Stoll, Manfred - Harmonic and Subharmonic

LONDON MATHEMATICAL SOCIETY LECTURE NOTE SERIES
Managing Editor: Professor M. Reid, Mathematics Institute,

University of Warwick, Coventry CV4 7AL, United Kingdom
The titles below are available from booksellers, or from Cambridge University Press at
http://www.cambridge.org/mathematics
311 Groups: topological, combinatorial and arithmetic aspects, T.W. MÜLLER (ed)
312 Foundations of computational mathematics, Minneapolis 2002, F. CUCKER et al (eds)
313 Transcendental aspects of algebraic cycles, S. MÜLLER-STACH & C. PETERS (eds)
314 Spectral generalizations of line graphs, D. CVETKOVIĆ, P. ROWLINSON & S. SIMIĆ
315 Structured ring spectra, A. BAKER & B. RICHTER (eds)
316 Linear logic in computer science, T. EHRHARD, P. RUET, J.-Y. GIRARD & P. SCOTT (eds)
317 Advances in elliptic curve cryptography, I.F. BLAKE, G. SEROUSSI & N.P. SMART (eds)
318 Perturbation of the boundary in boundary-value problems of partial differential equations, D.HENRY
319 Double affine Hecke algebras, I. CHEREDNIK
320 L-functions and Galois representations, D. BURNS, K. BUZZARD & J. NEKOVÁŘ (eds)
321 Surveys in modern mathematics, V. PRASOLOV & Y. ILYASHENKO (eds)
322 Recent perspectives in random matrix theory and number theory, F. MEZZADRI & N.C. SNAITH (eds)
323 Poisson geometry, deformation quantisation and group representations, S. GUTT et al (eds)
324 Singularities and computer algebra, C. LOSSEN & G. PFISTER (eds)
325 Lectures on the Ricci flow, P. TOPPING
326 Modular representations of finite groups of Lie type, J.E. HUMPHREYS
327 Surveys in combinatorics 2005, B.S. WEBB (ed)
328 Fundamentals of hyperbolic manifolds, R. CANARY, D. EPSTEIN & A. MARDEN (eds)
329 Spaces of Kleinian groups, Y. MINSKY, M. SAKUMA & C. SERIES (eds)
330 Noncommutative localization in algebra and topology, A. RANICKI (ed)
331 Foundations of computational mathematics, Santander 2005, L.M PARDO, A. PINKUS, E. SÜLI &
M.J. TODD (eds)
332 Handbook of tilting theory, L. ANGELERI HÜGEL, D. HAPPEL & H. KRAUSE (eds)
333 Synthetic differential geometry (2nd Edition), A. KOCK
334 The Navier–Stokes equations, N. RILEY & P. DRAZIN
335 Lectures on the combinatorics of free probability, A. NICA & R. SPEICHER
336 Integral closure of ideals, rings, and modules, I. SWANSON & C. HUNEKE
337 Methods in Banach space theory, J.M.F. CASTILLO & W.B. JOHNSON (eds)
338 Surveys in geometry and number theory, N. YOUNG (ed)
339 Groups St Andrews 2005 I, C.M. CAMPBELL, M.R. QUICK, E.F. ROBERTSON & G.C. SMITH (eds)
340 Groups St Andrews 2005 II, C.M. CAMPBELL, M.R. QUICK, E.F. ROBERTSON & G.C. SMITH (eds)
341 Ranks of elliptic curves and random matrix theory, J.B. CONREY, D.W. FARMER, F. MEZZADRI &
N.C. SNAITH (eds)
342 Elliptic cohomology, H.R. MILLER & D.C. RAVENEL (eds)
343 Algebraic cycles and motives I, J. NAGEL & C. PETERS (eds)
344 Algebraic cycles and motives II, J. NAGEL & C. PETERS (eds)
345 Algebraic and analytic geometry, A. NEEMAN
346 Surveys in combinatorics 2007, A. HILTON & J. TALBOT (eds)
347 Surveys in contemporary mathematics, N. YOUNG & Y. CHOI (eds)
348 Transcendental dynamics and complex analysis, P.J. RIPPON & G.M. STALLARD (eds)
349 Model theory with applications to algebra and analysis I, Z. CHATZIDAKIS, D. MACPHERSON, A. PILLAY
& A. WILKIE (eds)
350 Model theory with applications to algebra and analysis II, Z. CHATZIDAKIS, D. MACPHERSON, A. PILLAY
& A. WILKIE (eds)
351 Finite von Neumann algebras and masas, A.M. SINCLAIR & R.R. SMITH
352 Number theory and polynomials, J. MCKEE & C. SMYTH (eds)
353 Trends in stochastic analysis, J. BLATH, P. MÖRTERS & M. SCHEUTZOW (eds)
354 Groups and analysis, K. TENT (ed)
355 Non-equilibrium statistical mechanics and turbulence, J. CARDY, G. FALKOVICH & K. GAWEDZKI
356 Elliptic curves and big Galois representations, D. DELBOURGO
357 Algebraic theory of differential equations, M.A.H. MACCALLUM & A.V. MIKHAILOV (eds)
358 Geometric and cohomological methods in group theory, M.R. BRIDSON, P.H. KROPHOLLER &
I.J. LEARY (eds)
359 Moduli spaces and vector bundles, L. BRAMBILA-PAZ, S.B. BRADLOW, O. GARCÍA-PRADA &
S. RAMANAN (eds)
360 Zariski geometries, B. ZILBER
361 Words: Notes on verbal width in groups, D. SEGAL
362 Differential tensor algebras and their module categories, R. BAUTISTA, L. SALMERÓN & R. ZUAZUA
363 Foundations of computational mathematics, Hong Kong 2008, F. CUCKER, A. PINKUS & M.J. TODD (eds)
364 Partial differential equations and fluid mechanics, J.C. ROBINSON & J.L. RODRIGO (eds)
365 Surveys in combinatorics 2009, S. HUCZYNSKA, J.D. MITCHELL & C.M. RONEY-DOUGAL (eds)
366 Highly oscillatory problems, B. ENGQUIST, A. FOKAS, E. HAIRER & A. ISERLES (eds)
367 Random matrices: High dimensional phenomena, G. BLOWER
368 Geometry of Riemann surfaces, F.P. GARDINER, G. GONZÁLEZ-DIEZ & C. KOUROUNIOTIS (eds)
369 Epidemics and rumours in complex networks, M. DRAIEF & L. MASSOULIÉ
370 Theory of p-adic distributions, S. ALBEVERIO, A.YU. KHRENNIKOV & V.M. SHELKOVICH
371 Conformal fractals, F. PRZYTYCKI & M. URBAŃSKI
372 Moonshine: The first quarter century and beyond, J. LEPOWSKY, J. MCKAY & M.P. TUITE (eds)
373 Smoothness, regularity and complete intersection, J. MAJADAS & A. G. RODICIO
374 Geometric analysis of hyperbolic differential equations: An introduction, S. ALINHAC
375 Triangulated categories, T. HOLM, P. JØRGENSEN & R. ROUQUIER (eds)
376 Permutation patterns, S. LINTON, N. RUŠKUC & V. VATTER (eds)
377 An introduction to Galois cohomology and its applications, G. BERHUY
378 Probability and mathematical genetics, N. H. BINGHAM & C. M. GOLDIE (eds)
379 Finite and algorithmic model theory, J. ESPARZA, C. MICHAUX & C. STEINHORN (eds)
380 Real and complex singularities, M. MANOEL, M.C. ROMERO FUSTER & C.T.C WALL (eds)
381 Symmetries and integrability of difference equations, D. LEVI, P. OLVER, Z. THOMOVA &
P. WINTERNITZ (eds)
382 Forcing with random variables and proof complexity, J. KRAJÍČEK
383 Motivic integration and its interactions with model theory and non-Archimedean geometry I, R. CLUCKERS,
J. NICAISE & J. SEBAG (eds)
384 Motivic integration and its interactions with model theory and non-Archimedean geometry II, R. CLUCKERS,
J. NICAISE & J. SEBAG (eds)
385 Entropy of hidden Markov processes and connections to dynamical systems, B. MARCUS, K. PETERSEN &
T. WEISSMAN (eds)
386 Independence-friendly logic, A.L. MANN, G. SANDU & M. SEVENSTER
387 Groups St Andrews 2009 in Bath I, C.M. CAMPBELL et al (eds)
388 Groups St Andrews 2009 in Bath II, C.M. CAMPBELL et al (eds)
389 Random fields on the sphere, D. MARINUCCI & G. PECCATI
390 Localization in periodic potentials, D.E. PELINOVSKY
391 Fusion systems in algebra and topology, M. ASCHBACHER, R. KESSAR & B. OLIVER
392 Surveys in combinatorics 2011, R. CHAPMAN (ed)
393 Non-abelian fundamental groups and Iwasawa theory, J. COATES et al (eds)
394 Variational problems in differential geometry, R. BIELAWSKI, K. HOUSTON & M. SPEIGHT (eds)
395 How groups grow, A. MANN
396 Arithmetic differential operators over the p-adic integers, C.C. RALPH & S.R. SIMANCA
397 Hyperbolic geometry and applications in quantum chaos and cosmology, J. BOLTE & F. STEINER (eds)
398 Mathematical models in contact mechanics, M. SOFONEA & A. MATEI
399 Circuit double cover of graphs, C.-Q. ZHANG
400 Dense sphere packings: a blueprint for formal proofs, T. HALES
401 A double Hall algebra approach to affine quantum Schur–Weyl theory, B. DENG, J. DU & Q. FU
402 Mathematical aspects of fluid mechanics, J.C. ROBINSON, J.L. RODRIGO & W. SADOWSKI (eds)
403 Foundations of computational mathematics, Budapest 2011, F. CUCKER, T. KRICK, A. PINKUS &
A. SZANTO (eds)
404 Operator methods for boundary value problems, S. HASSI, H.S.V. DE SNOO & F.H. SZAFRANIEC (eds)
405 Torsors, étale homotopy and applications to rational points, A.N. SKOROBOGATOV (ed)
406 Appalachian set theory, J. CUMMINGS & E. SCHIMMERLING (eds)
407 The maximal subgroups of the low-dimensional finite classical groups, J.N. BRAY, D.F. HOLT &
C.M. RONEY-DOUGAL
408 Complexity science: The Warwick master’s course, R. BALL, V. KOLOKOLTSOV & R.S. MACKAY (eds)
409 Surveys in combinatorics 2013, S.R. BLACKBURN, S. GERKE & M. WILDON (eds)
410 Representation theory and harmonic analysis of wreath products of finite groups,
T. CECCHERINI-SILBERSTEIN, F. SCARABOTTI & F. TOLLI
411 Moduli spaces, L. BRAMBILA-PAZ, O. GARCÍA-PRADA, P. NEWSTEAD & R.P. THOMAS (eds)
412 Automorphisms and equivalence relations in topological dynamics, D.B. ELLIS & R. ELLIS
413 Optimal transportation, Y. OLLIVIER, H. PAJOT & C. VILLANI (eds)
414 Automorphic forms and Galois representations I, F. DIAMOND, P.L. KASSAEI & M. KIM (eds)
415 Automorphic forms and Galois representations II, F. DIAMOND, P.L. KASSAEI & M. KIM (eds)
416 Reversibility in dynamics and group theory, A.G. O’FARRELL & I. SHORT
417 Recent advances in algebraic geometry, C.D. HACON, M. MUSTAŢĂ & M. POPA (eds)
418 The Bloch–Kato conjecture for the Riemann zeta function, J. COATES, A. RAGHURAM, A. SAIKIA &
R. SUJATHA (eds)
419 The Cauchy problem for non-Lipschitz semi-linear parabolic partial differential equations, J.C. MEYER &
D.J. NEEDHAM
420 Arithmetic and geometry, L. DIEULEFAIT et al (eds)
421 O-minimality and Diophantine geometry, G.O. JONES & A.J. WILKIE (eds)
422 Groups St Andrews 2013, C.M. CAMPBELL et al (eds)
423 Inequalities for graph eigenvalues, Z. STANIĆ
424 Surveys in combinatorics 2015, A. CZUMAJ et al (eds)
425 Geometry, topology and dynamics in negative curvature, C.S. ARAVINDA, F.T. FARRELL &
J.-F. LAFONT (eds)
426 Lectures on the theory of water waves, T. BRIDGES, M. GROVES & D. NICHOLLS (eds)
427 Recent advances in Hodge theory, M. KERR & G. PEARLSTEIN (eds)
428 Geometry in a Fréchet context, C. T. J. DODSON, G. GALANIS & E. VASSILIOU
429 Sheaves and functions modulo p, L. TAELMAN
430 Recent progress in the theory of the Euler and Navier–Stokes equations, J.C. ROBINSON, J.L. RODRIGO, W.
SADOWSKI & A. VIDAL-LÓPEZ (eds)
431 Harmonic and subharmonic function theory on the hyperbolic ball, M. STOLL
432 Topics in graph automorphisms and reconstruction (2nd Edition), J. LAURI & R. SCAPELLATO
433 Regular and irregular holonomic D-modules, M. KASHIWARA & P. SCHAPIRA
434 Analytic semigroups and semilinear initial boundary value problems (2nd Edition), K. TAIRA
435 Graded rings and graded Grothendieck groups, R. HAZRAT
London Mathematical Society Lecture Note Series: 431
Harmonic and Subharmonic

Function Theory on the
Hyperbolic Ball
M A N F R E D S TO L L
University of South Carolina
University Printing House, Cambridge CB2 8BS, United Kingdom
Cambridge University Press is part of the University of Cambridge.

It furthers the University’s mission by disseminating knowledge in the pursuit of
education, learning, and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9781107541481
© Manfred Stoll 2016
This publication is in copyright. Subject to statutory exception
and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.
First published 2016
Printed in the United Kingdom by Clays, St Ives plc
A catalogue record for this publication is available from the British Library.
Library of Congress Cataloguing in Publication Data
Names: Stoll, Manfred.
Title: Harmonic and subharmonic function theory on the hyperbolic ball /
Manfred Stoll, University of South Carolina.
Description: Cambridge : Cambridge University Press, 2016. | Series: London
Mathematical Society lecture note series ; 431 | Includes bibliographical
references and index.
Identifiers: LCCN 2015049530 | ISBN 9781107541481 (pbk.)
Subjects: LCSH: Harmonic functions. | Subharmonic functions. |
Hyperbolic spaces.
Classification: LCC QA405 .S76 2016 | DDC 515/.53–dc23
LC record available at http://lccn.loc.gov/2015049530
ISBN 978-1-107-54148-1 Paperback
Cambridge University Press has no responsibility for the persistence or accuracy
of URLs for external or third-party Internet Web sites referred to in this publication
and does not guarantee that any content on such Web sites is, or will remain,
accurate or appropriate.
To Mary Lee
Contents
Preface page xi
1 Möbius Transformations 1
1.1 Notation 1
1.2 Inversion in Spheres and Planes 2
1.3 Möbius Transformations 4
2 Möbius Self-Maps of the Unit Ball 6
2.1 Möbius Transformations of B 6
2.2 The Hyperbolic Metric on B 9
2.3 Hyperbolic Half-Space H 12
2.4 Exercises 15
3 The Invariant Laplacian, Gradient, and Measure 17
3.1 The Invariant Laplacian and Gradient 17
3.2 The Fundamental Solution of Δh 19
3.3 The Invariant Measure on B 21
3.4 The Invariant Convolution on B 24
3.5 Exercises 27
4 H-Harmonic and H-Subharmonic Functions 31
4.1 The Invariant Mean-Value Property 31
4.2 The Special Case n = 2 35
4.3 H-Subharmonic Functions 37
4.4 Properties of H-Subharmonic Functions 41
4.5 Approximation by C∞ H-Subharmonic Functions 45
4.6 The Weak Laplacian and Riesz Measure 48
4.7 Quasi-Nearly H-Subharmonic Functions 51
4.8 Exercises 56
vii
viii Contents
5 The Poisson Kernel and Poisson Integrals 59

5.1 The Poisson Kernel for Δh 59
5.2 Relationship between the Euclidean and Hyperbolic Poisson
Kernel 62
5.3 The Dirichlet Problem for B 64
5.4 The Dirichlet Problem for Br 68
5.5 Eigenfunctions of Δh 70
5.6 The Poisson Kernel on H 76
5.7 Exercises 78
6 Spherical Harmonic Expansions 82
6.1 Dirichlet Problem for Spherical Harmonics 83
6.2 Zonal Harmonic Expansion of the Poisson Kernel 86
6.3 Spherical Harmonic Expansion of H-Harmonic Functions 90
6.4 Exercises 94
7 Hardy-Type Spaces of H-Subharmonic Functions 96
7.1 A Poisson Integral Formula for Functions in Hp , 1 ≤ p ≤ ∞ 97
7.2 Completeness of Hp , 0 < p ≤ ∞ 101
7.3 H-Harmonic Majorants for H-Subharmonic Functions 103
7.4 Hardy–Orlicz Spaces of H-Subharmonic Functions 109
7.5 Exercises 112
8 Boundary Behavior of Poisson Integrals 114
8.1 Maximal Functions 114
8.2 Non-tangential and Radial Maximal Function 120
8.3 Fatou’s Theorem 125
8.4 A Local Fatou Theorem for H-Harmonic Functions 127
8.5 An L p Inequality for Mα f for 0 < p ≤ 1 131
8.6 Example 134
8.7 Exercises 137
9 The Riesz Decomposition Theorem for H-Subharmonic
Functions 139
9.1 The Riesz Decomposition Theorem 140
9.2 Applications of the Riesz Decomposition Theorem 143
9.3 Integrability of H-Superharmonic Functions 149
9.4 Boundary Limits of Green Potentials 155
9.5 Non-tangential Limits of H-Subharmonic Functions 162
9.6 Exercises 169
Contents ix
10 Bergman and Dirichlet Spaces of H-Harmonic Functions 173

p p
10.1 Properties of Dγ and Bγ 174
10.2 Möbius Invariant Spaces 178
p p
10.3 Equivalence of Bγ and Dγ for γ > (n − 1) 180
p p
10.4 Integrability of Functions in Bγ and Dγ 186
10.5 Integrability of Eigenfunctions of Δh 193
10.6 Three Theorems of Hardy and Littlewood 198
10.7 Littlewood–Paley Inequalities 205
10.8 Exercises 211
References 216
Index of Symbols 221
Index 223
Preface
The intent of these notes is to provide a detailed and comprehensive treatment

of harmonic and subharmonic function theory on hyperbolic space in Rn .
Although our primary emphasis will be in the setting of the unit ball B with
hyperbolic metric ds given by
2|dx|
ds = , (1)
1 − |x|2
we will also consider the analogue of many of the results in the hyperbolic
half-space H. Undoubtedly some of the results are known, either in the setting
of rank one noncompact symmetric spaces (e.g. [38]), or more generally, in
Riemannian spaces (e.g. [13]). An excellent introduction to harmonic function
theory on noncompact symmetric spaces can be found in the survey article [47]
by A. Koranyi. The 1973 paper by K. Minemura [57] provides an introduction
to harmonic function theory on real hyperbolic space considered as a rank one
noncompact symmetric space. Other contributions to the subject area in this
setting will be indicated in the text.
With the goal of making these notes accessible to a broad audience,
our approach does not require any knowledge of Lie groups and only a
limited knowledge of differential geometry. The development of the theory
is analogous to the approach taken by W. Rudin [72] and by the author [84]
in their development of Möbius invariant harmonic function theory on the
hermitian ball in Cn . Although our primary emphasis is on harmonic function
theory on the ball, we do include many relevant results for the hyperbolic
upper half-space H, both in the text and in the exercises. With only one or
two exceptions, the notes are self-contained with the only prerequisites being
a standard beginning graduate course in real analysis.
In Chapter 1 we provide a brief review of Möbius transformation in Rn .
This is followed in Chapter 2 by a characterization of the group M(B) of
xi
xii Preface
Möbius self-maps of the unit ball B in Rn . As in [72] we define a family {ϕa :

a ∈ B} of Möbius transformations of B satisfying ϕa (0) = a, ϕa (a) = 0,
and ϕa (ϕa (x)) = x for all x ∈ B. Furthermore, for every ψ ∈ M(B), it is
proved that there exists a ∈ B and an orthogonal transformation A such that
ψ = Aϕa . When n = 2, the mappings ϕa correspond to the usual analytic
Möbius transformations of the unit disc D given by
a−z
ϕa (z) = . (2)
1 − az
Some of the properties of the mappings {ϕa } and of functions in M(B) are
developed in Section 2.1. In this chapter we also introduce the hyperbolic
metric in B and in the hyperbolic half-space H. Most of the results of these
two sections are contained in the works of L. V. Ahlfors [4], [5] , and the text
by A. F. Beardon [11].
In Chapter 3 we derive the Laplacian, gradient, and measure on B that are
invariant under M(B). Even though the formula for the Laplacian can be
derived from the hyperbolic metric, we will follow the approach of W. Rudin
[72, Chapter 4]. For f ∈ C2 (B) we define Δh f by
Δh f (a) = Δ( f ◦ ϕa )(0),
where Δ is the usual Laplacian in Rn . The operator Δh is shown to satisfy
Δh ( f ◦ ψ)(x) = (Δh f )(ψ(x)) for all ψ ∈ M(B). Furthermore, an explicit
computation gives
Δh f (x) = (1 − |x|2 )2 Δf (x) + 2(n − 2)(1 − |x|2 )x, ∇f (x),
where ∇f is the Euclidean gradient of the function f . In this chapter it is also
proved that the Green’s function for Δh is given by Gh (x, y) = g(|ϕx (y)|),
where g is the radial function on B defined by

1 1 (1 − s2 )n−2
g(r) = ds.
n r sn−1
In Theorem 3.3.1 we prove that for ψ ∈ M(B), the Jacobian Jψ of the
mapping ψ satisfies
(1 − |ψ(x)|2 )n
|Jψ (x)| = .
(1 − |x|2 )n
From this it now follows that the Möbius invariant measure τ on B is given by
dτ (x) = (1 − |x|2 )−n dν(x),
where ν is the normalized volume measure on B. In the exercises we develop
the invariant Laplacian, Green’s function, and invariant measure on H.
Preface xiii
A real-valued C2 function f on B is defined to be either H-harmonic or

H-subharmonic on B depending on whether Δh f = 0 or Δh f ≥ 0. It is well
known that a continuous function f is harmonic in the unit disc D if and only
if for all r, 0 < r < 1, and w ∈ D,
2π
1
f (w) = f (ϕw (reit )) dt, (3)
2π 0
where ϕw is the Möbius transformation of D given by (2). The above is called
the invariant mean-value property. One of the first results proved in Chapter
4 is the following analogue of the invariant mean-value property: A real-valued
C2 function f is H-subharmonic on B if and only if for all a ∈ B and 0 < r < 1,

f (a) ≤ f (ϕa (rt))dσ (t), (4)
S
with equality if and only if f is H-harmonic on B. In the above, S is the unit

sphere in Rn , σ is normalized surface measure on S, and ϕa is the Möbius
transformation of B mapping 0 to a with ϕa (ϕa (x)) = x. The integral in (4) is
an average of f over the hyperbolic or non-Euclidean sphere {ϕa (rt) : t ∈ S}
whose hyperbolic center is a. Inequality (4) is then used in Section 4.3 to
extend the definition of H-subharmonic to the class of upper semicontinu-
ous functions on B. The remainder of the chapter is devoted to extending
some of the standard results about subharmonic functions to H-subharmonic
functions on B. We conclude the chapter with a discussion of quasi-nearly
H-subharmonic functions and prove several inequalities involving these func-
tions that will prove useful later in the text.
The Poisson kernel Ph for Δh is introduced in Chapter 5. In Section 5.1 we
prove using Green’s formula that for (a, t) ∈ B × S,
Ph (a, t) = − lim nrn−1 (1 − r2 )2−n ∇Ga (rt), t,

r→1
where Ga (rt) = Gh (a, rt) is the Green’s function for Δh . This immediately
gives
n−1
1 − |x|2
Ph (x, t) = , (x, t) ∈ B × S.
|x − t|2
The standard results for Poisson integrals of continuous functions are included
in Section 5.3, and in Section 5.2 we prove a result of P. Jaming [43] that
provides an integral representation of the Euclidean Poisson kernel in terms of
the hyperbolic Poisson kernel. In Section 5.5 we investigate the eigenfunctions
of Δh . We close the section with a brief discussion of the Poisson kernel
on H.
xiv Preface
In Chapter 6 we consider the spherical harmonic expansions of H-harmonic

functions. One of the key results of this section is that if pα is a spherical
harmonic of degree α on S, then the Poisson integral Ph [pα ] of pα is given by

Ph [pα ](x) = |x|α Sn,α (|x|)pα |x|
x
,
where Sn,α is given by a hypergeometric function. Interestingly, when n is

even, Sn,α (r) is simply a polynomial in r of degree n − 2. These results are
then used to show how the Poisson integral Ph [q] can be computed for any
polynomial q on S. As an example, in R4 , the H-harmonic function with
boundary values t12 is given by Ph [t12 ](x) = 14 + (2 − |x|2 )(x12 − 14 |x|2 ). In
contrast, the Euclidean harmonic function h with boundary values t12 is given
by h(x) = 14 (1 − |x|2 ) + x12 . Finally, in Section 6.3 we follow the methods of
P. Ahern, J. Bruna, and C. Cascante [2] to derive the spherical harmonic
expansion of H-harmonic functions on B.
Chapter 7 is devoted to the study of Hardy and Hardy–Orlicz type spaces of
H-harmonic and H-subharmonic functions on B. In Chapter 8, we study the
boundary behavior of Poisson integrals on B. This chapter contains many of
the standard results concerning non-tangential and radial maximal functions.
In addition to proving the usual Fatou theorem (Theorem 8.3.3) concerning
non-tangential limits of Poisson integrals of measures, we also include a proof
of a local Fatou theorem of I. Privalov [68] for H-harmonic functions on B.
The Riesz decomposition theorem for H-subharmonic functions is proved
in Chapter 9. The main result of this chapter (Corollary 9.1.3) proves that if
f is H-subharmonic on B and f has an H-harmonic majorant, then

f (x) = Ff (x) − Gh (x, y)dμf (y),
B
where μf is the Riesz measure of f and Ff is the least H-harmonic majorant
of f . In Section 9.2 we include several applications of the Riesz decomposition
theorem, including a Hardy–Stein identity for non-negative H-subharmonic
functions for which f p , p ≥ 1, has an H-harmonic majorant on B. In Section
9.3 we extend a result of D. H. Armitage [8] concerning the integrability of
non-negative superharmonic functions. We conclude the chapter by proving
that invariant Green potentials of measures have radial limit zero almost
everywhere on S, and provide an example of a measure μ for which the Green
potential of μ has non-tangential limit +∞ almost everywhere on S.
Finally, in Chapter 10 we introduce and investigate basic properties of
weighted Bergman- and Dirichlet-type spaces of H-harmonic functions
p p
on B, denoted respectively by Bγ and Dγ . These spaces consist of the set
of H-harmonic functions f on B for which f , respectively |∇ h f |, are in
Preface xv
Lp ((1 − |x|2 )γ dτ (x)), 0 0, where τ is the invariant measure
on B and ∇ h is the invariant gradient on B. One of the main results of this
p p
chapter is that if γ > (n − 1), then f ∈ Bγ if and only if f ∈ Dγ for all
p, 0 < p < ∞. In Section 10.4 we investigate the integrability of functions in
p p
Bγ and Dγ for γ ≤ (n − 1). This chapter also contains a discussion of Möbius
invariant spaces of H-harmonic functions and the Berezin transform on B.
We conclude the chapter with three theorems of Hardy and Littlewood for H-
harmonic functions, and the Littlewood–Paley inequalities for H-subharmonic
functions.
At the end of each chapter, I have included a set of exercises dealing with
the topics discussed. Many of these problems involve routine computations
and inequalities not included in the text. They also provide examples relevant
to the topics of the chapter. Also included are problems whose solutions may
be suitable for possible publication. The latter are marked with an asterisk.
Acknowledgments
The preliminary draft of these notes comprising most of Chapters 1–3 and parts
of Chapters 5 and 6 was written while the author was on sabbatical leave at the
CRM at the University of Montreal during fall 1999. Since 1999 there has been
considerable research activity on hyperbolic function theory on B. Included in
this has been the work of P. Jaming [41], [42], [43], S. Grellier and P. Jaming
[31], M. Jevtić [44], and the author [89], among others.
I would like to thank my good friend Paul Gauthier at the University of
Montreal for providing the motivation for this project. His many questions
about harmonic functions on real hyperbolic space, to which I did not know
the answers, encouraged me to learn more about this interesting subject. I am
also indebted to Paul for using the preliminary draft of these notes in one
of his courses and for pointing out a number of errors. I would also like to
express my appreciation to John Taylor at McGill for providing me with a
copy of his notes on rank one symmetric spaces, and to thank the Director and
Staff at the CRM for their hospitality during my visit. Finally, I would like to
thank Wolfgang Woess Universität Graz, for bringing my original manuscript
to the attention of Sam Harrison, Editor at Cambridge University Press, and
the staff at Cambridge University Press and SPi Global for their assistance in
the completion of the project.
Manfred Stoll
Columbia, SC
stoll@math.sc.edu
1
Möbius Transformations
In this chapter we provide a brief review of Möbius transformations on

n-dimensional Euclidean space Rn (n ≥ 2). A good reference for these topics
is the monograph by A. F. Beardon [11]. First, however, we begin with a review
of notation that will be used throughout these notes.
1.1 Notation
n
For x, y ∈ Rn we let x, y = j=1 xj yj denote the usual inner product on Rn
√
and |x| = x, x the length of the vector x. For a ∈ Rn and r > 0, the ball
B(a, r) and sphere S(a, r) are given respectively by
B(a, r) = {x ∈ Rn : |x − a| < r},
S(a, r) = {x ∈ Rn : |x − a| = r}.
The unit ball and unit sphere with center at the origin will simply be denoted
by B and S respectively.1 The one-point compactification of Rn , denoted R̂n ,
is obtained by appending the point ∞ to Rn . A subset U of R̂n = Rn ∪ {∞} is
open if it is an open subset of Rn , or if U is the complement in R̂n of a compact
subset C of Rn . With this topology R̂n is compact.
For a subset D of Rn , D denotes the closure of D, Int (D) the interior of D,
∂D the boundary of D, and D the complement of D in Rn . Also if E and F are
sets, E \ F denotes the complement of F in E, that is, E \ F = E ∩ F.
The study of functions of n-variables is simplified with the use of multi-index
notation. For an ordered n-tuple α = (α1 , . . . , αn ), where each αj is a non-
negative integer, the following notational conventions will be used throughout:
1 If we wish to emphasize the dimension n, we will use the notation B and S to denote the unit
n n
ball and sphere in Rn .
1
2 Möbius Transformations
|α| = α1 + · · · + αn , α! = α1 ! · · · αn !, xα = x1α1 · · · xnαn ,

and
∂ |α| f
Dα f = .
∂x1α1 · · · ∂xnαn
If is an open subset of Rn , we denote by Ck (), k = 0, 1, 2, . . . the set of

real-valued (or complex-valued) functions f on for which Dα f exists and is
continuous for all multi-indices α with |α| ≤ k. Thus C0 (), or simply C(),
denotes the set of real-valued (or complex-valued) continuous functions on ,
and C∞ () the set of infinitely differentiable functions on . Also, the set of
functions f ∈ Ck () for which Dα f , |α| ≤ k, has a continuous extension to
will be denoted by Ck (). If f : → R, then the support of f , denoted
supp f , is defined as
supp f = {x ∈ : f (x) = 0}.
The set of continuous functions on with compact support will be denoted by
Cc (). The notations Cck () and Cc∞ () have the obvious meanings.
A linear transformation A : Rn → Rn is said to be orthogonal if |Ax| = |x|
for all x ∈ Rn . The set of orthogonal transformations of Rn will be denoted by
O(n). If A is represented by the n × n matrix (ai, j ), then A is orthogonal if and
only if

n
1 i = j,
ai,k aj,k = δi, j =
k=1
0, i = j.
If ψ(x) = (ψ1 (x), . . . , ψn (x)) is a C1 mapping of an open subset of Rn into

Rn , then the derivative ψ (x) is the n × n matrix given by

∂ψi n
ψ (x) = ,
∂xj i,j=1
and the Jacobian Jψ of the transformation ψ is given by Jψ (x) = det ψ (x).
1.2 Inversion in Spheres and Planes

Definition 1.2.1 The inversion2 (or reflection) in the sphere S(a, r) is the
function φ(x) defined by
2 Although we will mainly be interested in the case n ≥ 2, the formulas for inversions in spheres
and planes are still meaningful when n = 1.
1.2 Inversion in Spheres and Planes 3
2
r
φ(x) = a + (x − a). (1.2.1)
|x − a|
The inversion in the unit sphere S is the mapping φ(x) = x∗ where
⎧ x
⎪
⎪ x = 0, ∞,
⎨ |x|2
∗
x = 0 x = ∞,
⎪
⎪
⎩
∞ x = 0.
Thus (1.2.1) can now be rewritten as
φ(x) = a + r2 (x − a)∗ .
The reflection φ(x) is not defined at x = a. Since |φ(x)| → ∞ as x → a we
set φ(a) = ∞. Also, since lim|x|→∞ |φ(x) − a| = 0, we set φ(∞) = a. Thus
φ is defined on all of R̂n , and it is easily shown that φ is continuous in the
topology of R̂n . A straightforward computation also shows that φ(φ(x)) = x
for all x ∈ R̂n . Thus φ is a one-to-one continuous map of R̂n onto R̂n satisfying
φ(x) = x if and only x ∈ S(a, r).
In addition to reflection in a sphere we also have reflection in a plane. For
a ∈ Rn , a = 0, and t ∈ R, the plane P(a, t) is defined by
P(a, t) = {x ∈ Rn : x, a = t}.
By convention ∞ belongs to every plane P(a, t).
Definition 1.2.2 The inversion (or reflection) in the plane P(a, t) is the
function ψ(x) defined by
ψ(x) = x + λa,
where λ ∈ R is chosen so that 12 (x + ψ(x)) ∈ P(a, t).
Solving for λ gives
ψ(x) = x − 2[x, a − t]a∗ , x ∈ Rn . (1.2.2)
For the mapping ψ we have
|ψ(x)|2 = |x|2 + O(|x|),
and as a consequence lim|x|→∞ |ψ(x)| = ∞. Thus as above we define
ψ(∞) = ∞. With this definition the mapping ψ again satisfies ψ(ψ(x)) = x
for all x ∈ R̂n . Thus ψ is a one-to-one continuous map of R̂n onto itself with
ψ(x) = x if and only if x ∈ P(a, t). It is well known that each inversion (in
a sphere or a plane) is orientation-reversing and conformal (see [11, Theorem
3.1.6]).
4 Möbius Transformations
1.3 Möbius Transformations

Definition 1.3.1 A Möbius transformation of R̂n is a finite composition of
inversions in spheres or planes.
Clearly the composition of two Möbius transformations is again a Möbius

transformation, as is the inverse of a Möbius transformation. The group of
Möbius transformations on R̂n is called the general Möbius group and is
denoted by GM (R̂n ). Although not immediately obvious, both translation
and magnification by a constant are Möbius transformations. The translation
x → x + a, a ∈ Rn , is the composition of inversion in the plane x, a = 0
followed by inversion in the plane x, a = 12 |a|2 . Likewise, the magnification
or scalar multiplication x → kx, k > 0, is also a Möbius√transformation in
that it is the inversion in S followed by the inversion in S(0, k). Furthermore,
every Euclidean isometry of Rn is a composition of at most n + 1 reflections
in planes ([11, Theorem 3.1.3]).
We conclude this section by showing that every Möbius transformation
maps a sphere or plane onto a sphere or plane. We will use the term “sphere”
to denote either a sphere of the form S(a, r) or a plane P(a, t). Since every
inversion ψ in a plane P(a, t) can be written as
ψ(x) = x + λa,
the mapping ψ clearly maps a “sphere” onto a “sphere.” To show that an

inversion φ in a sphere S(a, r) preserves “spheres,” it suffices to show that
the mapping x∗ preserves “spheres.”
For any set E ⊂ Rn , we let E∗ = {x∗ : x ∈ E}. A set E ⊂ Rn is a sphere or a
plane if and only if
E = {x ∈ Rn : b|x|2 − 2x, a + c = 0},
where b and c are real and a ∈ Rn . By convention, ∞ satisfies this equation

if and only if b = 0, that is, E is a plane. Now it is easily seen that E∗ has the
same form with the roles of b and c reversed. Finally, it is an easy exercise to
show that for a ∈ Rn and r > 0,
⎧
⎨S a r
, if 0 ∈ S(a, r),
S∗ (a, r) = (|a|2 − r2 ) ||a|2 − r2 | (1.3.1)
⎩
P(a, 12 ) if 0 ∈ S(a, r).
1.3 Möbius Transformations 5
We conclude this section with one more useful formula that will be required
later. If φ is inversion in the sphere S(a, r), then a straightforward computation
gives
r2 |y − x|
|φ(y) − φ(x)| = . (1.3.2)
|x − a||y − a|
For details on the above the reader is referred to [11].
2
Möbius Self-Maps of the Unit Ball
In this chapter we will provide a characterization of the Möbius

transformations of R̂n mapping the unit ball B onto B that is similar to
the characterization of the Möbius mappings of the unit disc in C onto itself.
In the complex plane C, every analytic Möbius transformation ψ mapping
the unit disc D onto itself can be written as ψ(z) = eiθ ϕw (z), where for w ∈ D,
w−z
ϕw (z) = .
1 − wz
The mappings ϕw (z) satisfy ϕw (0) = w, ϕw (w) = 0, and ϕw (ϕw (z)) = z for all
z ∈ D. Furthermore, the mapping ϕw (z) also satisfies
(1 − |z|2 )(1 − |w|2 )

1 − |ϕw (z)|2 = .
|1 − wz|2
2.1 Möbius Transformations of B

In this section we define an analogous family of Möbius transformations
{ϕa : a ∈ B} mapping B onto B having the property that every Möbius
transformation ψ mapping B onto itself can be written as ψ = A ◦ ϕa , where
a ∈ B and A ∈ O(n). For a ∈ B, we first set
ψa (x) = a + (1 − |a|2 )(a − x)∗ . (2.1.1)
Since the mapping ψa is a composition of Möbius transformations, ψa is a

Möbius transformation of R̂n mapping 0 to a∗ and a to ∞. By a straightforward
computation we have
|a − x|2 + (1 − |a|2 )(1 − |x|2 )

|ψa (x)|2 = , (2.1.2)
|a − x|2
6
2.1 Möbius Transformations of B 7
and as a consequence
(1 − |a|2 )(1 − |x|2 )
|ψa (x)|2 − 1 = . (2.1.3)
|x − a|2
From the above it follows immediately that ψa maps B onto R̂n \ B.
We now define the mapping ϕa by
ψa (x)
ϕa (x) = ψa (x)∗ = . (2.1.4)
|ψa (x)|2
If we set1
ρ(x, a) = |x − a|2 + (1 − |a|2 )(1 − |x|2 ) = |a|2 |a∗ − x|2 , (2.1.5)
then the mapping ϕa can be expressed as
a|x − a|2 + (1 − |a|2 )(a − x)
ϕa (x) = . (2.1.6)
ρ(x, a)
As a consequence of (2.1.3)
(1 − |x|2 )(1 − |a|2 )
1 − |ϕa (x)|2 = . (2.1.7)
ρ(x, a)
Thus ϕa is a Möbius transformation mapping B onto B with ϕa (0) = a and
ϕa (a) = 0. That ϕa maps B onto B follows immediately from the fact that ψa
maps B onto R̂n \ B and that x∗ maps R̂n \ B onto B. We will shortly prove that
ϕa also satisfies ϕa (ϕa (x)) = x for all x ∈ B. In the unit disk D, for z, w ∈ D,
ρ(z, w) = |1 − wz|2 and the mappings ϕw (z) as defined by (2.1.6) are precisely
the functions (w − z) (1 − wz).
One of the advantages of the mappings ϕa is that the function (a, x) → ϕa (x)
is not only continuous on B × B but also differentiable in each of the variables.
At this point we will include several computations involving derivatives of the
mappings ϕa that will be required in the proof of Theorem 2.1.2 and also later
in the sequel. Let yj (x) denote the jth coordinate of y(x) = ϕa (x). Then by
straightforward computations we have
∂yj ∂yj −δi,j
(0) = −δi, j (1 − |a|2 ), (a) = , (2.1.8)
∂xi ∂xi (1 − |a|2 )
∂ 2 yj
(0) = (1 − |a|2 )[2aj − 4ai δi,j ]. (2.1.9)
∂xi2
Hence
ϕa (0) = −(1 − |a|2 ) I and ϕa (a) = −(1 − |a|2 )−1 I,
where I is the n × n identity matrix.
1 In [11] the function √ρ(x, a) is denoted by [x, a].
8 Möbius Self-Maps of the Unit Ball
Since the following theorem is well known, we state it without proof.

A proof may be found in [11, Theorem 3.4.1].
Theorem 2.1.1 Let ψ be a Möbius transformation of R̂n satisfying ψ(0) = 0

and ψ(B) = B. Then ψ(x) = Ax for some orthogonal transformation A.
We denote by M(B) the set of all Möbius transformations of B onto B.

It is an immediate consequence of the following theorem that the set M(B)
forms a group called the Möbius group of B.
Theorem 2.1.2 For a ∈ B, let ϕa be defined by (2.1.6). Then

(a) ϕa is a one-to-one Möbius mapping of B onto B satisfying
ϕa (0) = a, ϕa (a) = 0, and ϕa (ϕa (x)) = x
for all x ∈ B.
(b) If ψ ∈ M(B), then there exists an orthogonal transformation A and
a ∈ B such that ψ(x) = Aϕa (x).
Proof. To prove (a) it only remains to be shown that ϕa (ϕa (x)) = x for all
x ∈ B. Set ψ(x) = (ϕa ◦ ϕa )(x). Then ψ is a Möbius transformation of R̂n
mapping B onto B satisfying ψ(0) = 0. Thus ψ(x) = Ax for some orthogonal
transformation A. But then A = ψ (0). On the other hand, by the chain rule
and Equations (2.1.8)
ψ (0) = ϕa (a)ϕa (0) = I.
Hence A = I and thus ϕa (ϕa (x)) = x for all x ∈ B.

(b) Let ψ ∈ M(B) and let a = ψ −1 (0). Then ψ ◦ ϕa is a Möbius
transformation of B that fixes the origin. Thus ψ ◦ ϕa (x) = Ax for some
orthogonal transformation A. But then by (a) we have ψ(x) = Aϕa (x).
Prior to introducing the hyperbolic metric on B we prove an identity for

mappings ψ ∈ M(B).
Theorem 2.1.3 If ψ ∈ M(B), then for all x, y ∈ B,
|ψ(x) − ψ(y)|2 |x − y|2

= .
(1 − |ψ(x)|2 )(1 − |ψ(y)|2 ) (1 − |x|2 )(1 − |y|2 )
Proof. Although this identity could be proved using the mappings ϕa , it
appears to be easier to use the mappings σa definedas follows: for a ∈ B,
a = 0, let σa denote the inversion in the sphere S(a∗ , |a∗ |2 − 1), that is,
σa (x) = a∗ + (|a∗ |2 − 1)(x − a∗ )∗ . (2.1.10)

Then σa (0) = a, σa (a) = 0, and since σa is an inversion, σa (σa (x)) = x for all
x ∈ R̂n . Also, by identity (1.3.2),
(|a∗ |2 − 1)2 |x − a|2
|σa (x)|2 = |σa (x) − σa (a)|2 = ,
|x − a∗ |2 |a − a∗ |2
which upon simplification gives
|x − a|2
|σa (x)|2 = .
ρ(x, a)
Thus
(1 − |x|2 )(1 − |a|2 )
1 − |σa (x)|2 = . (2.1.11)
ρ(x, a)
Hence σa ∈ M(B).2 Again by (1.3.2) we obtain
(1 − |a|2 )2 |x − y|2
|σa (x) − σa (y)|2 = .
ρ(x, a)ρ(y, a)
Combining this with (2.1.11) now gives
|σa (x) − σa (y)|2 |x − y|2
= .
(1 − |σa (x)|2 )(1 − |σa (y)|2 ) (1 − |x|2 )(1 − |y|2 )
Finally, as in the proof of Theorem 2.1.2(b), every ψ ∈ M(B) can be expressed
as ψ(x) = Aσa (x) for some A ∈ O(n) and a ∈ B. From this the result now
follows.
As a consequence of the identity in Theorem 2.1.3, for ψ ∈ M(B),
|ψ(y) − ψ(x)| 1 − |ψ(x)|2
lim = . (2.1.12)
y→x |y − x| 1 − |x|2
This result will be required in proving the M-invariance of the hyperbolic
metric on B.
2.2 The Hyperbolic Metric on B

The element of arclength ds for the hyperbolic metric dh on B is given by
2|dx|
ds = . (2.2.1)
1 − |x|2
Thus if γ : [0, 1] → B is a C1 curve in B, the hyperbolic length L(γ ) of γ is
given by
2 Even though the mappings σ are easier to work with, they have the disadvantage that
a
lima→0 σa (x) does not exist.
1 2|γ (t)| dt
L(γ ) = ,
0 1 − |γ (t)|2
and for a, b ∈ B, the hyperbolic distance dh (a, b) between a and b is
defined by
dh (a, b) = inf L(γ ),
γ
where the infimum is taken over all C1 curves γ : [0, 1] → B with γ (0) = a
and γ (1) = b. From this we immediately obtain that for x ∈ B,

1 + |x|
dh (0, x) = log . (2.2.2)
1 − |x|
Theorem 2.2.1 For all ψ ∈ M(B) and a, b ∈ B, dh (ψ(a), ψ(b)) = dh (a, b).
Proof. To prove the theorem it suffices to prove that L(ψ ◦ γ ) = L(γ ) for all
C1 curves γ and ψ ∈ M(B). If we set σ (t) = ψ(γ (t)), then σ is a C1 curve
and

σ (t + h) − σ (t)

|σ (t)| = lim
h→0 h
|ψ(γ (t + h)) − ψ(γ (t))|
= lim ,
h→0 |h|
which by (2.1.12)

1 − |ψ(γ (t))|2
= |γ (t)| .
1 − |γ (t)|2
Thus
|σ (t)| |γ (t)|
= .
1 − |σ (t)|2 1 − |γ (t)|2
From this it now follows that L(σ ) = L(γ ), thus proving the claim.
As a consequence of (2.2.2) and Theorem 2.2.1, for a, b ∈ B,

1 + |ϕa (b)|
dh (a, b) = dh (0, ϕa (b)) = log . (2.2.3)
1 − |ϕa (b)|
Some brief computations also give
|a − b|2
sinh2 12 dh (a, b) = , (2.2.4)
(1 − |a|2 )(1 − |b|2 )
or

|a − b|
dh (a, b) = 2 Arcsinh
(1 − |a|2 )(1 − |b|2 )

|a − b|
= 2 Arctanh √ ,
ρ(a, b)
where ρ(a, b) is defined by (2.1.5).
For 0 < r < 1 we will denote B(0, r) and S(0, r) by Br and Sr respectively.
As in [72, p. 29], for a ∈ B and 0 < r < 1, we let E(a, r) = ϕa (Br ). Since ϕa
is an involution,

E(a, r) = {x ∈ B : |ϕa (x)| < r} = x ∈ B : dh (a, x) < log 1+r
1−r . (2.2.5)
Thus E(a, r) is a hyperbolic ball with hyperbolic center a and hyperbolic

radius

1+r
ρ = log = 2 Arctanh r.
1−r
However, E(a, r) is also a Euclidean ball whose center and radius are given in
the following theorem.
Theorem 2.2.2 For a ∈ B and 0 < r < 1, E(a, r) = B(ca , ρa ) where

(1 − r2 )a r(1 − |a|2 )
ca = and ρa = .
(1 − |a|2 r2 ) (1 − |a|2 r2 )
Proof. To prove the result we first determine the image of Sr under the
mapping ϕa . Let ψa be the mapping defined by (2.1.1). If |a| = r, then by
(1.3.1)

(1 − r2 )a (1 − |a|2 )r
ψa (Sr ) = S , .
(|a|2 − r2 ) ||a|2 − r2 |
Since ϕa = ψa∗ , using (1.3.1) again gives

(1 − r2 )a (1 − |a|2 )r
ϕa (Sr ) = S , . (2.2.6)
(1 − r2 |a|2 ) (1 − r2 |a|2 )
On the other hand, if |a| = r, ψa (Sr ) = P(a, 12 (1 + |a|2 )). Taking the inversion
of ψa (Sr ) gives
ϕa (Sr ) = {x : (1 + |a|2 )|x|2 − 2a, x = 0}.
This, however, is simply the equation of the sphere given in (2.2.6) with
r = |a|. Since ϕa is continuous and a ∈ B(ca , ρa ), ϕa (Br ) ⊂ B(ca , ρa ). Finally,
since ϕa is an involution, ϕa (Br ) = B(ca , ρa ).
We conclude this section with the pseudo-hyperbolic metric on B. For

a, b ∈ B set
dph (a, b) = |ϕa (b)|. (2.2.7)
Theorem 2.2.3 dph is a metric on B.

Proof. Since ϕa is one-to-one the only fact that needs to be proved is that dph
satisfies the triangle inequality, that is,
|ϕa (b)| ≤ |ϕa (x)| + |ϕb (x)|
for all a, b, x ∈ B. We first note that dph (a, b) = f (dh (a, b)) where
et − 1
f (t) = , t ∈ [0, ∞).
et + 1
Since f is increasing on [0, ∞), we have
dph (a, b) = f (dh (a, b)) ≤ f (dh (a, x) + dh (x, b)).
Finally, since
f (x + y) ≤ f (x) + f (y) for all x, y ∈ [0, ∞), (2.2.8)
we have
f (dh (a, x) + dh (x, b)) ≤ f (dh (a, x)) + f (dh (x, b)) = dph (a, x) + dph (x, b),
which proves the result. For the proof of (2.2.8) we first note that
ex − 1 ey − 1 2(ex+y − 1)
+ = .
ex + 1 ey + 1 (ex + 1)(ey + 1)
Since ex+y − ex − ey + 1 = (ex − 1)(ey − 1) ≥ 0, we have
2(ex+y + 1) − (ex + 1)(ey + 1) ≥ 0.
Therefore
2 1
≥ x+y
(ex + 1)(ey + 1) e +1
which proves (2.2.8) and thus the result.
2.3 Hyperbolic Half-Space H

In this final section we briefly consider hyperbolic half-space in Rn .
2.3 Hyperbolic Half-Space H 13
Definition 2.3.1 For n ≥ 2, the upper half-space H or Hn in Rn is defined by
H = {x ∈ Rn : xn > 0}.
For each x ∈ Rn−1 , let x̃ ∈ Rn be defined by
x̃ = (x, 0) = (x1 , . . . , xn−1 , 0).
For each inversion φ on R̂n−1 , we define an inversion φ̃ acting on R̂n as

follows. If φ is an inversion in S(a, r), then φ̃ is the inversion in S(ã, r); if
φ is an inversion in P(a, t), then φ̃ is the inversion in P(ã, t). If x ∈ Rn−1 , then

φ̃(x̃) = φ̃(x, 0) = (φ(x), 0) = φ(x).
The function φ̃ is called the Poincaré extension of φ.

Suppose φ is an inversion in S(a, r), a ∈ Rn−1 . Then for x ∈ Rn ,
φ̃(x) = ã + r2 (x − ã)∗ .
If [φ̃(x)]j denotes the jth coordinate function of φ̃(x), then
r2 xn
[φ̃(x)]n = .
|x − ã|2
By (1.3.2) and the above
|φ̃(x) − φ̃(y)|2 |x − y|2
= .
[φ̃(x)]n [φ̃(y)]n xn yn
As a consequence the mapping φ̃ leaves

|x − y|2
(2.3.1)
xn yn
invariant. If φ is reflection in the plane P(a, t), then φ̃ is a Euclidean isometry
of Rn with [φ̃(x)]n = xn . Thus (2.3.1) is also invariant under φ̃. Furthermore,
in both cases we have φ̃(H) = H.
If φ is any Möbius transformation acting on R̂n−1 , i.e., φ = φ1 ◦ · · · ◦ φm ,
where each φj is an inversion in Rn−1 , then φ̃ = φ̃1 ◦ · · · ◦ φ̃m is an extension
of φ to a Möbius transformation of R̂n which preserves H. By Theorem 3.2.4
of [11] this extension is unique. Also, since each φ̃j leaves (2.3.1) invariant,
so does the mapping φ̃. As a consequence the Poincaré extension φ̃ of any
φ ∈ GM(R̂n−1 ) is an isometry of the half-space Hn when endowed with the
Riemannian metric d given by
|dx|
ds = .
xn
This metric is invariant under φ̃ for each φ ∈ GM(R̂n−1 ). For this metric, we
have for x, y ∈ Hn ,
|x − y|2
sinh2 12 dH (x, y) = ,
4xn yn
or

|x − y|
dH (x, y) = 2 Arcsinh √ .
2 xn yn
We conclude this section by considering the Möbius transformation that

maps H√onto B. Set en = (0, . . . , 0, 1) and let denote the inversion in
S(−en , 2), that is,
2(x + en )
(x) = −en + . (2.3.2)
|x + en |2
Then
4 4en , x + en
|(x)|2 = 1 + −
|x + en |2 |x + en |2
4xn
=1− ,
|x + en |2
or
4xn
1 − |(x)|2 = . (2.3.3)
|x + en |2
Since is an inversion, is a one-to-one map of R̂n onto R̂n satisfying

((x)) = x for all x ∈ Rn . As a consequence of (2.3.3), maps H onto
B and B onto H. Also, since |(x)| = 1 when xn = 0, maps ∂H (in R̂n )
onto S. √
Since is the inversion in S(−en , 2), by identity (1.3.2)
2|y − x|
|(y) − (x)| = ,
|y + en ||x + en |
and thus
|(y) − (x)| 2
lim = (2.3.4)
y→x |y − x| |x + en |2
1 − |(x)|2
= . (2.3.5)
2xn
For φ(x) = x∗ , we have
φ (x) = |x|−4 [|x|2 I − 2Q(x)],

2.4 Exercises 15
where Q(x) is the n × n symmetric matrix (xi xj )ni, j=1 . Since the characteristic
polynomial of 2Q(x) is sn−1 (s − 2|x|2 ), taking s = |x|2 gives
det φ (x) = −|x|−2n .
But (x) = −en + 2φ(x + en ). Thus (x) = 2φ (x + en ) and hence
2n
J (x) = det (x) = − . (2.3.6)
|x + en |2n
2.4 Exercises
2.4.1. (a) With ρ(x, a) as defined by (2.1.5), prove that
(1 − |x||a|)2 ≤ ρ(x, a) ≤ (1 + |x||a|)2 .
(b) If y = ϕa (x), prove that
(1 − |x|) (1 + |x|)
(1 − |a|2 ) ≤ (1 − |y|2 ) ≤ (1 − |a|2 ).
(1 + |x|) (1 − |x|)
2.4.2. For x ∈ E(rζ , δ), show that
1+δ
|x − ζ | < (1 − r).
1−δ
2.4.3. Let a ∈ B, η ∈ S, and ζ = ϕa (η). Prove that
(1 − |a|2 )
|a − ζ | = .
|a − η|
Hint: Consider |a − ζ |2 and use (2.1.6).
|x − y|2
2.4.4. Show that |ϕy (x)|2 = .
ρ(x, y)
2.4.5. If A ∈ O(n), prove that A ◦ ϕa = ϕAa ◦ B for some B ∈ O(n).
Exercises on the Upper Half-Space H

2.4.6. Let be the mapping defined by (2.3.2).
(a) Show that

2x1 2xn−1 1 − |x|2
(x1 , . . . , xn ) = ,..., , .
|x + en | 2 |x + en |2 |x + en |2
(b) Prove the following.
i. is a one-to-one map of Rn \ {en } onto itself.
ii. ((x)) = x.
iii. maps B onto H and H onto B.
iv. maps S \ {en } onto Rn−1 and Rn−1 onto S \ {en }.

