Morrey Spaces

Monographs and Research Notes in Mathemacs

Series Editors:
John A. Burns, Thomas J. Tucker, Miklos Bona, Michael Ruzhansky

About the Series

This series is designed to capture new developments and summarize what is known over the
en!re field of mathema!cs, both pure and applied. It will include a broad range of monographs
and research notes on current and developing topics that will appeal to academics, graduate
students, and prac!!oners. Interdisciplinary books appealing not only to the mathema!cal
community, but also to engineers, physicists, and computer scien!sts are encouraged.

This series will maintain the highest editorial standards, publishing well-developed monographs
as well as research notes on new topics that are final, but not yet refined into a formal
monograph. The notes are meant to be a rapid means of publica!on for current material where
the style of exposi!on reflects a developing topic.

Spectral Methods Using Mulvariate Polynomials On The Unit Ball

Kendall Atkinson, David Chien, and Olaf Hansen

Glider Representaons
Frederik Caenepeel, Fred Van Oystaeyen

La!ce Point Idenes and Shannon-Type Sampling

Willi Freeden, M. Zuhair Nashed

Summable Spaces and Their Duals, Matrix Transformaons and Geometric Properes
Feyzi Basar, Hemen Du!a

Spectral Geometry of Paral Differenal Operators (Open Access)

Michael Ruzhansky, Makhmud Sadybekov, Durvudkhan Suragan

Linear Groups: The Accent on Infinite Dimensionality

Martyn Russel Dixon, Leonard A. Kurdachenko, Igor Yakov Subbo"n

Morrey Spaces: Introducon and Applicaons to Integral Operators and PDE’s, Volume I
Yoshihiro Sawano, Giuseppe Di Fazio, Denny Ivanal Hakim

Morrey Spaces: Introducon and Applicaons to Integral Operators and PDE’s, Volume II
Yoshihiro Sawano, Giuseppe Di Fazio, Denny Ivanal Hakim
For more informa!on about this series please visit: h$ps://www.crcpress.com/Chapman--HallCRC-Monographs-and-Research-Notes-in-Mathema!cs/book-
series/CRCMONRESNOT
 Morrey Spaces
Introduction and Applications to
Integral Operators and PDE’s,
Volume I

Yoshihiro Sawano
Chuo University
Giuseppe Di Fazio
University of Catania
Denny Ivanal Hakim
Bandung Institute of Technology
First edition published 2020
by CRC Press
6000 Broken Sound Parkway NW, Suite 300, Boca Raton, FL 33487-2742

and by CRC Press

2 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN

c 2020 Taylor & Francis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, LLC

Reasonable efforts have been made to publish reliable data and information, but the author
and publisher cannot assume responsibility for the validity of all materials or the conse-
quences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if
permission to publish in this form has not been obtained. If any copyright material has not
been acknowledged please write and let us know so we may rectify in any future reprint.

Except as permitted under U.S. Copyright Law, no part of this book may be reprinted,
reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other
means, now known or hereafter invented, including photocopying, microfilming, and record-
ing, or in any information storage or retrieval system, without written permission from the
publishers.

For permission to photocopy or use material electronically from this work, access
www.copyright.com or contact the Copyright Clearance Center, Inc. (CCC), 222 Rose-
wood Drive, Danvers, MA 01923, 978-750-8400. For works that are not available on CCC
please contact mpkbookspermissions@tandf.co.uk

Trademark notice: Product or corporate names may be trademarks or registered trade-

marks, and are used only for identification and explanation without intent to infringe.

Library of Congress Cataloging-in-Publication Data

Names: Sawano, Yoshihiro, author. | Di Fazio, Giuseppe, 1963- author. |
Hakim, Denny Ivanal, author.
Title: Morrey spaces : introduction and applications to integral operators
and PDE’s / Yoshihiro Sawano, Giuseppe Di Fazio, Denny Ivanal Hakim.
Description: First edition. | Boca Raton : C&H/CRC Press, 2020. | Series:
Chapman & Hall/CRC monographs and research notes in mathematics |
Includes bibliographical references and index. | Contents: Banach
function lattices -- Fundamental facts in functional analysis --
Polynomials and harmonic functions -- Various operators in Lebesgue
spaces -- BMO spaces and Morrey-Campanato spaces -- BMO spaces and
Morrey-Campanato spaces -- General metric measure spaces -- Weighted
Lebesgue spaces -- Approximations in Morrey spaces -- Predual of Morrey
spaces -- Linear and sublinear operators in Morrey spaces.
Identifiers: LCCN 2019060107 | ISBN 9781498765510 (v. 1 ; hardback) | ISBN
9780367459154 (v. 2 ; hardback) | ISBN 9780429085925 (v. 1 ; ebook) |
ISBN 9781003029076 (v. 2 ; ebook)
Subjects: LCSH: Banach spaces. | Harmonic analysis. | Differential
equations, Partial--Numerical solutions. | Differential equations,
Elliptic--Numerical solutions. | Integral operators.
Classification: LCC QA322.2 .S29 2020 | DDC 515/.732--dc23
LC record available at https://lccn.loc.gov/2019060107

ISBN: 9781498765510 (hbk)

ISBN: 9780429085925 (ebk)

Typeset in CMR
by Nova Techset Private Limited, Bengaluru & Chennai, India
Contents

Preface xi

Acknowledgement xv

Notation in this book xvii

1 Banach function lattices 1

1.1 Lp spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Measure space . . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Integration theorems . . . . . . . . . . . . . . . . . . 3
1.1.3 Fubini theorem and Lebesgue spaces . . . . . . . . . 4
1.1.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2 Morrey spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.2.1 Morrey norms . . . . . . . . . . . . . . . . . . . . . . 13
1.2.2 Examples of functions in Morrey spaces . . . . . . . 16
1.2.3 The role of the parameters . . . . . . . . . . . . . . . 23
1.2.4 Inclusions in Morrey spaces . . . . . . . . . . . . . . 24
1.2.5 Weak Morrey spaces . . . . . . . . . . . . . . . . . . 25
1.2.6 Morrey spaces and ball Banach function spaces . . . 27
1.2.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.3 Local Morrey spaces, Bσ -spaces, Herz spaces and Herz–
Morrey spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.3.1 Local Morrey spaces . . . . . . . . . . . . . . . . . . 32
1.3.2 Herz spaces and Herz–Morrey spaces . . . . . . . . . 34
1.3.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 36
1.4 Distributions and Lorentz spaces . . . . . . . . . . . . . . . 37
1.4.1 Distribution function . . . . . . . . . . . . . . . . . . 37
1.4.2 Lorentz spaces . . . . . . . . . . . . . . . . . . . . . . 42
1.4.3 Hardy operators and Hardy’s inequality . . . . . . . 43
1.4.4 Inequalities for monotone functions and their
applications to Lorentz norms . . . . . . . . . . . . . 46
1.4.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 48
1.5 Young functions and Orlicz spaces . . . . . . . . . . . . . . 49
1.5.1 Young functions . . . . . . . . . . . . . . . . . . . . . 50
1.5.2 Orlicz spaces . . . . . . . . . . . . . . . . . . . . . . . 57

v
vi Contents

1.5.3 Orlicz-averages . . . . . . . . . . . . . . . . . . . . . 60
1.5.4 Lebesgue spaces with a variable exponent . . . . . . 64
1.5.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 64
1.6 Smoothness function spaces . . . . . . . . . . . . . . . . . . 64
1.6.1 Sobolev spaces . . . . . . . . . . . . . . . . . . . . . . 65
1.6.2 Hölder–Zygmund spaces . . . . . . . . . . . . . . . . 73
1.7 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

2 Fundamental facts in functional analysis 87

2.1 Normed spaces and Banach spaces . . . . . . . . . . . . . . 87

2.1.1 Hahn–Banach theorem and Banach–Alaoglu theorem 87
2.1.2 Refinement of the triangle inequality . . . . . . . . . 88
2.1.3 Sum and intersection of Banach spaces . . . . . . . . 89
2.1.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 91
2.2 Hilbert spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 92
2.2.1 Komlos theorem . . . . . . . . . . . . . . . . . . . . . 92
2.2.2 Cotlar’s lemma . . . . . . . . . . . . . . . . . . . . . 93
2.2.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 94
2.3 Bochner integral . . . . . . . . . . . . . . . . . . . . . . . . 95
2.3.1 Measurable functions . . . . . . . . . . . . . . . . . . 95
2.3.2 Convergence theorems . . . . . . . . . . . . . . . . . 102
2.3.3 Fubini’s theorem for Bochner integral . . . . . . . . . 103
2.3.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 106
2.4 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

3 Polynomials and harmonic functions 109

3.1 Preliminary facts on polynomials . . . . . . . . . . . . . . . 109

3.1.1 The space Pk (Rn ) . . . . . . . . . . . . . . . . . . . . 109
3.1.2 Moment inequalities . . . . . . . . . . . . . . . . . . 111
3.1.3 Control of derivatives by integrals . . . . . . . . . . . 116
3.1.4 Best approximation . . . . . . . . . . . . . . . . . . . 118
3.1.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 120
3.2 Spherical harmonic functions . . . . . . . . . . . . . . . . . 120
3.2.1 The spaces Hk (Rn ) and Ḣk (Rn ) . . . . . . . . . . . . 120
3.2.2 Norm estimates for spherical harmonics . . . . . . . . 122
3.2.3 Laplacian and integration by parts formula . . . . . . 125
3.2.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 126
3.3 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

4 Various operators in Lebesgue spaces 129

4.1 Maximal operators . . . . . . . . . . . . . . . . . . . . . . . 129

4.1.1 Hardy–Littlewood maximal operator . . . . . . . . . 130
 Contents vii

4.1.2 Hardy–Littlewood maximal inequality . . . . . . . . 133

4.1.3 Local estimates for the Hardy–Littlewood maximal
operator . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.1.4 Fefferman–Stein vector-valued maximal inequality . . 143
4.1.5 Orlicz-maximal operators . . . . . . . . . . . . . . . . 147
4.1.6 Composition of the maximal operators . . . . . . . . 151
4.1.7 Local boundednss of the Φ-maximal operators . . . . 154
4.1.8 Estimates for convolutions . . . . . . . . . . . . . . . 157
4.1.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 160
4.2 Sharp maximal operators . . . . . . . . . . . . . . . . . . . 161
4.2.1 Sharp-maximal inequalities . . . . . . . . . . . . . . . 161
4.2.2 Distributional maximal function and median . . . . . 163
4.2.3 Generalized dyadic grid and the Lerner–Hytönen
decomposition . . . . . . . . . . . . . . . . . . . . . . 166
4.2.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 173
4.3 Fractional maximal operators . . . . . . . . . . . . . . . . . 174
4.3.1 Fractional maximal operators . . . . . . . . . . . . . 174
4.3.2 Local estimates for the maximal operators and the
fractional maximal operators . . . . . . . . . . . . . . 175
4.3.3 Sparse estimate for fractional maximal operators . . 177
4.3.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 178
4.4 Fractional integral operators . . . . . . . . . . . . . . . . . . 178
4.4.1 Fractional integral operators on Lebesgue spaces . . . 178
4.4.2 Local estimates for the fractional integral operators . 181
4.4.3 Sparse estimate of the fractional integral operators . 185
4.4.4 Fundamental solution to the elliptic differential
operators . . . . . . . . . . . . . . . . . . . . . . . . . 186
s
4.4.5 The Bessel potential operator (1 − ∆)− 2 , s > 0 . . . 188
4.4.6 Morrey’s lemma . . . . . . . . . . . . . . . . . . . . . 191
4.4.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 192
4.5 Singular integral operators . . . . . . . . . . . . . . . . . . . 192
4.5.1 Riesz transform . . . . . . . . . . . . . . . . . . . . . 193
4.5.2 Calderón–Zygmund operators . . . . . . . . . . . . . 195
4.5.3 Calderón–Zgymund decomposition . . . . . . . . . . 200
4.5.4 Weak-(1, 1) boundedness and strong-(p, p)
boundedness . . . . . . . . . . . . . . . . . . . . . . . 203
4.5.5 Truncation and pointwise convergence . . . . . . . . 205
4.5.6 Examples of singular integral operators . . . . . . . . 208
4.5.7 Sparse estimate of singular integral operators . . . . 212
4.5.8 Local estimates for singular integral operators . . . . 213
4.5.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 213
4.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
viii Contents

5 BMO spaces and Morrey–Campanato spaces 221

5.1 The space BMO(Rn ) and commutators . . . . . . . . . . . 221

5.1.1 The space BMO . . . . . . . . . . . . . . . . . . . . . 222
5.1.2 John–Nirenberg inequality . . . . . . . . . . . . . . . 224
5.1.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 227
5.2 Commutators . . . . . . . . . . . . . . . . . . . . . . . . . . 228
5.2.1 Commutators generated by BMO and singular integral
operators . . . . . . . . . . . . . . . . . . . . . . . . . 228
5.2.2 Commutators generated by BMO and fractional
integral operators . . . . . . . . . . . . . . . . . . . . 237
5.2.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 238
5.3 Morrey–Campanato spaces . . . . . . . . . . . . . . . . . . 238
5.3.1 Morrey–Campanato spaces . . . . . . . . . . . . . . . 238
5.3.2 Morrey–Campanato spaces and Hölder–Zygmund
spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 241
5.3.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 243
5.4 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244

6 General metric measure spaces 247

6.1 Maximal operators on Euclidean spaces with general Radon

measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
6.1.1 Covering lemmas on Euclidean spaces . . . . . . . . . 249
6.1.2 Maximal operators on Euclidean spaces with general
Radon measures . . . . . . . . . . . . . . . . . . . . . 255
6.1.3 Differentiation theorem . . . . . . . . . . . . . . . . . 259
6.1.4 Universal estimates . . . . . . . . . . . . . . . . . . . 261
6.1.5 Examples of metric measure spaces . . . . . . . . . . 266
6.1.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 273
6.2 Maximal operators on metric measure spaces with general
Radon measures . . . . . . . . . . . . . . . . . . . . . . . . 274
6.2.1 Weak-(1, 1) estimate and strong-(p, p) estimate . . . 274
6.2.2 Vector-valued bounededness of the Hardy–Littlewood
maximal operators . . . . . . . . . . . . . . . . . . . 278
6.2.3 Examples of metric measure spaces which require
modification . . . . . . . . . . . . . . . . . . . . . . . 280
6.2.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 286
6.3 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290

7 Weighted Lebesgue spaces 293

7.1 One-weighted norm inequality . . . . . . . . . . . . . . . . . 295

7.1.1 The class A1 . . . . . . . . . . . . . . . . . . . . . . . 295
7.1.2 The class Ap . . . . . . . . . . . . . . . . . . . . . . . 299
 Contents ix

7.1.3 The class A∞ . . . . . . . . . . . . . . . . . . . . . . 306

7.1.4 The class Ap,q . . . . . . . . . . . . . . . . . . . . . . 311
7.1.5 Extrapolation . . . . . . . . . . . . . . . . . . . . . . 314
7.1.6 A2 -theorem . . . . . . . . . . . . . . . . . . . . . . . 317
7.1.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 324
7.2 Two-weight norm inequality . . . . . . . . . . . . . . . . . . 324
7.2.1 Weighted estimates for the Hardy operator . . . . . . 325
7.2.2 Two-weight norm inequality for fractional maximal
operators . . . . . . . . . . . . . . . . . . . . . . . . . 332
7.2.3 Two-weight norm inequality for singular integral
operators . . . . . . . . . . . . . . . . . . . . . . . . . 335
7.2.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 338
7.3 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339

8 Approximations in Morrey spaces 343

8.1 Various closed subspaces of Morrey spaces . . . . . . . . . . 343

8.1.1 Closed subspaces generated by linear lattices . . . . . 343
8.1.2 Closed subspaces generated by the translation . . . . 346
8.1.3 Inclusions in closed subspaces of Morrey spaces . . . 347
8.1.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 348
8.2 Approximation in Morrey spaces . . . . . . . . . . . . . . . 348
8.2.1 Characterization of M fpq (Rn ) . . . . . . . . . . . . . . 349
8.2.2 Approximations and closed subspaces . . . . . . . . . 351
8.2.3 Examples of functions in closed subspaces . . . . . . 353
8.2.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 358
8.3 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359

9 Predual of Morrey spaces 363

9.1 Predual of Morrey spaces . . . . . . . . . . . . . . . . . . . 363

9.1.1 Definition of block spaces and examples . . . . . . . 364
9.1.2 Finite decomposition and a dense subspace . . . . . . 370
9.1.3 Duality–block spaces and Morrey spaces . . . . . . . 371
9.1.4 Fatou property of block spaces . . . . . . . . . . . . . 373
9.1.5 Köthe dual of Morrey spaces . . . . . . . . . . . . . . 375
9.1.6 Decomposition and averaging technique in Morrey
spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 382
9.1.7 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 388
9.2 Choquet integral and predual spaces . . . . . . . . . . . . . 389
9.2.1 Hausdorff capacity . . . . . . . . . . . . . . . . . . . 389
9.2.2 Choquet integral . . . . . . . . . . . . . . . . . . . . 397
9.2.3 Predual spaces of Morrey spaces by way of the
Choquet integral . . . . . . . . . . . . . . . . . . . . 402
9.2.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 406
9.3 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 406
x Contents

10 Linear and sublinear operators in Morrey spaces 409

10.1 Maximal operators in Morrey spaces . . . . . . . . . . . . . 409

10.1.1 Maximal operator in Morrey spaces . . . . . . . . . . 409
10.1.2 Maximal operator in local Morrey spaces . . . . . . . 413
10.1.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 414
10.2 Sharp maximal operators in Morrey spaces . . . . . . . . . 414
10.2.1 Sharp maximal inequalities for Morrey spaces . . . . 414
10.2.2 Sharp maximal inequalities for local Morrey spaces . 416
10.2.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 416
10.3 Fractional integral operators in Morrey spaces . . . . . . . . 416
10.3.1 Fractional integral operators in Morrey spaces . . . . 417
10.3.2 Fractional integral operators in local Morrey spaces . 422
10.3.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 423
10.4 Singular integral operators in Morrey spaces . . . . . . . . . 424
10.4.1 Singular integral operators in Morrey spaces . . . . . 424
10.4.2 Singular integral operators in local Morrey spaces . . 430
10.4.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 431
10.5 Commutators in Morrey spaces . . . . . . . . . . . . . . . . 431
10.5.1 Commutators in Morrey spaces . . . . . . . . . . . . 432
10.5.2 Commutators in local Morrey spaces . . . . . . . . . 433
10.5.3 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 434
10.6 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 434

Bibliography 441

Index 475
Preface

Morrey spaces were introduced by Charles Morrey, who considered an estimate

of the difference of functions in 1938 in order to investigate the local behav-
ior of solutions to second order elliptic partial differential equations [323].
The technique is very useful in many areas in mathematics, in particular in
harmonic analysis, potential theory, partial differential equations and math-
ematical physics. Our emphasis in this book is mainly on harmonic analysis.
We consider Morrey spaces in connection with potential theory.
We describe some milestone results in this book. Chiarenza and Frasca
showed the boundedness of the Hardy–Littlewood maximal operator, singular
integral operators and fractional integral operators in Morrey spaces [75]. The
boundedness of fractional maximal operators and fractional integral operators
is originally due to Adams [2]. Di Fazio and Ragusa generalized some results
on commutators [123] due to Coifman, Rochberg and Weiss [84] and due to
Chanillo in [66]. Mizuhara and Nakai considered generalized Morrey spaces.
They showed the boundedness of the Hardy–Littlewood maximal operator,
singular integral operators and fractional integral operators [316, 327]. Inde-
pendently Guliyev considered generalized weak Morrey spaces. Sawano and
Tanaka considered Morrey spaces for general Radon measures [394]. Komori-
Furuya and Shirai considered weighted Morrey spaces as a special case of
this general framework of Radon measures [254]. The framework by Komori-
Furuya and Shirai was carried over to the one in generalized weighted Morrey
spaces by Guliyev and Mustafayev [168]. Independently of Komori-Furuya
and Shirai, Samko considered the weighted Morrey spaces [376, 377]. There
are many works on weighted Morrey spaces under the condition similar to
the one for Lebesgue spaces. However, as is pointed out by Tanaka [425] and
Duoandiotxea and Rosental [106], as well as Sawano and Nakamura [334, 335],
weighted Morrey spaces go beyond weighted Lebesgue spaces in the sense that
some weights which do not satisfy the Muckenhoupt condition make the oper-
ators bounded.
Morrey-type spaces originated from the idea developed by Guliyev and
Burenkov [45]. Morrey-type spaces became a prototype of real interpolation.
Burenkov and Nursultanov showed that real interpolation yields general local
Morrey-type spaces [49].
Interpolation of Morrey spaces is one of the difficult problems in har-
monic analysis. Roughly there are two interpolation methods, the real inter-
polation and the complex interpolation. We still have open problems in both

xi
xii Preface

interpolations of Morrey spaces. Stampaccia obtained a partial result in 1964

[413]. Cobos et. al also showed a result on the complex interpolation functor
in 1998 [80]. Blasco, Ruiz and Vega provided a prominent counterexample on
the interpolation of Morrey spaces [37]. Despite all the recent progress in the
area of interpolation of Morrey spaces, there are still many important open
problems. In the last decade more and more works on interpolation of Mor-
rey spaces emerged. Lemarié-Rieusset pointed out that Morrey spaces are not
closed under the first complex interpolation [265, 266]. Later on Yang, Yuan
and Lu described complex interpolation of Morrey spaces [294]. Their descrip-
tion was solidified by Sawano and Hakim [179]. Recently Sawano and Mastylo
gave an explicit formula of the complex interpolation of Morrey spaces [305].
However, to the best knowledge of the authors, we have no general information
on what this complex interpolation spaces look like.
Another important aspect is that Morrey spaces arise naturally when we
consider the multiplier space from a Besov space to Lebesgue spaces [263].
One of the important aims in this book is to discuss the current state of
the art and perspectives of developments of this theory of Morrey spaces. We
want to provide a shelf to contain all these developments. We did our best to
include as many references as we can. However, by no means do we hope that
this book is merely a database. We strongly believe that there is always room
in discussing function spaces no matter how big the theory becomes. What is
important is that we can handle (unknown) sets of functions in a reasonable
manner. Probably, knowing all that is written in this book is insufficient to
launch a new study on function spaces. Nor do we need to read it completely.
We wrote this book to be able to get a smooth access to all of these topics.
We chose a topic from which the readers can learn to handle function spaces.
In the long study of Morrey spaces, some incorrect propositions were believed
to be true. For example, including Sawano, many people believed that Morrey
spaces are closed under the first complex interpolation, which was disproved
by Lemarié-Riuesset. We would like to explain why such a mistake happened
for example.
In the 1980’s and 1990’s, we obtained a number of results on the bound-
edness of classical operators in Morrey spaces and related function spaces.
Meanwhile around the beginning of the 21st century there were new active
developments in this area. They resulted in necessary and sufficient conditions
in many cases.
We explain the organization of the first half of this book, which consists
of 10 chapters.
Chapter 1 starts with defining some fundamental function spaces such as
Lebesgue spaces, Morrey spaces, Lorentz spaces and Orlicz spaces together
with related function spaces. We suppose that we are familiar with Lebesgue
spaces. See [156] for example. Our main function spaces, Morrey spaces, will
appear in Chapter 1 so as not to get lost regarding what we are discussing.
One of the important purposes of introducing various function spaces is that
we can grasp as many functions as we can. The function spaces we will take
 Preface xiii

up in these chapters will play such a role. We can say that this note expands
the editorial [386].
Chapters 2 and 3 are of preliminary nature, which is interesting in its own
right. Chapter 2 collects some auxiliary observations in functional analysis.
Chapter 3 takes up polynomials. The facts that we prove in Chapters 2 and 3
will appear not so many times. The facts on functional analysis in this book
are quite fundamental and can be found in some textbooks such as [310].
Chapters 4, 5, 6 and 7 are also of preliminary nature. However, unlike
Chapters 2 and 3, we will frequently use what we prove in Chapters 4, 5, 6
and 7. This part largely overlaps the existing textbooks such as [146, 176, 156,
157, 415, 416, 382]. Although a thorough explanation of fundamental facts in
harmonic analysis makes the book thicker, we decided to include it to the
minimum so as to be self-contained.
When we consider some problems in mathematics, it is not sufficient to
be able to handle functions themselves. Usually, functions we will consider
are transformed by some mappings described by means of integrals. So, we
introduce some important operators here. Chapter 4 deals with the funda-
mental operators in many branches of mathematics. We consider the Hardy–
Littlewood maximal operator, singular integral operators fractional integral
operators and fractional maximal operators. Chapter 5 includes commutators
generated by these operators.
The theory of Morrey spaces can be staged on metric measure spaces; it is
enough to have a distance function and a Borel measure to develop a theory
of Morrey spaces. So, we work in general metric measure spaces in Chapter
6. As a special case of metric measure spaces, Chapter 7 deals with weighted
function spaces. Chapter 7 contains the solution of the A2 conjecture.
We embark on the study of Morrey spaces, especially operators acting on
Morrey spaces in Chapters 8, 9 and 10. In some sense, Morrey spaces are
introduced so as to cover what Lebesgue spaces cannot do. We seek to have a
better understanding of mathematical transforms by the use of Morrey spaces.
For example, the fractional integral operator is given by f 7→ | · |−n+α ∗ f .
Here 0 < α < n. The well-known theorem by Hardy–Littlewood–Sobolev
says that this transform maps Lp to Lq under a certain condition. Since the
fractional integral operator is the convolution operator, the Young inequality
will be a useful tool. However, the kernel | · |−n+α never belongs to Lp for
any 0 < p ≤ ∞. We expect Morrey spaces to explain why this convolution
mapping is bounded. In many cases we resort to the density argument in func-
tional analysis when we define operators acting on function spaces. Chapter
8 precisely discuss whether this is possible in Morrey spaces. For example,
L∞ n
c (R ), the space of all compactly supported essentially bounded functions,
is not dense in Morrey spaces. Many other plausible candidates fail to be dense
too. Another way of investigating the boundedness of operators in function
spaces is to employ duality. Chapter 9 invesitigates duality. Unfortunately, we
do not have any description of the dual space of Morrey spaces. However, Mor-
rey spaces are realized as the dual spaces of certain function spaces. Based
xiv Preface

on what we investigated in Chapters 8 and 9, Chapter 10 investigates the

boundedness of operators in Morrey spaces. The Hardy–Littlewood maximal
operator can be defined for any measurable functions since it is defined via
the integral for non-negative measurable functions. So, with ease we can show
that the Hardy–Littlewood maximal operator is bounded on Morrey spaces.
Almost the same applies to fractional maximal operators and fractional inte-
gral operators since the integral kernels of these operators are non-negative.
However, we have to be careful when handling singular integral operators and
commutators generated by singular integral operators. As we will see in Chap-
ter 8, Morrey spaces do not have L∞ n
c (R ) as a dense subspace. So, we cannot
resort to the density argument. Instead, we can use duality results obtained
in Chapter 9. An alternative way to define singular integral operators is to
investigate carefully the size of the integrals.
Acknowledgement

We are thankful to the following individuals: Ali Akbulut, Ryutaro Arai,

Tserendorj Batbold, Neal Bez, Victor I. Burenkov, David Cruz-Uribe SFO,
Fatih Deringoz, Javier Duoandikoetxea, Eridani, Maria Stella Fanciullo,
Bruno Franchi, Michele Frasca, Sadek Gala, Arash M. Ghorbanalizadeh,
Loukas Georgios Grafakos, Vagif Guliyev, Hendra Gunawan, Cristian Gutier-
rez, Sabir G. Hasanov, Naoya Hatano, Kwok-Pun Ho, Takeshi Iida, Mitsuo
Izuki, Takashi Izumi, Eder Kikianty, Yasuo Komori-Furuya, Pier Domenico
Lamberti, Ermanno Lanconelli, Sang Hyuk Lee, Yiyu Liang, Liguang Liu,
Fumi-Yuki Maeda, Mieczyslaw Mastylo, Katsuo Matsuoka, Akihiko Miyachi,
Alexander Meskhi, Yoshihiro Mizuta, Eiichi Nakai, Shohei Nakamura, Van
Hanh Nguyen, Toru Nogayama, Takahiro Noi, Mehriban Omarova, Takahiro
Ono, Maria Alessandra Ragusa, Hiroki Saito, Daniel Salim, Enji Sato, Christo-
pher Schwanke, Raul Serapioni, Saad R. El-Shabrawy, Minglei Shi, Tetsu
Shimomura, Satoru Shirai, Idha Sihwaningrum, Takuya Sobukawa, Sakoto
Sugano, Hitoshi Tanaka, Ryotaro Tanaka, Naohito Tomita, Lorenzo Tuccari,
Yohei Tsutsui, Tino Ullrich, Hidemitsu Wadade, Wen Yuan, Kozo Yabuta,
Shotaro Yanagishita, Da Chun Yang, Dong Yong Yang, Sibei Yang, Kaoru
Yoneda, Hiroko Yoshida, Pietro Zamboni, Ciqiang Zhuo.
We wish to thank the following institutions: Faculty of Mechanics and
Mathematics, L. N. Gumilyov Eurasian National University, Astana, Kaza-
khstan, University of Catania, Bandung Institute of Technology, Tokyo
Metropolitan University and Instituto Nazionale di alta Matematica (Indam)
for financial support. We thank Professor Pidatella for the permission to use
a hand-writing of Professor Chiarenza as cover of the book.
Special thanks to Professor Filippo Chiarenza who has been a friend and
a mentor to Giuseppe Di Fazio.

xv
Notation in this book

Sets and set functions

(1) We write #A to denote the cardinality of a set A.
(2) The metric open ball defined by `2 is usually called a ball. We denote
by B(x, r) the ball centered at x of radius r. Namely, we write

B(x, r) ≡ {y ∈ Rn : kx − yk < r}

when x ∈ Rn and r > 0. Given a ball B, we denote by c(B) its center

and by r(B) its radius. We write B(r) instead of B(o, r), where o ≡
(0, 0, . . . , 0).
(3) By a “cube” we mean a compact cube whose edges are parallel to the
coordinate axes. The metric closed ball defined by `∞ is called a cube. If
a cube has center x and radius r, we denote it by Q(x, r). Namely, we
write
 
Q(x, r) ≡ y = (y1 , y2 , . . . , yn ) ∈ Rn : max |xj − yj | ≤ r
j=1,2,...,n

when x = (x1 , x2 , . . . , xn ) ∈ Rn and r > 0. For a cube Q or a right-open

cube Q, define Q(Q) to be the set of all cubes of the same type contained
in Q and Q] (Q) to be the set of all cubes of the same type containing
Q. From the definition of Q(x, r), its volume is (2r)n . We write Q(r)
instead of Q(o, r). Given a cube Q, we denote by c(Q) the center of Q
and by `(Q) the sidelength of Q: `(Q) = |Q|1/n , where |Q| denotes the
volume of the cube Q. The symbol Q = Q(Rn ) denotes the set of all
cubes.
(4) Given Q ∈ Q and k > 0, k Q means the cube concentric to Q with
sidelength k `(Q). Given a ball B and k > 0, we denote by k B the ball
concentric to B with radius k r(B).
(5) For ν ∈ Z and m = (m1 , m2 , . . . , mn ) ∈ Zn , we define Qνm ≡
n  
Y mj mj + 1
ν
, ν
. Denote by D = D(Rn ) the set of such cubes.
j=1
2 2
The elements in D(Rn ) are called dyadic cubes.

xvii
xviii Notation in this book

(6) Let E be a measurable set. Then we denote its indicator function by

χE . If E has positive measure and E is integrable over f , Then denote
by mE (f ) the average of f over E. |E| denotes the volume of E.
(7) If we are working on Rn , then B denotes the set of all balls in Rn , while
Q denotes the set of all cubes in Rn . Be careful because B can be used for
a different purpose: When we are working on a measure space (X, B, µ),
then B stands for the set of all Borel sets.

(8) The symbol 2X denotes the set of all subsets in X.

(9) A tacit understanding is that by a “cube” we mean a closed cube whose
edges are parallel to the coordinate axes. However, we say that (right-
open) dyadic cubes are also cubes.
(10) We define the upper half space Rn+ and the lower half space Rn− by

Rn± ≡ {(x0 , xn ) ∈ Rn : ±xn > 0}. (0.1)

Numbers
(1) Let a ∈ R. Then write a+ ≡ max(a, 0) and a− = a ∧ 0 ≡ min(a, 0).
Correspondingly, given an R-valued function f , f+ and f− are functions
given by f+ (x) ≡ max(f (x), 0) and f− (x) ≡ min(f (x), 0), respectively.
(2) Let a, b ∈ R. Then write a ∧ b ≡ min(a, b). Correspondingly, given
R-valued functions f, g, f ∧ g are functions given by f ∧ g(x) ≡
min(f (x), g(x)).
(3) The constants C and c denote positive constants that may change from
one occurrence to another. Because the two constants c can be different,
the inequality 0 < 2c < c is by no means a contradiction. When we add
a subscript, for example, this means that the constant c depends upon
the parameter. It can happen that the constants with subscript differ
according to the above rule. In particular, we prefer to use cn for various
constants that depend on n, when we do not want to specify its precise
value.
(4) Let A, B ≥ 0. Then A . B and B & A mean that there exists a constant
C > 0 such that A ≤ CB, where C depends only on the parameters of
importance. The symbol A ∼ B means that A . B and B . A happen
simultaneously, while A ' B means that there exists a constant C > 0
such that A = CB.
(5) When we need to emphasize or keep in mind that the constant C depends
on the parameters α, β, γ etc:
(1) Instead of A . B, we write A .α,β,γ,... B.
 Notation in this book xix

(2) Instead of A & B, we write A &α,β,γ,... B.

(3) Instead of A ∼ B, we write A ∼α,β,γ,... B.
(4) Instead of A ' B, we write A 'α,β,γ,... B.
(a) We define N ≡ {1, 2, . . .}, Z ≡ {0, ±1, ±2, . . .} and N0 ≡ {0, 1, . . .}.
p
(b) For a ∈ Rn , we write hai ≡ 1 + |a|2 .
(c) We occasionally identify Rm+n with Rm × Rn .

Function spaces
(1) We use · for functions; f = f (·).
(2) The function spaces are tacitly on Rn unless otherwise stated.
(3) Let X be a Banach space. We denote its norm by k · kX .
(4) Let Ω be an open set in Rn . Then Cc∞ (Ω) denotes the set of smooth
functions with compact support in Ω.
(5) Let 1 ≤ j ≤ n. The symbol xj denotes not only the j-th coordinate but
also the function x = (x1 , x2 , . . . , xn ) 7→ xj .
(6) The space L2 (Rn ) is the Hilbert space of square integrable functions on
Rn whose inner product is given by
Z
hf, gi = f (x)g(x)dx. (0.2)
Rn

(7) Let E be a measurable set and f be a measurable function

Z with respect
1
to the Lebesgue measure. Then write mE (f ) ≡ f.
|E| E
(8) Let 0 < η < ∞, E be a measurable set, and f be a positive measurable
(η)
function with respect to the Lebesgue measure. Then write mE (f ) ≡
1
mE (f η ) η .
(9) Let 0 < η < ∞. We define the powered Hardy–Littlewood maximal oper-
ator M (η) by
Z ! η1
(η) 1 η
M f (x) ≡ sup |f (y)| dy .
R>0 |B(x, R)| B(x,R)

(10) The space C(Rn ) denotes the set of all continuous functions on Rn .
(11) The space BC(Rn ) denotes the set of all bounded continuous functions
on Rn .
xx Notation in this book

(12) Occasionally we identify the value of functions with functions. For exam-
ple sin x denotes the function on R defined by x 7→ sin x.

(13) Given a Banach space X , we denote by X ∗ its dual space.

(14) Let µ be a measure on a measure space (X, B, µ). Given a µ-measurable
set A with positive µ-measure and a function f , we write
Z
1
mQ (f ) ≡ f (x)dµ(x).
µ(A) A
(η) 1
Let 0 < η < ∞. Then define mQ (f ) ≡ mQ (f η ) η whenever f is positive.
(15) If notational confusion seems likely, Then we use [ ] to denote
M f (x) = M [f ](x), T ϕ(ξ) = T [ϕ](ξ), etc.
∂
(16) For j = 1, 2, . . . , n, ∂xj = stands for the partial derivative. The
∂xj
n
∂2
P
symbol ∆ stands for the Laplacian ∂xj 2 .
j=1

(17) We denote the Lp (Rn )-norm by k·kLp . For other function spaces such as
Hölder’s continuous function space C γ (Rn ) of order γ > 0, we use k · kC γ
to stress the function spaces.
(18) When we consider function spaces on a domain Ω, we denote by C γ (Ω)
the Hölder continuous function space of order γ.
(19) A quasi-norm over a linear space X enjoys positivity, homogeneity,
and quasi-triangle inequality: for some α ≥ 1, kf + gkX ≤ α( kf kX +
kgkX ) (f, g ∈ X ). However, to simplify, we frequently omit the word
“quasi”. Likewise, we abbreviate the word “quasi-Banach space” to
Banach space.
The Kronecker delta function defined on a set X is given by δjk ≡
(20) (
1 (j = k),
for j, k ∈ X.
0 (j 6= k).

(21) When two normed spaces X, Y are isomorphic with equivalence of

norms, we write X ≈ Y .
(22) For subsets A, B of a linear space V and v ∈ V , define the Minkovski
sum by

v + A ≡ {v + a : a ∈ A}, A + B ≡ { a + b : a ∈ A, b ∈ B}.

There is an exception: If k > 0 and Q is a right-open/compact cube,

we denote by kQ the k-times expansion of Q, that is, kQ is a cube
concentric to Q and |kQ| = k n |Q|.
 Notation in this book xxi

(23) When two topological spaces X , Y are homeomorphic, we write X ≈ Y.

(24) When A and B are sets, A ⊂ B stands for the inclusion of sets. If,
in addition, both A and B are topological spaces, and if the natural
embedding mapping A → B is continuous, we write A ,→ B in the sense
of continuous embedding. If A and B are quasi-normed spaces with the
k
embedding constant k > 0, then A ,→ B.
Chapter 1
Banach function lattices

This is a preliminary part of this book. We recall some fundamental inequal-

ities as well as important theorems. The starting point is the review of the
theory of Lebesgue spaces in Section 1.1. Historically, as early as 1925 people
came to realize Lebesgue spaces themselves are insufficient to describe the
boundedness properties of operators. For example, Kolmogorov noticed that
L1 is not sufficient for the Hilbert transform. So we are interested in the tools
to overcome the problem. Roughly speaking, Banach function spaces, Banach
lattices are notions to indicate that we have a certain linear space of functions
together with a norm.
Since Morrey spaces are main function spaces in this book, we take up
Morrey spaces in Section 1.2. Having defined Morrey spaces and related func-
tion spaces in Section 1.2, Section 1.3 considers local Morrey spaces, which are
closely related to Morrey spaces. However, local Morrey spaces differ seriously
from Morrey spaces. We will consider different function spaces (Banach lattices)
in Sections 1.4 and 1.5. Although we consider Morrey spaces in this book, we will
need some other related function spaces to consider Morrey spaces deeply and
to consider the boundedness properties of operators. We consider distribution
functions of measurable functions in Section 1.4 to replace Lebesgue spaces with
other function spaces, while as another method of replacing Lebesgue spaces we
deal with Orlicz spaces in Section 1.5. Section 1.6, which is different from other
sections but still fundamental, introduces Sobolev spaces and Hölder–Zygmund
spaces, although they are not Banach lattices.

1.1 Lp spaces
Here we recall some fundamental properties on integration theory includ-
ing Lebesgue spaces. This will be useful when we consider Morrey spaces since
Lebesgue spaces are realized as a special case of Morrey spaces. Conversely
in this book it will be important what happens when Morrey spaces differ
from Lebesgue spaces. Section 1.1.1 deals with measure spaces. Section 1.1.2
handles integration theorems. Section 1.1.3 considers Fubini’s theorem. The-
orem 5 is a fundamental tool to express the Lebesgue norm of functions. The
Lebesgue space Lp initially appears in Section 1.1.3. Among other theorems,

1
2 Morrey Spaces

we postpone the proof of Theorem 12 till Section 4.1. Many of the results in
Section 1.1 are standard; see some textbooks on integration theory.

1.1.1 Measure space

Here, we collect some terminologies needed in this book. We state the
theorems without any detailed proof.
Let us start with σ-algebras.
Definition 1 (σ-algebra). Let X be a set. A non-void family B ⊂ 2X is a
σ-algebra if the following properties hold:
(1) B is closed under complement. That is, A ∈ B =⇒ Ac ∈ B.
∞
S
(2) B is closed under countable union. Namely, Aj ∈ B, whenever
j=1
A1 , A2 , . . . , Aj , . . . ∈ B.
We adopt the following standard definition of measures:
Definition 2 (Measure). Let B be a σ-algebra over X. A mapping µ : B →
[0, ∞] is a measure if the following two properties hold.
(1) µ(∅) = 0.
(2) If A1 , A2 , . . . , Aj , . . . ∈ B are disjoint, that is, Ai ∩ Aj = ∅ for all i 6= j,
then  
∞
X [∞
µ(Aj ) = µ  Aj  .
j=1 j=1

Let L0 (µ) be the space of all measurable functions. When we consider the
case of Euclidean space, L0 (Rn ) denotes the set of all Lebesgue measurable
functions. It may happen that the value of f ∈ L0 (Rn ) lies in C or [0, ∞]. We
need other terminologies for later consideration.
Definition 3. A regular measure µ means that µ has the inner and outer
regularities. To be precise, if µ is a regular Borel measure (Radon measure),
then for each Borel set E of finite measure, we have the inner regularity

µ(E) = sup{µ(K) : K runs over all compact subsets of E}

and the outer regularity

µ(E) = inf{µ(U ) : U runs over all open sets containing E}. (1.1)

See [191, p. 3] or [319, Theorem 2.1] for example.

 Banach function lattices 3

1.1.2 Integration theorems

As for convergence theorems, we assume that the readers are familiar with
Theorems 1 and 2 below. Let M+ (Rn ) be the cone of all non-negative Lebesgue
measurable functions and M+ (µ) be the cone of all non-negative µ-measurable
functions defined on a measure space (X, B, µ). In particular, we define

M+ (Rn ) ≡ {f : Rn → [0, ∞] : f is measurable }.

Theorem 1. Suppose that (X, B, µ) is a measure space. Let {fj }∞

j=1 be a
sequence of R-valued µ-measurable functions.
(1) (Monotone convergence theorem.) Suppose that {fj }∞ 0
j=1 ⊂ L (µ) is an
increasing sequence : We have 0 ≤ fj ≤ fj+1 , µ−a.e.. Then
Z Z
lim fj (x)dµ(x) = lim fj (x)dµ(x). (1.2)
j→∞ X X j→∞

(2) (Fatou’s lemma.) For {fj }∞ +

j=1 ⊂ M (µ), we have
Z Z
lim inf fj (x)dµ(x) ≤ lim inf fj (x)dµ(x).
X j→∞ j→∞ X

(3) (Dominated convergence theorem, Lebesgue’s convergence theorem.)

Suppose that {fj }∞
j=1 converges µ-almost everywhere to f . Assume fur-
ther that there
Z exists g ∈ L1Z(µ) such that |fj | ≤ g a.e. for all j ∈ N.
Then lim fj (x)dµ(x) = f (x)dµ(x).
j→∞ X X

See [236, Chapter 13] or [319, Chapter 1], for example.

Theorem 2 (Change of integration and differentiation). Let (X, B, µ) be a
measure space, and we suppose that a function f : X × (a, b) → C satisfies the
following:
(1) For each t ∈ (a, b), f (·, t) ∈ L1 (µ).
(2) For µ-almost all x ∈ X the function t 7→ f (x, t) is differentiable for all
t ∈ (a, b).
∂f
(3) There exists g ∈ L1 (µ) such that (x, t) ≤ g(x) for all t ∈ (a, b) and
∂t
for µ-almost every x ∈ X.
Z Z
d ∂f
Then f (x, t)dµ(x) = (x, t)dµ(x).
dt X X ∂t
4 Morrey Spaces

1.1.3 Fubini theorem and Lebesgue spaces

Fubini’s theorem is one of the important tools in this book. We now con-
sider the product of measures. We omit the proof here; see [236, Chapter 16]
or [319, §1.8] for example.
Definition 4 ((µ ⊗ ν)∗ ). Let (X, B, µ) and (Y, B 0 , ν) be a couple of measure
spaces. Then define an outer measure (µ ⊗ ν)∗ on X × Y by (µ ⊗ ν)∗ (∅) = 0
and
 
X∞ ∞
[ 
(µ ⊗ ν)∗ (A) = inf µ(Ej )ν(Fj ) : A ⊂ Ej × Fj , A ∈ 2X×Y \ {∅}.
 
j=1 j=1

We say that E ⊂ X × Y is (µ ⊗ ν)∗ -measurable, if for all G ⊂ X × Y ,

(µ ⊗ ν)∗ (G) = (µ ⊗ ν)∗ (E ∩ G) + (µ ⊗ ν)∗ (E c ∩ G).
The following theorems are fundamental, whose proof we do not recall.
Theorem 3. The set of the form E × F with E ∈ M and F ∈ N is (µ ⊗ ν)∗ -
measurable and (µ ⊗ ν)∗ (E × F ) = µ(E)ν(F ). Here it will be understood that
0 × ∞ = ∞ × 0 = 0.
Consequently, we are led to the measure µ ⊗ ν: If E is (µ ⊗ ν)∗ -measurable,
then we write µ ⊗ ν(E) = (µ ⊗ ν)∗ (E).
Let (X, B, µ) be a measure space. Recall that a measure µ is σ-finite, if X
can be partitioned into a countable collection of disjoint subsets with finite
µ-measure.
From Definition 4, we can create another measure µ ⊗ ν. About this new
measure, we can say the following:
Theorem 4 (Fubini’s theorem). Suppose that (X, B, µ) and (Y, N 0 , ν) are
σ-finite measure spaces. Then
ZZ Z Z 
f (x, y)dµ ⊗ ν(x, y) = f (x, y)dν(y) dµ(x)
X×Y X Y
Z Z 
= f (x, y)dµ(x) dν(y),
Y X

provided f is a non-negative (M⊗N )-measurable function or f ∈ L1 (M⊗N ).

We move on to the Lebesgue spaces. Let (X, B, µ) be a measure space
because it will be necessary to state our results in full generality although we
mainly work on Euclidean spaces. One of the simplest ways to measure the
size of functions is the use of Lebesgue spaces. What is important here is that
we consider the case of 0 < p < 1. Sometimes considering the case of 0 < p < 1
will be useful. For example, when we consider the product f · g of L1 functions
1
f and g, we notice that f · g is nothing but L 2 functions. Once again, since
we use Lebesgue spaces on general measure spaces for later consideration, we
state our results in full generality.
 Banach function lattices 5

Let (X, B, µ) be a measure space, and let 0 < p ≤ ∞. We tolerate the

case p = ∞. The space Lp (µ) denotes the set of all f ∈ L0 (µ) for which the
Z  p1
p
quantity kf kLp (µ) ≡ |f (x)| dµ(x) is finite if p < ∞ and kf kL∞ (µ) ≡
X
ess.supx∈X |f (x)| is finite if p = ∞. As usual if (X, B, µ) is the Lebesgue
measure, then we write Lp (Rn ) = Lp (µ). Define
Lp (µ) ≡ {f ∈ L0 (µ) : kf kLp (µ) < ∞}/ ∼ .
Here, the equivalence relation ∼ is defined by f ∼ g ⇐⇒ f = g a.e.. Below
omit this equivalence in defining function spaces.
Example 1. Let γ ∈ R and 0 < p < ∞. We define
fγ (x) ≡ |x|γ χB(1) (x), gγ (x) ≡ |x|γ χB(1)c (x) (x ∈ Rn ). (1.3)
n
(1) fγ ∈ Lp (Rn ) if and only if γ > − .
p
n
(2) gγ ∈ Lp (Rn ) if and only if γ < − .
p
n
(3) For any γ ∈ R, fγ + gγ = | · |− p never belongs to Lp (Rn ).
It is noteworthy that the functions do not belong to Lp (Rn ) in the border case
n
γ=− .
p
In this book, by using other function spaces, we consider this limit situa-
tion.
When we consider the integral, it matters how large the level set is. See
Theorem 5 below. This motivates us to give the following definition.
Definition 5 (Distribution). Let f : X → C be a measurable function. Then
the distribution function λf : [0, ∞) → [0, ∞] is a function defined by
λf (t) = λf,µ (t) ≡ µ{x ∈ X : |f (x)| > t} (t ≥ 0).
The so called “Layer Cake Formula” below will be important when we
analyze the Lebesgue norm of functions.
Theorem 5 (Layer-Cake Formula). Let (X, B, µ) be a σ-finite measure space.
Let 0 < p < ∞ and f be a µ-measurable function. Then we have
 Z ∞  p1
kf kLp (µ) = p tp−1 λf,µ (t)dt . (1.4)
0

More generally, if Φ : [0, ∞) → [0, ∞) is a C 1 -increasing function such that

Φ(0) = 0, then
Z Z ∞
Φ(|f (x)|)dµ(x) = Φ0 (t)λf,µ (t)dt. (1.5)
X 0
6 Morrey Spaces

Proof We concentrate on (1.4): The proof of (1.5) is merely a slight adap-

tation of (1.4). We observe
Z |f (x)| Z ∞
p p−1
|f (x)| = pt dt = p tp−1 χ(t,∞) (|f (x)|)dt
0 0

for x ∈ X. By Fubini’s theorem (Theorem 4) for non-negative functions, we

calculate that
Z
(kf kLp (µ) )p = |f (x)|p dµ(x)
ZX∞ Z 
= p tp−1 χ(t,∞) (|f (x)|)dµ(x) dt
0 X
Z ∞ Z 
p−1
= pt χ(t,∞) (|f (x)|)dµ(x) dt
Z0 ∞ X

p−1
= p t λf,µ (t)dt.
0

By taking the p-th root, we obtain (1.4).

We move on to the norm inequalities of Lebesgue spaces. Let (X, B, µ)
be a measure space. Also let 1 ≤ p ≤ ∞, and define the conjugate of p by
p 1 1
p0 ≡ , if 1 ≤ p < ∞. If p = ∞, then define p0 = 1. Note that + 0 = 1,
p−1 p p
whenever 1 ≤ p ≤ ∞. A direct calculation shows that (p0 )0 = p. The Hölder
inequality is fundamental throughout this book. In fact, it occasionally matters
how we use it.
Theorem 6 (Hölder’s inequality). Let (X, B, µ) be a measure space, and let
1 ≤ p ≤ ∞. Then kf · gkL1 (µ) ≤ kf kLp (µ) kgkLp0 (µ) for all f, g ∈ L0 (µ).

See [236, Theorem 17.3] or [319, Theorem 12.3], for example.

According to Minkowski’s inequality, which we recall below, we see that
Lp (µ) is a normed space. Recall that a linear space X is a normed space, if
it comes with a function k · kX : X → [0, ∞), which satisfies the following
property. In this case k · kX : X → [0, ∞) is a norm on a linear space X .
(1) Let x ∈ X . Then kxkX = 0 implies x = 0.
(2) ka · xkX = |a| · kxkX for all a ∈ K and x ∈ X .
(3) Let x, y ∈ X . Then the triangle inequality kx + ykX ≤ kxkX + kykX
holds.
We now recall Minkowski’s inequality; see [236, Theorem 17.4] or [319, Theo-
rem 12.4], for example.
Theorem 7 (Minkowski’s inequality). Let (X, B, µ) be a measure space. Let
1 ≤ p ≤ ∞. Then kf + gkLp (µ) ≤ kf kLp (µ) + kgkLp (µ) for all f, g ∈ L0 (µ).
 Banach function lattices 7

When 0 < p < 1, the triangle inequality fails to be true. Instead, we use
the following p-convexity or p-triangle inequality:
Proposition 8 (p-convexity). Let (X, B, µ) be a measure space. Let 0 < p ≤ 1.
Then kf + gkLp (µ) p ≤ kf kLp (µ) p + kgkLp (µ) p for all f, g ∈ L0 (µ).
The proof, which uses Exercise 6, is left for the readers.
Duality is an important tool to estimate the size of functions. Let 1 ≤ p <
∞ and (X, B, µ) be a σ-finite measure space. Then we can describe the dual
0
of Lp (µ) by way of Lp (µ) as follows:
0
Theorem 9 (Duality Lp (µ)-Lp (µ)). Let 1 ≤ p < ∞ and (X, B, µ) be a
0
σ-finite measure space. Then Lp (µ)∗ = Lp (µ) with coincidence of norms.
0
Speaking precisely, let f ∈ Lp (µ) and define Ff ∈ Lp (µ)∗ by
Z
p
g ∈ L (µ) 7→ f (x)g(x)dµ(x) ∈ C.
X
p0 p ∗
Then f ∈ L (µ) 7→ Ff ∈ L (µ) is well defined and kf kLp0 (µ) = kFf kLp (µ)∗ .
Furthermore if F : Lp (µ) → K is a continuous linear functional, i.e., there
exists M > 0 such that |F (h)| ≤ M khkLp (µ) for all h ∈ Lp (µ), then F is
0
realized as F = Ff with some f ∈ Lp (µ).
Proof We do not recall the proof of this fundamental theorem. For the
proof, see [81, Example 3.5.4] and [319, Theorem 12.7], for example.
In addition to Theorem 9, it is important to write the Lebesgue norm in
the dual form.
Theorem 10. Let 1 ≤ p ≤ ∞ and (X, B, µ) be a measure space. Then for all
f ∈ L0 (µ), kf kLp (µ) = sup{kf · gkL1 (µ) : kgkLp0 (µ) = 1}.
Proof Again we do not recall the proof of this fundamental theorem. For
the proof see [258, Theorem 3.1.6], for example.
Local integrability will be important in this book. This will be made clearer
after we define Morrey spaces.
Definition 6 (Lploc (Rn )). Let 1 ≤ p ≤ ∞. The space Lploc (Rn ) collects all
f ∈ L0 (Rn ) such that f ∈ Lp (K) for each compact set K, or equivalently,
χB ·f ∈ Lp (Rn ) for each ball B inR Rn . The topology of Lploc (Rn ) is generated by
the functional f ∈ Lploc (Rn ) 7→ |f (y)|p dy, where K moves over all compact
K
sets.
Example 2. Let γ ∈ R and 0 < p < ∞. We define fγ (x) ≡ |x|γ for x ∈ Rn .
n
Since fγ ∈ Lp (Rn ) if and only if γ > − , it follows that f ∈ Lploc (Rn ) if and
p
n n
only if γ > − . To compensate for the failure in the case of γ = − , we
p p
will define the weak Lebesgue space WLp (Rn ) as the set of all g ∈ L0 (Rn ) for
which kgkWLp ≡ sup λkχ(λ,∞] (|g|)kLp is finite. See Definition 7 below.
λ>0
8 Morrey Spaces

By combining Hölder’s inequality with Definition 6, we can locally embed

Lebesgue spaces at the cost of the integrability index.
Corollary 11. Let 0 < q < p ≤ ∞. Then Lp (Rn ) ,→ Lqloc (Rn ).
Although we will get out of the framework of Lebesgue spaces, we still
often have some local integrability. In this case, it is important to consider the
average. For a measurable set E and f ∈ L1 (E) over which f is integrable,
we define mE (f ) to be the average of f over E:
Z
1
mE (f ) ≡ f (y)dy. (1.6)
|E| E
Since the Lebesgue differentiation theorem will be important, we state it with-
out the proof. Keeping the elementary inequalities in mind, we will apply
the boundedness of the Hardy–Littlewood maximal operator to establish the
Lebesgue differentiation theorem. The Lebesgue differentiation theorem is
important because we can “visualize” the value f (x) through integrals. Since
we often handle integral operators, we have much information on the quantity
described by integrals. So, the Lebesgue differentiation theorem is important.
Theorem 12 (Lebesgue differentiation theorem). Let f ∈ L1loc (Rn ). Then

lim mB(x,r) (|f − f (x)|) = 0 (1.7)

r↓0

for almost all x ∈ Rn . A point x for which (1.7) holds is called a Lebesgue
point of f .
We will prove Theorem 12 after we prove the boundedness of the Hardy-
Littlewood maximal operator:
n
We postpone its proof.
The function | · |− p is not a member in Lp (Rn ). Hence,
n
we are interested
in function spaces close to Lp (Rn ) which contain | · |− p . We will see that the
weak Lp (Rn )-space (the weak Lebesgue spaces, the Marcinkiewicz space) will
serve this purpose.
Definition 7 (Marcinkiewicz space, weak Lebesgue space).
(1) Let 0 < p < ∞. Then the weak Lp (µ)-space (or the Marcinkiewicz
space) WLp (µ) is the space of all functions f : X → C which satisfies
1 1
kf kWLp (µ) ≡ sup tλf (t) p = sup t p f ∗ (t) < ∞ if p < ∞.
t>0 t>0

∞ ∞
(2) Define WL (µ) ≡ L (µ).
(3) As usual if (X, B, µ) is the Lebesgue measure, then we write Lp (Rn ) =
Lp (µ) and k · kWLp = k · kWLp (µ) .
Example 3. Let γ ∈ R and 0 < p < ∞. Define fγ and gγ as in Example 1.
n
Then fγ ∈ WLp (Rn ) if and only if γ ≥ − and gγ ∈ WLp (Rn ) if and only if
p
 Banach function lattices 9
n n
γ ≤ − , while fγ ∈ Lp (Rn ) if and only if γ > − and gγ ∈ Lp (Rn ) if and
p p
n n
only if γ < − . Note f− np + g− np = | · |− p ∈ WLp (Rn ).
p
We will investigate an embedding relation for decreasing functions. Here,
recall that a continuous (Borel) measure is a measure µ which does not charge
any point in the space; µ({a}) = 0 for any point a. Here and below w ∈ B(I)
means that w is Borel measurable. Here, we state our result in full generality
using continuous measures. We use Lemma 13 to prove Theorem 315. Here
and below denote by M↓ (0, ∞), the set of all non-negative decreasing functions
on (0, ∞).
Lemma 13. Let 0 < p ≤ p q < ∞, and let µ, ν be continuous Borel measures
q
µ(0, s)
on (0, ∞). If B ≡ sup p
p
< ∞, then kf kLq (µ) ≤ Bkf kLp (ν) for all
s>0 ν(0, s)
f ∈ M↓ (0, ∞).
Proof We may approximate f with simple decreasing functions. Let g ≡
K
X
f p . Then for the function g of the form g = ak χ(0,bk ) , with a1 , a2 , . . . , aK >
k=1
a0 = 0 and b0 = 0 < b1 < b2 < · · · < bK , we have
K Z ∞  pq
X q
kgk q ≤ ak p χ(0,bk ) (t)dµ(t)
Lp (µ) 0
k=1

since q ≥ p. Again since q ≥ p, we have

K
X
kgk q ≤ Bp ak ν(0, bk ) = B p kgkL1 (ν) .
Lp (µ)
k=1

So the proof is complete.

Let M+ (0, ∞) denote the set of all non-negative measurable functions in
(0, ∞). Likewise, it is not so hard to show that
Z ∞  p1
p
sup f (t)w(t) . f (t) v(t)dt (1.8)
t>0 0

for all f ∈ M↓ (0, ∞) and for all 0 < p ≤ 1, as long as v, w ∈ M+ (0, ∞) satisfy
! !
1 1
sup sup w(t) = sup sup w(t) < ∞,
a>0 v(0, a) t∈(0,a) a>0 kvkL1 (0,a) t∈(0,a)

since we can reduce the matters to the case p = 1.

10 Morrey Spaces

Lemma 14. Let −∞ < a < b < ∞ and θ ∈ (0, 1). For all f ∈ M↑ (0, ∞), we
Z b !θ Z b
have g(t)dt ≤θ (t − a)θ−1 g(t)θ dt.
a a

Proof Assume for the time being that g is a function of the form:
N
X
g= λj χ(aj−1 ,aj ]
j=1

with a0 = a < a1 < · · · < aN = b and 0 ≤ λ1 < λ2 < · · · < λN , then

!θ  θ
Z b XN
g(t)dt = λj (aj − aj−1 )
a j=1

and
Z b N
X
θ (t − a)θ−1 g(t)θ dθ = λj θ ((aj − a)θ − (aj−1 − a)θ ).
a j=1

To prove the conclusion, we induct on N by handling λN and b = aN as if they

were variables as well. If N = 1, there is nothing to prove. Hence, assume that
the assertion is true for N − 1 and suppose that we have f described above.
−1
NP
By translation, we may assume a = 0. Let B ≡ λj (aj − aj−1 ) ≥ 0. Since
j=1
aN θ ≤ aN −1 θ + (aN − aN −1 )θ and 0 < θ ≤ 1, we have
 
N
d X θ θ
λj (aj − aj−1 θ ) − (λN (aN − aN −1 ) + B)θ 
dλN j=1

= θ(aN θ − aN −1 θ )λN θ−1 − θ(aN − aN −1 )(λN (aN − aN −1 ) + B)θ−1

≥ θ(aN θ − aN −1 θ )λN θ−1 − θ(aN − aN −1 )(λN (aN − aN −1 ))θ−1
≥ 0.

Thus, assuming that the result is true for N − 1, we see that

!θ
Z b Z b
g(t)dt ≤θ (t − a)θ−1 g(t)θ dt.
a a

In the general case, we decompose and approximate g by using functions of

P∞
the form λj χ(0,aj ) .
j=1

We end this section with a variant of Lebesgue spaces called mixed

Lebesgue spaces. Let 0 < p1 , p2 , . . . , pn ≤ ∞ be constants. Write p ≡
 Banach function lattices 11

(p1 , p2 , . . . , pn ). Then define the mixed Lebesgue norm k · kLp by

 ! pp3  p1n
Z Z Z  pp2 2
1
kf kLp ≡  ··· |f (x1 , x2 , . . . , xn )|p1 dx1 dx2 · · · dxn  .
R R R

A natural modification for xj is made when pj = ∞, j = 1, 2, . . . , n. We

define the mixed Lebesgue space Lp (Rn ) to be the set of all f ∈ L0 (Rn ) with
kf kLp < ∞.

1.1.4 Exercises
Exercise 1. Let f : X → C be a complex-valued measurable function. Denote
h+ = max(h, 0) and h− = max(−h, 0) for a real-valued measurable function
h. Show that the following are equivalent:
(1) f ∈ L1 (µ).
(2) Re(f ) ∈ L1 (µ) and Im(f ) ∈ L1 (µ).
(3) Re(f )± ∈ L1 (µ) and Im(f )± ∈ L1 (µ).
Exercise 2.
(1) Choose functions integrable on [0, ∞) with respect to the Lebesgue mea-
sure among the functions listed below.
(2) Do the same thing replacing [0, ∞) with [1, ∞).
(3) Do the same thing replacing [0, ∞) with [0, 1).
(A) e−x (B) xα (α < −1) (C) x−1 (D) xα (−1 < α < 0) (E) 1
√ −x
e
(F ) xα (α > 0) (G) xe−x (H) x4 sin x (I) x e k − x
(J)
x−1
1 2
log(x + 1) 2 e−x
(K) 4 (L) p .
x3 |x − 2|
Exercise 3. Find the necessary and sufficient condition for a, b ∈ R to satisfy
|x|a
Z
b
dx < ∞.
Rn 1 + |x|

Exercise 4. Let (X, B, µ) be a finite measure space, and let f ∈ L0 (µ). Show
that lim kf kLp (µ) = kf kL∞ (µ) and that
p→∞
 Z 
1 1
lim µ(X)− p kf kLp = exp log |f (x)|dµ(x) .
p↓0 µ(µ) X
12 Morrey Spaces
2
Re log x
Exercise 5. Calculate the Lebesgue integral x dx.
e
Exercise 6.
(1) Show that (a + b)p ≤ ap + bp for all a, b > 0 and 0 < p ≤ 1.
(2) Prove Proposition 8.
Exercise 7. Show the following scaling nlaw in Lebesgue spaces Lp (Rn ) : For
all f ∈ Lp (Rn ) and t > 0 kf (t·)kLp = t− p kf kLp .
Exercise 8. Show that the Lebesgue space Lp ([0, 1]) with 0 < p < 1 is not a
normed space. What is the property of the normed space that fails?
Exercise 9. Suppose that h : R → C is a continuous function such that
lim h(t) = A exists.
|t|→∞
Z T
1
(1) Show that lim h(t)dt = A.
T →∞ 2T −T
Z T  
1 T
(2) Show that lim log h(t)dt = A. Hint : Let M+ (R) denote
T →∞ 2T −T |t|
the set of all non-negative measurable functions defined on R. Justify
that we can assume that h ∈ M+ (R) is an even function. Then use
Z T 
1 T
 Z  
1 T T
log h(t)dt = log h(t)dt
2T −T |t| T 0 t
!
1 T
Z Z T
ds
= h(t)dt
T 0 t s
1 T 1 s
Z  Z 
= h(t)dt ds,
T 0 s 0
where for the last equality we have used Fubini’s theorem.
Exercise 10. Let M+ (µ) stand for the set of all non-negative µ-measurable
functions. Let (X, B, µ) be a measure space, and let f ∈ M+ (µ). Define induc-
tively
1
f0 ≡ f, fj+1 ≡ fj − χ 1 (fj ).
j + 1 ( j+1 ,∞)
Then show that lim fj = 0. Consequently, we have a decomposition f =
j→∞
∞
1
P
j+1 χEj for some measurable set Ej .
j=0

1.2 Morrey spaces

The aim of this book is investigating Morrey spaces keeping in mind
that Morrey spaces cover Lebesgue spaces and hence that Morrey spaces are
 Banach function lattices 13

expected to enjoy nicer properties than Lebesgue spaces. Here, we define Mor-
rey spaces in Section 1.2.1 and give some functions in Morrey spaces in Section
1.2.2.
Section 1.2.3 also considers some examples of functions in Morrey spaces
by paying attention to the parameters in Morrey spaces. Since Morrey spaces
have two parameters, we expect that Morrey spaces with different parameters
are connected with each other. We check that Morrey spaces are nested in
Section 1.2.4. Section 1.2.5 is a different nature from the previous sections.
We consider weak Morrey spaces as an analogue of Morrey spaces. We handle
weak function spaces. Weak Lebesgue spaces appeared in Example 2. We
expand the idea there. As shown, weak Morrey spaces are used to compensate
for the failure of the boundedness of operators. We return to Morrey spaces
in Section 1.2.6. There are many function spaces other than Morrey spaces.
Some of them are Banach lattices. The notion of Banach lattices is used to
cover many (solid) function spaces. One of the other useful tools to classify
function spaces is the use of Banach function spaces. However, as it turns out,
Morrey spaces are not Banach function spaces in general. Hence, we propose a
notion which substitutes for Banach function spaces. We actually define ball
Banach function spaces in Section 1.2.6. Once again a different function in
Morrey spaces appears there.

1.2.1 Morrey norms

Section 1.2.1 defines and investigates (classical) Morrey spaces. There are
several notation to denote Morrey spaces but we use the letter M instead of
L consistently in this book.
Morrey spaces are the most fundamental in this book. Having defined
Lebesgue spaces, we now turn to Morrey spaces. Let us define the norm. We
write Rn+1
+ ≡ Rn × (0, ∞).
Definition 8 (Mpq (Rn )). Let 0 < q ≤ p ≤ ∞. For an Lqloc (Rn )-function f , its
(classical) Morrey norm is defined by
Z ! q1
1 1
kf kMpq = kf kB
Mpq
≡ sup |B(x, r)| p − q |f (y)|q dy . (1.1)
(x,r)∈Rn+1
+
B(x,r)

If there is need to stress the dimension we work in, we write kf kMpq (Rn ) instead
of kf kMpq . The (classical) Morrey space Mpq (Rn ) is the set of all f ∈ Lqloc (Rn )
for which the norm kf kMpq is finite.
If 0 < q ≤ p = ∞, then the Lebesgue differentiation theorem shows that
Mpq (Rn ) = L∞ (Rn ), so we exclude this case.
In other words, f ∈ Mpq (Rn ) if and only if there exists D ≥ 0 such that
1 1
kf kLq (B(x,r)) ≤ D|B(x, r)| p − q for all x ∈ Rn and r > 0. The minimal value
of D in this inequality is kf kMpq .
Since f · g ∈ Mpq (Rn ) for any f ∈ Mpq (Rn ) and g ∈ L∞ (Rn ), Mpq (Rn ) is
not a space of functions possessing any kind of common smoothness of any
order.
14 Morrey Spaces

Before we go further, a couple of remarks on the definition and the notation

of the Morrey norm may be in order. For j ∈ Z and k = (k1 , k2 , . . . , kn ) ∈ Zn
define Qjk ∈ Q by
    n  
k1 k1 + 1 kn kn + 1 Y kl kl + 1
Qjk ≡ , × ··· × , = , .
2j 2j 2j 2j 2j 2j
l=1

A dyadic cube is a set of the form Qjk for some j ∈ Z, k = (k1 , k2 , . . . , kn ) ∈

Zn . Finally, D(Rn ) stands for the set of all dyadic cubes.

Remark 1.
(1) Define (temporarily) the Morrey norms kf kB
Mp
0
q
and kf kQ
Mpq
of f ∈
0 n
L (R ) by

Z ! q1
n n
kf kB0
≡ sup r p−q q
|f (y)| dy (1.2)
Mp q
(x,r)∈Rn+1
+
B(x,r)

and
Z ! q1
1 1
kf kQ ≡ sup |Q(x, r)| p−q q
|f (y)| dy . (1.3)
Mpq
(x,r)∈Rn+1
+
Q(x,r)

The dyadic Morrey norm k · kD

Mpq
is defined by

Z  q1
1 1
kf kD
Mpq
≡ sup |Q| p − q |f (y)|q dy .
Q∈D(Rn ) Q

Then kf kMpq ∼ kf kD
Mpq
for all f ∈ L0 (Rn ). Then kf kMpq ∼n,p,q
kf kB
Mp
0
q
∼n,p,q kf kQ
Mpq
∼n,p,q kf kD
Mpq
for all f ∈ L0 (Rn ). This means that
we can define the Morrey space Mpq (Rn ) by using the norms k · kB0
Mp q
,
k · kQ
Mpq
k · kD
Mpq
: The resulting space will be the same. We sometimes
identify all of them; the superindexes B, B0 , D are sometimes omitted.
See Exercise 13.
(2) In this book, all norms k · kMpq , k · kB
Mp
0
q
and k · kQ
Mpq
are used. Usually
there will be no confusion.
(3) There are some traditions of how to express Morrey norms. Some prefer
to use the notation:
Z ! q1
1
kf kLqλ ≡ sup |f (y)|q dy .
(x,r)∈Rn+1 rλ B(x,r)
+
 Banach function lattices 15

Here, 0 < q < ∞ and 0 < λ < n. See the classical but important work
of Peetre [353]. By letting
 
q
λ≡ 1− n (1.4)
p

we obtain kf kLqλ = kf kMpq = kf kB0

Mp q
. However, in this book one prefers
p
to use the notation Mq .
As is seen from this remark, it is impossible to find the precise value of the
norm kf kMpq for a given f . Consequently, in many cases we give up calculating
kf kMpq : We often content ourselves with estimating kf kMpq .
The case where 0 < p < q < ∞ is excluded because this case boils down
to a triviality.
Proposition 15. The space Mpq (Rn ) is trivial in the following sense: Let
x0 ∈ Rn be fixed. If f ∈ L0 (Rn ) satisfies
Z ! q1
1 1
A ≡ sup |B(x0 , r)| p−q q
|f (y)| dy <∞
r>0 B(x0 ,r)

for some p < q, then f (x) = 0 for almost all x ∈ Rn .

In this negative assertion, we can freeze the point x0 instead of allowing
it to vary among all points in Rn .
Proof From the definition of A,
Z ! q1
1 1
|f (y)|q dy ≤ |B(x0 , r)| q − p A.
B(x0 ,r)

Letting r → ∞, we have kf kLq = 0. Thus, f (x) = 0 for almost all x ∈ Rn .

A quasi-norm over a linear space X enjoys positivity, homogeneity, and
quasi-triangle inequality: for some α ≥ 1, kf +gkX ≤ α( kf kX +kgkX ) (f, g ∈
X ). We explain what happens if p and/or q is less than 1.
Remark 2. It can happen that p and/or q is less than 1. In this case, we
formally define the space Mpq (Rn ) by (1.1) to obtain a quasi-Banach space,
where the triangle inequality is replaced by the quasi-triangle inequality;
1
kf + gkMpq ≤ 2max( q −1,0) (kf kMpq + kgkMpq ) (1.5)

for all f, g ∈ L0 (Rn ). Note that (1.5) follows from the triangle inequality
immediately when 1 ≤ q < ∞. Meanwhile, when 0 < q < 1, (1.5) follows from
the inequality (a + b)q ≤ 2max(q−1,0) (aq + bq ) for all a, b ≥ 0.
As the following theorem shows, we can say that Lebesgue spaces are
realized as a special case of Morrey spaces.
16 Morrey Spaces

Theorem 16. For 0 < p < ∞, Mpp (Rn ) = Lp (Rn ) with coincidence of norms.
Proof Let f ∈ L0 (Rn ). We write down the definition of the norm kf kMpp ;
1 1
kf kMpp ≡ sup |B(x, r)| p − p kf kLp (B(x,r)) = sup kf kLp (B(x,r)) .
(x,r)∈Rn+1
+ (x,r)∈Rn+1
+

Then, we have
Z  p1
p
kf kMpp ≤ |f (y)| dy = kf kLp . (1.6)
Rn
Meanwhile, by the monotone convergence theorem, for any x ∈ Rn , we have
kf kLp = lim kf kLp (B(x,m)) ≤ sup kf kLp (B(x,r)) = kf kMpp . (1.7)
m→∞ r>0

Combining (1.6) and (1.7), we obtain the desired result.

A cone in a linear space is a subset closed under addition and multiplication
by positive scalars. We remark that when q = 1, we can consider measures
instead of functions as follows:
Definition 9 (Mp1 (Rn )). The space Mn p1 (Rn ) consists of all Radon measures µ
on Rn satisfying kµkMp1 ≡ sup r p −n |µ|(B(x, r)) < ∞, where |µ| denotes
(x,r)∈Rn+1
+
the total variation of µ.
Here, we content ourselves with examples.
Example 4.
(1) Let 1 ≤ p < ∞. Then Mp1 (Rn ) ,→ Mp1 (Rn ).
(2) Let µ be the Hausdorff measure on Rn−1 ⊂ Rn . More precisely, for a
Borel measurable set E ⊂ Rn we let
Z
µ(E) = χE (x0 , 0)dx0 .
Rn−1
n
Then µ ∈ M1 n−1
(Rn ).

1.2.2 Examples of functions in Morrey spaces

Our main concern is what type of functions does belong to Morrey spaces.
Here, we will collect some examples of Morrey functions. As we saw in
Theorem 16, the Morrey space Mpp (Rn ) with 0 < p < ∞ is nothing but
Lp (Rn ). Hence, we may ask ourselves how the Morrey space Mpq (Rn ) with
0 < q < p < ∞ differs from Lp (Rn ). One of the big gaps between them is
that Lp (Rn ) requires only one cube “Rn ”, while we need to consider all the
cubes in Rn in the definition of the norm in the Morrey space Mpq (Rn ) with
0 < q < p < ∞. This difference actually yields many strange phenomena
below.
We calculate the Morrey norm of the indicator functions of the balls.
 Banach function lattices 17

Example 5. Let 0 < q ≤ p < ∞. Let B be an open ball. Then

kχB kMpq = kχB kLp . (1.8)

In fact, it is easy to see that

kχB kMpq ≤ kχB kLp (1.9)

from Theorem 16 and Hölder’s inequality. (See also Theorem 19 to follow.) If

we write out the norm kχB kMpq in full, then

Z ! q1
1 1
kχB kMpq = sup |B(x, r)| p−q |χB (y)|q dy .
(x,r)∈Rn+1
+
B(x,r)

We can calculate and evaluate the integral precisely. The result is:
1 1 1 1
kχB kMpq = sup |B(x, r)| p − q |B(x, r) ∩ B| q ≥ |B| p = kχB kLp . (1.10)
(x,r)∈Rn+1
+

Combining (1.9) and (1.10), we obtain (1.8).

A major problem concerning simple/fundamental function spaces BC(Rn )
and Lp (Rn ) is that neither of them contains | · |−α for any α ∈ R. However,
this simplest function appears everywhere but to handle this function, we need
n
n
to restrict its domain. The Morrey space Mqα (Rn ) with 1 < q < α contains
−α
| · | . Here and below, B(r) abbreviates B(x, r) with x the origin.
n
n
Example 6. The Morrey space Mqα (Rn ) with 1 < q < α contains fα (x) ≡
|x|−α , x ∈ Rn . To check this, we observe that
1 1 1 1
sup |B(x, r)| p − q kfα kLq (B(x,r)) = |B(r)| p − q kfα kLq (B(r)) . (1.11)
x∈Rn

As a corollary of this example, we have the following relation:

Example 7. Let 0 < q < p < ∞, and let α, β ∈ R.
(1) Let f (x) = fα (x) ≡ |x|α , x ∈ Rn . Then f ∈ Mpq (Rn ) if and only if
n
α=− .
p
n
(2) Let g ≡ f χB(1) . Then g ∈ Mpq (Rn ) if and only if α ≥ − . Likewise let
p
p n n
h ≡ f χRn \B(1) . Then h ∈ Mq (R ) if and only if α ≤ − .
p
(3) Let k ≡ fα χB(1) + fβ χRn \B(1) . From (2), we see that k ∈ Mpq (Rn ) if and
n
only if α ≥ − ≥ β.
p
18 Morrey Spaces

We present an example of functions to detect the difference of q in Mpq (Rn ).

Example 8 (Toothbrush). Let 0 < q < p < ∞. We choose a very small

 α  p1 1
number α so that = α q for each N ∈ N. We divide equally [0, 1]n into
N n
N
N cubes to have QN
n N N
P
1 , Q 2 , . . . , Q N n . We will consider f = fN ≡ χαQN
j
∈
j=1
L∞ n
c (R ). We claim that kf kMq ∼ kf kLq . To this end we use
p

Z ! q1
1 1
kf kMpq ≡ kf kQ = sup |Q(x, r)| p−q q
|f (y)| dy .
Mpq
(x,r)∈Rn+1
+
Q(x,r)

Since f is supported on [0, 1]n , we may assume that Q(x, r) runs over
all cubes contained in [0, 1]n . We first note that the supremum is almost
attained by letting r = 2α: If we choose x ∈ Rn suitably, then we have
1 1 1
|Q(x, r)| p − q kf kLq (Q(x,r)) . |Q(r)| p ' kχαQj kLp = kf kLq . If 2r < α instead,
1 1 1
then we have |Q(x, r)| p − q kf kLq (Q(x,r)) ≤ |Q(x, r)| p < kf kLq no matter where
x is. Thus, to show the result, we have only to consider the cubes Q(x, r)
with α ≤ 2r ≤ 1. If Q(x, r) intersects QN N
j and Qk with 1 ≤ j < k ≤ N ,
n

then there exists R ∈ Q such that R ∩ supp(f ) is realized as the union

{QN n
j }j∈J for some J ⊂ {1, 2, . . . , N }, that R contains Q(x, r) and that
|R| . |Q(x, r)|. Hence, by translating R if necessary, we may assume that
R = [0, kN −1 ]n for some k = 1, 2, . . . , N . For this R, we can readily check
1 1
that |Q(x, r)| p − q kf kLq (Q(x,r)) . kf kLq .
We consider another example whose motivation comes from the ternary
Cantor set. The parameter q seems to reflect the local and geometric structure
as the next example shows:
Example 9 (Self-similar decreasing sequence for Mpq (Rn )). We let p > q > 0
and R > 1 be fixed so that
1 1 1
(1 + R)− p = 2 q (1 + R)− q . (1.12)

It is noteworthy that the case where R = 2 corresponds to the ternary Cantor

set. For a vector ε ∈ {0, 1}n , we define an affine transformation Tε by

1 R
Tε (x) ≡ x+ ε (x ∈ Rn ).
1+R 1+R

Let E0 ≡ [0, 1]n and Ej,0 ≡ [0, (1 + R)−j ]n .SSuppose that we have defined
E0 , E1 , E2 , . . . , Ej , j ∈ N. Define Ej+1 ≡ Tε (Ej ). Then we will show
ε∈{0,1}n
that
n
kχEj kMpq ∼ (1 + R)−j p = kχEj,0 kMpq = kχEj,0 kLp = kχEj kLq , (1.13)
 Banach function lattices 19

where the implicit constants in ∼ are independent of j but can depend on p

and q and the Morrey norm k · kMpq used here is defined by (1.3). A direct cal-
n
culation shows kχEj kMpq ≥ kχEj,0 kMpq = (1 + R)−j p = kχEj,0 kLp = kχEj kLq .
Thus, we need to show kχEj kMpq . kχEj,0 kLp . Let us calculate kχEj kMpq ∼
1 1 1
sup |S| p − q |S ∩ Ej | q , where Q denotes the set of all cubes. Fix j ∈ N. Let
S∈Q
us temporarily say that Q ∈ Q is wasteful, if Q does not contribute to the
supremum, that is, if there exists S ∈ Q such that
1 1 1 1 1 1
|Q| p − q |Q ∩ Ej | q < |S| p − q |S ∩ Ej | q .

Thus, by definition, if `(Q) > 1, then

1 1 1 1 1 1 1
|Q| p − q |Q ∩ Ej | q < |Ej | q = |[0, 1]n | p − q |[0, 1]n ∩ Ej | q .

In addition, if the side-length of Q is less than (1 + R)−j , then it is wasteful.

Indeed, then the equality
n 1 1 1
o
sup |Q| p − q |Q ∩ Ej | q : Q ∈ Q, |Q| ≤ (1 + R)−jn
1 1 1
= sup{|Q| p − q |Q ∩ Ej | q : Q ∈ Q, Q ⊂ Ej }
n 1 o
= sup |Q| p : Q ∈ Q, Q ⊂ Ej
1
= |Ej,0 | p

holds. Here, we obtain the first equality by translating Q so that Q is included

1 1 1 1
in Ej,0 . This calculation shows that |Q| p − q |Q ∩ Ej | q < |Ej,0 | p for any such
−j
cube. Thus, if the cube Q is not wasteful, then (1 + R) ≤ `(Q) ≤ 1. Thus,
there exists k = 1, 2, . . . , n such that (1 + R)−kn ≤ |Q| ≤ (1 + R)−(k−1)n .
In this case, since any connected component Z satisfies (1 + R)−kn = |Z| ≤
|Q| ≤ (1 + R)−(k−1)n , 3Q contains a connected component of Ek . Hence, it
follows that

kχEj kMpq
1 1 1
∼ sup{|Q| p − q |Q ∩ Ej | q : Q contains a connected component of Ej }.

Let S be a cube which contains a connected component of Ej and is not

wasteful. By symmetry, we may assume S = I × I × · · · × I for some interval
I. We define
[ 
S ∗ ≡ co {W : W is a connected component of Ej intersecting S} ,

where co(A) stands for the smallest convex set containing a set A. Then a
geometric observation shows that S ∗ engulfs k n connected component of Ej
for some 1 ≤ k ≤ 2j . Take an integer l so that 2l−1 ≤ k ≤ 2l . Then

|S ∗ ∩ Ej | = k n (1 + R)−jn , |S ∗ | ∼ (1 + R)−jn+ln .
20 Morrey Spaces

Consequently, from (1.12) we have

1 1 1 ln n n n
|S ∗ | p − q |S ∗ ∩ Ej | q ∼ 2 q (1 + R)−j q (1 + R)(−j+l)( p − q )
= 2 q (1 + R)−j p +l( p − q )
ln n n n

= 2 q (1 + R)l( p − q ) (1 + R)−j p
ln n n n

n
= (1 + R)−j p .

Thus, we obtain (1.13).

We will present a similar example using dyadic cubes.

Example 10 (Dyadic rearrangement). Let 0 < q ≤ p < ∞. Fix j ∈ N. We
define
q
a0 ≡ N (j) ≡ [2jn(1− p ) ], ak+1 ≡ 1 + [2−n ak ], (k = 0, 1, . . .).

Note that ak+1 ≤ 1 + 2−n ak for each k and hence

 
−nk 1 1
ak ≤ 2 a0 − + .
1 − 2−n 1 − 2−n

We place disjoint cubes Q1 , Q2 , . . . , QN (j) ∈ Dj ([0, 1)n ) so that each Q ∈

Dk ([0, 1)n ) contains at most ak cubes. Thus, writing Ej ≡ Q1 ∪Q2 ∪· · ·∪QN (j) ,
we obtain
1 1 jn jn 1 1 1
|Q| p − q k2 p χEj kLq (Q) ≤ 2 p (2−kn ) p − q (2−jn ak ) q
jn 1 1 q 1
.2 p (2−kn ) p − q (max(2−jn , 2−jn−kn+jn(1− p ) )) q
.1

for all Q ∈ Dk (Rn ). However

jn jn jn q 1
k2 p χQ1 ∪Q2 ∪···∪QN (j) kLr (Q) = 2 p − q [2jn(1− p ) ] r → ∞

if q < r ≤ p.
The following is a sort of normalization: It can be arranged that Ej ⊂
[0, 2−j )n for each j ∈ N.
Hence, we can say that the Morrey norm k · kMpq reflects local regularity
of functions more precisely than the Lebesgue norm k · kLp .
A chain of equalities in (1.13) is the motivation of choosing R in (1.12).
Based on Example 9, we present another example. Although this is just a
matter of scaling, we sometimes find it convenient to use increasing sequences.
Example 11 (Self-similar increasing sequence for Mpq (Rn )). Let R > 1 solve
n n n
(1 + R) p − q 2 q = 1. In Example 9, if one defines Fj ≡ {x ∈ Rn : (1 + R)−j x ∈
 Banach function lattices 21

Ej }, then {Fj }∞
j1 is an increasing sequence of sets and each Fj is made up of
a disjoint union of cubes of length 1. It is noteworthy that
j
( )
X
Fj ≡ y + Rk ak : {ak }jk=1 ∈ {0, 1}j , y ∈ [0, 1]n .
k=1

If we denote by Q(Fj ), the set of all connected components of Fj , then

]Q(Fj ) = 2jn . By the scaling law of Mpq (Rn ), we have kχFj kMpq ∼ 1. If
we let F ≡ {y + Ra1 + R2 a2 + · · · : {aj }∞ 1 n
j=1 ∈ {0, 1} ∩ ` (N), y ∈ [0, 1] }.
N

Then we have Fj = F ∩ [0, (1 + R)j ]n for all j ∈ N. It is noteworthy that

∞
S
kχQ kMpq ∼ kχFj kMpq ∼ 1 for any Q ∈ Q(F ), where Q(F ) = Q(Fj ).
j=1

Example 12. Let 0 < q < p < ∞. Set

∞
[ p p
G≡ (k − 1 + k p−q , k + k p−q ). (1.14)
k=1

We show that χG ∈ Mpq (R)(∩L∞ (R)). In fact, keeping in mind that G ⊂

(0, ∞), we observe
n 1 1 1
o
kχG kMpq = sup (b − a) p − q |(a, b) ∩ G| q : 0 < a < b < ∞
n 1 1 1
o
≤ sup (b − a) p − q |(a, b) ∩ G| q : a, b ∈ R, a < b < a + 2
n 1 1 1
o
+ sup (b − a) p − q |(a, b) ∩ G| q : a, b ∈ R, 0 < a < b − 2 < ∞
n 1 1 1
o
. 1 + sup (b − a) p − q |(a, b) ∩ G| q : a, b ∈ R, 0 < a < b − 2 < ∞ .

Here, we used a geometric observation that (a, a + 2) can intersect at most 2

connected componentsp
of G. p
Let ak ≡ k +k p−q and bl ≡ l −1+l p−q for (k, l) ∈ N0 ×N with k ≤ l. Given
a, b ∈ (0, ∞) with a + 2 < b, we choose k ∈ N and l ∈ N so that ak−1 < a ≤ ak
and that bl < b ≤ bl+1 . If b ≤ bk+1 , we note that
1 1 1 1 1
(b − a) p − q |(b − a) ∩ G| q ≤ (b − a) p − q ≤ 1.
Hence, we are interested mainly in the case of b > bk+1 . Then k < l. Note
that
(a, b) ⊂ (ak−1 , bl+1 ), bl+1 − ak−1 ≤ 3(b − a), |G ∩ (ak−1 , bl+1 )| = l − k + 1.
Thus, it follows that
n 1 1 1
o
kχG kMpq . 1 + sup (bl − ak ) p − q |(ak , bl ) ∩ G| q : k, l ∈ N, k ≤ ll (1.15)
n p p 1 1 1
o
≤ 1 + sup (l p−q − k p−q + 1) p − q (l − k) q : k, l ∈ N, 2k ≤ l
n p p 1 1 1
o
+ sup (l p−q − k p−q + 1) p − q (l − k) q : k, l ∈ N, k < l < 2k .
22 Morrey Spaces

It is easy to see that

p p p
l p−q − k p−q + 1 ∼ l p−q , l−k ∼l (1.16)
when l ≥ 2k. Also, when k < l < 2k,
p p p
l p−q − k p−q + 1 ∼ l p−q (l − k). (1.17)
Hence, inserting (1.16) and (1.17) into the above expression, we conclude
f ∈ Mpq (R).
Here and below for functions f : X → C and g : Y → C, defined on
different sets X and Y , respectively, we define their tensor product f ⊗ g :
X × Y → C by f ⊗ g(x, y) = f (x)g(y) for (x, y) ∈ X × Y . Likewise we define
the n-fold tensor product ⊗n f : X n → C.
Morrey spaces in higher dimensions can describe more precisely the singu-
larity of lower dimensions. The next example is an attempt at this approach.
Example 13. Let f ∈ Mpq (Rn ). Define g ≡ f ⊗ 1. That is,
g(x, y) = f (x) (x ∈ Rn , y ∈ Rm ).
Choose v > 0 so that
 
n+m 1 1 m 1 1
− + = −
n v q qn p q
Then g ∈ Mvq (Rn+m ). In fact, let r > 0, x0 ∈ Rn and y0 ∈ Rm . We calculate
1 1
|(x0 , y0 ) + [−r, r]n+m | v − q kgkLq ((x0 ,y0 )+[−r,r]n+m )
! q1
  n+m Z
n v1 − q1 n m
q
= |[−r, r] | × |Q(r)| qn |f (x)| dxdy ,
[−r,r]n

which shows g ∈ Mvq (Rn+m ).

