/2 (k |x|) 1
Φk (x) := cn k (n−2)/2 , with cn := , (6.1.7)
|x| (n−2)/2 4i(2π)(n−2)/2

where, generally speaking, Hλ(1) (·) denotes the Hankel function of the first kind with
index λ ∈ R (cf. [1, § 9.1]).
To state our first major result in this section, recall the truncated nontangential
maximal operator from (A.0.92).

Theorem 6.1.3 Let n ∈ N, n ≥ 2, and fix a wave number k ∈ (0, ∞). Assume

Ω is an exterior domain in Rn with a lower Ahlfors regular boundary, such that
σ := H n−1 ∂Ω is a doubling measure on ∂Ω. This implies that Ω is a set of locally
finite perimeter, and its geometric measure theoretic outward unit normal ν is defined
σ-a.e. on ∂∗ Ω.
6.1 Integral Representations for Null-Solutions of the Helmholtz Operator 883

Suppose u, w ∈ 𝒞∞ (Ω) are two null-solutions of the Helmholtz operator Δ + k 2

in Ω which are radiating at infinity and such that, for some parameters κ, κ > 0,
and ε > 0, satisfy

Nκε u · Nκε (∇w) ∈ L 1 (∂Ω, σ), Nκε (∇u) · Nκε w ∈ L 1 (∂Ω, σ),
κ−n.t. κ−n.t. κ −n.t. κ −n.t.
and the traces u∂Ω , (∇u)∂Ω , w ∂Ω , (∇w)∂Ω (6.1.8)
exist at σ-almost every point on ∂nta Ω.
κ −n.t. κ −n.t. κ −n.t.
Then for any other κ > 0 the nontangential traces u∂Ω , (∇u)∂Ω , w ∂Ω ,
κ −n.t.
(∇w)∂Ω also exist at σ-a.e. point on ∂nta Ω, and are actually independent of κ .
Moreover, with the dependence on the parameter κ dropped, the following Green
type formula holds:
∫ ∫
 n.t.   n.t.   n.t.   n.t. 
u∂Ω ν · (∇w)∂Ω dσ = w ∂Ω ν · (∇u)∂Ω dσ. (6.1.9)
∂∗ Ω ∂∗ Ω

Parenthetically we wish to note that the conditions imposed in the first line of
(6.1.8) are automatically satisfied if

Nκε u ∈ L p (∂Ω, σ), Nκε (∇u) ∈ L q (∂Ω, σ),

(6.1.10)
Nκε w ∈ L q (∂Ω, σ), Nκε (∇w) ∈ L p (∂Ω, σ),

for some p, p , q, q ∈ [1, ∞] with 1/p + 1/p = 1/q + 1/q = 1.

Proof of Theorem 6.1.3 The first conclusion in the statement of the theorem is a
direct consequence of assumptions and the comment following the statement of [112,
Proposition 8.9.8], so we focus on establishing the Green type formula formulated
in (6.1.9). To this end, consider the vector field
 n
F := u∇w − w∇u ∈ 𝒞∞ (Ω) (6.1.11)

and note that F is divergence-free since

divF = ∇u · ∇w + uΔw − ∇w · ∇u − wΔu

= u(Δ + k 2 )w − w(Δ + k 2 )u = 0 in Ω. (6.1.12)

In addition, the properties listed in (6.1.8) imply (with the dependence on the aperture
parameter dropped)
n.t.
N ε F ∈ L 1 (∂Ω, σ) and F ∂Ω exists σ-a.e. on ∂nta Ω. (6.1.13)

In fact, at σ-a.e. point on ∂∗ Ω we have

 n.t.   n.t.   n.t.   n.t.   n.t. 
ν · F ∂Ω = u∂Ω ν · (∇w)∂Ω − w ∂Ω ν · (∇u)∂Ω . (6.1.14)
884 6 Scattering by Rough Obstacles

On account of (6.1.11)-(6.1.14), the Green type identity (6.1.9) is implied by the

Divergence Formula [112, (1.4.5)] as soon as we prove that the contribution on F at
infinity, i.e., the complex number (cf. [112, (4.7.5)])
∫
[F]∞ = lim 
x · F(x) dH n−1 (x), (6.1.15)
R→∞ |x |=R

presently vanishes. To see that this is the case, we write

∫ 
 
 
x · F(x) dH n−1 (x)
|x |=R
∫  
 
= x · (∇w)(x) − w(x)
u(x) x · (∇u)(x) dH n−1 (x)
|x |=R
∫  

= u(x) x · (∇w)(x) − ikw(x)
|x |=R
  

− w(x) x · (∇u)(x) − iku(x) dH n−1 (x)
∫ 1/2
≤ |u(x)| 2 dH n−1 (x) ×
|x |=R
∫
 2 1/2
× x · (∇w)(x) − ikw(x) dH n−1 (x)
|x |=R
∫ 1/2
+ |w(x)| 2 dH n−1 (x) ×
|x |=R
∫
 2 1/2
× x · (∇u)(x) − iku(x) dH n−1 (x)
|x |=R

= O(1) · o(1) + O(1) · o(1) = o(1) as R → ∞, (6.1.16)

where the last equality is based on the Sommerfeld’s finiteness and radiation condi-
tions for u and w (cf. (6.1.2)-(6.1.3)). From (6.1.15) and (6.1.16) we then conclude
that [F]∞ = 0, which finishes the proof of (6.1.9). 

In the next theorem we present a boundary layer integral representation formula

for null-solutions of the Helmholtz operator which are radiating at infinity.

Theorem 6.1.4 Fix n ∈ N, n ≥ 2, along with a wave number k ∈ (0, ∞). Suppose
Ω is an exterior domain in Rn with a lower Ahlfors regular boundary and such that
σ := H n−1 ∂Ω is a doubling measure on ∂Ω. In particular, Ω is a set of locally
finite perimeter, and its geometric measure theoretic outward unit normal ν is defined
σ-a.e. on ∂∗ Ω.
Assume u ∈ 𝒞∞ (Ω) is a null-solution of the Helmholtz operator Δ+ k 2 in Ω which
is radiating at infinity and such that, for some parameters κ > 0, ε > 0, satisfies
6.1 Integral Representations for Null-Solutions of the Helmholtz Operator 885

Nκε u ∈ L 1 (∂Ω, σ), Nκε (∇u) ∈ L 1 (∂Ω, σ),

κ−n.t. κ−n.t. (6.1.17)
and u∂Ω , (∇u)∂Ω exist σ-a.e. on ∂nta Ω.
κ −n.t. κ −n.t.
Then for any other κ > 0 the nontangential traces u∂Ω , (∇u)∂Ω also exist
at σ-a.e. point on ∂nta Ω, and are actually independent of κ . Furthermore, with
the dependence on the parameter κ dropped, the function u admits the following
boundary layer integral representation formula:
∫
   n.t. 
u(x) = ν(y) · ∇y Φk (x − y) u∂Ω (y) dσ(y)
∂∗ Ω
∫
 n.t. 
− Φk (x − y)ν(y) · (∇u)∂Ω (y) dσ(y) for each x ∈ Ω. (6.1.18)
∂∗ Ω

As a preamble to the proof of this theorem, we record the following lemma (cf.
[91]), pertaining to the nature of the function Φk .
Lemma 6.1.5 Fix k ∈ (0, ∞) and n ∈ N, n ≥ 2. Then for each fixed R ∈ (0, ∞) there
exists some finite constant C = C(R, n, k) > 0 with the property that the function Φk
from (6.1.7) satisfies
⎧   
⎪
⎨C 1 +  ln |x|  , n = 2,
⎪
|Φk (x)| ≤ and |(∇Φk )(x)| ≤ C|x| 1−n,
⎪
⎪C|x| 2−n, n ≥ 3,
⎩
(6.1.19)
whenever x ∈ Rn is such that 0 < |x| < R.
We now turn to the proof of Theorem 6.1.4.
Proof of Theorem 6.1.4 The first conclusion in the statement of the theorem is given
by [112, Proposition 8.9.8] (see the comment following its statement), so we focus
on proving the integral representation formula (6.1.18). With this goal in mind,
fix an arbitrary point x ∈ Ω. The idea is to adapt the proof of Theorem 6.1.3 to
the case when w(y) := Φk (x − y) for each y ∈ Ω \ {x}. While this has a locally
integrable singularity at x, the function w may be nonetheless canonically viewed as
a distribution in D (Ω) with compact singular support, which satisfies

(Δ + k 2 )w = δx in D (Ω), (6.1.20)

where δx is the Dirac distribution in Ω with mass at x, and which radiates at infinity.
If analogously to (6.1.11) we now define the vector field
 
F(y) := u(y)∇y Φk (x − y) − Φk (x − y)(∇u)(y), ∀y ∈ Ω \ {x}, (6.1.21)
 1 n
then Lemma 6.1.5 implies that F ∈ Lloc (Ω, L n ) and, in place of (6.1.12), we
now have

divF = u(x)δx in D (Ω). (6.1.22)

886 6 Scattering by Rough Obstacles

The current assumptions also ensure that F has the properties listed in (6.1.13). In
fact, at σ-a.e. point y ∈ ∂∗ Ω we presently have
 n.t.     n.t. 
ν(y) · F ∂Ω (y) = ν(y) · ∇y Φk (x − y) u∂Ω (y)
 n.t. 
− Φk (x − y)ν(y) · (∇u)∂Ω (y). (6.1.23)

Finally, the contribution of F at infinity may be estimated as in the past (cf. (6.1.16))
bearing in mind that, as noted earlier, Φk (x − ·) radiates at infinity. Hence, as before,
we see that [F]∞ = 0. Granted this, the Divergence Formula [112, (1.4.5)] then
yields (6.1.18) on account of (6.1.22) and (6.1.23). 

6.2 Radiation Conditions: Motivation

Throughout, assume n ∈ N satisfies n ≥ 2, and fix a wave number k ∈ (0, ∞).

Suppose Ω ⊆ Rn is an exterior domain (i.e., a set whose complement is compact)
which is connected, and consider the Dirichlet Problem for the Helmholtz operator:

⎧
⎪ u ∈ 𝒞∞ (Ω),
⎪
⎨
⎪
(Δ + k 2 )u = 0 in Ω, (6.2.1)
⎪
⎪ u = f on ∂Ω.
⎪
⎩ ∂Ω

Typically this is accompanied by a suitable decay condition for u at infinity, designed

to ensure uniqueness.
With Jλ denoting the regular Bessel function of the first kind of order λ ∈ R, let
r > 0 be such that J(n−2)/2 (r) = 0. If we then define1

J(n−2)/2 (k |x|)
Ω := Rn \ B(0, r/k) and u(x) := , ∀x ∈ Ω, (6.2.2)
|x| (n−2)/2

then u ∈ 𝒞∞ (Ω), (Δ + k 2 )u = 0 in Ω, u∂Ω = 0, and
 
(∂ α u)(x) = O |x| −(n−1)/2 as |x| → ∞,
(6.2.3)
for each multi-index α ∈ N0n .

This goes to show that imposing the decay condition

 
u(x) = O |x| −(n−1)/2 as |x| → ∞ (6.2.4)

in the boundary value problem (6.2.1) does not lead to uniqueness. Since (6.2.4) is
too weak, one may be tempted to ask
sin(k | x |)
1 for n = 3 this takes the form u(x) = |x| for x ∈ Ω := R3 \ B(0, π/k)
6.2 Radiation Conditions: Motivation 887
 
u(x) = o |x| −(n−1)/2 as |x| → ∞. (6.2.5)

However, in such a scenario Rellich’s uniqueness lemma forces u ≡ 0 in Ω. Hence,

(6.2.5) amounts to too much decay in order to have a nontrivial solution.
In order to ensure uniqueness of the solutions of the Dirichlet Problem for the
Helmholtz operator in exterior domains (cf. (6.2.1)), in 1912 A. Sommerfeld has
introduced in [164] the condition

n
 
iku(x) − x j (∂j u)(x) = o |x| −(n−1)/2 as |x| → ∞,
 (6.2.6)
j=1

where  x := |xx | ∈ S n−1 is the normalization of x ∈ Rn \ {0} to a unit vector (cf.

(6.1.1)). This is referred to as Sommerfeld’s radiation condition.
Let us briefly review some other known radiation conditions in the literature.
Radiation conditions for the Maxwell’s system have been identified by C. Müler (in
1945; cf. [140]) and S. Silver (in 1949; cf. [162]). Specifically, if E = (E1, E2, E3 )
and H = (H1, H2, H3 ) are two vector fields satisfying the time-harmonic Maxwell
system in a three-dimensional exterior domain Ω

curl E − ik H = 0 and curl H + ik E = 0 in Ω, (6.2.7)

the classical Silver-Müller radiation conditions read

x × H(x) + E(x) = o(|x| −1 ) as |x| → ∞,

 (6.2.8)

x × E(x) − H(x) = o(|x| −1 ) as |x| → ∞.

 (6.2.9)

Another known radiation condition in the literature is that for the perturbed Dirac
equation, identified by A. McIntosh and M. Mitrea (in 1999; cf. [102]). To elaborate
on this topic, let (C n+1, +, ) be the Clifford algebra generated by n + 1 anti-
commuting imaginary units (ei )1≤i ≤n+1 (see the discussion in [112, §6.4]). Recall
that
n
D := e j  ∂j (6.2.10)
j=1

is the homogeneous the Dirac operator in Rn and let

Dk := D + ken+1 (6.2.11)

 perturbed Dirac operator in R (hence, Dk u = Du + ken+1  u for each

be the n

u = I u I eI ). Then the McIntosh-Mitrea radiation condition for Clifford algebra-

valued null-solutions u of the perturbed Dirac operator Dk in an exterior domain
Ω ⊆ Rn reads
 
x  en+1 )  u(x) = o |x| −(n−1)/2 as |x| → ∞.
(i −  (6.2.12)

Observe that
888 6 Scattering by Rough Obstacles

any null-solution of the Maxwell system or the perturbed Dirac operator

(6.2.13)
is also a null-solution of the (vector) Helmholtz operator Δ + k 2 .
This prompts the question:
given that both the Maxwell system and the perturbed Dirac equation
are “sub-problems” of the (vector) Helmholtz equation, why do the ra- (6.2.14)
diation conditions identified for each of these PDE’s look so different?
Our answer is that, as we shall see, Sommerfeld’s radiation condition is just one of
a more flexible and inclusive family of (equivalent) radiation conditions for null-
solutions of the Helmholtz equation. Our goal is to find them all. To this end, we
shall need to introduce some proper algebraic formalism.
For some fixed M ∈ N, consider the vector Laplacian Δ in Rn , viewed as a
second-order
  M × M system in Rn , acting componentwise on vector-valued functions
u = u I 1≤I ≤M according to
 
Δu = Δu I ≤I ≤M . (6.2.15)

In other words, we identify

 
Δ ≡ Δ I M×M = δJ I Δ 1≤J,I ≤M . (6.2.16)
 
Suppose a coefficient tensor A := arJsI 1≤J,I ≤M with complex constant entries has
1≤r,s ≤n
been chosen so that we may express said vector Laplacian as

n
Δ= arJsI ∂r ∂s . (6.2.17)
1≤J,I ≤M
r,s=1

Henceforth, we agree to succinctly write

Δ = divA∇ (6.2.18)

in place of (6.2.17). For further use, observe that

arI sJ ξs ξr = δI J |ξ | 2 for each ξ = (ξ1, . . . , ξn ) ∈ Rn

(6.2.19)
and each I, J ∈ {1, . . . , M }.

Indeed, since for each I, J ∈ {1, . . . , M } we know from (6.2.16)-(6.2.17) that

arI sJ ∂r ∂s and δI J ∂j ∂j are identical partial differential operators, they have identical
principal symbols, so (6.2.19) follows. In particular, (6.2.19) implies

arI sJ 
xs 
xr = δI J for each x ∈ Rn \ {0} and each I, J ∈ {1, . . . , M }. (6.2.20)

Going forward, the conormal derivative ∂ξA associated with each coefficient tensor
 
A := arJsI 1≤J,I ≤M in the writing of the vector Laplacian as Δ = divA∇ and each
1≤r,s ≤n
6.3 The Family of Radiation Conditions (RC A ) 889

vector ξ = (ξr )r ∈ Cn is defined as


M 
n
∂ξAu := arJsI ξr ∂s u I (6.2.21)
1≤J ≤M
I=1 r,s=1
 
for each given vector-valued function u = u I 1≤I ≤M . Finally, we note that
 
for A0 := δJ I δr s 1≤J,I ≤M we may express Δ = divA0 ∇, and we have
  1≤r,s ≤n (6.2.22)
∂ξA0 u = ξr ∂r u I 1≤I ≤M for each function u = (u I )1≤I ≤M and each
vector ξ = (ξr )r ∈ Cn .
Next,
 fix  an arbitrary integer M ∈ N and consider a complex coefficient tensor
A := arJsI 1≤J,I ≤M in the writing of the vector Laplacian as Δ = divA∇. For each
1≤r,s ≤n
(doubly indexed) vector ζ = (ζsI ) 1≤s ≤n with complex components define
1≤I ≤M


M 
n
Aζ := arJsI ζsI 1≤r ≤n
. (6.2.23)
I=1 r,s=1 1≤J ≤M

Consequently, if ·, · denotes the real-pairing of vectors with complex components,

we have
M  n
Aζ, ζ = arJsI ζsI ζr J . (6.2.24)
I,J=1 r,s=1

This is going to be relevant later on (e.g., when discussing L 2 -finiteness results).

6.3 The Family of Radiation Conditions (RC A )

 
Fix M ∈ N arbitrary. For each complex coefficient tensor A = arJsI 1≤J,I ≤M in
1≤r,s ≤n
the writing of the vector Laplacian as Δ = divA∇ we shall consider the radiation
condition
 
n
 
arJsI 
xr ∂s u I (x) − ikuJ (x) = o |x| −(n−1)/2 as |x| → ∞.
1≤J ≤M
1≤I ≤M r,s=1
(6.3.1)
Equivalently, this may be expressed in terms of the conormal derivative (6.2.21) as
   
iku(x) − ∂xAu (x) = o |x| −(n−1)/2 as |x| → ∞. (6.3.2)

We shall say that u satisfies the radiation condition (RC A) provided (6.3.2) holds.
890 6 Scattering by Rough Obstacles

As is apparent from (6.2.21), the radiation condition (6.3.2) involves the conormal
derivative associated with the factorization of the vector Laplacian Δ = divA∇,
evaluated at unit vector S n−1 in the direction of x.
In summary: The vector Laplacian has many writings as Δ = divA∇ for various
choices of a complex coefficient tensor A and, associated with each of those writings,
we have proposed the radiation condition (RC A). We therefore have a family of
radiation conditions indexed by these complex coefficient tensors A.
In relation to this, we seek to address the following two questions:
Question 1: How are the radiation conditions (RC A), corresponding to various
choices of of complex coefficient tensor A (used to represent vector Laplacian as
Δ = divA∇) related to one another? In particular, how do they relate to the classical
Sommerfeld’s (pointwise) radiation condition?
Question 2: Under what kind of assumptions on its behavior at infinity and near
the boundary does a given null-solution of the (vector) Helmholtz operator Δ + k 2
in an exterior domain admit a Green type integral representation formula?
We shall provide satisfactory answers to both questions in Theorem 6.6.1, for-
mulated a little later. For now, recall that Φk , introduced in (6.1.7), is the unique
distribution in Rn which is a fundamental solution of Δ + k 2 in Rn and satisfies
Sommerfeld’s radiation condition. Corresponding to the limiting case k = 0 (when
the Helmholtz operator Δ + k 2 is simply the Laplacian Δ), let us define
⎧
⎪ 1
⎪
⎨ (2 − n)ωn−1 |x| , x ∈ R \{0}, if n ≥ 3,
2−n n
⎪
⎪
Φ0 (x) := Φ0 (x; n) := (6.3.3)
⎪
⎪ 1
⎪
⎪ ln |x|, x ∈ R2 \{0}, if n = 2.
⎩ 2π
We shall temporarily adopt a more general point of view and work with the Helmholtz
operator Δ + z involving a generic wave number z ∈ C \ (−∞, 0]. To be specific, fix
n ∈ N satisfying n ≥ 2. For each z ∈ C \ (−∞, 0], let Φz (x; n) be the fundamental
solution of the Helmholtz operator Δ + z in Rn given by (compare with (6.1.7))

i z 1/2 (n−2)/2
(1)  1/2 
Φz (x; n) := − H(n−2)/2 z |x| for all x ∈ Rn \ {0}, (6.3.4)
4 2π|x|

where Hλ(1) (·) denotes the Hankel function of the first kind with index λ ∈ R (cf. [1,
 point x ∈ R \ {0}, the function Φz (x; n) happens to have a
§ 9.1]). For n
 each fixed
limit as C \ (−∞, 0]  z → 0 in higher dimensions, namely

1
lim Φz (x; n) = Φ0 (x; n) = |x| 2−n if n ≥ 3, (6.3.5)
z→0 (2 − n)ωn−1

but this connection between Φz (x; n) and Φ0 (x; n) fails to materialize for n = 2,
since
6.3 The Family of Radiation Conditions (RC A ) 891
     
Φz (x; 2) = 1
2π ln z 1/2 |x|/2 1 + O z|x| 2 − 2π ψ(1)
1
+ O z|x| ,
2
(6.3.6)
as z → 0, uniformly for x in compact subsets of R2 \ {0}.

Here ψ(w) := Γ (w)/Γ(w) denotes the digamma  function (cf. [1, § 6.3, (6.3.1),
p. 258]), where Γ(w) with w ∈ C \ − n : n ∈ N is the classical Gamma function
(cf., e.g., [109, § 14.5, p. 566]).
In the lemma below (which expands on [50, Lemma C.1]) we elucidate the
asymptotic behavior at zero and infinity of the fundamental solution of the Helmholtz
operator recalled in (6.3.4).

Lemma 6.3.1 Fix n ∈ N with n ≥ 2, along with z ∈ C \ (−∞, 0] and R ∈ (0, ∞). In
addition, select some arbitrary multi-index α ∈ N0n . Then there exists some constant
C = C(n, z, R, α) ∈ (0, ∞) with the following significance. If |α| is even then for each
x ∈ Rn with 0 < |x| < R one has

⎧
⎪ C if n = 2, 3 and |α| = 0,
⎪
⎪
 α  ⎨ 
⎪  
∂ Φz (x; n) − ∂ Φ0 (x; n) ≤ C  ln |x|  + 1
α
if n + |α| = 4, (6.3.7)
x x
⎪
⎪
⎪
⎪C|x| 4−n− |α |
⎩ if n + |α| ≥ 5,

while if |α| is odd then for each x ∈ Rn with 0 < |x| < R one has

 α  C if n + |α| = 3,
∂ Φz (x; n) − ∂ α Φ0 (x; n) ≤ (6.3.8)
x x
C|x| 4−n− |α | if n + |α| ≥ 4.

Finally, for large values of |x|, formula (6.3.4) implies (cf. [1, § 9.1]) the following
asymptotic behavior at infinity:
 1/2  (n−1)/2  
i z i z 1/2 |x |−π((n−1)/4)  
Φz (x; n) = − 1/2
e 1 + O |x| −1
2z 2π|x| (6.3.9)
as |x| → ∞, for all z ∈ C \ (−∞, 0] and n ≥ 2.
 
In particular, as long as Im z 1/2 > 0 it follows that Φz (x; n) decays exponentially
with respect to x ∈ Rn \ {0} as |x| → ∞.

Proof We begin by recalling (cf. [1, § 9.1]) that

(1)
H(n−2)/2 (·) = J(n−2)/2 (·) + iY(n−2)/2 (·), (6.3.10)

where Jλ and Yλ denote the regular and irregular Bessel functions, respectively. Let
us also recall the following absolutely convergent expansions (cf. [1, § 9.1, (9.1.10)-
(9.1.11), p. 360]):
892 6 Scattering by Rough Obstacles
 λ 
∞
ζ (−1) j ζ 2j
Jλ (ζ) = , ζ ∈ C \ (−∞, 0], λ ∈ R \ (−N),
2 j=0 4 j!Γ(λ + j + 1)
j

(6.3.11)
J−m (ζ) = (−1)m Jm (ζ), ζ ∈ C, m ∈ N0, (6.3.12)
Jλ (ζ) cos(λπ) − J−λ (ζ)
Yλ (ζ) = , ζ ∈ C \ (−∞, 0], λ ∈ (0, ∞) \ N, (6.3.13)
sin(λπ)

ζ −m  (m − j − 1)! ζ 2j 2
m−1
Ym (ζ) = − · j + Jm (ζ) ln(ζ/2)
2m π j=0 j! 4 π

ζm 
∞
(−1) j ζ 2j
− [ψ( j + 1) + ψ(m + j + 1)] j , (6.3.14)
2 π j=0
m 4 j!(m + j)!

for ζ ∈ C \ (−∞, 0], m ∈ N0 .

We note that all functions in (6.3.11), (6.3.13), and (6.3.14) are analytic in C\(−∞, 0]
and that Jm (·) is entire for m ∈ Z. In addition, all functions in (6.3.11)-(6.3.14) have
continuous limits as ζ ∈ C \ (−∞, 0] approaches nontangentially points in the cut
(−∞, 0], with generally different values on either side of the cut (−∞, 0] (due to the
presence of the functions ζ λ and ln(ζ)).
Due to the presence of the logarithmic term for even dimensions we next distin-
guish two cases, corresponding to even and odd values of n:
(i) Assume n = 2m + 2 for some m ∈ N0 . Then for each z ∈ C \ (−∞, 0] fixed, for
each x ∈ Rn \ {0} we have
 m
i z 1/2 (1)  1/2 
Φz (x; 2m + 2) = − Hm z |x|
4 2π|x|
 m
i z 1/2   1/2   
=− Jm z |x| + iYm z 1/2 |x| . (6.3.15)
4 2π|x|

This implies that, as |x| → 0,

 m  1/2 
i z 1/2   2i z |x|  m 
Φz (x; 2m + 2) = − O |x| m + ln O |x|
4 2π|x| π 2
 m   
i 2 z|x| 2
− (1 − δ m,0 ) (m − 1)! + (1 − δ m,1 )(m − 2)!
π z 1/2 |x| 4
 m 
i 2  
− (1 − δm,0 )O |x| 4 , (6.3.16)
π z |x|
1/2

where δ j,k is the Kronecker symbol.

6.3 The Family of Radiation Conditions (RC A ) 893

(ii) Assume n = 2m + 1 for some m ∈ N. Then for each z ∈ C \ (−∞, 0] fixed,

making use of [1, (10.1.1), p. 437], [1, (10.1.16), p. 439], and [1, (10.1.9), p. 437] we
may write
  m−(1/2)
i z 1/2 (1)  1/2 
Φz (x; 2m + 1) = − Hm−(1/2) z |x|
4 2π|x|
i (1)  1/2 
= − z m/2 π −m 21−m |x| 1−m hm−1 z |x| (6.3.17)
4
 1/2  m  (m + j − 1)! 
m−1
i z  −j
eiz |x |
1/2
= − 1/2 − 2iz 1/2 |x|
2z 2πi|x| j=0
j!(m − j − 1)!

for each x ∈ Rn \ {0}, where h(1) (·) is the first Spherical Bessel function of the third
kind, defined in [1, § 10.1]. Note that
1/2 |x |  
eiz = 1 + iz 1/2 |x| + O |x| 2 as |x| → 0, (6.3.18)

and, as |x| → 0,


m−1
(m + j − 1)!   −j
− 2iz 1/2 |x| (6.3.19)
j=0
j!(m − j − 1)!

  (m + j − 1)! 
 1−m m−1  m−1−j
= − 2iz 1/2 |x| − 2iz 1/2 |x|
j=0
j!(m − j − 1)!

  1−m (2m − 2)! (2m − 3)!  

= − 2iz 1/2 |x| + − 2iz 1/2 |x| + O(|x| 2 ) .
(m − 1)! (m − 2)!

