Professional Documents
Culture Documents
Stoll, Manfred - Harmonic and Subharmonic
Stoll, Manfred - Harmonic and Subharmonic
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
M A N F R E D S TO L L
University of South Carolina
University Printing House, Cambridge CB2 8BS, United Kingdom
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
References 216
Index of Symbols 221
Index 223
Preface
xi
xii Preface
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
Lp ((1 − |x|2 )γ dτ (x)), 0 < p < ∞, γ > 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
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
n
1 i = j,
ai,k aj,k = δi, j =
k=1
0, i = j.
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
ψ(x) = x + λa,
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
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
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)
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.
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.
|a − b|2
sinh2 12 dh (a, b) = , (2.2.4)
(1 − |a|2 )(1 − |b|2 )
2.2 The Hyperbolic Metric on B 11
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)
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 ).
12 Möbius Self-Maps of the Unit Ball
H = {x ∈ Rn : xn > 0}.
φ̃(x) = ã + r2 (x − ã)∗ .
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
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
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).
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
and
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
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).
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) ≥ 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
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).
3.3 The Invariant Measure on B 23
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.
(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
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
24 The Invariant Laplacian, Gradient, and Measure
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 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
1 1
q
≤ g1 p
f ( y) g(x)(h ◦ ϕy ) (x)dτ (x)
q
dτ ( y).
B B
26 The Invariant Laplacian, Gradient, and Measure
1
1
q
( f ∗ g)h1 ≤ g1 f p
p
g(x)(h ◦ ϕy ) (x)dτ (x)dτ ( y)
q
.
B B
Therefore,
1 1
( f ∗ g)h1 ≤ g1p f p g1q hq .
f ∗ gp ≤ f p g1 .
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].
ϕϕa ( y) ◦ ϕa ◦ ϕy = A
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
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
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,
∂yj
= δi,j (|a|2 + 1) − 2ai aj ,
∂xi
x=0
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)),
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
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
Therefore
d 1
f (ρζ ) dσ (ζ ) = ρ 1−n (1 − ρ 2 )n−2 Δh f (x) dτ (x),
dρ S n Bρ
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
Thus h(x) < for all x ∈ . Finally, since f (x) ≤ h(x) for all x, letting → 0
gives f (x) ≤ 0 on .
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.
4.3 H-Subharmonic Functions 37
If f (xo ) > −∞, then the above definition is equivalent to given > 0, there
exists a δ > 0 such that
is upper semicontinuous on S.
(c) If { fα }α∈A are upper semicontinuous on S, then
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
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 + .
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.
4.3 H-Subharmonic Functions 39
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,
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.
( f ◦ ψ)(ϕa (rt)) dσ (t) = f (ϕb (Art)) dσ (t)
S
S
= f (ϕb (rt)) dσ (t) ≥ f (b) = ( f ◦ ψ)(a).
S
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)
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
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
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.
44 H-Harmonic and H-Subharmonic Functions
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 45
for all x ∈ 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 .
( f ∗ χk ) ∗ r = ( f ∗ r ) ∗ χk ≥ f ∗ χk .
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
= 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).
holds for all ψ ∈ Cc (B) with ψ ≥ 0, then there exists an H-subharmonic
2
( f ∗ [(Δh ψ) ◦ J])(0) ≥ 0
for all ψ ∈ Cc2 (B) with ψ ≥ 0. It follows from the definition that
fm ∗ χj = ( f ∗ χm ) ∗ χ j = ( f ∗ χj ) ∗ χ m
≥ ( f ∗ χj ) ∗ χ k = f k ∗ χj .
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 51
Thus
1
f p (a) ≤ f p (y)dτ (y).
rn E(a,r)
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 )).
52 H-Harmonic and H-Subharmonic Functions
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 ))
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)
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−)
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
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)
56 H-Harmonic and H-Subharmonic Functions
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
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
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.
58 H-Harmonic and H-Subharmonic Functions
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?
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
nrn−1
dσ̃ (rζ ) = dσ (ζ ). (5.1.3)
(1 − r2 )n−1
Clearly
lim gh Dñ (u ◦ ϕa )dσ̃ = 0.
→0 S
(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
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),
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
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.
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
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
Therefore,
1 1
I(r) = η(r, s)ds ≤ cn (1 − r ) 2
(1 − sr2 )−2 sn/2−1 ds,
0 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
1 Recall that the Borel subsets of S are the smallest σ -algebra of subsets of S such that every
continuous function is measurable.
66 The Poisson Kernel and Poisson Integrals
≤ 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
Ph [ f ](x), x ∈ B,
F(x) =
f (x), x ∈ S,
Similarly, set
Sr = {y : |y| = r} = {rt : t ∈ S}.