2.4.7. Show that if n = 2 and z = x1 + ix2 , e2 = i,
1 − iz
(z) =
z−i
for every z ∈ C \ {−i}.
2.4.8. Möbius transformations of H. Let be a one-to-one map of H onto
H. Prove that
((x)) = (Aϕa (x))
for some a ∈ B and A ∈ O(n).
2.4.9. For y, y1 ∈ H, set
y1 (y) = (ϕ(y1 ) ((y))).
Show that y1 is a one-to-one map of H onto H with
(a) y1 (y1 ) = en ,
(b) y1 (en ) = y1 , and
(c) y1 (y1 (y)) = y.
3
The Invariant Laplacian, Gradient, and Measure
In order to study harmonic function theory on the hyperbolic ball B, we first

need to determine the Laplacian Δh , gradient ∇ h , and measure τ on B that are
invariant under the group M(B) of Möbius transformations of B. Although
these are well known in the setting of rank one noncompact symmetric spaces,
we follow the approach of Rudin [72, Chapter 4] to determine Δh and ∇ h .
3.1 The Invariant Laplacian and Gradient

Definition 3.1.1 Suppose is an open subset of B, f ∈ C2 (), and a ∈ .
We define
(Δh f )(a) = Δ( f ◦ ϕa )(0),
∂2
where ϕa is the involution defined by (2.1.4) and Δ = is the usual
∂xj2
Laplacian on Rn .
The operator Δh is called the invariant Laplacian or
Laplace–Beltrami operator on B. Also, for f ∈ C1 (), we define the invariant
gradient ∇ h by
(∇ h f )(a) = −∇( f ◦ ϕa )(0),

where ∇ = ∂x∂ 1 , . . . , ∂x∂ n is the usual gradient.1
Suppose f is a C1 (or C2 ) function on B and y = ψ(x) is a C1 (or C2 )

mapping of B into B. Then if g = f ◦ ψ, by the chain rule
∇g(x) = ∇f (ψ(x))ψ (x),
1 The choice of the minus sign in the definition of ∇ h will assure that ∇ h f is in the same direction
to denote the Möbius invariant
and ∇
as ∇f . Throughout these notes we will use the notations Δ
Laplacian and gradient on the unit ball B of Cm . The use of the term “invariant” is made clear
in Theorem 3.1.2.
17
18 The Invariant Laplacian, Gradient, and Measure

where ψ is the n × n matrix
∂yj
∂xk . Also

n
∂ 2f
n
∂f
Δg(x) = ∇yi , ∇yj + Δyj .
∂yi ∂yj ∂yj
i,j=1 j=1
Hence if y = ϕa (x), from Equations (2.1.8) it now follows that
∇ h f (a) = (1 − |a|2 )∇f (a) (3.1.1)
and
Δh f (a) = (1 − |a|2 )2 Δf (a) + 2(n − 2)(1 − |a|2 )a, ∇f (a). (3.1.2)
If f is a radial function, that is, f (x) = g(|x|), then with r = |x|,

x
∇ h f (x) = (1 − r2 )g (r) , (3.1.3)
r
and thus |∇ h f (x)| = (1 − r2 )|g (r)|. Also,

2 g (r)
Δh f (x) = (1 − r ) (1 − r )g (r) +
2
{(n − 1)(1 − r ) + 2(n − 2)r } .
2 2
r
(3.1.4)
The Möbius invariant Laplacian on the Hermitian ball B in Cn (see [72,
Chapter 4]) is given by

n
∂ 2 f (z)
(z) = 4(1 − |z|2 )
Δf (δi,j − zi zj ) .
∂zi ∂zj
i,j=1
In contrast to the Laplacian on real-hyperbolic space, this operator has no linear

terms. The following theorem justifies the term “invariant” in reference to the
operator Δh and the gradient ∇ h .
Theorem 3.1.2 For f ∈ C2 () and ψ ∈ M(B),
Δh ( f ◦ ψ) = (Δh f ) ◦ ψ and |∇ h ( f ◦ ψ)| = |(∇ h f ) ◦ ψ|.
Proof. As in [72, Theorem 4.12], let b ∈ B and put a = ψ(b). Then ϕa ◦ψ ◦ϕb
is a Möbius transformation of B fixing 0. Thus ψ ◦ ϕb = ϕa ◦ A for some
orthogonal transformation A. Hence
Δh ( f ◦ ψ)(b) = Δ( f ◦ ψ ◦ ϕb )(0) = Δ( f ◦ ϕa ◦ A)(0).
But for any orthogonal transformation A, by the computations following

Definition 3.1.1,
Δ(g ◦ A)(0) = (Δg)(0).
3.2 The Fundamental Solution of Δh 19
Therefore, Δ( f ◦ϕa ◦A)(0) = Δ( f ◦ϕa )(0) = Δh f (a), that is, Δh ( f ◦ψ)(b) =

(Δh f )(ψ(b)). An analogous argument proves that |∇ h ( f ◦ ψ)| =
|(∇ h f ) ◦ ψ|.
Remark 3.1.3 If the hyperbolic metric (2.2.1) is expressed in standard

Riemannian notation as

ds2 = gi,j dxi dxj ,
i,j
where gi,j = 4δi,j (1 − |x|2 )−2 , then the Laplace–Beltrami operator L on B is

defined by
1 √
L( f ) = √ ∂i (gi,j g∂j f ), (3.1.5)
g
i,j
where g = det (gi,j ), and (gi,j ) is the inverse matrix of (gi,j ). A brief computation
shows that except for a factor of 14 , this agrees with the operator Δh given by
(3.1.2). Likewise, the gradient of a function f is the vector field grad f defined by
∂f ∂
grad f = gi,j . (3.1.6)
∂xi ∂xj
In particular
1 ∂f ∂ n
grad f = (1 − |x|2 )2 ,
4 ∂xj ∂xj
j=1
and
1
grad f , grad g = ( grad f )g = ( grad g)f = (1 − |x|2 )2 ∇f , ∇g.
4
Hence
1
| grad f | = grad f , grad f = (1 − |x|2 )|∇f |.
2
Thus, except for the factor of 12 , | grad f | agrees with |∇ h f | as given by (3.1.1).
3.2 The Fundamental Solution of Δh

Suppose g is a radial solution of Δh g = 0. If we let v(r) = g (r), then by
(3.1.4) the function v must satisfy
v(r)
(1 − r2 )v (r) + {(n − 1)(1 − r2 ) + 2(n − 2)r2 } =0
r
or
v (r) 1 2r
= −(n − 1) − (n − 2) .
v(r) r (1 − r2 )
Solving this differential equation for v(r) gives
(1 − r2 )n−2
v(r) = c
rn−1
for some constant c. Thus the fundamental solution gh for Δh is given by

1 1 (1 − s2 )n−2
gh (r) = ds. (3.2.1)
n r sn−1
The choice of the constant 1n will become apparent in Theorem 4.1.1. When
n = 2, this gives the usual solution
1 1
gh (r) = log .
2 r
Using the inequality
r
< log(1 + r) < r, r > −1, r = 0,
1+r
we have that when n = 2, the fundamental solution g(|x|) = − 12 log |x|
satisfies
1 1 (1 − |x|2 )
(1 − |x|2 ) ≤ gh (|x|) ≤ . (3.2.2)
4 2 |x|
When n > 2, we estimate the integral (3.2.1) above and below as follows.
For an upper estimate we have
1 1
(1 − s2 )n−2 ds
n−1
ds ≤ (1 − r 2 n−2
) n−1
r s r s

1 1
= (1 − r2 )n−2 n−2 − 1 .
(n − 2) r
But

1 1 − rn−2 (1 − r) (1 − r2 )
−1= ≤ (n − 2) ≤ (n − 2) .
rn−2 rn−2 rn−2 rn−2
Therefore, gh (r) ≤ 1n (1 − r2 )n−1 r2−n . For a lower estimate, we have

1 1 (1 − s2 )n−2 1 1 (1 − s2 )n−2 s
ds = ds
n r sn−1 n r sn
1
1 1
≥ (1 − s2 )n−2 sds = (1 − r2 )n−1 .
n r 2n(n − 1)
Therefore, gh (r) ≥ 1
2n(n−1) (1 − r2 )n−1 . Thus the fundamental solution gh (|x|)
satisfies
1 1 (1 − |x|2 )n−1
(1 − |x|2 )n−1 ≤ gh (|x|) ≤ (3.2.3)
2n(n − 1) n |x|n−2
for all x ∈ B, x = 0.
Definition 3.2.1 For x, y ∈ B, x = y, the Green’s function Gh (x, y) for Δh is
defined by

1 1 (1 − s2 )n−2
Gh (x, y) = gh (|ϕy (x)|) = ds.
n |ϕy (x)| sn−1
Since |ϕy (x)| = |ϕx ( y)|, we have that Gh (x, y) = Gh ( y, x) for all x, y
with x = y. Also, if y ∈ B is fixed, the function x → Gh (x, y) satisfies
Δh Gh (x, y) = 0 on B \ {y}. Furthermore, by (3.2.3), for n > 2,
1 (1 − |ϕy (x)|2 )n−1
cn (1 − |ϕy (x)|2 )n−1 ≤ Gh (x, y) ≤ (3.2.4)
n |ϕy (x)|n−2
for all x, y ∈ B, x = y with cn = 1
2n(n−1) . Also, by (2.1.6) and (2.1.7), for
n > 2 we have
1 (1 − |x|2 )n−1 (1 − |y|2 )n−1
Gh (x, y) ≤ , (3.2.5)
n |x − y|n−2 {|x − y|2 + (1 − |x|2 )(1 − |y|2 )}n/2
and
(1 − |x|2 )n−1 (1 − |y|2 )n−1
Gh (x, y) ≥ cn . (3.2.6)
(1 + |x||y|)2(n−1)
In comparison, the Euclidean Green’s function, Ge , for the Laplacian Δ
on B is given for n ≥ 3 by

|x − y|2−n − (|y||x − y∗ |)2−n , y ∈ B \ {0},

G (x, y) = cn
e
(3.2.7)
|x| 2−n − 1, y = 0,
for a suitable constant cn . When n = 2, we have

1 |y||x − y∗ |
G (x, y) = Gh (x, y) = log
e
. (3.2.8)
2 |x − y|
3.3 The Invariant Measure on B

Our next step is to determine the Möbius invariant measure τ on B. First,
however, we introduce some notation and formulas concerning integration
on B. We denote by ν a Lebesgue measure in Rn normalized so that ν(B) = 1.

Also, we denote by σ or σn surface measure on S again normalized such that
σ (S) = 1. Then by integration in polar coordinates we have
1
f (x) dν(x) = n rn−1 f (rζ ) dσ (ζ )dr. (3.3.1)
B 0 S
The measure σ is invariant under O(n), that is,

f (Aζ )dσ (ζ ) = f (ζ )dσ (ζ ) (3.3.2)
S S
for all A ∈ O(n) and f ∈ L1 (S). Furthermore, if K is any compact subgroup of

O(n),

f (ζ )dσ (ζ ) = f (kζ ) dkdσ (ζ ), (3.3.3)
S S K
where dk denotes Haar measure2 on K. In particular, with K = O(n) one

obtains
f (ζ )dσ (ζ ) = f (Aζ )dA, (3.3.4)
S O(n)
where dA is the Haar measure on O(n).

To determine the invariant measure τ on B we assume that dτ (x) =
ρ(x)dν(x), where ρ is a radial function on B. Then for f ∈ L1 (B, τ ) and the
mapping ϕa ∈ M(B) we have

( f ◦ ϕa )dτ = f (ϕa (x))ρ(x)dν(x),
B B
which by the change of variables formula for Rn

= f (x)ρ(ϕa (x))|Jϕa (x)|dν(x),
B
where Jϕa is the Jacobian of the mapping ϕa . Thus in order that

( f ◦ ϕa )dτ = fdτ
for all f ∈ L1 (B, τ ) and ϕa , we must have ρ(ϕa (x))|Jϕa (x)| = ρ(x). In particular
with x = 0, by (2.1.8), ρ(a) = (1 − |a|2 )−n ρ(0). Hence we define the measure
τ on B by
2 A Borel measure k on a compact group K is said to be a Haar measure if k(K) = 1 and

f (kk1 )dk = f (k)dk = f (k1 k)dk
K K K
for each k1 ∈ K and f ∈ C(K).
dν(x)
dτ (x) = . (3.3.5)
(1 − |x|2 )n
In the following theorem we prove that τ is the Möbius invariant measure
on B.
Theorem 3.3.1 (a) If ψ ∈ M(B), then the Jacobian Jψ of ψ satisfies
(1 − |ψ(x)|2 )n
|Jψ (x)| =
(1 − |x|2 )n
for all x ∈ B.
(b) The measure τ defined by (3.3.5) satisfies

f dτ = ( f ◦ ψ) dτ
B B
for every f ∈ L1 (B, τ ) and ψ ∈ M(B).
Proof. (a) For ψ ∈ M(B) and a ∈ B, let b = ψ(a). Then ϕb ◦ ψ ◦ ϕa ∈ M(B)

and fixes 0. Thus ψ(x) = ϕb ◦ A ◦ ϕa (x) for some A ∈ O(n). But then ψ (x) =
ϕb (Aϕa (x))Aϕa (x), or
ψ (a) = ϕb (0)Aϕa (a).
Hence by (2.1.8)
(1 − |b|2 )
ψ (a) = A.
(1 − |a|2 )
Since A is orthogonal, | det A| = 1. Therefore
(1 − |ψ(a)|2 )n
|Jψ (a)| = | det ψ (a)| = .
(1 − |a|2 )n
For (b), if f ∈ L1 (B, τ ) and ψ ∈ M(B), then

f dτ = f (w)(1 − |w|2 )−n dν(w),
B B
which by the change of variables formula

= f (ψ(x))(1 − |ψ(x)|2 )−n |Jψ (x)|dν(x)
B

2 −n
= f (ψ(x))(1 − |x| ) dν(x) = ( f ◦ ψ) dτ .
B B

For future reference we estimate τ (E(a, r)), 0 < r < 1. Since E(a, r) =
ϕa (Br ),
r
ρ n−1
τ (E(a, r)) = τ (Br ) = n dρ. (3.3.6)
0 (1 − ρ )
2 n
For n ≥ 2, set
rn
cn (r) = . (3.3.7)
(1 − r2 )n−1
Then c2 (r) = τ (E(a, r)), and for n > 2, by L’Hospital’s rule,
τ (Br ) n n
lim = lim = .
r→1− cn (r) r→1 n(1 − r ) + 2(n − 1)r
− 2 2 2(n − 1)
Likewise,
τ (Br )
lim = 1.
r→0+ cn (r)
Thus for all n ≥ 2,
rn
τ (Br ) = τ (E(a, r)) ≈ 3 , (3.3.8)
(1 − r2 )n−1
with equality when n = 2.
3.4 The Invariant Convolution on B

For 0 < p < ∞, we denote by L p (B, τ ) the space of measurable functions f
on B for which
f pp = | f (x)|p dτ (x) < ∞.
B
p
Also, Lloc (B) will denote the space of measurable functions f on B that are
locally p-integrable, that is,

| f (x)|p dτ (x) < ∞
K
for every compact subset K of B.4

Definition 3.4.1 For measurable functions f , g on B, the invariant convolu-
tion f ∗ g of f and g is defined by
3 A(r) ≈ B(r) means that there exist positive constants c and c such that c A(r) ≤ B(r) ≤
1 2 1
c2 A(r) for all r.
4 It should be obvious that in the definition of local integrability the measure τ may be replaced
by the Lebesgue measure ν.
3.4 The Invariant Convolution on B 25

( f ∗ g)( y) = f (x)g(ϕy (x)) dτ (x), y ∈ B,
B
provided this integral exists.
By the invariance of τ we have ( f ∗ g)( y) = (g ∗ f )( y). Although the
convolution as defined is not the usual definition for convolution of functions
on a topological group, the following analogue of the standard convolution
inequalities is still valid.
Theorem 3.4.2 Let p ∈ [1, +∞) and let q be defined by 1
p + 1
q = 1. If f ∈
Lp (B, τ ), then
f ∗ gp ≤ f p g1
for all radial functions g ∈ L1 (B, τ ), and
f ∗ g∞ ≤ f p gq
for all radial functions g ∈ Lq (B, τ ).
Proof. Let g ∈ L1 (B, τ ) be a radial function, and h ∈ Lq (B, τ ). Without loss of
generality we may assume that f , g and h are non-negative. Thus by Definition
3.4.1 and Tonelli’s theorem

f (x)( f ∗ g)(x)dτ (x) = h(x) f ( y)(g ◦ ϕx )( y)dτ ( y)dτ (x)
B
B B
= f ( y) h(x)(g(ϕx ( y))dτ (x)dτ ( y).
B B
Since ρ(x, y) = ρ( y, x), by (2.1.7), we have that |ϕx ( y)| = |ϕy (x)|. Thus since
g is radial,

h(x)g(ϕx ( y))dτ (x) = h(x)g(ϕy (x))dτ (x),
B B
which by the invariance of τ

= g(x)h(ϕy (x))dτ (x).
B
Therefore

1 1
( f ∗ g)h1 ≤ f ( y) g p (x)g q (x)(h ◦ ϕy )(x)dτ (x) dτ ( y),
B B
which by Hölder’s inequality
1 1
q
≤ g1 p
f ( y) g(x)(h ◦ ϕy ) (x)dτ (x)
q
dτ ( y).
B B
One more application of Hölder’s inequality now gives
1
1
q
( f ∗ g)h1 ≤ g1 f p
p
g(x)(h ◦ ϕy ) (x)dτ (x)dτ ( y)
q
.
B B
Also, since τ is invariant and g is radial,

g(x)(h ◦ ϕy ) (x)dτ (x)dτ ( y) =
q
(g ◦ ϕy )(x)hq (x)dτ (x)dτ ( y)
B B B B

= (g ◦ ϕx )( y)hq (x)dτ ( y)dτ (x)
B B
= g( y)dτ ( y) hq (x)dτ (x).
B B
Therefore,
1 1
( f ∗ g)h1 ≤ g1p f p g1q hq .
Taking the supremum over all h with hq ≤ 1 gives
f ∗ gp ≤ f p g1 .
The inequality f ∗ g∞ ≤ f p gq is an easy consequence of Hölder’s

inequality and the invariance of the measure τ .
p
As a consequence of Theorem 3.4.2, if g ∈ Lloc1 (B) is radial and f ∈ L (B)
loc
then f ∗ g is defined a.e. on B. There is one additional property of the above
convolution that will be needed in the proof of Theorem 4.5.4. The analogue
of the theorem for functions in the unit ball in Cn is due to Ullrich [94].
Theorem 3.4.3 If f , χ , h ∈ L1 (B, τ ) and χ is radial, then ( f ∗ χ ) ∗ h =

f ∗ (χ ∗ h).
Proof. Suppose a, y ∈ B. Since ϕϕa ( y) ◦ ϕa ◦ ϕy (0) = 0, by Theorem 2.1.1
ϕϕa ( y) ◦ ϕa ◦ ϕy = A
for some A ∈ O(n). Therefore
ϕϕa ( y) = A ◦ ϕy ◦ ϕa .
3.5 Exercises 27
Consequently |ϕϕa ( y) (x)| = |ϕy (ϕa (x))|. By the invariance of τ and the fact
that χ is radial,

χ (x)h(ϕϕa ( y) (x))dτ (x) = χ (ϕϕa ( y) (x))h(x)dτ (x)
B
B
= χ (ϕy (ϕa (x)))h(x)dτ (x)
B
= χ (ϕy (x))h(ϕa (x))dτ (x).
B
Therefore,

(( f ∗ χ ) ∗ h)(a) = f (x)χ (ϕy (x))h(ϕa ( y))dτ (x)dτ ( y)
B B
= f (x)χ ( y)h(ϕϕa (x) ( y))dτ ( y)dτ (x)
B B
= f (x)(χ ∗ h)(ϕa (x))dτ (x)
B B
= ( f ∗ (χ ∗ h))(a).
3.5 Exercises
3.5.1. For a function u defined on B and 0 < r < 1, set ur (x) = u(rx).
(a) If u is Euclidean harmonic on B, that is, u = 0, show that ur is
Euclidean harmonic on B for all n ≥ 2.
(b) If n ≥ 3 and u is H-harmonic on B, show by example that ur need
not be H-harmonic.
3.5.2. Show that g(Ax) = g(x) for all A ∈ O(n).
3.5.3. (a) Suppose f , g ∈ C2 (B). Prove that
Δh ( fg) = f Δh g + gΔh f + 2∇ h f , ∇ h g.
(b) If f ∈ C2 (R) and g ∈ C2 (B), prove that
Δh f (g(x)) = f (g(x))|∇ h g(x)|2 + f (g(x))Δh g(x).
3.5.4. (a) Prove that for all r, 0 < r < 1,

1 rn n rn
≤ τ (Br ) ≤ .
2n (1 − r2 )n−1 2(n − 1) (1 − r2 )n−1
(b) With ρ = tanh r, prove that

tanh−1 ρ
τ (Br ) = n sinhn−1 t coshn−1 tdt.
0
3.5.5. Let f ∈ C2 (B) and let Zf = {x ∈ B : f (x) = 0}. Show that for p > 1
Δh | f |p = p(p − 1)| f |p−2 |∇ h f |2 + p| f |p−2 f Δh f (3.5.1)
on B \ Zf .
3.5.6. For a C1 function f , the normal derivative operator N is defined by
Nf (x) = x, ∇f (x). If r = |x|, show that
(1 − r2 )
Δh = (1 − r 2
)N 2
+ (n − 2)(1 + r 2
)N + (1 − r 2
)Δ σ ,
r2
where Δσ is the tangential part of the Euclidean Laplacian given in
Cartesian coordinates by
∂ ∂
Δσ = 2
Li,j , where Li,j = xi − xj .
∂xj ∂xi
i<j
3.5.7. As in (3.2.7) and (3.2.8), let Ge denote the Euclidean Green’s function
on B.
(a) Prove that

− log |x − y|, n = 2,
G (x, y) ≤ K
e
|x − y|2−n , n ≥ 3.
(b) Prove that
Ge (x, y) ≤ KGe (0, y)|x − y|1−n . (3.5.2)
(c) For fixed yo ∈ B, prove that
Ge (x, yo ) ≥ c(1 − |x|) for all x ∈ B,
and
Ge (x, yo ) ≤ cδ (1 − |x|) for all x ∈ B \ B( yo , δ),

where δ < (1 − |yo |).
3.5.8. Fix δ, 0 < δ < 1. Prove that
⎧
⎪
⎨c G (0, y) (1 − |x| )
2 n−1
δ h , for all x ∈ B \ E( y, δ),
Gh (x, y) ≤ ρ(x, y)n−1
⎪
⎩c G (0, y)|x − y|1−n ,
n h for all x ∈ B.
3.5 Exercises 29
1 (1 − |x|2 )n−1
3.5.9. (a) Show that |∇ h g(x)| = .
n |x|n−1
(b) Using (a) compute |∇xh Gh (x, y)|.
3.5.10. Consider ϕx (a) as defined by (2.1.6), that is,
x|x − a|2 + (1 − |a|2 )(x − a)

ϕx (a) = .
ρ(x, a)
(a) Let yj be the jth coordinate of y = ϕx (a). Show that
∂yj
= δi,j (|a|2 + 1) − 2ai aj ,
∂xi
x=0
where δi,j = 1 if i = j, 0 otherwise.

(b) If f is a C1 function, prove that
∇x f (ϕx (a))|x=0 = −2a, ∇f (−a)a+(1+|a|2 )∇f (−a). (3.5.3)

3.5.11. The invariant Laplacian and gradient on H. Using formulas (3.1.5)
and (3.1.6), show that for the upper half-space H with the metric ds =
y−1
n |dy|,
∂f
LH f ( y) = y2n f ( y) − (n − 2)yn
∂yn
and
( grad f )g = y2n ∇f , ∇g = yn ∇f , yn ∇g,
or ∇ h f ( y) = yn ∇f ( y).
3.5.12. Let U be a real-valued function on H, and let : B → H be given by
(2.3.2). Let V(x) be defined on B by V(x) = U((x)).
(a) Prove that |∇ h V(x)|2 = 4y2n |∇U( y)|2 .
(b) Prove that Δh V(x) = 4LH U( y).
3.5.13. The Green’s function on H. For y = ( y , yn ) ∈ H, set
gH ( y) = gh (( y)),
where : H → B is given by (2.3.2). Show that

(a) LH gH ( y) = 0 for all y ∈ H \ {en } with gH (en ) = +∞.
yn−1
n 4n−1 yn−1
n
(b) cn 4n−1 ≤ gH ( y) ≤ .
|y + en |2(n−1) n |y + en |n |y − en |n−2
3.5.14. As in the previous exercise, for yo , y ∈ H, set GH ( y, yo ) = Gh (( y),

( yo )). Then GH is the Green’s function for LH on H with singularity
at yo . Using inequalities (3.2.5) and (3.2.6) find upper and lower
estimates for GH ( y, yo ).
3.5.15. The invariant measure on H. Let f ∈ Cc (H), and let : B → H be
the map given by (2.3.2).
(a) Using the change of variable formula and (2.3.6), prove that

f ((x))
f ( y)dy = 2n dx.
B |x + en |
2n
H
(b) Using (a), show that

f ((x))
f ( y)dy = 2n
dx.
H B (1 − |x|2 )n
(c) Set dτH ( y) = y−n
n dy. If is a one-to-one map of H onto H prove
that
f (( y))dτH (( y)) = f ( y)dτH ( y).
H H
3.5.16. For any positive measure ν on S, define ν ∗ = ν◦ on the Borel subsets
of Rn−1 by
ν ∗ (B) = (ν ◦ )(B) = ν((B)).
Prove that ν ∗ is a measure on Rn−1 and

fdν = ( f ◦ )d(ν ◦ )
S\{en } Rn−1
for every positive Borel measurable function f on S \ {en }.

3.5.17. Change of variables formulas for S. [10, p. 148] Use the previous
exercise with ν = σ to
(a) prove that

f ((t))
fdσ = c 2
n n−2
dt
S Rn−1 (1 + |t|2 )n−1
for every positive Borel measurable function on S.
(b) Prove that

2 f ((ζ ))
f (t)dt = dσ (ζ )
R n−1 cn S (1 + ζn )n−1
for every positive Borel measurable function on Rn−1 .
4
H-Harmonic and H-Subharmonic Functions
In this chapter we consider the class of functions on B that are harmonic or

subharmonic with respect to the operator Δh . We begin with the following
definition.
Definition 4.0.1 Let be an open subset of B. A real (or complex)-valued
function f ∈ C2 () is said to be H-harmonic (or invariant harmonic) on
if Δh f (x) = 0 for all x ∈ . Also, a real-valued function f ∈ C2 () is said
to be H-subharmonic (or invariant subharmonic) on if Δh f (x) ≥ 0 for all
x ∈ .1
Clearly if f is H-harmonic or H-subharmonic on B, then by the
M-invariance of Δh so is f ◦ ψ for all ψ ∈ M(B). In Section 4.1 we
derive the invariant mean-value property for C2 H-subharmonic functions,
and in Section 4.3 we extend the definition of H-subharmonic functions to
the class of upper semicontinuous functions. Sections 4.4, 4.5, and 4.6 contain
basic properties of H-subharmonic functions. We conclude the chapter with a
discussion of quasi-nearly H-subharmonic functions, a concept that will play
a significant role in subsequent chapters.
4.1 The Invariant Mean-Value Property

Recall that a function f on B is radial if f (x) = g(|x|) for some function g on
[0, 1). This is equivalent to f (Ax) = f (x) for all A ∈ O(n). For a continuous
function f on B we define the radialization f of f by
1 In the text by W. Rudin ([72]) functions that are harmonic or subharmonic with respect to
the Möbius invariant Laplacian are referred to as M-harmonic and M-subharmonic. In these
notes I use the terms H-harmonic and H-subharmonic to denote functions that are harmonic or
subharmonic with respect to the hyperbolic Laplacian.
31
32 H-Harmonic and H-Subharmonic Functions

f (x) = f (Ax) dA = f (|x|ζ )dσ (ζ ), x ∈ B, (4.1.1)
O(n) S
where dA is the Haar measure on O(n). Clearly f is a radial function on B,

and if f ∈ C2 (B),

Δh f (x) =

(Δh f )(Ax)dA.
O(n)
Prior to proving the mean-value property for H-harmonic and

H-subharmonic functions we prove the following analogue of Lemma 2.5
of [62].
Theorem 4.1.1
If f ∈ C2 (B), then
d
(a) f (rζ )dσ (ζ ) = n r (1 − r )
1 1−n 2 n−2
Δh f (x)dτ (x), and
dr S Br
(b) f (0) = f (rζ )dσ (ζ ) − g(|x|, r)Δh f (x) dτ (x), where

S Br

1 r (1 − s2 )n−2
g(|x|, r) = ds. (4.1.2)
n |x| sn−1
Proof. Since (Δh f ) = Δh ( f ), to prove (a) it suffices to assume that f is
a radial function on B. As in [62], suppose f (x) = u(|x|2 ), where u is a C2
function on [0, 1). Then with r = |x|,
Δf (x) = 4r2 u (r2 ) + 2nu (r2 ) and x, ∇f (x) = 2r2 u (r2 ).
Thus by (3.1.2)
Δh f (x) = 4r2 (1 − r2 )2 u (r2 ) + 2n(1 − r2 )2 u (r2 ) + 4(n − 2)r2 (1 − r2 )u (r2 ).
(4.1.3)
If we let
n
v(t) = t 2 (1 − t)2−n u (t),
then Δh f (x) = 4r2−n (1 − r2 )n v (r2 ). Therefore
ρ
Δh f dτ = n rn−1 (1 − r2 )−n Δh f (r) dr
Bρ 0
ρ
= 4n rv (r2 ) dr
0
= 2nρ (1 − ρ 2 )2−n u (ρ 2 ).
n
On the other hand,

d
f (ρζ ) dσ (ζ ) = 2ρu (ρ 2 ).
dρ S
4.1 The Invariant Mean-Value Property 33
Therefore

d 1
f (ρζ ) dσ (ζ ) = ρ 1−n (1 − ρ 2 )n−2 Δh f (x) dτ (x),
dρ S n Bρ
which proves (a). Integrating the above from 0 to r gives

r
1 1−n
f (rζ )dσ (ζ ) − f (0) = ρ (1 − ρ 2 )n−2 Δh f (x) dτ (x) dρ,
S 0 n Bρ
which upon changing the order of integration

= g(|x|, r)Δh f (x) dτ (x),
Br
where g(|x|, r) is defined by (4.1.2).

Remark 4.1.2 Although we assumed in Theorem 4.1.1 that f ∈ C2 (B), the
conclusions are still valid if f ∈ C2 (), where is an open subset of B with
0 ∈ . In this case the conclusions hold for all r such that B(0, r) ⊂ .
As a corollary of Theorem 4.1.1 we obtain the following mean-value
property for H-harmonic and H-subharmonic functions.
Corollary 4.1.3 (Invariant Mean-Value Property) Let be an open subset
of B. A real-valued C2 function f is H-subharmonic on if and only if for all
a ∈ ,

f (a) ≤ f (ϕa (rt)) dσ (t) (4.1.4)
S
for all r > 0 such that E(a, r) ⊂ . Furthermore, f is H-harmonic on B if and
only if equality holds in (4.1.4).
Proof. Suppose f is H-subharmonic on and a ∈ . Then f ◦ ϕa is
H-subharmonic on ϕa (). Since Δh ( f ◦ ϕa ) ≥ 0 on ϕa (), by part (b) of
the previous theorem we have

f (a) = ( f ◦ ϕa )(0) ≤ ( f ◦ ϕa )(rt) dσ (t)
S
for all r > 0 such that B(0, r) ⊂ ϕa (). Clearly, if f is H-harmonic on B, then
equality holds in (4.1.4).
Conversely, suppose f ∈ C2 () satisfies (4.1.4) for all a ∈ and r > 0
such that E(a, r) ⊂ . Let a ∈ be arbitrary, and set h(x) = ( f ◦ ϕa )(x). Then
there exists ro > 0 such that

h(0) ≤ h(rt)dσ (t)
S
for all r, 0 < r < ro . Since h is C2 in a neighborhood of 0, as a consequence

of the Taylor expansion of h about 0,

r2
{h(rt) − h(0)} dσ (t) = Δh(0) + O(r3 ). (4.1.5)
S 2n
Therefore,

2n
Δh(0) = lim 2 {h(rt) − h(0)} dσ (t),
r→0 r S
and thus Δh f (a) = Δh(0) ≥ 0. Hence f is H-subharmonic on B.
Remarks 4.1.4 (a) If f ∈ C2 (B) is H-subharmonic on B, then inequality

(4.1.4) holds for all r, 0 < r < 1. Furthermore, as a consequence of Theorem
4.1.1 the integral mean

M( f , r) = f (rt)dσ (t)
S
is a non-decreasing function of r, 0 < r < 1.

(b) As a consequence of the above proof, for f ∈ C2 (), we have

2n
Δh f (a) = lim 2 { f (ϕa (rt)) − f (a)}dσ (t). (4.1.6)
r→0 r S
One other consequence of Theorem 4.1.1 is the following corollary.
Corollary 4.1.5 If f ∈ Cc2 (B), then for all a ∈ B,

f (a) = − Gh (a, x)Δh f (x)dτ (x),
B
where Gh is the hyperbolic Green’s function for B.
Proof. Since f has compact support in B, letting r → 1 in Theorem 4.1.1(b)

gives

f (0) = − gh (|x|)Δh f (x)dτ (x).
B
Applying this to f ◦ ϕa gives the desired result.
We conclude this section with the following version of the maximum

principle.
Theorem 4.1.6 (Maximum Principle) Suppose is an open subset of B and

that f ∈ C2 () is H-subharmonic in and continuous on . If f ≤ 0 on ∂,
then f ≤ 0 in .
4.2 The Special Case n = 2 35
Proof. Set h(x) = f (x) + |x|2 . Then h ≤ on ∂ and for all x ∈ ,
Δh h(x) = Δh f (x) + 2[n(1 − |x|2 )2 + 2(n − 2)|x|2 (1 − |x|2 )].
Thus Δh h(x) > 0 for all x ∈ . If h has a local maximum at some point x ∈ ,
then h ◦ ϕx has a local maximum at 0. This, however, is impossible since
Δ(h ◦ ϕx )(0) = Δh h(x) > 0.
Thus h(x) < for all x ∈ . Finally, since f (x) ≤ h(x) for all x, letting → 0
gives f (x) ≤ 0 on .
4.2 The Special Case n = 2

In this section we consider some of the previous results in the setting of the
unit disc D in the complex plane C. When n = 2 with z = x + iy,
∂ 2 f (z)
Δh f (z) = (1 − |z|2 )2 f (z) = 4(1 − |z|2 )2 .
∂z∂z
Thus f is H-subharmonic (H-harmonic) on D if and only if f is Euclidean
subharmonic (Euclidean harmonic) on D.
When n = 2, the Möbius transformation ϕw (z) becomes
w−z
ϕw (z) = ,
1 − wz
and the Green’s function Gh (z, w) and invariant measure τ are

1 1 − wz
Gh (z, w) = log and dτ (z) = 1 (1 − |z|2 )−2 dA(z),
2 w−z π
where dA denotes area measure in D. Thus Corollary 4.1.5 becomes

1 1 − wz
f (z) = −
log Δf (w) dA(w)
2π D w−z
for all f ∈ C2 (D) with compact support.
Suppose now that f is subharmonic on D. With ϕa (z) = (a − z) (1 − az),
the mean-value property 4.1.4 becomes
2π
1
f (a) ≤ f (ϕa (reit ))dt (4.2.1)
2π 0
for all r, 0 < r < 1. Here, the integral is the average of f over the non-Euclidean
circle {ϕa (reit ): 0 ≤ t < 2π } with non-Euclidean center a. In contrast, the
Euclidean mean-value property is that
2π
1
f (a) ≤ f (a + reit )dt, (4.2.2)
2π 0
valid for all r such that B(a, r) ⊂ D. Here the integral is over the Euclidean
circle {a + reit : 0 ≤ t < 2π }. If f is harmonic, then equality holds in both
(4.2.1) and (4.2.2).
Suppose now that f is subharmonic and continuous on D. Since f ◦ ϕa is also
continuous on D, letting r → 1 in (4.2.1) gives
2π
1 1 f (ϕa (ζ ))
f (a) ≤ f (ϕa (eit ))dt = dζ .
2π 0 2π i |ζ |=1 ζ
Letting w = ϕa (ζ ) we have ζ = ϕa (w). Thus dζ = ϕa (w)dw and

1 f (ϕa (ζ )) 1 ϕ (w)
dζ = f (w) a dw
2π i |ζ |=1 ζ 2π i |w|=1 ϕa (w)
2π
1 ϕ (eit )eit
= f (eit ) a it dt.
2π 0 ϕa (e )
But
ϕa (eit )eit −(1 − |a|2 )eit (1 − |a|2 )
= = .
ϕa (e )it (a − e )(1 − ae )
it it |a − eit |2
Therefore
2π
1
f (a) ≤ P(a, eit )f (eit )dt, (4.2.3)
2π 0
where
1 − |a|2
P(a, eit ) =
|a − eit |2
is the Poisson kernel on D. If f is harmonic on D, then equality holds in the
above and (4.2.3) is the Poisson integral formula for f . In Section 5.3 we will
establish the Poisson integral formula for H-harmonic functions on B.
Suppose now that f is harmonic on D. When n = 2, the function fr (z) =
f (rz) is also harmonic on D for all r, 0 < r < 1. Applying (4.2.3) to fr we have
2π
1
fr (a) = P(a, eit )fr (eit )dt.
2π 0
For w ∈ C with |w| < r, set a = w/r. Then
2π 2
1 r − |w|2
f (w) = f (reit )dt, (4.2.4)
2π 0 |w − reit |2
which is the classical Poisson integral formula for the disc |w| < r.
If f is C2 , then as above f is H-subharmonic on D if and only if f is

subharmonic on D. As in the proof of (4.2.4), we then have
2π 2
1 r − |w|2
f (w) ≤ f (reit )dt
2π 0 |w − reit |2
for all w, |w| < r. In this case, the function
2π 2
1 r − |w|2
H(w) = f (reit )dt
2π 0 |w − reit |2
is harmonic in |w| < r, continuous on |w| ≤ r with boundary values f (reit ).
An excellent reference for potential theory in the complex plane is the
monograph by T. Ransford [69].
4.3 H-Subharmonic Functions

We now extend our definition of H-subharmonic functions to the class of
upper semicontinuous functions. We begin with a brief review of upper
semicontinuous functions.
Definition 4.3.1 Let D be a subset of Rn . A function f : D → [−∞, ∞) is

upper semicontinuous at xo ∈ D if for every α ∈ R with α > f (xo ) there
exists a δ > 0 such that
f (x) < α for all x ∈ D ∩ B(xo , δ).
The function f is upper semicontinuous on D if f is upper semicontinuous at

each point of D. A function g : D → (−∞, ∞] is lower semicontinuous if −g
is upper semicontinuous.
If f (xo ) > −∞, then the above definition is equivalent to given > 0, there
exists a δ > 0 such that
f (x) < f (xo ) + for all x ∈ D ∩ B(xo , δ).
From the definition it follows that f is upper semicontinuous on D if and only

if {x ∈ D : f (x) < α} is open in D for each α ∈ R. Characteristic functions of
closed sets are upper semicontinuous whereas characteristic functions of open
sets are lower semicontinuous. It is an easy exercise to show that f is upper
semicontinuous at xo if and only if
lim sup f (x) ≤ f (xo ).

x→xo
The following theorem provides some elementary properties of upper

semicontinuous functions.
Theorem 4.3.2 Let S be a non-empty subset of Rn .

(a) If f is upper semicontinuous on S and K is a compact subset of S, then
sup f (x) < ∞

x∈K
and the supremum is assumed at some xo ∈ K.

(b) If f1 , . . . , fn are upper semicontinuous on S, then
f (x) = max{ f1 (x), . . . , fn (x)}
is upper semicontinuous on S.
(c) If { fα }α∈A are upper semicontinuous on S, then
f (x) = inf { fα (x)}

α∈A
is upper semicontinuous on S.
(d) If f is upper semicontinuous and bounded above on S, then there exists
a decreasing sequence { fn } of continuous real-valued functions on S, bounded
above, with
f (x) = lim fn (x) for all x ∈ S.
n→∞
Proof. For the proof of (a) let α = supx∈K f (x), and let {xn } be a sequence in K
such that f (xn ) → α. By compactness there exists a subsequence {xnk } of {xn }
that converges to some xo ∈ K. Since f (xo ) is finite, by upper semicontinuity
of the function f , given > 0 there exists a δ > 0 such that
f (x) < f (xo ) +
for all x ∈ B(x, δ) ∩ S. In particular, there exists ko such that f (xnk ) < f (xo ) +
for all k ≥ ko Thus
α ≤ f (xo ) + ≤ α + .
Since > 0 was arbitrary, we have f (xo ) = α.

(d) Let M ∈ R be such that f (x) ≤ M for all x ∈ S. For each positive integer
n define
fn (x) = sup{ f (w) − n|x − w|}, x ∈ S.
w∈S
Then fn (x) ≤ M for all n, fn (x) ≥ f (x), and fn (x) ≥ fm (x) for all positive
integers m, n, with m ≥ n. Hence it remains only to be shown that each fn is
continuous on S and that fn (x) → f (x) for all x ∈ S.
Fix n and let > 0 be given. If x1 , x2 ∈ S with |x1 − x2 | < /n, then
1
|x2 − w| < |x1 − w| + ,
n
and thus
f (w) − n|x1 − w| < f (w) − n|x2 − w| + .
Therefore, fn (x1 ) < fn (x2 ) + . By symmetry we have |fn (x1 ) − fn (x2 )| < ,
and thus fn is continuous (in fact uniformly continuous) on S.
We next prove that fn (x) → f (x) pointwise on S. Fix xo ∈ S and suppose
f (xo ) > −∞. Without loss of generality we can assume that supw f (w) ≤ 1
and f (xo ) = 0. Let > 0 be given. Since f is upper semicontinuous there exists
δ > 0 such that f (w) < for all w ∈ S with |xo − w| < δ. If |xo − w| ≥ δ, then
for all n with n > 1/δ,
f (w) − n|xo − w| ≤ 1 − 1 = 0.
If |xo − w| < δ, then f (w) − n|xo − w| < for all n. Hence for all n > 1/δ,
0 = f (xo ) ≤ fn (xo ) = sup{ f (w) − n|xo − w|} < .

w
Thus fn (xo ) → f (xo ). If f (xo ) = −∞, then a simple modification of the above
shows that fn (xo ) → −∞.
The proofs of (b) and (c) are straightforward and are omitted.
Definition 4.3.3 Let be an open subset of B. An upper semicontinuous

function f : → [−∞, ∞), with f ≡ −∞, is H-subharmonic on if

f (a) ≤ f (ϕa (rt)) dσ (t) (4.3.1)
S
for all a ∈ and all r sufficiently small. A function f is H-superharmonic if

−f is H-subharmonic.
Inequality (4.3.1) is the Möbius invariant mean-value inequality. Definition

4.3.3 is of course equivalent to Definition 4.0.1 for the class of C2 functions.
In the following theorem we prove that with this definition the class of
H-subharmonic functions is again invariant under M.
Theorem 4.3.4 If f is H-subharmonic on B, then f ◦ ψ is H-subharmonic for

all ψ ∈ M(B).
Proof. For ψ ∈ M and a ∈ B, let b = ψ(a). Then (ϕb ◦ ψ ◦ ϕa )(0) = 0,

and thus ψ(ϕa (x)) = ϕb (Ax) for some A ∈ O(n). Consequently, by the O(n)
invariance of σ ,

( f ◦ ψ)(ϕa (rt)) dσ (t) = f (ϕb (Art)) dσ (t)
S
S
= f (ϕb (rt)) dσ (t) ≥ f (b) = ( f ◦ ψ)(a).
S
There is also a volume version of the invariant mean-value inequality.
Theorem 4.3.5 Let be an open subset of B. If f is H-subharmonic on ,

then for each a ∈ ,

1
f (a) ≤ f (x)dτ (x) (4.3.2)
τ (Br ) E(a,r)
for all r sufficiently small such that E(a, r) ⊂ . If f is H-harmonic on , then
equality holds in (4.3.2).
Proof. The inequality follows from the M-invariance of τ , and is obtained by

multiplying both sides of (4.3.1) by n ρ n−1 (1 − ρ 2 )−n and integrating from
0 to r.
Remark 4.3.6 If f is H-harmonic on B, then

f (ψ(0)) = f (ψ(rt)) dσ (t)
S
for all ψ ∈ M(B). By multiplying by nρ n−1 and integrating we also have

f (ψ(0)) = f (ψ(w)) dν(w) (4.3.3)
B
for all ψ ∈ M(B). This leads to the following question.

Question: If f ∈ L1 (B) satisfies (4.3.3) for every ψ ∈ M(B), is f
H-harmonic?
It is known that the answer is yes when n = 2. On the Hermitian ball B in Cm
the answer to the analogous question is very surprising. In [3] P. Ahern, M.
Flores, and W. Rudin answered the question in the affirmative when m ≤ 11,
and in the negative when m ≥ 12.
For C2 H-subharmonic functions f there is also an invariant volume version

of (4.1.6). If we set h(x) = ( f ◦ ϕa )(x) and integrate (4.1.5), then for all r
sufficiently small,

{h(x) − h(0)} dτ (x) = 12 Δh(0)c(r) + O(rn+3 ),
Br
where r
c(r) = ρ n+1 (1 − ρ 2 )−n dρ.
0
By L’Hospital’s rule,

r2 τ (Br ) τ (Br )
lim = n + 2 lim (1 − r ) n
2
,
r→0 c(r) r→0 r (1 − r2 )n−1
which by (3.3.6)
= n + 2.
Also, since r2 τ (Br ) ≈ rn+2 for r sufficiently small, we have

2(n + 2)
Δh f (a) = lim 2 { f (x) − f (a)} dτ (x). (4.3.4)
r→0 r τ (Br ) E(a,r)
We now express identity (4.3.4) in terms of the invariant convolution given

in Definition 3.4.1. If we define r by
⎧
⎨ 1 , |x| ≤ r,
r (x) = τ (Br )
⎩
0, |x| > r,
then
1
( f ∗ r )(a) − f (a) = { f (x) − f (a)} dτ (x).
τ (Br ) E(a,r)
Thus by (4.3.4), if f ∈ C2 (B), then

2(n + 2)
(Δh f )(a) = lim [( f ∗ r )(a) − f (a)] . (4.3.5)
r→0 r2
If f has compact support, then the convergence is uniform on B.
4.4 Properties of H-Subharmonic Functions

In this section we prove several properties of H-subharmonic functions. Most
of these are an immediate consequence of the definition.
Theorem 4.4.1 (a) If u1 , u2 are H-subharmonic on B, then
u(x) = max{u1 (x), u2 (x)}
is H-subharmonic on B.
(b) If u is H-subharmonic on B and ϕ : [−∞, ∞) is an increasing convex
function, then ϕ ◦ u is H-subharmonic on B.
(c) If un is a decreasing sequence of H-subharmonic functions, then
u(x) = lim un (x)

n→∞
is either identically −∞ or H-subharmonic on B.
Proof. The proof of (a) is straightforward and is left as an exercise. The proof
of (b) is an immediate consequence of Jensen’s inequality. Recall that a function
ϕ : (a, b) → R is convex on (a, b) if whenever x1 , x2 ∈ (a, b) we have
ϕ(tx1 + (1 − t)x2 ) ≤ tϕ(x1 ) + (1 − t)ϕ(x2 )
for all t ∈ [0, 1]. It is easily shown that a C2 function ϕ is convex if and only if
ϕ (x) ≥ 0. Jensen’s inequality is as follows: if (X, μ) is a measure space with
μ(X) = 1 and ϕ is a convex function on (−∞, ∞), then

ϕ f dμ ≤ ϕ ◦ f dμ
X X
whenever f is an integrable function on X (see [71, Proposition 20]).

(c) Assume that u ≡ −∞. For each α ∈ R we have
∞

{x : u(x) < α} = {x : un (x) < α}.
n=1
Hence u is upper semicontinuous on B. Also, for each n ∈ N and a ∈ B, we

have

un (a) ≤ un (ϕa (rt)) dσ (t).
S
Hence by the monotone convergence theorem,

u(a) ≤ u(ϕa (rt))dσ (t).
S
Thus u is H-subharmonic on B.
As a consequence of the previous theorem, if f is H-harmonic on B, then

| f |p is H-subharmonic for all p, 1 ≤ p < ∞. Likewise, if f is H-subharmonic
on B, then ecf , c > 0, and f + (x) = max{ f (x), 0} are H-subharmonic on B.
Since many of the properties of subharmonic functions follow from the
Euclidean analogue of (4.3.1) or (4.3.2), those same properties are still valid
for H-subharmonic functions on B. In particular one has the following versions
of the maximum principle.
Theorem 4.4.2 (Maximum Principle) Let be an open connected subset

of B.
(a) (Version I) If f is a non-constant H-subharmonic function on , then

f (x) < sup f (y) for all x ∈ .
y∈
(b) (Version II) If f is H-subharmonic on and

lim sup f (x) ≤ M for all ζ ∈ ∂,
x→ζ
x∈
then f (x) ≤ M for all x ∈ .
Proof. (a) Let M = supx∈ f (x). If M = ∞ then the result is certainly true.
Hence assume M < ∞ and let E = {y ∈ : f (y) = M}. Since f (y) ≤ M for
all y, E = \ {y : f (y) < M} and thus E is closed. We now show that E is
open. Suppose E = φ. Let a ∈ E and let ra > 0 be such that E(a, ra ) ⊂ and
that (4.3.2) holds for all r with 0 < r < ra . Then

1
M = f (a) ≤ f (x)dτ (x),
τ (E(a, r))
E(a,r)
and thus
[ f (x) − M] dτ (x) = 0.
E(a,r)
Since f (x) ≤ M, the above implies that f (x) = M a.e. on E(a, r). But since f
is upper semicontinuous, f (x) = M for all x ∈ E(a, r). Therefore E(a, r) ⊂ E
and thus E is open. Since is connected, E = or E = φ. Finally, since f is
non-constant, E = φ, which proves the result.
(b) Let α = supx∈ f (x), and let {xn } be a sequence in such that f (xn ) →
α. Since {xn } is bounded, there exists a subsequence {xnk } of {xn } that converges
to some xo ∈ . But by part (a), xo ∈ ∂. But then α = limk→∞ f (xnk ) ≤
lim supx→xo f (x) ≤ M.
Another consequence of the mean-value inequality for H-subharmonic
functions is local integrability. As in Section 3.4, a measurable function f
defined on an open subset of Rn is locally integrable if

| f (x)|dν(x) < ∞
K
for every compact subset K of . The set of locally integrable functions on
1 ().
is denoted by Lloc
Theorem 4.4.3 Let be an open, connected subset of B. If f is
H-subharmonic on , then f is locally integrable on and {x ∈ : f (x) =
−∞} has measure zero.
Proof. By compactness it suffices to show that f is integrable on every

hyperbolic ball E(a, r) with E(a, r) ⊂ . Set

U= x∈: | f | dτ < ∞ for some r > 0 ,
E(x,r)
and let
P = {x ∈ : f (x) = −∞}.

If a ∈ U and r > 0 is such that E(a,r) | f | dτ < ∞, then for every y ∈ E(a, r),

| f | dτ < ∞
E(y,δ)
for every δ, 0 < δ < r − |ϕa (x)|. Thus U is an open subset of . We next show
that U is also non-empty.
Since f is not identically −∞, \ P = φ. Let a ∈ \ P and choose r > 0
such that E(a, r) ⊂ . Since f (a) > −∞, by inequality (4.3.2)

f (x) dτ (x) > −∞.
E(a,r)
Let M = sup{ f (x): x ∈ E(a, r)}, which is finite since f is upper semicontinuous.
Since
| f (x)| ≤ |M| + M − f (x)

for all x ∈ E(a, r), we have E(a,r) | f | dτ < ∞. Therefore U is non-empty.
We now show that U is also closed. Let a ∈ \ U, and let r > 0 be such
that E(a, 3r) ⊂ . Since a ∈ \ U we have

f (x) dτ (x) = −∞.
E(a,r)
Let M < ∞ be such that f (x) ≤ M for all x ∈ E(a, 3r). Then M − f (x) ≥ 0 for
all x ∈ E(a, 3r). If x ∈ E(a, r), then by the triangle inequality for the pseudo-
hyperbolic metric, E(a, r) ⊂ E(x, 2r). Therefore,

[M − f ] dτ ≥ [M − f ] dτ = ∞.
E(x,2r) E(a,r)
As a consequence

f dτ = −∞
E(x,2r)
for all x ∈ E(a, r). Thus \U is open; that is, U is closed. Since is connected
and U = φ, we have U = . Thus f (x) > −∞ τ - a.e. and τ (P) = 0.
4.5 Approximation by C∞ H-Subharmonic Functions

We begin this section with the definition of an approximate identity.
Definition 4.5.1 A sequence of non-negative continuous functions {hj } is an
approximate
identity in L1 (B, τ ) if
(a) hj dτ = 1, and
B
(b) lim hj dτ = 0 for every δ > 0.
j→∞ B\B(0,δ)
Definition 4.5.2 Let {rk } be a decreasing sequence with rk → 0 as k → ∞.

For each k, let χk be a non-negative C∞ radial function on B with support
contained in {x : rk+1 < |x| < rk } satisfying

χk (x) dτ (x) = 1.
B
It is easily seen that the sequence {χk }∞
k=1 is a C
∞ approximate identity for
1
L (B, τ ).
The following lemma shows the importance of an approximate identity, also
sometimes referred to as an approximate unit.
Lemma 4.5.3 Let {χk } be defined as above. Then
lim (h ∗ χk ) = h,
k→∞
p
uniformly on B if h ∈ Cc (B), and locally in L p if h ∈ Lloc (B), 1 ≤ p < ∞.
Proof. Suppose h ∈ Cc (B) with K = supp h. Since K is compact, h is
uniformly continuous on K and hence on B. Thus given > 0 there exists
a δ > 0 such that
|h(x) − h(y)| <
for all x, y ∈ B with y ∈ E(x, δ). For each k let hk = h ∗ χk where ∗ is given in
Definition 3.4.1. Since

hk (x) = h(y)χk (φx (y))dτ (y),
B
we have

|h(x) − hk (x)| ≤ |h(x) − h(y)|χk (ϕx (y))dτ (y)
B

≤ |h(x) − h(y)|χk (ϕx (y))dτ (y) + 2h∞ χk (ϕx (y))dτ (y)
E(x,δ) B\E(x,δ)

< + 2h∞ χk (y)dτ (y).
B\B(0,δ)
Hence since {χk } is an approximate identity, limk→∞ hk (x) = h(x) uniformly

on B.
p
Suppose h ∈ Lloc (B), 1 ≤ p < ∞. Let K be a compact subset of B. Redefine
h on B such that h(x) = 0 for all x ∈ K. For > 0, choose g ∈ Cc (B) with
supp g ⊂ K and

|h − g|p dτ = |h − g|p dτ < p .
K B
Let gk = g ∗ χk . Then by Minkowski’s inequality,
h − hk p ≤ h − gp + hk − gk p + gk − gp .
Since g is continuous, gk → g uniformly on B. Furthermore, since supp g is

compact and supp χk ⊂ {x : rk+1 < |x| < rk } we have that supp gk , as well
as hk , is also compact for sufficiently large k. Hence there exists ko such that
gk − gp < for all k ≥ ko . Thus for k ≥ ko we have
1/p
|h − hk |dτ ≤ h − hk p ≤ 2 + hk − gk p ,
K
which by Theorem 3.4.2
≤ 2 + h − gp χk 1 ≤ 3,
from which the result now follows.
We are now in a position to prove the following theorem.
Theorem 4.5.4 Let {χk } be a C∞ approximate identity as above. If f is

H-subharmonic on B, then { f ∗ χk }∞
k=1 is a non-increasing sequence of C
∞
H-subharmonic functions on B satisfying
( f ∗ χk )(x) ≥ f (x) and lim ( f ∗ χk )(x) = f (x) (4.5.1)

k→∞
for all x ∈ B.
Proof. Since χk is C∞ , the function f ∗ χk is also C∞ . By integration in polar

coordinates,
1
rn−1
( f ∗ χk )(y) = n χ k (r) f (ϕy (rt)) dσ (t) dr (4.5.2)
0 (1 − r )
2 n
S

≥ f (y) χk dτ = f (y).
B
Fix a ∈ B, and let α > f (a). Since f is upper semicontinuous there exists
r > 0 such that f (w) < α for all w ∈ E(a, r). Hence if rk < r,

( f ∗ χk )(a) = f (w)χk (ϕa (w)) dτ (w) ≤ α χk dτ = α.
B B
Thus lim supk→∞ ( f ∗ χk )(a) ≤ f (a), which when combined with the above
proves (4.5.1).
To show that f ∗ χk is H-subharmonic we use (4.3.5). Since χk and r are
both radial, by Lemma 3.4.3
( f ∗ χk ) ∗ r = ( f ∗ r ) ∗ χk .
But as above, by integration in polar coordinates ( f ∗ r ) ≥ f . Therefore
( f ∗ χk ) ∗ r = ( f ∗ r ) ∗ χk ≥ f ∗ χk .
Thus by (4.3.5) Δh ( f ∗ χk ) ≥ 0. Therefore f ∗ χk is H-subharmonic.

Finally it remains to be shown that the sequence { f ∗ χk } is non-increasing.
For the proof we require that an H-subharmonic function satisfies

f (rt) dσ (t) ≤ f (ρt) dσ (t) (4.5.3)
S S
whenever 0 < r < ρ < 1. By Remark 4.1.4 following the proof of Corollary
4.1.3, inequality (4.5.3) is valid whenever f is a C2 H-subharmonic function.
Thus since f ≤ f ∗ χk ,

f (rt)dσ (t) ≤ ( f ∗ χk )(rt)dσ (t) ≤ ( f ∗ χk )(ρt)dσ (t).
S S S
But since H-subharmonic functions are locally integrable and bounded above
on compact sets, we have

lim sup ( f ∗ χk )(ρt)dσ (t) ≤ f (ρt)dσ (t),
k→∞ S S
which proves (4.5.3).