Consequently, from (6.3.17)-(6.3.19) and algebra we see that, as |x| → 0,

⎧   
⎪
⎨ −(4π|x|)−1 1 + iz 1/2 |x| + O |x| 2 ,
⎪ m = 1,
Φz (x; 2m + 1) =    (6.3.20)
⎪
⎪ −[(2m − 1)ω2m ]−1 |x| 1−2m 1 + O |x| 2 , m ≥ 2.
⎩
We now claim that for each fixed n ∈ N, each fixed z ∈ C \ (−∞, 0], each fixed
R > 0 there exists a constant C = C(n, z, R) ∈ (0, ∞) with the property that whenever
0 < |x| < R we have

⎧
⎪ n = 2, 3,
⎪C,
  ⎪⎨ 
⎪ 
Φz (x; n) − Φ0 (x; n) ≤ C | ln(|x|)| + 1 , n = 4, (6.3.21)
⎪
⎪
⎪
⎪C|x| 4−n,
⎩ n ≥ 5.
894 6 Scattering by Rough Obstacles

Indeed, when n = 2 this follows from (6.3.6) and (6.3.3), while if n ≥ 3 this may be
seen from (6.3.3), (6.3.16), and (6.3.20).
Next, we observe that

∂x j Φz (x; n) = −2π x j Φz (x; n + 2), for all

(6.3.22)
z ∈ C \ (−∞, 0), x ∈ Rn \ {0}, j ∈ {1, . . . , n}, n ≥ 2.

Indeed, if z = 0 this is seen straight from definitions, while for z ∈ C \ (−∞, 0]

the recursion relation (6.3.22) is a consequence of the well-known identity (cf. [1,
§ 9.1])
d  −λ 
ζ Cλ (ζ) = −ζ −λ Cλ+1 (ζ), ζ ∈ C \ {0}, λ ∈ R, (6.3.23)
dζ

where Cλ (·) denotes any linear combination of Bessel functions of order λ with
coefficients independent of ζ and λ.
Given an arbitrary multi-index α ∈ N0n , based on (6.3.22) and induction we see
that there exist homogeneous polynomials P(α) in Rn of degree ∈ N0 such that if
|α| = 2m for m ∈ N then

∂xα Φz (x; n) = P0(α) (x) · Φz (x; n + |α|) + P2(α) (x) · Φz (x; n + |α| + 2)
(α)
+ · · · + P2m−2 (x) · Φz (x; n + |α| + 2m − 2)

+ (−2π) |α | x α Φz (x; n + 2|α|), (6.3.24)

for all z ∈ C \ (−∞, 0), x ∈ Rn \ {0}, n ≥ 2,

and if |α| = 2m + 1 for m ∈ N0 then

∂xα Φz (x; n) = P1(α) (x) · Φz (x; n + |α| + 1) + P3(α) (x) · Φz (x; n + |α| + 3)
(α)
+ · · · + P2m−1 (x) · Φz (x; n + |α| + 2m − 1)

+ (−2π) |α | x α Φz (x; n + 2|α|), (6.3.25)

for all z ∈ C \ (−∞, 0), x ∈ Rn \ {0}, n ≥ 2.

Hence, in the case when |α| = 2m for some m ∈ N, for each z ∈ C \ (−∞, 0],
x ∈ Rn \ {0}, and n ≥ 2, we may use (6.3.24) to estimate
6.3 The Family of Radiation Conditions (RC A ) 895
 α 
∂ Φz (x; n) − ∂ α Φ0 (x; n)
x x
   
≤ P0(α) (x) · Φz (x; n + |α|) − Φ0 (x; n + |α|)
   
+ P2(α) (x) · Φz (x; n + |α| + 2) − Φ0 (x; n + |α| + 2)
 (α)   
+ · · · + P2m (x) · Φz (x; n + 2|α|) − Φ0 (x; n + 2|α|). (6.3.26)

Now, (6.3.7) follows from this and (6.3.21). Finally, (6.3.8) is justified in a similar
fashion, making use of (6.3.25) and (6.3.21). 

Shortly, we shall need the information regarding the asymptotic behavior of Φk

and its derivatives at infinity contained in the next lemma (see [91] for a proof).

Lemma 6.3.2 Pick k ∈ (0, ∞) and consider n ∈ N with n ≥ 2. Then for each fixed
x ∈ Rn one has  
Φk (x − y) = O |y| −(n−1)/2 as |y| → ∞. (6.3.27)
Also, for each multi-index α ∈ N0n with |α| > 0 one has
 
(∂ α Φk )(x − y) = (−ik
y )α Φk (x − y) + O |y| −(n+1)/2 as |y| → ∞,
(6.3.28)
uniformly for x in compact subsets of Rn .

Finally, for any two multi-indices α, β ∈ N0n one has

 
(∂ α+β Φk )(x − y) = (ik
x )α (∂ β Φk )(x − y) + O |x| −(n+1)/2 as |x| → ∞,
(6.3.29)
uniformly for y in compact subsets of Rn .

We next use the asymptotics of Φk to study the behavior at infinity of functions

that are representable as formal convolution of derivatives of Φk with finite measures.

Lemma 6.3.3 Fix k ∈ (0, ∞) and pick n ∈ N with n ≥ 2. Let Ω ⊆ Rn be an exterior

domain and consider a Borel measure μ on ∂Ω. Given f ∈ L 1 (∂Ω, μ) and α ∈ N0n
arbitrary define
∫
u(x) := (∂ α Φk )(x − y) f (y) dμ(y), ∀x ∈ Ω. (6.3.30)
∂Ω

Then u belongs to 𝒞∞ (Ω), satisfies (Δ + k 2 )u = 0 in Ω, and

   
ikx j u(x) − ∂j u (x) = O |x| −(n+1)/2 as |x| → ∞,
(6.3.31)
for each index j ∈ {1, . . . , n}.

Proof The background assumptions readily imply that u ∈ 𝒞∞ (Ω) and for each
multi-index β ∈ N0n we have
896 6 Scattering by Rough Obstacles
∫
(∂ β u)(x) = (∂ α+β Φk )(x − y) f (y) dμ(y), ∀x ∈ Ω. (6.3.32)
∂Ω

In particular, (Δ + k 2 )u = 0 in Ω. To show that u also satisfies (6.3.31), fix some

j ∈ {1, . . . , n}. Then based on (6.3.32) and (6.3.29) we may write (keeping in mind
the fact that f ∈ L 1 (∂Ω, μ))
∫
(∂j u)(x) = (∂ α+e j Φk )(x − y) f (y) dμ(y)
∂Ω
∫
 
= x j (∂ α Φk )(x − y) + O |x| −(n+1)/2 f (y) dμ(y)
ik
∂Ω
 
x j u(x) + O |x| −(n+1)/2 as |x| → ∞,
= ik (6.3.33)

from which (6.3.31) follows. 

It turns out that derivatives of radiating functions are radiating themselves. Here
is a formal statement:

Proposition 6.3.4 If u is radiates at infinity (in the sense of Definition 6.1.1), then
for each multi-index α ∈ Rn the function ∂ α u also radiates at infinity.

Proof This is a consequence of the integral representation formula from Theo-

rem 6.1.4 (applied with Ω := Rn \ B(0, R) for a sufficiently large R ∈ (0, ∞)), and
Lemma 6.3.3. 
For further use, we isolate a technical result in the following lemma.

Lemma 6.3.5 Suppose k ∈ (0, ∞) and consider n ∈ N with n ≥ 2 along with

M ∈ N arbitrary. Let u = (uJ )1≤J ≤M be vector-valued function defined in an
exterior domain Ω ⊆ Rn whose components
 JI  are in 𝒞∞ (Ω) and which satisfies
(Δ + k )u = 0 in Ω. Also, let A = ar s 1≤J,I ≤M be a complex coefficient tensor
2
1≤r,s ≤n
which allows expressing the vector Laplacian as Δ = divA∇. Then for each fixed
point x ∈ Rn the limit
∫
[u]∞
A
(x) := lim 
ys arI sJ (∂r Φk )(x − y)uJ (y) (6.3.34)
R→∞ |y |=R

+ asr
IJ
Φk (x − y)(∂r uJ )(y) dH n−1 (x)
1≤I ≤M

exists in C M (in fact there exists some Rx ∈ (0, ∞) with the property that the
expression in the round parentheses is independent of R if R ≥ Rx ).

Proof Pick an arbitrary point x ∈ Rn and consider a radius Rx > |x| with the
property that ∂Ω ⊆ B(x, Rx ). Also, fix an index I ∈ {1, . . . , M } and suppose R1, R2
are such that Rx < R1 < R2 < ∞. We may then use the standard version of the
6.4 Single and Double Acoustic Layer Potentials 897

Divergence Theorem to express the difference between the expression in the round
parentheses in (6.3.34) written, respectively, for R = R2 and R = R1 , as
∫
∂ys arI sJ (∂r Φk )(x − y)uJ (y) + asr
IJ
Φk (x − y)(∂r uJ )(y) dL n (x).
R1 < |y |<R2
(6.3.35)

Since arJsI ∂r ∂s = δJ I Δ for each index J ∈ {1, . . . , M } (cf. (6.2.17)), on the domain
of integration we have

∂ys arI sJ (∂r Φk )(x − y)uJ (y) + asr

IJ
Φk (x − y)(∂r uJ )(y)

= −arI sJ (∂s ∂r Φk )(x − y)uJ (y)

+ arI sJ (∂r Φk )(x − y)(∂s uJ )(y) − asr

IJ
(∂s Φk )(x − y)(∂r uJ )(y)

+ asr
IJ
Φk (x − y)(∂s ∂r uJ )(y)

= −δJ I (ΔΦk )(x − y)uJ (y) + δI J Φk (x − y)(ΔuJ )(y)

= k 2 Φk (x − y)u I (y) − k 2 Φk (x − y)u I (y) = 0, (6.3.36)

so the desired conclusion follows from (6.3.35). 

6.4 Single and Double Acoustic Layer Potentials

Fix an arbitrary wave number k ∈ (0, ∞) and consider n ∈ N satisfying n ≥ 2 along

with some M ∈ N arbitrary. Suppose Ω ⊆ Rn is an exterior domain which is a set of
locally finite perimeter. Abbreviate σ := H n−1 ∂Ω and denote by ν = (ν1, . . . , νn )
the geometric measure theoretic outward unit normal to Ω. In this setting, define
the action of the (vector) acoustic single layer potential operator on any function
 M
f = ( fI )1≤I ≤M ∈ L 1 (∂∗ Ω, σ) as
∫
𝒮 f (x) := Φk (x − y) fI (y) dσ(y) for all x ∈ Ω. (6.4.1)
∂∗ Ω 1≤I ≤M

 
Also, having fixed a complex coefficient tensor A := arJsI 1≤J,I ≤M that allows us
1≤r,s ≤n
to express the vector Laplacian as Δ = divA∇, associated the acoustic double layer
potential operator
 ∫ 
D A f (x) := − νs (y)ar s (∂r Φk )(x − y) fJ (y) dσ(y)
IJ
for all x ∈ Ω,
∂∗ Ω 1≤I ≤M
(6.4.2)
898 6 Scattering by Rough Obstacles
M
for each vector-valued function f = ( fJ )1≤J ≤M ∈ [L 1 (∂∗ Ω, σ) .
In [115, §1.5] we shall discuss fundamental properties of acoustic boundary layer
potentials considered in open sets with uniformly rectifiable boundaries. For now, in
the theorem below, the goal is to describe a basic Green-type integral representation
formula for null-solutions of the Helmholtz operator in an exterior domain in which
no decay demands for the functions involved are imposed (cf. (6.4.5)).

Theorem 6.4.1 Select k ∈ (0, ∞) and consider n ∈ N with n ≥ 2 along with some
arbitrary integer M ∈ N. Let Ω ⊆ Rn be an exterior domain with a lower Ahlfors
regular boundary and such that σ := H n−1 ∂Ω is a doubling measure on ∂Ω.
 the vector Laplacian as Δ = divA∇ for some complex coefficient tensor
Also, write
A = arJsI 1≤J,I ≤M . In this setting, consider a vector-valued function
1≤r,s ≤n
 M
u = (u I )1≤I ≤M ∈ 𝒞∞ (Ω) with (Δ + k 2 )u = 0 in Ω,
κ−n.t. κ−n.t.
such that u∂Ω , (A∇u)∂Ω exist σ-a.e. on ∂nta Ω, (6.4.3)
ρ ρ
and Nκ u, Nκ (A∇u) ∈ L 1 (∂Ω, σ) for some κ, ρ > 0,

(with A∇u understood in the sense of (6.2.23)). Finally, recall (6.3.34) and, with
ν = (ν1, . . . , νn ) denoting the geometric measure theoretic outward unit normal to
Ω, at σ-a.e. point on ∂∗ Ω define (compare with (6.2.21))

n
κ−n.t.
∂νAu := νr (A∇u)r,J ∂Ω
1≤J ≤M
r=1


n 
M 
n κ−n.t.

= νr arJsI ∂s u I  . (6.4.4)
∂Ω 1≤J ≤M
r=1 I=1 s=1

Then for each x ∈ Ω one has

 κ−n.t.   
u(x) = D A u∂Ω (x) − 𝒮 ∂νAu (x) − [u]∞
A
(x). (6.4.5)

Furthermore, if one additionally assumes that

∫
   
iku(x) − ∂ Au (x)2 dH n−1 (x) = o(1) as R → ∞, (6.4.6)

x
|x |=R

and ∫
|u(x)| 2 dH n−1 (x) = O(1) as R → ∞, (6.4.7)
|x |=R

it follows that [u]∞

A (x) = 0 for each x ∈ Ω, in which case (6.4.5) becomes the integral

representation formula
 κ−n.t.   
u = D A u∂Ω − 𝒮 ∂νAu in Ω. (6.4.8)
6.4 Single and Double Acoustic Layer Potentials 899

 a point x ∈ Ω and an index I ∈ {1, . . . , M }. Consider the vector field

Proof Fix
F = Fs 1≤s ≤n whose components defined at L n -a.e. point y ∈ Ω according to

Fs (y) := −arI sJ (∂r Φk )(x − y)uJ (y) − asr

IJ
Φk (x − y)∂r uJ (y), 1 ≤ s ≤ n. (6.4.9)

Then this definition readily implies that

   n
F belongs to 𝒞∞ Ω \ {x} ∩ Lloc
1
(Ω, L n ) , (6.4.10)
 
Nκε F ∈ L 1 (∂Ω, σ) if ε ∈ 0, min{ρ, dist(x, ∂Ω)} , (6.4.11)

and the nontangential trace

κ−n.t.
F ∂Ω exists at σ-a.e. point on ∂nta Ω. (6.4.12)

We shall next compute divF in the sense of distributions in Ω. Concretely,

 
divF(y) = arI sJ ∂yr ∂ys Φk (x − y) uJ (y)

− arI sJ (∂r Φk )(x − y)∂s uJ (y) + asr

IJ
(∂s Φk )(x − y)∂r uJ (y)

− asr
IJ
Φk (x − y)∂s ∂r uJ (y)

=: I + II + III + IV. (6.4.13)

Note that II + III = 0. Also, in view of the fact that Δu = −k 2 u we may write
 
IV = − Φk (x − y)(Δu)(y) I = k 2 Φk (x − y)u I (y). (6.4.14)

Finally, based on (6.2.17) and the fact that Φk is a fundamental solution of the
Helmholtz operator Δ + k 2 , for each index J ∈ {1, . . . , M } we may write
   
arI sJ ∂yr ∂ys Φk (x − ·) = δI J Δy Φk (x − ·)

= −k 2 Φk (x − ·)δI J + δx δI J in D (Ω). (6.4.15)

Thus,

I = −k 2 Φk (x − y)u I (y) + δx (y)u I (y) in D (Ω). (6.4.16)

Altogether, this proves that

divF = u I (x)δx in D (Ω). (6.4.17)

In particular, divF ∈ ℰ (Ω).

Granted the aforementioned properties, the version of the Divergence Theorem
established in [112, Corollary 1.5.2] applies. First, this guarantees that the contri-
900 6 Scattering by Rough Obstacles

bution of F at infinity is meaningfully and unambiguously defined. Based on [112,

Proposition 4.7.1] (cf. also [112, (4.7.6)]) as well as (6.4.9) and (6.3.34) we may
then write
∫
[F]∞ = lim ys Fs (y) dH n−1 (y)
R→∞ |y |=R
∫
= lim ys −arI sJ (∂r Φk )(x − y)uJ (y)

R→∞ |y |=R

−asr
IJ
Φk (x − y)∂r uJ (y) dH n−1 (y)
 A 
= − [u]∞ (x) I . (6.4.18)

In view of (6.4.16), the Divergence Formula recorded in [112, (1.5.19)] presently

gives
∫
   κ−n.t. 
u I (x) = (𝒞∞ (Ω)) divF, 1 𝒞∞ (Ω) =
∗ νs Fs ∂Ω dσ + [F]∞ . (6.4.19)
b b ∂∗ Ω

In view of (6.4.9), (6.4.4), (6.4.1), and (6.4.2) we may further write

∫
 κ−n.t. 
νs (y) Fs ∂Ω (y) dσ(y)
∂∗ Ω
∫
 κ−n.t. 
=− νs (y)arI sJ (∂r Φk )(x − y) uJ ∂Ω (y) dσ(y)
∂∗ Ω
∫
κ−n.t.
− Φk (x − y)νs (y) (asr
IJ
∂r uJ )∂Ω (y) dσ(y)
∂∗ Ω
 κ−n.t.   
= D A u∂Ω (x) − 𝒮 ∂νAu (x) . (6.4.20)
I

Collectively, (6.4.18), (6.4.19), and (6.4.20) imply that

 κ−n.t.   
u I (x) = D A u∂Ω (x) − 𝒮 ∂νAu (x) − [u]∞
A
(x) . (6.4.21)
I

Since the index I ∈ {1, . . . , M } was arbitrary, this completes the proof of the
Green-type integral representation formula (with a correction taking into account
the behavior at infinity) for u claimed in (6.4.5).
Next we further analyze the contribution at infinity [u]∞A (x) defined in (6.3.34).

The idea is to use the asymptotic formulas for Hankel functions and their derivatives
from Lemma 6.3.2. Specifically, for each x ∈ Ω fixed we may use (6.3.34) and
(6.3.28) to write
6.5 L 2 -Finiteness Results 901
∫
[u]∞
A
(x) = lim 
ys arI sJ (∂r Φk )(x − y)uJ (y)
R→∞ |y |=R
+ asr
IJ
Φk (x − y)∂r uJ (y) dH n−1 (y)
∫
= lim Φk (x − y) − ikarI sJ 
ys 
yr uJ (y) + asr
IJ

ys ∂r uJ (y) dH n−1 (y)
R→∞ |y |=R
∫
   
+ lim O R−(n+1)/2 O |u(y)| dH n−1 (y) =: I + II. (6.4.22)
R→∞ |y |=R

Assume in addition that u satisfies (6.4.6). Then the Cauchy-Schwarz inequality and
(6.4.7) permit us to conclude that
 ∫ 
 −1  1/2
II = lim O R |u(y)| dH (y)
2 n−1
= 0. (6.4.23)
R→∞ |y |=R

As for I, keeping (6.2.20) in mind we may express

∫
 
I = lim Φk (x − y) −iku I (y) + ∂yAu I (y) dH n−1 (y). (6.4.24)
R→∞ |y |=R

Hence, using the decay of Φk from the Lemma 6.3.2, and keeping in mind that u
satisfies (6.4.6), we obtain
∫  1/2
|I| ≤ lim |Φk (x − y)| dH 2 n−1
(y) ×
R→∞ |y |=R

∫  12
   
× iku(y) − ∂ Au (y)2 dH n−1 (y)

y
|y |=R

≤ lim sup O(1) · o(1) = 0. (6.4.25)
R→∞

A (x) = 0 whenever u also satisfies (6.4.6).

This proves that [u]∞ 

6.5 L 2 -Finiteness Results

As is apparent from Lemma 6.1.2, sometimes it happens that radiation conditions,

such as (6.4.6), imply L 2 -finiteness properties, like the one in (6.4.7). Our goal in
this section is to understand when this is the case.
With the piece of notation introduced in (6.2.24), we have the following basic
L 2 -finiteness result.
902 6 Scattering by Rough Obstacles

Proposition 6.5.1 Pick k ∈ (0, ∞) and consider n ∈ N with n ≥ 2 along with M ∈ N

arbitrary. Write the vector Laplacian as Δ = divA∇ for some coefficient tensor A
satisfying Im Aζ, ζ ≤ 0 for every complex vector ζ. Also, let u be a C M -valued
null-solution of the vector Helmholtz operator Δ + k 2 in an exterior domain Ω ⊆ Rn
with the property that
∫
 
iku(x) − (∂ Au)(x)2 dH n−1 (x) = O(1) as R → ∞. (6.5.1)

x
|x |=R

Then ∫
|u(x)| 2 dH n−1 (x) = O(1) as R → ∞, (6.5.2)
|x |=R

and ∫
 A 
(∂ u)(x)2 dH n−1 (x) = O(1) as R → ∞. (6.5.3)

x
|x |=R

A few comments in relation to this result are in order. First, since it is clear
from the triangle inequality that (6.5.1)-(6.5.2) imply (6.5.3), the conclusion in
Proposition 6.5.1 amounts to a reverse triangle inequality. Second, replacing O(1) by
o(1) in the hypothesis formulated in (6.5.1) does not imply o(1) in the conclusions
formulated in (6.5.2), (6.5.3).
As a preamble to the proof of Proposition 6.5.1, we establish the following
auxiliary result.

Lemma 6.5.2 Let k ∈ (0, ∞) and consider n ∈ N with n ≥ 2. Suppose D ⊆ Rn

is a bounded 𝒞1 domain with surface measure σ and outward unit normal ν, and
 M
u ∈ 𝒞1 (D)  , for  some M ∈ N, is a vector-valued function such that (Δ + k )u = 0
2
J I
in D. If A := ar s 1≤J,I ≤M is a complex coefficient tensor in the writing of the vector
1≤r,s ≤n
Laplacian as Δ = divA∇ with the property that

Im Aζ, ζ ≤ 0 for every complex vector ζ ∈ Cn M , (6.5.4)

then
∫ ∫
   2 
iku − ∂ Au2 dσ ≥ k 2 |u| 2 + ∂νAu dσ. (6.5.5)
ν
∂D ∂D

Proof Expand the left-hand side of (6.5.5) as

∫ ∫ ∫
   A 2  
iku − ∂ Au2 dσ =  
k |u| + ∂ν u dσ − 2kIm
2 2
∂νAu, u dσ.
ν
∂D ∂D ∂D
(6.5.6)

Hence, since k > 0, it suffices to show that

∫
 A
Im ∂ν u, u dσ ≤ 0. (6.5.7)
∂D
6.5 L 2 -Finiteness Results 903

Integration by parts and using the fact that (Δ + k 2 )u = 0 in D, we obtain (with L n

denoting the Lebesgue measure in Rn , and with the convention of summation for
repeated indices in effect)
∫ ∫

A∇u, ∇u dL n = arJsI (∂s u I )∂r uJ dL n
D D
∫ ∫
=− arJsI (∂r ∂s u I )uJ dL + n
arJsI νr (∂s u I )uJ dσ
D ∂D
∫ ∫

= k 2 |u| 2 dL n + ∂νAu, u dσ. (6.5.8)
D ∂D

Taking the imaginary parts of the most extreme sides of this identity yields, bearing
in mind that Im Aζ, ζ ≤ 0 for every vector ζ with complex components,
∫

0≥ Im A∇u, ∇u dL n
D
∫ ∫

= Im k 2 |u| 2 dL n + Im ∂νAu, u dσ
D ∂D
∫

= Im ∂νAu, u dσ, (6.5.9)
∂D

since k is real. This establishes (6.5.7) and finishes the proof of the lemma. 

We are now prepared to present the proof of the L 2 -finiteness result from Propo-
sition 6.5.1.
Proof of Proposition 6.5.1 Apply Lemma 6.5.2 with D := B(0, R) \ B(0, r), where
0 < r < R are such that D ⊂ Ω. Since ∂D = ∂B(0, R) ∪ ∂B(0, r), we have
∫
 
iku(x) − (∂ Au)(x)2 dH n−1 (x)

x
|x |=R
∫
 
+ iku(x) + (∂ Au)(x)2 dH n−1 (x)

x
|x |=r
∫ ∫
 A 
≥ k 2 |u(x)| 2 dH n−1 (x) + (∂ u)(x)2 dH n−1 (x)

x
|x |=R |x |=R
∫
 2 
+ k 2 |u(x)| 2 + (∂ Au)(x) dH n−1 (x).

x (6.5.10)
|x |=r

Since for each r fixed the left-hand side of (6.5.10) is O(1) as R → ∞, it follows
that for each r fixed the right-hand side of (6.5.10) must be O(1) as R → ∞. From
this, the desired conclusion follows. 
904 6 Scattering by Rough Obstacles

6.6 The Principal Result

The main result identifying all radiation conditions for the vector Helmholtz op-
erator is presented next. Among other things, this provides satisfactory answers to
Questions 1-2 formulated earlier in this section.

Theorem 6.6.1 Pick k ∈ (0, ∞), fix n ∈ N with n ≥ 2, and let M ∈ N be arbitrary.

 M
Suppose Ω ⊆ Rn and consider a function u ∈ 𝒞∞ (Ω) satisfying (Δ + k 2 )u = 0
in Ω. Then items (1)-(6) below are equivalent:
(1) The function u satisfies
   
x j u(x) − ∂j u (x) = O |x| −(n+1)/2 as |x| → ∞,
ik
(6.6.1)
for each index j ∈ {1, . . . , n}.

(2) The function u satisfies

   
x j u(x) − ∂j u (x) = o |x| −(n−1)/2 as |x| → ∞,
ik
(6.6.2)
for each index j ∈ {1, . . . , n}.

(3) For some, or any, complex coefficient tensor A such that Δ = divA∇ and with
the property that Im Aζ, ζ ≤ 0 for every vector ζ with complex components,
one has    
iku(x) − ∂xAu (x) = o |x| −(n−1)/2 as |x| → ∞. (6.6.3)
(4) For some, or any, complex coefficient tensor A such that Δ = divA∇ and with
the property that Im Aζ, ζ ≤ 0 for every vector ζ with complex components,
one has    
iku(x) − ∂xAu (x) = O |x| −(n+1)/2 as |x| → ∞. (6.6.4)
(5) For some, or any, complex coefficient tensor A such that Δ = divA∇ and with
the property that Im Aζ, ζ ≤ 0 for every vector ζ with complex components,
one has
∫
   
iku(x) − ∂ Au (x)2 dH n−1 (x) = o(1) as R → ∞. (6.6.5)

x
|x |=R

(6) For some, or any, complex coefficient tensor A such that Δ = divA∇ and with
the property that Im Aζ, ζ ≤ 0 for every vector ζ with complex components,
one has
∫
   
iku(x) − ∂ Au (x)2 dH n−1 (x) = O(R−2 ) as R → ∞. (6.6.6)

x
|x |=R

Strengthen the hypotheses on the underlying set Ω by additionally assuming that

∂Ω is a lower Ahlfors regular set and σ := H n−1 ∂Ω is a doubling measure on ∂Ω.
Also, denote by ν the geometric measure theoretic outward unit normal to Ω and fix
6.6 The Principal Result 905

some aperture parameter κ ∈ (0, ∞) along with a truncation parameter ρ ∈ (0, ∞).
Then if u also satisfies
κ−n.t. κ−n.t.
the traces u∂Ω , (∇u)∂Ω exist σ-a.e. on ∂nta Ω,
(6.6.7)
ρ ρ
and Nκ u, Nκ (∇u) ∈ L 1 (∂Ω, σ),

it follows that each of the radiation conditions described in items (1)-(6) is further
equivalent with having
 κ−n.t.   
u = D A u∂Ω − 𝒮 ∂νAu in Ω
(6.6.8)
for some, or any, A such that Δ = divA∇,

where D A, 𝒮 are the acoustic double and single layer potential operators associated
with the wave number k, the coefficient tensor A, and the exterior domain Ω as in
(6.4.1)-(6.4.2).
Finally, if the function u is radiating at infinity (cf. Definition 6.1.1) then for any
given complex coefficient tensor A such that Δ = divA∇ one has
 κ−n.t.    κ−n.t. κ−n.t.
u = D A u∂Ω − 𝒮 ∂νAu in Ω if both traces u∂Ω , (A∇u)∂Ω (6.6.9)
ρ ρ
exist σ-a.e. on ∂nta Ω, and if Nκ u, Nκ (A∇u) ∈ L 1 (∂Ω, σ).