The next example solidifies the above idea more precisely.
Example 14. Let n ≥ 2. Define
χ[−1,1] (2x1 )
f (x1 , x2 , . . . , xn ) ≡ ((x1 , x2 , . . . , xn ) ∈ Rn ).
|x1 |(log |x1 |)2
Z 1
dt
Mn1 (Rn ) Lploc (Rn ).
S
Then f ∈ \ In fact, = ∞ for any p > 1.
p>1 0 tp | log t|2p
Z 1
2 1
Meanwhile, kf kMn1 = dx1 < ∞.
− 12 |x1 |(log |x1 |)2
We can parametrize the above example as follows:
 
q q
Example 15. Let 0 < q < p < ∞, N ∈ N, and let a ∈ − ,− .
p−q p
 Banach function lattices 23
n an 1 1
(1) We have kfN kMpq ∼ N p + N = |[0, N ]n | p − q kfN kLq ([0,N ]n ) for fN
0
≡
N
χ(j−1,j−1+j a ) and fN = ⊗N f0N . See Example 12.
P
j=1

∞
X 1 a
(2) We set f0N ≡ N − p − q fN (· − N !) and fN ≡ ⊗n fN
0 0
= fN 0
⊗ fN ⊗···⊗
N =1
0
fN . Then f ∈ Mpq (Rn ).

We end this section with other functions defined on the real line.
N
X
Example 16. Let N ∈ N and ε ∈ (0, 1). Define f0N ≡ χ(l,l+l−ε ) . We
l=1
Z N +1
1 1 1 ε
have |[0, N + 1]| p − q kf0N kLq ([0,N +1]) ∼ N p − q , since f0N (t)dt ∼ N 1−ε .
0
1 ε
Consequently, kf0N kMpq (R) ∼ max(1, N p − q ).

1.2.3 The role of the parameters

As we have been seeing, Morrey spaces have two parameters p and q. Here,
we will ask ourselves what are their roles. One of the intuitive but important
obsevations on the parameter q is that q plays the role of local integrability
because Mpq (Rn ) is a subset of Lqloc (Rn ). This observation will be made clearer
in Theorem 18. The parameter p appears as the dilation index as we will see
in Theorem 17 below. To this end we next show the scaling law in Morrey
space Mpq (Rn ) for all 0 < q ≤ p < ∞;
Theorem 17. For all 0 < q ≤ p < ∞, we have a scaling law: kf (t·)kMpq =
n
t− p kf kMpq for all f ∈ Mpq (Rn ) and t > 0.
Proof We write out the norm in full:
Z ! q1
n n
kf (t·)kMpq = sup |B(x, r)| p−q |f (ty)| dyq
.
(x,r)∈Rn+1
+
B(x,r)

Changing variables, we have

Z ! q1
1 1 q
kf (t·)kMpq = sup |B(x, r)| p |f (y)| dy .
(x,r)∈Rn+1 |B(tx, tr)| B(tx,tr)
+

Since tn |B(x, r)| = |B(tx, tr)| for any balls B(x, r), we have the desired result.
As we discussed before, the parameter q seems to serve to describe the
local integrability of functions. We will clarify here the role of the parameter
q. Denote by L0c (Rn ) the set of all compactly supported measurable functions.
24 Morrey Spaces

Theorem 18 (The local integrability of q). Let 0 < q < r < p < ∞. Then
Mpq (Rn ) ∩ L0c (Rn ) \ Lrloc (Rn ) 6= ∅.
Proof We use the function f ≡ fN in Example 8. With α fixed, we have
  !
1 kfN kLr 1 kfN kLr
lim inf log = lim inf log > 1.
N →∞ N kfN kLq N →∞ N kfN kMpq
∞
X 1
Thus, f ≡ fN ∈ Mpq (Rn ) ∩ L0c (Rn ) \ Lrloc (Rn ).
N 2 kfN kMpq
N =1

1.2.4 Inclusions in Morrey spaces

Here, we are concerned with the relation of two different Morrey spaces or
of a Morrey space and a different function space. We ask ourselves about the
embedding between two different Morrey spaces. The first easy but important
thing to notice is that Morrey spaces are nested as is seen from Hölder’s
inequality:
Theorem 19. For all f ∈ L0 (Rn ) and 0 < q1 ≤ q0 ≤ p < ∞, kf kMpq1 ≤
1
kf kMpq0 . In particular, Mpq0 (Rn ) ,→ Mpq1 (Rn ).
Proof Using (1.1), we write out the norms in full:
! q1
Z 1
1 1
p − q1 q1
kf kMpq1 = sup |B(x, r)| |f (y)| dy (1.18)
(x,r)∈Rn+1
+
B(x,r)
! q1
Z 0
1 1
p − q0 q0
kf kMpq0 = sup |B(x, r)| |f (y)| dy (1.19)
(x,r)∈Rn+1
+
B(x,r)

By Hölder’s inequality (for probability measures), we have

! q1 ! q1
Z 1 Z 0
1 q1 1 q0
|f (y)| dy ≤ |f (y)| dy . (1.20)
|B(x, r)| B(x,r) |B(x, r)| B(x,r)

Thus, by inserting inequality (1.20) into (1.18) and (1.19), we obtain the
desired result.
The following theorem indicates that the embedding Mpq (Rn ) ,→ Mpq̃ (Rn )
is not dense unlike Lebesgue spaces over a set of finite measure.
Theorem 20. Let 1 < q < q̃ < p. Then Mpq̃ (Rn ) is not dense in Mpq (Rn ).
This result in particular shows that L∞ n p n
c (R ) is not dense in Mq (R ).

Proof Let F be the set defined in Example 11. The function χF belongs
to Mpq (Rn ) thanks to Example 11.
 Banach function lattices 25

We prove that χF is not in the closure of Mpq̃ (Rn ) by showing f ∈

/ Mpq̃ (Rn )
if f ∈ Mpq (Rn ) satisfies k2χF − f kMpq < 1. Indeed, if K is one of the con-
nected components, then kf kLq̃ (K) ≥ kf kLq (K) > 1 since 1 > k2χF − f kMpq ≥
1 1
k2 − f kLq (K) ≥ 2 − kf kLq (K) . Thus, kf kMpq̃ ≥ |[0, Rj ]n | p − q̃ kf kLq̃ ([0,Rj ]n ) ≥
jn jn 1 jn jn jn jn
R p − q̃ |Fj | q̃ = 2 q̃ − q R q − q̃ / Mpq̃ (Rn ), since this
for all j ∈ N. Hence, f ∈
is valid for all j ∈ N.
The following proposition shows that Mpq (Rn ) with 1 ≤ q < p < ∞ is
large unlike Lebesgue spaces. Denote by L1 (Rn ) + L∞ (Rn ) the linear space
of all functions which can be expressed as a sum of L1 (Rn )-functions and
L∞ (Rn )-functions. We discuss the structure of L1 (Rn ) + L∞ (Rn ) as a normed
space in Section 2.1.3.
Proposition 21. Let 1 ≤ q < p < ∞. Then Mpq (Rn ) is not included
in L1 (Rn ) + L∞ (Rn ). Namely, there exists f ∈ Mpq (Rn ) which cannot be
expressed as a sum of an integrable function and an essentially bounded func-
tion.
Proof The function f in Example 21 below belongs to f ∈ Mpq (Rn ) \
(L (Rn ) + L∞ (Rn )).
1

Recall that a quasi-metric space (X, d) is separable if there exists a count-

able set which is dense in X. Typically, (X, d) is a nondense quasi-metric mea-
sure space, if there exists an infinite set E ⊂ X such that inf d(x, y) > 0.
x,y∈E,x6=y
We will disprove that Morrey spaces are separable in general.
Proposition 22. Let 0 < q ≤ p < ∞.
(1) The Morrey space Mpq (Rn ) does not have L∞ n
c (R ) as dense subspace.

(2) The Morrey space Mpq (Rn ) is not separable.

Proof We recall the calculation for f = fn/p in Example 6.
n
(1) Observe that |·|− p cannot be approximated by the functions in L∞ n
c (R ).

(2) Let fj ≡ χB(2j ) f , j ∈ N. Then kfj − fk kMpq = kf1 − f0 kMpq > 0 for all
k, j ∈ N with k < j.

1.2.5 Weak Morrey spaces

As we did for Lebesgue spaces, we can introduce weak Morrey spaces. The
definition is as follows:
Definition 10 (Weak Morrey spaces). Let 0 < q ≤ p < ∞. For f ∈ L0 (Rn )
its weak Morrey norm is defined by kf kWMpq ≡ sup kλχ(λ,∞] (|f |)kMpq . The
λ>0
weak Morrey space Mpq (Rn ) is the set of all f ∈ L0 (Rn ) for which the norm
kf kWMpq is finite.
26 Morrey Spaces

The following example explains the gap between Mpq (Rn ) and WMpq (Rn ):
Example 17. Let 0 < q < p < ∞. Let F be the set in Example 11. Define
χF ((1 + R)−k ·)
f ≡ sup −k ·)k p
.
k∈N kχF ((1 + R) Mq

Then for all j ∈ N,

Z ! q1
1 1
j n
|[0, (1 + R) ] | p−q f (x) dxq
[0,(1+R)j ]n
(Z !q ) q1
1 1
p−q
χF ((1 + R)−k x)
≥ |[0, (1 + R)j ]n | sup −k ·)k p
dx
[0,(1+R)j ]n k∈N∩[1,j] kχF ((1 + R) Mq
1
& jq.
This disproves f ∈ Mpq (Rn ). Meanwhile

kf kWMpq = sup λkχ(λ,∞] (|f |)kWMpq

λ>0
1
= sup −k ·)k p
kχ[1,∞) (kχF ((1 + R)−k ·)kMpq |f |)kWMpq
k∈N kχF ((1 + R) Mq
1
= sup −k ·)k p
kχF ((1 + R)−k ·))kWMpq
k∈N kχF ((1 + R) Mq

= 1,
so that f ∈ WMpq (Rn ).
1
We can refine the embedding Mpr (Rn ) ,→ Mpq (Rn ) for 0 < q < r ≤ p < ∞.
Lemma 23. Let 0 < q < r ≤ p < ∞. Then WMpr (Rn ) ,→ Mpq (Rn ).
Proof Let f ∈ WMpr (Rn ). Also let Q ∈ Q. We need to estimate
Z  q1
1 1
|Q| p − q |f (x)|q dx .
Q

We use Theorem 5 to have

Z Z ∞
|f (x)|q dx = qλq−1 |{x ∈ Q : |f (x)| > λ}|dλ.
Q 0
1 1 1
Since |Q| p−r|{x ∈ Q : |f (x)| > λ}| r ≤ λ−1 kf kWMpr , we obtain
Z Z ∞
|f (x)|q dx ≤ qλq−1 min(|Q|, λ−r kf kWMpr r )dλ.
Q 0

If we calculate the right-hand side, we obtain f ∈ Mpq (Rn ).

We compare weak Lebesgue spaces and Morrey spaces.
 Banach function lattices 27

Theorem 24. Let 0 < q < p < ∞. Then WLp (Rn ) ,→ Mpq (Rn ).
Proof Let f ∈ WLp (Rn ). Then for any cube Q,
Z Z ∞
q q
|Q| p −1 |f (x)|q dx = |Q| p −1 qλq−1 |{x ∈ Q : |f (x)| > λ}|dλ
Q 0
q
Z ∞
1
≤ |Q| p qλq−1 min(1, (|Q|− p λ−1 kf kWLp )p )dλ
0
' (kf kWLp )q ,
as required.

1.2.6 Morrey spaces and ball Banach function spaces

We now consider Morrey spaces from the viewpoint of Banach function
spaces. As a starting point, define Banach lattices.
Definition 11 (Banach lattice). A Banach lattice on a measure space
(X, B, µ) is a Banach space (E(µ), k · k) contained in L0 (µ) such that, for
all f, g ∈ E(µ), the implication “|f | ≤ |g| → kf kE(µ) ≤ kgkE(µ) ” holds.
We recall that M+ (µ) is the cone of all non-negative measurable functions
on a measure space (X, B, µ).
Definition 12 (Banach function norm). A mapping ρ : M+ (µ) → [0, ∞] is
called a Banach function norm if, for all f, g, fj , (j = 1, 2, 3, . . .) in M+ (µ),
for all constants a ≥ 0 and for all measurable subsets E ∈ B, the following
properties hold:
(P1) ρ(f ) = 0 ⇔ f = 0 a.e., ρ(af ) = aρ(f ), ρ(f + g) ≤ ρ(f ) + ρ(g),
(P2) ρ(g) ≤ ρ(f ) if 0 ≤ g ≤ f a.e.,
(P3) the Fatou property; ρ(fj ) ↑ ρ(f ) holds whenever 0 ≤ fj ↑ f a.e.,
(P4) ρ(χE ) < ∞ whenever µ(E) < ∞,
(P5) whenever µ(E) < ∞, kχE f kL1 (µ) .E ρ(f ) with the implicit constant
depending on E and ρ but independent of f .
Quasi-Banach function norms are defined by replacing (P1) with a weaker
requirement (P1)0 for some α ≥ 1.
(P1)0 ρ(f ) = 0 ⇔ f = 0 a.e., ρ(af ) = aρ(f ), ρ(f + g) ≤ α(ρ(f ) + ρ(g)).
In this case, ρ is called a quasi-Banach function norm.
Here are examples of Banach function spaces.
Example 18. Let 1 ≤ p ≤ ∞. Then `p (N) and Lp (N) are Banach function
spaces.
Example 19. Let Φ be a Young function such that Φ : [0, ∞) → [0, ∞) is a
convex homeomorphism. Suppose that we have a function Φ : [0, ∞) → (0, ∞).
28 Morrey Spaces

Then define the Orlicz norm by

   
|f (x)|
Z
kf kLΦ ≡ inf λ > 0 : Φ ≤1
Rn λ
0 n
for f ∈ L (R ). See also Definition 25 below.
The symbol L0 (µ) denotes the collection of all extended scalar-valued (real
or complex) measurable functions over (X, B, µ). As usual, any two functions
coinciding a.e. will be identified.
Definition 13 (Banach function space). Let ρ be a Banach function norm
over a measure space (X, B, µ). The collection X = X (ρ) of all functions
f ∈ L0 (µ) for which ρ(|f |) < ∞ is called a Banach function space. For each
f ∈ X , define kf kX ≡ ρ(|f |).
Example 20. We go back to Example 19. The space LΦ (Rn ) is the set of all
f ∈ L0 (Rn ) for which the norm kf kLΦ is finite. Then LΦ (Rn ) is an example
of Banach function spaces.
Unfortunately, unlike Orlicz spaces, Morrey spaces are not Banach function
spaces. Let 1 ≤ q < p < ∞. We aim here to prove that Mpq (Rn ) is not a Banach
function space. More precisely, we let ρ(f ) ≡ kf kMpq for f ∈ M+ (Rn ). Then
ρ fails the property (P5); it can happen that |E| < ∞ and f ∈ Mpq (Rn ) while
kf kL1 (E) = ∞.
Example 21. Here, we exhibit an example showing that k·kMpq fails property
(P5) when 1 < q < p < ∞. For simplicity, we let n = 1 and 1 < q < 2 = p;
other cases are dealt with analogously. Let us consider the sequence
 
1 1 1 1 1 1
(a1 , a2 , . . .) = 1, , , , , , , . . . .
4 4 16 16 16 16
That is, {aj }∞
j=1 is a decreasing sequence and 4
−l
appears 2l times for l ∈ N0 .
∞
[
We define E ≡ (10j , 10j + aj ) ⊂ R. Observe that
j=1
∞
X 1 1
|E| = aj = 1 + ×2+ × 4 + · · · = 2. (1.21)
j=1
4 16
∞
X
We also define a function f by f ≡ (aj )−1/2 χ(10j ,10j +aj ) . Then f belongs
j=1
to Mpq (R). In fact, since f is supported on (0, ∞), we have
  q1
∞
1 1 X q
kf kMpq = sup (b − a) 2 − q  aj − 2 |(a, b) ∩ (10j , 10j + aj )| .
0<a<b<∞ j=1

We distinguish two cases 2a < b and 2a > b. For the latter case, there is at
most one j such that (a, b) ∩ (10j , 10j + aj ) 6= ∅. This implies
∞
X q
aj − 2 |(a, b) ∩ (10j , 10j + aj )| ≤ 1.
j=1
 Banach function lattices 29

Consequently
  q1
∞
1 1 X 1
kf kMpq ≤ 1 + sup (b − a) 2 − q  j j
q |(a, b) ∩ (10 , 10 + aj )|

0<2a<b<∞ j=1
aj 2

  q1
∞
1 1 1 1 X 1
≤ 1 + 2 q − 2 sup b 2 − q  j j
q |(0, b) ∩ (10 , 10 + aj )|

0<b<∞ a
j=1 j
2

  q1
∞
1 1 1 1 X
≤1+2 q−2 sup b 2−q  χ(0,b+3) (10j + aj )
1≤b<∞ j=1
1 1 1 1 1
≤1+2 q−2 sup b 2−q (log10 (b + 3)) q < ∞.
1≤b<∞

Thus, we have
f ∈ Mpq (R) (1.22)
Meanwhile, a direct calculation shows that:
Z ∞
1
X
f (t)dt = (4−j ) 2 · 2j = ∞. (1.23)
E j=1

Thus from (1.21)–(1.23) we see that (P5) fails.

It should be noted that the function f above is not expressed as the sum
of L1 (Rn ) + L∞ (Rn ) since (f − h)χsupp(f ) ∈
/ L1 (Rn ) for any h ∈ L1 (Rn ).
Motivated by Example 21, we present the following definition:
Definition 14 (Ball Banach function norm). A mapping ρ : M+ (Rn ) →
[0, ∞] is called a ball Banach function norm (over Rn ) if, for all f, g, fk (k =
1, 2, 3, . . .) in M+ (Rn ), for all constants a ≥ 0 and for all balls B in Rn , the
following properties hold:
(P1) ρ(f ) = 0 ⇔ f = 0 a.e.; ρ(af ) = aρ(f ), ρ(f + g) ≤ ρ(f ) + ρ(g),
(P2) ρ(g) ≤ ρ(f ) if 0 ≤ g ≤ f a.e.,
(P3) the Fatou property; ρ(fj ) ↑ ρ(f ) holds whenever 0 ≤ fj ↑ f a.e.,
(P4)0 ρ(χB ) < ∞,
(P5)0 kχB f kL1 .B ρ(f ) with the implicit constant depending on B and ρ but
independent of f .
Accordingly, the space generated by such ρ is called a ball Banach function
space. The notions of quasi-ball Banach function norms and quasi-ball Banach
function spaces are defined analogously by replacing (P1) with (P1)0 in Defi-
nition 12.
30 Morrey Spaces

Example 22. When 1 ≤ q ≤ p < ∞, then Mpq (Rn ) is a ball Banach function
1 1
space as is seen from kχB kMpq = |B| p and kf kL1 (B) ≤ |B|1− p kf kMpq .
It is noteworthy that mixed Lebesgue spaces are not Banach function
spaces but that they are ball Banach function spaces [470]. See Exercise 19.

1.2.7 Exercises
Exercise 11. Let 0 < q < u ≤ p < ∞.
(1) [393, Proposition 2.3] Let Fj be the set in Example 11. Then show that
  jn − jn
1+R q u
kχFj kMpu ≥ by considering the convex hull of Fj .
2
Mpr (Rn ) is not closed in Mpq (Rn )
S
(2) [393, Proposition 2.5] Show that
q<r≤p
by mimicking the proof of Theorem 20.
Mpr (Rn ) is not dense in Mpq (Rn )
S
(2) [393, Proposition 3.1] Show that
q<r≤p
by mimicking the proof of Theorem 20.
Exercise 12. [265, Lemma 2] Let 0 < q ≤ p < ∞. Let Ej be the set con-
q
structed in Example 10, so that |Ej | = 2−jn [2jn(1− p ) ] and that each Ej con-
tain [0, 2−j )n . Then for a positive sequence {aj }∞
j=1 , the following are equiv-
alent:
∞
aj χEj ∈ Lq (Rn ).
P
(1)
j=1

∞ jn
(2− aj )q < ∞.
P
(2) p

j=1

Exercise 13. Let f ∈ L0 (Rn ), and let 0 < q ≤ p < ∞. Recall that the
Morrey norm kf kMpq = kf kB
Mpq
is defined by Definition 8. Define

Z ! q1
1 1
kf kQ ≡ sup |Q(x, r)| p−q q
|f (y)| dy . (1.24)
Mpq
(x,r)∈Rn+1
+
Q(x,r)

Then show that kf kMpq ∼ kf kQ

Mpq
for all f ∈ L0 (Rn ).
Exercise 14.
(1) Which is true, M42 (Rn ) ⊂ L4 (Rn ), M42 (Rn ) ⊃ L4 (Rn )?
(2) Which is true for any f ∈ L0 (Rn ), kf kM42 ≤ kf k4 , kf kM42 ≥ kf k4 ?
k |f |r kMpq
Exercise 15. Let r > 1 and 1 ≤ q ≤ p < ∞. Calculate for
(kf kMpr qr
)r
f ∈ Mpr n
qr (R ) \ {0}.
 Banach function lattices 31

Exercise 16. We define the quantity kf kMpq by the following formula:

Z  q1
1 1
kf kMpq ≡ sup |I| p−q q
|f (y)| dy
I I

for f ∈ L0 (R), where I runs over all intervals in R.

(1) Let f ∈ L0 (R). Write kf kM42 out in full.
(2) Let f1 ≡ χ[0,1] . Show that kf1 kM42 = 1.
√
4
(3) Let f2 ≡ χ[0,1]∪[1,2] . Show that kf2 kM42 = 2.
q
4
(4) Let f3 ≡ χ[0,1]∪[2,3] . Show that kf3 kM42 = 4
3.

(5) Let f4 ≡ χ[0,1]∪[3,4] . Show that kf4 kM42 = 1.

(6) Let f5 ≡ χ[0,1]∪[4,5] . Show that kf5 kM42 = 1.
Exercise 17. [18, Proposition] Let 1 ≤ q ≤ p < ∞ and let g ∈ Lq (0, 1) ∩
M↓ (0, 1). Then by using the idea of the proof of Proposition 21, find a com-
pactly supported function f ∈ Mpq (Rn ) such that λf (t) & λg (t) for all t ≥ 1.
L0 (Rn ) → 
Exercise 18. Let ρ :  [0, ∞] be a ball Banach function norm. Show
that lim inf ρ(fk ) ≥ ρ lim inf fk whenever {fk }∞ + n
k=1 ⊂ M (R ).
k→∞ k→∞

Exercise 19. [470, Remark 7.21] Let 1 ≤ p1 , p2 , . . . , pn ≤ ∞ be constants.

Suppose that pj 6= pk for some 1 ≤ j < k ≤ n. Write p ≡ (p1 , p2 , . . . , pn ).
Show that the mixed Lebesgue space Lp (Rn ) is not a ball Banach function
space. Hint: Suppose n = 2 for simplicity and p1 = 2, p2 = 1. In this case,
∞
[m, m + m−1 ] × [m, m + m−1/2 ].
S
consider E ≡
m=1

1.3 Local Morrey spaces, Bσ -spaces, Herz spaces and

Herz–Morrey spaces
Let 0 < q ≤ p < ∞. Recall that the Morrey space Mpq (Rn ) is the set of
1 1
all functions f ∈ L0 (Rn ) for which kf kMpq ≡ sup |Q| p − q kf kLq (Q) is finite. In
Q∈Q
Section 1.3.1, by introducing local Morrey spaces, we consider what happens
if we assume that the center of cubes is the origin. Section 1.3.2 considers
equivalent expressions to local Morrey spaces. We will briefly consider function
spaces related to local Morrey spaces.
32 Morrey Spaces

1.3.1 Local Morrey spaces

We will solidify our plan above; we will define local Morrey spaces by fixing
the center of the cube in the definition of Morrey norms and then investigate
the relation between Morrey spaces and local Morrey spaces. Local Morrey
spaces have different names, which we will see in Section 1.3.1.
Definition 15 (Local Morrey norm). Let 0 < q ≤ p ≤ ∞. The local Morrey
norm k · kLMpq is defined by
1 1
kf kLMpq ≡ sup |Q(r)| p − q kf kLq (Q(r)) (1.1)
r>0

for f ∈ L0 (Rn ). The local Morrey space LMpq (Rn ) is the set of all f ∈ L0 (Rn )
for which kf kLMpq is finite. Likewise, define the weak local Morrey space
WLMpq (Rn ) as the set of all f ∈ L0 (Rn ) for which the norm kf kWLMpq =
sup λkχ(λ,∞] (|f |)kMpq is finite.
λ>0

Unlike Morrey spaces, we may tolerate the case where 0 < p, q ≤ ∞ or

−∞ < p < 0 < q ≤ ∞. However, we feel that it is enough to suppose
0 < q ≤ p < ∞ for our main purpose.
Example 23. Let 0 < p, q ≤ ∞ or −∞ < p < 0 < q ≤ ∞. Define the local
Morrey norm formally by (1.1).
(1) Let 0 < q ≤ p < ∞. Any element in Mpq (Rn ) belongs to LMpq (Rn ).
(2) Let 0 < q ≤ p < ∞. Any Lq (Rn )-function supported on a cube which
does not contain the origin is an element in LMpq (Rn ). This is a crucial
difference from the Morrey space Mpq (Rn ); this is a good contrast to
Theorem 18. Let f ∈ L0 (Rn ) be supported on B(2) \ B(1), for example.
Then f ∈ LMpq (Rn ) if and only if f ∈ Lq (Rn ). If f ∈ Lqc (Rn ), 0 < q <
∞, is supported away from the origin, then f ∈ LM∞ n
q (R ).

(3) If 0 < p < q ≤ ∞ or −∞ < p < 0 < q ≤ ∞, then LMpq (Rn ) = {0},
which is similar to Mpq (Rn ) = {0}.
We collect some elementary inclusion properties of local Morrey spaces.
Proposition 25.
(1) Let 0 < p < ∞. Then LMpp (Rn ) = Lp (Rn ) with coincidence of norms.
1
(2) If 0 < q1 ≤ q2 ≤ p < ∞, then LMpq2 (Rn ) ,→ LMpq1 (Rn ).
(3) Let 1 ≤ q ≤ p < ∞. Then WLMpq (Rn ) and LMpq (Rn ) are complete. We
have the following inclusions:
1 1 1
WMpq (Rn ) ,→ WLMpq (Rn ), Mpq (Rn ) ,→ LMpq (Rn ) ,→ WLMpq (Rn ).
 Banach function lattices 33
n
(4) Let 1 ≤ q < p < ∞. Then | · |− p ∈ LMpq (Rn )
Proof Overall, the proofs are careful reexaminations of the case of Morrey
spaces.
(1) Reexamine the proof of Theorem 16.
(2) Reexamine the proof of Theorem 19.
(3) Completeness of these spaces is clear because these spaces are continu-
ously embedded into WLqloc (Rn ). Embeddings follow from the definition
of the norms.
1
(4) This is a simple consequence of the embedding Mpq (Rn ) ,→ LMpq (Rn )
and Example 6.
Some people call local Morrey spaces Bσ spaces. This simply reformulates
the definition.
Definition 16. Let E be a ball quasi-Banach function space. Then, for σ ∈
[0, ∞), we define the E-based nonhomogeneous Bσ space Bσ (E)(Rn ) and the
E-based homogeneous Bσ space Ḃσ (E)(Rn ) as the sets of f ∈ L0 (Rn ) for which
kf kBσ (E) ≡ sup r−σ kf χQ(r) kE < ∞ and kf kḂσ (E) ≡ sup r−σ kf χQ(r) kE < ∞,
r≥1 r>0
respectively.
Example 24. As the candidate of E(Rn ), we can list the following function
spaces:
(1) E = Lp (Rn ) with 1 ≤ p ≤ ∞. In this case the corresponding func-
tion spaces Bσ (E)(Rn ) and Ḃσ (E)(Rn ) are written Bσ (Lp )(Rn ) and
Ḃσ (Lp )(Rn ) and are called a homogeneous Bσ -Lp space and a nonho-
mogeneous Bσ -Lp space, respectively. It is noteworthy that
Ḃ nq − np (Lq )(Rn ) ≈ LMpq (Rn )
with equivalence of norms for any 1 < q ≤ p < ∞. We define
Z ! p1
1
kf kḂ p,λ ≡ sup r−λ |f (x)|p dx
r>0 |Q(r)| Q(r)
and ! p1
Z
1
kf kB p,λ ≡ sup r−λ |f (x)|p dx .
r≥1 |Q(r)| Q(r)

(2) E = WLp (Rn ) with 1 ≤ p ≤ ∞. In this case the corresponding func-

tion spaces Bσ (E)(Rn ) and Ḃσ (E)(Rn ) are written Bσ (WLp )(Rn ) and
Ḃσ (WLp )(Rn ) and are called a homogeneous Bσ -weak Lp space and a
nonhomogeneous Bσ -weak Lp space, respectively. It is noteworthy that
Ḃ nq − np (WLq )(Rn ) ≈ WLMpq (Rn )
with equivalence of norms for any 1 < q ≤ p < ∞.
34 Morrey Spaces

(3) E = Mpq (Rn ) with 1 ≤ q ≤ p < ∞. In this case the corresponding

function spaces Bσ (E)(Rn ) and Ḃσ (E)(Rn ) are written Bσ (Mpq )(Rn )
and Ḃσ (Mpq )(Rn ) and are called a homogeneous Bσ -Morrey space and
a nonhomogeneous Bσ -Morrey space, respectively.
(4) E = WMpq (Rn ) with 1 ≤ q ≤ p < ∞. In this case the corresponding
function spaces Bσ (E)(Rn ) and Ḃσ (E)(Rn ) are written Bσ (WMpq )(Rn )
and Ḃσ (WMpq )(Rn ) and are called a homogeneous Bσ -weak Morrey
space and a nonhomogeneous Bσ -weak Morrey space, respectively.
Before we conclude this section, we give an example of an element in
LMpq (Rn ).
Example 25. Let 0 < q ≤ p < ∞ and 0 < u < ∞ and define
n
f (x) ≡ |x − 3e1 |− u χB(3e1 ,2) (x) (x ∈ Rn ).

Then f belongs to LMpq (Rn ) if and only if u < q.

1.3.2 Herz spaces and Herz–Morrey spaces

We also obtain a norm equivalent to the local Morrey norm k·kLMpq . In fact,
we will show that Herz spaces are isomorphic to local Morrey spaces. To define
Herz spaces, we set up notation. Write Q0 ≡ Q(1) and Cj ≡ Q(2j ) \ Q(2j−1 )
for j ∈ Z.
Here, we content ourselves with the definition of Herz spaces and a compar-
ison with local Morrey spaces. To simplify the notation, we use the following
norm: For a system {fj }j∈J ⊂ L0 (Rn ) of functions, define
  q1
X
kfj k`q (Lp ) ≡ k{fj }j∈J k`q (Lp ) = kfj kLp q  .
j∈J

The space `q (Lp , Rn ) is the set of all collections {fj }∞

j=1 for which the quantity
k{fj }j∈J k`q (Lp ) is finite.
Definition 17. Let 0 < p, q ≤ ∞ and α ∈ R.
α
(1) The nonhomogeneous Herz space Kpq (Rn ) is the set of all f ∈ L0 (Rn ) for
which the following norm kf kKpq
α ≡ kχQ f kLp + k{2
0
jα
χCj f }∞
j=1 k`q (Lp )
is finite.
α
(2) The homogeneous Herz space K̇pq (Rn ) collects all f ∈ L0 (Rn ) for which
the following norm kf kK̇ α ≡ k{2 χCj f }∞
jα
j=−∞ k`q (Lp ) is finite.
pq

Example 26. Let α ∈ R, 0 < p ≤ ∞ and 0 < q < ∞.

α
(1) χ[1,2]n ∈ K̇pq (Rn ) ∩ Kpq
α
(Rn ).
 Banach function lattices 35
α
(2) χB(1) ∈ K̇pq (Rn ) if and only if either one of the following conditions
holds:
n
(i) α + p ≥ 0 and 0 < q < ∞.
n
(ii) α + p = 0 and q = ∞.
α
(2) χB(1) ∈ Kpq (Rn ).
α
(3) 1 − χB(1) ∈ K̇pq (Rn ) if and only if either one of the following conditions
holds:
n
(i) α + p ≤ 0 and 0 < q < ∞.
n
(ii) α + p = 0 and q = ∞.
α n
(4) 1 ∈ K̇pq (Rn ) if and only if α + p = 0 and q = ∞.
n
(5) Let β ∈ R. | · |β ∈ K̇pq
α
(Rn ) if and only if α + β + p = 0 and q = ∞.
One of the important things is that Herz spaces and local Morrey spaces
are isomorphic as the following theorem shows. So, we can transplant mutually
the results for these function spaces [126].
n n
Theorem 26. Let 0 < q < p < ∞. Write α ≡ − < 0. Then
p q
α
K̇p∞ (Rn ) ≈ LMpq (Rn )
with equivalent norms.
Proof It is easy to see that LMpq (Rn ) ⊂ K̇p∞
α
(Rn ). Let f ∈ K̇p∞
α
(Rn ).
Then
1 1
kf kLMpq = sup |Q(r)| p − q kf kLq (Q(r))
r>0
0
X 1 1
≤ sup |Q(r)| p − q kf kLq ([−2j r,2j r]n \[−2j−1 r,2j−1 r]n )
r>0
j=−∞
0
X n n
≤ 2j q −j p kf kK̇ α
p∞
j=−∞

∼ kf kK̇ α .
p∞

Motivated by Theorem 26, we define the base LQ of cubes by

LQ ≡ {Q ∈ D : dist∞ ({0}, Q) = `(Q)}. (1.2)
A local cube is a cube in LQ. A direct consequence of Theorem 26 is that
1 1
kf kLMpq ∼ sup u(Q) p − q kf χQ kLq for any f ∈ L0 (Rn ) whenever 1 < q < p <
Q∈LQ
∞.
As a similar object, we consider Herz–Morrey spaces. Here, we content
ourselves with the definition and some examples of functions in Herz–Morrey
spaces.
36 Morrey Spaces

Definition 18. Let 0 < p, q ≤ ∞ and α, λ ∈ R. The Herz–Morrey space

α,λ
M Kpq (Rn ) is defined as the set of all f ∈ L0 (Rn ) for which the following
norm
  q1
L
X
−Lλ 
kf kM Kpq
α,λ ≡ kχQ0 f kLp + sup 2 (2jα kχCj f kLp )q 
L∈N0 j=1

is finite.
Example 27. Let 0 < p, q ≤ ∞ and α, λ ∈ R. Let f (x) ≡ |x|β χRn \B(1) (x),
n
x ∈ Rn with β ∈ Rn . Then 2jα kχCj f kLp ∼ 2j(α+β+ p ) .
n
(1) If α + β + p
α,λ
> 0, then f ∈ M Kpq (Rn ) if and only if λ ≥ α + β + np .
n α,λ
(2) If α + β + p = 0, then f ∈ M Kpq (Rn ) if and only if λ > 0.
n α,λ
(3) If α + β + p ≤ 0, then f ∈ M Kpq (Rn ) if and only if λ ≥ 0.

1.3.3 Exercises
Exercise 20. Let 1 ≤ p ≤ ∞ and σ ≥ 0.
(1) Show that χQ(1) ∈ Bσ (Lp )(Rn ) for all σ ≥ 0 and 1 ≤ p ≤ ∞.
n
(2) Show that χQ(1) ∈ Ḃσ (Lp )(Rn ) if and only if σ ≤ .
p
(3) Show that χ[1,2]n ∈ Bσ (Lp )(Rn ) for all σ ≥ 0 and 1 ≤ p ≤ ∞.

(4) Show that χ[1,2]n ∈ Ḃσ (Lp )(Rn ) for all σ ≥ 0 and 1 ≤ p ≤ ∞.
Exercise 21. Let n
0 < q ≤ p < ∞ and f ∈ L0 (Rn ). Then show that
−
kf (t·)kLMpq = t kf (·)kLMpq for all t > 0.
p

Exercise 22 (Diversity of local Morrey spaces). Let 0 < q < r ≤ p < ∞. Let
F be the set defined in Example 11. Then using χF , show that LMpq (Rn ) and
LMpr (Rn ) are different.
n
Exercise 23. Let 0 < q < p < ∞, and let f (x) = |x|− p .
(1) Show that kf kLMpq = kf kMpq .
(2) Show that dist(f, L∞ n
c (R )) = kf kLMq .
p

(3) Conclude that L∞ n p n

c (R ) is not dense in LMq (R ).

α,λ
Exercise 24. Let 0 < p, q ≤ ∞ and α, λ ∈ R. Show that M Kpq (Rn ) is a
quasi-ball Banach function space.
n
Exercise 25. Let 0 < q < p < ∞, and let f (x) = |x|− p .
 Banach function lattices 37

(1) Show that there exists an increasing sequence {Rj }∞ j=1 of positive real
numbers such that R1 = 1, kf χB(Rj+1 )\B(Rj ) kLMpq ≥ (1 − 2−j )kf kLMpq .
(2) Define
∞
X
Φ : {aj }∞ ∞
j=1 ∈ ` (N) 7→ aj f χB(Rj+1 )\B(Rj ) ∈ LMpq (Rn ).
j=1

Establish that kΦk`∞ (N)→LMpq (Rn ) = 1.

(3) Define
∞
X
Ψ : {aj }∞ ∞
j=1 ∈ ` (N) 7→ aj f χB(Rj+1 )\B(Rj ) ∈ Mpq (Rn ).
j=1

Establish that kΨk`∞ (N)→Mpq (Rn ) = 1.

1.4 Distributions and Lorentz spaces

To analyze operators appearing in many problems, the pointwise estimate
often fails to work well. In view of the Layer-Cake formula theorem, it seems
to make sense to analyze µ{x ∈ X : |f (x)| > λ} for each λ. Hence, we are
interested in the size of this set.
Section 1.4 collects the way of measuring the size of functions using this
type of sets in Section 1.4.2; Lorentz spaces are defined there. To this end we
define distributions in Section 1.4.1. Keeping the structure of the (quasi-)norm
defined in Section 1.4.2, we investigate the Hardy operator in Section 1.4.3.
Since the decreasing rearrangement is a decreasing function, we collect some
useful inequalities for monotone functions and apply it to Lorentz spaces in
Section 1.4.4.

1.4.1 Distribution function

n
For any 0 < p < ∞n | · |− p ∈/ Lp (Rn ). On many occasions, this will be a
−p
problem because | · | appears in many places. To overcome this problem,
we seek to find function spaces. Lorentz spaces can be used for this purpose.
We need the distribution function to define Lorentz spaces. Hence, we define
the distribution function of measurable functions and then we collect some
examples.
As a setup, recall the definition of λf = λf,µ for f ∈ L0 (µ) given in
Definition 5 via some examples.
38 Morrey Spaces

Example 28. Consider the Euclidean space Rn here. Let γ 6= 0. Let fγ and
gγ be as in Example 1. For each t > 0, we have:

(1) (i) If γ > 0, then λfγ (t) = max(vn (1 − tn/γ ), 0).

(ii) If γ < 0, then λfγ (t) = min(vn , vn tn/γ ).
(2) (i) If γ > 0, then λgγ (t) = ∞.
(ii) If γ < 0, then λgγ (t) = max(vn (tn/γ − 1), 0)
(3) λχ[0,1] (t) = χ[0,1) (t).
Example 29. Let E1 , E2 , . . . , EN be a sequence of disjoint measurable sets.
PN
Also let 0 < α1 < α2 < · · · < αN . Then if we set f ≡ αj χEj , then
j=1

N
X −1
f ∗ = αN χ(0,µ(EN )) + αk χ[µ(EN ∪EN −1 ∪···∪Ek+1 ),µ(EN ∪EN −1 ∪···∪Ek )) .
k=1

Let I be an interval. By M↓ (I) we mean the set of all non-negative non-

increasing functions defined on I and by M↑ (I) we mean the set of all non-
negative nondecreasing functions defined on I. We collect a fundamental prop-
erty of λf .
Theorem 27. Let f ∈ L0 (µ). Then the distribution function λf = λf,µ
belongs to M↓ (0, ∞) and is continuous from the right.
Proof It follows from Definition 5 that λf is decreasing. Let t0 ∈ [0, ∞).
Then for any decreasing sequence {tk }∞
k=1 satisfying lim tk = t0 ,
k→∞

∞
[
{x ∈ Rn : |f (x)| > t0 } = {x ∈ Rn : |f (x)| > tk }
k=1

and

{x ∈ Rn : |f (x)| > tk } ⊂ {x ∈ Rn : |f (x)| > tk+1 } (k ∈ N).

Thus, we are in the position of using the monotone convergence theorem.

Although the function f 7→ λf (t) is not linear for fixed t, we still have the
following substitute:
Lemma 28. Let f, g ∈ L0 (µ). Then λf +g (t1 + t2 ) ≤ λf (t1 ) + λg (t2 ) for any
t1 , t2 ≥ 0.
Proof If |f (x) + g(x)| > t1 + t2 , then |f (x)| > t1 or |g(x)| > t2 . Thus
{|f + g| > t1 + t2 } ⊂ {|f | > t1 } ∪ {|g| > t2 }. If λf +g (t1 + t2 ) > λf (t1 ) + λg (t2 ),
then there would exist a1 , a2 ∈ (0, ∞) such that λf +g (t1 + t2 ) > a1 + a2 ,
 Banach function lattices 39

λf (t1 ) < a1 and λf (t2 ) < a2 . This means that µ{|f + g| > t1 + t2 } > a1 + a2
and that µ{|f | > t1 } ≤ a1 , µ{|f | > t2 } ≤ a2 . This is a contradiction.
The distribution function is used to define Lorentz spaces. The definition
is as follows:

Definition 19. Let f ∈ L0 (Rn ). Then its decreasing rearrangement f ∗ is the

function defined on (0, ∞) by

f ∗ (t) ≡ inf({s ∈ [0, ∞) : λf (s) ≤ t} ∪ {∞}) (t > 0).

We have two possibilities in the above since λf is monotone:

{s ∈ [0, ∞) : λf (s) ≤ t} ∪ {∞} = (f ∗ (t), ∞] or [f ∗ (t), ∞].

It is also clear from the definition of distribution functions that f ∗ (t) ≤ g ∗ (t)
if f, g ∈ L0 (Rn ) satisfy |f | ≤ |g|.
We consider a couple of examples.
Example 30. Let γ < 0. Let fγ and gγ be as in Example 1. Then
 v γ/n
n
fγ∗ (t) = χ[0,vn ) (t) (0 < t < ∞)
t
and  γ/n
t
gγ∗ (t) = +1 (0 < t < ∞).
vn
Example 31. Since λχ[0,1] = χ[0,1) ,
(
[1, ∞) 0 < t < 1,
{s ∈ [0, ∞) : λχ[0,1] (s) ≤ t} =
[0, ∞) t ≥ 1.

Thus (χ[0,1] )∗ = χ(0,1) .

Example 32. Let f be as in Example 32. Then for all t > 0,
 ∗
XN
 αj χEj  (t)
j=1
N
X −1
= αN χ(0,µ(EN )) + αk χ[µ(EN ∪EN −1 ∪···∪Ek+1 ),µ(EN ∪EN −1 ∪···∪Ek )) .
k=1

The following lemma shows the relation between f ∗ and λf :

Lemma 29. Let f ∈ L0 (µ). Then for any t > 0, λf (f ∗ (t)) ≤ t, namely,
µ{x ∈ X : |f (x)| > f ∗ (t)} ≤ t.
40 Morrey Spaces

Proof We may assume that f ∗ (t) < ∞; otherwise the conclusion follows
readily from 0 = λf (∞) ≤ t. Suppose f ∗ (t) < ∞. Then, since f ∗ (t) = inf{s ∈
[0, ∞) : λf (s) ≤ t}, we have λf (s) ≤ t for all s > f ∗ (t). Therefore, by
Theorem 27, λf (f ∗ (t)) = lim
∗
λf (s) ≤ t. Thus, the proof is complete.
s↓f (t)

An analogy to Lemma 29 holds. Note that the inequality reverses.

Lemma 30. For any t > 0 and f ∈ L0 (µ), µ{x ∈ X : |f (x)| ≥ f ∗ (t)} ≥ t.
Proof Since f ∗ (t) = inf({s ∈ [0, ∞) : λf (s) ≤ t}∪{∞}), we have λf (s) ≥
t for all s < f ∗ (t). Therefore
µ{x ∈ X : |f (x)| ≥ f ∗ (t)} = lim
∗
µ{x ∈ X : |f (x)| > s} ≥ t.
s↑f (t)

Thus, the proof is complete.

Lemmas 29 and 30 are important when we consider distributions of func-
tions.
The following proposition together with Example 32 can be used to approx-
imate the distribution functions.
Proposition 31. Let {fj }∞ 0
j=1 ⊂ L (µ) satisfy 0 ≤ fj ≤ fj+1 . Then f ≡
∗ ∗
lim fj satisfies f (t) = lim (fj ) (t) for all t > 0.
j→∞ j→∞

Proof It is by the monotonicity clear that f ∗ (t) ≥ lim (fj )∗ (t). So, we
j→∞
need to establish that f ∗ (t) ≤ lim (fj )∗ (t). Assume otherwise: a = f ∗ (t) −
j→∞
lim (fj )∗ (t) > 0. Then f ∗ (t) − a ∈
/ {s ∈ [0, ∞) : λf (s) ≤ t}, or equivalently,
j→∞
λf (f ∗ (t) − a) > t. Consequently, there exists b > 0 such that λf (f ∗ (t) − a) >
b + t. Thus, λfj (f ∗ (t) − a) > b + t by the monotone convergence theorem
as long as j  1. Consequently, λfj ((fj )∗ (t)) > b + t for such j. This is a
contradiction to Lemma 29.
Example 33. Let f ∈ M↓ (0, ∞). Then we can approximate f almost every-
where with simple decreasing functions. Since fj is a simple decreasing func-
tion, then (fj )∗ = fj as is seen from Example 32. Thanks to Proposition 31,
f ∗ = f almost everywhere.
In a special case like Example 30, we can regard λf as the inverse of f ∗ .
Proposition 32. Let f ∈ L0 (µ). Assume that λf is positive, strictly decreas-
ing and continuous. Then;
(1) λ−1 ∗
f (t) = f (t) for all t > 0,

(2) λf (f ∗ (t)) = t for all t > 0.

Proof
(1) This is clear since
f ∗ (t) = inf({s ∈ [0, ∞) : λf (s) ≤ t}) = inf({s ∈ [0, ∞) : s ≤ λ−1
f (t)}).
 Banach function lattices 41

(2) Simply observe that λf (λ−1

f (t)) = t and use (1).

In general, it may happen that λf (f ∗ (t)) 6= t as the example of the function

f = χ[0,2] and the point t = 1 shows.
The distribution is important because we can express the Lp norm in terms
of distributions.
Theorem 33. For any f ∈ L0 (µ) and 0 < p ≤ ∞, kf kLp (µ) = kf ∗ kLp (0,∞) .

Proof Simply consider the case where f is a simple function. In this case,
the proof is direct.
As an application of Theorem 33, we justify the name of WLp (µ):
1
Lemma 34. Let 0 < p < ∞. Then Lp (µ) ,→ WLp (µ).
Proof Let f ∈ Lp (µ). According to Theorem 33, we have
Z ∞  p1 Z t  p1
∗ p ∗ p
kf kLp = f (s) ds ≥ f (s) ds .
0 0

Since f ∗ ∈ M↓ (0, ∞), we have

Z t  p1
1
∗
kf kLp ≥ f (t) ds p
= t p f ∗ (t).
0

1
Hence, kf kLp ≥ sup t f ∗ (t) = kf kWLp . Thus, the proof is complete.
p

t>0

The following example is a useful application of Theorem 33.

Example 34. Let (X, B, µ) be a measure space, and let E be a measurable
Z |E|
subset of X. Then kχE f kL1 (µ) ≤ f ∗ (s)ds for g ∈ L0 (µ). In fact,
0
Z Z
|f (x)|dµ(x) = χE (x)|f (x)|dµ(x)
E
ZX∞
= (χE f )∗ (t)dt
0
Z |E|
= (χE f )∗ (t)dt
0
Z |E|
≤ f ∗ (s)ds.
0

To conclude Section 1.4.1, we state an inequality connecting f ∗ and g ∗

with (f + g)∗ .
42 Morrey Spaces

Proposition 35. For any t1 , t2 > 0 and f, g ∈ L0 (µ), (f + g)∗ (t1 + t2 ) ≤

f ∗ (t1 ) + g ∗ (t2 ). In particular, for t > 0
   
∗ ∗ t ∗ t
(f + g) (t) ≤ f +g .
2 2

Proof By Lemma 29, λf (f ∗ (t1 )) ≤ t1 and λg (g ∗ (t2 )) ≤ t2 . Thanks to

Lemma 28, λf +g (f ∗ (t1 ) + g ∗ (t2 )) ≤ t1 + t2 . Hence

f ∗ (t1 ) + g ∗ (t2 ) ∈ {s ∈ [0, ∞) : λf +g (s) ≤ t1 + t2 },

which implies f ∗ (t1 ) + g ∗ (t2 ) ≤ (f + g)∗ (t1 + t2 ). Thus, the proof is complete.

1.4.2 Lorentz spaces

Section 1.4.2 works in a measure space (X, B, µ) because we can readily
pass our definitions and our results to (X, B, µ). As a generalization of weak
Lebesgue-spaces, we also define Lorentz spaces. We will see that Lorentz spaces
generalize Lebesgue spaces.
Definition 20.
(1) Let 0 < p < ∞ and 0 < q < ∞. Then the Lorentz space Lp,q (µ) is the
set of all f ∈ L0 (µ) for which the quasi-norm
Z ∞  q1
1
∗ q dt
kf k Lp,q (µ) ≡ (t f (t))
p

0 t

is finite.
(2) The space Lq,∞ (µ) stands for L∞ (µ) for any 0 < q ≤ ∞.
(3) If 0 < p < ∞, then the Lorentz space Lp,∞ (µ) denotes the weak Lp (µ)-
space: Lp,∞ (µ) = WLp (µ).
As usual if (X, B, µ) is the Lebesgue measure, then we write Lp,q (Rn ) =
Lp,q (µ) and k · kWLp,q = k · kWLp,q (µ) .
Theorem 33 shows that Lp,p (µ) = Lp (µ) with coincidence of norms. Indeed,
Z ∞  p1 Z ∞  p1
1
∗ p dt ∗ p
kf kLp,p = (t f (t))
p = f (t) dt = kf kLp .
0 t 0

When we consider integral inequalities in (0, ∞), we frequently face the fol-
lowing norm:
 Banach function lattices 43

Definition 21 (Φλ,q (0, ∞)).

(1) Let λ ∈ R and 0 < q < ∞. Then Φλ,q (0, ∞) denotes the set of all
ϕ ∈ M+ (0, ∞) for which
Z ∞ 1
dt q
kϕkΦλ,q (0,∞) ≡ (t−λ ϕ(t))q < ∞.
0 t
(2) Let λ ∈ R. Then Φλ,∞ (0, ∞) denotes the set of all ϕ ∈ M+ (0, ∞) for
which kϕkΦλ,∞ (0,∞) ≡ sup t−λ ϕ(t) < ∞.
t>0

In terms of Definition 21, kf kLp,q = kf ∗ kΦp−1 ,q (0,∞) for all f ∈ L0 (Rn ).

Example 35. Let λ, a ∈ R and 0 < q < ∞.
(1) χ[1,2] ∈ Φλ,q (0, ∞).
(2) f (t) ≡ ta , t > 0 never belongs to Φλ,q (0, ∞).
(3) g(t) ≡ e−t belongs to Φλ,µ (0, ∞) if and only if λ < 0.
(4) If f ∈ M↓ (0, ∞), then f ∗ (t) = f (t) for all t > 0, so that kf kLp,q (0,∞) =
Z ∞  q1
1
q dt
(t f (t))
p .
0 t
Example 36. Let q < ∞ and λ < 0. If ϕ ∈ M↓ (0, ∞) and ϕ(t0 ) > 0 for some
t0 > 0, then kϕkΦλ,q (0,∞) = ∞. Indeed,
Z ∞  q1
kϕkΦλ,q (0,∞) ≥ ϕ(t0 ) t−λq−1 dt = ∞.
t0
So far we do not go into the detail of the theory of Lorentz spaces because
Lorentz spaces are not the main objects of this book. However, in the context
of real interpolation, Lorentz spaces will appear in this book; see Chapter 18.

1.4.3 Hardy operators and Hardy’s inequality

The Hardy operator, which is defined by
1 t
Z
Hf (t) ≡ f (s)ds (1.1)
t 0
for suitable f ∈ L0 (0, ∞), is used to control functions defined on (0, ∞). The
Hardy operator will play the role of a toy model of the Hardy–Littlewood
maximal operator, which we take up later. This operator will be used to
consider the function spaces of local type in this book. Hardy’s inequality is
a useful inequality that can be used to control some integrals.
The multiplication by monomials in the set Φλ,q (0, ∞) is described as
follows:
Example 37. Let λ, a ∈ R and 0 < q ≤ ∞, and let ϕ ∈ Φλ,q (0, ∞). Then
Z ∞  q1
a (λ−a)q q dτ
k · ϕkΦλ,q (0,∞) = kϕkΦλ−a,q (0,∞) = τ ϕ(τ ) .
0 τ
Similarly, we have the dilation law.
44 Morrey Spaces

Example 38. Let λ ∈ R and 0 < q ≤ ∞. Then kϕ(t·)kΦλ,µ = tλ kϕkΦλ,µ for

all ϕ ∈ Φλ,µ (0, ∞) and t > 0.
The scaling law of Φλ,q (0, ∞) is summarized in the following lemma:
Lemma 36. Let q, r ∈ (0, ∞), λ0 , λ1 , λ ∈ (0, ∞) and 0 < θ < 1. Suppose that
λ = (1 − θ)λ0 + θλ1 . Then for all ϕ ∈ L0 (0, ∞), we have
  r1
kϕkΦλ,q (0,∞) = k ·−λ0 r−1 |ϕ|r kΦθ(λ q (0,∞) .
1 −λ0 )r−1, r

Proof We calculate
  r1
k ·−λ0 r−1 |ϕ|r kΦθ(λ −λ )r−1, q (0,∞)
1 0 r
Z ∞  q1
−θ(λ1 −λ0 )r−λ0 r rq q dη
= (η ) ϕ(η)
0 η
Z ∞  q1
dη
= η −θ(λ1 −λ0 )q−λ0 q ϕ(η)q
0 η
Z ∞  q1
dη
= η −λq ϕ(η)q
0 η
= kϕkΦλ,q (0,∞) .

We now will consider the space Φ↑λ,q (0, ∞) ≡ Φλ,q (0, ∞) ∩ M↑ (0, ∞). We
define kf kΦλ,q (0,∞) ≡ kf kΦλ,q (0,∞) for f ∈ Φ↑λ,q (0, ∞). One of the promi-
nent properties in the scale {Φλ,q (0, ∞)}q>0 is the monotonicity in the second
parameter q.
Lemma 37. Let 0 < q0 < q1 ≤ ∞, and let 0 < λ < ∞. Then Φ↑λ,q0 (0, ∞) ⊂
1 1
Φ↑λ,q1 (0, ∞). More quantitatively, kϕkΦλ,q1 (0,∞) ≤ (λq0 ) q0 − q1 kϕkΦλ,q0 (0,∞) for
all ϕ ∈ Φ↑λ,q0 (0, ∞).
Proof First let q1 = ∞. Then
Z ∞  q1
−λ
1 dτ 0
kϕkΦλ,∞ (0,∞) = sup t ϕ(t) = (λq0 ) q0
sup ϕ(t).
t>0 t>0 t τ 1+λq0

Since ϕ ∈ M↑ (0, ∞), we have

Z ∞  q1
1 dτ 0 1
kϕkΦλ,∞ (0,∞) ≤ (λq0 ) q0 sup ϕ(τ )q0 = (λq0 ) q0 kϕkΦλ,q0 (0,∞) .
t>0 t τ 1+λq0
q0 q0
If q1 < ∞, then kϕkΦλ,q1 (0,∞) ≤ (kϕkΦλ,∞ (0,∞) )1− q1 (kϕkΦλ,q0 (0,∞) ) q1 by
virtue of Hölder’s inequality. Thanks to what we have proven in the case
1 1
q1 = ∞ kϕkΦλ,q1 (0,∞) ≤ (λq0 ) q0 − q1 kϕkΦλ,q0 (0,∞) , as desired.
 Banach function lattices 45

We move on to Hardy operators. Let f ∈ L1loc (0, ∞). Then define

Z t Z ∞
1 1
Hf (t) ≡ f (s)ds, Hf (t) ≡ f (s)ds (t > 0)
t 0 t t

as long as the integral makes sense. Both operators are called Hardy operators.
min(1, t) max(1 − t, 0)
Example 39. Since Hχ(0,1) (t) = and Hχ(0,1) (t) =
t t
for t > 0. So H and H fail to be L1 (0, ∞)-bounded.

Chapters 7, 13 and 18 use Hardy’s inequalities.

Theorem 38 (Hardy’s inequality). Let 1 ≤ q ≤ ∞.
1
(1) For g ∈ Φµ,q (0, ∞) and µ > −1, kHgkΦµ,q (0,∞) ≤ kgkΦµ,q (0,∞) .
µ+1
1
(2) For g ∈ Φ−µ,q (0, ∞) and µ > 1, kHgkΦ−µ,q (0,∞) ≤ kgkΦ−µ,q (0,∞) .
µ−1
Proof We omit the proof of the easy case where q = ∞. Also, if q = 1,
there will be no need to use Minkowski’s inequality. In this case we use Fubini’s
theorem instead; see Exercise 31.
(1) We write out what we need to prove:
(Z q ) q1  q1
∞  Z t Z ∞
−µ−1 dt 1 −qµ q dt
t g(η)dη ≤ t g(t) .
0 0 t µ+1 0 t

We change variables: η = tτ to obtain

(Z q ) q1 (Z  q ) q1
∞  Z t ∞ Z 1
dt dt
t−µ−1 g(η)dη = t−µ g(tτ )dτ .
0 0 t 0 0 t

By Minkowski’s inequality we have

(Z q ) q1  q1
∞  Z t Z 1 Z ∞
−µ−1 dt −µ q dt
t g(η)dη ≤ (t g(tτ )) dτ.
0 0 t 0 0 t

If we change variables once again: η = tτ , then

(Z q ) q1  q1
∞  Z t Z 1 Z ∞
−µ dt µ −µ q dt
t g(η)dη ≤ τ (t g(t)) dτ.
0 0 t 0 0 t

Since µ > −1, we obtain the desired result once we integrate the right-
hand side against τ .
46 Morrey Spaces

(2) Let ν > −1 satisfy −µ = −ν − 2. Then we simply change variables in

1
kHgkΦν,q ≤ kgkΦν,q . We consider s = t−1 .
1+ν
It is important that we do not tolerate the case µ = −1 and q = 1. In
fact, if there exists the estimate kHgkL1 (0,∞) . kgkL1 (0,∞) , then this would
contradict Example 39.
A quasi-normed space is said to be normable if there exists a norm equiv-
alent to the quasi-norm. We provide an application of Hardy’s inequality.
Corollary 39. Let 1 < p < ∞ and 1 ≤ q ≤ ∞. Let (X, B, µ) be a measure
space. Then for all f ∈ L0 (µ), kf kLp,q (µ) ∼ kf ∗∗ kΦp−1 ,q (0,∞) , so that Lp,q (µ)
is metrizable.
Proof The norm equivalence is a consequence of Hardy’s inequality. Since
we can check (f + g)∗∗ ≤ f ∗∗ + g ∗∗ for any simple functions f and g, we can
pass to a limit to have (f + g)∗∗ ≤ f ∗∗ + g ∗∗ for any f, g ∈ L0 (µ). Thus,
f 7→ kf ∗∗ kΦp−1 ,q (0,∞) satisfies the triangle inequality.
We end some applications of the idea used in Section 1.4.3.
Example 40. Let 1 < p < ∞.
Z ∞
f (s)
(1) Let K0 f (t) ≡ ds for f ∈ M+ (0, ∞). Then since for any t > 0,
Z ∞ 0 s + t Z ∞
f (ts) ds
K0 f (t) ≡ ds, kK0 f kLp (0,∞) ≤ kf kLp (0,∞) 1 . So
0 s+1 0 (s + 1)s p
K0 can be extended to a linear operator bounded on Lp (0, ∞).
(2) For x = (x0 , xn ) ∈ Rn+ , we write x̃ ≡ (x 0
Z , −xn ). We define the reflected
f (y)
singular integral operator by Kf (x) ≡ dy for f ∈ M+ (Rn+ ).
R+ n |x̃ − y|n
If we use the Fubini theorem and the boundedness of K0 above, we see
that K is bounded on Lp (Rn+ ).

1.4.4 Inequalities for monotone functions and their

applications to Lorentz norms
We present an idea of discretizing the integral here to prove Chiarenza’s
lemma.
Lemma 40. (Chiarenza’s lemma) Let q > 0, ρ > 0 and 0 < θ < 1, and let
µ ∈ C(0, ∞) be such that lim µ(r) = 0 and that µ is strictly increasing. Denote
r↓0
by ν = µ−1 its inverse. Then the following statements are equivalent:
∞
X
θj log ν µ(ρ)θqj



(1) The sum K ≡ converges.
j=0
 Banach function lattices 47
1
Z ρ
µ(t) q
(2) M ≡ dt < ∞.
0 t
Proof We start with a setup: Write aj ≡ θj log ν(µ(ρ)θqj ) for each j ∈ N0 ,
so that θaj ≤ aj+1 .
Suppose that (1) holds. After a change of variables, we decompose
Z ρ 1 Z µ(ρ) 1 ∞ Z µ(ρ)θ qj 1
µ(t) q sq X sq
dt = ds = ds.
0 t 0 ν(s)µ0 (ν(s)) j=0 µ(ρ)θ
0
q(j+1) ν(s)µ (ν(s))

We estimate
Z ρ 1 ∞ Z µ(ρ)θ qj
µ(t) q X 1
dt ≤ µ(ρ)θj ds. (1.2)
0 t j=0 µ(ρ)θ q(j+1) ν(s)µ0 (ν(s))

By the fundamental theorem of calculus, we have

Z ρ 1 ∞ Z µ(ρ)θqj
µ(t) q X
j 1
dt ≤ µ(ρ)θ 0 (ν(s))
ds
0 t j=0 µ(ρ)θ q(j+1) ν(s)µ
∞  
X 1
= µ(ρ) q θj log(ν(µ(ρ)θqj )) − log(ν(µ(ρ)θq(j+1) ))
j=0
1 ∞ 1
µ(ρ) q X 1 1 K µ(ρ) q
= (θaj − aj+1 ) = µ(ρ) K − µ(ρ)
q q a0 , (1.3)
θ j=0 θ θ

implying that (2) holds.

We now assume that (2) holds. If we switch inequality (1.2) by a reasonable
manner, we have
∞ Z ρ 1
1 X µ(t) q
µ(ρ) q (θaj − aj+1 ) ≤ dt. (1.4)
j=0 0 t

Then, from (1.3) and (1.4), it follows that

∞
X
(θaj − aj+1 ) ≤ M. (1.5)
j=0

Since θ ∈ (0, 1), aj < 0 for large j. Thus, we have either K = −∞ or K is

∞
P
finite. If K were not finite, then we would have (θaj − aj+1 ) = ∞. This is
j=0
a contradiction to (1.5).
We present an example of discretization.
Example 41. Let {ak }∞
k=1 be a non-negative decreasing sequence. If we
write A1 ≡ B(1) and Ak ≡ B(2k ) \ B(2k−1 ) for all k ∈ N ∩ [2, ∞), then
∞ X∞  r  r1
X n
ak χAk ∼ 2 p k ak , since f ∗ (2kn |A1 |) = ak .
k=1 Lp,r k=1
48 Morrey Spaces

1.4.5 Exercises
Exercise 26. Let 0 < p < ∞ and 0 < q ≤ ∞.
(1) Argue similarly to Example 28(3) to show that χ∗B(R) = χ[0,|B(R)|) .

(2) Show that χB(R) ∈ Lp,q (Rn ).

(3) Show that L∞ n
c (R ) ⊂ L
p,q
(Rn ).
Exercise 27. Let 0 < p < ∞.
(1) Use the Lebesgue convergence theorem to show that lim t|{x ∈ Rn :
t↓0
1 1
n
|f (x)| > t}| p = 0 and that lim t|{x ∈ R : |f (x)| > t}| p = 0 for all
t↓∞
f ∈ Lp (Rn ).
1
(2) Show by an example that lim t|{x ∈ Rn : |f (x)| > t}| p = 0 and
t↓∞
1
lim t|{x ∈ Rn : |f (x)| > t}| p = 0 fails for all f ∈ WLp (Rn ). Hint:
t↓∞
n
Consider f (x) = |x|− p , x ∈ Rn .
n
Exercise 28. Let 0 < α < n, and let 1 < q < α. By using (1.11), prove
n n
−α n −α n
|·| ∈ Mq (R ). Also disprove | · |
α
∈ L (R ) by a direct calculation.
α

Exercise 29. Let p and q be as in Definition 20. Find the necessary and
sufficient condition on the parameter γ ∈ R ensuring that fγ ∈ Lp,q (Rn ) and
gγ ∈ Lp,q (Rn ).
Exercise 30. By using the set E defined by (6.25), prove that L4,∞ (R) is
strictly wider than M42 (R). Here, the norm of L4,∞ (R) is given by
q
kf kL4,∞ ≡ sup λ 4 λf (λ)
λ>0

for f ∈ L0 (R).
Exercise 31.
(1) For g ∈ Φµ,∞ (0, ∞) and µ > −1, show that kHgkΦµ,∞ (0,∞) ≤
Z t
kgkΦµ,∞ (0,∞) 1
using sµ ds = .
µ+1 0 µ+1
Z ∞
1
(2) For g ∈ Φ−µ,∞ (0, ∞) and µ > 1, using s−µ ds = , show that
t µ−1
kgkΦ−µ,∞ (0,∞)
kHgkΦ−µ,∞ (0,∞) ≤ .
µ−1
kgkΦµ,1 (0,∞)
(3) For g ∈ Φµ,1 (0, ∞) and µ > −1, show that kHgkΦµ,1 (0,∞) ≤
µ+1
using Fubini’s theorem.
 Banach function lattices 49
kgkΦ−µ,1 (0,∞)
(4) Show that kHgkΦ−µ,1 (0,∞) ≤ for g ∈ Φ−µ,∞ (0, ∞) and
µ−1
µ > 1 using Fubini’s theorem.

Exercise 32. Let 1 < p < ∞. Work in R2 = Rt × Rs .

1
(1) Show that L1s Lpt (R2 ) = L(p,1) (R2t,s ) ,→ Lpt L1s (R2 ) = L(1,p) (R2s,t ). Show
also that F = f ⊗ g for some functions f, g ∈ M+ (0, ∞) if F ∈ L0 (R2 )
0 < kF kLpt L1s = kF kL1s Lpt < ∞.
(2) Establish that kHf kLp (0,∞) ≤ p0 kf kLp (0,∞) for all f ∈ Lp (0, ∞).

1 kHf kLp (0,∞)

(3) Let fε (t) = t− p +ε χ(0,1) (t) for t > 0. Show that lim = p0 .
ε↓0 kf kLp (0,∞)

(4) Show that there is no measurable function f on (0, ∞) such that 0 <
kHf kLp (0,∞) = p0 kf kLp (0,∞) < ∞.
Exercise 33. [156, Exercise 1.4.13] Let 0 < p < ∞, and let f ∈ L0 (Rn ). Use
Lemma 29 to show that f ∈ Lp,1 (Rn ) if and only if there exist a decreas-
ing sequence {aj }∞ ∞ ∞ n
j=−∞ ⊂ [0, ∞), {ϕj }j=−∞ ⊂ L (R ) and a sequence
{Ej }∞ j
j=−∞ of measurable sets safisfying |ϕj | ≤ aj χEj and |Ej | ≤ 2 for each
∞
P ∞
P j
j ∈ Z such that f = ϕj and 2 p aj < ∞.
j=−∞ j=−∞

1.5 Young functions and Orlicz spaces

One of the fundamental techniques in the theory of function spaces is
considering function spaces slightly different from Lebesgue spaces. This tech-
nique is used especially to compensate for the failure of the boundedness of
operators. More precisely we consider Young functions and Orlicz spaces which
Young functions generate. Actually in this book we need to use Orlicz spaces
for the purpose above. The Young functions will generalize the parameter p
in the Lebesgue space Lp .
We investigate Young functions in Section 1.5.1. Orlicz spaces will be dealt
with in Section 1.5.2. As an application of Orlicz functions, we generalize the
average of functions Section 1.5.3. The results in Section 1.5 will be very
subtle. However, the Orlicz spaces are nice tools to describe the boundedness
properties of operators even when we consider Morrey spaces. Later we will
combine Morrey spaces and Orlicz spaces to generate Orlicz–Morrey spaces.
Orlicz–Morrey spaces are effectively used for the purpose above.
50 Morrey Spaces

1.5.1 Young functions

Functions defined over [0, ∞) or (0, ∞) will serve various purposes of mea-
suring the size of other functions. For example, many integral operators are
not always Lp (Rn )-bounded. Hence, we are faced with the modification of the
definition of Lp (Rn )-spaces. Young functions, which we deal with in Section
1.5.1, will be used for this purpose. By definition Young functions are convex,
so that they are almost everywhere differentiable. We collect some inequalities
on Young functions and their derivatives. We also define classes ∆2 and ∇2 to
discuss the boundedness properties of operators acting on Orlicz spaces which
Young functions generate.
Definition 22 (Young functions). A function Φ : [0, ∞) → [0, ∞) is a Young
function, if it satisfies the following conditions:
(1) Φ(0) = 0.
(2) Φ is continuous.
(3) Φ is convex. That is, Φ((1 − θ)t1 + θt2 ) ≤ (1 − θ)Φ(t1 ) + θΦ(t2 ) for all
t1 , t2 ∈ [0, ∞) and 0 < θ < 1.
By convention define Φ(∞) ≡ ∞.
Motivated by the above examples it is convenient to introduce the class
M↑ (0, a). Let M↑ (0, a) denote the space of all non-negative and non-decreasing
functions on (0, a) for a ∈ [0, ∞]. We do not tolerate the function Φ(t) =
∞χ[1,∞) (t), although this is a convex function satisfying Φ(t) = 0.
Example 42. The following functions are all Young functions.
(1) tp , p ≥ 1.
(2) et − 1
(3) t log(t + 1).
There are other examples.
(4) Let β > 0. If Φ satisfies Φ(r) = r for 0 < r ≤ 1 and Φ(r) = r(log r)β for
r  1, then Φ can be arranged to be a Young function.
(5) Let p > 0. Define
(
exp(− exp(r−p )) 0 < r  1,
ep (r) ≡
exp(rp ) r  1.

Then ep can be arranged to be a Young function.

Z t
ds
(6) Define tan−1 t ≡ 2
for t ∈ R. Then Φ(t) = t − tan−1 t, t ≥ 0
0 1+s
can be arranged to be an Orlicz function.
 Banach function lattices 51

(7) The function Φ(t) = t exp(−t−1 ), t > 0 has the limit Φ(+0) = lim Φ(t) =
t↓0
0. Consequently, if we extend Φ continuously to [0, ∞), then Φ is a Young
function since Φ00 (t) = t−4 Φ(t) for all t > 0.
Let us start with the following simple estimates for Young functions.
Proposition 41. For a Young function Φ : [0, ∞) → [0, ∞) and t ≥ 0,
θΦ(t) ≥ Φ(θt) if 0 < θ < 1, or equivalently, θΦ(t) ≤ Φ(θt) if 1 < θ < ∞.
Proof Since Φ is convex and Φ(0) = 0, for 0 < θ < 1,

Φ(θt) = Φ((1 − θ)0 + θt) ≤ (1 − θ)Φ(0) + θΦ(t) = θΦ(t).

As a corollary, we can show that the tangent of the line connecting (0, 0)
and (a, Φ(a)) is increasing in a ≥ 0.
Lemma 42. Let Φ : [0, ∞) → [0, ∞) be a Young function. Then mapping
t ∈ (0, ∞) 7→ t−1 Φ(t) ∈ [0, ∞) is increasing.
Proof This can be derived easily from Proposition 41.
Since Young functions are locally absolutely continuous, they have density:
Any Young function can be expressed for some ϕ ∈ M↑ (0, ∞) as
Z t
Φ(t) = ϕ(s)ds (t ≥ 0).
0

We call this equality the canonical representation of Φ. Next, we will treat the
information on the density function ϕ of a Young function.
Lemma 43. Let Φ : [0, ∞) → [0, ∞) be a Young function that is expressed as
Z t
Φ(t) = ϕ(s)ds (t ≥ 0).
0

Then for t > 0:

Φ(t) Φ(2t)
≤ ϕ(t) ≤ , (1.1)
t t
  Z t
1 t Φ(s)
ϕ ≤ ds. (1.2)
4 2 0 s
Proof All the estimates are easy to derive:

1 t 1 t
Z Z
Φ(t)
= ϕ(s)ds ≤ ϕ(t)ds = ϕ(t)
t t 0 t 0
and
Z 2t Z 2t Z 2t
1 1 1 Φ(2t)
ϕ(t) = ϕ(t)dt ≤ ϕ(s)ds ≤ ϕ(s)ds = ,
t t t t t 0 t
52 Morrey Spaces

which proves (1.1).

To obtain (1.2), we use Fubini’s theorem and log 2 = 0.69 · · · to see
Z t Z t Z s  Z t  
Φ(s) 1 t 1 t
ds = ϕ(u)du ds = ϕ(u) log du ≥ ϕ .
0 s 0 s 0 0 u 4 2

Thus, the proof is therefore complete.

The following definition can be regarded as a passage of the harmonic
conjugate of real numbers to functions.
Definition 23 (Conjugate). For a Young function Φ : [0, ∞) → [0, ∞), its
(Legendre) conjugate function or its Fenchel–Legendre transform is defined by
Z t
Φ∗ (t) ≡ ϕ∗ (s)ds (t ≥ 0), (1.3)
0
where Z ∞
ϕ∗ (v) ≡ χϕ−1 ([0,v]) (t)dt = |ϕ−1 ([0, v])| (1.4)
0
for every v > 0.
We provide a rather general but useful formula on the Fenchel–Legendre
transform.
Example 43. Let 1 < p < ∞, q1 , q2 ∈ R. Also let Φ be a Young function
satisfying Φ(t) = tp (log(e + t))q1 (log(e + t−1 ))q2 for large t and t slightly
0
larger than 0. Then since lim t log t = lim t−1 log t = 0, Φ∗ (t) ∼ tp (log(e +
t↓0 t→∞
q1 q2
t))− p−1 (log(e + t−1 ))− p−1 for t ≥ 0.
Concerning the definition above, we have the following observation:
Lemma 44. Maintain the notation in Definition 23. Let x, v ≥ 0.
(1) If ϕ(x) < v, then x ≤ ϕ∗ (v).
(2) If ϕ(x) > v, then x ≥ ϕ∗ (v).
Proof

(1) Note that [0, x] ⊂ ϕ−1 ([0, v]). Thus,

Z ∞ Z ∞
ϕ∗ (v) = χϕ−1 ([0,v]) (t)dt ≥ χ[0,x] (t)dt = x.
0 0

(2) Note that [0, x] ⊃ ϕ−1 ([0, v]). Thus, we can argue as in (1).
We have the following expression of Φ∗ :
 Banach function lattices 53

Theorem 45. Let Φ be a Young function that is, Φ : [0, ∞) → [0, ∞) is a

homeomorphism, and let Φ∗ be its conjugate. Then

Φ∗ (t) = sup{st − Φ(s) : s ∈ [0, ∞)} (t ≥ 0).

Hence, Theorem 45 asserts that Φ∗ is the Fenchel–Legendre transform of

Φ.
Proof Fix s, t ≥ 0. We are going to show

Φ(t) + Φ∗ (s) ≥ st (1.5)

for all s, t ≥ 0 and that for each s ≥ 0, there exists t ≥ 0 for which equality
in (1.5) holds.
If st = 0, then (1.5) is trivial. Therefore, we may assume that s, t > 0.
Inserting (1.4) into the canonical representation, we have
Z s Z ∞ 
Φ∗ (s) = χϕ−1 ([0,v]) (x)dx dv
Z0Z 0

= χ{(x,v)∈[0,∞)2 : 0≤v≤s, ϕ(x)<v} (x, v)dvdx.

[0,∞)2

Note that “0 ≤ v ≤ s and x ∈ ϕ−1 ([0, v])” is not equivalent to “0 ≤ v ≤ s and

ϕ(x) < v”.
Meanwhile,
Z t ZZ
Φ(t) = ϕ(x)dx = χ{(x,v)∈[0,∞)2 : 0≤x≤t, 0≤v≤ϕ(x)} (x, v)dvdx.
0 [0,∞)2

Observe that

[0, t] × [0, s] ⊂ {(x, v) : 0 ≤ v ≤ s, ϕ(x) ≤ v or 0 ≤ x ≤ t, 0 ≤ v ≤ ϕ(x)}.

(1.6)
Therefore we conclude
Z ∞Z ∞
∗
Φ (s) + Φ(t) ≥ χ[0,t]×[0,s] (x, v)dxdv = st.
0 0

To show that equality holds for some t, we set t ≡ ϕ∗ (s). We obtain a kind
of reverse inclusion in (1.6). We aim to show

{(x, v) ∈ (0, ∞) × (0, ∞) : 0 < v < s, ϕ(x) < v} ⊂ [0, t] × [0, s]

and that

{(x, v) ∈ (0, ∞) × (0, ∞) : 0 < x < t, 0 < v < ϕ(x)} ⊂ [0, t] × [0, s].

Let 0 < v < s satisfy ϕ(x) < v. Then t = ϕ∗ (s) ≥ ϕ∗ (v) ≥ x ≥ 0 thanks to
Lemma 44. Meanwhile, let 0 ≤ x < t and 0 ≤ v < ϕ(x). Since x < t = ϕ∗ (s),
we have ϕ(x) ≤ s thanks to Lemma 44 once again. Therefore v ≤ s.
54 Morrey Spaces

Carrying out the same calculation as before, we conclude that Φ∗ (s) +

Φ(t) = st.
Let Φ : [0, ∞) → [0, ∞) be a Young function. It turns out that the conju-
gate function of a convex bijection Φ : [0, ∞) → [0, ∞) is given by

Φ∗ (t) ≡ sup{st − Φ(s) : s ≥ 0} (t ≥ 0).

The following theorem shows that the operation of conjugation is an involu-

tion: if we take the conjugate twice then we go back to the original function.
Theorem 46. Let Φ : [0, ∞) → [0, ∞) be a Young function. Then Φ∗∗ = Φ.
Proof Fix x ∈ [0, ∞). We claim that ϕ∗∗ = ϕ except at countable points
in [0, ∞). Once this is achieved, since Φ∗∗ and Φ have the same canonical
representation, it follows that Φ∗∗ = Φ.
Let v > ϕ(x). Then x ≤ ϕ∗ (v) thanks to Lemma 44. Consequently, for all
ε > 0, ϕ∗∗ (x − ε) ≤ v again thanks to Lemma 44. Since v > ϕ(x) is arbitrary,
ϕ∗∗ (x − ε) ≤ ϕ(x). Since ε > 0 is also arbitrary, ϕ∗∗ (x) ≤ ϕ(x) for all x at
which ϕ∗∗ is continuous.
Conversely, let v < ϕ(x). Then x ≥ ϕ∗ (v) thanks to Lemma 44. Thus,
for all ε > 0, ϕ∗∗ (x + ε) ≥ v again thanks to Lemma 44. Since v < ϕ(x) is
arbitrary, ϕ∗∗ (x + ε) ≥ ϕ(x). Since ε > 0 is also arbitrary and ϕ∗∗ is right-
continuous, it follows that ϕ∗∗ (x) ≥ ϕ(x).
If we reexamine the proof, we notice that ϕ∗∗ = ϕ on (0, ∞), since both
functions are right-continuous.
Lemma 47. Let Φ : [0, ∞) → [0, ∞) be a Young function, and let Φ∗ be its
conjugate. Then the following:
(1) The inequality
   
Φ(t) 2Φ(t)
Φ∗ ≤ Φ(t) ≤ Φ∗ (1.7)
t t
holds for all t > 0.
(2) For all t, s > 0 such that st ≤ Φ(t), we have

Φ∗ (s) ≤ st. (1.8)

(3) For all t, s > 0 such that st ≥ Φ(t), we have

st ≤ Φ∗ (2s) (1.9)

(4) Let s, t, λ > 0. If Φ(t) = Φ∗ (s) = λ, then

λ ≤ ts ≤ 2λ. (1.10)
 Banach function lattices 55

Proof
(1) Recall that t ∈ (0, ∞) 7→ t−1 Φ(t) ∈ (0, ∞) is increasing in t thanks to
Lemma 42. Thus,
     
∗ Φ(t) Φ(t) Φ(s) Φ(t) Φ(s)
Φ = sup s − ≤ t sup − ≤ Φ(t).
t 0<s<t t s 0<s<t t s
The right inequality is easier to prove. We have only to use Theorem 45:
   
2Φ(t) 2Φ(t) Φ(s)
Φ∗ = sup s − ≥ Φ(t).
t s>0 t s
Thus, (1.7) is proven.
(2) Let st ≤ Φ(t). Since t−1 Φ(t) is increasing in t > 0 again thanks to
Lemma 42, we have
Φ∗ (s)
 
Φ(u)
= sup u −
s u>0 s
 
u Φ(u)
≤ t sup −
u>0 t Φ(t)
 
u Φ(u)
= t sup −
0<u<t t Φ(t)
≤ t.

Thus, the proof of (1.8) is complete.

(3) Suppose instead that st ≥ Φ(t). Then
Φ∗ (2s)
 
Φ(u) Φ(t)
= sup 2u − ≥ 2t − ≥ t.
s u>0 s s
Thus, the proof of (1.9) is complete.
(4) It follows from the Young inequality that st ≤ Φ(s) + Φ(t) = 2λ. Mean-
while, if ϕ(t) ≤ s, then
Z t
λ = Φ(t) = ϕ(u)du ≤ st.
0
∗
If ϕ(t) >Z s, then ϕ (s) ≤ t from Lemma 44. Thus it follows that λ =
s
Φ∗ (s) = ϕ∗ (u)du ≤ st. Therefore (1.10) is established.
0

We will be mainly interested in the boundedness of linear or sublinear

operators. To obtain the boundedness of operators, we need some conditions
on Φ. If Φ is a Young function, it is easy to see that Φ(2t) ≥ 2Φ(t) for all
t ≥ 0 using Lemma 42. Here, we define two important classes keeping this in
mind.
56 Morrey Spaces

Definition 24 (∆2 , ∇2 ).

(1) A function ϕ : (0, ∞) → [0, ∞) or ϕ : [0, ∞) → [0, ∞) is a doubling

function, if there exists a constant C > 0, called a doubling constant,
such that C −1 ϕ(t) ≤ ϕ(s) ≤ Cϕ(t) for all t and s in the domain of ϕ
satisfying s ≤ t ≤ 2s.
(2) Denote by ∆2 the set of all convex bijections Φ : [0, ∞) → [0, ∞) sat-
isfying the doubling condition; Φ(2r) . Φ(r) for r > 0. The implicit
constant is again called a doubling constant. In this case Φ also satisfies
the ∆2 -condition.
(3) The set ∇2 is the set of all convex bijections Φ : [0, ∞) → [0, ∞) such
that there exists C > 1 such that Φ(2t) ≥ 2CΦ(t) for all t ≥ 0. In this
case one says that Φ satisfies the ∇2 condition.
Example 44. We can list the following examples:
(1) Let Φ(t) = tq where 1 < q < ∞. Then, Φ ∈ ∆2 ∩ ∇2 .
(2) Let Φ(t) = t. Then, Φ ∈ ∆2 but Φ ∈
/ ∇2 .
(3) Let Φ(t) = et − t − 1. Then, Φ ∈ ∇2 but Φ ∈
/ ∆2 .
Any function Φ ∈ ∆2 grows like a polynomial as the following proposition
shows:
Lemma 48. Let Φ ∈ ∆2 . There exists q ∈ [2, ∞) such that the following
properties hold:
Φ(s) Φ(t)
(1) For all 0 < t ≤ s < ∞, . q−1 .
sq−1 t
Z ∞
Φ(t) Φ(r)
(2) For all r > 0, q+1
dt . q .
r t r
Proof Let D ≥ 2 be a doubling constant of Φ. Take q ≡ 1 + log2 D ≥ 1.
(1) For all 0 < t ≤ s < ∞, we choose a number k ∈ N0 such that 2−k−1 s ≤
t ≤ 2−k s. By using convexity and the ∆2 condition of Φ, we obtain
Φ(s) Φ(2k+1 t) Φ(t)
q−1
≤ k(q−1) q−1
≤ 2−k(q−1) Dk+1 = q−1 .
s 2 t t
(2) From the above computation, we have
Z ∞ Z ∞
Φ(r) ∞ 1
Z
Φ(t) Φ(t) dt Φ(r)
q+1
dt = q−1 t2
. q−1 2
dt = q .
r t r t r r t r

Any function Φ ∈ ∇2 grows faster than Φ(t) = t, as the following propo-

sition shows:
 Banach function lattices 57

Lemma 49. Let Φ ∈ ∇2 .

Φ(s) Φ(t)
(1) There exists θ ∈ (1, ∞) such that for all 0 < t ≤ s < ∞ . θ .
sθ t
Z r
Φ(t) Φ(r)
(2) There exists θ† ∈ (1, ∞) such that for all r > 0, † +1 dt . .
0 t θ rθ†
Proof Let D > 1 satisfy Φ(2t) ≥ 2DΦ(t). Take
(
1
+ log D (D ≥ 2),
τ≡ 2 1 2
1 + 2 log2 D (1 < D < 2).

Define an auxiliary parameter q̃ by

(
1
2 (D ≥ 2),
q̃ ≡ 1
1− 2 log2 D (1 < D < 2).

(1) We claim that θ ≡ τ + 1 − q̃ does the job. For all 0 < t ≤ s < ∞, we
choose a number k ∈ N0 such that 2−k−1 s ≤ t ≤ 2−k s. By using the ∇2
condition, we obtain
 s 1+log2 D Φ(t)
Φ(s) ≥ Φ 2k t ≥ (2D)k Φ(t) = 2(1+log2 D)k Φ(t) ≥


,
t 21+log2 D
or equivalently,
Φ(t) Φ(s)
θ
≤ 2θ θ .
t s

(2) We claim that θ† ≡ τ does the job. Since q̃ ∈ (0, 1), we have
Z r Z r
Φ(r) r 1
Z
Φ(t) Φ(t) dt Φ(r)
τ +1
dt = θ q̃
. θ q̃
dt ∼ τ
0 t 0 t t r 0 t r
thanks to (1). Thus, the proof is complete.

1.5.2 Orlicz spaces

We define Orlicz spaces here and then investigate some properties similar
to Lebesgue spaces; triangle inequality Hölder’s inequality, duality and so on.
For the time being we will work on a σ-finite measure space (X, B, µ). For
f ∈ L0 (µ), we rewrite the norm kf kLp (µ) as follows:
 Z  p  
|f (x)|
kf kLp (µ) ≡ inf λ ∈ (0, ∞) : dµ(x) ≤ 1 ∪ {∞} .
X λ

After writing the norm in this way, we generalize the Lp (µ)-norm. We define
the Orlicz space LΦ (µ) as follows:
58 Morrey Spaces

Definition 25 (Luxemburg–Nakano norm). Let Φ : [0, ∞) → [0, ∞) be a

Young function. Then define the Luxemburg–Nakano norm k · kLΦ (µ) by
    
|f (x)|
Z
kf kLΦ (µ) ≡ inf λ ∈ (0, ∞) : Φ dµ(x) ≤ 1 ∪ {∞}
X λ

for f ∈ L0 (µ). The Orlicz space LΦ (µ) over X is the set of all f ∈ L0 (µ) for
which kf kLΦ (µ) is finite.

Here and below for the sake of simplicity, we assume that Φ is nice.
Example 45.
(1) If Φ(r) = rp , 1 ≤ p < ∞, then LΦ (µ) = Lp (µ) with coincidence of
norms.
(2) Let p > 1, q > 0 be parameters. The most important example of Orlicz
spaces is generated by a function satisfying Φ(t) ∼ tp [log(3 + t)]q for all
t ≥ 0. Denote by Lp logq L(µ) such a function space. The space L log L(µ)
is occasionally called the Hardy–Littlewood space. If p = 1, then write
L logq L(µ) instead of Lp logq L(µ).
(3) If Φ(t) = et − 1 for t ≥ 0, then we write exp(µ) instead of LΦ (µ).
(4) Let E be a measurable set, and let Φ : [0, ∞) → [0, ∞) be a Young
1
function. Then kχE kLΦ (µ) = −1 . In fact, assuming that
Φ (µ(E)−1 )  
1
µ(E) < ∞, we see that λ ≡ kχE kLΦ (µ) solves µ(E)Φ = 1.
λ
As the following theorem shows, k · kLΦ (µ) is a complete norm.