Since ( f ∗ χk )(a) = (( f ◦ ϕa ) ∗ χk )(0), it suffices to prove that the sequence
{( f ∗ χk )(0)} is non-increasing. Suppose m > k. Since the support of χk is
contained in {rk+1 < |x| < rk } and rk+1 ≥ rm , we have
rk
2 −n
( f ∗ χk )(0) = n r (1 − r ) χk (r) f (rt) dσ (t) dr
n−1
r S
k+1
≥ f (rm t) dσ (t) χk dτ
S B

= f (rm t) dσ (t) χm dτ
S B
rm
2 −n
≥n r (1 − r ) χm (r) f (rt) dσ (t) dr
n−1
rm+1 S
= ( f ∗ χm )(0).
Corollary 4.5.5 If f is a continuous function on B that is both H-subharmonic

and H-superharmonic on B, then f is C∞ on B and satisfies Δh f (x) = 0 for
all x ∈ B.
Proof. If f satisfies equality in (4.3.1), then by (4.5.2) ( f ∗ χk )(x) = f (x) for

all x ∈ B. Thus f is C∞ on B and satisfies Δh f = 0.
4.6 The Weak Laplacian and Riesz Measure

Our goal in this section is to provide a characterization of H-subharmonic
functions in terms of the weak Laplacian. For this we require Green’s identity
(see [74]) for the invariant Laplacian. If f , g ∈ C2 (B) and one of them has
compact support, then

f Δh g dτ = gΔh f dτ . (4.6.1)
B B

Thus if f is a C2 H-subharmonic function, we have B f Δh ψ dτ ≥ 0 for all
ψ ∈ Cc2 (B) with ψ ≥ 0. Thus we make the following definition.
Definition 4.6.1 For f ∈ Lloc

1 we say that Δ f ≥ 0 in the weak sense if
h

f Δh ψ dτ ≥ 0
B
for all ψ ∈ Cc2 (B) with ψ ≥ 0.
Theorem 4.6.2 If f is H-subharmonic on B, then

f (z)Δh ψ(z) dτ (z) ≥ 0 (4.6.2)
B
for all ψ ∈ Cc2 (B) with ψ ≥ 0. Conversely, if f ∈ Lloc

1 (B) is such that (4.6.2)
holds for all ψ ∈ Cc (B) with ψ ≥ 0, then there exists an H-subharmonic
2
function F on B such that F = f a.e. on B. Furthermore, if B f Δh ψdτ = 0,

then there exists an H-harmonic function F on B such that F = f a.e. on B.
4.6 The Weak Laplacian and Riesz Measure 49
Proof. Let {χk } be a C∞ approximate identity as given in Definition 4.5.2.

Suppose f is H-subharmonic on B. Set fk = f ∗ χk . Then by Theorem 4.5.4,
{ fk } is a non-increasing sequence of C∞ H-subharmonic functions on B that
converges to f everywhere on B. Thus by Green’s identity (4.6.1) and the
monotone convergence theorem,

f Δh ψ dτ = lim fk Δh ψ dτ
B k→∞ B

= lim ψ Δh fk dτ ≥ 0
k→∞ B
for all ψ ∈ Cc2 (B) with ψ ≥ 0. This proves (4.6.2).

Conversely, suppose f ∈ Lloc1 (B) satisfies (4.6.2). Let f be defined as above.
k
Since ϕ0 (x) = −x, if we set J = −I (minus the identity on B), then

( f ∗ [(Δh ψ) ◦ J])(0) = f (x)Δh ψ(x) dτ (x).
B
Thus in the notation of convolutions, hypothesis (4.6.2) is just
( f ∗ [(Δh ψ) ◦ J])(0) ≥ 0
for all ψ ∈ Cc2 (B) with ψ ≥ 0. It follows from the definition that
Δh fk (a) = Δh ( f ∗ χk )(a) = ( f ∗ [(Δh (χk ◦ ϕa )) ◦ J])(0).
Thus since χk ◦ ϕa is C∞ with compact support, Δh fk (a) ≥ 0. Therefore fk is

H-subharmonic on B.
We now show that the sequence { fk } is non-increasing. Suppose k > m and
j is arbitrary. Since fj is H-subharmonic and χm is radial, by Theorem 3.4.3 we
have
fm ∗ χj = ( f ∗ χm ) ∗ χ j = ( f ∗ χj ) ∗ χ m
≥ ( f ∗ χj ) ∗ χ k = f k ∗ χj .
Since fm , fk are continuous, limj→∞ fm ∗ χj = fm , with a similar result for fk .

Thus the sequence { fk } is non-increasing. Define
F(x) = lim fk (x),

k→∞
which exists everywhere on B. As a consequence of the mean-value property,

F is either H-subharmonic on B or F ≡ −∞. But by Lemma 4.5.3, { fk }
converges to f locally in L1 , and thus F = f a.e. on B.
Theorem 4.6.3 If f is H-subharmonic on B, then there exists a unique regular

Borel measure μf on B such that

ψ dμf = f Δh ψ dτ (4.6.3)
B B
for all ψ ∈ Cc2 (B).
Definition 4.6.4 If f is H-subharmonic on B, the unique regular Borel mea-

sure μf satisfying (4.6.3) is called the Riesz measure of f .
Proof. Let f be H-subharmonic on B. By (4.6.2),

L(ψ) = f Δh ψ dτ
B
defines a non-negative linear functional on Cc∞ (B). We extend L to Cc (B) as

follows. Let ψ ∈ Cc (B). Choose a sequence {ψk } ⊂ Cc∞ (B) such that ψk → ψ
uniformly on B. Choose a compact subset K of B such that the supports of ψ
and ψk , k = 1, 2, . . . are contained in K. Let V be a relatively compact open
subset of B such that K ⊂ V, and let h ∈ Cc∞ (B), 0 ≤ h ≤ 1, be such that
h ≡ 1 on K and the support of h is contained in V. Set
k,m = sup |ψk (x) − ψm (x)|.

x∈K
Then for all x ∈ B,
−k,m h(x) ≤ ψk (x) − ψm (x) ≤ k,m h(x).
Thus since L is positive,
|L(ψk ) − L(ψm )| ≤ k,m L(h).
Therefore {L(ψk )} is Cauchy. Define
L(ψ) = lim L(ψk ).

k→∞
It is easy to show that L(ψ) is independent of the choice of {ψk }, and thus
defines a non-negative linear functional on Cc (B). The result now follows by
the Riesz representation theorem for non-negative linear functionals on Cc (B)
([39, Theorem 12.6]).
4.7 Quasi-Nearly H-Subharmonic Functions

Following the notation introduced by M. Pavlović and J. Riihentaus [66] we
make the following definition.
Definition 4.7.1 A non-negative locally integrable function g on B is said to

be quasi-nearly H-subharmonic if there exists a constant C such that

C
g(a) ≤ n g(x)dτ (x)
r E(a,r)
for all a ∈ B and all r, 0 < r < ro for some ro > 0.
If f is a non-negative H-subharmonic function, then so is f p for all p ≥ 1.

Thus by Theorem 4.3.5

1
f p (a) ≤ f p (y)dτ (y).
τ (Br ) E(a,r)
But by (3.3.6)
r r
ρ n−1
τ (Br ) = n dρ ≥ n ρ n−1 dρ = rn .
0 (1 − ρ 2 )n 0
Thus

1
f p (a) ≤ f p (y)dτ (y).
rn E(a,r)
Hence f p is quasi-nearly H-subharmonic for all p ≥ 1. Furthermore, as a

consequence of Theorem 4.7.3, if in addition f is continuous, then the result
is also valid for all p, 0 < p < 1. Thus if h is H-harmonic on B, then |h|p is
quasi-nearly H-subharmonic for all p, 0 < p < ∞.
Remarks 4.7.2 (i) As a consequence of Exercise 4.8.3, a non-negative locally

integrable function g is quasi-nearly H-subharmonic if and only if for each
δ, 0 < δ < δo , for some δo , 0 < δo < 12 , there exists a constant Cδ such that

g(a) ≤ Cδ g(y)dτ (y), (4.7.1)
B(a,cδ (1−|a|2 ))
where cδ = δ/(1 − δ), or equivalently,

Cδ
g(a) ≤ g(y)dν(y). (4.7.2)
(1 − |a|2 )n B(a,cδ (1−|a|2 ))
The last inequality follows from the fact that (1 − |y|2 ) ≈ (1 − |a|2 ) for all y ∈
B(a, δ(1 − |a|2 )). Also, for 0 < δ < 12 , B(a, cδ (1 − |a|2 )) ⊂ B(a, 2δ(1 − |a|2 )).
(ii) A non-negative locally integrable function g is said to be (Euclidean)

quasi-nearly subharmonic if

C
g(a) ≤ n g(y)dν(y) (4.7.3)
r B(a,r)
for all r > 0 such that B(a, r) ⊂ B. If we take r = δ(1 − |a|2 ) for fixed
δ, 0 < δ < 12 , then

Cδ
g(a) ≤ g(y)dν(y). (4.7.4)
(1 − |a|2 )n B(a,δ(1−|a|2 ))
Thus g satisfies the same inequality (4.7.2) as quasi-nearly H-subharmonic

functions. In fact, the two concepts are the same (Exercise 4.8.5).
Quasi-nearly subharmonic functions, perhaps with a different terminology,

have previously been considered by many authors. The concept itself dates
back to C. Fefferman and E. M. Stein [21], who proved that if h is (Euclidean)
harmonic, then |h|p is quasi-nearly subharmonic for all p, 0 < p < 1. This
result was also proved independently by Ü. Kuran [49]. For a non-negative
subharmonic function f , J. Riihentaus [70] and N. Suzuki [92] independently
proved that f p is quasi-nearly subharmonic for all p, 0 < p < 1.
We first prove that if f ≥ 0 is a continuous quasi-nearly H-subharmonic
function, then f p is quasi-nearly subharmonic for all p, 0 < p < ∞. We will
also prove an analogous result for |∇ h H(x)| where H is H-harmonic on B. The
proofs of both Theorems 4.7.3 and 4.7.4 are similar to the proofs given by M.
Pavlović [62] for M-harmonic functions in the unit ball of Cm .
Theorem 4.7.3 Let f be a continuous non-negative quasi-nearly

H-subharmonic function on B. Then for all p, 0 < p < ∞, there exists a
constant C(n, p, r), independent of f , such that

C(n, p, r)
f (a) ≤
p
f p (x)dτ (x)
rn E(a,r)
for all a ∈ B and all r, 0 < r < 1, where

cn Cp (1 − r2 )−(n−1)(p−1) if p ≥ 1,
C(n, p, r) =
C1/p 2n/p if 0 1, by Theorem 4.3.5 and Hölder’s inequality,

1/p
C
f (a) ≤ n f p (x)dτ (x) τ (Br )1/q .
r E(a,r)
But by inequality (3.3.8)

rn
τ (Br ) ≤ cn .
(1 − r2 )n−1
Therefore

1 cn Cp
f p (a) ≤ f p (y)dτ (y).
r (1 − r )(n−1)(p−1)
n 2
E(a,r)
For the case 0 < p < 1 we use a variation of the argument given in
[62, Lemma 5]. Without loss of generality we take C = 1. Let a ∈ B and
fix an r, 0 < r < 1. Set

Cpp = f p (x)dτ (x),
E(a,r)

and let g(x) = f (x)/Cp . Then p
E(a,r) g dτ ≤ 1 and

1
g(x) ≤ g(y)dτ (y) (4.7.5)
sn E(x,s)
for all x ∈ B and all s, 0 < s < 1. Define
A = sup{gp (x)(r − )n : x ∈ E(a, ), 0 < < r}.
Since g is continuous, A is finite. Let x ∈ E(a, ) and let < s < r. Then by
(4.7.5)

g(x)(s − ) ≤n
g(y)dτ (y).
E(x,s−)
Since dρh (x, y) = |ϕx (y)| is a metric on B (Theorem 2.2.3),
E(x, s − ) ⊂ E(a, s) ⊂ E(a, r).
Therefore

g(x)(s − )n ≤ gp (y)g1−p (y)dτ (y)
E(a,s)
(1−p)/p
A
≤ gp (y)dτ (y)
(r − s)n E(a,s)
(1−p)/p
A
≤ .
(r − s)n
Take s = ( + r)/2. Then s − = r − s = (r − )/2, and thus
gp (x)(r − )n ≤ 2n A1−p .

Taking the supremum over x ∈ E(a, ), 0 < < r, gives

A ≤ 2n A1−p or A ≤ 2n/p .
Therefore gp (a)(r − )n ≤ 2n/p for all , 0 < < r. Thus

2n/p
f p (a) ≤ n f p (y)dτ (y),
r E(a,r)
which is the desired inequality.

Theorem 4.7.4 Let H be H-harmonic on B. Then for all δ, 0 < δ < 12 , and
all p, 0 < p < ∞,
Cn,p
(a) |∇ H(a)| ≤ n
h p
|∇ h H(y)|p dτ (y), and
δ
E(a,δ)
Cn,p,δ
(b) |∇ h H(a)|p ≤ |H(y)|p dτ (y),
δn E(a,δ)
where Cn,p and Cn,p,δ are constants depending only on p, n, and δ.
Proof. (a) Fix δ, 0 < δ < 1
2, and let 0 < r < δ. Since y → H(ϕx (y)) is
H-harmonic, we have

H(x) = H(ϕx (0)) = H(ϕx (rt))dσ (t).
S
Hence
∂H(x) ∂H(ϕx (rt))
= dσ (t).
∂xi S ∂xi

Since |y| ≤ i |yi | ≤ n|y| for a vector y = (y1 , . . . , yn ), we have

|∇H(x)| ≤ n |∇x H(ϕx (rt))|dσ (t),
S
and thus
|∇H(0)| ≤ n |∇x H(ϕx (rt))|x=0 dσ (t).
S
However, by Exercise 3.5.10
|∇x H(ϕx (rt))|x=0 ≤ 2r2 |∇H(−rt)| + (1 + r2 )|∇H(−rt)| ≤ 4|∇H(−rt)|.
Therefore

|∇H(0)| ≤ 4n |∇H(−rt)|dσ (t) = 4n |∇H(rt)|dσ (t),
S S
and by Hölder’s inequality

|∇H(0)| ≤ Cn p
|∇H(rt)|p dσ (t)
S
for all p ≥ 1. Multiplying by rn−1 and integrating from 0 to δ gives

Cn
|∇H(0)| ≤ n
p
|∇H(x)|p dν(x).
δ E(0,δ)
But for x ∈ E(0, δ) with δ < we have 34 < (1 − |x|2 ) < 1. Therefore
1
2

Cn
|∇ H(0)| ≤ n
h p
|∇ h H(x)|p dτ (x),
δ E(0,δ)
where Cn is a constant depending only on n. The desired inequality for p ≥ 1
is finally obtained by applying the above to H ◦ ϕa . The result for 0 < p < 1
now follows by Theorem 4.7.3.
∞
(b) Let χ be a non-negative C radial function with supp χ ⊂ Bδ/2 and
B χ dτ = 1. Then by Theorem 4.5.4,

H(x) = (H ∗ χ )(x) = H(y)χ (ϕy (x))dτ (y).
B
Thus
∇ h H(x) = ∇xh χ (ϕy (x))H(y)dτ (y).
B
Hence,
|∇ h H(x)| ≤ Cδ sup |H(y)|,
y∈E(x,δ/2)
where

Cδ = |∇xh χ (ϕy (x))|dτ (y),
E(x,δ/2)
which by the invariance of ∇ h and τ

= |(∇ h h)(ϕx (y))|dτ (y) = |∇ h χ |dτ .
E(x,δ/2) Bδ/2
Thus the constant Cδ is independent of x. Since |H| is H-subharmonic, by

Theorem 4.7.3,

Cn,p
|H(y)| ≤ n
p
|H(y)|p dτ (y),
δ E(y,δ/2)
for all p, 0 < p < ∞, and Cn,p is a constant depending only on n, p. If y ∈

E(x, δ/2), then E(y, δ/2) ⊂ E(x, δ). Therefore

Cn,p,δ
|∇ h H(x)|p ≤ |H(y)|p dτ (y).
δn E(x,δ)

4.8 Exercises
4.8.1. For 1 ≤ i, j ≤ n set
∂ ∂
Ti,j = xi − xj .
∂xj ∂xi
(a) If f is H-harmonic on B, prove that Ti,j f is H-harmonic on B.
(b) Prove that

|x|2 |∇f (x)|2 = |x, ∇f (x)|2 + |Ti,j f (x)|2 .
i<j
4.8.2. (a) For α = 0 compute Δh (1 − |x|2 )α and show that (1 − |x|2 )α is

H-superharmonic for 0 < α ≤ (n − 1), and H-subharmonic for
all α < 0.
(b) Using (a), show that Δh (1 − |x|2 )n−1 = −2n(n − 1)(1 − |x|2 )n .
4.8.3. (a) For 0 < δ < 12 , show that
E(x, δ) ⊂ B(x, cδ (1 − |x|2 )),
where cδ = δ/(1 − δ), and
(b)
B(x, (1 − |x|2 )) ⊂ E(x, δ)
for all , 0 < < δ(1 − δ).
4.8.4. Show that |∇ h f | is quasi-nearly H-subharmonic if and only if |∇f | is
quasi-nearly H-subharmonic.
4.8.5. Prove that a non-negative locally integrable function g is quasi-nearly
H-subharmonic if and only if g is quasi-nearly subharmonic.
4.8.6. If f is an eigenfunction of Δh with non-zero eigenvalue, that is, Δh f =
λf , λ = 0, prove that for 0 < δ < 12 , 0 < p < ∞,

Cn,p,δ
| f (x)| ≤
p
|∇ h f (y)|p dτ (y),
δn E(x,δ)
where Cn,p,δ is a constant depending only on δ, p, and n.
Definition 4.8.1 If μ is a regular Borel measure2 on B, the (invariant)

Green potential of the measure μ, denoted Gμ , is defined on B by

Gμ (x) = Gh (x, y) dμ(y).
B
2 A measure μ defined on the Borel subsets B of a locally compact Hausdorff space is called a
regular Borel measure if (i) μ(K) < ∞ for every compact set K, and for every Borel set B, (ii)
μ(B) = inf{μ(V) : B ⊂ V, V open }, and (iii) μ(B) = sup{μ(K) : K ⊂ B, K compact}.
4.8 Exercises 57
4.8.7. (a) If Gμ ≡ +∞, prove that the function Gμ is H-superharmonic on

B and H-harmonic on B \ supp μ.3
(b) If supp μ is a compact subset of B, prove that
lim Gμ (x) = 0 for every t ∈ S.
x→t
4.8.8. Prove that Gh (·, x) ∈ L p (B, ν) for all p, 0 n, and let

Gf (x) = f (y)Gh (x, y)dν(y).
B
Prove that lim|x|→1 Gf (x) = 0.
4.8.10. (*) (a) If f is continuous on B and satisfies (4.3.3) for all ψ ∈ M(B),
is f H-harmonic on B?
(b) If f ∈ L1 (B) and f satisfies (4.3.3) for all ψ ∈ M(B), determine
the values of the dimension n > 2, if any, for which f is H-harmonic
on B ⊂ Rn .
4.8.11. Let μ be a regular Borel measure on B satisfying

(1 − |x|)dμ(x) < ∞,
B
and set V(x) = Geμ (x), where Ge is the Euclidean Green’s function on
B. In [8] D. H. Armitage proved that V ∈ L p (B) for all p, 0 < p <
n/(n − 1), with

V p (x)dx ≤ A(n, p)V p (0).
B
Investigate the analogous theorem for the invariant Green poten-
tial Gμ .
4.8.12. (*) Let V be as in the previous exercise. In [25] S. J. Gardiner proved
that
(a) If n ≥ 2 and 1 ≤ p < (n − 1)/(n − 2), then

lim (1 − r) (n−1)(p−1)
V p (rt)dσ (t) = 0.
r→1 S
(b) If n ≥ 3 and (n − 1)/(n − 2) ≤ p < (n − 1)/(n − 3), then

lim inf(1 − r) (n−1)(p−1)
V p (rt)dσ (t) = 0.
r→1 S
3 The support of μ, denoted supp μ, is the unique relatively closed subset F of B such that μ(B \
F) = 0 and μ( F ∩ O) > 0 for every open set O for which O ∩ F = φ.
The analogous result for Green potentials with regard to the Laplace–
Beltrami operator on the unit ball in Cm was investigated by the
author in [81] and [82]. Investigate this problem for hyperbolic Green
potentials on B. In this setting a suitable hypothesis on the measure μ
is that it satisfies

(1 − |x|2 )n−1 dμ(x) < ∞.
B
4.8.13. (*) (a) Find necessary and sufficient conditions on an H-harmonic
function f such that |∇ h f |2 is H-subharmonic.
(b) If u is a Euclidean harmonic function on B, then it is known that
log |∇u| is subharmonic when n = 2, and that |∇u|p is subharmonic
whenever p ≥ (n − 2)/(n − 1) when n ≥ 3 [76, Theorem 4.14].
When n ≥ 3, do there exist analogous results for |∇ h f | where f is
H-harmonic on B?

4.8.14. Let : H → B be the mapping given by (2.3.2). If U is H-harmonic
on B, prove that V(x) = U((x)) is H-harmonic on H.
4.8.15. If U is a function on ⊂ Rn \ {en }, define K(U) on () by
K(U) = 2(n−2)/2 |x + en |2−n U((x)).
Prove that
(a) K[K(u)] = u.
(b) K[aU + bV] = aK[U] + bK[V] for a, b ∈ R.
(c) If U is a (Euclidean) harmonic function on , then K[U] is a
(Euclidean) harmonic function on ().
5
The Poisson Kernel and Poisson Integrals
We begin the chapter by using Green’s formula to derive the hyperbolic

Poisson kernel Ph on B. In Section 5.2 we include a theorem of P. Jaming that
expresses the Euclidean Poisson kernel as an integral involving the hyperbolic
Poisson kernel. In Section 5.3 we solve the Dirichlet problem for B, and in
Section 5.4 we use a method due to W. Rudin to solve the Dirichlet problem
for Br . As in [72] we characterize the eigenfunctions of Δh in Section 5.5, and
in Section 5.6 we derive the Poisson kernel for the upper half-space H.
5.1 The Poisson Kernel for Δh

In this section we provide a heuristic argument for the formula for the Poisson
kernel on B. We begin by stating Green’s formula (see [74]) for the invariant
Laplacian: if is an open subset of B, ⊂ B, whose boundary is sufficiently
smooth, then if u, v ∈ C2 () ∩ C1 (),

(uΔh v − vΔh u)dτ = (uDñ v − vDñ u) dσ̃ , (5.1.1)
∂
where σ̃ is the surface element on ∂ with respect to the hyperbolic metric,
and Dñ denotes the normal derivative in the outward normal direction again
with respect to the hyperbolic metric. With = Br , the above becomes

(uΔh v − vΔh u)dτ = (uDñ v − vDñ u) dσ̃ . (5.1.2)
Br Sr
For the surface Sr given by f (x) = 0, where f (x) = |x|2 − r2 , the normal
derivative Dñ of v is given by
∇ h v(rζ ), ∇ h f (rζ )
Dñ v(rζ ) = = (1 − r2 )∇v(rζ ), ζ .
|∇ h f (rζ )|
59
60 The Poisson Kernel and Poisson Integrals
Also, by Theorem 4.1.1 (a),

2 2−n d
Δh u(x)dτ (x) = nr (1 − r )
n−1
f (rζ ) dσ (ζ )
Br dr S

= nrn−1 (1 − r2 )2−n ∇u(rζ ), ζ dσ (ζ )
S
= nr (1 − r )
n−1 2 1−n
Dñ u(rζ ) dσ (ζ ).
S
Thus the surface element dσ̃ on Sr is given by
nrn−1
dσ̃ (rζ ) = dσ (ζ ). (5.1.3)
(1 − r2 )n−1
Suppose now that u is H-harmonic on B, and for the sake of simplicity C1

on B. Let a ∈ B be arbitrary, and let ro > 0 be such that a ∈ Br for all
r, ro ≤ r < 1. Choose > 0 such that E(a, ) ⊂ Bro . Then by Green’s
formula applied to = Br \ E(a, ) with v(x) = gh (ϕa (x)) where gh is given
by (3.2.1),

{uDñ v − vDñ u} dσ̃ = {uDñ v − vDñ u}dσ̃ ,
Sr ϕa (S )
which by the M-invariance of σ̃

= {(u ◦ ϕa )Dñ (v ◦ ϕa ) − (v ◦ ϕa )Dñ (u ◦ ϕa )} dσ̃
S

= {(u ◦ ϕa )Dñ gh − gh Dñ (u ◦ ϕa )}dσ̃ .
S
Clearly

lim gh Dñ (u ◦ ϕa )dσ̃ = 0.
→0 S
On the other hand, since gh is radial,
(1 − 2 )n−1
Dñ gh (ζ ) = (1 − 2 )gh () = − .
n n−1
Thus by (5.1.3)

lim (u ◦ ϕa )Dñ gh dσ̃ = − lim u(ϕa (ζ )) dσ (ζ )
→0 S →0 S
= −u(a).
5.1 The Poisson Kernel for Δh 61
Hence for any r > ro ,

u(a) = − [uDñ v − vDñ u] dσ̃ .
Sr
Since Gh (a, r) ≈ (1 − r2 )n−1 and u ∈ C1 (B),

lim vDñ u dσ̃ ≈ lim (1 − r ) ∇u(rζ ), ζ dσ (ζ ) = 0.
2
r→1 Sr r→1 S
Thus setting Ga (x) = Gh (a, x),

u(a) = − lim uDñ Ga dσ̃
r→1 Sr

= − lim nrn−1 (1 − r2 )2−n u(rζ )∇Ga (rζ ), ζ dσ (ζ )
r→1 S

= Ph (a, ζ )u(ζ ) dσ (ζ ),
S
where
Ph (a, ζ ) = − lim nrn−1 (1 − r2 )2−n ∇Ga (rζ ), ζ . (5.1.4)
r→1
Our next step is to compute the above limit. Since Ga (x) = gh (|ϕa (x)|),
where g is the radial function defined by (3.2.1),
(1 − |ϕa (x)|2 )n−2

∇Ga (x) = − ∇|ϕa (x)|
n|ϕa (x)|n−1
(1 − |a|2 )n−2 (1 − |x|2 )n−2
=− 1
∇|ϕa (x)|.
n|x − a|n−1 ρ(x, a) 2 (n−3)
Using identity (2.1.7) we have

|x − a|2
2|ϕa (x)|∇|ϕa (x)| = ∇ .
ρ(x, a)
From this it now follows that
(x − a)(1 − |x|2 )(1 − |a|2 ) + x|x − a|2 (1 − |a|2 )
∇|ϕa (x)| = 3
.
|x − a|ρ(x, a) 2
Therefore
− nrn−1 (1 − r2 )2−n ∇Ga (rζ ), ζ

rn−1 (1 − |a|2 )n−2
= n (1 − r )(1 − |a| )rζ − a, ζ +r|rζ − a| (1 − |a| ) .
2 2 2 2
|rζ − a|n ρ(rζ , a) 2
Thus
(1 − |a|2 )n−1 |ζ − a|2
− lim rn−1 (1 − r2 )2−n ∇Ga (rζ ), ζ = n ,
r→1 |ζ − a|n ρ(ζ , a) 2
which since ρ(ζ , a) = |ζ − a|2 gives
n−1
1 − |a|2
Ph (a, ζ ) = , (a, ζ ) ∈ B × S. (5.1.5)
|ζ − a|2
Definition 5.1.1 The function Ph on B × S defined by (5.1.5) is called the

invariant Poisson kernel for Δh on B.
A tedious, but straightforward computation proves that for each t ∈ S, the

function x → Ph (x, t) is H-harmonic on B. This is outlined in greater detail in
Section 5.5.
In contrast to the above, the Poisson kernel Pe for the ordinary Laplacian Δ
on B × S is given by
1 − |x|2
Pe (x, t) = , (x, t) ∈ B × S,
|x − t|n
for the invariant Laplacian Δ
whereas the Poisson kernel P on the Hermitian
ball B in C is given by
n
t) = (1 − |z| ) ,
2 n
P(z, (z, t) ∈ B × S.
|1 − z, t|2n
It is only in the unit disc D in R2 that all three agree.
5.2 Relationship between the Euclidean

and Hyperbolic Poisson Kernel
Writing the hyperbolic Poisson kernel Ph as
(1 − |x|2 ) (1 − |x|2 )n−2
Ph (x, t) = ,
|x − t|n |x − t|n−2
we obtain that
(1 − |x|)n−2 Pe (x, t) ≤ Ph (x, t) ≤ 2n−2 Pe (x, t).
In this section we prove the following much stronger result due to P. Jaming
[42].
5.2 Relationship between the Euclidean and Hyperbolic Poisson Kernel 63
Theorem 5.2.1 There exists a function η : [0, 1] × [0, 1] → R+ such that

1
(a) Pe (rζ , t) = η(r, ρ)Ph (ρrζ , t)dρ, and
0
(b) there exists a constant C, independent of r, such that
1
η(r, ρ)dρ ≤ C
0
for every r ∈ [0, 1].
Proof. Since
∞ tn/2−2 ( n2 − 1)( n2 )
= ,
0 (1 + t)n−1 (n − 1)
where is the gamma function, we have
∞
1 sn/2−2
n = cn ds.
(x + y) 2 0 (x + y + s)n−1
Since
1 − r2
Pe (rζ , t) = n ,
[(1 − r)2 + 2r(1 − ζ , t)] 2
writing X = 2(1 − ζ , t) we obtain
(1 − r2 ) (1 − r2 )
Pe (rζ , t) = n = n n
|(1 − r)2 + rX] r [(1 − r)2 /r + X] 2
2 2

(1 − r2 ) ∞ sn/2−2 ds
= cn n .
r2 0 [X + (1 − r)2 /r + s]n−1
The change of variable
(1 − ρ)2 (1 − r)2 (r − ρ)(1 − ρr)
s= − =
ρ r ρr
gives

(1 − r2 ) r [(r − ρ)(1 − ρr)]n/2−2 (1 − ρ 2 )
Pe (rζ , t) = cn n !n−1 dρ
r 2 0 X + (1 − ρ)2 /ρ (ρr)n/2−2 ρ2

(1 − r2 )
[(r − ρ)(1 − ρr)]n/2−2 ρ n/2−1 (1 − ρ 2 )
r
= cn dρ
rn−2 0 [ρX + (1 − ρ)2 ]n−1

(1 − r)2 r [(r − ρ)(1 − ρr)]n/2−2 ρ n/2−1
= cn n−2 Ph (ρζ , t) dρ.
r 0 (1 − ρ 2 )n−2
Finally with the change of variable ρ = rs, s ∈ (0, 1), we obtain
1
Pe (rζ , t) = Ph (rsζ , t) η(r, s) ds,
0
where
[(1 − s)(1 − sr2 )]n/2−2 sn/2−1
η(r, s) = cn (1 − r2 ) . (5.2.1)
(1 − r2 s2 )n−2
For the proof of (b) we first note that
(1 − r2 s2 )n−2 ≥ (1 − s)n/2−2 (1 − sr2 )n/2 .
Therefore,
1 1
I(r) = η(r, s)ds ≤ cn (1 − r ) 2
(1 − sr2 )−2 sn/2−1 ds,
0 0
which by the change of variable u = sr2

r2
(1 − r2 )
= cn (1 − u)−2 un/2−1 du.
rn 0
For r2 < 12 ,
r2
(1 − r2 ) 2
I(r) ≤ Cn un/2−1 du ≤ Cn (1 − r2 ) ≤ Cn .
rn 0 n
1
On the other hand, for r2 > 2,
r2
I(r) ≤ Cn (1 − r ) 2
(1 − u)−2 du ≤ Cn .
0
Thus I(r) ≤ C for all r ∈ [0, 1].
Remark 5.2.2 One other interesting relationship between H-harmonic and

Euclidean harmonic functions is the following result of H. Samii [73]: if u
is H-harmonic on B, then there exists a Euclidean harmonic function v with
v(0) = 0 such that
1
dt
u(x) = u(0) + v(tx)[(1 − t)(1 − tr2 )]n/2−1 (5.2.2)
0 t
for all x ∈ B.
5.3 The Dirichlet Problem for B

We summarize some of the properties of the invariant Poisson kernel in the
following lemma. These are analogous to the properties of the Poisson kernel
for Δ on B.
Lemma 5.3.1 The Poisson kernel Ph on B × S satisfies the following:

(a) For fixed t ∈ S, x → Ph (x, t) is H-harmonic on B,
(b) Ph (rζ , t) = Ph (rt, ζ ) for all t, ζ ∈ S,
(c) Ph (x, t) dσ (t) = 1, and
S
(d) For fixed ζ ∈ S and δ > 0, lim Ph (x, t) dσ (t) = 0.
x→ζ |t−ζ |>δ
x∈B
Proof. As indicated above, (a) follows by computation, and (b) is almost

obvious. Writing x = rζ , (c) follows by (b) and the mean-value property for
H-harmonic functions. The proof of (d) is again standard.
Definition 5.3.2 For f ∈ L1 (S), the Poisson integral of f denoted Ph [ f ] is
defined by

Ph [ f ](x) = Ph (x, t)f (t) dσ (t).
S
Similarly, if μ is a finite signed Borel1 measure on S, the Poisson integral of μ

will be denoted by Ph [μ], that is,

Ph [μ](x) = Ph (x, t) dμ(t).
S
Since the function t → Ph (x, t) is continuous on S, the above integrals exist
and are finite for all x ∈ B. Furthermore, as a consequence of the mean-value
property, the function Ph [μ](x) is H-harmonic on B. Since the Poisson kernel
is an approximate identity, we have the following theorem.
Theorem 5.3.3 Let f be a bounded measurable function on S. Then
lim Ph [ f ](x) = f (ζ )
x→ζ
x∈B
at each ζ ∈ S where f is continuous.

Proof. Suppose | f (t)| ≤ M for all t ∈ S and that f is continuous at ζ ∈ S.
Then given > 0 there exists a δ > 0 such that | f (t) − f (ζ )| < for all t ∈ S
with |t − ζ | < δ. Set Q(ζ , δ) = {t ∈ S : |t − ζ | < δ}. Then

|Ph [ f ](x) − f (ζ )| = Ph (x, t)f (t)dσ (t) − f (ζ ) ,

S
which by Lemma 5.3.1(c),
1 Recall that the Borel subsets of S are the smallest σ -algebra of subsets of S such that every
continuous function is measurable.

≤ Ph (x, t)| f (t) − f (ζ )| dσ (ζ )
S

≤ Ph (x, t)| f (t) − f (ζ )|dσ (t) + 2M Ph (x, t)dσ (t)
Q(ζ ,δ) S\Q(ζ ,δ)

≤ + 2M Ph (x, t)dσ (t).
|t−ζ |≥δ
The result now follows by Lemma 5.3.1(d).

Corollary 5.3.4 (a) If f ∈ C(S), then the function F defined by

Ph [ f ](x), x ∈ B,
F(x) =
f (x), x ∈ S,
is H-harmonic on B and continuous on B with F∞ = f ∞ .2

(b) Conversely, if f is H-harmonic on B and continuous on B, then f (x) =
Ph [ f ](x).
Proof. Part (a) is an immediate consequence of the previous theorem, and
the proof of (b) is a consequence of the maximum principle (Theorem 4.1.6)
applied to F(x) = f (x) − Ph [ f ](x).
An immediate consequence of the previous is the following analogue of [72,
Theorem 3.3.8].
Theorem 5.3.5 If f ∈ L1 (S), then Ph [ f ◦ ψ] = Ph [ f ] ◦ ψ for all ψ ∈ M(B).
Proof. Since C(S) is dense in L1 (S), it suffices to prove the result for
continuous functions on S. If f ∈ C(S), then by the previous theorem F(x) =
Ph [ f ](x) is H-harmonic on B and continuous on B. Suppose ψ ∈ M(B). Since
ψ is continuous on B, F ◦ ψ is also H-harmonic on B, continuous on B with
limx→ζ (F ◦ ψ)(x) = f (ψ(ζ )) for all ζ ∈ S. Thus (F ◦ ψ)(x) = Ph [ f ◦ ψ](x).
On the other hand, (F ◦ ψ)(x) = F(ψ(x)) = Ph [ f ](ψ(x)), which proves the
result.
If in the above proof we take ψ = ϕa and x = 0, then

f (ϕa (t)) dσ (t) = Ph (a, t)f (t) dσ (t) (5.3.1)
S S
for all f ∈ L1 (S).
2 See (5.4.1) for the definition of .

∞
Theorem 5.3.6 Let ν be a finite signed Borel measure on S. Then

(a) for all ψ ∈ C(S),

lim Ph [ν](rζ )ψ(ζ )dσ (ζ ) = ψ(ζ )dν(ζ ).
r→1 S S
(b) If Ph [ν](x) = 0 for all x ∈ B, then ν = 0.
Proof. (a) Since Ph (rt, ζ ) = Ph (rζ , t), by Fubini’s theorem we have

lim Ph [ν](rt)ψ(t)dσ (t) = lim Ph [ψ](rζ ) dν(ζ ),
r→1 S r→1 S
which by Corollary 5.3.4

= ψ(ζ ) dν(ζ ).
S

(b) By (a), the hypothesis implies that S ψdν = 0 for every continuous
function ψ on S. Suppose ν = 0. Let (P, P) be a Hahn decomposition
for ν and let ν + and ν − be the positive and negative variation of ν, that is,
ν + (E) = ν(E ∩ P) and ν − (E) = −ν(E ∩
P)
for all measurable sets E. (See [71] for details.) Since ν = 0 we can assume,
without loss of generality, that ν + (P) > 0. Choose F a compact subset of P
and > 0 such that ν + (F) − > 0. Since all Borel measures on S are regular,
there exists a compact subset C of P \ C) < . Let U =
P such that ν − ( C. Then
F ⊂ U, and by Urysohn’s lemma (see [71, Chapter 3]) there exists a continuous
function ψ on S with 0 ≤ ψ ≤ 1, ψ = 1 on F, and supp ψ ⊂ U. Therefore

+ +
ν (F) = ψ dν ≤ ψ dν
F P
and
ψ dν = ψ dν − ≤ ν − (
P \ C) < .

P
P∩U
Therefore,

0 < ν + (F) − ≤ ψ dν − ψ dν = ψ dν = 0,
P
P S
which is a contradiction.
There is a significant difference between invariant Poisson integrals and
solutions of the classical Dirichlet problem for the Laplacian Δ. Since Δ is
uniformly elliptic, if f ∈ C∞ (S) and F is the Euclidean Poisson integral of f ,
then F ∈ C∞ (B). The following example shows that this fails dramatically for
invariant Poisson integrals.
Example 5.3.7 To illustrate the above we compute the invariant Poisson

integral of the function f (t) = t12 for n = 3. In this case

t12
F(x) = (1 − |x|2 )2 dσ (t).
S |x − t|
4
With x = re1 , where e1 = (1, 0, 0),

t12
F(re1 ) = (1 − r )
2 2
dσ (t),
S (1 + r − 2rt1 )
2 2
which since the integrand is a function of t1 only (see [10, p. 216])

1
x2
= 12 (1 − r2 )2 dx
−1 (1 + r − 2rx)
2 2

1 1−r
= 3 2r(1 + r ) + (1 + r )(1 − r ) log
4 2 2 2
.
4r 1+r
Note, at r = 0, the term in brackets is of the form 43 r3 + O(r5 ), and thus F(re1 )
is indeed continuous at 0. Even though f (t) = t12 is C∞ on S, the function
F(re1 ) is not C2 at the boundary point e1 . A formula valid for all x ∈ B will be
given in Example 6.1.3.
5.4 The Dirichlet Problem for Br

In Theorem 5.3.4 we proved that the Dirichlet problem for Δh was solvable for
the unit ball B. In the following theorem we prove that the Dirichlet problem
is also solvable for Br , where for 0 < r < 1,
Br = {y : |y| < r} = {rx : x ∈ B}.
Similarly, set
Sr = {y : |y| = r} = {rt : t ∈ S}.
Our method of proof is analogous to that used by W. Rudin in [72] in proving

the same theorem for M-harmonic functions in the unit ball of Cn .
Theorem 5.4.1 Fix r, 0 < r < 1. If f ∈ C(Sr ), then there exists F ∈ C(Br )
such that
(a) Δh F = 0 in Br ,
(b) F(rt) =f (rt) for all t ∈ S, and
(c) F(0) = f (rt)dσ (t).
S
5.4 The Dirichlet Problem for Br 69
Proof. Let H(Br ) denote the class of functions F which are H-harmonic in Br
and continuous on Br . For F ∈ H(Br ) set
F∞ = sup{| F(y)| : y ∈ Br } = sup{| F(rt)| : t ∈ S}. (5.4.1)
The last equality follows from the maximum principle for H-subharmonic
functions. Let
H(Sr ) = { f ∈ C(Sr ) : f (rt) = F(rt) for some F ∈ H(Br )}.
Then H(Sr ) with the sup norm · ∞ is a closed subspace of C(Sr ), which by
(5.4.1) is isomorphic to H(Br ).
We now show that H(Sr ) = C(Sr ). If not, then by the Hahn–Banach theorem
there exists a non-trivial continuous linear functional γ on C(Sr ) such that
γ ( f ) = 0 for all f ∈ H(Sr ). Thus by the Riesz representation theorem there
exists a finite signed Borel measure ν on S such that

f (rt) dν(t) = 0 for all f ∈ H(Sr ).
S
In particular,

Ph [ψ](rt) dν(t) = 0 for all ψ ∈ C(S).
S
Thus by Fubini’s theorem

Ph [ψ](rt) dν(t) = Ph [ν](rζ )ψ(ζ )dσ (ζ ) = 0
S S
for all ψ ∈ C(S). Therefore Ph [ν](rζ ) = 0 for all ζ ∈ S. Hence by the
maximum principle Ph [ν](x) = 0 for all x ∈ Br . Since H-harmonic functions
are real analytic,3 Ph [ν](x) = 0 for all x ∈ B. Thus by Theorem 5.3.6 we have
that ν = 0.
Finally, to show that F(0) = S f (rt)dσ (t), define γ0 on H(Br ) by γ0 (F) =
F(0). This defines a continuous linear functional on C(Sr ) in the obvious way.
Thus there exists a signed Borel measure ν on D such that

F(rt)dν(t) = F(0) (5.4.2)
S
for all F ∈ H(Br ). Let μ = ν − σ . Then if ψ ∈ C(S),

Ph [μ](rt)ψ(t)dσ (t) = Ph (rt, ζ )dν(ζ ) − Ph (0, ζ ) ψ(t)dσ (t) = 0.
S S S
3 The fact that H-harmonic functions are real analytic follows from the following theorem of
Hörmander [40, Theorem 7.5.1]: if L is an elliptic differential operator with real analytic
coefficients, then every solution of Lu = 0 is real analytic.
The last equality follows by (5.4.2) since Ph (·, t) ∈ H(Br ). Thus, as above,
μ = 0, that is, ν = σ .
In Remark 4.1.4 it was pointed out that if f ∈ C2 (B) is H-subharmonic on

B, then the integral mean

M( f , r) = f (rt)dσ (t)
S
is a non-decreasing function of r, 0 < r < 1. We now use the previous theorem

to prove that this in fact holds for all H-subharmonic functions.
Theorem 5.4.2 If f is H-subharmonic on B, then for every a ∈ B,

f (ϕa (rt))dσ (t)
S
is a non-decreasing function of r, 0 < r < 1.
Proof. Without loss of generality we take a = 0. Suppose 0 < r1 < r2 < 1.

We first suppose that f is continuous on B. By Theorem 5.4.1 there exists a
function G which is H-harmonic on Br2 with G(r2 t) = f (r2 t). Thus by the
maximum principle f (x) ≤ G(x) for all x ∈ Br2 . Therefore

f (r1 t)dσ (t) ≤ G(r1 t)dσ (t) = G(0) = G(r2 t)dσ (t) = f (r2 t)dσ (t).
S S S S
For arbitrary f , since f is upper semicontinuous, by Theorem 4.3.2(d) there

exists a decreasing sequence { fn } of continuous functions on Sr2 with
fn (r2 t) → f (r2 t) for all t ∈ S. For each n let Fn ∈ H(Br2 ) be such that
Fn (r2 t) = fn (r2 t) for all t ∈ S. Since Fn ≥ f on Sr2 , by the maximum principle
Fn ≥ f on Br2 . Therefore,

f (r1 t)dσ (t) ≤ Fn (r2 t)dσ (t) = fn (r2 t)dσ (t).
S S S
The result now follows by letting n → ∞.
5.5 Eigenfunctions of Δh
In this section we consider the eigenfunctions of the invariant Laplacian Δh . In
the setting of rank one noncompact symmetric spaces this topic has previously
been considered by K. Minemura [58]. Our approach follows that of W. Rudin
in [72].
For fixed t ∈ S and α ∈ R, consider the function Pα (x, t). Since x → Ph (x, t)
is H-harmonic, by Exercise 3.5.3(b)
Δh Pαh (x, t) = α(α − 1)Ph (x, t)α−2 |∇xh Ph (x, t)|2 .
But |∇xh Ph (x, t)|2 = (1 − |x|2 )2 |∇x Ph (x, t)|2 . It is left as an exercise (Exercise
5.7.3) to show that
(1 − |x|2 )2n−4
|∇x Ph (x, t)|2 = 4(n − 1)2 . (5.5.1)
|x − t|4n−4
Therefore
(1 − |x|2 )2n−2
|∇xh Ph (x, t)|2 = 4(n − 1)2
|x − t|4n−4
= 4(n − 1)2 Ph (x, t)2
and
Δh Pαh (x, t) = 4(n − 1)2 α(α − 1)Pαh (x, t). (5.5.2)
Thus Pαh (x, t) is an eigenfunction of Δh with eigenvalue 4(n − 1)2 α(α − 1).
Definition 5.5.1 For λ ∈ R, let Hλ denote the space of all f ∈ C2 (B) that
satisfy
Δh f = λf .
As a consequence of the above we have the following analogue of [72,
Theorem 4.2.2].
Theorem 5.5.2 If α and λ are related by λ = 4(n − 1)2 α(α − 1), then Hλ
contains every f of the form

f (x) = Pαh (x, t) dμ(t), x ∈ B, (5.5.3)
S
where μ is a signed Borel measure on S. In particular, Hλ contains the radial
function gα defined by

gα (x) = Pαh (x, t)dσ (t), x ∈ B. (5.5.4)
S
We now proceed to prove the following analogues of [72, Theorems 4.2.3,
4.2.4]. The proofs, with minor changes, follow those given by W. Rudin in [72]
In the following two theorems we will assume that α
for eigenfunction of Δ.
and λ are related by
λ = 4(n − 1)2 α(α − 1).
Theorem 5.5.3 If f ∈ Hλ and f is radial, then f (x) = f (0)gα (x).
Proof. Let f ∈ C2 (B) be a radial function satisfying Δh f = λf . Write f (x) =

u(|x|2 ). Then by (4.1.3)
Δh f (x) = 4r2 (1 − r2 )2 u (r2 ) + 2n(1 − r2 )2 u (r2 ) + 4(n − 2)r2 (1 − r2 )u (r2 ).
Hence the equation Δh f = λf converts to
Lu = λu, (5.5.5)
where
L(u)(t) = a(t)u (t) + b(t)u (t),
with
a(t) = 4t(1 − t)2 and b(t) = 2n(1 − t)2 + 4(n − 2)t(1 − t). (5.5.6)
The equation L(u) = 0 has a solution (see Exercise 5.7.5(a))

1
1
u0 (t) = s− 2 n (1 − s)n−2 , 0 < t < 1, (5.5.7)
t
which is unbounded as t → 0.
Suppose uλ is a solution of (5.5.5) with uλ (0) = 1, and v is a solution of
u2λ v = u0 for small t > 0 where uλ (t) = 0. Set w = uλ v. Then Lw = λw
(see Exercise 5.7.5(b)). Since v is unbounded as t → 0, we also have that w
is unbounded at 0. Since the solution space of (5.5.5) is two dimensional, the
solutions of (5.5.5) that are bounded as t → 0 form a space of dimension one.
Since gα is a bounded solution of Δh f = λf , every f in C2 (B) ∩ Hλ satisfies
f (x) = agα (x) for some constant a. Since gα (0) = 1 the result follows.
Corollary 5.5.4 gα = g1−α .
Corollary 5.5.4 follows since the definition of λ is unchanged if α is replaced

by 1 − α.
Theorem 5.5.5 Every f ∈ Hλ satisfies

f (ψ(rt))dσ (t) = gα (rη)f (ψ(0)) (5.5.8)
S
for every ψ ∈ M(B), 0 ≤ r < 1, η ∈ S. Conversely, if f ∈ C(B) and f satisfies

(5.5.8), then f ∈ C∞ (B) and f ∈ Hλ .
Proof. Suppose f ∈ Hλ and let f be the radialization of f given by (4.1.1).

Since f ∈ Hλ we have f = f (0)gα . If we replace f by f ◦ ψ, ψ ∈ M(B), we
obtain (5.5.8).
Conversely, suppose f ∈ C(B) satisfies (5.5.8). Let h ∈ Cc∞ (B) be a radial

function such that

h(y)gα (y)dτ (y) = 1.
B
In (5.5.8) take ψ = ϕx . Then with r = |y|, by (3.2.5)

f (ϕx (Ay))dA = f (ϕx (rζ ))dσ (ζ ) = gα (y)f (x). (5.5.9)
O(n) S
Multiplying the above by h(y)dτ (y) and integrating gives

f (x) = f (x)gα (y)h(y)dτ (y) = f (ϕx (Ay))h(y)dAdτ (y),
B B O(n)
which by Fubini’s theorem and the change of variable w = ϕx (Ay)

= h(A−1 ϕx (w))f (w)dτ (w)dA.
O(n) B
Thus since h is radial, we have

f (x) = f (w)h(ϕx (w))dτ (w).
B
As a consequence, f is C∞ on B.
If in (5.5.9) we compute Δh with respect to y we obtain
λgα (y)f (x) = (Δh )y [gα (y)f (x)]

= (Δh )y [ f ◦ ϕx ◦ A](y)dA.
O(n)
But by the invariance of Δh ,
(Δh )y [ f ◦ ϕx ◦ A](y) = (Δh f )(ϕx (Ay)).
Therefore

λgα (y)f (x) = (Δh f )(ϕx (Ay))dA.
O(n)
Setting y = 0 gives

λ f (x) = (Δh f )(ϕx (0))dA = Δh f (x),
O(n)
which proves the result.
The following corollary provides another proof of Corollary 4.5.5.

Corollary 5.5.6 If f is a continuous function on B satisfying

f (ψ(0)) = f (ψ(rt))dσ (t)
S
for all ψ ∈ (B) and 0 < r < 1, then f ∈ C∞ (B) with Δh f = 0.
We close this section by obtaining asymptotic estimates for the function gα .

By Equation (5.5.4) the function gα is given by

dσ (ζ )
gα (x) = (1 − |x| )
2 α(n−1)
. (5.5.10)
S |x − ζ |
2α(n−1)
For γ > 0, 0 < r < 1, and t ∈ S, set

Iγ (r) = |rt − ζ |−γ dσ (ζ ).
S
The following asymptotic estimates for Iγ appear to be well known, but are
usually stated without a reference to a proof.
Theorem 5.5.7 For 0 < r < 1 and γ > 0,

⎧
⎪
⎨ 1 if γ < (n − 1),
Iγ (r) ≈ 1
log (1−r) if γ = (n − 1),
⎪
⎩ (1 − r)−γ +(n−1) if γ > (n − 1).
Proof. Since the result is well known when n = 2 ([72, Proposition 1.4.10]),
we will assume that n ≥ 3. Set e1 = (1, 0 ) where 0 is the zero in Rn−1 . Then
since σ is invariant under O(n),

Iγ (r) = |re1 − ζ |−γ dσ (ζ ),
S
which by writing ζ = (ζ1 , ζ ), ζ ∈ Rn−1 ,

dσ (ζ )
= .
S [|r − ζ1 |2 + |ζ |2 ]γ /2
Consider the map : (−1, 1) × Sn−1 → Sn defined by

(x, ζ ) = (x, 1 − x2 ζ ).
The map is one-to-one and range = Sn \ {(x, ζ ) : ζ = 0}. By [10,

Appendix A.4],
n−3
dσn ((x, ζ )) = (1 − x2 ) 2 dx dσn−1 (ζ )
and
1
n−3
fdσn = cn (1 − x2 ) 2 f (x, 1 − x2 ζ )dσn−1 (ζ )dx, (5.5.11)
Sn −1 Sn−1
where cn is a constant depending only on n. Therefore

1 n−3
(1 − x2 ) 2 dx
Iγ (r) = cn .
−1 [|r − x| + (1 − x )]
2 2 γ /2
Since (r − x)2 + (1 − x2 ) = 1 − 2rx + r2

1 n−3
(1 − x2 ) 2 dx
Iγ (r) = cn . (5.5.12)
−1 [1 − 2rx + r ]
2 γ /2
With the change of variable x = 1 − t,

2 n−3 n−3
t 2 (2 − t) 2
Iγ (r) = cn dt.
0 [(1 − r) + 2rt]
2 γ /2
Since we are only interested in the behavior of Iγ (r) as r → 1, we assume that

r ≥ 12 . Thus since n ≥ 3 we have
2 n−3
n−3 t 2 dt
Iγ (r) ≤ 2 2 cn
0 [(1 − r)2 + t]γ /2
2 n−3 γ
≤ Cn t 2 −2 dt,
0
which is finite provided γ < (n − 1).
Suppose now that γ ≥ (n − 1). We now make the change of variable t =
(1 − r)2 w to obtain
2/(1−r)2 n−3
w 2 dw
Iγ (r) ≤ Cn (1 − r)n−1−γ .
0 [1 + w]γ /2
Therefore

1 2/(1−r)2 1
−γ /2
Iγ (r) ≤ Cn (1 − r) n−1−γ
(1 + w) dw + w 2 (n−3−γ ) dw
0 1

2/(1−r)2 1
≤ Cn (1 − r) n−1−γ
Cγ + w 2 (n−3−γ ) dw .
1
If γ = (n − 1), then

2/(1−r)2 1 1
Iγ (r) ≤ Cn Cγ + dw ≤ Cn,γ log .
1 w (1 − r)
If γ > (n − 1), then 12 (n − 3 − γ ) < −1 and thus

∞
1
Iγ (r) ≤ Cn (1 − r) n−1−γ
Cγ + w 2 (n−3−γ )
dw ≤ Cn,γ (1 − r)n−1−γ .
1
For the lower estimate we have that

2 n−3 n−3
t 2 (2 − t) 2
Iγ (r) = cn dt
0 [(1 − r) + 2rt]
2 γ /2
1
cn 1
≥ γ /2 t 2 (n−3−γ ) dt.
2 0
If γ < (n − 1), then the above integral is finite. For γ ≥ (n − 1) we have
1
1
Iγ (r) ≥ Cn,γ t 2 (n−3−γ ) dt,
(1−r)
from which the result follows.
Combining the above with (5.5.10) gives the following corollary.
Corollary 5.5.8
⎧
⎨ (1 − |x| ) 1
⎪ if α < 12 ,
2 α(n−1)
gα (x) ≈ (1 − |x|2 ) 2 (n−1) log (1−|x|

1
if α = 12 ,
⎪
⎩
2)
(1 − |x| )
2 (1−α)(n−1) if α > 12 .
As a consequence of the above we have
(a) If 0 < α ≤ 12 , then gα is continuous on B with lim|x|→1 gα (x) = 0, that
is, gα ∈ C0 (B).
(b) If α = 0, then gα is bounded on B but gα ∈ C0 (B).
(c) If α < 0, then gα is unbounded.
5.6 The Poisson Kernel on H

Recall that the upper half-space H in Rn is the set
H = {z = (x, y) ∈ Rn : x ∈ Rn−1 , y > 0}.
As is usual we identify Rn−1 with Rn−1 × {0}. With this convention we have
that ∂H = Rn−1 .
For z = (x, y) ∈ H, x ∈ Rn−1 , y > 0, and t ∈ Rn−1 , set
n−1
y
PH (z, t) = cn , (5.6.1)
|x − t|2 + y2
5.6 The Poisson Kernel on H 77

where cn is chosen so that Rn−1 PH (z, t)dt = 1. The function PH is called
the Poisson kernel for the upper half-space H. As a consequence of Exercises
5.7.16 and 5.7.17 it follows that for fixed t ∈ Rn−1 the function x → PH (x, t)
is H-harmonic on H.
For y > 0 set
n−1
y
vy (x) = cn .
|x|2 + y2
The function vy has the following property: v1 ( xy ) = yn−1 vy (x). Consider

Rn−1 vy (x)dx, where dx denotes Lebesgue measure on R
n−1 . By the change

of variable x = yx we have

vy (x)dx = vy (yx )yn−1 dx
R
R
n−1 n−1

dx
= v1 (x )dx = cn .
Rn−1 (|x| + 1)
2 n−1
Rn−1

Hence vy (x)dx is independent of y. Since v1 (x)dx < ∞, we can choose cn
such that
dx
vy (x)dx = cn 2 + 1)n−1
= 1. (5.6.2)
R n−1 R n−1 (|x|
By expressing the integral in (5.6.2) in polar coordinates it is easily shown that
2(n − 1)
cn = ,
(n − 1)V(Bn−1 ) 2 ( 12 (n − 1))
where is the gamma function (Exercise 5.7.11).
The above computations show that PH (z, t) is positive, H-harmonic on H,
and with the above choice of cn satisfies

PH (z, t) dt = 1, z = (x, y) ∈ H. (5.6.3)
Rn−1
The next result proves that PH is an approximate identity on H.
Lemma 5.6.1 For every a ∈ Rn−1 and δ > 0,

lim PH (z, t)dt = 0.
z→a |t−a|>δ
Proof. We prove the result for n ≥ 3, with the obvious modifications when
n = 2. Let δ > 0 be given. Since PH (z, t) ≤ cn yn−1 |x − t|−2(n−1) , for |x − a| <
n−1 |a − t|−2(n−1) . Therefore
2 , we have PH (x, t) ≤ Cn y
δ

dt
PH (z, t) dt ≤ Cn yn−1 .
|a−t|>δ |a − t|
2(n−1)
|a−t|>δ
Expressing the above integral in polar coordinates yields

dt Cn
= ,
|a−t|>δ |a − t| 2(n−1) (n − 2)δ n−2
For 1 ≤ p ≤ ∞ let L p (Rn−1 ) denote the space of Lebesgue measurable
functions on Rn−1 for which f p < ∞, where for 1 ≤ p < ∞
1/p
f p = | f (x)|p dx ,
Rn−1
and for p = ∞, f ∞ denotes the essential supremum of f . The Poisson
integral of f ∈ L p (Rn−1 ), 1 ≤ p ≤ ∞, is the function

PH [ f ](z) = PH (z, t)f (t)dt.
Rn−1
Since PH (z, ·) ∈ Lq (Rn−1 ) for every q ∈ [1, ∞] (Exercise 5.7.13), PH [ f ](z) is

well defined for every z ∈ H.
Theorem 5.6.2 (Dirichlet problem for H) Suppose f is continuous and
bounded on Rn−1 . For z = (x, y) ∈ H, define F on H by

PH [ f ](z), z ∈ H,
F(z) =
f (x), x ∈ Rn−1 .
Then F is continuous on H and H-harmonic on H with | F(z)| ≤ f ∞ for all

z ∈ H.
Proof. That | F(z)| ≤ f ∞ is an immediate consequence of (5.6.3). Let a ∈
Rn−1 and δ > 0. Then

| F(z) − f (a)| = PH (z, t)( f (t) − f (a))dt
R
n−1

≤ PH (z, t)| f (t) − f (a)|dt + 2 f ∞ PH (z, t)dt
|t−a|≤δ |t−a|>δ
for all z ∈ H. The result now follows as in Theorem 5.3.3 from the continuity
of f at a and Lemma 5.6.1.
5.7 Exercises
5.7.1. (a) If fˆ ∈ L p , 1 ≤ p < ∞, prove that

lim |Ph [ fˆ](rt) − fˆ(t)| p dσ (t) = 0.
r→1 S
5.7 Exercises 79
(b) If fˆ ∈ L∞ (S), prove that

lim Ph [ fˆ](rt)g(t)dσ (t) = fˆ(t)g(t)dσ (t)
r→1 S S
for all g ∈ L1 (S).