A few comments shedding further light on this theorem are in order. First, note
that (6.6.1), (6.6.2) are intrinsic conditions (independent of the choice of a coefficient
tensor A used to write the vector Laplacian as Δ = divA∇), which may be equivalently
be considered on each of the coefficients of the vector-valued
 function u.
Second, writing Δ = divA∇ with A := A0 = δJ I δr s 1≤J,I ≤M the identity, we see
1≤r,s ≤n
from (6.2.22) that any of the items (1)-(6) in Theorem 6.6.1 is further equivalent to
any of the conditions

n
 
iku(x) − x j (∂j u)(x) = o |x| −(n−1)/2 as |x| → ∞,
 (6.6.10)
j=1


n
 
iku(x) − iku(x) − x j (∂j u)(x) = O |x| −(n+1)/2 as |x| → ∞,
 (6.6.11)
j=1

∫  
n 2
 
iku(x) − 
x j (∂j u)(x) dH n−1 (x) = o(1) as R → ∞, (6.6.12)
|x |=R j=1

∫  
n 2
 
iku(x) − x j (∂j u)(x) dH n−1 (x) = O(R−2 ) as R → ∞.
 (6.6.13)
|x |=R j=1

In particular, any of the items (1)-(6) in Theorem 6.6.1 is further equivalent to

Sommerfeld’s radiation condition for the function u.
906 6 Scattering by Rough Obstacles

Third, as is apparent from (A.0.118), for any function u satisfying the radiation
condition (6.6.2) and any homogeneous first-order system D with constant complex
coefficients we have
 
k Sym(D;  x )u(x) − (Du)(x) = O |x| −(n+1)/2 as |x| → ∞. (6.6.14)

Here is the proof of Theorem 6.6.1.

Proof of Theorem 6.6.1 To set the stage for the subsequent discussion, fix some
sufficiently large radius R ∈ (0, ∞) (with the property that ∂Ω ⊆ B(0, R)) and
introduce the open set ΩR := Rn \ B(0, R). Then ΩR is an exterior domain in Rn
whose closure is contained in Ω. In particular, all components of u are in 𝒞∞ (ΩR ).
Let us now begin in earnest. The fact that (1) ⇒ (2) is obvious. Assume now that
the function u satisfies (6.6.2). Then for each index I ∈ {1, . . . , M } we may write

n
iku I (x) − 
x · (∇u I )(x) = iku I (x) − 
x j (∂j u I )(x)
j=1


n
 
= iku I (x) −  x j u I (x) + o |x| −(n−1)/2
x j ik
j=1
 
= o |x| −(n−1)/2 as |x| → ∞, (6.6.15)

i.e.,
 
x · (∇u I )(x) = o |x| −(n−1)/2 as |x| → ∞.
iku I (x) −  (6.6.16)

This guarantees that Sommerfeld’s radiation condition (6.1.2) is satisfied by each

component of u. As such, we may invoke Theorem 6.1.4 for each component of u
restricted to ΩR and obtain the integral representation formula (6.1.18) written for
ΩR in place of Ω for said function. In turn, from this and Lemma 6.3.3 it follows
that (6.6.1) holds, hence (2) ⇒ (1). The argument so far shows that (1) ⇔ (2).
Assume now that u is a function
  as in item (1) and consider an arbitrary com-
plex coefficient tensor A := arJsI 1≤J,I ≤M which permits us to express the vector
1≤r,s ≤n
Laplacian as Δ = divA∇. Thanks to (6.2.21), (6.6.2), and (6.2.20) we may then write
   
iku(x) − ∂xAu (x) = iku(x) − arJsI 
xr ∂s u I (x) J
   
= iku(x) − ik arJsI  xs u I (x) J + O |x| −(n+1)/2
xr 
 2   
= iku(x) − ik |x | δJ I u I (x) J + O |x| −(n+1)/2
 
= O |x| −(n+1)/2 as |x| → ∞. (6.6.17)
6.6 The Principal Result 907

This goes to show that (6.6.4) holds, i.e., (1) implies the strongest version of (4),
which clearly implies the strongest version of (6), which further implies the strongest
version of (5).
Going further, suppose A is a complex coefficient tensor such that Δ = divA∇
and with the property that Im Aζ, ζ ≤ 0 for every vector ζ with complex com-
ponents. Also, assume u satisfies (6.6.5) for said coefficient tensor A. Then from
Proposition 6.5.1 and Theorem 6.4.1 applied to u restricted to ΩR we conclude that
the integral representation formula (6.4.8) written for ΩR in place of Ω is valid. In
concert with Lemma 6.3.3, this implies that (6.6.1) holds, hence (5) ⇒ (1). At this
stage, we have proved that items (1)-(6) are equivalent.
For the remainder of the proof strengthen the hypotheses on the underlying set
Ω as indicated in the statement of the theorem. For now, assume that u has the
additional qualities listed in (6.6.7). Then, as may be seen from Lemma 6.3.3, the
property recorded in (6.6.8) implies the claim made in item (1).
To deal with the opposite implication, assume u satisfies the radiation condition
formulated in (6.6.1) and pick a complex coefficient tensor A which permits us to
express the vector Laplacian as Δ = divA∇. It suffices to prove the claim in (6.6.9).
κ−n.t. κ−n.t.
To this end, assume that the nontangential traces u∂Ω , (A∇u)∂Ω exist at σ-a.e.
ρ ρ
point on ∂nta Ω, and that Nκ u, Nκ (A∇u) ∈ L 1 (∂Ω, σ). As proved earlier in (6.6.17),
from (6.6.1) we deduce that
   
iku(x) − ∂xAu (x) = O |x| −(n+1)/2 as |x| → ∞. (6.6.18)

As a consequence,
∫
   
iku(x) − ∂ Au (x)2 dH n−1 (x) = O(R−2 ) as R → ∞. (6.6.19)

x
|x |=R

We already know that (6.6.1) implies the version of (6.6.5) written for the identity
coefficient tensor=, i.e., A := δJ I δr s 1≤J,I ≤M . In view of (6.2.22) this yields
1≤r,s ≤n

∫  
n 2
 
iku(x) − 
x j (∂j u)(x) dH n−1 (x) = o(1) as R → ∞. (6.6.20)
|x |=R j=1

In concert with Lemma 6.1.2, this also implies

∫
|u(x)| 2 dH n−1 (x) = O(1) as R → ∞. (6.6.21)
|x |=R

Granted (6.6.19) and (6.6.21), we may invoke Theorem 6.4.1 to conclude that (6.4.8)
holds. This establishes the claim in (6.6.9).
To finish the proof of the theorem there remains to observe that if u radiates
at infinity in the sense of Definition 6.1.1 (as assumed in the preamble to (6.6.9))
then the
 weakest
 version of the claim in item (3) (corresponding to the choice
A := δJ I δr s 1≤J,I ≤M ) holds and, as already proved above, this implies (6.6.9). 
1≤r,s ≤n
908 6 Scattering by Rough Obstacles

6.7 Coordinate-Free Formalism and Examples

Select some k ∈ (0, ∞) and pick some n ∈ N with n ≥ 2. Also, fix M, N ∈ N

arbitrary. We are interested in the case when the vector Laplacian2 in Rn factors as
!
Δ = DD (6.7.1)

where D is a first-order constant complex constant coefficient N × M system, acting

on vector-valued function u = (u I )1≤I ≤M according to
 
Du = bsL I ∂s u I + cL I u I 1≤L ≤ N (6.7.2)

for some constant coefficients bsL I , cL I ∈ C, with 1 ≤ L ≤ N, 1 ≤ I ≤ M,

1 ≤ s ≤ n, and where D! is a first-order constant complex constant coefficient M × N
system acting on vector-valued function w = (wJ )1≤J ≤ N according to
 LJ 
! = !
Dw br ∂r w L + !cL J wJ 1≤J ≤M . (6.7.3)

Introducing the complex coefficient tensor

 
A := arJsI 1≤J,I ≤M with arJsI := !
brL J bsL I ∈ C
1≤r,s ≤n (6.7.4)
for all J, I ∈ {1, . . . , M } and r, s ∈ {1, . . . , n},

then permits us to express the vector Laplacian as Δ = divA∇. In this regard, note
that

the condition ImAζ, ζ ≤ 0 for all vectors ζ with complex entries is

! = λD for some λ ∈ C with Im λ ≥ 0
automatically satisfied if D (6.7.5)
(where ‘bar’ denotes complex conjugation, and where  stands for real
transposition).

Indeed, if all coefficients brL J are real numbers and if there exists λ ∈ C with
Im λ ≥ 0 such that
!
brL J = −λbrL J for all J, I ∈ {1, . . . , M } and r, s ∈ {1, . . . , n}, (6.7.6)

then for each (doubly indexed) vector ζ = (ζsI ) 1≤s ≤n with complex components
1≤I ≤M
(6.2.24) becomes


M 
N 
n N 
 M 
n 2
 
Aζ, ζ = −λ brL J bsL I ζsI ζr J = −λ  bsL I ζsI  (6.7.7)
I,J=1 L=1 r,s=1 L=1 I=1 s=1

which proves that in this case we indeed have

2 recall that we tacitly identify Δ with the M × M system ΔI M ×M , where I M ×M is the M × M
identity matrix
6.7 Coordinate-Free Formalism and Examples 909

N 
 M 
n 2
 
ImAζ, ζ = (−Im λ)  bsL I ζsI  ≤ 0. (6.7.8)
L=1 I=1 s=1

Next, bearing in mind (6.7.2)-(6.7.4), (6.2.21), and (A.0.118) we may express

the conormal derivative associated with the coefficient tensor A from (6.7.4) in an
invariant fashion, namely
 
! ξ D for each ξ ∈ Cn .
∂ξA = (−i)Sym D; (6.7.9)

This further gives

 
! 
∂xA = (−i)Sym D; x D for each x ∈ Rn \ {0}. (6.7.10)

As such, the corresponding radiation condition (RC A) (associated as in (6.3.2) with

the coefficient tensor A from (6.7.4)) for a vector-valued null-solution u of the vector
Helmholtz operator Δ + k 2 in an exterior domain in Rn reads
   
iku(x) − (−i)Sym D;! 
x (Du)(x) = o |x| −(n−1)/2 as |x| → ∞. (6.7.11)

For further use let us also remark that, thanks to (6.2.23), (A.0.48), (6.7.4), and
(6.7.2), for each function u = (u I )1≤I ≤M we have
   LJ LI 
A∇u = arJsI ∂s u I 1≤J ≤M = !br bs ∂s u I 1≤J ≤M
1≤r ≤n 1≤r ≤n
 LJ 
= !br (Du) L 1≤J ≤M . (6.7.12)
1≤r ≤n

In turn, if Ω ⊆ Rn is an arbitrary open set, σ := H n−1 ∂Ω, and κ, ρ ∈ (0, ∞) are

arbitrary, then (6.7.12) implies (also bearing in mind [112, (8.2.28)]) that for each
 M
function u ∈ 𝒞∞ (Ω) we see that
κ−n.t.
if (Du)∂Ω (x) exists at some point x ∈ ∂nta Ω, then the nontangen-
κ−n.t.
tial trace (A∇u)∂Ω (x) also exists, and if Nκ (Du) ∈ L 1 (∂Ω, σ)
ρ
(6.7.13)
ρ
then the nontangential maximal function Nκ (A∇u) also belongs to
L 1 (∂Ω, σ).
Next we reconsider the acoustic double and single layer potential operators in the
formalism associated with the factorization (6.7.1). First, let us take a look at the
integrand in (6.4.2). Observe that for each point x ∈ Ω and σ-a.e. y ∈ ∂∗ Ω we may
write (keeping (6.7.2)-(6.7.4) and (A.0.118) in mind)
910 6 Scattering by Rough Obstacles

arI sJ (∂r Φk )(x − y)νs (y) fJ (y)

1≤I ≤M

= !
bsL I νs (y)brL J (∂r Φk )(x − y)νs (y) fJ (y)
1≤I ≤M
 
! ν(y) Dx [Φk (x − y) f (y)]
= (−i)Sym D;
   
! ν(y) Sym D; (∇Φk )(x − y) f (y).
= − Sym D; (6.7.14)

In view of this and (6.4.4), the integral representation formula

 κ−n.t.   
u = D A u∂Ω − 𝒮 ∂νAu in Ω (6.7.15)

appearing in (6.6.8) may be re-fashioned as

∫
     κ−n.t. 
u(x) = Sym D;! ν(y) Sym D; (∇Φk )(x − y) u (y) dσ(y)
∂Ω
∂∗ Ω
∫
  κ−n.t. 
+i ! ν(y) (Du)
Φk (x − y)Sym D; (y) dσ(y), ∀x ∈ Ω.
∂Ω
∂∗ Ω
(6.7.16)

Collectively, Theorem 6.6.1 and (6.7.13) then prove the following result:
Theorem 6.7.1 Pick k ∈ (0, ∞) and choose n ∈ N with n ≥ 2. Also, fix M, N ∈ N
arbitrary and consider a first-order constant complex constant coefficient M × N
! along with a first-order constant complex constant coefficient N × M system
system D
!
D with the property that the vector Laplacian in Rn may be factored as Δ = DD.
Then the integral representation formula (6.7.16) holds whenever
Ω ⊆ Rn is an exterior domain with a lower Ahlfors regular boundary
(6.7.17)
and such that σ := H n−1 ∂Ω is a doubling measure on ∂Ω,

and the function u is a C M -valued null-solution in Ω of the vector Helmholtz operator

Δ + k 2 which radiates at infinity (in the sense of Definition 6.1.1), and satisfies
κ−n.t. κ−n.t.
the traces u∂Ω , (Du)∂Ω exist σ-a.e. on ∂nta Ω,
(6.7.18)
ρ ρ
and Nκ u, Nκ (Du) ∈ L 1 (∂Ω, σ) for some κ, ρ > 0.

Example 6.7.2 Start with Δ = ∂r ∂r acting on scalar functions u, which we interpret

! with
as the factorization Δ = DD
∂1
".%
D := ## .. && (6.7.19)
$∂n '
and
! := −D = (∂1, . . . , ∂n ).
D (6.7.20)
6.7 Coordinate-Free Formalism and Examples 911
 
Since now we have (−i)Sym D; ! x (Du)(x) = 
x · (∇u)(x) for every scalar function
u, the corresponding radiation condition (RC A) (originally introduced as in (6.3.2)
and subsequently equivalently
 rephrased
 in (6.7.11)) in this scenario simply becomes
iku(x) −  x · (∇u)(x) = o |x| −(n−1)/2 as |x| → ∞, which is precisely the pointwise
version of Sommerfeld’s radiation condition.
Also, the integral representation formula (6.7.16) becomes precisely (6.1.18).

Example 6.7.3 Start with Δ = −δd − dδ acting on differential forms (see the discus-
sion in [112, §6.4]; cf. [112, (6.4.145)] in particular), where d is the exterior derivative
operator, and δ its formal transpose (cf. [112, (6.4.140)]). We may interpret this as
the factorization Δ = DD ! with
 
d
D := (6.7.21)
δ
(which is a first-order system with constant real coefficients) and
! := −D = (−δ, −d).
D (6.7.22)
 
In this case, (−i)Sym D;! 
x (Du)(x) = x ∨ (du)(x) − x ∧ (δu)(x) for each differential
form u, with ∧ and ∨ denoting the exterior and interior product, respectively, of
differential forms (again, see [112, §6.4]). As such, the corresponding radiation
condition (RC A) (originally introduced in (6.3.2) and then equivalently rephrased in
(6.7.11)) presently takes the form
 
iku(x) −  x ∧ (δu)(x) = o |x| −(n−1)/2 as |x| → ∞.
x ∨ (du)(x) +  (6.7.23)

In view of (6.7.21) and (6.7.5), Theorem 6.6.1 ensures that for any differential form
u which is a null-solution of the Helmholtz operator (6.7.23) is equivalent to the
demand that u radiates at infinity (in the sense of Definition 6.1.1). In fact, according
to (6.6.4)-(6.6.6), if u is a differential form which is a null-solution of the Helmholtz
operator then any of the conditions
 
iku(x) − 
x ∨ (du)(x) +  x ∧ (δu)(x) = O |x| −(n+1)/2 as |x| → ∞, (6.7.24)

∫  2
 
iku(x) − 
x ∨ (du)(x) + 
x ∧ (δu)(x) dH n−1 (x) = o(1) as R → ∞, (6.7.25)
|x |=R
∫  2
 
x ∧(δu)(x) dH n−1 (x) = O(R−2 ) as R → ∞, (6.7.26)
x ∨(du)(x)+
iku(x)−
|x |=R

is also equivalent to the demand that u radiates at infinity in the sense of Defini-
tion 6.1.1.
In addition, the integral representation formula (6.7.16) presently becomes
912 6 Scattering by Rough Obstacles
∫
 κ−n.t. 
u(x) = − ν(y) ∨ (∇Φk )(x − y) ∧ u∂Ω (y) dσ(y)
∂∗ Ω
∫
 κ−n.t. 
− ν(y) ∧ (∇Φk )(x − y) ∨ u∂Ω (y) dσ(y)
∂∗ Ω
∫
 κ−n.t.   κ−n.t. 
+ Φk (x − y) ν ∧ (δu)∂Ω (y) − ν ∨ (du)∂Ω (y) dσ(y) (6.7.27)
∂∗ Ω

which is valid at each point x ∈ Ω under the assumption that Ω is as in (6.7.17) and u
is a differential form in Ω which is a null-solution of the Helmholtz operator Δ + k 2 ,
radiates at infinity (in the sense of Definition 6.1.1 or, equivalently, in the sense of
any of the conditions in (6.7.23), (6.7.24)-(6.7.26)), and satisfies
κ−n.t. κ−n.t. κ−n.t.
the traces u∂Ω , (du)∂Ω (δu)∂Ω exist σ-a.e. on ∂nta Ω,
(6.7.28)
ρ ρ ρ
and Nκ u, Nκ (du), Nκ (δu) ∈ L 1 (∂Ω, σ) for some κ, ρ > 0.

Example 6.7.4 Suppose E and H are differential forms satisfying the higher-
dimensional version of the Maxwell system for differential forms in an exterior
domain Ω ⊆ Rn . For a given wave number k ∈ (0, ∞), this reads (cf. [68])

δE − ikH = 0 and dH + ikE = 0 in Ω. (6.7.29)

Since d2 = 0 and δ = 0 (cf. [112, (6.4.145)]), this forces
2

dE = 0 and δH = 0 in Ω, (6.7.30)

as well as

dδE = ikdH = k 2 E and δdH = −ikδE = k 2 H in Ω. (6.7.31)

As a consequence of (6.7.30), (6.7.32), and [112, (6.4.145)], we have

(Δ + k 2 )E = −δdE − dδE + k 2 E = 0 and

(6.7.32)
(Δ + k 2 )H = −δdH − dδH + k 2 H = 0 in Ω.

Hence, E and H may be regarded as differential forms which, individually, are

null-solutions of the Helmholtz operator Δ + k 2 in Ω. From (6.7.29), (6.7.32), and
Proposition 6.3.4 we then see that
E radiates at infinity if and only H radiates
(6.7.33)
at infinity (in the sense of Definition 6.1.1).
Granted (6.7.29)-(6.7.30), the radiation condition (6.7.23) written for the choice
u := E presently becomes equivalent to
 
E(x) + x ∧ H(x) = o |x| −(n−1)/2 as |x| → ∞, (6.7.34)
6.7 Coordinate-Free Formalism and Examples 913

whereas the conditions in (6.7.24)-(6.7.26) presently become equivalent to, respec-

tively,  
E(x) + x ∧ H(x) = O |x| −(n+1)/2 as |x| → ∞, (6.7.35)
∫
 2
E(x) + 
x ∧ H(x) dH n−1 (x) = o(1) as R → ∞, (6.7.36)
|x |=R
∫
 2
E(x) + 
x ∧ H(x) dH n−1 (x) = O(R−2 ) as R → ∞. (6.7.37)
|x |=R

From the discussion in Example 6.7.3 above we then conclude that

any of the conditions (6.7.34)-(6.7.37) is equivalent to the demand
(6.7.38)
that E radiates at infinity (in the sense of Definition 6.1.1).
Moreover, the integral representation formula (6.7.27) written for u := E presently
becomes
∫
 κ−n.t. 
E(x) = − ν(y) ∨ (∇Φk )(x − y) ∧ E ∂Ω (y) dσ(y)
∂∗ Ω
∫
 κ−n.t. 
− ν(y) ∧ (∇Φk )(x − y) ∨ E ∂Ω (y) dσ(y)
∂∗ Ω
∫
 κ−n.t. 
+ ik Φk (x − y) ν ∧ H ∂Ω (y) dσ(y), (6.7.39)
∂∗ Ω

which is valid at each point x ∈ Ω under the assumption that Ω is as in (6.7.17) and
the differential forms E, H solve the Maxwell system (6.7.29), radiate at infinity (in
the sense of Definition 6.1.1 or, equivalently, in the sense of any of the conditions in
(6.7.34)-(6.7.37)), and satisfy
κ−n.t. κ−n.t.
the traces E ∂Ω , H ∂Ω exist σ-a.e. on ∂nta Ω,
(6.7.40)
ρ ρ
and Nκ E, Nκ H ∈ L 1 (∂Ω, σ) for some κ, ρ > 0.

Likewise, thanks to (6.7.29)-(6.7.30) the radiation condition (6.7.23) written for

u := H presently becomes equivalent to
 
H(x) + x ∨ E(x) = o |x| −(n−1)/2 as |x| → ∞, (6.7.41)

while the conditions in (6.7.24)-(6.7.26) presently become equivalent to, respectively,

 
H(x) +  x ∨ E(x) = O |x| −(n+1)/2 as |x| → ∞, (6.7.42)

∫
 2
H(x) + 
x ∨ E(x) dH n−1 (x) = o(1) as R → ∞, (6.7.43)
|x |=R
914 6 Scattering by Rough Obstacles
∫
 2
H(x) + 
x ∨ E(x) dH n−1 (x) = O(R−2 ) as R → ∞. (6.7.44)
|x |=R

Then the discussion in Example 6.7.3 above permits us to conclude that

any of the conditions (6.7.41)-(6.7.44) is equivalent to the demand
(6.7.45)
that H radiates at infinity (in the sense of Definition 6.1.1).
Furthermore, the integral representation formula (6.7.27) written for u:=H presently
becomes
∫
 κ−n.t. 
H(x) = − ν(y) ∨ (∇Φk )(x − y) ∧ H ∂Ω (y) dσ(y)
∂∗ Ω
∫
 κ−n.t. 
− ν(y) ∧ (∇Φk )(x − y) ∨ H ∂Ω (y) dσ(y)
∂∗ Ω
∫
 κ−n.t. 
+ ik Φk (x − y) ν ∧ E ∂Ω (y) dσ(y) ∀x ∈ Ω, (6.7.46)
∂∗ Ω

under the same assumptions on Ω, E, H as in the case of (6.7.39).

Finally, we wish to note that (6.7.34) and (6.7.41) considered collectively, i.e.,
  
E(x) + x ∧ H(x) = o |x| −(n−1)/2 as |x| → ∞,
  (6.7.47)
H(x) + x ∨ E(x) = o |x| −(n−1)/2 as |x| → ∞,

constitute a higher-dimensional generalization of the classical three-dimensional

Sylver-Müler radiation condition from (6.2.8)-(6.2.9), and that the integral represen-
tation formulas in (6.7.39) and (6.7.46) are higher dimensional generalizations of
the classical Stratton-Chu integral representation formulas for solutions the standard
Maxwell system in the three-dimensional setting (cf. [172]).

Example 6.7.5 Denote by


n
D := e j  ∂j (6.7.48)
j=1

the standard homogeneous Dirac operator in Rn , and observe that [112, (6.4.52)]
entails
D = D. (6.7.49)
Consider the Laplacian acting componentwise on Clifford algebra-valued functions.
Then (6.7.49) and [112, (6.4.56)] imply that Δ = −DD. We shall interpret this
factorization as
! with D
Δ = DD ! := −D and D := D. (6.7.50)
 
In this case, (−i)Sym D; ! 
x (Du)(x) = − x  (Du)(x) hence the corresponding
radiation condition (RC A) (originally introduced in (6.3.2) and then equivalently
rephrased in (6.7.11)) presently has the form
6.7 Coordinate-Free Formalism and Examples 915
 
x  (Du)(x) = o |x| −(n−1)/2 as |x| → ∞.
iku(x) +  (6.7.51)

Note that (6.7.49) and the choices made in (6.7.50) imply that we have
! = −D,
D (6.7.52)

which is relevant in the context of (6.7.5). With this in mind, Theorem 6.6.1 then
guarantees that, for any Clifford algebra-valued function u which is a null-solution
of the Helmholtz operator, the radiation condition recorded in (6.7.51) is equivalent
to the demand that u radiates at infinity (in the sense of Definition 6.1.1). In fact,
according to (6.6.4)-(6.6.6), if u is a Clifford algebra-valued function which is a
null-solution of the Helmholtz operator then any of the conditions
 
iku(x) + x  (Du)(x) = O |x| −(n+1)/2 as |x| → ∞, (6.7.53)

∫
 2
iku(x) + 
x  (Du)(x) dH n−1 (x) = o(1) as R → ∞, (6.7.54)
|x |=R
∫
 2
iku(x) + 
x  (Du)(x) dH n−1 (x) = O(R−2 ) as R → ∞, (6.7.55)
|x |=R

is also equivalent to the demand that u radiates at infinity in the sense of Defini-
tion 6.1.1.
Finally, the integral representation formula (6.7.16) currently takes the form
∫
 κ−n.t. 
u(x) = (DΦk )(x − y)  ν(y)  u∂Ω (y) dσ(y)
∂∗ Ω
∫
 κ−n.t. 
+ Φk (x − y)  ν(y)  (Du)∂Ω (y) dσ(y), ∀x ∈ Ω, (6.7.56)
∂∗ Ω

assuming that Ω is as in (6.7.17) and that u is a Clifford algebra-valued function in

Ω which is a null-solution of the Helmholtz operator Δ + k 2 in Ω, radiates at infinity
(in the sense of Definition 6.1.1 or, equivalently, in the sense of any of the conditions
in (6.7.51), (6.7.53)-(6.7.55)), and satisfies
κ−n.t. κ−n.t.
the traces u∂Ω , (Du)∂Ω exist σ-a.e. on ∂nta Ω,
(6.7.57)
ρ ρ
and Nκ u, Nκ (Du) ∈ L 1 (∂Ω, σ) for some κ, ρ > 0.

Example 6.7.6 Recall that a k-monogenic function in an open set Ω ⊆ Rn is a

smooth Clifford algebra-valued function u in Ω satisfying Du + ken+1  u = 0 in Ω
(where D is the standard homogeneous Dirac operator in Rn ; cf. (6.7.48)). Since
 2
D + ken+1 = −(Δ + k 2 ) (6.7.58)
it follows that any k-monogenic function u is a null-solution of Δ + k 2 with the
additional property that Du = −ken+1  u. In such a case, (6.7.51) reduces precisely
916 6 Scattering by Rough Obstacles

to the McIntosh-Mitrea radiation condition for k-monogenic functions, namely

 
x  en+1 )  u(x) = o |x| −(n−1)/2 as |x| → ∞,
(i −  (6.7.59)

whereas (6.7.53)-(6.7.55) become

 
x  en+1 )  u(x) = O |x| −(n+1)/2 as |x| → ∞,
(i −  (6.7.60)

∫
 2
x  en+1 )  u(x) dH n−1 (x) = o(1) as R → ∞,
(i −  (6.7.61)
|x |=R
∫
 2
(i − 
x  en+1 )  u(x) dH n−1 (x) = O(R−2 ) as R → ∞. (6.7.62)
|x |=R

Each of these conditions is equivalent to the demand that the k-monogenic function
u radiates at infinity in the sense of Definition 6.1.1.
Moreover, for a k-monogenic function u in Ω which radiates at infinity (in the
sense of Definition 6.1.1 or, equivalently, in the sense of any of the conditions in
(6.7.59)-(6.7.62)) and satisfies
κ−n.t.
the nontangential trace u∂Ω exists σ-a.e. on ∂nta Ω
(6.7.63)
ρ
and Nκ u ∈ L 1 (∂Ω, σ) for some κ, ρ > 0,

the integral representation formula (6.7.56) further reduces to

∫
   κ−n.t. 
u(x) = (DΦk )(x − y) + kΦk (x − y)en+1  ν(y)  u∂Ω (y) dσ(y)
∂∗ Ω
(6.7.64)

at each x ∈ Ω.
We continue by discussing a general integral representation result for radiating
null-solutions of the Helmholtz operator in an exterior domain, whose specific format
depends on the manner in which the Helmholtz operator is factored as the product
of two first-order systems.