Theorem 50. Let Φ : [0, ∞) → [0, ∞) be a Young function. Then LΦ (µ) is a

Banach space.
Proof We content ourselves with proving that k · kLΦ (µ) is a norm. Let
f, g ∈ LΦ (µ). We are now going to show that

kf + gkLΦ (µ) ≤ kf kLΦ (µ) + kgkLΦ (µ) . (1.11)

Let ε > 0 be taken arbitrarily. Then (1.11) can be reduced to showing

kf + gkLΦ (µ) ≤ kf kLΦ (µ) + kgkLΦ (µ) + ε. (1.12)

From the definition of the set defining the norm kf kLΦ (µ) , we deduce
   
|f (x)|
Z
1
kf kLΦ (µ) + ε ∈ λ>0 : Φ dµ(x) ≤ 1 .
2 X λ
 Banach function lattices 59

Namely, we have
 
|f (x)|
Z
Φ dµ(x) ≤ 1.
X kf kLΦ (µ) + 2−1 ε

The same can be said for g;

 
|g(x)|
Z
Φ dµ(x) ≤ 1.
X kgkLΦ (µ) + 2−1 ε

1 1 β
Set α ≡ kf kLΦ (µ) + ε, β ≡ kgkLΦ (µ) + ε and θ ≡ .
2 2 α+β
Since Φ is convex, we have a pointwise estimate: for all x ∈ X,
   
|f (x) + g(x)| |f (x)| |g(x)|
Φ = Φ (1 − θ) +θ
α+β α β
   
|f (x)| |g(x)|
≤ (1 − θ) Φ +θΦ . (1.13)
α β

Integrating (1.13) over X, we obtain

   
|f (x) + g(x)| |f (x) + g(x)|
Z Z
Φ dµ(x) = Φ dµ(x) ≤ 1,
X kf kLΦ (µ) + kgkLΦ (µ) + ε X α+β

which is equivalent to (1.12).

As is the case with the people who learned Riemann integration theory,
occasionally we are not interested in the exact value of integrals. Indeed, in
many cases, it is again next to impossible to calculate precisely the value of
an integral. Instead, we are mainly interested in the size of functions. Here,
we remark that the Orlicz norm has the following equivalent expression:
Theorem 51. Let f ∈ L0 (µ), and let Φ : [0, ∞) → [0, ∞) be a Young function.
Then
   
|f (x)|
Z
kf kLΦ ≤ inf λ + λ Φ dx : λ > 0 ≤ 2kf kLΦ .
X λ

Proof By the Fatou lemma, we may assume that f ∈ L∞ (µ) and that
µ{f 6= 0} < ∞. We may assume that kf kLΦ = 1 by normalization.  By  con-
|f (x)|
Z
sidering two cases; λ ≤ 1 and λ ≥ 1, we obtain 1 ≤ λ + λ Φ dx
X λ
for all λ > 0. The opposite inequality can be obtained by taking λ ≡ 1.
We next prove the dual inequality for Orlicz spaces.
Theorem 52. Let Φ : [0, ∞) → [0, ∞) be a Young function, and let Φ∗
∗
be its conjugate. Then for all f ∈ LΦ (µ) and g ∈ LΦ (µ), kf · gkL1 (µ) ≤
2kf kLΦ (µ) kgkLΦ∗ (µ) .
60 Morrey Spaces

Proof We may assume that kf kLΦ (µ) = kgkLΦ∗ (µ) = 1. In this case, by
integrating |f (x)g(x)| ≤ Φ(|f (x)|) + Φ∗ (|g(x)|) over X we obtain the desired
result.
We end Section 1.5.2 with a duality result on Orlicz spaces.
Theorem 53. Let Φ ∈ ∆2 . Then the dual space of LΦ (µ) is isomorphic to
∗
LΦ (µ) in the following sense:
∗ ∗
(1) For all f ∈ LΦ (µ) and g ∈ LΦ (µ), f ·g ∈ L1 (µ), Φ
R so that any g ∈ L (µ)
Φ
defines a continuous functional h ∈ L (µ) 7→ h(x)g(x)dµ(x) ∈ C.
X
Φ
(2) Conversely any continuous functional on L (µ) is realized in this way.
Proof From Theorem 52, (1) follows. We prove (2). Let L : LΦ (µ) → C
be continuous. Since X is σ-finite, we may assume µ(X) < ∞ by a routine
truncation procedure.
Let f ∈ LΦ (µ) with kf kLΦ (µ) = 1. We remark that g ≡ χ{f 6=0} |f |−1 Φ(|f |)
satisfies
Z Z
kg · f kL1 (µ) = 1, Φ∗ (g(x))dµ(x) ≤ Φ(|f (x)|)dµ(x) = 1
X X

thanks to Lemma 47(1). Thus, for any f ∈ LΦ (µ) with kf kLΦ (µ) = 1, there
exists a simple function g satisfying kgkLΦ∗ (µ) ≤ 1 such that kf · gkL1 (µ) = 1.
Since Φ ∈ ∆2 , we can find a number p > 0 such that Φ(t) . tp for t ≥ 1.
Thus, since µ(X) < ∞, the mapping f ∈ Lp (µ) 7→ L(f ) ∈ C is a continuous
0
linear functional, so that there exists g ∈ Lp (µ) such that
Z
L(f ) = f (x)g(x)dµ(x)
X
p
for all f ∈ L (µ). Consequently, by a simple truncation procedure once again,
we have
kf · gkL1 (µ) ≤ kLkLΦ (µ)→C kf kLΦ (µ)
∗
for all f ∈ LΦ (µ). Once we show that g ∈ LΦ (µ) with the estimate
kgkLΦ∗ (µ) . kLkLΦ (µ)→C , we see that g realizes L. Again by a simple trunca-
∗
tion procedure, we may assume that g is a simple function, so that g ∈ LΦ (µ).
This means that we can concentrate on the norm estimate. It remains to test
this estimate on f = χ{g6=0} |g|−1 Φ∗ (|g|).

1.5.3 Orlicz-averages
(p)
On many occasions it is important to consider the powered average mQ (f )
of functions f over a cube Q for 0 < p < ∞. We will apply the idea of
generalizing p to a Young function Φ : [0, ∞) → [0, ∞) to generalize the
above notion of the powered average. Here, we are concerned with the case
 Banach function lattices 61

where (X, F, µ) is a probabilty space. More precisely, let Q be a cube with

the uniform probability measure |Q|−1 dx.
Definition 26 (Orlicz-average). Let Φ : [0, ∞) → (0, ∞) a Young function.
For a cube Q, define its Φ-average, its Orlicz-average or its the mean Luxem-
berg norm over Q by
   
|f (x)|
Z
1
kf kΦ;Q ≡ inf λ > 0 : Φ dx ≤ 1 . (1.14)
|Q| Q λ

We do not have to stick to cubes. Instead we can use balls.

For applications, we envisage the following Young functions:
Example 46. Let Q ∈ Q and f ∈ L0 (Q).
(1) If Φ(t) = t, then kf kΦ;Q = mQ (|f |).
(p)
(2) If Φ(t) = tp , then kf kΦ;Q = mQ (|f |).

Example 47. Let Q ∈ Q and f ∈ L0 (Q).

(1) If Φ(t) = L log L(t) ≡ t log(e + t), then kf kL log L;Q ≡ kf kΦ;Q .
(2) If Φ(t) = L log Lα (t) ≡ t(log(e + t))α , then kf kL log Lα ;Q ≡ kf kΦ;Q .
The following is an example of calculating the Φ-average of the indicator
functions. We say that a Young function Φ is normalized if Φ(1) = 1.
Example 48. Let Φ : [0, ∞) → [0, ∞) be a bijective Young function. Then
for all cubes Q and s ≥ 1, kχQ kΦ;sQ = Φ−11(sn ) . Thus, for a normalized Young
function Φ, kχQ kΦ;Q = 1.
The dilation relation is summarized in the following lemma:
Lemma 54. Let Q ∈ Q, and let κ > 0. Define R ≡ {x ∈ Rn : κx ∈ Q}.
Then kf kΦ;Q = kf (κ·)kΦ;R for all f ∈ L0 (Q).
Proof Indeed, by a change of variables,
   
|f (x)|
Z
1
kf kΦ;Q = inf λ > 0 : Φ dx ≤ 1
|Q| Q λ
   
|f (κx)|
Z
1
= inf λ > 0 : Φ dx ≤ 1
|R| R λ
= kf (κ·)kΦ;R .
(p)
We can generalize the inequality mQ (|f |) ≤ mQ (|f |) for 1 ≤ p < ∞,
Q ∈ Q and f ∈ L0 (Q).
Lemma 55. For a cube Q and f ∈ L0 (Q), mQ (|f |) . kf kΦ;Q .
62 Morrey Spaces

Proof We may suppose that kf kΦ;Q = 1. Then

mQ (|f |)
Z Z
1 1
≤ χ[0,1] (|f (x)|)|f (x)|dx + χ(1,∞] (|f (x)|)Φ(|f (x)|)dx
|Q| Q Φ(1)|Q| Q
1
≤1+ .
Φ(1)

In Lemma 55, we used the fact that Φ(1)t ≤ Φ(t) for t ≥ 1. Note that
Φ(1)t ≥ Φ(t) for 0 ≤ t ≤ 1. As is guessed from this fact, the value Φ(t) with
0 ≤ t ≤ 1 does not affect very much. This fact can be summarized in the
following proposition:
Proposition 56. Let Φ1 , Φ2 : [0, ∞) → [0, ∞) be Young functions. Assume
that Φ1 (t) ≤ Φ2 (t) for all t > t0 , where t0 > 0 is a fixed number. Then for all
cubes Q and all f ∈ L0 (Q), kf kΦ1 ;Q . kf kΦ2 ;Q .
Proof Assume that kf kΦ2 ;Q = 1. Then
Z Z
1 1
Φ1 (|f (x)|χ(t0 ,∞] (|f (x)|))dx = Φ2 (|f (x)|χ(t0 ,∞] (|f (x)|))dx
|Q| Q |Q| Q
≤ 1,

which implies kf χ(t0 ,∞] (|f |)kΦ1 ;Q ≤ 1. Meanwhile

|f (x)|χ[0,t0 ] (|f (x)|))

Z    
1 t0
Φ1 dx ≤ Φ1 ≤ 1.
|Q| Q Φ1 (t0 ) Φ1 (t0 )

Thus, kf kΦ1 ;Q . kf kΦ2 ;Q .

Here, we list a couple of properties of Φ-averages. The first one extends
(p)
the fact that |Q| · mQ (f ) is increasing in Q for all p ≥ 1 and f ∈ L0 (Q).

Lemma 57. Let f ∈ L0 (Rn ). Then for all cubes Q1 ⊂ Q2 ,

|Q1 | · kf kΦ;Q1 . |Q2 | · kf kΦ;Q2 .

Proof We may assume that the right-hand side is finite or more strongly
equals 1. Then Z
1
Φ(|Q2 | · |f (x)|)dx ≤ 1.
|Q2 | Q2
Since Φ is a Young function and Q1 ⊂ Q2 , we have
1 1
Φ(|Q1 | · |f (x)|) ≤ Φ(|Q2 | · |f (x)|) (x ∈ Rn ).
|Q1 | |Q2 |
It remains to integrate this pointwise estimate over Q2 using Lemma 42 and
discard the contribution of Q2 \ Q1 in the left-hand side.
 Banach function lattices 63

Let Q ∈ Q, and let U be any partition of Q into cubes. Also let f ∈ L1 (Q).
Then by the pigeon-hole principle, we have mR (|f |) ≥ mQ (|f |) for some R ∈
U. We will generalize this fact to the Φ-average as follows:
Lemma 58. Let f ∈ L0 (Rn ). Then kf kΦ;Q ≤ 2n sup kf kΦ;Q0 for
Q0 ∈Q(Q) : `(Q0 )=t0
any cube Q ∈ Q and any positive number t0 ≤ `(Q).
Proof By the monotone convergence theorem, we may assume f ∈
L∞ (Rn ). By the normalization, we may also assume kf kΦ;Q = 1. Let t0 ∈
(0, `(Q)] be fixed. Let
 
`(Q)
N = 1+ 0 ∈ [1, ∞).
t
n
Y
We write Q = [aj , aj +`(Q)], so that (a1 , a2 , . . . , an ) is the “bottom” corner
j=1
of Q. Define (
0 (mj < N ),
δ(mj ) ≡ 0
N t − `(Q) (mj = N )
for mj ∈ {1, 2, . . . , N } and
n
Y
Qm ≡ [aj + (mj − 1)t0 − δ(mj ), aj + mj t0 − δ(mj )]
j=1

for m = (m1 , m2 , . . . , mn ) ∈ {1, 2, . . . , N }n . Then

X
1≤ χQm ≤ 2n
m∈{1,2,...,N }n

almost everywhere. Consequently,

Z
1
1= Φ(|f (x)|)dx
|Q| Q
Z
X 1
≤ Φ(|f (x)|)dx
|Q| Qm
m∈{1,2,...,N }n

t0n 1
X Z
= Φ(|f (x)|)dx.
n
|Q| |Qm | Qm
m∈{1,2,...,N }

Since
t0 N t0 t0 t0
   
`(Q) `(Q)
= 1+ 0 ≤ 1+ 0 ≤ + 1 ≤ 2,
`(Q) `(Q) t `(Q) t `(Q)
it follows that
Z Z
1 n 1
Φ(|f (x)|)dx ≤ 2 max Φ(|f (x)|)dx.
|Q| Q m∈{1,2,...,N }n |Qm | Qm

Thus, the right-hand side equals ∞. If the left-hand side is finite, then we
argue similarly.
64 Morrey Spaces

1.5.4 Lebesgue spaces with a variable exponent

Here, we content ourselves with the definitions. Lebesgue spaces with a
variable exponent can be defined similarly to Orlicz spaces.
Definition 27 (Variable (exponent) Lebesgue space). Let (X, B, µ) be a mea-
sure space, and let p(·) : X → [1, ∞] be a measurable function. Then the
variable exponent Lebesgue space Lp(·) (µ) with a variable exponent is defined
by [
Lp(·) (µ) ≡ {f ∈ L0 (µ) : ρp (λ−1 f ) < ∞},
λ>0

where
ρp (f ) ≡ k |f |p(·) χ(0,∞) (p)kL1 (µ) + kf χ{∞} (p)kL∞ (µ) .
Moreover, for f ∈ Lp(·) (µ) one defines the variable Lebesgue norm by

kf kLp(·) (µ) ≡ inf λ ∈ (0, ∞) : ρp (λ−1 f ) ≤ 1 ∪ {∞} .



1.5.5 Exercises
Exercise 34. Show that any homeomorphism from [0, ∞) to [0, ∞) leaves 0
invariant. Hint: Count the number of connected components of the set [0, ∞)\
{a} for a ∈ [0, ∞).
Exercise 35.
(1) Let p > 1 and t > 0. Calculate sup(st − sp ).
s≥0
s
(2) Calculate sup(st − e + 1) for t > 1.
s≥0

Exercise 36. [192, Theorem 1.3] Let Φ : [0, ∞) → [0, ∞) be a Young function,
and let f ∈ L0 (Rn ).
kf χB(x,r) kLΦ
(1) Show that lim inf ≥ |f (x)| for almost all x ∈ Rn .
r↓0 kχB(x,r) kLΦ
kf χB(x,r) kLΦ
(2) Assume that Φ ∈ ∆2 . Then prove that lim inf = |f (x)| for
r↓0 kχB(x,r) kLΦ
almost all x ∈ Rn .

1.6 Smoothness function spaces

The simplest function spaces appearing in analysis seems BC(X), the set of
all complex-valued continuous functions defined on a metric space (X, d). This
function space appears as an example or an exercise of complete metric spaces.
 Banach function lattices 65

The next example we encounter in the course of analysis is the Lebesgue

space Lp (Rn ). Unlike the space BC(X), it had been very difficult to prove its
completeness. One of the reasons why the theory of function spaces is prevail-
ing is that the triangle inequality is available in addition to the completeness.
Everything is fine as long as we have only to work within the framework of
BC(X) or Lp (Rn ).
However, things are not so nice once we consider many other mathematical
problems. For example, these spaces are not sufficient, when we consider the
Laplace equation −∆u = f . For the purpose of tackling this equation, we
are led to thinking of Sobolev spaces. Indeed, in this case, it is convenient to
introduce
v
u X n
u
kf kW 1,2 = tkf kL2 2 + k∂j f kL2 2 (Sobolev space)
j=1

because we can use the integration by parts.

We define Sobolev spaces in Section 1.6.1 as an application of Lebesgue
spaces. We define Hölder–Zygmund spaces in Section 1.6.2. Remark that we
do not recall the definition of multi-indexes; see [117, Appendix A] or [258,
Chapter 2].

1.6.1 Sobolev spaces

The aim of this section is very modest. We intend to define the Sobolev
space W m,p (Rn ) and investigate some elementary properties. We do not go
into detail, for example, the trace theorem, the extension theorem and so on
are not taken up here. We content ourselves with making a brief view of the
function spaces.
Since L1loc (Rn ) is too large, it may be surprising that we can consider the
derivatives of L1loc (Rn ). It may be indirect to define them in the next definition.
We work in a domain Ω to consider the function spaces defined there. Here by
a domain we mean an open set in Rn . But this indirect definition does work
as we shall see.
Definition 28 (Partial derivative). Let f ∈ L1loc (Ω) and αZ∈ N0 n . A weak par-
tial α-derivative of f , if it exists, is g ∈ L1loc (Ω) such that f (x)∂ α ϕ(x) dx =
Z Ω

(−1)|α| g(x)ϕ(x) dx for all ϕ ∈ Cc∞ (Ω).

Ω

The following lemma is easy to prove.

Lemma 59. Let f ∈ L1loc (Rn ) and α ∈ N0 n .
(1) The weak partial α-derivative of f , if it exists, is unique.
(2) If f ∈ C |α| (Rn ), then ∂ α f coincides with the usual partial α-derivative.
66 Morrey Spaces

Proof The proof is left as an exercise; see Exercise 37.

Example 49. Let ϕ ∈ Cc∞ (Rn ) and define ψ(x) ≡ |x|ϕ(x) for x ∈ Rn . Then
we have
xj
∂j ψ(x) = |x|∂j ϕ(x) + ϕ(x) (1.1)
|x|
in the sense of weak derivatives. To check this, we take η ∈ Cc∞ (Rn ) arbi-
trarily. For the sake of simplicity we can assume j = n. Then if we carry out
integration by parts, we obtain
Z
ψ(x) · ∂n η(x) dx
|xn |>ε
Z
= (ψ(x0 + εen )η(x0 + εen ) − ψ(x0 − εen )η(x0 − εen ) ) dx0
Rn−1
Z
− ∂n ψ(x) · η(x) dx.
|xn |>ε
Z
If we let ε ↓ 0, then the left-hand side tends to ψ(x)∂n η(x) dx and the
Rn
boundary terms of the right-hand side cancel. Therefore
Z Z
ψ(x)∂n η(x) dx = − lim ∂n ψ(x)η(x) dx
Rn ε↓0 |xn |>ε
Z  
xj
= − lim |x|∂j ϕ(x) + ϕ(x) η(x) dx.
ε↓0 Rn |x|
/ C 1 (Rn ) in general!
Thus, we obtain (1.1), although ψ ∈
As we have seen before, Cc∞ (Rn ) is not complete, if we endow it with the
L (Rn )-norm. If p is finite, then Lp (Rn ) is a completion of Cc∞ (Rn ) with the
p

Lp (Rn )-norm. The space L∞ (Rn ) is a natural extension of the Lp (Rn )-norm.
The failure of completeness is the case, if we endow Cc∞ (Rn ) with a norm
given by X
kf kW m,p ≡ k∂ α f kLp .
α∈N0 n
|α|≤m

We can say that W m,p (Rn ), whose precise definition we are about to present,
overcomes this defect, if p < ∞.
Definition 29 (The Sobolev spaces in Lp (Rn ) with m derivatives). Let m ∈
N0 , and let 1 ≤ p ≤ ∞. Then define the Sobolev space W m,p (Rn ) by:
W m,p (Rn )
\
≡ {f ∈ Lp (Rn ) : ∂ α f exists in weak sense and belongs to Lp (Rn ) }.
|α|≤m
X
The norm of f ∈ W m,p (Rn ) is defined by: kf kW m,p ≡ k∂ α f kLp .
n
α∈N0
|α|≤m
 Banach function lattices 67

Let 1 ≤ p ≤ ∞. From the viewpoint of W m,p (Rn ), we see that f ∈ Lp (Rn ),

if and only if ∂ α f ∈ Lp (Rn ) for all α ∈ N0 n with |α| ≤ m.

Theorem 60. Let 1 ≤ p ≤ ∞ and m ∈ N. Then the Sobolev space W m,p (Rn )
is a Banach space.
Proof The proof is left as an exercise; see Exercise 38.
Next, we are going to discuss the mollification and consider the dense
subspace of W m,p (Rn ).
Lemma 61. Let 1 ≤ p < ∞. Let ϕ ∈ Cc∞ (Rn ) be a positive function with
integral 1. Then for given f ∈ W m,p (Rn ) and t > 0, we set ft ≡ ϕt ∗ f , where
1 ·
ϕt = n ϕ . Then we have lim ft = f in the topology of W m,p (Rn ).
t t t↓0

Proof Note that ∂j ft = ϕt ∗∂j f for j = 1, 2, . . . , n because f ∈ W 1,p (Rn ).

Therefore, this lemma is true for |α| = 1. In the general case we have only to
induct on |α|.
Corollary 62. Let 1 ≤ p < ∞. Then Cc∞ (Rn ) is dense in W m,p (Rn ).
Proof By Lemma 61, we have only to approximate smooth elements f ∈
W m,p (Rn ) ∩ C ∞ (Rn ). Choose ∞ n

a truncation function η ∈ Cc (R ) that equals
·
1 on B(1). Set fm ≡ f η m . Then by the Lebesgue convergence theorem fm
tends to f in W m,p (Rn ).
We next consider the relation between the difference quotient and the
weak-partial derivative. Classically the differentiation was defined by the limit,
that is, it was given by the limit of the difference quotient. However, the weak
derivative we take up here seems to have nothing to do with the difference
quotient. In this paragraph, we investigate the connection between the dif-
ference quotient and the weak-partial derivative. The vector ek ≡ (δkj )nj=1
stands for the k-th elementary vector.
Definition 30 (difference quotient). Let h ∈ R\{0} and k = 1, 2, . . . , n. Then
f (· + hek ) − f
for f ∈ L1loc (Rn ), define its difference quotient Dkh f by Dkh f ≡ .
h
Theorem 63. Let 1 < p < ∞. Then f ∈ Lp (Rn ) belongs to W 1,p (Rn ) if and
only if sup sup kDjh f kLp . 1.
j=1,2,...,n h∈R\{0}

Proof The proof uses the weak compactness of Lp (Rn ) and is left as an
exercise; see Exercise 39.
Now we consider the chain rule. What is totally different from the classical
analysis is that if f ∈ W 1,p (Rn ), then we have |f | ∈ W 1,p (Rn ). This property
fails for the space C 1 (Rn ) because the limit of the difference quotient comes
into play. The next theorem is the first step for this purpose.
68 Morrey Spaces

Theorem 64. Let 1 < p < ∞. Assume that η ∈ BC1 (C) with η(0) = 0. Also
let f ∈ W 1,p (Rn ). Then η ◦ f ∈ W 1,p (Rn ).
Here if f (x) = ±∞, then it will be understood that η ◦ f (x) = 0 and below
we shall disregard such a point. We shall not allude to this point later.
Proof The proof is left as an exercise; see Exercise 40.
We next consider the Sobolev embedding theorem. Suppose that f : R → R
is
Z x smooth function with the first derivative integrable over R. Then f (x) =
a
f 0 (t) dt + f (0) and hence f is bounded. In this way if we have some infor-
0
mation on the partial derivatives, then we can say more about the function.
The Sobolev embedding theorem quantifies such a situation.
We write Du ≡ (∂x1 u, ∂x2 u, . . . , ∂xn u) for a function u. For a Banach
lattice X, we define kDukX ≡ k |Du| kX .
Theorem 65 (Gagliardo-Nirenberg-Sobolev inequality). Assume 1 ≤ p < n.
1 1 1
Define q by = − . Then kukLq .p,n kDukLp for all u ∈ Cc1 (Rn ).
q p n
Proof The proof is left as an exercise; see Exercise 41.
Function spaces are tools with which to describe the size and the smooth-
ness of functions. Let p, q, m ∈ [1, ∞] and suppose that m is an integer.
For example, Lq (Rn ) measures the size. Since the pointwise multiplication
of L∞ (Rn )-functions is closed in Lq (Rn ), Lq (Rn ) never measures the smooth-
ness of functions. However, W m,p (Rn ) can control somehow the smoothness
of functions. What counts about size and smoothness is that the control of
smoothness can be transformed into that of size. This aspect of these two
quantities can be illustrated by the next theorem.
n n n
Theorem 66. Let m ∈ N and 1 ≤ p < . Then define q by m − = − .
m p q
Then W m,p (Rn ) ,→ Lq (Rn ).
Proof The proof is left as an exercise; see Exercise 41.
n
The quantity m − in the space W m,p (Rn ) has a special name.
p
Definition 31 (Differential/Sobolev index). In W m,p (Rn ), the quantity m −
n
is referred to as the differential/Sobolev index of functions.
p
We move on to the extension property. The extension problem is a classical
problem in the theory of function spaces with important applications in many
fields of mathematical analysis, in particular harmonic analysis and the theory
of partial differential equations. Broadly speaking, the problem consists in
extending the elements of a space of functions defined on a given subset of
Rn to the whole of Rn , preserving certain differentiability and summability
properties.
 Banach function lattices 69

We consider elementary Lipschitz domains Ω in Rn of the form

Ω = Ωϕ ≡ {x = (x0 , xn ) ∈ Rn : xn < ϕ(x0 ), x0 ∈ Rn−1 } , (1.2)

where ϕ : Rn−1 → R is a Lipschitz continuous function.

We consider function spaces on Lipschitz domains above as a model case of
bounded domains having a Lipschitz boundary. Since ϕ is a Lipschitz function,
x ∈ Ωϕ and yn > L|y 0 | implies x + y ∈ Ωϕ . The following lemma takes full
advantage of the property of Lipschitz domains.
Lemma 67. Let 1 ≤ p < ∞. Let Ω be an elementary domain as above. Then
W 2,p (Ω) ∩ C ∞ (Ω) is dense in W 2,p (Ω).
Proof We start with the setup: Let ψ ∈ Cc∞ (Rn ) ∩ M+ (Rn ) be such that
it is supported in a cone {yn > L|y 0 |}. Define ψt (x) ≡ t−n ψ(t−1 x) for x ∈ Rn
and t > 0.
For f ∈ W 2,p (Ω), we define ft ≡ f ∗ψt . Then Dk ft = ψt ∗Dk f for k = 0, 1, 2
and ft ∈ C ∞ (Ω). Thus, we see that ft → f in W 2,p (Rn ) as t ↓ 0.
Let G ≡ Rn \Ω. For every k ∈ Z, set Gk ≡ {x ∈ G : 2−k−1 < ρn (x) ≤ 2−k },
where x ∈ Rn 7→ ρn (x) ≡ xn − ϕ(x0 ) is the signed distance from x ∈ Rn to
∂G in the xn -direction. Let G̃k ≡ Gk−1 ∪ Gk ∪ Gk+1 . Clearly, ρn (x) ≥ 0 for
all x ∈ G. Write M ≡ kϕkLip and A ≡ 16(M + 1).
In the sequel, we need the partition of unity subordinate to {Gk }∞
k=−∞ .

Lemma 68. There exists a collection {ψk }∞ ∞ n + n

k=−∞ ⊂ C (R ) ∩ M (R ) sat-
isfying the following conditions:
∞
P ∞
P
(1) χG = ψk ≤ χsupp(ψk ) ≤ 2χG ;
k=−∞ k=−∞

(2) |Dα ψk | .α 2k|α| χΩ̃k for all k ∈ Z, α ∈ Nn0 .

Proof Let τ ∈ Cc∞ (B(r)) ∩ M+ (Rn ) for some small r > 0 and kτ kL1 = 1.
Set ψ̃k ≡ 2kn τ (2k ·) ∗ χρ−1
n (2
n
−k−1 ,2−k ) for each k ∈ Z. Since ϕ ∈ Lip(R ), if we

choose r  1, then we have supp(ψ̃k ) ⊂ ρ−1

n (0.9 · 2
−k−1
, 1.1 · 2−k ).
α k|α| n
We note |D ψ̃k | .α 2 χΩ̃k for all k ∈ Z, α ∈ N0 . Thus, if we set ψ̃ ≡
∞
P ψ̃k
ψ̃l and ψk ≡ for each k ∈ Z, then we obtain the desired function.
l=−∞ ψ̃
Define
∞
X
d(x) ≡ 2k ψk (x) (x ∈ Rn ).
k=−∞

We are now ready to define Burenkov’s extension operator for an elementary

Lipschitz domain Ω as in (1.2). Let l ∈ N0 and 1 ≤ p ≤ ∞. Write M ≡ kϕkLip
and A ≡ 16(M + 1). The following lemma explains the role of A:
70 Morrey Spaces

Lemma 69. Let a > 0 be fixed. Then (x0 − 2−k z 0 , xn − A2−k zn ) ∈ Ω for all
k ∈ Z, x ∈ G and z ∈ (−a, a)n−1 × (2a, 4a).
Proof This is clear from the definition of the Lipschitz norm.
We are now ready for the construction of Burenkov’s extension operator.
Choose ω ∈ Cc∞ ((−a, a)n−1 ×(2a, 4a))∩M+ (Rn ) so that kωkL1 = 1. For every
f ∈ L1loc (Ω) and x ∈ G, we set
Z
fk (x) ≡ f (x0 − 2−k z 0 , xn − A2−k zn )ω(z)dz
(−a,a)n−1 ×(2a,4a)
Z
= A−1 2kn ω(2k (x0 − y 0 ), A−1 2k (xn − yn ))f (y)dy
Ω

and 

f (x), if x ∈ Ω,

∞
T f (x) ≡ X (1.3)


 ψk (x)fk (x), if x ∈ G ≡ Rn \ Ω.
k=−∞
Note that ∂Ω has Lebesgue measure zero, so that T f is defined almost every-
where on Rn . We note that the operator T in (1.3) is defined by means of a
sequence of mollifiers with discrete steps and has a local nature in the sense
that the values of the extended function T f around a point in Rn \ Ω depend
only on the values of f localized around certain “reflected” points inside Ω.
Theorem 70. Let l ∈ N. Let f ∈ C l (Ω) satisfy ∂ α f ∈ L∞ (Ω) for any α ∈ N0 n
with |α| ≤ l, so that its derivative up to any order extends continuously to the
boundary. Then T f ∈ C l (Rn ).
Proof Since T f ∈ C ∞ (Rn \ ∂Ω), it suffices to prove lim ∂ α T f (x) =
x∈G,x→x0
∂ α f (x0 ) for all x0 ∈ ∂Ω and |α| ≤ l. We let x = (x0 , xn ) ∈ G. We set
Z(x)
≡ (∂ α f (x0 − 2−k z 0 , xn − A2−k zn ) − ∂ α f (x0 , ϕ(x0 )))ω(z),
Z 0 (x)
Z
≡ (∂ β f (x0 − 2−k z 0 , xn − A2−k zn ) − ∂ β f (x0 , ϕ(x0 )))ω(z)dz
(−a,a)n−1 ×(2a,4a)

ζ(s) = ζ(s; k, x, z, A, β)
≡ ∂ β f (x0 − 2−k sz 0 , s(xn − A2−k zn ) + (1 − s)ϕ(x0 )) − ∂ β f (x0 , ϕ(x0 )).
∞
∂ γ ψk (x) = 0 for any γ ∈ N0 n \ {0}, we have
P
Since
k=−∞

Dα T f (x) − ∂ α f (x0 , ϕ(x0 ))

X∞ X α ∞
X Z
α−β 0
= ∂ ψk (x)Z (x) + ψk (x) Z(z)dz.
β (−a,a)n−1 ×(2a,4a)
k=−∞ 0<β≤α k=−∞
 Banach function lattices 71

Denote by ζl = ζl (·; k, x, z, A, β) the remainder term of the Taylor expansion

up to order l at s = 0. Since ω satisfies ω ∈ Cc∞ ((−a, a)n−1 × (2a, 4a)) ∩
∞
M+ (Rn ) and ∂ β ψk ≡ 0 of any multi-index β ∈ N0 n \ {0}, we obtain
P
k=−∞

Dα T f (x) − ∂ α f (x0 , ϕ(x))

X∞ X α  Z
= ∂ α−β ψk (x) ζl (1; k, x, z, A, β)ω(z)dz
β (−a,a)n−1 ×(2a,4a)
k=−∞ 0<β≤α
∞
X Z
+ ψk (x) Z(z)dz.
k=−∞ (−a,a)n−1 ×(2a,4a)

Consequently, we obtain |Dα T f (x) − ∂ α f (x0 , ϕ(x))| . d(x) for all x ∈ G, as

required.
Lemma 71. Let x ∈ G, and let f ∈ L1loc (Ω). Then
Z
|T f (x)| . |f (x0 − (xn − ϕ(x0 ))z 0 , xn − (xn − ϕ(x0 ))zn )|dz.
(−2a,2a)n−1 ×(aA,8aA)

Proof It suffices to handle each piece fk , k ∈ Z. To handle fk (x), x ∈ G,

we may assume x ∈ Gk , so that 2−k−1 ≤ xn − ϕ(x0 ) ≤ 2−k . Observe that

|fk (x)|
Z
. 2kn χ(−a,a)n−1 ×(2a,4a) (2k (x0 − y 0 ), A−1 2k (xn − yn ))|f (y)|dy
Ω
Z
kn
=2 |f (x − w)|dw
(−2−k a,2−k a)n−1 ×(2−k+1 aA,2−k+2 aA)
Z
≤ 2kn |f (x − w)|dw
(−2axn +2aϕ(x0 ),2axn −2aϕ(x0 )))n−1 ×(A(xn −ϕ(x0 )),8A(xn −ϕ(x0 )))
Z
∼ |f (x0 − (xn − ϕ(x0 ))z 0 , xn − (xn − ϕ(x0 ))zn )|dz.
(−2a,2a)n−1 ×(aA,8aA)

We investigate the transformation in Lemma 71.

Lemma 72. Suppose 0 < a  1. Let z ∈ (−2a, 2a)n−1 × (aA, 8aA). If a is
small enough, then the mapping

Φz : x ∈ Rn 7→ (x0 − (xn − ϕ(x0 ))z 0 , xn − (xn − ϕ(x0 ))zn ) ∈ Rn ,

is bi-Lipschitz with the constants independent of z.

Proof Since the Lipschitz norm of the mapping x ∈ Rn 7→ (−(xn −
ϕ(x0 ))z 0 , −(xn − ϕ(x0 ))zn ) ∈ Rn is small as long as a is small, the conclusion
is clear.
72 Morrey Spaces

Note that Φz is the composition Φ2z ◦ Φ1z , where

Φ1z (x) ≡ (x0 , xn − ϕ(x0 ))

Φ2z (x) ≡ (x0 − xn z 0 , xn + ϕ(x0 ) − xn zn ).

As a consequence, the mapping Φz is a homeomorphism (see Exercise 44).

We move on to the interpolation property. We have the interpolation for-
mula:
Theorem 73. Let ϕ be a Lipschitz function. Define Ωϕ as above. Then for all
f ∈ W 2,p (Ωϕ ), kDf kLp (Ωϕ ) . kD2 f kLp (Ωϕ ) + kf kLp (Ωϕ ) , where the implicit
constant depends only on kϕkLip .
We need an inequality for functions defined in a compact interval [0, 1].
This inequality also explains the meaning of interpolation.
Lemma 74. Let f ∈ C 2 [0, 1]. Then |f 0 (0)| . kf 00 kL1 ([0,1]) + kf kL1 ([0,1]) .
Proof If x∈ 43 , 1 and y ∈ 14 , 12 , then f (x) − f (y) = (x − y)f 0 (z) for
   

some z ∈ 21 , 34 by the mean-value theorem. Thus,

|f 0 (0)| ≤ |f 0 (0) − f 0 (z)| + |f 0 (z)|

≤ kf 00 kL1 ([0,1]) + 4|f (x) − f (y)|
≤ kf 00 kL1 ([0,1]) + 4|f (x)| + |4f (y)|.

If we take the average over x ∈ 34 , 1 and y ∈ 41 , 12 , we obtain the desired

   

result.
Proof of Theorem 73 Write Σ ≡ {z ∈ Rn−1 : |z| = 1, zn < −M |z 0 |}.
Equip Σ with the (hyper)surface measure σ. We may suppose that f ∈
C ∞ (Ωϕ ) thanks to Lemma 67. Then we have
Z
|v · Df (x)|p dσ(v) ∼ |Df (x)|p (x ∈ Ω).
Σ

With v ∈ Σ and x ∈ Ωϕ fixed, we consider u(t) ≡ f (x + tσ). Then thanks to

Lemma 74,
Z 1 Z 1
|v · Df (x)| = |u0 (0)| . |u(t)|dt + |u00 (t)|dt,
0 0

or equivalently,
Z 1 Z 1
|v · Df (x)| . |f (x + tσ)|dt + |D2 f (x + tσ)|dt.
0 0

Consequently,
Z Z 1 Z Z 1
|Df (x)|p . |f (x + tσ)|p dtdσ(v) + |D2 f (x + tσ)|p dtdσ(v).
Σ 0 Σ 0
 Banach function lattices 73

If we integrate this inequality against x ∈ Ωϕ , we obtain

Z
|Df (x)|p dx
Ω
Z Z Z 1 Z Z Z 1
. |f (x + tσ)|p dtdσ(v)dx + |D2 f (x + tσ)|p dtdσ(v)dx
ZΩ Σ 0 Z Ω Σ 0

. |f (x)|p dx + |D2 f (x)|p dx.

Ω Ω

Thus, we obtain the desired result.

Corollary 75. Let 1 ≤ p < ∞. For all f ∈ W 2,p (Ωϕ ) and ε > 0,

kDf kLp (Ωϕ ) . εkD2 f kLp (Ωϕ ) + ε−1 kf kLp (Ωϕ ) ,

where the implicit constant depends only on kϕkLip .

Proof Let ϕε (x0 ) ≡ ε−1 ϕ(εx0 ), x0 ∈ Rn−1 . Then we have

kDgkLp (Ωϕε ) . kD2 gkLp (Ωϕε ) + kgkLp (Ωϕε )

for all g ∈ W 2,p (Ωϕε ). For f ∈ W 2,p (Ωϕ ), the function g = f (ε·) belongs to
W 2,p (Ωϕε ). Thus,

εkDf (ε·)kLp (Ωϕε ) . ε2 kD2 f (ε·)kLp (Ωϕε ) + kf (ε·)kLp (Ωϕε ) .

If we change variables, we obtain

kDf kLp (Ωϕ ) . εkD2 f kLp (Ωϕ ) + ε−1 kf kLp (Ωϕ ) .

So, we obtain the desired result.

1.6.2 Hölder–Zygmund spaces

Unlike the class C r (Rn ), r ∈ N0 , we need to take a couple of steps to define
homogeneous Hölder–Zygmund spaces according to the parameter γ which is
equipped with the space C˙γ (Rn ). We deal with C˙γ (Rn ) for 0 < γ < 2 and then
we pass to the general parameters.
We want to define the homogeneous Hölder–Zygmund space C˙γ (Rn ) for
0 < γ < 1 using the first order difference as follows.
Definition 32 (Homogeneous Lipschitz space). For f ∈ C(Rn ) and 0 < γ <
1, define
|f (x) − f (y)|
kf kLipγ ≡ sup = sup |y|−γ |f (x + y) − f (x)|.
x,y∈Rn ,x6=y |x − y|γ n
x,y∈R ,y6=0

The homogeneous Lipschitz space Lipγ (Rn ) is the set of all continuous func-
tions f for which the norm kf kLipγ is finite.
74 Morrey Spaces

This definition of Lipγ (Rn ) is reasonable since the space Lipγ (Rn ) gen-
eralizes the familiar space Lip(Rn ) made up of all functions f satisfying
kf kLip ≡ sup |x − y|−1 |f (x) − f (y)| < ∞. The trouble is that we cannot
x,y∈Rn ,x6=y
extend this definition naturally to γ > 1; the space Lipγ (Rn ) is nothing but
P0 (Rn ) once we use the same formula using the first order difference.
So we consider another expression of Lipγ (Rn ). We define C˙γ (Rn ) with
0 < γ < 2 using the difference operator of order 2.
Definition 33 (Homogeneous Hölder–Zygmund space). For 0 < γ < 2 and a
continuous function f : Rn → C, define

kf kĊ γ ≡ sup |y|−γ |f (x + y) − 2f (x) + f (x − y)|.

x,y∈Rn ,y6=0

The homogeneous Hölder–Zygmund space C˙γ (Rn ) is the set of all continuous
functions f for which the norm kf kĊ γ is finite.
Remark that if f ∈ C(Rn ) satisfies kf kĊ γ = 0 in Definition 33, then
f ∈ P1 (Rn ), so that we regard C˙γ (Rn ) as a normed space modulo P1 (Rn ). See
Exercise 42.
We will show that Lipγ (Rn ) and C˙γ (Rn ) are isomorphic for 0 < γ < 1 in
Theorem 82.
We employ Definition 33 to investigate the property of Hölder–Zygmund
spaces. Let us now choose ψ ∈ Cc∞ (Rn ) so that

χB(1) ≤ ψ = ψ(−·) ≤ χB(2) . (1.4)

For j ∈ Z, define

ψ j ≡ 2jn ψ(2j ·), ϕj ≡ ψ j − ψ j−1 . (1.5)

We show how to use the difference operator of order 2 in the next lemma.
Lemma 76. Let f ∈ C˙γ (Rn ) with 0 < γ < 2. Then kϕj ∗ f kL∞ . 2−jγ kf kĊ γ
for all j ∈ Z.
Z
1
Proof We note that ϕj ∗f (x) = ϕj (y)(f (x−y)−2f (x)+f (x+y))dy
2 Rn
since ϕj has zero integral and ϕj is an even function. It remains to use the
triangle inequality and the definition of the norm in Definition 33.
Before we go further, we will develop the argument in Lemma 76.
Let β be a multi-index with length 2. By the use of the Taylor expan-
sion we see that ϕj ∗ ϕk and ∂ β ϕj are even functions having zero inte-
gral and satisfying |ϕj ∗ ϕk | . 2min(j,k)n−2|j−k| χB(2min(j,k)+2 ) , |∂ β ϕj | .
2j(2+n) χB(2j+1 ) and |∂ α ϕj | . 2j(n+|α|) χB(2j+1 ) , which yields kϕj ∗ϕk ∗f kL∞ .
2min(j,k)n−2|j−k| kf kĊ γ , k∂ β ψ j ∗ f kL∞ . 2j(2−γ) kf kĊ γ , and k∂ α (ϕj ∗ ϕk ∗
f )kL∞ . min(2j|α|−kγ , 2k|α|−jγ )kf kĊ γ for all f ∈ C˙γ (Rn ). So we are led to
the following estimate:
 Banach function lattices 75

Lemma 77. Let f ∈ C˙γ (Rn ) with 0 < γ < 2. Then kϕk ∗ ϕj ∗ f kL∞ .
∞
2−2|j−k|−min(j,k)γ kf kĊ γ for all j, k ∈ Z. In particular, ϕj ∗ϕk ∗f = ϕk ∗f
P
j=−∞
and k∂ α (ϕk ∗ f )kL∞ . 2k|α|−kγ kf kĊ γ for all k ∈ Z, α ∈ N0 n and f ∈ C˙γ (Rn )
with 0 < γ < 2 and for all multi-indexes β with |β| = 2, lim ∂ β [ψ j ∗ f ] = 0
j→−∞
in L∞ (Rn ).
∞ ∞
kϕj ∗ f kL∞ . 2−jγ kf kĊ γ ∼ kf kĊ γ from Lemma 76, which
P P
Hence
j=0 j=0
leads us to the following definition:
∞
Definition 34. Let f ∈ C˙γ (Rn ) with 0 < γ < 2. The function H ≡ ϕj ∗ f
P
j=0
∞
ϕ−j ∗f ,
P
is called the high frequency part of f . Meanwhile, the function G̃ ≡
j=1
which is defined formally, is called the low frequency part of f .
The trouble is that there is no guarantee that the right-hand side defining
G̃ is convergent in some suitable topology. To circumvent this problem, we
consider its derivative as follows:
Lemma 78. Let f ∈ C˙γ (Rn ) with 0 < γ < 2. Let ψ ∈ Cc∞ (Rn ) satisfy (1.4).
0
Define ϕj by (1.5) for j ∈ Z. Then Gα ≡ ∂ α [ϕj ∗ f ] is convergent in
P
j=−∞
L∞ (Rn ) whenever |α| ≥ 2. Hence, Gα is smooth.
Proof Simply use Lemma 77.
To construct a substitute of the lower part, we need to depend on a geo-
metric property of Rn ; HDR
1
(Rn ) = 0. Applying this fact, we can prove the
following:
Lemma 79. Let N ∈ N. Suppose that we have {fα }α∈N0 n ,|α|=N ⊂ C ∞ (Rn )
satisfying the curl-free condition: ∂β fα = ∂β 0 fα0 for all α, α0 , β, β 0 ∈ N0 n with
|α| = |α0 | = N and α + β = α0 + β 0 . Then there exists f ∈ C ∞ (Rn ) such that
∂ α f = fα .
Proof We prove this lemma by induction on N . If N = 1, then the result
1
is immediate from the Poincaré lemma, or equivalently, HDR (Rn ) = 0. Let
N ≥ 2. Define gjγ ≡ fej +γ if j = 1, 2, . . . , n and |γ| = N − 1. Let ej ≡
(0, . . . , 0, 1, 0, . . . , 0), where 1 is in the j-th lot. Then ∂ δ gjγ = ∂ δ fej +γ =
0 0
∂ δ fej0 +γ = ∂ δ gj 0 γ for all j, j 0 = 1, 2, . . . , n, γ, δ, δ 0 ∈ N0 n with |γ| = N − 1
and ej + δ = ej 0 + δ 0 . Thus, we are in the position of applying the Poincaré
lemma to have gγ ∈ C ∞ (Rn ) satisfying gjγ = ∂ j gγ for all γ with |γ| = N − 1
and j = 1, 2, . . . , n.
Let α, α0 , γ, γ 0 ∈ N satisfy |γ| = |γ 0 | = N − 1 and |α| = |α| > 0 and α + γ =
α + γ 0 . Suppose α − ej , α0 − ej 0 ≥ 0. Then ∂ α gγ = ∂ α−ej ∂ j gγ = ∂ α−ej gjγ =
0
76 Morrey Spaces
0 0 0 0 0
∂ α−ej fej +γ and ∂ α gγ 0 = ∂ α −ej0 ∂ j gγ 0 = ∂ α −ej0 gj 0 γ 0 = ∂ α −ej0 fej0 +γ 0 . Since
0
(α − ej ) + (ej + γ) = (α0 − ej 0 ) + (ej 0 + γ 0 ), we obtain ∂ α gγ = ∂ α gγ 0 , thus,
∞ n
by the induction assumption to have a function f ∈ C (R ) such that gγ =
∂ γ f . Consequently, if |α| = N , then by choosing j so that ej ≤ α we have
fα = ∂j gα−ej = ∂j ∂ α−ej f = ∂ α f .
We go back to the analysis of the low frequency part. As before choose
ψ ∈ Cc∞ (Rn ) so that χB(1) ≤ ψ = ψ(−·) ≤ χB(2) . Define ϕj ≡ 2jn ψ(2j ·) −
2(j−1)n ψ(2j−1 ·) for j ∈ Z. By using Lemma 79, we have the following control
of the low frequency part:
Corollary 80. Let f ∈ C˙γ (Rn ) with 0 < γ < 2. Let ψ ∈ Cc∞ (Rn ) satisfy
(1.4). Define ϕj by (1.5) for j ∈ Z. There exists a function G ∈ C ∞ (Rn ) such
0  
that ∂ α G = ∂ α ϕj ∗ f for all α ∈ N0 with |α| ≥ 2.
P
j=−∞

We further investigate the property of G in Corollary 80.

Proposition 81. Let 0 < γ < 2 and f ∈ C˙γ (Rn ). Define the low frequency
part G and the high frequency part H as in Definition 34 and Corollary 80.
Then f − G − H ∈ P1 (Rn ).
Proof Let α ∈ N0 n satisfy |α| = 2. Note that f − H = ψ ∗ f , so that
0
∂ α (f − H) = ∂ α ψ ∗ f = ∂ α ψl ∗ f = ∂ α G thanks to Lemma 76. Thus,
P
l=−∞
f − H differs from G by a linear function.
We show that two different definitions Lipγ (Rn ) in Definition 32 and
˙γ
C (Rn ) in Definition 33 are actually the same for 0 < γ < 1.
Theorem 82. The two sets Lipγ (Rn ) and C˙γ (Rn ) are the same for 0 < γ < 1.
Furthermore, their two norms are equivalent.
It is clear that Lipγ (Rn ) is a subset of C˙γ (Rn ), since the difference of the
second order can be decomposed into the difference of the difference of the
first order. So, the heart of the matter is to show the opposite inclusion.
Proof Let f ∈ C˙γ (Rn ). We will show that F ≡ G + H ∈ Lipγ (Rn ).
We need to show that

|F (x + y) − F (x)| . |y|γ kf kĊ γ . (1.6)

Let j0 ∈ Z be chosen so that

1 ≤ 2j0 |y| < 2. (1.7)

∞
ϕj ∗ f , Gj0 ≡ G + H − Hj0 . Once we prove
P
We define Hj0 ≡
j=j0

|Gj0 (x + y) − Gj0 (x)| . |y|γ kf kĊ γ (1.8)

 Banach function lattices 77

and
|Hj0 (x + y) − Hj0 (x)| . |y|γ kf kĊ γ , (1.9)
then we have (1.6).
To prove (1.8), by Lemma 77 we estimate

|Gj0 (x + y) − Gj0 (x)| . |y| · k∇Gj0 kL∞

0 −1
jX
∇ ϕj ∗ f
 
= |y|
j=−∞
L∞
0 −1
jX
|y| · ∇ ϕj ∗ f
 
≤ L∞
.
j=−∞

Using Lemma 77 and (1.7), we have

0 −1
jX
|Gj0 (x + y) − Gj0 (x)|
. 2j(1−γ) |y|1−γ kf kĊ γ ∼ kf kĊ γ .
|y|γ j=−∞

To prove (1.9) we use Lemma 76 to give

∞
X ∞
X
|Hj0 (x + y) − Hj0 (x)| . kϕj ∗ f k∞ . 2−jγ kf kĊ γ ∼ |y|γ kf kĊ γ ,
j=j0 j=j0

as was to be shown.
From now on, we can identify Lipγ (Rn ) with C˙γ (Rn ) for 0 < γ < 1.
So far, we do not consider the meaning of the parameter γ. If γ ∈ (1, 2),
we can differentiate functions in C˙γ (Rn ).
Theorem 83. Let f ∈ C˙γ (Rn ), and let 1 < γ < 2. Then f ∈ C 1 (Rn ) and
∂j f ∈ C˙γ−1 (Rn ) for all j = 1, 2, . . . , n.
Proof We use the functions G and H in the proof of Theorem 82. Arguing
∞
∂j [ϕl ∗ f ], j = 1, 2, . . . , n, converges uniformly,
P
as before, we can show that
l=0
so that H ∈ C 1 (Rn ). Since f differs from G + H by a linear function, we see
that f ∈ C 1 (Rn ). We can show that ∂j (G + H) ∈ C˙γ−1 (Rn ) by an argument
similar to Theorem 82. Hence, ∂j f ∈ C˙γ−1 (Rn ).
We consider the converse of Theorem 83.
Proposition 84. Let f ∈ C 1 (Rn ), and let 1 < γ < 2. If ∂j f ∈ C˙γ−1 (Rn ) for
each j = 1, 2, . . . , n, then f ∈ C˙γ (Rn ).
Proof Let x, y ∈ Rn . We may assume that f is real-valued. By the mean-
value theorem,

f (x + y) − 2f (x) + f (x − y) = y · (∇f (x + ty) − ∇f (x − ty))

78 Morrey Spaces

for some 0 < t < 1. If we use ∂j f ∈ C˙γ−1 (Rn ) for each j = 1, 2, . . . , n, then
n
we obtain |f (x + y) − 2f (x) + f (x − y)| . |y|γ
P
k∂j f kC γ−1 .
j=1

We can define C˙γ (Rn ) for any γ > 0.

Definition 35 (C γ (Rn )). Let γ > 1. Then f ∈ C γ (Rn ) if and only if f ∈
C−1−[−γ] (Rn ) and ∂ α f ∈ C γ+1+[−γ] (Rn ) for all α ∈ N0 n with |α| = −1 − [−γ].
Note that Theorem 83 and Proposition 84 justify Definition 35.
The spaces handled above are called the homogeneous spaces. We can also
define inhomogeneous spaces by adding the lower terms: The inhomogeneous
Hölder–Zygmund space C γ (Rn ), γ > 0, is the set of all continuous functions f
P theα derivative up to order L, which satisfies L < γ, such that kf kC =
with γ

k∂ f kL∞ + kf kC [γ] is finite. Likewise it is worth defining inhomogeneous

|α|<γ
Lipschitz spaces. Let 0 < α < 1. Write BCα (Rn ) = L∞ (Rn )∩Lipα (Rn ). Equip
BCα (Rn ) with the norm kf kBCα (Rn ) = kf kL∞ + kf kLipα .

Exercises
Exercise 37.
(1) Prove Lemma 59(1) by the use of the Lebesgue differentiation theorem.
(2) Prove Lemma 59(2) by the use of the Stokes theorem.
Exercise 38. Let 1 ≤ p ≤ ∞ and m ∈ N.
(1) Show that W m,p (Rn ) is a normed space using the fact that Lp (Rn ) is a
normed space.
(2) Prove Theorem 60.
Exercise 39. Using the Banach–Alaoglu theorem (see Theorem 89), prove
Theorem 63.
Exercise 40. Prove Theorem 64 for f ∈ Cc∞ (Rn ) and then complete the
proof of Theorem 64.
Exercise 41.
(1) Prove Theorem 65 by using Theorems 182 and 193 to follow.
(2) Prove Theorem 66 by inducting on m. Hint: The base case m = 1 is
Theorem 65.
Exercise 42. Let f : Rn → C be a continuous function satisfying f (x + y) +
f (x − y) = 2f (x) for all x, y ∈ Rn .
 Banach function lattices 79

(1) Establish that there exist uniquely a0 , a1 , a2 , . . . , an ∈ C such that

f (m1 , m2 , . . . , mn ) = a0 + a1 m1 + · · · + an mn for all (m1 , m2 , . . . , mn ) ∈
Zn . Here and below we fix such a0 , a1 , a2 , . . . , an ∈ C.
(2) Establish that f (q1 , q2 , . . . , qn ) = a0 + a1 q1 + · · · + an qn for all
(q1 , q2 , . . . , qn ) ∈ Qn .
(3) Conclude that f ∈ P1 (Rn ) by the continuity argument.

Exercise 43. Let k, m ∈ N.

dk −t2 2
(1) Show that k
e = pk (t)e−t for some pk ∈ Pk (Rn ).
dt
dk √ 2 1−2k
(2) Show that k
t + 1 = pk (t)(t2 + 1) 2 for some pk ∈ Pk (Rn ).
dt
dk 1 pk (t)
(3) Show that = for some pk ∈ P(m−1)k (Rn ).
dtk (1 + t2 )m (1 + t2 )m(k+1)
dk − 1 pk (t) 1
(4) Show that k
e t = k+1 e− t for some pk ∈ Pk−1 (Rn ). for t 6= 0.
dt t
Exercise 44. Let F : Rn → Rn be a bi-Lipschitz function. That is, |F (x) −
F (y)| ∼ |x − y| for all x, y ∈ Rn .
(1) Show that F (Rn ) is closed.
(2) Show that F (Rn ) is open.
(3) Show that F is a homeomorphism.
(4) Suppose we have a bi-Lipschitz mapping Φ defined on `2 (N). Can we
say that f is a homeomorphism?

1.7 Notes
Section 1.1
General remarks and textbooks in Section 1.1
There are many books on Lebesgue spaces; see [156, 258, 319] for example.
We refer to the exhaustive textbooks [29] and [296] for Banach function spaces.
We refer to the textbooks [189, 319] for an introduction of integration theory.
We refer to the recent textbook [203] for mixed Morrey spaces. The authors
hit upon Exercise 9 after the lecture of Kaoru Yoneda.
80 Morrey Spaces

Section 1.1.1 and 1.1.2

The materials dealt with here are standard and we do not recall to whom
the definitions belong.

Section 1.1.3
See [302] for Marcinkiewicz spaces.
In 1961, Benedek and Panzone introduced Lebesgue spaces with mixed
norm [27]. See [24, 128, 195, 196, 211] for the studies of mixed Lebesgue
spaces such as the boundedness of the Hardy–Littlewood and strong maximal
operators, multivariate rearrangements, the theory of variable exponents, and
interpolation theory.

Section 1.2
General remarks and textbooks in Section 1.2
Morrey spaces go back to the work of Morrey [323], which handled elliptic
differential equations. Actually Morrey pointed out that the function is con-
tinuous once the derivative satisfies a weak integrability condition. Peetre and
Stampaccia considered Morrey and Campanato spaces [353, 413]. We refer to
the textbooks [235, Chapter 13], [258, Chapter 5], [438] and [382, §6.1.5] for
Morrey spaces. We refer to [29], [258, Chapter 6] for Banach function spaces.

Section 1.2.1
n
There are many papers dealing with | · |− p in the context of Morrey
spaces. Zorko considered this function to explain that smooth functions cannot
approximate this function [472, p. 587]. See [304, Proposition 4.1] for Example
6, for example.
The space Mp1 (Rn ) was defined by Adams, Xiao, Giga and Miyakawa.
See [1, (4)], [149, p. 582, §2] as well as [8, p.1631] for Mp1 (Rn ), Definition 9.
Among others, Adams used Mp1 (Rn ) to characterize the boundedness property
of the convolution operator [1, Theorems 1 and 2]. We can find Proposition
15, asserting that Mpq (Rn ) is trivial for 0 < p < q < ∞, in [42, p. 13].

Section 1.2.2
We calculated kχB(x,r) kMpq ; see Example 6 [252, Lemma 1], for example.
The dyadic rearrangement (Example 10) is due to Olsen [342]. Olsen worked
in R3 and constructed an example in [342, Theorem 10]. Example 10 is a
passage to Rn of [342, Theorem 10].
We considered the power function in Example 7. Example 7(1) can be
found in [472], while Example 7(2) and (3) can be found in [42, Examples 1
and 2].
In [390], Sawano, Sugano and Tanaka considered the self-similar increas-
ing/decreasing sequence for Morrey spaces. Examples 9 and 11 come from [390,
 Banach function lattices 81

Proposition 4.1], based on the proof of [64, Theorem 2.1]. See also [304, Lemma
4.6] and [391, Proposition 2.1] for Examples 9 and 11. The term “wasteful”
comes from the proof of the survey paper [391, Proposition 2.1].

Section 1.2.3
Giaquinta pointed out the role of the parameter q; see [148, p. 67] for
Theorem 18.
As we saw in this book, we have an approach using a self-similar decreasing
sequence for Morrey spaces. See [390, Proposition 4.1] together with detailed
calculation [304, p.120] and [393, Lemma 2.1] for Example 9.
As we saw in Theorem 20, Mpq̃ (Rn ) is not dense in Mpq (Rn ) whenever
1 < q < q̃ < p; see the paper [380].

Section 1.2.4
Triebel pointed out that the Morrey space Mpq (Rn ) with 0 < q < p < ∞
does not have L∞ n
c (R ) and the Schwartz space as dense subspaces; see [438,
Proposition 2.16] for Proposition 22.
As a modification of Morrey spaces, we can consider the Stummel class
considered by Ragusa and Zamboni [365]. See [442] for the recent development.
Eridani and Gunawan, pointed out the inclusion between Mp1 (Rn ) and the
Stummel class [115]. See [116] for a further generalization to various directions.
As we saw in Proposition 21, Morrey spaces are not rearrangment invari-
ant, that is, it can happen that kf kMpq > kgkMpq if λf = λg . Furusho and Ono
proposed to restrict the class of Morrey functions to a special case by using a
finite subdivision of cubes; see [137, Definition 1] and [345, 346, 347]. See also
[347] for the inclusionship similar to Theorem 19.

Section 1.2.5
The definition of weak Morrey spaces can be found in [405]; see also [2,
Section 5]. Applications to PDE can be found in [314, 360]. See [304, Lemma
4.7(ii)] for Example 17. See also [173] for Example 17, a more quantitative
approach for the difference between WMpq (Rn ) and Mpq (Rn ).
In [172], we can find Lemma 23, which compares WMpr (Rn ) and Mpq (Rn )
for 0 < q < r ≤ p.
As we saw in Theorem 24, the weak Morrey space WMpq (Rn ) is smaller
than Mpq (Rn ); see [139, p. 2143], [149, Proposition 2.2], [440, p. 4] and [471,
p.86] as well as [172]. We refer to [186] for a weighted version.

Section 1.2.6
The terminology of ball Banach function spaces can be tracked back to
[178, Definition 2]. See [398, Example 3.3] for an account of why Morrey
spaces are not Banach function spaces.
82 Morrey Spaces

Section 1.3
General remarks and textbooks in Section 1.3
Theory of Herz spaces goes back to the work of Herz [193]. See the textbook
[458] for Herz spaces.

Section 1.3.1
Although Wiener did not consider local Morrey spaces when his paper [447]
was published, the idea of defining local Morrey spaces can be found in [447]. In
fact, Wiener considered the case of p = 1, 2 in the context of Fourier analysis.
Beurling [31] introduced the spaces B p (Rn ) and Ḃ p (Rn ), whose norm is given
(p) (p)
by kf kB p ≡ sup mQ(r) (f ) and kf kḂ p ≡ sup mQ(r) (f ) together with the pred-
r≥1 r>0
ual Ap (Rn ) so-called a Beurling algebra. In this case B p (Rn ) = B np (Lp )(Rn )
and Ḃ p (Rn ) = Ḃ np (Lp )(Rn ). Later, to extend Wiener’s ideas [447, 448] which
describe the behavior of functions at infinity, Feichtinger [126] gave an equiv-
alent norm on B p (Rn ), which is a special case of norms to describe non-
α
homogeneous Herz spaces Kp,r (Rn ) introduced in [193]. Garcı́a-Cuerva and
Herrero [142] and Alvarez, Guzmán-Partida and Lakey [17] introduced the
non-homogeneous central Morrey space B p,λ (Rn ) (the non-homogeneous Bσ -
Lebesgue space/ the non-homogeneous central Morrey space) and the central
Morrey space Ḃ p,λ (Rn ) (the homogeneous Bσ -Lebesgue space/ the homoge-
neous central Morrey space).
  p1
−λ− n p
kf kWB p,λ ≡ sup r p sup t |{|f | > t} ∩ Q(r)|
r≥1 t>0

and
  p1
−λ− n p
kf kWḂ p,λ ≡ sup r p sup t |{|f | > t} ∩ Q(r)| .
r>0 t>0

Bσ -Lipschitz and Bσ -Morrey-Campanato spaces were introduced in [248] and

[306], respectively. Komori-Furuya, Matsuoka and Nakai studied the bounded-
ness of fractional integral operators on these spaces. Since Morrey and Cam-
panato spaces realize Lebesgue spaces, BMO and Lipschitz spaces as special
cases, Bσ -Morrey-Campanato spaces cover B p,λ (Rn ), Ḃ p,λ (Rn ), CMOp,λ (Rn ),
CBMOp,λ (Rn ), as well as the usual Morrey, Campanato and Lipschitz spaces.
The word “CMO” stands for “central mean oscillations” and “CBMO” stands
for “central bounded mean oscillations”. See Example 102 for [249].
Komori-Furuya, Matsuoka, Nakai and Sawano considered a new framework
to grasp central Morrey spaces, Campanato spaces and so on [250, §2]. See
[250, 251, 399] for more about Bσ -spaces.
 Banach function lattices 83

Section 1.3.2
Herz–Morrey spaces are defined by replacing the Lp (Rn )-norm with the
Morrey norm k · kMpq in some sense. Lu and Xu defined Herz–Morrey spaces
[295, Definition 1.1]. Wang and Shu considered the Littlewood–Paley gλµ -
function in Herz–Morrey spaces [445]. Herz–Morrey spaces are investigated
by Izuki in the setting of variable exponents. For details see [214, 215].
Wu investigated the boundedness property of Hardy operators as well as
the Riesz transform in [452, 453, 454]. We refer to [431] for the multilinear
case.
Local Morrey spaces and Herz spaces overlap largely [167, Lemma 8.1]; see
Theorem 26. Mizuta and Ohno took an approach similar to Theorem 26 [321].
Gao, Tang, Xue and Zhou investigated the boundedness properties of com-
mutators generated by Hardy operators and BMO functions [140, 429].
Herz–Morrey spaces can be defined over locally compact groups. More
precisely, Wu and Liu worked in locally compact Vilenkin groups in [451, 455].
Dzhabrailov and Khaligova worked in the anisotropic setting in [110, 111].
Fan and Gao considered the boundedness property of the bilinear fractional
integral operator acting on Herz–Morrey spaces [120].
Kuang considered Hilbert/Hausdorff operators acting on Herz–Morrey
spaces; see [256, 257].

Section 1.4
General remarks and textbooks in Section 1.4
Edmunds, Kokilashvili and Meskhi dealt with Hardy operators. Especially,
[112, Chapter 1] deals intensively with this topic [112].
See [373, Chapter 2] for Hardy’s inequality.
See [30, §1.3], [29, Chapter 2 §1], [156, Chapter 1], [147, §1.4], [258, Chapter
8] and [223, Chapter 0, §II] for the distribution of functions and Lorentz spaces.
In [147, Chapter 2], we can find some information on the Hardy–Littlewood
maximal operator. See [30], [156, §1.4], [417, Chapter V, §3] for a detailed
explanation on Lorentz spaces.

Sections 1.4.1 and 1.4.2

Lorentz defined Lorentz spaces in [288, 289]. Luxenberg investigated them
as well. The work of Luxenberg on Lorentz spaces is just a part of his paper
[296].

Section 1.4.3
We can find inclusion Φ↑λ,q0 (0, ∞) ⊂ Φ↑λ,q1 (0, ∞); see [50, Lemma 1] for
Lemma 37. We considered the boundedness of K0 ; see [77, Theorem 2.5] for
Example 40.
84 Morrey Spaces

Section 1.4.4
See [364] Lemma 40, Chiarenza’s lemma.
We polished [32, §2] to state Example 41.

Section 1.5
General remarks and textbooks in Section 1.5
See the exhaustive textbook and seminar note [29, 299, 366] as well as
[147, pp. 22–23], [258, Chapter 4] and [382, §6.1.6] for Orlicz spaces and Young
functions. Trudinger pointed out that Orlicz spaces can be used to compensate
for the failure of the endpoint case for integral operators. See [439]. We can
find some maximal estimates in [147, §5.2] (including [147, Chapter 6]) and
[382] and some singular integral estimates in [147, §5.6]. We refer to [258,
Chapter 13] for Lebesgue spaces with variable exponents. See also the survey
[216].

Section 1.5.1
Young proved an inequality for the Young function and its conjugate [466];
see Theorem 45. We obtained an estimate for ∇2 -functions. See [232, Lemma
1.3.2] for Lemma 49(1) and [387, Lemma 4.3] for Lemma 49(2). We borrowed
[297, Example 2.3] for Example 43.

Section 1.5.2
Koshi and Shimogaki obtained an equivalent expression of Orlicz norms
[238]; see Theorem 51.
Kita established similar results to Proposition 156 for Ψ(t) ≡ t on LΦ (Rn ).
See [228, 229].

Section 1.5.3
Carro, Pérez, F. Soria and J. Soria defined the Φ-average in [64, §3]; see
Definition 26. See [64, (3.1)] for Definition 26 including Example 46.

Section 1.5.4
Orlicz initially considered Lebesgue spaces with a variable exponent [348,
§2]. Later on Nakano considered variable Lebesgue norms [336, 337].
In particular, function spaces with variable exponents are necessary in
the field of electronic fluid mechanics [374] and the applications to image
restoration [69, 187, 276]. Kovacik and Rakosnik [243] gave an application of
generalized Lebesgue spaces with variable exponents to Dirichlet boundary
value problems for nonlinear partial differential equations with coefficients
of a variable growth. Another simple example of the application to differen-
tial equations can be found in [121, p. 438, Example], where Fan and Zhao
 Banach function lattices 85

implicitly showed that the variable Lebesgue spaces can be used to control
the non-linear term of differential equations.
See the survey paper [216] for a detailed explanation of the old books
[336, 337].

Section 1.6
General remarks and textbooks in Section 1.6
There are many textbooks dealing with Sobolev spaces; see the exhaustive
textbooks [12, 13, 309] as well as the textbooks [5], [117, §5.2], [118, Chapter
4], [415, §V.2] and [410, §6.3] for example.
See [117, §5.1], [415, Chapter V, §4] and [416, Chapter VI, §5] for a char-
acterization of Hölder–Zygmund spaces. For the extension, we refer to the
textbook [41].

Section 1.6.1
We followed the idea of Burenkov [41]. See [40] for Lemma 68. Theorem
70 is due to Burenkov [41]; see also the textbook [41, Lemma 18]. See [411]
for Theorem 66.

Section 1.6.2
See [415] for Definition 33. We followed [384] overall in this section.
Chapter 2
Fundamental facts in functional
analysis

This chapter collects some preliminary facts. The readers may choose to skip
Chapter 2. We indicate where we use each theorem in the other parts of this
book. We start with normed spaces in Section 2.1, while we concentrate on
Hilbert spaces in Section 2.2. Section 2.3 is used for the last chapter of the
second book. We will consider integration theory for measurable functions
assuming their values in a Banach space.

2.1 Normed spaces and Banach spaces

We do not go into a detailed discussion of normed spaces in Section 2.1.
Section 2.1.1 recalls various forms of the Hahn–Banach theorem. Section 2.1.2
is interesting in its own right. We consider a self-improvement of the triangle
inequality. Section 2.1.3 is oriented to the operations for Banach spaces.

2.1.1 Hahn–Banach theorem and Banach–Alaoglu theorem

We assume that readers are familiar with the Hahn–Banach theorem and
Banach–Alaoglu theorem. We list a couple of variants of the Hahn–Banach
theorem.
Theorem 85 (Hahn–Banach theorem/analytic form). Let V be a complex
vector space and p : V → C a function satisfying
p(u + v) ≤ p(u) + p(v), p(αu) = |α| p(u)
for all α ∈ C and u, v ∈ V . Assume that l0 : W0 → R, which is defined on
a subspace W0 , is a linear mapping satisfying l0 (u) ≤ p(u) for all u ∈ W0 .
Then there exists a linear mapping L : V → R such that L|W0 = l0 and that
|L(u)| ≤ p(u) for all u ∈ V .
Theorem 86 (Hahn–Banach extension). Let Y be a closed subspace of a
normed space X . Then a bounded functional f : Y → C can be extended to a
bounded functional F defined on X so that k F kX ∗ = k f kY ∗ .

87
88 Morrey Spaces

Theorem 87 (Existence of the norm attainer). Let X be a normed space and

x ∈ X . Then x has a norm attainer. Namely, there exists x∗ ∈ X ∗ with unit
norm such that x∗ (x) = kxkX .
Theorem 88 (Hahn–Banach theorem/separation version). Suppose that A is
a closed convex set in a normed space X . If b is a point outside A, then there
exists a continuous functional x∗ : X → C such that

sup Re x∗ (a) < x∗ (b).

a∈A

We next recall the Banach–Alaoglu theorem. Recall that the metric space
is separable if it is realized as the closure of a countable subset.
Theorem 89 (Banach and Alaoglu). Let X be a separable normed space.
Assume that {xj ∗ }∞ ∗
j=1 belongs to the closed unit ball B in X . Then there
∗
exists a subsequence {xjk }k∈N convergent with respect to weak-∗ topology.
That is,
lim xjk ∗ (x) = x∗ (x)
k→∞

for all x ∈ X .
We do not recall the proof of Theorems 85, 86, 87, 88 and 89. See [310] for
example.

2.1.2 Refinement of the triangle inequality

We will use Section 2.1.2 in Section 3.1.4. We prove an inequality valid for
any normed space, which is interesting in its own right and refines the original
triangle inequality.
Theorem 90. We have
 
x y
kx + ykX ≤ kxkX + kykX − 2 − + min(kxkX , kykX ).
kxkX kykX X

for any nonzero elements x and y in a normed space (X , k · kX ).

Proof Without loss of generality we may assume kxkX ≤ kykX . Then
 
kxkX kxkX kxkX
kx + ykX = x+ y+ 1− y
kxkX kykX kykX X
 
kxkX kxkX kxkX
≤ x+ y + 1− y
kxkX kykX X kykX X
x y
= kxkX + + kykX − kxkX
kxkX kykX X
 
x y
= kxkX + kykX − 2 − + min(kxkX , kykX .
kxkX kykX X
 Fundamental facts in functional analysis 89

2.1.3 Sum and intersection of Banach spaces

Keeping in mind the definition of Orlicz spaces, we consider the sum and
the intersection of quasi-Banach spaces. We will show that duality connects
these two notions of the sum and the intersection. We will use Section 2.1.3
mainly in Chapters 18 and 19.
Suppose that we have quasi-Banach spaces X0 , X1 , Y0 , Y1 , X , Y such that

X0 , X1 ,→ X Y0 , Y1 ,→ Y,

in the sense that X0 ⊂ X , X1 ⊂ X , Y0 ⊂ Y and Y1 ⊂ Y and that there exists

M > 0 such that

kx0 kX ≤ M kx0 kX0 , kx1 kX ≤ M kx1 kX1 ,

ky0 kY ≤ M ky0 kY0 , ky1 kY ≤ M ky1 kY1

for all x0 ∈ X0 , x1 ∈ X1 , y0 ∈ Y0 and y1 ∈ Y1 .
Recall that a linear topology on a linear space X is a topology of X in
which the multiplication and the addition are continuous. Such a space X is
called a topological vector space. We may assume that X and Y are merely
topological vector spaces in the sense that X and Y are topological spaces
equipped with continuous addition and continuous scalar multiplication.
In any of these cases, X0 and X1 are called compatible and we can deal
with elements in X0 and X1 as if they belonged to X . The same applies to
Y0 and Y1 . Thus, we can naturally define X0 + X1 and Y0 + Y1 . For example,
X0 + X1 is given by

X0 + X1 ≡ {x ∈ X : x = x0 + x1 , x0 ∈ X0 , x1 ∈ X1 }.

We automatically make X0 + X1 into a normed space. The norm is given by

kxkX0 +X1 ≡ inf{kx0 kX0 + kx1 kX1 : x0 ∈ X0 , x1 ∈ X1 , x = x0 + x1 }.

Likewise, the norm of X0 ∩ X1 is given by kxkX0 ∩X1 = max(kxkX0 , kxkX1 ). It

is routine to show that X0 + X1 and X0 ∩ X1 are Banach spaces. Suppose that
we have a linear operator T : X0 + X1 → Y0 + Y1 with the properties that
T (X0 ) ⊂ Y0 , T (X1 ) ⊂ Y1 and

kT x0 kY0 ≤ C0 kx0 kX0 , kT x1 kY1 ≤ C1 kx1 kX1

for all x0 ∈ X0 and x1 ∈ X1 . Here, C0 and C1 are independent of x0 and x1 ,

respectively.
Example 50. We defined L1 (Rn )+L∞ (Rn ) as the linear space of all functions
which can be expressed as a sum of L1 (Rn )-functions and L∞ (Rn )-functions in
Proposition 21. For f ∈ L1 (Rn )+L∞ (Rn ), a simple computation using the def-
inition of the norm, its norm is given by kf kL1 +L∞ = inf (L+kχ{|f |>L} f kL1 ).
L>0
See Exercise 48.
90 Morrey Spaces

Note that for a given x ∈ X0 + X1 representation x = x0 + x1 may not be

unique. In fact, we can charaterize this condition as follows:
Lemma 91. Let X0 , X1 be linear subsets of X . Then for any x ∈ X0 + X1
the representation x = x0 + x1 , x0 ∈ X0 , x1 ∈ X1 is unique if and only if
X0 ∩ X1 = {0}.
Proof Assume that X0 ∩ X1 ) {0}, or equivalently, there exists b ∈ X0 ∩
X1 \ {0}. Then 0 = 0 + 0 = b + (−b), which implies that the representation of
0 is not unique.
Suppose that X0 ∩ X1 = {0}. Assume that x0 , y0 ∈ X0 , x1 , y1 ∈ X1 and
z = x0 + x1 = y0 + y1 . Then x0 − y0 = y1 − x1 ∈ X0 ∩ X1 . Hence x0 − y0 ∈
X0 ∩ X1 = {0}, which implies y0 = x0 and y1 = x1 .
Example 51. Let 1 ≤ p0 , p1 ≤ ∞. Then

Lp0 (Rn ) + Lp1 (Rn ) = {f = f0 + f1 : f0 ∈ Lp0 (Rn ), f1 ∈ Lp1 (Rn )}.

Clearly, Lp0 (Rn ) ∩ Lp1 (Rn ) 6= {0}. Hence, the representation f = f0 + f1 is

not unique. In this case X is the set of all f ∈ L0 (Rn ), X0 = Lp0 (Rn ), X1 =
Lp1 (Rn ). Remark that L0 (Rn ) can be made into a topological vector space.
The topology is given by the functional f ∈ L0 (Rn ) 7→ E min(1, |f (x)|)dx ∈
R

R, where E moves over all measurable sets of finite measure in Rn .

We remark that X0 +X1 ,→ X and we are led to subspaces of X0 +X1 when
we consider interpolation spaces of spaces X0 , X1 . Among other properties, we
use the following duality fact:
Theorem 92 (Duality theorem). Suppose that (X0 , X1 ) is a compatible couple
of Banach spaces. Assume in addition that X0 ∩ X1 is dense in X0 and X1 .
Then (X0 ∩ X1 )∗ = X0 ∗ + X1 ∗ and (X0 + X1 )∗ = X0 ∗ ∩ X1 ∗ with coincidence
of norms; namely,
| hx∗ , xi |
kx∗ kX0 ∗ +X1 ∗ = sup for all x∗ ∈ X0 ∗ + X1 ∗ ,
x∈X0 ∩X1 \{0} kxkX0 ∩X1

| hx∗ , xi |
kx∗ kX0 ∗ ∩X1 ∗ = sup for all x∗ ∈ X0 ∗ ∩ X1 ∗ .
x∈X0 +X1 \{0} kxkX0 +X1

Proof Note that

(X0 ∩ X1 )∗ ←- X0 ∗ + X1 ∗ , (X0 + X1 )∗ ≈ X0 ∗ ∩ X1 ∗ (2.1)

and that
| hx∗ , xi |
kx∗ kX0 ∗ +X1 ∗ ≥ sup , (2.2)
x∈X0 ∩X1 kxkX0 ∩X1
| hx∗ , xi |
kx∗ kX0 ∗ ∩X1 ∗ = sup . (2.3)
x∈X0 +X1 kxkX0 +X1
 Fundamental facts in functional analysis 91

Let us prove the reverse inclusion of (2.1). To this end, we take x∗ ∈

(X0 ∩ X1 )∗ . The norm of X0 ⊕ X1 is defined by

k (x0 , x1 ) kX0 ⊕X1 = max(kx0 kX0 , kx1 kX1 ) ((x0 , x1 ) ∈ X0 ⊕ X1 ),

which immediately makes E ≡ {(x0 , x1 ) ∈ X0 ⊕ X1 : x0 = x1 ∈ X0 ∩ X1 } into

a closed subspace of X0 ⊕ X1 . Furthermore, the dual of X0 ⊕ X1 is canonically
identified with X0 ∗ ⊕ X1 ∗ , whose norm is given by

k (x0 ∗ , x1 ∗ ) kX0 ⊕X1 = kx0 ∗ kX0 + kx1 ∗ kX1 .

Then l : (x0 , x1 ) ∈ E 7→ 2−1 x∗ (x0 + x1 ) ∈ C is a continuous functional which

is dominated by the norm of X0 ⊕ X1 . Therefore, l extends to a continuous
linear functional L on X0 ⊕ X1 thanks to the Hahn–Banach theorem, Theorem
86, in such a way that kLk(X0 ⊕X1 )∗ ≤ kx∗ k(X0 ∩X1 )∗ As a result we obtain x0 ∗
and x1 ∗ such that

kx0 ∗ kX0 ∗ + kx1 ∗ kX1 ∗ = kLk(X0 ⊕X1 )∗ ≤ kx∗ k(X0 ∩X1 )∗ (2.4)

and that L(x0 , x1 ) = hx0 ∗ , x0 i + hx1 ∗ , x1 i for all (x0 , x1 ) ∈ X0 ⊕ X1 . Letting

x0 = x1 = x, we obtain hx∗ , xi = L(x, x) = hx0 ∗ , xi + hx1 ∗ , xi. Thus, x∗ =
x0 ∗ |X0 ∩ X1 + x1 ∗ |X0 ∩ X1 and we deduce from (2.4) that kx∗ kX0 ∗ +X1 ∗ ≤
kx0 ∗ kX0 ∗ + kx1 ∗ kX1 ∗ ≤ kx∗ k(X0 ∩X1 )∗ . This is the desired converse inequality.

2.1.4 Exercises
Exercise 45. [225] Let X and Y be Banach spaces and T : X → Y a linear
operator. Suppose that there exist constants M > 0 and 0 < ε < 1 with the
following property: For every y ∈ Y with unit norm, there exists x ∈ X with
kxkX ≤ M and kT x − ykY < ε. Then show that T is onto.
Exercise 46. Let X , Y be normed spaces, and let T : X → Y be a lin-
ear operator. Define kT kX →Y ≡ sup{kT xkY : kxkX = 1}. Then show that
kT kX →Y = sup{kT xkY : kxkX < 1}.
Exercise 47 (Clarkson’s inequality). [319, Theorem 12.6] Let (X, B, µ) be a
measure space, and let 1 < p ≤ 2.
p0 p0  p  1
a+b a−b |a| + |b|p p−1
(1) Show that + ≤ for all a, b ∈ R.
2 2 2
(Hint: We may assume b = 1 ≥ a.)
!p0 !p0
f +g f −g
(2) Show that +
2 Lp (µ) 2 Lp (µ)
0
p p p −1
(kf kLp (µ) ) + (kgkLp (µ) )
 
≤ for all f, g ∈ L0 (µ). (Hint: Prove
2
0 0 0
first that kf kLp0 (µ) p + kgkLp0 (µ) p ≤ kf + gkLp0 (µ) p for f, g ∈ M+ (µ).)
92 Morrey Spaces
p0 p0 0 0
a+b a−b |a|p + |b|p
(3) Show that + ≤ for all a, b ∈ R.
2 2 2
!p0 !p0
f +g f −g
(4) Show that +
2 Lp0 (µ) 2 Lp0 (µ)
0 0
(kf kLp0 (µ) )p + (kgkLp0 (µ) )p
≤ for all f, g ∈ L0 (µ).
2
Exercise 48. In Example 50, we considered the norm of f ∈ L1 (Rn ) +
L∞ (Rn ). Let L ≥ 0. Then find the minimum of kf − hkL1 , where h ∈ L0 (Rn )
moves over all functions satisfying khkL∞ = L.
Exercise 49. Let p(·) : Rn → [1, ∞) be a variable exponent. Let p+ ≡
ess.supx∈Rn p(x) and p− ≡ ess.inf x∈Rn p(x). Then establish that Lp+ (Rn ) ∩
Lp− (Rn ) ⊂ Lp(·) (Rn ) ⊂ Lp+ (Rn ) + Lp− (Rn ).

2.2 Hilbert spaces

A Banach space (H, k · kH ) is said to be a Hilbert space, if it comes with a
mapping h·1 , ·2 iH : H×H → C, with the following properties for all x, y, z ∈ H
and a ∈ C. The mapping h·1 , ·2 iH is an inner product.
(1) hx, xiH = kxkH 2 for all x ∈ H.
(2) hx, yiH = hy, xiH for all x, y ∈ H.
(3) hx + y, ziH = hx, ziH + hy, ziH for all x, y, z ∈ H.
(4) ha · x, yiH = a · hx, yiH .
Let H be a Hilbert space in Section 2.2. Section 2.2.1 handles bounded
sequences in Hilbert spaces, while Section 2.2.2 considers the uniform bound-
edness of operators acting on Hilbert spaces.

2.2.1 Komlos theorem

The Komlos theorem is a tool to create the limit when we have a bounded
sequence in a normed space. We do not consider the Komlos theorem itself
here. In this book we will use the following simple variant, which we use in
Chapter 17.
Proposition 93. Let H be a Hilbert space.
(1) Let C be a closed convex set in H. Then there exists an element h0 ∈ C
such that kh0 kH = inf{khkH : h ∈ C}.
 Fundamental facts in functional analysis 93

(2) Let f1 , f2 , . . . be a sequence in the closed unit ball of H. Choose gk ∈

co({fk+1 , fk+2 , . . .}) so that kgk kH = min{kgkH ∈ co({fk+1 , fk+2 , . . .})}
for each k ∈ N. Then {gk }∞ k=1 converges to an element in H.

Proof
(1) This is a standard result in functional analysis. See [409, Theorem 3.2.1],
for example.
(2) Write vk ≡ min{kgkH ∈ co({fk+1 , fk+2 , . . .})}. Note that kgk −gk0 kH 2 =
2kgk kH 2 + 2kgk0 kH 2 − kgk + gk0 kH 2 ≤ 2vk 2 + 2vk0 2 − 4vmin(k,k0 ) 2 , since
gk + gk0
∈ co({fmin(k,k0 )+1 , fmin(k,k0 )+2 , . . .})}.
2
Note that {vk }∞
k=1 is a bounded increasing sequence. Hence, it converges
and we obtain the desired result.

2.2.2 Cotlar’s lemma

Here, we will collect an orthogonality lemma. For a Hilbert space H denote
by P (H) the set of all projections; a bounded linear operator P : H → H
belongs to P (H) if and only if P = P ∗ and P 2 = P . Suppose that we have a
sequence of projections {Pj }∞j=1 ∈ P (H). If Pj Pk = 0, j 6= k, then

kP1 + P2 + · · · + PN kH→H ≤ 1. (2.1)

Of course, in general, (2.1) fails. A typical case is P1 = P2 = · · · . In a very

simple case, we do have Pj Pk = 0, j 6= k in many applications. However, with
a slight perturbation, Pj Pk = 0, j 6= k no longer holds. What we are going to
consider is a substitute of (2.1), when a situation slightly differs from the case
Pj Pk = 0, j 6= k.
Theorem 94. Suppose that {Tj }j∈J ⊂ B(H) is a finite collection, where
J ⊂ Z is a finite subset. Assume that a sequence of positive constants {γj }j∈Z
satisfies  
X X
A ≡ max  sup γj−j 0 , sup γj 0 −j  < ∞ (2.2)
j 0 ∈J j 0 ∈J
j∈J j∈J

and that

k Tj∗0 Tj kH→H + k Tj 0 Tj∗ kH→H ≤ γj 0 −j 2 (j, j 0 ∈ J). (2.3)

P
Then the operator TJ ≡ Tj satisfies k TJ kH→H ≤ A.
j∈J

Proof If we write out in full, then

X
(TJ∗ TJ )m = Tj∗1 Tj2 Tj∗3 · · · Tj2m . (2.4)
j1 ,j2 ,...,j2m ∈J
94 Morrey Spaces

We will estimate this sum by majorizing the norms of the individual sum-
mands.
First, associating the factors in each summand as
m
Y
Tj∗1 Tj2 Tj∗3 · · · Tj2m = (Tj∗2l−1 Tj2l ),
l=1

and using inequality in (2.3), we obtain

m
Y
k Tj∗1 Tj2 Tj∗3 · · · Tj2m kH→H ≤ γj2l−1 −j2l 2 . (2.5)
l=1

Alternatively, we can associate the factors as

m−1
!
Y
Tj∗1 Tj2 Tj∗3 · · · Tj2m = Tj∗1 (Tj2l Tj∗2l+1 ) Tj2m .
l=1

Since kTj1 kH→H ≤ γ0 ≤ A, and similarly kTj2m kH→H ≤ A, thanks to (2.3)

we obtain
m−1
Y
k Tj∗1 Tj2 Tj∗3 · · · Tj2m kH→H ≤ A2 γj2l −j2l+1 2 . (2.6)
l=1

We take the geometric mean of (2.5) and (2.6) and insert this into (2.4). The
result is
X 2m−1
Y
k (TJ∗ TJ )m kH→H ≤ A γjl −jl+1 .
j1 ,j2 ,...,j2m ∈J l=1
P
In the above, we first sum in j1 and use the fact that γj1 −j2 ≤ A. Next, we
P j1 ∈J
sum in j2 , using γj2 −j3 ≤ A. Continuing in this way for j1 , j2 , . . . , j2m−1
j2 ∈J
yields k (TJ∗ TJ )m kH→H ≤ A2m−1 ]J. Recall that we assumed J finite. Thus,
1
kTJ kH→H ≤ A(]J) 2m . Finally, we let m → ∞, proving the theorem.

2.2.3 Exercises
Exercise 50. Let 1 < p < ∞. By using Clarkson’s inequality (see Exercise
47), prove an analogue in Lp ([0, 1]) to Proposition 93.
Exercise 51. Suppose that we have {Tj }j∈Z ⊂ B(H). and a sequence of
positive constants {γj }j∈Z satisfying (2.3) with J = Z and
∞
X
A≡ γj < ∞. (2.7)
j=−∞

∞
Tj Tk∗ x converges for all x ∈ H and k ∈ Z.
P
(1) Show that
j=−∞
 Fundamental facts in functional analysis 95
∞
P S
(2) Show that Tj z converges for all z ∈ Im(Tj ).
j=−∞ j∈Z

∞
P
(3) Construct the limit of Tj in an appropriate sense. Hint: What
j=−∞
∞
T
happens on ker(Tj )?
j=−∞

2.3 Bochner integral

The Bochner integral is an advanced topic for beginners. In a word the
Bochner integral replaces R or C in the range by Banach spaces. In the context
of partial differential equations, it appears naturally: Suppose that we are
trying to solve a PDE given by

ut (x, t) = ∆u(x, t) + f (x, t), u0 (x) = u0 (x).