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
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.
70 The Poisson Kernel and Poisson Integrals
The last equality follows by (5.4.2) since Ph (·, t) ∈ H(Br ). Thus, as above,
μ = 0, that is, ν = σ .
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].
5.5 Eigenfunctions of Δh 71
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).
72 The Poisson Kernel and Poisson Integrals
Δh f (x) = 4r2 (1 − r2 )2 u (r2 ) + 2n(1 − r2 )2 u (r2 ) + 4(n − 2)r2 (1 − r2 )u (r2 ).
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)
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.
As a consequence, f is C∞ on B.
If in (5.5.9) we compute Δh with respect to y we obtain
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)
The following asymptotic estimates for Iγ appear to be well known, but are
usually stated without a reference to a proof.
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
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
If γ = (n − 1), then
2/(1−r)2 1 1
Iγ (r) ≤ Cn Cγ + dw ≤ Cn,γ log .
1 w (1 − r)
76 The Poisson Kernel and Poisson Integrals
Corollary 5.5.8
⎧
⎨ (1 − |x| ) 1
⎪ if α < 12 ,
2 α(n−1)
(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.
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
78 The Poisson Kernel and Poisson Integrals
PH [ f ](z), z ∈ H,
F(z) =
f (x), x ∈ Rn−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
that is, for each f ∈ L2 (S), there exists pm ∈ Hm (S) such that
∞
f = pm , (6.0.1)
m=0
82
6.1 Dirichlet Problem for Spherical Harmonics 83
2(1 − r2 )r2 g (r2 ) + {(n + 2α)(1 − r2 ) + 2r2 (n − 2)}g (r2 ) + α(n − 2)g(r2 ) = 0,
or
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)
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.
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)
6.1 Dirichlet Problem for Spherical Harmonics 85
( 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],
m
Ph [q](x) = Ph [pk ](x).
k=0
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 ).
Ph [q](x) = 1
3 + S3,2 (|x|)(x12 − 13 |x|2 ).
86 Spherical Harmonic Expansions
Unfortunately, however, there is no simple expression for S3,2 (r). The function
S3,2 (r) is given by
Thus
∞
(k + 1)(k − 12 )
Ph [t12 ](x) = 1
3 − 32 (x12 − 13 |x|2 ) |x|2k .
k=0
(k + 72 )
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
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!
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
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)!
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)
6.2 Zonal Harmonic Expansion of the Poisson Kernel 89
where Cn and Dn are constants depending only on n. Also, for all ζ , t ∈ S,
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
∞ (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
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
= (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 ).
92 Spherical Harmonic Expansions
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!
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
(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
6.3 Spherical Harmonic Expansion of H-Harmonic Functions 93
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
k=0
(α + 12 n)k k
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
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→∞
94 Spherical Harmonic Expansions
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
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
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
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
Hp = { f : f is H-harmonic with | f | ∈ S p }.
96
7.1 Poisson Integral Formula for Functions in Hp , 1 ≤ p ≤ ∞ 97
with f p = fˆp .
(b) If p = 1, then there exists a Borel measure νf on B such that
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
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)
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
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 )| <
100 Hardy-Type Spaces of H-Subharmonic Functions
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
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→∞
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
102 Hardy-Type Spaces of H-Subharmonic Functions
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)
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→∞
≤ 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 .
The following theorem provides necessary and sufficient conditions for the
existence of a least H-harmonic majorant.
For the proof of the theorem we will need the following generalization of
Harnack’s inequality for H-harmonic functions.
104 Hardy-Type Spaces of H-Subharmonic Functions
Since
lim Fn (0) = lim f (rn t)dσ (t) < ∞,
n→∞ n→∞ S
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
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.
Likewise
Hϕ = { f : f is H-harmonic and | f | ∈ Sϕ }.
Since νs ≤ 0,
f (x) ≤ Ph [ fˆ](x). (7.4.4)
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 < p ≤ 1, but not for
any p > 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
114
8.1 Maximal Functions 115
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
For the proof of Theorem 8.1.2 we require the following lemma.
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
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
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
118 Boundary Behavior of Poisson Integrals
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 ),
8.1 Maximal Functions 119
Tf (ζ ) ≤ | f (ζ ) − g(ζ )| + M[ f − g](ζ ).
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 < p < ∞, α > 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.
8.2 Non-tangential and Radial Maximal Function 121
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
But
4 8
|x − y| ≤ 2rx < (1 − |x|2 ) ≤ (1 − |x|).
3 3
Also, by Exercise 2.4.1(b),
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|))
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|))
Theorem 8.2.3 For every α > 1, there exists a constant Aα such that
Mα Ph [μ] ≤ Aα Mμ
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},
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
8.2 Non-tangential and Radial Maximal Function 123
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 |.
|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)
for any signed Borel measure μ on S. We now prove that the reverse inequality
also holds for positive Borel measures on S.