5.7.2. Suppose U is a positive H-harmonic function on B and that U extends
continuously to B \ {ζ }, ζ ∈ S, with U = 0 on S \ {ζ }. Prove that
U(x) = cPh (x, ζ ) for some c > 0.
5.7.3. Prove that
(1 − |x|2 )2(n−2)
|∇x Ph (x, t)|2 = 4(n − 1)2 .
|x − t|4(n−1)
5.7.4. If f ∈ Hλ , λ ≥ 0, prove that | f | p is H-subharmonic for all p,
1 ≤ p < ∞.
5.7.5. (a) Prove that u0 (t) given by (5.5.7) is a solution of
au + bu = 0
where a and b are given by (5.5.6).
(b) Let uλ and v be as given in the proof of Theorem 5.5.3, and let
w = uλ v. Prove that Lw = λw.
5.7.6. Let { fn } be a sequence in Hλ that converges to f uniformly on compact
subsets of B. Prove that f ∈ Hλ .
5.7.7. Given α, −∞ < α ≤ 12 , determine all values of p such that gα ∈
L p (B, τ ).
5.7.8. Suppose f (x) = Ph [ fˆ](x) where fˆ ∈ L1 (S). Prove that
|∇ h f (x)| ≤ 2(n − 1)Ph [| fˆ|](x),
5.7.9. For f ∈ L1 (S), let u(x) = Ph [ f ](x) and v(x) = Pe [ f ](x). Show that
1
v(x) = η(|x|, s)u(sx)ds,
0
where η is given by (5.2.1).
5.7.10. One radius theorem [72, Theorem 4.3.4]. The following theorem is
true for M-harmonic functions on the unit ball in Cn .
Theorem Suppose u ∈ C(B) and suppose that to every z ∈ B there
corresponds one radius r(z), 0 < r(z) < 1, such that

u(z) = u(ϕz (r(z)ζ )dσ (ζ ).
S
Then u is M-harmonic in B.
The analogue of this result for Euclidean harmonic functions may be
found in [10].
Question (*): Is the analogous result true for H-harmonic functions
in B?

5.7.11. Compute the value of cn in Equation (5.6.2).
5.7.12. Prove Lemma 5.6.1.
5.7.13. Prove that PH (z, ·) ∈ Lq (Rn−1 ) for every q ∈ [1, ∞].
For a function U on H and y > 0, we let Uy denote the function on
Rn−1 defined by
Uy (x) = U(x, y).
The functions Uy play the same role on H that the dilations fr , 0 <
r < 1, play on the unit ball B.
5.7.14. Suppose 1 ≤ p < ∞. If f ∈ L p (Rn−1 ) and U = PH [ f ], prove that
(a) Uy p ≤ f p for all y > 0,
f p
(b) |Uy (x)| ≤ C (n−1)/p , and
y
(c) limy→0 Uy − f p = 0.
5.7.15. For z = (x, y) ∈ H, x ∈ Rn−1 , y > 0, let : H → B be the mapping
given by (2.3.2), that is,
2(z + en )
(z) = −en + ,
|z + en |2
where en = (0, 1), 0 ∈ Rn−1 . For w ∈ B, consider
1 − |w|2
U(w) = .
|en − w|2
Show that
y
U((z)) = .
|x|2 + y2
5.7.16. For x ∈ Rn−1 , y > 0, set
y
V(x, y) = .
|x|2 + y2
Establish each of the following:
1
(a) |∇V|2 = .
(|x| + y2 )2
2
−2y(n − 2)
(b) V = .
(|x|2 + y2 )2
5.7.17. Let h(x, y) = (V(x, y))n−1 and let
∂h
LH = y2 h − (n − 2)y
∂y
be the invariant Laplacian on H.
5.7 Exercises 81
(a) Prove that

∂V
LH = y(n − 1)V n−3
y(n − 2)|∇V| + yVV − (n − 2)V
2
.
∂y
(b) Using the results of the previous exercise show that Lh = 0 on H.
5.7.18. Fix a ∈ Rn−1 . Let U be a positive H-harmonic function on H that
extends continuously to H \ {a} with boundary values 0 on Rn−1 \ {a}.
Suppose also that
U(0, y)
lim = 0.
y→∞ yn−1
Prove that U(x) = cPH (x, a) for some c > 0.

5.7.19. Suppose that U is a positive H-harmonic function on H, continuous on
H with boundary value 0 on Rn−1 . Prove that there exists a constant
c > 0 such that U(x, y) = cyn−1 for all (x, y) ∈ H.
6
Spherical Harmonic Expansions
In this chapter we provide a brief survey of spherical harmonic and zonal

harmonic functions in order to obtain the zonal harmonic expansion of the
Poisson kernel Ph and the spherical harmonic expansion of H-harmonic
functions. For details concerning the general theory of spherical harmonics
the reader is referred to the text [10] by S. Axler, P. Bourdon, and W. Ramey.
Throughout this chapter we will assume n ≥ 3. The results for n = 2
are well known. As in [10, Chapter 5], for m = 0, 1, 2, . . . , we denote by
Hm (Rn ) the space of all (Euclidean) homogeneous1 harmonic polynomials
of degree m on Rn . A spherical harmonic of degree m is the restriction to S
of a harmonic polynomial in Hm (Rn ). The collection of all spherical harmonic
polynomials of degree m will be denoted by Hm (S). Every element of Hm (S)
has a unique extension to Hm (Rn ). If m = k, then Hm (S) and Hk (S) are
orthogonal in L2 (S), that is,

p, q = p(t)q(t) dσ (t) = 0
S
for all p ∈ Hm (S), q ∈ Hk (S). Furthermore,
∞
"
L2 (S) = Hm (S),
m=0
that is, for each f ∈ L2 (S), there exists pm ∈ Hm (S) such that
∞

f = pm , (6.0.1)
m=0
where the series converges in L2 (S).
1 A polynomial p on Rn is homogeneous of degree m if p(λx) = λm p(x) for all λ ∈ R.
82
6.1 Dirichlet Problem for Spherical Harmonics

Our first goal is to solve the Dirichlet problem for pα ∈ Hα (S). Since each pα
has a unique extension to Hα (Rn ), we can assume that pα (x) ∈ Hα (Rn ). Set
f (x) = g(r2 )pα (x), where r2 = |x|2 . Since x, ∇pα = αpα (x), we have
x, ∇f (x) = pα (x)x, ∇g + g(r2 )x, ∇pα

= 2r2 g (r2 )pα (x) + αg(r2 )pα (x),
and since pα is harmonic,
Δf (x) = 2∇pα , ∇g + pα (x)Δg(r2 )

= 4αpα (x)g (r2 ) + pα (x)[2ng (r2 ) + 4r2 g (r2 )].
Therefore
Δh f (x) = 2(1 − r2 )pα (x)

× 2(1 − r2 )r2 g (r2 ) + {(n + 2α)(1 − r2 ) + 2r2 (n − 2)}g (r2 ) + α(n − 2)g(r2 ) .
Thus in order that Δh f (x) = 0 we must have
2(1 − r2 )r2 g (r2 ) + {(n + 2α)(1 − r2 ) + 2r2 (n − 2)}g (r2 ) + α(n − 2)g(r2 ) = 0,
or
(1 − r2 )r2 g (r2 ) + {(α + 12 n) − (α + 2 − 12 n)r2 }g (r2 ) − α(1 − 12 n)g(r2 ) = 0.
If we set t = r2 , a = α, b = 1 − 12 n, and c = α + 12 n, then the above equation

can be rewritten as
t(1 − t)g (t) + {c − (a + b + 1)t}g (t) − abg(t) = 0. (6.1.1)
Equation (6.1.1), however, is the hypergeometric equation, for which a

particular solution is given by the hypergeometric function F(a, b; c; t)
[1, Identity 15.5.1], [20, Chapter II], [50] defined by
∞
(a)k (b)k zk
F(a, b; c; z) = , |z| < 1 (6.1.2)
(c)k k!
k=0
In the above, (a)0 = 1 and for k = 1, 2, . . . ,
(a)k = a(a + 1) · · · (a + k − 1).
If a is not a negative integer, then

(a)k = (a + k) (a),
84 Spherical Harmonic Expansions
where is the gamma function defined on C \ {0, −1, −2, ...}. If c − a − b > 0,
then the series in (6.1.2) converges absolutely for all z, |z| ≤ 1. For the above
values of a, b, and c, we have c − a − b = n − 1. Thus the function g(r2 ) is
given by
g(r2 ) = cα F(α, 1 − 12 n; α + 12 n; r2 ), (6.1.3)
for an arbitrary constant cα . Define Sn,α (r) by
F(α, 1 − 12 n; α + 12 n; r2 )
Sn,α (r) = . (6.1.4)
F(α, 1 − 12 n; α + 12 n; 1)
Then Sn,α (1) = 1 and f (x) = Sα (|x|)pα (x) is a solution of Δh f (x) = 0 that is
continuous on B with f (ζ ) = pα (ζ ) for all ζ ∈ S. This proves the following
theorem.
Theorem 6.1.1 If pα ∈ Hα (S), α = 0, 1, 2, . . . , then for all t ∈ S,
Ph [pα ](rt) = rα Sn,α (r)pα (t),
where Sn,α is defined by (6.1.4).
Example 6.1.2 If n is even, say n = 2m, then b = 1 − m and thus (b)k = 0 for
all k ≥ m. Hence g(r2 ) is a polynomial of degree n − 2. When n = 4, b = −1
and

α
S4,α (r) = 12 (α + 2) 1 − r2 , α = 0, 1, 2, . . . .
α+2
When n = 6, b = −2 and
2α 2 α(α + 1)
F(α, −2; α + 3; r2 ) = 1 − r + r4 .
α+3 (α + 3)(α + 4)
Thus for α = 0, 1, 2, . . . ,

2α 2 α(α + 1)
S6,α (r) = 1
12 (α + 3)(α + 4) 1 − r + 4
r .
α+3 (α + 3)(α + 4)
By [50, Identity 9.3.4], if c − a − b > 0 then
(c)(c − a − b)
lim F(a, b; c; t) = . (6.1.5)
t→1− (c − a)(c − b)
Therefore
(α + 12 n)(n − 1)
F(α, 1 − 12 n; α + 12 n; 1) = ,
( 12 n)(α + n − 1)
and hence Sn,α (r) = cn,α F(α, 1 − 12 n; α + 12 n; r2 ), where
( 12 n)(α + n − 1)
cn,α = . (6.1.6)
(α + 12 n)(n − 1)
Also, using the transformation [1, Identity 15.3.3], [50, Identity 9.5.3],
F(a, b; c; t) = (1 − t)c−a−b F(c − a, c − b; c; t),
we can express Sn,α (r) as
Sn,α (r) = cn,α (1 − r2 )n−1 F( 12 n, α + n − 1; α + 12 n; r2 ). (6.1.7)
Theorem 6.1.1 can be used to compute the invariant Poisson integral of a

polynomial q on S. By [10, Corollary 5.7], if q is a polynomial on Rn of degree
m, then the restriction of q to S is a sum of spherical harmonics of degree at
most m. That is, there exist pk ∈ Hk (S), k = 0, 1, . . . , m, such that q(t) =
m
k=0 pk (t) for all t ∈ S. Hence

m
Ph [q](x) = Ph [pk ](x).
k=0
But by Theorem 6.1.1, Ph [pk ](x) = Sn,k (|x|)pk (x). Thus

m
Ph [q](x) = Sn,k (|x|)pk (x).
k=0
The above computations are particularly easy when n is even. These computa-
tions are illustrated in the following examples.
Examples 6.1.3 (a) For our first example we consider the function q(t) = t12
in R4 . Then for t ∈ S, q(t) = p0 (t) + p2 (t), where p0 (x) = 14 and p2 (x) =
x12 − 14 |x|2 . Thus by Example 6.1.2, in R4
Ph [t12 ](x) = 1
4 + S4,2 (|x|)p2 (x)
= 1
4 + (2 − |x|2 )(x12 − 14 |x|2 ).
The above function is easily shown to be H-harmonic on B.

(b) In spaces of odd dimension these computations are much more compli-
cated. As in Example 5.3.7 consider q(t) = t12 in R3 . Then q(t) = p0 (t) + p2 (t)
where p0 (x) = 13 and p2 (x) = x12 − 13 |x|2 . Hence
Ph [q](x) = 1
3 + S3,2 (|x|)(x12 − 13 |x|2 ).
Unfortunately, however, there is no simple expression for S3,2 (r). The function
S3,2 (r) is given by
S3,2 (r) = c3,2 F(2, − 12 ; 72 ; r2 )

∞
( 32 )(4) (k + 2)(k − 12 )
= r2k
(− 12 ) k=0
(k + 72 )k!
∞
3 (k + 1)(k − 12 ) 2k
=− r .
2 (k + 72 )
k=0
Thus
∞
(k + 1)(k − 12 )
Ph [t12 ](x) = 1
3 − 32 (x12 − 13 |x|2 ) |x|2k .
k=0
(k + 72 )
Using (6.1.7) we also have

∞
(k + 3)(k + 2)(k + 1)
S3,2 (r) = 4(1 − r2 )2 r2k .
(2k + 5)(2k + 3)
k=0
6.2 Zonal Harmonic Expansion of the Poisson Kernel

Our next goal is to obtain an expansion of the Poisson kernel Ph in terms of
the zonal harmonics. Fix a point η ∈ S. By considering the linear functional
γ : Hm (S) → R defined by γ (p) = p(η), it follows from the Riesz represen-
(m)
tation theorem that there exists a unique function Zη ∈ Hm (S) such that

p(η) = p(t)Zη(m) (t) dσ (t)
S
for all p ∈ Hm (S). The spherical harmonic Zη(m) is called the zonal harmonic
of degree m with pole η. Set Zm (η, ζ ) = Zη(m) (ζ ). It is an easy exercise to show
that the zonal harmonic Zm satisfies
Zm (η, ζ ) = Zm (ζ , η), (6.2.1)

Zm (Aη, Aζ ) = Zm (η, ζ ) for all A ∈ O(n), and (6.2.2)
Zm (η, η) = Zη(m) 22 = hm = dim Hm (S). (6.2.3)
(m)
Also, for fixed η ∈ S, the function Zη has a unique extension to a harmonic
function on Rn . This function will again be denoted by Zm (x, η). An explicit
formula for Zm is given in [10, Theorem 5.2.4], which for completeness we
state as a theorem without proof.
Theorem 6.2.1 Let x ∈ Rn and let η ∈ S, then

[m/2]
n(n + 2) · · · (n + 2m − 2k − 4)
Zm (x, η) = (n + 2m − 2) (−1)k x, ηm−2k |x|2k .
2k k!(m − 2k)!
k=0
In terms of zonal harmonics, the decomposition (6.0.1) of L2 (S) can now be

expressed as follows: for f ∈ L2 (S)
∞ ∞

f (η) = Zη , f =
(α)
Zα (η, ζ )f (ζ )dσ (ζ ), (6.2.4)
α=0 α=0 S
where the series converges in L2 (S).

We now derive the following expansion of the invariant Poisson kernel on B
in terms of the zonal harmonics Zα .
Theorem 6.2.2 For x ∈ B, t ∈ S,

∞
∞

Ph (x, t) = |x|α Sn,α (|x|)Zα ( |x|
x
, t) = Sn,α (|x|)Zα (x, t), (6.2.5)
α=0 α=0
where the series converges absolutely, and uniformly on compact subsets of B.
Proof. We first prove that the series (6.2.5) converges absolutely, and uni-
formly on compact subsets of B. Consider Sn,α (r) given by
∞
(α)k (1 − 12 n)k
Sn,α (r) = cn,α r2k .
k=0
(α + 12 n)k k!
Let m = [n/2], and set Sn,α (r) = cn,α [Pm (r) + Qm (r)], where

m−1 ∞

(α)k (1 − 12 n)k (α)k (1 − 12 n)k
Pm (r) = r2k ; Qm (r) = r2k . (6.2.6)
k=0
(α + 12 n)k k! k=m
(α + 12 n)k k!
If n is even, then (1 − 12 n)k = 0 for all k ≥ m, and thus Qm (r) ≡ 0. Now

m−1
(α)k |(1 − 12 n)k |
|Pm (r)| ≤ r2k .
k=0
(α + 12 n)k k!

Since (α)k (α + 12 n)k ≤ 1 for all k,

m−1
|(1 − 12 n)k |
|Pm (r)| ≤ = Cn , (6.2.7)
k!
k=0
where Cn is a constant depending only on n.

Our next step is to obtain an estimate for Qm (r) when n is odd. For k ≥ m
we have
(γ )k = (γ )m (γ + m)k−m .
Thus
∞
(α)k (1 − 12 n)k
Qm (r) = r2k (6.2.8)
k=m
(α + 12 n)k k!
∞

(α)m (1 − 12 n)m (α + m)j (1 + m − 12 n)j
= r2m r2j .
(α + 12 n)m j=0
(α + m + 12 n)j (m + j)!
As (m + j)! ≥ j! and (1 + m − 12 n) > 0,
|(1 − 12 n)m |(α + m)(α + 12 n)

|Qm (r)| ≤ F(α+m, 1+m− 12 n; α+m+ 12 n; r2 ).
(α)(α + 12 n + m)
But F(α + m, 1 + m − 12 n; α + m + 12 n; r2 ) is an increasing function of r. Thus

by identity (6.1.5)
(α + m + 12 n)(n − 1 − m)
F(α + m, 1 + m − 12 n; α + m + 12 n; r2 ) ≤ .
( 12 n)(α + n − 1)
Therefore
|(1 − 12 n)m | (n − 1 − m) (α + m)(α + 12 n)
|Qm (r)| ≤ . (6.2.9)
( 12 n) (α)(α + n − 1)
But by (6.1.6)
( 12 n)(α + n − 1)
cn,α = .
(n − 1)(α + 12 n)
Thus
(α + n − 1) (α + m)
|Sn,α (r)| ≤ Cn + Dn ,
(α + 12 n) (α)
where again Cn and Dn are constants depending only on n. Using the fact that
([1, Identity 6.1.46])
(α + a)
lim α b−a = 1,
α→∞ (α + b)
we have
(α + a)
≈ α a−b . (6.2.10)
(α + b)
Hence by the above
|Qm (r)| ≤ Dn α 1/2 and |Sn,α (r)| ≤ Dn α [n/2] , (6.2.11)
where Cn and Dn are constants depending only on n. Also, for all ζ , t ∈ S,
|Zα (ζ , t)| ≤ Zα 22 = hα ,
where hα = dim(Hα (S)). By [10, Chapter 5]

n+α−1 n+α−3
hα = − .
n−1 n−1
But then hα ≤ Cα n−2 . Hence

∞
∞

|Sn,α (|x|)| |Zα (x, ζ )| ≤ C |x|α α p ,
α=0 α=0
where p = n + [n/2] − 2. The series on the right however converges for all
x, |x| < 1, and uniformly for 0 ≤ |x| ≤ ρ, whenever 0 < ρ < 1 is fixed. This
proves our assertion.
It only remains to be shown that the series converges to Ph (x, t). By (6.2.4),
if f ∈ L2 (S), then
∞

f (η) = f , Zη(α)
α=0
in L2 (S). In particular, by Theorem 6.1.1, for fixed x ∈ B,

∞
∞

(α)
Ph (x, t) = Ph [Zt ](x) = Sn,α (|x|)Zα (x, t),
α α=0
An immediate consequence of the previous theorem is the following

corollary.
Corollary 6.2.3 If f ∈ L2 (S),

∞

Ph [ f ](x) = Sn,α (|x|) Zα (x, t)f (t) dσ (t),
α=0 S
where the series converges absolutely, and uniformly on compact subsets of B.

∞ (α)
Proof. For f ∈ L2 (S), f (η) = α=0 f , Zη in L2 (S). Thus
∞

Ph (x, η)f (η) dσ (η) = Ph (x, η) f , Zη(α) dσ (η)
S α=0 S
∞
= Ph (x, η)f (t)Zα (η, t) dσ (t)dσ (η),
α=0 S S
which by Fubini’s theorem and Theorem 6.1.1

∞

= Sn,α (|x|) f (t)Zα (x, t) dσ (t).
α=0 S
6.3 Spherical Harmonic Expansion

of H-Harmonic Functions
In this section we follow the methods of P. Ahern, J. Bruna, and C. Cascante
[2] to obtain the following spherical harmonic expansion of H-harmonic
functions. The analogous result in [2] was proved for M-harmonic functions
on the unit ball of Cn . The result has previously been proved by P. Jaming
using the methods of [2] in his doctoral dissertation [41], and also earlier by
K. Minemura in [57].
Theorem 6.3.1 If u is an H-harmonic function on B, then

u(rζ ) = F(α, 1 − 12 n; α + 12 n; r2 )rα ϕα (ζ ), ζ ∈ S,
α
where ϕα is a spherical harmonic of degree α. Moreover, the series converges

absolutely and uniformly on every compact subset of B.
Proof. For each r, 0 < r < 1, by (6.2.4) the L2 -decomposition in harmonic

polynomials of u(rζ ) is given by
∞
u(rζ ) = Zα (ζ , η)u(rη)dσ (η).
α=0 S
Let ζ ∈ S and α ∈ N be fixed, and for r ∈ (−1, 1) let

fζ (r) = Zα (ζ , η)u(rη)dσ (η).
S
Set
Lf = (1 − r2 )N 2 f + (n − 2)(1 + r2 )Nf , where
df
Nf = x, ∇f = r .
dr
Also, let
∂ ∂
Δσ = Li,2 j where Li, j = xi − xj .
∂xj ∂xi
i<j
Since u is H-harmonic, by Exercise 5.7.6

Lfζ (r) = −(1 − r ) Zα (ζ , η)σ u(rη)dσ (η),
2
S
which by Exercises 6.4.3 and 6.4.4
= (1 − r2 )cα,n fζ (r),
where cα,n = α[α + (n − 2)].
Since fζ (−r) = (−1)α fζ (r), the function fζ (r)/rα is even on (−1, 1). Hence,
there exists g(x) defined on (0, 1) such that fζ (r) = rα g(r2 ). Substituting into
Lfζ (r) − cα,n (1 − r2 )fζ (r) = 0
and simplifying gives
2r2 (1 − r2 )g (r2 ) + [(n + 2α) + (n − 2α − 4)r2 ]g (r2 ) + α(n − 2)g(r2 ) = 0,
which upon replacing r2 by x and dividing by 2 yields the hypergeometric
equation
x(1 − x)g (x) + [(α + n2 ) − (α − n
2 + 2)x]g (x) − α(1 − n2 )g(x) = 0.
With a = α, b = 1 − 12 n, and c = α + 12 n, this equation can be rewritten in
standard form as
x(1 − x)g (x) + {c − (a + b + 1)x}g (x) − abg(x) = 0.
By Frobenius’ theorem [30, Section 6.5], every solution of the above is a linear
combination of two functions g1 (x) and g2 (x), whose behavior near x = 0 is
respectively like 1 and x1−c . Since g(x) is bounded near 0, g(x) is a multiple of
the hypergeometric function (6.1.2). Therefore,
g(x) = cα (ζ )F(α, 1 − 12 n; α + 12 n; x)
and
fζ (r) = cα (ζ )rα F(α, 1 − 12 n; α + 12 n; r2 ).
For fixed r, as a function of ζ , fζ (r) ∈ Hα (S). Hence there exists ϕα ∈ Hα (S)

so that cα (ζ ) = ϕα (ζ ). Therefore

u(rζ ) = F(α, 1 − 12 n; α + 12 n; r2 )rα ϕα (ζ ). (6.3.1)
α
It remains to be shown that the series (6.3.1) converges absolutely and

uniformly on compact subsets of B. For notational convenience, set Fα (r) =
F(α, 1 − 12 n; α + 12 n; r2 ). Also, as in Theorem 6.2.2 set m = [n/2] and

m−1 ∞

(α)k (1 − 12 n)k (α)k (1 − 12 n)k
Pm (α, r) = 2k
r ; Qm (α, r) = r2k .
k=0
(α + 12 n)k k! m (α + 12 n)k k!
Let K be a compact subset of B. Choose ro , 0 < ro < 1, such that K ⊂ Bro .

Then for |ρ| < ro ,
Fα (ρ) ρ α
Fα (ρ)ρ ϕα (ζ ) =
α
Fα (ro )roα ϕα (ζ ).
α α
Fα (r o ) r o
But

|Fα (ro )roα ϕα (ζ )| = Zα (ζ , η)u(ro η)dσ (η)

S
≤ Zα (ζ , ·)2 uro 2
1 n
≤ hα2 uro 2 ≤ Cα 2 −1 uro 2 .
Therefore
|Fα (ρ)| |ρ| α n
|Fα (ρ)ρ α ϕα (ζ )| ≤ Curo 2 α 2 −1 .
α α
|Fα (r o )| r o
Suppose n is even. Then Qm (α, ρ) = 0 for all ρ, and as in Theorem 6.2.2,

|Pm (α, ρ)| ≤ Cn . Also, since
(m − 1)!
(1 − m)k = (−1)k
(m − k − 1)!
(α)k
and lim = 1, we have
α→∞ (α + 12 n)k

m−1
(m − 1)!
lim P(α, ro ) = (−1)k r2k = (1 − ro2 )m−1 .
α→∞ (m − k − 1)!k! o
k=0
Hence there exists an integer αo such that |Pm (α, ro )| ≥ 12 (1 − ro2 )m−1 for all
α ≥ αo . Therefore
∞ ∞
Curo 2 |ρ| α n −1
|Fα (ρ)ρ α ϕα (ζ )| ≤ α2 , (6.3.2)
α=α
(1 − ro2 )m−1 α=α ro
o o
which converges for all ρ, |ρ| < ro . # $

Suppose n is odd. Using the notation of [39, Theorem 7.25], set γ0 = 1,
and for k = 1, 2, . . . ,

γ γ (γ − 1)(γ − 2) · · · (γ − k + 1)
= .
k k!
Since (1 − 12 n)k = (−1)k ( 12 n − 1)( 12 n − 2) · · · ( 12 n − k), the series for Fα (r)

can be expressed as
∞
1
Fα (r) =
(α)k 2 n − 1 (−1)k r 2k .
k=0
(α + 12 n)k k
By Theorem 7.25 of [39], the series

∞ 1
2n − 1
(6.3.3)
k
k=0
converges, and
∞ 1

2 n − 1 (−1)k r 2k = (1 − r 2 ) 12 n−1 (6.3.4)
k
k=0
for all r ∈ [0, 1], with the series converging uniformly and absolutely in [0, 1].
For N > 12 n, let PN (α, r) and QN (α, r) be as defined in (6.2.6). Then
|Fα (ro )| ≥ |PN (α, ro )| − |QN (α, ro )|.
Since (α)k < (α + 12 n)k ,

∞ 1
2n − 1
|QN (α, r)| ≤ ,
k
k=N
and since the series converges, given > 0, there exists an integer N1 such that
|QN (α, r)| < for all N ≥ N1 and all non-negative integers α and r ∈ [0, 1].
Also, since
1
lim |PN (ro )| = (1 − ro2 ) 2 n−1 ,
N→∞
1
there exists an integer N2 such that |PN (ro )| ≥ (1−ro2 ) 2 n−1 − for all N ≥ N2 .
Fix an N ≥ max{N1 , N2 }. For this N, since
lim |PN (α, ro )| = |PN (ro )|,

α→∞
there exists an integer αo such that |PN (α, ro )| > |PN (ro )| − for all α ≥ αo .
1
Take = 16 (1 − ro2 ) 2 n−1 . Then for this epsilon, there exists an αo such that
1
|Fα (ro )| > 12 (1 − ro2 ) 2 n−1
for all α ≥ αo . Finally, as a consequence of (6.2.7) and (6.2.11), we have that

1
|Fα (ρ)| ≤ Cn α 2 for all ρ ∈ (−1, 1). Hence, as in (6.3.2), we have that the
series converges absolutely and uniformly for all ρ, |ρ| < ro .
6.4 Exercises
6.4.1. (a) Find the zonal harmonic expansion of the Euclidean Poisson kernel.
(b) Let Q be a polynomial on Rn of degree m. Prove that the Euclidean
Poisson integral Pe [Q] of Q is a polynomial of degree at most m,
and that
m
Pe [Q](x) = Zk (x, t)Q(t)dσ (t).
k=0 S
6.4.2. Let Q be a polynomial on S, and let Ph [Q] denote the Poisson integral
of Q. Prove that |∇Ph [Q](x)| is continuous on B.
6.4.3. As in Exercise 3.5.6 let
∂ ∂
Δσ = 2
Li,j where Li,j = xi − xj .
∂xj ∂xi
i<j
(a) Prove that for f ∈ C2 (B)
Δσ f (x) = |x|2 Δf (x) − (n − 2)x, ∇f (x) − x, ∇x, ∇f (x).
(b) With Zα , α = 0, 1, 2, . . . as given in Theorem 6.2.1, prove that
Δσ Zα (x, η) = −α[α + (n − 2)]Zα (x, η).

6.4.4. (a) If uα (x) = Zα (x, ζ )u(ζ )dσ (ζ ), prove that Δσ uα (x) = −cn,α uα (x)
S
where cn,α = α[α + (n − 2)].
(b) If u, v ∈ L2 (S), prove that u, Δσ v = Δσ u, v.
6.4 Exercises 95
6.4.5. Prove that for c > b > 0,

1
(c)
F(a, b; c; r) = tb−1 (1 − t)c−b−1 (1 − rt)−a dt.
(b)(c − b) 0
6.4.6. Prove the following result of H. Samii [73]. If u is H-harmonic on B,
then there exists a Euclidean harmonic function v with v(0) = 0 such
that
1
dt
u(x) = u(0) + v(tx)[(1 − t)(1 − tr2 )]n/2−1 .
0 t
7
Hardy-Type Spaces of H-Subharmonic

Functions
In this chapter we consider Hardy-type Hp spaces of both H-harmonic and

H-subharmonic functions on B. The main result of Section 7.1 is a Poisson
integral formula for H-harmonic functions in the space Hp , 1 ≤ p ≤ ∞.
At the same time we also prove the existence of an H-harmonic majorant for
an H-subharmonic function satisfying an H p growth condition. In Section 7.3
we take up the general question of the existence of H-harmonic majorants for
H-subharmonic functions as well as the existence of a least H-harmonic majo-
rant. In Section 7.4 we consider a generalization of a theorem of L. Garding
and L. Hörmander [26] concerning Hardy–Orlicz spaces of H-subharmonic
functions.
We begin the chapter with the following definitions.
Definition 7.0.1 For 0 < p ≤ ∞, we denote by S p the Hardy-type space of

non-negative continuous H-subharmonic functions f on B for which

f pp = sup f p (rt)dσ (t) < ∞. (7.0.1)
0<r<1 S
When p = ∞, we set f ∞ = supx∈B f (x).
In the case 0 < p < 1, as a general rule we will only be interested in non-
negative H-subharmonic functions for which f p is also H-subharmonic. In the
case that f p is H-subharmonic, then by Theorem 5.4.2

f p = lim f p (rt)dσ (t).
p
r→1 S
Likewise, for 0 < p < ∞, we let
Hp = { f : f is H-harmonic with | f | ∈ S p }.
96
7.1 Poisson Integral Formula for Functions in Hp , 1 ≤ p ≤ ∞ 97
The space Hp is called the Hardy space of H-harmonic functions on B. Also,

for λ ≥ 0 and 0 < p < ∞ we let
p
Hλ = { f ∈ Hλ : | f | ∈ S p }.
By Exercise 5.7.4, if f ∈ Hλ , λ ≥ 0, then | f | is H-subharmonic on B.

p
As a general rule, in discussions of Hp or Hλ we will primarily be interested
in the case p ≥ 1. However, the definitions still make sense for 0 < p < 1.
In fact, if f is a positive H-harmonic function on B, then for 0 < p < 1, f p is
H-superharmonic and thus f ∈ Hp for all p, 0 < p ≤ 1.
7.1 A Poisson Integral Formula for Functions in

Hp , 1 ≤ p ≤ ∞
We now prove the Poisson integral formula for H-harmonic functions in
Hp , 1 ≤ p ≤ ∞, while at the same time proving the existence of an
H-harmonic majorant for an H-subharmonic function f in S p , that is, f (x) ≤
H(x) for some H-harmonic function H.
Theorem 7.1.1 Let f be a continuous non-negative H-subharmonic function

on B with f ∈ S p for some p, 1 ≤ p ≤ ∞.
(a) If 1 < p ≤ ∞, then there exists fˆ ∈ L p (S) such that
f (x) ≤ Ph [ fˆ](x) for all x ∈ B (7.1.1)
with f p = fˆp .
(b) If p = 1, then there exists a Borel measure νf on B such that
f (x) ≤ Ph [νf ](x) for all x ∈ B. (7.1.2)
(c) If f is H-harmonic on B with | f | ∈ S p , that is, f ∈ Hp , for some p,

1 ≤ p ≤ ∞, then equality holds in (7.1.1) and (7.1.2). Furthermore, in this
case the measure νf is a signed Borel measure on S.
Since non-negative H-harmonic functions satisfy (7.0.1) with p = 1, we

have the following corollary.
Corollary 7.1.2 If h is a non-negative H-harmonic function on B, then there

exists a Borel measure μ on S such that h(x) = Ph [μ](x) for all x ∈ B.
Remarks 7.1.3 (1) For H-harmonic functions f , the function fˆ and measure
νf are called the boundary function and boundary measure of f . Further-
more, if
dνf = fˆ dσ + νs
98 Hardy-Type Spaces of H-Subharmonic Functions
is the Lebesgue decomposition of f (see [71, p. 278]) where fˆ ∈ L1 (S) and νs

is singular with respect to σ , then as we will prove in Theorem 8.3.3, for every
α > 1,
lim f (x) = fˆ(ζ ) σ - a.e. on S, (7.1.3)
x→ζ
x ∈ α (ζ )
where α (ζ ) = {x ∈ B : |x − ζ | < α(1 − |x|)}.
(2) For invariant harmonic functions, the case p = ∞ was originally proved
by H. Furstenberg ([24]) for symmetric spaces of noncompact type. The L p
statements were proved by A. Koranyi in [46] for M-harmonic functions in
the unit ball of Cm and by the author [78] for symmetric spaces of noncompact
type. In the setting of rank one symmetric spaces, Hp spaces have also been
considered by P. Cifuentis in [14], [15]. The proof given here is a variation of
an elementary proof using an equicontinuity argument which for M-harmonic
functions is due to D. Ullrich [93]. (See also [72, Theorem 4.3.3] and [84,
Theorem 5.8].) The result for Euclidean subharmonic functions on the unit
ball in Rn is due to L. Garding and L. Hörmander [26]. An alternate proof
of Theorem 7.1.1 for H-harmonic functions was given by P. Jaming in [42,
Proposition 2]. In [31] the Hp spaces were defined for all p, 0 < p < ∞, in
terms of the radial maximal function (see Definition 8.2.1).
Proof. Let h : O(n) → [0, ∞) be a continuous function satisfying

h(A)dA = 1,
O(n)
where dA is the normalized Haar measure on O(n). Let f ∈ S p (Hp ) for some
p, 1 ≤ p ≤ ∞. Define the function G on B by

G(x) = f (Ax)h(A)dA. (7.1.4)
O(n)
Consider

G(ϕa (rt))dσ (t) = f (Aϕa (rt))h(A)dAdσ (t),
S S O(n)
which by Fubini’s theorem

= f (Aϕa (rt))dσ (t) h(A)dA.
O(n) S
By Exercise 2.4.5, Aϕa = ϕAa ◦ B for some B ∈ O(n). Therefore,

f (Aϕa (rt))dσ (t) = f (ϕAa (rBt))dσ (t),
S S
7.1 Poisson Integral Formula for Functions in Hp , 1 ≤ p ≤ ∞ 99
which since σ is invariant under O(n)

= f (ϕAa (rt))dσ (t) ≥ f (Aa),
S
with equality if f is H-harmonic. Therefore

G(ϕa (rt))dσ (t) ≥ f (Aa)h(A)dA = G(a),
S O(n)
again with equality if f is H-harmonic. Thus G is H-subharmonic (H-

harmonic).
Since f p < ∞, by Hölder’s inequality applied to (7.1.4) for x = rζ , ζ ∈ S,
1/p
|G(x)| ≤ | f (Arζ )|p dA hq ≤ f p hq
O(n)
where 1
p + 1
q = 1. In the above we have used the fact that

| f (Arζ )|p dA = | f (Arζ )|p dAdσ (ζ ),
O(n) S O(n)
which by (3.3.3)

= | f (rζ )|p dσ (ζ ).
S
Also, by the continuous version of Minkowski’s inequality ([18, VI.11.13]),

for 1 ≤ p < ∞,
1/p 1/p
|G(rζ )|p dσ (ζ ) ≤ h(A) | f (Arζ )|p dσ (ζ ) dA
S O(n) S
≤ f p . (7.1.5)
Clearly the result is also true for p = ∞.

We now show that the family {Gr : 0 < r < 1} , where Gr (t) = G(rt), is an
equicontinuous family on S; that is, to each > 0 there exists a δ > 0 such
that
|Gr (x) − Gr (y)| <
for all r, 0 < r < 1, and all x, y ∈ S with |x − y| < δ. Let > 0 be fixed. Since
h is continuous on O(n), there exists a neighborhood V of the identity I such
that
|h(A) − h(AB−1 )| <
for all A ∈ O(n), B ∈ V. Furthermore, since

Gr (Bx) = f (rABx)h(A)dA = f (rAx)h(AB−1 )dA,
O(n) O(n)
we have

|Gr (x) − Gr (Bx)| ≤ | f (rAx)||h(A) − h(AB−1 )|dA
O(n)
≤ f 1 ≤ f ||p
for all B ∈ V. The mapping B → Bx is a one-to-one mapping of O(n) onto S.
Thus there exists a δ > 0 such that |x − y| < δ implies y = Bx for some B ∈ V.
Therefore
|Gr (x) − Gr (y)| < f p
for all x, y ∈ S with |x − y| < δ and all r, 0 < r < 1. Hence {Gr } is
equicontinuous on S. Hence by the Ascoli–Arzelá theorem [71], there exists
a sequence rk → 1 such that Grk converges uniformly to a continuous function
g on S.
Let
k = sup |G(rk t) − Ph [g](rk t)|.
t∈S
Since Grk → g uniformly on S, and since g is continuous, from the proof of

Theorem 5.3.3 we also have that Ph [g](rk t) → g(t) uniformly. Thus k → 0
as k → ∞. Hence by the maximum principle
G(x) ≤ Ph [g](x) + k for all x, |x| ≤ rk .
If f is H-harmonic, then
|G(x) − Ph [g](x)| ≤ k for all x, |x| ≤ rk .
Letting k → ∞ we have
G(x) ≤ Ph [g](x) (7.1.6)
for all x ∈ B with equality if f is H-harmonic. Since

|G(rk ζ )|p dσ (ζ ) ≤ f pp
S
and Grk → g uniformly on S, we have
gp ≤ f p , 1 ≤ p ≤ ∞.
To conclude the proof, let {hj } be a sequence of continuous functions
on O(n) which forms an approximate identity, that is, hj ≥ 0 for all j,
7.2 Completeness of Hp , 0 < p ≤ ∞ 101

O(n) hj dA = 1, and limj→∞ O(n)\V hj dA = 0 for every neighborhood V
of the identity. For each j let Gj be as defined by (7.1.4). Since f is continuous,
as in the proof of Theorem 5.3.3,
lim Gj (x) = f (x) for all x ∈ B. (7.1.7)
j→∞
By the above, each Gj = Ph [gj ] for some gj ∈ L p (S) with gj p ≤ f p . If

p > 1, then by the Alaoglu theorem (see [71, p. 237]), some subsequence of
{gj }, which we denote by {gj }, converges
in the weak* topology of L p (S) to
some fˆ ∈ L (S), that is, gj hdσ → fˆhdσ for every h ∈ Lq (S), 1p + 1q = 1.
p
In particular limj→∞ Ph [gj ] = Ph [ fˆ]. Thus by (7.1.6) and (7.1.7),

f (x) ≤ Ph [ fˆ](x)
with equality if f is H-harmonic. Furthermore, since gj p ≤ f p we have
fˆp ≤ f p .
Also, since f (x) ≤ Ph [ fˆ](x) (or f (x) = Ph [ fˆ](x) for H-harmonic f ) we also
have f p ≤ fˆp .
For the case p = 1, some subsequence of {gj } converges weak* in the
dual of C(S). Thus f (x) ≤ Ph [νf ](x) for some Borel measure νf on S. If f is
H-harmonic, then f (x) = Ph [νf ](x) for some signed Borel measure νf on S.
7.2 Completeness of Hp , 0 < p ≤ ∞

For 0 < p ≤ ∞, f , g ∈ Hp , define

f − gp , p ≥ 1,
dp ( f , g) = p (7.2.1)
f − gp , 0 < p < 1.
Then dp is a metric on Hp . This follows from the fact that for p ≥ 1, · p is a
p
norm on Hp , while for 0 < p < 1, f p is a p-norm. If X is a vector space (of
functions), a function · : X → [0, ∞) is a norm on X if (i) f = 0 if and
only if f = 0, (ii) f +g ≤ f +g for all f , g ∈ X, and (iii) af = |a| f
for all a ∈ R and f ∈ X. For 0 < p < 1, · is a p-norm if · p satisfies (i)
and (ii) above, and in addition (iii*) af p = |a|p f p for all a ∈ R, f ∈ X.
In this section we prove that the metric space (Hp , dp ) is complete for all
p, 0 < p < ∞. The case p = ∞ is left to the exercises. We begin with the
following lemma.
Lemma 7.2.1 If f ∈ S p , 0 < p < ∞, then there exists a constant C,
independent of f, such that
C
f p (x) ≤ f pp
(1 − |x|2 )n−1
for all x ∈ B.
Proof. For 1 ≤ p < ∞, the result is an immediate consequence of the Poisson
integral formula for functions in S p (see Exercise 7.5.3).
Fix δ, 0 < δ < 12 . Since f is H-subharmonic, by Theorem 4.7.3 there exists
a constant Cδ such that

f p (a) ≤ Cδ,p f p (x)dτ (x).
E(a,δ)
By Exercise 2.4.1(b), if x ∈ E(a, δ), then

cδ (1 − |a|2 ) ≤ (1 − |x|2 ) ≤ c−1
δ (1 − |a| ),
2
where cδ = (1 − δ)/(1 + δ). Let

Aδ (a) = {rζ : ζ ∈ S, cδ (1 − |a|2 ) ≤ (1 − r2 ) ≤ c−1
δ (1 − |a| )}.
2
Thus

f p (a) ≤ Cδ,p f p (x)dτ (x)
Aδ (a)
p √1−cδ (1−|a|2 )
C f p
≤ % rn−1 dr
(1 − |a|2 )n 1−c−1 2
δ (1−|a| )
p
C f p
≤ .
(1 − |a|2 )n−1

Theorem 7.2.2 The metric space (Hp , dp ) is complete for all p, 0 < p < ∞.
Proof. The case p = ∞ is left as an exercise (Exercise 7.5.4). Suppose { fn }
is a Cauchy sequence in (Hp , dp ) and K is a compact subset of B. By Lemma
7.2.1, there exists a constant CK such that
| fn (x) − fm (x)| ≤ CK fn − fm pp
for all n, m ∈ N and all x ∈ K. Thus { fn } is a uniform Cauchy sequence
on each compact subset of B and as a consequence converges uniformly on
compact sets to a function f . Let
f (x) = lim fm (x).
m→∞
Since the convergence is uniform on compact sets, f is H-harmonic on B.

Furthermore, by Fatou’s lemma, f ∈ Hp . It remains to be shown that fn → f
in · p . Let > 0 be given. Since { fn } is Cauchy, there exists an integer no

p
such that fn − fm p < for all n, m ≥ no . But for each r ∈ (0, 1),

| f (rt) − fm (rt)|p dσ (t) ≤ lim | fn (rt) − fm (rt)|p dσ (t)
S n→∞ S
≤ lim fn − fm pp ≤ .
n→∞
p
Taking the supremum over r ∈ (0, 1) gives f − fm p ≤ for all m ≥ no ; that
is, { fn } converges to f in the metric of Hp .
7.3 H-Harmonic Majorants for H-Subharmonic Functions

In Theorem 7.1.1 we proved that if f is a continuous non-negative H-
subharmonic function satisfying

lim f (rt)dσ (t) < ∞,
r→1 S
then f has an H-harmonic majorant of the form Ph [ν] where ν is a Borel

measure on S. In this section we consider the existence of H-harmonic
majorants for general H-subharmonic functions on B.
Definition 7.3.1 An H-subharmonic function f on B has an H-harmonic

majorant if there exists an H-harmonic function h such that f (x) ≤ h(x) for
all x ∈ B. Furthermore, if there exists an H-harmonic function F satisfying
(a) f (x) ≤ F(x) for all x ∈ B, and
(b) F(x) ≤ G(x) for any H-harmonic majorant G of f ,
then F is called the least H-harmonic majorant of f and will be denoted
by Ff .
The following theorem provides necessary and sufficient conditions for the
existence of a least H-harmonic majorant.
Theorem 7.3.2 Let f be H-subharmonic on B. Then the following are

equivalent:
(a) f has a least H-harmonic majorant on B.
(b) f has an H-harmonic majorant on B.
(c) lim f (rt)dσ (t) < ∞.
r→1 S
For the proof of the theorem we will need the following generalization of
Harnack’s inequality for H-harmonic functions.
Lemma 7.3.3 (Harnack’s Inequality) Let be an open-connected subset of

B and a ∈ . If K is a compact subset of , then there exists a constant CK
such that
h(x) ≤ CK h(a)
for all x ∈ K and all non-negative H-harmonic functions on .
Proof. Let y ∈ be arbitrary and δ > 0 such that E(y, 4δ) ⊂ . If y1 , y2 ∈
E(y, δ), then
E(y1 , δ) ⊂ E(y, 2δ) ⊂ E(y2 , 3δ).
Therefore, by (4.3.2),

1
h(y1 ) = h(x)dτ (x),
τ (Bδ ) E(y1 ,δ)
which since h is non-negative

1
≤ h(x)dτ (x)
τ (Bδ ) E(y2 ,3δ)
τ (B3δ )
= h(y2 ) = Cδ h(y2 ).
τ (Bδ )
Choose δ > 0 such that disth (K, ∂) > 4δ where for sets A and B
disth (A, B) = inf{dh (x, y) : x ∈ A, y ∈ B}.
Since K is compact and ∂ is closed, one can easily show that
disth (K, ∂) > 0. Since K is compact, K can be covered by a finite number of
hyperbolic balls E(yi , δ), i = 1, . . . , N. Hence for any y ∈ K, h(y) ≤ Cδ h(yi )
for some i ∈ {1, . . . , N}.
Since is connected, for each i choose an arc Ai ⊂ joining a to yi and
δi > 0 such that disth (Ai , ∂) > 4δi . Each arc Ai can be covered by a finite
number Ni of balls E(xj , δi ), j = 1, . . . , Ni . Hence
h(yi ) ≤ CδNi i h(a).
If we let CK = max{Cδ CδNi i : i = 1, . . . , N}, then for any y ∈ K,

h(y) ≤ CK h(a).

Lemma 7.3.4 Let f be H-subharmonic on B. Then for each r, 0 < r < 1,

there exists an H-harmonic function F r on Br such that
(a) f(x) ≤ F r (x) for all x ∈ Br , and
(b) f (rt)dσ (t) = F r (0).
S
Furthermore, if F is H-harmonic on an open subset of B with Br ⊂ and

F(x) ≥ f (x) for all x ∈ , then
(c) F r (x) ≤ F(x) for all x ∈ Br .
(d) In particular, if 0 < r < ρ < 1, then F r (x) ≤ F ρ (x) for all x ∈ Br .
Proof. Fix r, 0 < r < 1. Since f is upper semicontinuous on Sr = rS,
there exists a decreasing sequence { fn } of continuous functions on Sr such
that limn→∞ fn (rt) = f (rt) for all t ∈ S. For each n, by Theorem 5.4.1 there
exists an H-harmonic function Hn on Br with Hn (rt) = fn (rt). By Harnack’s
inequality, for each compact subset K of Br there exists a constant CK such
that for m ≥ n we have

0 ≤ Hm (x) − Hn (x) ≤ CK [Hm (0) − Hn (0)] = CK [ fm (rt) − fn (rt)] dσ (t).
S
Therefore,
H r (x) = lim Hn (x)
n→∞
exists for all x ∈ Br and is H-harmonic on Br with

H (0) = lim
r
fn (rt)dσ (t) = f (rt)dσ (t). (7.3.1)
n→∞ S S
Suppose F is H-harmonic on a domain ⊃ Br with F(x) ≥ f (x) for all
x ∈ . With { fn } as above, set
gn (rt) = min{ fn (rt), F(rt)},
and let Gn be the corresponding H-harmonic function on Br . Since gn ≤ fn on
Sr we have Gn (x) ≤ Fn (x) for all x ∈ Br . But {gn } is a non-increasing sequence
of functions on Sr with lim gn (rt) = f (rt) for all t ∈ S. Thus by (7.3.1),

lim Gn (0) = f (rt) dσ (t) = lim Fn (0).
n→∞ S n→∞
Therefore as a consequence of Harnack’s inequality,

F r (x) = lim Fn (x) = lim Gn (x).
n→∞ n→∞
However, by the maximum principle, Gn (x) ≤ F(x) for all n and all x ∈ Br .
Thus F r (x) ≤ F(x) for all x ∈ Br . The result (d) is an immediate consequence
of (c).
Proof of Theorem 7.3.2. Clearly (a) ⇒ (b) ⇒ (c). Suppose that (c) holds.
Choose an increasing sequence {rn } with rn → 1. For each n, let F (n) be the
H-harmonic function on Brn satisfying the conclusion of Lemma 7.3.4. By
part (a) of the lemma we have
F (n) (x) ≤ F (n+1) (x) for all x, |x| < rn .
Since
lim Fn (0) = lim f (rn t)dσ (t) < ∞,
n→∞ n→∞ S
by Harnack’s theorem (Exercise 7.5.1),

Ff (x) = lim F (n) (x) (7.3.2)
n→∞
is H-harmonic on B and satisfies Ff (x) ≥ f (x) for all x ∈ B. Suppose G is
an H-harmonic majorant of f . Then by part (c) of the lemma G(x) ≥ F (n) (x)
for all x ∈ Brn . Therefore G(x) ≥ Ff (x) for all x ∈ B, and thus Ff is the least
H-harmonic majorant of f .
Corollary 7.3.5 Let f ≤ 0 be H-subharmonic on B. Then the least
H-harmonic majorant of f is the zero function if and only if

lim f (rt) dσ (t) = 0.
r→1 S
If Ff is the least H-harmonic majorant of f given by (7.3.2), then since

Ff (0) = lim f (rt)dσ (t)
r→1 S
we have
lim | Ff (rt) − f (rt)|dσ (t) = 0. (7.3.3)
r→1 S
For a function f on B, let

f + (x) = max{ f (x), 0}.
The function f + is called the positive part of f . If f is H-subharmonic on B,
then by Theorem 4.4.1 the function f + is also H-subharmonic on B.
Theorem 7.3.6 Let f be H-subharmonic on B. Then the following are
equivalent:
(a) lim f + (rt)dσ (t) < ∞.
r→1 S
(b) f (x) ≤ Ph [ν](x) for some signed Borel measure ν on S.
(c) f has a least H-harmonic majorant of the form Ph [νf ] for some signed
measure νf on S.
The measure νf of Theorem 7.3.6 is called the boundary measure of f .
Proof. Assume (a) holds. Since f + is H-subharmonic, by Theorem 7.3.2 f +
has a least H-harmonic majorant H on B. Since H is non-negative, by Theorem
7.1.1 H(x) = Ph [ν](x) for some non-negative measure ν on S. Thus since
f (x) ≤ f + (x) ≤ Ph [ν](x),
we obtain (b).
Assume that (b) holds, that is, f (x) ≤ Ph [ν](x) for some signed measure ν
on S. Then for all r, 0 < r < 1,

f (rt)dσ (t) ≤ Ph [ν](rt)dσ (t) = Ph [ν](0),
S S

which is finite. Hence limr→1− S f (rt)dσ (t) ≤ C < ∞. Thus by Theorem
7.3.2 f has a least H-harmonic majorant Hf on B. Since Hf is the least
H-harmonic majorant of f ,
Hf (x) ≤ Ph [ν](x) ≤ Ph [ν + ](x),
where ν + is the positive variation of ν. Therefore Hf+ (x) ≤ Ph [ν + ](x). Since
|Hf | = 2Hf+ − Hf ,

|Hf (rt)|dσ (t) = 2 Hf+ (rt)dσ (t) − Hf (rt)dσ (t)
S S S
+
≤ 2ν (S) − Hf (0).
Therefore Hf satisfies the hypothesis of Theorem 7.1.1 and as a consequence,
Hf = Ph [νf ] for some signed Borel measure νf on S. The implication (c) ⇒ (a)
is obvious.
Remark 7.3.7 If f (0) > −∞ and f satisfies Theorem 7.3.6(a), then since
| f (x)| = 2f + (x) − f (x) we have

sup | f (rt)|dσ (t) ≤ 2 lim f + (rt)dσ (t) − f (0) < ∞.
0<r<1 S r→1 S
Suppose f satisfies (a) of Theorem 7.3.6 and Ff = Ph [νf ] is the least

H-harmonic majorant of f . Let g be continuous on S, and set G(x) = Ph [g](x).
Since Ph (rζ , t) = Ph (rt, ζ ) we have

Ff (rt)g(t)dσ (t) = Ph [νf ](rt)g(t)dσ (t)
S
S
= Ph (rt, ζ )g(t)dνf (ζ ) dσ (t)
S S

= Ph (rζ , t)g(t)dσ (t) dνf (ζ )
S S
= G(rζ )dνf (ζ ),
S

where the right side tends to S g(ζ )dνf (ζ ) as r → 1− . Hence dνf (t) is the
weak limit of Ff (rt)dσ (t). By (7.3.3) it now also follows that

lim Ff (rt)g(t)dσ (t) = lim f (rt)g(t)dσ (t) = g(t)dνf (t) (7.3.4)
r→1 S r→1− S S
for every g ∈ C(S).
For the proof of Theorem 7.4.2 we require the following theorem.