Here, f (·, t) ∈ Lp (Rd ). Then the solution is given by

Z t
t∆
u(x, t) = e u0 (x) + e(t−s)∆ [f (·, s)]ds.
0

The integral appearing in the right-hand side of this formula is strictly speak-
ing the Bochner integral.
For the sake of simplicity, we assume that (X, B, µ) is a complete σ-finite
measure space in Section 2.3.
Section 2.3 is interested mainly in Bochner integrals. Section 2.3.1 extends
the notion of measurability of functions to Banach spaces-valued functions.
Section 2.3.2 considers some convergence theorems in analogy with the ones
in the theory of the Lebesgue integral. Section 2.3.3 is devoted to the theory
of repeated integrals.

2.3.1 Measurable functions

Let X be a Banach space. Our present aim, to which we have been refer-
ring, is the construction of the integration theory for X -valued functions. The
theory of the underlying measure space (X, B, µ) is already set up, so it seems
appropriate that we start from the definition of the measurability of X -valued
functions.
Definition 36. Let X be a Banach space, and let (X, B, µ) be a measure
space. Suppose that ϕ : X → X is a function.
(1) The function ϕ is weakly measurable, if b∗ ◦ ϕ : X → C is measurable for
all b∗ ∈ X ∗ .
96 Morrey Spaces

(2) The function ϕ is simple, if there exist b1 , b2 , . . . ∈ X and A1 , A2 , . . . ∈ B

P∞
such that ϕ(x) = χAj (x) · bj for µ-almost all x ∈ X.
j=1

(3) The function ϕ is strongly measurable, if there exists a sequence of simple

functions {ϕj }∞
j=1 such that lim ϕj (x) = ϕ(x) for µ-a.e. x ∈ X.
j→∞

(4) The function ϕ is separably valued, if there exists a measurable set X0

such that ϕ(X0 ) is separable and that µ(X \ X0 ) = 0.
The notion of weak measurablity corresponds to the separation of the real-
valued functions into the positive part and the negative part. The notion of
countably valuedness is used to guarantee that this operation can be completed
within countably infinite steps.
The next lemma is useful, when we consider separably valued functions.
Lemma 95. If Y is a separable Banach space, then there exists a countable
subset Z ∗ of the unit ball (Y ∗ )1 of Y ∗ such that kykY = sup |z ∗ (y)| for all
z ∗ ∈Z ∗
y ∈ Y.
Proof Pick a countable dense set Y0 = {yj }∞ j=1 . For each j ∈ N by virtue
of the Hahn–Banach theorem, Theorem 87, we can find a norm attainer zj∗ ∈
(Y ∗ )1 of yj . We have only to set Z ∗ ≡ {zj∗ : j ∈ N}.
Below, given an X -valued function ϕ : X → X , let us write kϕkX ≡
kϕ(·)kX , so that kϕkX is a non-negative function. We characterize the strong
measurablity.
Theorem 96 (Pettis). Let ϕ : X → X be a function. Then ϕ is strongly
measurable, if and only if ϕ is separably valued and weakly measurable.
Proof It is straightforward to prove the “only if ” part and we leave
this for the readers. See Exercise 54. Assume that ϕ is separably valued and
weakly measurable. We may assume by disregarding a set of measure zero in
addition that ϕ(X) itself is separable. Let Y = Span(ϕ(X)) and {yk }∞ k=−∞ be
a countable dense set in Y . Then, by Lemma 95, we obtain z1∗ , z2∗ , . . . , zk∗ , . . . ∈
(Y ∗ )1 such that kykY = sup |zk∗ (y)|. From this we conclude that kϕ − zkX is
j∈N
a measurable function.
Let j ∈ N. Then define Ak,j ≡ {kϕ − yk kX < j −1 } ⊂ X. Set

B1,j ≡ A1,j , Bk,j ≡ Ak,j \ (A1,j ∪ A2,j ∪ · · · ∪ Ak−1,j ), k ≥ 2.

∞
P
Then we have only to set ϕj (x) ≡ χBk,j (x) · yk for x ∈ X . From the
k=1
property of Ak,j , k, j ∈ N, we conclude that ϕj (x) tends to ϕ(x) for µ-almost
every x ∈ X . Hence, ϕ is separably valued.
 Fundamental facts in functional analysis 97

Having clarified the definition of measurability, we turn to the definition

of the Bochner integral.
Now we are going to define integrals for countably simple functions. As we
did for usual measurable functions, it is convenient to start from the definition
of countably simple functions. But we need be careful : It will not work if we
define the countably simple functions as the one taking finitely many values.
What differs from our earlier job is that countably simple functions are allowed
to take countably many values.
Definition 37 (Integrable countably simple functions). Let (X, B, µ) be a
measure space, and let X be a Banach space. A countably simple function ϕ
∞
P
X → X is integrable, if it admits a representation ϕ(x) = χEj (x)bj µ-a.e.
j=1
∞
x ∈ X where {bj }∞ ∞
P
j=1 ⊂ X and {Ej }j=1 ⊂ B satisfy kbj kX · µ(Ej ) < ∞.
j=1
The above representation is called an integrable representation of ϕ and one
∞
P
writes ϕ ' χEj bj .
j=1

Lemma 97. Suppose that ϕ is a countably simple integrableZfunction. Then

there exists a unique element Φ ∈ X , which we will write Φ ≡ ϕ(x)dµ(x) ∈
Z X

X , such that b∗ (Φ) = b∗ (ϕ(x))dµ(x) for all b∗ ∈ X ∗ .

X
Proof We will prove the existence of Φ. Choose an integrable representa-
∞
P ∞
P
tion ϕ ' χEj aj . Then define an element Φ ≡ aj µ(Ej ) ∈ X . Note that
j=1 j=1
the convergence is absolute. Let b∗ ∈ X ∗ . Then
X∞ Z
∗ ∗
b (Φ) = b (aj )µ(Ej ) = b∗ ◦ ϕ(x)dµ(x),
j=1 X

where we have used the Lebesgue convergence theorem for the second equality.
Uniqueness of Φ follows from the Hahn–Banach theorem, Theorem 87.
We generalize the triangle inequality so that it is adapted to our current
setting.
Lemma 98. Let ϕ : X → X be an integrable countably simple function. Then
Z Z
ϕ(x)dµ(x) ≤ kϕkX (x)dµ(x). (2.1)
X X X
∞
P
In particular, if ϕ has an integrable representation ϕ ' aj χEj , then
j=1
Z ∞
X
ϕ(x)dµ(x) ≤ kaj kX µ(Ej ). (2.2)
X X j=1
98 Morrey Spaces

Proof By the Hahn–Banach theorem, Theorem 87, we can choose y ∗ from

the closed unit ball of X ∗ so that
Z Z  Z
∗
ϕ(x)dµ(x) =y ϕ(x)dµ(x) = y ∗ ◦ ϕ(x)dµ(x).
X X X X

By the definition of the dual norm,

Z Z
ϕ(x)dµ(x) ≤ kϕkX (x)dµ(x),
X X X

∞
P
showing (2.1). Since kϕkX (x) ≤ kaj kX χEj (x) for µ-almost every x ∈ X,
j=1
(2.2) follows. Therefore, the proof is now complete.
The next lemma shows that the choice of the integrable representation can
be made not so wasteful.
Lemma 99. Suppose that ϕ : X → X is a countably simple measurable
∞
P
function and ε > 0. Then ϕ has an integrable representation ϕ ' aj χEj
j=1
such that
∞
X Z
kaj kX µ(Ej ) < kϕkX (x)dµ(x) + ε. (2.3)
j=1 X

∞
P
Proof Suppose that ϕ ' bj χFj is an integrable representation. Then
j=1
there exists an increasing sequence of integers N1 < N2 < · · · such that
∞
X ε
kbj kX µ(Fj ) < . (2.4)
2k+1
j=Nk

for each k. Let N0 = M0 = 1. By partitioning FNk−1 , FNk−1 +1 , . . . , FNk , we

can find a collection of disjoint measurable sets EMk−1 , EMk−1 +1 , . . . , EMk and
aMk−1 , aMk−1 +1 , . . . , aMk ∈ X
Mk −1
X k −1
NX
al χEl = bj χFj . (2.5)
l=Mk−1 j=Nk−1

k −1
Then the disjointness of {Ej }M
j=Mk−1 yields

Z k −1
NX Z Mk −1
X
bj χFj (x) dµ(x) = al χEl (x) dµ(x)
X j=Nk−1 X l=Mk−1
X X
Mk −1
X
= kal kX µ(El ).
l=Mk−1
 Fundamental facts in functional analysis 99

We also have
∞
X ∞ Z
X k −1
NX
kal kX µ(El ) = bj χFj (x) dµ(x).
l=1 k=1 X j=Nk−1
X

By the triangle inequality,

∞
X Z 1 −1
NX ∞ X
X ∞
kal kX µ(El ) ≤ bj χFj (x) dµ(x) + kbj kX µ(Fj ).
l=1 X j=1 k=1 j=Nk
X

We rewrite the right-hand side using f :

∞
X Z ∞
X X ∞
∞ X
kal kX µ(El ) ≤ f (x) − bj χFj (x) dµ(x) + kbj kX µ(Fj ).
l=1 X j=N1 k=1 j=Nk
X

Therefore, we obtain
∞
X Z X ∞
∞ X
kal kX µ(El ) ≤ kf (x)kX dµ(x) + 2 kbj kX µ(Fj )
l=1 X k=1 j=Nk
Z
≤ kf (x)kX dµ(x) + ε.
X

The number ε > 0 being arbitrary, this is the desired result.

Keeping our observations above, we pass to the general c ase. We are going
to define the integrable functions. In the R-valued case we have allowed the
integral to take the value ±∞. However, in the Bochner integral we rule out
the possibility for the integral to diverge.
Definition 38 (Bochner integrable functions). A function ϕ : X → X is
Bochner integrable, if there exists a sequence of countably simple functions
{ϕj }∞
j=1 such that
lim ϕj (x) = ϕ(x) (2.6)
j→∞

for µ-almost every x ∈ X and that

Z
lim kϕ(x) − ϕj (x)kX dµ(x) = 0. (2.7)
j→∞ X

If this is the case, define the integral of ϕ on X by:

Z Z Z
ϕ(x)dµ(x) = ϕ(x)dµ(x) ≡ lim ϕj (x)dµ(x). (2.8)
X X j→∞ X

The symbol L1 (X; X ) denotes the set of all Bochner integrable functions.
100 Morrey Spaces
R
Lemma 100. The definition of the integral ϕ(x)dµ(x) makes sense. That
R X
is, in that the limit (2.8) defining ϕ(x)dµ(x) does exist and the element
X
ϕ(x)dµ(x) ∈ X is independent of the choice of {ϕj }∞
R
j=1 .
X

Proof We will show that the limit (2.8) exists. To see this, we observe, if
j, k ≥ J
Z Z Z
ϕj (x)dµ(x) − ϕk (x)dµ(x) ≤ 2 sup kϕ(x) − ϕl (x)kX dµ(x).
X X X l≥J X
Z ∞
Therefore ϕj (x)dµ(x) is a Cauchy sequence in X and the limit there-
X j=1
fore does exist.
We will show that the integral is independent of the admissible representa-
tion of ϕ. Suppose {ψj }∞
j=1 is another sequenceZof countably simple integrable

functions such that lim ψj = ϕ and that lim kϕ(x) − ψj (x)kX dµ(x) = 0.
j→∞ j→∞ X
Then by the triangle inequality we obtain
Z Z
ϕj (x)dµ(x) − ψj (x)dµ(x)
X X X
Z
≤ (kϕ(x) − ϕj (x)kX + kϕ(x) − ψj (x)kX ) dµ(x),
X
Z ∞
which tends to 0 as j → ∞. Thus, it follows that both ϕj (x)dµ(x)
X j=1
Z ∞
and ψj (x)dµ(x) tend to the same limit. Therefore, the integral is
X j=1
independent of the admissible representations of ϕ.
Lemma 101. In addition to the same setting as above, we let ϕ : X → X be
a Bochner integrable function. Then
Z Z
ϕ(x)dµ(x) ≤ kϕkX (x)dµ(x).
X X X

Proof It is just a simple matter to pass to the limit of the countably simple
functions. However, it is worth considering the case when X = C. Writing out
the statement in full with X = C, we can easily imagine what is the heart of
the matter in the proof of Lemma 101.
We characterize Bochner integrable functions.
Theorem 102 (Bochner). The function ϕ : X → X is Bochner integrable, if
and only if ϕ is an integrable countably simple function.
 Fundamental facts in functional analysis 101

Proof Suppose that ϕ is an integrable countably simple function. Then

∞
P j
P
we can take an integrable representation ϕ ' aj χEj . Set ϕj ≡ ak χEk
j=1 k=1
for each j ∈ N. Then each ϕj is an integrable countably simple function. By
the Fatou lemma we have

Z Z ∞
X
kϕ(x) − ϕj (x)kX dµ(x) = ak χEk (x) dµ(x)
X X k=j+1
X
∞
X
≤ kak kX µ(Ek ).
k=j+1

The most right-hand side tending to 0 as k → ∞, so does the left-hand side.

Therefore, ϕ is Bochner integrable.
Suppose that ϕ is Bochner integrable. Then there exists a sequence of
countably simple integrable functions {ϕj }∞
j=1 such that
Z
lim kϕ(x) − ϕj (x)kX dµ(x) = 0.
j→∞ X

A passage to the subsequence, we can assume

Z
kϕ(x) − ϕj (x)kX dµ(x) < 2−j−1 .
X

In this case, by the triangle inequality, we have

Z
3
kϕj+1 (x) − ϕj (x)kX dµ(x) < · 2−j .
X 4

If we invoke Lemma 99, then for each j ∈ N, we can find an integrable repre-
sentation

∞
X ∞
X
ϕj+1 − ϕj ' aj+1,k χEj+1,k with kaj,k kX µ(Ej,k ) < 2−j . (2.9)
k=1 k=1

∞
P
We also have an integrable representation of ϕ1 : ϕ1 ' a1,k χE1,k . Using
k=1
(2.9), we conclude that ϕ has the following integrable representation
∞
X
ϕ' aj,k χEj,k .
j,k=1

Therefore ϕ is an integrable countably simple function as well.

102 Morrey Spaces

2.3.2 Convergence theorems

Having set down the definition of the integrals, we now turn to the con-
vergence theorems such as the dominated convergence theorem in integration
theory.

Theorem 103. Suppose that {ϕj }∞ j=1 is a sequence of strongly integrable

functions such that {ϕj (x)}∞
j=1 is a Cauchy sequence for almost every x ∈ X.
Let ϕ be a weakly measurable function such that ϕ(x) = lim ϕ(x) for µ-almost
j→∞
every x ∈ X. Assume in addition there exists ψ ∈ L1 (µ) ∩ M+ (µ) such that

kϕj (x)kX ≤ ψ(x)

for µ-a.e. x ∈ X. Then, ϕ ∈ L1 (X; X ) and

Z Z
lim ϕj (x)dµ(x) = ϕ(x)dµ(x). (2.10)
j→∞ X X

∞
S
Proof Since Yj is separable, whenever each Yj is separable, ϕ is
j=1
strongly measurable.
We claim that ϕ ∈ L1 (X; X ) and (2.10) holds. Since ϕj ∈ L1 (X; X ), by
Theorem 102 we conclude that ϕj is an integrable countably simple function.
Furthermore, kϕ(x)−ϕj (x)kX ≤ 2ψ(x) for µ-almost everywhere x ∈ X. Thus,
if we invoke Lebesgue’s convergence theorem, we obtain
Z
lim kϕ(x) − ϕj (x)kX dµ(x) = 0.
j→∞ X

As a result ϕ ∈ L1 (X; X ) and (2.10) holds.

We will show that the space L1 (X; X ) is a Banach space.
Theorem 104. Let X be a Banach space. The space L1 (X; X ) is complete,
once we identify the functions with the same integrable representation.
Proof Let {ϕj }∞ 1
j=1 ⊂ L (X; X ) be arbitrary. We have only to prove that,
∞
P ∞
P
if kϕj kL1 (X ;X) converges, then so does ϕj . To do this we pick an inte-
j=1 j=1
grable representation
∞
X ∞
X
ϕj ' aj,k χEj,k with kaj,k kX µ(Ej,k ) < kϕj kL1 (X;X ) + 2−j .
k=1 k=1

∞
P
Then for x ∈ X, we define ψ(x) ≡ aj,k χEj,k (x), provided the series
j,k=1
converges and ψ(x) = 0 otherwise. Then ψ has an integrable representation
 Fundamental facts in functional analysis 103
∞ l
aj,k χEj,k . Therefore ψ ∈ L1 (X; X ). Since ψ −
P P
ψ ' ϕj has an inte-
j,k=1 j=1
l
P ∞
P ∞
P ∞
P
grable representation ψ− ϕj ' aj,k χEj,k , we conclude ψ = ϕj .
j=1 j=l+1 k=1 j=1

Finally let us characterize the Bochner integral in terms of the dual spaces.
Proposition
Z 105. Let ϕ : X → X be a Bochner integrable function. Then
b≡ ϕ(x)dµ(x) is a unique element in X satisfying
X
Z
b∗ (b) = b∗ (ϕ(x))dµ(x)
X
∗ ∗
for all b ∈ X .
Proof Now that “Bochner integrable function” and “integrable countably
simple function” are synonymous, this proposition is Lemma 97 itself.

2.3.3 Fubini’s theorem for Bochner integral

In this subsection to discuss Fubini’s theorem for Bochner integrals, let
(X, M, µ) and (Y, N , ν) be σ-finite measure spaces.
Theorem 106. Suppose that ϕ ∈ L1 (X × Y ; X ). Then we have the following:
(1) • For almost every x ∈ X, the function ϕ(x, ·) belongs to L1 (Y ; X ).
• For almost every y ∈ Y , the function ϕ(·, y) belongs to L1 (X; X ).
R
Below it will be understood tacitly that ϕ(x, y)dν(y) = 0 and that
R Y
ϕ(x, y)dµ(x) = 0, if the integrand is not integrable.
X

(2) We have
ZZ Z Z 
ϕ(x, y)dµ ⊗ ν(x, y) = ϕ(x, y)dν(y) dµ(x)
X×Y X Y
Z Z 
= ϕ(x, y)dµ(y) dν(y).
Y X

Proof
ZZ
(1) Since kϕ(x, y)kX dµ ⊗ ν(x, y) < ∞, assertion (1) follows imme-
X×Y
diately.
∞
P
(2) Pick an integrable representation of ϕ: ϕ ' aj χGj , where each Gj is
j=1
a µ ⊗ ν-measurable set. By disregarding a set of measure zero we may
104 Morrey Spaces
∞
P R
assume kaj kX · Y
χGj (x, y)dν(y) < ∞ for all x ∈ X. Then for all
j=1
Z ∞
X Z
x ∈ X, ϕ(x, y)dµ(y) = aj χGj (x, y)dν(y). Since
Y j=1 Y

∞ Z
X Z ∞
X Z
aj χGj (x, y)dν(y) dµ(y) = kaj kX χGj (x, y)dν(y)
j=1 X Y X j=1 Y

< ∞,
we are in the position of using the Lebesgue convergence theorem (The-
orem 103) for the Bochner integral to obtain
Z Z  X∞ Z
ϕ(x, y)dµ(y) dν(x) = aj χGj (x, y)dµ(x)dν(y)
X Y j=1 X×Y
ZZ
= ϕ(x, y)dµ ⊗ ν(x, y).
X×Y

This is the desired result.

Now we pass our theory of Lebesgue integral to the Lp (X; X )-space.
Definition 39 (Lp (X; X )). Let 1 ≤ p < ∞. The space Lp (X; X ) is defined
as the set of all simple functions f : X → X for which the norm
Z  p1
p
kf kLp (X;X ) ≡ kf kX (x) dµ < ∞.
X
∞
The space L (X; X ) is the set of all simple functions satisfying
kf kL∞ (X;X ) ≡ k kf kX kL∞ (X) < ∞.
Now we will collect some duality results.
Theorem 107. Let X be a separable Banach space. Then the dual space of
L1 (R; X ) is canonically identified with L∞ (R; X ∗ ) in the following manner:
(1) If g ∈ L∞ (R; X ∗ ), then there exists a linear functional Lg such that
Z
Lg (χA · b) = hb, g(x)iX ,X ∗ dx
A

for all b ∈ X and measurable sets A.

(2) Let ϕ ∈ L1 (R; X )∗ . Then there exists g ∈ L∞ (R; X ∗ ) such that Lg = ϕ.
Proof It is not so hard to see that L∞ (R; X ∗ ) defines a continuous func-
tional on L1 (R; X ). Suppose that ϕ ∈ L1 (R : X )∗ . Then define Φ : R → X ∗
by

ϕ(χ[0,t] · b) t ≥ 0,
hb, Ψ(t)iX ,X ≡
∗ (b ∈ X ).
−ϕ(χ[t,0] · b) t ≤ 0
 Fundamental facts in functional analysis 105

For −∞ < s < t < ∞, we note that

hb, Ψ(t)iX ,X ∗ − hb, Ψ(s)iX ,X ∗ = ϕ(χ[s,t] · b).
Hence, denoting X1 by the unit ball of X , we obtain
kΨ(t) − Ψ(s)k∗ = sup |hb, Ψ(t)iX ,X ∗ − hb, Ψ(s)iX ,X ∗ | (2.11)
b∈X1

= sup |ϕ(χ[s,t] · b)|

b∈X1

≤ kϕk∗ · |t − s|.
Let us define
Φj (t) ≡ 2j (Ψ(t + 2−j ) − Ψ(t)) (t ∈ R).
Then
kΦj (t)kX ∗ ≤ kϕk∗ (t ∈ R)
by virtue of (2.11). Therefore, by the Banach Alaoglu theorem, Theorem 89,
there exists a subsequence {Φjk }k∈N such that the weak-∗ limit
Φ(t) ≡ lim Φjk (t)
k→∞

exists for all t ∈ Q. If t, s ∈ Q, then

| hb, Φ(t) − Φ(s)iX ,X ∗ |
kΦ(t) − Φ(s)kX ∗ = sup ≤ kϕk∗ |t − s|.
b∈X \{0} kbkX
Therefore, Φ can be extended to a continuous function from R to X ∗ satisfying
kΦ(t) − Φ(s)kX ∗ ≤ kϕk∗ |t − s|.
This implies
|hb, Φ(t) − Φjk (t)iX ,X ∗ |
≤ |hb, Φ(t) − Φ(s) − Φjk (t) + Φjk (s)iX ,X ∗ | + |hb, Φ(s) − Φjk (s)iX ,X ∗ |
≤ 2kϕk∗ |t − s| + |hb, Φ(s) − Φjk (s)iX ,X ∗ |
for all t ∈ R and s ∈ Q. Thus, we obtain
lim sup kΦ(t)−Φjk (t)kX ∗ = lim sup sup |hb, Φ(t)−Φjk (t)iX ,X ∗ | ≤ 2kϕk∗ |t−s|.
j→∞ j→∞ b∈X1

Since s ∈ Q is arbitrary, this implies lim Φjk (t) = Φ(t) for all t ∈ R. As a
k→∞
result, for each g ∈ L∞ d
c (R ; X ),
Z Z
hΦ(t), g(t)idt = lim hg(t), Φjk (t)iX ,X ∗ dt = ϕ(g).
R k→∞ R

Since L∞ d 1 d
c (R ; X ) is dense in L (R ; X ), we see that ϕ is realized by Φ.

We conclude Section 2.3.3 with another duality result. We recall that

a Banach space X is reflexive if the canonical mapping Q : X → X ∗∗ is
106 Morrey Spaces

surjective; see [310, 1.11.6]. The proof is similar to Theorem 107 and we omit
the proof.
Theorem 108. Let (X, B, µ) be a σ-finite measure space, and let 1 ≤ p < ∞.
0
Suppose that X is a reflexive Banach space. Then Lp (X; X )∗ = Lp (X; X ∗ ).
Let X be a Banach space, and let Lip(R; X ) be the set of all Lipschitz
continuous functions f : R → X for which the quantity kf kLip(R;X ) ≡
sup |t−s|−1 kf (t)−f (s)kX is finite. If X = C, then abbreviate Lip(R; X )
−∞<s<t<∞
to Lip(R).
Lemma 109. Let X be a Banach space.
(1) Let f ∈ L1 (R; X ) ∩ C(R; X ) and g ∈ Lip(R; X ∗ ). Then the limit
∞
X
Lg (f ) ≡ lim hg(N −1 (j + 1)) − g(N −1 j), f (N −1 j)i
N →∞
j=−∞

exists and satisfies |Lg (f )| ≤ kf kL1 (R;X ) kgkLip(R;X ∗ ) . Thus, any g ∈

Lip(R∗ ; X ) induces a bounded linear functional on L1 (R; X ).
(2) Conversely, any bounded linear functional L on L1 (R; X ) is realized as
L = Lg for some g ∈ Lip(R; X ∗ ).
Proof Assertion (1) is straightforward and we omit the proof. For (2), we
define (
L(χ[0,t] x) t ≥ 0,
hg(t), xi ≡
−L(χ[t,0] x) t < 0
for x ∈ X . Then g ∈ Lip(R, X ∗ ). It is not so hard to check L = Lg , since
functions of the form χ[a,b] x with a, b ∈ R and x ∈ X span a dense subspace
in L1 (R; X ).

2.3.4 Exercises
Exercise 52. Let (X, B, µ) be a measure space, and let X be a Banach space.
Suppose that {ϕj }∞ 1
j=1 ∈ L (X; X ) satisfies
∞ Z
X
kϕj (x)kX dµ(x) < ∞.
j=1 X

∞
P
Then show that ϕj (x) converges for µ-almost every x ∈ X. Modifying
j=1
∞ ∞
ϕj ∈ L1 (X; X ) and
P P
ϕj (x) over a set of measure zero, show that
j=1 j=1
 
Z X ∞ ∞ Z
X
 ϕj dµ =
 ϕj (x)dµ(x).
X j=1 j=1 X
 Fundamental facts in functional analysis 107

Exercise 53. Let (X, B, µ) be a measure space. Also let X be a Banach space.

(1) Let 1 ≤ p ≤ ∞. Then show that Lp (X; X ) is a Banach space.

(2) Let E(µ) be a Banach function space. Denote by E(µ; X ) the set of all
strongly measurable functions f : X → X such that kf (·)kX ∈ E(µ).
Equip E(µ; X ) with the norm;

kf kE(µ;X ) ≡ k kf (·)kX kE(µ) (f ∈ E(µ; X )).

Then show that E(µ; X) is a Banach space.

Exercise 54. Let X be a Banach space, and let (X, B, µ) be a measure space.
Suppose ϕ : X → X is a strongly measurable function.
(1) Prove that ϕ is separably valued and weakly measurable, if ϕ is simple.
(2) Prove that ϕ is separably valued and weakly measurable.
Exercise 55. Let X be a Banach space. Show that C 1 ([t0 , t1 ]; X ) is dense in
L1 ([t0 , t1 ]; X ).

2.4 Notes
Section 2.1
General remarks and textbooks in Section 2.1
See the textbooks [241, Chapter 14], [310] and [409, Chapter 5] for func-
tional analysis, for example.

Section 2.1.1
We can consult [310, 409]. We refer to [310, Theorem 1.9.8] and [310,
Proposition 1.9.15] for Theorems 87 and 88, respectively or see [409, Theo-
rem 5.5.5], [409, Theorem 5.5.8] and [409, Theorem 5.5.9] for Hahn–Banach
extensions, Theorems 85, 86 and 87, respectively.
See [239] as well.
See [310, 2.6.18 and 2.6.19] for Theorem 89 for the Banach Alaoglu
theorem.

Section 2.1.2
Many people strengthened the triangle inequality. We employed one of
them in Theorem 90; see [300, Theorem 1].
108 Morrey Spaces

Section 2.1.3
Section 2.1.3 heavily depends on Lions and Peetre [280]. See [280, Chapter
premier] for the definition of sum spaces and intersection spaces. We refer to
[280, 3. I] for Theorem 92.

Section 2.2
General remarks and textbooks in Section 2.2
There are many textbooks on Hilbert spaces. Any one of them will suffice
for this book. Here we content ourselves with the book by Yoshida [465].

Section 2.2.1
Beigböck, Schachermayer and Veliyev considered a variant of Komlós’ the-
orem. See [26, §2.1] for Proposition 93, while the original Komlós theorem is
due to Komlós [244].

Section 2.2.2
Cotlar proved the “so called” Cotlar’s lemma, Theorem 94, in [88].

Section 2.3
General remarks and textbooks in Section 2.3
The Bochner integral is named after Salomon Bochner [38]. We refer to
[81, Appendix E] and [465] for Bochner integrals. See also the lecture note
[223, Chapter 0, §I].

Section 2.3.1
Pettis characterized the strong measurability [357, Theorem 1.1]; see The-
orem 96. We followed Pettis [357, Definition 2.1] when we defined the integral
of functions assuming their values in a Banach space; see Definition 37.

Section 2.3.2
See the paper by Bochner [38, §3] for Theorem 103. The completeness of
the Bochner integrable spaces can be found in [357, Theorem 4.2]; see Theorem
104. Pettis obtained a dense subspace; see [357, Theorem 4.3].

Section 2.3.3
See the paper by Bochner [38, §4] for Theorem 106. Pettis considered the
duality with respect to Bochner integral [357, Theorem 3.3]; see Theorem 107.
Calderón specified the dual space of L1 (R; X ) [52, §32.1]; see Lemma 109.
Chapter 3
Polynomials and harmonic functions

This chapter collects some preliminary facts. The readers may skip Chapter
3. We indicate where we use each theorem in the other parts of this book.
As a preliminary step, in Section 3.1 we collect some fundamental facts
on polynomials. Section 3.2 is the main part of Chapter 3, where we consider
harmonic functions.

3.1 Preliminary facts on polynomials

Here, we collect some facts on polynomials. Although we use them to
investigate Morrey spaces, they are interesting in their own right. We aim
to set up some notation and then consider some fundamental properties in
Section 3.1.1. We consider the convolution estimates in Section 3.1.2, where
the moment conditions come into play. Section 3.1.3 compares the norm of the
derivatives and the Lq -norm. In Section 3.1.4 we consider the approximation
of functions by polynomials.

3.1.1 The space Pk (Rn )

Here, we set up some notation for polynomials used in this book. We will
equip these spaces with the structure of a Hilbert space. We also investigate
the dimension of the space we define here.
Definition 40 (P(Rn ), Pk (Rn )). Define P−1 (Rn ) ≡ {0}. Denote by P =
P(Rn ) the set of all polynomials. Denote by Pk (Rn ), k ∈ N0 , the set of all
polynomial functions with degree less than or equal to k.
∞
By the definition P(Rn ) ≡ Pd (Rn ). Unlike algebra, we regard each
S
d=0
element in P(Rn ) as a function.
We also use the notion of homogeneous polynomials.
Definition 41. Let k ∈ N0 ∪{−1}. The set Ṗk (Rn ) is the set of all polynomials
P such that P (tx) = tk P (x) for all x ∈ Rn and t > 0. The space Ṗk (Rn ) is

109
110 Morrey Spaces

called the homogeneous polynomial space of order k, while Pk (Rn ) is called

the nonhomogeneous polynomial space of order k.
Here and in the sequel, n Cr denotes the combination of choosing r objects
out of n.
Corollary 110. For each k ∈ N0 , one has dimC (Pk (Rn )) = k+n Cn and
dimC (Ṗk (Rn )) = k+n−1 Cn−1 .
Proof We will deal with the nonhomogeneous spaces due to similarity.
We leave the proof for interested readers; see Exercise 56. We calculate

dimC (Pk (Rn )) = ]{(m0 , m1 , . . . , mn ) ∈ N0 n+1 : m0 + m1 + · · · + mn = k}

= k+n Cn .

We will equip Pk (Rn ) with the structure of a Hilbert space.

P Observe that
any polynomial P defines a differential operator P (D) ≡ aα ∂ α , where
|α|≤d
X
P (x) = aα xα .
|α|≤d

Lemma 111. For P, Q ∈X Ṗk (Rn ), P (D)Q is a complex

X number. More pre-
cisely, if we let P (x) ≡ aα xα and Q(x) ≡ bα xα , then P (D)Q =
|α|=k |α|=k
X
α!aα bα .
|α|=k

Proof It suffices to deal with monomials. In this case, the conclusion is

clear.
Consequently, for any P, Q ∈ P(Rn ) we can define an inner product.
Definition 42. One defines an inner product in P by hP, QiP ≡ P (D)Q̄(0).
Here we present an example of taking advantage of the finiteness of
dimC (Pl (Rn )).
Example 52. Let l ∈ZN0 , and let V ⊂ Rn be an open set. Then there exists
ω ∈ Cc∞ (V ) satisfying ω(x)xα dx = δ0α for all α ∈ Nn0 with |α| ≤ l. In fact,
V
equip Pl (Rn ) with the inner product
Z
hP, Qi = P (x)Q(x)Ω(x)dx (P, Q ∈ Pl (Rn )),
Rn

where Ω ∈ (M+ (Rn ) ∩ Cc∞ (Rn )) \ {0} is a fixed function supported on V . Set
U ≡ {P ∈ Pl (Rn ) : P (0) = 0} and
 Z 
U ⊥ ≡ R ∈ L1 (V ) : R(x)P (x)Ω(x)dx = 0, P ∈ U .
Rn
 Polynomials and harmonic functions 111
Z
Then any function R ∈ U ⊥ ∩ Pl (Rn ) \ {0} satisfies R(x)Ω(x)dx 6= 0;
V
n
otherwise R is orthogonal to any element in Pl (R ) with respect to this
inner
Z product, that is, R = 0. Choosing R suitably, we can assume that
R(x)Ω(x)dx = 1. If we set ω ≡ R · Ω, then we have the desired function.
V

3.1.2 Moment inequalities

The moment condition defined in Definition 43 below is an important
condition which will lead us to a nontrivial estimate; the triangle inequality
is beyond our reach. Sometimes we are faced with the situation where the
cancellation
p of functions must be taken into account. For a ∈ Rn , we write
hai ≡ 1 + |a|2 .
Definition 43 (PL⊥ (Rn )). Let L ∈ N0 . TheZset PL⊥ (Rn ) denotes the set of all
f ∈ L0 (Rn ) for which h·iL f ∈ L1 (Rn ) and xα f (x)dx = 0 for all α ∈ N0 n
Rn
with |α| ≤ L. Such a function f satisfies the moment condition of order L. In
this case, one also writes f ⊥ PL (Rn ). If f ⊥ PL (Rn ) for all L ∈ N, one writes
f ⊥ P(Rn ).
Example 53. If a ∈ Cc∞ (Rn ) and α ∈ N0 n with |α| = L ∈ N0 , then ∂ α a ∈
PL−1 (Rn ).
 
Example 54. Let ρ ∈ C ∞ (R) ∩ M+ (R) be supported in − k+1 1 1
, k+1 and
 
R α 1 t
suppose that kρkL1 = 1. Let mα ≡ t ρ(t)dt for α ∈ N0 . Set ρk (t) ≡ ρ
R k k
s+1
for t ∈ R. Observe that t ρk (t)dt = k α mα . If we let ψ ≡
R α P
ck ρk , then we
R k=1
s+1
α α
R P
have t ψ(t)dt = mα ck k . If we choose the ck ’s suitably by using the
R k=1
Vandermonde determinant, we can arrange
Z
tα ψ(t)dt = δαβ (β = 0, 1, . . . , k), supp(ψ) ⊂ (−1, 1).
R

A passage to higher dimension can be done with the tensor product. Thus, it
∞ n ⊥ n
R there exists ψ ∈ C n (R ) ∩ Pk (R ⊥) supported
follows that in a cube (−1, 1)n
n
such that ψ(x)dx = 1 and 2 ψ(2·) − ψ ∈ Pk (R ).
Rn

We start with the inequality without the moment condition. Among the
convolutions of functions, we consider those generated by non-negative func-
tions with polynomial decay. When we handle integral operators, we occasion-
ally want to regard h·i−n−1 and χ[0,1]n as the same functions. Although the
former spreads more than the latter, the effect away from the origin of the
former is not so strong. This idea can be quantified in the following theorem:
112 Morrey Spaces

Theorem 112. Let λ > n. Then for 0 < t ≤ 1,

Z D E−λ
1 y
hx − yi−λ dy ∼λ hxi−λ (x ∈ Rn ) (3.1)
tn Rn t
with constants independent of t.
Proof We decompose the integral on the left-hand side
√ of (3.1) according
to B(t). On B(t), we use Peetre’s inequality hx + yi ≤ 2hxihyi to obtain
1
Z D y E−λ 1
Z
−λ
hx − yi dy ∼ n hx − yi−λ dy
tn B(t) t t B(t)
Z
1
∼ n hxi−λ dy
t B(t)
' hxi−λ .
From this, we obtain estimate (3.1) from below.
Hence it suffices to prove
1
Z D y E−λ
n
hx − yi−λ dy . hxi−λ (3.2)
t Rn \B(t) t

to complete the proof of (3.1). When |x| ≤ 10, we use kf ∗gkL1 ≤ kf kL∞ kgkL1
for f ∈ L∞ (Rn ) and g ∈ L1 (Rn ) and the result follows.
|x|
Hence, it can be assumed that |x| ≥ 10. Write M (x; t) ≡ −1 + log2 and
t
[M (x;t)]
X 1 Z hx − yi−λ dy
I ≡ n
,
j=1
t B(2j t) 2λj
∞
hx − yi−λ dy
Z
X 1
II ≡ .
tn B(2j t) 2λj
j=[1+M (x;t)]

Then
Z D y E−λ ∞ Z D y E−λ
1 −λ
X 1
hx − yi dy = hx − yi−λ dy
tn Rn \B(t) t j=1
t n
B(2 j t)\B(t) t
∞ Z
X 1 1
. n λj
hx − yi−λ dy = I + II.
j=1
t j
B(2 t) 2

|x|
Note that |x| ≥ max(10t, 2j+1 t) as long as 1 ≤ j ≤ M (x; t) = −1 + log2 .
t
j
Thus, |x| ≥ 10t, y ∈ B(2 t) and 1 ≤ j ≤ M (x; t), provided 2|x − y| ≥ |x|.
Consequently,
[M (x;t)] Z D x E−λ 1 [M (x;t)]
X 1 X 1 1
I≤ dy . hxi−λ j(λ−n) . .
j=1
tn B(2j t) 2 2λj
j=1
2 hxiλ
 Polynomials and harmonic functions 113

Meanwhile,
∞ ∞
tλ−n
Z
X 1 dy X 1
II ≤ ∼ . .
tn Rn 2λj hx − yiλ 2λj tn hxiλ
j=[1+M (x;t)] j=[1+M (x;t)]

If we combine I and II, then we obtain the desired inequality.

The following estimates are used many times in this book.
Theorem 113. Let N ∈ N be a constant. Suppose that λ ∈ R satisfies λ >
n + N − 1. Also assume that a ∈ C N (Rn ) satisfies the differential inequality:

|∂ α a(x)| ≤ hxi−λ (x ∈ Rn ) (3.3)

for |α| ≤ N and that η ⊥ PN −1 satisfies the differential inequality:

|η(x)| ≤ hxi−λ (x ∈ Rn ). (3.4)

1 ·
Set η(·; t) ≡ η for 0 < t ≤ 1. Then
tn t
Z Z 2
a(y)η(x − y; t)dy .λ tN hxi−λ rλ−N −n−1 dr.
Rn t

In particular, when λ > N + n,

Z
a(y)η(x − y; t)dy . tN hxi−λ .
Rn

Proof When 81 ≤ t ≤ 1, Theorem 112 covers Theorem 113. Hence, we

assume t ≤ 18 . Change the variables of the integral on the left-hand side and
then subtract the Taylor polynomial of a with degree N − 1 from a;
Z
a(y)η(x − y; t)dy
Rn
Z
= a(x − t y)η(y)dy
Rn
 
Z X 1
= a(x − t y) − ∂ α a(x)(−ty)α  η(y)dy. (3.5)
Rn n
α!
α∈N0 ,|α|≤N −1

We estimate the integral on the most right-hand side of (3.5) by decom-

posing the integral of variable y according to B(t−1 ) and obtaining (3.6) and
(3.7) below.
Let us begin with working on B(t−1 ). Here by assumption (3.3), the
reminder term of the Taylor expansion is estimated by (3.4):

X 1 α
a(x − t y) − ∂ a(x)(−ty)α . |t y|N hxi−λ (3.6)
α!
α∈N0 n ,|α|≤N −1
114 Morrey Spaces

if |y| ≤ t−1 . We write min(a, b) ≡ a ∧ b. Then

t−1
|y|N dy
Z Z Z
≤ |y|N ∧ |y|N −λ dy ' rn+N −1 (1 ∧ r−λ )dr.
B(t−1 ) hyiλ B(t−1 ) 0

Changing the variables: r 7→ r−1 on the integral on the right-hand side yields
Z t−1 Z ∞
rn+N −1 (1 ∧ r−λ )dr = r−n−N −1 (1 ∧ rλ )dr.
0 t

Consequently,
 
Z X 1 α
a(x − t y) − ∂ a(x)(−ty)α  η(y)dy
B(t−1 ) α!
α∈N0 n ,|α|≤N −1
Z 2
dr
. .
t rN +n+1−λ

Assume instead that |y| ≥ t−1 . Then we obtain

X 1 α
a(x − t y) − ∂ a(x)(−ty)α
α!
α∈N0 n ,|α|≤N −1
X 1 α
≤ |a(x − t y)| + ∂ a(x)(−ty)α
α!
α∈N0 n ,|α|≤N −1
−λ
+ |t y|N −1 hxi−λ
. hx − t yi (3.7)
√
by Peetre’s inequality, hx + yi ≤ 2hxihyi and (3.3). Hence
Z
a(y)η(x − y; t)dy
Rn \B(t−1 )
Z
hx − t yi−λ + |t y|N −1 hxi−λ hyi−λ dy.


.
Rn \B(t−1 )

We change the variables y 7→ z = ty:

Z Z
hx − t yi−λ hyi−λ dy ≤ hx − t yi−λ |y|−λ dy
Rn \B(t−1 ) Rn \B(t−1 )
Z
= tλ−n hx − zi−λ |z|−λ dz
Rn \B(t−1 )
Z
. tλ−n hx − zi−λ hzi−λ dz
Rn \B(1)
Z
≤ tλ−n hx − zi−λ hzi−λ dz.
Rn
 Polynomials and harmonic functions 115

From Theorem 112, we obtain

Z
hx − t yi−λ hyi−λ dy . tλ−n hxi−λ .
Rn \B(t−1 )

Recall also that λ > N + n − 1. With this in mind we change the variables:
y 7→ z = ty to obtain
Z Z
N −1−λ λ−N −n+1
|y| dy = t |z|N −1−λ dz ∼ tλ−N −n+1 .
Rn \B(t−1 ) Rn \B(t−1 )

Consequently,
Z Z
|t y|N −1 hyi−λ dy < tN −1 |y|N −1−λ dy
Rn \B(t−1 ) Rn \B(t−1 )

' tλ−n
Z 2
. tN rλ−N −n−1 dr.
t

Hence
 
Z X 1 α
a(x − t y) − ∂ a(x)(−ty)α  η(y)dy
Rn \B(t−1 ) α!
α∈N0 n ,|α|≤N −1
Z 2
. tN hxi−λ rλ−N −n−1 dr.
t

Thus, the proof is complete.

The proof of Theorem 113 used the mean-value theorem. We transform
Theorem 113 so that we can easily handle it later.
Theorem 114. Maintain the same assumption on λ, N > 0 and on a, η ∈
L1 (Rn ) as Theorem 113. For j ≤ ν we write aj ≡ 2jn a(2j ·), η ν ≡ 2νn η(2ν ·).
Z 2
Then |aj ∗ η ν (x)| . 2jn+(j−ν)N h2j xi−λ rλ−N −n−1 dr for x ∈ Rn . In
2j−ν
particular, when λ > N + n, |aj ∗ η ν (x)| . 2jn+(j−ν)N h2j xi−λ .
Proof Since aj ∗ η ν = (a ∗ η ν−j )j , simply use Theorem 113 with t = 2j−ν .
Example 55. Let {ajν }j∈Z,ν∈Zn ∈ `2 (Z × Zn ). Also let ψ ∈ Cc∞ (Rn ). For
jn
∈ Zn , we set ϕjν (x) ≡ 2 2 ∆ψ(2j x − ν), x ∈ Rn . Then we claim
j ∈ Z and ν P
that f ≡ ajν ϕjν converges in L2 (Rn ) and satisfies
j∈Z,ν∈Zn

  12
X
kf kL2 .  |ajν |2  . (3.8)
j∈Z,ν∈Zn
116 Morrey Spaces

In fact, thanks to Theorem 114, we have

Z
ϕjν (x)ψj̃ ν̃ (x)dx
Rn
1
. 2min(j,j̃)n− 2 (j+j̃)n−2|j−j̃| (1 + 2min(j,j̃) |2−j ν − 2−j̃ ν̃|)−2n−2 .

Fix j̃ and ν̃. Then

1
X
2min(j,j̃)n− 2 (j+j̃)n−2|j−j̃| (1 + 2min(j,j̃) |2−j ν − 2−j̃ ν̃|)−2n−2
j∈Z,ν∈Zn
XZ 1
∼ 2min(j,j̃)n− 2 (j+j̃)n−2|j−j̃| (1 + 2min(j,j̃) |2−j x − 2−j̃ ν̃|)−2n−2 dx
j∈Z Rn
1
X
∼ 2 2 (j−j̃)n−2|j−j̃|
j∈Z

. 1.

Thus,
X X Z
|ajν aj̃ ν̃ | ϕjν (x)ψj̃ ν̃ (x)dx
j∈Z,ν∈Zn j̃∈Z,ν̃∈Zn Rn

X X Z
≤2 |aj̃ ν̃ |2 ϕjν (x)ψj̃ ν̃ (x)dx
j∈Z,ν∈Zn j̃∈Z,ν̃∈Zn Rn
X
. |ajν |2 .
j∈Z,ν∈Zn

As a result, the sum converges in L2 (Rn ) and (3.8) holds.

Duality entails
s X
|hf, ϕjν iL2 |2 . kf kL2 . (3.9)
j∈Z,ν∈Zn

Combining (3.8) and (3.9), we conclude that

X
f ∈ L2 (Rn ) 7→ hf, ϕjν iL2 ϕjν
j∈Z,ν∈Zn

is a bounded linear operator.

3.1.3 Control of derivatives by integrals

If E is not too small, then we can control derivatives of polynomials in
Pk (Rn ) by the Lq -norm defined over E. For an open set Ω and a ball B(x0 , ρ),
we write Ω(x0 , ρ) ≡ Ω ∩ B(x0 , ρ).
 Polynomials and harmonic functions 117

Lemma 115. Let k ∈ N0 ∪ {−1}, q ≥ 1 and A ∈ (0, 1). Also let x0 ∈ Ω and
0 < ρ < diam(Ω). If E is a measurable subset in Ω(x0 ; ρ) such that |E| ≥
 Z  q1
β 1 q
A|B(ρ)|, then |∂ P (x0 )| .β,A |P (x)| dx for all P ∈ Pk (Rn )
ρn+|β|q E
and β ∈ N0 n .
It should be noted that if E = B(x0 , ρ), then Lemma 115 is trivial from
the dilation and translation.
Proof By the translation and the dilation we may assume that B(x0 , ρ) =
B(1). The symbol Tk denotes the set of all polynomials in Pk (Rn ) of the form
X
P (x) = aα xα
|α|≤k

with X
|aα |2 = 1. (3.10)
|α|≤k

Let F be the set of all measurable functions f defined on Rn such that 0 ≤

f ≤ χB(1) and that
Z Z
f (x)dx = χB(1) (x)f (x)dx ≥ A.
Rn Rn

Let us define Z
γ(A) ≡ inf |P (x)|q f (x)dx.
P ∈Tk ,f ∈F B(1)
Let us prove that the infimum is attained;
Z
γ(A) = min |P (x)|q f (x)dx.
P ∈Tk ,f ∈F B(1)

By the definition of γ(A), we can find a sequence of polynomials {Pl }∞

l=1 in
Tk and a sequence of measurable function {fl }∞ l=1 in F such that
Z
1
γ(A) ≤ |Pl (x)|q fl (x)dx ≤ γ(A) + .
B(1) l
Recall that we are assuming (3.10). Define N ≡ ]{α ∈ N0 n : |α| ≤ k}. Since
the closed unit ball in RN is compact for all N ∈ N and the unit ball of
L2 (B(1)) is weakly compact, we can suppose that Pl → P ∈ Tk uniformly
over B(1) and that fl → f ∈ F weakly in L2 (B(1)). Thus the pair (P, f ) is
the one we are looking for. Z
Since χE ∈ F, it follows that |P (x)|q dx ≥ γ(A). Also, since all lin-
E
ear operators in a finite dimensional normed space Pk (Rn ) are automati-
  12
X
cally bounded, we have |∂ β P (0)| .  |aα |2  ∼ kP kLq (B(1)) when P
|α|≤k
118 Morrey Spaces

aα xα . Thus, we obtain the desired result when

P
takes the form; P (x) =
|α|≤k
B(x0 , ρ) = B(1). In the general case, we can use a scaling argument as we
mentioned in the beginning.

3.1.4 Best approximation

We define the best approximation, which we use in Chapters 5 and 14. We
will work in an open set Ω in Rn . For x0 ∈ Rn and ρ > 0, write Ω(x0 , ρ) ≡
Ω ∩ B(x0 , ρ). Denote by diam(Ω) its diameter.
Lemma 116. Let 1 < q < ∞ and k ∈ N0 ∪ {−1}. Also let f ∈ Lqloc (Ω).
Then, for all x0 ∈ Ω and ρ ∈ (0, diam(Ω)), we can find a unique polynomial
Pk (·; x0 , ρ, f ) ∈ Pk (Rn ) such that

inf kf − P kLq (Ω(x0 ;ρ)) = kf − Pk (·; x0 , ρ, f )kLq (Ω(x0 ;ρ)) .

P ∈Pk (Rn )

The polynomial Pk (·; x0 , ρ, f ) is called the best approximation of f .

Proof The uniqueness is a consequence of Theorem 90. In fact, if
Pk (·; x0 , ρ, f ) and Qk (·; x0 , ρ, f ) in Pk (Rn ) attains the infimum, then we have

kf − Pk (·; x0 , ρ, f ) + f − Qk (·; x0 , ρ, f )kLq (Ω(x0 ,ρ))

= kf − Pk (·; x0 , ρ, f )kLq (Ω(x0 ,ρ)) + kf − Qk (·; x0 , ρ, f )kLq (Ω(x0 ,ρ)) .

This means that f −Pk (·; x0 , ρ, f ) and f −Qk (·; x0 , ρ, f ) are linearly dependent
thanks to Theorem 90 and the constant K satisfying f −Pk (·; x0 , ρ, f ) = K(f −
Qk (·; x0 , ρ, f )) is positive. Since f − Pk (·; x0 , ρ, f ) and f − Qk (·; x0 , ρ, f ) have
the same Lq (Ω(x0 , ρ))-norm, we see that K = 1, implying that Pk (·; x0 , ρ, f )
is unique.
We prove the existence. Set
X aα
P (x) ≡ (x − x0 )α (x ∈ Rn ).
α!
|α|≤k

Observe that   21
X
 |aα |2  ∼ kP kLq (x0 ;ρ)
|α|≤k

again thanks to the fact that the norms in a finite dimensional normed space
Pk (Rn ) are equivalent. Thus, when a sequence {Pl }∞ n
l=0 ∈ Pk (R ) satisfies

Z ! q1
q
inf kf − P kLq (Ω(x0 ;ρ)) = lim |f (x) − Pl (x)| dx ,
P ∈Pk (Rn ) l→∞ Ω(x0 ;ρ)
 Polynomials and harmonic functions 119

|al,α |2 < ∞, where

P
then sup
l∈N |α|≤k

X
Pl (x) = al,α xα (x ∈ Rn ).
|α|≤k

Consequently, we can resort to the compactness argument.

Example 56. Let f ∈ Lqloc (Rn ) and let 1 ≤ q < ∞ and k ∈ N0 . Also let
(x0 , ρ) ∈ Rn+1
+ .

(1) If q = 1, we can still show that such a polynomial exists by going

through a similar argument. However, we do not have the uniqueness of
Pk (·; x0 , ρ, f ). In fact, let k = 0. Let f : [0, 2] → [0, ∞) be a measurable
function. Then any number between f ∗ (1), f ∗ (1 + 0) can be chosen as
P.
(2) Let q = 2 and k = 0, and let f be a square integrable function defined
on a cube Q. Then Pk (·; x0 , ρ, f )(x) = mQ (u) for x ∈ Q.
(3) In general, for f ∈ L1loc (Rn ) and P ∈ Pk (Rn ), we have Pk (·; x0 , ρ, f +
P ) = Pk (·; x0 , ρ, f ) + P .
Before we go further, a helpful remark may be in order.
Remark 3. Let 1 < q < ∞ and k ∈ N0 ∪ {−1}. Also let f ∈
Lqloc (Ω), x0 ∈ Ω and ρ ∈ (0, diam(Ω)). For any polynomial P ∈ Pk (Rn ),
kPk (·; x0 , ρ, f )kLq (Ω(x0 ;ρ)) . kf kLq (Ω(x0 ;ρ)) .
There is another technique of approximating functions. Let k ∈ N0 be
1 n
R ball. For all fR ∈ kLloc (R ) there uniquely exists
fixed and let B be a fixed
PB f ∈ Pk (R ) such that f (x)g(x)dx = PB f (x)g(x)dx for all g ∈ Pk (Rn )
k n
B B
since {xα }|α|≤k is independent in L2 (B). The polynomial PBk f is called the
Gram–Schmidt polynomial of order k for B. Remark that we can also define
PQk f for cubes Q in a similar manner.
Proposition 117. Let B = B(x0 , r) be a ball. Also let k ∈ N.
 
· − x0
(1) For all f ∈ L1loc (Rn ), PBk f = r−n PB(1)
k
f .
r

(2) If f ∈ Pk (Rn ), then PBk f = f .

Z
2n
(3) There exists kB ∈ P(R ) such that PBk f (x) = kB (x, y)f (y)dy and
Rn
−1 n
that |kB (x, y)| . |B| for all x, y ∈ R and f ∈ Lloc (Rn ).
1

Proof The proof is routine; see Exercise 57.

120 Morrey Spaces

3.1.5 Exercises
Exercise 56. Prove dimC (Ṗk (Rn )) = k+n−1 Cn−1 by considering the number
of the integer solutions of x1 + x2 + · · · + xn = k.
Exercise 57. Prove Proposition 117 as follows:

(1) Use the change of variables to prove Proposition 117(1).

(2) Use (1) to prove Proposition 117(2).
2
(3) Use the Riesz representation theorem
Z for L (B) to prove that there exists
kB ∈ P(R2n ) such that PBk f (x) = kB (x, y)f (y)dy and f ∈ L1loc (Rn ).
Rn

(4) Use (1) to show that |kB (x, y)| . |B|−1 for all x, y ∈ Rn .

3.2 Spherical harmonic functions

When we consider the application of Morrey spaces to partial differential
equations, we need the expansion by way of the spherical harmonic functions.
This section is a preliminary section for applications to PDE.
Section 3.2.1 deals with harmonic polynomials. Section 3.2.2 considers the
norm estimates. We obtain an integration by parts formula in Section 3.2.3.

3.2.1 The spaces Hk (Rn ) and Ḣk (Rn )

Here, we are interested in polynomials such that ∆P (x) = 0. Such polyno-
mials are called harmonic polynomials. We are interested in the structure of
the polynomial linear spaces. Among others, we are interested in the dimen-
sion of the linear spaces of harmonic polynomials. For a start, we investigate
the image of ∆ : Pk (Rn ) → Pk−2 (Rn ). We can locate Section 3.2.1 as a setup
for the discussion in Section 3.2.2.
Definition 44. Let k ∈ N0 ∪ {−1}. The space Hk (Rn ) stands for the set of
∞
[
all harmonic polynomials in Ṗk (Rn ). Also define H(Rn ) ≡ Hk (Rn ). The
k=0
space Ḣk (Rn ) is called the homogeneous harmonic polynomial space of order
k, while Hk (Rn ) is called the nonhomogeneous harmonic polynomial space of
order k.
Algebraically, we can say that Hk (Rn ) = ker(∆ : Pk (Rn ) → Pk−2 (Rn )),
Ḣk (Rn ) = ker(∆ : Ṗk (Rn ) → Ṗk−2 (Rn )) and H(Rn ) = ker(∆ : P(Rn ) →
P(Rn )).
 Polynomials and harmonic functions 121

Example 57. We have Ḣ0 (Rn ) = Ṗ0 (Rn ), H1 (Rn ) = P1 (Rn ) and Ḣ1 (Rn ) =
Ṗ1 (Rn ).
We are interested in the structure of these polynomial linear spaces. For a
start, we investigate the image of ∆ : Pk (Rn ) → Pk−2 (Rn ).
Lemma 118. The Laplacian ∆ maps Pk (Rn ) onto Pk−2 (Rn ) for all k ∈ N.
Proof Since the polynomials of the form (a1 x1 + a2 x2 + · · · + an xn )k ,
a1 , a2 , . . . , an ∈ R span Pk (Rn ), we have only to show that such a polynomial
is realized as the image of ∆ : Pk (Rn ) → Pk−2 (Rn ). Since

∆(a1 x1 + a2 x2 + · · · + an xn )k+2
= (k + 2)(k + 1)(a1 2 + · · · + an 2 )(a1 x1 + a2 x2 + · · · + an xn )k ,

this is clear.
Since Pk (Rn ) is a finite dimensional linear space, so are Hk (Rn ) and
Ḣk (Rn ). In higher dimensions, it is difficult to take a basis. Consequently,
we do not go in this direction here. However, we will need to know how large
the spaces Hk (Rn ) and Ḣk (Rn ) are.
Corollary 119. Let k ∈ N0 . Then dimC (Hk (Rn )) = k+n Cn − k+n−2 Cn =
O(k n−1 ) and dimC (Ḣk (Rn )) = k+n−1 Cn−1 − k+n−3 Cn−1 = O(k n−2 ).
Proof We will deal with the nonhomogeneous spaces; the homogeneous
spaces can be handled similarly. See Exercise 59. Using the definition of
Hk (Rn ) and Lemma 118, we calculate

dimC (Hk (Rn )) = dim ker(∆ : Pk (Rn ) → Pk−1 (Rn ))

= dimC (Pk (Rn )) − dimC (Pk−2 (Rn ))
= k+n Cn − k+n−2 Cn .

Along this inner product, we decompose Pk (Rn ).

Proposition 120. We define a linear mapping M : P(Rn ) → P(Rn ) by
M P (x) ≡ |x|2 P (x) for P ∈ P(Rn ). Then for all k ≥ 2, we have Pk (Rn ) =
Hk (Rn ) ⊕ Im(M : Pk−2 (Rn ) → Pk (Rn )) and Ṗk (Rn ) = Ḣk (Rn ) ⊕ Im(M :
Ṗk−2 (Rn ) → Ṗk (Rn )).
Proof We consider the nonhomogeneous case; we will deal with the first
equality since the second one follows immediately from the first one; see Exer-
cise 59. First, let us check that Hk (Rn ) and Im(M : Pk−2 (Rn ) → Pk (Rn )) are
orthogonal. Indeed, for all P ∈ Pk (Rn ) and Q ∈ Pk−2 (Rn ),

hP, M QiP = hM Q, P iP = Q(D)[∆P ] = Q(D)0 = 0.

Consequently, Ḣk (Rn ) and Im(M : Pk−2 (Rn ) → Pk (Rn )) are orthogonal.
To check that Ḣk (Rn ) and Im(M : Pk−2 (Rn ) → Pk (Rn )) span Pk (Rn ),
we can compare the dimension of these spaces using Corollaries 110 and 119.
122 Morrey Spaces

3.2.2 Norm estimates for spherical harmonics

A spherical harmonic is a function obtained by restricting harmonic func-
tions to the unit sphere. Here and below we write S n−1 ≡ {x ∈ Rn : |x| = 1}.
Here, we summarize some useful estimates for spherical harmonics. The esti-
mates will be used to show that some integral operators are Lp (Rn )-bounded.
In addition, write ωn−1 ≡ |S n−1 |.
Definition 45. Equip H(Rn ) with an inner product given by
Z
hP, QiH = P (x)Q(x)dσ(x),
S n−1

where dσ denotes the surface measure. Fix a complete orthonormal system

dim (Ḣ (Rn ))
{Yjk }j=1C k .

Example 58. In R2 , if we set P (x, y) ≡ x2 − y 2 , then

Z Z 2π
2 2 2
hP, P iH = (x − y ) dσ(x, y) = cos2 2θdθ = π.
x2 +y 2 =1 0

Meanwhile, according to Definition 42,

 2
∂2

∂
hP, P iP = P (D)P (0) = − 2 (x2 − y 2 ) = 4.
∂x2 ∂y

Example 59. We work in R2 again. Let f (x, y) ≡ (x + iy)m for m ∈ N Then

Z
hf, f iH = (x2 + y 2 )2m dσ(x, y) = 2π,
S1

while
 m
∂ ∂
hf, f iP = +i (x − iy)m = 2m m!.
∂x ∂y

Concerning this new inner product, we have a sort of the Pythagorean

equality.
Proposition 121. Let k ∈ N0 . Then

dimC (Ḣk (Rn ))

X dimC (Ḣk (Rn )) 2k
|Yjk (x)|2 = |x|
j=1
ωn−1

for all x ∈ Rn .
In the proof, we use the orthogonal group of degree n; O(n) denotes the
set of all orthogonal matrices.
 Polynomials and harmonic functions 123

Proof Let z ∈ Rn be fixed. We will evaluate both sides at x = z. Thanks

to homogeneity, we can assume that z ∈ S n−1 .
If we set
dimC (Ḣk (Rn ))
X
Zz ≡ Yjk (z)Yjk ∈ Ḣk (Rn ), (3.1)
j=1

which is called zonal harmonic of degree k with pole at x, then Y (z) = hY, Zz iH
for all Y ∈ Ḣk . By the uniqueness of the Riesz representation theorem for
Hilbert spaces, we see that

dimC (Ḣk (Rn )) dimC (Ḣk (Rn ))

X X
Yjk (z)Yjk (x) = Yjk (Az)Yjk (Ax)
j=1 j=1

Ḣk (Rn ))
dimC (P
for all A ∈ O(n). Thus, |Yjk (z)|2 is a constant. To determine the
j=1
precise value of this constant, we integrate over S n−1 and use the fact that
dim (Ḣ (Rn ))
{Yjk }j=1C k is an orthonormal system.
dim (Ḣk (Rn ))
The family {Yjk }j=1C has the following growth in L∞ (S n−1 )-norm.
s
dimC (Ḣk (Rn )) n
Corollary 122. |Yjk (x)| ≤ = O(k 2 −1 ) for all x ∈ S n−1 ,
ωn−1
k ∈ N0 and j = 1, 2, . . . , dimC (Ḣk (Rn )).
Proof Simply use Proposition 121:

dimC (Ḣk (Rn ))

2
X dimC (Ḣk (Rn )) 2k dimC (Ḣk (Rn ))
|Yjk (x)| ≤ |Ylk (x)|2 = |x| = .
ωn−1 ωn−1
l=1

We investigate the relation between the L2 -norm of functions in P ∈

Ḣk (Rn ) and that for their gradient.

Lemma 123. Let P ∈ Ḣk (Rn ). Then

Z Z
|grad(P )(x)|2 dx = k |P (x)|2 dσ(x).
B(1) S n−1

In particular,
Z Z
k(n + 2k − 2) |P (x)|2 dσ(x) = |grad(P )(x)|2 dσ(x) (3.2)
S n−1 S n−1

for k ≥ 1.
124 Morrey Spaces

Proof Since P is harmonic, we obtain

|grad(P )(x)|2 = |grad(P )(x)|2 + P (x)∆P (x) = div(P (x)grad(P )(x)).

By the Stokes theorem, we have

Z Z
|grad(P )(x)|2 dx = div(P (x)grad(P )(x))dx
B(1) B(1)
Z
x
= (P (x)grad(P )(x)) · dσ(x).
S n−1 |x|

Since grad(P )(x) · x = kP (x), we obtain the desired result.

When P ∈ Ḣk (Rn ), grad(P ) has the following non-trivial bound:
Corollary 124. Let P ∈ Ḣk (Rn ). Then
s
k(n + 2k − 2) dimC (Ḣk (Rn ))
|grad(P )(x)| ≤ kP kL2 (S n−1 ) |x|k−1 (x ∈ Rn ).
ωn−1

Ḣk (Rn ))
dimC (P
Proof Let x ∈ Rn be fixed. Since P = hP, Yjk iH Yjk , we obtain
j=1
Ḣk (Rn ))
dimC (P
grad(P )(x) = hP, Yjk iH grad(Yjk )(x). Notice that there exists a
j=1
constant A = Ak > 0 such that
dimC (Ḣk (Rn ))
X
|grad(Yjk )(x)|2 = A|x|2k−2 . (3.3)
j=1

Ḣk (Rn ))
dimC (P Ḣk (Rn ))
dimC (P
Indeed, |grad(Yjk )(M x)|2 = |grad(Yjk )(x)|2 for all
j=1 j=1
M ∈ O(n). To calculate the precise value of A, we integrate (3.3) over S n−1 :

Z dimC (Ḣk (Rn ))

1 X
A= |grad(Yjk )(x)|2 dx.
ωn−1 S n−1 j=1

By (3.2), we obtain

dimC (Ḣk (Rn ))

k(n + 2k − 2)
Z X
A= |Yjk (x)|2 dx
ωn−1 S n−1 j=1
k(n + 2k − 2)
= dimC (Ḣk (Rn )).
ωn−1
 Polynomials and harmonic functions 125

Thus, by Hölder’s inequality and the Bessel inequality, we obtain

v
u
Ḣk (Rn ))
udimC (X
|grad(Yjk )(x)|2
u
|grad(P )(x)| ≤ kP kL2 (S n−1 ) t
j=1
s
k(n + 2k − 2) dimC (Ḣk (Rn ))
≤ kP kL2 (S n−1 ) |x|k−1 .
ωn−1

Consequently, we obtain the desired result.

3.2.3 Laplacian and integration by parts formula

We consider some equalities for homogeneous harmonic functions. Propo-
sition 125 can be located as the cancellation condition which the functions in
Ḣk (Rn ) have.

Proposition 125. Let k ∈ N0 . Then ∆ P|x|(x)

k = k(2−k−n)
|x|k+2
P (x) for all P ∈
n
Ḣk (R ).
n
P
Proof We remark that ∂j P (x)xj = kP (x) by Euler’s formula. (See
j=1
Exercise 60.) Hence
n
P (x) 1 X xj
∆ = P (x)∆ k − 2k ∂j P (x) · k+2
|x|k |x| j=1
|x|
n n
k(k + 2) X k X xj
= P (x) k+2
− P (x) k+2
− 2k ∂j P (x) · k+2
|x| j=1
|x| j=1
|x|
P (x)
= k(2 − k − n) .
|x|k+2

Thus, we obtain the desired result.

Example 60. We work in R2 ; n = 2 (and k = 4). We calcluate

x4 + y 4 − 6x2 y 2 x4 + y 4 − 6x2 y 2
∆ 2 2 2
= −16 .
(x + y ) (x2 + y 2 )3

Recall that a function f defined on Rn \ {0} is of homogeneous degree 0,

if f (tx) = f (x) for all t > 0 and x ∈ Rn \ {0}. We will need Proposition 126
below to prove Lemma 214.
n
Proposition 126. Let Z f, g : R \ {0} → R be aZ smooth function of homoge-
neous degree 0. Then f (x)∆g(x)dσ(x) = g(x)∆f (x)dσ(x).
S n−1 S n−1
126 Morrey Spaces
∂
Proof Denote by the normal derivative of the smooth surface. Since
∂n Z  
∂f ∂g ∂g ∂f
= = 0, and hence f (x) (x) − g(x) (x) dσ(x) = 0,
∂n ∂n ∂(∆(2)\∆(1)) ∂n ∂n
by the Stokes theorem, we have
Z Z
f (x)∆g(x)dx = g(x)∆f (x)dx.
B(2)\B(1) B(2)\B(1)

If we write this by the spherical coordinate, then we obtain the desired result.

3.2.4 Exercises
Exercise 58. Show that P1 (Rn ) = Ḣ0 (Rn ) ⊕ Ḣ1 (Rn ). Hint: Can we write
these spaces explicitly ?
Exercise 59. Let k, n ∈ N.
(1) Fix n ∈ N. Prove dimC (Ḣk (Rn )) = k+n−1 Cn−1 − k+n−3 Cn−1 = O(k n−2 )
as k → ∞ by reexaming the proof of Corollary 119.
(2) Prove Ṗk (Rn ) = Ḣk (Rn ) ⊕ Im(M : Ṗk−2 (Rn ) → Ṗk (Rn )) from the
decomposition Pk (Rn ) = Hk (Rn )⊕Im(M : Pk−2 (Rn ) → Pk (Rn )). Hint:
How do M in Proposition 120 and ∆ act on Pk (Rn ) ?
Exercise 60 (Euler’s formula). Let α ∈ R and f ∈ C ∞ (Rn \ {0}) be a
function such that f (tx1 , tx2 , . . . , txn ) = tα f (x1 , x2 , . . . , xn ) for all t > 0 and
(x1 , x2 , . . . , xn ) ∈ Rn \ {0}. Then, by differentiating the above equation in t
n
xj fxj (x) = αf (x) for all x = (x1 , x2 , . . . , xn ) ∈ Rn \ {0}.
P
show that
j=1

3.3 Notes
Section 3.1
General remarks and textbooks in Section 3.1
See [85] for example.

Section 3.1.1
See [85, Theorem 3.8] for the expression of P (D)Q, Lemma 111. Janson,
Taibleson and Weiss used the Gram–Schmidt polynomial. See [220, p.111] for
Proposition 117, for example.

Section 3.1.2
See [129, Lemma 3.3].
 Polynomials and harmonic functions 127

Section 3.1.3
We can find some quantitative information on the relation between the
Lq (Rn )-norm and the L∞ (Rn )-norm of the derivative; see [59, Lemma 2.I] for
Lemma 115.

Section 3.1.4
We followed [59, (3.1)] to find the best approximation; see Lemma 116.

Section 3.2
General remarks and textbooks in Section 3.2
The properties and uses of spherical harmonics can be found in the books
by [415, 417] and [410, §3.5].

Section 3.2.1
Corollary 119, which calculates dimC (Hk (Rn )), can be found in [127,
(2.55), (2.56)]. Proposition 120, which decomposes Pk (Rn ) can be found in
[127, (2.49)].

Section 3.2.2
Proposition 121 can be found in [127, (2.57)].

Section 3.2.3
The functional equation in Proposition 125 can be found in [127, (2.62)].
The self-adjointness of the spherical Laplacian can be found in [76, Lemma
2.9]; see Proposition 126.
Chapter 4
Various operators in Lebesgue spaces

One of the primary aims in this book is the boundedness properties of the
integral operators. Here, we collect the properties of the linear operators acting
on Lebesgue spaces. The boundedness obtained here will be used for analysis
of operators acting on Morrey spaces. We take up the following operators.
(1) Hardy–Littlewood maximal operators, including the related maximal
operators (Sections 4.1 and 4.2).
(2) Fractional maximal operators (Section 4.3).
(3) Singular integral operators (Section 4.4).
(4) Fractional integral operators (Section 4.5).
The first three operators are used for the purpose of controlling the remaining
two operators and these two operators are mainly used to express the solution
of the differential equations.

4.1 Maximal operators

Here, we present fundamental results for the Hardy–Littlewood maximal
operator.
Section 4.1.1 defines the Hardy–Littlewood maximal operator. We give
some examples of calculations and address some important problems concern-
ing its boundedness. In Section 4.1.2 we consider fundamental results of the
boundedness of the Hardy–Littlewood maximal operator (Theorems 134 and
so on). We develop what we did in Section 4.1.2; Section 4.1.3 will investigate
the local counterpart of the results in Section 4.1.2. Section 4.1.4 deals with
the Fefferman–Stein vector-valued inequality (Theorem 145). In Section 4.1.5
we apply what we considered in Section 1.5.3 to the definition of the maximal
operator. We will obtain a new maximal operator called the Orlicz-maximal
operator. As an application of the Orlicz-maximal operator defined in Section
4.1.5, we consider the composition of the maximal operators in Section 4.1.6.
We investigate the boundedness property of the Orlicz maximal operator in
Section 4.1.7. As it turns out, the Hardy–Littlewood maximal operator can

129
130 Morrey Spaces

be regarded as the supremum of the convolution of some classes of functions,

and we collect some convolution estimates in Section 4.1.8.

4.1.1 Hardy–Littlewood maximal operator

One of the fundamental tools in the theory of function spaces is the Hardy–
Littlewood maximal operator. Here, we define the Hardy–Littlewood maximal
operator. We will have a look at the history and some applications here and
then do some calculations.
The Hardy–Littlewood maximal operator appears in many mathematical
contexts. Here, let us see three examples after giving definitions.
Definition 46 ((classical) Hardy–Littlewood maximal operator). For
f ∈ L0 (RZn ), one defines a function M f by M B f (x) = M f (x) ≡
χB (x)
sup |f (y)|dy for x ∈ Rn . The mapping M : f 7→ M f is called the
B∈B |B| B
(ball based) (uncentered) (Hardy–Littlewood) maximal operator or the maxi-
mal function for short.
A sublinear operator T : L∞ n 0 n
c (R ) → L (R ) is by definition an opera-
tor with the property |T (f + g)(x)| ≤ |T f (x)| + |T g(x)| and |T (af )(x)| =
|a||T f (x)| for all f, g ∈ L∞ n
c (R ) and a ∈ C. In this sense, the Hardy–
Littlewood maximal operator is sublinear. Before we show the calculation,
we verify that M commutes with dilation; we have the following scaling law,
whose proof is straightforward.
Lemma 127. For any f ∈ L0 (µ) and r > 0 M [f (r·)] = M f (r·).
Proof Simply use the change of variables x = ry.
Denote by Q the set of all cubes whose edges are parallel to coordinate
axes.
Remark 4. One may replace B with Q to obtain M Q f (= M f ), which one
occasionally does; one can define M Q f = M f ≡ sup mQ (|f |)χQ . However for
Q∈Q
the time being one will study the Hardy–Littlewood maximal operator defined
by B unless otherwise stated.
Let f ∈ L0 (Rn ) and λ ∈ R. According to Definition 46 of M f , we note
that {M f > λ} is an open set. Hence, M f ∈ L0 (Rn ). Let us explain why the
Hardy–Littlewood maximal operator is important.
Example 61. The first example is closely related to PDE. Let Iα be the
fractional integral operator or the Riesz potential given by
Z
f (y)
Iα f (x) ≡ n−α
dy (x ∈ Rn ) (4.1)
Rn |x − y|
 Various operators in Lebesgue spaces 131
n 1 1 α
for f ∈ M+ (Rn ). Let 1 ≤ p < and define q so that = − . Due to its
α q p n
p p
importance in this book, we will show Iα f (x) . M f (x) q kf kLp 1− q for x ∈ Rn
in Lemma 181.
Here, we work on R for the time being. We find M χ(−R,R) (t) for t ∈ R,
R > 0 for the uncentered maximal operator M .
Proposition 128. We have M χ(−R,R) (t) = min 1, 2R(R +
|t|)−1 , t ∈ R,

for all R > 0. In particular, M χ[−1,1] (t) = min 1, 2(1 + |t|)−1 for t ∈ R.
Proof Since we are considering the average of the function of χ(−R,R) , we
have M χ(−R,R) (t) ≤ 1 for all t ∈ R. Let |t| < R. Then
Z t+r Z t+r
1 1
M χ(−R,R) (t) ≥ lim χ(−R,R) (y)dy = lim dy = 1.
r↓0 2r t−r r↓0 2r t−r

|(s, t) ∩ (−R, R)| 2R

For t ≥ R, we use max = . For t ≤ −R, we use
s∈(−R,R) |(s, t)| R +t
symmetry.
Z ∞
1
Thus, M χ[−1,1] is not an L (R)-function: M χ[−1,1] (t)dt = ∞. Mim-
−∞
/ L1 (Rn ) when f ∈ L0 (Rn ) \
icking this argument, we can establish that M f ∈
{0}.
We generalize Proposition 128 as follows:
Example 62. Let f ∈ M+ (R) be such that f (t) = 0 if t ≤ 0 and that
1 t
Z
f (t) ≥ f (s) if t > s > 0. Then M f (t) = f (s)ds for all t > 0.
t 0
Although we could find M χ[−1,1] (t), t ∈ R in Proposition 128, it is in
general hard to find M f (x), x ∈ Rn for f ∈ L0 (Rn ).
Next, we pass Proposition 128 to higher dimensions to have the following
useful expression.
|R|
Proposition 129. For R ∈ Q and x ∈ Rn , M Q χR (x) ∼n ,
|R| + |x − c(R)|n
where the implicit constant
 does not
n depend only on n. In particular, for all
r
x ∈ Rn , M B χB(r) (x) ∼n .
|x| + r
Proof The proof is based on a geometric observation. We content our-
selves with its outline. We will distinguish two cases.

(1) Let x ∈ 3R. In this case, we can show that

1 |x − c(R)|
≤ M Q χR ≤ 1, 1≤1+ ≤ 1 + 3n.
3n `(R)
132 Morrey Spaces

(2) Let x ∈ 3l+1 R \ 3l R for some l ∈ N. In this case, we can show that

1 4n 3l |x − c(R)|
≤ M Q χR (x) ≤ , ≤1+ ≤ 1 + n · 3l+1 .
3(l+1)n 3ln 2 `(R)

Since M is not bounded on L1 (Rn ), we need something more to describe

the boundedness of M near L1 (Rn ). Orlicz spaces are used for this purpose;
see Chapter 13.
The following crude estimate is useful on many occasions.
Example 63. Let Q ∈ Q, and let E be a measurable set containing x ∈ Rn
|E|
and contained by Q. Observe that M Q χE ≥ on Q. Indeed, for x ∈ Q,
|Q|
 
Q |E ∩ S| |E|
M χE (x) = sup : S satisfies x ∈ S ∈ Q ≥ .
|S| |Q|

We can readily replace cubes by balls.

The following technique is used to localize the problem: M Q [χRn \5Q f ](y),
y ∈ Q has a standard estimate while M Q [χ5Q f ](y), y ∈ Q must be handled
in a more careful manner. Let Q ∈ Q. The symbol Q] (Q) stands for the set
of all cubes containing Q.
Lemma 130 (Nonlocal/global estimate of the maximal operator). For all
f ∈ L0 (Rn ) and cubes Q, we have
∞
X
M [χRn \5Q f ](x) ∼ sup mR (χRn \5Q |f |) . m2k+1 Q (|f |) (x ∈ Q).
R∈Q] (Q) k=1

Remark that cubes can be replaced by balls and that 5Q can be readily
replaced by κQ for any κ > 1 once the implicit constant can be allowed to
depend on κ. The right inequality is clear but we frequently use it. However,
it will turn out that we need to pay attention to the difference between the
summation and the supremum.
Proof The right inequality is clear, so we prove the left equivalence. One
inequality is clear from the definition of M . We write out M [χRn \5Q f ](x) in
full: Z
χR (x)
M [χRn \5Q f ](x) = sup |f (y)|dy.
R∈Q |R| R\5Q
Z
In order that χR (x) |f (y)|dy be not zero, we need to have x ∈ R and
R\5Q
R \ 5Q 6= ∅. Thus, R meets both Q and Rn \ 5Q. If R ∈ Q is a cube that meets
both Q and Rn \ 5Q, then `(R) ≥ 2`(Q) and 2R ⊃ Q. Thus, the desired result
follows.
 Various operators in Lebesgue spaces 133

The next theorem is an example of controls of linear operators by means

of the Hardy–Littlewood maximal operator. A radial and decreasing function
means a function Φ such that there exists ϕ ∈ M↓ (0, ∞) such that Φ(x) =
ϕ(|x|) for all x ∈ Rn .
Theorem 131. Let Φ : Rn → [0, ∞) be an integrable, continuous and radial
decreasing function. Then |Φ∗f (x)| ≤ kΦkL1 M f (x) holds for all f ∈ L1loc (Rn )
and x ∈ Rn .
Proof We may assume that f ∈ M+ (Rn ) by the triangle inequality for
integrals. Let ϕ : [0, ∞) → [0, ∞) be a function taking the form ϕ(r) =
k
P
aj χ[0,rj ] (r) for r ≥ 0. Then
j=1

k Z
X 1
|Φ ∗ f (x)| ≤ aj |B(x, rj )| |f (y)|dy ≤ kΦkL1 M f (x).
j=1
|B(x, rj )| B(x,rj )

A passage to the limit, makes this inequality valid for any integrable, contin-
uous and radial decreasing function Φ.
Example 64. We list some examples of functions to which Theorem 131 is
applicable.
(1) Φ = χB(1) .

|x|2
 
1
(2) Gaussian: Φ(x) = p exp − for x ∈ Rn .
(4π)n 4

(3) Let j ∈ N0 and m > n. Define ηj,m (x) ≡ 2jn (1 + 2j |x|)−m for x ∈ Rn .
The function ηj,m is called the η-function.

4.1.2 Hardy–Littlewood maximal inequality

In view of Proposition 128, we see that M f never belongs to L1 (Rn ) even
when f ∈ L1 (Rn ). What can be said for L1 (Rn ) functions? We will answer
this question by proving the weak-(1, 1) estimate. As an application of the
weak-(1, 1) estimate and the trivial L∞ (Rn )-estimate, we will obtain various
important inequalities.
First of all, we state a geometrical lemma. It is important that the covering
lemma can be located as an assertion on geometry but that it comes into play
in analysis. Consequently, the covering lemma below can be viewed as an
interface of geometry and analysis.
Theorem 132 (Finite Vitali’s 5r-covering lemma). Suppose that we have a
finite family {Bλ }λ∈Λ of J balls. If we relabel 1, 2, . . . , J, then, for the newly
labeled family B1 , B2 , . . . , BJ , there exists 1 ≤ K ≤ J such that the following
condition is fulfilled;
134 Morrey Spaces

(1) B1 , B2 , . . . , BK are disjoint,

(2) there exists a mapping ι : {1, 2, . . . , J} → {1, 2, . . . , K} such that Bj ⊂

3 Bι(j) .

We can readily replace balls by cubes.

Proof We induct on J. When J = 1, K = 1 trivially does the job and
Theorem 132 is true in this case.
Let J0 ∈ N be fixed. Suppose that we have finite family {Bλ∗ }λ∈Λ of J balls
with J ≤ J0 and assume that we can relabel 1, 2, . . . , J to have B1∗ , B2∗ , . . . , BJ∗
so that;
(1) B1∗ , B2∗ , . . . , BK
∗
are disjoint,
(2) there exists a mapping ι : {1, 2, . . . , J} → {1, 2, . . . , K} such that Bj∗ ⊂
∗
3 Bι(j) .
Now suppose that we have J0 + 1 balls. We relabel B1 , B2 , . . . , BJ0 +1 so
that B1 has the largest radius. Again, if necessary, we change the label of
B2 , B3 , . . . , BJ0 +1 . If there are more than one ball with the maximum radius,
then choose one of them among B2 , B3 , . . . , BJ0 +1 . If B1 intersects Bj for
some j = 2, 3, . . . , J0 + 1, then Bj ⊂ 3 B1 since r(Bj ) ≤ r(B1 ).
Among {Bj }Jj=1 0 +1
, we can find at most J0 balls such that Bj ⊂ 3 B1 does
not hold, since B1 is automatically excluded. Then such a ball Bj does not
intersect B1 . Thus, among {Bj }Jj=1 0 +1
, let {Bj }J+1
j=2 be a family of balls such
that Bj ∩ B1 = ∅ for all j ∈ [2, J + 1]. Note that Bj ⊂ 3B1 for j > J in view
of the definition of J.
If we apply the induction assumption to {Bj }J+1 j=2 , a relabeling yields
B2 , . . . , BK and J(j) = 2, 3, . . . , K such that Bj ⊂ 3 BJ(j) unless 2 ≤ j ≤ J.
Then define a function ι : N → {1, 2, . . . , n} by

 1, (j = 1),
ι(j) ≡ J(j), (2 ≤ j ≤ J + 1),
1, (j > J + 1).


Thus, the balls B1 , B2 , . . . , BK satisfy the condition of the theorem and the
proof is complete.
Usually, we mean by the “5r-covering lemma” the following theorem (due
to Wiener), whose proof we omit due to similarity:
Theorem 133 (Infinite Vitali’s 5r-covering lemma). Suppose that we have a
family {Bλ }λ∈Λ of balls of general cardinality. If the radii of the balls in the
family is bounded from above by a positive constant, then we can find Λ0 ⊂ Λ
such that
(1) {Bλ }λ∈Λ0 is disjoint,
 Various operators in Lebesgue spaces 135

(2) there exists a mapping ι : Λ0 → Λ such that Bλ ⊂ 5 Bι(λ) .

Now we investigate M f for L1 (Rn )-functions f .
We prove the following important estimate called the weak-(1, 1) bounded-
ness or the L1 (Rn )-L1,∞ (Rn ) boundedness of M .
Theorem 134 (Hardy–Littlewood maximal inequality). For all f ∈ L1 (Rn )
and λ > 0, λ|{M f > λ} | ≤ 3n kf kL1 .
We will not discuss whether we can replace 3n by a smaller number. The
best constant C that can be used instead of 3n is called the weak-(1, 1) con-
stant.
Proof By the inner regularity (1.1) of the Lebesgue measure, it suffices
to show
λ|K| ≤ 3n kf kL1 (4.2)
for any compact set K contained in {M f > λ}.
Let x ∈ K. By the definition of M = M B , there exists a ball Bx centered
at x such that
mBx (|f |) > λ. (4.3)
By the compactness of K, there exists a finite collection x1 , x2 , . . . , xJ of
[J
points such that K ⊂ Bxj . To simplify the notation, let Bj ≡ Bxj . By
j=1
Theorem 132, if we relabel the Bj ’s, for some L ≤ J, we have the following:
(1) The family {Bj }L
j=1 of balls is disjoint. That is,

L
X
χBj ≤ 1, (4.4)
j=1

(2) We have inclusion:

J
[ L
[
K⊂ Bj ⊂ 3 Bj . (4.5)
j=1 j=1

Hence, if we use (4.5), (4.3) and (4.4) in that order, then

L L L
3n 3n
X X X Z
|K| ≤ |3Bj | = 3n |Bj | ≤ |f (x)|dx ≤ kf kL1 .
j=1 j=1 j=1
λ Bj λ

Thus, (4.2) and hence Theorem 134 are proven.

If we define the weak Lp (Rn )-space WLp (Rn ) = Lp,∞ (Rn ), p ∈ [1, ∞) as
1
the set of all f ∈ L0 (Rn ) for which kf kLp,∞ ≡ sup λ|{|f | > λ} p is finite, then
λ>0
it follows that M is bounded from L1 (Rn ) to L1,∞ (Rn ).
136 Morrey Spaces

Corollary 135. Suppose that f is a locally integrable function. Then

3n
Z
|{y ∈ Rn : M f (y) > 2λ}| ≤ |f (x)|dx. (4.6)
λ {y∈Rn : |f (y)|>λ}
Proof We calculate
|{y ∈ Rn : M f (y) > 2λ}| ≤ |{y ∈ Rn : M [χ(λ,∞] (|f |)f ](y) > λ}|
3n
Z
≤ |f (x)|dx.
λ {y∈Rn : |f (y)|>λ}
To investigate the Hardy–Littlewood maximal operator further, it is some-
times useful to consider its dyadic version. The next lemma is almost trivial
and we omit its proof.
Lemma 136 (The nesting structure of dyadic cubes). When two dyadic cubes
intersect, one is contained in the other.
Recall that the symbol D(Rn ) denotes the set of all dyadic cubes:
D = D(Rn ) ≡ {Qjk : j ∈ Z, k ∈ Zn } .
Qn
Likewise, for a right-open cube Q = [aj , bj ), the set D1 (Q) of the children of
j=1
Q is the set of all right-open
S cubes obtained by bisecting Q. Define inductively
Dk (Q) by Dk (Q) ≡ D1 (R). Most likely, we let Q = [−N, N )n or
R∈Dk−1 (Q)
Q = [0, N )n .
Theorem 137 (Sunrise lemma). Suppose that f ∈ L1loc (Rn ). Then for all
λ > 0,
Z
1
|{y ∈ Rn : M f (y) > λ}| ≥ n |f (x)|dx.
2 λ {y∈Rn : |f (y)|>λ}
Proof We may assume that f ∈ L1 (Rn ) by using the monotone conver-
gence theorem. Then we will obtain a disjoint family {Qj }j∈J of dyadic cubes
such that λ < mQj (|f |). By maximality, we have λ < mQj (|f |) ≤ 2n λ.
From the Lebesgue differentiation theorem, Theorem 12,
X
{y ∈ Rn : |f (y)| > λ} ⊂ {y ∈ Rn : M D f (y) > λ} = Qj
j∈J

a.e.. Thus, we have

Z Z
1 1
|f (x)|dx ≤ |f (x)|dx
λ {y∈Rn : |f (y)|>λ} λ P Qj
j∈J
X
≤ 2n |Qj |
j∈J

= 2n |{y ∈ Rn : M D f (y) > λ}|

≤ 2n |{y ∈ Rn : M f (y) > λ}|.
Thus, Theorem 137 is proven.
 Various operators in Lebesgue spaces 137

We are now oriented to the localized estimate. If we consider the bound-

edness property of the operators acting on Morrey spaces or Lebesgue spaces,
it is sometimes important to use the information for their behavior around
cubes or balls.
Example 65 (Sunrise lemma). For all f ∈ L0 (Rn ), λ > 0 and cubes Q,
Z
1
|{z ∈ Q : M [χQ f ](z) > λ}| ≥ n |f (y)|dy
2 λ {z∈Q : |f (z)|>λ}

We can modify the argument above to have an estimate for the maximal
function of the measure.
Example 66. Let µ be a non-negative finite Radon measure and define (tem-
porarily) the Hardy–Littlewood maximal operator of measures by
 
µ(Q)
M µ(x) ≡ sup : Q ∈ Q, c(Q) = x (x ∈ Rn ). (4.7)
|Q|

Then modifying the argument above, we can show

λ|{x ∈ Rn : M µ(x) > λ}| . µ(Rn ) (4.8)

for λ > 0. Let δ0 be the measure massed at the origin. It is also noteworthy
that M δ0 (x) ∼ |x|−n for x ∈ Rn . The proof of (4.8) is left as an exercise; see
Exercise 63.
Although in Proposition 128 we explained that the function M f is never
integrable, we still substitute this inequality for the weak variant.
Theorem 138. Let 0 < p < 1, E ⊂ Rn be a measurable set with finite
1
measure, and let f ∈ L1 (Rn ). Then kM f kLp (E) . |E| p −1 kf kL1 .
The idea that lies behind it is the use of the finiteness of the measure
and the weak-(1, 1) boundedness of operators. This estimate is called the Kol-
mogorov inequality. Note that this type of estimate is valid for any weak-(1, 1)
bounded linear operator.
Proof This is a consequence of
Z Z ∞
M f (x)p dx = pλp−1 |{x ∈ E : M f (x) > λ}|dλ,
E 0

which follows from Theorem 5, and

|{x ∈ E : M f (x) > λ}| . min{|E|, λ−1 kf kL1 } (λ > 0),

which in turn follows from Theorem 134.