Theorem 6.7.7 Let Ω ⊆ Rn , where n ∈ N satisfies n ≥ 2, be an exterior domain

with a lower Ahlfors regular boundary, and such that σ := H n−1 ∂Ω is a doubling
measure on ∂Ω. Denote by ν the geometric measure theoretic outward unit normal
to Ω, and fix an aperture parameter κ ∈ (0, ∞) along with a truncation parameter
ρ ∈ (0, ∞).
Next, fix a wave number k ∈ (0, ∞) and select N, M ∈ N arbitrary. Assume the
vector Helmholtz operator Δ + k 2 , regarded as a second-order M × M system in Rn ,
factors as
!
Δ + k 2 = DD (6.7.65)
6.7 Coordinate-Free Formalism and Examples 917

for some first-order M × N system D ! in Rn and some N × M system D in Rn

(not necessarily homogeneous; see [112, (1.7.12)]) both with constant complex
coefficients. Finally, pick a vector-valued function u satisfying
 M
u ∈ 𝒞∞ (Ω) , (Δ + k 2 )u = 0 in Ω,
u radiates at infinity (cf. Definition 6.1.1),
ρ ρ (6.7.66)
Nκ u and Nκ (Du) belong to L 1 (∂Ω, σ),
κ−n.t. κ−n.t.
u∂Ω and (Du)∂Ω exist σ-a.e. on ∂nta Ω.
κ −n.t. κ −n.t.
Then for any κ > 0 the nontangential traces u∂Ω and (Du)∂Ω also exist at σ-
a.e. point on ∂nta Ω and are actually independent of κ . Moreover, with the dependence
on the aperture parameter dropped, the following integral representation formula
holds:
∫
     n.t. 
u(x) = Sym D; ! ν(y) Sym D; (∇Φk )(x − y) u (y) dσ(y)
∂Ω
∂∗ Ω
∫
  n.t. 
+i ! ν(y) (Du) (y) dσ(y),
Φk (x − y)Sym D; ∀x ∈ Ω.
∂Ω
∂∗ Ω
(6.7.67)

Before presenting the proof of this result we wish to note that the integral rep-
resentation formula (6.7.67) may be though of a generalization of (6.7.16). Indeed,
given a first-order M × N system D ! along with a first-order N × M system D having
constant complex coefficients and satisfying (6.7.1), it follows that3
 
  D
! := D,
D ! k I M×M and D := (6.7.68)
k I M×M

are constant complex coefficient first-order (M × (N + M) and (N + M) × M,

respectively) systems satisfying
 
  D
! !
DD = D, k I M×M ! + k 2 = Δ2 + k 2 .
= DD (6.7.69)
k I M×M

! and D as in (6.7.68). Since for this

Hence, the factorization (6.7.65) is valid for D
choice in (6.7.68) we also have that (6.7.18) implies the properties in the last two
lines of (6.7.66), formula (6.7.67) holds and (6.7.16) readily follows from it.
Let us now turn to the proof of Theorem 6.7.7.
Proof of Theorem 6.7.7 The fact that, for any other aperture parameter κ > 0, the
κ −n.t. κ −n.t.
nontangential traces u∂Ω and (Du)∂Ω exist at σ-a.e. point on ∂nta Ω and are actu-
ally independent of κ is seen from the present hypotheses and [112, Corollary 8.9.9].
3 recall that I M ×M denotes the M × M identity matrix
918 6 Scattering by Rough Obstacles

To proceed, with I M×M denoting the M × M identity matrix, observe that if we define
the first-order (M × (N + M) and (N + M) × M, respectively) systems
 
  D
! := D,
D ! ik I M×M and D := (6.7.70)
ik I M×M

then
 
  D
! = D, ik I M×M
DD = D D − k 2 = (Δ2 + k 2 ) − k 2 = Δ. (6.7.71)
ik I M×M

Also, for each ξ ∈ Cn ,

 
 
Sym( D; ! ξ), 0 and Sym(D; ξ) = Sym(D; ξ) .
! ξ) = Sym(D; (6.7.72)
0

Moreover, since  
Du
Du = , (6.7.73)
iku
the last two lines in (6.7.66) guarantee that
ρ ρ
Nκ u and Nκ (Du) belong to L 1 (∂Ω, σ),
κ−n.t. κ−n.t. (6.7.74)
u∂Ω and (Du)∂Ω exist σ-a.e. on ∂nta Ω.

Then from (6.7.74), (6.7.72), and (6.7.73) we see that at σ-a.e. point y ∈ ∂∗ Ω we
have
  κ−n.t.    κ−n.t. 
Sym D;! ν(y) (Du) ! ν(y) (Du)
(y) = Sym D; (y), (6.7.75)
∂Ω ∂Ω

and
     κ−n.t. 
! ν(y) Sym D; (∇Φk )(x − y) u
Sym D; (y)
∂Ω
     κ−n.t. 
! ν(y) Sym D; (∇Φk )(x − y) u
= Sym D; (y). (6.7.76)
∂Ω

Granted (6.7.71), (6.7.74), (6.7.75), (6.7.76), we may invoke Theorem 6.7.1 and
conclude from (6.7.16) that the integral representation formula claimed in (6.7.67)
holds. 

It is of interest to single out the following special case of Theorem 6.7.7:

Corollary 6.7.8 Pick a wave number k ∈ (0, ∞), a dimension n ∈ N with n ≥ 2, and
select N, M ∈ N arbitrary. Let Ω ⊆ Rn be an exterior domain with a lower Ahlfors
regular boundary, and such that σ := H n−1 ∂Ω is a doubling measure. Denote by
ν the geometric measure theoretic outward unit normal to Ω, and fix an aperture
parameter κ ∈ (0, ∞) along with a truncation parameter ρ ∈ (0, ∞). Next, assume
the vector Helmholtz operator Δ + k 2 , regarded as a second-order M × M system in
6.7 Coordinate-Free Formalism and Examples 919

Rn , factors as
!
Δ + k 2 = DD (6.7.77)
for some first-order M × N system D ! in Rn and some N × M system D in Rn
(not necessarily homogeneous; see [112, (1.7.12)]) both with constant complex
coefficients. Finally, consider a vector-valued function u satisfying
 M
u ∈ 𝒞∞ (Ω) , Du = 0 in Ω,
u radiates at infinity (cf. Definition 6.1.1),
ρ (6.7.78)
Nκ u belongs to the space L 1 (∂Ω, σ),
κ−n.t.
and the trace u∂Ω exists σ-a.e. on ∂nta Ω.
κ −n.t.
Then for any other aperture parameter κ > 0 the nontangential trace u∂Ω
also exists at σ-a.e. point on ∂nta Ω and, in fact, is independent of κ . Furthermore,
with the dependence on the aperture parameter dropped, the following integral
representation formula holds at each point x ∈ Ω:
∫
     n.t. 
u(x) = Sym D; ! ν(y) Sym D; (∇Φk )(x − y) u (y) dσ(y). (6.7.79)
∂Ω
∂∗ Ω

Proof Since having Du = 0 in Ω entails

!
(Δ + k 2 )u = DDu = 0 in Ω, (6.7.80)

this is a direct consequence of Theorem 6.7.7. 

To offer an example, with D as in (6.7.48) abbreviate

Dk := D + ken+1 . (6.7.81)

In this piece of notation, (6.7.58) becomes

D2k = −(Δ + k 2 ). (6.7.82)

We may regard this as the factorization (6.7.77) written for

! := −Dk and D := Dk .
D (6.7.83)

Then Corollary 6.7.8 gives that any k-monogenic function u in Ω (i.e., any Clifford
algebra-valued null-solution of Dk in Ω) which radiates at infinity (in the sense of
Definition 6.1.1 or, equivalently, in the sense of any of the conditions in (6.7.59)-
κ−n.t.
(6.7.62)) and with the property that the nontangential trace u∂Ω exists σ-a.e. on
ρ
∂nta Ω and Nκ u ∈ L 1 (∂Ω, σ) (for some κ, ρ > 0), may be represented as in (6.7.79)
! D as in (6.7.83). A cursory inspection shows that said integral representation
for D,
formula is precisely the one recorded in (6.7.64).
920 6 Scattering by Rough Obstacles

In closing we wish to remark that the analysis in this section has brought to
prominence a number of singular integral operators naturally associated with the
Helmholtz equation, such as (6.4.2). We shall take a closer look at some of these in
[115, §1.5].
Appendix A
Terms and notation used in Volume III

A
Ahlfors regular domain (cf. [112, Definition 5.9.15]):

a nonempty open subset Ω of Rn such that ∂Ω is

(A.0.1)
an Ahlfors regular set and H n−1 (∂Ω \ ∂∗ Ω) = 0

Aκ (∂Ω), the κ-accessible (from within Ω) subset of ∂Ω (cf. [112, (8.8.2)]):

 
Aκ (∂Ω) := x ∈ ∂Ω : x ∈ Γκ (x) (A.0.2)

[w] A p , the characteristic of the weight w on a space of homogeneous type (X, ρ, μ)

(cf. [112, §7.7]):
⨏ ⨏  p−1
[w] A p := sup w dμ w −1/(p−1) dμ (A.0.3)
B ρ -ball B B

Ap (X, ρ, μ), the Muckenhoupt Ap -class on a space of homogeneous type (X, ρ, μ)

(cf. [112, §7.7]):
 
Ap (X, ρ, μ) := w weight function : [w] A p < ∞ (A.0.4)

A∞ (X, ρ, μ), the Muckenhoupt A∞ -class on a space of homogeneous type (X, ρ, μ)

(cf. [112, §7.7]): 
A∞ (X, ρ, μ) := Ap (X, μ) (A.0.5)
1≤p<∞

V ⊥, the annihilator V⊥ of a subspace V of a Banach space X:

 
V ⊥ := Λ ∈ X ∗ : Λ(x) = 0 for all x ∈ V (A.0.6)
⊥ W, the annihilator of a subspace W of X ∗ (where X is Banach):

© The Editor(s) (if applicable) and The Author(s), under exclusive license to 921
Springer Nature Switzerland AG 2023
D. Mitrea et al., Geometric Harmonic Analysis II, Developments in Mathematics 74,
https://doi.org/10.1007/978-3-031-22735-6
922 A Terms and notation used in Volume III

⊥
 
W := x ∈ X : Λ(x) = 0 for all Λ ∈ W (A.0.7)

T ∗ , the adjoint of the operator T ∈ L(X → Y ):

T ∗ : Y ∗ −→ X ∗, T ∗ (Λ) := Λ ◦ T for each Λ ∈ Y ∗ (A.0.8)

A p , the absolutely p-convex hull of a subset A of a vector space X, with p ∈ (0, 1],
defined as (cf. [112, §7.8.3]):

M
A p := λ j v j : M ∈ N, {v j }1≤ j ≤M ⊆ A,
j=1


M
{λ j }1≤ j ≤M ⊆ C with |λ j | p ≤ 1 (A.0.9)
j=1

a ⊗ b, the tensor product of vectors a = (a j )1≤ j ≤ N ∈ C N and b = (bk )1≤k ≤M ∈ C M ,

defined as the N × M matrix defined as in (1.3.7):

a ⊗ b := a j bk 1≤ j ≤ N
1≤k ≤M

αβ
A := ar s 1≤r,s ≤n , a coefficient tensor (of type (n × n, M × N) with complex number
1≤α ≤M
1≤β ≤ N
entries (1.7.1)
βα αβ
A , the transpose of a coefficient tensor A defined as A sr := ar s for all α, β, r, s
αβ
if A = ar s (1.7.3)
αβ αβ
A∇u := ar s ∂s uβ 1≤α ≤M , the action of the linear mapping A = ar s 1≤r,s ≤n on the
1≤r ≤n 1≤α ≤M
1≤β ≤ N
n×N N
Jacobian matrix ∇u ∈ D (Ω) of the distribution u = (uβ )1≤β ≤ N ∈ D (Ω)
(1.7.5)
 
A·, · , the bilinear form associated with the coefficient tensor A as in (1.7.6):
  αβ β β
Aζ, η := ar s ζs ηrα for all ζ := (ζs )β,s ∈ C N ×n and η := (ηrα )α,r ∈ C M×n

A q,κ , the L q -based area-function, defined as in (2.4.31):

∫
1/q
(A q,κ u)(x) := c
|(∇u)(y)| q |x − y| q−n dy
ΓκΣ (x)

1,1 n
for x ∈ Σ if u ∈ Wloc (R \ Σ)
p,q
As (Ω), the Besov/Triebel-Lizorkin space in the open set Ω ⊆ Rn (with A := B
corresponding to Besov spaces, and with A := F corresponding to Triebel-Lizorkin
spaces), with 0 < p, q ≤ ∞ and s ∈ R, defined as (cf. [113, §9.2]):
A Terms and notation used in Volume III 923
  
As (Ω) := u ∈ D (Ω) : there exists U ∈ As (Rn ) such that U Ω = u (A.0.10)
p,q p,q

and equipped with the quasi-norm

 
(Rn ), U Ω = u
p,q
u Ap, q
s (Ω)
:= inf U  Ap, q n : U ∈ As
s (R )
(A.0.11)


As (Ω)bdd , the space of all distributions u in Ω with the property that ψ Ω u belongs
p,q

to As (Ω) for each cutoff function ψ ∈ 𝒞∞

p,q
c (R ) (cf. [113, Convention 8.3.7] and
n

(A.0.132)).
B
Bρ (x, r), the ρ-ball with center at x ∈ X and radius r > 0 in the quasi-metric space
(X, ρ) (cf. [112, §7.1]):

Bρ (x, r) := {y ∈ X : ρ(x, y) < r } (A.0.12)

 
Bn−1 (x , r) := y  ∈ Rn−1 : |y  − x  | < r , the (n − 1)-dimensional (open) ball in
Rn−1 centered at x  ∈ Rn−1 and of radius r ∈ (0, ∞)
BMO1 , the BMO-based Sobolev spaces of order one (locally integrable functions
with distributional first-order partial derivatives in BMO)
 f ∗ (Δ), the local BMO norm of the function f on the surface ball Δ (cf. [112,
§7.4]): ⨏
 f ∗ (Δ) := sup | f − fΔ | dμ (A.0.13)
Δ ⊆Δ Δ

.
 f BMO(X,μ) , the homogeneous BMO semi-norm of the function f in the context of
a space of homogeneous type (X, ρ, μ) (cf. [112, §7.4]):
⨏
 
.
 f BMO(X,μ) := sup  f − fB (x,r)  dμ (A.0.14)
ρ
x ∈X, r >0 B ρ (x,r)

 · BMO(X,μ) , the inhomogeneous BMO “norm” in the context of a space of homo-

geneous type (X, ρ, μ) (cf. [112, §7.4]):
⎧
⎪ .
 f BMO(X,μ) if X is unbounded
⎪
⎨
⎪
 f BMO(X,μ) :=  ∫  (A.0.15)
⎪  
⎪ .
⎪  f BMO(X,μ) + f dμ if X is bounded
⎩ X

BMO X, μ , the space of functions of bounded mean oscillations for a space of

homogeneous type (X, ρ, μ) (cf. [112, §7.4]):

BMO X, μ := f ∈ Lloc 1
(X, μ) :  f BMO(X,μ) < +∞ (A.0.16)
924 A Terms and notation used in Volume III


BMO(X, μ), the space BMO modulo constants for a space of homogeneous type
(X, ρ, μ) (cf. [112, (7.4.96)]):
  

BMO(X, μ) := BMO(X, μ) ∼ = [ f ] : f ∈ BMO(X, μ) (A.0.17)

∂E, the topological boundary of the set E

Borelτ (X), the Borelians of the topological space (X, τ)
[A; B] := [A, B] := AB − BA, the commutator of A and B
{ A; B} := AB + BA, the anti-commutator of A and B
Bd(X → Y ), the space of linear and bounded operators from the quasi-normed
vector space X,  · X into the quasi-normed vector space Y,  · Y (cf. [113,
§1.2]):
 
Bd(X → Y ) := T : X → Y : T linear mapping with T X→Y < +∞ (A.0.18)

Bd(X), the space of linear and bounded operators from the quasi-normed vector
space X into itself:

Bd(X) := Bd(X → X) (A.0.19)

B(X → Y ), the space of linear and bounded operators from X to Y , where X, Y are
two linear topological spaces (cf. [113, §1.1]):
 
B(X → Y ) := T : X −→ Y : T linear and bounded (A.0.20)

B q,λ (Σ, σ), the block space on the Ahlfors regular set Σ ⊆ Rn , defined for q ∈ (1, ∞)
and λ ∈ (0, n − 1) as (cf. [113, §6.2]):

B q,λ (Σ, σ) (A.0.21)



:= f ∈ Lipc (Σ) : there exist a numerical sequence {λ j } j ∈N ∈  1 (N)
and a family {b j } j ∈N of B q,λ -blocks on Σ so that
∞

f = λ j b j with convergence in Lipc (Σ)
j=1

and equipped with the norm

 f  B q, λ (Σ,σ) (A.0.22)

∞ 
∞

:= inf |λ j | : f = λ j b j in Lipc (Σ) with
j=1 j=1

{λ j } j ∈N ∈  1 (N) and each b j a B q,λ -block on Σ

A Terms and notation used in Volume III 925
p,q
Bs (Σ, σ), the inhomogeneous Besov space on the Ahlfors regular set Σ ⊆ Rn ,
defined (cf. [113, §7.1]) for
 
s ∈ (−1, 1), n , n+s < p ≤ ∞,
max n−1 n−1
0 < q ≤ ∞,
    
max (s)+ , −s + (n − 1) p1 − 1 < β < 1, (n − 1) p1 − 1 < γ < 1,
+ +
(A.0.23)

as the collection of all “distributions” f on Σ such that, if {Ek }k ∈Z, k ≥κΣ is the family
of conditional expectation operators on the Ahlfors regular set Σ, then
  N
(
κΣ,τ)   p  1/p
f p, q
B s (Σ,σ) := σ(QτκΣ,ν ) mQ κΣ ,ν |EκΣ f |
τ
τ ∈IκΣ ν=1

  1/q
 q
+ 2 Ek f  L p (Σ,σ)
ks
<∞ (A.0.24)
k ∈Z
k ≥
κΣ +1

p,q
Bs (Rn ), the Besov space in Rn equipped with the quasi-norm  · Bsp, q (Rn ) for
0 < p, q ≤ ∞ and s ∈ R (cf. [113, §9.1])
p,q
Bs (Ω), the Besov space in the open set Ω ⊆ Rn with 0 < p, q ≤ ∞ and s ∈ R,
defined as the collection of u ∈ D (Ω) for which there exists U ∈ Bs (Rn ) such
p,q

that U Ω = u, and equipped with the quasi-norm (cf. [113, (9.2.1)])
 
uBsp, q (Ω) := inf U Bsp, q (Rn ) : U ∈ Bs (Rn ), U Ω = u
p,q

C
U, the closure of the set U ⊆ Rn
𝒞k (Ω), the space of functions of class 𝒞k in an open neighborhood of Ω
𝒞kc (Ω), the space of functions of class 𝒞k with compact support in the open set Ω
𝒞kb (Ω), the space of bounded functions of class 𝒞k in Ω
 ∗
𝒞∞ ∞
b (Ω) , the algebraic dual of 𝒞b (Ω)
CBM(Ω), the space of complex Borel measures in the open set Ω ⊆ Rn
CBM(X, τ), the space of complex Borel measures in the topological space (X, τ)
Cρ , the triangle inequality “penalty” constant associated with the quasi-distance ρ
as follows (cf. [112, (7.1.3)]):
ρ(x, y)
Cρ := sup ∈ [1, ∞) (A.0.25)
x,y,z ∈X max{ρ(x, z), ρ(z, y)}
not all equal
926 A Terms and notation used in Volume III

ρ , the symmetry “penalty” constant associated with the quasi-distance ρ as follows

C
(cf. [112, (7.1.4)]):
ρ := sup ρ(y, x) ∈ [1, ∞)
C (A.0.26)
x,y ∈X ρ(x, y)
xy

 · 𝒞. α (U,ρ) , the homogeneous Hölder space semi-norm of order α > 0 in the set
U ⊆ X, in the context of a quasi-metric space (X, ρ), defined for each function
f : U → R as (cf. [112, (7.3.2)]):

| f (x) − f (y)|
 f 𝒞. α (U,ρ) := sup (A.0.27)
x,y ∈U ρ(x, y)α
xy

.
𝒞α (U, ρ), the homogeneous Hölder space of order α > 0 in the set U ⊆ X, defined
in the context of a quasi-metric space (X, ρ) as (cf. [112, (7.3.1)]):
. 
𝒞α (U, ρ) := f : U → R :  f 𝒞. α (U,ρ) < +∞ (A.0.28)

.
𝒞α (U, ρ)/∼, the homogeneous Hölder space of order α > 0 modulo constants, in
the set U ⊆ X, defined in the context of a quasi-metric space (X, ρ) as (cf. [112,
(7.3.6)]): .  . 
𝒞α (U, ρ)/∼ := [ f ] : f ∈ 𝒞α (U, ρ) (A.0.29)
. α (U, ρ), the local homogeneous Hölder space of order α > 0 in the set U ⊆ X,
𝒞loc
defined in the context of a quasi-metric space (X, ρ) as (cf. [112, (7.3.7)]):
.α   .
𝒞loc (U, ρ) := f : U → C : f Bρ (x,r)∩U ∈ 𝒞α Bρ (x, r) ∩ U, ρ

for each x ∈ U and r ∈ (0, ∞) (A.0.30)

 · 𝒞α (U,ρ) , the inhomogeneous Hölder space norm of order α > 0 in the set U ⊆ X,
in the context of a quasi-metric space (X, ρ), defined for each function f : U → R
as (cf. [112, (7.3.20)]):

 f 𝒞α (U,ρ) := sup | f | +  f 𝒞. α (U,ρ), ∀ f ∈ 𝒞α (U, ρ) (A.0.31)

U

𝒞α (U, ρ), the inhomogeneous Hölder space of order α > 0 in the set U ⊆ X, defined
in the context of a quasi-metric space (X, ρ) as (cf. [112, (7.3.19)]):
 . 
𝒞α (U, ρ) := f ∈ 𝒞α (U, ρ) : f is bounded in U (A.0.32)

𝒞αc (U, ρ), the space of Hölder functions of order α > 0 with ρ-bounded support in
the set U ⊆ X, defined in the context of a quasi-metric space (X, ρ) as (cf. [112,
A Terms and notation used in Volume III 927

(7.3.26), (7.3.27)]):
 . 
𝒞αc (U, ρ) := f ∈ 𝒞α (U, ρ) : f vanishes outside of a ρ-bounded subset of U
(A.0.33)
Cp(X → Y ), the space of compact linear operators from the topological vector space
X into the topological vector space Y :
 
Cp(X → Y ) := T : X → Y : T linear compact mapping (A.0.34)

Cp(X), the space of compact linear operators from the topological vector space X
into itself:

Cp(X) := Cp(X → X) (A.0.35)

cap (K),∫ the capacity of a compact set K ⊂ Rn defined as the infimum of all
values |∇ϕ| 2 dL n corresponding to ϕ ∈ 𝒞∞
c (R ) satisfying ϕ ≥ 1 on K (cf.
n
Rn
Definition 5.2.5)
Cn , the Clifford algebra (Cn, +, ) generated by n imaginary units defined as the
minimal enlargement of Rn to a unitary real algebra which is not generated (as
an algebra) by any proper subspace of Rn , and such that x  x = −|x| 2 for each
x ∈ Rn → Cn (cf. [112, §6.4])
C(= CL ), the (left-handed) boundary-to-domain Cauchy-Clifford integral operator
whose action on each f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
⊗ Cn at each x ∈ Ω̊ is given by (2.5.282)
∫
1 x−y
C f (x) :=  ν(y)  f (y) dσ∗ (y)
ωn−1 |x − y| n
∂∗ Ω

C(= C L ), the (left-handed) boundary-to-boundary Cauchy-Clifford integral operator

whose action on each f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
⊗ Cn at σ∗ -a.e. point x ∈ ∂∗ Ω is given
by (2.5.284)
∫
1 x−y
C f (x) := lim+  ν(y)  f (y) dσ∗ (y)
ε→0 ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε

Cmax , the maximal Cauchy-Clifford integral operator whose action on each function
f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
⊗ Cn at each x ∈ ∂Ω is given by (2.5.283)
 1 ∫ 
 x−y 
Cmax f (x) := sup   ν(y)  f (y) dσ∗ (y) 
ε>0 ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε
928 A Terms and notation used in Volume III

CR , the right-handed boundary-to-domain Cauchy-Clifford integral operator whose

action on each f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
⊗ Cn at each x ∈ Ω̊ is given by (2.5.285)
∫
1 x−y
CR f (x) := f (y)  ν(y)  dσ∗ (y)
ωn−1 |x − y| n
∂∗ Ω

C R , the right-handed boundary-to-boundary Cauchy-Clifford integral operator

whose action on f ∈ L 1 ∂∗ Ω, 1+σ|x∗ (x)
| n−1
⊗ Cn at σ∗ -a.e. point x ∈ ∂∗ Ω is given
by (2.5.286)
∫
1 x−y
C R f (x) := lim+ f (y)  ν(y)  dσ∗ (y)
ε→0 ωn−1 |x − y| n
y ∈∂∗ Ω
|x−y |>ε

C # , the transpose Cauchy-Clifford integral operator whose action on each function

f ∈ L 1 ∂Ω, 1+σ(y)
|y | n−1
⊗ Cn at each point σ-a.e. x ∈ ∂∗ Ω is given by
∫
−1 x−y
C # f (x) := lim+ ν(x)  |x−y | n  f (y) dσ(y) (A.0.36)
ε→0 ω n−1
y ∈∂Ω
|x−y |>ε

D
u · w = u, w , the dot product of two vectors u, w ∈ Rn
 the divergence of the vector field F
divF,
divg , the differential geometric divergence (associated with the metric tensor g; cf.
[112, §1.11])
dg (x, y), the geodesic distance (induced by the metric tensor g) between x and y (cf.
[112, §1.11])
D (Ω), the space of distributions in the open set Ω
D  (Ω) ·, · D(Ω) , the distributional pairing in Ω
Δ := ∂12 + · · · + ∂n2 , the Laplace operator in Rn
δx , the Dirac distribution with mass at x
D, the classical (homogeneous) Dirac operator in Rn defined as (cf. [112, (6.4.139)]):

n
D= e j  ∂j (A.0.37)
j=1
A Terms and notation used in Volume III 929
 
αβ
D= n
j=1 a j ∂j + bαβ 1≤α ≤ N , a (generic) N × N  first-order system
1≤β ≤ N 
D , the (real) transpose of the first-order system D:
 n 
αβ
D := − a j ∂j + bαβ 1≤β ≤ N  (A.0.38)
j=1 1≤α ≤ N

D, the complex conjugate of the first-order system D

D∗ , the Hermitian adjoint of the first-order system D
d, the exterior derivative operator acting on the differential form u = J uJ dx J
according to (cf. [112, (1.11.32)], and also [112, (6.4.140)-(6.4.141)] for the Clifford
algebra context):
 n 
∂uJ
du = dx j ∧ dx J (A.0.39)
j=1 J
∂ x j

δ, the formal adjoint of the exterior derivative operator d on differential forms (see
also [112, (6.4.142)] for the Clifford algebra context)
δ jk , the Kronecker symbol, i.e., δ jk := 1 if j = k and δ jk := 0 if j  k
δF (·), the distance function to the set F
δ∂Ω (·), the distance function to the boundary of Ω
UV := (U \ V) ∪ (V \ U), the symmetric difference of the sets U and V
Δ(x, r) := B(x, r) ∩ ∂Ω, the surface ball on ∂Ω with center at x ∈ ∂Ω and radius
r>0
Δ(x, r) := B(x, r) ∩ Σ, the surface ball on Σ with center at x ∈ Σ and radius r > 0
D L , the Dirac operator acting from the left on each Clifford algebra-valued distribu-
tion u ∈ D (Ω) ⊗ Cn according to (cf. [112, (6.4.48)]):

n
D L u := e j  (∂j u) (A.0.40)
j=1

DR , the Dirac operator acting from the right on each Clifford algebra-valued distri-
bution u ∈ D (Ω) ⊗ Cn according to (cf. [112, (6.4.49)]):

n
DR u := (∂j u)  e j (A.0.41)
j=1

diamρ (A), the ρ-diameter of the set A ⊆ X, in the context of a quasi-metric space
(X, ρ), defined as (cf. [112, (7.1.6)]):

diamρ (A) := sup {ρ(x, y) : x, y ∈ A} (A.0.42)

930 A Terms and notation used in Volume III

Dk (X), the k-th generation of dyadic cubes in the geometrically doubling quasi-
metric space X, defined as in [112, Proposition 7.5.4]:

Dk (X) := {Q αk }α∈Ik (A.0.43)

D(X), the dyadic grid on the geometrically doubling quasi-metric space X, defined
as in [112, Proposition 7.5.4]:

D(X) := Dk (X) (A.0.44)
k ∈Z, k ≥κ X

dim X, the dimension of a vector space X

dζ := iν(ζ) dσ(ζ), the complex arc-length (1.1.3)
Def u := 12 ∂j uk + ∂k u j 1≤ j,k ≤n , the deformation tensor acting on the vector-
distribution u = (u1, . . . , un ) (1.3.49)
Dk := D + ken+1 , the perturbed Dirac operator in Rn , where k ∈ (0, ∞) (6.2.11)
Δ + k 2 , the Helmholtz operator with wave number k
diag (X) := {(x, y) ∈ X × X : x = y}, the diagonal in the Cartesian product X × X
(2.2.2)
E
(ε, δ)-domain (cf. [112, Definition 5.11.8]):
a nonempty, open, proper subset Ω of Rn such that for any x, y ∈ Ω
with |x − y| < δ there exists γ : [0, 1] → Ω rectifiable curve satisfying
γ(0) = x, γ(1) = y, as well as the conditions length(γ) ≤ ε1 |x − y| (A.0.45)
and |z−x|x−y
| |z−y |
| ≤ ε1 dist(z, ∂Ω) for each z ∈ γ([0, 1])

e j := (δ jk )1≤k ≤n ∈ Rn , where δ jk is the Kronecker symbol, standard unit vector in

the j-th direction in Rn
{e j }1≤ j ≤n , the standard orthonormal basis in Rn
EΔ , the standard fundamental solution for the Laplacian in Rn (1.5.56)
EΔ2 , the standard fundamental solution for the bi-Laplacian (1.4.93)
E = E L = Eαβ 1≤α,β ≤M , the canonical fundamental solution of the weakly elliptic
M × M system L (1.4.18)
 (Ω), the space of distributions in Ω supported in the compact set K ⊂ Ω
ℰK
ℰ(Ω), the space of distributions compactly supported in the open set Ω ⊆ Rn
[x]X/Y := x + Y , the equivalence class of the vector x ∈ X in the quotient space X/Y
A Terms and notation used in Volume III 931

E p (X), the p-envelope (with p ∈ (0, 1]) of a quasi-normed spaceX whose dual
separates points, defined as the completion of X in the quasi-norm  ·  p (cf. [113,
Definition 7.8.4])
Ex∂Ω→Ω , the extension operator from ∂Ω to Ω (cf. [113, Theorem 8.4.1])
F
 ∞ , the contribution of the vector field F at infinity, defined for any family {φ R }R>0
[F]
(referred to as a system of auxiliary functions) consisting of smooth compactly
supported functions in Rn which are globally bounded and progressively equal to 1
on compact sets in a uniform fashion, as (cf. [112, (1.3.2)-(1.3.3)]):
∫
 ∞ := − lim
[F] ∇φ R · F dL n (A.0.46)
R→∞ Ω

fE∗ , the non-increasing rearrangement of f : E → R defines as (cf. [112, (6.2.2)]):


fE∗ (t) := inf λ ≥ 0 : μ {x ∈ E : | f (x)| > λ} ≤ t , ∀ t ∈ [0, ∞) (A.0.47)

Φ(X → Y ), the collection of Fredholm operators from the linear topological space
X into the linear topological space Y (cf. [113, Definition 2.2.1])
 
fγ (x) := sup  f , ψ , the Fefferman-Stein grand maximal function (at the point
ψ ∈ Tγ (x)

x ∈ Σ) of the “distribution” f ∈ Lipc (Σ) on the Ahlfors regular set Σ (cf. [113,
(4.1.6)])
p,q
Fs (Σ, σ), the inhomoageneous Triebel-Lizorkin space on the Ahlfors regular set
Σ ⊆ Rn , defined as the collection of all “distributions” f on Σ with the property that
 f Fsp, q (Σ,σ) < +∞ (cf. [113, Definition 7.1.2])
p,q
Fs (Rn ), the (inhomogeneous) Triebel-Lizorkin space in Rn for 0 < p, q ≤ ∞ and
s ∈ R, equipped with the quasi-norm  · Fsp, q (Rn ) (cf. [113, (9.1.4)-(9.1.5)])
  
Fs (Ω) := u ∈ D (Ω) : there exists U ∈ Fs (Rn ) such that U Ω = u , the
p,q p,q

Triebel-Lizorkin space in the (arbitrary) open set Ω ⊆ Rn for 0 < p, q ≤ ∞ and

s ∈ R, equipped with the quasi-norm (cf. [113, (9.2.1)])
 
uFsp, q (Ω) := inf U Fsp, q (Rn ) : U ∈ Fs (Rn ), U Ω = u
p,q

∫
φ!(ξ) := (ℱφ)(x) := e−i x,ξ φ(x) dx, the Fourier transform of φ in Rn (1.4.14)
Rn
Φk (·), the (unique) radiating fundamental solution of the Helmholtz operator Δ + k 2
in Rn , given by formula (6.1.7)
G
g= g jk dx j ⊗ dxk , the Riemannian metric tensor
1≤ j,k ≤n
932 A Terms and notation used in Volume III

∇u, the gradient (Jacobian matrix) of a C M -valued function u = (uα )1≤α ≤M defined
in an open subset of Rn , defined as:
⎡ ∂1 u1 · · · ∂n u1 ⎤⎥
⎢
⎢ .. ⎥
∇u := ∂j uα 1≤α ≤M = ⎢ ... ..
. . ⎥⎥ (A.0.48)
1≤ j ≤n ⎢
⎢ ∂1 u M · · · ∂n u M ⎥⎦
⎣

∇ , the gradient operator in Rn−1

Γκ (x) = ΓΩ,κ (x), the (κ-)nontangential approach region with vertex at x ∈ ∂Ω,
defined as (cf. [112, (8.1.2)]):
 
Γκ (x) = ΓΩ,κ (x) := y ∈ Ω : |x − y| < (1 + κ)δ∂Ω (y) , ∀x ∈ ∂Ω (A.0.49)

X, a Generalized Banach Function Space on the measure space (X, M, μ) equipped

with the norm  · X (cf. [113, Definition 5.1.4])
∞ (X, μ)  · X ∞
X̊ := Lcomp , the closure of Lcomp in the Generalized Banach Function
Space X,  · X (cf. [113, Definition 5.2.6])
∇tan f , the tangential gradient of the function f ∈ L1,loc
1 (∂ Ω, σ ) (cf. [113, §11.4]):
∗ ∗


n 
∇tan f := νk ∂τk j f at σ∗ -a.e. point on ∂∗ Ω (A.0.50)
1≤ j ≤n
k=1

∇tan
A , the tangential gradient of u = (u ) associated with the coefficient tensor
β β 
κ−n.t. 
A := ar s 1≤α,β ≤M , defined by ∇tan u := νr ar s ∇tan uβ ∂Ω s
αβ A αβ
where
1≤r,s ≤n 1≤α ≤M
∇tan is the tangential gradient (acting on scalar functions) along ∂Ω (1.7.59), (1.7.60)

GΩ (·, ·), the Green function for the Laplacian, where Ω is a bounded open set in Rn
Theorem 5.3.1
H
H n−1 , the (n − 1)-dimensional Hausdorff measure in Rn
Hgn−1 , the (n − 1)-dimensional Hausdorff measure induced by the metric tensor g
H s , the s-dimensional Hausdorff measure in Rn
H∗s , the s-dimensional Hausdorff outer measure in Rn
H p (Σ, σ), the Lebesgue-based Hardy space on the Ahlfors regular set Σ ⊆ Rn ,
defined for p ∈ n−1n , ∞ and γ ∈ (n − 1) p − 1 +, 1 as (cf. [113, Definition 4.2.1])
1



H p (Σ, σ) := f ∈ Lipc (Σ) : fγ ∈ L p (Σ, σ) (A.0.51)

and equipped with the quasi-norm

A Terms and notation used in Volume III 933
( (
 f  H p (Σ,σ) := ( fγ ( ,
L p (Σ,σ)
∀ f ∈ H p (Σ, σ) (A.0.52)

H p,q (Σ, σ), the Lorentz-based Hardy space on the Ahlfors regular set Σ ⊆ Rn ,
defined for p ∈ n−1 n , ∞ , p ∈ (0, ∞], and γ ∈ (n − 1) p − 1 +, 1 as (cf. [113,
1

Definition 4.2.3])


H p,q (Σ, σ) := f ∈ Lipc (Σ) : fγ ∈ L p,q (Σ, σ) (A.0.53)

and equipped with the quasi-norm

( (
 f  H p, q (Σ,σ) := ( fγ ( L p, q (Σ,σ), ∀ f ∈ H p,q (Σ, σ) (A.0.54)

p,q
Hfin (Σ, σ), the vector space of all finite linear combinations of (p, q)-atoms on
Σ equipped with the quasi-norm  f  H p, q (Σ,σ) defined as the infimum of all
 N  1/p N
fin

|λ j | p such that f = λ j a j for {λ j }1≤ j ≤ N ⊆ C and (p, q)-atoms

j=1 j=1
{a j }1≤ j ≤ N (cf. [113, (4.4.113)])
H, the L p -filtering operator, acting on each given distribution
 f ∈ H p (Σ, σ) with
n < p < ∞ according to (H f )(x) := lim+ (H (Σ,σ)) St (x, ·), f H (Σ,σ) at each
n−1 p ∗ p
t→0
x ∈ Σ, where {St (·, ·)}t are the integral kernels of a suitable approximation to the
identity (cf. [113, Theorem 4.9.1])
ℋq,λ (Σ, σ), the space defined for any given q ∈ (1, ∞) and λ ∈ (0, n − 1) as (cf. [113,
§6.1]):

ℋq,λ (Σ, σ) (A.0.55)



:= f ∈ Lipc (Σ) : there exist a numerical sequence {λ j } j ∈N ∈  1 (N)
and a family {a j } j ∈N of ℋq,λ -atoms on Σ so that
∞

f = λ j a j with convergence in Lipc (Σ) ,
j=1

and equipped with the norm

 f ℋq, λ (Σ,σ) (A.0.56)


∞ 
∞

:= inf |λ j | : f = λ j a j in Lipc (Σ) with
j=1 j=1

{λ j } j ∈N ∈  1 (N) and each a j a ℋq,λ -atom on Σ

H s (Ω), the L 2 -based fractional Sobolev space of order s ∈ R in an open set Ω ⊆ Rn

(cf. [113, §9.2]):
934 A Terms and notation used in Volume III

   
H s (Ω) := U Ω : U ∈ H s (Rn ) = U Ω : U ∈ Fs2,2 (Rn ) , ∀s ∈ R, (A.0.57)

and equipped with the norm

 
u H s (Ω) := inf U  H s (Rn ) : U ∈ H s (Rn ) such that u = U Ω
 
≈ inf U F 2,2 (Rn ) : U ∈ Fs2,2 (Rn ) such that u = U Ω (A.0.58)
s

p
Hκ (Ω), the Hardy space of harmonic functions u in the open set Ω ⊆ Rn satisfying
Nκ u ∈ L p (∂Ω, σ) (5.5.4)
Hλ(1) (·), the Hankel function of the first kind with index λ ∈ R
H p (Ω; D), the Hardy space in Ω associated with the first-order N × M system D,
M
defined as the collection of all u ∈ 𝒞∞ (Ω) satisfying Nκ u ∈ L p (∂Ω, σ) and
Du = 0 in Ω (which also vanish at infinity
( (when Ω is an exterior domain), equipped
with the quasi-norm u H p (Ω;D) := (Nκ u( L p (∂Ω,σ) (3.2.2), (3.2.3)
p
ℋ• (∂Ω; D) := {ν  • u : u ∈ H p (Ω; D)}, the “bullet” boundary Hardy space as-
sociated with the Dirac operator D := nj=1 e j  ∂j , equipped with the quasi-norm
inherited from H p (∂Ω, σ) ⊗ Cn (3.2.21)
I
·, · , the (real) inner product in C M defined for any vectors u = (uk )1≤k ≤M ∈ C M
and w = (wk )1≤k ≤M ∈ C M as:


M
u, w := u k wk (A.0.59)
k=1

√
i := −1 ∈ C, the complex imaginary unit
Ů, the interior of the set U ⊆ Rn
⨏ ∫
E
1
f dμ := μ(E) E
f dμ, the integral average of the function f on the set E ⊆ X,
in a measure space (X, μ)
fBρ (x,r) , the integral average of f over the ρ-ball Bρ (x, r), in the context of a space
of homogeneous type (X, ρ, μ), defined as (cf. [112, (7.4.9)]):
⨏ ∫
1
fBρ (x,r) := f dμ := f (y) dμ(y) (A.0.60)
B ρ (x,r) μ Bρ (x, r) Bρ (x,r)
⨏
fΔ := Δ
f dσ, the integral average of f in the “surface ball” Δ
IE,α , the fractional integral operator of order α on the set E contained in a metric
space (X, ρ) equipped with an upper d-dimensional Borel measure μ on (X, τρ ),
μ(x)
acting on functions f ∈ L 1 E, 1+ρ(x,x ) d−α
according to (cf. [112, (7.8.3)]):
0
A Terms and notation used in Volume III 935
∫
f (y)
IE,α f (x) := dμ(y) for μ-a.e. x ∈ E (A.0.61)
E ρ(x, y)d−α

Im (T : X → Y ) := {T x : x ∈ X }, the image (or range) of the operator T

ind T := dim (Ker T) − codim (Im T) the index of a (semi-)Fredholm operator T (cf.
[113, Definition 2.1.1])
ln hΦ (t) ln hΦ (t)
i(Φ) := sup ln t = lim+ ln t where hΦ (t) := sup Φ(st)
Φ(s) for t ∈ (0, ∞), the
0<t<1 t→0 s>0
lower dilation index of the Young function Φ (cf. [113, (5.3.14)])
ln hΦ (t) ln hΦ (t)
I(Φ) := inf ln t = lim ln t where hΦ (t) := sup Φ(st)
Φ(s) for t ∈ (0, ∞), the
1<t<∞ t→∞ s>0
upper dilation index of the Young function Φ (cf. [113, (5.3.15)])
K
KΔ , the boundary-to-boundary harmonic double layer potential, defined as (cf. [112,
(1.1.32)]):
∫
1 ν(y), y − x
KΔ f (x) := lim+ f (y) dH n−1 (y), x ∈ ∂Ω
ε→0 ωn−1 ∂Ω\B(x,ε) |x − y| n
(A.0.62)
KΔ# , the transpose harmonic double layer potential, defined as (cf. [112, (1.1.33)]):
∫
1 ν(x), x − y
KΔ f (x) := lim+
#
f (y) dH n−1 (y), x ∈ ∂Ω
ε→0 ωn−1 ∂Ω\B(x,ε) |x − y| n
(A.0.63)
Ker (T : X → Y ) := {x ∈ X : T x = 0} the kernel (or null-space) of the operator T

M
Ker L := u ∈ 𝒞∞ (Ω) : Lu = 0 in Ω the null-space of the M × M system L,
in an open set Ω
k x0 Poisson kernel for the Laplacian for the domain Ω, defined as the Radon-
Nikodym derivative of the harmonic measure ω x0 with pole at x0 ∈ Ω with respect
to the surface measure σ on ∂Ω (5.3.38)
L
local John condition, satisfied by an open set Ω ⊆ Rn (cf. [112, Definition 5.11.7]):

there exist θ ∈ (0, 1), Mo ∈ (1, ∞), and R ∈ 0, diam ∂Ω (the

latter required to be ∞ if ∂Ω is unbounded) such that for every
x ∈ ∂Ω and r ∈ (0, R∗ ) one can find a point xr ∈ B(x, r) ∩ Ω
with the property that B(xr , θr) ⊆ Ω and for each y ∈ Δ(x, r) it (A.0.64)
is possible to find a rectifiable path γy : [0, 1] → Ω whose length
is ≤ Mo · r, which satisfies γy (0) = y, γy (1) = xr , and such that
dist γy (t), ∂Ω > θ · |γy (t) − y| for every t ∈ (0, 1]

L n , the (n-dimensional) Lebesgue measure in Rn

∞ , the space of essentially bounded functions with compact support
Lcomp
936 A Terms and notation used in Volume III
√
dLgn := g dL n Lebesgue measure induced by the metric tensor g
Λ T M, the -th exterior power of the vector bundle on the manifold M
L 0 (X, μ), the space of measurable functions which are pointwise finite μ-a.e. on X
p
Lbdd (Ω, L n ), the space of functions which are p-th power integrable on bounded
subsets of Ω (cf. [112, (4.2.4)]):
p
Lbdd (Ω, L n ) be the collection of all L n -measurable functions defined
in Ω which are p-th power absolutely integrable with respect to the (A.0.65)
Lebesgue measure on each bounded L n -measurable subset of the set Ω.

 · Lip(X) , the natural semi-norm on Lip(X), defined in the context of a metric space
(𝒳, d) as (cf. [112, (3.7.1)]):

| f (x) − f (y)|
 f Lip(X) := sup (A.0.66)
x,y ∈X, xy d(x, y)

Lipc (X), the space of Lipschitz functions with bounded support in the (quasi-)metric
space X

Lipc (Σ) , the space distributions on a given set Σ ⊆ Rn , defined as (cf. [112,
(4.1.34)]):
the topological dual of Lipc (Σ), τ𝒟 (A.0.67)

(Lip c (Σ)) ·, · Lip c (Σ) , or simply ·, · , the distributional pairing on the set Σ
· L p, q (X,μ) , the Lorentz space quasi-norm, defined as (cf. [112, (6.2.14)]):

⎧ ∫ ∞ q dt  1/q
⎪
⎪ t 1/p fX∗ (t) if 0 < p, q < ∞,
⎪
⎪
⎨ 0
⎪ t
 f  L p, q (X,μ) := sup 1/p f ∗ (t) if 0 < p ≤ ∞, q = ∞, (A.0.68)
⎪
⎪ t>0 t
⎪
⎪
X
⎪f ∞ if p = ∞, 0 < q ≤ ∞
⎩ L (X,μ)

L p,q (X, μ), the Lorentz space on X with respect to the measure μ defined as (cf.
[112, (6.2.13)]):

L p,q (X, μ) := f : X → R μ-measurable :  f  L p, q (X,μ) < +∞ (A.0.69)

p,q
L (Ω, μ), the maximal Lorentz space with respect to the Borel measure μ in the
open set Ω ⊆ Rn , defined as (cf. [112, (6.6.41)]):
p,q  
L (Ω, μ) := u : Ω → C : u is L n -measurable and u,θ ∈ L p,q (Ω, μ) (A.0.70)

p
L (Ω, μ), the maximal Lebesgue space with respect to the Borel measure μ in the
open set Ω ⊆ Rn , defined as (cf. [112, (6.6.43)]):
A Terms and notation used in Volume III 937
p p, p
L (Ω, μ) := L (Ω, μ) (A.0.71)
 
= u : Ω → C : u is L n -measurable and u,θ ∈ L p (Ω, μ)

log+ , the positive part of ln, defined for each t ∈ [0, ∞) as (cf. [112, (7.6.68)]):

0 if t ∈ [0, 1],
log+ t := (A.0.72)
ln t if t ∈ [1, ∞)

L(X → Y ) the space of linear and continuous operators from X to Y

L Φ (X, μ), the Orlicz space on the sigma-finite measure space (X, M, μ) associated
with a Young function Φ (cf. [113, §5.3]):
 
L Φ (X, μ) := f ∈ ℳ(X, μ) :  f  L Φ (X,μ) < ∞ (A.0.73)

where the Luxemburg norm  ·  L Φ (X,μ) is defined as

 ∫
 f  L Φ (X,μ) := inf λ > 0 : Φ(| f (x)|/λ) dμ(x) ≤ 1 (A.0.74)
X
 ∫ 
∞ 
= inf λ > 0 : μ x ∈ X : | f (x)| > λΦ−1 (t) dt ≤ 1 (A.0.75)
0

 ∫ α
L p (log L)α (X, μ) = f ∈ ℳ(X, μ) : X | f (x)| p ln(e + | f (x)|) dμ(x) < +∞
for p ∈ (1, ∞) and α ∈ R, Zygmund’s space (cf. [113, §5.3])
L p (Ω, wL n ), the weighted L p Lebesgue space on the set Ω ⊆ Rn , equipped with
∫  1/p
the natural norm u L p (Ω, w L n ) := Ω |u| p w dL n (cf. [113, §8.3])
 
Ls (Ω) := U Ω : U ∈ Ls (Rn ) , the Bessel potential space in an open set Ω ⊆ Rn
p p

for p ∈ (1, ∞) and s ∈ R, equipped with the norm (cf. [113, §9.2]):
 
u p := inf U  p n : U ∈ Ls (Rn ), u = U 
p
L s (Ω) L s (R ) Ω

p p
L1,loc (∂∗ Ω, σ∗ ) = L1,loc (∂∗ Ω, σ), the local Sobolev space of order one on ∂∗ Ω,
defined for p ∈ [1, ∞] as (cf. [113, Definition 11.1.2]):

p p p
L1,loc (∂∗ Ω, σ∗ ) := f ∈ Lloc (∂∗ Ω, σ∗ ) : ∂τ j k f exists in Lloc (∂∗ Ω, σ∗ )

for each j, k ∈ {1, . . . , n} (A.0.76)

938 A Terms and notation used in Volume III
.p
L1 (∂Ω, σ), the L p -based homogeneous Sobolev space of order one on ∂Ω as (cf.
[113, Definition 11.5.3]):

.p  σ(x) 
L1 (∂Ω, σ) := f ∈ L 1 ∂Ω, : ∂τ j k f ∈ L p (∂∗ Ω, σ)
1 + |x| n

for each j, k ∈ {1, . . . , n} (A.0.77)

and equipped with the semi-norm

.p 
n
L1 (∂Ω, σ)  f →  f  L. p (∂Ω,σ) := ∂τ j k f  L p (∂∗ Ω,σ) (A.0.78)
1
j,k=1

p
L1 (∂∗ Ω, σ∗ ), the L p -based Sobolev space of order one on ∂∗ Ω, defined for each
p ∈ [1, ∞] as (cf. [113, Definition 11.1.2]):

p
L1 (∂∗ Ω, σ∗ ) := f ∈ L p (∂∗ Ω, σ∗ ) : ∂τ j k f exists in L p (∂∗ Ω, σ∗ )

for each j, k ∈ {1, . . . , n} (A.0.79)

and equipped with the norm (cf. [113, Proposition 11.1.9]):


p
L1 (∂∗ Ω, σ∗ )  f →  f  L p (∂∗ Ω,σ∗ ) :=  f  L p (∂∗ Ω,σ∗ ) + ∂τ j k f  L p (∂∗ Ω,σ∗ )
1
1≤ j,k ≤n
(A.0.80)
p
L−1 (∂∗ Ω, σ∗ ),
the negative Sobolev space of order minus one on ∂∗ Ω
(L p -based)
defined for p, p ∈ (1, ∞) with p1 + p1 = 1 as (cf. [113, Definition 11.8.1]):
  ∗
p p
L−1 (∂∗ Ω, σ∗ ) := L1 (∂∗ Ω, σ∗ ) (A.0.81)

αβ
L = ar s ∂r ∂s 1≤α,β ≤M homogeneous, second-order, constant (complex) coeffi-
cient, M × M system in Rn (1.3.1)
 
αβ
L A := ar s ∂r ∂s 1≤α ≤M second-order system in Rn canonically associated with a
1≤β ≤ N
αβ
given coefficient tensor A := ar s 1≤r,s ≤n (1.7.7)
1≤α ≤M
1≤β ≤ N
L(ξ), the characteristic matrix of a given homogeneous (constant-coefficient) higher-
order system L = ∂ α Aαβ ∂ β , defined as (cf. [112, (6.5.39)]):
|α |= |β |=m

L(ξ) := (−1)m ξ α+β Aαβ, ∀ξ ∈ Rn (A.0.82)
|α |= |β |=m
A Terms and notation used in Volume III 939
αβ
L(ξ) := − ar s ξr ξs 1≤α,β ≤M for all ξ = (ξr )1≤r ≤n ∈ Rn , the characteristic matrix
of the homogeneous, complex constant coefficient, second-order M × M system
αβ
L = ar s ∂r ∂s 1≤α,β ≤M in Rn (1.3.2)
Lλ,μ := μ Δ + (λ + μ)∇div the complex Lamé system, with Lamé moduli λ, μ ∈ C
(1.3.6)
Lλ,μ (ξ) := −μ|ξ | 2 In×n − (λ + μ)ξ ⊗ ξ for each ξ ∈ Rn , the characteristic matrix of
the complex Lamé system Lλ,μ := μ Δ + (λ + μ)∇div (1.3.8)
M
M X,s , the L s -based Hardy-Littlewood maximal operator in the space of homoge-
neous type (X, ρ, μ), defined for each μ-measurable function f on X as (cf. [112,
(7.6.7)]):
⨏  s1
M X,s f (x) := sup | f | s dμ , ∀ x ∈ X (A.0.83)
r >0 B ρ (x,r)

M X , the Hardy-Littlewood maximal operator of the function f on the space of

homogeneous type (X, ρ, μ), defined for each μ-measurable function f on X as (cf.
[112, (7.6.16)]):
∫
1
M X f (x) := sup | f | dμ, ∀ x ∈ X (A.0.84)
r ∈(0,∞) μ Bρ (x, r) B ρ (x,r)

M , the tangential maximal function of u (with exponent M), defined at each x ∈ ∂Ω

umax
as (cf. [112, (8.5.2)]):

M : ∂Ω −→ [0, +∞] defined by

umax
(  δ (y)  M ( (A.0.85)
( ∂Ω (
M (x) := (u(y)
umax ( ∞ for each x ∈ ∂Ω
|x − y| L y (Ω, L n )

ℳ(X, μ), the collection of μ-measurable functions defined on an arbitrary measure

space (X, M, μ)
ℳ+ (X, μ) the collection of non-negative μ-measurable functions defined on an
arbitrary measure space (X, M, μ)
M p,λ (Σ, σ), the Morrey space on the Ahlfors regular set Σ ⊆ Rn , defined for each
p ∈ (0, ∞) and λ ∈ (0, n − 1) as (cf. [113, §6.2]):

M p,λ (Σ, σ) := f : Σ → C : f is σ-measurable and  f  M p, λ (Σ,σ) < +∞
(A.0.86)
where, for each σ-measurable function f on Σ,
 
n−1−λ  ⨏  p1
 f  M p, λ (Σ,σ) := sup R p | f | dσ
p
(A.0.87)
x ∈Σ and Σ∩B(x,R)
0<R<2 diam(Σ)
940 A Terms and notation used in Volume III

M̊ p,λ (Σ, σ), the vanishing Morrey space on the Ahlfors regular set Σ ⊆ Rn with
p ∈ (0, ∞) and λ ∈ (0, n − 1) (cf. [113, §6.2]):
p(n−1)
M̊ p,λ (Σ, σ) := the closure of L s (Σ, σ) with s := n−1−λ in M p,λ (Σ, σ) (A.0.88)

ℳq,λ (Σ, σ), the (q, λ)-mid space on the Ahlfors regular set Σ ⊆ Rn , defined for
q ∈ (1, ∞) and λ ∈ (0, n − 1) as (cf. [113, §6.2]):

ℳq,λ (Σ, σ) (A.0.89)



:= f ∈ Lipc (Σ) : there exist a sequence {λ j } j ∈N ∈  1 (N) and a family
{m j } j ∈N of (q, λ)-dull molecules on Σ such that
∞

f = λ j m j with convergence in Lipc (Σ)
j=1

and equipped with the norm

 f ℳq, λ (Σ,σ) (A.0.90)

 ∞ 
∞

:= inf |λ j | : f = λ j m j in Lipc (Σ) with {λ j } j ∈N ∈  1 (N)
j=1 j=1

and each m j a (q, λ)-dull molecule on Σ

Mb , the operator of pointwise multiplication by the function b ∈ L ∞ (Σ, σ)

[Mb, T], the commutator of Mb (the pointwise multiplication by b) with the operator
T, defined by [Mb, T] f := b(T f ) − T(b f ) (2.7.4)
N
N0 := N ∪ {0} = {0, 1, 2, . . . }
NTA domain: a nonempty open subset Ω of Rn satisfying a two-sided corkscrew
condition as well as a Harnack chain condition (cf. [112, Definition 5.11.1])
two-sided NTA domain: a nonempty open subset Ω of Rn satisfying a two-sided
corkscrew condition as well as a two-sided Harnack chain condition (cf. [112, Defi-
nition 5.11.1])
one-sided NTA domain (or interior NTA domain): a nonempty open subset Ω of Rn
satisfying an interior corkscrew condition as well as a Harnack chain condition (cf.
[112, Definition 5.11.1])
Nκ u, the (κ-)nontangential maximal operator acting on the function measurable
u : Ω → Rn according to (cf. [112, (8.2.1)]):