Theorem 7.3.8 Let F = Ph [ν] where ν is a finite signed measure on S. If
dν = fˆdσ + dνs
is the Lebesgue decomposition of ν where νs is singular with respect to σ and
fˆ ∈ L1 (S), then for every sequence rk ↑ 1, there exists a subsequence of {rk },
denoted {rk }, such that
lim F(rk t) = fˆ(t) for almost every t ∈ S.
k→∞
Remark 7.3.9 In Theorem 8.3.3 of the next chapter we prove a much stronger
version of the above; namely, we prove that fˆ is the non-tangential limit of F
a.e. on S. The much weaker version stated above is required for the proof of
Theorem 7.4.2.
For the proof of Theorem 7.3.8 we require the following lemma.
Lemma 7.3.10 Let νs be singular with respect to σ and let H(x) = Ph [νs ](x).
Then Hr → 0 in measure as r → 1; that is, for every > 0,
lim σ ({t ∈ S : |H(rt)| ≥ }) = 0.
r→1
Proof. Since the total variation measure |νs | is also singular with respect to σ
and |H(x)| ≤ Ph [|νs |](x) we can without loss of generality assume that νs ≥ 0.
Since Hr converges weakly to νs we have

lim H(rt)g(t)dσ (t) = g(t)dνs (t)
r→1 S S
for every g ∈ C(S). Let C be a closed subset of S. Since the characteristic
function χC is upper semicontinuous, by Theorem 4.3.2 there exists a sequence
{gk } of continuous functions such that gk ↓ χC . Then

lim sup Hr dσ ≤ lim Hr gk dσ = gk (t)dνs (t).
r→1 C r→1 S S
Since the above holds for all k we have

lim sup Hr dσ ≤ νs (C)
r→1 C
for every closed subset C of S.
Since νs is singular, there exists a measurable set A ⊂ S such that σ (A) = 0
and νs (S \ A) = 0. Let U be any open subset of S containing A. Then for any
> 0,

1
σ ({t ∈ S \ U : H(rt) ≥ }) ≤ H(rt)dσ (t).
S\U
Since S \ U is closed and contained in S \ A,
lim sup σ ({t ∈ S \ U : H(rt) ≥ }) = 0. (7.3.5)

r→1
Since (7.3.5) is valid for all > 0 and all open sets U containing A, Hr → 0
in measure on S \ A and hence on S.
Proof of Theorem 7.3.8 Let {rk } be an increasing sequence in (0, 1) with

rk → 1. By Exercise 5.7.1, Ph [ fˆ](rt) → fˆ(t) in L1 , and hence also in measure.
Thus by [71, Proposition 18] there exists a subsequence of {rk }, which without
loss of generality is denoted by {rk }, such that
lim Ph [ fˆ](rk t) = fˆ(t) and lim Ph [νs ](rk t) = 0

k→∞ k→∞
for almost every t ∈ S. Therefore
lim F(rk t) = fˆ(t)

k→∞
for almost every t ∈ S.
7.4 Hardy–Orlicz Spaces of H-Subharmonic Functions

If f is H-subharmonic and φ is a convex, non-decreasing function on R
with φ(−∞) = limt→−∞ φ(t), then φ( f ) is H-subharmonic. A non-negative
convex function φ is strongly convex if φ is non-decreasing, φ(t) → φ(−∞)
as t → −∞, and φ(t)/t → ∞ as t → ∞.
Definition 7.4.1 For a strongly convex function ϕ, set

Sϕ = f : f is H-subharmonic and sup ϕ( f (rt))dσ (t) < ∞ .
0<r<1 S
Likewise
Hϕ = { f : f is H-harmonic and | f | ∈ Sϕ }.
The spaces Sϕ and Hϕ are called the Hardy–Orlicz spaces of H-subharmonic

and H-harmonic functions corresponding to ϕ.
Since we assume that ϕ is non-negative, in the definition of Sϕ we do not
assume that the H-subharmonic function is non-negative. If for p > 1 we take
p
x if x ≥ 0,
ϕp (x) =
0 if x < 0,
then ϕp is strongly convex and

Sϕp = { f : f is H-subharmonic with f + ∈ S p }.
On the other hand Hϕp = Hp . We are now ready to prove the following
analogue for H-subharmonic functions of a theorem of L. Garding and
L. Hörmander [26].
Theorem 7.4.2 Let f be H-subharmonic on B and let φ be a strongly convex
function on R. Assume that

φ( f (rt))dσ (t) ≤ C < ∞, 0 ≤ r < 1. (7.4.1)
S
Then f has a boundary measure dνf = fˆdσ + dνs with fˆ ∈ L1 and νs ≤ 0.

Furthermore, the boundary measure of φ( f ) is absolutely continuous and
equals φ( fˆ)dσ . In particular,

lim |φ( f (rt)) − φ( fˆ(t))|dσ (t) = 0. (7.4.2)
r→1− S
Proof. Since t/φ(t) → 0 as t → ∞, we have f (rt) ≤ cφ( f (rt)) for some

positive constant c. Since φ ≥ 0, we also have f + (rt) ≤ cφ( f (rt)). Hence by
Theorem 7.3.6 f has a least H-harmonic majorant of the form Ph [νf ].
Let O ⊂ S be open and let 0 ≤ g ≤ 1 be continuous and vanishing on S \ O.
Set
α(s) = sup(t/φ(t)), t ≥ s > 0.
Then f (rt) ≤ α(s)φ( f (rt)) whenever f (rt) ≥ s. Therefore

f (rt)g(t)dσ (t) ≤ α(s) φ( f (rt))g(t)dσ (t) + s g(t)dσ (t).
S S O
Hence by (7.3.4)

gdνf = lim f (rt)g(t)dσ (t) ≤ Cα(s) + sσ (O).
S r→1− S
Since χO is lower semicontinuous, by Theorem 4.3.2 there exists an increasing

sequence of continuous functions gn with gn → χO . Thus by the above
νf (O) ≤ Cα(s) + sσ (O).
Set s = σ (O)−1/2 . Then s → ∞ as σ (O) → 0. Thus α(s) → 0 and the right
side above goes to zero. Therefore
lim νf (O) ≤ 0
σ (O)→0
for every open set O.

Let dνf = fˆdσ + dνs be the Lebesgue decomposition of νf where fˆ is in

L1 (S)and νs is singular with respect to σ . Since

gfˆdσ ≤ | fˆ|dσ ,

S O
which goes to zero as σ (O) → 0, we obtain
lim νf (O) = lim νs (O) ≤ 0.

σ (O)→0 σ (O)→0
From this it now follows that νs ≤ 0.

Let ν be the boundary measure of φ( f ). Then

lim φ( f (rt))g(t)dσ (t) = g(t)dν(t) (7.4.3)
r→1− S S
for every continuous function g. As a consequence of (7.3.3) and Theorem

7.3.8 there exists a sequence rj → 1 such that f (rj t) → fˆ(t) a.e. Hence, taking
g ≥ 0 in (7.4.3) and applying Fatou’s lemma we have

ˆ
φ( f )gdσ ≤ gdν.
Hence φ( fˆ) ∈ L1 (S) and

φ( fˆ)dσ ≤ dν.
Since νs ≤ 0,
f (x) ≤ Ph [ fˆ](x). (7.4.4)
Hence by Jensen’s inequality
φ( f (x)) ≤ Ph [φ( fˆ)](x). (7.4.5)
Since Ph [ν] is the least H-harmonic majorant of φ( f ), we have by (7.4.5) that

dν ≤ φ( fˆ)dσ . Hence ν is absolutely continuous with dν = φ( fˆ)dσ . Finally,
since

lim |φ( f (rt)) − φ( fˆ(t))|dσ (t)
r→1 S

≤ lim |Ph [φ( fˆ)](rt)dσ (t) − φ( f (rt))|dσ (t)
r→1 S

+ lim |Ph [φ( fˆ)](rt) − φ( fˆ(t))|dσ (t),
r→1 S
the result (7.4.2) follows by (7.3.3) and Exercise 5.7.6.
As a consequence of Theorem 7.4.2 we have the following theorem.

Theorem 7.4.3 Suppose f ≥ 0 is such that f po is H-subharmonic for some

po , 0 < po ≤ 1. If f ∈ S p for some p > po , then there exists a function
fˆ ∈ L p (S) such that

f p (x) ≤ Ph (x, ζ )fˆp (ζ )dσ (ζ ) (7.4.6)
S
and
lim | f p (rζ ) − fˆp (ζ )|dσ (ζ ) = 0. (7.4.7)
r→1 S
Proof. If po = 1, then we apply the previous theorem to f with φ(x) = xp , x ≥

0, which is strongly convex for p > 1. If po < 1, then we consider the H-
subharmonic function g = f po which is in S r where r = p/po . By (7.4.4),
f p (x) = gr (x) ≤ Ph [ĝr ](x)
where ĝ is the boundary function of g. Finally, since

lim | f p (ρζ ) − ĝr (ζ )|dσ (ζ ) = 0,
ρ→1 S
we have that fˆ exists a.e. on S and fˆp = ĝr a.e. on S.

Remark 7.4.4 As a consequence of (7.4.7), if f po is H-subharmonic for some
po , 0 < po ≤ 1, and f ∈ S p for some p > po , then

f pp = lim f p (rζ )dσ (ζ ) = fˆpp . (7.4.8)
r→1 S
7.5 Exercises
7.5.1. Harnack’s theorem. Let { fn } be a non-decreasing sequence of H-
harmonic functions on an open connected set ⊂ B. Prove that either
fn (x) → ∞ for all x ∈ or that { fn } converges uniformly on compact
subsets of to an H-harmonic function f .
7.5.2. (a) Fix ζ ∈ S. Prove that Ph (·, ζ ) ∈ Hp for all p, 0 1.
p
(b) For α ≥ 1 show that Pαh (·, ζ ) ∈ Hλα for all p, 0 < p ≤ α1 , where
λα = 4(n − 1)2 α(α − 1).
7.5.3. (a) Let f be a non-negative H-subharmonic function with least H-
harmonic majorant Ff . Prove that

1 + |x| n−1
f (x) ≤ Ff (0).
1 − |x|
7.5 Exercises 113
(b) Prove that

n−1
1 + |x|
| f (x)| ≤
p
f pp
1 − |x|
for all x ∈ B and all f ∈ Hp , 1 ≤ p < ∞, and

1 + |x| n−1
| f (x)| ≤ f ∞ , p = ∞.
1 − |x|
7.5.4. Prove that (H∞ , d∞ ) is a complete metric space.
Definition 7.5.1 A family of functions F ⊂ L1 (S) is said to be
uniformly
integrable if for every > 0 there exists a δ > 0 such
that E | f |dσ < whenever f ∈ F and σ (E) < δ.
7.5.5. (a) If F ⊂ Sϕ for some strongly convex function ϕ, prove that { f + :
f ∈ F} is uniformly integrable on S.
(b) Let f be H-subharmonic on B. If { fr+ : 0 < r < 1} is uniformly
integrable on S, prove that νs ≤ 0 where νs is the singular part of
the boundary measure νf of f .
7.5.6. Let
H∗1 = { f ∈ H1 : {| fr |}0<r<1 is uniformly integrable on S}.
Prove that f ∈ H∗1 if and only if f = Ph [ fˆ] where fˆ ∈ L1 (S).

7.5.7. Suppose U is positive and H-harmonic on H. Prove that there exists a
positive Borel measure μ on Rn−1 and a non-negative constant c such
that
U(x, y) = PH [μ](x, y) + cyn−1 .
Hint: Consider V(x) = U((x)) where is the Möbius transformation
given by 2.3.2. (See [10, Theorem 7.24] for the analogous result for
Euclidean harmonic functions on H.)
8
Boundary Behavior of Poisson Integrals
In Theorem 5.3.3 we proved that if f is a continuous function on S, then

limx→ζ Ph [ f ](x) = f (ζ ) for every ζ ∈ S. In this chapter we investigate
the non-tangential boundary behavior of Poisson integrals of functions in
L p (S), 1 ≤ p < ∞. We begin with the Hardy–Littlewood maximal function
in Section 8.1 for functions and measures on S. The results of this section are
standard and are included primarily for completeness. Section 8.2 contains
the main results concerning radial and non-tangential maximal function of
continuous functions on B. In Section 8.3 we prove Fatou’s theorem on the
existence of non-tangential limits of Poisson integrals of measures, and in
Section 8.4 we prove a local Fatou theorem for H-harmonic functions. One
of the results of Section 8.2 is the equivalence of the L p (S) norm of the radial
and non-tangential maximal function for all p, 0 < p < ∞. This result is then
used in Section 8.5 to obtain an L p inequality for the non-tangential maximal
function valid for all p, 0 0, let
S(ζ , δ) = {t ∈ S : |t − ζ | < δ}. (8.1.1)
Definition 8.1.1 If μ is a signed Borel measure on S, the maximal function
of μ, denoted Mμ, is the function on S defined by
|μ|(S(ζ , δ))
Mμ(ζ ) = sup ,
δ>0 σ (S(ζ , δ))
114
where |μ| denotes the total variation of μ. Likewise, if f ∈ L1 (S), the maximal
function of f , denoted Mf ,1 is defined by

1
Mf (ζ ) = sup | f | dσ .
δ>0 σ (S(ζ , δ)) S(ζ ,δ)
Theorem 8.1.2 There exists a constant Bn , depending only on n, such that for
any signed measure μ on S,
|μ|(S)
σ ({ζ ∈ S : Mμ(ζ ) > t}) ≤ Bn .
t
Lemma 8.1.3 Suppose E ⊂ S is the union of a finite collection = {S(ζi , δi )}
of balls in S. Then there exists a finite disjoint collection {S(ζik , δik )}m
k=1 such
that
m
E⊂ S(ζik , 3δik )
k=1
and

m
σ (E) ≤ Bn σ (S(ζik , δik )),
k=1
where Bn is a constant depending only on n.

Remark 8.1.4 Although we prove Lemma 8.1.3 for a finite collection, with
minor modifications, the result is also true if is a countable collection. In
this case, the conclusion states that there exists a finite or countable disjoint
sub-collection of such that the conclusions of the lemma hold.
Proof. Without loss of generality we can assume that the balls {S(ζi , δi )} have
been ordered such that δi+1 ≤ δi . Set Si = S(ζi , δi ).
Suppose k = {Si1 , . . . , Sik } have been chosen. If for each i > ik , Si ∩ Sij =
φ for some j = 1, . . . , k, the process stops. If not, let ik+1 > ik be the smallest
integer such that
&
Sik+1 Sij = φ for all j = 1, . . . , k.
Set k+1 = k ∪ {Sik+1 }. This process terminates since is finite.
Let Si ∈ and let m = {Si1 , . . . , Sim }. If i > im , then Si ∩ Sij = φ for some
ij , j ∈ {1, . . . , m}. Since δij ≤ δi , Si ⊂ 3Sij .2 If i < im , then there exists a k such
1 The maximal function Mf is usually referred to as the Hardy–Littlewood maximal function.

Also, in the definition of Mμ and Mf one could have used δ n−1 instead of σ (S(ζ , δ)).
2 For a ball S = S(ζ , δ), 3S = S(ζ , 3δ).
116 Boundary Behavior of Poisson Integrals
that ik ≤ i < ik+1 . But then Si ∩ Sij = φ for some j ∈ {1, . . . , k}. Since δij ≥ δi
we have Si ⊂ 3Sij . Thus
m
E⊂ 3Sij
j=1
and by the sub-additivity of the measure σ ,

m
σ (E) ≤ σ (3Sij ).
j=1
Finally, since σ (S(ζ , δ)) ≈ δ n−1 (see Exercise 8.7.1), we have

m
σ (E) ≤ Bn σ (S(ζij , δij )),
j=1
for some constant Bn .
Proof of Theorem 8.1.2. We fix t > 0. We first note that for fixed δ > 0 the
function
|μ|(S(ζ , δ))
ζ →
σ (S(ζ , δ))
is lower semicontinuous on S. Thus by Theorem 4.3.2 Mμ(ζ ) is also lower
semicontinuous on S. Therefore
E(t) = {ζ ∈ S : Mμ(ζ ) > t}
is an open subset of S. Let K be a compact subset of E(t). For each ζ ∈ K,
there exists a δζ > 0 such that
|μ|(S(ζ , δζ )) > σ (S(ζ , δζ )) t.
By compactness of K there exists a finite sub-collection {S(ζi , δi )} which also
covers K. Thus by Lemma 8.1.3 there exists a finite disjoint sub-collection
{S(ζij , δij )} such that

σ (K) ≤ Bn σ (S(ζij , δij ))
j
Bn # $
≤ |μ| S(ζij , δij )
t
j
⎛ ⎞
Bn
≤ |μ| ⎝ S(ζij , δij )⎠
t
j
Bn
≤ |μ|(S).
t
Taking the supremum over K ⊂ E(t) proves the result.
As a consequence of Theorem 8.1.2 we

obtain the following corollary by
defining the measure μ on S by μ(E) = E fdσ , f ∈ L1 (S).
Corollary 8.1.5 Let f ∈ L1 (S). Then

Cn
σ ({ζ ∈ S : Mf (ζ ) > t}) ≤ | f |dσ .
t S
Theorem 8.1.6 For 1 < p ≤ ∞,

|Mf |p dσ ≤ Ap | f |p dσ
S S
p
for all f ∈ L p (S), where Ap = Cn p−1 for 1 < p < ∞ and A∞ = 1.
For the proof of Theorem 8.1.6 we require the following lemma, the proof
of which is left as an exercise (Exercise 8.7.2).
Lemma 8.1.7 For f ∈ L p (S) we have
∞
| f |p dσ = p t p−1 λ(t)dt,
S 0
where
λ(t) = σ ({ζ ∈ S : | f (ζ )| > t}).
Proof of Theorem 8.1.6. For the proof of this theorem we follow the method
used by E. M. Stein in [75, Theorem 1, p. 5]. For p = ∞ we have Mf ∞ ≤
f ∞ . Assume now that 1 0 define f1 by

f (x) if | f (x)| ≥ 12 t,
f1 (x) =
0 otherwise.
Then | f (x)| ≤ | f1 (x)| + 12 t and

1 t
| f |dσ ≤ Mf1 + .
S(ζ , δ) S(ζ ,δ) 2
Thus Mf ≤ Mf1 + 12 t, and by Corollary 8.1.5
2Cn
σ (Et ) = σ ({ζ : Mf (ζ ) > t}) ≤ f1 1 ,
t
or equivalently,

2Cn
σ (Et ) = σ ({ζ : Mf (ζ ) > t}) ≤ | f |dσ . (8.1.2)
t | f |>t/2
Set g = Mf and λ(t) = σ ({ζ : g(ζ ) > t}). Then by Lemma 8.1.7
∞
(Mf ) dσ = p
p
t p−1 λ(t)dt.
S 0
But by (8.1.2) we have

∞
Mf pp = p t p−1 λ(t)dt
0
∞
p−1 2Cn
≤p t | f |dσ dt,
0 t | f |>t/2
which by interchanging the order of integration

2| f (ζ )|
= 2pCn | f (ζ )| t p−2 dtdσ (ζ ).
S 0
But
2| f (ζ )| 1
t p−2 dt = |2f (x)|p−1 .
0 (p − 1)
Therefore
p
Mf pp ≤ 2Cn | f |p dσ ,
(p − 1) S
Another application of Theorem 8.1.6 is the following differentiation
theorem.
Theorem 8.1.8 If f ∈ L1 (S), then

1
lim | f − f (ζ )|dσ = 0 (8.1.3)
δ→0+ σ (S(ζ , δ)) S(ζ ,δ)
for almost every ζ ∈ S. Hence

1
f (ζ ) = lim fdσ a.e. on S. (8.1.4)
δ→0+ σ (S(ζ , δ)) S(ζ ,δ)
Remark 8.1.9 The points ζ for which (8.1.3) holds are called the Lebesgue
points of f . If E ⊂ S and f = χE , the characteristic function of E, then every
point of E that is a Lebesgue point of f is called a point of density of E.
Proof. If f is continuous, then (8.1.3) holds for every ζ ∈ S. For arbitrary
f ∈ L1 (S) and > 0, choose g ∈ C(S) such that f − g1 < .
Let
1
Tf (ζ ) = lim sup | f − f (ζ )|dσ .
δ→0+ σ (S(ζ , δ)) S(ζ ,δ)
Then
(a) Tf (ζ ) ≤ Tg (ζ ) + Tf −g (ζ ), and
(b) Tf −g (ζ ) ≤ | f (ζ ) − g(ζ )| + M[ f − g](ζ ),
where M[ f − g] is the maximal function of f − g. Since g is continuous,

Tg (ζ ) = 0 for all ζ ∈ S. Thus
Tf (ζ ) ≤ | f (ζ ) − g(ζ )| + M[ f − g](ζ ).
Thus for t > 0,
σ ({ζ : Tf (ζ ) > t}) ≤ σ ({ζ : | f (ζ ) − g(ζ )| > 2t })

+ σ ({ζ : M[ f − g](ζ ) > 2t }).
If we let Et = {ζ : | f (ζ ) − g(ζ )| > 2t } then

t
f − g1 ≥ | f − g|dσ ≥ σ (Et ).
Et 2
Therefore,
2 2
σ ({ζ : | f (ζ ) − g(ζ )| > 2t }) ≤ f − g1 ≤ .
t t
Also, by Corollary 8.1.5
2Cn 2Cn
σ ({ζ : M[ f − g](ζ ) > 2t }) ≤ f − g1 ≤ .
t t
Therefore,
C
σ ({ζ : Tf (ζ ) > t}) ≤ .
t
Since > 0 was arbitrary we have
σ ({ζ : Tf (ζ ) > t}) = 0
for every t > 0. Thus Tf (ζ ) = 0 a.e. on S which proves (8.1.3) as well as

(8.1.4).
Definition 8.1.10 If ν is a finite positive measure on S, define the upper

derivate Dν of ν on S by
ν(S(ζ , δ))
Dν(ζ ) = lim sup . (8.1.5)
δ→0+ σ (S(ζ , δ))
If the limit in (8.1.5) exists, we write
ν(S(ζ , δ))
Dν(ζ ) = lim .
δ→0+ σ (S(ζ , δ))
One of the advantages in considering Dν is that unlike Mν it depends only
on the behavior of ν(S(ζ , δ)) for small values of δ.
Lemma 8.1.11 If ν is a finite positive singular measure on S, then

Dν = 0 σ - a.e. on S.
Proof. Suppose the conclusion is false. Then since Dμ is lower semicontinu-
ous on S, there exists a positive constant a, a compact set K ⊂ S with σ (K) > 0
and ν(K) = 0, such that Dν(ζ ) > a for all ζ ∈ K.
Let > 0 be arbitrary. Choose an open set O ⊂ S such that K ⊂ O and
ν(O) < . Since Dν(ζ ) > a for every ζ ∈ K, we can cover K by finitely many
S(ζi , δi ) ⊂ O and ν(S(ζi , δi )) > aσ (S(ζi , δi )). By Lemma 8.1.3 there exists a
finite disjoint sub-collection S(ζik , δik ) such that

σ (K) ≤ Bn σ (S(ζik , δik ))
k
Bn
≤ ν(S(ζik , δik ))
a
k
Bn Bn
≤ ν(O) ≤ .
a a
Since > 0 was arbitrary, we have σ (K) = 0, which is a contradiction.
8.2 Non-tangential and Radial Maximal Function

As in the Euclidean case, for ζ ∈ S and α > 1, we denote by α (ζ ) the
non-tangential approach region at ζ defined by
α (ζ ) = {y ∈ B : |y − ζ | < α(1 − |y|)}. (8.2.1)
Definition 8.2.1 Let f be a continuous function on B. For α > 1, the non-
tangential maximal function of f , denoted Mα f , is defined on S by
Mα f (ζ ) = sup{| f (x)| : x ∈ α (ζ )}.
Also, the radial maximal function of f , denoted Mrad f , is the function on S
defined by
Mrad f (ζ ) = sup | f (rζ )|.
0≤r<1
One clearly has that for any f ∈ C(B), Mrad f ≤ Mα f . In the reverse
direction we have the following theorem.
Theorem 8.2.2 For 0 1, there exists a constant Cp,α such that

(Mα f )p dσ ≤ Cp,α (Mrad f )p dσ
S S
for all non-negative, continuous, quasi-nearly H-subharmonic functions.
The proof of Theorem 8.2.2 is similar to the proof given for harmonic
functions in [27, Theorem 3.6] as well as Lemma 8 of [90].
Proof. Throughout the proof we let E(x) = E(x, 12 ) = {y : |ϕx (y)| < 12 }. By
Theorem 2.2.2, E(x) is also a Euclidean ball B(cx , rx ) with center cx and radius
rx given by
3x 2(1 − |x|2 )
cx = and rx = .
(4 − |x|2 ) (4 − |x|2 )
Hence if y ∈ E(x) we have
2|x| + 1
|y| ≤ |cx | + rx ≤ <1
2 + |x|
for all x, |x| < 1. Also, if we write x = rη and y = ρt, then
|t − η| ≤ (1 − |y|) + (1 − |x|) + |x − y|.
But
4 8
|x − y| ≤ 2rx < (1 − |x|2 ) ≤ (1 − |x|).
3 3
Also, by Exercise 2.4.1(b),
(1 − |y|) ≤ 3(1 − |x|2 ) ≤ 6(1 − |x|).
Therefore |t − η| < c(1 − |x|) for a fixed positive constant c > 0. Thus

2|x| + 1
E(x) ⊂ 0, × S(η, c(1 − |x|)). (8.2.2)
2 + |x|
Let x ∈ α (ζ ), ζ ∈ S, and let 0 < q < p. Since f is quasi-nearly
H-subharmonic on B,

f q (x) ≤ 2n C f q (y)dτ (y),
E(x)
which by (8.2.2)
(2|x|+1)/(2+|x|)
≤ Cn (1 − r)−n f q (rt)dσ (t)dr.
0 S(η,c(1−|x|))
Hence by Fubini’s theorem,

(2|x|+1)/(2+|x|)
f (x) ≤ Cn
q q
(Mrad f ) (t)dσ (t) (1 − r)−n dr.
0
S(η,c(1−|x|))
But
(2|x|+1)/(2+|x|) cn
(1 − r)−n dr ≤
0 (1 − |x|)n−1
and S(η, c(1 − |x|)) ⊂ S(ζ , cα (1 − |x|)) for some fixed constant cα > 0.
Therefore

Cn,α
f q (x) ≤ (Mrad f )q (t)dσ (t) ≤ Mg(ζ ),
(1 − |x|)n−1
S(ζ ,cα (1−|x|))
where g = (Mrad f )q . Therefore

(Mα f )p (ζ ) ≤ Cn,α (Mg)p/q (ζ ).
Since p/q > 1, by Theorem 8.1.6

(Mα f ) dσ ≤ Cp,α,n g dσ = Cp,α,n (Mrad f )p dσ ,
p p/q
S S S
provided Mrad f ∈ L p (S).
Theorem 8.2.3 For every α > 1, there exists a constant Aα such that
Mα Ph [μ] ≤ Aα Mμ
for every signed Borel measure μ on S.
The proof of the Theorem 8.2.3 follows standard techniques. A proof of the
analogous result for the M-harmonic Poisson kernel P on the unit ball in Cn
can be found in [72], whereas for the Euclidean Poisson kernel a proof of the
result may be found in [10].
Proof. Since |Ph [μ]| ≤ Ph [|μ|] we can assume without loss of generality that
μ is a finite Borel measure on S. Fix a ζ ∈ S for which Mμ(ζ ) < ∞, and fix
x ∈ α (ζ ). Set β = α(1−|x|). Let N be the smallest integer such that 2n β > 2.
Set
V0 = {τ ∈ S : |τ − ζ | < β},
and for k = 1, . . . , N set

Vk = {τ ∈ S : 2k−1 β ≤ |τ − ζ | < 2k β}.
Then

Ph [μ](x) = Ph (x, τ )dμ(τ )
S
N

= Ph (x, τ )dμ(τ ) + Ph (x, τ )dμ(τ ).
V0 k=1 Vk
For τ ∈ V0 we have
(1 − |x|2 )n−1 2n−1
Ph (x, τ ) = ≤ .
|x − τ |2(n−1) (1 − |x|)n−1
Therefore,

(2α)n−1
Ph (x, τ )dμ(τ ) ≤ dμ ≤ (2α)n−1 Mμ(ζ ).
V0 β n−1 S(ζ ,β)
For x ∈ α (ζ ) we have
|τ − ζ | ≤ |x − ζ | + |x − τ |
< α(1 − |x|) + |x − τ | < (α + 1)|x − τ |.
Therefore |x − τ | > cα |τ − ζ |, which for τ ∈ Vk gives
|x − τ | > cα 2k−1 β
and thus
2n−1 (1 − |x|)n−1
Ph (x, τ ) ≤ 2(n−1) k−1 2(n−1) 2(n−1)
cα (2 ) β
Cα,n
≤ .
(2n−1 )k (2k β)n−1
Therefore

Cα,n 1 Cα,n
Ph (x, τ )dμ(τ ) ≤ dμ ≤ Mμ(ζ ).
Vk (2 ) (2 β)n−1
n−1 k k (2n−1 )k
S(ζ ,2k β)
Summing the above over k gives that for x ∈ α (ζ ),

Ph [μ](x) = Ph (x, τ )dμ(τ ) ≤ Cn,α Mμ(ζ ).
S
Thus Mα Ph [μ](ζ ) ≤ Cn,α Mμ(ζ ).
As a consequence of Theorem 8.2.3 we always have that
Mrad Ph [μ] ≤ A Mμ (8.2.3)
for any signed Borel measure μ on S. We now prove that the reverse inequality
also holds for positive Borel measures on S.
Theorem 8.2.4 If μ is a positive Borel measure on S, then
Mμ ≤ Cn Mrad Ph [μ]
for a positive constant Cn depending only on n.

Proof. Fix ζ ∈ S and for 0 < δ < 2 set (1 − r) = 12 δ. Then

(1 − r2 )n−1
Ph [μ](rζ ) = dσ (η)
S |rζ − η|
2(n−1)

(1 − r)n−1
≥ dσ (η).
S(ζ ,δ) |rζ − η|
2(n−1)
But for η ∈ S(ζ , δ), |rζ − η| ≤ (1 − r) + |ζ − η| < 32 δ. Therefore
2n−1 1
Ph [μ](rζ ) ≥ μ(S(ζ , δ)).
32(n−1) δ n−1
Since σ (S(ζ , δ)) ≈ δ n−1 (see Exercise 8.7.1)
μ(S(ζ , δ))
Ph [μ](rζ ) ≥ Cn ,
σ (S(ζ , δ))
Thus if μ ≥ 0 and if one of Mμ, Mrad Ph [μ], Mα Ph [μ] is finite at some

ζ ∈ S, then so are the others.
Combining Theorems 8.1.6 and 8.2.3 gives the following theorem.
Theorem 8.2.5 If 1 < p < ∞ and f ∈ L p (S), then

|Mα Ph [ f ]| dσ ≤ Aα,p | f |p dσ ,
p
S S
where Aα,p is a constant independent of f .
As a corollary we have the following characterization of functions in S p ,

and thus also Hp .
Corollary 8.2.6 Suppose f ≥ 0 is such that f po is H-subharmonic for some

po , 0 < po ≤ 1. Then f ∈ S p for some p > po if and only if Mα f ∈ L p (S). If
this is the case, then
f p ≤ Mα f p ≤ Ap,α f p .
Proof. Since | f (rζ )|p ≤ (Mα f )p (ζ ), we have f ∈ S p whenever Mα f ∈ L p

with f p ≤ Mα f p . Suppose now that f ∈ S p for some p > po . Set g = f po .
Then g ∈ S r with r = p/po . By inequality (7.4.4), g(x) ≤ Ph [ĝ](x) where ĝ is
the boundary function of g. Thus
f p (x) = gr (x) ≤ (Ph [ĝ])r (x),

8.3 Fatou’s Theorem 125
and by Theorem 8.2.5

(Mα f ) dσ ≤ (Mα Ph [ĝ]) dσ ≤ Aα,r ĝr dσ .
p r
S S S
But ĝr = fˆp a.e. and fˆp = f p , which proves the result.
Remark 8.2.7 For p > 1 we have that f ∈ Hp if and only if Mα f ∈ L p . Using
this fact it is possible to define Hp for 0 < p ≤ 1 by
Hp = { f : f is H-harmonic and Mα f ∈ L p (S)}. (8.2.4)
For further details the reader is referred to the papers [42] by P. Jaming and
[31] by S. Grellier and P. Jaming.
8.3 Fatou’s Theorem

In this section we will prove Fatou’s theorem3 on the existence of non-
tangential limits of Poisson integrals of measures. We begin with the following
definition.
Definition 8.3.1 A continuous function f on B has a non-tangential limit at
ζ ∈ S if there exists a real number L such that
lim f (x) = L
x→ζ
x∈α (ζ )
for every α > 1.

In Theorem 8.3.3 we prove that if f is the Poisson integral of a measure on S,
then f has a non-tangential limit at a.e. ζ ∈ S. We begin with the following
theorem.
Theorem 8.3.2 If ν is a Borel measure on S satisfying Dν(ζ ) = 0, ζ ∈ S, then
for every α > 1,
lim Ph [ν](x) = 0.
x→ζ
x∈α (ζ )
Proof. Let > 0 be arbitrary. Since Dν(ζ ) = 0, there exists a δo > 0 such
that
ν(S(ζ , δ)) < σ (S(ζ , δ)) (8.3.1)
3 In the setting of rank one symmetric spaces, Fatou’s theorem was proved originally by
A. W. Knapp in [45].
for all δ, 0 < δ < δo . Let So = S(ζ , δo ). Set

νo = ν|So and ν1 = ν − νo .
If t ∈ S \ So then |t − ζ | ≥ δo . Also, as in the proof of Theorem 8.2.3, |t − ζ | ≤
(α + 1)|x − t|. Therefore for x ∈ α (ζ ),

(α + 1)2(n−1) (1 − |x|2 )n−1 α + 1 2(n−1)
Ph (x, t) ≤ ≤ (1 − |x|2 )n−1 .
|t − ζ |2(n−1) δo
Thus
lim Ph [ν1 ](x) = 0.
x→ζ
x∈α (ζ )
Also, by (8.3.1), M[νo ](ζ ) < . Therefore by Theorem 8.2.3,

lim sup Mα Ph [νo ](x) ≤ Aα ,
x→ζ
x∈α (ζ )
from which the result follows.

As a consequence of Theorem 8.1.8 we have the following result concerning
non-tangential boundary limits of Poisson integrals.
Theorem 8.3.3 (Fatou’s Theorem)
(a) If f ∈ L1 (S), then for every α > 1,
lim Ph [ f ](x) = f (ζ ) σ - a.e. on S.
x→ζ
x∈α (ζ )
(b) If ν is a signed Borel measure on S which is singular with respect to σ ,

then for every α > 1,
lim Ph [ν](x) = 0 σ - a.e. on S.
x→ζ
x∈α (ζ )
Proof. We first prove part (b) of the theorem. Suppose ν is a signed Borel
measure on S which is singular with respect to σ . Since the total variation
measure |ν| is also singular, by Lemma 8.1.11, D|ν|(ζ ) = 0 σ - a.e. on S. Thus
by Theorem 8.3.2,
lim Ph [|ν|](x) = 0 σ - a.e. on S.
x→ζ
x∈α (ζ )
Since |Ph [ν]| ≤ Ph [|ν|] we obtain (b).

For the proof of (a), if f ∈ L1 (S), by Theorem 8.3.2

1
lim | f − f (ζ )|dσ = 0
δ→0+ σ (S(ζ , δ)) S(ζ ,δ)
for almost every ζ ∈ S. Fix such a ζ and define μ on S by

μ(E) = | f − f (ζ )|dσ
E
for all Borel subsets E of S. Then Dμ(ζ ) = 0 and
|Ph [ f ](x) − f (ζ )| ≤ Ph [μ](x).
But then by Theorem 8.3.2 again,
lim |Ph [ f ](x) − f (ζ )| ≤ lim Ph [μ](x) = 0,

x→ζ x→ζ
x∈α (ζ ) x∈α (ζ )
which completes the proof.
8.4 A Local Fatou Theorem for H-Harmonic Functions

As a consequence of Theorems 7.1.1 and 8.3.3, if f is a bounded H-harmonic
function on B, then f has a non-tangential limit at almost every ζ ∈ S. In this
section we will prove a local Fatou theorem of I. Privalov [68] for (weakly)
non-tangentially bounded H-harmonic functions on S. The result of Privalov
was first extended to several variables for harmonic functions in half-spaces
by A. P. Calderón [12]. See also [10, Chapter 7]. The result was subsequently
proved by A. Koranyi and R. P. Putz [48] for harmonic functions on rank one
symmetric spaces, which includes H-harmonic functions in B.
Our method of proof follows the techniques used, with minor modifications,
by W. Rudin [72, Section 5.5] for M-harmonic functions.
Definition 8.4.1 For E ⊂ S and α > 1, let

(E, α) = α (ζ ).
ζ ∈E
Our main result of this section is the following theorem.
Theorem 8.4.2 If E ⊂ S is measurable with σ (E) < 1, α > 1, and u is a

bounded function on (E, α) such that Δh u = 0, then
lim u(x) exists

x→ζ
x∈β (ζ )
for almost every ζ ∈ E and every β > 1.
For the proof of Theorem 8.4.2 we require several lemmas.

Lemma 8.4.3 If E ⊂ S, σ (E) = m < 1, α > 1, then there exists a constant

c = c(α, m) > 0 such that
Ph [χS\E ](x) ≥ c
for all x ∈ B \ (E, α).
Proof. Let V = S \ E and f = χV . Let x ∈ B \ (E, α). Write x = rη, η ∈ S.
If r ≤ α1 , then

(1 − r2 )n−1 1 − r n−1
Ph (rη, ζ ) = ≥ .
|rη − ζ |2(n−1) 1+r
Therefore,

1 − r n−1 α − 1 n−1
Ph (rη, ζ )dσ (ζ ) ≥ σ (V) ≥ (1 − m).
S\E 1+r α+1
Assume α1 < r < 1, and let ϕx be the automorphism of B satisfying Theorem

2.1.2. Then ( f ◦ ϕx )(ζ ) = χϕx (V) (ζ ). Hence

Ph [ f ](x) = Ph [ f ◦ ϕx ](0) = ( f ◦ ϕx )dσ = σ (ϕx (V)).
S
Let
Gα = {t ∈ S : 1 + 1
α < |t − η| ≤ 2}.
Suppose t ∈ Gα . Then
|x − t| = |rη − t| ≥ |η − t| − (1 − r),
which since r > 1

α and t ∈ Gα
>1+ 1
α − (1 − α1 ) = α2 .
Set ζ = ϕx (t). By Exercise 2.4.3,
(1 − |x|2 )
|x − ζ | = .
|x − t|
Therefore,
|x − ζ | < α2 (1 − |x|2 ) < α(1 − |x|).
Hence x ∈ α (ζ ). Since x ∈ (E, α), ζ ∈ V. Therefore ϕx (Gα ) ⊂ V. Hence
Gα ⊂ ϕx (V) and σ (ϕx (V)) > σ (Gα ) > 0. Therefore
Ph [ f ](x) ≥ σ (Gα ) ≥ cα > 0 for |x| > α1 .
Hence Ph [ f ](x) ≥ c = c(α, m) for all x ∈ S \ (E, α).
Lemma 8.4.4 If ζ ∈ E is a point of density of E, α > 1, and β > 1, then there

exists r = r(E, ζ , α, β) < 1 such that
{|x| > r} ∩ β (ζ ) ⊂ (E, α).
Proof. Without loss of generality assume that σ (E) < 1. Define the measure
ν on the Borel subsets of S by

ν(B) = |χE (t) − χE (ζ )|dσ (t).
B
Since ζ is a Lebesgue point of E, Dν(ζ ) = 0. Hence by Theorem 8.3.2,
lim Ph [ν](x) = 0.
x→ζ
x∈β (ζ )
But
|Ph [χE ](x) − χE (ζ )| ≤ Ph [ν](x).
Therefore
lim Ph [χE ](x) = 1 and lim Ph [χS\E ](x) = 0. (8.4.1)

x→ζ x→ζ
x∈β (ζ ) x∈β (ζ )
Let {rj } be an increasing sequence with rj ↑ 1. If the conclusion of the lemma is

false, then for each j, there exists xj ∈ {|x| > rj } ∩ β (ζ ) with xj ∈ B \ (E, α).
But then by Lemma 8.4.3, Ph [χS\E ](xj ) ≥ c > 0, which is a contradiction.
Lemma 8.4.5 Let 0 < r < 1 be fixed. If F is a uniformly bounded family

of functions which are H-harmonic on Br , then F is equicontinuous on each
compact subset K of Br .
Proof. Let K be a compact subset of Br . Choose δ > 0 such that ϕx (Bδ ) ⊂ Br

for all x ∈ K. Let h be a non-negative C∞ radial function with supp h ⊂ Bδ
and

hdτ = 1.
B
Let x ∈ K and f ∈ F. Without loss of generality we may assume that | f | ≤ 1

for all f ∈ F. Since f is H-harmonic on Br ,

f (x) = f (ϕx (ρζ ))dσ (ζ )
S
for all ρ, 0 < ρ < δ. Therefore

1
ρ n−1
f (x) = n f (x)h(ρ) dρ
0 (1 − ρ 2 )n
1
ρ n−1
=n h(ρζ )f (ϕx (ρζ ))dσ (ζ )dρ
0 (1 − ρ )
2 n
S

= h(y)f (ϕx (y)) dτ (y) = f (y)h(ϕx (y)) dτ (y).
B B
Therefore, for x, y ∈ K,

| f (x) − f (y)| ≤ |h(ϕx (w)) − h(ϕy (w))|dτ (y),
B
which since h is radial

= |(h ◦ ϕw )(x) − (h ◦ ϕw )(y)| dτ (y).
B
Since (w, x) → (h ◦ ϕw )(x) is continuous on Br × K, given > 0, there exists

a δ > 0, 0 < δ < δ, such that |(h ◦ ϕw )(x) − (h ◦ ϕw )(y)| < for all x, y ∈ K
with ρh (x, y) < δ and all w ∈ Br . Therefore
| f (x) − f (y)| < τ (Br )
for all f ∈ F and y ∈ K with ρh (x, y) < δ . Hence the family F is

equicontinuous on K.
Proof of Theorem 8.4.2 Without loss of generality we can assume that

0 ≤ u ≤ 1. Also, by replacing α by a slightly smaller α we can assume u
is continuous on (E, α) ∩ B. Set = (E, α). Since u is continuous on
∩ B, by Tietze’s extension theorem [71, p. 179], u extends to a continuous
function on B, also denoted u, with 0 ≤ u ≤ 1.
Let rj ↑ 1. By Theorem 5.4.1 there exists gj ∈ C(Brj ) such that gj = u on
Srj and Δh gj = 0 in Brj with 0 ≤ gj ≤ 1. By Lemma 8.4.5 the family {gj }j ≥ jo
is equicontinuous on Brjo . Hence by the Ascoli–Arzela theorem [71, p. 169]
there exists a subsequence, denoted {gj }, that converges uniformly on compact
subsets of B to an H-harmonic function g on B.
Let v = Ph [χS\E ]. Then by Lemma 8.4.3, v ≥ c on B∩∂. Also, |u−gj | ≤ 1
in Brj with u − gj = 0 on Srj . Therefore
1
|u − gj | ≤ v on ∂( ∩ Brj ). (8.4.2)
c
By the maximum principle (8.4.2) holds at every x ∈ ∩ Brj . Letting j → ∞

gives
1
|u − g| ≤ v in . (8.4.3)
c
Since g is bounded, by Theorem 7.1.1, g = Ph [ĝ] for some bounded function
ĝ on S. Hence by Theorem 8.3.3, the non-tangential limit of g exists at almost
every ζ ∈ S. If ζ is a point of density of E, then by (8.4.3) and (8.4.1)
lim g(x) = lim u(x).

x→ζ x→ζ
x∈β (ζ ) x∈β (ζ )
Therefore g and u have the same non-tangential limits a.e. on E.
Definition 8.4.6 A function f : B → R is said to be (weakly) non-tangentially

bounded at ζ ∈ S if there exists an α > 1 such that
sup | f (x)| < ∞. (8.4.4)

x∈α (ζ )
Likewise, f is non-tangentially bounded at ζ ∈ S if (8.4.4) holds for all α > 1.
Note: The term “weakly” is used to emphasize that (8.4.4) holds only for one
α > 1.
Theorem 8.4.7 If u is H-harmonic in B, E is a measurable subset of S, and

u is weakly non-tangentially bounded at every ζ ∈ E, then the non-tangential
limit of u exists for almost every ζ ∈ E.
Proof. For i = 1, 2, 3, . . . , let
Ei = {ζ ∈ S : |u(x)| ≤ i for all x ∈ 1+ 1 (ζ )}.

i
Since u is continuous in B, each Ei is closed. By Theorem 8.4.2, u has a non-

+
tangential limit at almost every point of each Ei . Since E ⊂ i Ei the result
follows.
8.5 An L p Inequality for Mα f for 0 1, and
to ∈ fixed, there exists a constant Cα,p , depending only on p and α, such that

(Mα f ) (ζ )dS(ζ ) ≤ Cα,p | f (to )| +
p p
δ(y) |∇f (y)| dy ,
p−1 p
∂
where δ(y) is the distance from y to ∂. In the case of the unit ball this result
has also been proved by O. Djordjević and M. Pavlović [17]. In the present
section, as an application of Theorem 8.2.2, we prove that the above inequality
also holds for H-harmonic functions on B.
Theorem 8.5.1 Let f be an H-harmonic function on B. Then for α > 1,

0 < p ≤ 1,

(Mα f ) dσ ≤ Cα,p | f (0)| + (1 − |x| ) |∇f (x)| dν(x) ,
p p 2 p−1 p
S B
where Cα,p is a constant independent of f .
Proof. By the fundamental theorem of calculus, for fixed r, 0 < r < 1, and
ζ ∈ S,
r r h
|∇ f (tζ )|

| f (rζ ) − f (0)| = ∇f (tζ ), ζ dt ≤ dt.
0 0 (1 − t)
Let rk = 1 − 21k , k = 0, 1, 2, . . . , and let N be the first integer such that rN > r.
Then
N
rk |∇ h f (tζ )|
| f (rζ )| ≤ | f (0)| + dt
(1 − t)
k=1 rk−1
N
≤ | f (0)| + (log 2) sup |∇ h f (tζ )|.
k=1 t∈[rk−1 ,rk )
Hence for 0 < p ≤ 1,

N
| f (rζ )|p ≤ | f (0)|p + C sup |∇ h f (tζ )|p .
k=1 t∈[rk−1 ,rk )
For N = 1, 2, 3, . . . and ζ ∈ S, set

N
SN (ζ ) = | f (0)|p + C sup |∇ h f (tζ )|p
k=1 t∈[rk−1 ,rk )
and
∞

S(ζ ) = | f (0)|p + C sup |∇ h f (tζ )|p .
k=1 t∈[rk−1 ,rk )
Then SN (ζ ) ↑ S(ζ ) as N → ∞ and

(Mrad f )p (ζ ) ≤ S(ζ )
for all ζ ∈ S for which S(ζ ) < ∞.
For x ∈ B let B(x) = B(x, 14 (1 − |x|2 )). Since |∇ h f (x)|p is quasi-nearly
H-subharmonic for all p > 0,

|∇ f (tζ )| ≤ Cn
h p
|∇ h f (x)|p dτ (x). (8.5.1)
B(tζ )
If x ∈ B(tζ ), t ∈ (rj−1 , rj ), then

1
|x − rj−1 ζ | ≤ |x − tζ | + (r − rj−1 ) < .
2 j−1
Therefore,
B(tζ ) ⊂ Bj (ζ ) = B(rj−1 ζ , 2−j+1 ).
It is easy to show (see Exercise 2.4.2) that there exists β > 0 such that Bj (ζ ) ⊂
β (ζ ). For j = 2, 3, 4, . . . , set

1 1
Aj = x ∈ β (ζ ) : j+2 < (1 − |x|) < j−2 .
2 2
Thus Bk (ζ ) ⊂ Ak for all k. Hence by (8.5.1),

sup |∇ h f (tζ )|p ≤ Cn |∇ h f (x)|p dτ (x).
t∈[rk−1 ,rk ) Ak
For each i = 1, 2, 3, 4, the sequence of sets {A4k+i }∞k=1 is pairwise disjoint.

Hence by Exercise 8.7.3,
N
SN (ζ ) ≤ | f (0)| + Cn
p
|∇ h f (x)|p dτ (x)
k=1 Ak

≤ | f (0)| + 4Cn
p
+∞ |∇ h f (x)|p dτ (x)
k=1 Ak

≤ | f (0)|p + 4Cn |∇ h f (x)|p dτ (x).
β (ζ )
Letting N → ∞ gives

S(ζ ) ≤ | f (0)| + Cn
p
|∇ h f (x)|p dτ (x).
β (ζ )
Hence

S(ζ )dσ (ζ ) ≤ | f (0)|p + C |∇ h f (x)|p dτ (x)dσ (ζ ),
S S β (ζ )

= |∇ h f (x)|p χ
β (x) (ζ )dσ (ζ ) dτ (x).
B S
In the above, for β > 1 and x ∈ B,

β (x) = {ζ ∈ S : x ∈ β (ζ )}.
But by Exercise 8.7.4, if x = |x|η, η ∈ S, then there exists a positive constant

c > 0, depending only on β, such that

β (x) ⊂ S(η, c(1 − |x|2 )).
Therefore

S(ζ )dσ (ζ ) ≤ | f (0)|p + C (1 − |x|2 )n−1 |∇ h f (x)|p dτ (x)
S
B
= | f (0)| + C
p
(1 − |x|2 )p−1 |∇f (x)|p dν(x).
B

B (1 − |x|
Hence if 2 )p−1 |∇f (x)|p dν(x) < ∞, then S(ζ ) is finite a.e. on S, and
thus

(Mrad f )p (ζ )dσ (ζ ) ≤ | f (0)|p + (1 − |x|2 )p−1 |∇f (x)|p dν(x).
S B
The conclusion now follows by Theorem 8.2.2.
Remark 8.5.2 In Section 10.7 we will prove that Theorem 8.5.1 is still valid
whenever 1 < p ≤ 2.
8.6 Example
In this final section we include an example of an H-harmonic function which
fails to have a finite radial limit at every ζ ∈ S. Our proof follows the methods
used by S. Axler, P. Bourdon, and W. Ramey in proving the analogous result
for Euclidean harmonic functions on B [10, p. 160].
Lemma 8.6.1 For ζ ∈ S, m = 1, 2, . . . , let fm (ζ ) = eimζ1 . Then Ph [ fm ] → 0

uniformly on compact subsets of B as m → ∞.
Proof. For g ∈ C(S) and t ∈ (−1, 1), set

n−3
G(t) = (1 − t2 ) 2 g(t, 1 − t2 ζ )dσn−1 (ζ ).
Sn−1
8.6 Example 135
By (5.5.11),
1
gfm dσ = cn G(t)eimt dt.
S −1

Thus by the Riemann–Lebesgue lemma ([71, p. 169]), limm→∞ fm gdσ = 0.
By taking g(ζ ) = Ph (x, ζ ) we have limm→∞ Ph [ fm ](x) = 0 for all x ∈ B.
Since | fm | ≡ 1 on S, we have |Ph [ fm ](x)| ≤ 1 on B. Thus by Lemma 8.4.5
the family {Ph [ fm ]} is equicontinuous on each compact subset K of B. Hence
by the Ascoli–Arzela theorem [71, p. 169] every subsequence of {Ph [ fm ]}
contains a subsequence converging uniformly on compact subsets of B. Since
{Ph [ fm ]} converges pointwise to 0, we have Ph [ fm ] → 0 uniformly on compact
subsets of B.
Remark 8.6.2 The above sequence {Ph [ fm ]} is interesting in that each func-
tion Ph [ fm ] extends continuously to S with boundary values of modulus one
everywhere on S, yet converges uniformly to zero on compact subsets of B.
Theorem 8.6.3 Let α : [0, 1) → [1, ∞) be an increasing function with

α(r) → ∞ as r → 1. Then there exists an H-harmonic function on B such
that
(a) |u(rζ )| < α(r) for all r ∈ [0, 1) and all ζ ∈ S;
(b) at every ζ ∈ S, u(rζ ) fails to have a finite limit as r → 1.
Proof. Choose an increasing sequence {sm } ⊂ [0, 1) such that α(sm ) > m + 1.
From the sequence {Ph [ fm ]} choose a subsequence {vm } such that |vm | < 2−m
on Bsm . Suppose r ∈ [sm , sm+1 ). Since each vm satisfies |vm | < 1 on B we have
∞

m ∞

|vk (rζ )| = |vk (rζ )| + |vk (rζ )|
k=1 k=1 k=m+1
∞

<m+ 2−k
k=m+1
<m+1
< α(sm )
≤ α(r).

Thus |vk (rζ )| < α(r) for all r ∈ (0, 1) and all ζ ∈ S. Furthermore the series

|vk | converges uniformly on compact subsets of B.
From the sequence {vm } we inductively extract a subsequence {um } in the

following manner: set u1 = v1 . Because u1 is continuous on B we may choose
r1 ∈ [0, 1) such that
1
|u1 (rζ ) − u1 (ζ )| <
4
for all r ∈ [r1 , 1] and all ζ ∈ S. Suppose that we have chosen functions
u1 , u2 , . . . , um from {vm } and radii 0 < r1 < · · · < rm < 1 such that

m
1
|uk (rζ ) − uk (sζ )| < for all r, s ∈ [rm , 1], ζ ∈ S. (8.6.1)
4
k=1
We now choose um+1 such that |um+1 | < 2−(m+1) on Brm . Now choose rm+1 ∈
(rm , 1) so that (8.6.1) holds with m + 1 in place of m. The radius rm+1 can be
chosen since um+1 is continuous on B. We now define u as
∞

u(x) = um (x).
m=1
Since the series converges uniformly on compact subsets of B, u is H-harmonic

on B. Furthermore, since |u| ≤ |vk | we have |u(rζ )| < α(r) for all r ∈ [0, 1)
and ζ ∈ S.
We now show that u fails to have a finite radial limit at each ζ ∈ S. For each
ζ ∈ S and each integer m we have
|u(rm+1 ζ ) − u(rm ζ )| ≥ |um+1 (rm+1 ζ ) − um+1 (rm ζ )|

− |uk (rm+1 ζ ) − uk (rm ζ )|
k=m+1
≥ |um+1 (ζ )| − |um+1 (rm+1 ζ ) − um+1 (ζ )|

∞

1
− |um+1 (rm ζ )| − −2 2−k
4
k=m+2
1 1
≥ 1 − − 2−(m+1) − − 2 2−k
4 4
k=m+2
1
≥ −2 2−k .
2
k=m+1
In the above we have used the fact that |um+1 (ζ )| = 1 for all ζ ∈ S. Thus since
|u(rm+1 ζ ) − u(rm ζ )| is bounded away from zero, the sequence {u(rm ζ )} fails
to have a finite limit as m → ∞. Hence u(rζ ) also fails to have a finite limit as
r → 1.
8.7 Exercises 137
8.7 Exercises
8.7.1. (a) Using the invariance of σ and (5.5.11) show that

n−3
σ (S(ζ , δ)) = cn (1 − s2 ) 2 ds
Q(δ)
where Q(δ) = {s : s < 1 and 1 − s < 12 δ 2 }.