138 Morrey Spaces

We extend Theorem 134. We consider the uncentered maximal operator

M generated by balls. For w ∈ M+ (Rn
R ) and measurable set E, we define the
weighted measure w(E) by w(E) ≡ w(x)dx. We will prove a sort of weighted
E
estimate, which will often be of importance in this book.
Proposition 139 (WeakZdual inequality of Stein-type). Let w ∈ M+ (Rn ).
3n
Then w{M f > λ} ≤ |f (x)|M w(x)dx for all f ∈ L0 (Rn ). More pre-
λ Z Rn
3n
cisely, w{M f > λ} ≤ |f (x)|M w(x)dx.
λ {M f >λ}

We also call the above inequality the weighted weak (1, 1)-inequality. In
Proposition 139, by letting w ≡ 1, we can recover Theorem 133.
Proof By the monotone convergence theorem, it can be assumed that w
is bounded. For λ > 0, set Eλ ≡ {M f > λ}. In view of the inner regularity of
the measure w(x)dx, it suffices to show

3n
Z
w(K) ≤ |f (x)|M w(x)dx
λ Rn

for any compact set K contained in Eλ .

If we combine the compactness of K and Theorem 132, we can define a
finite collection B1 , B2 , . . . , BM of balls such that:
(1) B1 , B2 , . . . , BM are disjoint:
M
X
χBj ≤ 1. (4.9)
j=1

(2) The inequality

mBj (|f |) > λ (4.10)
holds for j = 1, 2, . . . , M .
(3) K is covered by the triple of the balls B1 , B2 , . . . , BM , namely,
M
[
K⊂ 3 Bj . (4.11)
j=1

By the definition of the Hardy–Littlewood maximal operator M , we have

w(3Bj )
≤ 3n inf M w(y). (4.12)
|Bj | y∈Bj
 Various operators in Lebesgue spaces 139
M
P
By (4.11), we have w(K) ≤ w(3 Bj ). If we use (4.10), (4.12), and (4.9),
j=1
in that order, then
M Z
1X w(3 Bj )
w(K) ≤ |f (x)|dx ·
λ j=1 Bj |Bj |
M Z
n X
3
≤ |f (x)|M w(x)dx
λ j=1 Bj

3n
Z
≤ |f (x)|M w(x)dx.
λ Rn

Thus, the proof is complete.

If we combine Proposition 139 with the trivial equality kM kL∞ →L∞ = 1,
we obtain the weighted Lp (Rn )-estimate.
Corollary 140 (Strong
Z dual inequality of Stein-type).
Z For all 1 < p < ∞,
p n p 0
0 n
and f, w ∈ L (R ), M f (x) |w(x)|dx ≤ 3 2 p |f (x)|p M w(x)dx.
Rn Rn

Proof By replacing w with |w|, we may suppose in addition that w ≥ 0

almost everywhere. First of all, whenever k f kL∞ ≤ λ, M f ≤ λ. Together
with the subadditivity M [f + g] ≤ M f + M g, we deduce that {M f > 2λ} ⊂
{M [ χ(λ,∞] (|f |)f ] > λ}. By Proposition 139,
Z
1
w{M f > 2λ} ≤ w{M [χ(λ,∞] (|f |)f ] > λ} ≤ |f (x)| · M w(x)dx.
λ {|f |>λ}

We use (1.4) for the function M f and the measure w(x)dx. If we change
variables: σ = 2λ, then
!
Z Z Z M f (x)
M f (x)p w(x)dx = p σ p−1 dσ w(x)dx
Rn Rn 0
2−1 M f (x)
Z Z !
= 2p p λp−1 dλ w(x)dx. (4.13)
Rn 0

By virtue of (4.13) and Proposition 139,

Z
M f (x)p w(x)dx
Rn
Z ∞ Z 
= 2p p χ(2λ,∞] (M f (x))w(x)dx λp−1 dλ
0 Rn
Z ∞ Z 
≤ 3n 2p p χ(λ,∞) (|f (x)|)|f (x)|M w(x)dx λp−2 dλ.
0 Rn
140 Morrey Spaces

By Fubini’s theorem (Theorem 4), we calculate the 1-dimensional integral

precisely to obtain
Z
M f (x)p w(x)dx
Rn
Z ∞ Z 
≤ 3n 2p p χ(λ,∞) (|f (x)|)|f (x)|M w(x)dx λp−2 dλ
0 Rn
Z Z ∞ 
= 3n 2p p χ(λ,∞) (|f (x)|)λp−2 dλ |f (x)|M w(x)dx
Rn 0
!
Z Z |f (x)|
= 3n 2p p λp−2 dλ |f (x)|M w(x)dx
Rn 0
Z
= 3n 2p p0 |f (x)|p M w(x)dx.
Rn

Therefore, the proof is complete.

In particular, we obtain the Lp (Rn )-inequality of the Hardy–Littlewood
maximal operator M as a corollary by letting w ≡ 1:
Theorem 141 (Lp (Rn )-inequality of M ). Let 1 < p < ∞. Then we have
kM f kLp ≤ 3n p0 kf kLp for all f ∈ L0 (Rn ).
Proof By the monotone convergence theorem, we may assume that f ∈
L∞ n
c (R ). Reexamine the proof of Corollary 140 above. We start with

3n
Z
|{M f > λ}| ≤ |f (x)|dx.
λ {M f >λ}

Going through the same argument using the Layer-Cake formula, we have
Z Z
M f (x)p dx ≤ 3n p0 M f (x)p−1 |f (x)|dx
Rn Rn
Z  p−1
p
Z  p1
≤ 3n p0 M f (x)p dx |f (x)|p dx .
Rn Rn

Since f ∈ L∞ n
c (R ), we can arrange this inequality to obtain the desired con-
clusion.
About the constant obtained in Theorem 141, we have the following clar-
ifying remark:
Remark 5. Theorem 141 is very important and we use it many many times
in this book. Usually, it is not necessary to learn the constant 3n p0 by heart.
However, it is not so difficult to keep track of the constants in the proof.
Example 67. Let p ∈ (1, ∞) and r ∈ [1, ∞). Let T : L∞ n 0 n
c (R ) → L (R ) be a
sublinear operator such that T can be extended to a bounded operator both
 Various operators in Lebesgue spaces 141

T : L1 (Rn ) → Lr (Rn ) and T : Lp (Rn ) → L∞ (Rn ). In analogy to Corollary

140, we can prove that T is bounded from Lpθ (µ) → Lqθ (µ), where pθ and qθ
are given by
1 θ 1 1−θ
=1−θ+ , =
pθ p qθ r
for some 0 < θ < 1.
The boundedness of the Hardy–Littlewood maximal operator M is impor-
tant because it is less likely to control many other operators in terms of the
given function itself. Since the Hardy–Littlewood maximal operator enlarges
the functions in a suitable way, we still have a hope to control many integral
operators using M .
We conclude Section 4.1.2 with the proof of the Lebesgue differentiation
theorem, Theorem 12. To this end, we have only to show that the Lebesgue
measure does not charge the set E of all points for which (1.7) fails. We
decompose E; we define
( )
n 1
Ek ≡ x ∈ R : lim sup mB(x,r) (|f − f (x)|) > (4.14)
r↓0 k

for k ∈ N. Once we prove Ek has Lebesgue measure 0, then E will have

Lebesgue measure 0. Condition (1.7) is local. Hence, we can consider χB(r) f
instead of f . We can assume that f ∈ L1 (Rn ).
Furthermore, we can argue by using the ε-δ argument to obtain (1.7) for
g ∈ L1 (Rn ) ∩ BC(Rn ).
Let ε > 0 be arbitrary. Let us conclude the proof of the theorem by showing
that |Ek | ≤ 2(3n + 1) k ε.
Choose g ∈ L1 (Rn ) ∩ BC(Rn ) by density so that kf − gkL1 ≤ ε. Then have
( )
n 1
Ek = x ∈ R : lim sup mB(x,r) (|f − g + g(x) − f (x)|) > .
r↓0 k

We estimate Ek from above to have

 
1
Ek ⊂ x ∈ Rn : M [f − g](x) + |f (x) − g(x)| > (4.15)
k
   
1 1
⊂ x ∈ Rn : M [f − g](x) > ∪ x ∈ Rn : |f (x) − g(x)| > .
2k 2k

If we combine (4.15) with the subadditivity, Chebychev’s inequality and the

weak (1, 1)-maximal inequality, |Ek | ≤ 2(3n + 1) k kf − gkL1 ≤ 2(3n + 1) k ε.
The number ε > 0 being arbitrary, we obtain (1.7).
Before we go further, we give a fundamental relation between M f and f .
Lemma 142. Let f ∈ L1loc (Rn ). Then M f (x) ≥ |f (x)| for almost all x ∈ Rn .
142 Morrey Spaces

We end this section with an analogy to Lebesgue differentiation theorem

for the polynomials.
Proof By the Lebesgue differentiation theorem,
Z
1
M f (x) ≥ lim |f (y)|dy = |f (x)|
r↓0 |B(x, r)| B(x,r)

for almost all x ∈ Rn .

We present another application of the idea employed in the proof above.
k
Lemma 143. Let k ∈ N0 and f ∈ L1loc (Rn ). Then lim PQ(x,r) f (x) = f (x) for
r↓0
almost all x ∈ Rn .
Proof Similar to the Lemma 142; see Exercise 64.

4.1.3 Local estimates for the Hardy–Littlewood maximal

operator
So far, we have obtained some global estimates. However, when handling
some linear operators, we will face the situation where some local estimates
are necessary. As an application of the dual inequality of Stein-type, we will
obtain some related estimates useful in the situation above.
Example 68. For all f ∈ L0 (Rn ) and balls B(x, r), we have
Z Z
p
M f (y) dy = M f (y)p χB(x,r) (y)dy
B(x,r) R n
Z
n 0
≤3 p |f (y)|p M χB(x,r) (y)dy,
Rn

which can be regarded as a local norm estimate at all scales.

It is noteworthy that the most right-hand side in Example 68 can be rewrit-
ten using Fubini’s theorem. We transform Example 68 to the form which we
use, which is called a local estimate of the Hardy–Littlewood maximal oper-
ator.
Theorem 144. Let 1 < p < ∞. Then for all f ∈ L0 (Rn ) and all balls B(x, r),
(Z ! ) p1
∞ Z
n
p dt
kM f kLp (B(x,r)) . r p |f (y)| dy .
r B(x,t) tn+1

Proof Simply combine Example 68 and Fubini’s theorem.

Example 69. Let 1 < p < ∞. Consider the inequality

kM f kLp (B(r)) ≤ c(r)kf kLp (r−n M χB(r) ) (f ∈ L0 (Rn )),

 Various operators in Lebesgue spaces 143

and let c∗ (r) be the nminimal value of c(r) in the inequality above. Then we
claim c∗ (r) = c∗ (1)r p , which redues matters to the case where r = 1. Indeed,
we calculate
Z Z
n
M f (x)p dx rp M f (ry)p dx
B(r) B(1)
c∗ (r)p = sup Z = sup Z
f |f (x)|p f |f (ry)|p
n
dx n
dy
Rn (|x| + r) Rn (|y| + 1)
Z Z
n
rp M (f (r·))(y)p dx M g(x)p dx
B(1) n B(1)
= sup = r p sup Z
|f (ry)|p |g(x)|p dx
Z
f g
n
dy n
Rn (|y| + 1) Rn (|x| + 1)
∗ p n
= c (1) r .

4.1.4 Fefferman–Stein vector-valued maximal inequality

The Fefferman–Stein vector-valued maximal inequality, which we prove in
Section 4.1.4, asserts that
  q1   q1
X∞ X∞
 M fj q  .p,q  |fj |q 
j=1 j=1
Lp Lp

for all sequences {fj }∞ 0 n

j=1 ⊂ L (R ). Note that this inequality differs from
  q1    q1
X∞ X∞
M  |fj |q   .p,q  |fj |q  ,
 
j=1 j=1
Lp Lp

which is a direct consequence of Theorem 141.

A basic idea of analysis is the decomposition of a function into a countable
sum of elementary pieces of functions. The sum must be countable; otherwise
the sum may fail to be measurable. Because the sum is made up of functions
which are not complicated, we pay attention to each summand. Therefore, we
handle each term separately. The Hardy–Littlewood maximal operator con-
trols in some sense functions whose structures are not clear. Therefore, the
Fefferman–Stein vector-valued maximal inequality, whose left-hand side con-
tains a countable sum of the Hardy–Littlewood maximal functions of func-
tions, is useful.
The Fefferman–Stein vector-valued maximal inequality is elementary and
it is used mainly in the study of Triebel–Lizorkin spaces. We cannot cover
Triebel–Lizorkin spaces since it is beyond the scope of this book. See [146,
176, 156, 157, 415, 416, 382] for more about this direction. However, recently
more and more applications of this inequality were found. Theorem 145 is
extremely important and is used hundreds of times in this book.
144 Morrey Spaces

Theorem 145 (Fefferman–Stein vector-valued maximal inequality). Let 1 <

p < ∞ and 1 < q ≤ ∞. Then for all sequences {fj }∞ 0 n
j=1 ⊂ L (R ),

  q1   q1
∞
X X∞
 M fj q  .p,q  |fj |q  .
j=1 j=1
Lp Lp

To formulate the vector-valued inequalities in general, we adopt the fol-

lowing notation; the index set J is usually taken as J = N, N0 , Z. However
here we present the definition generally.
Definition 47 (Vector-valued norm). Let 0 < p, q ≤ ∞, and let J be a
countable set.
(1) For a system {fj }j∈J ⊂ L0 (Rn ) of functions, define
  q1
X
k{fj }j∈J k`q (Lp ) ≡ kfj kLp q  .
j∈J

The space `q (Lp , Rn ) is the set of all collections {fj }j∈J for which the
quantity k{fj }j∈J k`q (Lp ) is finite.
(2) For a sequence {fj }j∈J of functions we define a function k{fj }j∈J k`q :
Rn → [0, ∞] by
  q1
X
k{fj }j∈J k`q ≡  |fj |q 
j∈J

(3) For a system {fj }j∈J ⊂ L0 (Rn ) of functions, define

k{fj }j∈J kLp (`q ) ≡ kk{fj }j∈J k`q kLp .

The space Lp (`q , Rn ) is the set of all collections {fj }j∈J for which the
quantity k{fj }j∈J k`q (Lp ) is finite.
(4) For a measure space (X, B, µ), define Lp (`q , µ) and `q (Lp , µ) similarly.
These two norms are called vector-valued norms. A natural modification is
made in the above when q = ∞.
Example 70. We extend Hölder’s inequality. Let 1 ≤ p ≤ ∞ and 1 ≤
q ≤ ∞. Let (X, B, µ) be a measure space. By original Hölder’s inequality
k{aj bj }∞ ∞ ∞ ∞ ∞
j=1 k`1 ≤ k{aj }j=1 k`q k{bj }j=1 k`q0 for sequences {aj }j=1 and {bj }j=1
0
and by the one kF · GkL1 (µ) ≤ kF kLp (µ) kGkLp0 (µ) for F, G ∈ L (µ),
k{Fj Gj }∞ ∞ ∞
j=1 kL1 (`1 ,µ) ≤ k{Fj }j=1 kLp (`q ,µ) k{Gj }j=1 kLp0 (`q0 ,µ) .

We plan to prove Theorem 145 after proving a lemma.

 Various operators in Lebesgue spaces 145

Lemma 146. Let 1 ≤ p, q ≤ ∞, and let {fj }∞ p

j=1 be a sequence of L (R )-
n
∞
functions such that fj = 0 a.e. if j is large enough. Then we can take {gj }j=1 ∈
∞ Z
0 0
X
Lp (`q ) with norm 1 such that k{fj }∞ j=1 k Lp (`q ) = fj (x)gj (x)dx. If
j=1 Rn

{fj }∞
j=1
+ n
⊂ M (R ), then we can arrange that {gj }∞
j=1 ⊂ M (Rn ).
+

Proof Suppose 1 < q < ∞ for simplicity. There is nothing to prove if

fj = 0 a.e. for all j ∈ N0 ; assume otherwise. In this case, we recall the construc-
0 1
tion of the duality Lp (Rn )-Lp (Rn ): Set gj ≡ sgn(fj )|fj |q−1 k{fj }∞ j=1 k`q
p−q
.
A 0
p−1 ∞ p
Here A is a normalization constant: A ≡ kfj kLp (`q ) . Since (k{gj }j=1 k`q0 ) =
X∞
(k{fj }∞ k
j=1 ` q ) p
, we have k{g } ∞
k 0 0
j j=1 Lp (`q ) = 1. Furthermore, since fj gj =
j=1
∞ Z
1 X
(k{fj }∞ p ∞
j=1 k`q ) , we have k{fj }j=1 kLp (`q ) = fj (x)gj (x)dx.
A j=1 Rn

We prove Theorem 145. First we rephrase Theorem 145. We will show

Z Z
k{M fj (x)}∞ k
j=1 ` q
p
dx .p,q k{fj (x)}∞ p
j=1 k`q dx.
Rn Rn

By the monotone convergence theorem, we can assume that fj ≡ 0 for

large j.
Proof A simple case 1: q = p, ∞.
If p = q, Theorem 145 is clear by the Lp (Rn )-boundedness of M and
the monotone convergence theorem. If q = ∞, we invoke a trivial pointwise
estimate: sup M fj (x) ≤ M sup |fj | (x) for x ∈ Rn .
j∈N j∈N
Case 2: p > q. Keeping in mind that the left-hand side is at least finite,
p 0
we resort to duality. Set r ≡ . Then there exists g ∈ Lr (Rn ) ∩ M+ (Rn ) with
q
Z Z X∞
norm 1 such that k{fj (x)}∞
j=1 k`
p
q dx = M fj (x)q g(x)dx by virtue
Rn Rn j=1
r n r0 n
of the duality L (R )-L (R ). By Corollary 140 and Hölder’s inequality, we
have
Z Z X ∞
k{M fj (x)}∞ k
j=1 ` q
p
dx . |fj (x)|q M g(x)dx
Rn Rn j=1
Z
. kM gkLr0 p k{fj (x)}∞ p
j=1 k`q dx.
Rn
0
Since we know that M is Lr (Rn )-bounded, kM gkLr0 . 1. Hence
Z Z
k{M fj (x)}∞ k
j=1 ` q
p
dx . k{fj (x)}∞ p
j=1 k`q dx.
Rn Rn
146 Morrey Spaces

Therefore, the proof is complete.

√ p q
Case 3: p < q < ∞. Let θ ≡ p, s ≡ and t ≡ . Then
θ θ
Z
k{M fj (x)}∞ p θ ∞ s
j=1 k`q dx = k{M fj }j=1 kLs (`t ) .
Rn
0 0
By Lemma 146, there exists {gj }∞ s t n
j=1 ⊂ L (` , R ) such that

∞ Z
X
k{M fj θ }∞
j=1 kLs (`t ) = M fj (x)θ gj (x)dx, k{gj }∞
j=1 kLs0 (`t0 ) = 1.
j=1 Rn

By Corollary 140, we can move M to the function gj . Using Hölder’s inequality,

we have
X∞ Z ∞ Z
X
M fj (x)θ gj (x)dx .p,q |fj (x)|θ M gj (x)dx
j=1 Rn j=1 Rn
s
≤ k{|fj |θ }∞ ∞
j=1 kLs (`t ) k{M gj }j=1 kLs0 (`t0 ) .
t

The relation for s, t is reversed if we pass to the conjugate index; s0 > t0 . Thus,
by Step 2, k{M gj }∞ j=1 kLs0 (`t0 ) . 1. Consequently, we have
Z Z
k{M fj (x)}∞ j=1 k`
p
q dx . k{fj (x)}∞ p
j=1 k`q dx.
Rn Rn

Therefore, the proof is complete.

We will show that the Fefferman–Stein vector-valued inequality carries
over to various function spaces in this book.
Example 71. Let Φ ∈ ∇2 ∩ ∆2 and 1 <Z u < ∞. Then using Lemmas 48 and
∞
Φ(s) Φ(r)
49, we can find θ, θ† ∈ (1, ∞) such that θ+1
ds . θ for all r > 0 and
Z r r s r
Φ(t) Φ(r)
that θ † +1
dt . θ† for all r > 0.
0 t r
∞
Let {fj }j=1 ∈ LΦ (`u , Rn ). Then by the Fefferman–Stein vector valued
inequality for Lebesgue spaces, we have

Vλ ≡ x ∈ Rn : k{M fj }∞

j=1 k`u (x) > λ (4.16)
Z
. λ−θ k{fj }∞ θ ∞
j=1 k`u (x) χ(0,λ) (k{fj }j=1 k`u (x))dx
Rn
Z
†
θ†
+ λ−θ k{fj }∞ ∞
j=1 k`u (x) χ[λ,∞] (k{fj }j=1 k`u (x))dx.
Rn

Since Φ ∈ ∆2 , by the Layer-Cake formula

Z Z
Φ(λ)
Φ(k{M fj }∞ k
j=1 ` u (x))dx . Vλ dλ, (4.17)
Rn Rn λ
 Various operators in Lebesgue spaces 147

if we insert (4.16) into (4.17), then we obtain the vector-valued modular

inequality
Z Z
∞
Φ(k{M fj }j=1 k` (x))dx .
u Φ(k{fj }∞
j=1 k`u (x))dx (4.18)
Rn Rn

holds. In particular, k{M fj }∞ ∞

j=1 kLΦ (`u ) . k{fj }j=1 kLΦ (`u ) holds for all
∞ Φ u n
{fj }j=1 ∈ L (` , R ).

4.1.5 Orlicz-maximal operators

As we will see, the Hardy–Littleowood maximal operators are used to
control various linear operators. Sometimes we need to consider the two-fold
iteration of this maximal operator. Since M is not L1 (Rn )-bounded, it does
not make sense to consider such an iteration in L1 (Rn ). However, we have an
equivalent expression of the composition of M in terms of the Orlicz-maximal
operator. We first define the Orlicz-maximal operator. Recall that we defined
(Φ; Q)-average kf kΦ;Q of f ∈ L0 (Q) in Definition 26. Motivated by Definition
26, we define the Φ-maximal operator.
Definition 48. Let Φ : [0, ∞) → (0, ∞) be a Young function. The (Hardy–
Littlewood) Φ-maximal operator is given by MΦQ f ≡ sup χQ kf kΦ;Q for f ∈
Q∈Q
L0 (Rn ). Likewise define MΦB . The symbol MΦ denotes either MΦQ or MΦB .
We do not have to stick to cubes. Instead we can use balls. However, here
we use cubes because we need to bisect them. Since LΦ (µ) is a normed space
for any measure space (X, F, µ), MΦ is a sublinear operator. Sometimes we
are faced with the powered maximal operator after we use Hölder’s inequality.
Let 0 < η < ∞, and let f ∈ L0 (Rn ). We define the powered Hardy–Littlewood
maximal operator M (η) = M (η),B by
Z ! η1
1 (η)
M (η) f (x) ≡ sup |f (y)|η dy = sup mB(x,R) (f )
R>0 |B(x, R)| B(x,R) R>0

for x ∈ Rn . Likewise M (η) = M (η),Q is defined. We have the following example:

Example 72. Let Φ : [0, ∞) → [0, ∞) be a Young function.
(1) Note that M and M (u) for 1 < u < ∞ fall under the scope of this class
of operators; simply take Φ(t) = tu for 1 ≤ u < ∞.
(2) Let α ∈ R and let Φ : [0, ∞) → [0, ∞) be a convex function satisfying
Φ(t) = t(log(e + t))α . In this case we write ML(log L)α ≡ MΦ . If Φ(t) =
t log(e + t), then we write ML log L ≡ ML(log L)α .
(3) The function Φ(t) = et −1 is convex. In this case we write Mexp L ≡ MΦ .
Let us start with comparison with M and the Hardy–Littlewood Φ-
maximal operator.
148 Morrey Spaces

Corollary 147. For any Young function Φ : [0, ∞) → [0, ∞) and f ∈ L0 (Rn )
M f . MΦ f .
Proof Simply compare the definition of M f and MΦ f using Lemma 55.
As before, the maximal operator MΦ is bigger than the identity operator.
Corollary 148. Suppose that f ∈ L1loc (Rn ) and that Φ is a Young function.
Then |f (x)| . MΦ f (x) for almost all x ∈ Rn .
Proof This is clear since |f (x)| ≤ M f (x) . MΦ f (x) for any Lebesgue
point x of f ; see Lemma 142 and Corollary 147.
We investigate the boundedness property of the Orlicz maximal operators.
We first establish the following weak-type inequality:
Proposition 149. Suppose that f ∈ L1loc (Rn ) and that Φ : [0, ∞) → [0, ∞)
is a Young function. Then for all λ > 0,
3n
 
|f (x)|
Z
|{y ∈ Rn : MΦ f (y) > λ}| ≤ Φ dx.
λ Rn λ
Proof By the inner regularity (1.1) of the Lebesgue measure, it suffices
to show  
|f (x)|
Z
n
|K| ≤ 3 Φ dx (4.19)
Rn λ
for any compact set K contained in {MΦ f > λ}.
Let x ∈ K be fixed. By the definition of M , there exists Qx ∈ Q containing
x such that
kf kΦ;Qx > λ. (4.20)
By expanding Qx slightly, we may assume x ∈ Int(Qx ).
By the compactness of K, there exists a finite collection x1 , x2 , . . . , xJ of
J
S
points such that K ⊂ Qxj . To simplify the notation, write Qj ≡ Qxj . By
j=1
Theorem 132, if we relabel the Qj ’s, for some L ≤ J, the collection {Qj }L
j=1
is disjoint, that is,
L
X
χQj ≤ 1, (4.21)
j=1

and
J
[ L
[
K⊂ Qj ⊂ 3 Qj . (4.22)
j=1 j=1

Hence, if we use (4.22), (4.20) and (4.21) in that order, then

L L Z
3n
   
|f (x)| |f (x)|
X X Z
|K| ≤ |3Qj | ≤ 3n Φ dx ≤ Φ dx.
j=1 j=1 Qj
λ λ Rn λ

Thus, (4.19) and hence Proposition 149 are proven.

 Various operators in Lebesgue spaces 149

We generalize (4.6) and Theorem 137 as follows:

Proposition 150. Suppose that f ∈ L1loc (Rn ) and that Φ : [0, ∞) → [0, ∞)
is a bijective Young function. Then for all λ > 0,
3n
 
|f (x)|
Z
n −1
|{y ∈ R : MΦ f (y) > (1 + Φ (1))λ}| ≤ Φ dx.
λ {y∈Rn : |f (y)|>λ} λ
Proof Since λ > 0 is fixed, we can go through a similar argument to (4.6)
using Example 48.
In Proposition 150 we also have the opposite inequality.
Proposition 151. Suppose that f ∈ L1loc (Rn ) and that Φ : [0, ∞) → [0, ∞)
is a Young function. Then there exists D  1 such that, for all λ > 0,
 
|f (x)|
Z
n Q
|{y ∈ R : MΦ f (y) > λ}| ≥ Φ dx. (4.23)
{y∈Rn : |f (y)|>Dλ} λ
For the proof, we employ the dyadic maximal operator. Let f ∈ L0 (Rn ).
We define dyadic maximal operator M D by
M D f (x) ≡ sup χQ (x)mQ (|f |) (x ∈ Rn ).
Q∈D(Rn )

As an extension of M D consider the dyadic Φ-maximal operator: for all f ∈

L0 (Rn ),
MΦD f ≡ sup kf kΦ;Q χQ . (4.24)
Q∈D(Rn )
n
Since Q(R ) ⊃ D(R ),n
MΦD f ≤ MΦQ f .
Proof We have an analogy to Corollary 148: For almost all x ∈ Rn
|f (x)|(≤ M D f (x)) . MΦD f (x). It suffices to show that
 
|f (x)|
Z
n D
|{y ∈ R : MΦ f (y) > λ}| ≥ Φ dx.
D f (y)>λ}
{y∈Rn : MΦ λ
We consider the maximal dyadic cubes Q contained in
Ωλ ≡ {y ∈ Rn : MΦD f (y) > λ}
satisfying kf kΦ;Q > λ. Let S be a collection of all such cubes. Then
X
|{y ∈ Rn : MΦD f (y) > λ}| = |Q|
Q∈S
 
XZ |f (x)|
≥ Φ dx
Q λ
Q∈S
 
|f (x)|
Z
= Φ dx,
D f (y)>λ}
{y∈Rn : MΦ λ
as required.
150 Morrey Spaces

So far, we have seen how the operator MΦ arises. Here, we are interested
in showing that MΦ is bounded on Lp (Rn ) under some suitable conditions.We
propose the following condition:
Definition 49. Let 1 ≤ Zp < ∞. A Young function Φ : [0, ∞) → [0, ∞)
∞
Φ(t)
belongs to the class Bp if dt < ∞. In this case write Φ ∈ Bp .
1 tp+1
Example 73. Let 1 < p < ∞.
(1) Since Φ(t) ≥ Φ(1)t for any Young function Φ : [0, ∞) → [0, ∞) and
t ≥ 0, we see that B1 = ∅.
(2) Let Φ1 (t) = tq for 1 ≤ q < ∞. Then Φ1 ∈ Bp if and only if q < p.
(3) Let 1 ≤ q < ∞ and r ∈ R. Let Φ2 be a Young function such that
Φ2 (t) ∼ tq log(3 + t) for t  1. Then Φ2 ∈ Bp if and only if q < p.
The class Bp is important because this class characterizes the boundedness
of MΦ on Lp (Rn ) for 1 < p < ∞.
Theorem 152. Let 1 < p < ∞, and let Φ : [0, ∞) → [0, ∞) be a bijective
Young function. Then MΦ is bounded on Lp (Rn ) if and only if Φ ∈ Bp .
Proof Suppose Φ ∈ Bp . Let C0 ≡ 1 + Φ−1 (1). Then by Theorem 5 and
Proposition 150,
Z Z ∞
p
MΦ f (y) dy ∼ λp−1 |{y ∈ Rn : MΦ f (y) > C0 λ}|dλ
Rn 0
Z ∞   !
|f (x)|
Z
p−2
. λ Φ dx dλ
0 {y∈Rn : |f (y)|>λ} λ
Z ∞ Z
Φ(t)
= dt · |f (y)|p dy.
1 tp+1 Rn

Conversely suppose that MΦ is bounded on Lp (Rn ). Let D be a constant

as in Proposition 151. Then
Z
∞> MΦ χ[0,1]n (y)p dy
R n
Z ∞
=p λp−1 |{y ∈ Rn : MΦ χ[0,1]n (y) > 2λ}|dλ
0
Z ∞  
χ[0,1]n (x)
Z
≥p λp−1 Φ dx
0 {y∈Rn : χ[0,1]n (y)>Dλ} λ
Z ∞
Φ(s)
∼ p+1
ds,
D s

implying Φ ∈ Bp .
 Various operators in Lebesgue spaces 151

4.1.6 Composition of the maximal operators

As is mentioned before, the Φ-maximal operators can be used for the
purpose of the composition of the maximal operators. To see this, we start
with localizing the sunrise lemma. To localize our estimate, it is useful to use
the dyadic cubes for a fixed compact or right-open cube Q.
Definition 50 (D(Q)). Let D(Q) be the collection of all dyadic subcubes of
Q. That is, all those cubes obtained by dividing Q into 2n congruent cubes
of half its length, dividing each of those into 2n congruent cubes, and so on.
For the cube Q and k = 0, 1, . . . Dk (Q) is the set of all cubes obtained by
∞
S
bisecting Q k times. Write D(Q) ≡ Dk (Q). In particular, by convention,
k=0
Q itself belongs to D0 (Q) and hence D(Q).
Example 74. Let Q ≡ [−2N , 2N )n for N ∈ Z. Then D1 (Q) ⊂ D(Rn ).
The following lemma is the localized version of the so-called “Wiener-Stein
equivalence”.
Lemma 153 (Sunrise lemma for MΦ ). Suppose that Φ : [0, ∞) → [0, ∞) is a
normalized Young function and that f ∈ L1loc (Rn )∩M+ (Rn ). Then, for Q ∈ Q
and any number t > kf kΦ,Q ,
Z  
f (y)
Φ dy ≤ |{x ∈ Q : MΦ [f χQ ](x) > t}|
{x∈Q: f (x)>t} 2n t
Z  
2f (y)
. Φ dy. (4.25)
{x∈Q: 2f (x)>t} t
In particular, if Φ ∈ ∆2 then
Z  
f (y)
Φ dy ≤ |{x ∈ Q : MΦ [f χQ ](x) > t}|
{x∈Q: f (x)>t} t
Z  
f (y)
. Φ dy.
{x∈Q: 2f (x)>t} t
Proof For x ∈ Rn write f1 (x) ≡ f (x) if 2f (x) > t and f1 (x) = 0 otherwise
and let f2 ≡ f − f1 . As we did in Proposition 150, using Example 48, we have
t
MΦ [f χQ ](x) ≤ MΦ [f1 χQ ](x) + .
2
This implies
 
t
{x ∈ Q : MΦ [f χQ ](x) > t} ⊂ x ∈ Q : MΦ [f1 χQ ](x) > .
2
It follows from the infinite 5r-covering lemma that there exists the set of
disjoint cubes {Qj } such that
  [
t t
Qj ⊂ Q, < kf1 kΦ, Qj , x ∈ Q : MΦ [f1 χQ ](x) > ⊂ 3Qj .
2 2 j
152 Morrey Spaces
Z  
1 2f1 (x)
We now see that Φ dx > 1; if it were not, then by the
|Qj | Qj t
t
definition of the mean Luxemberg norm we would have ≥ kf1 kΦ, Qj , which
Z  2 
2f1 (x)
is a contradiction. This implies that |Qj | < Φ dx and that the
Qj t
right inequality of (4.25).
Since, t > kf kΦ, Q , there exists a disjoint collection of maximal (with
inclusion) dyadic cubes {Qj } ⊂ D(Q) \ {Q} such that
[
t < kf kΦ, Qj , {x ∈ Q : MΦ [f χQ ](x) > t} ⊃ Qj .
j

Let Q̃j ∈ D(Q) be the dyadic parent of Qj , the unique dyadic cube contain-
ing Qj with side-length twice that of Qj . Thus, from Example 47, we have
kf kΦ, Qj ≤ 2n kf kΦ, Q̃j and by the maximality of Qj kf kΦ, Qj ≤ 2n kf kΦ, Q̃j ≤
2n t. It follows from the definition of the
 mean
 Luxemberg
 normand  the 
fact
↑ f f
that Φ ∈ M [0, ∞) satisfies 1 ≥ mQj Φ ≥ mQj Φ .
  kf kΦ, Qj 2n t
Z
f (x)
This yields that |Qj | ≥ Φ dx. Thus, we obtain the left inequality
Qj 2n t
S
of (4.25) by observing further that Qj ⊃ {x ∈ Q : f (x) > t}, which holds
j
by the Lebesgue differential theorem.
We use the following criteria to check that MΨ is bounded.
Lemma 154. Let Ψ : [0, ∞) → [0, ∞) be a normalized Young function. Sup-
pose that normalized Young functions Φ, Φ1 , Φ2 : [0, ∞) → [0, ∞) fulfill, for
some positive constants C1 and C2 ,
Z t  
t
Φ1 (C1 t) − 1 ≤ Ψ Φ0 (s)ds ≤ Φ2 (C2 t) (t > 1).
1 s

Then, for any compact Q ∈ Q and any f ∈ M+ (Rn ),

kf kΦ1 , Q . kMΨ [f χQ ]kΦ;Q . kf kΦ2 , Q .

Hence, MΦ1 f . MΦ ◦ MΨ f . MΦ2 f . In particular, the boundedness of MΨ is

independent of the values of Φ(t) and Ψ(t), 0 < t ≤ 1.
Proof First, we verify kMΨ [f χQ ]kΦ;Q . kf Z
kΦ2 , Q . Let t > 0. By  virtue
2f (x)
of Lemma 153, |{x ∈ Q : MΨ [f χQ ](x) > t}| . Ψ dx. It
Q∩{2f >t} t
 Various operators in Lebesgue spaces 153

follows from this inequality that

Z
Φ (MΨ [f χQ ](x)) dx
Q
Z ∞
= |{x ∈ Q : MΨ [f χQ ](x) > t}|Φ0 (t)dt
0
Z ∞ Z   !
2f (x)
≤ |Q| + C Ψ dx Φ0 (t)dt
1 Q∩{2f >t} t
Z Z 2f (x)   !
2f (x) 0
= |Q| + C Ψ Φ (t)dt dx
Q∩{2f >1} 1 t
Z
≤ |Q| + C Φ2 (2C2 f (x))dx.
Q

For any λ > 0, replacing f (x) by f (x)

λ , we have
     
MΨ [f χQ ] 2C2 f
mQ Φ ≤ 1 + CmQ Φ2 .
λ λ

This yields, by the definition of the mean Luxemburg norm, for some C > 1
Z  
1 MΨ [f χQ ](x)
Φ dx ≤ 1,
|Q| Q Ckf kΦ2 , Q

which proves the desired inequality.

Next, we verify the remaining inequality. Without loss of generality, we
may assume kMΨ [f χQ ]kΦ;Q = 1. This means that
Z
1
Φ (MΨ [f χQ ](x)) dx ≤ 1.
|Q| Q

We now claim that then kf kΨ, Q ≤ 1. If it were not, then we must have
that the above integral mean is bigger than one, which contradicts our
normalization above, by virtue of the fact that for almost every x ∈ Q
kf kΨ, Q ≤ MΨ [f χQ ](x).
We claim
kf kΦ1 , Q . 1. (4.26)
In fact, since kf kΨ, Q ≤ 1,
Z
|Q| ≥ Φ (MΨ [f χQ ](x)) dx
Q
Z ∞
= |{x ∈ Q : MΨ [f χQ ](x) > t}|Φ0 (t)dt.
0
154 Morrey Spaces

Thanks to Lemma 153,

Z ∞ Z  ! 
f (x)
|Q| ≥ Ψ nt
dx Φ0 (t)dt
1 Q∩{f >t} 2
Z Z f (x)   !
f (x)
= Ψ Φ0 (t)dt dx.
Q∩{f >1} 1 2n t

By our assumption,
f (x)/2n
Z Z   !
f (x) 0
|Q| ≥ Ψ Φ (t)dt dx
Q∩{f >2n } 1 2n t
Z
≥ Φ1 (2−n C1 f (x))dx − |Q ∩ {f > 2n }|
Q∩{f >2n }
Z
≥ Φ1 (2−n C1 f (x))dx − (Φ1 (C1 ) + 1)|Q|.
Q
Z  
Φ1 (C1 ) + 2 1 f (x)
Let C ≡ . Then the above estimate reads Φ1 dx ≤
2−n C1 |Q| Q C
1 and, hence, (4.26) follows. The proof is now complete.
Example 75. Let Q ∈ Q, and let f ∈ L0 (Q). Then mQ (M [χ3Q f ]) .
kf kL log L,3Q from Lemma 75 with Φ(t) = Ψ(t) = t, t ≥ 0.
As a special case of Φ1 (t) = Φ2 (t) = t log(e + t) and Φ(t) = Ψ(t) = t,
t ≥ 0, we have the equivalent expression of the composition of the maximal
operators.
Corollary 155. One has M ◦ M f ' ML log L f for any f ∈ L0 (Rn ).
Proof Simply combine Example 65 and Lemma 75 with Φ(t) = Ψ(t) = t
and Φ1 (t) = Φ2 (t) = t log(e + t), t ≥ 0.
Example 76. For any f ∈ L0 (Rn ) and λ > 0,
 
|f (x)| |f (x)|
Z
n
|{x ∈ R : M ◦ M f (x) > λ}| . log 3 + dx
Rn λ λ

from Proposition 150 and Corollary 155.

4.1.7 Local boundednss of the Φ-maximal operators

As an application of the results in the previous section, we further inves-
tigate the property of MΨ ; we will obtain its local boundedness. We recall
that by a normalized Young function, we mean a Young function Φ satisfying
Φ(1) = 1.
 Various operators in Lebesgue spaces 155

Proposition 156. Let Φ, Ψ : [0, ∞) → [0, ∞) be normalized Young functions.

Then the following are equivalent.

(1) The maximal operator MΨ is locally bounded in the norm determined by

Φ. That is,

kMΨ f kΦ;Q . kf kΦ;Q (4.27)

for all measurable functions f and cubes Q.

(2) There exists a constant C > 0 such that
Z t  
t
Ψ Φ0 (s)ds ≤ Φ(Ct) (t ≥ 1). (4.28)
1 s

(3) There exists a constant C > 0 such that

Z ∞  
t
Φ Ψ0 (s)ds ≤ Φ(Ct) (t ≥ 1). (4.29)
t s

Proof We have already proven that (4.28) implies (4.27); see Lemma 154.
First, we verify (4.27) implies (4.28). Suppose
Z that (4.27), namely,
 assume
MΨ [f χQ ](x)
that there exists a constant C0 ≥ 1 such that Φ dx ≤ |Q|.
Q C0 kf kΦ;Q
We now claim that then kf kΨ, Q . C0 kf kΦ;Q . If it were not, then we would
have that the above integral mean is bigger than one, which contradicts our
normalization above, by virtue of the fact that kf kΨ, Q ≤ MΨ [f χQ ](x) for
almost every x ∈ Q. Thus we have by Lemma 154
Z  
MΨ [f χQ ](x)
|Q| ≥ Φ dx
Q C0 kf kΦ;Q
Z ∞
= |{x ∈ Q : MΨ [f χQ ](x) > C0 kf kΦ;Q · s}| Φ0 (s)ds
0
Z ∞Z  
|f (x)|
≥ Ψ dx Φ0 (s)ds
1 Q∩{|f |>C0 kf kΦ;Q ·s} 2n C0 kf kΦ;Q · s
Z 2n C|fkf
(x)|  
|f (x)|
Z kΦ;Q
0
≥ Ψ Φ0 (s)dsdx.
Q∩{|f |>2n C0 kf kΦ;Q } 1 2n C0 kf kΦ;Q · s

If we test the above inequality on f ≡ χ with some appropriate R ∈ Q(Q)

Z t R 
t |Q|
n −1
and let t ≡ (2 C0 kf kΦ;Q ) , then Ψ Φ0 (s)ds ≤ . Observing that
  1 s |R|
1 |Q|
Φ(2n C0 t) = Φ = , we obtain (4.28).
kf kΦ;Q |R|
156 Morrey Spaces

Next, we verify (4.29) implies (4.28). Carrying out integration by parts,

we have
Z t      t Z t  
t t t ds
Ψ Φ0 (s)ds = Ψ Φ(s) + t Ψ0 Φ(s) 2
1 s s 1 s s
Z1 ∞  
t
= Φ(t) − Ψ(t) + Φ Ψ0 (s)ds
t s
≤ Φ(t) + Φ(Ct) ≤ Φ(Ct).
t
Here, we have used changed variables s 7→ s. The converse, (4.28) implies
(4.29), is similar, once we notice that
Z 2 

1 2t
Ψ(t) ≤ Ψ Φ0 (s)ds
Φ(2) − Φ(1) 1 s
Z 2t  
1 2t
≤ Ψ Φ0 (s)ds
Φ(2) − Φ(1) 1 s
1
≤ Φ(2Ct)
Φ(2) − Φ(1)
≤ Φ(Ct), t > 1.

Thus, the proof is complete.

Earlier we investigated the condition on Φ for which MΦ is bounded on
Lp (Rn ). Here, we present the opposite type of boundedness: We investigate
when the maximal operator M (p) is locally bounded in the norm determined
by Φ.
Theorem 157. Let Φ : [0, ∞) → [0, ∞) be a normalized Young function, and
let 1 ≤ p < ∞. Suppose that s1−p Φ0 (s), s > 1, is nondecreasing. Then the
following are equivalent:
(a) The maximal operator M (p) is locally bounded in the norm determined
by Φ.
Z t 0
Φ (s)
(b) There exists a constant C > 0 such that tp ds ≤ Φ(Ct) for all
1 sp
t > 1.
(c) There exists some constant K > 1 such that 2K p Φ(t) ≤ Φ(Kt) for all
t > 1.
Proof We have already proven the equivalence between (a) and (b) in
Proposition 156; simply let Ψ(t) ≡ tp . We verify that (b) implies (c). By
assumption, s1−p Φ0 (s) is nondecreasing, we have for µ > 1
µt µt
Φ0 (s)
Z Z
1
Φ(Cµt) ≥ (µt)p ds ≥ µp tΦ0 (t) ds ≥ µp Φ(t) log µ,
t sp t s
 Various operators in Lebesgue spaces 157

where we have used the fact that Φ(t) 0

t ≤ Φ (t) which holds from the convexity.
It follows by letting µ be big enough so that C −p log µ > 2 and then, by setting
K = Cµ so that
2K p Φ(t) ≤ Φ(Kt).
This is the desired inequality.
We verify the converse: We verify (c) implies (b). If we let N ≡ [1+logK t],
then
t N K −j t N
Φ0 (s) Φ0 (s) X Φ(K −j t)
Z X Z
ds ≤ ds ≤
1 s p
j=0 K −j−1 t s p
j=0
(K −j−1 t)p
N N
K p X pj −j Kp X 1
= p K Φ(K t) ≤ p Φ(t) 2−j ≤ p Φ(2K p t).
t j=0 t j=0
t

This proves the desired inequality.

As a special case we can recover the classical result on the ∇2 -condition.
Corollary 158. Let Φ : [0, ∞) → [0, ∞) be a Young function.
(1) Let Φ ∈ ∇2 . Then kM [f χQ ]kΦ;Q ≈ kf kΦ;Q for all f ∈ L0 (Rn ).
(2) Conversely, if the above estimate is true, the Φ ∈ ∇2 .
Proof
(1) We can readily check that Φ satisfies (b) with p = 1 in Theorem 157
using Lemma 49.
(2) Simply observe that (c) with p = 1 in Theorem 157 is equivalent to
Φ ∈ ∇2 .

4.1.8 Estimates for convolutions

Here we collect the estimates for convolutions. Recall that the convolution
f ∗ g is given by Z
f ∗ g(x) = f (x − y)g(y)dy
Rn
as long as the integral makes sense.
We present an auxiliary estimate using the Hardy–Littlewood maximal
operator.

Proposition 159. Suppose that ρ ∈ M↓ [0, ∞) and that τ ∈ M+ (Rn ) is given

by
τ (x) = ρ(|x|) for all x ∈ Rn . (4.30)
Then for all t > 0 |τt ∗ f (x)| ≤ kτ kL1 M f (x). for all f ∈ Lp (Rn ), 1 ≤ p ≤ ∞.
158 Morrey Spaces

Proof This simply paraphrases Theorem 131.

We will provide a method to transform |x − y|−β to another expression
here.
The following lemma is interesting in its own right.
Lemma 160. Let β > 0. Then for all x, y ∈ Rn ,
Z ∞
dl |x − y|−β
χB(x,l) (y) β+1 = .
0 l β
Proof The proof is simple. Indeed, we have
Z ∞ Z ∞ Z ∞
dl dl dl |x − y|−β
χB(x,l) (y) β+1 = χ(|x−y|,∞) (l) β+1 = β+1
=
0 l 0 l |x−y| l β
by using Fubini’s theorem, which is the desired result.
Here we present applications of Lemma 160.
Lemma 161. Let β ≥ 0. Let ϕ ∈ M+ (Rn ). Then for all r ≥ 0,
Z Z ∞ Z !
ϕ(x) d`
β
dx = β ϕ(x)dx β+1 .
n
R \B(r) |x| r B(t)\B(r) `

In particular, for all z ∈ Rn ,

Z Z ∞ Z !
ϕ(x) d`
β
dx = β ϕ(x)dx .
Rn |x − z| 0 B(z,t) `β+1

Proof By Lemma 160 and Fubini’s theorem, we obtain

Z Z Z ∞ 
ϕ(x) dl
β
dx = β ϕ(x) χB(l) (x) β+1 dx
Rn \B(r) |x| Rn \B(r) 0 l
Z ∞ Z 
dx
=β ϕ(x)χB(l)\B(r) (x) β+1 dl
0 Rn l
Z ∞ Z !
dx
=β ϕ(x) β+1 dl.
r B(l)\B(r) l

The second formula is a translation of the first formula with r = 0 by z.

We transform Lemma 161 to handle a different type of function.
Corollary 162. Let β > 0 and r > 0. Let ϕ ∈ M+ (Rn ). Then
Z ∞ Z ! Z
−β dt ϕ(x)dx
β2 ϕ(x)dx β+1 ≤ β
r B(t) t Rn (|x| + r)
Z ∞ Z !
dt
.β ϕ(x)dx β+1 .
r B(t) t
 Various operators in Lebesgue spaces 159

Proof We note that

Z Z Z
ϕ(x) −β ϕ(x)
β
dx ≤ r ϕ(x)dx + β
dx
Rn (|x| + r) B(r) Rn \B(r) |x|

and that Z Z
ϕ(x) −β −β
β
dx ≥ 2 r ϕ(x)dx.
Rn (|x| + r) B(r)
If we decompose the integral dyadically in the second term in the right-hand
side and use Lemma 161 and
Z ∞ Z ! Z
d` −β
β ϕ(x)dx β+1 = r ϕ(x)dx,
r B(r) ` B(r)

then we obtain the desired result.

We collect some other convolution estimates of convolution type.
Proposition 163. Let γ ∈ R. Then for all x ∈ Rn and r, ρ > 0,
Z
|x − y|γ dy
B(r)\B(x,ρ)


1 (γ < −n),
|x| + r 

 
log 3 + (γ = −n),

γ n n
. ρ min(r , ρ )χ(0,|x|+r) (ρ) ρ
 γ+n

 |x| + r

 (γ > −n).
ρ
Proof We may suppose ρ < |x| + r; otherwise B(x, ρ) ⊃ B(r). Let m =
log2 |x|+r
 

ρ +
. For k > m, we note B(r) ∩ B(x, 2k+1 ρ)\B(x, 2k ρ) = ∅, since
2k ρ ≥ |x| + r. Hence B(r) ⊂ B(x, 2k ρ). Thus,
Z m Z
X
|x − y|γ dy = |x − y|γ dy
B(r)\B(x,ρ) k=0 B(r)∩(B(x,2k+1 ρ)\B(x,2k ρ))

Xm Z
. (2k ρ)γ dy
k=0 B(r)∩B(x,2k+1 ρ)

Xm
. (2k ρ)γ min(rn , 2(k+1)n ρn )
k=0
m
X
. ργ min(rn , ρn ) 2k(γ+n) ,
k=0
and the desired result follows since for any t ∈ R,
 1
 (t < 0),
 1 − 2t
m


X
kt
2 ≤ m+1 (t = 0),
(m+1)t
2

k=0 

 (t > 0).
2t − 1
We also collect another type of estimate.
160 Morrey Spaces

Proposition 164. Let β ≥ 0, x ∈ Rn , ρ > 0 and r > 0.

(1) If γ < −β − n, then
Z
β
|x − y|γ (|y| + r) dy . (|x| + ρ + r)β ργ+n .
Rn \B(x,ρ)

(2) If γ < −β, then

β
sup |x − y|γ (|y| + r) ≤ 2β (|x| + ρ + r)β ργ .
y∈Rn \B(x,ρ)

Proof
(1) We decompose
Z
I≡ |x − y|γ (|y| + r)β dy
Rn \B(x,ρ)
∞ Z
X
= |x − y|γ (|y| + r)β dy
k=0 B(x,2k+1 ρ)\B(x,2k ρ)

Note that
∞
X Z
γ kγ
I≤ρ 2 (|y| + r)β dy
k=0 B(x,2k+1 ρ)\B(x,2k ρ)
∞
X
n
β
. ργ 2kγ 2k+1 ρ |x| + 2k+1 ρ + r
k=0
∞
X
β
= 2n+β ργ+n (|x| + ρ + r) 2k(γ+β+n)
k=0
γ+n β
∼ρ (|x| + ρ + r) .

(2) Mimic the proof above.

4.1.9 Exercises
Exercise 61. Let f : R → R be a measurable function. Also let a > 0 and
g : R → [0, ∞) is a function such that g|[a, ∞) is decreasing and g|(−∞, a] is
increasing. Then show that kf ·gkL1 ≤ kgkL1 M f (a) by mimicking the proof of
Proposition 159. Here M denotes the uncentered Hardy–Littlewood maximal
operator.
Exercise 62. Let 0 < a < b < 1, and let Q be a cube. Then show that
M Q χbQ\aQ ≥ (bn − an )χQ by considering mQ (χbQ\aQ ).
 Various operators in Lebesgue spaces 161

Exercise 63. Let µ be a finite Radon measure on Rn and λ > 0.

(1) Let K be a compact set. Show that there exists a finite collection
J˜
˜
{Q̃j }Jj=1 of cubes such that K ⊂
S
Q̃j and that µ(Q̃j ) ≥ λ|Q̃j | for
j=1
˜
all j = 1, 2, . . . , J.
(2) Let K be a compact set. Use the finite 5r-covering lemma (Theorem 132)
to show that there exists a disjoint finite collection {Qj }Jj=1 of cubes such
J
S
that K ⊂ 3Qj and that µ(Qj ) ≥ λ|Qj | for all j = 1, 2, . . . , J.
j=1

(3) Prove (4.8).

k
Exercise 64. Let k ∈ N0 and denote by PQ(x,r) f the Gram–Schmidt polyno-
mial of order k for a cube Q(x, r) for f ∈ Lloc (Rn ).
1

k
(1) Prove that |PQ(x,r) f (x)| . M f (x) using Proposition 117.
(2) Prove Lemma 143 by mimicking the proof of Lemma 142.

4.2 Sharp maximal operators

On many occasions, we are faced with the necessity of estimating the
norm of functions. Especially we need to control the image of functions in
terms of the norms. It is often difficult to do this because the definition of
operators is complicated. However, when we can use the mean-value theorem,
the sharp maximal operator, which is defined by the difference of functions, is
a useful tool. In some sense, we decompose measurable functions into the sum
of less oscillating functions. Here, we investigate the sharp maximal operator
to be able to deal with and take full advantage of the oscillating properties of
functions and operators.
We define the sharp maximal operator and then investigate some funda-
mental inequality in Section 4.2.1. In Section 4.2.2 we will expand the idea
employed in Section 4.2.1 so that it is applicable to any measurable functions.
Section 4.2.3 is a further development of Sections 4.2.1 and 4.2.2 and considers
a fundamental pointwise estimate of measurable functions.

4.2.1 Sharp-maximal inequalities

Here, we present an important estimate from the viewpoint given above.
Let f ∈ L1loc (Rn ). We define the sharp maximal operator M ] f by
Z
χQ
M ] f ≡ sup |f (y) − mQ (f )|dy = sup mQ (|f − mQ (f )|)χQ .
Q∈Q |Q| Q Q∈Q
162 Morrey Spaces

A trivial but interesting observation is that

Z Z
|f (x) − mQlj (f )|dx ≤ M ] f (x)dx. (4.1)
Qlj Qlj

We prove an estimate which is useful when we control the singularity of

functions:
Theorem 165 (Sharp-maximal inequality). Let f, g ∈ L1loc (Rn ). Then if

|{|f | > λ}| < ∞ (4.2)

for all λ > 0, kf · gkL1 ≤ 6kM ] f · M D gkL1 .

Proof Let k ∈ Z be fixed. We can assume that f ∈ L∞ (Rn ) ∩ M+ (Rn )
and that g ∈ L∞ n + n
c (R ) ∩ M (R ) by the truncation argument. We partition
D
{M g > 2 } into a collection of maximal dyadic cubes such that mQkj (g) > 2k .
k
Y
We write {M D g > 2k } = Qkj , where {Qjk }j∈J k is a disjoint collection of
j∈J k
dyadic cubes. We will form a Calderón–Zygmund decomposition with respect
to D. Define
X
Gk ≡ χ{M D g≤2k } g + mQj (g)χQkj , B k ≡ g − Gk .
k
j∈J k

We will move k ∈ Z now. For eachZj 0 ∈ J k+1 , there exists j ∈ J k such

that Qk+1
j0 ⊂ Qkj . Therefore, we have (B k+1 (x) − B k (x))dx = 0 for all
Qk
j

j ∈ J k . In view of the Lebesgue differentiation theorem, |B k (x) − B k+1 (x)| =

|Gk (x) − Gk+1 (x)| ≤ 3 · 2n+k for almost all x ∈ Rn . Since g ∈ L∞ (Rn ), B k ≡ 0
if k  1. Therefore,
Z Z Z
k
f (x)g(x)dx = f (x)G (x)dx + f (x)B k (x)dx
Rn Rn Rn
Z ∞
XZ
= f (x)Gk (x)dx + f (x)(B l (x) − B l+1 (x))dx
Rn l=k Rn

=: Ik + IIk .

We will show that lim Ik = 0. Let ε > 0 be arbitrary. We split

k→−∞
Z Z
Ik = f (x)Gk (x)dx + f (x)Gk (x)dx.
{f >ε} {f ≤ε}

Choose k ∈ Z ∩ (−∞, log2 ε). Note again that |Gk (x)| ≤ 2k+n almost every-
where by the maximality of the cubes Qkj and the Lebesgue differentiation
theorem. Since
Ik ≤ 2k+n kf kL∞ |{f > ε}| + εkgkL1 ,
 Various operators in Lebesgue spaces 163

it follows from (4.2) that lim Ik = 0. As for IIk , we have

k→−∞

∞ X Z
X
|IIk | ≤ |f (x) − mQlj (f )| · |B l (x) − B l+1 (x)|dx
l=k j∈J l Qlj
∞
X XZ
n+l
≤3 2 |f (x) − mQlj (f )|dx.
l=k Qlj
j∈J l

∞
X Z
We deduce |IIk | ≤ 3 2 n+l
M ] f (x)dx ≤ 6kM ] f · M D gkL1 from
l=−∞ {M D g>2l }
(4.1). The proof is therefore complete.
We transform Theorem 165 into the form which we actually use.
Theorem 166 (Sharp maximal inequality). Let 1 < p < ∞. Then kf kLp .p
kM ] f kLp for f ∈ L0 (Rn ) with min(M f, 1) ∈ Lp (Rn ).
We really have to assume that min(1, M f ) ∈ Lp (Rn ) as the example of
the function f ≡ 1 shows.
Proof By truncation, we may suppose that f ∈ L∞ (Rn ). In this case,
M f ∈ Lp (Rn ). Hence by Corollary 148, we have f ∈ Lp (Rn ), so that (4.2) is
0
satisfied. Let w ∈ Lp (Rn ) ∩ M+ (Rn ) be a function with norm 1 satisfying
Z
kf kLp ≤ 2 f (x)w(x)dx.
Rn

By Theorem 165, we have kf kLp ≤ 12kM ] f · M D wkL1 . It remains to use

Hölder’s inequality and the boundedness of M D .
We consider a pointwise estimate of the oscillation of functions.
We use the following notation: D(Q0 ) is the set of all dyadic cubes with
respect to a cube Q0 . The mean oscillation of f ∈ L0 (Q) of level λ ∈ (0, 1)
is given by ωλ (f ; Q) ≡ inf ((f − c)χQ )∗ (λ|Q|), where ∗ denotes the decreasing
c∈C
rearrangement for functions.
],D
Definition 51 (Mλ;Q 0
). Let 0 < λ < 1 and Q0 ∈ Q be fixed, and define
the local distributional dyadic sharp maximal operator or the distributional
],D
dyadic sharp maximal operator in Q0 by Mλ;Q 0
f ≡ sup ωλ (f ; Q)χQ for
Q∈D(Q0 )
f ∈ L0 (Rn ).

4.2.2 Distributional maximal function and median

When we consider the Hardy–Littlewood maximal operator, we need to
assume that f is locally integrable; otherwise M f ≡ ∞. This is somewhat
inconvenient. For example, M f, f ∈ L1 (Rn ) \ {0} is merely in the space
164 Morrey Spaces

WL1 (Rn ) = L1,∞ (Rn ), so that it is not always locally integrable. Hence,
we are interested in the maximal operator which yields something for any
measurable function. Recall that the decreasing rearrangement of f ∈ L0 (Rn )
is given as follows:

f ∗ (t) ≡ inf{λ > 0 : |{|f | > λ}| < t} (t > 0).

See Definition 19. We will consider a maximal operator which can be useful
for any measurable function.
For Q ∈ Q, the median of f ∈ L0 (Q), which is close to (χQ f )∗ 2−1 |Q| ,

will be an important role in the sequel.

Definition 52 (Median). Let Q ∈ Q, and f ∈ L0 (Q) be a real-valued func-
tion. Define MED(f ; Q) ≡ {a ∈ R : a satisfies (4.3)} where condition (4.3) is
given by
1
|{x ∈ Q : f (x) > a}|, |{x ∈ Q : f (x) < a}| ≤ |Q|. (4.3)
2
Denote by Med(f ; Q) any element in MED(f ; Q). The quantity Med(f ; Q)
is called the median of f over Q. One can regard Med(f ; Q), as if it were a
mapping Q 7→ Med(f ; Q).
Note that MED(f ; Q) is an interval. As the following example shows,
MED(f ; Q) contains more than 1 point.
Example 77. MED(χ[0,1]×[0,2]n−1 + 2χ[1,2]×[0,2]n−1 ; [0, 2]n ) = [1, 2].
We collect some elementary properties of the median.
Proposition 167. Let f, g ∈ L0 (Q) be real-valued functions defined on a cube
Q. Assume that f, g are finite almost everywhere on Q.
(1) MED(f ; Q) 6= ∅.
(2) Let Φ : R → R be a strictly increasing function. Extend Φ to a continuous
mapping from R to itself. Then

MED(Φ ◦ f ; Q) = {Φ(t) : t ∈ MED(f ; Q)}.

In particular, MED(f + c; Q) = {c + α : α ∈ MED(f ; Q)}.

(3) If f ≤ g, then sup MED(f ; Q) ≤ sup MED(g; Q).
(4) Let 0 < a < 1. If λ ∈ MED(f ; Q), then |λ| ≤ f ∗ (2−1 a|Q|).
Proof
(1) Since lim |{x ∈ Q : f (x) > λ}| = 0, the number
λ→∞
  
|Q|
λ0 ≡ inf λ > 0 : |{x ∈ Q : f (x) > λ}| < ∪ {∞}
2
 Various operators in Lebesgue spaces 165

is finite. If λ < λ0 , then 2|{x ∈ Q : f (x) ≥ λ}| ≥ 2|{x ∈ Q : f (x) >

λ}| ≥ |Q|. Letting λ ↑ λ0 , we have 2|{x ∈ Q : f (x) ≥ λ0 }| ≥ |Q|. Hence,
by passing to the complement again, we obtain 2|{x ∈ Q : f (x) <
λ0 }| ≤ |Q|. If λ > λ0 , then 2|{x ∈ Q : f (x) > λ}| ≤ |Q|. Letting λ ↓ λ0 ,
we obtain 2|{x ∈ Q : f (x) > λ0 }| ≤ |Q|. Thus, λ0 ∈ MED(f ; Q).
(2) This is clear from the definition.

(3) Assume otherwise; sup MED(f ; Q) > sup MED(g; Q). We choose β ∈
(sup MED(g; Q), sup MED(f ; Q)). Then
1
|{x ∈ Q : g(x) > β}| ≤ |{x ∈ Q : g(x) > Med(g; Q)}| ≤ |Q|.
2
Meanwhile,
1
|{x ∈ Q : g(x) < β}| > |Q| ≥ |{x ∈ Q : f (x) < β}|
2
≥ |{x ∈ Q : g(x) < β}|,

which is a contradiction.
(4) Assume otherwise, assume λ > f ∗ (2−1 a|Q|) or λ < −f ∗ (2−1 a|Q|). Then
if λ > f ∗ (2−1 a|Q|), then
1
|{x ∈ Q : f (x) ≤ f ∗ (2−1 a|Q|)}| ≤ |{x ∈ Q : f (x) < λ}| ≤ |Q|
2
since λ ∈ MED(f ; Q). Meanwhile, thanks to Lemma 29,

|{x ∈ Q : f (x) > f ∗ (2−1 a|Q|)}| ≤ |{x ∈ Q : |f (x)| > f ∗ (2−1 a|Q|)}|
≤ λf (f ∗ (2−1 a|Q|))
a
≤ |Q|,
2
which is a contradiction since |Q| > 21 |Q| + a2 |Q|.
If λ < −f ∗ (2−1 a|Q|), then we consider −f instead of f . Thanks to the
above argument, we obtain a contradiction.

Roughly speaking, (χQ f )∗ (τ |Q|) is an average of the function f ∈ L0 (Q)

which is not always locally integrable. We have the following version of the
Lebesgue differentiation theorem:
Lemma 168 (Fujii’s lemma). Let 0 < τ < 1. Let f ∈ L0 (Rn ) be real-valued.
Then lim (f χQ )∗ (τ |Q|) = |f (x)| for almost every x ∈ Rn . That is, for all
Q→x
ε > 0, there exists δ > 0 such that ||f (x)| − (f χQ )∗ (τ |Q|)| < ε whenever a
cube Q satisfies |Q| < δ and x ∈ Q.
166 Morrey Spaces

Note that there is no need to assume that f is integrable.

Proof Since the proof is similar to that of Corollary 169 to follow, we
omit the proof.
The following equality is a version of the Lebesgue differentiation theorem
for the medium.
Corollary 169 (Fujii’s lemma II). Let f ∈ L0 (Rn ) be a real-valued function.
Then there exists a set E of measure 0 with the following property: For all
x ∈ Rn \ E and ε > 0, there exists r > 0 such that for all cubes Q satisfying
x ∈ Q ⊂ Q(x, r) and λ ∈ MED(f ; Q), |f (x) − λ| < ε.
Proof For every j ∈ N, we define
∞
X
sj (x) = 2−j (k − 1)χEj,k (x),
k=−∞

where
Ej,k ≡ x ∈ Rn : 2−j (k − 1) ≤ f (x) < 2−j k .


Then
0 ≤ f (x) − sj (x) ≤ 2−j , Inf(MED(sj ; Q)) ≤ Med(f ; Q)
for every Q ∈ Q(Rn ). Let
∞ [∞  
\ |Ej,k ∩ Q|
F ≡ x ∈ Ej,k : lim =1 .
j=1
|Q|→0,x∈Q |Q|
k=1

Then F differs from Rn by a set E of measure zero.

If x ∈ F = Rn \ E, then by the definition of F , for every j ∈ N, we have
x ∈ Ej,k and 3|Ej,k ∩ Q| > 2|Q| for some k ∈ N and for some sufficiently small
Q. Fix such a cube Q, then MED(sj ; Q) = {2−j (k − 1)}. Since 2|{y ∈ Q :
f (y) ≥ Med(f ; Q)}| ≥ |Q|, there exists y ∈ Ej,k such that Med(f ; Q) ≥ f (y).
Since 2−j (k − 1) = sj (x) = sj (y), we have
|Med(f ; Q) − f (x)| ≤ |Med(f ; Q) − sj (y)| + |f (x) − sj (y)|
≤ Med(f ; Q) − 2−j (k − 1) + f (x) − sj (x)
≤ 2−j+1 .
Since j is arbitrary, we conclude the proof.

4.2.3 Generalized dyadic grid and the Lerner–Hytönen

decomposition
As an application of the distributional maximal operator, we obtain the
Lerner–Hytönen decomposition. Roughly speaking, given any f ∈ L0 (Rn ), we
want its decomposition into better functions. The Calderón–Zygmund decom-
position, dealt with in Section 4.5.2, is insufficient because it requires that f
be locally integrable.
 Various operators in Lebesgue spaces 167

The decomposition is not linear. So we suppose that the decomposition

may heavily depend on f . To have such a decomposition we need the following
observation below.
Theorem 170. Let Q0 be a (right-open) cube. For h ∈ L∞ (Q0 ) ∩ M+ (Rn ),
we write γ0 ≡ mQ0 (h). Fix τ ≥ 2n+1 . Define

Zj ≡ Q : Q ∈ D(Q0 ) : mQ (h) > γ0 τ j



for j ∈ N and Z0 = {Q0k }k∈K0 ≡ {Q0 }. Let Dj ≡

S S
Zj = Q for
Q∈Zj
j ∈ N0 . Considering the maximal cubes {Qjk }k∈Kj ⊂ D(Q0 ) in Zj with
Qjk for j ∈ N0 , where the family
S
respect to inclusion, we can write Dj =
k∈Kj
{Qjk }k∈Kj ⊂ D(Q0 ) is disjoint for each j ∈ N0 . Let Ekj ≡ Qjk \Dj+1 for j ∈ N0
and k ∈ Kj .
(1) We have γ0 τ j < mQj (h) ≤ 2n γ0 τ j for k ∈ Kj with j = 1, 2, . . . and
k
γ0 = mQ0k (h) for k ∈ K0 .

(2) The family E ≡ {Ekj }j∈N0 ,k∈Kj is disjoint, decomposes Q0 and satisfies
τ |Ekj | 1
|Qjk | ≤ ≤ |Ekj | for all j ∈ N0 and k ∈ Kj .
τ − 2n 2
Proof
(1) If j = 0, then the assertion is clear from the maximality. Let us suppose
j ≥ 1. The left inequality is a consequence of the fact that we chose Qjk
from Zj . If we consider the dyadic parent R of Qjk , then mR (h) ≤ γ0 τ j .
Otherwise, instead of Qjk R would have been chosen as an element of
Zj . Since mQj,k (h) ≤ 2n mR (h), we obtain the right inequality.
(2) A geometric observation shows that two dyadic cubes never intersect
unless one is not included in the other. We freeze j ∈ N and k ∈ Kj .
From the observation above, we have
[ n j+1 o
Dj+1 ∩ Qjk = Qk∗ : k ∗ ∈ Kj+1 , Qj+1
k ∗ ⊂ Qj
k

because Qj+1 j
k∗ ) Qk never happens thanks to the maximality
Z of Qjk and
the fact that τ ≥ 2n+1 . Note that |Qj+1
k∗ |γ0 τ
j+1
≤ h(x)dx, since
Qj+1
k∗

Qj+1
k∗
∗
∈ Zj+1 . If we add this estimate over Zall k ∈ Kj+1 such that
j+1
Qk∗ ⊂ Qjk , we obtain |Dj+1 ∩ Qjk |γ0 τ j+1 ≤ h(x)dx ≤ |Qjk |2n γ0 τ j .
Qjk
Consequently, if we consider the complement of Dj+1 , then we obtain
the desired result.
168 Morrey Spaces

Although the set of dyadic cubes has a good geometric property, there
are many dyadic cubes that dyadic cubes cannot cover: there does not exist a
dyadic cube R such that Q(1) cannot be covered by R. However, if we consider
2n translated families, we can have this property.
Definition 53 (Da ). Let a = (a1 , a2 , . . . , an ) ∈ {0, 1/3}n . The generalized
dyadic grid Da is defined by Da ≡ {Q : Q − (−1)log2 `(Q) `(Q)a ∈ D}. If
n = 1, then we write D+ ≡ D(1) .

Example 78. Let n = 1. Then

       
1 4 4 7 7 10 10 13
..., , , , , , , , , . . . ∈ D+
3 3 3 3 3 3 3 3

and
         
1 1 1 5 5 4 4 11 11 7
..., − , , , , , , , , , , . . . ∈ D+ .
6 3 3 6 6 3 3 6 6 3

Furthermore, [−1, 1] ⊂ − 38 , 43 ∈ D+ .


The family Da , a ∈ {0, 1/3}n is useful because we can control any cubes
in the following sense:
Lemma 171. Let Q be any cube. Then there exist a = (a1 , a2 , . . . , an ) ∈
{0, 1/3}n and R ∈ Da ∩ Q] (Q) such that |R| ≤ 6n |Q|.
Proof We work in R; n = 1. By translation, we may assume Q = (a, b),
where 0 < a < b. Choose k0 ∈ Z such that 2−k0 −1 ≤ 3(b−a) < 2−k0 . If (a, b)∩
2−k0 Z = ∅, then (a, b) is contained in an interval of the form 2−k0 [m0 , m0 +1).
Thus, we may assume (a, b) ∩ 2−k0 Z 6= ∅. Let m0 2−k0 ∈ (a, b). Then m2−k0 ∈ /
(a, b); otherwise b − a ≥ 2−k0 and this is a contradiction. Then

(a, b) ⊂ (m0 2−k0 − 2−k0 /3, m0 2−k0 + 2−k0 /3)

⊂ (m0 2−k0 − 21−k0 /3, m0 2−k0 + 2−k0 /3).

Thus, if we choose R = (m0 2−k0 − 21−k0 /3, m0 2−k0 + 2−k0 /3) ∈ D+ , then we
have `(R) = 2−k0 ≤ 6(b − a) = 6`(Q).
Motivated by the generalized dyadic grid Da , we introduce the notion of
generalized dyadic grids.
Definition 54 (Generalized dyadic grid). A generalized dyadic grid D is the
set of all cubes of the form

x + [0, 2l )n (x ∈ Rn , l ∈ Z)

with the property: If Q, R ∈ D, then Q ∩ R ∈ {Q, R, ∅}.

 Various operators in Lebesgue spaces 169

In addition to generalized dyadic grids, we need the notion of sparse fami-

lies. Roughly speaking, the notion of sparse families extends the one of disjoint
families. A sparse family is an “almost disjoint” one. We need some flexibility
in the disjointness because in many cases it is difficult to obtain the disjoint-
ness.
Definition 55 (Sparse family).
(1) Let τ > 0. A set of cubes A is τ -sparse, if there exists a disjoint collection
{K(Q)}Q∈A such that K(Q) contained in Q and that satisfies |K(Q)| ≥
τ |Q| for each Q ∈ A. If there is no need to specify the precise value of
τ , A is called simply sparse. Each K(Q) is called the nutshell of Q.
(2) Let a ∈ {0, 1/3}n be fixed, and let τ > 0. Suppose that we have a family
Sj = {Qjk }k∈Kj of disjoint cubes in Da for each j ∈ N0 . Assume in
Qj+1 Qjk and that |Qj+1
S S P
addition that k ⊂ k ∩ R| ≤ τ |R|
k∈Kj+1 k∈Kj k∈Kj+1
∞
S
for all R ∈ Sj , j ∈ N0 . Then Sj is a τ -sparse family with a level
j=0
structure.
If there is no need to specify τ , we omit τ .
Example 79.
(1) Any family Q of disjoint cubes is 1-sparse. In fact for any cube Q we
can choose its nutshell K(Q) = Q. In particular, if Q is a cube, then
{Q} is a sparse family.
(2) In Theorem 170, the family {Q0 } ∪ {Qjk }j∈N,k∈Kj is a sparse family of
cubes with a level structure. Furthermore, we have a decomposition in
the sense of the almost everywhere pointwise estimate:
∞ X
X
χ Q0 M f ≤ γ0 τ χ E0 + γ0 τ k+1 χE j ,
k
j=1 k∈Kj

where E 0 is the nutshell of Q0 and each Ekj is the nutshell of Qjk .

(3) The family {3Q}Q∈D0 is a 3−n -sparse family with K(Q) = Q. This is a
misleading but important example.
(4) Let 0 < η ≤ 1. Suppose that famlies Sj , j ∈ N0 of disjoint
P we haveP
dyadic cubes satisfying χQ ≤ χQ ≤ 1 for all j ∈ N0 and
Q∈Sj+1 Q∈Sj
P ∞
S
|Q ∩ R| ≤ η|R| for all R ∈ Sj and j ∈ N0 . Then Sj is an
Q∈Sj+1 j=0
η-sparse family with a level structure.
We generalize Theorem 170 as follows:
170 Morrey Spaces

Theorem 172. Let Q0 be a (right-open) cube, and let 0 ≤ α < n. Write a0 ≡

α
|Q0 | n m3Q0 (f ) and a ≡ 18n+1 . For each j ∈ N0 and f ∈ L∞ 0 + n
c (Q ) ∩ M (R ),
define
α
[
Dj ≡ Q ∈ D(Q0 ) : |Q| n m3Q (f ) ≥ a0 aj (⊂ Q0 ).

Considering the maximal cubes Qjk , k ∈ Kj with respect to inclusion, we can

∞
[
write Dj = Qjk , where the cubes Qjk , k ∈ Kj are pairwise disjoint. Then
k=1
S ≡ {Qjk }j∈N0 ,k∈Kj is sparse.
Note that D0 = Q0 in the above.
Proof Let j ∈ N and k ∈ Kj . By the maximality of each Qjk , we see that
α
a0 aj ≤ |Qjk | n m3Qj (f ) ≤ 2n a0 aj .
k

Let Ekj≡ Qjk\ Dj+1 . Then the family {Ekj }j∈N0 ,k∈Kj is pairwise disjoint and
decomposes Q0 . Consequently, we need only verify that 2|Ekj | ≥ |Qjk |. Notice
that, if k 0 ∈ Kj+1 satisfies Qj+1 j
k0 ⊂ Qk , then
α α α
a0 ak+1 ≤ |Qj+1
k0 | m3Qj+1
n
0
(f ) ≤ |Qjk | n m3Qj+1
0
(f ) ≤ |Qjk | n M [f χ3Qj ](x)
k k k

for all x ∈ Qj+1

k0 . This entails
  ( )
k+1
[ a0 a
Qjk ∩ Dj+1 j
= Qk ∩  j+1
Qk̃  ⊂ x ∈ Rn : M [f χ3Qj ](x) ≥ .
|Qjk | n
α
k
k̃∈Kj+1

The weak-(1, 1) boundedness of M (see Theorem 134) yields

α
|Qjk | n
Z
j n
|Qk ∩ Dj+1 | ≤ 3 · · f (x)dx
a0 ak+1 3Qjk
α
n n−α k |Qjk | n 3n |Qjk |
≤3 ·2 a0 a · ·
a0 ak+1 |Qjk | αn
18n j
= |Qk |
a
1
≤ |Qjk |,
2
which implies 2|Ekj | ≥ |Qjk |. Thus S ≡ {Qjk }j∈N0 ,k∈Kj is sparse.
Likewise, we can prove the following:
Theorem 173. Let Q0 be a (right-open) cube, and let 0 ≤ α < n. Write a0 ≡
α
|Q0 | n mQ0 (f ) and a ≡ 18n+1 . For each j ∈ N0 and f ∈ L∞ 0 + n
c (Q ) ∩ M (R ).
define
α
[
Dj ≡ Q ∈ D(Q0 ) : |Q| n mQ (f ) ≥ a0 aj (⊂ Q0 ).
 Various operators in Lebesgue spaces 171

Considering the maximal cubes Qjk , k ∈ Kj with respect to inclusion, we can

[ j
write Dj = Qk , where the cubes Qjk , k ∈ Kj are pairwise disjoint. Then
k∈Kj
j
S≡ {Qk }j∈N0 ,k∈Kj is sparse.
Note again that D0 = Q0 in the above.
The next inequality is a very strong theorem to control the singularity of
functions. The following theorem is called the Lerner–Hytönen decomposition
of the function, which is important. For a cube R we define
ω2−n−2 (f ; R) ≡ (χR (f − Med(f ; R)))∗ (2−n−2 |R|).
Although there is ambiguity of the definition of Med(f ; R), we fix one of them
n
and define ω2−n−2 (f ; R). For a right-open cube Q0 ≡
Q
[aj , bj ), recall that
j=1
we defined the dyadic cubes D(Q0 ) with respect to Q0 by bisecting Q0 in a
finite number of steps.
Theorem 174 (Lerner–Hytönen decomposition). Let f : Q0 → R be a mea-
surable function defined on a right-open cube Q0 . Then there exists a sparse
family {Qjk }j∈N0 , k∈Kj ⊂ D(Q0 ) with a level structure such that {Q0k }k∈K0 =
{Q0 } and that
∞ X
X
χQ0 |f − Med(f ; Q0 )| ≤ ω2−n−2 (f ; Qjk )χQj .
k
j=0 k∈Kj

Here, the inequality is understood as the one for almost every point.
For the proof we use the following notation: Recall that given a cube Q,
D1 (Q) denotes the set of all dyadic children. We rely on a somewhat distorted
stopping time argument. For any fixed measurable function f on a fixed cube
Q0 and Q ∈ D(Q0 ), let S0 (Q) ≡ {Q}, and define S1 (Q) ⊂ D(Q0 ) to be the
collection of all maximal dyadic subcubes R of Q such that
max |Med(f, Q0 ) − Med(f ; Q0 )| > ω2−n−2 (f ; Q). (4.4)
Q0 ∈D1 (R)

Suppose that we have defined S0 (Q), S

S1 (Q), S2 (Q), . . . , Sk−1 (Q) for each
Q ∈ D(Q0 ). Then define Sk (Q) ≡ S1 (R) for each Q ∈ D(Q0 ).
R∈Sk−1 (Q)
The proof of Theorem 174 hinges on the repeated use of the following
lemma:
Lemma 175. Let f ∈ L0 (Rn ) be a real-valued function defined on a cube Q0 .
(1) Almost everywhere
χQ0 |f − Med(f ; Q0 )|
X
≤ ω2−n−2 (f ; Q0 )χQ0 + χQ |f − Med(f ; Q)|.
Q∈S1 (Q0 )
172 Morrey Spaces
[
(2) Write Ω(Q0 ) ≡ Q ⊂ Q0 . Then 2|Ω(Q0 )| ≤ |Q0 |.
Q∈S1 (Q0 )

Once we prove Lemma 175, for each J ∈ N, we have

J
X X X
χQ0 |f − Med(f ; Q0 )| ≤ ω(f ; Q)χQ + χQ |f − Med(f ; Q)|
j=0 Q∈Sj (Q) Q∈SJ (Q)

|Q| ≤ 2−J |Q0 |. Hence, letting J → ∞, we have

P
and
Q∈SJ (Q)

∞
X X
χQ0 |f − Med(f ; Q0 )| ≤ ω(f ; Q)χQ
j=0 Q∈Sj (Q)

almost everywhere. Thus, the proof of Theorem 174 will be complete once we
prove Lemma 175.
Proof of Lemma 175 We may assume that Med(f ; Q0 ) = 0 by con-
sidering f − Med(f ; Q0 ) instead of f itself. In this case ω2−n−2 (f ; Q0 ) =
(χQ0 f )∗ (2−n−2 |Q0 |).
(1) We distinguish two cases. We first assume x ∈ Ω(Q0 ) and then assume
x ∈ Q0 \ Ω(Q0 ).
Fix x ∈ Ω(Q0 ). For the disjoint subcubes S1 (Q0 ) of Q0 , we write
X
A(x) ≡ Med(f ; Q)χQ (x),
Q∈S1 (Q0 )
X
B(x) ≡ χQ (x)(f (x) − Med(f ; Q)),
Q∈S1 (Q0 )

so that χΩ(Q0 ) (x)f (x) = A(x) + B(x). Note that the quantity B(x)
matches the second term in the right-hand side of the conclusion; we
have only to handle A(x). From the maximality it follows that any Q ∈
S1 (Q0 ) satisfies the opposite estimate in place of Q0 ∈ D(Q). That is,
|Med(f ; Q)| ≤ ω2−n−2 (f ; Q0 ). Thus, since S1 (Q0 ) is disjoint, |A(x)| ≤
ω(f ; Q0 )χΩ(Q0 ) (x).
Let x ∈ Q0 \ Ω(Q0 ) instead. Then we have |Med(f ; R)| ≤ ω2−n−2 (f ; Q0 )
for any cube R ∈ D(Q0 ) containing x. It should be noted that, for any
x ∈ Q0 \ Ω(Q0 ), such a cube R can be chosen as small as we wish. Thus,
|f (x)| ≤ ω2−n−2 (f ; Q0 ) for almost all x ∈ Q0 \Ω(Q0 ) by the Fujii lemma;
see Corollary 169.
(2) Let Q ∈ S1 (Q0 ) be arbitrary, so that some Q0 ∈ D1 (Q) satisfies

|Med(f ; Q0 )| > ω2−n−2 (f ; Q0 ). (4.5)

 Various operators in Lebesgue spaces 173
1 1
Let < ν < . Since Q0 is a dyadic child of Q, |Q| = 2n |Q0 |. Hence
4 2
|Med(f ; Q0 )| ≤ (χQ0 f )∗ (ν|Q0 |) ≤ (χQ f )∗ (2−n ν|Q|). (4.6)
Thus we obtain ω2−n−2 (f ; Q0 ) < (χQ f )∗ (2−n ν|Q|) combining (4.5) and
(4.6). In view of Lemma 30, we have
ν|Q|
|Q ∩ {|f | > ω2−n−2 (f ; Q0 )}| ≥ |Q ∩ {|f | ≥ (χQ f )∗ (2−n ν|Q|)}| ≥
.
2n
If we add this estimate over Q ∈ S1 (Q0 ) and use Lemma 29, then we
obtain
ν ν X
n
|Ω(Q0 )| = n |Q|
2 2 0
Q∈S1 (Q )

≤ |Q ∩ {|f | > ω2−n−2 (f ; Q0 )}|

0

= |Q0 ∩ {|f | > (χQ0 f )∗ (2−n−2 |Q0 |)}|

≤ 2−n−2 |Q0 |.
It remains to let ν ↑ 12 .
We conclude Section 4.2.3 with the remark on the quantity ω2−n−2 (f ; Q).
Remark 6. Let f be a measurable function defined on a cube Q. Then for
any c ∈ R, ω2−n−1 (f ; Q) ≤ 2(χQ (f − c))∗ (2−n−2 |Q|). In fact,
χQ (x)|f (x) − Med(f ; Q)| ≤ χQ (x)|f (x) − c| + |Med(f ; Q) − c|
≤ χQ (x)|f (x) − c| + (χQ (f − c))∗ (2−n−2 |Q|).
If one considers the distribution function, then one obtains the desired result.

4.2.4 Exercises
Exercise 65. We let
D2k ≡ {[j · 4k − 2 · 4k /3, j · 4k + 4k /3)}j∈Z
D2k+1 ≡ {[2j · 4k − 2 · 4k /3, 2j · 4k + 2 · 4k /3)}j∈Z
for k ∈ Z. Let 0 < q ≤ p < ∞.
(1) [273] Show that D ≡ {Dk }∞
k=−∞ is a generalized dyadic grid.

(2) [273] Show that for any compact set K there exist k ∈ Z and Q ∈ Dk
such that K ⊂ Q.
Z  q1
1
− q1 q
(3) Show that kf kMq ∼ sup sup |Q|
p p |f (y)| dy for f ∈ L0 (Rn ).
k∈Z Q∈Dk Q
0
Exercise 66. Let f ∈ L (Q0 ) with Q0 ∈ Q, and let λ ∈ (0, 1).
(1) Show that ωλ (f ; Q0 ) = 0 if and only if f is constant.
],D
(2) Show that Mλ;Q 0
f = 0 if and only if f is constant.
174 Morrey Spaces

4.3 Fractional maximal operators

The fractional maximal operator, which we define below in Section 4.3.1,
substitutes for fractional integral operators. These operators behave similarly.
However, in this book, we encounter the difference between these two things.
We will obtain the local estimate and the sparse estimates in Sections 4.3.2
and 4.3.3, respectively.