Nκ u : ∂Ω −→ [0, +∞], (Nκ u)(x) := u L ∞ (Γκ (x), L n ) for all x ∈ ∂Ω (A.0.91)
A Terms and notation used in Volume III 941

Nκε u, the (κ-)nontangential maximal operator truncated at height ε > 0, acting on

the function u : Ω → Rn according to (cf. [112, (1.5.5)]):
 
Nκε u := Nκ (u · 1 Oε ) where Oε := x ∈ Ω : dist(x, ∂Ω) < ε . (A.0.92)

NκE u, the restricted nontangential maximal function of u : Ω → R, relative to the

set E (cf. [112, (8.2.4)]):

NκE u : ∂Ω −→ [0, +∞] defined as

(A.0.93)
(NκE u)(x) := u L ∞ (Γκ (x)∩E, L n ) for each x ∈ ∂Ω

ν, the geometric measure theoretic outward unit normal to a set Ω ⊆ Rn of locally

finite perimeter, defined via the requirement that (cf. [112, (5.6.2)-(5.6.3)]):
there exist a locally finite Borel-regular measure σ∗ in Rn and a
n
vector-valued function ν ∈ L ∞ (Rn, σ∗ ) satisfying |ν(x)| = 1 at (A.0.94)
n
σ∗ -a.e. x ∈ Rn and with such that ∇1Ω = −νσ∗ in D (Rn )

νg , the geometric measure theoretic outward unit normal induced by the metric
tensor g
ν E , the geometric measure theoretic outward unit normal induced by the standard
Euclidean metric
ν • F,  the “bullet” product involving a vector field F ∈ L 1 (Ω, L n ) n (where
bdd
Ω is an arbitrary open subset of Rn ) whose divergence, considered in the sense of
distributions in Ω, satisfies divF ∈ Lbdd
1 (Ω, L n ), defined as a functional acting on

each ψ ∈ Lipc (∂Ω) according to (cf. [112, Proposition 4.2.3]):

∫ ∫
 
 ψ :=
ν • F, F · ∇Ψ dL n + (divF)Ψ  dL n, (A.0.95)
Ω Ω

where Ψ is any complex-valued function satisfying


Ψ ∈ Lip(Ω), Ψ∂Ω = ψ, and Ψ ≡ 0 out-
(A.0.96)
side of some compact subset of Ω

ν
• u, the Clifford bullet product of ν with u (cf. [113, (10.2.100)]):

ν
• u := (−i)Sym(D; ν) • u (A.0.97)

p
Nκ (Ω; μ), the space of measurable functions in Ω with a p-th power integrable
non-tangential maximal function on ∂Ω with respect to the measure μ (cf. [112,
(8.3.31)]):
942 A Terms and notation used in Volume III
p 
Nκ (Ω; μ) := u : Ω → C : u is L n -measurable, and

u Nκp (Ω;μ) := Nκ u L p (∂Ω, μ) < +∞ (A.0.98)

κ−n.t.
u|∂Ω (x), the nontangential trace of the function u : Ω → R at the point x ∈ ∂Ω
such that x ∈ Γκ (x), defined as (cf. [112, Definition 8.9.1]):
κ−n.t.
u|∂Ω (x) is the number a ∈ R with the property that for every
ε > 0 there exists some r > 0 such that |u(y) − a| < ε for L n -a.e. (A.0.99)
point y ∈ Γκ (x) ∩ B(x, r)

T X→Y , the operator “norm” of a positively homogeneous mapping T acting from

the quasi-normed vector space X,  · X with values in the quasi-normed vector
space Y,  · Y
 
T X→Y := sup TuY : u ∈ X, uX = 1 ∈ [0, +∞] (A.0.100)

T Bd(X→Y) := T X→Y
T Bd(X) , the “norm” of a positively homogeneous mapping T acting from the
quasi-normed vector space X,  · X into itself

T Bd(X) := T Bd(X→X) = sup T xX (A.0.101)

x ∈X, x  X =1

ess
T X→Y , the essential norm of the operator T ∈ Bd(X → Y ), where X, Y are
quasi-normed spaces (cf. [113, §1.2]):
ess
T X→Y : = dist T, Cp(X → Y )
 
= inf T − K X→Y : K ∈ Cp(X → Y ) (A.0.102)

O
1E , the characteristic function of a given set E
ωn−1 := H n−1 (S n−1 ), the surface area of S n−1 (the (n − 1)-dimensional sphere in
Rn )
 
Oε := x ∈ Ω : δ∂Ω (x) < ε , the one-sided collar neighborhood of ∂Ω of “width”
ε>0
ω x , the harmonic measure on ∂Ω with pole at x
P
P, the P-maximal operator acting on a Lebesgue measurable function u : Ω → Rn
at the point x ∈ ∂Ω according to (cf. [113, §10.1]):
A Terms and notation used in Volume III 943
 ∫ 
1
(Pu)(x) := sup |u| dL n (A.0.103)
0<r <2 diam(∂Ω) σ ∂Ω ∩ B(x, r) Ω∩B(x,r)

 
·, · E , the pointwise (real) pairing in the fibers of the Hermitian vector bundle E
 
ℰ (Ω) ·, · ℰ(Ω) , the pairing between a compactly supported distribution u in Ω and a
smooth function f ∈ 𝒞∞ (supp u), say f ∈ 𝒞∞ (O) with O ⊆ Ω open set containing
supp u, defined for each F ∈ 𝒞∞ (Ω) with the property that F = f near supp u as
(cf. [112, (2.2.33)]):
   
ℰ (Ω) u, f ℰ(Ω) := ℰ (Ω) u, F ℰ(Ω) (A.0.104)

·, · Λ TM, the (real) pointwise pairing on Λ T M

∧, the exterior product of differential forms
∨, the interior product of differential forms
∂τXY , the tangential derivative operator on the boundary of a subset Ω (of locally
finite perimeter) of a Riemannian manifold (M, g), associated with two vector fields
p
X, Y ∈ T M, acting on any function f ∈ L1 (∂∗ Ω, σg ) ⊗ E according to (cf. [112,
Definition 1.12.7]):

∂τXY f := −hXY − νg (X) divgY − νg (Y ) divg X + νg ([X, Y ]) f (A.0.105)

at σg -a.e. point on ∂∗ Ω, where hXY ∈ L p (∂∗ Ω, σg ) ⊗ E is the unique function with

the property that for each section ϕ ∈ 𝒞1c (M, E) one has
∫ ∫
    
f , ∂τXY ϕ E dσg = hXY , ϕ∂∗ Ω E dσg (A.0.106)
∂∗ Ω ∂∗ Ω

.
∂τ j k u, the weak tangential derivative of a given function u ∈ Lloc
1 (Ω, L n ) with the
n
property that ∇u ∈ Lbdd 1 (Ω, L n ) , defined as (cf. [112, Example 4.2.4]):

.
∂τ j k u := ν • (∂k u)e j − (∂j u)ek , (A.0.107)

i.e., the “bullet” product ν • Fjk

u , where

Fjk
u
:= (∂k u)e j − (∂j u)ek (A.0.108)

∂∗ E, the measure theoretic boundary of a Lebesgue measurable set E ⊆ Rn , defined

as (cf. [112, (5.2.1)]):
944 A Terms and notation used in Volume III

 L n (B(x, r) ∩ E)
∂∗ E := x ∈ Rn : lim sup > 0 and
r→0+ rn
L n (B(x, r) \ E)
lim sup >0 (A.0.109)
r→0+ rn

∂ ∗ E, the reduced boundary of a set E ⊆ Rn of locally finite perimeter, defined as

(cf. [112, (5.6.13)]):
∂ ∗ E consists of all points x ∈ ∂E satisfying the following three
properties:
⨏ 0 < σ∗ (B(x, r)) < +∞ for each r ∈ (0, ∞), formula (A.0.110)
lim+ B(x,r) ν dσ∗ = ν(x) is valid, and |ν(x)| = 1.
r→0

∂nta Ω, the nontangentially accessible boundary of an open proper subset Ω of Rn ,

defined as (cf. [112, Definition 8.8.5]):
)  
∂nta Ω := Aκ (∂Ω) = x ∈ ∂Ω : x ∈ Γκ (x) for each κ > 0 (A.0.111)
κ>0

(a)+ := max{a, 0}, the positive part of the number a ∈ R

πκ (E) = πΩ,κ (E), the “shadow” (or projection) of a given set E ⊆ Ω onto ∂Ω,
defined as (cf. [112, (8.1.15)]):
 
πκ (E) = πΩ,κ (E) := x ∈ ∂Ω : Γκ (x) ∩ E  (A.0.112)

pX , the lower Boyd index of a rearrangement invariant Banach function space X on

a non-atomic sigma-finite measure space (X, M, μ)
∂νA(·, ·), the conormal derivative operator with respect to the coefficient tensor
A acting from weighted Sobolev spaces (cf. [113, Proposition 8.5.3]), and from
Besov/Triebel-Lizorkin spaces (cf. [113, Proposition 9.5.2])
.
∂νAu, the weak conormal derivative of u with respect to the coefficient tensor A
defined (cf. [113, Definition 10.2.18]) by first introducing
αβ
Fα := (A∇u)α = ar s ∂s uβ 1≤r ≤n for each α ∈ {1, . . . , M } (A.0.113)

and then setting

.  
 M
∂νAu := ν • Fα 1≤α ≤M ∈ Lipc (∂Ω) (A.0.114)

∂τ j k ϕ with ϕ ∈ 𝒞1 (O), the pointwise tangential derivative operator (cf. [113,

§11.1]):
 
∂τ j k ϕ := ν j ∂k ϕ  O∩∂∗ Ω − νk ∂j ϕ  O∩∂∗ Ω at σ∗ -a.e. point on O ∩ ∂∗ Ω (A.0.115)
A Terms and notation used in Volume III 945

∂τ j k , the tangential derivative operator defined in a weak, or distributional, sense (cf.

[113, (11.1.17) and Definition 11.1.2], and [113, Definition 11.2.1])
∂ := ∂z̄ := 12 (∂x + i∂y ), the Cauchy-Riemann operator (1.1.2)
∂ := ∂z := 12 (∂x − i∂y ), the (complex) conjugate of the Cauchy-Riemann operator
(1.1.2)
∂νA pointwise conormal derivative operator with respect to the coefficient tensor
αβ N
A := ar s 1≤α,β ≤M whose action on u = (uβ )1≤β ≤ N ∈ D (Ω) is defined as
 1≤r,s ≤n κ−n.t. 
∂νAu := νr ar s ∂s uβ ∂Ω
αβ
at H n−1 -a.e. point on ∂∗ Ω (1.7.9)
1≤α ≤M

∂νD,D ,
the conormal derivative operator associated with the factorization of the
 where D
second-order system L as DD,  and D are homogeneous, constant complex
 κ−n.t.
 ν)(Du)
coefficient, first-order systems in R , defined as ∂νD,D u := (−i)Sym( D;
n
∂Ω
N
for u ∈ D (Ω) (1.7.36)
 M n 
∂ξAu := arJsI ξr ∂s u I , the conormal derivative of u = (u I )I , associ-
I=1 r,s=1 1≤J ≤M
ated with the coefficient tensor A := arJsI 1≤J,I ≤M used in the writing of the vector
1≤r,s ≤n
Laplacian as Δ = divA∇ and along ξ = (ξr )r ∈ Cn (6.2.21)
pΩ , the Hölder conjugate exponent of qΩ (5.7.46)
 
𝒫δ,b (x, h) := y ∈ B(x, R) : |y − x|δ(|y − x|) < h · (y − x) < b ⊆ Rn , the pseudo-
ball in Rn with apex at x ∈ Rn , symmetry axis along h ∈ S n−1 , truncated at height
b ∈ (0, ∞), whose profile is modeled by some “shape function" δ : [0, R] → [0, ∞)
(5.8.20)
PI, the Poisson integral operator (5.2.6)

P.V. b k(x − ·)Σ , the principal-value distribution associated with the smooth odd
kernel k and the bounded function b on the countably rectifiable upper Ahlfors
regular set Σ ⊆ Rn , acting on each test function φ ∈ Lipc (Σ) as (cf. [113, Proposi-
tion 11.9.1]):
∫
  
P.V. b k(x − ·)Σ , φ := lim+ b(y)k(x − y)φ(y) dσ(y)
ε→0
y ∈Σ
|y−x |>ε

Q
X/Y , the quotient space of a vector space X and a linear subspace Y of X
qX , the upper Boyd index of a rearrangement invariant Banach function space X on
a non-atomic sigma-finite measure space (X, M, μ)
 
qΩ := sup q ∈ (1, ∞) : ω x0 ∈ Bq (σ) ∈ (1, ∞] where ω x0 is the harmonic
measure with pole at x0 ∈ Ω and Bq (σ) consisting of all locally finite Borel measures
946 A Terms and notation used in Volume III

λ on ∂Ω satisfying λ << σ and for which dλ/dσ belongs to the reverse Hölder class
RHq with respect to σ (5.7.45)
R
R+n , the (open) upper half-space in Rn
R−n , the (open) lower half-space in Rn
Res( f , z), the residue of the function f at the point z (1.1.77)
R j , the j-th boundary-to-boundary Riesz transform, with j ∈ {1, . . . , n}, acting on
f ∈ L 1 Σ, 1+ |σ· | n−1 at σ-a.e. x ∈ Σ as in (2.5.336):
∫
2 xj − yj
R j f (x) := lim+ f (y) dσ(y)
ε→0 ωn−1 |x − y| n
|x−y |>ε
y ∈Σ

R j , the j-th boundary-to-domain Riesz transform, with j ∈ {1, . . . , n}, acting on

each f ∈ L 1 Σ, 1+ |σ· | n−1 at every point x ∈ Rn \ Σ as in (2.5.335):
∫
2 xj − yj
R j f (x) := f (y) dσ(y)
ωn−1 |x − y| n
Σ

R jk , the boundary-to-domain integral operator, with j, k ∈ {1, . . . , n}, acting on

each f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
at any point x ∈ Ω as in (2.5.256):
∫
−1 ν j (y)(xk − yk ) − νk (y)(x j − y j )
R jk f (x) := f (y) dσ∗ (y)
ωn−1 |x − y| n
∂∗ Ω

R jk , the principal-value singular integral operator, with j, k ∈ {1, . . . , n}, acting on

each f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
at σ∗ -a.e. point x ∈ ∂∗ Ω as in (2.5.268):
∫  
R jk f (x) := lim+ ν j (y)(∂k EΔ )(y − x) − νk (y)(∂j EΔ )(y − x) f (y) dσ∗ (y)
ε→0
y ∈∂∗ Ω
|x−y |>ε

(R jk )max , the maximal operator, with j, k ∈ {1, . . . , n}, acting on each function
f ∈ L 1 ∂∗ Ω, 1+σ(x)
|x | n−1
at any point x ∈ ∂∗ Ω as in (2.5.270):

(R jk )max f (x)
 ∫ 
 
:= sup  ν j (y)(∂k EΔ )(y − x) − νk (y)(∂j EΔ )(y − x) f (y) dσ∗ (y)
ε>0
y ∈∂∗ Ω
|x−y |>ε
A Terms and notation used in Volume III 947

R C , the boundary-to-domain Clifford-Riesz transform, acting on any given Clifford

algebra-valued function f ∈ L 1 ∂Ω, 1+σ(x)
|x | n−1
⊗ Cn at each point x ∈ Ω as
∫
2 x−y
R C f (x) :=  f (y) dσ(y) (A.0.116)
ωn−1 ∂Ω |x − y| n

mod
R j , the j-th modified Riesz transform (with j ∈ {1, . . . , n − 1}) in the Euclidean

space Rn−1 , acting on any given function f ∈ L 1 Rn−1, 1+dx |x  | n at L
n−1 -a.e. point

x  ∈ Rn−1 as
∫ 
mod 2 x j −y j 
R j f (x) := lim+ |x  −y  | n 1R n−1 \B n−1 (x ,ε) (y ) (A.0.117)
ε→0 ωn−1
R n−1
−y j 
− |−y  | n 1R n−1 \B n−1 (0,1) (y ) f (y ) dy 

where, generally speaking, Bn−1 (z , r) is the (n − 1)-dimensional ball centered at

z  ∈ Rn−1 and of radius r
S
σ := H n−1 !∂Ω, the surface measure on ∂Ω
σ∗ := H n−1 !∂∗ Ω, the surface measure on ∂∗ Ω
σ := H n−1 !Σ, the surface measure on the closed Ahlfors regular set Σ ⊆ Rn
σg , the surface measure induced by the metric tensor g
σ E , the surface measure induced by the standard Euclidean metric
 n 
αβ
Sym(D; ξ), the principal symbol of the first-order system D = a j ∂j 1≤α ≤ N :
j=1 1≤β ≤M


n 
αβ
Sym(D; ξ) := i aj ξj 1≤α ≤ N (A.0.118)
j=1 1≤β ≤M

∗, the Hodge star operator

ξ → ξ  , X → X  , the musical isomorphisms (metric identifications via lowering
and rising indices), defined as (cf. [112, (1.12.139)]):

T M  X = X j ∂j −→ X  := g jk Xk dx j ∈ T ∗ M,
(A.0.119)
T ∗ M  ξ = ξ j dx j −→ ξ  := g jk ξ j ∂k ∈ T M

S n−1 := ∂B(0, 1), the unit sphere in Rn

𝒞k -singsup u, the smallest closed set outside of which the distribution u is of class
𝒞k
948 A Terms and notation used in Volume III

supp f , the support of the μ-measurable function f defined on a topological space

(X, τ) (where μ is a Borel measure measure on (X, τ)), defined as (cf. [112, (3.8.7)
in Definition 3.8.3]):
 ∫
supp f := x ∈ X : | f | dμ > 0 for each open set O ⊆ X with x ∈ O (A.0.120)
O

𝒮(Rn ), the space of (smooth, rapidly decreasing) Schwartz functions in Rn

𝒮(Rn ), the space of tempered distributions in Rn
u,θ , the solid maximal function of u : Ω → C, defined at each point x ∈ Ω
according to (cf. [112, (6.6.2)]):

u,θ (x) := u L ∞ (B(x,θ δ∂Ω (x)), L n ) ∈ [0, ∞] (A.0.121)

ρ# , the regularized version of the quasi-distance ρ for a quasi-metric space (X, ρ),
defined as (cf. [112, (7.1.17) in Theorem 7.1.2]):

ρ# := (ρsym )α for α := (log2 Cρ )−1 ∈ (0, ∞]. (A.0.122)

fp# , the L p -based Fefferman-Stein sharp maximal function of f ∈ Lloc 1 (X, μ), defined

as (cf. [112, (7.4.110)]):

⨏    1/p
fp (x) := sup
#  f (y) − fB (x,r)  p dμ(y) , ∀x ∈ X (A.0.123)
ρ
r >0 B ρ (x,r)

T
Tγ (x) the family of “bump” (i.e., localized, and normalized in the Hölder norm)
functions centered at x ∈ Σ (cf. [113, §4.1])
τρ , the topology induced by the quasi-distance ρ in a quasi-metric space (X, ρ),
defined as (cf. [112, (7.1.7)]):
def
O is open in τρ ⇐⇒ for any x ∈ O there is r > 0 with Bρ (x, r) ⊆ O (A.0.124)

TrRn →Σ , the trace operator from Rn to Σ whose action on each given function
ap
u ∈ W 1, p Rn, δΣ L n is the limit (cf. [113, Theorem 8.1.1]):
⨏
(TrR →Σ u)(x) := [u]R (x) := lim+
n n u(y) dy (A.0.125)
r→0 B(x,r)

TrΩ→∂Ω , the trace operator from Rn to ∂Ω whose action on each given function
1, p ap
u ∈ Wa (Ω) := W 1, p Ω, δ∂Ω L n is the limit (cf. [113, Theorem 8.3.6]):
⨏
(TrΩ→∂Ω u)(x) := [u]Ω (x) := lim+ u(y) dy (A.0.126)
r→0 B(x,r)∩Ω
A Terms and notation used in Volume III 949

TrΩ→∂Ω , the trace operator from the open set Ω ⊆ Rn to its boundary
 ∂Ω, acting on
each given u ∈ Aα (Ω) (and with w ∈ Aα (Rn ) such that w Ω = u) according to
p,q p,q

(cf. [113, Theorem 9.4.5]):

⨏
TrΩ→∂Ω u (x) := lim+ w dL n at σ-a.e. x ∈ ∂Ω (A.0.127)
r→0 B(x,r)

Tε , the truncated singular integral operator (at the threshold ε > 0), defined as in
(2.3.6): ∫
(Tε f )(x) := k(x − y) f (y) dσ(y)
y ∈Σ, |x−y |>ε

σ(x)
for each function f ∈ L 1 Σ, 1+ |x | n−1
and each point x ∈ Σ
 
Tmax , the maximal operator defined by (Tmax f )(x) := supε>0 (Tε f )(x) for all
f ∈ L 1 Σ, 1+σ(x)
|x | n−1
and all x ∈ Σ (2.3.5)
Tmod , the modified principal-value singular integral operator acting on each function
f ∈ L 1 Σ, 1+σ(x)
|x | n at σ-a.e. point in Σ as in (2.3.31):
∫
 
Tmod f := lim+ k ε (· − y) − k1 (−y) f (y) dσ(y)
ε→0
Σ

where k ε := k · 1Rn \B(0,ε) for each ε > 0

Tmod , the boundary-to-domain modified integral operator acting on each function
f ∈ L 1 Σ, 1+σ(x)
|x | n at every x ∈ Ω as in (2.5.29):
∫
 
(Tmod f )(x) := k(x − y) − k1 (−y) f (y) dσ(y)
∂Ω

where k1 := k · 1Rn \B(1,0)

U
UR set (cf. [112, Definition 5.10.1]): a closed set Σ ⊂ Rn which is (upper) Ahlfors
regular and has Big Pieces of Lipschitz Images (in a uniform, quantitative, scale-
invariant fashion)
UR domain (cf. [112, Definition 5.10.6]): a nonempty open subset Ω of Rn such that
∂Ω is a UR set and
H n−1 (∂Ω \ ∂∗ Ω) = 0 (A.0.128)

U $ V, the union of two disjoint sets U, V

UC(X, ρ), the space of uniformly continuous functions on the metric space (X, ρ)
[u]∞
A , the contribution at infinity of a null-solution u for the vector-Helmholtz oper-

ator, associated with the coefficient tensor A := arJsI 1≤J,I ≤M in the writing of the
1≤r,s ≤n
950 A Terms and notation used in Volume III

vector Laplacian as Δ = divA∇, defined at each x ∈ Rn as in (6.3.34):

∫ 
lim !
ys arI sJ (∂r Φk )(x − y)uJ (y)
R→∞ |y |=R
 
+ asr
IJ
Φk (x − y)(∂r uJ )(y) dH n−1 (x)
1≤I ≤M

where Φk is the radiating fundamental solution of the Helmholtz operator Δ + k 2

V
√
dVg := g dx1 ∧ · · · ∧ dxn , the volume element induced by the metric tensor g
VMO(X, μ), the Sarason space of functions of vanishing mean oscillations on the
measure metric space (X, ρ, μ) defined (with UC(X, ρ) denoting the space of uni-
formly continuous functions on (X, ρ)) as the space (cf. [113, §3.1]):

VMO(X, μ) := the closure of UC(X, ρ) ∩ BMO(X, μ) in BMO(X, μ) (A.0.129)

W
weakly elliptic system: a system L whose characteristic matrix satisfies the condition
det [L(ξ)]  0 for each ξ ∈ Rn \ {0}
weakly elliptic coefficient tensor: a coefficient tensor A with the property that the
canonically associated second-order system L A is weakly elliptic, i.e., we have
det [L A(ξ)]  0 for each ξ ∈ Rn \ {0}
W k, p (Ω), the L p -based Sobolev space of order k in Ω (intrinsically defined)
k, p
Wloc (Ω), the local L p -based Sobolev space of order k in Ω
k, p
Wbdd (Ω), the space of Sobolev functions on any bounded measurable subset of Ω
(cf. [112, (3.0.4)]):
k, p k, p
Wbdd (Ω) denote the space of functions u ∈ Wloc (Ω) with the property
that ∂ α u ∈ L p (O, L n ) for each α ∈ N0n with |α| ≤ k and each bounded (A.0.130)
Lebesgue measurable subset O of Ω.

k, p
Wa (Ω), the weighted Sobolev space in Ω defined as in [113, (8.3.5)] for the weight
ap
w := δ∂Ω , with k ∈ N0 , p ∈ (0, ∞), a ∈ R, and equipped with the quasi-norm (cf.
[113, Definition 8.3.4]):
 ∫  1/p
uW k, p (Ω) := |(∂ α u)(x)| p δ∂Ω (x)ap dx (A.0.131)
a
|α | ≤k Ω

X
∗
X ∗ ·, · X , the duality pairing between a vector space X and its algebraic dual X

X , the associated space (aka Köthe dual) of the Generalized Banach Function Space
X, equipped with the norm  · X (cf. [113, Definition 5.1.11])
A Terms and notation used in Volume III 951
      
 x  :=  ·  := inf λ > 0 : λ−1 x ∈ BX (0, 1) p for all x ∈ X, the Minkowski
p X, p
functional associated with the absolutely p-convex hull of the unit ball in X (cf. [113,
(7.8.6)])
𝒳bdd (Ω), or 𝒳(Ω)bdd , the space of distributions in the open set Ω ⊆ Rn defined as
(cf. [113, Convention 8.3.7]):
 
u ∈ D (Ω) : ψ Ω u ∈ 𝒳(Ω) for each ψ ∈ 𝒞∞ c (R )
n
(A.0.132)

x := x/|x| for each x ∈ Rn \ {0} (6.1.1)

!
References

1. M. Abramowitz and I.A. Stegun, Handbook of Mathematical Functions, Dover, New York,
1972.
2. S. Agmon, A. Douglis, and L. Nirenberg, Estimates near the boundary for solutions of elliptic
partial differential equations satisfying general boundary conditions, I, Comm. Pure Appl.
Math., 12 (1959), 623–727.
3. S. Agmon, A. Douglis, and L. Nirenberg, Estimates near the boundary for solutions of elliptic
partial differential equations satisfying general boundary conditions, II, Comm. Pure Appl.
Math., 17 (1964), 35–92.
4. L.V. Ahlfors, Remarks on the Neumann-Poincaré integral equation, Pacific J. Math., 2 (1952),
no. 3, 271–280.
5. R. Alvarado, D. Brigham, V. Maz’ya, M. Mitrea, and E. Ziadé, On the regularity of domains
satisfying a uniform hour-glass condition and a sharp version of the Hopf-Oleinik Boundary
Point Principle, Journal of Mathematical Sciences, 176 (2011), no. 3, 281–360.
6. R. Alvarado, I. Mitrea, and M. Mitrea, Whitney-type extensions in geometrically doubling
quasi-metric spaces, Communications in Pure and Applied Analysis, 12 (2013), no. 1, 59–88.
7. R. Alvarado and M. Mitrea, Hardy Spaces on Ahlfors-Regular Quasi-Metric Spaces. A Sharp
Theory, Lecture Notes in Mathematics, Vol. 2142, Springer, 2015.
8. P. Auscher and T. Hytönen, Orthonormal bases of regular wavelets in spaces of homogeneous
type, Applied and Computational Harmonic Analysis, 34 (2013), no. 2, 266–296.
9. J. Azzam, Semi-uniform domains and a characterization of the A∞ property for harmonic
measure, preprint, (2017). arXiv:1711.03088v2
10. J. Azzam, S. Hofmann, J.M. Martell, K. Nyström, and T. Toro, A new characterization of
chord-arc domains, J. Eur. Math. Soc., 19 (2017), no. 4, 967–981.
11. B. Bennewitz and J.L. Lewis, On weak reverse Hölder inequalities for nondoubling harmonic
measures, Complex Var. Theory Appl., 49 (2004), no. 7-9, 571–582.
12. C.A. Berenstein, An inverse spectral theorem and its relation to the Pompeiu problem, J. Anal.
Math., 37 (1980), 128–144.
13. H. Brezis and L. Nirenberg, Positive solutions of nonlinear elliptic equations involving critical
Sobolev exponents, Comm. Pure Appl. Math., 36 (1983), 437–477.
14. A.P. Calderón, Cauchy integrals on Lipschitz curves and related operators, Proc. Nat. Acad.
Sci. USA., 74 (1977), no. 4, 1324–1327.
15. A.P. Calderón, Commutators, singular integrals on Lipschitz curves and applications, pp. 85–
96 in “Proceedings of the International Congress of Mathematicians” (Helsinki, 1978), Acad.
Sci. Fennica, Helsinki, 1980.