Mμ ≤ Cn Mrad Ph [μ]
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))
from which the result now follows.
f p ≤ Mα f p ≤ Ap,α f p .
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.
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].
126 Boundary Behavior of Poisson Integrals
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 )
>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).
8.4 A Local Fatou Theorem for H-Harmonic Functions 129
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
lim Ph [ν](x) = 0.
x→ζ
x∈β (ζ )
But
|Ph [χE ](x) − χE (ζ )| ≤ Ph [ν](x).
Therefore
Therefore, for x, y ∈ K,
| f (x) − f (y)| ≤ |h(ϕx (w)) − h(ϕy (w))|dτ (y),
B
1
|u − gj | ≤ v on ∂( ∩ Brj ). (8.4.2)
c
8.5 An L p Inequality for Mα f for 0 < p ≤ 1 131
Note: The term “weakly” is used to emphasize that (8.4.4) holds only for one
α > 1.
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.
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 )
N
| f (rζ )|p ≤ | f (0)|p + C sup |∇ h f (tζ )|p .
k=1 t∈[rk−1 ,rk )
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 )
8.5 An L p Inequality for Mα f for 0 < p ≤ 1 133
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 )
134 Boundary Behavior of Poisson Integrals
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
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].
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.
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.
136 Boundary Behavior of Poisson Integrals
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
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 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
139
140 The Riesz Decomposition Theorem for H-Subharmonic Functions
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
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 }.
As a consequence,
ψdμh = (ψ ◦ ϕx )dμf
B B
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 ,
which proves the result.
Proof. Set ψ(x) = 1 − (1 − |x|2 )n−1 . Using the radial form of Δh we have
Therefore,
Gh (x, y)Δh ψ(x)dτ (x) = 2n(n − 1) Gh (x, y)dν(x).
B B
144 The Riesz Decomposition Theorem for H-Subharmonic Functions
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.
146 The Riesz Decomposition Theorem for H-Subharmonic Functions
Hence
f pp = Hf p (0) = f (0) +
p
gh (y)dμf p (y) (9.2.3)
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
9.2 Applications of the Riesz Decomposition Theorem 147
Proof. If h satisfies the hypothesis, then |h|p ∈ C2 (B), 1 < p < ∞, with
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,
Therefore
lim sup V2 (rt)dσ (t) < .
r→1 S
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
U p (x)dν(x) ≤ A(n, p, a)U p (a)
B
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
However, the above integral is finite if and only if p < n/(n − 1).
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) .
152 The Riesz Decomposition Theorem for H-Subharmonic Functions
Suppose now that p > 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
≤ 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
9.3 Integrability of H-Superharmonic Functions 153
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),
from which the result now follows.
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 < p ≤ 1 we have
p
Gμ (y)dν(y) ≤
p
Gμ (y)dν(y)
B B
, -p
1
= n rn−1 Gμ (rt)dσ (t)dr ≤ Gpμ (0).
0 S
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
9.4 Boundary Limits of Green Potentials 155
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
156 The Riesz Decomposition Theorem for H-Subharmonic Functions
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
and thus
≤ Cn Aα M[νR ](ζ ),
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 )
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)
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 }.
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
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
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 )
Thus
M rad V1 (ζ ) ≤ Cn 2j Mδj μ(ζ ).
j=n−2
9.4 Boundary Limits of Green Potentials 161
By the above,
∞
λ −(1+α)j
Eλ ⊂ ζ ∈ S : Mδj μ(ζ ) > 2 .
Cn
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
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
where
Vf (x) = Gh (x, y)f (y)dν(y).
B
≤ (c + 23 )(1 − |x|2 ).
The second part of the theorem follows immediately from the fact that
Sα∗ μ ∈ L1 (S).
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)
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
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 )
∞
, -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
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
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
lim V(x) = L
x→ζ
x∈α,τ (ζ )
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
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.
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
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
10.1 Properties of Dγ and Bγ 175
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 < p < ∞, and Banach spaces for 1 ≤ p < ∞.
We close this section with two integral formulas for functions in Bγ1 , γ >
(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
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
f ◦ ϕ = f for all f ∈ X, ϕ ∈ M.
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,
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 < p < ∞, and all γ > 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)
χEδ (y) (x)(1 − |y|2 )γ dτ (y) ≈ (1 − |x|2 )γ dτ (y) = (1 − |x|2 )γ τ (Bδ ),
B Eδ (x)
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
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)
(b) For γ > (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
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)
|∇ 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
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
184 Bergman and Dirichlet Spaces of H-Harmonic Functions
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
| f (ρζ )| p dσ (ζ ) ≤ | f (0)| p + C (1 − |x|)p−1 ρ p |∇f (ρx)|dν(x),
S B
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.