(b) Using the above show that there exist positive constants c1 and c2
such that
c1 δ n−1 ≤ σ (S(ζ , δ)) ≤ c2 δ n−1
√
for all δ, 0 < δ < 2.
8.7.2. Let (X, μ) be a measure space. If f is a measurable function on X, define
the distribution function λ by
λ(t) = μ({x ∈ X : | f (x)| > t}).
Prove that for f ∈ L p (X, μ), 1 < p < ∞,
∞
| f | dμ = p
p
t p−1 λ(t)dt.
X 0
Hint: First prove the result for the case where s is a non-negative simple
function given by
m
s(x) = αj χEj (x),
j=1
where the {Ej } are pairwise disjoint in X, and where without loss of
generality αi < αi+1 for all i, i = 1, . . . , m − 1.
8.7.3. Let (X, μ) be a measure space and let {Ak } be a sequence of measurable
sets such that {A4k+i }∞
k=0 are pairwise disjoint for each i = 0, 1, 2, 3.
Prove that for f ∈ L1 (μ),
∞
| f | dμ ≤ 4 | f | dμ.
k=0 Ak +∞
k=0 Ak
8.7.4. For β > 1, x ∈ B, let

β (x) = {ζ ∈ S : x ∈ β (ζ )}. Prove that there
exists a constant c > 0, depending only on β such that if x = |x|η,
η ∈ S,

β (x) ⊂ S(η, c(1 − |x|2 )).
8.7.5. Weighted radial limits of Poisson integrals. Let ν be a non-negative
singular measure on S. Prove that
lim (1 − r)n−1 Ph [ν](rζ ) = 2n−1 ν({ζ }) for every ζ ∈ S.
r→1
The following exercise is a generalization of an inequality due to Hardy

and Littlewood [35] for n = 2. A statement of the inequality, along with
a proof using derivatives, may be found in [99, p. 101].
8.7.6. For U ∈ L1 (S), U ≥ 0, let

1
U ∗ (ζ ) = sup n−1 U(t)dσ (t).
0<ρ<1 ρ S(ζ ,ρ)
If U(x) = Ph [U](x), x ∈ B, prove that there exists a constant C,

independent of U, such that

∗ |η − ζ | n−1
U(rη) ≤ CU (ζ ) 1 + .
1−r
Hint: Consider the two cases |η − ζ | > c(1 − r) and |η − ζ | ≤ c(1 − r),
where c > 0 is a constant.
8.7.7. For α > 12 and ζ ∈ S, let
Aα (ζ ) = {x ∈ B : |x − ζ | < α(1 − |x|2 )}.
If U is a continuous function on B, prove that U has a non-tangential
limit L at ζ ∈ S if and only if
lim U(x) = L.
x→ζ
x∈Aα (ζ )

8.7.8. For α > 0 and a ∈ Rn−1 , set
α (a) = {(x, y) ∈ H : |x − a| < αy}.
A function V on H has a non-tangential limit at a if and only if
lim V(x) exists.
x→a
x∈α (a)
Prove that V has a non-tangential limit at a ∈ Rn−1 if and only if U(x) =

V((x)) has a non-tangential limit at (a).
9
The Riesz Decomposition Theorem for

H-Subharmonic Functions
One version of the classical Riesz decomposition theorem for subharmonic

functions on the unit ball B is as follows: if f ∈ C2 (B) ∩ C(B) is subharmonic
on B, then

f (x) = H(x) − Ge (x, y)f (y)dν(y),
B
where H is harmonic on B and Ge is the Euclidean Green’s function on B. For
the invariant Laplacian Δh , by Theorem 4.1.1 we have

f (x) = ( f ◦ ϕx )(0) = f (ϕx (rt))dσ (t) − g(|y|, r)Δh ( f ◦ ϕx )(y)dτ (y),
S B
where

1 r (1 − s2 )n−2
g(|y|, r) = ds.
n |y| sn−1
But by Theorem 5.3.5, if f ∈ C(B),

lim f (ϕx (rt))dσ (t) = Ph (x, t)f (t)dσ (t).
r→1 S S
Also, limr→1 g(|y|, r) = gh (y), the Green’s function on B. Hence if f ∈
C2 (B) ∩ C(B), then

f (x) = Ph (x, t)f (t)dσ (t) − gh (y)Δh ( f ◦ ϕx )(y)dτ (y),
S B

= Ph (x, t)f (t)dσ (t) − Gh (x, y)Δh f (y)dτ (y).
S B
In particular, the above is valid if f is H-subharmonic on B, in which case Δh f
is non-negative on B.
139
140 The Riesz Decomposition Theorem for H-Subharmonic Functions
In Section 9.1 we extend the above to H-subharmonic functions on B

without the hypothesis that f ∈ C2 (B) ∩ C(B). Our proof follows the
methods used by D. Ullrich in his dissertation [93], and published in [94],
for the analogous results for M-subharmonic functions on B ⊂ Cn . Since
the results for n = 2 are well known and can be found elsewhere, we will
assume throughout this chapter that n ≥ 3. In Section 9.3 we extend a
result of D. H. Armitage [8] concerning the integrability of superharmonic
functions to H-superharmonic functions on B, and in Sections 9.4 and 9.5 we
investigate boundary limits of Green potentials and non-tangential limits of
H-subharmonic functions.
9.1 The Riesz Decomposition Theorem

As a consequence of the above (see also Corollary 4.1.5) if f is C2 in B with
compact support, then

f (a) = − Gh (a, x)Δh f (x)dτ (x). (9.1.1)
B
In our next theorem we prove the following generalization of the above.
Theorem 9.1.1 Suppose f ≤ 0 is H-subharmonic on B satisfying

lim f (rt) dσ (t) = 0. (9.1.2)
r→1 S
Then for all x ∈ B,

f (x) = − Gh (x, y)dμf (y),
B
where μf is the Riesz measure of f .
Lemma 9.1.2 Suppose χ is a C2 function with compact support which
satisfies B χ dτ = 0. Let v = −gh ∗ χ . Then
v ∈ Cc2 (B) and Δh v = χ .
Proof. In the above, gh is the Green’s function on B given in (3.2.1) and ∗ is
the invariant convolution on B.
For fixed x, the function y → Gh (x, y) is H-harmonic in B(0, |x|). Thus for
all ρ, 0 < ρ < |x|,

Gh (x, ρt)dσ (t) = Gh (x, 0) = gh (x).
S
9.1 The Riesz Decomposition Theorem 141
Choose r, 0 < r < 1, such that χ has support in Br . Then for all x, |x| > r,

(gh ∗ χ )(x) = Gh (x, y)χ (y)dτ (y)
B
r
2 −n
=n ρ (1 − ρ ) χ (ρ) Gh (x, ρt)dσ (t)dρ
n−1
S
0

= gh (x) χ (y)dτ (y) = 0.
Br
Thus supp v ⊂ B|x| .

Let ψ ∈ Cc2 (B) be arbitrary. Then by Green’s formula (see (5.1.1)) and
Fubini’s theorem,

ψ(y)Δh v(y)dτ (y) = v(y)Δh ψ(y)dτ (y)
B B

=− Gh (x, y)χ (x)dτ (x) Δh ψ(y)dτ (y)
B B

= − χ (x) Gh (x, y)Δh ψ(y)dτ (y) dτ (x),
B B
which by (9.1.1)

= χ (x)ψ(x)dτ (x).
B
Therefore
ψ(x)Δh v(x)dτ (x) = ψ(x)χ (x)dτ (x).
B B
Since v is C2 and the above holds for all ψ ∈ Cc2 (B) we have Δh v = χ .
Proof of Theorem 9.1.1 Choose a sequence {rj }, 0 < rj < 12 , which decreases
to 0. Let
A1j = {x : rj+1 ≤ |x| ≤ rj } and A2j = {x : 1 − rj ≤ |x| ≤ 1 − rj+1 }.
For each j = 1, 2, . . . and k = 1, 2, let χjk ≥ 0 be a C2 radial function with

supp χjk ⊂ Akj and χjk dτ = 1.
Since f is H-subharmonic, by Theorem 4.5.4,

f (0) = lim ( f ∗ χj )(0) = lim
1
f (x)χj1 (x)dτ (x).
j→∞ j→∞ B
Also, for each j, if ρj = 1 − rj ,

f (ρj t)dσ (t) ≤ f (x)χj (x)dτ (x) ≤ f (ρj+1 t)dσ (t).
2
S B S
Hence by our hypothesis on f ,

lim f (x)χj2 (x)dτ (x) = 0.
j→∞ B
Hence if χj = χj1 − χj2 ,

lim f (x)χj (x)dτ (x) = f (0).
j→∞ B
Furthermore, since the Green’s function gh is H-superharmonic on B, gh ∗ χj1

increases to gh . Also, a similar argument as in Theorem 4.5.4 shows that gh ∗χj2
decreases to zero. Thus gh ∗ χj increases to gh .
For each j let vj = −(gh ∗ χj ). By Lemma 9.1.2, vj ∈ Cc2 (B) with Δh vj = χj .
Thus

f (0) = lim f (x)χj (x)dτ (x)
j→∞ B

= lim f (x)Δh vj (x)dτ (x),
j→∞ B

= lim vj (x)dμf (x),
j→∞ B
where μf is the Riesz measure of f . Therefore,

f (0) = − lim (gh ∗ χj )(x)dμf (x),
j→∞ B
which by the monotone convergence theorem

= − gh (x)dμf (x).
B
Let x ∈ B be arbitrary and let h = f ◦ ϕx . By Corollary 7.3.5 the least
H-harmonic majorant of f , and hence also of h, is the zero function. Thus h
satisfies (9.1.1), and therefore

f (x) = h(0) = − gh (y)dμh (y)
B
where μh is the Riesz measure of h. But if ψ ∈ Cc2 (B), by the invariance of Δh

and τ ,

ψdμh = ( f ◦ ϕx )Δh ψdτ
B
B
= f Δh (ψ ◦ ϕx )dτ = (ψ ◦ ϕx )dμf .
B B
As a consequence,

ψdμh = (ψ ◦ ϕx )dμf
B B
for any non-negative Borel measurable function ψ. Therefore

f (x) = − gh (ϕx (y))dμf (y) = − Gh (x, y)dμf (y),
B B
Theorem 9.1.3 (Riesz Decomposition Theorem) Suppose f is H-sub-

harmonic on B and f has an H-harmonic majorant. Then

f (x) = Ff (x) − Gh (x, y)dμf (y),
B
where μf is the Riesz measure of f and Ff is the least H-harmonic majorant

of f .
Proof. Let Ff be the least H-harmonic majorant of f and let h(x) = f (x) −
Ff (x). Then h satisfies (9.1.2). Furthermore, since Ff is H-harmonic, μh = μf ,
9.2 Applications of the Riesz Decomposition Theorem

In this section we prove several consequences of the Riesz decomposition
theorem. We begin with the following lemma, which in itself is an application
of Theorem 9.1.3.
Lemma 9.2.1 For all y ∈ B,

1
Gh (x, y)dν(x) = (1 − |y|2 )n−1 .
B 2n(n − 1)
Proof. Set ψ(x) = 1 − (1 − |x|2 )n−1 . Using the radial form of Δh we have
Δh ψ(x) = 2n(n − 1)(1 − |x|2 )n .
Therefore,

Gh (x, y)Δh ψ(x)dτ (x) = 2n(n − 1) Gh (x, y)dν(x).
B B
But ψ is a non-negative C2 H-subharmonic function with least H-harmonic

majorant Fψ (x) = 1. Thus by Theorem 9.1.3,

Gh (x, y)Δh ψ(x)dτ (x) = Fψ (y) − ψ(y) = (1 − |y|2 )n−1 ,
B
Theorem 9.2.2 Let f be a non-negative H-subharmonic function with least
H-harmonic majorant Ff and Riesz measure μf . Then

1
Ff (0) = f (y) dν(y) + (1 − |y|2 )n−1 dμf (y).
B 2n(n − 1) B
Proof. Since Ff is H-harmonic,

Ff (0) = Ff (rt)dσ (t).
S
Multiplying the above by nrn−1 and integrating from 0 to 1 gives

Ff (0) = Ff (x)dν(x),
B
which by Theorem 9.1.3 and Fubini’s theorem

= f (x)dν(x) + Gh (x, y)dν(x) dμf (y).
B B B
The result now follows by Lemma 9.2.1
Theorem 9.2.3 Let f ≡ −∞ be an H-subharmonic function on B with Riesz
measure μf . Then the following are equivalent.
(a) f has an H-harmonic majorant on B.
(b) Gh (yo , x)dμf (x) < ∞ for some yo ∈ B.
B
(c) (1 − |x|2 )n−1 dμf (x) < ∞.
B
Proof. If (a) holds, then by the Riesz decomposition theorem

f (y) = Hf (y) − G(y, x)dμf (x).
B

By Theorem 4.4.3 f (y) > −∞ a.e. on B. Thus B Gh (y, x)dμf (x) < ∞ for
almost every y ∈ B.
Suppose now that (b) holds for some yo ∈ B. By (3.2.4),
(1 − |yo |2 )n−1 (1 − |x|2 )n−1
Gh (yo , x) ≥ cn (1 − |ϕyo (x)|)n−1 = cn .
ρ(yo , x)
But by Exercise 2.4.1, ρ(yo , x) ≤ (1 + |yo ||x|)2 ≤ 4. Therefore
Gh (yo , x) ≥ cn (1 − |x|2 )n−1 (1 − |yo |2 )n−1 ,
from which (c) follows.

Finally, suppose that (c) holds. Fix c, 0 < c < 1, and define V1 and V2 on
B by

V1 (x) = Gh (y, x)dμf (x),
Bc

V2 (x) = Gh (y, x)dμf (x).
B\Bc
Since μf |B is a finite measure on B, V1 is H-superharmonic on B. Also, by

c
(3.2.3),
1
Gh (0, x) ≤ n−2 (1 − |x|2 )n−1 for all x ∈ B \ Bc .
nc
Therefore, V2 (0) < ∞ and thus V2 is H-superharmonic on B. Hence Vf =
V1 + V2 is H-superharmonic on B.
Set h(x) = f (x) + Vf (x), which is defined
a.e. on B and is locally integrable.
Let ψ ∈ Cc2 (B), ψ ≥ 0, and consider hΔh ψdτ . By (9.1.1)

Gh (y, x)Δh ψ(x)dτ (x) = −ψ(y).
B
Thus by Fubini’s theorem and the definition of μf ,

Vf Δh ψdτ = − ψdμf = − f Δh ψdτ .
B B B
Therefore,

h(x)Δh ψ(x)dτ (x) = 0
B
for all ψ ∈ Cc2 (B) with ψ ≥ 0. Thus by Theorem 4.6.2 there exists an
H-harmonic function H such that h(x) = H(x) a.e. on B. Since Vf ≥ 0 we
have f (x) ≤ H(x) a.e. However, as a consequence of Theorem 4.3.5, if a ∈ B
and r > 0 is sufficiently small,

1
f (a) ≤ f (x)dτ (x)
τ (E(a, r)) E(a,r)

1
≤ H(x)dτ (x) = H(a).
τ (E(a, r)) E(a,r)
Therefore f (a) ≤ H(a) for all a ∈ B, which proves the result.
Theorem 9.2.4 Let f be a non-negative

H-subharmonic function on B. Then
(a) f ∈ S p , p ≥ 1, if and only if (1 − |y|2 )n−1 dμf p (y) < ∞, where μf p is
B
the Riesz measure of f p .
(b) If this is the case, then

f pp = f p (0) + gh (y)dμf p (y). (9.2.1)
B
Proof. The result (a) is just a restatement of Theorem 9.2.3.

(b) Suppose f ∈ S p , p ≥ 1. Then the least H-harmonic majorant Hf p of f p
is given by

Hf (x) = Ph (x, t)fˆp (t)dσ (t), p > 1,
p
S
where fˆ ∈ L p (S) is the boundary function of f with f p = f̂ p . When p = 1,

Hf is the Poisson integral of a measure νf where νf is the boundary measure
of f . In this case, f 1 = νf (S) = Hf (0). By the Riesz decomposition theorem,

Hf (x) = f (x) + Gh (x, y)dμf p (y).
p
p
(9.2.2)
B
Hence

f pp = Hf p (0) = f (0) +
p
gh (y)dμf p (y) (9.2.3)
B

Remarks 9.2.5 (a) Equation (9.2.1) is sometimes referred to as the Hardy–
Stein identity.
(b) Although we assumed p ≥ 1 in the previous theorem, the conclu-
sion is still valid for 0 < p < 1 provided we assume in addition that f p is
H-subharmonic. In this case f p (x) ≤ Ph [νf p ](x) for some non-negative
p
measure νf p on S with f p = νf p (S) = Hf p (0).
(c) Since gh (y) ≥ cn (1 − |y|2 )n−1 we have

f p ≥ f (0) + cn (1 − |y|2 )n−1 dμf p (y).
p p
B
In the other direction, integrating (9.2.2) over Bro , 0 < ro < 12 , gives

1
Hf p (0) = n f (x)dν(x) +
p
Gh (x, y)dν(x)dμf p (y) ,
ro Bro B Bro
which by Lemma 9.2.1

1
≤ n f (x) + Cn (1 − |y| ) dμf p (y) .
p 2 n−1
ro Bro B
Suppose in addition that f p is C2 and Δh f p has quasi-nearly H-subharmonic

behavior. Then by Theorem 4.1.1, for 0 < r < ro ,

f p (rζ )dσ (ζ ) ≤ f p (0) + Cn ro2 sup Δh f p (x).
S x∈Bro

f pp ≤ f p (0) + Cn,ro (1 − |y|2 )n−1 Δh f p (y)dτ (y).
B
Corollary 9.2.6 Let h be H-harmonic on B. If p ≥ 2, or if h is non-zero when

1 < p < 2, then h ∈ Hp if and only if

(1 − |x|2 )n−1 |h(x)|p−2 |∇ h h(x)|2 dτ (x) < ∞.
B
Proof. If h satisfies the hypothesis, then |h|p ∈ C2 (B), 1 < p < ∞, with
Δh |h|p = p(p − 1)|h|p−2 |∇ h h|2 .
The conclusion now follows by the previous theorem.
A similar result also holds for Hϕ , where ϕ is strongly convex on (−∞, ∞).
In this case, if h is non-zero on B and ϕ is C2 , then by Exercise 3.5.3,
Δh ϕ(|h|) = ϕ (|h|)|∇ h h|2 .
Definition 9.2.7 If μ is a regular Borel measure on B, the function Gμ

defined by

Gμ (x) = Gh (x, y) dμ(y)
B
is called the (invariant) Green potential of μ provided Gμ ≡ +∞.
If Gμ ≡ +∞, then since x → Gh (x, y) is H-superharmonic, by Tonelli’s

theorem so is the function Gμ . Furthermore, by Exercise 4.8.7, Gμ is
H-harmonic on B \ supp μ. Also, as in the proof of Theorem 9.2.3 it is
easy to show that Gμ ≡ +∞ if and only if

(1 − |x|2 )n−1 dμ(x) < ∞. (9.2.4)
B
Theorem 9.2.8 A non-negative H-superharmonic function V is a Green

potential on B if and only if

lim V(rt)dσ (t) = 0. (9.2.5)
r→1 S
Proof. If V satisfies (9.2.5), then by the Riesz decomposition theorem the

function V is the Green potential of its Riesz measure.
Conversely, suppose V = Gμ where μ satisfies (9.2.4). Without loss of
generality we assume n > 2. Fix δ, 0 < δ < 12 . By inequality (3.2.3),
Gh (0, x) ≤ Cδ (1 − |x|2 )n−1
for all x, |x| ≥ δ. Therefore, since V(x) < ∞ a.e., by (9.2.4)

Gh (0, x)dμ(x) < ∞.
B\Bδ
Let > 0 be given. Choose R, δ < R < 1, such that

Gh (0, x)dμ(x) < .
B\BR
For this R write

V(y) = Gh (y, x)dμ(x) + Gh (y, x)dμ(x)
BR B\BR
= V1 (y) + V2 (y).
Since y → Gh (y, x) is H-superharmonic, by Tonelli’s theorem

V2 (rt)dσ (t) = Gh (rt, x)dσ (t)dμ(x)
S B\B S
R
≤ Gh (0, x)dμ(x) < .
B\BR
Therefore

lim sup V2 (rt)dσ (t) < .
r→1 S
Consider V1 (rt). By (3.2.5),
(1 − r2 )n−1 (1 − |x|2 )n−1

Gh (rt, x) ≤
n|rt − x|n−2 {|rt − x|2 + (1 − r2 )(1 − |x|2 )}n/2
(1 − r2 )n−1 (1 − |x|2 )n−1
≤ .
n|rt − x|2n−2
But for x ∈ BR and (1 + R)/2 < r < 1, |rt − x| ≥ r − |x| > (1 − R)/2. Hence
2n−2
2 1
Gh (rt, x) ≤ (1 − r2 )n−1 (1 − |x|2 )n−1 ,
1−R n
and thus

V1 (rt) ≤ CR (1 − r ) 2 n−1
(1 − |x|2 )n−1 dμ(x).
BR
Therefore limr→1 V1 (rt) = 0 uniformly on B and thus

lim sup V(rt)dσ (t) < ,
r→1 S
9.3 Integrability of H-Superharmonic Functions

In [8] D. H. Armitage proved that if V(x) = Geμ (x), where Ge is the Euclidean
Green’s function on B and μ is a measure on B satisfying

(1 − |x|)dμ(x) < ∞.
B
Then V ∈ L p (B, ν) for all p, 0 < p < n/(n − 1), with

V p (x)dx ≤ A(n, p)V p (0).
B
The analogous result for M-superharmonic functions on the unit ball B in Cm

was proved by S. Zhao in [96]. The result of Zhao states that if V is a non-
negative M-superharmonic function on B ⊂ Cm = R2m , then V ∈ L p (B, ν)
for all p, 0 1, the constant (m + 1)/m
is strictly greater than 2m/(2m − 1). Of course, when m = 1, the two upper
bounds agree.
In this section we prove the analogous result for non-negative H-
superharmonic functions on B ⊂ Rn . As we will see, our upper bounds
are similar to those obtained for Euclidean superharmonic functions. Our main
result of this section is the following analogue of Theorem 3 of D. H. Armitage
[8]. Our method of proof, with minor modifications, is similar to that used by
S. Zhao in [96].
Theorem 9.3.1 Let U be a non-negative H-superharmonic function on B.

Then for each a ∈ B, there exists a constant A(n, p, a), independent of U,
such that

U p (x)dν(x) ≤ A(n, p, a)U p (a)
B
for all p, 0 < p < n/(n − 1).
We begin by proving several results concerning integrability of non-negative

H-harmonic functions.
Lemma 9.3.2 For all ζ ∈ S, there exists a finite constant A(n, p), such that

p
Ph (x, ζ )dν(x) ≤ A(n, p)
B
for all p, 0 < p < n/(n − 1). Furthermore, for p = n/(n − 1)

p
Ph (x, ζ )dν(x) = +∞.
B
In the above, ν denotes normalized Lebesgue measure on B.

p
Proof. Since x → Ph (x, ζ ) is H-superharmonic on B for all p, 0 1. Hence
1
p dσ (t)
Ph (x, ζ )dν(x) = n rn−1 (1 − r2 )p(n−1) dr,
S |rt − ζ |
2p(n−1)
B 0
which by Theorem 5.5.7 for p > 1

1
≈ rn−1 (1 − r2 )p(n−1) (1 − r)−2p(n−1)+(n−1) dr
0
1
≈ rn−1 (1 − r)(n−1)−p(n−1) dr.
0
However, the above integral is finite if and only if p < n/(n − 1).
Theorem 9.3.3 Let h(x) be a non-negative H-harmonic function on B. Then

hp (x)dν(x) ≤ A(n, p)hp (0)
B
for all p, 0 1. By Theorem 7.1.1, h(x) = Ph [μ](x) where μ
is a non-negative measure on S. Therefore
p
h (x)dν(x) =
p
Ph (x, ζ )dμ(ζ ) dν(x),
B B S
which by Hölder’s inequality and Lemma 9.3.2

p−1
p
≤ Ph (x, ζ )dμ(ζ ) dμ(ζ ) dν(x)
B S S

p
= hp−1 (0) Ph (x, ζ )dν(x) dμ(ζ ),
S B
which for p < n/(n − 1)

≤ A(n, p)hp−1 (0) dμ(ζ ) = A(n, p)hp (0).
S

We next turn our attention to the integrability of Green potentials. For the
proofs we require the following lemma.
Lemma 9.3.4 Fix δ, 0 < δ < 1. Then
⎧
⎪
⎪ (1 − |x|2 )n−1
⎨Cδ Gh (0, y) for all x ∈ B \ E(y, δ),
ρ(x, y)n−1
Gh (x, y) ≤
⎪
⎪ 1
⎩Cn Gh (0, y) for all x ∈ B.
|x − y|n−1
Proof. For x ∈ B \ E(y, δ), by (3.2.4)
1 (1 − |y|2 )n−1 (1 − |x|2 )n−1
Gh (x, y) ≤ .
nδ n−2 ρ(x, y)n−1
But by (3.2.3) (1 − |y|2 )n−1 ≤ cn Gh (0, y), from which the first inequality
follows. On the other hand, by (3.2.5),
(1 − |y|2 )n−1 (1 − |x|2 )n−1
Gh (x, y) ≤ n .
n|x − y|n−2 ρ(x, y) 2
But
n n−1 1 n−1
ρ(x, y) 2 = ρ(x, y) 2 ρ(x, y) 2 ≥ |x − y|ρ(x, y) 2 .
Also
ρ(x, y) ≥ ||x| − |y||2 + (1 − |x|2 )(1 − |y|2 ) = (1 − |x||y|)2 ≥ (1 − |x|)2 .
Therefore
n−1
ρ(x, y) 2 ≥ (1 − |x|)n−1 ,
and thus
Gh (x, y) ≤ Cn Gh (0, y)|x − y|−(n−1) .

Theorem 9.3.5 For all p, 0 < p < n/(n − 1),

p
Gh (x, y)dν(x) ≤ A(n, p)Gp (0, y).
B
Proof. Suppose first that 0 1. Set E(y) = E(y, δ) where 0 < δ < 12 , and write

p p p
Gh (x, y)dν(x) = Gh (x, y)dν(x) + Gh (x, y)dν(x).
B E(y) B\E(y)
By Lemma 9.3.4

p p
1 (1 − r2 )p(n−1)
Gh (x, y)dν(x) ≤ Cδ,n Gh (0, y) rn−1 dσ (t) dr.
B\E(y) 0 S ρ(rt, y)p(n−1)
Write y = sη, η ∈ S. Since
ρ(rt, sη) = ρ(t, rsη) = |t − rsη|2 ,
we have by Theorem 5.5.7

dσ (t) dσ (t)
= ≤ C(1 − rs)−2p(n−1)+(n−1)
S ρ(rt, y) p(n−1)
S |t − rsη| 2p(n−1)
≤ C(1 − r)−2p(n−1)+(n−1) .
Therefore
1
p p
Gh (x, y)dν(x) ≤ Cδ,n Gh (0, y) rn−1 (1 − r)(n−1)−p(n−1) dr.
B\E(y) 0
The above integral, however, is finite if and only if p < n/(n − 1).
Suppose x ∈ E(y). Since E(y, δ) is also a Euclidean ball with center cy and
radius ρy given by
(1 − δ 2 )y δ(1 − |y|2 )
cy = and ρy = ,
1 − |y|2 δ 2 1 − |y|2 δ 2
1
we have for 0 < δ < 2 that
|y| + δ
|x| < |cy | + |x − cy | = .
1 + |y|δ
Therefore E(y, δ) ⊂ B(0, ρ) with ρ = (|y| + δ)/(1 + |y|δ). Hence by Lemma

9.3.4
p p dν(x)
Gh (x, y)dν(x) ≤ C Gh (0, y) .
E(y) B(0,ρ) |x − y|p(n−1)
But by the change of variables w = x − y and integration in polar coordinates,
2
dν(x) dν(w)
≤ =n rn−1−p(n−1) ,
B(0,ρ) |x − y| B(0,2) |w|
p(n−1) p(n−1)
0
which is finite provided p < n/(n − 1).
As an immediate consequence of the previous theorem, we have the
following corollary.
Corollary 9.3.6 Let f ∈ Lq (B, ν) for some q > n, and let

Vf (x) = f (y)Gh (x, y)dν(y). (9.3.1)
B
Then
lim Vf (x) = 0.
|x|→1
Proof. Let p denote the conjugate exponent of n. Since q > n we have 1 <
p < n/(n − 1). Hence by Hölder’s inequality,
1/p
p
|Vf (x)| ≤ f q Gh (x, y)dν(y)
B
≤ A(n, p) f q Gh (0, x),
Theorem 9.3.7 Let Gμ be the potential of a measure μ satisfying (9.2.4). Then

Gpμ (y)dν(y) ≤ A(n, p)Gpμ (0)
B
for all p, 0 < p < n/(n − 1).
Proof. The theorem is certainly true if Gμ (0) = +∞. Hence, assume that
Gμ (0) < ∞. For 0 1, by the continuous version of Minowski’s inequality [18,

VI.11.13],
p
Gμ (x)dν(x) =
p
Gh (x, y)dμ(y) dν(x),
B
B B
p
≤ Gh (x, y)dμ(x) dν(y),
B B
which by the previous theorem

p
≤ A(n, p) Gh (0, y)dμ(y) = A(n, p)Gpμ (0).
B
Proof of Theorem 9.3.1 Fix a ∈ B and let V = U ◦ ϕa . Then V is

H-superharmonic and

V (x)dν(x) = (1 − |x|2 )n U p (ϕa (x)) dτ (x),
p
B B

= (1 − |ϕa (x)|2 )n U p (x) dτ (x).
B
But
(1 − |x|2 )(1 − |a|2 ) (1 − |x|2 )(1 − |a|2 )
1 − |ϕa (x)|2 = ≥ .
ρ(x, a) (1 + |a|)2
Hence
n
1 + |a|
U p (x)dν(x) ≤ V p (x)dν(x).
B 1 − |a| B
Since V is a non-negative H-superharmonic function, by the Riesz decom-

position theorem
V(x) = h(x) + Gμ (x),
where h is H-harmonic and Gμ is the Green potential of a measure μ. Thus by

Theorems 9.3.3 and 9.3.7

. /
V p (x)dν(x) ≤ A(n, p) hp (0) + Gpμ (0)
B
≤ A(n, p)V p (0) = A(n, p)U p (a).

9.4 Boundary Limits of Green Potentials

In Corollary 9.3.6 we proved that if f ∈ Lq (B, ν) for some q > n, then
lim Gf (x) = 0,
|x|→1
where Gf was given by (9.3.1). Also, it is easily shown that if μ has compact
support K, then

Gμ (x) ≤ CK (1 − |x| )
2 n−1
(1 − |y|2 )n−1 dμ(y),
K
and thus lim|x|→1 Gμ (x) = 0 uniformly.
Our main result of this section is the following theorem concerning radial
limits of Green potentials.
Theorem 9.4.1 Let Gμ be the Green potential of a measure μ satisfying
(9.2.4). Then
lim Gμ (rt) = 0 for almost every t ∈ S.
r→1
We will also give an example of a measure μ satisfying (9.2.4) for which
lim Gμ (x) = +∞ for almost every t ∈ S
x→ζ
x∈α (ζ )
for every α > 1.
Our method of proof of Theorem 9.4.1 will follow the technique used by
D. Ullrich in proving the analogous result for invariant potentials on the unit
ball in Cm [93], [94]. (See also [84, Theorem 8.1].) For a regular Borel measure
μ on B satisfying (9.2.4) define the functions V1 and V2 on B by

V1 (x) = Gh (x, y)dμ(y), and
E(x)

V2 (x) = Gh (x, y)dμ(y),
B\E(x)
where E(x) = E(x, 12 ). In our first theorem we prove that the function V2 has
non-tangential limit zero at almost every point of S.
Theorem 9.4.2 Let μ and V2 be as above. Then
lim V2 (x) = 0 for almost every ζ ∈ S.
x→ζ
x∈α (ζ )
Proof. For y ∈ B \ E(x) we have

(1 − |x|2 )n−1 (1 − |y|2 )n−1
V2 (x) ≤ dμ(y).
B ρ(x, y)n−1
Fix an R, 0 < R < 1. Since
ρ(x, y) = |x − y|2 + (1 − |x|2 )(1 − |y|2 ) ≥ (1 − |x||y|)2 ,
for y ∈ BR we have ρ(x, y) ≥ (1 − R)2 . Therefore,

(1 − |x|2 )n−1 (1 − |y|2 )n−1 (1 − |x|2 )n−1
dμ(y) ≤ (1 − |y|2 )n−1 dμ(y)
BR ρ(x, y)n−1 (1 − R)2(n−1) BR
= CR (1 − |x|2 )n−1 .
Let AR = {y : R < |y| < 1}. Define a measure νR on S as follows: for

h ∈ C(S),

y
h(t)dνR (t) = h |y| (1 − |y|2 )n−1 dμ(y).
S AR
Then νR is a finite Borel measure on S. For y ∈ AR , write y = rη, η ∈ S. Then
ρ(x, rη) = |rx − η|2 ≥ 14 |x − η|2 .
Therefore

(1 − |x|2 )n−1 (1 − |y|2 )n−1 (1 − |x|2 )n−1
dμ(y) ≤ 2(n−1) dμ(y)
ρ(x, y) AR y
AR
x − |y|

= Cn Ph (x, t)dνR (t),
S
where Cn is a constant depending only on n. Therefore
V2 (x) ≤ CR (1 − |x|2 )n−1 + Cn Ph [νR ](x),
and thus
lim sup V2 (x) ≤ Cn lim sup Ph [νR ]x

x→ζ x→ζ
x∈α (ζ ) x∈α (ζ )
≤ sup Cn Ph [νR ](x),

x∈α (ζ )
≤ Cn Aα M[νR ](ζ ),
where Aα is a constant depending only on α. But by Theorem 8.1.2 there exists

a constant Bn , depending only n, such that
Bn
σ ({ζ ∈ S : M[νR ](ζ ) > λ}) < νR (S).
λ
Therefore, there exists a constant Cn,α , depending only on α and n, such that
Cn,α
σ (Eλ ) ≤ νR (S),
λ
where
Eλ = {ζ ∈ S : lim sup V2 (x) > λ}.
x→ζ
x∈α (ζ )
Suppose
lim sup V2 (x) > 0
x→ζ
x∈α (ζ )
on a set of positive measure E. Thus E ⊂ ∪λ>0 Eλ . Suppose there exists a

λ > 0 such that σ (Eλ ) > 0. For this λ, choose R, 0 < R < 1, such that
Cn,α νR (S) < 12 λσ (Eλ ) to obtain a contradiction.
For the proof that V1 (x) has radial limit zero at almost every point of S we
require some preliminary notation and lemmas. As in (2.2.5), for 0 < δ < 1,
x ∈ B, set
Eδ (x) = E(x, δ) = {y ∈ B : |ϕx (y)| < δ}.
Also, for x ∈ B, x = 0, let (x) denote the projection of x onto S, that is,
x
(x) = .
|x|
For δ, r ∈ (0, 1), set
Vδr (ζ ) = (Eδ (rζ )). (9.4.1)
Lemma 9.4.3 For ζ ∈ S, 0 < δ ≤ 12 , 3

4 < r < 1,
σ (Vδr (ζ )) ≈ (1 − r2 )n−1 δ n−1 .
Proof. By invariance under the orthogonal group O(n) we can without loss of
generality take ζ = e1 = (1, 0, . . . , 0). For 0 < δ ≤ 12 and ρ small, let
ρ
δ = {t : 0 < 1 − t < ρ 2 δ 2 }
ρ ρ
Nδ = {ζ ∈ S : ζ1 ∈ δ }.
We will show below that there exist positive constants c, c , independent of δ

and ρ, such that
ρ ρ
Ncδ ⊂ Vδr (e1 ) ⊂ Nc δ where ρ 2 = 1 − r2 . (9.4.2)
ρ
Assuming that (9.4.2) holds we then obtain that σ (Vδr (e1 )) ≈ σ (Nδ ). But for
n ≥ 3, by (5.5.11) and Exercise 8.7.1,

ρ n−3
σ (Nδ ) = cn (1 − s2 ) 2 ds
ρ
δ
≈δ n−1 n−1
ρ .
ρ
To prove (9.4.2) we first show that for an appropriate choice of c, if ζ ∈ Ncδ
then rζ ∈ Eδ (re1 ). By identity (2.1.11), rζ ∈ Eδ (re1 ) if and only if
(1 − r2 )2
= 1 − |ϕre1 (rζ )|2 > 1 − δ 2 ,
ρ(re1 , rζ )
or
1
|1 − r2 ζ1 |2 = ρ(re1 , rζ ) < (1 − r2 )2 .
1 − δ2
But
|1 − r2 ζ1 |2 = 1 − 2r2 ζ1 + r4 ζ12 = (r2 (1 − ζ1 ) + (1 − r2 ))2 .
ρ
Since ζ ∈ Ncδ , with ρ 2 = (1 − r2 ), we have 1 − ζ1 < c2 δ 2 (1 − r2 ). Therefore
|1 − r2 ζ12 | < (c2 δ 2 r2 (1 − r2 ) + (1 − r2 ))2

< (1 − r2 )2 (1 + c2 δ 2 )2 ,
which if 2c2 ≤ 1
< (1 − r2 )2 (1 + δ 2 + δ 4 )
1
< (1 − r2 )2 .
1 − δ2
ρ
Hence for c2 ≤ 12 we have Ncδ ⊂ Vδr (e1 ).
Suppose ζ = (ζ1 , ζ ) ∈ Vδr (e1 ). Then ζ = x/|x| for some x ∈ Eδ (re1 ) for
0 < δ ≤ 12 and r ≥ 34 . By Theorem 2.2.2 x is in the Euclidean ball B(cre1 , ρr )
where
(1 − δ 2 )re1 δ(1 − r2 )
cre1 = and ρr = .
1 − δ 2 r2 1 − δ 2 r2
Set x = (x1 , x ). Then
δ 2 (1 − r2 )2
|x − cre1 |2 = |x1 − cre1 |2 + |x |2 < .
(1 − δ 2 r2 )2
But for 0 < δ ≤ 12 , (1 − δ 2 r2 )2 ≥ 14 (1 − r2 ). Therefore
|x |2 < (2δ)2 (1 − r2 ).

Also, since |x1 − cre1 | < ρr , we have for 0 < δ ≤ 1

2 and r ≥ 3
4
r−δ
|x1 | > |cre1 | − ρr = ≥ c1 > 0.
1 − δr
Therefore we also have |x| ≥ c1 . Hence, since x = |x|ζ ,
x1 x2 |x |2
1 − ζ1 = 1 − ≤ 1 − 12 = < (c δ)2 (1 − r2 )
|x| |x| |x|2
for an appropriate choice of c . Therefore ζ ∈ Nc δ .
ρ

Definition 9.4.4 For μ a non-negative measure on B, 0 < δ < 13 , ζ ∈ S, set
Mδ μ(ζ ) = sup μ(Eδ (rζ )). (9.4.3)

0<r<1
Lemma 9.4.5 Let μ be a regular Borel measure on B satisfying (9.2.4), and

let λ > 0. Then there exists a constant C, independent of δ, λ, and μ, such that

δ n−1
σ ({ζ ∈ S : Mδ μ(ζ ) > λ}) ≤ C (1 − |x|2 )n−1 dμ(x).
λ B
Proof. Define the finite measure μ∗ on B by
dμ∗ (x) = (1 − |x|2 )n−1 dμ(x).
Fix λ > 0 and let

Eλ = {x ∈ B : μ(Eδ (x)) > λ}.
If y ∈ Eδ (x), 0 < δ < 1

3, then (1 − |y|2 ) > 1
2 (1 − |x|2 ). Therefore, for all
x ∈ Eλ ,
μ∗ (Eδ (x)) > c(1 − |x|2 )n−1 λ,
where c is a constant independent of δ. Since x → μ(Eδ (x)) is lower

semicontinuous, Eλ is an open subset of B. As in Lemma 8.1.3 there exists a
countable collection of points {xi } ⊂ Eλ such that {Eδ (xi )} is pairwise disjoint
and

Eλ ⊂ E3δ (xi ).
i
ri
Set ζi = xi /|xi | and ri = |xi |. Also, let Vi = V3δ (ζi ). Suppose Mδ μ(ζ ) > λ.
Then there exists a ρ such that ρζ ∈ Eλ . But then ρζ ∈ E3δ (xi ) for some i.
Therefore ζ ∈ Vi and thus

{ζ ∈ S : Mδ μ(ζ ) > λ} ⊂ Vi .
i
Therefore,

σ ({ζ ∈ S : Mδ μ(ζ ) > λ}) ≤ σ (Vi ) ≤ Cδ n−1 (1 − ri2 )n−1
i i
δ n−1
≤C μ∗ (Eδ (xi ))
λ

δ n−1
≤C (1 − |x|2 )n−1 dμ(x).
λ B

Lemma 9.4.6 Let μ be a non-negative regular Borel measure on B satisfying
(9.2.4), and let λ > 0. Then there exists a positive constant C, independent of
λ and μ, such that

C
σ ({ζ ∈ S : (M rad V1 )(ζ ) > λ}) ≤ (1 − |x|2 )n−1 dμ(x),
λ B
where
V1 (x) = Gh (x, y)dμ(y).
E(x, 21 )
Proof. By inequality (3.2.3),

1
V1 (ρζ ) ≤ |ϕx (ρζ )|2−n dμ(x),
n E(ρζ )
where E(ρζ ) = E(ρζ , 12 ). Set δj = 2−j/(n−2) . Then

∞

E(ρζ ) = Eδj (ρζ ) \ Eδj+1 (ρζ ).
j=n−2
If x ∈ Eδj (ρζ ) \ Eδj+1 (ρζ ), then

n−2
|ϕx (ρζ )|n−2 ≥ δj+1 = 2−(j+1) .
Therefore,
∞

V1 (ρζ ) ≤ Cn 2j μ(Eδj (ρζ ))
j=n−2

≤ Cn 2j Mδj μ(ζ ).
j=n−2
Thus

M rad V1 (ζ ) ≤ Cn 2j Mδj μ(ζ ).
j=n−2
Choose α, 0 < α < 1/(n − 2). Then

∞

M rad V1 (ζ ) ≤ Cn 2−αj 2(1+α)j Mδj μ(ζ ),
j=(n−2)

which since 2−αj < ∞
≤ Cn sup 2(1+α)j Mδj μ(ζ ).

j≥(n−2)
For λ > 0 let

Eλ = {ζ ∈ S : (M rad V1 )(ζ ) > λ}.
By the above,
∞

λ −(1+α)j
Eλ ⊂ ζ ∈ S : Mδj μ(ζ ) > 2 .
Cn
j=(n−2)
Therefore, by Lemma 9.4.5

⎛ ⎞
∞

Cn
σ (Eλ ) ≤ (1 − |x|2 )n−1 dμ(x) ⎝ 2(1+α)j δj(n−1) ⎠ .
λ B j=(n−2)
= 2−βj where
(n−1)
Since 2(1+α)j δj
n−1 1 − α(n − 2)
β= − (1 + α) = > 0,
n−2 n−2
we have

Cn
σ (Eλ ) ≤ (1 − |x|2 )n−1 dμ(x).
λ B

Proof of Theorem 9.4.1. Let μ be a non-negative regular Borel measure on

B satisfying (9.2.4) and let V1 be as defined in Lemma 9.4.6. Let > 0 be
arbitrary and let C be the constant of Lemma 9.4.6. Choose R, 0 < R < 1,
such that

C (1 − |y|2 )n−1 dμ(y) < 2 .
AR
Let μR be the measure μ restricted to AR and let

VR (x) = Gh (x, y)dμR (y).
E(x)
Suppose y ∈ E(x). Let cx and rx be the Euclidean center and radius of E(x)
as given in Theorem 2.2.2. Then
2|x| − 1
|y| ≥ |cx | − rx = .
2 − |x|
Hence for all x, 1 > |x| > (2R + 1)/(R + 2) we have y ∈ AR . Hence there
exists R , 0 < R < 1, such that E(x) ⊂ AR for all x, |x| > R . Hence by
Lemma 9.4.6,
σ ({ζ ∈ S : lim sup V1 (rζ ) > }) ≤ σ ({ζ ∈ S : (M rad VR )(ζ ) > })
r→1

C
≤ (1 − |y|2 )n−1 dμ(y) < .
AR
Since > 0 was arbitrary, the result follows.
Example 9.4.7 We now provide an example of a measure μ satisfying (9.2.4)
for which
lim Gμ (x) = +∞
x→ζ
x∈α (ζ )
for every ζ ∈ S and α > 1. Let {xj } be a countable infinite subset of B with
|xj | → 1 such that α (ζ ) contains infinitely many xj for every ζ ∈ S and
α > 1. Such a sequence can be obtained by taking a countable dense subset
{ζj } of S and a sequence {rj } increasing to 1, and setting xj = rj ζj . Now choose
cj > 0 such that
∞

cj (1 − |xj |2 )n−1 < ∞.
j=1
Define the measure μ on B by

∞

μ= cj δxj ,
j=1
where δa is point mass measure at a. Then Gμ is a potential on B satisfying

Gμ (xj ) = ∞ for all j.
9.5 Non-tangential Limits of H-Subharmonic Functions

In Corollary 9.3.6 we proved that if f ∈ Lq (B, ν) for some q > n, then
lim Vf (x) = 0,
|x|→1
where

Vf (x) = Gh (x, y)f (y)dν(y).
B
If f is a non-negative measurable function on B satisfying

(1 − |x|2 )n−1 f (x)dτ (x) < ∞, (9.5.1)
B
then the Green potential Gf of f is the H-superharmonic function on B

defined by

Gf (x) = Gh (x, y)f (y)dτ (y). (9.5.2)
B
When n = 2, M. Arsove and A. Huber [9] provided sufficient conditions

for the existence of non-tangential limits of subharmonic functions in the unit
disc. The results were subsequently extended by J. Cima and C. S. Stanton to
M-subharmonic functions on the unit ball in Cm [16]. We modify their
techniques in proving analogous results for H-subharmonic functions on B.
Theorem 9.5.1 Let f be a non-negative measurable function on B satisfying

(9.5.1). If in addition

(1 − |x|2 )n−1 f q (y)dτ (y) < ∞
B
for some q > n2 , then
lim Gf (x) = 0 for almost every ζ ∈ S.

x→ζ
x∈α (ζ )
As a consequence of the Riesz decomposition theorem and Theorem 8.3.3

we also have the following.
Corollary 9.5.2 Let f be an H-subharmonic function on B with an

H-harmonic majorant on B. If the Riesz measure μf of f is absolutely
continuous and satisfies

(1 − |x|2 )n−1 (Δh f (x))q dτ (x) < ∞
B
for some q > n2 , then f has non-tangential limits at almost every ζ ∈ S.
For a non-negative measure μ on B, α > 1, and ζ ∈ S, set
Sα∗ μ(ζ ) = μ(α (ζ )). (9.5.3)

Also, for x ∈ B, set

α (x) = {η ∈ S : x ∈ α (ζ )}. (9.5.4)
Lemma 9.5.3 For x ∈ B, set x = |x|ζ , ζ ∈ S. Then
(a)
α (x) ⊂ S(ζ , 2α(1 − |x|2 )) for all α > 1.
(b) Given α > 1, there exists c > 0 such that S(ζ , c(1 − |x|2 )) ⊂
α (x) for
all x, |x| > 12 .
Proof. For (a), if η ∈
α (x), then
|η − ζ | ≤ |η − x| + |x − ζ | < 2α(1 − |x|2 ).
For the proof of (b), suppose η ∈ S(ζ , c(1−|x|2 )) for c > 0 to be determined.
Then
|x − η| ≤ ||x|ζ − ζ | + |ζ − η|
< (1 − |x|) + c(1 − |x|2 )

1
= c+ (1 − |x|2 ),
1 + |x|
which for |x| > 1

2
≤ (c + 23 )(1 − |x|2 ).
Now given α > 1 choose c > 0 such that c + 2

3 < α.
Theorem 9.5.4 Let μ be a non-negative regular Borel measure on B. Then

(1 − |x|2 )n−1 dμ(x) < ∞ if and only if Sα∗ μ ∈ L1 (S).
B
If this is the case, then for every α > 1, μ(α (ζ )) < ∞ for almost every
ζ ∈ S, and thus
lim μ(α (ζ ) ∩ AR ) = 0 for almost every ζ ∈ S.
R→1
Proof. By Tonelli’s theorem,

∗
Sα (ζ )dσ (ζ ) = χα (ζ ) (x)dμ(x)dσ (ζ )
S
S B
= χα (x) (ζ )dσ (ζ )dμ(x)
B S
= σ (
α (x))dμ(x).
B
By Lemma 9.5.3, σ ( α (x)) ≤ σ (S( |x|

x
, 2α(1 − |x|2 )) ≤ Cα (1 − |x|2 )n−1 , and
thus Sα∗ μ ∈ L1 (S) whenever μ satisfies (9.2.4).
Conversely, since μ is regular, μ satisfies (9.2.4) if and only if

(1 − |x|2 )n−1 dμ(x) < ∞.
A1
2
Thus by Lemma 9.5.3,

(1 − |x|2 )n−1 dμ(x) ≤ C σ (
α (x))dμ(x) ≤ C Sα∗ (ζ )dσ (ζ ).
A1 A1 S
2 2
The second part of the theorem follows immediately from the fact that
Sα∗ μ ∈ L1 (S).
Proof of Theorem 9.5.1. Suppose f is a non-negative measurable function

satisfying the hypothesis of Theorem 9.5.1. Set Gf = V1 + V2 , where

V1 (x) = Gh (x, y)f (y)dτ (y) and V2 (x) = Gh (x, y)f (y)dτ (y).
E(x) B\E(x)
Since dμ = fdτ satisfies (9.2.4), by Theorem 9.4.2 the function V2 (x) has
non-tangential limits at almost every ζ ∈ S.
Consider the function

V1 (x) = gh (ϕx (y))f (y)dτ (y),
E(x)
which by the change of variable w = ϕx (y) and the invariance of τ

= gh (w)f (ϕx (w))dτ (w).
B1
2
Let p denote the conjugate exponent of q. Then by Hölder’s inequality

⎡ ⎤1/p ⎡ ⎤1/q

V1 (x) ≤ ⎣ gh (w)dτ (w)⎦ ⎣ f q (ϕx (w))dτ (w)⎦
p
.
B1 B1
2 2
By inequality (3.2.3) and integration in polar coordinates

1
p 2
gh (w)dτ (w) ≤ C rn−1−p(n−2) dr.
B1 0
2
The above integral, however, is finite if and only if p < n/(n − 2), that is,
q > n/2. Thus for q > n/2,
⎡ ⎤1/q
1/q
V1 (x) ≤ C ⎣ f q (ϕx (w))dτ (w)⎦ = C f q (y)dτ (y) .
B1 E(x)
2
Suppose x ∈ α (ζ ), α > 1. If y ∈ E(x), then |y| ≥ (2|x| − 1)/(2 − |x|).

Hence, as in the proof of Theorem 9.4.1, given R, 0 < R < 1, E(x) ⊂ AR for
all x, |x| > (2R + 1)/(R + 2). Let cx and rx denote the Euclidean center and
radius of E(x). Then for y ∈ E(x),
|y − ζ | ≤ |y − cx | + |cx − x| + |x − ζ |
≤ 2rx + α(1 − |x|)
≤ 43 (1 − |x|2 ) + α(1 − |x|) ≤ (α + 83 )(1 − |x|).
Hence, given 0 < R < 1, α > 1, there exists R , 0 < R < 1, and βα > 1 such
that E(x) ⊂ βα (ζ ) ∩ AR . As a consequence,
1/q
V1 (x) ≤ C q
f (y)dτ (y)
βα (ζ )∩AR
for all x ∈ α (ζ ), |x| > R . The result now follows by Theorem 9.5.4 with
dμ = f q dτ .
Our final theorem of this section, although valid for all n ≥ 2, is of particular
interest when n = 2.
Theorem 9.5.5 Let {aj } be a sequence in B satisfying
∞

(1 − |aj |2 )n−1 < ∞, (9.5.5)
j=1
∞
and let μ = j=1 δaj , where δa denotes point mass measure at a. Then for
every α > 1,
lim Gμ (x) = 0 for almost every ζ ∈ S.
x→ζ
x∈α (ζ )
Proof. As a consequence of Theorem 9.4.2, we again only need to consider

the function
V1 (x) = Gh (x, y)dμ(y).
E(x)
For each α > 1,

μ(α (ζ )) = |{ j ∈ N : aj ∈ α (ζ )}|,
where for each subset J of N, |J| denotes the number of elements in J. If

μ(α (ζ )) < ∞, then α (ζ ) contains only a finite number of aj . Consequently,
for such a ζ ,
lim V1 (x) = 0.
x→ζ
x∈α (ζ )
But by Theorem 9.5.4, μ(α (ζ )) < ∞ for almost every ζ ∈ S.
Remark 9.5.6 In the case when n = 2, the sequence {aj } is a subset of a

unit disc D, and as such we will assume they are in complex form. Writing
z = x1 + ix2 , (x1 , x2 ) ∈ B, we have
∞
1 − aj z
Gμ (z) =
log = − log |B(z)|,
z − aj
j=1
where B is the Blaschke product with (non-zero) zeros {aj } given by

∞
4 |aj | z − aj
B(z) = .
aj 1 − aj z
j=1
When n = 2, the hypothesis (9.5.5) is known as the Blaschke condition,

which is necessary and sufficient for a bounded analytic function in D having
prescribed zeros {aj }.
Example 9.5.7 We conclude this section by providing an example to show

that the exponent q > n/2 is the best possible. Specifically, we construct an
example of a function f satisfying the hypothesis of Theorem 9.5.1 with q =
n/2 for which
lim Gf (x) = +∞ for every ζ ∈ S. (9.5.6)
x→ζ
x∈α (ζ )
As in the example of the previous section, let {xj } be a countably infinite

subset of B with |xj | → 1 as j → ∞ such that α (ζ ) contains infinitely many
xj for every ζ ∈ S and α > 1.
For each j, choose δj , 0 < δj < 12 , and cj > 0 such that
(a) the family {E(xj , δj )} is pairwise disjoint, and
∞
2/n
(b) (1 − |xj |2 )cj < ∞.
j=1
Let {aj } be a sequence of positive numbers with aj → ∞. For each j, choose a
non-negative measurable function fj satisfying
(1) supp fj ⊂ E(xj , δj ),

n/2
(2) fj dτ < cj , and

(3) fj (y)Gh (xj , y)dτ (y) > aj .
B
Clearly for each j we can find an fj satisfying (1) and (2). Fix a j and let
Ej = E(xj , δj ). If we cannot find an fj satisfying (1)–(3), then we have

2/n
f (y)Gh (xj , y)dτ (y) ≤ aj cj
Ej
n/2 dτ ≤
for all non-negative measurable functions satisfying Ej f 1. By
duality, this, however, implies that y → Gh (xj , y) ∈ whichL n/(n−2) (Ej , τ ),
is a contradiction.

Let f = fj . We now show that f satisfies the hypothesis of Theorem 9.5.1
with q = n/2. First, by (3.3.8)
δjn
τ (E(xj , δj )) ≈ ≤ Cn .
(1 − δj2 )n−1
Also, for y ∈ E(xj , δj ) we have (1 − |y|2 ) ≤ 3(1 − |xj |2 ). Thus by Hölder’s

inequality
∞

(1 − |y|2 )n−1 f (y)dτ (y) ≤ 3 (1 − |xj |2 )n−1 fj dτ
B j=1 Ej
∞
, -2/n
n/2
≤3 (1 − |xj | ) (τ (Ej ))
2 n−1 (n−2)/n
fj dτ
j=1 Ej
∞
2/n
≤ 3Cn (1 − |xj |2 )n−1 cj < ∞.
j=1
Thus the measure dμ = fdτ satisfies (9.5.1). Also,

∞

(1 − |y|2 )f n/2 (y)dτ (y) ≤ 3 (1 − |xj |2 )n−1 cj .
B j=1
2/n
By convergence of the series in (b), there exists jo such that (1 − |xj |2 )cj ≤1
for all j ≥ jo . Hence, for j ≥ jo , and all n ≥ 2,

2/n n/2 2/n
(1 − |xj |2 )n−1 cj ≤ (1 − |xj |2 )cj ≤ (1 − |xj |2 )cj .
9.6 Exercises 169
Thus
∞
2/n
(1 − |y|2 )n−1 f n/2 (y)dτ (y) ≤ C (1 − |x|2j )cj < ∞.
B j=1
Therefore f satisfies the hypothesis of Theorem 9.5.1 with q = n

2. Finally,
since
Gf (xj ) ≥ fj (y)Gh (xj , y)dτ (y) > aj ,
B
the potential Gf satisfies (9.5.6) at every ζ ∈ S. By choosing the fj to be C∞
functions we obtain the existence of a C∞ potential Gf satisfying (9.5.6) at
every ζ ∈ S.
9.6 Exercises
9.6.1. Construct
a measure μ on B ⊂ R3 such that e1 is a limit point of supp μ,
B (1 − |x| ) dμ(x) < ∞, with Gμ (re1 ) = +∞ for all r, 0 < r < 1.
2 2
9.6.2. (*) In [8] D. H. Armitage provided an example of a Euclidean harmonic

function h on B such that |h| ∈ L p (B) for any p > 0. Construct an
example of an H-harmonic function h on B such that |h| ∈ L p (B, ν) for
any p > 0.
Definition 9.6.1 A function V(x) on B has non-tangential limit L at
ζ ∈ S in L p , p > 0, if for every α > 1,

1
lim |V(x) − L|p dν(x) = 0.
δ→0 ν(α,δ (ζ )) α,δ (ζ )
where α,δ (ζ ) = α (ζ ) ∩ B(ζ , δ).

In [98] L. Ziomek proved the following theorem. Suppose that V is
subharmonic in B and

sup V + (rt)dσ (t) < ∞. (9.6.1)
0<r<1 S
Then V has a non-tangential limit in L p , 1 ≤ p < n/(n − 2) at almost

every ζ ∈ S. This limit coincides with the ordinary radial limit of V.
An analogue of this result for M-harmonic functions on the unit ball in
Cm has also been proved by J. Cima and C. S. Stanton [16].
9.6.3. (*) Prove the following Conjecture. Let V be H-subharmonic in B
satisfying (9.6.1). Then V has a non-tangential limit in L p , 0 1, τ > 1, ζ ∈ S, set
α,τ (ζ ) = {x ∈ B : |x − ζ |τ < α(1 − |x|)}.
The set α,τ (ζ ) is called a tangential approach region at ζ ∈ S. It can

be shown that for τ > 1 the regions α,τ (ζ ) have tangential contact in
all directions at ζ ∈ S (see [86]).
Definition 9.6.2 A function V on B is said to have tangential limit L

of order τ at ζ ∈ S if
lim V(x) = L
x→ζ
x∈α,τ (ζ )
for every α > 1. Likewise, V has tangential limit L of order τ in L p at

ζ ∈ S if

1
lim |V(x) − L|p dν(x) = 0,
δ→0 ν(α,τ ,δ (ζ )) α,τ ,δ (ζ )
where α,τ ,δ (ζ ) = α,τ (ζ ) ∩ B(ζ , δ).
Tangential boundary limits of harmonic functions or Green potentials

have been considered by many authors, including H. Aikawa [6], M.
Arsove and H. Huber [9], Y. Mizuta [59, 60], A. Nagel, W. Rudin, and
J. H. Shapiro [61], J.-M. G. Wu [95], and the author [83, 86], among
many others.
Investigate the following two conjectures for H-superharmonic func-
tions on B.
(a) Conjecture 1. If f is a non-negative measurable function on B
satisfying

(1 − |x|2 )γ f q (x)dτ (x) < ∞
B
for some γ , 0 < γ < (n − 1), and some q > n2 , then Gf has
tangential limit 0 of order τ , 1 ≤ τ ≤ (n − 1)/γ , at a.e. ζ ∈ S.
(b) Conjecture 2. If μ is a non-negative regular Borel measure on B
satisfying

(1 − |x|2 )γ dμ(x) < ∞
B
for some γ , 0 < γ < (n − 1), then Gμ has tangential limit 0 of

order τ , 1 ≤ τ ≤ (n − 1)/γ , in L p , 0 < p < n/(n − 2), at a.e.
ζ ∈ S.
9.6 Exercises 171
9.6.5. (*) Weighted boundary limits of non-negative H-subharmonic func-

tions. In [28] F. W. Gehring proved that if w is a non-negative
subharmonic function in the unit disc D satisfying

wp (z)dxdy < ∞
D
for some p > 1, then
lim (1 − |z|)wp (z) = 0 for a.e. θ .

z→ζ
z∈α (eiθ )
This result was extended by D. Hallenbeck [33] to 0 0, as well as subharmonic
functions on bounded domains with C1 boundary [86]. The paper [86]
also contains results concerning tangential approach regions. Analogous
results for M-subharmonic functions have also been proved by the
author in [85].
Investigate weighted boundary limits of non-negative H-subharmonic
p
functions f on B satisfying f ∈ Lγ (τ ) for some p > 0 and some
γ > (n − 1) min{1, p}.
9.6.6. (*) Weighted boundary limits of potentials. This area of investigation
is motivated by an old result of M. Heins concerning weighted limits of
log |B(z)|, where B(z) is a convergent Blaschke product in the unit disc
D. In [37] M. Heins proved that if B is a Blaschke product in D, then
1
lim inf(1 − r) log =0 (9.6.2)
r→1 |B(reiθ )|
for all θ , 0 ≤ θ < 2π . This result was extended by the author [80] to
potentials on D in the following theorem.
Theorem 9.6.3 If Gμ is the potential of a measure μ satisfying (9.2.4)

with n = 2, then for all curves γ : [0, 1) → D with limt→1 γ (t) = 1,
lim inf(1 − |γ (t)|)Gμ (γ (t)eiθ ) = 0

t→1
for all θ , 0 ≤ θ < 2π .
Exercise 9.6.1 shows that the analogue of (9.6.2) is false for n ≥ 3.