© The Editor(s) (if applicable) and The Author(s), under exclusive license to 953
Springer Nature Switzerland AG 2023
D. Mitrea et al., Geometric Harmonic Analysis II, Developments in Mathematics 74,
https://doi.org/10.1007/978-3-031-22735-6
954 References

16. L. Capogna, C.E. Kenig, and L. Lanzani, Harmonic Measure: Geometric and Analytic Points
of View, University Lecture Series, Vol. 35, Amer. Math. Soc., Providence, Rhode Island,
2005.
17. A.L. Cauchy, Sur la mécanique céleste et sur un nouveau calcul appelé calcul des limites,
Turin, 1831.
18. M. Christ, A T (b) theorem with remarks on analytic capacity and the Cauchy integral, Colloq.
Math., LX/LXI (1990), 601–628.
19. R.R. Coifman and C. Fefferman, Weighted norm inequalities for maximal functions and sin-
gular integrals, Studia Math., 51 (1974), 241–250.
20. R. Coifman, A. McIntosh, and Y. Meyer, L’intégrale de Cauchy definit un opérateur borné
sur L 2 pour les courbes lipschitziennes, Ann. Math., 116 (1982), 361–388.
21. R.R. Coifman, R. Rochberg and G. Weiss, Factorization theorems for Hardy spaces in several
variables, Ann. of Math., 103 (1976), no. 3, 611–635.
22. R.R. Coifman and G. Weiss, Analyse Harmonique Non-Commutative sur Certains Espaces
Homogènes, Lecture Notes in Mathematics No. 242, Springer-Verlag, 1971.
23. R. Courant and D. Hilbert, Methods of Mathematical Physics, Vol. I-II, New York, Inter-
science, 1962.
24. D. Cruz-Uribe, J.M. Martell, and C. Pérez, Weights, extrapolation and the theory of Rubio de
Francia, Operator Theory: Advances and Applications, Vol. 215, Birkhäuser/Springer Basel
AG, Basel, 2011.
25. B.E.J. Dahlberg, Estimates of harmonic measure, Arch. Rational Mech. Anal., 65 (1977),
no. 3, 275–288.
26. B.E.J. Dahlberg, On the Poisson integral for Lipschitz and C 1 domains, Studia Math., 66
(1979), 13–24.
27. B.E.J. Dahlberg, D.S. Jerison and C.E. Kenig, Area integral estimates for elliptic differential
operators with non-smooth coefficients, Ark. Mat., 22 (1984), no. 1-2, 97–108.
28. B.E.J. Dahlberg and C.E. Kenig, Hardy spaces and the Neumann problem in L p for Laplace’s
equation in a Lipschitz domain, Ann. of Math., 125 (1987), 437–465.
29. B.E.J. Dahlberg, C.E. Kenig, and G.C. Verchota, Boundary value problems for the systems of
elastostatics in Lipschitz domains, Duke Math. J., 57 (1988), no. 3, 795–818.
30. G. David, Opérateurs intégraux singulière sur certaines courbes du plan complexe, Ann.
Scient. École Norm. Sup. (4) 17 (1984), 157–189.
31. G. David, Opérateurs d’intégrale singulière sur les surfaces régulières, Ann. Scient. Ecole
Norm. Sup., 21 (1988), 225–258.
32. G. David and D. Jerison, Lipschitz approximation to hypersurfaces, harmonic measure, and
singular integrals, Indiana Univ. Math. J., 39 (1990), no. 3, 831–845.
33. G. David and J.L. Journé, A boundedness criterion for generalized Calderón-Zygmund oper-
ators, Annals of Math., 120 (1984), 371–397.
34. G. David, J.L. Journé, and S. Semmes, Opérateurs de Calderón-Zygmund, fonctions para-
accrétives et interpolation, Rev. Mat. Iberoamericana, 1 (1985), 1–56.
35. M. Dindoš, D. Mitrea, I. Mitrea, and M. Mitrea, Green function estimates for elliptic systems
in the upper half-space, preprint, (2018).
36. J. Duoandikoetxea, Forty years of Muckenhoupt weights, pp. 23–75 in “Function Spaces and
Inequalities,” Lecture Notes Paseky nad Jizerou 2013, J. Lukes and L. Pick ed., Matfyzpress,
Praga, 2013.
37. X.T. Duong, S. Hofmann, D. Mitrea, M. Mitrea, and L. Yan, Hardy spaces and regularity for
the inhomogeneous Dirichlet and Neumann problems, Revista Matemática Iberoamericana,
29 (2013), no. 1, 183–236.
38. L. Escauriaza and M. Mitrea, Transmission problems and spectral theory for singular integral
operators on Lipschitz domains, J. Funct. Anal., 216 (2004), no. 1, 141–171.
39. L.C. Evans and R.F. Gariepy, Measure Theory and Fine Properties of Functions, Studies in
Advanced Mathematics, CRC Press, Boca Raton, FL, 1992.
40. E.B. Fabes, Layer potential methods for boundary value problems on Lipschitz domains,
pp. 55—80, Lecture Notes in Mathematics, Vol. 1344, Springer, Berlin-Heidelberg-New York,
1988.
References 955

41. E.B. Fabes, M. Jodeit Jr., and N.M. Rivière, Potential techniques for boundary value problems
on C 1 -domains, Acta Math., 141 (1978), no. 3–4, 165–186.
42. E. Fabes, C. Kenig, and G.C. Verchota, The Dirichlet problem for the Stokes system on Lips-
chitz domains, Duke Math. J., 57 (1988), no. 3, 769–793.
43. E. Fabes, O. Mendez, and M. Mitrea, Boundary layers on Sobolev-Besov spaces and Poisson’s
equation for the Laplacian in Lipschitz domains, J. Funct. Anal., 159 (1998), no. 2, 323–368.
44. H. Federer, Curvature measures, Trans. Amer. Math. Soc., 93 (1959), 418–491.
45. C. Fefferman and E.M. Stein, H p spaces of several variables, Acta Math., 129 (1972), no. 3-4,
137–193.
46. G.B. Folland, Real Analysis, Modern Techniques and Their Applications, 2-nd edition, John
Wiley and Sons, Inc., New York, 1999.
47. L. Friedlander, Some inequalities between Dirichlet and Neumann eigenvalues, Arch. Ration.
Mech. Anal., 116 (1991), 153–160.
48. J. García-Cuerva and J.L. Rubio de Francia, Weighted Norm Inequalities, North Holland,
Mathematics Studies Vol. 116, 1985.
49. J. Garnett and D.E. Marshall, Harmonic Measure, Cambridge University Press, 2005.
50. F. Gesztesy and M. Mitrea, Generalized Robin boundary conditions, Robin-to-Dirichlet maps,
and Krein-type resolvent formulas for Schrödinger operators on bounded Lipschitz domains,
pp. 105–173 in the Proceedings of Symposia in Pure Mathematics, Vol. 79, Amer. Math. Soc.,
2008.
51. D. Gilbarg and N. Trudinger, Elliptic Partial Differential Equations of Second Order, 2nd
edition, Springer-Verlag, Berlin, Heidelberg, 1983.
52. I. Gohberg and N. Krupnik, One-Dimensional Linear Singular Integral Equations, Vol. I-II,
Operator Theory: Advances and Applications Vol. 53-54, Birkhäuser Verlag Basel, 1992.
53. G. Green, An Essay on the Application of Mathematical Analysis to the Theories of Electricity
and Magnetism, Nottingham, England, T. Wheelhouse, 1828.
54. P. Grisvard, Elliptic problems in nonsmooth domains, Monographs and Studies in Mathemat-
ics, Vol. 24 Pitman Boston, MA, 1985.
55. M. Grüter and K.-O., Widman, The Green function for uniformly elliptic equations,
Manuscripta Math., 37 (1982), no. 3, 303–342.
56. N.M. Günter, Potential Theory and its Applications to Basic Problems of Mathematical
Physics, F. Ungar, 1967.
57. A. Harnack, Beiträge zur Theorie des Cauchy’schen Integral, Berichte d. k. Sächs. Ges. d.
Wiss., Math.-Phys. Classe, 37 (1885), 379–398.
58. L.L. Helms, Potential Theory, 2nd edition, Universitext, Springer, London, 2014.
59. K. Hoffman, Banach Spaces of Analytic Functions, Dover Publications Inc., New York, 1962.
60. S. Hofmann and S. Kim, The Green function estimates for strongly elliptic systems of second
order, Manuscripta Math., 124 (2007) no. 2, 139–172.
61. S. Hofmann, D. Mitrea, M. Mitrea, and A.J. Morris, L p -Square Function Estimates on Spaces
of Homogeneous Type and on Uniformly Rectifiable Sets, Memoirs of the American Mathe-
matical Society, Vol. 245, No. 1159, 2017.
62. S. Hofmann, M. Mitrea, and M. Taylor, Geometric and transformational properties of Lips-
chitz domains, Semmes-Kenig-Toro domains, and other classes of finite perimeter domains,
J. Geom. Anal., 17 (2007), no. 4, 593–647.
63. S. Hofmann, M. Mitrea, and M. Taylor, Singular Integrals and Elliptic Boundary Problems on
Regular Semmes-Kenig-Toro Domains, Int. Math. Res. Not. IMRN (2010), no. 14, 2567–2865.
64. S. Hofmann, M. Mitrea, and M. Taylor, Symbol calculus for operators of layer potential type
on Lipschitz surfaces with VMO normals, and related pseudodifferential operator calculus,
Analysis and PDE, 8 (2015), no. 1, 115–181.
65. L Hörmander, The Analysis of Linear Partial Differential Operators, Vol. 3, Springer-Verlag,
New York, 1985.
66. L. Hörmander, The Analysis of Linear Partial Differential Operators. I. Distribution Theory
and Fourier Analysis, reprint of the second (1990) edition, Classics in Mathematics, Springer-
Verlag, Berlin, 2003.
956 References

67. T. Jakab, I. Mitrea, and M. Mitrea, Traces of functions in Hardy and Besov spaces on Lipschitz
domains with applications to compensated compactness and the theory of Hardy and Bergman
type spaces, J. Funct. Anal., 246 (2007), no. 1, 50–112.
68. B. Jawerth and M. Mitrea, Higher dimensional scattering theory on C 1 and Lipschitz domains,
Amer. J. Math., 117 (1995), no. 4, 929–963.
69. D. Jerison and C. Kenig, An identity with applications to harmonic measure, Bull. Amer.
Math. Soc., 2 (1980), 3, 447–451.
70. D. Jerison and C. Kenig, The Neumann problem on Lipschitz domains, Bull. Amer. Math.
Soc., 4 (1981), 2, 203–207.
71. D.S. Jerison and C.E. Kenig, Boundary behavior of harmonic functions in nontangentially
accessible domains, Advances in Mathematics, 46 (1982), no. 1, 80–147.
72. D.S. Jerison and C.E. Kenig, Hardy spaces, A∞ , and singular integrals on chord-arc domains,
Math. Scand., 50 (1982), 221–247.
73. D. Jerison and C.E. Kenig, The inhomogeneous Dirichlet problem in Lipschitz domains, J.
Funct. Anal., 130 (1995), 161–219.
74. F. John, Plane Waves and Spherical Means Applied to Partial Differential Equations, Inter-
science Publishers, New York-London, 1955.
75. J.-L. Journé, Calderón-Zygmund Operators, Pseudodifferential Operators, and the Cauchy
Integral of Calderón, Lecture Notes in Mathematics, Vol. 994, Springer-Verlag, 1983.
76. N. Kalton, S. Mayboroda, and M. Mitrea, Interpolation of Hardy-Sobolev-Besov-Triebel-
Lizorkin spaces and applications to problems in partial differential equations, pp. 121–177 in
“Interpolation Theory and Applications,” AMS Contemporary Mathematics, Vol. 445, 2007.
77. N. Kalton and M. Mitrea, Stability of Fredholm properties on interpolation scales of quasi-
Banach spaces and applications, Trans. Amer. Math. Soc., 350, (1998), 3837–3901.
78. C.E. Kenig, Weighted H p spaces on Lipschitz domains, Amer. J. Math., 102 (1980), 129–163.
79. C.E. Kenig, Harmonic Analysis Techniques for Second Order Elliptic Boundary Value Prob-
lems, CBMS Regional Conference Series in Mathematics, Vol. 83, AMS, Providence, RI,
1994.
80. C. Kenig and T. Toro, Free boundary regularity for harmonic measures and Poisson kernels,
Annals of Mathematics, 150 (1999), 369–454.
81. C. Kenig and T. Toro, Poisson kernel characterization of Reifenberg flat chord arc domains,
Ann. Sci. École Norm. Sup. (4), Vol. 36 (2003), 323–401.
82. C. Kenig and T. Toro, Free boundary regularity below the continuous threshold: 2-phase
problems, J. Reine Angew. Math., 596 (2006), 1–44.
83. B.V. Khvedelidze, Linear discontinuous boundary value problems of function theory, singular
integral equations, and some of their applications, (Russian), Trudy Tbil. Matem. Instituta
Akademii Nauk GSSR., 23 (1956), 3–158.
84. P. Koosis, Introduction to H p Spaces, Cambridge University Press, 1998.
85. V.A. Kozlov, V.G. Maz’ya and J. Rossmann, Elliptic Boundary Value Problems in Domains
with Point Singularities, Mathematical Surveys and Monographs, Vol. 52, Amer. Math. Soc.,
Providence, RI, 1997.
86. V.D. Kupradze, Potential Methods in the Theory of Elasticity, Daniel Davey & Co., Inc., New
York 1965.
87. O.A. Ladyzhenskaya, The Mathematical Theory of Viscous Incompressible Flow, Gordon and
Breach Science Publishers, New York-London, 1963.
88. J.-L. Lions and E. Magenes, Non-Homogeneous Boundary Value Problems and Applications,
Vol. 1, Springer-Verlag, Berlin, Heidelberg, New York, 1972.
89. W. Littman, G. Stampacchia, and H.F. Weinberger, Regular points for elliptic equations with
discontinuous coefficients, Ann. Scuola Norm. Sup. Pisa, (3) 17 (1963), 43–77.
90. A.M. Lyapunov, Papers on potential theory, Gostekhizdat, Moscow-Leningrad (1949).
91. E. Marmolejo-Olea, D. Mitrea, I. Mitrea, and M. Mitrea, Radiation conditions and integral
representations for Clifford algebra-valued null-solutions of the Helmholtz operator, Journal
of Mathematical Sciences, 231, (2018), no. 3, 367–472.
References 957

92. E. Marmolejo-Olea and M. Mitrea, Harmonic analysis for general first order differential op-
erators in Lipschitz domains, pp. 91–114 in “Clifford Algebras: Application to Mathematics,
Physics, and Engineering”, Birkhäuser Progress in Mathematical Physics Series, 2003.
93. J.M. Martell, Sharp maximal functions associated with approximations of the identity in
spaces of homogeneous type and applications, Studia Mathematica, 161 (2004), no. 2, 113–
145.
94. J. Martell, D. Mitrea, I. Mitrea, and M. Mitrea, The higher order regularity Dirichlet problem
for elliptic systems in the upper half-space, pp. 123–141 in “Harmonic Analysis and Partial
Differential Equations,” Contemporary Mathematics, Vol. 612, Amer. Math. Soc., Providence,
RI, 2014.
95. J.M. Martell, D. Mitrea, I. Mitrea, and M. Mitrea, The Dirichlet problem for elliptic systems
with data in Köthe function spaces, Revista Matemática Iberoamericana, 32 (2016), no. 3,
913–970.
96. S. Mayboroda and M. Mitrea, The solution of the Chang-Krein-Stein conjecture, pp. 61-154
in the Proceedings of the Conference in Harmonic Analysis and its Applications, Tokyo
Woman’s Christian University, Akihiko Miyachi editor, March 24-26, 2007, Tokyo, Japan.
97. S. Mayboroda and M. Mitrea, Sharp estimates for Green potentials on non-smooth domains,
Math. Res. Lett., 11 (2004), no. 4, 481–492.
98. V. Maz’ya, Differential Equations of My Young Years, Birkhüser, 2014.
99. V. Maz’ya, M. Mitrea, and T. Shaposhnikova, The Dirichlet problem in Lipschitz domains for
higher order elliptic systems with rough coefficients, Journal d’Analyse Mathématique, 110
(2010), no. 1, 167–239.
100. V. G. Maz’ya and T. O. Shaposhnikova, Theory of Multipliers in Spaces of Differentiable
Functions, Monographs and Studies in Mathematics, Vol. 23, Pitman, Advanced Publishing
Program, Boston, MA, 1985.
101. R. Mazzeo, Remarks on a paper of Friedlander concerning inequalities between Neumann
and Dirichlet eigenvalues, Int. Math. Res. Not., 4 (1991), 41–48.
102. A. McIntosh and M. Mitrea, Clifford algebras and Maxwell’s equations in Lipschitz domains,
Math. Meth. Appl. Sci., 22 (1999) 1599–1620.
103. W. McLean, Strongly Elliptic Systems and Boundary Integral Equations, Cambridge Univer-
sity Press, 2000.
104. Y. Meyer, Ondelettes et Opérateurs II. Opérateurs de Calderón-Zygmund, Hermann, Paris,
1990.
105. Y. Meyer, Ondelettes et Opérateurs III. Opérateurs multilinéarires, Hermann, Paris, 1990.
106. S.G. Mikhlin, Singular integral equations, Uspekhi Matem. Nauk. III, 25 (1948), no. 3,
29–112.
107. C. Miranda, Partial Differential Equations of Elliptic Type, second edition, Springer-Verlag,
Berlin, Heidelberg, New York, 1970.
108. D. Mitrea, Distributions, Partial Differential Equations, and Harmonic Analysis, Springer,
New York, 2013.
109. D. Mitrea, Distributions, Partial Differential Equations, and Harmonic Analysis, Second
edition, Springer Nature Switzerland, 2018.
110. D. Mitrea, I. Mitrea, and M. Mitrea, On the Geometry of Sets Satisfying Uniform Ball Condi-
tions, preprint, (2014).
111. D. Mitrea, I. Mitrea, and M. Mitrea, A sharp divergence theorem with nontangential traces,
Notices of the American Mathematical Society, 67 (2020), no. 9, 1295–1305.
112. D. Mitrea, I. Mitrea, and M. Mitrea, Geometric Harmonic Analysis I: A Sharp Divergence The-
orem with Nontangential Pointwise Traces, Developments in Mathematics Vol. 72, Springer
Nature, Switzerland, 2022.
113. D. Mitrea, I. Mitrea, and M. Mitrea, Geometric Harmonic Analysis II: Function Spaces Mea-
suring Size and Smoothness on Rough Sets, Developments in Mathematics Vol. 73, Springer
Nature, Switzerland, 2022.
114. D. Mitrea, I. Mitrea, and M. Mitrea, Geometric Harmonic Analysis III: Integral Representa-
tions, Calderón-Zygmund Theory, Fatou Theorems, and Applications to Scattering, Develop-
ments in Mathematics Vol. 74, Springer Nature, Switzerland, 2022.
958 References

115. D. Mitrea, I. Mitrea, and M. Mitrea, Geometric Harmonic Analysis IV: Boundary Layer Poten-
tials in Uniformly Rectifiable Domains, and Applications to Complex Analysis, Developments
in Mathematics Vol. 75, Springer Nature, Switzerland, 2022.
116. D. Mitrea, I. Mitrea, and M. Mitrea, Geometric Harmonic Analysis V: Fredholm Theory and
Finer Estimates for Integral Operators, with Applications to Boundary Problems, Develop-
ments in Mathematics Vol. 76, Springer Nature, Switzerland, 2022.
117. D. Mitrea, I. Mitrea, M. Mitrea, and S. Monniaux, Groupoid Metrization Theory with Appli-
cations to Analysis on Quasi-Metric Spaces and Functional Analysis, Birkhäuser, 2013.
118. D. Mitrea, I. Mitrea, M. Mitrea, and M. Taylor, The Hodge-Laplacian: Boundary Value Prob-
lems on Riemannian Manifolds, Studies in Mathematics, Vol. 64, De Gruyter, 2016.
119. D. Mitrea, M. Mitrea, and M. Taylor, Layer Potentials, the Hodge Laplacian and Global
Boundary Problems in Non-Smooth Riemannian Manifolds, Memoirs of the American Math-
ematical Society, Vol. 150, No. 713, Providence RI, 2001.
120. D. Mitrea, M. Mitrea, and J. Verdera, Characterizing regularity of domains via the Riesz
transforms on their boundaries, Analysis and PDE, 9 (2016), no. 4, 955–1018.
121. D. Mitrea, M. Mitrea, and L. Yan, Boundary value problems for the Laplacian in convex
domains and semiconvex domains, Journal of Functional Analysis, 258 (2010), no. 8, 2507–
2585.
122. I. Mitrea and M. Mitrea, Multi-Layer Potentials and Boundary Problems for Higher-Order
Elliptic Systems in Lipschitz Domains, Lecture Notes in Mathematics, Vol. 2063, Springer,
Berlin, 2013.
123. I. Mitrea, M. Mitrea, and M. Taylor, Cauchy integrals, Calderon projectors, and Toeplitz
operators on uniformly rectifiable domains, Adv. in Math., 268 (2015), 666–757.
124. I. Mitrea, M. Mitrea, and M. Taylor, Riemann-Hilbert Problems, Cauchy Integrals, and
Toeplitz Operators on Uniformly Rectifiable Domains, book manuscript, in preparation
(2017).
125. M. Mitrea, Clifford Wavelets, Singular Integrals, and Hardy Spaces, Lecture Notes in Math-
ematics, Vol. 1575, Springer-Verlag, Berlin, 1994.
126. M. Mitrea, The method of layer potentials in electro-magnetic scattering theory on non-smooth
domains, Duke Math. J., 77 (1995), no. 1, 111–133.
127. M. Mitrea, Generalized Dirac operators on non-smooth manifolds and Maxwell’s equations,
J. Fourier Anal. Appl., 7 (2001), 207–256.
128. M. Mitrea and B. Schmutzler, The regularity problem in rough subdomains of Riemannian
manifolds, Chapter 36, pp. 427–440 in “Integral Methods in Science and Engineering: The-
oretical and Computational Advances", C. Constanda and A. Kirsch editors, Birkhäuser,
2015.
129. M. Mitrea and M. Taylor, Boundary layer methods for Lipschitz domains in Riemannian
manifolds, J. Funct. Anal., 163 (1999), no. 2, 181–251.
130. M. Mitrea and M. Taylor, Potential theory on Lipschitz domains in Riemannian manifolds:
Hölder continuous metric tensors, Communications in Partial Differential Equations, 25
(2000), no. 7-8, 148
131. M. Mitrea and M. Taylor, Potential theory on Lipschitz domains in Riemannian manifolds:
Sobolev-Besov space results and the Poisson problem, J. Funct. Anal., 176 (2000), no. 1,
1–79.
132. M. Mitrea and M. Taylor, Potential theory on Lipschitz domains in Riemannian manifolds:
L p , Hardy, and Hölder space results, Communications in Analysis and Geometry, 9 (2001),
no. 2, 369–421.
133. M. Mitrea and M. Taylor, Boundary problems for the Navier-Stokes system on non-smooth
Riemannian manifolds, Math. Annalen, 321 (2001), no. 4, 955–987.
134. M. Mitrea and M. Taylor, Potential theory for Lipschitz subdomains of Riemannian manifolds
equipped with Dini continuous metric tensors, Transactions of the American Mathematical
Society, 355 (2003), no. 5, 1961–1985.
135. M. Mitrea and M. Wright, Boundary Value Problems for the Stokes System in Arbitrary
Lipschitz Domains, Astérisque, Soc. Math. de France, Vol. 344, 2012.
References 959

136. G. Moisil, Sur la généralisation des fonctions conjuguées, Rendiconti della Reale Accad. Naz.
dei Lincei, 14 (1931), 401–408.
137. Gr.C. Moisil, Les fonctions monogènes dans les espaces à plusieurs dimensions, p. 129 in
Comptes Rendus LXIV Congrès des Soc. Savantes Clermont-Ferrant, 1931.
138. G.C. Moisil and N. Teodorescu, Fonctions holomorphes dans l’espace, Mathematica (Cluj),
5 (1931), 142–159.
139. C.B. Morrey, Second order elliptic systems of differential equations. Contributions to the
theory of partial differential equations, Ann. Math. Studies, 33 (1954), 101–159.
140. C. Müller, Zur mathematischen Theorie elektromagnetischer Schwingungen, Abh. deutsch.
Akad. Wiss. Berlin, 3 (1945/1946), 5–56.
141. J. Nečas, Direct Methods in the Theory of Elliptic Equations, corrected 2nd printing, Springer
Monographs in Mathematics, Springer, Heidelberg, Dordrecht, London, New York, 2012.
142. N. Ortner and P. Wagner, Fundamental Solutions of Linear Partial Differential Operators.
Theory and Practice, Springer, 2015.
143. V.A. Paatashvili and G.A. Khuskivadze, On the boundedness of the singular Cauchy integral
on Lebesgue spaces in the case on non-smooth contours (Russian), Trudy Tbilisk. Matem.
Inst. AN GSSR., 69 (1982), 93–107.
144. L.E. Payne and H.F. Weinberger, New bounds in harmonic and biharmonic problems, J. Math.
Phys., 33 (1954), 291–307.
145. J. Plemelj, Ein Ergänzungssatz zur Cauchyschen Integraldarstellung analytischer Funktionen,
Randwerte betreffend, Monatshefte für Mathematik und Physik, 19, (1908), 205–210.
146. J. Plemelj, Riemannsche Funktionenscharen mit gegebener Monodromiegruppe, Monatshefte
für Mathematik und Physik, 19, (1908), 211–246.
147. H. Poincaré, Sur les equations aux dérivée partielles de la physique mathématiques, Amer. J.
Math., 12 (1890), 211–294.
148. D. Pompeiu, Sur une classe d’intégrales doubles, Bull. de la Sect. Scient. de l’Académie
Roumaine, 1 (1912), 128–131.
149. D. Pompeiu, Sur une classe d’intégrales doubles (deuxième note), Bull. de la Sect. Scient. de
l’Académie Roumaine, 1 (1912), 265–271.
150. D. Pompeiu, Sur une classe d’intégrales doubles (troisième note), Bull. de la Sect. Scient. de
l’Académie Roumaine, 1 (1912), 289–296.
151. D. Pompeiu, Sur une classe de fonctions d’une variable complexe et sur certaines équations
intégrales, Rendiconti del Circolo Matematico di Palermo, 35 (1913), 277–281.
152. P. Pucci and J. Serrin, Critical exponents and critical dimensions for polyharmonic operators,
J. Math. Pures Appl., 69 (1990), 55–83.
153. J. Rauch and M. Taylor, Regularity of functions smooth along foliations, and elliptic regularity,
J. Funct. Anal., 225 (2005), 74–93.
154. F. Rellich, Darstellung der Eigenwerte von Δu + λu = 0 durch ein Randintegral, Math. Z.,
46 (1940), 635–636.
155. W. Rudin, Functions Theory in the Unit Ball of C n , Springer-Verlag, New York, Heidelberg,
Berlin, 1980.
156. W. Rudin, Real and Complex Analysis, 3-rd edition, McGraw-Hill, Boston, Massachusetts,
1987.
157. H.A. Schwarz, Zur integration der partiellen differentialgleichung ∂2 u/∂x 2 + ∂2 u/∂y 2 = 0,
J. Reine Angew. Math., 74 (1872), 218–253.
158. A. Seeger, A note on Triebel-Lizorkin spaces, Approximation and Function Spaces, pp. 391–
400, Banach Center Publ., No. 22, PWN, Warsaw, 1989.
159. S.W. Semmes, A criterion for the boundedness of singular integrals on hypersurfaces, Trans.
Amer. Math. Soc., 311 (1989), no. 2, 501–513.
160. S. Semmes, Analysis vs. geometry on a class of rectifiable hypersurfaces in R n , Indiana Univ.
Math. J., 39 (1990), no. 4, 1005–1035.
161. Z. Shapiro, On elliptical systems of partial differential equations, C.R. (Doklady) Acad. Sci.
URSS (N. S.), 46 (1945).
162. S. Silver, Microwave Antenna Theory and Design, M.I.T. Radiation Laboratory Series, Vol. 12,
McGraw-Hill, New York, 1949.
960 References