4.3.1 Fractional maximal operators

The fractional maximal operator Mα of order α ∈ [0, n) is defined by
Z
Mα f (x) ≡ sup χQ (x)`(Q)α−n |f (y)|dy
Q∈Q Q

0 n
for f ∈ L (R ), so that M = M0 is the Hardy–Littlewood maximal operator.
In connection with the Hardy–Littlewood maximal operator, we will consider
this generalization in Section 4.3.1.
Example 80. It seems that Morrey spaces are useful when we consider the
boundedness property of the fractional maximal operator Mα , 0 < α < n. In
n
fact, Mα maps M1α (Rn ) boundedly to L∞ (Rn ).
As is the case with the Hardy–Littlewood maximal operator, we have the
endpoint weak estimate.
Theorem 176. Let 0 < α < n. Then kMα f kWL n−α
n . kf kL1 for all f ∈
1 n
L (R ).
Proof Simply modify the case of α = 0; see Exercise 67.
The following theorem, called the Hardy–Littlewood–Sobolev theorem, is
fundamental.
1 1
Theorem 177. Let 1 < p < q < ∞ and 0 < α < n satisfy q = p −α
n . Then
kMα f kLq . kf kLp for all f ∈ Lp (Rn ).
Before the proof we remark that arithmetic shows
   0
n (n − α)p
q− × = q.
n−α nu
Proof By the monotone convergence theorem we may assume that f ∈
L∞ n
c (R ); see Exercise 68. For λ > 0, we let Ωλ ≡ {Mα f > λ}. We observe
that
Z ∞ Z ∞
n n
(kMα f kLq )q = q λq−1 |Ωλ |dλ . λq−1− n−α (kf χΩλ kL1 ) n−α dλ
0 0
 Various operators in Lebesgue spaces 175

by the use of Theorem 176. We choose  u ∈ (0, 1) sothat

u
n(1 − u) = pα. Then
by Hölder’s inequality, kf χΩλ kL ≤ kf χΩλ kL n−α
1 nu (kf kLp )1−u . Thus,
Z ∞  nu
 n−α
n n(1−u)
(kMα f kL ) .
q
q
λq−1− n−α kf χΩλ kL n−α nu (kf kLp ) n−α dλ
0
Z
n(1−u) nu n
' (kf kLp ) n−α |f (x)| n−α Mα f (x)q− n−α dx.
Rn

By Hölder’s inequality, we obtain

Z
nu n nu n
|f (x)| n−α Mα f (x)q− n−α dx ≤ (kf kLp ) n−α (kMα f kLq )q− n−α .
Rn
n n
As a result, (kMα f kLq )q . (kf kLp ) n−α (kMα f kLq )q− n−α . If we arrange this
inequality, we conclude kMα f kLq . kf kLp .

4.3.2 Local estimates for the maximal operators and the

fractional maximal operators
Here, we collect the local estimates for Mα for 0 ≤ α < n. We tolerate the
case α = 0 here.
Theorem 178. Let 0 ≤ α < n, 1 < p1 < ∞, 0 < p2 < ∞ and γ ∈ R.
(1) Suppose 1 < p1 ≤ p2 < ∞. Let W (x) ≡ (1 + |x|)p1 γ , x ∈ Rn . Then the
inequality
kMα f kLp2 (Q(1)) . kf kLp1 (W ) = kf · W 1/p1 kLp1 , (4.1)
where the implicit constant is independent of f holds for all f ∈ L0 (Rn )
if and only if  
1 1 n
n − ≤α≤ + γ. (4.2)
p1 p2 + p1
(2) Let 1 < p1 < ∞ and 0 < p2 < ∞ satisfy (4.2). For r > 0, we let
p1 γ
Wr (x) ≡ (r + |x|)p1 γ ∼ rp1 γ M χB(r) (x)− n for x ∈ Rn . Then
n n
kMα f kLp2 (Q(r)) . rα− p1 + p2 −γ kf kLp1 (Wr ) (4.3)
for all r > 0 and for all f ∈ L0 (Rn ). In particular,
 p1
|f (x)|p1
Z
n 1
kMα f kLp2 (Q(r)) . r 2
p
n−αp
dx ,
Rn (|x| + r) 1

or equivalently
(Z ! ) p1
∞ Z 1
n
p1 dt
kMα f kLp2 (Q(r)) . r p2
|f (x)| dx (4.4)
r B(t) tn−αp1 +1

for all r > 0 and for all f ∈ L0 (Rn ).

176 Morrey Spaces

Proof Inequality (4.3) is a conseqneuce of (4.1) and the dilation argument.

Consequently, we discuss the equivalence between (4.1) and (4.2).
n n
Assume (4.2). Choose q1 ∈ (1, p1 ] and q2 ∈ [p2 , ∞) so that α = − .
q1 q2
n n
Such a choice of q1 and q2 is possible because − ≤ α Then by the
p1 p2
Hardy–Littlewood–Sobolev theorem and Hölder’s inequality,

kMα [χQ(3) f ]kLp2 (Q(1)) . kMα [χQ(3) f ]kLq2 (Q(1))

. kf kLq1 (Q(3))
. kf kLp1 (Q(3))
. kf kLp1 (W ) ,

and similar to Lemma 130

Z
1
kMα [χRn \Q(3) f ]kLp2 (Q(1)) . sup |f (y)|dy
r≥3 rn−α Q(r)\Q(3)

. sup rα mpQ(r)
1
(f )
r≥1

. kf kLp1 (W ) ,

Thus, (4.1) holds.

Conversely, assume that (4.1) holds. Then
Z
1
|f (y)|dy . kf kLp1 (W )
rn−α Q(r)\Q(3)

for all f ∈ L0 (Rn ), or equivalently,

Z
1
|g(y)|(1 + |y|)−γ dy . kgkLp1
rn−α Q(r)\Q(3)

for all g ∈ L0 (Rn ) and r ≥ 3. Consequently, we obtain

! 1
Z p01
1 −p01 γ
sup (1 + |y|) dy < ∞,
r>0 rn−α Q(r)\Q(3)

n
which forces −n + α − γ + ≤ 0. If p1 6= p2 , use the truncated power function
p01
n n
| · |β χB(1) to have α ≥ p1 − p2 .
Example 81. Let 1 < p < ∞. Fix r > 0 and x ∈ Rn . We consider the
following “partial maximal function” M r f (x) ≡ sup mB(x,t) (|f |) together with
t≥r
(p) (p)
its powered version: M r f (x) ≡ sup mB(x,t) (f ). Then one can prove that
t≥r

n
kM f kLp (B(x,r)) ' kM (f χB(x,2r) )kLp (B(x,r)) + r p M 2r f (x).
 Various operators in Lebesgue spaces 177
n n (p)
Hence r p M r f (x) . kM f kLp (B(x,r)) . r p M 2r f (x). Note that
! ! p1
Z ∞ Z
(p) n
p dt
M 2r f (x) .r p |f (y)| dy
r B(x,t) tn+1

thanks to (4.4).

4.3.3 Sparse estimate for fractional maximal operators

Here, we obtain a sparse estimate for fractional maximal operators. As
shown later, the sparse estimate will be of great use.
Theorem 179. Let 0 ≤ α < n. Suppose that we have f ∈ L∞ n +
c (R ) ∩ M (R )
n
0 0
and a (right-open) cube Q . For a sparse family S ⊂ D(Q ) with the nutshell
{EQ }Q∈S , write
α
D ] (Q0 ) α
X
LSα f ≡ |Q| n m3Q (f )χEQ and c∞,α (f ) ≡ sup |Q| n m3Q (f ).
Q∈S Q∈D ] (Q0 )

Then, there exists a sparse family S = Sf ⊂ D(Q0 ) that depends on f such

D ] (Q0 )
that Mα f (x) .α,n LSα f (x) + c∞,α (f ) for all x ∈ Q0 .
It matters that the sparse family depends on f .
Proof We follow the construction in Theorem 172. Recall that D(Q0 )
is the set of dyadic cubes of Q0 . We may suppose that f is not zero
almost everywhere; otherwise simply take S = {Q0 }. We write M fα f (x) ≡
α
sup χQ (x)|Q| n m3Q (f ) for each x ∈ Rn , so that for all x ∈ Q0 , Mα f (x) ≤
Q∈D(Q0 )
D ] (Q0 )
fα f (x) + c∞,α (f ). We set S ≡ {Qj }j∈N ,k∈K . Since f ∈ L∞ (Rn ) \ {0},
CM k 0 j c
we have x ∈ Dj \ Dj+1 for some j ∈ Z. Since x ∈ Dj , x ∈ Qjk for some k ∈ Kj .
fα f (x) ≤ a0 aj+1 ≤ a|Qj | αn m j (f ).
Thus, x ∈ Ekj ⊂ Rn \ Dj+1 . In this case, M k 3Q k

Since the family {Ekj }j∈N0 ,k∈Kj is pairwise disjoint and decomposes Q0 , we
fα f (x) . LSα f (x) for all x ∈ Q0 . This completes the proof.
conclude that M
We have the following variant of Theorem 179:
Remark 7. Let Da be the dyadic grid with a ∈ {0, 1/3}n . Denote by D]a (Q0 )
the set of all cubes in Da containing a cube Q0 . Then define
D] (Q0 )
Mα a f≡ sup `(Q)α mQ (|f |)χQ
Q∈D]a (Q0 )

for f ∈ L0 (Rn ). For a sparse family S ⊂ D]a (Q0 ) with the nutshell {EQ }Q∈S ,
write
α D]a (Q0 ) α
X
LSα f ≡ |Q| n mQ (f )χEQ and c∞,α (f ) ≡ sup |Q| n mQ (f ).
Q∈S Q∈D]a (Q0 )
178 Morrey Spaces

Then, for any fixed cube Q0 ∈ Q, there exists a sparse family S = Sf ⊂ D(Q0 )
D] (Q0 )
such that, for all x ∈ Q0 , MαDa f (x) .α,n LSα f (x) + c∞,α
a
(f ).

4.3.4 Exercises
n−α n−α n−α
Exercise 67. Prove Theorem 176 by the use of (a + b) n ≤a n +b n

for a, b ≥ 0 and 0 ≤ α < n.

Exercise 68. Let {fj }∞
j=1 ⊂ M
+
(Rn ) be an increasing sequence, and let
 
0 ≤ α < n. Then show that Mα lim fj = lim Mα [fj ].
j→∞ j→∞

4.4 Fractional integral operators

One of the crucial properties of Morrey spaces is that Morrey spaces can
describe the boundedness property of fractional integral operators more pre-
cisely than Lebesgue spaces. The fractional integral operators come about
when we consider the inverse of Laplacian: the fractional integral operator of
order 2 is a fundamental solution of the Laplace equation.
The aim of Section 4.4 is to introduce fractional integral operators and
then to discuss their Lp (Rn )-boundedness property.
We define and investigate fractional maximal operators in Section 4.4.1.
We expand what we obtained in Section 4.4.1 to the local setting and to the
sparse setting in Sections 4.4.2 and 4.4.3. Sections 4.4.4 and 4.4.5 will explain
why fractional integral operators are useful. Section 4.4.5, which handles the
Bessel operator, does not fall under the scope of other sections in Section 4.4.
However, since it is related to fractional integral operators, we deal with the
Bessel operator in Section 4.4.5.

4.4.1 Fractional integral operators on Lebesgue spaces

Another important class of singular integral operators is the fractional
integral operator Iα (of order α) defined by
Z
f (y)
Iα f (x) ≡ n−α
dy (x ∈ Rn ). (4.1)
R n |x − y|

Here, 0 < α < n. Section 4.4.1 examines the close connection between Mα
and Iα .
Example 82. Let 0 < α < n and 0 < β < n. If f (x) ≡ |x|−β χB(1) (x) for
x ∈ Rn , then a simple calculation shows Iα f (x) ∼ |x|−α−β as x → 0.
 Various operators in Lebesgue spaces 179

Example 83. Let 0 < α < n. Observe that Iα χB(1) (x) > 0 for all x ∈
Rn and Iα χB(1) ∈ L∞ (Rn ). Furthermore, Iα χB(1) (x) ∼ |x|α−n for |x| ≥ 2.
Consequently, Iα χB(1) (x) ∼ (|x| + 1)α−n for x ∈ Rn . Thus a dilation and
translation argument yields Iα χB(x0 ,r) (x) ∼ rα (|x − x0 | + r)α−n . A similar
conclusion Iα χQ (x) ∼ `(Q)n (|x − c(Q)| + `(Q))α−n holds for cubes.
Example 84. Let 0 < α < n and f ∈ M+ (Rn ). Using Lemma 161, we write
Z ∞ Z !
1
Iα f (x) = (n − α) f (y)dy dl.
0 ln+1−α B(x,l)

This formula will be used frequently to analyze Iα f .

One of the fundamental questions about Iα f (x) is whether the integral
defining Iα f (x) converges absolutely.
Here, we will present the necessary and sufficient condition for the integral
above to converge for almost all x ∈ Rn .
Lemma 180. Let f ∈ L1loc (Rn ). The integral defining Iα f (x) converges abso-
lutely for almost all x ∈ Rn if and only if
|f (y)|
Z
dy < ∞. (4.2)
R n 1 + |y|n−α

Proof Suppose first that (4.2) holds. Let us check that the integral con-
verges for almost all y ∈ B(x, 1), where x ∈ Rn is chosen arbitrarily. It is not
so hard to see that
Z
Iα [χB(x,1) |f |](z)dz < ∞,
B(x,1)

which implies Iα [χB(x,1) |f |](y) < ∞ for almost all y ∈ B(x, 1) thanks to Theo-
rem 182(1) to follow. Therefore, it remains to show that Iα [χRn \B(x,1) |f |](y) <
∞ for almost all y ∈ B(x, 1).
Note that (4.2) implies

|f (y)|
Z
dy < ∞. (4.3)
R n 1 + |x − y|n−α
Thus
|f (y)|
Z
Iα [χRn \B(x,1) |f |](y) ≤ dy
Rn \B(x,1) |x − y|n−α
|f (y)|
Z
≤ dy
Rn \B(x,1) 1 + |x − y|n−α
< ∞.

Thus, Iα [|f |](y) < ∞ for almost all y ∈ B(x, 1).

180 Morrey Spaces

Assume instead that Iα [|f |](x) < ∞ for almost all x ∈ Rn . Then choose a
point x0 ∈ Rn such that Iα [|f |](x0 ) < ∞. Then, we have
|f (y)|dy |f (y)|dy
Z Z
n−α
n−α
≤ 2 n−α
Rn \B(x0 ,2|x0 |) 1 + |y| Rn \B(x0 ,2|x0 |) |x0 − y|

≤ 2n−α Iα [|f |](x0 ) < ∞. (4.4)

Since f is assumed locally integrable, we obtain

|f (y)|
Z Z
n−α
dy ≤ |f (y)|dy < ∞. (4.5)
B(x0 ,2|x0 |) 1 + |y| B(x0 ,2|x0 |)

Combining (4.4) and (4.5) gives (4.2).

The part (2) of the next lemma is known as the Hedberg lemma or the
Hedberg inequality [190].
n 1 1 α
Lemma 181. Let f ∈ Lp (Rn ) with 1 ≤ p < . Define s > 0 by = − .
α s p n
(1) The integral defining Iα f (x) converges absolutely for almost everywhere
x ∈ Rn .
p p
(2) The estimate |Iα f (x)| . M f (x) s (kf kLp )1− s holds for a.e. x ∈ Rn .
Proof It suffices to prove (2); (1) follows from the Lp (Rn )-boundedness
of M if p > 1 and from the weak Lp (Rn )-boundedness of M if p = 1. We can
also assume that f ∈ M+ (Rn ) since the integral kernel of Iα is positive. We
have only to insert
Z Z Z ! p1
1 1 1 p
f (y)dy . M f (x), n f (y)dy . f (y) dy .
ln B(x,l) l B(x,l) ln B(x,l)

into Example 84 to obtain the pointwise estimate. More precisely

Z ∞ Z !
1
Iα f (x) ≤ (n − α) f (y)dy d`
0 `n−α+1 B(x,`)
Z ∞

Z ! p1 
1 d`
. min M f (x), n f (y)p dy  1−α
0 l B(x,l) `
Z ∞  
1 d`
. min M f (x), n kf kLp 1−α
0 lp `
p p
. M f (x) s (kf kLp )1− s .

Thus, the proof is complete.

Thanks to the Lp (Rn )-boundedness of M with 1 < p < ∞ and the weak
L1 (Rn )-boundedness of M , we obtain:
 Various operators in Lebesgue spaces 181

Theorem 182 (Hardy–Littlewood-Sobolev). Let 0 < α < n.

n−α
(1) For all f ∈ L1 (Rn ) and λ > 0, λ|{x ∈ Rn : |Iα f (x)| > λ}| n . kf kL1 .
n 1 1 α
(2) Assume that the parameters p and q satisfy 1 < p < and = − .
α q p n
Then kIα f kLq . kf kLp for all f ∈ Lp (Rn ).
Proof Simply combine Theorems 134 and 141 and Lemma 181.
We end this section with the relation between the fractional integral oper-
ator Iα and the fractional maximal operators Mα+ε and Mα−ε , which is a
good contrast to the relation between the fractional integral operator Iα and
the fractional maximal operator Mα .
Lemma 183. Let 0 < α < n, and let ε ∈ (0, min(α, n − α)). Then we have
p
Iα f (x) . Mα−ε f (x)Mα+ε f (x) (x ∈ Rn )

for all f ∈ M+ (Rn ).

Proof Simply insert the estimate
Z
1
f (y)dy ≤ min(`−ε Mα−ε f (x), `ε Mα+ε f (x)) (` > 0)
`n−α B(x,`)

into Example 84.

We can generalize Lemma 183 as follows: Let 0 ≤ α1 < α < α2 < n. Then,
for all f ∈ M+ (Rn ),
α2 −α α−α1
Iα f (x) . Mα1 f (x) α2 −α1 Mα2 f (x) α2 −α1 (x ∈ Rn ).

4.4.2 Local estimates for the fractional integral operators

We collect local estimates for fractional integral operators.
Theorem 184. Let 0 < α < n, 1 ≤ p1 ≤ p2 ≤ ∞ and γ ∈ R. We define
W (x) ≡ (1 + |x|)p1 γ , x ∈ Rn .
(1) If 1 < p1 ≤ p2 < ∞, then the inequality
1
kIα f kLp2 (Q(1)) . kf kLp1 (W ) ≡ kf · W p1 kLp1 , (4.6)
+ n
 constantis independent of f holds for all f ∈ M (R ),
where the implicit
1 1 n
if and only if n − ≤α< + γ.
p1 p2 p1
(2) If p1 = 1 ≤ p
2 < ∞, then inequality (4.6) holds for all f ∈ M+ (Rn ) if
1
and only if n 1 − < α ≤ n + γ.
p2
182 Morrey Spaces

(3) If 1 < p1 ≤ ∞, then inequality (4.6) with p2 = ∞ holds for all f ∈

n n
M+ (Rn ) if and only if <α< + γ.
p1 p1
(4) If p1 = 1 and p2 = ∞, then (4.6) fails for any 0 < α < n and γ ∈ R.
Remark that (4) reveals the difficulty of fractional integral operators and
that at the same time (4) is the source of the interest of analyzing fractional
integral operators.
1 1 α
Proof Define p3 by = − . Let f ∈ M+ (Rn ).
p3 p1 n
 
1 1 n
(1) Assume n − ≤ α < + γ, so that p2 ≤ p3 . Consequently,
p1 p2 p1
by Hölder’s inequality and the Hardy–Littlewood–Sobolev theorem, we
obtain kIα [χQ(3) f ]kLp2 (Q(1)) . kIα [χQ(3) f ]kLp3 . kf kLp3 ≤ kf kLp1 (W ) .
Meanwhile,
Z
f (y)dy
Iα [χRn \Q(3) f ](x) ∼
Rn (1 + |y|)n−α
Z  1
dy p01
. kf kLp1 (W ) 0
Rn (1 + |y|)(n−α+γ)p1
∼ kf kLp1 (W )
n
for any x ∈ Q(1), since α < + γ.
p1
Assume (4.6) instead. Then since |x − y| . |y| +Z 1 for any x ∈ Q(1),
Z
f (y)dy g(y)dy
n−α
. kf kLp1 (W ) , or equivalently, .
R n (1 + |y|) R n (1 + |y|)n−α+γ
kgkLp1 for all g ∈ M+ (Rn ). Thus, duality entails
Z
1
p01 (n−α+γ)
dy < ∞. (4.7)
R n (1 + |y|)
n n n
This implies α < +γ. Finally, from Theorem 178, we deduce − ≤
p1 p1 p2
α. Thus, the proof is complete.
 
1
(2) Assume n 1 − < α ≤ n + γ. Then going through a similar
p2
argument to (1) and using the Hardy–Littlewood–Sobolev theorem, we
obtain

kIα [χQ(3) f ]kLp2 (Q(1)) . kIα [χQ(3) f ]kWLp3 (Q(1)) . kf kLp1 (W ) .

We estimate Iα [χRn \Q(3) f ] similarly to (1). We thus obtain (4.6) keeping

in mind that the case where α = n + γ is included in (4.7).
 Various operators in Lebesgue spaces 183
n
Conversely, assume that (4.6) holds with p2 = . Then for any
n−α
n −n+α
0 < r  1, Iα [χB(r) ](x) ∼ r |x| on |x| > 2r thanks to Example 83.
As a result, kIα [χB(r) ]kLp2 (Q(1)) & rn log(r−1 ), while kχB(r) kL1 ∼ rn .
n
This contradicts (4.6). Thus, p2 6= n−α . If we go through the same
 
1
argument as before, we conclude n 1 − < α ≤ n + γ. See Exercise
p2
70.
n
(3) If < α, then by Hölder’s inequality we can show that f ∈ Lp1 (Rn ) 7→
p1
χQ(1) Iα [χQ(1) f ] ∈ L∞ (Rn ) is a bounded linear operator. Consequently,
n
we can go through the same argument as before. If = α, then
p1
n 1
f (x) ≡ χB(e−1 ) (|x|)|x| p1 log (x ∈ Rn )
|x|
is an Lp1 (Rn )-function which is mapped to an unbounded function by
Iα . Other parts are the same as the model case (1).
(4) Should kIα f kL∞ (Q(1)) . kf kL1 (W ) hold for all f ∈ L1 (W ), then we
would have |Iα [gW −1 ](x)| . kgkL1 for all g ∈ L1 (Rn ). Hence, by duality
for any x ∈ Rn ,
1
∈ L∞ (Rn ).
|x − ·|n−α (1 + | · |)γ
This is clearly false for any x ∈ Rn , since n − α > 0.
We extend Theorem 184 to the case of p1 = 1 and 0 < p2 < ∞.
 
Corollary 185. Let 0 < p < ∞ and n 1 − p1 < α < n. Then for all δ > 0,
+
r > 0 and for all f ∈ M+ (Rn )
Z
n f (x)
kIα f kLp (B(r)) . r p dx.
Rn (|x| + r)n−α
Proof By Hölder’s inequality, we may assume p > 1. Then we are in the
position of using Theorem 184 and the dilation argument.
We consider what happens for f ∈ Lploc (Rn ) in Corollary 185.
Corollary 186. Let 1 < p1 ≤ ∞, 0 < p2 ≤ ∞ and
1 1
n − < α < n,
p1 p2 +
or let 1 < p1 < p2 < ∞ and
n n
α= − .
p1 p2
184 Morrey Spaces

Then for all δ > 0

 p1
f (x)p1
Z
n 1
−δ
kIα f kLp2 (B(r)) . r p2
n−(α+δ)p1
dx (4.8)
R n (|x| + r)
n
for all r > 0 and for all f ∈ M+ (Rn ). In particular, if δ = − α, then
p1
 
α−n p1 − p1
kIα f kLp2 (B(r)) . r 1 2 kf kLp1 (4.9)
n
for all f ∈ Lp1 (Rn ) and r > 0 and if 0 < δ < − α, then
p1
(Z ! ) p1
∞ Z 1
n
−δ p1 dt
kIα f kLp2 (B(r)) . r p2
f (x) dx n−(α+δ)p +1 (4.10)
r B(t) t 1

for all r > 0 and for all f ∈ M+ (Rn ).

Proof By Hölder’s inequality, we may assume p2 > p1 to prove (4.8).
Then we can use the scaling argument. See Exercise 71 for more.
Let p1 > 1 and δ = 0 in (4.8). We consider f (x) ≡ |x|−α (log |x|)−1 χB(2) (x)
for x ∈ Rn . Then the right-hand side of (4.8) is finite whilst (Iα f )(x) = ∞
for all x ∈ Rn . Hence (4.8) cannot hold with δ = 0.
 
Remark 8. If 1 < p1 < ∞, 0 < p2 < ∞ and n p11 − p12 ≤ α < n, then
+
 p1
|f (x)|p1
Z
n 1
kMα f kLp2 (B(r)) . r p2
n−αp
dx ,
Rn (|x| + r) 1

for all r > 0 and f ∈ L0 (Rn ). Consequently if Iα is replaced by Mα then

inequality (4.10) also holds for δ = 0.
In the same spirit we can prove the following estimate:
Lemma 187. Let 0 < α < n. Let 1 < q < t < ∞ satisfy
1 1 α
= − .
t q n
Let B ≡ B(x, r) be a ball. Then for all f ∈ M+ (Rn ),
Z ∞
n ds
kIα f kLt (B) . kf kLq (2B) + r t sα−n kf kL1 (B(x,s)) .
r s
Proof We calculate
kIα f kLt (B) . kIα [χ2B f ]kLt (B) + kIα [f − χ2B f ]kLt (B)
. kf kLq (2B) + kIα [f − χ2B f ]kLt (B)
Z ∞
n ds
. kf kLq (B) + r t sα−n kf kL1 (B(x,s)) .
r s
 Various operators in Lebesgue spaces 185

4.4.3 Sparse estimate of the fractional integral operators

We are now interested in the sparse estimate of Iα .
Let 0 < α < n. We start with an auxiliary operator. Suppose that we have
a cube Q0 and a family S. Define the discrete fractional integral operator IαS f
generated by S and the global operator Cα,∞ (f ) by
X ∞
X
IαS f ≡ `(Q)α m3Q (f )χQ and Cα,∞ (f ) ≡ `(2k Q0 )α m2k Q0 (f ).
Q∈S k=0

If S = D(Rn ), then IαS f is simply called the discrete fractional integral oper-
ator. In particular, for S = D(Q0 ), we have
0 X
IαD(Q ) f ≡ `(Q)α m3Q (f )χQ .
Q∈D(Q0 )

We begin with a preparatory but useful estimate.

Proposition 188. Let 0 < α < n. For f ∈ M+ (Rn ), Iα f ∼ IαD f.
Proof It suffices
P to compare the integral kernel after we fix x and y in Rn :
α−n
Let K(x, y) ≡ `(Q) χQ (x)χ3Q (y). Let ν0 be the largest integer such
Q∈D(Rn )
that x ∈ Q and y ∈ 3Q for some Q ∈ Dν (Rn ). Observe that 2−ν0 ∼ |x − y|.
In fact, if we let Q0 ∈ Dν+1 (Rn ) be the unique cube such that x ∈ Q0 ,
then y ∈ / 3Q0 . Thus, |x − y| & `(Q0 ) ∼ 2−ν0 . Since x, y ∈ 3Q, we have
ν0
|x − y| . `(Q) = 2−ν0 . Thus, K(x, y) ∼ 2−j(n−α) ∼ |x − y|−n+α .
P
j=−∞
Consequently, the integral kernels of these operators are equivalent.
We can stop tripling Q in Proposition 188 by the use of the maximal
operator. We define
Dj~e (Rn ) ≡ {2j ([0, 1)n + m + (−1)j 3−1~e) : m ∈ Zn },
for ~e ∈ {0, 1, −1}n and j ∈ Z and
[
D~e (Rn ) = Dj~e (Rn )
j∈Z

for ~e ∈ {0, 1, −1}n .

X
Corollary 189. For f ∈ M+ (Rn ), Iα f . sup `(Q)α mQ (f )χQ
e∈{0,1,−1}n
~
Q∈D(Rn )
holds.
Proof An important observation is that for any cube Q ∈ Q there exists
P ∈ ~e∈{0,1,−1}n D~e such that P ⊂ Q and |Q| ≤ 3n |P |. If we combine this
S
observation with Proposition 188, we obtain the desired result.
The following is the sparse decomposition of Iα .
186 Morrey Spaces

Lemma 190. Let 0 < α < n and f ∈ L∞ n + n

c (R ) ∩ M (R ). Then, for any cube
Q ∈ Q, there exists a sparse family S ⊂ D(Q ) such that, for all x ∈ Q0 ,
0 0

Iα f (x) . IαS f (x) + Cα,∞ (f ).

D(Q0 )
Proof For all x ∈ Q0 we have Iα f (x) . Iα f (x) + Cα,∞ (f ). We use
the construction in Theorem 172 with α > 0 there. Write Dkj ≡ {Q ∈ D(Qjk ) :
mQ (f ) ≤ a0 aj+1 }. Fix x ∈ Qjk , j ∈ N0 and k ∈ Kj . Since α > 0, we have
χQ (x)`(Q)α . `(Qjk )α . Thus, we have
P
j
Q∈Dk

X X
χQ (x)`(Q)α m3Q (f ) ≤ a0 aj+1 χQ (x)`(Q)α
j j
Q∈Dk Q∈Dk

. a0 aj+1 `(Qjk )α
. `(Qjk )α m3Qj (f )
k

0
D(Q )
for all Qjk ∈ S. Consequently, Iα χQ (x)`(Q)α m3Q (f ) for all
P
f (x) .
Q∈S
x ∈ Q0 . This completes the proof.
Following the procedure of Theorem 170 with h replaced by f , we can find
a sparse family S such that the following estimate holds:
Theorem 191. For f ∈ M+X (Rn ) and 0 < α < n, there exists a sparse family
S = Sf such that Iα f (x) ∼ `(Q)α m3Q (f ) · χQ (x) for x ∈ Rn .
Q∈S

Proof Since the proof is similar to Lemma 190, we omit the proof.

4.4.4 Fundamental solution to the elliptic differential

operators
Let us see how Iα with α = 2 comes about in the context of other fields
of mathematics. This operator can be regarded as a majorant operator of the
fundamental solution to the elliptic differential equations.
Theorem 192. Let n ≥ 3 and let A = {aij }ni,j=1 be a positive definite matrix
with inverse A−1 = {aij }ni,j=1 . Let L be a constant differential operator given
n
P
by L ≡ − aij ∂i ∂j . Define
i,j=1

 1− n2
n
1 X
Γ(x) ≡ √  aij xi xj  (x ∈ Rn ).
(n − 2)ωn−1 det A i,j=1

Then ϕ = L[Γ ∗ ϕ] = Γ ∗ [Lϕ] for all ϕ ∈ Cc∞ (Rn ).

 Various operators in Lebesgue spaces 187

Proof A direct calculation shows that L[Γ ∗ ϕ](x) = Γ ∗ [Lϕ](x) in the

classical case. Since the conclusion is translation invariant, we may suppose
that x = 0. We write out this quantity in full:
 1− n2
Z n
1 X
Γ ∗ [Lϕ](0) = √  aij yi yj  Lϕ(y)dy.
Rn (n − 2)ωn−1 det A i,j=1
√
Denote by A = {bij }ni,j=1 the unique positive definite matrix such that
√
( A)2 = A. Then a change of variables yields:
Z
1 n √
Γ ∗ [Lϕ](0) = |y|1− 2 Lϕ( Ay)dy.
Rn (n − 2)ωn−1
√
Write ϕA (y) ≡ ϕ( Ay). Note that
n X
n X
n
X √ √
∆ϕA (y) = [bkj blj [∂l ∂k ϕ]( Ay)] = −Lϕ( Ay).
j=1 k=1 l=1

Thus, we obtain
Z
1
Γ ∗ [Lϕ](0) = |y|2−n (−∆ϕA )(y)dy.
(n − 2)ωn−1 Rn

Let τ ∈ C ∞ ((−∞, ∞)) be a radial function such that χ(1,∞) ≤ τ ≤ χ(0,∞) .

To apply the Stokes theorem, we delete the singularity:
 
|y|
Z
1
Γ ∗ [Lϕ](0) = lim τ |y|2−n (−∆ϕA )(y)dy.
(n − 2)ωn−1 ε↓0 Rn ε

By the Stokes theorem, we have

   
|y|
Z
1 2−n
Γ ∗ [Lϕ](0) = − lim ∆ τ |y| ϕA (y)dy.
(n − 2)ωn−1 ε↓0 Rn ε

Note that
   
|y| 1
∆ τ |y| 2−n
. ε−n χB(ε) (y).
ε (n − 2)ωn−1

Thus, thanks to the continuity of ϕ at 0,

   
|y|
Z
1 2−n
Γ ∗ [Lϕ](0) = − lim ∆ τ |y| ϕA (0)dy
ε↓0 Rn ε (n − 2)ωn−1
   
|y|
Z
1 2−n
= −ϕA (0) lim ∆ τ |y| dy.
ε↓0 Rn ε (n − 2)ωn−1
188 Morrey Spaces

By the use of the polar coordinate, we have

ϕA (0) ε n−1 d2 r2−n
    
n−1 d
Z
r
Γ ∗ [Lϕ](0) = − r + τ dr
n−2 0 dr2 r dr ε (n − 2)ωn−1
ϕA (0) ε d
Z  
d r
=− rn−1 r2−n τ dr
n − 2 0 dr dr ε
ϕA (0) ε d
Z  r 1  
n−1 1−n 2−n 0 r
=− r (2 − n)r τ + r τ dr
n − 2 0 dr ε ε ε
= ϕA (0) = ϕ(0).
Thus, we obtain the desired result.
As an application of Theorem 192, we obtain an estimate of functions in
terms of their derivative. The fractional integral operator I1 of order 1 is also
important as the following theorem shows:
Theorem 193. Let n ≥ 2. Then |f | . I1 [|∇f |] for all f ∈ Cc∞ (Rn ).
Proof Fix x ∈ Rn . We suppose n ≥ 3; the case of n = 2 can be handled
similarly except in that we need to handle the kernel log |x − y|. We omit
the proof for the case of n = 2. Thanks to Theorem 192, we have f (x) 'n
I2 [∆f ](x). Since n ≥ 3, we can perform the integration by parts to have
n Z
X xk − yk
f (x) 'n ∂k f (y)dy,
R n |x − y|n
k=1

so that by the triangle inequality for integrals, we have the desired result.

s
4.4.5 The Bessel potential operator (1 − ∆)− 2 , s > 0
The operator dealt with in Section 4.4.5 lies outside the main stream of
Section 4.4. However due to the similarity and the strong connection with
fractional integral operartors, we consider the Bessel potential operator (1 −
s
∆)− 2 , s > 0. We will show the multiplicative law (1 − ∆)−s1 (1 − ∆)−s2 = (1 −
∆)−s1 −s2 and the inverse formula (1−∆)(1−∆)−1 f = (1−∆)−1 (1−∆)f = f
for suitable functions f .
We start with the definition of the Bessel kernel.
Definition 56. Let s > 0. Then we define the Bessel kernel of order s by
exp(−|εξ|2 )eix·ξ
Z
1
Gs (x) ≡ lim s dξ (x ∈ Rn ).
ε↓0 (2π)n Rn (1 + |ξ|2 ) 2
eix·ξ
Z
Roughly speaking, Gs (x) is the integral 2 s
dξ. However, if
Z Rn (1 + |ξ| ) 2
1
0 < s ≤ n, then 2 s
dξ = ∞, so that we need to take an indirect
Rn (1 + |ξ| ) 2
way to define Gs . We need to justify:
 Various operators in Lebesgue spaces 189

Lemma 194. Let s > 0. Then the limit defining Gs (x) exists for each x ∈ Rn .
Proof We use integration by parts to the expression. To this end we trans-
form
exp(−|εξ|2 )eix·ξ exp(−|εξ|2 )(−∆ξ )n eix·ξ
Z Z
1
s dξ = s dξ.
Rn (1 + |ξ|2 ) 2 |x|2n Rn (1 + |ξ|2 ) 2
Since sup tm exp(−t) < ∞ for each m ∈ N,
t>0

exp(−|εξ|2 )(−∆ξ )n eix·ξ

Z
s dξ
Rn (1 + |ξ|2 ) 2
Z
s
exp(−|εξ|2 )eix·ξ (−∆ξ )n (1 + |ξ|2 )− 2 dξ + o(1)

=
Rn

as ε ↓ 0. Thanks to the Lebesgue convergence theroem, we have

Z
s
exp(−|εξ|2 )eix·ξ (−∆ξ )n (1 + |ξ|2 )− 2 dξ

lim
ε↓0 Rn
Z
s
eix·ξ (−∆ξ )n (1 + |ξ|2 )− 2 dξ.

=
Rn

As a result, we have an expression:

Z
1 s
eix·ξ (−∆ξ )n (1 + |ξ|2 )− 2 dξ (x ∈ Rn ).

Gs (x) =
(2π)n |x|2n Rn
Likewise, for any l  1, we have
Z
1 s
eix·ξ (−∆ξ )l (1 + |ξ|2 )− 2 dξ (x ∈ Rn ).

Gs (x) = n 2l
(2π) |x| Rn

From this expression, we learn:

Corollary 195. Let s > 0. Then Gs ∈ C ∞ (Rn \ {0}) and satisfies Gs (x) =
O(|x|−l ) as x → ∞ for any l ∈ N.
Hence, we are interested in the behavior of this function near the origin.
We obtain an asymptotic expression of Gs near the origin.
Proposition 196. Let s > 0. Then Gs (x) = O(|x|−n+s ) as x → 0.
Proof Let η(x) ≡ exp(−|x|2 ), x ∈ Rn . Fix x ∈ B(1). We write

η(ξ)eix·ξ
Z
1
gs,0 (x) ≡ n s dξ (x ∈ Rn )
(2π) Rn (1 + |ξ|2 ) 2
and for l ∈ N
(η(2−l ξ) − η(21−l ξ))eix·ξ
Z
1
gs,l (x) ≡ s dξ (x ∈ Rn ).
(2π)n Rn (1 + |ξ|2 ) 2
190 Morrey Spaces
∞
P
We decompose Gs (x) = gs,l (x). It is easy to see that gs,0 is a bounded
l=0
function, so that we have only to take care of gs,l . If we perform integration
by parts, |gs,l (x)| . (2l |x|)−N 2l(n−s) for any N ∈ N and x ∈ Rn . If we add
this estimate over l ∈ N, we obtain
∞
X ∞
X
|gs,l (x)| . min(2l(n−s) , 2−s |x|−n ) = O(|x|−n+s )
l=1 l=1

as x → 0.
As a result, although Gs is defined in a tricky manner, this is integrable.
Corollary 197. Let s > 0. Then Gs ∈ L1 (Rn ).
If we use Corollary 197, then we see that the operator f ∈ L1 (Rn ) 7→
Gs ∗ f ∈ L1 (Rn ) is a bounded linear operator.
As we will prove, we have the following multiplicative law:
Theorem 198. Let s1 , s2 > 0. Then Gs1 +s2 = Gs1 ∗ Gs2 .
Proof We remark that in the course of the proof of Corollary 197, we
have showed that

exp(−|εξ|2 )eix·ξ
Z
1
sup n 2 ) 2s
dξ . min(|x|−n+s , |x|−n−1 ).
ε>0 (2π) R n (1 + |ξ|

We calculate

(2π)2n Gs1 ∗ Gs2 (x)

Z
= (2π)2n Gs1 (y)Gs2 (x−y) dy
Rn
exp(−|εξ|2 )eiy·ξ exp(−|εζ|2 )ei(x−y)·ζ
Z Z  Z 
= lim s1 dξ s2 dζ dy
ε↓0 Rn Rn (1 + |ξ|2 ) 2 Rn (1 + |ζ|2 ) 2

in the topology of L1 (Rn ). We observe

exp(−|εξ|2 )eiy·ξ exp(−|εζ|2 )ei(x−y)·ζ

Z Z  Z 
s1 dξ s2 dζ dy
Rn Rn (1 + |ξ|2 ) 2 Rn (1 + |ζ|2 ) 2
2
! Z 2
!
eiy·ξ−|εξ| dξ ei(x−y)·ζ−|εζ| dζ
Z Z
dy
= lim s1 s2
δ↓0 Rn Rn (1 + |ξ|2 ) 2 Rn (1 + |ζ|2 ) 2 exp(δ|y|2 )
2 2 2
Z 
exp(−ε (|ξ| + |ζ| ))
ZZ
dy
= lim s1 s2 ei(x−y)·ζ+iy·ξ dξ dζ
δ↓0 Rn ×Rn (1 + |ξ|2 ) 2 (1 + |ζ|2 ) 2 Rn exp(δ|y|2 )
 Various operators in Lebesgue spaces 191

thanks to Corollary 195, Proposition 196 and Fubini’s theorem. Using the heat
kernel, we calculate
exp(−|εξ|2 )eiy·ξ exp(−|εζ|2 )ei(x−y)·ζ
Z Z  Z 
s1 dξ s2 dζ dy
Rn Rn (1 + |ξ|2 ) 2 Rn (1 + |ζ|2 ) 2
 π  n2 Z Z exp(−ε (|ξ| + |ζ| ))eix·ζ
2 2 2
 
1 2
= lim s s exp − |ξ − ζ| dξ dζ
δ↓0 δ 2 1 2 2 4δ
Rn ×Rn (1 + |ξ| ) 2 (1 + |ζ| ) 2

exp(−|εξ|2 )2 eix·ζ
Z
= (2π)n s1 +s2 dξ.
Rn (1 + |ξ|2 ) 2

Thus, Gs1 ∗ Gs2 = Gs1 +s2 follows.

Among Gs , s > 0, G2 is important because this is used to express the
solution u of (1 − ∆)u = f .
Theorem 199. For all f ∈ Cc∞ (Rn ), (1 − ∆)[G2 ∗ f ] = G2 ∗ [(1 − ∆)f ] = f .
Proof Let x ∈ Rn be fixed. We calculate
exp(−ε|ξ|2 )ei(x−y)·ξ
ZZ
1
(1 − ∆)f (y) dξdy
(2π)n Rn ×Rn (1 + |ξ|2 )
ZZ
1
= f (y) exp(−ε|ξ|2 )ei(x−y)·ξ dξdy
(2π)n Rn ×Rn
√
ε−n π n
Z  
1 2
= f (y) exp − |x − y| dξdy → f (x)
(2π)n Rn 4ε
as ε ↓ 0, showing that G2 ∗ [(1 − ∆)f ] = f . Likewise, We can show that
(1 − ∆)[G2 ∗ f ](x)
exp(−ε|ξ|2 )ei(x−y)·ξ
ZZ
1
= lim n
(1 − ∆)f (y) dξdy
ε↓0 (2π) Rn ×Rn (1 + |ξ|2 )
= f (x).

4.4.6 Morrey’s lemma

We sharpen Theorem 193. Here we investigate an inequality which even-
tually led to the definition of Morrey spaces.
Lemma 200. Let B(x, r) be a ball. Then for all u ∈ C 1 (B(x, r)),
rn |∇u(y)|
Z Z
|u(x) − u(y)|dy ≤ dy.
B(x,r) n B(x,r) |x − y|n−1

Proof Fix any point w ∈ S n−1 and s ∈ (0, r]. Then

Z s Z s
|u(x + sw) − u(x)| = w · ∇u(x + tw)dt ≤ |∇u(x + tw)| dt.
0 0
192 Morrey Spaces

Hence for s ∈ (0, r],

Z Z s 
ku(x + s·) − u(x)kL1 (S n−1 ) ≤ t−n+1 |∇u(x + tw)| tn−1 dt dσ(w)
n−1
ZS 0

= |y|−n+1 |∇u(x + y)| dy

B(s)
Z
= |x − y|−n+1 |∇u(y)| dy
B(x,s)
Z
≤ |x − y|−n+1 |∇u(y)| dy.
B(x,r)

Multiply by sn−1 and integrate from 0 to r against s to have

rn
Z Z
|u(x) − u(y)|dy ≤ |x − y|−n+1 |∇u(y)| dy.
B(x,r) n B(x,r)

4.4.7 Exercises
0
θp θ−p
Exercise 69. Let f (θ) ≡ A + 0 B with A, B > 0. Find the minimum of
p p
f : [0, ∞) → [0, ∞).
Exercise 70. Let f (x) ≡ (1 + |x|)θ , x ∈ Rn with θ ∈ R. Let p0 = 1 < p1 < ∞.
(1) Find the necessary and sufficient condition on θ for f ∈ L∞ (Rn ).
(2) Find an analogue to (4.7) to show α ≤ n + γ if (4.6) holds.
Exercise 71. Prove Corollary 186. Hint: Use Theorem 184(1).
Exercise 72. [424, p. 252] Let Q be a dyadic cube, and let 0 < α < n. Then
for any f ∈ M+ (Rn ), show that
X Z Z
`(R)α f (x)dx . `(Q)α f (x)dx
R∈D(Q) R Q

using the decomposition

X Z ∞
X X Z
`(R)α f (x)dx = `(R)α f (x)dx
R∈D(Q) R l=0 R∈Dl (Q) R

4.5 Singular integral operators

As well as the Hardy–Littlewood maximal operator, singular integral oper-
ators are important in harmonic analysis. One of our goals of this book is the
 Various operators in Lebesgue spaces 193

consideration of the boundedness of integral operators. Section 4.5, presents

the definition of singular integral operators which are already general enough
to applications. Then we consider the Lp (Rn )-boundedness of singular inte-
gral operators. This boundedness will be a key to the boundedness of singular
integral operators acting on Morrey spaces.
We take up examples of singular integral operators in Section 4.5.1. The
Hilbert transform and the Riesz transform will be typical examples of singu-
lar integral operators. Section 4.5.2 generalizes what we considered in Section
4.5.1. Here, we define Calderón–Zygmund operators and then we consider
their fundamental properties. We do not embark on the boundedness prop-
erties other than the L2 (Rn )-boundedness here. Section 4.5.3 is a prepara-
tory but important step to deal with singular operators. Here, we obtain a
decomposition method called the Calderón–Zgymund decomposition. Based
on the preliminary observation made in Sections 4.5.1 – 4.5.3, we investigate
the boundedness properties in Section 4.5.4. As the name suggests, singular
integral operators have the singularity. More precisely, as it turns out, the
integral kernels have the singularity at the diagonal {(x, x) : x ∈ Rn }. Conse-
quently, it is natural to avoid such a singularity by truncating the integration
domains. Section 4.5.5 is devoted to such an attempt. Additional examples
of singular integral operators will be provided in Section 4.5.6. Finally, we
investigate the sparse estimates and the local estimates in Sections 4.5.7 and
4.5.8, respectively.

4.5.1 Riesz transform

Before we give the precise definition of singular integral operators, we
will consider one concrete example of the Riesz transform. We will prove the
following theorem:
Theorem 201. Let j = 1, 2, . . . , n. Then the linear operator, called the Riesz
transform,
xj − yj
Z
Rj f (x) ≡ lim f (y)dy (x ∈ Rn ),
ε↓0 Rn \B(x,ε) |x − y|n+1

which is initially defined for f ∈ Cc∞ (Rn ), extends to a bounded linear operator
on L2 (Rn ).
We will prove Theorem 201 as the special case of Example 92. It is not so
hard to show that the limit defining Rj f exists for f ∈ Cc∞ (Rn ); see Exercise
74.
Example 85. The most prominent case is the case where n = 1. In this case,
Z −ε Z ∞ 
f (t − s) f (t − s)
Hf (t) ≡ lim ds + ds (t ∈ R).
ε↓0 −∞ s ε s
194 Morrey Spaces

This transform is called the Hilbert transform. In particular, for f = χ[−1,1] ,

t+1
we have Hf (t) = log t−1 for almost all t ∈ R.

The Riesz transform arises naturally.

Example 86. Let n ≥ 2. Let fZ ∈ Cc∞ (Rn ). Let I1 be the fractional integral
1
operator of order 1; I1 f (x) = n−1
f (y)dy, x ∈ Rn . Then for any
Rn |x − y|
x ∈ Rn
xj − yj
Z
∂j I1 f (x) = − lim(n − 1) f (y)dy.
ε↓0 n
R \B(x,r) |x − y|n+1
f (x − y) ∂j f (x − y)
Z Z
In fact, since I1 f (x) = n−1
dy, ∂ I
j 1 f (x) = dy. Let
Rn |y| Rn |y|n−1
η ∈ C ∞ (R) ∩ M↑ (R) satisfy χ(1,∞) ≤ η ≤ χ(1/2,∞) . Then
Z
η(ε|y|)
∂j I1 f (x) = lim ∂j f (x − y)dy.
ε↓0 Rn |y|n−1

By integration by parts, we have

Z  
η(ε|y|)
∂j I1 f (x) = lim ∂j f (x − y)dy.
ε↓0 Rn |y|n−1
By the Leibniz rule,
εyj η 0 (ε|y|)
 
η(ε|y|) η(ε|y|)yj
∂j n−1
= (1 − n) n+1
+ .
|y| |y| |y|n+1
Since y →
7 yj is an odd function,
εyj η 0 (ε|y|) εyj η 0 (ε|y|)
Z Z
f (x − y)dy = (f (x − y) − f (x))dy.
Rn |y|n+1 B(ε)\B ( 2ε ) |y|n+1

By the triangle inequality, we have,

εyj η 0 (ε|y|)
Z
f (x − y)dy
Rn |y|n+1
ε|yj |η 0 (ε|y|)
Z
≤ |f (x − y) − f (x)|dy
B(ε)\B ( 2ε ) |y|n+1
ε|yj |
Z
0
≤ kη kL (0,∞) sup |f (x − y) − f (x)|
∞
n+1
dy
B(ε)\B ( 2 )
ε B(ε)\B ( 2 )
ε |y|
ε|yj |
Z
0
= kη kL∞ (0,∞) sup |f (x − y) − f (x)| n+1
dy.
B(ε)\B ( ε )
1
2 <|y|<1
|y|
2

As a result,
Z
yj η(ε|y|)
∂j I1 f (x) = (1 − n) lim f (x − y)dy
ε↓0 Rn |y|n+1
 Various operators in Lebesgue spaces 195

by the continuity of f . We decompose

Z
yj η(ε|y|)
n+1
f (x − y)dy
Rn |y|
Z Z
yj η(ε|y|) yj
= n+1
f (x − y)dy + n+1
f (x − y)dy.
B(ε)\B ( 2ε ) |y| Rn \B(ε) |y|

Since y 7→ yj is an odd function, we have

Z Z
yj η(ε|y|) yj η(ε|y|)
n+1
f (x − y)dy = n+1
(f (x − y) − f (x))dy.
B(ε)\B ( 2 )
ε |y| B(ε)\B ( 2 ) |y|
ε

As a result,
Z
yj η(ε|y|)
n+1
f (x − y)dy
B(ε)\B ( 2ε ) |y|

|yj |η(ε|y|)
Z
≤ |f (x − y) − f (x)|dy
B(ε)\B ( 2 )
ε |y|n+1
|yj |
Z
≤ sup |f (x − y) − f (x)| n+1
dy
y∈B(ε)\B ( 2 )ε B(ε)\B ( 2 )
ε |y|
|yj |
Z
= sup |f (x − y) − f (x)| n+1
dy.
y∈B(ε)\B ( 2 )ε B(1)\B ( 2 )
1 |y|

xj − yj
Z
Consequently, ∂j I1 f (x) = −(n − 1) lim f (y)dy.
ε↓0 Rn \B(x,ε) |x − y|n+1

4.5.2 Calderón–Zygmund operators

Now we start a general theory keeping in mind the example of the Riesz
transform. The generalized Calderón–Zygmund operator and the theory that
lies behind are key tools in this book. The definition, presented below, is
highly generalized in view of the singular integral operators we have been
dealing with.
Definition 57 (Calderon–Zygmund operator). An L2 (Rn )-bounded linear
operator T is a (generalized) Calderón–Zygmund operator (with the kernel
K), if it satisfies the following conditions:
(1) There exists

K ∈ L1loc (Rn × Rn \ {(x, y) ∈ Rn × Rn : x = y})

such that, for all f ∈ L2c (Rn ), we have

Z
T f (x) = K(x, y)f (y)dy for almost all x ∈
/ supp(f ). (4.1)
Rn
196 Morrey Spaces

(2) There exist constants D1 and D2 such that K satisfies the following
estimates.
(i) Size condition:
|K(x, y)| ≤ D1 |x − y|−n , (4.2)
if x 6= y.
(ii) Hörmander’s condition:
|x − y|
|K(x, z) − K(y, z)| + |K(z, x) − K(z, y)| ≤ D2 , (4.3)
|x − z|n+1
if 0 < 2|x − y| < |z − x|.
If K(x, y) = K0 (x − y) for some function K0 , then T is called a singular
integral operator of convolution type.
Example 87. Let {ajν }j∈Z,ν∈Zn ∈ `2 (Z × Zn ). Also let ψ ∈ Cc∞ (Rn ) be
jn
supported on Q(3). For j ∈ Z and ν ∈ Zn , we set ϕjν (x) ≡ 2 2 ∆ψ(2j x −
ν), x ∈ Rn . Since
X
`(Q)−a−n χ3Q (x)χ3Q (y) ∼ |x − y|−n−a
Q∈D(Rn )

for all a > 0, according to Example 55, the operator

X
f ∈ L2 (Rn ) 7→ hf, ϕjν iL2 ϕjν
j∈Z,ν∈Zn

is a Calderón–Zygmund operator, which is not convolution type.

Let S n−1 ≡ {x ∈ Rn : |x| = 1} as before and equip S n−1 with the surface
measure σ.
Definition 58. A function K ∈ L0 (Rn ) ∩ C ∞ (Rn \ {0}) is called a homo-
geneous smooth Calderón–Zygmund
Z kernel if K is a function of homogeneous
of degree zero satisfying K(x)dσ(x) = 0. For such a function K, the
S n−1
corresponding
R singular integral operator T of convolution type is defined by
T f (x) = K(x − y)f (y)dy as long as the definition makes sense.
Rn

Example 88. Let j = 1, 2, . . . , n. The j-th Riesz transform, given by

xj − yj
Z
Rj f (x) ≡ lim f (y)dy (x ∈ Rn ), (4.4)
ε↓0 Rn \B(x,ε) |x − y|n+1

is an example of singular integral operators of convolution type with the homo-

x
geneous smooth Calderón–Zygmund kernel K(x) = |x|j . We write

xj − yj
Z
Rj f (x) ≡ p.v. n+1
f (y)dy (x ∈ Rn ).
R n |x − y|
 Various operators in Lebesgue spaces 197

When there is a singularity in the integral kernel, we use p.v. in this way to
approximate the integral.
Example 89. Surprisingly enough, the identity operator idL2 is a generalized
singular integral operator with the kernel K = 0.
∞ n
Example 90. Let n ≥ 3 and f ∈ C Zc (R ). Also let I2 be the fractional
f (y)
integral operator of order 2; I2 f (x) ≡ dy (x ∈ Rn ). A similar
R n |x − y|n−2
technique to Example 86 shows that
(xj − yj )(xk − yk )f (y)
Z
∂j ∂k I2 f (x) = n(n − 2) lim dy (4.5)
ε↓0 Rn \B(x,ε) |x − y|n+2
if 1 ≤ j < k ≤ n and
(n(xj − yj )2 − |x − y|2 )f (y)
Z
∂j 2 I2 f (x) = (n − 2) lim dy (4.6)
ε↓0 Rn \B(x,ε) |x − y|n+2
if 1 ≤ j ≤ n. See Exercise 73. Since ∂j ∂k I2 is L2 (Rn )-bounded thanks to
Example 92 to follow, we see that ∂j ∂k I2 is a singular integral operator of
convolution type.
In what follows we abbreviate this type of operator to CZ-operator. The
next proposition explains why we want to pose symmetric condition (4.3).
Lemma 202. For any CZ-operator T and f ∈ L2c (Rn ), its dual T ∗ on the
complex Hilbert space L2 (Rn ) satisfies
Z
∗
T f (x) = K(y, x)f (y)dy for a.e. x ∈
/ supp(f ). (4.7)
Rn

Proof We take a function g ∈ L2c (Rn ) whose support does not meet that
of f . Then by (4.1)
Z Z
∗
f (x)T g(x)dx = T f (x)g(x)dx
Rn n
ZRZ
= K(x, y)f (y)g(x)dxdy
Rn ×Rn
Z Z 
= f (y) K(x, y)g(x)dx dy.
Rn Rn

Thus, since g is an arbitrary L2c (Rn )-function with its support contained in
Rn \ supp(f ), we have
Z
T ∗ f (x) = K(y, x)f (y)dy for almost everywhere x ∈ / supp(f ).
Rn

This is the desired result.

Note that the condition on the kernel K is symmetric in the variables x
and y.
198 Morrey Spaces

Corollary 203. For a CZ-operator T , its adjoint T ∗ is a CZ-operator again.

We are interested in what is truly required for the boundedness of singular

integral operators T . However, it may happen that T is the zero mapping or
the identity. These cases should be avoided. Consequently, we are interested
in operators which are appropriately non-degenerate. Denote by SO(n) the
set of all orthonormal matrices with determinant 1.

Definition 59. One says that T is a genuine Calderon–Zygmund operator if

it is a Calderon–Zygmund operator with the kernel K and for n ≥ 2 there
exist a > 0 and a rotation R ∈ SO(n) such that K(x, y) & |x − y|−n for all
x ∈ Rn and for all y ∈ Cx ≡ x + R(C), where

C ≡ {y = (y 0 , yn ) ≡ (y1 , y2 , . . . , yn ) ∈ Rn : yn > a |y 0 |} .

If n = 1 then we assume that K(x, y) & |x − y|−1 for all x ∈ R and for all
y > x or for all x ∈ R and for all y < x.
A function F defined on Rn \ {0} is said to be homogeneous of degree k if
F (tx) = tk F (x) for t > 0 and x ∈ Rn \ {0}.
Example 91.
1
(1) The Hilbert transform H, in which case K(x, y) ≡ x−y , is a genuine
−1
Calderon–Zygmund operator because K(x, y) ≥ |x − y| for all x ∈ R
and for all y < x.
n−1
R Ω ∈ C(S
(2) Suppose that we have )\{0}, which is homogeneous of degree
zero and satisfies S n−1 Ω(η)dη = 0. We let T be a Calderon–Zygmund
 
1 x−y
operator whose kernel K takes the form K(x, y) = nΩ .
|x − y| |x − y|
n−1
The properties of Ω imply that there exist η0 ∈ S and δ > 0 such
that Ω(η) & 1 for all η ∈ S n−1 ∩ B (η0 , δ). So, T is a genuine singular
integral operator.
We will show that there are Calderón–Zygmund operators amply. Here,
we present an application of Theorem 114 to create an example to which the
results in Section 4.5.1 are available.
Example 92. Let K ∈ C 1 (S n−1 ), and let ϕ ∈ Cc∞ (Rn \ {0}) be such that
Z ∞
X
K(x)dσ(x) = 0, ϕ(2l x) = χRn \{0} (x) (x ∈ Rn ).
S n−1 l=−∞

In order that we apply Cotlar’s inequality, we set

 
x−y
Z
Tk (x) ≡ ϕ(2k (x − y))K f (y)dy (x ∈ Rn ) (4.8)
Rn |x − y|
 Various operators in Lebesgue spaces 199

for f ∈ L2 (R). By means of the integral kernel Kk1 ,k2 , we write

Z
Tk1 ◦ Tk2 f (x) = Kk1 ,k2 (x, y)f (y)dy
Rn

for x ∈ Rn and for and k1 , k2 ∈ Z. Then, thanks to Theorem 114 with N = 1

and λ = n + 1,
2n max(k1 ,k2 )−|k1 −k2 | X
|Kk1 ,k2 (x, y)| . max(k ,k ) n+1
k∂ α KkL∞ (S n−1 ) (x, y ∈ Rn )
(1 + 2 1 2 |x − y|)
|α|≤1

for all k1 , k2 ∈ Z. Consequenlty,

 by the use of Young’s
 inequality we have
X
kTk1 ◦ Tk2 f kL2 . 2−|k1 −k2 |  k∂ α KkL∞ (S n−1 )  kf kL2 . Thus, we are in
|α|≤1
the position of using Cotlar’s inequality (Theorem 94) to obtain
 
Xk2 X
Tk f . k∂ α KkL∞ (S n−1 )  kf kL2
k=−k1 L2 |α|≤1

for all k1 , k2 ∈ Z. Meanwhile for f ∈ Cc∞ (Rn ), we know that the limit T f (x) =
k2
Tk f (x) exists for all x ∈ Rn . By the Fatou lemma, we have
P
lim
k1 ,k2 →∞ k=−k1
kT f kL2 . kf kL2 . Since Cc∞ (Rn ) is dense in L2 (Rn ), T extends to a bounded
linear operator on L2 (Rn ) with the bound kT kL2 →L2 . k∂ α KkL∞ (S n−1 ) ,
P
|α|≤1
in particular proving Theorem 201.
Recall that Ḣk (Rn ) denotes the homogeneous harmonic polynomial space
dim (Ḣk (Rn ))
of order k ∈ N0 . Fix an orthonormal basis {Yjk }j=1C in Ḣk (Rn ).
Example 93. Let k ∈ N0 and j = 1, 2, . . . , dimC (Ḣk (Rn )). Fix x ∈ Rn . Then
s
dimC (Ḣk (Rn )) k n
|Yjk (x)| ≤ |x| . (k + 1) 2 −1 |x|k
ωn−1
thanks to Corollary 122 and
s
k(n + 2k − 2) dimC (Ḣk (Rn )) k−1 n
|grad(Yjk )(x)| ≤ |x| . (k + 1) 2 |x|k−1
ωn−1
thanks to Corollary 124. Consequently, if we consider the integral operator
Tjk given by
Yjk (x − y)
Z
Tjk f (x) ≡ n+k
f (y)dy (x ∈ Rn )
Rn |x − y|
n
for k ≥ 1 and j = 1, 2, . . . , dimC (Ḣk (Rn )), then kTjk f kL2 . k 2 kf kL2 for all
f ∈ L2 (Rn ) with the implicit constants independent of such j and k.
200 Morrey Spaces

4.5.3 Calderón–Zgymund decomposition

In principle we are oriented to the boundedness property of singular inte-
gral operators. However, due to the strong singularity of the integral kernel,
it is difficult to handle these operators. Hence, we need some weapons to be
able to deal with singular integral operators. As we have mentioned, we are
fascinated with the method of decomposition of functions into countable ele-
mentary pieces. That is, we will look into each function in much depth. How-
ever no matter how deeply we investigate the functions, we need to stop at
a certain stage. Consequently, we have to judge whether we stop or continue.
To make such a decision, we need a tool, a collection of dyadic cubes.
With the definition of the dyadic cubes in mind, we define the dyadic
average operators.
Definition 60 (Dyadic average operator). Let f ∈ L1loc (Rn ) and j ∈ Z. The
dyadic average operator Ej of generation j, or shortly, the dyadic averaging
function x 7→ Ej f (x) stands for the average of f over Pa unique cube Q ∈
Dj (Rn ) containing x. That is, Ej f (x) ≡ Ej [f ](x) ≡ χQ (x)mQ (f ) for
Q∈Dj (Rn )
n
x∈R .
Example 94. For f ∈ L1loc (Rn ),
X X
E0 f = mQ (f )χQ = χm+[0,1)n (f )χm+[0,1)n .
Q∈D0 (Rn ) m∈Zn

If f ∈ L1 (Rn ), we have |Ej f (x)| ≤ 2jn kf kL1 for all j ∈ Z and x ∈ Rn .

Hence, the next lemma holds trivially.
Lemma 204. We have lim Ej f (x) = 0 for all f ∈ L1 (Rn ) and x ∈ Rn .
j→−∞

The next theorem is the heart of the matter. To formulate this theorem,
we define the quadrant as a set of the form

{(x1 , x2 , . . . , xn ) ∈ Rn : (a1 x1 , a2 x2 , . . . , an xn ) ∈ (0, ∞)n }

for some (a1 , a2 , . . . , an ) ∈ {−1, 1}n .

Theorem 205 (Construction of Calderón–Zygmund cubes). Let λ > 0, and
let f ∈ L1loc (Rn ). Assume that Eλ ≡ {M D f > λ} does not contain any
quadrant and is not empty. Then we have a disjoint collection {Qj }j∈J ⊂
D(Rn ) such that Eλ = Qj and that λ < mQj ( |f | ) ≤ 2n λ for all j ∈ J.
S
j∈J

Lp (Rn )\{0}, then the assumption

S
By the maximal inequality, if f ∈
1≤p<∞
on quadrants in Theorem 205 is satisfied.
Proof By the definition of the dyadic
[ maximal operator, in particular, by
the property of “sup”, we have Eλ = {Ej [|f |] > λ}. From assumption on
j∈Z
 Various operators in Lebesgue spaces 201

quadrants, the set of j satisfying x ∈ {Ej [|f |] > λ} is bounded from below for
each x ∈ Eλ . Hence, there exists j ∗ such that Ej [ |f | ](x) ≤ λ for all j < j ∗ .
Hence, we can partition Eλ :
[ \
Eλ = {x ∈ Rn : Ej [ |f | ](x) > λ, Ek [ |f | ](x) ≤ λ}.
j∈Z k∈Z∩(−∞,j)

From the property of dyadic cubes we can partition

\
{x ∈ Rn : Ej [ |f | ](x) > λ, Ek [ |f | ](x) ≤ λ} (4.9)
k∈Z∩(−∞,j)

into the disjoint sum of dyadic cubes Qjm ∈ Dj (Rn ) P

with m ∈ Mj . A rear-
rangement of Qjm ’s yields a partition of Eλ : Eλ = Qj and {Qj }j∈J =
j∈J
{Qjm }m∈Mj , j∈Z , where Mj is at most countable.
Let x ∈ Qj =: Qj0 m0 . From the choice of Qj0 m0 , we have mQj (|f |) =
Ej0 [ |f | ](x) > λ.
Let Q∗j ∈ Dj0 −1 (Rn ) be the dyadic parent of the dyadic cube Qj . That is,
the unique cube containing Qj with volume as 2n times Qj . From (4.9), the
choice of Qj0 m0 , we have mQ∗j ( |f | ) = Ej0 −1 [ |f | ](x) ≤ λ. From this formula,
we conclude mQj ( |f | ) ≤ 2n mQ∗j ( |f | ) = 2n Ej0 −1 [ |f | ](x) ≤ 2n λ.
We consider the Calderón–Zygmund decomposition keeping the setup
above in mind. The Calderón–Zygmund decomposition is one of the fundamen-
tal tools to analyze locally integrable functions. Another important decompo-
sition is the Lerner–Hytönen decomposition considered in Theorem 174.
Theorem 206 (Calderón–Zygmund decomposition). Let f ∈ L1 (Rn ), and
let λ > 0. Then we can find a set {Qj }j∈J of dyadic cubes and a family
{g} ∪ {bj }j∈J of countable collections of L1 (Rn )-functions such that:
P
(1) f admits the following decomposition: f = g + bj .
j∈J

(2) (The L1 (Rn )-condition) The L1 (Rn )-norm of g is less than that of f :

kgkL1 ≤ kf kL1 . (4.10)

(3) (The L∞ (Rn )-condition) The function g is L∞ (Rn )-bounded:

kgkL∞ ≤ 2n λ. (4.11)

(4) (The support condition) The function bj is supported on the closure of

Qj ; namely supp(bj ) ⊂ Qj .
(5) (The moment condition) bj ⊥ P0 (Rn ).
202 Morrey Spaces

(6) (Partition of thePset {M D f > λ}) The family {Qj }j∈J is disjoint and
D
{M f > λ} = Qj .
j∈J

This decomposition is called the Calderón–Zygmund decomposition (at height

λ).
Proof If f is zero, then we simply set J = ∅ and f = g. So we assume
that f 6= 0 and hence Eλ ≡ { M D f > λ } is empty. Then by the Lebesgue
differentiation theorem, Theorem 12, we see that g = f and b = 0 suffice.
See also Lemma 142 below. Assume otherwise. Choose a disjoint conllection
{Qj }j∈J of dyadic cubes such that
X
λ ≤ mQj (|f |) ≤ 2n λ, Eλ = Qj (4.12)
j∈J

according to TheoremP 205. Define bj ≡ χQj (f − mQj (f )). Based on bj , we

define g by g ≡ f − bj , which yields a decomposition of f . Here note that
j∈J
the sum makes sense in the topology of L1 (Rn ) and in the sense of almost
everywhere convergence. Since the supports of bj ’s are nonoverlapping, from
the definition of bj , the moment condition (4.10) follows. It remains to show
(4.11).
Let x ∈ Qj . Since {Qj }j∈J is disjoint, we have g(x) = mQj (f ). Hence
|g(x)| ≤ 2n λ by (4.12). P
Conversely, let x ∈ / Qj . Inequality (4.11) follows since for almost all
j∈J
x ∈ Rn , |g(x)| = |f (x)| ≤ M D f (x) ≤ λ by the Lebesgue differentiation
theorem.
A helpful remark may be in order.
1
Remark 9. Let f ∈ LP (Rn ) ∩ L2 (Rn ). If one reexamines the above proof,
then one notices b = bj in the topology of L1 (Rn ) ∩ L2 (Rn ) and that
j∈J
g ∈ L1 (Rn ) ∩ L2 (Rn ).
Theorems 205 and 206 are widely known as the Calderón–Zygmund decom-
position or the Calderón–Zygmund theory and these theorems have been
applied to harmonic analysis.
Example 95. Let λ > 0, and let D be a generalized dyadic grid or a family
D(Q) for some right-open cube. We consider Eλ ≡ {M D f > λ} for f ∈ L0 (Rn )
or L0 (Q), where M D denotes the maximal operator generated by D. Assume
that Eλ is not empty. Choose the family N of maximal cubes Q ∈ D satisfying
3Q ⊂ Eλ . Note that N is disjoint. Since Eλ is an open set, if x ∈ Eλ , then
P x ∈ Q ⊂ 3Q ⊂ Eλ . If necessary we
we can find a dyadic cube Q such that
can assume Q ∈ N . Thus, Eλ = Q. We define bQ ≡ (f − mQ (f ))χQ
Q∈N
 Various operators in Lebesgue spaces 203
P
for Q ∈ N . Set b ≡ bQ and g ≡ f − b. Since 10Q \ Eλ =
6 ∅, we have
Q∈N
mQ (|f |) . λ if we go through a similar argument. Thus, |g| . λ. If we let
D = D(Q) for a compact cube Q, we will obtain a family with disjoint interior.

4.5.4 Weak-(1, 1) boundedness and strong-(p, p)

boundedness
With our weapon obtained in Section 4.5.3 we are going to prove two
theorems for CZ operators, the weak-(1, 1) boundedness and the strong-(p, p)
boundedness, for 1 < p < ∞. Recall that given f ∈ L1 (Rn ) and λ > 0. There
is a (possibly empty) family of disjoint (dyadic) cubes {Qj }j such that
P λ≤
mQj (|f |) ≤ 2n λ. Using these cubes we have a decomposition f = g + j∈J bj
which enjoys the following properties: bj ∈ L1 (Qj )∩P0⊥ (Rn ) and g ∈ L∞ (Rn ).
From the definition of g we deduce kgkL1 ≤ kf kL1 .
Although T fails to be L1 (Rn ), more precisely, T cannot be extended
continuously to an L1 (Rn )-bounded operator, we have a substitute called the
weak-L1 (Rn ) estimate.
Theorem 207. Suppose that T is a singular integral operator. Let D1 and D2
be constants in (4.2) and (4.3), respectively. Then T is weak-(1, 1) bounded.
That is, λ |{ |T f | > λ}| . (kT kL2 →L2 + D1 + D2 )kf kL1 for all f ∈ L1 (Rn ) ∩
L2 (Rn ) and λ > 0.
Proof If necessary, we may suppose that kT kL2 →L2 + D1 + D2 ≤ 1 by
T
considering . Keeping to the above notation, we have
kT kL2 →L2 + D1 + D2
   
 
λ  X λ 
|{|T f | ≥ λ}| ≤ |T g| ≥ + T bj  ≥ . .
2  2 
j∈J

The first term is easy to estimate. In fact, using the L2 (Rn )-boundedness
as usual, we have
  Z Z
λ 1 1 1 1
|T g| ≥ . 2 |T g(x)|2 dx . 2 |g(x)|2 dx . kgkL1 . kf kL1 .
2 λ Rn λ Rn λ λ

Hence  
λ 1
|T g| ≥ . kf kL1 . (4.13)
2 λ
To estimate the second term we need the following observation:
Z
Claim 208. For any j ∈ J, √
|T bj (x)|dx . kbj kL1 .
Rn \2 n Qj
204 Morrey Spaces

S Admitting
√ Claim 208, let us finish the proof of the theorem. Write Z ≡
2 nQj . Firstly we proceed as follows:
j∈J
       
 X λ X λ
T bj  ≥ ≤ (Rn \ Z) ∩ T  bj  ≥ + |Z|
 2  2
j∈J j∈J
 
1 X X
. T bj  + |Qk |.
λ
j∈J k∈J
L1 (Rn \Z)

X [ 1
Recall that |Qk | = Qk = |{M D f > λ}| . kf kL1 . By the trian-
λ
k∈J k∈J
gle inequality we obtain
 
X X XZ
T bj  ≤ kT bj kL1 (Rn \Z) ≤ √
|T bj (x)|dx.
j∈J j∈J j∈J Rn \2 nQj
L1 (Rn \Z)
P
We can estimate kbj kL1 (Qj ) because the bj have support in Qj and the Qj
j∈J
are disjoint. If we invoke Claim 208, then we obtain
 
X X Z
T bj  . kbj kL1 (Qk ) . (|f (x)| + |g(x)|)dx . kf kL1 .
j∈J j∈J Rn
L1 (Rn \Z)

Thus, the proof is over modulo Claim 208.

Let us prove Claim 208. For this purpose we can write out T bj (x) in full:
√
Z
T bj (x) = K(x, y)bj (y)dy for all for a.e. x ∈ Rn \ 2 n Qj .
Rn
We can deduce from the moment condition and the Hörmander condition
|y − c(Qj )|
Z Z
|T bj (x)| = (K(x, y) − K(x, c(Qj )))bj (y)dy . n+1 j
|b (y)|dy.
Rn Qj |x − y|
√
By integrating over Rn \ 2 n Qj we have, with the aid of Fubini’s theorem,
!
|y − c(Qj )|
Z Z Z
√
|T bj (x)|dx . √
|b (y)|dy dx
n+1 j
Rn \2 n Qj Rn \2 n Qj Qj |x − y|
!
|y − c(Qj )|
Z Z
= √ n+1
dx |bj (y)|dy
Qj Rn \2 n Qj |x − y|
!
|y − c(Qj )|
Z Z
. n+1
dx |bj (y)|dy
Qj |x−y|>|x−c(Qj )| |x − y|
Z
. |bj (y)|dy.
Qj
 Various operators in Lebesgue spaces 205

Thus, we have finished the proof of the claim and hence the theorem.
Although it is impossible to have a pointwise control of singular integral
operators in terms of given functions, we can control their integral in terms
of the original functions in Lp (Rn )-norm.

Theorem 209. Let 1 < p < ∞ and T be a singular integral operator, which
is initially defined on L2 (Rn ). Let D1 and D2 be constants in (4.2) and (4.3),
respectively. Then T can be extended to Lp (Rn ) for all 1 < p < ∞ so that
kT f kLp .p (kT kL2 →L2 + D1 + D2 )kf kLp for all f ∈ Lp (Rn ) ∩ L2 (Rn ).
We still write T defined on Lp (Rn ).
Proof We may suppose that kT kL2 →L2 + D1 + D2 ≤ 1 as before. We
distinguish two cases keeping in mind that the case p = 2 is contained in the
assumption.
Case 1 : 1 < p < 2 In this case we will use interpolation of L2 (Rn )-
boundedness and weak-(1, 1) boundedness, the same technique that we used
for the maximal operator.
0
Case 2 : p > 2 In this case we use the duality Lp (Rn )-Lp (Rn ). Since
0
Cc∞ (Rn ) is dense in Lp (Rn ), for all f ∈ L2 (Rn ) ∩ Lp (Rn ),
Z 
kT f kLp = sup T f (x)g(x)dx : g ∈ Cc∞ (Rn ), kgkLp0 ≤ 1
n
 ZR 
= sup f (x)T ∗ g(x)dx : g ∈ Cc∞ (Rn ), kgkLp0 ≤ 1
Rn
∗
≤ kf kLp kT gkLp0 .

Recall that T ∗ is again a singular integral operator and that 1 < p0 < 2. Thus
0
for all g ∈ Lp (Rn ) ∩ L2 (Rn ), kT ∗ gkLp0 . kgkLp0 .
Combining these estimates allows us to conclude that T can be extended
to a bounded operator on Lp (Rn ).

4.5.5 Truncation and pointwise convergence

Having proven the boundedness property, we now turn to the maximal
operator of the truncated singular integral to check that the definition of
singular integral operators can be realized as the limit of truncated singular
integral operators: We now turn to the problem of pointwise convergence. To
do this, it is very effective that we use the maximal operator, as this is the
case when we investigated the Lebesgue point of measurable functions. The
maximal operator associated with our present situation is the following one.

Definition 61 (Truncated maximal singular integral operator). Let T be a

singular integral operator with the kernel K and define the truncated maximal
206 Morrey Spaces

singular integral operator T̃ by

Z
T̃ f (x) ≡ sup K(x, y)f (y)dy (x ∈ Rn )
ε>0 Rn \B(x,ε)

for f ∈ L0 (Rn ) as long as the definition makes sense.

Importantly, S
at this moment, we suppose that T itself is defined for any
function f ∈ Lp (Rn ), in which case we are confortable with defining
1≤p<∞
T̃ f for such functions.
Example 96. If f ∈ Lp (Rn ) for some 1 ≤ p < ∞, then the definition of
T̃ f (x) makes sense for any x ∈ Rn thanks to the size condition (4.2).
The Cotlar inequality below will control the truncated maximal singular
integral operators.
Theorem 210 (Cotlar). Suppose that T is a singular integral operator and
0 < ν ≤ 1. Let D1 and D2 be constants in (4.2) and (4.3), respectively. Then

T̃ f (x) .ν (kT kL2 →L2 + D1 + D2 )(M (ν) [T f ](x) + M f (x)) (x ∈ Rn ) (4.14)

Lp (Rn ).
S
for all f ∈
1≤p<∞

Proof We may suppose that kT R kL2 →L2 +D1 +D2 ≤ 1 as before. Fix ε > 0
and x ∈ Rn . We will estimate K(x, y)f (y)dy independently of ε.
Rn \B(x,ε)
 
1
Let z ∈ B x, ε . We will decompose f according to B(x, 2ε): Let f0 ≡
2
χB(x,ε) · f and f1 ≡ f − f0 . As for the estimate of f0 , we will employ the size
condition |K(x, y)| . |x − y|−n . It follows from the definition of f0 that
Z Z
K(x, y)f (y)dy = K(x, y)f0 (y)dy . M f (x).
Rn \B(x,ε) B(x,2ε)\B(x,ε)

Next we will consider the estimate of f1 . We decompose the estimate fur-

ther:
Z
K(x, y)f1 (y)dy
Rn \B(x,ε)
Z Z
= (K(x, y) − K(z, x))f1 (y)dy + K(z, y)f1 (y)dy.
Rn \B(x,ε) Rn \B(x,ε)

By the Hörmander condition (4.3), the first term is readily estimated by the
Hardy–Littlewood maximal function. Thus, we have
Z
K(x, y)f1 (y)dy . M f (x) + |T f1 (z)|. (4.15)
Rn \B(x,ε)
 Various operators in Lebesgue spaces 207

Thus, from (4.5.5) and (4.15) we have obtained

Z
K(x, y)f (y)dy . M f (x) + |T f1 (z)| . M f (x) + |T f0 (z)| + |T f (z)|
Rn \B(x,ε)

as long as x and z satisfy |x − y| ≤ 2ε . Taking the ν-power and then averaging

this estimate over B(x, ε), we have
Z ν
(ν) (ν)
K(x, y)f (y)dy . M f (x)ν + mB(x,ε) (T f0 )ν + mB(x,ε) (T f )ν .
Rn \B(x,ε)
(4.16)
Since 0 < ν < 1, we are in the position of applying Kolmogorov’s inequality
(Theorem 138), to have
(ν)
mB(x,ε) (T f0 ) . M f (x). (4.17)

Combining (4.16) and (4.17) gives (4.14).

We now prove the a.e. convergence of singular integral operators, keep-
ing the estimate of the maximal singular integral operators in mind. As a
preparatory step, we will show that many functions behave well.
Theorem 211. Let T be a singular integral operator with the kernel K.
Assume that the limit
Z
lim K(x, y)dy (4.18)
ε↓0 B(x,1)\B(x,ε)

exists for almost all x ∈ Rn . Then for all f ∈ Cc∞ (Rn ),

Z
lim K(x, y)f (y)dy (4.19)
ε↓0 Rn \B(x,ε)

exists for such x ∈ Rn . In particular, if K(x, x − y) = −K(x, y − x), then

(4.18) and hence (4.19) hold.
Proof By the kernel condition the matter is reduced to the convergence
of Z
lim K(x, y)f (y)dy. (4.20)
ε↓0 B(x,1)\B(x,ε)

However, by assumption (4.18), (4.20) is reduced to showing

Z
lim K(x, y)(f (x) − f (y))dy. (4.21)
ε↓0 B(x,1)\B(x,ε)

By the mean-value theorem we have |f (x) − f (y)| . |x − y|. Combining

this with the size condition |K(x, y)| . |x − y|−n , we see that the integrand
of (4.21) is integrable. Therefore, the limit in question does exist.
208 Morrey Spaces

Combining Cotlar’s inequality with Theorem 211, we can prove that the
truncated singular integral operator converges if and only if K enjoys a sort
of cancellation condition.
Theorem 212. Suppose that T is a singular integral kernel and that K is an
associated kernel. Then the limit
Z
T0 f (x) ≡ lim K(x, y)f (y)dy
ε↓0 Rn \B(x,ε)

exists for almost all x ∈ Rn , if and only if the limit

Z
lim K(x, y)dy
ε↓0 B(x,1)\B(x,ε)

exists for almost every x ∈ Rn . If this is the case, then for all 1 < p < ∞
there exists a constant cp such that

kT0 f kLp ≤ cp kf kLp , f ∈ Lp (Rn )

and there exists a constant c1 > 0 such that

c1
|{|T0 f | > λ}| ≤ kf kL1 , for all f ∈ L1 (Rn ) and λ > 0.
λ
In particular, K(x, x − y) = −K(x, y − x), then these estimates hold.
We remark that singular integral operators satisfy the condition of the
above theorem. Hence, we have the weak-(1, 1) estimate as well. Below let us
say that a generalized singular integral operator is standard if the cancellation
condition (4.18) is fulfilled. In this case we say that the kernel is standard as
well. As an example of standard singular integral operators, we can list the
Riesz transform.

4.5.6 Examples of singular integral operators

We have seen that the Riesz transform and hence the Hilbert transform
fall under the scope of Section 4.5. We will present further examples. When we
consider the integration by parts involving functions with a certain singularity,
singular integral operators arise naturally. Proposition 213 naturally extends
Example 90.
Proposition 213. Let n ≥ 3 and let A = {aij }ni,j=1 be a positive definite
matrix with inverse A−1 = {aij }ni,j=1 . Let τ ∈ C ∞ (R) be such that χ(0,∞) ≤
τ ≤ χ(1,∞) . Define
 1− n2
n
1 X
Γ(x) ≡ √  aij xi xj 
(n − 2)ωn−1 det A i,j=1
 Various operators in Lebesgue spaces 209

for x ∈ Rn . For each i, j = 1, 2, . . . , n, write

Z
cij ≡ {∂ij [τ (|x|)]Γ(x) + ∂j [τ (|x|)]∂i Γ(x) + ∂i [τ (|x|)]∂j Γ(x)} dx.
Rn

We define Z Z
p.v. ∂ij Γ(x)dx ≡ lim ∂ij Γ(x)dx.
Rn ε↓0 Rn \B(ε)
∞
Then for all ϕ ∈ Cc (Rn ),
Z Z
Γ(x)∂ij ϕ(x)dx = p.v. ∂ij Γ(x)dx + cij ϕ(0).
Rn Rn

Proof Let ϕ ∈ Cc∞ (Rn ) with ϕ(0) = 1 by normalization. Then

Z Z
Γ(x)∂ij ϕ(x)dx − p.v. ∂ij Γ(x) · ϕ(x)dx
Rn Rn
 
|x|
Z Z
= lim τ Γ(x)∂ij ϕ(x)dx − p.v. ∂ij Γ(x)dx
ε↓0 Rn ε Rn
Z       
|x| |x|
Z
= lim ∂ij τ Γ(x)ϕ(x)dx + ∂j τ ∂i Γ(x)ϕ(x)dx
ε↓0 Rn ε Rn ε
  
|x|
Z
+ lim ∂i τ ∂j Γ(x)ϕ(x)dx.
ε↓0 Rn ε
As before, in view of the size of the support, using the change of variables, we
have
Z Z
Γ(x)∂ij ϕ(x)dx − p.v. ∂ij Γ(x) · ϕ(x)dx
Rn n
Z    R    
|x| |x|
Z
= lim ∂ij τ Γ(x)dx + ∂j τ ∂i Γ(x)dx
ε↓0 Rn ε Rn ε
  
|x|
Z
+ lim ∂i τ ∂j Γ(x)dx
ε↓0 Rn ε
Z
= {∂ij [τ (|x|)]Γ(x) + ∂j [τ (|x|)]∂i Γ(x) + ∂i [τ (|x|)]∂j Γ(x)} dx.
Rn

This is the desired result.

We integrate by parts to obtain the decay of the inner product. We fix a
dim (Ḣ (Rn ))
complete orthonormal system {Yjk }j=1C k as we did in Definition 45.
Lemma 214. Let K : Rn \ {0} → C be a homogeneous smooth Calderón–
Zygmund kernel.
Also let L ≡ |x|2 ∆ be a differential operator. We set
Z
ajk ≡ K(z)Yjk (z)dσ(z) (k ∈ N0 , 1 ≤ j ≤ dimC (Ḣk (Rn ))).
S n−1

Then |ajk | .n sup kLj [| · |n K]kL∞ k −2n .

|k|≤n
210 Morrey Spaces

Proof By Proposition 126 we have

Z Z  
n Yjk (x) n Yjk (x)
L[|x| K(x)] · dσ(x) = |x| K(x)L dσ(x).
S n−1 |x|k S n−1 |x|k

Thanks to Proposition 125, we have

Z
Yjk (x)
L[|x|n K(x)] · dσ(x)
S n−1 |x|k
Z
Yjk (x)
= −k(k + n − 2) |x|n K(x)|x|2 dσ(x).
S n−1 |x|k

From the definition of S n−1 , we have

Z
Yjk (x)
L[|x|n K(x)] · dσ(x)
S n−1 |x|k
Z
Yjk (x)
= −k(k + n − 2) |x|n K(x) dσ(x).
S n−1 |x|k

If we go through the same argument n times, we obtain

Z
ajk = (−k)−n (k + n − 2)−n Lj [|x|n K(x)]Ykm (x)dσ(x).
S n−1

It remains to use k + n − 2 ≥ k.
With these preliminary observations in mind, we present an example of sin-
gular integral operators. Roughly speaking, we will see that we can handle the
operator of the form T f (x) = k(x)T0 f (x), where T0 is a singular integral oper-
ator and k ∈ L∞ (Rn ). We consider the kernel (x, z) ∈ Rn × Rn 7→ K(x, z) ∈ C
below.
Theorem 215. Let K ∈ L0 (Rn × Rn ) satisfy the following conditions:
(1) K(x, ·) ∈ C ∞ (Rn \ {0}) for all x ∈ Rn ;
(2) K(x, tz) = t−n K(x, z) for every t > 0 and (x, z) ∈ Rn × S n−1 ;
Z
(3) K(x, z)dσ(z) = 0;
S n−1

∂αK
(4) max < ∞.
|α|≤2n ∂z α L∞ (Rn ×S n−1 )

Define Z
T f (x) ≡ K(x, x − y)f (y)dy (x ∈ Rn )
Rn

for f ∈ Cc∞ (Rn ). Then T extends to an Lp (Rn )-bounded linear operator.

 Various operators in Lebesgue spaces 211
dim (Ḣk (Rn )) ∞
Proof We expand the kernel K using {{Yjk }j=1C }k=0 . Due to (3),
we have
n
Ḣk (R
∞ dimC (X )) Z
X Yjk (z)
K(x, z) = K(x, y)Yjk (y)dσ(y) · .
j=1 S n−1 |z|k+n
k=1

Along this decomposition, we decompose T to write

n
Ḣk (R
∞ dimC (X ))
X
T f (x) = Tjk f (x). (4.22)
k=0 j=1

We claim that the series converges absolutely thanks to the strong decay in
each summand. Due to Example 93,
Z
Yjk (z) n

k+n
f (· − z)dz . (1 + k) 2 kf kLp .
Rn |z| Lp

We use (1), (2) and (4). Since we have

Z
K(·, y)Yjk (y)dσ(y) . k −2n
S n−1 L∞

due to Lemma 214, we conclude from Corollary 124 that (4.22) takes place in
the topology of the operator norm k · kLp →Lp . Thus, T is Lp (Rn )-bounded.
We can also discuss the pointwise convergence for this class of operators.
Theorem 216. Let K ∈ L0 (Rn × Rn ) satisfy the following conditions:
(1) K(x, ·) ∈ C ∞ (Rn \ {0}) for all x ∈ Rn ;
(2) K(x, tz) = t−n K(x, z) for every t > 0 and (x, z) ∈ Rn × S n−1 ;
Z
(3) K(x, z)dσ(z) = 0 for all x ∈ Rn ;
S n−1

∂αK
(4) max < ∞.
|α|≤2n ∂z α L∞ (Rn ×S n−1 )

Let f ∈ Lp (Rn ) with 1 < p < ∞. For any ε > 0, we set

Z
Tε f (x) ≡ K(x, x − y)f (y)dy (x ∈ Rn ).
Rn \B(x,ε)

Then there exists a function T f ∈ Lp (Rn ) such that lim kTε f − T f kLp = 0.
ε↓0

Proof This is a general fact on singular integral operators together with

the decomposition (4.22). Use Theorem 211 as well.
212 Morrey Spaces

4.5.7 Sparse estimate of singular integral operators

Having set down some elementary properties, we now embark on the sparse
estimate. Hence, we are now converting our results above to some forms useful
to applications. We will need to estimate the quantities related to T f , where
f ∈ L∞ n
c (R ). First, we handle the local oscillation ω2−n−2 (T f ; Q).

Lemma 217. Let T be a singular integral operator, and let Q be a cube. Then
∞
X
ω(Re(T f ); Q) + ω(Im(T f ); Q) . 2−l m2l Q (|f |)
l=0

for all f ∈ L∞ n
c (R ).