There are, however, several extensions of Theorem 9.6.3 that are worthy
of consideration. The first involves the following question. Suppose F
is a relatively closed subset of B such that t ∈ S is a limit point of F.

What are necessary and sufficient conditions on F such that
lim inf(1 − |x|2 )n−1 Gμ (x) = 0? (9.6.3)
x→t
x∈F
When n = 2, this question was answered by D. H. Luecking [52], and

subsequently extended to invariant potentials on the unit ball in Cm by
K. T. Hahn and D. Singman [32]. It was proved that (9.6.3) holds if and
only if the capacity of the sets
F ∩ {x ∈ B : |z − t| < }, > 0,
is bounded away from zero.
Here, the capacity c(K) of a compact set K is defined as
c(K) = sup{μ(K) : supp μ ⊂ K and Gμ ≤ 1 on K}.
For an arbitrary set A,
c(A) = sup{c(K) : K is a compact subset of A}.
In a different direction, one can show that when n = 2, Theorem 9.6.3
is equivalent to
lim inf(1 − r)M∞ (Gμ , r),
r→1
where M∞ (Gμ , r) = sup|z|=r Gμ (z). Again, Exercise 9.6.1 shows

that this is false for n ≥ 3. The above result was extended by
S. Gardiner [25] to Euclidean potentials on B ⊂ Rn (see Exercise
4.8.12), and by the author to invariant potentials [81], [82] on B ⊂ Cm .
10
Bergman and Dirichlet Spaces of

H-Harmonic Functions
In this chapter we consider Bergman and Dirichlet spaces of H-harmonic

functions on B. For p > 0 and γ ∈ R, let

Lγp (τ ) = f measurable : (1 − |x|2 )γ | f (x)|p dτ (x) < ∞ ,
B
with
1/p
f γ ,p = (1 − |x|2 )γ | f (x)|p dτ (x) . (10.0.1)
B
p
The H-harmonic weighted Bergman space Bγ , 0 < p < ∞, γ ∈ R, is
p
defined as the space of H-harmonic functions f on B for which f ∈ Lγ (τ ).
p
For γ = n we obtain the classical Bergman space B p = Bn of H-harmonic
functions f on B with f ∈ L (ν).
p
p
Also, for 0 < p < ∞, γ ∈ R, Dγ denotes the weighted Dirichlet space of
p
H-harmonic functions f for which |∇ h f | ∈ Lγ with
1/p
f Dγp = | f (0)| + (1 − |x|2 )γ |∇ h f (x)|p dτ (x) . (10.0.2)
B
Since |∇ h f (x)| = (1 − |x|2 )|∇f (x),

2 γ +p−n
Dγp = f H-harmonic : (1 − |x| ) |∇f (x)| dν(x) < ∞ .
p
B
p p
In Section 10.1 we investigate basic properties of the spaces Bγ and Dγ , and
in Section 10.2 we provide a brief discussion of Möbius invariant spaces on B.
This leads to several open problems concerning the Möbius invariant Hilbert
space of H-harmonic functions on B. In Section 10.3 we prove that for γ >
p p
n − 1 the spaces Bγ and Dγ are equivalent for all p, 0 < p < ∞. Sections 10.4
p
and 10.5 deal with questions concerning the integrability of functions in Bγ
p
and Dγ as well as of eigenfunctions of Δh . Here we are primarily concerned
173
174 Bergman and Dirichlet Spaces of H-Harmonic Functions
with the question of given p, 0 < p < ∞, for what values of γ are the spaces
p p
Bγ and Dγ non-trivial? Finally, in Section 10.6 we prove generalizations of
three theorems of Hardy and Littlewood for H-harmonic functions on B, and
in Section 10.7 we prove a generalization of the Littlewood–Paley inequalities.
Other generalizations and results are provided in the exercises.
p p
10.1 Properties of Dγ and Bγ
p p p p
We first prove that if f ∈ Bγ (or Dγ ), then f ◦ ϕa ∈ Bγ (Dγ respectively). We
restrict ourself to the case γ > 0, which as we will later see will always be
necessary in order that these spaces be non-trivial.
p p
Theorem 10.1.1 Let f ∈ Bγ (respectively Dγ ) with γ > 0, 0 < p < ∞, then
p p
f ◦ ϕa ∈ Bγ (respectively Dγ ) with
γ
1 + |a| p
f ◦ ϕa γ ,p ≤ f γ ,p , (10.1.1)
1 − |a|
or
γ +1
1 + |a| p
f ◦ ϕa Dγp ≤ C f Dγp , (10.1.2)
1 − |a|
for some constant C independent of f and a.
Proof. For a ∈ B,

f ◦ ϕa γp,p = (1 − |x|2 )γ | f (ϕa (x))| p dτ (x),
B

= (1 − |ϕa (x)|2 )γ | f (x)| p dτ (x).
B
However, by (2.1.7),
(1 − |x|2 )γ (1 − |a|2 )γ
(1 − |ϕa (x)|2 )γ =
ρ(x, a)γ
(1 − |x|2 )γ (1 − |a|2 )γ
≤
(1 − |x||a|)2γ
(1 − |x|2 )γ (1 + |a|)γ
≤ ,
(1 − |a|)γ
from which (10.1.1) follows.
p p
For inequality (10.1.2), as above, by the invariance of ∇ h and τ we have

1 + |a| γ /p h
f ◦ ϕa Dγp ≤ | f (a)| + ∇ f γ ,p . (10.1.3)
1 − |a|
By the fundamental theorem of calculus,
|a| h
|∇ f (rt)|
| f (a)| ≤ | f (0)| + dr
0 (1 − r2 )
1
≤ | f (0)| + sup{|∇ h f (rt)| : 0 < r ≤ |a|}.
(1 − |a|2 )
But for x ∈ B|a| and 0 < δ < 1

2 fixed, by Theorem 4.7.4,

|∇ h f (x)| p ≤ Cδ |∇ h f (y)| p dτ (y),
Eδ (x)
which since (1 − |x|2 ) ≈ (1 − |y|2 ) for y ∈ Eδ (x)

Cδ
≤ (1 − |y|2 )γ |∇ h f (y)| p dτ (y)
(1 − |x|2 )γ Eδ (x)

Cδ
≤ (1 − |y|2 )γ |∇ h f (y)| p dτ (y).
(1 − |a|2 )γ B
Therefore,
Cδ
| f (a) ≤ | f (0)| + ∇ h f γ ,p ,
(1 − |a|2 )γ /p+1
from which (10.1.2) follows.
p
In the above we have proved that if f ∈ Dγ , 0 0, then for all
a ∈ B,
C
| f (a)| ≤ f Dγp , (10.1.4)
(1 − |a|2 )γ /p+1
where C is a constant depending on a fixed δ, 0 < δ < 1

2. Likewise, if f ∈
p
Bγ , 0 0, then
C
| f (a)| ≤ f γ ,p . (10.1.5)
(1 − |a|2 )γ /p
p
It is easy to show that ·γ ,p (respectively ·Dγp ) is a norm on Bγ (respectively
p p p
Dγ ) when 1 ≤ p < ∞, whereas · γ ,p and · D p are p-norms when
γ
0 < p < 1.
p
As in (7.2.1) define dBγp on Bγ by

f − gγ ,p , 1 ≤ p < ∞,
dBγp ( f , g) = p (10.1.6)
f − gγ ,p , 0 < p < 1.
p p
Then dBγp is a metric on Bγ , 0 < p < ∞. The metric dDγp on Dγ is defined
similarly. As in Section 7.2 we have the following theorem, the proof of which
is similar to the proof of Theorem 7.2.2 and thus is omitted.
p p
Theorem 10.1.2 The spaces (Bγ , dBγp ) and (Dγ , dDγp ) are complete metric
spaces for all p, 0 
(n − 1), and Bn1 respectively. Our first result is an analogue of [79, Theorem 1].
Theorem 10.1.3 For f ∈ Bλ1 , λ > (n − 1),

1
f (a) = (1 − |ϕa (y)|2 )λ f (y)dτ (y), a ∈ B,
cλ B
where cλ is a constant depending only on λ.
Proof. Define the function F on O(n) by

F(A) = (1 − |x|2 )λ f (ϕa (Ax))dτ (x),
B
which by the change of variables y = Ax

= (1 − |y|2 )λ f (ϕa (y))dτ (y)
B
= F(I).
Therefore F(A) = F(I) for all A ∈ O(n). Hence by Fubini’s theorem,

F(I) = F(A) dA = f (ϕa (Ax))dA dτ (x),
O(n) B O(n)
which since f is H-harmonic on B

= f (a) (1 − |x|2 )λ dτ (x)
B
= cλ f (a),

where cλ = B (1 − |x|
2 )λ dτ (x).
Therefore,

1 1
f (a) = F(I) = (1 − |x|2 )λ f (ϕa (x))dτ (x).
cλ cλ B
p p
The conclusion of the theorem now follows by the change of variable

y = ϕa (x).
Our second theorem is an analogue of [79, Theorem 2].
Theorem 10.1.4 If f ∈ Bn1 , then

(1 − |x|2 )n
f (x) = n
f (y)dν(y). (10.1.7)
B ρ(y, x)
Proof. Since f is H-harmonic, by Equation (4.3.3),

f (a) = f (ϕa (0)) = f (ϕa (x))dν(x),
B
which by the change of variable y = ϕa (x) and Theorem 3.3.1

= f (y)|Jϕa (y)|dν(y)
B

(1 − |ϕa (y)|2 )n
= f (y)dν(y)
B (1 − |y|2 )n

(1 − |a|2 )n
= n
f (y)dν(y).
B ρ(y, a)

When n = 2, Equation (10.1.7) becomes

(1 − |z|2 )2
f (z) = f (w)dA(w). (10.1.8)
D |1 − zw|
4
Definition 10.1.5 For f ∈ L1 (B, ν), define

(1 − |x|2 )n
Bf (x) = n
f (y)dν(y), x ∈ B.
B ρ(x, y)
In analogy to the case n = 2, the operator B will be called the Berezin

transform. For properties of the Berezin tansform when n = 2 the reader
is referred to the text [36] by H. Hedenmalm, B. Korenblum, and K. Zhu.
By Theorem 10.1.3, for α > −1,

1
f (0) = (1 − |y|2 )α f (y)dν(y),
cα B
where
1 1 ( n2 )(α + 1)
cα = (1 − |y|2 )α dν(y) = n rn−1 (1 − r2 )α dr = .
B 0 2 ( n2 + α + 1)
Hence as above,

1
f (a) = f (ϕa (0)) = (1 − |y|2 )α f (ϕa (y))dν(y)
cα B

1
= (1 − |ϕa (x)|2 )α |Jϕa (x)| f (x)dν(x)
cα B

1 (1 − |x|2 )α (1 − |a|2 )α+n
= f (x)dν(x).
cα ρ(x, a)α+n
Thus, for f ∈ Lα1 (B, ν), α > −1, we write

1 (1 − |x|2 )α+n (1 − |y|2 )α
Bα f (x) = f (y)dν(y). (10.1.9)
cα B ρ(x, y)α+n
If f is H-harmonic on B with f ∈ Lα1 (B, ν), then Bα f = f . The question posed

in Chapter 4 following Remark 4.3.6 can now be restated as follows:
Question. If f ∈ L1 (B, ν) and B f (x) = f (x) for all x ∈ B, is f H-harmonic
on B?
10.2 Möbius Invariant Spaces

In this section we provide a brief introduction to Möbius invariant function
spaces on B.
Definition 10.2.1 Let X be a linear space of functions on B with a complete

seminorm · on X, that is, f + g ≤ f + g and αf = |α| f for all
α ∈ R (or C) and f , g ∈ X. The space (X, ) is said to be Möbius invariant
if f ◦ φ ∈ X and
f ◦ ϕ = f for all f ∈ X, ϕ ∈ M.
The natural candidates for Möbius invariant spaces of H-harmonic functions

p p
on B are Bγ and Dγ with γ = 0. In this case

p
f ◦ ϕ0,p = | f (ϕ(x))| p dτ (x),
B

p
= | f | p dτ = f 0,p .
B
10.2 Möbius Invariant Spaces 179
p
Furthermore, for p ≥ 1 f 0,p is a norm on B0 . Likewise, by the invariance of
∇ h and τ ,

|∇ h f (ϕ(x))| p dτ (x) = |∇ h f | p dτ .
B B
In this case,
1/p
| f |Dγp = (1 − |x|2 )γ |∇ h f (x)| p dτ (x)
B
p
is a seminorm on Dγ . There is only one problem with these spaces. As we will
prove in Section 10.4, for p ≥ 1,
Bγp = {0} if and only if γ > n − 1, and

Dγp = {constants} if and only if γ > (n − 1) − p.
p p
Thus B0 = {0} for all p ≥ 1, and thus none of the spaces B0 is a Möbius
invariant Banach space.
p
The spaces D0 , however, are all non-trivial Möbius invariant Banach spaces
p
whenever p > (n − 1). In particular, when n = 2, D0 is a Möbius invariant
Banach space for all p > 1. Furthermore, when p = 2,

D0 = h harmonic on D :
2
|∇h(z)| dA(z) < ∞
2
D
is a Möbius invariant Hilbert space√ on D. Note, if h = Re f where f is

analytic on D, then since |∇h| = 2| f (z)|, the Dirichlet space of harmonic
and analytic functions is equivalent. Furthermore, J. Arazy and S. Fisher [7]
proved that when n = 2 (real dimension), the Dirichlet space D02 was unique
among Möbius invariant Hilbert spaces of analytic functions on D. However,
for n ≥ 3, the spaces D02 are trivial. Similar results are also true for the
analogous spaces of holomorphic functions on the unit ball in Cm , m ≥ 1.
K. Zhu [97] and independently M. Peloso [67] proved that for m ≥ 2, there
exists a unique Möbius invariant Hilbert space H of holomorphic functions on
the unit ball in Cm . Furthermore, Zhu provided the following characterization
of the space H and the Möbius invariant inner product on H. Zhu proved that

α!
H = f (z) = aα zα : |aα |2 |α| < ∞
α α
|α|!
with Möbius invariant (semi-) inner product

α!
f , g = c aα bα |α|.
α
|α|!
In the above, for a multi-index α = (α1 , . . . , αm ), α! = α1 ! · · · αm ! and

|α| = α1 + · · · + αm . Also, Peloso proved that the dual of H is the Bloch
space B consisting of the space of holomorphic functions f on B for which
(z)| < ∞, where ∇
sup |∇f is the invariant gradient on B. Furthermore, Peloso
also proved that the reproducing kernel1 of H is given by
1
K(z, w) = log
(1 − z, w)
where z, w is the usual inner product on Cm .
This leads to three interesting questions.
(1) For n ≥ 3, does there exist a Möbius invariant Hilbert space H of
H-harmonic functions on B?
(2) What is the characterization of H?
(3) What is the reproducing kernel of H?
p p
10.3 Equivalence of Bγ and Dγ for γ > (n − 1)
Our main result of this section is Theorem 10.3.3 in which we prove that the
p p
spaces Bγ and Dγ are equivalent for all p, 0 n − 1. We
first prove the following lemma.
Lemma 10.3.1 For f measurable on B, γ ∈ R, and 0 < δ < 12 ,

(1 − |x| ) | f (x)| dτ (x) ≈ (1 − |w| )
2 γ 2 γ
| f (x)| dτ (x) dτ (w).
B B Eδ (w)
Proof. For E ⊂ , let χE denote the characteristic function of E. Thus

(1 − |y| )
2 γ
| f (x)| dτ (x) dτ (y)
B Eδ (y)

= (1 − |y|2 )γ χEδ (y) (x)| f (x)| dτ (x) dτ (y),
B B

= | f (x)| χEδ (y) (x)(1 − |y|2 )γ dτ (y) dτ (x).
B B
Since |ϕy (x)| = |ϕx (y)|, we have χEδ (y) (x) = χEδ (x) (y). Also, for y ∈ Eδ (x),
(1 − |y|2 )γ ≈ (1 − |x|2 )γ . Therefore,
1 See Exercise 10.8.11 for the definition of reproducing kernel.

p p

χEδ (y) (x)(1 − |y|2 )γ dτ (y) ≈ (1 − |x|2 )γ dτ (y) = (1 − |x|2 )γ τ (Bδ ),
B Eδ (x)
Lemma 10.3.2 Let g be a non-negative locally integrable function on B.

(a) If g is quasi-nearly H-subharmonic, then g is quasi-nearly subharmonic,
that is, for all a ∈ B,

C
g(a) ≤ n g(x)dν(x) (10.3.1)
r B(a,r)
for all r < ro with B(a, r) ⊂ B and some constant C.

(b) If g is quasi-nearly subharmonic, then for all ρ, 0 < ρ < 1, gρ is quasi-
nearly subharmonic with constant C independent of ρ.
Proof. (a) Since g is quasi-nearly H-subharmonic, for all δ, 0 < δ < 12 , there
exists a constant C independent of g and δ, such that

C dν(x)
g(a) ≤ n g(x) .
δ E(a,δ) (1 − |x|2 )n
Since δ < 12 , by Exercise 2.4.1,

1
(1 − |a|2 ) ≤ (1 − |x|2 ) ≤ 3(1 − |a|2 )
3
for all x ∈ E(a, δ). Therefore

2n C
g(a) ≤ g(x)dν(x).
(2δ(1 − |a|2 ))n E(a,δ)
But by Exercise 4.8.3,
E(a, δ) ⊂ B(a, cδ (1 − |a|2 ))
with cδ = δ/(1 − δ). Since δ < 12 , cδ < 2δ. Let r, 0 < r < (1 − |a|). Choose
δ < 12 such that 2δ(1 − |a|2 ) = r. Then

2n C
g(a) ≤ n g(x)dν(x),
r B(a,r)
which proves (10.3.1).

(b) Suppose g is quasi-nearly subharmonic. Fix ρ, 0 < ρ < 1. Then

C
gρ (a) ≤ n g(x)dν(x),
r B(ρa,r)
which by the change of variable x = ρy

ρn
=C n gρ (y)dν(y)
r B(a, ρr )

C
= n gρ (y)dν(y).
δ B(a,δ)

Theorem 10.3.3 Let f be H-harmonic on B.

(a) For all p, 0 (n − 1) and 0 < p < ∞, there exists a constant C, independent
of f , such that

(1 − |x| ) | f (x)| dτ (x) ≤ C | f (0)| + (1 − |x| ) |∇ f (x)| dτ (x) .
2 γ p p 2 γ h p
B B
Proof. (a) Fix δ, 0 < δ < 12 . Then by Theorem 4.7.4(b)

|∇ h f (x)| p ≤ Cδ,p | f (y)| p dτ (y).
Eδ (x)
Therefore,

(1 − |x| ) |∇ f (x)| dτ (x) ≤ C (1 − |x| )
2 γ h p 2 γ
| f (y)| dτ (y) dτ (x),
p
B B Eδ (x)

≤C (1 − |x|2 )γ | f (x)| p dτ (x).
B
This proves (a).

(b) (i) The case 1 < p < ∞. By the fundamental theorem of calculus,
|x| |x| h
|∇ f (tζ )|
| f (x) − f (0)| ≤ |∇f (tζ )|dt ≤ dt, (10.3.2)
0 0 (1 − t2 )
where x = |x|ζ , ζ ∈ S. Hence for |x| ≤ 12 ,

|x|
| f (x) − f (0)| ≤ C |∇ h f (tζ )|dt ≤ C sup{|∇ h f (tζ )| : 0 ≤ t ≤ |x|}.
0
p p
But by Theorem 4.7.4(a), for fixed δ, 0 < δ < 12 , and t ≤ |x|,

|∇ h f (tζ )| p ≤ Cδ |∇ h f (y)| p dτ (y)
Eδ (tζ )

Cδ
≤ (1 − |y|2 )γ |∇ h f (y)| p dτ (y)
(1 − t2 )γ Eδ (tζ )

Cδ
≤ (1 − |y|2 )γ |∇ h f (y)| p dτ (y).
(1 − |x|2 )γ B

(1 − |x|2 )γ | f (x)| p dτ (x) ≤ C | f (0)| p + (1 − |x|2 )γ |∇ h f (x)| p dτ (x) .
B1 B
2
(10.3.3)
We now consider the integral over { 12 < |x| < 1}. Choose α > 0 such that
γ − n − αp > −1. Suppose 1 < p < ∞. Let q denote the conjugate exponent
of p. Then by Hölder’s inequality,
p p/q
r |∇ h f (tζ )| r
2 −αq−1
r
dt ≤ (1 − t ) dt (1 − t2 )αp−1 |∇ h f (tζ )| p dt
0 (1 − t2 ) 0 0
r
2 −αp
≤ (1 − r ) (1 − t2 )αp−1 |∇ h f (tζ )| p dt.
0
Hence
1
2 γ −n
r n−1
(1 − r ) | f (rt) − f (0)| p dσ (t) dr
1
2 S
1 r
2 γ −n−αp
≤C r n−1
(1 − r ) (1 − t ) 2 αp−1
Mpp (|∇ h f |, t)dt dr
1
2 0
= C(I1 + I2 ),
p
where Mp (|∇ h f |, t) = S |∇
h f (tζ )| p dσ (ζ ), and I1 and I2 are given by
1 1
2
I1 = rn−1 (1 − r2 )γ −n−αp (1 − t2 )αp−1 Mpp (|∇ h f |, t)dt dr,
1
2 0
1 r
I2 = rn−1 (1 − r2 )γ −n−αp (1 − t2 )αp−1 Mpp (|∇ h f |, t)dt dr.
1 1
2 2
We first consider the integral I1 . As above, for all t, 0 ≤ t ≤ 12 , and fixed

δ, 0 < δ < 12 ,

Cδ
|∇ h f (tζ )| p ≤ (1 − |y|2 )γ |∇ h f (y)| p dτ (y)
(1 − t2 )γ Eδ (tζ )

Cδ
≤ (1 − |y|2 )γ |∇ h f (y)| p dτ (y).
(1 − t2 )γ B
Since the remaining integrals with respect to t and r are all finite,

I1 ≤ C (1 − |y|2 )γ |∇ h f (y)| p dτ (y).
B
For the integral I2 , by interchanging the order of integration, we have
1 1
I2 ≤ (1 − t2 )αp−1 Mpp (|∇ h f |, t) rn−1 (1 − r2 )γ −n−αp dr dt.
1
2 t
But for t ≥ 12 ,
1
rn−1 (1 − r2 )γ −n−αp dr ≤ Ctn−1 (1 − t2 )γ −n−αp+1 .
t
Therefore,
1
I2 ≤ C tn−1 (1 − t2 )γ −n |∇ h f (tζ )| p dσ (ζ ) dt
1
S
2
≤C (1 − |y|2 )γ |∇ h f (y)| p dτ (y).

B
Thus

(1 − |y|2 )γ | f (y)| p dτ (y)
{ 12 ≤|y|<1}

≤ C | f (0)| + p
(1 − |y| ) |∇ f (y)| dτ (y) .
2 γ h p
B
Combining this with (10.3.3) proves the result for p > 1.
(ii) The case 0 < p ≤ 1. For this part of the proof we use Theorem 8.5.1.
An examination of that proof shows that the result is true whenever |∇f | is
quasi-nearly subharmonic (see also [90, Theorem 4]). Let f be H-harmonic
on B. Then by Lemma 10.3.2 |∇fρ | is quasi-nearly subharmonic2 for all
ρ, 0 < ρ < 1. Hence as in the proof of Theorem 8.5.1,
2 An alternate proof that |∇f | is quasi-nearly subharmonic whenever f is H-harmonic can be

ρ
found in the two papers [63] of M. Pavlović and [17] by O. Djordjević and M. Pavlović.
p p

| f (ρζ )| p dσ (ζ ) ≤ | f (0)| p + C (1 − |x|)p−1 ρ p |∇f (ρx)|dν(x),
S B
which by the change of variable x → x

ρ

= | f (0)| + Cρp 1−n
|∇g(x)| p (ρ − |x|)p−1 dν(x).
Bρ
Therefore, for γ > n − 1,

(1 − |x|2 )γ | f (x)| p dτ (x)
B
1
2 γ −n
=n ρ n−1
(1 − ρ ) | f (ρζ )| p dσ (ζ )dρ
0 S

1
γ −n
≤ Cn,γ | f (0)| + p
(1 − ρ) (ρ − |x|) p−1
|∇f (x)| dν(x)dρ ,
p
0 Bρ
which upon interchanging the order of integration

1
γ −n
= Cn,γ | f (0)| + p
|∇f (x)| p
(1 − ρ) (ρ − |x|) p−1
dρdν(x) .
B |x|
But
1
(1 − ρ)γ −n (ρ − |x|)p−1 dρ ≤ Cn,p,γ (1 − |x|)γ −n+p , (10.3.4)
|x|
from which the result now follows. For γ − n ≥ 0 the proof of inequality
(10.3.4) is straightforward. When −1 < γ − n < 0, split the integral up into
integrals over [|x|, 12 (1+|x|)] and [ 12 (1+|x|), 1] respectively, and then estimate
each of the integrals.
To conclude this section we consider the analogue of Theorem 10.3.3 for
eigenfunctions of Δh with non-zero eigenvalues. As in Definition 5.5.1, for
λ ∈ R,
Hλ = { f ∈ C2 (B) : Δh f = λ f }.
Our first result is the following analogue of Theorem 10.3.3.
Theorem 10.3.4 Let f ∈ Hλ , λ = 0. Then for all p, 0 < p < ∞ and γ ∈ R,
f ∈ Lγp (τ ) ⇐⇒ |∇ h f | ∈ Lγp (τ ),
with equivalence of norms.
p p
Proof. The proof that f ∈ Lγ implies that |∇ h f | ∈ Lγ is the same as in Theorem
10.3.3. The reverse implication follows likewise from Exercise 4.8.6.
p p
10.4 Integrability of Functions in Bγ and Dγ
p p
In the previous section we proved that up to constants, the spaces Bγ and Dγ
are equivalent for all p, 0 n − 1. In this section we
p
determine the values of γ for a given p, 0 < p < ∞, for which the spaces Bγ
p p
and Dγ are trivial. Our first theorem concerns the spaces Bγ . However, we first
p
prove a theorem for H-subharmonic functions in Lγ (τ ), from which our result
p
concerning Bγ follows. See [85] for the analogous result for M-subharmonic
functions on the unit ball in Cm .
Theorem 10.4.1 (a) Let 0 < p < ∞. If f is a non-negative H-subharmonic

p
function with f ∈ Lγ (τ ) for some γ ≤ min{ p(n − 1), (n − 1)}, then f ≡ 0.
p
(b) If 0 < p < 1 and f is an H-subharmonic function on B with f ∈ Lγ (τ )
for some γ ≤ min{ p(n − 1), (1 − p)(n − 1)}, then f ≡ 0.
p
Proof. (a) Suppose p ≥ 1 and assume γ ≤ n − 1 and that f ∈ Lγ (τ ) with
p
f ≡ 0. Hence f (a) = 0 for some a ∈ B. Since f ◦ ϕa ∈ Lγ (τ ), we can
without loss of generality assume that a = 0. Since f is a non-negative H-
p
subharmonic function, for 0 < ρ < 1,

ρ
f γp,p ≥ n rn−1 (1 − r2 )γ −n f (rt) p dσ (t)dr
0 S
ρ
2 γ −n
≥ f (0) n
p
r (1 − r )
n−1
dr.
0
However,
ρ
lim rn−1 (1 − r2 )γ −n dr
ρ→1 0
is finite if and only if γ − n > −1, that is, γ > n − 1. Thus f (0) = 0, which is
a contradiction.
p
Suppose 0 < p < 1. Let f ∈ Lγ (τ ) with f ≥ 0 and f ≡ 0. As above we can
p
without loss of generality assume f (0) = 0. Since f ∈ Lγ (τ ), as in (10.1.5) we
have
C
f (x) ≤ f γ ,p
(1 − |x|2 )γ /p
for all x ∈ B. Thus for all x with f (x) = 0 we have
γ
(1−p)
f (x)p−1 ≥ c(1 − |x|2 ) p
for a positive constant c. Therefore

γ
(1−p)
f (rt) dσ (t) = f (rt)f (rt)p−1 dσ (t) ≥ cf (0)(1 − r2 ) p
p
S S
p p
and
ρ
2 γ −n
f γp,p ≥n r n−1
(1 − r ) f (rt) p dσ (t)dr
0 S
ρ γ
−n
≥ nf (0) rn−1 (1 − r2 ) p dr.
0
However,
ρ γ
−n
lim rn−1 (1 − r2 ) p dr
ρ→1 0
is finite if and only if γp − n > −1, that is, γ > p(n − 1).
p
(b) Suppose 0 < p < 1 and f is H-subharmonic with f ∈ Lγ (τ ) for some
γ ≤ min{ p(n−1), (1−p)(n−1)}. Let f (x) = max{ f (x), 0}. Then f + is a non-
+
negative H-subharmonic function with f + ∈ Lγ (τ ) for some γ ≤ p(n − 1).

p
Thus f + ≡ 0 and thus | f | = −f . Since −f is a non-negative H-superharmonic

function on B, by the Riesz decomposition theorem

| f (x)| = Gh (x, y)dμ(y) + Ph (x, t)dα(t),
B S

where μ is a regular Borel measure on B with B (1 − |y|2 )n−1 dμ < ∞ and α
is a non-negative finite Borel measure on S. Since
Ph (x, t) ≥ C(1 − |x|2 )n−1 and Gh (x, y) ≥ C(1 − |x|2 )n−1 (1 − |y|2 )n−1 ,
we have that
| f (x)| ≥ C(1 − |x|2 )n−1
for all x ∈ B. Hence, if f is not identically zero,

1
f γ ,p ≥ C
p
rn−1 (1 − r2 )γ −n−p(n−1) = +∞
0
for any γ satisfying γ ≤ (1 − p)(n − 1).
As an immediate consequence of the previous theorem we have the

following corollary.
p
Corollary 10.4.2 (a) If f ∈ Bγ for some p ≥ 1 and γ ≤ n − 1, then f ≡ 0.
p
(b) If f ∈ Bγ for some p, 0 < p < 1, and γ ≤ p(n − 1), then f ≡ 0.
Proof. (a) Since | f | is a non-negative H-subharmonic function, the result for

p ≥ 1 follows from part (a) of Theorem 10.4.1.
(b) Since f is H-harmonic, both f + (x) = max{ f (x), 0} and f − (x) =
max{−f (x), 0} are non-negative H-subharmonic functions on B. Hence if
f ∈ Bγ for some γ ≤ p(n − 1), both f + and f − are in Lτ with γ ≤ p(n − 1)

p p
and thus by (a) of Theorem 10.4.1 are identically zero.
The proof of part (b) of Theorem 10.4.1 also proves the following.
Theorem 10.4.3 If f is a non-negative H-superharmonic function with f ∈

p
Lγ , 0 < p < 1, γ ≤ (1 − p)(n − 1), then f ≡ 0.
For non-negative H-harmonic functions we have the following theorem.
Theorem 10.4.4 Let 0 < p < 1. If h is a non-negative H-harmonic function

p
with h ∈ Bγ for some γ satisfying
γ ≤ max{ p(n − 1), (1 − p)(n − 1)},
then h ≡ 0.
Proof. If 12 ≤ p < 1, then max{ p(n − 1), (1 − p)(n − 1)} = p(n − 1),
and the conclusion follows from Theorem 10.4.1. Suppose 0 < p < 12 , then
max{ p(n − 1), (1 − p)(n − 1)} = (1 − p)(n − 1) and the conclusion follows
from Theorem 10.4.3.
Remark 10.4.5 Theorems 10.4.1 and 10.4.3 are analogues of results proved
by N. Suzuki [92] for Euclidean subharmonic functions on domains in Rn . For
example, Suzuki proved that if S is a non-negative (Euclidean) subharmonic
function on a bounded domain D with C1,1 boundary satisfying

S p (x)
1+β( p)
dx < ∞,
D δ(x)
then S ≡ 0. In the above β( p) = max{(n−1)(1−p), 0} and δ(x) is the distance

from x to the boundary of D. Furthermore, an example is given to show that
this is the best possible.
Examples 10.4.6 In these examples we show that some of the conclusions of

the previous theorems for 0 (1 − p)(n − 1) there exists
p
a non-negative H-superharmonic function in Lγ . Let f (x) = (1 − |x|2 )α . By
Exercise 4.8.2 the function f is H-superharmonic for all α, 0 < α ≤ (n − 1).
Furthermore,
1
(1 − |x|2 )γ f p (x)dτ (x) = n rn−1 (1 − r2 )γ −n+pα dr,
B 0
which is finite provided γ > (n−1)−pα. Taking α = (n−1) gives the desired
example. Thus the conclusion of Theorem 10.4.3 is the best possible.
p p
(ii) For Theorem 10.4.3, take f (x) = Ph (x, e1 ). Then

1
rn−1 (1 − r2 )γ −n Ph (rt, e1 )dσ (t)dr.
p
f γp,p = n
0 S
But

p p
Ph (rt, e1 )dσ (t) = Ph (re1 , t)dσ (t),
S S
1
which by Corollary 5.5.8 for 2 <p<1
≈ (1 − r2 )(1−p)(n−1) .
Therefore,
1
f γp,p ≈ rn−1 (1 − r2 )γ −n+(1−p)(n−1) dr.
0
The above integral, however, is finite if and only if γ > p(n − 1). When p = 12 ,
then again by Corollary 5.5.8
1
1 1
f γp,p ≈ rn−1 (1 − r2 )γ −n+ 2 (n−1) log dr,
0 (1 − r2 )
and the above integral again is finite if and only if γ > 12 (n − 1).
For 0 (1 − p)(n − 1). Hence for each
γ not satisfying the hypothesis of Theorem 10.4.4 there exists a non-negative
p
H-harmonic function in Bγ .
(iii) In this example we show that for each γ > max{ p(n − 1), 12 (n − 1)}
p
there exists a non-negative H-subharmonic function in Lγ .
If 12 p(n − 1).
Suppose 0 < 2 2 (n − 1). Then we can choose β > 1 such
1 1
that
γ 1
>β> .
p(n − 1) 2p
β
Then 2pβ > 1 and γ > βp(n − 1). In this case take f (x) = Ph (x, e1 ). Then f
is H-subharmonic and
1
2 γ −n βp
(1 − |x| ) f (x) = n
2 γ p
r (1 − r )
n−1
Ph (re1 , t)dσ (t),
B 0 S
which by Corollary 5.5.8, since βp > 12 ,

1
≈ rn−1 (1 − r2 )γ −n+(1−βp)(n−1) dr.
0
The above integral is finite provided γ > βp(n − 1).
Remark 10.4.7 As mentioned in Remark 10.4.5, N. Suzuki proved that for 0 p(n − 1), there exists a non-negative Euclidean subharmonic
p
function f on B such that f ∈ Lγ (τ ). Since the two concepts coincide when
n = 2, we provide a short proof of the example given by the author in [85] in
the following theorem. This still leaves the question of the existence of a non-
p
negative H-subharmonic function f on B with f ∈ Lγ (τ ), γ > p(n − 1), when
n ≥ 3.
Theorem 10.4.8 [85] If n = 2 and 0 p there
p
exists a non-negative subharmonic function fγ ∈ Lγ (τ ).
Proof. For 0 < β < π/2, let Sβ be the angular region with vertex at 1 defined
by
Sβ = {z ∈ D : | arg (1 − z)| < β, |1 − z| < cos β}.
Let ϕβ be a conformal mapping of Sβ onto D mapping the boundary of Sβ onto

the boundary of D with ϕβ (1) = 1. Let gβ be the function defined on D by

Ph (ϕβ (z), 1), z ∈ Sβ

gβ (z) =
0, z ∈ D \ Sβ .
Then the function gβ is subharmonic on D. By [55, Lemma 2.2] there exists a

non-zero analytic function h defined on a neighborhood N of 1 such that
1 − ϕβ (z) = (1 − z)b h(z)
for all z ∈ N ∩ Sβ , where b = π/(2β). Set b = 1 + (β) where (β) → 0 as

β → π/2. Then, for all z ∈ Sβ ,
C
gβ (z) ≤ Ph (ϕβ (z), 1) ≤ .
|1 − z|1+(β)
Hence,
1
2 γ p 2 γ −2
2π χSβ (reiθ )
(1 − |z| ) gβ (z)dτ (z) ≤ C (1 − r ) dθ r dr
D 0 0 |1 − reiθ |p+p(β)
1
≤C (1 − r2 )γ −p−p(β)−1 r dr.
0
p p
If γ > p we can choose β sufficiently close to π/2 such that γ −p−p(β) > 0,
in which case the above integral is finite. For such a β, define fγ = gβ .
p
We now consider the spaces Dγ , 0 < p < ∞, γ ≤ n − 1. By Exercise 4.8.1,

|x|2 |∇ h f (x)|2 = (1 − |x|2 )2 |x, ∇f (x)|2 + (1 − |x|2 )2 |Ti,j f (x)|2 , (10.4.1)
i<j
p
where Ti,j f (x) = xi (∂f /∂xj ) − xj (∂f /∂xi ). Thus f ∈ Dγ if and only if
p
|x, ∇f (x)| and |Ti,j f | are in Lγ +p for all i, j. This leads to the following
theorem.
p
Theorem 10.4.9 (a) If f ∈ Dγ for some p ≥ 1 and γ ≤ (n − 1) − p, then f is
constant.
p
(b) If f ∈ Dγ for some p, 0 < p < 1, and γ ≤ p(n − 1) − p, then f is
constant.
p
Proof. Since Ti,j f is H-harmonic whenever f is H-harmonic, if f ∈ Dγ , 0 <
p
p < ∞, then Ti,j f ∈ Bγ +p for all i, j. Hence if f satisfies either (a) or (b), then
by Corollary 10.4.2 Ti,j f = 0 for all i, j.
To conclude the proof we show that this implies that f is constant. If
Ti,j f (x) = 0 for all i, j and all x ∈ B, then by (10.4.1),
|x, ∇f (x)| = |x||∇f (x)|,
and as a consequence ∇f (x) = λ(x) x for some scalar function λ. We now show
that this implies that f is a radial function.
Consider the sphere Sr of radius r and let p1 and p2 be any two points on Sr .
Also let φ(t), 0 ≤ t ≤ 1, be any C1 curve in Sr from p1 to p2 . Since |φ(t)| = r,
by differentiation it follows that φ (t), φ(t) = 0. Thus
1
f ( p2 ) − f ( p1 ) = ∇f (φ(t)), φ (t)dt
0
1
= λ(φ(t))φ(t), φ (t)dt = 0.
0
Thus f ( p2 ) = f ( p1 ); that is, f is a radial function. However, any radial

H-harmonic function is a constant. To see this, if f is radial and satisfies
Δh f = 0, then either f is constant or f is a constant multiple of the Green’s
function gh , which is not H-harmonic on B.
Example 10.4.10 (i) Suppose f is H-harmonic on B such that |∇f (x)| ≤ C.

Then |∇ h f (x)| ≤ C(1 − |x|2 ) and
1
(1 − |x|2 )γ |∇ h f (x)| p dτ (x) ≤ C rn−1 (1 − r2 )γ −n−p dr,
B 0
which is finite if and only if γ > (n − 1) − p.

If the dimension of Rn is even, then as in Example 6.1.3, Ph [q](x) is a
polynomial whenever q is a polynomial on S. Thus |∇Ph [q](x)| ≤ C for all
x ∈ B.
If the dimension of Rn is odd, it is still possible to find a function f which is
H-harmonic on B for which |∇f | ≤ C on B. We illustrate this with n = 3. If
as above we let q(t) = t12 , then by Example 6.1.3
Ph [q](x) = 1
3 + (x12 − 13 |x|2 )S3,2 (|x|),
where
S3,2 (r) = c3,2 F(2, − 12 ; 72 ; r2 ).
Using the fact that ([1])

d ab
F(a, b; c; z) = F(a + 1, b + 1; c + 1; z),
dz c
and that F(a, b; c; z) converges absolutely on |z| = 1 whenever c − a − b > 0,
we have that

S3,2 (r) = c rF(3, 12 ; 92 ; r2 ),
and the series converges absolutely for all r, 0 ≤ r ≤ 1. Thus |∇Ph [q](x)|
is bounded on B. Note, even though Ph [q](x) is C1 on B, it is not C2 . The
does not converge absolutely on |z| = 1. For n > 3, the result
function S3,2
follows by Exercise 6.4.2.
(ii) Suppose n = 2 and u√is harmonic on D. If we let u = Re f where f is
analytic on D, then |∇u| = 2| f (z)|. Thus u ∈ Dγ if and only if
p

(1 − |z|2 )γ −2+p | f (z)| p dA(z) < ∞. (10.4.2)
D
However, since | f (z)| p

is subharmonic for all p > 0,
1
(1 − |z|2 )γ −2+p | f (z)| p dA(z) ≥ Mpp ( f , R) r(1 − r2 )γ −2+p dr
R<|z|<1 R
2 γ +p−1
≥ Cγ ,p (1 − R ) .
Therefore,

Mpp ( f , R) ≤ C(1 − R )2 −γ −p+1
(1 − |z|2 )γ −2+p | f (z)| p dτ (z).
R<|z|<1
Hence if f ∈ Dγ for γ ≤ 1 − p, limR→1 Mp ( f , R) = 0. Thus f is constant on

p p
p
B. Therefore when n = 2, if u ∈ Dγ for γ ≤ (1 − p), then u is constant. Thus
when n = 2 the result of Theorem 11.8.2 is not best possible. Also, for n = 2,
p
f (z) = z is in Dγ for all γ > (1 − p).
(iii) For p > 0 and γ ∈ R, set
Dγp = { f : f is Euclidean harmonic with |∇ h f | ∈ Lγp (τ )}.

p
Then f ∈ Dγ if and only if

(1 − |x|2 )γ −n+p |∇f (x)| p dν(x) < ∞.
B
By [76, Theorem 4.1.4], when n ≥ 3, |∇f | p is subharmonic when p ≥ (n − 2)/
(n − 1). Hence as in Example (ii), for p ≥ (n − 2)/(n − 1),

Mpp (|∇f |, R) ≤ Cγ , p (1 − R2 )n−γ −p−1 (1 − |x|2 )γ −n+p |∇f (x)| p dν(x).
R<|x|<1
Thus if γ ≤ (n − 1) − p we have
lim Mpp (|∇f |, R) = 0

R→1
from which it follows that f is constant on B. The function f (x) = x12 − x22 is
p
harmonic on B with |∇f (x)| ≤ 2. Thus f ∈ Dγ for any γ > (n − 1) − p.
Remark 10.4.11 In view of the above examples it is conjectured that if

p
f ∈ Dγ , 0 < p < ∞, for any γ ≤ (n − 1) − p, then f is constant.
10.5 Integrability of Eigenfunctions of Δh

In this section we investigate the integrability of eigenfunctions of Δh with real
non-zero eigenvalues λ. If λ and α are related by
λ = 4(n − 1)2 α(α − 1), α = 0, 1, (10.5.1)
then each Phα (x, t), t ∈ S, is an eigenfunction of Δh with eigenvalue λ. Solving

(10.5.1) for α gives
1 1
α= ± λ + (n − 1)2 . (10.5.2)
2 2(n − 1)
Thus λ and α are real if and only if λ ≥ −(n − 1)2 . Our first theorem is the
following.
Theorem 10.5.1 Let 0 (n − 1) + p2 λ + (n − 1)2 − (n − 1) .
Proof. If f ∈ Hλ is radial, then by Theorem 5.5.3,
f (x) = f (0)gα (x),
related by (10.5.1) and gα is given by (5.5.4). Hence with

where λ and α are
α = 12 + 2(n−1)
1
λ + (n − 1)2 ,

(1 − |x| ) f (x)dτ (x) = f (0) (1 − |x|2 )γ −n gαp (x)dν(x). (10.5.3)
2 γ p p
B B
Since α > 12 , by Corollary 5.5.8 gα (x) p ≈ (1 − |x|2 )p(1−α)(n−1) . Thus the

integral in (10.5.3) is finite if and only if γ > (n − 1) − p(1 − α)(n − 1) or
p
γ > (n − 1) + (n − 1)2 + λ − (n − 1) .
2

Theorem 10.5.2 Let 0 < p < ∞ and suppose λ is real with λ ≥ −(n − 1)2 .
Then
Lγp (τ ) ∩ Hλ = {0}

(a) for all γ ≤ (n − 1) + p2 (n − 1)2 + λ − (n − 1) when p ≥ 1, and

(b) for all γ ≤ p2 (n − 1) + (n − 1)2 + λ when 0 < p < 1.
p p
Proof. (a) Suppose p ≥ 1 and f ∈ Lγ ∩ Hλ . Then f ◦ ϕa ∈ Lγ for all a ∈ B.
Hence by integration in polar coordinates
1
f ◦ ϕa γp,p = n rn−1 (1 − r2 )γ −n | f (ϕa (rζ ))| p dσ (ζ )dr,
0 S
which by Hölder’s inequality

p
1
2 γ −n

≥n r n−1
(1 − r )
f (ϕa (rt))dσ (t) dr.
0 S
But by Theorem 5.5.5, since f ∈ Hλ ,

f (ϕa (rt))dσ (t) = gα (r)f (a).
S

Therefore with α = 1
2 + 1
2(n−1) (n − 1)2 + λ,
1
f ◦ ϕa γp, p ≥ n| f (a)| p rn−1 (1 − r2 )γ −n gαp (r)dr
0
1
≥ C| f (a)| p
rn−1 (1 − r2 )γ −n+p(1−α)(n−1) dr.
0
The above integral is finite, however, if and only if

p
γ > (n − 1) − p(1 − α)(n − 1) = (n − 1) + (n − 1)2 + λ − (n − 1) .
2

Thus if f ◦ϕa γ , p < ∞ for some γ ≤ (n−1)+ p2 (n − 1)2 + λ − (n − 1) ,

we must have f (a) = 0, hence the result. Taking α = 12 − 2(n−1)
1
(n − 1)2 + λ
gives the same result since (1 − α ) = α. If α = 1
2 (λ = −(n − 1)2 ), then
1 p
p 1
f ◦ ϕa γp, p ≥ C| f (a)| p rn−1 (1 − r2 )γ −n− 2 (n−1) log dr,
0 (1 − r2 )
and the above integral again is finite if and only if γ > (n − 1) − p2 (n − 1).
p
(b) Suppose now that 0 < p < 1 and f ∈ Lγ ∩ Hλ . Let y ∈ B. Then since
| f | is quasi-nearly H-subharmonic, for 0 < δ < 12 ,

| f (y)| p ≤ Cδ | f (x)| p dτ (x),
Eδ (y)
which since (1 − |x|2 ) ≈ (1 − |y|2 ) for all x ∈ Eδ (y),

Cδ
≤ (1 − |x|2 )γ | f (x)| p dτ (x)
(1 − |y|2 )γ Eδ (y)
Cδ
≤ f γp, p .
(1 − |y|2 )γ
Thus for all y with f (y) = 0,

γ
(1−p)
| f (y)|p−1 ≥ C(1 − |y|2 ) p .
Therefore,

| f (rt)| dσ (t) =
p
| f (rt)|| f (rt)|p−1 dσ (t)
S S
γ

(1−p)
≥ C(1 − r2 ) p | f (rt)|dσ (t)
S
γ
≥ C(1 − r ) 2 p −γ
| f (0)|gα (r).

Therefore, if λ > −(n − 1)2 and α = 1
2 + 2(n−1)
1
(n − 1)2 + λ, then
1 γ
−n+(1−α)(n−1)
f γp,p ≥ C| f (0)| rn−1 (1 − r2 ) p dr.
0
The above integral is finite if and only if

γ 1
> (n − 1) − (1 − α)(n − 1) = (n − 1) + (n − 102 + λ ,
p 2
from which the result follows. If λ = −(n − 1)2 , that is, α = 12 , then the result
follows as in (a).
Example 10.5.3 In this example we show that the result of Theorem 10.5.2 is
sharp for all
(n − 1)
p≥ , λ ≥ −(n − 1)2 .
(n − 1) + (n − 1)2 + λ

(i) Suppose first that p ≥ 1. Then with α = 12 + 2(n−1)
1
(n − 1)2 + λ, by
p
Theorem 10.5.1, the function gα ∈ Lγ if and only if
p
γ > (n − 1) + (n − 1)2 + λ − (n − 1) .
2
(ii) Suppose 0 12 , then gpα (r) ≈ (1 − r2 )(1−pα)(n−1) , and the above integral is finite if
and only if
p
γ > pα(n − 1) = (n − 1) + (n − 1)2 + λ .
2
1
Likewise, if pα = 12 , gpα (r) ≈ (1 − r2 ) 2 (n−1) log (1−r
1
2 ) , and the integral again
is finite if and only if
1 p
γ > (n − 1) = (n − 1) + (n − 1)2 + λ .
2 2
But pα ≥ 1
2 if and only if
(n − 1)
p≥ .
(n − 1) + (n − 1)2 + λ
Whether the conclusion of Theorem 10.5.2 is
the best possible for the values
of p satisfying 0 < p < (n − 1)/((n − 1) + (n − 1)2 + λ) is not known to
the author. As the following theorem proves, for non-negative functions in Hλ
it is not. For λ = 0 this is Theorem 10.4.3.
Theorem 10.5.4 Let 0 −(n − 1)2 . If f is a
p
non-negative function in Hλ ∩ Lγ (τ ) for some γ satisfying

γ ≤ max 12 p (n − 1) + (n − 1)2 + λ ,

(n − 1) − 12 p (n − 1) + (n − 1)2 + λ ,
then f ≡ 0 on B. The result is the best possible for all p, 0 < p < 1.
p

Proof. If f ∈ Lγ ∩ Hλ for some λ ≤ 12 p((n − 1) + (n − 1)2 + λ), then by
Theorem 10.5.2 f ≡ 0 on B. On the other hand, if f ∈ Hλ is non-negative, then
by Theorem 5.5.2,

f (x) = Pαh (x, t)dμ(t)
S

for some non-negative measure μ on S and α = 12 + 2(n−1) 1
(n − 1)2 + λ.
Since α > 0 we have Ph (x, t) ≥ C(1 − |x| )
α 2 α(n−1) , and thus
1
(1 − |x| ) | f (x)| dτ (x) ≥ C[μ(S)]
2 γ p p
rn−1 (1 − r2 )γ −n+αp(n−1) dr.
B 0
But the integral on the right is finite if and only if

γ > (n − 1) − αp(n − 1) = (n − 1) − 12 p (n − 1) + (n − 1)2 + λ .
p

Hence if f ∈ Lγ for some γ ≤ (n − 1) − 12 p((n − 1) + (n − 1)2 + λ), we
have f ≡ 0. The function Phα (x, e1 ) shows that the result is the best possible.
p
Our final result of this section concerns the space Hλ . Recall that for 0 0, then f ≡ 0.
p
Proof. Since Hλ ⊂ Hλ1 for all p ≥ 1, it suffices to prove the result for p = 1.
Without loss of generality we assume f is real valued, and for > 0, set
f = f + i. Then | f | is C2 on B. Although f ∈ Hλ1 , the function | f | is
H-subharmonic with least H-harmonic majorant F| f | satisfying F| f | ≤ F| f | +
where F| f | is the least H-harmonic majorant of | f |. Since Δh f = λ f ,
Δh | f | = 2 | f |−3 |∇ h f |2 + λ| f |−1 | f |2 ≥ λ| f |−1 | f |2 .
Since | f | is C2 and has an H-harmonic majorant, we have by Theorem 9.2.4

and the above that

(1 − |y|2 )n−1 | f (y)|−1 | f (y)|2 dτ (y)
B

≤ λ−1 (1 − |y|2 )n−1 Δh | f (y)|dτ (y) ≤ CF| f | (0) ≤ C,
B
where C is a constant independent of . Hence by Fatou’s lemma,

(1 − |y|2 )n−1 | f (y)|dτ (y) < ∞.
B
Thus f ∈ Lγ1 (τ ) ∩ Hλ , λ > 0 with γ = (n − 1), which is strictly less than

2 [(n − 1) + (n − 1)2 + λ]. Thus by Theorem 10.5.2, f ≡ 0.
1
Example 10.5.6 The conclusion of the previous theorem is false when 0 1, Ph ∈ Hλα for all p, 0 < p ≤ α1 .
10.6 Three Theorems of Hardy and Littlewood

In this section we prove three analogues of well-known results of Hardy and
Littlewood concerning the means Mp ( f , r) of analytic functions in the unit
disc. Our first result concerns the comparative rate of growth of Mp ( f , r) and
Mp ( f , r), where for 0 < p < ∞ and 0 < r < 1,
1/p
Mp ( f , r) = | f (rt)| p dσ (t) .
S
As usual M∞ ( f , r) = sup{| f (rt)| : t ∈ S}. When n = 2, proofs of the results
for analytic functions in the unit disc may be found in [19]. The classical result
of Hardy and Littlewood [34] on the unit disc D is as follows. If f is analytic
in D, 0 0, then
# $
Mp ( f , r) = O (1 − r)−α ⇐⇒ Mp ( f , r) = O (1 − r)−(α+1) .
The proofs of the extension to H-harmonic functions on B and the invariant

gradient ∇ h will rely heavily on the fact that | f | p and |∇ h f | p are quasi-nearly
H-subharmonic for all p, 0 0, then

Mp ( f , r) = O (1 − r2 )−α ⇐⇒ Mp (|∇ h f |, r) = O (1 − r2 )−α
as r → 1.
This result was originally proved by M. Pavlović [62] for M-harmonic
functions on the unit ball in Cm (see also [84, Theorem 10.7]).
Proof. For the proof we will assume that 0 < p < ∞. The case p = ∞ follows
similarly. Suppose Mp ( f , r) = O((1 − r2 )−α ). Then by Theorem 4.7.4(b), for
A ∈ O(n),

|∇ h f (rAe1 )| p ≤ Cδ, p | f (y)| p dτ (y),
Eδ (rAe1 )

≤ Cδ, p | f (Ay)| p dτ (y).
Eδ (re1 )
Thus

Mpp (|∇ h f |, r) = |∇ h f (rAe1 )| p dA
O(n)

≤ Cδ, p |f (Ay)| p dA dτ (y).
Eδ (re1 ) O(n)
But
|f (Ay)| p dA = O (1 − |y|2 )−α .
O(n)
But for y ∈ E(re1 , δ), (1 − |y|2 ) ≈ (1 − r2 ). Therefore

Mpp (|∇ h f |, r) = O (1 − r2 )−α .
# $
Conversely, suppose Mp (|∇ h f |, r) = O (1 − r2 )−α . Without loss of gener-
p
ality we assume f (0) = 0. By the fundamental theorem of calculus,

r h
|∇ f (sζ )|
| f (rζ )| ≤ ds.
0 (1 − s )
2
For j = 0, 1, 2, . . . , set rj = 1 − 2−j . Also, for 0 < r < 1, let m ∈ N be such

that rm−1 < r ≤ rm . Then
rj+1 |∇ h f (sζ )|
m−1
| f (rζ )| ≤ ds
rj (1 − s2 )
j=0

m−1
≤C sup{|∇ h f (sζ )| : s ∈ (rj , rj+1 )}.
j=0
If rj < s < rj+1 , then by (2.1.7)

(1 − rj s)2 − (1 − rj2 )(1 − s2 )
|ϕrj ζ (sζ )|2 =
(1 − rj s)2
(s − rj )2 1
≤ ≤ .
(1 − rj )2 4
Thus sζ ∈ E(rj ζ , 12 ) for all s, rj < s < rj+1 . As a consequence, E(sζ , 14 ) ⊂
E(rj ζ , 34 ) for all such ζ . Since |∇ h f | is quasi-nearly H-subharmonic,

sup{|∇ h f (sζ )| p : rj < s < rj+1 } ≤ Cp |∇ h f (y)| p dτ (y).
E(rj ζ , 34 )
Therefore

sup{|∇ h f (sAe1 )| p : rj < s < rj+1 }dA
O(n)

≤ Cp |∇ h f (Ay)| p dAdτ (y)
E 3 (rj e1 ) O(n)
4
= Cp Mpp (|∇ h f |, |y|)dτ (y).