163. Yu.V. Sokhotski, On definite integrals and functions employed in expansions into series, St.
Petersburg, 1873.
164. A. Sommerfeld, Die Greensche Funktion der Schwingungsgleichung, Jahresbericht der
Deutschen Mathematiker-Vereinigung, 21 (1912), 309–353. Reprinted in Gesammelte
Schriften, Vol. 1, pp. 272–316.
165. E.M. Stein, Conjugate harmonic functions in several variables, pp. 414–420 in International
Congress of Mathematicians, Stockholm, 1962.
166. E.M. Stein, Singular Integrals and Differentiability Properties of Functions, Princeton Math-
ematical Series, No. 30, Princeton University Press, Princeton, NJ, 1970.
167. E.M. Stein, Harmonic Analysis: Real-Variable Methods, Orthogonality, and Oscillatory In-
tegrals, Princeton Mathematical Series, Vol. 43, Monographs in Harmonic Analysis, III,
Princeton University Press, Princeton, NJ, 1993.
168. E.M. Stein, Singular Integrals: The Roles of Calderón and Zygmund, Notices of AMS, 45
(1998), no. 9, 1130–1140.
169. E.M. Stein and G. Weiss, On the theory of harmonic functions of several variables, I. The
theory of H p spaces, Acta Math., 103 (1960), 25–62.
170. E.M. Stein and G. Weiss, Generalization of the Cauchy-Riemann equations and representa-
tions of the rotation group, Amer. J. Math., 90 (1968), no. 1., 163–196.
171. E.M. Stein and G. Weiss, Introduction to Fourier Analysis on Euclidean Spaces, Princeton
University Press, Princeton, N.J., 1971.
172. J.A. Stratton and L.J. Chu, Diffraction theory of electromagnetic waves, Physical Review, 56
(1939), 99–107.
173. Ş. Strătilă and L. Zsidó Lectures on von Neumann Algebras, Editura Academiei, Bucureşti,
Romania/Abacus Press, Tunbridge Wells, Kent, England, 1979.
174. M.E. Taylor, Pseudodifferential Operators and Nonlinear PDE, Birkhäuser, Boston, 1991.
175. M. Taylor, Partial Differential Equations, Vol. I, Springer-Verlag, 1996 (2nd ed., 2011).
176. M. Taylor, Partial Differential Equations, Vol. II, Springer-Verlag, 1996 (2nd ed., 2011).
177. M. Taylor, Regularity for a class of elliptic operators with Dini continuous coefficients, J.
Geom. Anal., 21 (2011), 174–194.
178. N. Théodoresco, La Dérivée Areolaire, Ann. Roumanine des Math., Cahier 3, Bucharest,
1936.
179. G. Verchota, Layer potentials and regularity for the Dirichlet problem for Laplace’s equation
in Lipschitz domains, J. Funct. Anal., 59 (1984), no. 3, 572–611.
180. N. Wiener, The Dirichlet problem, J. Math. Phys., 3 (1924), 127–146.
181. T.H. Wolff, Counterexamples with harmonic gradients in R3 , pp. 321–384 in “Essays on
Fourier Analysis in Honor of Elias M. Stein,” Princeton Math. Ser., Vol. 42, Princeton Univ.
Press, Princeton, NJ, 1995.
Subject Index

(ε, δ)-domain, 930 boundary-to-domain, left-handed,

P-maximal function, 942 461
Pg -maximal function, 732 boundary-to-domain,
right-handed, 461
Ahlfors regular Cauchy-Riemann equations
domain, 921 generalized, 46
Characteristic matrix of the
Barrier function, 842 (complex) Lamé system, 48
Besov space Classical pseudodifferential operator,
inhomogeneous, 925 561
Boundary Coefficient tensor
measure theoretic, 943 complex, 188
Boundary-to-boundary Commutator, 531
Cauchy-Clifford integral operator, Complementing (system), 51
460 Complex arc-length (measure), 3
Boundary-to-domain Conormal derivative (for a generic
Cauchy-Clifford integral operator, system, in a pointwise sense),
460 189
Bullet product, 941
Dirac operator
Carleson measure homogeneous, 887
in an open set, 382 perturbed, 887
Cauchy-Clifford integral operator Distribution
boundary-to-boundary, 460 on an arbitrary set, 936
boundary-to-boundary, Domain
left-handed, 461 (ε, δ), 930
boundary-to-boundary, Ahlfors regular, 921
right-handed, 461 Dual system
boundary-to-domain, 460 definition, 200
© The Editor(s) (if applicable) and The Author(s), under exclusive license to 961
Springer Nature Switzerland AG 2023
D. Mitrea et al., Geometric Harmonic Analysis II, Developments in Mathematics 74,
https://doi.org/10.1007/978-3-031-22735-6
962 SUBJECT INDEX

regular, 200 Lower class (of functions), 836

Luxemburg norm on an Orlicz space,
Exterior derivative operator, 929 937

Fourier transform, 62 Maximal Cauchy-Clifford integral

Function operator, 460
barrier, 842 Maxwell system, 887
subharmonic, 833 McIntosh-Mitrea radiation condition,
superharmonic, 833 887
Fundamental solution of the Measure
Helmholtz operator (radiating), harmonic, 840
882 Meromorphic function, 15
Modified Moisil-Teodorescu system,
Generalized 53, 661
Cauchy-Riemann equations, 46, Modified principal-value singular
661 integral operator, 351
Cauchy-Riemann system, 46 Moisil-Teodorescu system, 46
Generalized Cauchy-Riemann modified, 53, 661
equations, 46
Green function for the Laplacian, 842 Nontangential
maximal operator, 940
Half-sphere lemma, 424 Nontangential maximal function
Hankel function of the first kind, 882 truncated, 941
Hardy space Nontangentially bounded from
“bullet” boundary, 686 below, 855
associated with an operator, 682 Norm
Lebesgue-based, 932 Luxemburg, on an Orlicz space,
Lorentz-based, 933 937
Harmonic double layer
boundary-to-boundary, 446 Operator
principal-value, 446 perturbed Dirac, in Rn , 887
Harmonic measure, 237, 840 Orlicz space, 937
Helmholtz operator, 881 Poisson integral, 837
Hodge star operator, 947 Poisson kernel
Homogeneous Dirac operator in Rn , for the Laplacian, 849
887 in an open set, 814
Principal symbol, 560
Inhomogeneous Besov space, 925
Principal symbol of a first-order
Injectively elliptic (first-order)
system, 50
system, 50
Proximally regular set, 873
Pseudo-ball, 875
Lamé system, 48, 939
Legendre-Hadamard (strong) Quasi-metric space
ellipticity, 48 d-Ahlfors regular, 273
Local (boundary) Sobolev space, 937
Local John condition, 935 Radiating at infinity, 881
SUBJECT INDEX 963

Radiation condition Standard kernel

McIntosh-Mitrea, 887 in the first variable, 276
Silver-Müller, 887 in the second variable, 276
Sommerfeld, 881 Strong ellipticity
Radon-Nikodym derivative of the (Legendre-Hadamard), 48
harmonic measure, 849 Subharmonic function, 833
Regular SKT domain, 871 Superharmonic function, 833
Resolutive boundary function, 837 Support of a measurable function,
Riemannian metric tensor, 234 948
Riesz transform Surface ball, 852
boundary-to-domain, 469 Surface measure
σ := H n−1 !Σ on Σ, 531
Sarason space VMO, 950
System
Schwarz-Pompeiu
Legendre-Hadamard elliptic, 48
formula, 24
strongly (Legendre-Hadamard)
Schwarz-Pompeiu formula, 24
elliptic, 48
Semiconvex function, 873
System of auxiliary function, 931
Set
proximally regular, 873 Tangential derivative
regular for the Dirichlet Problem, weak, 943
840 Tangential derivative operator, 943
Silver-Müller radiation conditions, in a pointwise sense, 944
887 Transpose kernel, 276
Sobolev space Truncated
L p -based homogeneous, 938 singular integral operator, 277,
homogeneous, 938 348
local, 937
negative Uniform exterior ball condition
definition, 938 (UEBC), 873
Sommerfeld radiation condition, 881 Uniform interior ball condition
Space (UIBC), 873
L p -based Sobolev, 938 Unique continuation property (UCP),
L p -based homogeneous Sobolev, 734
938 Upper class (of functions), 836
Hardy (Lebesgue-based), 932
Hardy (Lorentz-based), 933 Vanishing at infinity (null-solution of
inhomogeneous Besov, 925 a first-order system), 171
negative Sobolev, 938 Vanishing mean oscillations, 950
of vanishing mean oscillations,
950 Weak boundedness property (WBP),
Orlicz, 937 336
Sarason, VMO, 950 Weak tangential derivative, 943
Standard fundamental solution for Weakly elliptic system, 47
the bi-Laplacian, 76 Weighted
Standard fundamental solution for Sobolev space in Ω, for the weight
ap
the Laplacian, 127, 930 w := δ∂Ω , 950
Symbol Index

a ⊗ b tensor product of vectors a, b, ·, · Λ T M (real) pointwise pairing on

48, 922 Λ T M, 943
∗ Hodge star operator, 947 u · w = u, w dot product of two
∧ exterior product of differential vectors u, w ∈ Cn , 934
forms, 943 u · w = u, w dot product of two
∨ interior product of differential vectors u, w ∈ Rn , 928
forms, 943 [Mb, T] commutator of Mb with T,
Δ + k Helmholtz operator, 881, 930
2
531, 940
 
Δ Laplace operator, 928 Aζ, η bilinear form associated with
∇u gradient (Jacobian matrix) of u, the coefficient tensor A, 189,
932 922
∇  gradient operator in Rn−1 , 932 [F]∞ contribution of F at infinity,

∇tan tangential gradient, 932 931
∇tan
A , 198, 932
[u]∞ A contribution at infinity of

A∇u, 189, 922 null-solution for the

Δ = Δ(x, r) surface ball, 852, 929 vector-Helmholtz operator, 896
UV symmetric difference of U and [u]∞ A contribution at infinity of

V, 929 null-solution for the

U $ V the union of two disjoint sets vector-Helmholtz operator, 950
U, V, 949  · Bd(X→Y) operator norm, 942
D  (Ω) ·, · D(Ω) distributional pairing  · Bd(X) operator norm, 942
in Ω, 928  · X→Y operator norm, 942
X ∗ ·, · X , 950 ·
ess
essential norm, 942
(Lip c (Σ)) ·, · Lip c (Σ) (or simply ·, · )
  X→Y
 ·  Minkowski functional of
p 
  distributional pairing, 936 BX (0, 1) p , 951
·, · E pointwise (real) pairing in the
fibers of Hermitian vector (a)+ := max{a, 0}, 944
bundle E, 943 X/Y quotient space, 945
© The Editor(s) (if applicable) and The Author(s), under exclusive license to 965
Springer Nature Switzerland AG 2023
D. Mitrea et al., Geometric Harmonic Analysis II, Developments in Mathematics 74,
https://doi.org/10.1007/978-3-031-22735-6
966 SYMBOL INDEX

[x]X/Y equivalence class of x ∈ X in ν • F the bullet product of ν with F,


X/Y , 930 941
!
x := x/|x|, 881, 951 ν • u the Clifford bullet product of ν
1E characteristic function of E, 942 with u, 941
⨏fΔ integral average of f in Δ, 934 ω x harmonic measure with pole at x,
E
f dμ the integral average of f on 839, 942
E, 934 ωn−1 surface area of S n−1 , 942
Ů interior of the set U, 934 ρ# the regularized version of ρ, 948
U closure of the set U, 925 σg surface measure induced by the
V ⊥ annihilator of a subspace V of a metric tensor g, 947
Banach space X, 921 σ E surface measure induced by the
⊥ W annihilator of a subspace W of standard Euclidean metric, 947
∗ σ∗ = H n−1 !∂∗ Ω surface measure,
√ X , where X is Banach, 921
i = −1 ∈ C complex imaginary 947
unit, 934 σ = H n−1 !∂Ω surface measure on
[A; B] := [A, B] := AB − BA the ∂Ω, 947
commutator of A and B, 924 σ := H n−1 !Σ surface measure on
{ A; B} := AB + BA the the closed Ahlfors regular set
anti-commutator of A and B, Σ ⊆ Rn , 531, 947
924 ∂nta Ω nontangentially accessible
dζ complex arc-length, 3, 930 boundary of Ω, 944
d exterior derivative operator, 929 ∂∗ E measure theoretic boundary of
δ formal adjoint of the exterior E, 943
derivative operator d, 929 ∂ ∗ E reduced boundary of E, 944
δ jk Kronecker symbol, 929 ∂νA conormal derivative operator
δx Dirac distribution with mass at x, with respect to the coefficient
928 tensor A acting from Besov and
δF distance function to the set F, 929 Triebel-Lizorkin spaces, 944
δ∂Ω (·) distance function to the ∂νA conormal derivative operator
boundary, 929 with respect to the coefficient
φ! = ℱφ Fourier transform of φ, 62, tensor A acting from weighted
931 . Sobolev spaces, 944
Φk (·) radiating fundamental solution ∂νA weak conormal derivative
of the Helmholtz operator, 882, operator with respect to the
931 coefficient tensor A, 944
Φ(X → Y ) Fredholm operators from ∂νA pointwise conormal derivative
X into Y , 931 operator with respect to the
Γκ (x) nontangential approach region, coefficient tensor A, 189, 945

932 ∂νD,D pointwise conormal derivative
Λ T M the -th exterior power of the associated with the
vector bundle on M, 936  194,
factorization L = DD,
νg GMT unit normal induced by the 945
metric tensor g, 941 ∂ξA conormal derivative along ξ, with
ν E GMT unit normal induced by the respect to the coefficient tensor
standard Euclidean metric, 941 A, 889, 945
SYMBOL INDEX 967

∂τ j k pointwise tangential derivative Bd X → Y linear and (norm)

operator, 944 bounded operators from X to
∂τ j k tangential derivative operator, Y , 924
BMO(X, μ) space of functions of
. 945
∂τ j k weak tangential derivative, 943 bounded mean oscillations, 923
∂τXY tangential derivative operator .
 · BMO(X,μ) , homogeneous BMO
on manifolds, 943 semi-norm, 923
∂, ∂z the conjugate of the  · BMO(X,μ) inhomogeneous BMO
Cauchy-Riemann operator, 2, “norm”, 923
945  f ∗ (Δ) local BMO norm of f on Δ,
∂, ∂z̄ the Cauchy-Riemann operator, 923
2, 945 
BMO(X, μ) the space BMO modulo
πκ (E), πΩ,κ (E) “shadow” (or constants, 924
projection) of E ⊆ Ω onto ∂Ω, Bn−1 (x , r) open ball with center x 
944 and radius r in Rn−1 , 923
τρ topology induced by the Bρ (x, r) ρ-ball with center at x and
quasi-distance ρ, 948 radius r, 923
p,q
αβ
A = ar s 1≤r,s ≤n coefficient tensor Bs (Σ, σ) (inhomogeneous) Besov
1≤α ≤M space on the set Σ, 925
1≤β ≤ N
(of type (n × n, M × N)), 188  · Bsp, q (Σ,σ) (inhomogeneous)
αβ Besov quasi-norm, 925
A = ar s 1≤r,s ≤n coefficient tensor p,q
1≤α ≤M Bs (Rn ) Besov space in Rn , 925
1≤β ≤ N
(of type (n × n, M × N)), 922  · Bsp, q (Rn ) quasi-norm in the Besov
space in Rn , 925
A (global) transpose of A, 188, 922
B (Σ, σ) block space, 924
q,λ
Aκ (∂Ω) accessibility set, 921
 ·  B q, λ (Σ,σ) norm on block space,
Ap (X, ρ, μ) Muckenhoupt class, 921
924
[w] A p characteristic of the
 · ℳq, λ (Σ,σ) norm on midway space,
Muckenhoupt weight w, 921
940
A∞ (X, ρ, μ) Muckenhoupt class, 921 Borelτ (X) Borelians of the
A p absolutely p-convex hull of the topological space (X, τ), 924
set A, 922 p,q
Bs (Ω) Besov space in Ω, 925
p,q
As (Ω) Besov/Triebel-Lizorkin  · Bsp, q (Ω) quasi-norm in the Besov
space in Ω, 923 space in Ω, 925
 ·  Ap, q
s (Ω)
quasi-norm in C boundary-to-domain
Besov/Triebel-Lizorkin space Cauchy-Clifford integral
in Ω, 923 operator, 460, 927
A q,κ L q -based area-function, 387, CL left-handed boundary-to-domain
922 Cauchy-Clifford integral
B(X → Y ) linear and (topologically) operator, 461, 927
bounded operators from X to CR right-handed
Y , 924 boundary-to-domain
Bd(X) linear and bounded operators Cauchy-Clifford integral
on X, 924 operator, 461, 928
968 SYMBOL INDEX

Cmax maximal Cauchy-Clifford CBM(X, τ) complex Borel measures

integral operator, 460, 928 in the topological space (X, τ),
C boundary-to-boundary 925
Cauchy-Clifford integral cap (K) capacity of the compact set
operator, 460, 927 K, 841, 927
C L left-handed Cn Clifford algebra generated by n
boundary-to-boundary imaginary units, 927
Cauchy-Clifford integral Cp(X → Y ) space of compact linear
operator, 461, 927 operators from X into Y , 927
C R right-handed Cp(X) space of compact linear
boundary-to-boundary operators from X into itself,
Cauchy-Clifford integral 927
operator, 461, 928 Cρ , 925
C # transpose Cauchy-Clifford ρ , 926
C
integral operator, 928 D(X) dyadic grid on X, 930
𝒞 (Ω) functions of class 𝒞k in an
k Dk (X), 930
open neighborhood of Ω, 925 D (Ω) space of distributions in Ω,
𝒞kc (Ω) functions of class 𝒞k with 928
compact support in the open D = nj=1 e j  ∂j Dirac operator in
set Ω, 925 Rn , 928
𝒞b (Ω) bounded functions of class
k D first-order system, 929
D (real) transpose of the first-order
𝒞k in Ω, 925
system D, 929
𝒞k -singsup u singular support of u,
D complex conjugate of the
947
∞ ∗ first-order system D, 929
𝒞b (Ω) the algebraic dual of D∗ Hermitian adjoint of the
∞
𝒞b (Ω), 925
. first-order system D, 929
𝒞α (U, ρ) homogeneous Hölder Def deformation tensor, 55, 930
space, 926 D L Dirac operator acting from the
 · 𝒞. α (U,ρ) homogeneous Hölder left, 929
space semi-norm, 926 DR Dirac operator acting from the
. right, 929
𝒞α (U, ρ)/∼ homogeneous Hölder
space modulo constants, 926 D homogeneous Dirac operator in
. α (U, ρ) local homogeneous
Rn , 887
𝒞loc Dk perturbed Dirac operator in Rn ,
Hölder space, 926 887, 930
α
𝒞 (U, ρ) inhomogeneous Hölder diag (X) diagonal in the Cartesian
space, 926 product X × X, 273, 930
 · 𝒞α (U,ρ) inhomogeneous Hölder diamρ (A) ρ-diameter of the set A,
space norm, 926 929
α
𝒞c (U, ρ) Hölder functions with dg (x, y) geodesic distance between x
ρ-bounded support, 927 and y, 928
CBM(Ω) complex Borel measures in divF the divergence of the vector
Ω, 925 field F, 928
SYMBOL INDEX 969
p,q
divg differential geometric Fs (Ω) Triebel-Lizorkin space in Ω,
divergence (associated with the 931
metric tensor g), 928  · Fs q (Ω) quasi-norm in
p,

dim X dimension of X, 930 Triebel-Lizorkin space in Ω,

dVg volume element on M induced 931
by the metric tensor g, 950 GΩ (·, ·) Green function for the
E = Eαβ 1≤α,β ≤M fundamental Laplacian, 842, 932
solution of the system L, 63, g = 1≤ j,k ≤n g jk dx j ⊗ dxk
930 Riemannian metric tensor, 931
EΔ standard fundamental solution for g = 1≤ j,k ≤n g jk dx j ⊗ dxk
the Laplacian, 127, 930 Riemannian metric tensor, 234
EΔ2 standard fundamental solution H n−1 the (n − 1)-dimensional
for the bi-Laplacian, 76, 930 Hausdorff measure in Rn , 932
ℰ(Ω) distributions compactly Hg (n − 1)-dimensional Hausdorff
n−1

supported in Ω, 930 measure associated with the

ℰK  (Ω) distributions in Ω supported
metric g, 932
in K, 930 H s s-dimensional Hausdorff
E p (X) p-envelope of X, 931 measure in Rn , 932
Ex∂Ω→Ω extension operator from ∂Ω H∗ s-dimensional Hausdorff outer
s

to Ω, 931 measure in Rn , 932

e j standard j-th unit vector in Rn , H L -filtering operator, 933
p

930 H s (Ω) L 2 -based fractional Sobolev

{e j }1≤ j ≤n standard orthonormal space in Ω, 933
basis in Rn , 930  ·  H s (Ω) norm in the L 2 -based
p,q
Fs (Rn ) Triebel-Lizorkin space in fractional Sobolev space in Ω,
Rn , 931 934
 · Fsp, q (Rn ) quasi-norm in the Hλ(1) (·) Hankel function of the first
Triebel-Lizorkin space in Rn , kind, 882, 934
p
931 ℋ• (∂Ω; D) “bullet” boundary Hardy
p,q
Fs (Σ, σ) (inhomogeneous) space, 686, 934
Triebel-Lizorkin space on the H p (Ω; D) Hardy space in Ω,
set Σ, 931 associated with the first-order
 · Fsp, q (Σ,σ) (inhomogeneous) operator D, 682, 934
Triebel-Lizorkin quasi-norm,  ·  H p (Ω;D) quasi-norm in the Hardy
931 space H p (Ω; D), 683, 934
ℱφ = φ! Fourier transform of φ, 62, H (Σ, σ) Hardy space, 932
p
931  ·  H p (Σ,σ) quasi-norm on Hardy
fBρ (x,r) integral average of f over space, 932
Bρ (x, r), 934 H p,q (Σ, σ) Lorentz-based Hardy
fE∗ non-increasing rearrangement of space, 933
f : E → R, 931  ·  H p, q (Σ,σ) quasi-norm on
fγ Fefferman-Stein grand maximal Lorentz-based Hardy space,
function of f , 931 933
p,q
fp# L p -based Fefferman-Stein sharp Hfin (Σ, σ) finite linear combinations
maximal function, 948 of atoms, 933
970 SYMBOL INDEX

 ·  H p, q (Σ,σ) quasi-norm on ∞
Lcomp essentially bounded functions
fin
p,q with compact support, 935
Hfin (Σ, σ), 933 p
p
Hκ (Ω) Hardy space of harmonic Lbdd (Ω, L n ) p-th power integrable
functions, 856, 934 functions over bounded subsets
ℋq,λ (Σ, σ), 933 of Ω, 936
 · ℋq, λ (Σ,σ) , 933 L p (Ω, wL n ) weighted L p Lebesgue
IE,α fractional integral operator of space over Ω, 937
p
order α on E, 934 L (Ω, μ) maximal Lebesgue space,
936
Im T : X → Y image of T : X → Y , p
L1 (∂∗ Ω, σ∗ ) L p -based (boundary)
935
Sobolev space, 938
index index function, 935
 ·  L (∂∗ Ω,σ∗ ) norm on Sobolev
p
i(Φ) lower dilation index of Φ, 935 1
space, 938
I(Φ) upper dilation index of Φ, 935 p
L1,loc (∂∗ Ω, σ∗ ) local (boundary)
KΔ boundary-to-boundary harmonic
double layer potential, 935 . p Sobolev space, 937
L1 (∂Ω, σ) homogeneous Sobolev
KΔ# transpose harmonic double layer
space, 938
potential, 935
L p,q (X, μ) Lorentz space on X with
Ker T : X → Y kernel of respect to the measure μ, 936
T : X → Y , 935
 ·  L p, q (X,μ) Lorentz space
Ker L null-space of the system L, 935 quasi-norm, 936
k x0 Radon-Nikodym derivative of p,q
L (Ω, μ) maximal Lorentz space,
ω x0 with respect to σ, 849, 935 936
L homogeneous second-order system p
Ls (Ω) Bessel potential space in Ω,
of differential operators, 47, 937
938  ·  Lsp (Ω) norm in the Bessel
L A second-order system associated potential space in Ω, 937
to a coefficient tensor A, 189, L Φ (X, μ) Orlicz space, 937
938  ·  L Φ (X,μ) Luxemburg norm on the
L(ξ) characteristic matrix of L, 47, Orlicz space L Φ (X, μ), 937
51, 628, 938, 939 L (log L)α Zygmund’s space, 937
p
L(X → Y ) linear and continuous p
L−1 (∂∗ Ω, σ∗ ) negative Sobolev
operators from X to Y , 937 space, 938
L Lebesgue measure in Rn , 935
n
log+ positive ln, 937
Lgn measure associated with the Mb operator of pointwise
n-form dVg , 936 multiplication by the function
Lμ,λ the Lamé system, 48, 939 b, 531, 940
Lλ,μ (ξ) characteristic matrix of the M p,λ (Σ, σ) Morrey space, 939
Lamé system, 48, 939  ·  M p, λ (Σ,σ) norm on Morrey space,
Lipc (X) space of Lipschitz functions 939
with bounded support in X, M̊ p,λ (Σ, σ) vanishing Morrey space,
936 940

Lipc (Σ) distributions on Σ, 936 ℳ+ (X, μ) non-negative
L (X, μ) measurable functions which
0 μ-measurable functions on X,
are a.e. pointwise finite, 936 277, 939
SYMBOL INDEX 971

ℳ(X, μ) μ-measurable functions on R C boundary-to-domain

X, 291, 939 Clifford-Riesz transform, 947
M X Hardy-Littlewood maximal S n−1 unit sphere in Rn , 947
operator on X, 939 Sym(D; ξ) principal symbol of the
M X,s L s -based Hardy-Littlewood first-order system D, 50, 947
maximal operator, 939 𝒮(Rn ) Schwartz functions, 948
ℳq,λ (Σ, σ) (q, λ)-mid space, 940 𝒮(Rn ) tempered distributions, 948
N0 = N ∪ {0}, 940 supp f support of the measurable
p
Nκ (Ω; μ), 941 function f , 948
Nκ nontangential maximal operator, T ∗ adjoint of T, 922
940 Tε truncated singular integral
NκE the nontangential maximal operator, 277, 348, 949
operator restricted to E, 941 Tmax maximal operator, 277, 348,
ε
Nκ the nontangential maximal 949
function truncated at height ε, Tmod modified principal-value
941 singular integral operator, 351,
Oε one-sided collar neighborhood of 949
∂Ω, 942 TrΩ→∂Ω trace operator from Ω to
pX lower Boyd index, 944 ∂Ω, 949
qX upper Boyd index, 945 TrRn →Σ trace operator from Rn to Σ,
P maximal function of Carleson 948
type, 942 TrΩ→∂Ω trace operator from Ω to
Pg maximal function of Carleson ∂Ω, 948
type on manifolds, 732 Tmod boundary-to-domain modified
PI Poisson integral operator, 837, integral operator, 413, 949
945 Tγ (x) bump functions centered at x,
P.V. b k(x − ·)|Σ principal-value 948
κ−n.t.
distribution on the set Σ, 945 u|∂Ω (x) nontangential trace of u
𝒫δ,b (x, h) pseudo-ball, 875, 945 at x ∈ ∂Ω, 942
pΩ conjugate exponent of qΩ , 868, u,θ solid maximal function of u,
945 948
qΩ , 868 UC(X, ρ) the space of uniformly
qΩ conjugate exponent of pΩ , 946 continuous functions on the
R+n upper half-space in Rn , 946 metric space (X, ρ), 949
R−n lower half-space in Rn , 946 umax
M tangential maximal function of
R j boundary-to-boundary Riesz u, 939
transform, 469, 946 VMO(X, μ) space of functions of
mod
R j modified Riesz transform, 947 vanishing mean oscillations,
R jk , 458, 946 950
(R jk )max , 458, 946 W k, p (Ω) L p -based Sobolev space of
Res( f , z) residue of f at z, 15, 946 order k in Ω, 950
k, p
R j boundary-to-domain Riesz Wbdd (Ω), 950
k, p
transform, 469, 946 Wloc (Ω) local L p -based Sobolev
R jk , 456, 946 space of order k in Ω, 950
972 SYMBOL INDEX

k, p
Wa (Ω) weighted Sobolev space in  · X norm on the Generalized
ap
Ω, for the weight w := δ∂Ω , 950 Banach Function Space X, 932
 · W k, p (Ω) quasi-norm in the X associated space of X, 950
a
weighted Sobolev space  · X norm on the associated space
k, p
Wa (Ω), 950 of X, 950
∞
X̊ closure of Lcomp in X, 932
𝒳bdd (Ω), 𝒳(Ω)bdd , 923, 951
X Generalized Banach Function
Space on the measure space X  , 947
(X, M, μ), 932 ξ  , 947