Proof We consider the real part solely. We set c ≡ Re(T [f ·χRn \3Q ](c(Q))).
Then there exist D1 , D2 > 0 such that
Z
`(Q)|f (y)|
|Re(T f (x)) − c| ≤ D1 |T [f χ3Q ](x)| + D1 dy
n
R \Q |y − c(Q)|n+1
∞
X
≤ D2 |T [f χ3Q ](x)| + D2 2−l m2l Q (|f |).
l=0

Since T is weak-(1, 1) bounded, there exists a constant D3 > 0 such that

Z
D3
|{x ∈ Q : |Re(T [f χ3Q ](x))| > λ}| ≤ |f (y)|dy
λ 3Q
for all λ > 0. Thus,
∞
( )
n+2
X m2l Q (|f |)
x ∈ Q : |Re(T f (x)) − c| > 9 D2 D3 m4Q (|f |) + D2
2l
l=0
|Q|
≤ .
2n+2
Consequently, we have the desired result.
As an application of the Lerner–Hytönen decomposition (see Theorem 174)
and Lemma 217, we obtain the following control of singular integral operators.
The following estimate, which can be viewed as a direct corollary of Lemma
217, converges singular integral operators to non-negative integral operators
in some sense. Consequently, this estimate is a bridge that connects singular
integral operators with non-negative integral operators.
Lemma 218. Let Q0 be a cube, and let f ∈ L1 (Q0 ). Then
∞ X
X
χQ0 |T f − Med(T f ; Q0 )| . 2−l χL m2l L (|f |).
l=0 L∈L

Here, L ⊂ D(Q0 ) is a sparse family of dyadic cubes that depends on f .

 Various operators in Lebesgue spaces 213

Proof If we use the Lerner–Hytönen decomposition, we can find a sparse

family L ≡ {Qjk }j∈N0 , k∈Kj ⊂ D(Q0 ) with a level structure such that
∞ X
X
χQ0 |T f − Med(T f ; Q0 )| ≤ ω2−n−2 (T f ; Qjk )χQj .
k
j=0 k∈Kj

It remains to resort to Lemma 217.

4.5.8 Local estimates for singular integral operators

As we did for other operators we obtain the local estimate of singular
integral operators.
Theorem 219. Let 1 < q < ∞. Then
Z ∞
n n
kT f kLq (B(x,r)) . r q t− q −1 kf kLq (B(x,t)) dt (4.23)
r

for all x ∈ Rn and r > 0 and for all f ∈ L∞ n

c (R ).

The idea of the proof is similar. The only difference from the case of the
maximal operator lies in the point where we handle the integral kernel with
the size condition.
Proof We split f = f1 + f2 , where f1 ≡ χB(x,2r) f and f2 ≡ f − f1 . For
f1 , apply the Lq (Rn )-boundedness of T and, for f2 , apply the size condition
together with the triangle inequality.
We transform Theorem 219 into the form which we use.
n
Corollary 220. Fix x ∈ Rn . Let 1 < q < ∞ and 0 < δ < q. Then
! ! q1
Z ∞ Z
n
q −δ
dt
kT f kLq (B(x,r)) . r |f (z)|q dz
r B(x,t) tn−δq+1

for all r > 0 and for all f ∈ L∞ n

c (R ).

Proof It suffices to apply Hölder’s inequality to the integral in the right-

hand side of (4.23).

4.5.9 Exercises
Exercise 73.
(1) Prove (4.5) and (4.6) using the Stokes theorem. Alternatively, we may
argue as follows: Let ρ ∈ C ∞ (Rn ) satisfy χ(1,∞) ≤ ρ ≤ χ(2,∞) . For
ρ(ε|x − y|)f (y)
Z
ε > 0, we set I2,ε f (x) ≡ dy (x ∈ Rn ). Prove first
Rn |x − y|n−2
that I2,ε f converges in some suitable topology.
214 Morrey Spaces

(2) Show that ∂j ∂k I2 is a Calderón–Zygmund operator if 1 ≤ j, k ≤ 2.

n
Z x ∈ R . Show that the limit Z
Exercise 74. Let defining Rj f (x) exists for f ∈
∞ n xj − yj xj − yj
Cc (R ) using n+1
f (y)dy = n+1
(f (y)−
B(1)\B(x,ε) |x − y| B(1)\B(x,ε) |x − y|
f (x))dy and the mean-value theorem.

4.6 Notes
Section 4.1
General remarks and textbooks in Section 4.1
See the fundamental textbooks and lecture notes [104, Chapter 2], [146,
Chapter II, §2], [156, §2.1], [223, Chapter 1] [293, §1.1], [319, §1.10], [417,
Chapter 2, §3], [416, Chapter I, §3 and Chapter II, §1], [382, §1.4] and [410,
§2.3] for the maximal operator considered in this book.
For the Fefferman–Stein vector-valued inequality, we refer to [146, Chapter
V, §4], where a different approach can be found.
See [118, §1.5] for various covering lemmas such as Vitali’s covering lemma
and Besicovitch’s covering lemma.
We viewed the Hedberg inequality in the context of Hedberg’s inequality;
see Lemma 181 for more about Example 61.

Section 4.1.1
The history of the Hardy–Littlewood maximal operator dates back to 1930
[182]. The original aim was to apply it to the functions on the unit disk ∆(1) on
the complex plane. Consequently, originally the Hardy–Littlewood maximal
operator was used for the analysis over the torus T = R/2πZ. Here, to study
the function spaces on R, we work in R. A natural passage to higher dimensions
is due to Wiener [449]. His purpose was to apply it to the ergodic theory.

Section 4.1.2
In retrospect, the proof of Theorem 141 essentially relies upon the interpo-
lation theorem due to Marcinkiewicz [302] whose proof Zygmund gave [473].
Wiener proved the sunrise lemma in [449]; see Theorem 137.

Section 4.1.3
We showed how to calculate the maximal operator on a cube Q for func-
tions supported outside 2Q. Lemma 130 is well known. See [387, Lemma 4.2]
for example. Fefferman and Stein considered Example 68.
 Various operators in Lebesgue spaces 215

See [43] for the relation between the Kantorovitch operator and the Hardy–
Littlewood maximal operator on [0, 1].

Section 4.1.4
See [124] for the Fefferman–Stein vector-valued inequality, Theorem 145.
Our proof depends on the method of [378, Theorem 1.3].

Section 4.1.5
Carro, Pérez, F. Soria and J. Soria used the maximal operator ML log L to
investigate the dual inequality of Stein-type; see [64, Theorem 3.1].

Section 4.1.6
Gogatishvili, Mustafayev and Agcayazi considered the composition of the
maximal operator and fractional maximal operators. See [152, Theorem 4.1]
with α = 0 for Corollary 155.

Section 4.1.7
Sawano, Sugano and Tanaka characterized the condition on the bounded-
ness for the maximal operator MΨ to be locally bounded in the norm deter-
mined by Φ including the powered maximal operator M (η) in [392, Proposition
5.7], [392, Claim 2.18] and [392, Corollary 2.19]; see Proposition 156, Theorem
157 and Corollary 158, respectively.

Section 4.1.8
Recall that we considered the integral expression of the integral kernel
| · |α−n . See the survey paper [391, Lemma 3.2] for Lemma 160. We recorded
a general estimate on convolution by Burenkov and Guliyev; see [46, Lemmas
5,6 and 7] for Propositions 163, 164(1) and 164(2), respectively.

Section 4.2
General remarks and textbooks in Section 4.2
See the fundamental textbooks [29, Chapter 5 §8], [104, Chapter 6], [146,
Chapter II, §3], [157, §3.4], [293], [416, Chapter IV, §2] and [410, §6.2].

Section 4.2.1
Lerner proved the sharp maximal inequality, Theorem 165, [267, Theorem
2.1]. The sharp maximal inequality, Theorem 166, is due to Fefferman and
Stein; see [125, Theorem 5] as well as [125, (4.1)].
216 Morrey Spaces

Section 4.2.2
Fujii refined the Lebesgue differentiation theorem as in Lemma 168 [136].

Section 4.2.3
Lerner considered the local mean oscillation and the local sharp maximal
function in [269, Section 4] see also [270, Section 2.3]. Theorem 174 has a
history. Lerner prove Theorem 174 with an extra term in [270]. Later on
Hytönen proved Theorem 174 as it stands. We followed [270, Theorem 4.5].
We followed observations on Da made by Lerner [270, Proposition 2.1]. See
[269, Lemma 3.1] for Lemma 171.
We constructed a sparse family in Theorem 170; see the work by Tanaka
[424].

Section 4.3
General remarks and textbooks in Section 4.3
We refer to the fundamental textbooks [156, Exercise 1.2.6] and [293, §3.2].
See also [104, p. 89].
The fractional maximal operator can control the operator of the form
V α (−∆ + V )−β with 0 ≤ α ≤ β ≤ 1 [421, Theorem 1].

Section 4.3.1
Adams and Xiao calculated Iα [| · |−β ] in [8, Example 5.5]; see Exam-
ple 82. Theoren 182 is known as the Hardy–Littlewood–Sobolev inequality
[184, 185, 411]. We employed the method of Hedberg [190]. Lemma 181,
known as the Hedberg inequality, is nowadays a standard method of prov-
ing the Hardy–Littlewood–Sobolev inequality. It is nowadays common to use
the maximal operator and fractional maximal operators to control fractional
integral operators. See [446, p.146] for Lemma 183 and [7, Proposition 3.1.2]
for its variants.
Next, we consider fractional integral operators with rough kernel.
Definition 62 (Fractional integral operators with rough kernel). Let s > 1,
and let Ω ∈ L0 (Rn ) be a function homogeneous of degree zero such that
Ω|S n−1 ∈ Ls (S n−1 ). Let 0 < α < n.
(1) One defines the fractional maximal operator MΩ,α with rough kernel Ω
by Z
1
MΩ,α f (x) ≡ sup n−α |Ω(y)f (x − y)|dy (x ∈ Rn ).
r>0 r B(r)

(2) One defines the fractional integral operator IΩ,α with rough kernel Ω by
Ω(y)f (x − y)
Z
IΩ,α f (x) ≡ n−α
dy (x ∈ Rn ).
Rn |x − y|
 Various operators in Lebesgue spaces 217

These operators go back to [101, p. 29]. See also the textbook [293, §3.6].
See [293, Chapter 2] for a counterpart to singular integrals together with a
detailed motivation. See [311, Theorem 1.1] for the Morrey–BMO boundedness
and [?, 375] for the Morrey boundedness.

Section 4.3.2
See [164, Theorem 4.1] for Theorem 178 with α = 0 and p2 = 1.

Section 4.3.3
See [91, Theorem 5.1] for the sparse estimate of the fractional maximal
operator Mα . Theorem 179, a sparse estimate for fractional maximal opera-
tors, is [334, Lemma 2.3].

Section 4.4
See [90, Proposition 2.3] for the sparse estimate of fractional integral oper-
ators Iα .

General remarks and textbooks in Section 4.4

See [4], [7, §1.2.2 and Chapter 3], [231], [293], [319] and [416, p. 354] for
fractional integral operators. See [117, Chapter 5, Theorem 4], [150, Theorem
7.19] for Morrey’s lemma.
There are many applications of fractional integral operators. For exam-
ple, we can find applications of fractional integral operators to the restriction
problem in [416, p. 352].
See [7, §1.2.3], [319, Chapter 7] and [415, Chapter III, §2] for the Bessel
kernel.

Section 4.4.1
Example 84 can be found in [320, 394] for example. See [16] for the Sobolev
embedding in terms of Morrey spaces over metric measure spaces. We followed
[319] for Lemma 180.

Section 4.4.4
We have some explicit formulas of the solution of the variable coefficient
elliptic differential equations; see [60] for Theorem 192.
We know that functions are controlled by the Riesz potential I1 of their
gradient; see [439, Lemma 1] as well as [415, Section V.2.3] for Theorem 193.

Section 4.4.2
We followed the works by Burenkov, Gogatishvili, Guliyev and Mustafayev
for the local estimate of fractional integral operators; see [46, Lemma 8], [44,
Lemma 3.3] and [44, Lemma 3.6] for Theorem 184.
218 Morrey Spaces

Section 4.4.3
The idea of decomposing fractional integral operators, as we did in Propo-
sition 188, dates back to 1995; see the paper [354, Section 3] by Pérez. See [199]
for a non-linear version of the discrete fractional integral operator appearing
in Proposition 188. The sparse estimate used here is [334, Lemma 3.1]; see
Theorem 191.

Section 4.4.5
The Bessel potential is a potential named after Friedrich Wilhelm Bessel,
a German astronomer, mathematician, physicist and geodesist.

Section 4.4.6
See the original paper [323] by Morrey for Morrey’s lemma. Cianchi and
Pick used Morrey’s lemma to show that embedding relation between Sobolev
(type) spaces and generalized Campanato spaces [79].

Section 4.5
General remarks and textbooks in Section 4.5
See the fundametal papers [28, 55] for the theory of singular integral oper-
ators. A general overview can be found in [53].
See [104, Theorem 2.11], [293, §1.2], [416, Chapter I, §4] (for general dou-
bling measures), [416, Chapter IV, §3] (using the dyadic cubes) [382] for the
Calderón–Zygmund decomposition.
For further results on singular integral operators considered in this book,
we refer to the textbooks [104, Chapters 3,4 and 5], [146, Chapter II, §5],
[147, Chapter 5], [223, Chapter 4], [293, §2.1], [415, Chpaters II and III], [417,
Chapter VI], [416, Chapter 1, §5], [382, §1.5] and [410, §6.4].
Although we did not consider the boundedness of singular integral oper-
ators from L∞ (Rn ) to BMO(Rn ), this is due to Spanne, Peetre and Stein
[352, 405, 414].

Section 4.5.1
The weak-(1, 1) boundedness of the Riesz transform is due to Kolmogorov
[237]. Its strong-(p, p) boundedness is due to M. Riesz [367, 368]. See also
[51, 287].

Section 4.5.2
We refer to [83] for Definition 57. We borrowed the term “genuine singular
integral operator” from [48]; see also [47, p. 34].
 Various operators in Lebesgue spaces 219

Section 4.5.3
We refer to [54, Lemmas 1 and 2] for Theorem 205.

Section 4.5.4
See [54, Lemma 2] for the weak estimate of singular integral operators,
Theorem 207.

Section 4.5.5
See [54, Chapter 4] for Theorem 211.

Section 4.5.6
Theorems 215 and 216 (see [76, Theorem 2.5]) including Lemma 214 (see
[76, (2.6)]) are some of the important results in the theory of elliptic differen-
tial operators; see [463] for an application of Theorem 215 together with the
technique of the proof.

Section 4.5.7
Lerner obtained a control of ω(Re(T f ); Q) and ω(Im(T f ); Q) as in Lemma
217; see [271, Proposition 2.3].

Section 4.5.8
We followed Guliyev’s works [162, 163] for the local estimates; see Theorem
219. Corollary 220, the local estimate for singular integral operators, is [47,
Corollary 1].
Chapter 5
BMO spaces and
Morrey–Campanato spaces

Let us look at Morrey’s original observation in 1938 [323]. For 0 < θ < 1, recall
that the Lipschitz space of order θ denotes the set of all continuous functions
F for which kF kLipθ ≡ sup |x − y|−θ |F (x) − F (y)| is finite. Roughly,
x,y∈Rn ,x6=y
he showed that if f ∈ L1loc (Rn ) has a weak derivative in Mp1 (Rn ) with p > n,
n
then f ∈ Lip1− p (Rn ).
However, it is not easy to guarantee that a function in L1loc (Rn ) has a weak
derivative, let alone that it belongs to Mp1 (Rn ) for some p ∈ [1, ∞). Often,
it is taken for granted that the functions are locally integrable. Consequently,
there is a gap between what is assumed and what needs to be proven.
It is desirable to obtain some smoothness of locally integrable functions
from inequalities derived from integration. Actually, this chapter is interested
in how to replace the condition for a derivative belonging to Mp1 (Rn ). Morrey–
Campanato spaces can be used for this purpose. As the limiting case, the
bounded mean oscillation (BMO) space, arises naturally.
Section 5.1 investigates the BMO space since it is a fundamental func-
tion space. Section 5.2 handles operators made up of BMO functions. We
define Morrey–Campanato spaces in Section 5.3. Section 5.3, which includes
an auxiliary observation to Morrey–Campanato spaces, shows that Morrey–
Campanato spaces are isomorphic to known function spaces.

5.1 The space BMO(Rn ) and commutators

Generally speaking, the commutator is an operator of the form [A, B] =
AB − BA, where A and B are linear operators. Here, we are mainly concerned
with the case where A is a singular integral operator defined and discussed in
Section 4.5 and B is a pointwise multiplier generated by BMO functions; for
b ∈ BMO(Rn ), consider the operator f 7→ b · f . In Section 5.1.1 we will define
BMO(Rn ) to be the set of all L1loc (Rn )-functions a for which
Z
1
kak∗ ≡ sup |a(x) − mQ (a)|dx
Q∈Q |Q| Q

221
222 Morrey Spaces

is finite. This mapping makes sense despite ambiguity caused by the additive
constant in BMO(Rn ) as long as we consider the commutator. This type of
operator will be used when we consider the elliptic differential operators. We
aim in Section 5.1 to define the commutator described above more precisely.
One of the prominent properties of BMO(Rn ) is the distributional inequality
which is called the John–Nirenberg inequality and is investigated in Section
5.1.2. The John–Nirenberg inequality will be a key tool when we consider
operators involoved with BMO.

5.1.1 The space BMO

As we saw in Example 85, some important operators dealt with in this
book fail to be L∞ (Rn )-bounded. The space BMO(Rn ) will compensate for
this problem.
Definition 63 (BMO(Rn )). Define kf k∗ ≡ sup mQ (|f − mQ (f )|) for f ∈
Q∈Q
L1loc (Rn ). One says that f ∈ L1loc (Rn ) has bounded mean oscillation (abbrevi-
ated to f ∈ BMO(Rn )), if kf k∗ < ∞. One also calls BMO(Rn ) the BMO(Rn )
space.
Concerning terminology, clarifying remarks may be in order.
Remark 10.
(1) Strictly speaking, the BMO(Rn )-norm is not a norm appearing in func-
tion analysis: in fact, the “BMO(Rn )-norm” annihilates nonzero con-
stants. Conversely, if a function f has 0 “BMO(Rn )-norm”, then f is a
constant almost everywhere. Consequently, consider the quotient linear
space, and redefine the “BMO(Rn )-norm” there. Then BMO(Rn ) as a
quotient linear space is a Banach space. However, the BMO(Rn ) space
is not regarded as a quotient space in this book; so there are nonzero
functions whose “BMO(Rn )-norm” is 0.
(2) Nevertheless, it is noteworthy that one can make the set BMO(Rn ) into
a Banach space (BMO(Rn ), k·kBMO+ ) by the norm given by kf kBMO+ ≡
kf kL1 (Q(1)) + kf kBMO . The price to pay here is to distort symmetry.

Example 97.
(1) In the sense of Remark 10, L∞ (Rn ) ,→ BMO(Rn ), since mQ (|f |) ≤
kf kL∞ for all f ∈ L∞ (Rn ).
(2) Let f (x) ≡ log |x| for x ∈ Rn . Then f ∈ BMO(Rn ). In fact, we need to
check mQ (|f −mQ (f )|) . 1 for all cubes Q. By the scaling f 7→ f (a·), we
may assume that |Q| = 1. If |Q| = 1 and |c(Q)| ≤ 2n, then mQ (|f |) . 1,
so that it is clear that mQ (|f − mQ (f )|) . 1. Meanwhile if |Q| = 1 and
|c(Q)| > 2n, then we can use | log |y| − log |z|| . 1 for all y, z ∈ Q.
 BMO spaces and Morrey–Campanato spaces 223

Note that f ∈ BMO(Rn ) if and only if M ] f ∈ L∞ (Rn ).

The next proposition reveals to us how large each BMO(Rn ) function is.
Here and in what follows we tacitly mean by f ∈ BMO(Rn ) that f belongs
to the normed space BMO(Rn ) or that f ∈ L1loc (Rn ) with kf k∗ < ∞. The
confusion can rarely occur.
|f (x) − mQ(1) (f )|
Z
n
Proposition 221. For all f ∈ BMO(R ), dx . kf k∗ .
Rn 1 + |x|n+1
Proof Let Q ∈ Q. Then

|mQ (f ) − m2Q (f )| = mQ (|f − m2Q (f )|) ≤ 2n m2Q (|f − m2Q (f )|) ≤ 2n kf k∗ .

As a result,
|mQ (f ) − m2j Q (f )| ≤ 2n jkf k∗ (5.1)
for j ∈ N. Therefore,
|f (x) − mQ(1) (f )|
Z
dx
Rn 1 + |x|n+1
∞ Z
|f (x) − mQ(1) (f )| |f (x) − mQ(1) (f )|
Z X
≤ n+1
dx + dx
Q(1) 1 + |x| j=1 Q(2 )
j 1 + |x|n+1
∞ Z
X 1
≤ |f (x) − mQ(1) (f )|dx
j=1
2(j−1)(d+1) Q(2j )
∞ Z !
X j 1
. · kf k∗ + |f (x) − mQ(2j ) (f )|dx .
j=1
2j |Q(2j )| Q(2j )

By the definition of the BMO(Rn )-norm, we have

∞
|f (x) − mQ(1) (f )|
Z X j
n+1
dx . · kf k∗ . kf k∗ .
Rn 1 + |x| j=1
2j

This is what we want.

In the next section we deal with more precise estimates. Despite the fact
that k1k∗ = 0, we are still eager to regard BMO(Rn ) as a Banach space
without distorting the symmetry. To do this, we first observe the following,
which is an obstacle when we regard BMO(Rn ) as a normed space.
Lemma 222. Suppose that f ∈ L1loc (Rn ) satisfies kf k∗ = 0. Then f is a.e.
constant. The converse is also true.
Proof This is a direct consequence of Proposition 221.
It should be noted that C is embedded into L∞ (Rn ) naturally. Conse-
quently, if we (re)define

BMO(Rn ) ≡ {f ∈ L1loc (Rn ) : kf k∗ < ∞}/C,

224 Morrey Spaces

then BMO(Rn ) is a normed space. We write f mod C, if we want to disregard

the difference of additive constants of f .
We can use Proposition 221 to show that BMO(Rn ) is a Banach space.
Theorem 223. The space BMO(Rn ) is a Banach space.
∞
Proof Let fj ∈ BMO(Rn ) for j = 1, 2, . . . and assume
P
kfj k∗ < ∞.
j=1
n+1
We assume mQ(1) (fj ) = 0 for each j, so that fj ∈ L1 ((M χQ(1) ) n ) and
∞
P
kfj k 1 n+1 . kfj k∗ thanks to Proposition 221. Thus F ≡ fj con-
L ((M χQ(1) )
n ) j=1
n+1
∞
verges in L1 ((M χQ(1) ) ). We claim that F ∈ BMO(Rn ) and that
P
n fj = F
j=1
in BMO(Rn ).
Given R ∈ Q, we have
∞
X ∞
X
mR (|F − mR (F )|) ≤ mR (|fj − mR (fj )|) ≤ kfj k∗
j=1 j=1

by Lebesgue’s convergence theorem. Taking the supremum over R ∈ Q, we

see that F ∈ BMO(Rn ).
In the same way we can prove
  
XJ XJ X∞
mR  F − fj − mR F − fj   ≤ kfj k∗ .
j=1 j=1 j=J+1

∞
fj = F in BMO(Rn ).
P
Thus, we have proven that
j=1
As a result, BMO(Rn ) is a Banach space.
Although we have remarked that we have to identify functions modulo
additive constants, we are still able to embed BMO(Rn ) to the set of all mea-
surable functions. To do this, we choose a special representative with integral
over Q(1) zero. Nevertheless, we usually regard BMO(Rn ) as a function space
modulo C.

5.1.2 John–Nirenberg inequality

We go into the detailed discussion of the property of BMO(Rn ). For the
time being, we are oriented to showing that BMO(Rn ) is close to L∞ (Rn ).
The John–Nirenberg inequality below is a fundamental tool when we handle
BMO(Rn )-functions.
Theorem 224 (John–Nirenberg inequality). There exists c > 0 such that
 
cλ
|{x ∈ Q : |b(x) − mQ (b)| > λ}| .n |Q| exp −
kbkBMO
 BMO spaces and Morrey–Campanato spaces 225

for all λ > 0, cubes Q, and non-constant BMO(Rn )-functions b, or equiva-

lently,
|{x ∈ Q : |b(x) − mQ (b)| > λkbkBMO }| .n |Q| exp (−c λ)
for all λ > 0, cubes Q, and BMO(Rn )-functions b.
Proof Here and below, we suppose that Q is a right-open cube. If λ is
small, then the assertion is trivial; the right-hand side is more than the con-
stant multiple of |Q| and the left-hand side never exceeds |Q|. With this in
mind, let us prove
x ∈ Q : |b(x) − mQ (b)| > k 2n+2 kbkBMO ≤ 21−k |Q|

(5.2)
by induction on k. Once this is achieved, we let λ = k 2n+2 kbkBMO and then
pass to the continuous variables, we obtain the inequality.
When k = 1, then (5.2) is trivial. Assume that
x ∈ R : |b(x) − mR (b)| > k 2n+1 kbkBMO ≤ 21−k |R|

(5.3)
for all cubes R. We aim to prove the inequality
x ∈ Q : |b(x) − mQ (b)| > (k + 1) 2n+1 kbkBMO ≤ 2−k |Q|

(5.4)
for all cubes Q. To this end, we can assume that
|{x ∈ Q : |b(x) − mQ (b)| > (k + 1) 2n+1 kbkBMO }| > 0;
otherwise there is nothing to prove.
By the Lebesgue differentiation theorem (see Theorem 253 below), there
exists a set Z of measure zero and S ∈ D(Q) \ {Q} containing x such that
mS (|b − mQ (b)|) > 2kbkBMO as long as x ∈ Q \ Z satisfies |b(x) − mQ (b)| >
(k + 1) 2n+1 kbkBMO . Choose a maximal cube S(x) among such cubes. By the
maximality, we have mS(x) (|b − mQ (b)|) ≤ 2n+1 kbkBMO . Hence, |mS(x) (b) −
mQ (b)| ≤ mS(x) (|b − mQ (b)|) ≤ 2n+1 kbkBMO . If S(x) and S(y) are disjoint,
maximality forces S(x) and S(y) to be identical. Thus we have a countable
collection Db,k+1 (Q) ≡ {Qλ }λ∈Λ ⊂ D(Q) of disjoint cubes such that
2−n−1 |mQλ (b) − mQ (b)| ≤ kbkBMO ≤ 2−1 mQλ (|b − mQ (b)|), (5.5)
[
{x ∈ Q : |b(x) − mQ (b)| > (k + 1) 2n+1 kbkBMO } ⊂ Qλ ∪ Z. (5.6)
λ∈Λ

Since {Qλ }λ∈Λ is a family of non-overlapping dyadic cubes with respect

to Q, and inequality (5.5) holds, we have
X 1 XZ
|Qλ | ≤ |b(x) − mQ (b)|dx
2kbkBMO
λ∈Λ λ∈Λ Qλ
Z
1
≤ |b(x) − mQ (b)|dx
2kbkBMO Q
|Q|
≤ . (5.7)
2
226 Morrey Spaces

Hence,

|{x ∈ Q : |b(x) − mQ (b)| > (k + 1) 2n+1 kbkBMO }|

X
= |{x ∈ Qλ : |b(x) − mQ (b)| > (k + 1) 2n+1 kbkBMO }|
λ∈Λ
(from (5.6) and the fact that {Qλ }λ∈Λ is non-overlapping)
X
≤ |{x ∈ Qλ : |b(x) − mQλ (b)| > k 2n+1 kbkBMO }| (from (5.5))
λ∈Λ
X
≤ 21−k |Qλ | (from the induction assumption (5.3) on k)
λ∈Λ
−k
≤ 2 |Q| (from (5.7)).

Thus (5.4) was proven.

The John–Nirenberg inequality yields a corollary. This shows that any
BMO(Rn ) function is Lp (Rn )-integrable for all 1 ≤ p < ∞, and as a result we
can say that BMO(Rn ) is close to L∞ (Rn ).
(p)
Corollary 225. Let 1 ≤ p < ∞. Then, mQ (b − mQ (b)) .p kbkBMO for
all b ∈ BMO(Rn ) and all cubes Q. More precisely, there exists a constant D
(p)
independent of p, Q and b such that mQ (b − mQ (b)) ≤ Dp kbkBMO .
Proof Denote by Γ the Gamma function. We use the Layer-Cake formula:
Z Z ∞
|b(x) − mQ (b)|p dx = p λp−1 |{x ∈ Q : |b(x) − mQ (b)| > λ}|dλ
Q 0
Z ∞  
p−1 c1 λ
. p λ exp − dλ
0 kbkBMO
' p Γ(p) · kbkBMO p
= Γ(p + 1) · kbkBMO p .
p
Since we know p
Γ(p + 1) ∼ p for 1 ≤ p < ∞, we obtain the desired result.
We have the following integrability estimate:

Corollary 226. There exists a constant θ > 0 depending on n such that the
following inequality:
 
|b(x) − mQ (b)|
Z
1
exp θ dx . 1
|Q| Q kbkBMO

holds for all non-constant BMO(Rn )-functions b and all cubes Q.

 BMO spaces and Morrey–Campanato spaces 227

j j!(j + 1)j+1
Proof By the Cauchy–Hadamard test, lim √ = lim =
j→∞ j j! j→∞ j j (j + 1)!
1
e < 3. Use a constant D in Corollary 225. Set θ ≡ . Then,
3D
∞
!
|b(x) − mQ (b)|j
 
|b(x) − mQ (b)|
Z Z
1 X 1
exp θ dx = θ dx
|Q| Q kbkBMO j=0
j!|Q| Q kbkjBMO
∞
X (Dθ)j j j
≤
j=0
j!
. 1.

Thus, the proof is complete.

5.1.3 Exercises
Exercise 75. Let Γ be the Gamma function given by
Z ∞
Γ(p) ≡ tp−1 e−t dt (p > 0).
0

(1) Show that Γ(p) = (p − 1)! for all p ∈ N.

p
(2) Show that p Γ(p + 1) ∼ p for p ≥ 1.
Exercise 76. Let b ∈ BMO(Rn ) and Q ∈ Q. Set Db,0 (Q) ≡ {Q}. For a
cube T ∈ Q and k ∈ N0 , choose Db,k+1 (T ) satisfying
S (5.5) and (5.6) with Q
replaced by T . Set S0 ≡ {Q} and Sj+1 ≡ Db,j+1 (T ) for j ∈ N0 . Then
T ∈Sj
reexamine the proof of Theorem 224 to show that {Sj }∞
j=0 is 1/2-sparse.

Exercise 77. [344, Lemma 2.1] Let 1 ≤ p < ∞ and Q ∈ Q. Then show that
Z Z p  p1 Z Z 1
p p
|u(x) − u(y)|dy ≤2 (u(x) − u(y))dy
Q Q Q Q

for all u ∈ Lp (Q). Hint: Use the decomposition u(x) − u(y) = u(x) − mQ (u) +
mQ (u) − u(y) to the left-hand side.
Exercise 78. [331, Lemma 2.2] Let F : C → C be a Lipschitz mapping.
(1) Let f ∈ L1loc (Rn ). Show that mQ (|F ◦ f (x) − mQ (F ◦ f )|) ≤
kF kLip mQ (|f (x) − mQ (f )|).
(2) If f ∈ BMO(Rn ). Then show that F ◦ f ∈ BMO(Rn ) together with the
estimate kF ◦ f k∗ ≤ kF kLip kf k∗ .
228 Morrey Spaces

5.2 Commutators
Commutators will play a key role when we consider elliptic differential
equations. Especially, commutators generated by BMO and singular integral
operators will arise as the solution operators of the elliptic differential equa-
tions. We will consider commutators generated by BMO and singular integral
operators and commutators generated by BMO and fractional integral oper-
ators in Sections 5.2.1 and 5.2.2, respectively.

5.2.1 Commutators generated by BMO and singular

integral operators
Section 5.2.1 takes up an operator of the form
Z
[a, T ]f (x) ≡ (a(x) − a(y))K(x, y)f (y)dy, (5.1)
Rn

where T is a singular integral operator whose kernel is K and a ∈ BMO(Rn ).

Such an operator [a, T ] is called a commutator. We will notice that [a, T ]
behaves similarly to T . Let 1 < p, q, r < ∞ and assume 1r = p1 + 1q .
Then if a ∈ Lq , then [a, T ] maps Lp (Rn ) boundedly into Lr (Rn ). But the
cancellation should give something else. Assume in addition that T0 f (x) ≡
defines an Lp (Rn )-bounded operator. We aim in Section
R
Rn
K(x, y)f (y)dy
5.2.1 to apply the sharp maximal inequality to prove the boundedness of this
operator. Because of the singularity of the kernel of [a, T ], we have to begin
with the task of making clear what (5.1) means.
Keeping in mind that we are going to use the sharp maximal operator M ] ,
let us fix a cube Q. Although BMO(Rn ) differs from L∞ (Rn ), it is still close to
L∞ (Rn ). Once we choose a cube Q, the restriction of any element in BMO(Rn )
to Q is in Lu (Q), for all u < ∞ because of the John Nirenberg inequality. In
view of this observation we are tempted to rewrite (5.1) as follows :
[a, T ]f (x) = (a(x) − mQ (a))T0 f (x) − T0 [(a − mQ (a))f ](x). (5.2)
p n
The first term of (5.2) seems nice, because K is assumed L (R )-bounded.
However, we are still at a loss because (a−mQ (a))f does not belong to Lp (Rn ).
Thus we have to work more. We decompose the above formula once more;
[a, T ]f (x) = (a(x) − mQ (a))T0 f (x) − T [(a − mQ (a))χ2Q · f ](x)
Z
− K(x, y)(a(y) − mQ (a))f (y)dy.
Rn \2Q

Now that x ∈ Q and (a − mQ (a))χ2Q · f ∈ Lq (Rn ) for some 1 < q < p by

virtue of the John–Nirenberg inequality, it looks nice. The integral of the third
term does not contain the singularity.
Speaking more precisely, we have the following information:
 BMO spaces and Morrey–Campanato spaces 229

Lemma 227. Let T be a singular integral operator. Then

Z
n
|K(x, y)(a(y) − mQ (y))f (y)|dy . `(Q)− p kak∗ · kf kLp .
Rn \2Q

Proof By the size condition

Z
|K(x, y)(a(y) − mQ (y))f (y)|dy
Rn \2Q
∞ Z
X 1
. |a(y) − mQ (y)| · |f (y)|dy
j=1
(2j `(Q))n 2j+1 Q\2j Q

∞
! 10
p
kf kLp
Z
p0
X
. |a(y) − mQ (y)| dy
j=1
(2j `(Q))n 2j+1 Q\2j Q
∞
X j
. n kak∗ kf kLp
j=1
(2j `(Q)) p
n
. `(Q)− p kak∗ · kf kLp .
Here for the second inequality we used (5.1).
Therefore, this is an appropriate candidate of the definition of [a, T ]f .
To justify the observation above, we will prove the following lemma.
Lp (Rn ), T be a singular integral operator and
S
Lemma 228. Let f ∈
1≤p<∞
a ∈ BMO(Rn ). Given a cube Q, we set
SQ f (x) ≡ (a(x) − mQ (a))T0 f (x) − T [(a − mQ (a))χ2Q · f ](x)
Z
− K(x, y)(a(y) − mQ (a))f (y)dy.
Rn \2Q

Then if Q and R are cubes with Q ⊂ R, then SQ f (x) = SR f (x) for almost
every x ∈ Q(⊂ R).
Proof Note that
SQ f (x) − SR f (x)
Z
= T [(a − mR (a))χ2R\2Q · f ](x) − K(x, y)(a(y) − mR (a))f (y)dy.
2R\2Q

By the kernel condition this is equal to 0. The lemma is therefore proven.

Definition 64 (Commutator). Given a ∈ BMO(Rn ) and a singular integral
operator T with associated kernel K, define the commutator [a, T ] by
[a, T ]f (x) ≡ lim (a(x) − mQ(j) (a))T0 f (x) − T [(a − mQ(j) (a))χ2Q(j) · f ](x)
j→∞
Z
− K(x, y)(a(y) − mQ(j) (a))f (y)dy.
Rn \2Q(j)
230 Morrey Spaces

Note that the above definition makes sense because the limit exists for a.e.
x ∈ Rn . In view of Lemma 228, the following proposition is clear.
Proposition 229. Let f ∈ L∞ n
c (R ). If R is a cube and T is a singular integral
operator with the kernel K, then
[a, T ]f (x) = (a(x) − mR (a))T0 f (x) − T [(a − mR (a))χ2R · f ](x)
Z
− K(x, y)(a(y) − mR (a))f (y)dy
Rn \2R

for a.e. x ∈ R.
Proof The proof is left as an exercise for the interested readers. See Exer-
cise 79.
We now prove the boundedness of commutators, where we present an appli-
cation of the sharp maximal inequality.
Lemma 230. Let T be a singular integral operator, and let r > 0, a ∈
BMO(Rn ) and f ∈ L∞ n
c (R ). Then
 
M ] ([a, T ]f )(x) . kak∗ M (r) ◦ T f (x) + M (r) f (x) (x ∈ Rn ). (5.3)

Proof Now let us estimate M ] ([a, T ]f )(x). To do this, we use the decom-
position of [a, T ]f for a given cube Q above. Firstly, we write
F1 (x) ≡ (a(x) − mQ (a))T0 f (x),
F2 (x) ≡ T [(a − mQ (a))χ2Q · f ](x),
Z
F3 (x) ≡ K(x, y)(a(y) − mR (a))f (y)dy
Rn \2Q

for the sake of simplicity. Let 1 < r < p be an auxiliary parameter fixed
throughout.
We treat F1 by using mQ (|F1 − mQ (F1 )|) ≤ 2mQ (|F1 |). Let us recall that
we have been writing
 Z  1t
(t) 1 t 1
mQ (G) ≡ |G(x)| dx , M (t) F (x) ≡ M [|F |t ](x) t
|Q| Q
for t > 0 and F, G ∈ L0 (Rn ). Under this notation we have
(r 0 ) (r)
mQ (|F1 − mQ (F1 )|) . mQ (|a − mQ (a)|) · mQ (|T f |) . kak∗ M (r) [T f ](x).
The most right-hand side matches
√ the first term of (5.3).
As for F2 , we use the L r (Rn )-boundedness of T as well.
mQ (|F2 − mQ (F2 )|) ≤ 2mQ (|F2 |)
(r)
≤ 2mQ (|F2 |)
2
≤ kT [(a − mQ (a))χ2Q · f ]kL√r .
|Q|
 BMO spaces and Morrey–Campanato spaces 231

We now employ the John–Nirenberg inequality :

√
( r)
mQ (|F2 − mQ (F2 )|) . m2Q (|(a − mQ (a))f |) . kak∗ M (r) f (x).

The most right-hand side matches the second term of (5.3).

Finally for the estimate of F3 , we take full advantage of the oscillation
property of M ] . Note that if x, y ∈ Q then
Z
|F3 (x) − F3 (y)| . |K(x, z) − K(x, y)| · |a(z) − mR (a)| · |f (z)|dz.
Rn \2Q

Let ` ≤ 2`(Q). Then, by the Hörmander condition of the kernel K,

Z ∞
1 χB(c(Q),`) (z)d`
n+1
= (n + 1) ,
|z − c(Q)| 0 `n+2

and the fact that (Rn \ 2Q) ∩ B(x, 21 `) = ∅, we have

|a(z) − mR (a)| · |f (z)|

Z
|F3 (x) − F3 (y)| . `(Q) dz
n
R \2Q |z − c(Q)|n+1
Z ∞ Z !
1
. `(Q) |a(z) − mR (a)| · |f (z)|dz d`.
2`(Q) `n+2 B(c(Q),`)

If we invoke the John–Nirenberg inequality, then we obtain

|F3 (x) − F3 (y)|

Z ∞ Z ! r1
d`
. `(Q) ka − mR (a)kLr0 (B(c(Q),`)) |f (z)|r dz
2`(Q) B(c(Q),`) `n+2
Z ∞  
(r) ` d`
. `(Q) kak∗ · M f (x) log 2 +
2`(Q) `(Q) `2
(r)
. kak∗ M f (x),

which leads us to (5.3).

We will prove that commutators of the above type are Lp (Rn )-bounded
for all 1 < p < ∞.
Theorem 231. Let 1 < p < ∞ and a ∈ BMO(Rn ). Then [a, T ]
extends to an Lp (Rn )-bounded operator so that it satisfies the norm estimate
k [a, T ] kLp →Lp .T kak∗ .
Proof By the truncation, we may assume that a ∈ L∞ (Rn ). We will show
that k[a, T ]f kLp .T kak∗ kf kLp for f ∈ L∞ n
c (R ).
p n
Now that min( |[a, T ]f |, 1 ) ∈ L (R ), we are in the position of applying
the sharp maximal inequality (see Theorem 166) to obtain

k [a, T ]f kLp . kM ] ([a, T ]f )kLp . (5.4)

232 Morrey Spaces
√
Let r = p. Since M and T are bounded on Lq (Rn ) for all 1 < q < ∞, it
follows that
 
kM ] ([a, T ]f )kLp . kak∗ kM (r) ◦ T f kLp + kM (r) f kLp . kak∗ kf kLp . (5.5)

Combining (5.4) and (5.5) gives the desired estimate.

We consider the convergence problem of the integral defining commutators.
Start with truncated commutators. Let ε > 0, a ∈ BMO(Rn ) and T be a
singular integral operator. Then define
Z
[a, T ]ε f (x) ≡ (a(x) − a(y))K(x, y)f (y)dy
Rn \B(ε)

as long as the right-hand side makes sense. If f ∈ L∞ n

c (R ) and a ∈ BMO(R ),
n
n
then [a, T ]f (x) = lim[a, T ]ε f (x) exists for almost all x ∈ R , since f, a · f ∈
ε↓0
L2 (Rn ) and
Z Z
[a, T ]f (x) = a(x) lim K(x, y)f (y)dy−lim a(y)K(x, y)f (y)dy.
ε↓0 Rn \B(ε) ε↓0 Rn \B(ε)

For a given f ∈ Lp (Rn ), we seek to consider the almost everywhere con-

vergence as ε ↓ 0. We need to consider the maximal operator of truncated
commutators.
Lemma 232. Let a ∈ BMO(Rn ). Also let ϕ ∈ Cc∞ (Rn ) ∩ M+ (Rn ). Then the
operator
 
x−y
Z
1
W f (x) ≡ sup n |a(x) − a(y)|ϕ |f (y)| dy (x ∈ Rn ),
ε>0 ε Rn ε

is a bounded operator on Lp (Rn ) for any 1 < p < ∞.

√
Proof Let f ∈ Lp (Rn ). Set r ≡ p. Write
 
x−y
Z
1
Wε f (x) ≡ n |a(x) − a(y)|ϕ f (y)dy (x ∈ Rn ).
ε Rn ε

Consider the operator given by WE ≡ sup |Wε f | for each subset E ⊂ (0, ∞), so
ε∈E
that W f (x) = W(0,∞) [|f |](x). By a simple approximation argument, we have
W f (x) = W(0,∞)∩Q [|f |](x). Furthermore, since WEj [|f |](x) ↑ W f (x) for any
∞
sequence of increasing finite subsets {Ej }∞
S
j=1 such that Ej = (0, ∞)∩Q, we
j=1
have only to prove kWE f kLp . kf kLp for any finite subset E ⊂ (0, ∞) with the
 BMO spaces and Morrey–Campanato spaces 233

implicit constant does not depend on E. Furthermore, by another truncation

argument, we may assume that a ∈ L∞ (Rn ). We will prove kW f kLp . kf kLp
for all f ∈ L∞ n
c (R ). We will resort to the sharp maximal operator estimate.
We define
ME] F~ (x) = sup χQ (x)mQ (kF~ − mQ (F~ )k`∞ (E) )
Q∈Q

for any vector-valued function F = {Fε }ε∈E ∈ L1loc (Rn )E .

We will choose Fε = Wε f for each ε ∈ E. In fact, we fix x ∈ Rn and a
√
cube Q and then, letting r = p, we will prove
 
mQ sup |Wε f − Wε [(a − mQ (a))χRn \2Q f ](c(Q))|
ε∈E

. kak∗ M (r) f (x) + kak∗ M (r) T∗ f (x).

Let z ∈ Q and ε ∈ E be arbitrary. By the triangle inequality,

   
z−y c(Q) − y
|a(z) − a(y)|ϕ − |mQ (a) − a(y)|ϕ χRn \2Q (y)
ε ε
   
z−y c(Q) − y
≤ |a(z) − a(y)|ϕ − |mQ (a) − a(y)|ϕ χRn \2Q (y)
ε ε
 
z−y
+|a(z) − a(y)|ϕ χ2Q (y).
ε

By the triangle inequality once again,

   
z−y c(Q) − y
|a(z) − a(y)|ϕ − |mQ (a) − a(y)|ϕ χRn \2Q (y)
ε ε
   
z−y c(Q) − y
≤ |mQ (a) − a(y)|ϕ − |mQ (a) − a(y)|ϕ χRn \2Q (y)
ε ε
   
z−y z−y
+ |a(z) − a(y)|ϕ − |mQ (a) − a(y)|ϕ χRn \2Q (y)
ε ε
 
1 z−y
+|a(z) − a(y)| · n ϕ χ2Q (y).
ε ε

By the triangle inequality and by the mean value theorem,

   
1 z−y c(Q) − y
|a(z) − a(y)|ϕ − |mQ (a) − a(y)|ϕ χRn \2Q (y)
εn ε ε
 
|a(y) − mQ (a)| 1 z−y
. `(Q) + |a(z) − mQ (a)| · n ϕ
(|y − c(Q)| + `(Q))n+1 ε ε
 
1 z−y
+|mQ (a) − a(y)| · n ϕ χ2Q (y).
ε ε
234 Morrey Spaces

Note that M is bounded on Lr (Rn ). We observe that

Z   
1 ·−y
mQ |mQ (a) − a(y)| · n ϕ χ2Q |f (y)|dy
Rn ε ε
≤ mQ (M [|mQ (a) − a| · χ2Q f ])
√
( r)
≤ mQ (M [|mQ (a) − a| · χ2Q f ])
√
( r)
. m2Q (|mQ (a) − a| · f ) .
Consequently, by the John–Nirenberg inequality,
|Wε f (z) − Wε [(a − mQ (a))χRn \2Q f ](c(Q))|
≤ kak∗ M (r) f (x) + |a(z) − mQ (a)|M f (z).

Thus, ME] [{Wε f }ε∈E ](x) . kak∗ M (r) f (x) + kak∗ M (r) ◦ M f (x) for all x ∈ Rn .
If we use the vector-valued sharp maximal inequality, then we have the desired
result.
Although the indicator function of balls is not smooth, we can take advan-
tage of the positivity of the integral kernel.
Corollary 233. Let a ∈ BMO(Rn ), and let 1 < p < ∞. Define
Z
χQ (x)
Ma f (x) ≡ sup |a(x) − a(y)| · |f (y)|dy (x ∈ Rn )
Q∈Q |Q| Q

for f ∈ Lp (Rn ). Then Ma is bounded on Lp (Rn ).

Proof Let ϕ be a smooth bump function at zero, that is ϕ ∈ Cc∞ (Rn )
satisfies χB(1) ≤ ϕ ≤ χB(2) . Then
 
x−y
Z
1
Ma f (x) . sup n |a(x) − a(y)|ϕ |f (y)|dy.
ε>0 ε Rn ε
Thus, we are in the position of using Lemma 232.
Let ϕ ∈ C ∞ (Rn ). We will consider a smooth variant of the truncation in
view of Corollary 233.
We write
 
x−y
Kϕ,ε (x, y) ≡ ϕ K(x, y) (x, y ∈ Rn ).
ε
Lemma 234. Let a ∈ BMO(Rn ). Let ϕ ∈ C ∞ (Rn ) satisfy 0 ∈ / supp(ϕ),
kϕkL∞ + k∇ϕkL∞ . 1 and χRn \B(1) ≤ ϕ ≤ 1. Also let T be a singular integral
operator with the kernel K. Then the operator
Z
[a, T ]ϕ
∗ f (x) ≡ sup (a(x) − a(y))Kϕ,ε (x, y)f (y)dy (x ∈ Rn ),
ε>0 Rn

is a bounded linear operator on Lp (Rn ) for any 1 < p < ∞.

 BMO spaces and Morrey–Campanato spaces 235

Proof It suffices to prove k[a, T ]ϕ ∞ n

∗ f kLp . kf kLp for all f ∈ Lc (R ). It
∞ n
can be assumed that a ∈ L (R ) by the truncation. As before, we replace
√
sup by sup for a finite set E ⊂ (0, ∞). Let r = p as before.
ε>0 ε∈E
Let f ∈ Lp (Rn ). We write
Z
[a, T ]ϕ,ε f (x) ≡ (a(x) − a(y))Kϕ,ε (x, y)f (y)dy
Rn

and
Z
ϕ,ε
T f (x) ≡ Kϕ,ε (x, y)f (y)dy, T ϕ,∗ f (x) = sup |T ϕ,ε f (x)|.
Rn ε>0

Consider the sharp maximal operator estimate. In fact, we will prove

mQ (|[a, T ]ϕ,ε f − T ϕ,ε [(a − mQ (a))χRn \2Q f ](c(Q))|)

. kak∗ M (r) f (x) + kak∗ M (r) ◦ T ϕ,∗ f (x)

for any cube Q containing x ∈ Rn .

Let z ∈ Rn . Then
Z
[a, T ]ϕ,ε f (z) − T ϕ,ε [(a − mQ (a))χRn \2Q f ](c(Q)) = Fε (z, y)dy,
Rn

where

Fε (z, y) ≡ −(a(y) − mQ (a)) (Kϕ,ε (z, y) − Kϕ,ε (c(Q), y)) χRn \2Q (y)f (y)
+ (a(z) − mQ (a))Kϕ,ε (z, y)f (y)
+ (mQ (a) − a(y))Kϕ,ε (z, y)χ2Q (y)f (y).

For the first term, we will use the John–Nirenberg inequality and the fact that
ϕ is smooth to give
Z
(a(y) − mQ (a)) (Kϕ,ε (z, y) − Kϕ,ε (c(Q), y)) χRn \2Q (y)f (y)dy
Rn
. kak∗ M (r) f (z).

For the second term, we will use the John–Nirenberg inequality again to give
Z Z
1
(a(z) − mQ (a))Kϕ,ε (z, y)f (y)dy dz . kak∗ M (r) [T ϕ,∗ f ](z)
|Q| Q Rn

For the third term, we will use the maximal singular integral operator to
give
Z
(mQ (a) − a(y))Kϕ,ε (z, y)χ2Q (y)f (y)dy . T ϕ,∗ [(a − mQ (a))χ2Q f ](z)
Rn
236 Morrey Spaces

If we integrate this estimate over z ∈ Q, we obtain

Z
(mQ (a) − a(y))Kϕ,ε (z, y)χ2Q (y)f (y)dy . kak∗ M (r) ◦ T ϕ,∗ f (x).
Rn

Thus,
M ] [[a, T ]ϕ
∗ f ](x) . kak∗ M
(r)
f (x) + kak∗ M ◦ T ϕ,∗ f (x).
Since a ∈ L∞ (Rn ) and f ∈ L∞ n
c (R ), we are in the position of using the
sharp maximal inequality to have the desired result.
We are oriented to the almost everywhere convergence of truncated com-
mutators. To this end, we will prove that the maximal operators of truncated
commutators are bounded.
Theorem 235. Let a ∈ BMO(Rn ). Also let T be a singular integral operator
with the kernel K. Then the operator
Z
[a, T ]∗ f (x) ≡ sup (a(x) − a(y))K(x, y)f (y)dy (x ∈ Rn )
ε>0 Rn \B(x,ε)

is a bounded linear operator on Lp (Rn ) for any 1 < p < ∞.

Proof Let ϕ ∈ C ∞ (Rn ) satisfy 0 ∈
/ supp(ϕ), kϕkL∞ + k∇ϕkL∞ . 1 and
χRn \B(1) ≤ ϕ ≤ 1 as in Lemma 234. We simply use
[a, T ]∗ f (x) . [a, T ]ϕ
∗ f (x) + Ma f (x) (x ∈ Rn )
and Lemma 234.
We apply Theorem 235 to some special cases. Although we do not take
up elliptic differential equations in this book, Theorem 236 can be used for
elliptic differential equations.
Theorem 236. Let a ∈ BMO(Rn ). Assume that K ∈ L∞ (Rn ×S n−1 ) satisfies
the following:
(1) K(x, ·) ∈ C ∞ (Rn \ {0}) for all x ∈ Rn ;
(2) K(x, tz) = t−n K(x, z) for every t > 0 and (x, z) ∈ Rn × S n−1 ;
Z
(3) K(x, z)dσ(z) = 0 for all x ∈ Rn ;
S n−1

∂αK
(4) max < ∞.
|α|≤2n ∂z α L∞ (Rn ×S n−1 )

Let f ∈ Lp (Rn ) with 1 < p < ∞. For any ε > 0, we set

Z
[a, T ]ε f (x) ≡ K(x, x − y)(a(x) − a(y))f (y)dy.
Rn \B(x,ε)

Then there exist [a, T ]f ∈ Lp (Rn ) such that lim k[a, T ]ε f − [a, T ]f kLp = 0.
ε↓0
 BMO spaces and Morrey–Campanato spaces 237

Proof The integral kernel K satisfies the cancellation condition. Hence,

we can use the expansion in Theorem 215.

5.2.2 Commutators generated by BMO and fractional

integral operators
In addition to the above commutator and the kernel K of T satisfying
the size condition, we consider the commutator generated by BMO and the
fractional integral operator Iα given by
(a(x) − a(y))
Z
[a, Iα ]f (x) ≡ lim f (y)dy (x ∈ Rn ). (5.6)
ε↓0 Rn \B(x,ε) |x − y|n−α

Needless to say what [a, T ] is to [a, Iα ] is what T is to Iα . Consequently, [a, Iα ]

can be located as the mixture of [a, T ] and Iα . As we will see the commutator
[a, Iα ] behaves like Iα .
Lemma 237. For all 0 < α < n, 1 < r < ∞ and f ∈ L∞ n
c (R ), we have

M ] ([a, Iα ]f )(x) . kak∗ (Mα(r) f (x) + M (r) Iα f (x)) (x ∈ Rn ).

Proof Let Q be a fixed cube, and let x ∈ Q. We will show that
mQ (|[a, Iα ]f −mQ (Iα [(a−mQ (a))χRn \3Q f ])|) . kak∗ (Mα(r) f (x)+M (r) Iα f (x))
for x ∈ Rn . To this end, by replacing a with a − mQ (a), we may assume
mQ (a) = 0. We decompose
[a, Iα ]f = a · Iα f − Iα [a · χ3Q f ] − Iα [a · χRn \3Q f ].
By the John–Nirenberg inequality we have
(r 0 ) (r)
mQ (|a · Iα f |) ≤ mQ (a)mQ (Iα f ) . kak∗ M (r) ◦ Iα f (x).
We also have
(r)
mQ (Iα [a · χ3Q f ]) . `(Q)α m3Q (|a · f |) . `(Q)α m3Q (f )kak∗ . kak∗ Mα(r) f (x)
again by the John–Nirenberg inequality. Finally, by the mean–value theorem
and the John–Nirenberg inequaity, for all x, y ∈ Q, we have
|Iα [(a − mQ (a))χRn \3Q f ](x) − Iα [(a − mQ (a))χRn \3Q f ](y)| . kak∗ Mα(r) f (x).
Thus, the proof is complete.
With this pointwise estimate, we can easily obtain the boundedness of
[a, Iα ].
1 1 α
Theorem 238. Let 1 < p < q < ∞ and 0 < α < n satisfy = − . Then
q p n
[a, Iα ], defined initially on L∞ n
c (R ), extends to a bounded linear operator from
Lp (Rn ) to Lq (Rn ).
Proof The proof is akin to that of Theorem 231. See Exercise 80.
238 Morrey Spaces

5.2.3 Exercises
Exercise 79. Prove Proposition 229 using (4.1).
Exercise 80. Let 1 < r < p. Using Lemma 237, the sharp maximal inequality
(r)
and the boundedess of Mα and Iα , prove Theorem 238.

5.3 Morrey–Campanato spaces

In 1938, C. Morrey investigated the elliptic differential operators. His tech-
nique became the theory of normed spaces. However, around the 1960’s, S.
Campanato proposed a different approach. Quite often, the Morrey norm has
an equivalent expression to this Campanato technique. He proposed to con-
sider the difference when we consider Campanato spaces. Later his technique
grew up to be another theory of normed spaces. Named after Campanato, the
normed spaces are named (Morrey–)Campanato spaces. We call them Cam-
panato spaces for short here and below. We will define (Morrey–)Campanato
spaces in Section 5.3.1. In Section 5.3.1, we will also establish that Morrey
spaces and Campanato spaces are equivalent. Campanato spaces are connected
with Hölder–Zygmund spaces, which we establish in Section 5.3.2.

5.3.1 Morrey–Campanato spaces

In the definition of the Morrey norm we did not tolerate the case where
p < 0. To consider such cases, we will need to consider function spaces modulo
Pk (Rn ). We formulate our idea.

Definition 65 (Lλ,q n
k (R )). Let λ ∈ R and 1 ≤ q ≤ ∞, and fix k ∈ N0 ∪ {−1}.
Then define the Morrey–Campanato space Lλ,q n
k (R ) as the set of all f ∈
q n
Lloc (R ) for which
 
1
kf kLλ,q ≡ sup inf n ρλ |B(ρ)|− q kf − P kLq (B(x,ρ)) < ∞.
k
(x,ρ)∈Rn+1 P ∈Pk (R )
+

Before we investigate Lλ,q n

k (R ), a helpful remark may be in order.

Remark 11. Let λ ∈ R, 1 ≤ q ≤ ∞, and k ∈ N0 ∪ {−1}. The function space

Lλ,q n n
k (R ) is a function space modulo Pk (R ). Hence, it is natural to write
λ,q
[f ] to denote the element in Lk (Rn ). However, to simplify the notation, one
often writes f instead of [f ].
Here we consider some concrete cases.
 BMO spaces and Morrey–Campanato spaces 239

Example 98. Let λ < 0 and 1 ≤ q < ∞. Then Lq,λ n

−1 (R ) = {0} by the
Lebesgue differentiation theorem.
Example 99. Let 1 ≤ q < ∞. Then L0,q n n
0 (R ) is nothing but BMO(R )
thanks to the John–Nirenberg inequality.
Example 100. Let λ ∈ R and 1 ≤ q < ∞. Then the Morrey–Campanato
space Lλ,q n q n
0 (R ) is the set of all f ∈ Lloc (R ) for which the quantity
 
λ − q1
kf kLλ,q ≡ sup inf ρ |B(ρ)| kf − ckLq (B(x,ρ)) < ∞.
0 c∈C
(x,ρ)∈Rn+1
+

In general thanks to Lemma 115, the infimum is attained: We have

 
λ − q1
kf kLλ,q ≡ sup min n ρ |B(ρ)| kf − P kLq (B(x,ρ))
k
(x,ρ)∈Rn+1 P ∈Pk (R )
+

in the above. We will use the following notation for this minimum.
Definition 66. Let λ ∈ R, 1 ≤ q < ∞ and k ∈ N0 ∪ {−1}. Also let f ∈
Lqloc (Rn ), x0 ∈ Rn and ρ > 0.
(1) Define Pk (·; x0 , ρ, f ) ≡ argminP ∈Pk (Rn ) kf − P kLq (B(x0 ;ρ)) , the best
approximation of u up to order k in the ball B(x0 , ρ). Regard this poly-
nomial as a function of the first variable. Other variables play the role
of the parameters.
(2) Define aα (x0 ; ρ, f ) ≡ ∂ α Pk (x0 ; x0 , ρ, f ) for any α ∈ N0 with |α| ≤ k.
Thanks to Remark
n 3, we have the following norm equivalence
o Lλ,q n
k (R ):
1
kf kLλ,q ∼ sup ρλ |B(ρ)|− q kf − PB(x,ρ)
k
f kLq (B(x,ρ)) .
k
(x,ρ)∈Rn+1
+

We investigate the property of aα (·; ρ, f ) keeping in mind that Lλ,q n

k (R ) is
monotone in q.
Lemma 239. Let λ ∈ R and k ∈ N0 ∪ {−1}. Also let f ∈ Lλ,1 n
k (R ). Then
−|α|−λ
|aα (x0 ; 2ρ, f ) − aα (x0 ; ρ, f )| . ρ kf kLλ,1 for all x0 ∈ R and ρ > 0.
k

Proof Thanks to Lemma 115, we have

|aα (x0 ; 2ρ, f ) − aα (x0 ; ρ, f )|
= |∂ α Pk (x0 ; x0 , 2ρ, f ) − ∂ α Pk (x0 ; x0 , ρ, f )|
. ρ−n−|α| kPk (·; x0 , 2ρ, f ) − Pk (·; x0 , ρ, f )kL1 (B(x0 ,ρ)) .
Meanwhile, by the definition of f , we have
kPk (·; x0 , 2ρ, f ) − Pk (·; x0 , ρ, f )kL1 (B(x0 ,ρ))
≤ kPk (·; x0 , 2ρ, f ) − ukL1 (B(x0 ,ρ)) + kf − Pk (·; x0 , ρ, f )kL1 (B(x0 ,ρ))
. ρn−λ kf kLλ,1 .
k
240 Morrey Spaces

Combining these two estimates gives the desired result.

In particular, if 1 ≤ q < ∞, then |aα (x0 ; 2ρ, f ) − aα (x0 ; ρ, f )| .
ρ−|α|−λ kf kLλ,1 for all f ∈ Lλ,q n
k (R ).
k
We have the following analogy to Lemma 239:
Lemma 240. Let λ ∈ R, 1 ≤ q < ∞ and k ∈ N0 ∪ {−1}. Also let f ∈
Lλ,q n
k (R ). Then |aα (x0 ; ρ, f )−aα (y0 ; ρ, f )| . ρ
−|α|−λ
kf kLλ,q for all x0 , y0 ∈ R
k
and ρ > 0 such that |x0 − y0 | < ρ.
Proof Using B(x0 , ρ) ∪ B(y0 , ρ) ⊂ B(x0 , 2ρ), we argue as in Lemma 239.
Importantly, the number k does not affect the definition of Lλ,q n
k (R ) as
long as k is large enough.
Theorem 241. Let λ ∈ R, and let 1 ≤ q ≤ ∞. Then for any l ∈ Z ∩
(max(−1, [−λ]), ∞), Lλ,q n λ,q n
max(−1,[−λ]) (R ) and Ll (R ) are isomorphic.

Proof It follows from the definition of the norm that the natural mapping
[f ] ∈ Lλ,q n λ,q n
max(−1,[−λ]) (R ) 7→ [f ] ∈ Ll (R ) is continuous.
To see that we can construct the inverse mapping, it suffices to prove that
l−1
there exists P = PB(x 0 ,ρ)
f ∈ Pl (Rn ) such that
1
ρλ |B(ρ)|− q kf − PB(x
l−1
0 ,ρ)
kLq (B(x,ρ)) . kf kLλ,q .
l

for any [f ] ∈ Lλ,q n

l (R ) and for any ball B = B(x0 , ρ).
We start with constructing the constant aα to this end. We choose the
best approximation Pl (x; x0 , ρ, f ) of f up to order l in the ball B(x0 , ρ).
As long as B(x0 , ρ) ⊃ B(y0 , ρ0 ) and ρ0 ≤ ρ ≤ 2ρ0 , |aα (x0 ; ρ, f ) −
aα (y0 ; ρ0 , f )| . ρ−|α|−λ kf kLλ,q due to Lemmas 239 and 240. As a result, aα ≡
l
lim aα (x0 ; ρ, f ) exists and is independent of x0 . Furthermore |aα (x0 ; ρ, f ) −
ρ→∞
aα | . ρ−|α|−λ kf kLλ,q .
l
n l
Let B ≡ B(x 0 , ρ) be a ball once again. For x ∈ R , write QB (x) ≡
(aα (x0 ; ρ, f ) − aα )(x − x0 )α and PBl (x) ≡ QlB (x) − P (x),
P
Pl (x; x0 , ρ, f ) −
|α|=l
so that f (x) − QlB (x) = f (x) − PBl (x) + (aα (x0 ; ρ, f ) − aα )(x − x0 )α . Thus,
P
|α|=l
(q) (q)
mB (f − QlB ) mB (f − PBl ) + |aα (B) − aα |ρl . ρ−λ kf kLλ,q . This means
P
.
l
|α|=l
that we can take P = QlB in the above.
Theorem 241 with λ > 0 shows that Morrey–Campanato spaces and Mor-
rey spaces coincide in the following sense:
Theorem 242. Let 1 ≤ q ≤ p < ∞.
n
,q
(1) In the sense of continuous embedding, Mpq (Rn ) ,→ Lkp (Rn ) for all
k ∈ N0 ∪ {−1}.
 BMO spaces and Morrey–Campanato spaces 241
n
,q
(2) Let f ∈ L0p (Rn ).
(i) Let x0 ∈ Rn . Then the quantity u∞ ≡ lim a0 (x0 ; R, u) ∈ C exists,
R→∞
where a0 (x0 ; R, u) is defined in Definition 66 with α = 0 and ρ = R.
Furthemore u∞ is independent of x0 .
(ii) The function u − u∞ is independent of the representative of u.
That is, if v differs from u by an additive constant and v∞ ≡
lim a0 (x0 , ρ; v), then v − v∞ = u − u∞ .
R→∞
(iii) We have ku − u∞ kMpq . kf k n ,q .
L0p

Proof
(1) This is trivial from the definition of the norms.
(2) This is nothing but Theorem 241 with λ > 0.

5.3.2 Morrey–Campanato spaces and Hölder–Zygmund

spaces
The homogeneous Hölder–Zygmund space C˙γ (Rn ) (see in Section 1.6.2)
and the Morrey–Campanato space Lλ,q n
k (R ) defined in Section 5.3.1 are closely
connected. We investigate the relation between Morrey–Campanato spaces
and Hölder–Zygmund spaces.
Lemma 243. Let ψ ∈ C ∞ (Rn ) be a function in Example 54. Write ψρ ≡
ρ−n ψ(ρ−1 ·) for ρ > 0. Let λ ∈ R, 1 ≤ q ≤ ∞ and k ∈ N0 ∪ {−1}, and let
f ∈ Lλ,q n
k (R ).

(1) Let ρ, ρ0 > 0 satisfy ρ0 ≤ ρ ≤ 2ρ0 , and let λ ∈ R. Then we have k∂ α (f ∗

ψρ ) − ∂ α (f ∗ ψρ0 )kL∞ . ρ−λ−|α| kf kLλ,q .
k

(2) Let ρ > 0 and λ < 0. Then whenever α ∈ N0 satisfies |α| < −λ,
f ∈ C [−λ] (Rn ) and k∂ α (f ∗ ψρ ) − ∂ α f kL∞ . ρ−λ−|α| kf kLλ,q .
k

Proof
(1) Since
k∂ α (f ∗ ψρ ) − ∂ α (f ∗ ψρ0 )kL∞
= k∂ α [(f − Pk (·; x0 , ρ, f ))] ∗ ψρ − ∂ α [(f − Pk (·; x0 , ρ, f ))] ∗ ψ2−n kL∞
≤ k(f − Pk (·; x0 , ρ, f )) ∗ ∂ α ψρ kL∞ + k(f − Pk (·; x0 , ρ, f )) ∗ ∂ α ψρ0 kL∞ ,

we have k∂ α (f ∗ ψρ ) − ∂ α (f ∗ ψρ0 )kL∞ . ρ−λ−|α| kf kLλ,q .

k

∞
X
(2) Since f − f ∗ ψρ = (f ∗ ψ2−L ρ − f ∗ ψ2−L+1 ρ ) converges in C [−λ] (Rn ),
L=1
f ∈ C [−λ] (Rn ) and f − f ∗ ψρ satisfy the desired estimate from (1).
242 Morrey Spaces

We will show that any function f ∈ Lλ,q n

k (R ) is represented by a continuous
function and that the value of q is not important in the definition of Lλ,q n
k (R )
if λ < 0.
Theorem 244. Let 1 ≤ q ≤ ∞, λ < 0 and k ∈ Z ∩ [−1, −λ). Then any
f ∈ Lλ,q n n
k (R ) is almost everywhere equal to a continuous function g ∈ C(R )
such that kgkLλ,∞ . kf kLλ,q .
k k

Proof Let B ≡ B(x0 , r) be a ball. Also let PBk f be the Gram–Schmidt

polynomial of f of order k over the ball B. Take ψ to be a function defined
in Example 54. Thanks to Lemma 243, f is almost everywhere equal to a
continuous function g ≡ lim f ∗ ψ2−l r ∈ C(Rn ). Meanwhile since we have
l→∞
kf ∗ ψr − PBk f kL∞ = k(f − PBk f ) ∗ ψr kL∞ . r−λ kf kLλ,q and g − PBk f =
k
∞
k
P
f ∗ ψr − PB f + f ∗ ψ2−l r − f ∗ ψ2−l+1 r . we obtain kgkLλ,∞ . kf kLλ,q .
k k
l=1

We investigate the differential feature of Morrey–Campanato spaces keep-

ing in mind that q does not affect the definition of the function spaces.
Theorem 245. Let λ < 0, and let K ∈ Z ∩ [0, −λ). Then f ∈ Lλ,∞ n
[−λ] (R )
if and only if f ∈ CK (Rn ) and ∂ α f ∈ Lλ+K,∞ n
[−λ−K] (R ) for all α ∈ N0
n
with
|α| ≤ K.
Proof We may suppose K = 1 and hence λ < −1; since Theorem 244
shows that f ∈ C(Rn ), the case of K = 0 is covered by Theorem 244. We
may induct on K for the case K ≥ 2. Let f ∈ Lλ,q n
[−λ] (R ). From Lemma
1 n
243, f ∈ C (R ). Since ∂j f − ∂j Pk (·; x0 , ρ, f ) = ∂j f − ∂j (f ∗ ψρ ) + (f −
Pk (·; x0 , ρ, f )) ∗ ∂j ψρ , we see that k∂j f − ∂j Pk (·; x0 , ρ, f )kL∞ . ρ−λ−1 kf kLλ,q .
k

Thus ∂ α f ∈ Lλ+1,∞ n n λ,∞

[−λ−1] (R ) for all α ∈ N0 with |α| ≤ 1 if f ∈ L[−λ] (R ).
n

Suppose that f ∈ C 1 (Rn ) and that each component of the gradient of f

belongs to Lλ+1,∞ n
[−λ−1] (R ). Let B ≡ B(x0 , ρ) be a ball. Then by the fundamental
Xn Z 1
theorem of calculus f (x) − f (x0 ) = (xj − xj,0 ) ∂j f (x0 + t(x − x0 ))dt.
j=1 0
n
X Z 1
Setting P (x) ≡ (xj − xj,0 ) P[−λ−1] (∂j f (x0 + t(· − x0 )); x0 , ρ)(x)dt, we
j=1 0
have

kf − f (x0 ) − P kL∞ (B)

Xn Z 1
≤r k∂j f (x0 + t(· − x0 )) − P[−λ] (∂j f (x0 + t(· − x0 )); x0 , ρ)kL∞ (B) dt
j=1 0
n
X
≤ nr−λ k∂j f kLλ+1,∞ .
[−λ−1]
j=1
 BMO spaces and Morrey–Campanato spaces 243

Thus, f ∈ Lλ,q n
[−λ] (R ).

As we mentioned, Morrey–Campanato spaces and Hölder–Zygmund spaces

are closely connected. More precisely, we have the following norm equivalence.
Theorem 246. Let λ > 0 and 1 ≤ q ≤ ∞. Then C˙λ (Rn ) ∼ L−λ,q (Rn ). [λ]

Proof We may assume 0 < λ < 2 in view of Theorem 245 and the struc-
ture of the class {C˙λ (Rn )}λ>0 .
Let f ∈ Lλ,q n
1 (R ). We know that f is continuous thanks to Lemmas 239
and 240. Let us obtain the norm estimate. Let x, y ∈ Rn be fixed. Suppose
x 6= y. Choose r so that r < |x − y|. Then we have
1 1 1
PB(x+y,r) f (x + y) − 2PB(x,r) f (x) + PB(x−y,r) f (x − y)
1 1 1 1
= PB(x+y,r) f (x + y) − PB(x,|x−y|) f (x + y) − 2(PB(x,r) f (x) − PB(x,|x−y|) f (x))
1 1
+ PB(x−y,r) f (x − y) − PB(x,|x−y|) f (x − y).
Since λ < 0, thanks to Lemmas 239 and 240, we have
1
|PB(x+y,r) 1
f (x + y) − PB(x,|x−y|) f (x + y)| . |x − y|−λ kf kLλ,q ,
k
1
|PB(x,r) 1
f (x) − PB(x,|x−y|) f (x)| . |x − y|−λ kf kLλ,q ,
k
1
|PB(x−y,r) 1
f (x − y) − PB(x,|x−y|) f (x − y)| . |x − y|−λ kf kLλ,q .
k

So, 1
|PB(x+y,r) 1
f (x + y) − 2PB(x,r) 1
f (x) + PB(x−y,r) f (x − y)| . |x − y|−λ kf kLλ,q .
k
Letting r ↓ 0, we obtain |f (x + y) − 2f (x) + f (x − y)| . |x − y|−λ kf kLλ,q
k

thanks to Lemma 143. Thus, f ∈ C˙−λ (Rn ).

Conversely, let f ∈ C˙−λ (Rn ). Then choose ψ j and ϕj for j ∈ Z as in (1.4)
and (1.5), respectively. Fix a cube Q = Q(z, r) for the purpose of estimat-
1
ing inf n ρλ |B(r)|− q kf − P kLq (B(z,r)) . If we replace r by a slightly larger
P ∈Pk (R )
number, we may assume that log2 r ∈ Z. Set r ≡ 2−j . Write f j ≡ ψ j ∗ f .
Then thanks to Lemma 76, we have kf j+1 − f j kL∞ . 2jλ kf kĊ −λ . Since f is
∞
continuous, there is a decomposition f − f j = (f k+1 − f k ) in L∞ (Rn ). So
P
k=j
kf − f j kL∞ . 2jλ kf kĊ −λ . Meanwhile, if P denotes the Taylor polynomial of
f j up to order 1 around z, then we have
fj − P L∞ (B(z,2−j ))
. sup 2−2j k∂ α f j kL∞ (B(z,2−j )) . 2λj kf kĊ −λ
|α|=2

thanks to Lemma 115. We see that f ∈ Lλ,q n

1 (R ) putting these observations
together.

5.3.3 Exercises
Exercise 81. Let f ∈ L1loc (Rn ). Show that |mQ (f ) − mk Q (f )| . log(k +
2)kf k∗ for any k ≥ 1 and any cube Q.
244 Morrey Spaces

Exercise 82. Let 0 < α < n, 1 ≤ p < n

α. If f ∈ Mp1 (Rn ), then show that
(4.2) is satisfied.

Exercise 83. [279, Theorem 1] Let λ ≤ 0 and k ∈ N0 , and let f ∈ Lλ,1 n

k (R ) \
k
{0}. For all cubes Q, write PQ f for the Gram–Schmidt polynomial of order k
for Q. Show that
!
λ k λ
|{x ∈ Q : |Q| |f (x) − PQ f (x)| > λ}| . exp − |Q|
kf kLλ,1
k

using Proposition 117.

5.4 Notes
Section 5.1
General remarks and textbooks in Section 5.1
See the textbooks and lecture notes [29, Chapter 5], [104, Chapter 6],
[146, Chapter II, §3], [147], [157, Chapter 3], [223, Chapter 3], [231], [293],
[416, Chapter IV], [382, §3.3] and [432] for the theory of BMO.

Section 5.1.1
Fefferman and Stein defined the space BMO in [125]. Theorem 223 is essen-
tially due to Fefferman and Stein; [125, (1.2)]. See [404] for the definition of
BMO+ (Rn ).
See [363] for the closure of Cc∞ (Rn ) in BMO(Rn ).

Section 5.1.2
See the original paper [221] by John and Nirenberg, where we can find a
variant for functions defined on cubes.

Section 5.2
General remarks and textbooks in Section 5.2
See the textbooks [157, §3.5] and [293].

Section 5.2.1
The boundedness of commutators [b, T ] appearing in this section is due to
Coifman, Meyer and Weiss [84]. Chanillo considered the boundedness property
 BMO spaces and Morrey–Campanato spaces 245

of [b, Iα ] in Chanillo [66]. See [84, Theorem 1] for Theorem 231. See [401,
Theorem 2] for Theorem 235 including Lemmas 232 and 234.
Chianrenza, Frasca and Longo pointed out that the commutators in The-
orem 236 arise naturally; see [76, Theorem 2.5] for Theorem 236. We refer to
[464] for the application of Theorem 236. See [161] for its generalization.
We can generalize the above commutators, given by (5.6), as follows: Let
~b = (b1 , b2 , . . . , bm ). Define the multicommutator T~
b
Z m
Y
T~b f (x) ≡ K(x, y)f (y) (bj (x) − bj (y))dy,
Rn j=1

where K is the Calderón–Zygmund kernel satisfying (4.1), (4.2) and (4.3).

Here, f is an appropriate function that needs to be specified after a calculation.
This operator dates back to the paper by Coifman, Rochberg and Weiss
in 1976 [84].

Section 5.2.2
See [36] for example.

Section 5.3
General remarks and textbooks in Section 5.3
The earliest overview of Morrey–Campanato spaces seems to be [353]. The
next one seems [423]. We refer to [258, Chapter 5] for Morrey–Campanato
spaces. We also refer to the survey [359, §4]. See [108] for the characterization
of Morrey spaces and Morrey–Campanato spaces in terms of the heat kernel.

Section 5.3.1
A series of results in this section are basically based on the works by
Campanato [57, 58, 59, 60]. As we saw in Theorem 241, Janson, Taibleson
and Weiss pointed out that the order of the polynomials does not affect so
much the definition of Morrey–Campanato spaces [220, Theorem 1]. Morrey–
Campanato spaces are known to be equivalent to Morrey spaces in many cases.
See [427, 460, 461] for weighted characterizations. Several characterizations of
weighted Morrey–Campanato spaces are obtained by Tang in [427].
The Morrey–Campanato space Lλ,q n θ
k (R ) and the Lipschitz space Lip (R )
n

are equivalent. This equivalence goes back to the works by Campanato and
Meyer; see [58, Theorema, p. 183] and [312, Theorem, p. 718], where both
authors showed that these norms are equivalent to the Lipθ (Rn ) norm. See
[353, p. 72] for an account of these facts. See also [329, Theorem 3.1].
Campanato and Meyers independently showed that Morrey–Campanato
spaces and Morrey spaces coincide in many cases as we saw in Theorem 242
see [57], [59, Theorem 6.II] and [313]. Duong, Xiao and Yan expanded this
idea by the use of the heat kernel in [109].
246 Morrey Spaces

Section 5.3.2
Jonsson, Sjögen and Wallin proved Theorem 244 [222]. Janson, Taible-
son and Weiss pointed out that Morrey–Campanato spaces are isomorphic to
Hölder–Zygmund spaces; see [220, Lemma (3.4)], [220, Theorem 2] and [220,
Theorem 3] for Lemma 243, Theorems 244 and 245, respectively. Grevholm
proved Theorem 246 in [159, Theorem 3.1]. Greenwald also gave a different
proof of Theorem 246 [160]. See the booklet by DeVore and Sharpley [100] and
the lecture note [220] by Janson, Taibleson and Weiss for Theorem 246. We
now aim to consider function spaces related to Morrey–Campanato spaces.
We content ourselves with definitions. We consider CMO and CBMO spaces.
Let d ∈ N0 ∪ {−1}, 1 ≤ p ≤ ∞ and − np ≤ λ < d, and define

Z ! p1
1 1
kf kCMOp,λ,d ≡ sup λ |f (x) − PQd f (x)|p dx
r≥1 r rn Q(r)

Z ! p1
1 1
kf kCBMOp,λ,d ≡ sup λ |f (x) − PQd f (x)|p dx .
r>0 r rn Q(r)

Example 101. Chen and Lau [70] and Garcı́a-Cuerva [141] introduced the
central mean oscillation space CMOp (Rn ) with the norm

Z ! p1
1
kf kCMOp ≡ sup |f (x) − fQ(r) |p dx ,
r≥1 |Q(r)| Q(r)

So, CMOp (Rn ) = CMOp,0,0 (Rn ).

Example 102. Lu and Yang [291, 292] introduced the central bounded mean
oscillation space CBMOp (Rn ) with the norm

Z ! p1
1 p
kf kCBMOp ≡ sup |f (x) − fQ(r) | dx .
r>0 |Q(r)| Q(r)

So, CBMOp (Rn ) = CBMOp,0,0 (Rn ). Let C be the space of all constant func-
tions and X /C mean the quotient space of X by C. Then CMOp (Rn ) and
CBMOp (Rn ) are expressed by Bσ (E)(Rn ) and Ḃσ (E)(Rn ), respectively, with
E = Lp (Rn )/C, kf kE(Q(r)) = kf − fQ(r) kLp (Q(r)) and σ = np .
Fu obtained λ-central BMO estimates for commutators of N -dimensional
Hardy operators [131, Theorem 1.1].
Chapter 6
General metric measure spaces

Morrey spaces can describe the local properties of functions over Rn . This
applies to a general metric measure space. We remark that a metric measure
space means a couple (X, d, µ) of a metric space (X, d) and a Borel measure µ.
We do not recall the notion of metric spaces; see [241, §4] for metric measure
spaces. Recall that a Borel measure µ is defined for the σ-field generated by
open sets.
This chapter considers two typical cases. The first is where the Lebesgue
measure is replaced by a general Radon measure on Rn . Recall that a Radon
measure in a metric space (X, d) is a Borel measure such that µ(K) < ∞ for
any compact set K. Here, as examples the fractal sets generated by systems
of iteration functions are envisioned. Although fractal theory is not discussed
due to its vastness, the ternary Cantor set is recalled.
The second is where the underlying space differs from Euclidean space.
Hence, any (X, d) will do as long as (X, d) is a metric space. Graphs are
typical examples. An (unoriented) graph is a pair G = (V, E) comprising a
set V of vertices together with a set E of edges, which are 2-element subsets
of V . The (graph) distance between x and y is the minimal integer k such
that there exist k − 1 elements in between x and y. Since graphs are not the
primary purpose of this book, this chapter does not go into detail. But let us
mention that we use graphs in Examples 109 and 110. One of our motivations
is an extension to the general metric measure setting. Metric measure spaces
have already been discussed in this book: A previous chapter considered the
weighted measure w(x)dx. This chapter is interested in cases where the norm
bound is independent of the weights. Such an estimate can be obtained if
the weighted case is considered. This aspect is examined later in this book.
Section 6.1 establishes the theory of Lebesgue spaces in Euclidean space with
general Radon measures, while Section 6.2 focuses on the theory of Lebesgue
spaces in metric measure spaces.

6.1 Maximal operators on Euclidean spaces with

general Radon measures
We suppose that we have a Radon measure µ on Rn . As we have seen,
the Hardy–Littlewood maximal operator M on Euclidean space Rn plays an
247
248 Morrey Spaces

important role in analysis. When we work on a metric measure space, we need

to generalize the definition of M . A natural candidate of this definition is as
follows:
Z
0 1
M f (x) ≡ sup |f (y)|dµ(y) (x ∈ Rn )
r>0 µ(B(x, r)) B(x,r)

and
χB(z,r) (x)
Z
M f (x) ≡ sup |f (y)|dµ(y) (x ∈ Rn ).
z∈Rn ,r>0 µ(B(z, r)) B(z,r)

Since we have no information on µ(B(x, r)) and µ(B(z, r)), these two opera-
tors must be treated differently. However, if we have enough information on
the relation between µ(B(x, r)) and µ(B(x, 2r)) for x ∈ Rn and r > 0, then we
can identify M and M 0 as follows: Assume that there exists a constant C > 0
such that the doubling condition µ(B(x, r)) ≤ µ(B(x, 2r)) ≤ Cµ(B(x, r))
holds for all x ∈ Rn and r > 0. Then M 0 f (x) ≤ M f (x) ≤ CM 0 f (x), x ∈ Rn
for any f ∈ L0 (µ). In fact, since
Z
1
M 0 f (x) = sup |f (y)|dµ(y),
z=x,r>0 µ(B(z, r)) B(z,r)

we have M 0 f (x) ≤ M f (x). Meanwhile, since B(z, r) ⊂ B(x, 2r) for all z ∈
B(x, r),

χB(z,r) (x)
Z
M f (x) = sup |f (y)|dµ(y)
z∈Rn ,r>0 µ(B(z, r)) B(z,r)
χB(z,r) (x)
Z
≤ sup |f (y)|dµ(y).
z∈Rn ,r>0 µ(B(z, r)) B(x,2r)

Thus, thanks to the doubling condition, we have

Z
1
M f (x) ≤ C sup |f (y)|dµ(y).
r>0 µ(B(x, 2r)) B(x,2r)

In addition to the identification of M and M 0 , we can prove the boundedness

of these operators. For all λ > 0 and f ∈ L1 (µ), we will show that

C0
Z
n
µ{x ∈ R : M f (x) > λ} ≤ |f (x)|dµ(x).
λ Rn

See Theorem 267. Consequently, if the measure is doubling, then we can

develop the theory of the maximal operator. A metric measure space is called
a space of homogeneous type if its measure is doubling.
However, there are many important measures which are not doubling. Here
are examples of non-doubling measures.
 General metric measure spaces 249

Example 103. Let ν ≡ (ν1 , ν2 , . . . , νnZ) ∈ Rn be a fixed vector. Denote by h·, ·i

the inner product. If we let µ(E) ≡ e2hx,νi dx for a Lebesgue measurable
E
set E, then we obtain a non-doubling measure µ. This measure seems to be
artificial but actually this measure naturally arises. In fact, if we consider
n  2 
∂ ∂
P
∆ν ≡ ∂xj 2 + 2νj ∂xj , then
j=1
Z
f (x)∆ν g(x)dµ(x) = −h∇f, ∇giL2 (µ) (6.1)
Rn

for all f, g ∈ Cc∞ (Rn ). See Exercise 84.

Thus, we are interested in the theory of metric spaces equipped with mea-
sures which are not always non-doubling. To handle general Radon measures,
we need a covering lemma, which will be obtained in Section 6.1.1. Using this
covering lemma, we investigate the boundedness property of these maximal
operators in Section 6.1.2. As an application of the boundedness property,
we establish a differentiation theorem in Section 6.1.3. Section 6.1.4 further
develops what we did in Sections 6.1.1 and 6.1.2. Instead of using balls or
cubes, we concentrate on the dyadic cubes. At first glance this seems to be a
mere generalization of what we have been doing. However, Chapter 7 will con-
vincingly explain why such a generalization is useful. We take up an example
of metric measure spaces in Section 6.1.5.