E 3 (rj e1 )
4
But for y ∈ E(rj e1 , 34 ),

1 # $pα
Mpp (|∇ h f |, |y|) ≤ C ≤ C 2j .
(1 − rj )
2 pα
Therefore,

sup{|∇ h f (sζ )| p : s ∈ (rj , rj+1 )}dσ (ζ ) ≤ C(2pα ) j .
S
Thus if 0 < p < 1,

m−1
| f (rζ )| p dσ (ζ ) ≤ sup{|∇ h f (sζ )| p : s ∈ (rj , rj+1 )}dσ (ζ )
S j=0 S

m−1
≤C (2pα ) j ≤ Cp,α (2m )pα
j=0
≤ Cp,α (1 − r2 )−pα .
On the other hand, for 1 ≤ p < ∞, by Minkowski’s inequality,

1/p m−1 1/p
| f (rζ )| p dσ (ζ ) ≤ sup{|∇ h f (sζ )| p : s ∈ (rj , rj+1 )}dσ (ζ )
S j=0 S

m−1
≤C (2α ) j ≤ Cα (2m )α
j=0
≤ Cα (1 − r2 )−α .
Hence, in either case, Mp ( f , r) ≤ C(1 − r2 )−α .
Our second theorem is an application of the previous one.
Theorem 10.6.2 [34] Let f be H-harmonic on B and suppose
C
Mp ( f , r) ≤ , 0 < p < ∞, β ≥ 0.
(1 − r)β
Then there exists a constant K, depending only on p, β, and n, such that
KC
Mq ( f , r) ≤ , 0 < p < q ≤ ∞.
β+(n−1)( 1p − 1q )
(1 − r)
Proof. We first prove the result for q = ∞. Our proof will use the notation of
Theorem 8.5.1. Let 0 < r < 1. Set rj = 1 − 21j , j = 0, 1, 2, . . . , N, where N is
the first integer such that rN > r. Then as in Theorem 8.5.1,

N
| f (rζ )| ≤ | f (0)| + log 2 sup |∇ h f (tζ )|.
k=1 t∈[rk−1 ,rk )
For x ∈ B let B(x) = B(x, 14 (1 − |x|2 )), and for j = 2, 3, 4, . . . , N, set

Aj = {x ∈ B : rj−2 < |x| < rj+2 }.
Then as in the proof of Theorem 8.5.1

1/p
sup |∇ h f (tζ )| ≤ Cn |∇ h f (x)| p dτ (x) .
t∈[rk−1 ,rk ) Ak
Suppose Mp ( f , r) ≤ C(1 − r)−β . Then by Theorem 10.3.4, Mp (|∇ h f |, r) ≤

C(1 − r)−β . Note, an examination of the proof of Theorem 10.3.4 shows that
this part of the result is still valid when β = 0. Therefore,
rk+2
|∇ h f (x)| p dτ (x) = n ρ n−1 (1 − ρ 2 )−n Mpp (|∇ h f |, ρ)dρ
Ak rk−2
rk+2
≤ Cn C p
(1 − ρ)−n−βp dρ
rk−2
k
≤ C p Cn,β, p 2(n−1)+βp .
Hence with K = Cn,β, p ,

N
k
(n−1)
+β
| f (rζ )| ≤ | f (0)| + C K 2 p
k=1
# $ (n−1) +β
≤ | f (0)| + C K 2N p
C K
≤ (n−1)
,
(1 − r) p +β
for some constant K . Taking the supremum over ζ ∈ S proves the result for
q = ∞.
Suppose 0 < p < q < ∞. Then
1/q
Mq ( f , r) = | f (rζ )| | f (rζ )|
p q−p
dσ (ζ )
S
1− qp p
≤ M∞ ( f , r) Mp ( f , r) q
KC
≤ ,
(1 − r)λ

where λ = β + (n − 1) 1p − 1q .
Our final theorem of this section is an application of the previous theorem.

The proof uses the Marcinkiewicz interpolation theorem (see [75, Appendix
B.1] or [99, (II) p.111]), and the method we use is similar to that used by T. M.
Flett in [23, Theorem 1].
Theorem 10.6.3 If 1 < p < q ≤ ∞, α = 1p − 1q , and p ≤ λ < ∞, then for all

fˆ ∈ L p (S) there exists a constant Cp,q , independent of fˆ, such that
1/λ
1
(1 − r) (n−1)λα−1
Mqλ ( f , r)dr ≤ Cp, q fˆp ,
0
where f (x) = Ph [ fˆ](x).
Proof. Define ω(r) = rn−1 . Also, define T on L p (S) by

− 1q
T fˆ(r) = ω(r) Mq ( f , 1 − r) 0 < r < 1, fˆ ∈ L p ,
where f (x) = Ph [ fˆ](x). By Minkowski’s inequality T is sub-additive, that is,

T( fˆ1 + fˆ2 )(r) ≤ T fˆ1 (r)+T fˆ2 (r) for all r ∈ (0, 1). Furthermore, for 1 ≤ p ≤ q,
by the previous theorem,
1
− 1p
Mq ( f , 1 − r) ≤ C fˆp ω(r) q .
Therefore,
− 1p
T fˆ(r) ≤ Cω(r) fˆp .
Hence,

C p fˆp
p
ˆ
{r : T f (r) > α} ⊂ r : ω(r) < .
α
Let denote the inverse function of ω. Then
{r : T fˆ(r) > α} ⊂ (0, b),
where

b = min 1, (C p fˆpp α −p ) .
Define the measure ν by dν = ω (r)dr. Then

b
C p fˆp
p
ˆ
ν({r : T f (r) > α}) ≤
ω (t)dt = ω(b) ≤ .
0 αp
Thus T is weak-type ( p, p). Hence by the Marcinkiewicz interpolation theo-
rem, T is of strong type ( p, p) for 1 < p < q, that is,
1/p 1/p
1 1
ˆ − pq
p
(T f ) dν = ω(r) ω (r)Mq ( f , 1 − r)dr
p
,
0 0
which by a change of variable

1/p
1
(1− qp )(n−1)−1
= cn (1 − r) Mqp ( f , r)dr
0
≤ A( p, q) fˆp .
Finally, suppose p < λ < ∞. Then
1
ω(r)λα−1 ω (r)Mqλ ( f , 1 − r)dr
0

. /λ−p 1
− pq
≤ sup ω(r) Mq ( f , 1 − r)
α
ω(r) ω (r)Mqp ( f , 1 − r)dr
r∈(0,1) 0
≤ A( p, q) fˆλ−p+p
p = A( p, q) fˆλp .
The result now follows by the change of variable r = (1 − ρ).
As an immediate consequence of the above we have the following.
Corollary 10.6.4 Let f ∈ H p , 1 1.
Proof. Take λ = q = pn/(n − 1) in the previous theorem.
Along the same lines, taking λ = q gives
Corollary 10.6.5 If f ∈ H p , 1 p,
1
p q
2 (n−1) q
(1 − |x| ) | f (x)| dτ (x) ≤ Ap,q f p ,
q
B
q
that is, H ⊂ p
B(n−1) q for all q > p > 1.
p
Example 10.6.6 We now use Theorem 8.6.3 to show that for 0 (n − 1), there exists h ∈ Bγ which is not in Hq for any q ≥ 1. Since
γ − n > −1, we can choose α > 0 such that γ − n − αp > −1. Take ω(r) =
(1 − r2 )−α . Then by Theorem 8.6.3 there exists an H-harmonic function on B
such that
(i) |h(rζ )| ≤ ω(r) for all r ∈ [0, 1) and all ζ ∈ S, and
(ii) h(rζ ) fails to have a finite limit as r → 1 at every ζ ∈ S.
p
As a consequence of (i) we have that h ∈ Bγ , but by (ii) h is not in Hq for any
q ≥ 1.
10.7 Littlewood–Paley Inequalities

The classical Littlewood–Paley inequalities for harmonic functions [51] in D
are as follows. Let h be harmonic on D. Then there exist positive constants
C1 , C2 , independent of h, such that
(a) for 1 < p ≤ 2,

hp ≤ C1 |h(0)| +
p p
(1 − |z|) |∇h(z)| dx dy .
p−1 p
(10.7.1)
D
(b) For p ≥ 2, if h ∈ H p , then

(1 − |z|)p−1 |∇h(z)| p dx dy ≤ C2 hpp . (10.7.2)
D
In 1956 T. M. Flett [22] proved that for analytic functions inequality (10.7.1)
is valid for all p, 0 < p ≤ 2. Hence if u = Re h, h analytic, then since |∇u| =
|h | it immediately follows that inequality (10.7.1) also holds for harmonic
functions in D for all p, 0 < p ≤ 2. A new proof of the Littlewood–Paley
inequalities for analytic functions was given by D. H. Luecking in [53]. Also,
a short proof of the inequalities for harmonic functions in D valid for all p, 0 1. In [77] S. Stević proved that for
n ≥ 3, inequality (10.7.1) is valid for all p ∈ [ n−2
n−1 , 1]. In [87], the author proved
that the analogue of (10.7.1) was valid for harmonic functions on bounded
domains with C1,1 boundary for all p, 0 < p ≤ 2. In the case of the unit ball,
this result was subsequently improved upon by O. Djordjević and M. Pavlović.
In [17] they proved that if u is a harmonic function on the unit ball B in RN
and if 0 < p ≤ 1, then

∗
u ( y) dσ ( y) ≤ C |u(0)| + (1 − |x|) |∇ u(x)| dx , (10.7.3)
p p p−1 h p
S B
where u∗ is the non-tangential maximal function of u. The Littlewood–Paley

inequalities have also been extended by the author to Hardy–Orlicz spaces of
harmonic functions on general domains in Rn , including Lipschitz domains
[88], as well as H-harmonic functions on the unit ball B [89].
In the present section we consider the Littlewood–Paley inequalities for
non-negative H-subharmonic functions on B for which Δh f is quasi-nearly
H-subharmonic on B. The results will then apply to the special case where

f = |h|2 , where h is H-harmonic on B, or more generally to f = |hn |2 , with
suitable convergence, where each hn is again H-harmonic on B. In each case,
Δh f is quasi-nearly H-subharmonic. Our main results of the section are the

following theorem and corollaries.
Theorem 10.7.1 Let f be a non-negative C2 H-subharmonic function on B for

which Δh f is quasi-nearly H-subharmonic on B. Then there exist constants C1
and C2 , independent of f , such that
(a) for 1 ≤ p < ∞,

(1 − |x|2 )n−1 (Δh f (x)) p dτ (x) ≤ C1 f p .
B
(b) For 0 < p ≤ 1, if in addition f p is C2 and H-subharmonic, then

f pp ≤ C2 f p (0) + (1 − |x|2 )n−1 (Δh f (x)) p dτ (x) .
B
Corollary 10.7.2 Let h be H-harmonic on B. Then there exist constants C1

and C2 , independent of h, such that
(a) if f ∈ H p , 2 ≤ p < ∞, then

(1 − |x|2 )n−1 |∇ h h(x)| p dτ (x) ≤ C1 hpp .
B
(b) For 0 1,

(Mα h) dσ ≤ C2 |h(0)| + (1 − |x| ) |∇ h(x)| dτ (x) .
p p 2 n−1 h p
S B
The previous corollary can also be stated as follows:
Corollary 10.7.3 Let h be H-harmonic on B.

p
(a) If h ∈ H p , 2 ≤ p < ∞, then h ∈ Dn−1 with hD p ≤ Chp .
n−1
p
(b) If h ∈ Dn−1 , 0 < p ≤ 2, then h ∈ Hp with Mα hp ≤ ChD p .
n−1
As a consequence of the previous corollary, a function h ∈ H2 if and only

(1 − |x|2 )n−1 |∇ h h(x)|2 dτ (x)
B
is finite. A similar result follows from Exercise 10.8.9. If for α > 1 we set
1
2
Sα (h, ζ ) = |∇ h h(x)|2 dτ (x) ,
α (ζ )
then it is easily shown that

Sα (h, ζ )dσ (ζ ) ≈ (1 − |x|2 )n−1 |∇ h h(x)|2 dτ (x).
2
S B
Thus h ∈ H2 if and only if Sα ∈ L2 (S). The function Sα (h, ζ ) is called the

square area integral of h. This characterization of H2 was also proved by
P. Cifuentis in [15]. For related problems the reader is referred to Exercise
10.8.10.
For the proof of the main theorem we require several preliminary lemmas.
Lemma 10.7.4 Let f be a positive C2 H-subharmonic function on B(0, ρ),

0 < ρ < 12 , such that Δh f is quasi-nearly H-subharmonic on B(0, ρ). Then
(a) for p ≥ 1,

(Δh f (y)) p dτ (y) ≤ cn ρ 2−2p Δh f p (y)dτ (y).
Bρ/2 Bρ
(b) Suppose 0 < p ≤ 1. If in addition, f p is C2 and H-subharmonic, then

Δh f (y)dτ (y) ≤ cn ρ
p 2p−2
(Δh f (y)) p dτ (y).
Bρ/4 Bρ
Proof. Without loss of generality we assume that n ≥ 3. The case n = 2 is

proved similarly. Recall from Theorem 4.1.1, for any r, 0 < r < 1, if f p is
H-subharmonic and C2 , then

f p (rζ )dσ (ζ ) − f p (0) = g(|x|, r)Δh f p (x)dτ (x), (10.7.4)
S Br
where

1 r (1 − s2 )n−2
g(|x|, r) = ds.
n |x| sn−1
We first note that for |x| ≤ 2δ ,
g(|x|, δ) ≥ cn,δ δ 2−n . (10.7.5)
(a) The case p ≥ 1. Set δ = 12 ρ. Since Δh f is quasi-nearly H-subharmonic,

, -p
(Δh f (0)) p ≤ cn ρ −n Δh f (x)dτ (x) ,
Bρ/2
which by inequality (10.7.5) and identity (10.7.4)

p
≤ cn,δ ρ −2 g(|x|, δ)Δh f (x)dτ (x)
Bδ
p
−2p
= cn,δ ρ f (δζ )dσ (ζ ) − f (0) .
S
Since (a − b) p ≤ a p − b p whenever p ≥ 1 and 0 ≤ b ≤ a,

p
(Δh f (0)) p ≤ cn,δ ρ −2p f (δζ )dσ (ζ ) − f p (0)
S

≤ cn,δ f (δζ ) − f p (0) .
p
S
The last inequality follows by Hölder’s inequality. Hence by (10.7.4) we have

−2p
(Δh f (0)) ≤ cn,δ ρ
p
g(|x|, δ)Δh f p (x)dτ (x).
Bδ
Replacing f by f ◦ ϕy , y ∈ Bδ , gives

(Δh f (y)) p ≤ cn,δ ρ −2p g(|x|, δ)Δh ( f (ϕy (x))) p dτ (x),
Bδ
which by the change of variable w = ϕy (x)

= cn,δ ρ −2p g(|ϕy (w)|, δ)Δh f p (w)dτ (w).
Eδ (y)
Since Eδ (y) ⊂ Bρ for all y ∈ Bδ ,

(Δh f (y)) p dτ (y) ≤ cn,δ ρ −2p g(|ϕy (w)|, δ)Δh f p (w)dτ (w)dτ (y),
Bδ Bδ Bρ

= cn,δ ρ −2p Δh f p (w) g(|ϕy (w)|, δ)dτ (y)dτ (w).
Bρ Bδ
But |ϕy (w)| = |ϕw (y)|. Therefore since E(w, δ) ⊂ B2ρ ,

g(|ϕw (y)|, δ) dτ (y) = g(|x|, δ) dτ (x) ≤ g(|x|, δ) dτ (x) ≤ Cn ρ 2 .
Bδ Eδ (w) B2ρ
Therefore,

(Δh f (y)) dτ (y) ≤ cn,δ ρ
p 2−2p
Δh f p (w)dτ (w),
Bρ/2 Bρ
which proves (a).

(b) The case 0 < p ≤ 1. For this case we have to assume in addition that f p
is C2 and H-subharmonic. As above, set δ = 12 ρ. Then by inequality (10.7.5),

f p (y)dτ (y) ≤ cn,δ δ n−2 g(|x|, δ)Δh f p (y)dτ (y),
Bδ/2 Bδ
which by identity (10.7.4)

= cn,δ δ n−2 f p (δζ )dσ (ζ ) − f p (0) .
S
Since 0 < p < 1,

p
f p (δζ )dσ (ζ ) − f p (0) ≤ f (δζ )dσ (ζ ) − f (0)
S S
p
= g(|x|, δ)Δh f (x)dτ (x)
Bδ
p
≤ sup (Δh f (x)) p
g(|x|, δ)dτ (x)
x∈Bδ Bδ
≤ Cn ρ 2p sup (Δh f (x)) p .

x∈Bδ
Since Δh f is quasi-nearly H-subharmonic and Eδ (x) ⊂ Bρ for all x ∈ Bδ ,

ρ (Δh f (x)) ≤ Cρ
2p p 2p−n
(Δh f (x)) p dτ (x).
Bρ
Therefore,

Δh f p (y)dτ (y) ≤ Cn,δ ρ 2p−2 (Δh f (x)) p dτ (x).
Bρ/4 Bρ
Lemma 10.7.5 Let f be a positive C2 H-subharmonic function on B such that

Δh f is quasi-nearly H-subharmonic on B. Fix δ, 0 < δ < 12 , and γ ∈ R. Then
(a) for p ≥ 1,

(1 − |x|2 )γ (Δh f (x)) p dτ (x) ≤ Cn,δ (1 − |x|2 )γ Δh f p (x)dτ (x)
B B
where Cn,δ is a constant independent of f .
(b) For 0 < p < 1, if in addition f p is H-subharmonic and C2 , then

(1 − |x|2 )γ Δh f p (x)dτ (x) ≤ Cn,δ (1 − |x|2 )γ (Δh f (x)) p dτ (x).
B B
Proof. (a) By Lemma 10.3.1

(1 − |x|2 )γ (Δh f (x)) p dτ (x)
B

≤ Cn (1 − |w| )
2 γ p
(Δh f (x)) dτ (x) dτ (w).
B Eδ/2 (w)
But by Lemma 10.7.4(a)

(Δh f (x)) p dτ (x) = (Δh ( f (ϕw (y))) p dτ (y)
Eδ/2 (w) Bδ/2

≤ Cn,δ Δh ( f ◦ ϕw ) p (y)dτ (y)
Bδ

= Cn,δ Δh f p (y)dτ (y).
Eδ (w)
Applying Lemma 10.3.1 again gives

2 γ
(1−|w| ) Δh f (y)dτ (y) dτ (w) ≤ Cn (1−|w|2 )γ Δh f p (w)dτ (w).
p
B Eδ (w) B
The proof of (b) follows similarly.

Proof of Theorem 10.7.1 (a) The case p ≥ 1. Assume first that f (x) > 0 for
all x ∈ B. Then f p is C2 for all p > 0. Fix δ, 0 < δ < 12 . Then with γ = n − 1,

(1 − |x|2 )n−1 (Δh f (x)) p dτ (x) ≤ Cn,δ (1 − |x|2 )n−1 Δh f p (x)dτ (x),
B B

≤ Cn,δ lim f p (rt)dσ (t) = Cn,δ f pp .
r→1 S
For the general case f (x) ≥ 0 take f (x) = f (x) + . Then Δh f = Δh f , and
by the above

(1 − |x|2 )n−1 (Δh f (x)) p dτ (x) ≤ Cn,δ f pp .
B
Letting → 0 proves the result.
(b) The case 0 < p ≤ 1. Assume that B (1−|x|2 )n−1 (Δh f (x)) p dτ (x) < ∞.
Then by Lemmas 10.3.2 and 10.7.5,

1 (1 − |x|2 )n−1
f p (rt)dσ (t) ≤ f p (0) + Δh f p (x)dτ (x)
S n B |x| n−2

≤ f (0) + sup Δh f p (x) + Cn (1 − |x|2 )n−1 (Δh f (x)) p dτ (x).
p
x∈B 1 B
4
Hence f ∈ S p (or alternately f p ∈ S 1 ) and thus has an H-harmonic majorant

p
Hf p on B with f p = Hf p (0). Hence by the Riesz decomposition theorem,

Hf p (x) = f (x) + Gh (x, y)Δh f p (y)dτ (y).
p
B
10.8 Exercises 211
Thus by Remark 9.2.5(c),

f pp ≤ f (0) + Cn
p
(1 − |y|2 )n−1 Δh f p (y)dτ (y),
B

≤ f p (0) + Cn,δ (1 − |y|2 )n−1 (Δh f (y)) p dτ (y).
B
Therefore

f pp ≤ f p (0) + Cn,δ (1 − |y|2 )n−1 (Δh f (y)) p dτ (y),
B
Proof of Corollary 10.7.2 For (a) take f = |h|2 . Then Δh f = 2|∇ h h|2 . For
h ∈ Hq , q ≥ 2, the result follows from Theorem 10.7.1(a) by taking p = q/2.
(b) For 0 < q ≤ 1 the result is Theorem 8.5.1. For 1 < q < 2 we consider
f = |h + i|2 , > 0. Then f is a positive C2 H-subharmonic function for
which Δh f = 2|∇ h f |2 . Thus by Theorem 10.7.1(b) with p = q/2,

f p ≤ C |h(0) + i| + (1 − |x| ) |∇ h(x)| dτ (x) .
p q 2 n−1 h q
B
The result now follows

by letting → 0 and using the fact that for q > 1,
(M h) q dσ ≤ A |h| q dσ .
S α q S
10.8 Exercises
10.8.1. If f ∈ H p , p ≥ 1 and p < q ≤ ∞, prove that

(n−1)( p1 − 1q )
Mq ( f , r) = o (1 − r) .
10.8.2. Fix β > 1. Show that there exists a non-negative continuous H-

superharmonic function V on B such that V(x) > 0 for all x ∈ β (e1 )
and V(x) = 0 on B \ β (e1 ).
10.8.3. (*) Fix β > 1. Question. Does there exist a positive H-harmonic
function h on β (e1 ) such that
(a) h(x) ≤ C|x − e1 |1−n−(β) , (β) > 0, where (β) → 0 as β → 1,
and
(b) h(x) = 0 on ∂β (e1 )?
If the answer to the previous exercise is yes, then it can be used to
provide a positive answer to the following question.
10.8.4. (*) For 0 p(n − 1), does there exist a non-negative
p
H-subharmonic function f with f ∈ Lγ (τ )?
10.8.5. (a) Suppose f is a non-negative C2 H-subharmonic function such
that f p is also C2 for all p > 1. Prove that

f γp,p ≈ | f (0)| p + (1 − |x|2 )γ Δh f p (x)dτ (x).
B
(b) Let h be an H-harmonic function on B. Prove that for p > 1,
p
h ∈ Bγ if and only if B (1−|x|2 )γ |h(x)|p−2 |∇ h h(x)|2 dτ (x) < ∞,
and if this is the case, then

hγ ,p ≈ |h(0)| + (1 − |x|2 )γ |h(x)|p−2 |∇ h h(x)|2 dτ (x).
p p
B
10.8.6. As in [36] investigate properties of the operators Bα .
10.8.7. (*) Using the methods of [64], prove Theorem 10.7.1 using the Riesz
measure μf rather than Δh f .
10.8.8. (*) Using the methods of [88], prove the following version of
the Littlewood–Paley inequalities for Hardy-Orlicz spaces of H-
harmonic functions.
Theorem Let ψ ≥ 0 be an increasing convex C2 function on [0, ∞)
satisfying ψ(0) = 0 and ψ(2x) ≤ cψ(x) for some positive constant
√
c. Set ϕ(t) = ψ( t). Then there exist constants C1 and C2 such that
for all H-harmonic functions f ,
(a) if ϕ is concave on [0, ∞),

f ψ ≤ C1 ψ(| f (0)|) + (1 − |x| ) ψ(|∇ f (x)|)dτ (x) .
2 n−1 h
B
(b) If ϕ is convex on [0, ∞), then

C2 ψ(| f (0)|) + (1 − |x|2 )n−1 ψ(|∇ h f (x)|)dτ (x) ≤ f ψ .
B

In the above, f ψ = lim ψ(| f (rt)|)dσ (t).
r→1 S
10.8.9. Hardy and Dirichlet spaces. [89] For γ ∈ R the Dirichlet space
Dγ is defined as the set of C1 functions on B for which |∇ h f | ∈ Lγ2 (τ )
with norm
f Dγ = | f (0)| + Dγ ( f )1/2 , (10.8.1)
where Dγ ( f ) is given by

Dγ ( f ) = (1 − |x|2 )γ |∇ h f (x)|2 dτ (x).
B
Prove the following.
10.8 Exercises 213
Theorem Let f be H-harmonic on B.

(a) If f ∈ Dγ for some γ , (n − 3) < γ ≤ (n − 1), (0 < γ ≤ 1 when
n = 2), then f ∈ H p for p = 2(n − 1)/γ , with
f p ≤ C1 f Dγ ,
where C1 is a positive constant independent of f .

(b) If f ∈ H p for some p, 1 < p ≤ 2, then f ∈ Dγ for γ = 2(n−1)/p
with
f Dγ ≤ C2 f p ,
where C2 is a positive constant independent of f .

10.8.10. Littlewood–Paley theory for H-subharmonic functions. In 1930,
N. Lusin [54] introduced the square area integral Sα , α > 1, for an
analytic function f on D given by

Sα ( f , ζ )2 = | f (z)|2 dA(z), ζ ∈ [0, 2π ).
α (ζ )
In 1938 J. Marcinkiewicz and A. Zygmund [56] proved that if f is in

the Hardy space H p , 0 1. For H-harmonic
functions h on B, the analogue of the Lusin square area integral
is the function Sα (h, ζ ) defined for ζ ∈ S by
1
2
Sα (h) = Sα (h, ζ ) = |∇ h(x)| dτ (x)
h 2
. (10.8.2)
α (ζ )
This function was considered by S. Grellier and P. Jaming in [31].

The main result of their paper was that Sα ∈ L p (S), 0 
1, if and only if the non-tangential maximal function Mα h ∈ L p (S),
with equivalence of norms. For a non-negative C2 H-subharmonic
function f , set
1
2
Sα ( f ) = Sα ( f , ζ ) = 2
Δh f (x)dτ (x) . (10.8.3)
α (ζ )
Taking f = |h| where h is H-harmonic on B gives, up to a constant,

(10.8.2).
(a) As in [91], prove the following theorem.
Theorem Let f be a non-negative C2 H-subharmonic function

on B such that f po is H-subharmonic for some po > 0. If f ∈ S p
for some p > po , then for for every α > 1,

Sαp ( f , ζ )dσ (ζ ) ≤ Aα f pp
S
for some constant Aα independent of f .

(b) (*) Prove the reverse inequality for H-subharmonic functions.
10.8.11. (a) For γ > (n − 1) prove that Bγ2 is a real Hilbert space with inner
product

f , gγ = (1 − |x|2 )γ f (x)g(x)dτ (x). (10.8.4)
B
(b) Using the fact that point evaluation is a bounded linear functional
on Bγ2 , prove that there exists a function Rγ (x, y) on B × B
satisfying the following:
i. Rγ (x, y) = Rγ (y, x).
γ
ii. For each y ∈ B, prove that the function Ry defined by
γ
Ry (x) = R (x, y) is in Bγ .
γ 2

iii. f (y) = f , Ry = (1 − |x|2 )γ Rγ (x, y)dτ (x) for all f ∈ Bγ2 .
γ
B
The function Rγ is called the reproducing kernel of Bγ2 .
(c) (*) Find the function Rγ (x, y). (See [10, Theorem 8.13] for the
analogous question for Euclidean harmonic functions on B.)
10.8.12. Define the Bloch space B of H-harmonic functions on B by

B = f H-harmonic : sup |∇ f (x)| < ∞
h
x∈B
with norm
f B = | f (0)| + sup |∇ h f (x)|.
x∈B
The little Bloch space B0 is defined as

B0 = f ∈ B : lim |∇ f (x)| = 0 .
h
|x|→1
Prove the following:

(a) B with the seminorm supx∈B |∇ h f (x)| is Möbius invariant.
(b) B is a Banach space.
(c) B0 is a closed subspace of B.
10.8 Exercises 215
10.8.13. (*) Does there exist a Möbius invariant Hilbert space H of H-

harmonic functions on B? If the answer is yes do the following:
(a) Provide a characterization of H.
(b) Find the reproducing kernel of H.
10.8.14. (*) For 0 < μ < (n − 1), let Kμ be the kernel on S × S given by
1
Kμ (ζ , η) = (1 − r2 )μ−1 Ph (rζ , η)dr.
0
Note: Kμ is the fractional integral of Ph . As in [29] investigate the
mapping

Kμ f (ζ ) = Kμ (ζ , η)f (η)dσ (η).
S
It is conjectured that the mapping f → Kμ f maps L p (S) to Lq (S)
where
1 1 μ
= − .
q p (n − 1)
10.8.15. (*) A classical theorem in the theory of several complex variables

states that if f is holomorphic in the unit ball B of Cn and Rf =
∂f
zj ∂zj ∈ H p for some p, 0 < p < n, then f ∈ H np/(n−p) (e.g., [29]).
Suppose f is H-harmonic on B ⊂ Rn . If Ti,j f ∈ H p for some p, what
can be said about f ?
References
[1] M. Abramowitz and I. Stegun. Handbook of Mathematical Functions. Applied

Math. Series 55. National Bureau of Standards, 1964.
[2] P. Ahern, J. Bruna, and C. Cascante. H p -theory for generalized M-harmonic
functions on the unit ball. Indiana Univ. Math. J., 45(1):103–135, 1996.
[3] P. Ahern, M. Flores, and W. Rudin. An invariant volume mean value property.
J. Funct. Analysis, 111:380–397, 1993.
[4] L. Ahlfors. Möbius Transformations in Several Dimensions. University of
Minnesota, School of Mathematics, 1981.
[5] L. V. Ahlfors. Hyperbolic motions. Nagoya Math. J., 29:163–166, 1967.
[6] H. Aikawa. Tangential behavior of Green potentials and contractive properties of
Lp -potentials. Tokyo J. Math., 9:223–245, 1986.
[7] J. Arazy and S. Fisher. The uniqueness of the Dirichlet space among Möbius
invariant function spaces. Illinois J. Math., 29:449–462, 1985.
[8] D. H. Armitage. On the global integrability of superharmonic functions. J. London
Math. Soc., 4:365–373, 1971.
[9] M. Arsove and A. Huber. On the existence of non-tangential limits of subharmonic
functions. J. London Math. Soc., 42:125–132, 1967.
[10] S. Axler, P. Bourdon, and W. Ramey. Harmonic Function Theory. Springer-
Verlag, New York, NY, 1992.
[11] A. F. Beardon. The Geometry of Discrete Groups. Springer-Verlag, New York,
NY, 1983.
[12] A. P. Calderón. Commutators of singular integral operators. Proc. Nat. Acad. Sci.
U. S. A., 53:1092–1099, 1965.
[13] I. Chavel. Eigenvalues in Riemannian Geometry. Academic Press, Orlando, FL,
1984.
[14] P. Cifuentis. H p classes on rank one symmetric spaces of noncompact type. II.
Nontangential maximal function and area integral. Bull. Sci. Math., 108:355–371,
1984.
[15] P. Cifuentis. A characterization of H 2 classes on rank one symmetric spaces of
noncompact type. Proc. Amer. Math. Soc., 106:519–525, 1989.
[16] J. A. Cima and C. S. Stanton. Admissible limits of M-subharmonic functions.
Michigan Math. J., 32:211–220, 1985.
216
References 217
[17] O. Djordjević and M. Pavlović. On a Littlewood–Paley type inequality. Proc.

Amer. Math. Soc., 135:3607–3611, 2007.
[18] N. Dunford and J. T. Schwartz. Linear Operators Part I. Interscience Publishers,
Inc., New York, NY, 1957.
[19] P. Duren. Theory of H p Spaces. Academic Press, New York, NY, 1970.
[20] A. Erdélyi, editor. Higher Transcendental Functions, Bateman Manuscript
Project, volume I. McGraw-Hill, New York, NY, 1953.
[21] C. Fefferman and E. Stein. H p spaces of several variables. Acta Math, 129:
137–193, 1972.
[22] T. M. Flett. On some theorems of Littlewood and Paley. J. London Math. Soc.,
31:336–344, 1956.
[23] T. M. Flett. On the rate of growth of mean values of holomorphic functions. Proc.
London Math. Soc., 20:749–768, 1970.
[24] H. Furstenberg. A Poisson formula for semisimple Lie groups. Ann. Math.,
77:335–386, 1963.
[25] S. J. Gardiner. Growth properties of potentials in the unit ball. Proc. Amer. Math.
Soc., 103:861–869, 1988.
[26] L. Garding and L. Hörmander. Strongly subharmonic functions. Math. Scand.,
15:93–96, 1964.
[27] J. B. Garnett. Bounded Analytic Functions. Pure and Applied Mathematics.
Academic Press, New York, NY, 1981.
[28] F. W. Gehring. On the radial order of subharmonic functions. J. Math. Soc. Japan,
9:77–79, 1957.
[29] I. Graham. The radial derviative, fractional integrals, and the comparitive growth
of means of holomorphic functions on the unit ball in Cn . Annals Math. Studies,
100:171–178, 1981.
[30] M. D. Greenberg. Ordinary Differential Equations. Wiley, Hoboken, NJ, 2014.
[31] S. Grellier and P. Jaming. Harmonic functions on the real hyperbolic ball II.
Hardy–Sobolev and Lipschitz spaces. Math. Nachr., 268:50–73, 2004.
[32] K. T. Hahn and D. Singman. Boundary behavior of invariant Green’s potentials
on the unit ball of Cn . Trans. Amer. Math. Soc., 309:339–354, 1988.
[33] D. J. Hallenbeck. Radial growth of subharmonic functions. Pitman Research
Notes, 262:113–121, 1992.
[34] G. H. Hardy and J. E. Littlewood. Some properties of fractional integrals, II. Math.
Z., 34:403–439, 1932.
[35] J. H. Hardy and J. E. Littlewood. The strong summability of Fourier series. Fund.
Math., 25:162–189, 1935.
[36] H. Hedenmalm, B. Korenblum, and K. Zhu. Theory of Bergman Spaces, volume
199 of Graduate Texts in Mathematics. Springer, New York, NY, 2000.
[37] M. Heins. The minimum modulus of a bounded analytic function. Duke Math. J.,
14:179–215, 1947.
[38] S. Helgason. Groups and Geometric Analysis. American Mathematical Society,
Providence, RI, 2000.
[39] E. Hewitt and K. Stromberg. Real and Abstract Analysis. Springer-Verlag,
New York, NY, 1965.
[40] L. Hörmander. Linear Partial Differential Operators. Springer-Verlag, New York,
NY, 1963.
218 References
[41] P. Jaming. Trois problémes d’analyse harmonique. PhD thesis, Université

d’Orléans, 1998.
[42] P. Jaming. Harmonic functions on the real hyperbolic ball I. Boundary values and
atomic decomposition of Hardy spaces. Colloq. Math., 80:63–82, 1999.
[43] P. Jaming. Harmonic functions on classical rank one balls. Boll. Unione Mat.
Italia, 8:685–702, 2001.
[44] M. Jevtić. Tangential characterizations of Hardy and mixed-norm spaces of
harmonic functions on the real hyperbolic ball. Acta Math. Hungar., 113:
119–131, 2006.
[45] A. W. Knapp. Fatou’s theorem for symmetric spaces, I. Ann. Math, 88(2):
106–127, 1968.
[46] A. Koranyi. Harmonic functions on Hermitian hyperbolic space. Trans. Amer.
Math. Soc., 135:507–516, 1969.
[47] A. Koranyi. Harmonic functions on symmetric spaces. In W. M. Boothby and
G. L. Weiss, editors, Symmetric Spaces, Marcel Dekker, Inc., New York, NY,
1972.
[48] A. Koranyi and R. P. Putz. Local Fatou theorem and area theorem for symmetric
spaces of rank one. Trans. Amer. Math. Soc., 224:157–168, 1976.
[49] Ü. Kuran. Subharmonic behavior of |h|p ( p > 0, h harmonic). J. London Math.
Soc., 8:529–538, 1974.
[50] N. N. Lebedev. Special Functions and their Applications. Dover Publications,
New York, NY, 1972.
[51] J. E. Littlewood and R. E. A. C. Paley. Theorems on Fourier series and power
series. Proc. London Math. Soc., 42:52–89, 1936.
[52] D. H. Luecking. Boundary behavior of Green potentials. Proc. Amer. Math. Soc.,
96:481–488, 1986.
[53] D. H. Luecking. A new proof of an inequality of Littlewood and Paley. Proc.
Amer. Math. Soc., 103(3):887–893, 1988.
[54] N. Lusin. Sur une propriété des fonctions à carré sommable. Bull. Calcutta Math.
Soc., 20:139–154, 1930.
[55] B. D. MacCluer. Compact composition operators of H p (Bn ). Michigan Math. J.,
32:237–248, 1985.
[56] J. Marcinkiewicz and A. Zygmund. A theorem of Lusin. Duke Math. J., 4:
473–485, 1938.
[57] K. Minemura. Harmonic functions on real hyperbolic spaces. Hiroshima Math. J.,
3:121–151, 1973.
[58] K. Minemura. Eigenfunctions of the Laplacian on a real hyperbolic space. J. Math.
Soc. Japan, 27(1):82–105, 1975.
[59] Y. Mizuta. On the boundary limits of harmonic functions with gradient in Lp . Ann.
Inst. Fourier, Grenoble, 34:99–109, 1984.
[60] Y. Mizuta. On the boundary limits of harmonic functions. Hiroshima Math. J.,
18:207–217, 1988.
[61] A. Nagel, W. Rudin, and J. H. Shapiro. Tangential boundary behavior of functions
in Dirichlet-type spaces. Ann. Math., 116:331–360, 1982.
[62] M. Pavlović. Inequalities for the gradient of eigenfunctions of the invariant
Laplacian in the unit ball. Indag. Mathem., N. S., 2:89–98, 1991.
References 219
[63] M. Pavlović. On subharmonic behavior and oscillation of functions in balls in Rn .

Publ. Inst. Math. (N.S.), 69:18–22, 1994.
[64] M. Pavlović. A Littlewood–Paley theorem for subharmonic functions. Publ. Inst.
Math. (Beograd), 68(82):77–82, 2000.
[65] M. Pavlović. A short proof of an inequality of Littlewood and Paley. Proc. Amer.
Math. Soc, 134:3625–3627, 2006.
[66] M. Pavlović and J. Riihentaus. Classes of quasi-nearly subharmonic functions.
Potential Analysis, 29:89–104, 2008.
[67] Marco M. Peloso. Möbius invariant spaces on the unit ball. Michigan Math. J,
39:509–536, 1992.
[68] I. Privalov. Sur une généralization du théorème de Fatou. Rec. Math. (Mat.
Sbornik), 31:232–235, 1923.
[69] T. Ransford. Potential Theory in the Complex Plane. London Math. Soc. Student
Texts 28, Cambridge University Press, 1995.
[70] J. Riihentaus. On a theorem of Avanissian–Arsove. Exposition. Math., 7:69–72,
1989.
[71] H. L. Royden. Real Analysis. Macmillan Publishing Co., New York, NY, third
edition, 1988.
[72] W. Rudin. Function Theory in the Unit Ball of Cn . Springer-Verlag, New York,
NY, 1980.
[73] H. Samii. Les transformations de Poisson dans le boule hyperbolic. PhD thesis,
Université Nancy 1, 1982.
[74] I. Sokolnikoff. Tensor Analysis. Wiley, New York, NY, 1964.
[75] E. M. Stein. Singular Integrals and Differentiability Properties of Functions.
Princeton University Press, Princeton, NJ, 1970.
[76] E. M. Stein and G. Weiss. Fourier Analysis on Euclidean Spaces. Princeton
University Press, Princeton, NJ, 1971.
[77] S. Stević. A Littlewood–Paley type inequality. Bull. Braz. Math. Soc., 34:1–7,
2003.
[78] M. Stoll. Hardy-type spaces of harmonic functions on symmetric spaces of
noncompact type. J. Reine Angew. Math., 271:63–76, 1974.
[79] M. Stoll. Mean value theorems for harmonic and holomorphic functions on
bounded symmetric domains. J. Reine Angew. Math., 290:191–198, 1977.
[80] M. Stoll. Boundary limits of Green potentials in the unit disc. Arch. Math.,
44:451–455, 1985.
[81] M. Stoll. Rate of growth of pth means of invariant potentials in the unit ball of Cn .
J. Math. Analysis & Appl., 143:480–499, 1989.
[82] M. Stoll. Rate of growth of pth means of invariant potentials in the unit ball
of Cn , II. J. Math. Analysis & Appl., 165:374–398, 1992.
[83] M. Stoll. Tangential boundary limits of invariant potentials in the unit ball of Cn .
J. Math. Anal. Appl., 177(2):553–571, 1993.
[84] M. Stoll. Invariant potential theory in the unit ball of Cn , volume 199 of
London Mathematical Society Lecture Note Series. Cambridge University Press,
Cambridge, 1994.
[85] M. Stoll. Boundary limits and non-integrability of M-subharmonic functions in
the unit ball of Cn (n ≥ 1). Trans. Amer. Math. Soc., 349(9):3773–3785, 1997.
220 References
[86] M. Stoll. Weighted tangential boundary limits of subharmonic functions on

domains in Rn (n ≥ 2). Math. Scand., 83(2):300–308, 1998.
[87] M. Stoll. On generalizations of the Littlewood–Paley inequalities to domains
in Rn (n ≥ 2). Unpublished manuscript, 2004. www.researchgate.net/profile/
Manfred Stoll/publications.
[88] M. Stoll. The Littlewood–Paley inequalities for Hardy–Orlicz spaces of harmonic
function on domains in Rn . Advanced Studies in Pure Mathematics, 44:363–376,
2006.
[89] M. Stoll. Weighted Dirichlet spaces of harmonic functions on the real hyperbolic
ball. Complex Var. and Elliptic Equ., 57(1):63–89, 2012.
[90] M. Stoll. On the Littlewood–Paley inequalities for subharmonic functions on
domains in Rn . In Recent Advances in Harmonic Analysis and Applications, pages
357–383. Springer–Verlag, New York, NY, 2013.
[91] M. Stoll. Littlewood–Paley theory for subharmonic functions on the unit ball in
Rn . J. Math. Analysis & Appl., 420:483–514, 2014.
[92] N. Suzuki. Nonintegrability of harmonic functions in a domain. Japan J. Math.,
16:269–278, 1990.
[93] D. Ullrich. Möbius-invariant potential theory in the unit ball of Cn . PhD thesis,
University of Wisconsin, 1981.
[94] D. Ullrich. Radial limits of M-subharmonic functions. Trans. Amer. Math. Soc.,
292:501–518, 1985.
[95] J.-M. G. Wu. Lp densities and boundary behavior of Green potentials. Indiana
Univ. Math. J., 28:895–911, 1979.
[96] S. Zhao. On the weighted Lp -integrability of nonnegative M-superharmonic
functions. Proc. Amer. Math. Soc, 113:677–685, 1992.
[97] K. Zhu. Möbius invariant Hilbert spaces of holomorphic functions in the unit ball
of Cn . Trans. Amer. Math. Soc., 323:823–842, 1991.
[98] L. Ziomek. On the boundary behavior in the metric Lp of subharmonic functions.
Studia Math., 29:97–105, 1967.
[99] A. Zygmund. Trigonometric Series. Cambridge University Press, London, 1968.
Index of Symbols
A(r) ≈ B(r), 24 Mα f , Mrad f , 120

AR , 156 Mδ μ(ζ ), 159
B(a, r), 1 Mμ, Mf , 115
Bα f (x), 178 Nf (x), 28
B f (x), 177 O(n), 2
C(), 2 P(a, t), 3
Cc (), 2 Pe (x, t), 62
Ck (), 2 Ph (a, t), 62
Ck (), 2 Ph [ f ], 65
C∞ (), 2 Ph [μ], 65
dB p ( f , g), dD p ( f , g), 176 PH [ f ](z), 78
γ γ
Rγ (x, y), 215
dh (a, b), 10
S(a, r), 1
dH (x, y), 14
S(ζ , δ), 114
dph (a, b), 12
Sα (h, ζ ), 207, 213

D, 1 Sn,α (r), 84
D, 1 x∗ , 3
Dν, Dν, 119 Zm (η, ζ ), 86
Dα f , 2 α (ζ ), 120
E(a, r), 11
β (x), 134, 164
E \ F, 1 ∂D, 1
E∗ , 4 Δh , 17
F(a, b; c; t), 83 Δσ , 28
f ∗ g, 24 ρ(x, a), 7
f + , 106 τ , 23
f , 31 τH , 30
f p , 24, 78 φ̃, 13
f ∞ , 69 ϕa , 7
Ge (x, y), 21 χE , 118
Gh (x, y), 21 ψ (x), 2
gH ( y), GH ( y, yo ), 29 ∇ h , 17
Jψ , 2 B, 1
L p (B, τ ), 24 D, 7, 35
L p (Rn−1 ), 78 H, Hn , 13
p
Lloc (B), 24 R̂n , 1
LH f , 29 S, 1
221
222 Index of Symbols
p
Bγ , Bp , 173 M(B), 8
p
Dγ , 173 t), 62
P(z,
Hm (Rn ), Hm (S), 82 S p , 96
Hp , 97 Sϕ , 109
p
Hλ , 97 disth (A, B) , 104
Hλ , 71 Int (D), 1
Hϕ , 109 supp f , 2
Index
Ahern, P., xiv, 40, 90 Dirichlet space, 173, 212

Ahlfors, L. V., xii distribution function, 137
Aikawa, H., 170 Djordjević, O., 132, 205
approximate identity, 45, 100
Arazy, J., 179 equicontinuous, 99
Armitage, D. H., 57, 140, 149, 169 Euclidean Green’s function, 21
Arsove, M., 163, 170
Axler, S., 82, 134 Fatou’s theorem, 125
Fefferman, C., 52
Beardon, A. F., xii Fisher, S., 179
Berezin transform, 177 Flett, T. M., 202, 205
Bergman space, 173 Flores, M., 40
Blaschke fundamental solution, 20
condition, 167 Furstenberg, H., 98
product, 167, 171
Bloch space, 214 Gardiner, S., 57, 172
little, 214 Garding, L., 98, 110
Borel sets, 65 Gehring, F. W., 171
boundary general Möbius group, 4
function of f , 97 gradient, 17
measure of f , 97, 106 invariant, 17
Bourdon, P., 82, 134 Green potential, 56, 147
Bruna, J., xiv, 90 of f , 163
Green’s formula, 59
Calderon, A. P., 127 Green’s function, 21
capacity, 172 on H, 29
Cascante, C., xiv, 90 Green’s identity, 48
Cifuentis, P., 98, 207 Grellier, S., xv, 125, 213
Cima, J., 163, 169
convex, 42 H-harmonic, 31
convolution, 24 majorant, 103
H-subharmonic, 31, 39
Dirichlet problem, 64 H-superharmonic, 39
for Br , 68 Haar measure, 22, 32, 98
for H, 78 Hahn, K. T., 172
223
224 Index
Hallenbeck, D., 171 Korenblum, B., 177

Hardy, G. H., 199 Kuran, Ü., 52
Hardy space, 97
Hardy–Orlicz space, 109 Laplace–Beltrami, 17
Hardy-Stein identity, 146 Laplacian, 17
Hardy-type space of H-subharmonic invariant, 17
functions, 96 least H-harmonic majorant, 103
harmonic majorant, 97 Lebesgue point, 118
harmonic polynomials, 82 Littlewood, J. E., 199
Harnack’s inequality, 104 Littlewood–Paley inequalities, 205, 212
Hedenmalm, H., 177 local Fatou theorem, 127
Heins, M., 171 locally integrable, 43
homogeneous, 82 locally p-integrable, 24
Hörmander, L., 98, 110 lower semicontinuous, 37
Huber, A., 163, 170 Luecking, D. H., 172, 205
hyperbolic Lusin, N., 213
ball, 11
distance, 10 Möbius
half-space, 12 group of B, 8
length, 9 invariant Laplacian, 18
metric, 9 invariant measure, 23
hypergeometric invariant space, 178, 214
equation, 83 Möbius transformation, 4, 6
function, 83 of B, 6, 8
of H, 16
invariant Marcinkiewicz, J., 213
convolution, 24 maximal function
gradient, 17 of f , 115
gradient on H, 29 of μ, 114
Green potential, 147 maximum principle, 34, 42
harmonic, 31 mean-value property, 33
Laplacian, 17, 18 Minemura, K., xi, 70, 90
Laplacian on H, 29 Mizuta, Y., 170
mean-value property, 33
measure, 23 Nagel, A., 170
measure on H, 30 non-tangential
Poisson kernel, 62 approach region, 120
subharmonic, 31 limit, 125
inversion limit in L p , 169
in plane, 3 maximal function, 120
in sphere, 2 non-tangentially bounded, 131
in the unit sphere S, 3 norm, 101
normal derivative, 59
Jacobian, 2 operator, 28
Jaming, P., xiii, xv, 62, 90, 98, 125,
213 one-point compactification, 1
Jensen’s inequality, 42 orthogonal transformation, 2
Jevtić, M., xv
p-integrable, 24
Knapp, A. W., 125 p-norm, 101
Koranyi, A., xi, 98, 127 Pavlović, M., 51, 52, 132, 199, 205
Index 225
Peloso, M., 179 Singman, D., 172

Poincaré extension, 13 spherical harmonic, 82
point of density, 118 square area integral, 207, 213
point mass measure, 162, 166 Stanton, C. S., 163, 169
Poisson integral, 65 Stein, E. M., 52, 117
Poisson kernel, 62, 77 Stević, S, 205
on H, 76 strongly convex, 109
positive part of f , 106 support of f , 2
Privalov, I., xiv, 127 support of a measure, 57
projection, 157 Suzuki, N., 52, 188
pseudo-hyperbolic metric, 12
Putz, R. P., 127 tangential approach region, 170
tangential limit, 170
quasi-nearly H-subharmonic, 51 in L p , 170
quasi-nearly subharmonic, 52
Ullrich, D., 98, 140, 155
radial function, 31
uniformly integrable, 113
radial maximal function, 120
upper derivate, 119
radialization, 31
upper half-space, 13
Ramey, W., 82, 134
upper semicontinuous, 37
Ransford, T., 37
reflection, 2
regular Borel measure, 56 weak Laplacian, 48
reproducing kernel, 214 weak sense, 48
Riesz Decomposition Theorem, 143 weighted boundary limits, 171
Riesz measure, 50 Wu, J.-M. G., 170
Riihentaus, J., 51, 52
Rudin, W., xi, 31, 40, 68, 70, 71, 127, 170 Zhao, S., 149
Zhu, K., 177, 179
Samii, H., 64 Ziomek, L., 169
seminorm, 178 zonal harmonic, 86
Shapiro, J. H., 170 Zygmund, A., 213

Stoll, Manfred - Harmonic and Subharmonic

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Stoll, Manfred - Harmonic and Subharmonic

Uploaded by

Copyright:

Available Formats

LONDON MATHEMATICAL SOCIETY LECTURE NOTE SERIES

Managing Editor: Professor M. Reid, Mathematics Institute,

Harmonic and Subharmonic

Cambridge University Press is part of the University of Cambridge.

5 The Poisson Kernel and Poisson Integrals 59

10 Bergman and Dirichlet Spaces of H-Harmonic Functions 173

The intent of these notes is to provide a detailed and comprehensive treatment

Möbius self-maps of the unit ball B in Rn . As in [72] we define a family {ϕa :

A real-valued C2 function f on B is defined to be either H-harmonic or

with equality if and only if f is H-harmonic on B. In the above, S is the unit

Ph (a, t) = − lim nrn−1 (1 − r2 )2−n ∇Ga (rt), t,

In Chapter 6 we consider the spherical harmonic expansions of H-harmonic

where Sn,α is given by a hypergeometric function. Interestingly, when n is

In this chapter we provide a brief review of Möbius transformations on

|α| = α1 + · · · + αn , α! = α1 ! · · · αn !, xα = x1α1 · · · xnαn ,

If is an open subset of Rn , we denote by Ck ( ), k = 0, 1, 2, . . . the set of

If ψ(x) = (ψ1 (x), . . . , ψn (x)) is a C1 mapping of an open subset of Rn into

and the Jacobian Jψ of the transformation ψ is given by Jψ (x) = det ψ  (x).

1.2 Inversion in Spheres and Planes

1.3 Möbius Transformations

Clearly the composition of two Möbius transformations is again a Möbius

the mapping ψ clearly maps a “sphere” onto a “sphere.” To show that an

E = {x ∈ Rn : b|x|2 − 2x, a + c = 0},

where b and c are real and a ∈ Rn . By convention, ∞ satisfies this equation

Möbius Self-Maps of the Unit Ball

In this chapter we will provide a characterization of the Möbius

(1 − |z|2 )(1 − |w|2 )

2.1 Möbius Transformations of B

ψa (x) = a + (1 − |a|2 )(a − x)∗ . (2.1.1)

Since the mapping ψa is a composition of Möbius transformations, ψa is a

|a − x|2 + (1 − |a|2 )(1 − |x|2 )

Since the following theorem is well known, we state it without proof.

Theorem 2.1.1 Let ψ be a Möbius transformation of R̂n satisfying ψ(0) = 0

We denote by M(B) the set of all Möbius transformations of B onto B.

Theorem 2.1.2 For a ∈ B, let ϕa be defined by (2.1.6). Then

ϕa (0) = a, ϕa (a) = 0, and ϕa (ϕa (x)) = x

ψ  (0) = ϕa (a)ϕa (0) = I.

Hence A = I and thus ϕa (ϕa (x)) = x for all x ∈ B.

Prior to introducing the hyperbolic metric on B we prove an identity for

Theorem 2.1.3 If ψ ∈ M(B), then for all x, y ∈ B,

|ψ(x) − ψ(y)|2 |x − y|2

σa (x) = a∗ + (|a∗ |2 − 1)(x − a∗ )∗ . (2.1.10)

2.2 The Hyperbolic Metric on B

As a consequence of (2.2.2) and Theorem 2.2.1, for a, b ∈ B,

Thus E(a, r) is a hyperbolic ball with hyperbolic center a and hyperbolic

Theorem 2.2.2 For a ∈ B and 0 < r < 1, E(a, r) = B(ca , ρa ) where

ϕa (Sr ) = {x : (1 + |a|2 )|x|2 − 2a, x = 0}.

We conclude this section with the pseudo-hyperbolic metric on B. For

Theorem 2.2.3 dph is a metric on B.

2.3 Hyperbolic Half-Space H

Definition 2.3.1 For n ≥ 2, the upper half-space H or Hn in Rn is defined by

For each x ∈ Rn−1 , let x̃ ∈ Rn be defined by

x̃ = (x, 0) = (x1 , . . . , xn−1 , 0).

For each inversion φ on R̂n−1 , we define an inversion φ̃ acting on R̂n as

The function φ̃ is called the Poincaré extension of φ.

If [φ̃(x)]j denotes the jth coordinate function of φ̃(x), then

As a consequence the mapping φ̃ leaves

We conclude this section by considering the Möbius transformation  that

Since  is an inversion,  is a one-to-one map of R̂n onto R̂n satisfying

φ  (x) = |x|−4 [|x|2 I − 2Q(x)],

Exercises on the Upper Half-Space H

iv.  maps S \ {en } onto Rn−1 and Rn−1 onto S \ {en }.

The Invariant Laplacian, Gradient, and Measure

Ph (a, t) = − lim nrn−1 (1 − r2 )2−n ∇Ga (rt), t,

If is an open subset of Rn , we denote by Ck (), k = 0, 1, 2, . . . the set of

and the Jacobian Jψ of the transformation ψ is given by Jψ (x) = det ψ (x).

E = {x ∈ Rn : b|x|2 − 2x, a + c = 0},

ψ (0) = ϕa (a)ϕa (0) = I.

ϕa (Sr ) = {x : (1 + |a|2 )|x|2 − 2a, x = 0}.

We conclude this section by considering the Möbius transformation that

Since is an inversion, is a one-to-one map of R̂n onto R̂n satisfying

φ (x) = |x|−4 [|x|2 I − 2Q(x)],

iv. maps S \ {en } onto Rn−1 and Rn−1 onto S \ {en }.

Δh f (a) = (1 − |a|2 )2 Δf (a) + 2(n − 2)(1 − |a|2 )a, ∇f (a). (3.1.2)

Theorem 3.1.2 For f ∈ C2 () and ψ ∈ M(B),

Taking the supremum over all h with hq ≤ 1 gives

The inequality f ∗ g∞ ≤ f p gq is an easy consequence of Hölder’s

Δh ( fg) = f Δh g + gΔh f + 2∇ h f , ∇ h g.

Δh f (g(x)) = f (g(x))|∇ h g(x)|2 + f (g(x))Δh g(x).

∇x f (ϕx (a))|x=0 = −2a, ∇f (−a)a+(1+|a|2 )∇f (−a). (3.5.3)

where : H → B is given by (2.3.2). Show that

3.5.14. As in the previous exercise, for yo , y ∈ H, set GH ( y, yo ) = Gh (( y),

where dA is the Haar measure on O(n). Clearly f is a radial function on B,