186 Bergman and Dirichlet Spaces of H-Harmonic Functions
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 < p < ∞, whenever γ > 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 .
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
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-
+
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
The proof of part (b) of Theorem 10.4.1 also proves the following.
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)
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
10.4 Integrability of Functions in Bγ and Dγ 189
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 < p < 12 ,
1
f γ ,p ≈
p
rn−1 (1 − r2 )γ −n+p(n−1) dr,
0
and this integral is finite if and only if γ > (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 < 1, let f (x) = Ph (x, e1 ). Then f is a non-negative H-subharmonic
p
function which by the previous example is in Lγ whenever γ > p(n − 1).
Suppose 0 < 2 < p and γ > 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
190 Bergman and Dirichlet Spaces of H-Harmonic Functions
Remark 10.4.7 As mentioned in Remark 10.4.5, N. Suzuki proved that for 0 <
p < 1 and γ > 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 < 1, then for each γ > 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}.
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),
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
Ph [q](x) = 1
3 + (x12 − 13 |x|2 )S3,2 (|x|),
where
S3,2 (r) = c3,2 F(2, − 12 ; 72 ; 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
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
Thus if γ ≤ (n − 1) − p we have
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.
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 < p < ∞ and suppose λ is real with λ ≥ −(n − 1)2
and λ = 0. If f is a radial function on B with f (0) = 0, then
f ∈ Lγp (τ ) ∩ Hλ
if and only if γ > (n − 1) + p2 λ + (n − 1)2 − (n − 1) .
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
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)
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
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 < p < 1. Consider the function Phα (x, e1 ) with α = 1
+
2
2(n−1) (n − 1) + λ. Then,
1 2
1
2 γ −n pα
Phα γp, p =n r n−1
(1 − r ) Ph (rt, e1 )dσ (t) dr
0 S
1
=n rn−1 (1 − r2 )γ −n gpα (r)dr.
0
If pα > 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
10.5 Integrability of Eigenfunctions of Δh 197
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 < p < 1 and let λ be real with λ > −(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
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.
198 Bergman and Dirichlet Spaces of H-Harmonic Functions
p
Our final result of this section concerns the space Hλ . Recall that for 0 <
p < ∞,
p
Hλ = f ∈ Hλ : sup | f (rt)| p dσ (t) < ∞ .
0<r<1 S
Example 10.5.6 The conclusion of the previous theorem is false when 0 <
p < 1. The function Phα (x, e1 ) is an eigenfunction of Δh with eigenvalue λα =
4(n − 1)2 α(α − 1). It is well known that Ph ∈ H p for all p, 0 < p < 1. Hence
p
if α > 1, Ph ∈ Hλα for all p, 0 < p ≤ α1 .
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 < p < ∞, and α > 0, then
# $
Mp ( f , r) = O (1 − r)−α ⇐⇒ Mp ( f , r) = O (1 − r)−(α+1) .
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 )
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)
200 Bergman and Dirichlet Spaces of H-Harmonic Functions
m−1
≤C sup{|∇ h f (sζ )| : s ∈ (rj , rj+1 )}.
j=0
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
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 )
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 .
Therefore,
− 1p
T fˆ(r) ≤ Cω(r) fˆp .
Hence,
C p fˆp
p
ˆ
{r : T f (r) > α} ⊂ r : ω(r) < .
α
where
b = min 1, (C p fˆpp α −p ) .
≤ 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 < p < ∞. Then
(n−1)
pn pn
| f (x)| (n−1) dν(x) ≤ Ap f p ,
B
pn
that is, H p ⊂ B (n−1) for all p > 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 < ∞, then for all q > 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 < p < ∞,
p
γ > (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 205
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 <
p < ∞, has been given recently by M. Pavlović in [65].
The Littlewood–Paley inequalities are also known to be valid for harmonic
functions in the unit ball in Rn when p > 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
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 )
where
1 r (1 − s2 )n−2
g(|x|, r) = ds.
n |x| sn−1
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
Therefore,
(Δh f (y)) dτ (y) ≤ cn,δ ρ
p 2−2p
Δh f p (w)dτ (w),
Bρ/2 Bρ
Therefore,
Δh f p (y)dτ (y) ≤ Cn,δ ρ 2p−2 (Δh f (x)) p dτ (x).
Bρ/4 Bρ
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
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) .
10.8.4. (*) For 0 < p < 12 and γ > 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
f p ≤ C1 f Dγ ,
and that the reverse inequality holds for p > 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 )
(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
216
References 217
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
223
224 Index