6.1.1 Covering lemmas on Euclidean spaces

Unlike the general setting, in the Euclidean case, we can take advantage
of the geometric structure of the space. We need a substitute of Besicovitch’s
covering lemma (Theorem 249). We start with the following sharp covering
lemma.
Theorem 247. For all k > 1 there exists an integer N = Nk , depending only
on the dimension and k, that satisfies the following:
Let {B(xλ , rλ )}λ∈L be a family of balls in Euclidean space endowed with a
standard distance. Suppose that sup rλ < ∞.
λ∈L
Then we can take disjoint subfamilies
{B(xρ , rρ )}ρ∈L1 , {B(xρ , rρ )}ρ∈L2 , . . . , {B(xρ , rρ )}ρ∈LN
[ [ [
such that B(xλ , rλ ) ⊂ B(xρ , krρ ).
λ∈L j=1,2,...,N ρ∈Lj

We aim to prove Theorem 247 via an auxiliary step. If k ≥ 5, we set Nk ≡ 1

and we can use Theorem 133, the 5r-covering lemma to prove Theorem 247.
Hence, for the proof of Theorem 247, we may assume that k ≤ 5. This may
be viewed also as a substitute of Vitali’s covering lemma. To prove Theorem
247, we prove it by posing another assumption on centers of balls.
250 Morrey Spaces

Lemma 248. Let k > √ 1, and let {B(xλ , rλ )}λ∈L be a family of balls and
assume that sup rλ ≤ k inf rλ < ∞. Then we can take disjoint subfamilies
λ∈L λ∈L
as in Theorem 247.
Proof As we mentioned, we may suppose that k ≤ 5; otherwise we can
use the classical 5r-covering lemma (Theorem 133).
First of all, a scaling allows us to assume sup rλ = 1. (We are working on
λ∈L
Euclidean space. Hence, we are able to multiply the scalar.)
In this part we divide the family of balls. More precisely we proceed as
follows: Let Q0 be a family of dyadic cubes of side length 1. Here, we are now
Yn
considering cubes of the form Q = [mj , mj +1), where the mj ’s are integers.
j=1
We abbreviate the dyadic cubes in Q0 to “cubes” for short. Let Q0 ≡ [0, 1)n
be a unit cube. We divide the cubes into subfamilies: If →
−
m = (m1 , m2 , . . . , mn )
n
is an element of N0 , we set

Q→ →
− → − 0 →
− n
m ≡ {Q ∈ Q0 : Q − ( p + m) = Q for some p ∈ (4Z) },
−

where 4Z ≡ {4l : l ∈ Z}. Note that the cubes in Q→ m satisfy the following
−
property: Suppose that Q and Q0 both belong to Q→ m and that Q and Q are
− 0

different, then the distance between the two cubes is larger than 3. Hence, if
the center of the ball B is in Q and the center of the ball B 0 is in Q0 , then B
and B 0 are disjoint.
Next we define →
− [
Lm ≡ {λ ∈ L : xλ ∈ Q}.
Q∈Q→
−
m

Taking into account the preceding paragraphs we may assume that all
centers of the balls are in Q0 . In fact once this is proven, by the last paragraph
we can take the balls satisfying the property of this lemma from Q→ m for any
−
→
− n →
− n
m ∈ N0 . For any m ∈ N0 , we obtain families B→
(1) (Nk )
, . . . , B→ . Translation
−
m −
m
shows the number Nk is not dependent on → −m. Hence, our desired family is
(j)
{B→− }→
− n
m m∈N0 ,j≤Nk
.
Hence, in what follows let us assume that all centers of the balls are in Q0
and that sup rλ = 1 by normalization.
λ∈L
First take a ball B(xλ1 , rλ1 ) arbitrarily from the family {B(xλ , rλ )}λ∈L .
1
The assumption k − 2 ≤ inf rλ ensures that the radius of each ball is between
λ∈L
1
k − 2 and 1. Thus, the ball
√ B(xλ1 , krλ1 ) contains all balls B(xλ , rλ ) such that
d(xλ , xλ1 ) is less than k − 1. √
Next take a ball B(xλ2 , rλ2 ) such that d(xλ2 , xλ1 ) ≥ k−1. We may choose
it arbitrarily as long as this condition is satisfied. As in the proceeding para-
graph, the ball√ B(xλ2 , krλ2 ) contains all balls B(xλ , rλ ) such that d(xλ , xλ2 )
is less than k − 1.
 General metric measure spaces 251

In this way we repeatedly take a ball B(xλj0 , rλj0 ) such that d(xλj0 , xλj ) ≥
√
k − 1 for all j = 1, 2, . . . , j 0 − 1. This procedure will be stopped at the J0 th
[ J0
[
step where we obtain B(xλ , rλ ) ⊂ B(xλj0 , krλj0 ).
λ∈L j 0 =1
In fact this procedure stops in finite times: Precisely speaking, we claim
that J0 appearing in the last part is bounded by the constant that depends
only on k > 1 and n. Since all radii of the balls are at most 1, all balls are √ con-
tained in [−1, 2]n . By the construction of {xλj }, we have d(xλj , xλj0 ) ≥ k −1
for all j < j 0 ≤ J0 . Thus, we have
( J0 disjoint balls
!) whose radii are more than
√ √
k−1 k−1
. Precisely speaking, B xj , is disjoint. Note
2 2
j=1,2,...,J0
√ !
k−1
that B xj , is contained in [−1, 2]n for all j = 1, 2, . . . , J0 . Hence,
2
√ !n
k−1
we have J0 |B(1)| ≤ 3n . Thus, J0 is bounded by the quantity
2
which depends only on k > 1 and n. We denote this bound by N ; J0 ≤ N. If
J0 is less than N , we formally define Lj ≡ ∅ for j > J0 . Thus, we obtain the
desired families.
Next we prove Theorem 247. That is, we want to eliminate the assumption
√
sup rλ ≤ k inf rλ < ∞.
λ∈L λ∈L

We use an abbreviation; k B stands for B(x, k r) when B = B(x, r) and k > 0.

Proof of Theorem 247 Again we may assume that sup rλ = 1. First we
λ∈L
take the subfamilies Bj,p inductively as follows (j runs through all positive
integers and p through [1, N ], where N is a number obtained in the Lemma
248): First we define X1 by
n 1
o
X1 ≡ B(xλ , rλ ) : λ ∈ L, rλ > k − 2 .

Let B1,p be families obtained from X1 , using the Lemma 248. Suppose we have
obtained the families of the balls Bl,p with l = 1, 2, . . . , j, p = 1, 2, . . . , N and
that Xl with l = 1, 2, . . . , j are defined as the subsets of {B(xλ , rλ )}λ∈L . Then
we define
Xj+1 (6.2)
 
 j+1 j [ [ 
≡ B = B(xλ , rλ ) : k − 2 < rλ ≤ k − 2 , B ⊂ kB 0 .
 
1≤p≤N,1≤l≤j B 0 ∈Bl,p

We apply the Lemma 248 to this subset to obtain Bj+1,p with p = 1, 2, . . . , N ,

which enjoy the following properties: {Bj+1,p }1≤p≤N are disjoint subfamilies
252 Morrey Spaces

and [ [ [
B⊂ k B.
B∈Xj+1 p=1,2,...,N B∈Bj+1,p

By the definition of Xj we have;

1
(1) whenever B(xλ , rλ ) ∈ Xj , rλ ≤ √ j−1 ,
k
j−1
[ [ N [
(2) B(xλ , rλ ) is not contained in kB.
l=1 p=1 B∈Bl,p

Next we claim that there is an integer N 0 , whichS depends only on k,

S that
satisfies the following: If |j − l| > N 0 , then B 0 ∈ Bj,p , and if B ∈ Bl,p ,
p p
then B ∩ B 0 = ∅.
In fact suppose that B ∩ B 0 3 x and l > j. Then by the property noted
above, there exists y ∈ B \ kB 0 . Let c be the center of B 0 . If E is a subset of
Rn , diam(E) denotes the diameter of E. Under this notation and setting we
have
k 1
d(x, y) ≤ diam(B), d(c, y) ≥ diam(B 0 ), and d(c, x) ≤ diam(B 0 ).
2 2
(k − 1)
Thus diam(B 0 ) < diam(B). By the construction of Xj , we have
2
2 2
diam(B) ≤ √ l−1 , diam(B 0 ) ≥ √ j ,
k k
2 (k − 1)
hence we have √ l−1 ≥ √ j . Since k > 1, this is possible only if the
k k √ √
difference of j and l is small. That is, log 2k−1k > (l − j) log k, implying that
h √ i
1
N = 1 + k − 2 log 2k−1k is enough.
S
Put Gj,p ≡ Bi,p . Then {Gj,p }j≤N 0 ,p≤N does satisfy all
i : i≡j (mod N 0 )
demands of the theorem thanks to the property of N 0 .
We will need a covering technique called Besicovitch’s covering lemma.
Theorem 249 (Besicovitch). Let G ≡ {Bλ }λ∈Λ be a family of balls. Suppose
the diameters of balls {Bλ }λ∈Λ are bounded. R ≡ sup r(Bλ ) < ∞. Then there
λ∈Λ
exists an integer N depending only on the dimension that enjoys the following
property:
We can pick N disjoint subfamilies G1 , G2 , . . . , GN from G so that, any
element in A ≡ {c(Bλ )}λ∈Λ , the set of centers of balls {Bλ }λ∈Λ , belongs to a
ball in some Gj .
 General metric measure spaces 253

Proof We may assume that A is bounded. We use the 5r-covering lemma

(see Theorem 132) for the family
 
∗ 9R
Λ0 ≡ λ : λ ∈ Λ, r(Bλ ) > .
10

Then we obtain a countable family of balls {Bλ }λ∈Λ0 satisfying two conditions
below:
[ 1 [
(1) Bλ ⊂ Bλ ,
∗
5
λ∈Λ0 λ∈Λ0
 
1
(2) Bλ is disjoint.
5 λ∈Λ0

Since we assume A is bounded, ]Λ0 is finite. Enumerate {Bλ }λ∈Λ0 : Then

[ 1
we obtain a subfamily B1 , B2 , . . . , BN (1) of {Bλ }λ∈Λ such that Bλ ⊂
5
λ∈Λ∗
0
N (1)  N (1)
[ 1
Bλ and that Bj is disjoint. This is the first generation of the
j=1
5 j=1
N (1)
[
balls. Let A1 ≡ A \ Bλ and Λ∗∗
1 ≡ {λ ∈ Λ : c(Bλ ) ∈ A1 }.
j=1
9R
Observe that sup r(Bλ ) ≤ . Therefore, we are in the position
λ∈Λ∗∗
1
10
of applying the above observation to {Bλ }Λ∗∗ 1
. Consequently, we can find
BN (1)+1 , BN (1)+1 , . . . , BN (2) from {Bλ }Λ∗∗
1
with the following properties.
81R
(1) If λ ∈ Λ satisfy r(Bλ ) ≥ , then there exists ρ ∈ Λ0 ∪ Λ1 so that
100
c(Bλ ) ∈ Bρ ,
 
1 1 1
(2) The family BN (1)+1 , BN (1)+1 , . . . , BN (2) is disjoint.
5 5 5
1 1
Note that Bj and Bk are disjoint, if j ∈ Λ0 and k ∈ Λ1 . Indeed, c(Bk ), the
5 5
81
center of Bk , does not belong to Bj and r(Bj ) ≤ r(Bk ) ≤ r(Bj ). This is
100
the second generation of the balls.
Repeating this procedure we can find a sequence of balls {Bj }j∈J from
{Bλ }λ∈Λ so that it satisfies the following properties.
(1) J = {1, 2, . . . , j0 } or J = N,
[
(2) A ⊂ Bj ,
j∈J
254 Morrey Spaces

(3) if J = N, then r(Bj ) ↓ 0 as j → ∞,

 
1
(4) Bj is disjoint.
5 j∈J

Note that (2) will guarantee that any element in A, belongs to a ball in some
Gj . The generation of the ball B is defined to be the number j satisfying
B = Bρ with ρ ∈ Λj .
A trivial but important observation is that different balls Bj and Bk inter-
sect, then either j > k or k < j holds. Let k ∈ N be fixed, keeping the fact
above in mind. Then we set

Ik ≡ {j ∈ N : j < k, Bj ∩ Bk 6= ∅}, Jk ≡ {j ∈ Ij : rj ≤ 3rk }.

We will prove that there exists Mn > 0 such that ]Ik ≤ Mn  by proving

1
](Ik \ Jk ) . 1 and ]Jk . 1. Let j ∈ Jk . Then rk ≤ rj ≤ 3rk . Since Bj
5 j∈J
is disjoint, it follows that Jk .n 1.
Let j1 , j2 ∈ Ik \ Jk . If the generation of Bj1 and that of Bj2 are different,
then the balls Bj1 and Bj2 meet in a point and c(Bj ) lies sufficiently close
to the boundary of Bj1 and Bj2 . A geometric observation therefore shows the
angle ∠c(Bj1 )c(Bj )c(Bj2 ) & 1.
For each j ∈ Ik \ Jk , consider a point at which Bj and the segment
c(Bj )c(Bk ) meet. Then the observation of the above paragraph says that such
intersection points are torn apart. Even from this observation, we conclude
that the number of generations appearing in Ik \ Jk is bounded by a number
depending on n.
Of course, for each generation, the number of balls that can intersect Bk
is also bounded by a number depending on n. Therefore, ](Ik \ Jk ) .n 1.
In view of this observation we conclude that ]Ik is bounded by a constant
independent of k, say Mn .
We define a mapping σ : N → {1, 2, . . . , Mn + 1} as follows : First we set
σ(j) ≡ j for j ≤ Mn + 1. If j ≥ Mn + 2, then we define inductively

σ(j) (6.3)
−1
≡ min{l = 1, 2, . . . , Mn + 1 : Bj ∩ Bm = ∅ for all m ∈ σ (l) ∩ [1, j − 1]}.

We remark that by the pigeon hole principle the set appearing in (6.3) is not
empty, which guarantees that σ(j) is well defined. From the property of Mn ,
we learn that {Bj }j∈σ−1 (l) is disjoint for each l = 1, 2, . . . , Mn + 1. Thus we
have {{Bj }j∈σ−1 (l) }M n +1
l=1 is the desired family.
 General metric measure spaces 255

6.1.2 Maximal operators on Euclidean spaces with general

Radon measures
We consider the maximal operator given by
Z
B χB (x)
Mκ,uc f (x) = Mκ,uc f (x) ≡ sup |f (y)|dµ(y) (x ∈ Rn ).
B∈B µ(κB) B

The parameter κ > 0 is called the modification rate.

We cannot hope for the boundedness of M1,uc as the following example
shows:
2
Example 104. Let µ(x) = e|x| dx. Here, we will show that M1,uc can be
unbounded by disproving
Z
C
µ{x ∈ Rn : M1,uc f (x) > λ} ≤ |f (x)|dµ(x) (f ∈ L1 (µ), λ > 0).
λ Rn
For simplicity we assume n = 2. Assume to the contrary that this weak-(1, 1)
χB(r)
maximal inequality holds. If we take f ≡ , r > 0 then
µ(B(r))
C
µ{x ∈ R2 : M1,uc f (x) > λ} ≤ .
λ
Letting r ↓ 0, we obtain
C
µ{x ∈ R2 : M1,uc δ(x) > λ} ≤ ,
λ
where
1
M1,uc δ(x) ≡ .
µ(B(x/2, |x|/2))
1
Let λ ≡ for r > 0. Then we claim
µ(B((r, 0), r))
{x ∈ R2 : M1,uc δ(x) > λ} ⊃ B(2r).
To check this, we let x ∈ B(2r) and prove M1,uc δ(x) > λ. To this end by rota-
tion we may assume the second component of x is 0 and the first component
of x is positive. Hence, we have x ∈ B((r, 0), r). Then
1 1
M1,uc δ(x) = > = λ.
µ(B(x/2, |x|/2)) µ(B((r, 0), r))
This proves the claim.
From this claim, for all r > 0, we have µ(B(2r)) ≤ Cµ(B((r, 0), r)),
namely, Z Z
2 2 2 2
es +t dsdt ≤ C es +t dsdt.
s2 +t2 ≤4r 2 (s−r)2 +t2 ≤r 2
256 Morrey Spaces

If we use the polar coordinate, then we obtain

Z 2r Z π2 Z 2r cos θ Z π
2
u2 u2 2
cos2 θ
2π ue du ≤ C ue dudθ = C (e4r − 1)dθ.
0 −π
2 0 −π
2

Thus,
Z π
2
−4r 2 2
sin2 θ 2
π(1 − e )≤C (e−4r − e−4r )dθ.
−π
2

If we let r → ∞, then we obtain 3.14 < π ≤ 0, which is a contradiction.

We modify the Hardy–Littlewood maximal operator. We now consider
Z
1
Mk,uc f (x) ≡ sup |f (y)|dµ(y),
x∈B(z,r),r>0 µ(B(z, kr)) B(z,r)

which is called the uncentered maximal operator on Rn . Here f ∈ L0 (µ). The

additional parameter k is necessary because the maximal operator M1,uc can
be unbounded as Example 104 shows.
Theorem 250. Let µ be a Radon measure on Rn . We have the dual inequality;
Z Z
1
|g(x)|dµ(x) .a,b Ma,uc g(x)|f (x)|dµ(x)
{Mb,uc f >λ} λ Rn

for b > a > 1 and f, g ∈ L0 (µ).

Using Theorem 247, we prove Theorem 250.
Proof Fix R > 0. We truncate the maximal function again. Put again
Z
R 1
Mb,uc f (x) ≡ sup |f (y)|dµ(y).
x∈B(z,r),R>r>0 µ(B(z, br)) B(z,r)

Fix λ > 0. We abbreviate Eb ≡ {x ∈ Rn : Mb,uc R

f (x) > λ}. For all x ∈ Eb
by its definition there exists rx ∈ (0, R) and yx ∈ Rn such that

x ∈ B(yx , rx ) (6.4)

and that Z
1
|f (z)|dµ(z) > λ. (6.5)
µ(B(yx , brx )) B(yx ,rx )

Note that sup rx is at most R. Hence, we can apply Theorem 247. Applying
x∈Eb
the theorem with k = b/a > 1, we obtain a countable subset A ⊂ Eb such
that {B(yx , rx )}x∈A satisfies
[ [  b

B(yx , rx ) ⊂ B xj , rj (6.6)
j
a
x∈Eb
 General metric measure spaces 257

and X
χB(xj ,rj ) .a,b 1. (6.7)
j

Using (6.4)–(6.7), we have

Z Z
g(x)dµ(x) ≤ g(x)dµ(x)
b
S
Eb B(xj , a rj )
j
Z
X 1
≤ g(x)dµ(x) × µ(B(xj , brj ))
j
µ(B(xj , brj )) B(xj , ab rj )
X  Z
1
R
≤ inf Ma,uc g(x) |f (y)|dµ(y)
j
x∈B(xj ,rj ) λ B(xj ,rj )
Z
Ca,b R
≤ |f (y)|Ma,uc g(y)dµ(y).
λ Rn

Tending R to ∞, we obtain the desired result.

As an application, we can prove the vector-valued inequality.
Theorem 251. If 1 < p, k < ∞ and 1 < q ≤ ∞, then
  q1   q1
X∞ X∞
 (Mk,uc fj )q  .p,q,k  |fj |q 
j=1 j=1
Lp (µ) Lp (µ)

for all {fj }∞ 0

j=1 ⊂ L (µ).

We leave the proof of Theorem 251; see Exercise 88.

Example 105. In considering the weighted norm inequalities it could not be
better if it held that
Z Z
|g(x)|dµ(x) ≤ Cκ |f (x)|Mκ,uc g(x)dµ(x), (6.8)
{x∈Rn : Mκ,uc f (x)>λ} Rn

but this fails in general. Here, we will disprove (6.8) with κ = 3. We consider
the case n = 2. We define a measure µ by posing a weight function f given
below :
∞
X 1
f (x) ≡ χRn \B(1) (x) + χB(2−m+1 )\B(2−m ) (x).
m=1
m!

Let µ ≡ f (x)dx.
We disprove (6.8) by reduction to the absurdity. Suppose we have inequal-
ity (6.8) with κ = 3. First of all fix an integer α. We are going to let α tend
to infinity later.
258 Morrey Spaces

Set Rm ≡ µ(B((2−m , 0), 3 · 2−m ))−1 . Then

B(21−m ) ⊂ {(x, y) ∈ R2 : M3 δ0 (x, y) > Rm },

where δ0 denotes a dirac measure massed on the origin. In fact, we let (x, y) ∈
B(21−m ). By the rotation invariance of the sets B(2 · 2−m ) and {(x, y) ∈
R2 : M3 δ0 (x, y) > λ}, we may assume that 0 ≤ x < 2−k+1 and y = 0.
Since 0, (x, 0) ∈ B((x/2, 0), (1 + s)x/2) for all s > 0, we have M3 δ0 (x, 0) >
µ(B((x/2, 0), (1+s)x/2))−1 . If s > 0 is sufficiently small, we have M3 δ0 (x, 0) >
µ(B((x/2, 0), (1 + s)x/2))−1 > Rk .
It follows from the claim that we have
Z
|g(x)|dµ(x) ≤ Cµ(B((2−m , 0), 3 · 2−m ))M3 g(0).
B(21−m )

Let ϕ ∈ M+ (µ) satisfy kϕkL1 (µ) = 1 supported on a small ball whose

center is the origin. For r  1/2 we take a function gr of the form gr =
α
X 1
2
ϕ(r2 · −xrj ), where α is chosen arbitrarily and xrj satisfies
j=1
r
 
2πj −m+1 2πj
lim xrj = 2−m+1 cos ,2 sin
r↓0 α α
and that

supp(gr )
 
α   
\ [ 2πj 2πj
⊂ B(2−m+1 )  B 2−m+1 cos , 2−m+1 sin ,r .
j=1
α α

Tending r ↓ 0, we have α . µ(B((2−m , 0), 3 · 2−m ))M3 µm,α (0), where

denoting δx as the dirac measure supported on x, we set
α
X
µm,p ≡ δ(2−m+1 cos 2πj ,2−m+1 sin 2πj ) .
α α
j=1

By the definition of M3 µm,α (0) we have

M3 µm,α (0)
] 1 ≤ j ≤ α : 2−k+1 cos 2πj 2πj

−k+1


α ,2 sin α ∈ B(y, r)
= sup .
(y,r)∈Rn+1 µ(B(y, 3r))
+
0∈B(y,r)

For a finite set J = {j1 , j2 , . . . , jm } with 1 ≤ j1 < j2 < . . . < jm ≤ α we

set     
2πj 2πj
Pj ≡ 2−k+1 cos , 2−k+1 sin (j ∈ J)
α α
 General metric measure spaces 259

SJ ≡ inf {µ(B(y, 3r)) : 0, Pj1 , . . . , Pjm ∈ B(y, r)} .

]J
Then M3 µk,α (0) can be written as M3 µk,p (0) = . max
SJ J⊂{1,2,...,α}
Fixing α, if ]J ≥ 2, we have by geometric observation
  that µ(SJ ) ≥
Cα −k −k 1
. Notice also that µ(B((2 , 0), 3 · 2 )) = O . Thus, we have
(m − 1)! m!
µ(B((2−k , 0), 3 · 2−k ))
lim = 0. Furthermore, S{j} ∼ µ(B((2−k , 0), 3 · 2−k ),
k→∞ SJ
where ∼ is independent of α and k. Thus, keeping α fixed, we have

α . lim sup M3 µk,p (0) . 1

k→∞

where the implicit constant is independent on α. Since α is arbitrary, we

obtained the desired contradiction and (6.8) is disproven.

6.1.3 Differentiation theorem

Having proven that the maximal operator Mk,uc is bounded when k >
1, we can now prove the differentiation theorem. A cube Q is doubling if
µ(2Q) ≤ 2n+1 µ(Q). Or more generally, for k > 1, a cube Q is a k-doubling
cube if µ(kQ) ≤ k n+1 µ(Q). Doubling balls are defined in a similar fashion.
Due to the geometric structure of Rn , we can prove that there are sufficiently
many doubling cubes no matter how irregular the measure µ is.
Lemma 252. Let µ be a Radon measure. Then there exists a µ-null set N
such that for all x ∈ Rn \ N , we can find a sequence {Q(x, rj )}∞j=1 of doubling
cubes or a sequence {B(x, rj )}∞j=1 of doubling balls such that rj ↓ 0.

Proof We consider balls; for cubes, a similar technique can be used. Let
us set  
n µ(B(x, r))
N ≡ x ∈ R : lim inf =0 . (6.9)
r↓0 rn
We claim that µ does not charge N . In fact, for all x ∈ N and ε > 0, there
exists rx ∈ (0, 1) such that µ(B(x, rx )) ≤ εrx n . Let R > 0 be arbitrary. By
Lemma 268 with δ = 1 (Vitali’s covering lemma), for any compact set K
contained in N ∩ B(R), we can find x1 , x2 , . . . , xL ∈ K such that
L
X L
[
χB(xl ,rxl /3) ≤ 1, K⊂ B(xl , rxl ).
l=1 l=1

In fact, we can apply Lemma 268 to {B(x, rx /3)}x∈K . Then

L L L  r 
x
X X X
µ(K) ≤ µ(B(xl , rxl )) ≤ ε rxl n . ε B xl , l . ε|B(x, 1 + R)|.
3
l=1 l=1 l=1
260 Morrey Spaces

Since K is an arbitrary compact set, we have µ(N ∩ B(R)) . ε|B(x, 1 + R)|.

Letting ε ↓ 0, we have µ(N ∩ B(R)) = 0. Finally, since R > 0 is arbitrary,
µ(N ) = 0.
Let us show that N is the set we are looking for. For all x ∈ Rn \ N , we
can find a non-negative decreasing sequence {rj }∞
j=1 such that
√ µ(B(x, ρ)) µ(B(x, 2rj ))
2 lim inf >
ρ↓0 ρn (2rj )n
and that
√ µ(B(x, r)) µ(B(x, ρ))
2 n
≥ lim inf >0
r ρ↓0 ρn
√
for all r ≤ 2 nr1 . Thus, it follows that
√
 
µ(B(x, ρ))
µ(B(x, 2rj )) ≤ 2 lim inf (2rj )n ≤ 2n+1 µ(B(x, rj )).
ρ↓0 ρn
Thus, we obtain the desired result.
A direct consequence of Lemma 252 is the Lebesgue differentiation theorem
for general measures, whose proof we have been postponed.
Theorem 253. Let µ be a Radon measure on Rn . Let f ∈ L1loc (µ). Then
 Z 
1 n+1
sup f (x) − f (y)dµ(y) : µ(2Q) ≤ 2 µ(Q), Q 3 x, |Q| ≤ r
µ(Q) Q
goes to 0 as r ↓ 0 for µ-a.e. x ∈ Rn .
In particular, when f is a Lebesgue measurable function on Rn ,
X χQ (x) Z
lim f (z)dz = f (x) (6.10)
j→∞
n
|Q| Q
Q∈Dj (R )

for almost all x ∈ Rn .

Proof The conclusion follows from by the use of M2,c .
A truncation of f allows us to assume that f ∈ L1 (µ). Let ε, ε0 > 0 be
arbitrary. Choose a function g ∈ L1 (µ) ∩ C(Rn ) so that kf − gkL1 (µ) < ε0 . Set
Eε ≡ Eε (f )
(   Z  )
1
≡ x ∈ Rn : lim sup sup f (x) − f (y)dµ(y) : ∗∗ > 2ε ,
r↓0 µ(Q) Q

where ∗∗ is an abbreviation of the condition µ(2Q) ≤ 2n+1 µ(Q), Q 3 x, |Q| ≤

r. Notice that Eε (f ) = Eε (f − g). In view of inclusion
Eε
(   Z  )
n 1
⊂ x ∈ R : lim sup sup f (y) − g(y)dµ(y) : ∗∗ >ε
r↓0 µ(Q) Q
n
∪ {x ∈ R : |f (x) − g(x)| > ε}
⊂ x ∈ Rn : M2,c [f − g](x) > 2−n−1 n−n ε ∪ {x ∈ Rn : |f (x) − g(x)| > ε} ,

 General metric measure spaces 261

Theorem 267 and the Chebychev inequality, we have

2n+1 nn + 1 ε0
µ(Eε ) ≤ kf − gkL1 (µ) ≤ (2n+1 nn + 1) .
ε ε
0
Since ε > 0 is arbitrary, it follows that µ(Eε ) = 0. Since ε > 0 is arbitrary,
we conclude the proof of Theorem 253.

6.1.4 Universal estimates

One of the strong thrusts of studying general metric measure spaces is to
obtain some estimates independent of the underlying measures; see Chapter
7. Sometimes such estimates are obtained with the constants independent of
the dimension. If the constant is independent of the dimension, we can make
the dimension as high as we wish. This fact is useful from the viewpoint of
applications in stochastic analysis and so on. Meanwhile, we intend to study
the general metric measure spaces which will be an application to analysis in
Euclidean spaces. The application will be obtained later in this book.
We start Section 6.1.4 by presenting a result on the dyadic maximal oper-
ator. We define
Z
D(µ) χQ (x)
M f (x) ≡ sup |f (z)|dµ(z) (x ∈ Rn ),
Q∈D(µ) µ(Q) Q

where D(µ) stands for the set of all cubes Q in D(Rn ) with µ(Q) > 0. By
using the fact that the dyadic cubes are nested, we can prove the following:
Theorem 254 (Universal weak L1 (µ)-estimate for M D(µ) ). Let µ be a Radon
measure on Rn , λ > 0 and f ∈ L1 (µ). Then
Z
1
µ{x ∈ Rn : M D(µ) f (x) > λ} ≤ |f (y)|dµ(y).
λ {x∈Rn : M D(µ) f (x)>λ}
Proof Mimic the proof of Theorem 134.
A direct consequence of Theorem 254 is the Lp (µ)-boundedness.
Theorem 255 (Universal Lp (µ)-estimate for M D(µ) ). Let µ be a Radon mea-
1
sure on Rn , and let 1 < p < ∞. Then kM D(µ) f kLp (µ) ≤ (p0 ) p kf kLp (µ) for all
f ∈ Lp (µ).
Proof If we integrate the inequality in Theorem 254 against λp−1 dλ over
(0, ∞), we first obtain
Z Z
D(µ) 0
M p
f (x) dµ(x) ≤ p M D(µ) f (x)p−1 |f (x)|dµ(x). (6.11)
Rn Rn
∞
We may assume that f ∈ L (µ) and that µ{f 6= 0} < ∞. In this case,
M D(µ) f ∈ Lp (µ). See Exercise 87. By Hölder’s inequality and (6.11), we obtain
Z Z
M D(µ) f (x)p dµ(x) ≤ p0 |f (x)|p dµ(x). (6.12)
Rn Rn
262 Morrey Spaces

Thus, we obtain the desired result.

Let f ∈ L0 (µ). We define
 Z 
D(µ) 1
Mlog f (x) ≡ sup χQ (x) exp log |f (y)|dµ(y) (x ∈ Rn ).
Q∈D(µ) µ(Q) Q
(6.13)
D(µ)
Theorem 256 (Universal Lp (µ)-estimate for Mlog ). Let 0 < p < ∞. Then
D(µ) 1
kMlog f kLp (µ) ≤ e kf kLp (µ) for all f ∈ L0 (µ).
p

Proof Let 0 < r < p. Then

  r1
D(µ) 1 p
kMlog f kLp (µ) ≤ k[M D(µ) (|f |r )] r kLp (µ) ≤ kf kLp (µ) .
p−r
If we send r ↓ 0, we obtain the desired result.
Define Q(µ) to be the set of all cubes with positive µ-measures. We let
D(µ; Q) ≡ D(Q) ∩ Q(µ). We consider the local universal estimates. For 0 ≤
D(µ;Q)
α < n, denote Mα the fractional maximal operator generated by dyadic
cubes in a cube Q. That is, for f ∈ L0 (Q) we define
Z
1
MαD(µ;Q) f (x) ≡ sup χR (x) α |f (z)|dµ(z) (x ∈ Rn ).
R∈D(µ;Q) µ(R)1− n R
We also define
 Z 
D(µ;Q) 1
Mlog f (x) ≡ sup χR (x) exp log |f (z)|dµ(z) (x ∈ Rn ).
R∈D(µ;Q) µ(R) R

D(Q) D(Q)
If µ is the Lebesgue measure, then we write Mα and Mlog instead of
D(µ;Q) D(µ;Q)
Mα and Mlog ,
respectively.
We have the following universal estimates:
Theorem 257 (Universal estimates over cubes). Let Q be a cube and 0 ≤
α < n.
D(µ;Q)
(1) Let f ∈ L1 (µ), λ > 0, and let Ωλ ≡ {x ∈ Q : Mα f (x) > λ}. Then
n−α
λµ(Ωλ ) n ≤ kf χΩλ kL1 (µ) . (6.14)

n 1 1 α
(2) Let f ∈ Lp (µ) with 1 < p ≤ . Set = − . Then
α q p n
! n−α
n
q
kMαD(µ;Q) f kLq (µ) ≤ n kf kLp (µ) .
q − n−α

In particular, if α = 0, then
kM D(µ;Q) f kLp (µ) ≤ p0 kf kLp (µ) . (6.15)
 General metric measure spaces 263
D(µ;Q)
(3) Let f ∈ L1 (µ). Then kMlog f kL1 (µ) ≤ ekf kL1 (µ) .
Proof Throughout the proof, we may assume that f ∈ L∞ (µ) by Fatou’s
lemma.
(1) We partition Ωλ by choosing maximal cubes contained Z in Ωλ : Ωλ =
S 1
Qθ , where each Qθ ∈ D(µ; Q) satisfies α |f (z)|dµ(z) >
θ∈Θ µ(Qθ )1− n Qθ
λ. Thus,
n−α X n−α X
λµ(Ωλ ) n ≤ λ µ(Qθ ) n ≤ kf χQθ kL1 (µ) ≤ kf χΩλ kL1 (µ) .
θ∈Θ θ∈Θ

(2) We observe that

Z ∞
(kMαD(µ;Q) f kLq (µ) )q = q λq−1 µ(Ωλ )dλ
Z0 ∞
n n
≤q λq−1− n−α (kf χΩλ kL1 (µ) ) n−α dλ.
0

We choose u ≥ 0 so that n(1 − u) = pα. Then by Hölder’s inequality

 u
kf χΩλ kL1 (µ) ≤ kf χΩλ kL n−α
nu
(µ)
(kf kLp (µ) )1−u .

Thus,

(kMαD(µ;Q) f kLq (µ) )q

Z ∞  nu
 n−α
n n(1−u)
≤q λq−1− n−α kf χΩλ kL n−α nu
(µ)
(kf kLp (µ) ) n−α dλ
0
Z
q n(1−u) nu n
= n (kf k p
L (µ) ) n−α |f (x)| n−α MαD(µ;Q) f (x)q− n−α dµ(x).
q − n−α Q

Observe that
   0
n (n − α)p q(p − 1)n
q− × = = q.
n−α nu (p − 1)n
By Hölder’s inequality, we obtain
Z
nu n
|f (x)| n−α MαD(µ;Q) f (x)q− n−α dµ(x)
Q
nu n
≤ (kf kLp (µ) ) n−α (kMαD(µ;Q) f kLq )q− n−α .

As a result,
q n n
(kMαD(µ;Q) f kLq (µ) )q ≤ n (kf kLp (µ) ) n−α (kMαD(µ;Q) f kLq )q− n−α .
q− n−α

If we arrange this inequality, we obtain the desired result.

264 Morrey Spaces

(3) Let u > 0. Note that

  u1
D(Q) D(µ;Q)
Mlog f (x) ≤ M D(µ;Q),(u) f (x) ≡ M0 [|f |u ](x) (x ∈ Rn ).

Thus, by (6.15)
D(Q)
kMlog f kL1 (µ) ≤ kM D(µ;Q),(u) f kL1 (µ)
1
= (kM D(µ;Q) [|f |u ]k u1 ) u
L (µ)
  u1
1
≤ kf kL1 (µ) .
1−u

If we let u ↓ 0, then we obtain the desired result.

D(µ)
Denote by Mα the fractional maximal operator for 0 ≤ α < n. That is,
for a measurable function f on Q we define
Z
1
MαD(µ) f (x) ≡ sup χR (x) α |f (z)|dµ(z).
R∈D(µ) µ(R)1− n R

We also define
 Z 
D(µ) 1
Mlog f (x) ≡ sup χR (x) exp log |f (z)|dµ(z) .
R∈D(µ) µ(R) R

We have the following universal estimates:

Theorem 258 (Universal estimates over Rn ). Let 0 ≤ α < n.
D(µ)
(1) Let f ∈ L1 (µ), λ > 0, and let Ωλ ≡ {x ∈ Q : Mα f (x) > λ}. Then
n−α
λµ(Ωλ ) n ≤ kf χΩλ kL1 (µ) .

n 1 1 α
(2) Let f ∈ Lp (µ) with 1 < p ≤ . Set = − . Then
α q p n
! n−α
n
q
kMαD(µ) f kLq (µ) ≤ n kf kLp (µ) .
q− n−α

In particular, kM D(µ) f kLp (µ) ≤ p0 kf kLp (µ) .

(3) Let f ∈ L1 (µ). Then
D(µ)
kMlog f kL1 (µ) ≤ ekf kL1 (µ) . (6.16)
 General metric measure spaces 265

Proof Let Qj ≡ [−2−j , 2j ) for j ∈ N and use Theorem 258. Since the
constants are independent of j, we can let j → ∞ to obtain the desired
results.
Instead of D(Rn ), it is sometimes convenient to use different families. We
do not go into detail; we content ourselves with the case n = 1.
Lemma 259 (Three lattice theorem). Write Zl (R) = {3Q : Q ∈ Dl (R)} for
each l ∈ Z. Then each Zl (R), l ∈ Z, is partitioned into 3 disjoint subfamilies
{Eelk (R)}3k=1 so that the following properties hold, where Elk = {Q ∈ Dl : 3Q ∈
Eelk (R)}.
∞
Eelk (R) which meet at a
S
(1) Let k = 1, 2, 3. For any cubes Q1 , Q2 ∈
l=−∞
point, one is contained in the other.
(2) For all Q ∈ Dl (R), k = 1, 2, 3 and l ∈ Z, there exists Q0 ∈ Elk (R) such
that Q ⊂ 3Q0 ⊂ 5Q.
P
(3) For any k = 1, 2, 3 and l ∈ Z, χQ = 1.
Q∈Eelk

Proof Let

We define Elk (R) with k = 1, 2, 3 and l = 0, −1, −2, −3 as follows:

E01 (R) ≡ {. . . , [−3, −2), [0, 1), [3, 4), [6, 7), . . .},
1
E−1 (R) ≡ {. . . , [−2, 0), [4, 6), [10, 12), [16, 18), . . .},
1
E−2 (R) ≡ {. . . , [0, 4), [12, 16), [24, 28), [36, 40), . . .},
1
E−3 (R) ≡ {. . . , [−8, 0), [16, 24), [40, 48), [64, 72), . . .},
E02 (R) ≡ {. . . , [−2, −1), [1, 2), [4, 5), [7, 8), . . .},
2
E−1 (R) ≡ {. . . , [0, 2), [6, 8), [12, 14), [18, 20), . . .},
2
E−2 (R) ≡ {. . . , [4, 8), [16, 20), [28, 32), [40, 44), . . .},
2
E−3 (R) ≡ {. . . , [0, 8), [24, 32), [48, 56), [72, 80), . . .},
E01 (R) ≡ {. . . , [−1, 0), [2, 3), [5, 6), [8, 9), . . .},
1
E−1 (R) ≡ {. . . , [2, 4), [8, 10), [14, 16), [20, 22), . . .},
1
E−2 (R) ≡ {. . . , [8, 12), [20, 24), [32, 36), [44, 48), . . .}
1
E−3 (R) ≡ {. . . , [8, 16), [32, 40), [56, 64), [80, 88), . . .}.
k k
From the definition of E2l (R) and E2l−1 (R) with k = 1, 2, 3 and l ∈ Z, we can
k
guess the defintion of each El (R). We leave the interested readers to check
that the family {Elk (R)}k=1,2,3,l∈Z satisfies (1)–(3).
266 Morrey Spaces

6.1.5 Examples of metric measure spaces

Here we take up some examples. The first one is the p-adic field.
Example 106. Let p ∈ N be a fixed prime number. Equip Q with a distance
function as follows: the distance between x, y ∈ Q is dp (x, y) = p−m , where
the integers q, r, m satisfy x − y = pm q/r and the integers q and r are not
multiples of p; see [155]. Denote by Qp the completion with respect to dp of
Q. Since Qp carries the structure of a locally compact Abelian group, Qp has
the invariant Haar measure µ.
We next consider the Gauss measure space as an example of metric measure
spaces. Denote by B(Rn ) the Borel σ-field over Rn . The Gauss measure space
is a triplet (Rn , | · |, γ), where
Z
2
−n
γ(E) ≡ π 2 e−|x| dx (E ∈ B(Rn )). (6.17)
E

The measure γ is called the Gauss measure. We distort the underlying measure
dx; the ambient space Rn together with the distance remains intact. Here, we
content ourselves with introducing the notion of maximal admissible balls and
the mother of a maximal admissible ball due to Mauceri and Meda.
Definition 67. Let a > 0 and define m(x) ≡ min(1, |x|−1 ).
(1) Define Ba ≡ {B(x, r) : (x, r) ∈ Rn+1
+ , 0 < r ≤ am(x)}.

(2) Write [0, x] ≡ {tx : t ∈ [0, 1]} for x ∈ Rn .

(3) Any ball B ∈ Ba is maximal if r(B) = am(c(B)).
(4) For each maximal ball B ∈ Ba which does not contain the origin, M (B)
denotes the maximal ball in Ba with center at a point on the segment
[0, c(B)] such that the boundary of M (B) contains c(B). Call M (B)
the mother of B. In other words, the relation between B = B(x, r)
and its mother M (B) = B(z, R) satisfy R = am(z), |z| + R = |x| and
|x|z = |z|x 6= 0. For notational convenience, we set M 0 (B) ≡ B.
(5) If M (B) does not contain the origin, then we may consider the mother
of M (B), which we denote by M 2 (B). Therefore, given any maximal
ball B in Ba , we find a chain of maximal balls, B, M (B), M 2 (B), . . . ,
M k (B), with the property that M j (B) is the mother of M j−1 (B) for
j ∈ {1, 2, . . . , k}, and M k (B) contains the origin.
Example 107. Let a > 0. Let {aj }∞ j=0 be a sequence defined in Lemma 262.
If the ball B = B(z, r) satisfies that |z| = aj and r = aaj for some j ∈ N, then
B is a maximal ball in Ba .
The class Ba has the following control:
 General metric measure spaces 267

Proposition 260. Let B = B(cB , r) ∈ Ba , a > 0, and let x ∈ B. Then

−a2 − 2a ≤ |c(B)|2 − |x|2 ≤ 2a.
Proof The proof is left as an exercise. See Exercise 85.
We establish some subtle geometric relations between the maximal admis-
sible balls and their mothers.
Lemma 261. Let a > 0. If the maximal ball B ∈ Ba does not contain 0, then
B ⊂ (a + 2)M (B) and M (B) ⊂ (2a + 2)B. Consequently, γ(M (B)) ∼ γ(B)
with implicit constants depending only on a and n.
Proof By the definition of M (B), we have r(M (B)) = am(c(M (B))),
r(B) = am(c(B)) and |c(M (B))| + r(M (B)) = |c(B)|. Using the fact that
c(B) is on the boundary of M (B), together with the continuity of the function
m, we see that (a + 1)−1 m(c(M (B))) ≤ m(c(B)) ≤ (a + 1)m(c(M (B))).
Hence, (a + 1)−1 r(M (B)) ≤ r(B) ≤ (a + 1)r(M (B)). This implies that for
any z ∈ B, |z −c(M (B))| ≤ |z −c(B)|+|c(B)−c(M (B))| < r(B)+r(M (B)) ≤
(a + 2)r(M (B)). That is, B ⊂ (a + 2)M (B). Meanwhile, for any z ∈ M (B),
we have
|z − c(B)| ≤ |z − c(M (B))| + |c(M (B)) − c(B)| < 2r(M (B)) ≤ 2(a + 1)r(B).
which in turn implies that M (B) ⊂ (2a + 2)B. Finally, we have γ(B) ≤
γ((a + 1)M (B)) . γ(M (B)) and γ(M (B)) ≤ γ((2a + 2)B) . γ(B), which
completes the proof of Lemma 261.
We next handle a special recurrence formula to deal with the Gauss mea-
sure space.
a
Lemma 262. Fix a constant a > 0. Let ϕ(s) ≡ s + for s > 0. Let {aj }∞ j=0
s
be the sequence recursively defined by
a
a0 ≡ 1 and aj+1 ≡ aj + = ϕ(aj ) for j ∈ N. (6.18)
aj
√ p
(1) For all j ∈ N, 2aj ≤ aj ≤ (a2 + 2a)j.
(2) For all i, j ∈ N, 2a(j − i) ≤ a2j − a2i ≤ (j − i)(a2 + 2a).
Proof
(1) Indeed, this follows from
p a p
s2 + 2a < s + < s2 + a2 + 2a, s ∈ [1, ∞),
s
√ p
and the fact that the sequences { 2aj }∞ 2 ∞
j=1 and { (a + 2a)j }j=1 can
be recursively defined by
( √ ( √
s1 = 2a s1 = a2 + 2a
q and q
sj+1 = s2j + 2a sj+1 = s2j + a2 + 2a,

respectively.
268 Morrey Spaces

a2
(2) For any j ∈ N0 , since aj+1 2 − aj 2 = 2a + from (6.18) we have
aj 2
2a ≤ aj+1 2 − aj 2 ≤ 2a + a2 . (6.19)
j−i
Consequently, for all j ≥ i ≥ 0, by aj 2 − ai 2 = (ak+i 2 − ak+i−1 2 ) and
P
k=1
(6.19), we see that (2) holds.
Lemma 263. Let a > 0 and B be a maximal ball in Ba . Suppose that the
sequence {aj }∞
j=0 is defined as in (6.18). Then, the following hold for each
j ∈ N:
(1) if |c(B)| ≤ 1 + a, then γ(B) ∼ 1;
(2) if aj ≤ |c(B)| < aj+1 for some j ∈ N, then either |c(M (B))| ≤ 1 + a or
aj−1 ≤ |c(M (B))| < aj ;
2 n
(3) if aj ≤ |c(B)| < aj+1 , then γ(B) ∼ e−aj j − 2 ;
(4) if |c(B)| > 1 + a and {M k (B) : 0 ≤ k ≤ k0 } is a chain of maximal
balls in Ba , with the property that M k (B) is the mother of M k−1 (B) for
k ∈ {1, 2, . . . , k0 }, where k0 is the smallest integer such that |cM k0 (B) | ≤
2
  n2
γ(B)
1 + a, then e−(a +2a)k 1 − k0k+1 . γ(M k (B)) . e
−2ak
for all k ∈
{1, 2, . . . , k0 }.
Here, the implicit positive constants in (1)–(4) depend only on a and n.
Proof
(1) The proof is left as an exercise; see Exercise 86.
(2) For the purpose of showing |c(M (B))| < aj if |c(M (B))| > 1 + a, we
show that, when |c(M (B))| > 1 + a, we have aj−1 ≤ |c(M (B))| < aj .
This follows from reductio ad absurdum. Indeed, if |c(M (B))| ≥ 1 + a
a
√ |c(M (B))| ≥ aj , then by observing that ϕ(s) = s + s increases on
and
[ a, ∞), we obtain
a
|c(B)| = |c(M (B))| + |c(M (B))|−1 a ≥ aj + = aj+1 ,
aj
which contradicts assumption |c(B)| < aj+1 . Therefore, we have
|c(M (B))| < aj . Likewise, we have aj−1 ≤ |c(M (B))|; the details are
left to the reader. This proves (2).
√ p
(3) Lemma 262 yields 2aj ≤ aj ≤ |c(B)| < aj+1 ≤ (a2 + 2a)(j + 1),
which, combined with |c(B)| ≥ aj > 1, implies r(B) = am(c(B)) =
1
|c(M (B))|−1 a ∼ j − 2 . From aj ≤ |c(B)| < aj+1 and (6.19), it follows
2 2 2 2 2 2 2
that e−a −2a < eaj −aj+1 < eaj −|c(B)| ≤ 1. Hence, eaj −|c(B)| ∼ 1 or,
2 2 1
equivalently, e−|c(B)| ∼ e−aj . This and the aforementioned r(B) ∼ j − 2 ,
2 2 n
yield that γ(B) ∼ e−|c(B)| r(B)n ∼ e−aj j − 2 . Hence, (3) holds.
 General metric measure spaces 269

(4) Without loss of generality, we may assume that k0 ≥ 2, since γ(B) =

γ(M 0 (B)) and γ(M (B)) ∼ γ(B) by Lemma 261. Notice that

|c(B)| > |c(M (B))| > · · · > |cM k0 −1 (B) | > 1 + a =: a1 . (6.20)

Thus, there exists some j ∈ N such that aj ≤ |c(B)| < aj+1 . Applying
(6.20) and (2) repeatedly, we conclude that, for all k ∈ {1, 2, . . . , k0 },

aj−k ≤ |cM k (B) | < aj+1−k . (6.21)

In particular, when k = k0 − 1, by (6.20) and (6.21), we have a1 <

|cM k0 −1 (B) | < aj−k0 +2 . From this and the fact that {aj }∞ j=0 is strictly
increasing, it follows that j ≥ k0 . For 1 ≤ k ≤ k0 − 1, by (3) and (6.21),
we obtain
2 n n
e−aj j − 2

γ(B) −(a2j −a2j−k ) k 2
∼ 2 ∼ e 1 − .
γ(M k (B)) e−aj−k (j − k)− 2
n
j

Moreover, from Lemma 262(2), we infer that

 n  n
2 k 2 2 2 k 2
e−k(a +2a) 1 − . e−(aj −aj−k ) 1 − . e−2ak .
k0 j
Thus, when 1 ≤ k ≤ k0 − 1,
  n2
−k(a2 +2a) k γ(B)
e 1− . . e−2ak .
k0 + 1 γ(M k (B))

By Lemma 261, we have γ(M k0 (B)) ∼ γ(M k0 −1 (B)). Thus

γ(B) γ(B)
∼ ,
k
γ(M (B))
0 γ(M k0 −1 (B))
which implies that
  n2
−k0 (a2 +2a) k0 γ(B)
e 1− . . e−2ak0 .
k0 + 1 γ(M k0 (B))
This concludes the proof of (4).
For later applications, we establish the following covering lemma. The key
point is that the number of balls that cover the annulus B(|aj+1 |) \ B(|aj |)
with j ∈ N0 can be estimated from above (see Lemma 264(D)are below),
which plays a key role in applications. For simplicity, we only consider the
balls in B1 in Lemma 264 below; similar results hold for the balls in Ba with
general a > 0.
Lemma 264. Let a ≡ 1 and {aj }∞ j=0 be defined as in (6.18), namely, a0 ≡ 1,
and aj+1 ≡ aj + a1j when j ∈ N. Then, there exists a family {Bj,i }j∈N0 , i∈Ij
of maximal balls in B1 with the following property:
270 Morrey Spaces
1
(A) Each Bj,i is centered at the boundary of B(aj ) and with radius aj .

∞ S
(B) Rn =
S
4Bj,i .
j=0 i∈Ij

(C) The family { 21 Bj,i }j∈N0 , i∈Ij is disjoint.

(D) For any j ∈ N0 , Ij is an index set such that ]Ij .n (j + 1)n with the
implicit positive constant depending only on n.
Proof Set A0 ≡ B(1). For j ∈ N, set Aj ≡ B(aj ) \ B(aj−1 ). Then,
∞
n
S
R = Aj . For each j ∈ N0 , let Aj = {Bj,i }i∈Ij , where Ij is an index set,
j=0
be the collection of maximal balls with the following property:
(a) For any ball B ∈ Aj , B is centered at the boundary of B(aj ) and with
radius a1j .

(b) Any two balls in Aj are disjoint.

(c) For any ball B ∈ / Aj but centered at the boundary of B(aj ) and with
radius a1j , there exists some ball B 0 ∈ Aj which intersects B.
The existence of such Aj is guaranteed by Zorn’s lemma. With this setup, we
establish that this collection satisfies the desired properties.
(A) Clearly, this is a consequence of (a).
(B) We distinguish two cases: x ∈ A0 and x ∈ Aj for some j ∈ N.
For any x ∈ A0 and B ∈ A0 , since |c(B)| = 1 and r(B) = 1, we have
|x − c(B)| ≤ |x| + |c(B)| < 2 ≤ 4r(B), which implies that
[ ∞ [
[
A0 ⊂ 4B0,i ⊂ 4Bj,i .
i∈I0 j=0 i∈Ij

We now assume that x ∈ Aj for some j ∈ N; to establish (B) it suffices

to show that there exists some ball B ∈ Aj such that x ∈ 4B. Indeed, if,
/ 4B, then |c(B)| = aj ≥ 1, r(B) = a1j
for all balls B ∈ Aj , we have x ∈
and |x − c(B)| ≥ 4r(B) = a4j . Consider the ball centered at aj |x| x
and
with radius a1j , denoted below by B ∗ , namely, B ∗ ≡ B(aj |x|
x
, a1j ). Note
1
that ||x| − aj | < aj − aj−1 = aj−1 when x ∈ Aj , and that a2j ≥ aj−1 1
.
Since
x x
|c(B) − z| ≥ |c(B) − x| − x − aj − aj −z ,
|x| |x|
for any z ∈ B ∗ and B ∈ Aj , we have
4 1 3 1 1
|c(B) − z| ≥ − ||x| − aj | − ≥ − ≥ .
aj aj aj aj−1 aj
 General metric measure spaces 271

Thus, B ∗ does not intersect any ball in Aj , which contradicts (c). Thus,
S ∞ S
S
for all j ∈ N, Aj ⊂ 4Bj,i ⊂ 4Bk,i . This shows the property
i∈Ij k=0 i∈Ik
(B).
(C) By virtue of (a) and (b), together with the definition of Aj , we see
that any ball B from Aj only
S intersects the balls in Aj−1 or Aj+1 .
If B ∈ Aj and B 0 ∈ Aj−1 Aj+1 , by noticing that the width of the
annulus {Aj }∞ 1 1 0
j=1 is decreasing, we see that 2 B ∩ 2 B = ∅. Hence, we
obtain the desired result.
(D) From (b) and a volume argument, it follows easily that ]Aj is bounded
by a positive constant multiple of |aj |2n . That is, ]Ij . j n ∼ (j + 1)n
for j ∈ N by virtue of Lemma 262. Similarly, we see that ]I0 . 1.
The Gauss measure γ is locally doubling in the following sense:
npq
1+ p−q
Lemma 265. Let Cp,q,n ≡ 32 7 log(8 ). If a ball B(x, r) satisfies that
npq
1+ p−q
γ(B(x, 8r)) ≤ 8 γ(B(x, r)), then either |x| < 2r or B(x, r) ∈ BCp,q,n .
Proof It suffices to establish that B(x, r) ∈ BCp,q,n when |x| ≥ 2r. If
x ∈ B(2), then 4m(x) ≥ 2, which together with the fact |x| ≥ 2r gives that
r ≤ 1 ≤ 2m(x). Thus, B(x, r) ∈ B2 ⊂ BCp,q,n .
Now assume that |x| ≥ 2. In this case, m(x) = |x|−1 . Notice that
( 2 )
2 2 2 5
min |y| = (|x| − r) , max |y| = max |x| − r
y∈B(x,r) y∈B (x− 2|x|
3r
x, 41 r ) 4

3r 1
and that B(x − 2|x| x, 4 r) ⊂ B(x, 8r). Consequently, we have
npq n 2 npq
81+ p−q π − 2 e−(|x|−r) |B(x, r)| ≥ 81+ p−q γ(B(x, r)) ≥ γ(B(x, 8r))
  
3r 1 n 5 2
≥γ B x− x, r ≥ π − 2 e−(|x|− 4 r) |B(x, r)|,
2|x| 4

which together with r ≤ 2−1 |x| yields that

2 2 npq
e 32 |x|r ≤ e 4 r(2|x|− 4 r) = e(|x|−r) −(|x|− 54 r )
7 1 9
≤ 81+ p−q .
npq
1+ p−q
Consequently, |x|r < 32 7 log(8 ) = Cp,q,n . Hence, r < Cp,q,n m(x). Thus
B(x, r) ∈ BCp,q,n , which completes the proof of Lemma 265.
We can generalize the above setting. Let (X, d, µ) be a metric measure
space. A function m : X → (0, ∞) is called admissible if for any given τ ∈
(0, ∞), there exists a constant Θτ ∈ [1, ∞) such that for all x, y ∈ X satisfying
d(x, y) ≤ τ m(x), it holds that

Θτ −1 m(y) ≤ m(x) ≤ Θτ m(y). (6.22)

272 Morrey Spaces

For any a > 0, denote by Ba the set of all balls B ⊂ X such that r(B) ≤
am(c(B)). Balls in Ba are referred to as admissible balls with scale a. If r(B) =
am(c(B)), then B is called maximal.
Definition 68. The triple (X, d, µ) is called a locally doubling measure metric
space if, for every a > 0, there exist constants Da ∈ (1, ∞) and Ra ∈ (1, ∞)
such that for all B ∈ Ba ,

µ(2B) ≤ Da µ(B) (6.23)

and
µ(2B) ≥ Ra µ(B). (6.24)
Conditions (6.23) and (6.24) are respectively referred to as the locally doubling
condition and the locally reverse doubling condition.
Example 108. A typical but non-trivial example of the locally doubling
measure metric spaces is the Gauss measure space (Rn , | · |, γ), where
Z
2
−n
γ(E) ≡ π 2 e−|x| dx.
E

For all x ∈ Rn , let m(x) ≡ min{1, |x|−1 }. As above, for any a > 0, we say
that B ∈ Ba if and only if r(B) ≤ am(c(B)). If B ∈ Ba and x ∈ B, then

(a + 1)−1 m(x) ≤ m(c(B)) ≤ (a + 1)m(x), (6.25)

which implies that m is an admissible function in the sense of (6.22). For all
B ∈ Ba , we have 2B ∈ B2a . Proposition 260 further implies that
Z
n 2
γ(2B) = π − 2 e−|x| dx
2B
n 2
≤ π − 2 e−|c(B)| +4a |2B|
Z
2 n 2
≤ 2n ea +6a π − 2 e−|x| dx
B
n a2 +6a
≤ 2 e γ(B),

whenever x ∈ B ∈ Ba . This shows that γ satisfies (6.23). Moreover, Proposi-

tion 266 below also shows that γ satisfies (6.24). Therefore, (Rn , | · |, γ) is a
locally doubling measure metric space.
Proposition 266. In (Rn , | · |, γ), for any a > 0, there exists Ca > 0 such
that γ(2B) ≥ (1 + Ca )γ(B) for all balls B ∈ Ba .
Proof Suppose that the ball B is centered at c(B) and with radius r(B).
We prove the conclusion by considering two cases: |c(B)| ≤ 2r(B) and |c(B)| >
2r(B).
 General metric measure spaces 273

First, we consider the case |c(B)| ≤ 2r(B). Since B ∈ Ba , we have r(B) ≤

am(c(B)) ≤ a. For all z ∈ 2B \ B, by |c(B)| ≤ 2r(B), we conclude that
|z| ≤ |z − c(B)| + |c(B)| < 4r(B) ≤ 4a, which in turn implies that
2
γ(2B) e−16a |2B \ B| 2
≥1+ = 1 + e−16a (2n − 1). (6.26)
γ(B) |B|
Thus, we obtain the desired result in this case.
We now turn to the opposite case |c(B)| > 2r(B). Notice that in this
case 2B does not contain the origin. Consider the ball with the center w ≡
|c(B)| − 2 r(B) |c(B)| and radius r(B)
3

c(B)
2 . A geometric observation implies that
B w, 2 ⊂ 2B \ B. Moreover, for any z ∈ B w, r(B)
r(B)


2 , we have

r(B) 3
|z| ≤ |z − w| + |w| < + |c(B)| − r(B) = |c(B)| − r(B),
2 2
which implies that
    
r(B) n 2 r(B)
γ B w, ≥ π − 2 e−(|c(B)|−r(B)) B w, .
2 2

Meanwhile, for all z ∈ B, we have |z| ≥ |c(B)| − |c(B) − z| > |c(B)| − r(B)
n 2
and γ(B) ≤ π − 2 e−(|c(B)|−r(B)) |B|. Thus

γ(2B) γ(2B \ B) γ(B(w, r(B)

2 )) |B(w, r(B)
2 )| 1
−1= ≥ ≥ = n. (6.27)
γ(B) γ(B) γ(B) |B| 2
Thus, we obtain the desired result in this case.
We then see from (6.26) and (6.27) that γ(2B) ≥ (1 + 2−n )γ(B) for all
B ∈ Ba . This finishes the proof of Proposition 266.

6.1.6 Exercises
Exercise 84. Prove (6.1) by integration by parts.
Exercise 85. Let a > 0. Use |x − c(B)| ≤ a min(1, |x|−1 ) for all x ∈ B such
that B is a ball in Ba to show −a2 − 2a ≤ |c(B)|2 − |x|2 ≤ 2a.
Exercise 86. Let a > 0 and define m(x) ≡ min(1, |x|−1 ). Write b ≡ a
a+1 .

(1) Prove that γ given by (6.17) is a probability measure.

(2) Let B be a maximal ball with c(B) ∈ B(1 + a).
(a) Show that r(B) ≥ b.
n 2
(b) Show that γ(B) ≥ ωn π − 2 e−2a −4a−1 n
b .
(c) Show that γ(B) ∼ |B| exp(−|c(B)|2 ) with the implicit constant
dependent on a.
274 Morrey Spaces

Exercise 87.

(1) Prove (6.11) using Theorem 254.

(2) Prove (6.12). Hint: We may suppose supp(f ) ⊂ Q ⊂ D(µ). Then find
χRn \Q M D(µ) f .
Exercise 88. Prove Theorem 251 using Theorems 247 and 250.

Exercise 89. Let f ∈ L0 (µ), where µ is a Radon measure on Rn . Then show

that |f (x)| ≤ k n Mk,uc f (x) for almost every x ∈ Rn by considering k-doubling
cubes.

6.2 Maximal operators on metric measure spaces with

general Radon measures
Keeping in mind what we did in Section 6.1, we work on a metric measure
space (X, d, µ). A triple (X, d, µ) is a metric measure space if (X, d) is a metric
space and µ is a Borel measure, where Borel sets in (X, d) are defined as the
σ-algebra generated by open sets. Then µ is doubling, if there exists a constant
C > 0 such that µ(B(x, 2r)) ≤ Cµ(B(x, r)) for all x ∈ X and r > 0. When we
work on a metric measure space, we need to generalize the definition of M . A
natural candidate of this definition is as follows:
Z
1
M 0 f (x) ≡ sup |f (y)|dµ(y) (x ∈ X)
r>0 µ(B(x, r)) B(x,r)

and
χB(z,r) (x)
Z
M f (x) ≡ sup |f (y)|dµ(y) (x ∈ X).
z∈X,r>0 µ(B(z, r)) B(z,r)

The former is the centered maximal operator, while the latter is the uncentered
maximal operator. Section 6.2.1 considers the boundedness of the maximal
operator. What differs from Section 6.1.2 is the lack of the strong covering
lemma. We have to use the 5r-covering lemma instead. Sections 6.2.1 and
6.2.2 parallel Section 6.1.2. We collect some estimates on the modified maximal
operators. In Section 6.2.3 with an example we explain that the modification
considered in Section 6.2 is actually necessary.

6.2.1 Weak-(1, 1) estimate and strong-(p, p) estimate

Section 6.2.1, considers the modified maximal operator in the separable
metric space (X, d, µ) such that all balls are non-degenerate, in the sense that
 General metric measure spaces 275

µ(B(x, r)) > 0 for all x ∈ X and r > 0. Define the centered modified maximal
operator
Z
1
Mk,c f (x) = Mk,c,µ f (x) ≡ sup |f (y)|dµ(y)
r>0 µ(B(x, kr)) B(x,r)

and the uncentered modified maximal operator

χB(y,r) (x)
Z
Mk,uc f (x) = Mk,uc,µ f (x) ≡ sup |f (z)|dµ(z).
y∈X,r>0 µ(B(y, kr)) B(y,r)

We will see how large k must be in order that we have the bounded-
ness of Mk,c . Section 6.2.1 investigates the property of the centered Hardy–
Littlewood maximal operator. We will omit the analysis of the uncentered
Hardy–Littlewood maximal operator. In general metric measure spaces, the
following theorem is fundamental.
Theorem 267. Let (X, d) be a separable metric space. Let µ be a Radon
measure such that µ(B(x, r)) < ∞. Then M2,c is weak-(1, 1) bounded and the
weak-(1, 1) constant is less than or equal to 1:
Z
1
µ{x ∈ X : M2,c f (x) > λ} ≤ |f (x)|dµ(x)
λ X

for all f ∈ L0 (µ).

Since M2,c ≥ M3,uc , we have a similar assertion for M3,uc . Sometimes one
abbreviates {x ∈ X : M2,c f (x) > λ} to {M2,c f > λ}. Since the proof of
Theorem 267 uses a delicate covering lemma, we postpone its proof.
Recall that the Radon property guarantees that any measurable set can
be approximated from inside by compact sets.
Remark 12. Let k > 0. By the Radon property, we can replace B(x, r) by
B(x, r) in the definition of Mk,c . Thus, for f ∈ L0 (µ) we have Mk+δ f (x) ↑
Mk,c f (x) as δ ↓ 0.
The next covering lemma refines Vitali’s covering lemma, which leads us
to obtain the weak-(1, 1) boundedness of M2,c .
To prove Theorem 267, we need to refine the “so called” 5r-covering lemma.
Lemma 268. Suppose that we have a finite family {B(xj , rj )}j∈{1,2,...,N } of
balls. Then for δ > 0, we can take a subfamily {B(xj , rj )}j∈A such that
(1) {B(xj , rj )}j∈A is disjoint,
N
S S
(2) B(xj , δrj ) ⊂ B(xj , (2 + δ)rj ).
j=1 j∈A
276 Morrey Spaces

This extends Vitali’s covering lemma: The lemma is precisely Vitali’s cov-
ering lemma if δ = 1.
Proof We select j1 ∈ {1, 2, . . . , N } so that rj1 = max{r1 , r2 , . . . , rN }. If
[
B(xj , δrj ) ⊂ B(xj1 , (2 + δ)rj1 ),
j∈{1,2,...,N }

we have nothing else to do. Let us assume otherwise in the sequel. We define

Λ1 ≡ {j ∈ {1, 2, . . . , N } : B(xj , δrj ) ⊂ B(xj1 , (2 + δ)rj1 )}.

We inductively define the subsets of {1, 2, . . . , N } and j1 , j2 , . . . , jp ∈

{1, 2, . . . , N } as follows:
Let q ≥ 2. Suppose that j1 , j2 , . . . , jq−1 and the subsets Λ1 , Λ2 , . . . , Λq−1 ⊂
{1, 2, . . . , N } are defined. Then we define
 
 jq−1 
[
Λq ≡ j ∈ {1, 2, . . . , N } : B(xj , δrj ) ∩ B(xj , (2 + δ)rj ) 6= B(xj , δrj ) .
 
k=j1

Next, we take jq ∈ Λq so that rjq = max rj and This procedure will terminate
j∈Λq
because we are dealing with the finite number of the balls. Suppose we have
stopped in the p-th step after we selected jp and Λp . We will verify that
A ≡ {j1 , j2 , . . . , jp } satisfies all requirements of the lemma.
To verify this we fix j ∈ {1, 2, . . . , N }. We have three possibilities.
(a) j ∈ {j1 , j2 , . . . , jp }.
(b) rj = rjp and j ∈
/ {j1 , j2 , . . . , jp }.
(c) rjk ≥ rj > rjk+1 for some k ∈ {1, 2, . . . , p − 1} and j ∈
/ {j1 , j2 , . . . , jp }.
S
We want to show that B(xj , δrj ) ⊂ B(xj , (2 + δ)rj ); once this is achieved,
j∈A
we have (2). If (a) happens, this inclusion is clear from the definition of
j1 , j2 , . . . , jp . Assume (c) in the sequel. Another possibility (b) can be dealt
with in a similar manner. A geometric observation S together with (c) shows
that j ∈ / Λk and hence B(xj , δrj ) ⊂ B(xk , (2 + δ)rk ) by the
k∈{j1 ,j2 ,...,jk }
definition of Λk and rjk+1 . Thus, our claim is justified.
It remains to show that the balls {B(xj , rj )}j∈A are disjoint. Indeed sup-
pose k < k 0 , so that we have

rjk ≥ rjk0 . (6.1)

By the definition of Λj , we have B(xjk0 , δrjk0 ) is not contained in B(xjk , (2 +

δ)rjk ), since jk < jk0 . Thus xjk0 ∈
/ B(xjk , 2rjk ). This implies

d(xjk , xjk0 ) ≥ 2rjk . (6.2)

 General metric measure spaces 277

Combining (6.1) and (6.2), we conclude that {B(xj , rj )}j∈A is disjoint. So,
(1) is verified.
We now prove Theorem 267.
Proof Fix λ > 0. We define Ek ≡ {x ∈ X : Mk,c f (x) > λ} for k > 2. By
Remark 12, it follows that
[
{x ∈ X : Mk,c f (x) > λ} = {x ∈ X : M2,c f (x) > λ}. (6.3)
k>2

and k ≡ 2 + δ. For all x ∈ Ek , by its definition there exists rx > 0

Let δ > 0 Z
such that |f (y)|dµ(y) > λµ(B(x, krx )). Since µ is a Radon measure,
B(x,rx )
Ek is an open set. Since X is separable, with the aid of the Linderöf covering
theorem, asserting that any covering of X has its countable subcovering as
long as X is separable, we can take xj ∈ Ek , j = 1, 2, . . . such that Ek ⊂
[∞
B(xj , δrxj ).
j=1
We fix N ∈ N. By Lemma 268 there exists a subset A ⊂ {1, 2, . . . , N } such
that
[N [
B(xj , δrxj ) ⊂ B(xl , (2 + δ)rxl ) (6.4)
j=1 l∈A

and that X
χB(xl ,rxl ) ≤ 1. (6.5)
l∈A

By the definition of Ek , we also have

Z
1
µ(B(xl , (2 + δ)rxl )) ≤ |f (x)|dµ(x) (6.6)
λ B(xl ,rxl )

for l ∈ A.
Combining (6.4)–(6.6) gives
  !
N
[ [
µ B(xj , δrxj ) ≤ µ B(xl , (2 + δ)rxl )
j=1 l∈A
X
≤ µ(B(xl , (2 + δ)rxl ))
l∈A
X1Z
≤ |f (x)|dµ(x)
λ B(xl ,rx )
l∈A l

1
≤ kf kL1 (µ) . (6.7)
λ
Letting N tend to infinity in (6.7), we derive λµ(Ek ) ≤ kf kL1 (µ) . Tending
k ↓ 2, we conclude the proof of Theorem 267.
278 Morrey Spaces

If we interpolate between Theorem 267 and an estimate kM2,c f kL∞ (µ) ≤

kf kL∞ (µ) , we obtain the following:
Theorem 269. Let p > 1 and (X, d, µ) be a separable metric measure space.
1
Then kM2,c f kLp (µ) ≤ 2 (p0 ) p kf kLp (µ) for all f ∈ L0 (µ).
Proof The technique of the proof is known as the interpolation. The proof
is similar to Theorem 140. See Exercise 91.

6.2.2 Vector-valued boundedness of the Hardy–Littlewood

maximal operators
As an application of Lemma 268, we will prove the following vector-valued
inequality for the Hardy–Littlewood maximal operator. It turns out that k ≥
22 suffices as the following theorem shows:
Theorem 270. If 1 < p < ∞ and 1 < q ≤ ∞, then
k{M22,c fj }∞
j=1 k`q Lp (µ)
.p,q k{fj }∞
j=1 k`q Lp (µ)
. (6.8)

When q = ∞, (6.8) reads;

sup M22,c fj .p sup |fj | . (6.9)

j∈N Lp (µ) j∈N Lp (µ)

We leave the proof of Theorem 270 to interested readers, since it is similar

to the classical case; see Exercise 93. If p > q, then we can say a little more;
see Theorem 274.
First of all, to prove Theorem 270, we obtain a weighted inequality of
Stein-type by using again Lemma 268.
Proposition 271 (Stein’s weak dual inequality). For all f, g ∈ L0 (µ) and
λ > 0, Z Z
1
|g(x)|dµ(x) ≤ |f (x)|M2,c g(x)dµ(x).
{M7,c f >λ} λ X
Proof Fix k > 7 keeping in mind Remark 12. Let E = {x ∈ X : Mk,c f >
λ}. By the definition of E, for all x ∈ E, there exists rx > 0 such that
Z
1
|f (x)|dµ(x) > λ.
µ(B(x, krx )) B(x,rx )

We will take a ≡ (k − 7)/2. Since E is an open set, again by the Linderöf

∞
S
covering theorem, there exist xj , j = 1, 2, . . . such that E ⊂ B(xj , arxj ).
j=1
Fix N ∈ N arbitrarily. We claim that
Z Z
1
|g(x)|dµ(x) ≤ |f (x)|M2,c g(x)dµ(x). (6.10)
S
B(xj ,arxj ) λ X
j=1,2,...,N
 General metric measure spaces 279

By Lemma 268 there exists a subfamily of balls {B(xj , arxj )}j∈J with J ⊂
{1, 2, . . . , N } that satisfies the following properties.
X
χB(xj ,rxj ) ≤ 1, (6.11)
j∈J

N
[ [
B(xj , arxj ) ⊂ B(xj , (2 + a)rxj ). (6.12)
j=1 j∈J

A geometric observation shows, for all x ∈ B(xj , rxj ),

B(xj , krxj ) ⊂ B(x, 3(2 + a)rxj ).

In fact, for all y ∈ B(x, (2 + a)rxj ),

d(xj , y) ≤ d(x, xj ) + d(y, x) ≤ rxj + 3(2 + a)rxj = (7 + 2a)rxj = krxj .

Consequently, for all x ∈ B(xj , rxj ),

Z
1
|g(y)|dµ(y)
µ(B(xj , krxj )) B(xj ,(2+a)rxj )
Z
1
≤ |g(y)|dµ(y)
µ(B(x, (6 + 2a)rxj )) B(x,(3+a)rxj )
≤ M2,c g(x). (6.13)

Using (6.11)–(6.13), we obtain

Z
S
|g(x)|dµ(x)
B(xj ,arxj )
j=1,2,...,N
Z XZ
≤ S
|g(x)|dµ(x) ≤ |g(x)|dµ(x)
B(xj ,(2+a)rxj ) j∈J B(xj ,(2+a)rxj )
j∈J
Z Z
1X 1
≤ |f (x)|dµ(x) · |g(x)|dµ(x)
λ µ(B(xj , krxj )) B(xj ,(2+a)rxj )
j∈J B(xj ,rxj )
Z Z
1X 1
≤ |f (x)|M2,c g(x)dµ(x) ≤ |f (x)|M2,c g(x)dµ(x).
λ B(xj ,rxj ) λ X
j∈J

Thus, (6.10) is proven. By the definition of E we have

Z Z
1
|g(x)|dµ(x) ≤ |f (x)|M2,c g(x)dµ(x).
{Mk,c f >λ} λ X

Letting k ↓ 7, we finally obtain Proposition 271. Thus, we have finished.

By interpolation, we obtain the strong inequality.
280 Morrey Spaces
0
Corollary
Z 272 (Stein’s strong dualZ inequality). If p > 1 and f ∈ L (µ), then
(M7,c f (x))p |g(x)|dµ(x) ≤ 2p p0 |f (x)|p M2,c g(x)dµ(x).
X X

Proof For a positive function w, k · kL∞ (wdµ) denotes the L∞ (µ)-norm of

functions with respect to the weighted measure wdµ. Since

kM7,c f kL∞ (|g|dµ) ≤ kf kL∞ (M2,c gdµ) (6.14)

is clear (see Exercise 90(2)), this is again just a matter of interpolation of

(6.14) and Proposition 271.
We use the following analogy, which is used below to obtain Theorem 270.
The proof is only the adaptation of the parameter k of Proposition 271 and
Corollary 272 respectively; we omit its proof.
Proposition 273. Let f, g ∈ L0 (µ).
Z Z
1
(a) The estimate |g(x)|dµ(x) ≤ |f (x)|M7,c g(x)dµ(x) holds
{M22,c f >λ} λ X
for all λ > 0.
Z Z
(b) If p > 1, then (M22,c f (x))p |g(x)|dµ(x) .p |f (x)|p M7,c g(x)dµ(x).
X X

At last we are in the position of proving Theorem 270. If p ≥ q > 1, a little

more can be said. We have the following:
Theorem 274. If p ≥ q > 1, then for all sequences {fj }∞ 0
j=1 ⊂ L (µ),

k{M7,c fj }∞
j=1 k`q Lp (µ)
.p,q k{fj }∞
j=1 k`q Lp (µ)
. (6.15)

Proof The proof is a reexamination of the proof of Theorem 145. See

Exercise 92.

6.2.3 Examples of metric measure spaces which require

modification
So far, k needs to be greater than or equal to 2 in order that Mk,c is
bounded. We will show by an example that the condition k ≥ 2 is optimal for
the centered modified maximal operator.
Example 109. Let l ∈ N and j ∈ {1, 2, . . . , l}. Consider the graph X with
vertices xl and xlj and edges joining xl with xl+1 and xl with each xlj . Let
µ be a counting measure on X; µ(E) = ]E. Consider (X, d, µ) as a metric
measure space. We will disprove that Mk,c is weak-(1, 1) bounded for any
0 < k < 2.
Fix l ∈ N ∩ [2, ∞) and j ∈ {1, 2, . . . , l}. Then B(xlj , kr) = {xl , xlj } for
r ∈ (1, 2 · k −1 ). Note that δa , the Dirac delta massed at any a ∈ X, is a
 General metric measure spaces 281
1
function defined in X. We have Mk,c δxl (xlj ) = for any j = 1, 2, . . . , l. We
2
1
also have Mk,c δxl (xl±1 ) = and Mk,c δxl (xl ) = 1. For any other point z ∈ X,
2
Mk,c δxl (z) = 0. Consequently, {x ∈ X : 2Mk,c δxl (x) ≥ 1} is made up of l + 3
points. Since the µ-measure of this set is l + 3 and δxl has the L1 (µ)-norm 1,
we see that Mk,c fails to be weak-(1, 1) bounded. Likewise, we see that Mk,c
fails to be strong-(p, p) bounded for any 1 < p < ∞.
We will show by an example that the condition k ≥ 3 is optimal for the
uncentered modified maximal operator.
Example 110. Let positive integers l and j move over the set {(l, j) ∈ N2 :
j ≤ l}. Consider the graph X with vertices xl , xlj and ylj and edges joining
xl with xl+1 , ylj with xlj and xl with xlj . Let µ0 be a counting measure on
X as before. Define
∞ ∞
X X 1
w(x) ≡ χ{xl ,yl1 ,yl2 ,...,yll } + χ{xl1 ,xl2 ,...,xll } .
l
l=1 l=1

Set µ ≡ wµ0 , the weighted measure. Consider (X, d, µ) as a metric measure

space. We will disprove that Mk,uc is weak-(1, 1) bounded for any 0 < k < 3.
If r ∈ (1, 3 · k −1 ), l ∈ N ∩ [3, ∞) and j = 1, 2, . . . , l, then

B(xlj , kr) = {xl , xl−1 , xl+1 , ylj } ∪ {xlm }lm=1 .

1
Thus, M δxl (ylm ) ≥ for any m = 1, 2, . . . , l. Thus, as before Mk,uc fails to
5
be weak-(1, 1) bounded and strong-(1, 1) bounded for any 1 < p < ∞.
We will present another example of compact metric measure spaces. First,
we define a set X on which we work.
Definition 69 (Definition of X). Define a set X as follows:
(1) One writes ∆(z, r) ≡ {w ∈ C : |w − z| < r}.
(2) Let Ak ≡ {z ∈ C : |z| = 3−k } for k ∈ N0 .
∞
[
(2) Define X0 ≡ {0} ∪ Ak ⊂ C.
k=0

(3) Let X ≡ X0 N ⊂ CN be the cross product.

(4) Let O ≡ (0, 0, . . .) ∈ X.
We draw graphs of A0 and X0 .
Note that A0 is an annulus and X0 is the union and X is a countable product
of X0 .
Here and below, until the end of this section, we adopt the following rules:
282 Morrey Spaces

(1) The letter z without subindex denotes the point in C.

(2) Points in X are written in bold letters such as x, y, z.
(3) Symbols such as xj , yj , zj and so on are complex numbers and they
denote the j-th component of elements in X.
The point is that we give a “singular” metric on X. The precise definition
is as follows:
(1) The integer
N0 (6.16)
is chosen so that log3 N0 is a big integer.
(2) Denote by [·] a Gauss symbol and define N (δ) ≡ max(1, [logN0 δ −1 ]) for
δ > 0.
(3) Let x = {xj }∞ ∞
j=1 and y = {yj }j=1 be points in X, and define the distance
d(x, y) of x and y by

d(x, y) ≡ inf{δ > 0 : |xj − yj | ≤ δ for all j ≤ N (δ)}.

(4) One defines a sphere Sk by

−k
Sk ≡ (Ak )N (3 )
× X0 × X0 × · · · (6.17)

for each k ∈ N.
At this moment, about the definition of the natural number N0 , we keep in
mind that N ∈ M↓ (0, ∞).
Note that

d(x, y) = inf{δ > 0 : |x1 − y1 | ≤ δ, |x2 − y2 | ≤ δ, . . . , |xN (δ) − yN (δ) | ≤ δ}

and that

Sk = {z = (z1 , z2 , . . .) ∈ X0 N : |z1 | = |z2 | = · · · = |zN (3−k ) | = 3−k }.

Thus, diam(X) ≡ sup d(x, y) = 2. Let us check that d is a metric function

x,y∈X
and that χSk ∈ L0 (µ).
Lemma 275. The function d is a metric function. That is,

0 ≤ d(x, y) < ∞ (x, y ∈ X), (6.18)

d(x, y) = 0 =⇒ x = y (x, y ∈ X), (6.19)
d(x, y) = d(y, x) (x, y ∈ X), (6.20)
d(x, z) ≤ d(x, y) + d(y, z) (x, y, z ∈ X). (6.21)

Furthermore, the d-topology is exactly the product topology of X.

 General metric measure spaces 283

Proof Since 2 ∈ {δ > 0 : |xj − yj | ≤ δ for all j ≤ N (δ)}, (6.18) is clear.

If x 6= y, then |xj0 − yj0 | > δ for some δ > 0 and j0 ∈ N. Therefore, if we
choose δ ∗ > 0 so that N (δ ∗ ) > j0 , then we have |xj0 − yj0 | > min(δ, δ ∗ ) and
j0 < N (min(δ, δ ∗ )). This implies
min(δ, δ ∗ ) ∈
/ {r > 0 : |x1 − y1 | ≤ r, |x2 − y2 | ≤ r, . . . , |xN (r) − yN (r) | ≤ r}.
Hence, d(x, y) ≥ min(δ, δ ∗ ), which shows (6.19). Equality (6.20) follows
immediately from the definition of d. Next, we check (6.21). To this end,
we take ε > 0. Then by the definition of d(x, y) and d(y, z), we can find
δ1 ∈ (d(x, y), d(x, y) + ε) and δ2 ∈ (d(y, z), d(y, z) + ε) so that |xj − yj | ≤ δ1
for all j ≤ N (δ1 ) and that |yj − zj | ≤ δ2 for all j ≤ N (δ2 ). Note that
N (δ1 + δ2 ) ≤ min(N (δ1 ), N (δ2 )), we have |xj − yj | ≤ δ1 , |yj − zj | ≤ δ2 for all
j ≤ N (δ1 + δ2 ). Hence |xj − zj | ≤ δ1 + δ2 for all j ≤ N (δ1 + δ2 ). Consequently,
d(x, z) ≤ δ1 + δ2 ≤ d(x, y) + d(y, z) + 2ε. Since ε > 0 is arbitrary, we obtain
(6.21).
Finally, we investigate the relation between two topologies. Since any d-
open set is open with respect to the product topology, the product topology
is not weaker than the d-topology. However, we can express ∆ as a union of
d-open balls as long as X is given by
X = ∆(z1 , r1 ) × ∆(z2 , r2 ) × · · · × ∆(zÑ , rÑ ) × X0 × X0 × · · ·

with some Ñ ∈ N, (z1 , z2 , . . . , zÑ ) ∈ X0 Ñ and (r1 , r2 , . . . , rÑ ) ∈ (0, ∞)Ñ .

Therefore, two topologies coincide.
Lemma 276. Let r > 0 and x = (x1 , x2 , . . .) ∈ X. Then
B(x, r) = ∆(x1 , r) × ∆(x2 , r) × · · · × ∆(xN (r) , r) × X0 × X0 × · · · .
NQ
(r)
In particular, µ(B(x, r)) = µ0 (∆(xj , r)).
j=1

Proof From the definition of the open ball, we have

B(x, r) = {y = {yj }∞ ∞ ∞
j=1 ∈ X : d({yj }j=1 , {xj }j=1 ) < r}
N (δ)
[ \
= {y = {yj }∞
j=1 ∈ X : |yj − xj | ≤ δ}.
δ∈(0,r) j=1

Since N : (0, ∞) → R is left-continuous and assumes its value in Z, if δ is

slightly less than r, N (δ) = N (r). Together with the monotonicity of the
most-right hand side of the above equality, we conclude
B(x, r)
= {y = {yj }∞
j=1 ∈ X : |y1 − x1 | < r, |y2 − x2 | < r, . . . , |yN (r) − xN (r) | < r}.

Consequently, Lemma 276 was proven.

The measure is given by way of the product: Let H1 denote the 1-
dimensional Hausdorff measure.
284 Morrey Spaces
∞
X 1
(1) One defines a function w0 : X0 → [0, ∞) by w0 ≡ γ χA , where
(k!)k k
k=0
w0 (z) dH1 (z) = 1.
R
γ is chosen so that
X0

(2) Define a measure on X0 by µ0 ≡ w0 dH1 .

(3) Define a measure µ on X by µ ≡ µ0 × µ0 × · · · = w0 dH1 × w0 dH1 × · · · .
As for the measures µ and µ0 , we prove the following estimate:
!
∞
S
Lemma 277. For k ∈ N0 , µ0 (Ak ) and µ0 Aj are comparable in the
j=k
following sense:
 
∞  
2πγ [ 2πγ 1
= µ0 (Ak ) ≤ µ0  Aj  ≤ 1+ .
(3 · k!)k (3 · k!)k γ · (k + 1)!
j=k

Proof From the definition of µ0 , we see

Z Z
γ 2πγ
µ0 (Ak ) = w0 (z) dH1 (z) = k
dH1 (z) = .
Ak (k!) |z|=3−k (3 · k!)k

Hence  
∞ ∞ ∞
[ X 2πγ 2πγ X (3 · k!)k
µ0  Aj  = = . (6.22)
(3 · j!)j (3 · k!)k (3 · j!)j
j=k j=k j=k
!
∞
S
This observation yields the lower bound for µ0 Aj . It remains to obtain
j=k
!
∞
S
the upper bound for µ0 Aj .
j=k
First of all, let us assume that k = 0. Note that µ0 (A0 ) = 2πγ. Hence,
when k = 0, from (6.22), we deduce
   
∞ ∞  
[ [ 1
µ0  Aj  = µ0  Aj  = 1 ≤ 2πγ 1 + .
j=0
γ
j=k

Note that µ0 (A1 ) = 2πγ

3 . Hence, when k = 1, again from (6.22), we deduce
   
∞ ∞  
[ [ 2πγ 1
µ0  Aj ≤ µ0
  Aj = 1 ≤
 1+ .
j=0
3 2γ
j=k
 General metric measure spaces 285

Let k ∈ N ∩ [2, ∞) below. Using (6.22), we calculate that

   
∞ ∞ k
[ 2πγ X (3 · k!)
µ0  Aj  ≤ 1 + .
(3 · k!)k (3 · j!)k (3 · j!)
j=k j=k+1

If j ≥ k + 1, then k! × j ≤ j!. Thus, we have

   
∞ ∞
[ 2πγ X 1
µ0  Aj  ≤ 1 + .
(3 · k!)k 3 · j k (k + 1)!
j=k j=k+1

Note that γ ∈ (0, 1) and that

∞ ∞
X 1 X 1 π2
≤ = −1<1
jk j=2
j2 6
j=k+1

for all k ∈ N ∩ [2, ∞). Thus, we obtain

 
∞    
[ 2πγ 1 2πγ 1
µ0  Aj  ≤ 1 + ≤ 1 + .
(3 · k!)k (k + 1)! (3 · k!)k γ · (k + 1)!
j=k

Thus, Lemma 277 was proven.

The next lemma is an easy consequence
of a simple geometic observation
and the inequality sin−1 r ≥ r for r ∈ 0, π2 . Let H1 (C) be the arc length of
the arc C.
Lemma 278. Let R > 3 and (x0 , y0 ) ∈ R2 . In R2 , suppose that two circles
x2 + y 2 = 1 and (x − x0 )2 + (y − y0 )2 = R2 intersect. Then the arclength L of
the arc, given by {(x, y) : x2 + y 2 = 9, (x − x0 )2 + (y − y0 )2 ≤ R2 }, is greater
than or equal to
2
6 cos−1 .
3
Proof A rotation allows us to assume y0 = 0 < x0 . Two circles x2 +y 2 = 1
and (x − x0 )2 + y 2 = R2 intersect if and only if R − 1 < x0 < 1 + R. Set
Θ(x0 , R) ≡ |{θ ∈ [−π, π] : (3 cos θ − x0 )2 + (3 sin θ)2 ≤ R2 }|
 2
x0 + 9 − R 2

= 2 cos−1 .
6x0
Then, L is given by
L = 3Θ(x0 , R). (6.23)
Since x0 < 1 + R and R > 3, we have
   
−1 10 + 2R −1 2
Θ(x0 , R) ≥ 2 cos ≥ 2 cos . (6.24)
6(1 + R) 3
Estimates (6.23) and (6.24) yield the desired lower bound of L.
The weak (1, 1)-constant in our general setting is sharp.
286 Morrey Spaces

Proposition 279. There exist a separable space (X, d) and a measure µ such
that Mk,c is bounded if and only if k ≥ 2. Furthermore, the weak-(1, 1) constant
of M2,c is 1 on this space.
Proof We may assume 32 < k < 2, since Mk,c is decreasing as k increases.
Suppose that Mk,c is bounded. We want to derive a contradiction. We begin
1B(0,r)
with constructing an approximation of Dirac delta. Let gr ≡ . Then
µ(B(0, r))
Mk,c gr tends pointwise to Mk,c δ0 as r ↓ 0, where δ0 is a point mass at 0. Let
a be a point defined as follows:
Put Kn ≡ 2 + [log10 3n ]. Define a = an = {3−n χ[1,Kn ] (j)}j∈N ∈ D.
1
Take λ ≡ . We have
µ(B(a, k · 3−n ))

{x ∈ X : Mk,c δ0 (x) > λ} ⊃ B(0, 3−n ).

Thus, we have µ(B(0, 3−n )) . λ−1 . That is, µ(B(0, 3−n )) . µ(B(a, k · 3−n )).
By the definition of ν, there are infinitely many integers n such that
√ (2+[log10 3n ])
ν({x + −1y ∈ D : x2 + y 2 ≤ 9−n })

√ . 1.
ν({x + −1y ∈ D : (x − 3−n )2 + y 2 ≤ k 2 · 9−n })

Take the limit of this quantity as n, running through such an integer, to

infinity. We obtain a contradiction, since the constant is strictly less than 1.
Likewise we can calculate the weak-(1, 1) constant M2,c .

6.2.4 Exercises
Exercise 90. Let (X, d, µ) be a metric measure space.
(1) Show that for any ball B(x, r) and for any y ∈ B(x, r),
Z
1
|f (y)|dµ(y) ≤ M2,c f (x).
µ(B(x, 2r)) B(x,r)

(2) Show that for any ball B(x, r) and for any y ∈ B(x, r),
Z
1
|f (y)|dµ(y) ≤ kf kL∞ (µ) .
µ(B(x, 2r)) B(x,r)

Exercise 91. Let 1 < p < ∞ and f ∈ L0 (µ).

(1) Write out the definition in full using Theorem 5 to have
Z ∞
kM2,c f kLp (µ) p = pλp−1 µ{M2,c f > λ}dλ. (6.25)
0
 General metric measure spaces 287

(2) Reexamine the proof of Theorem 267 to give

Z
1
µ{M2,c f > λ} ≤ χ(λ,∞] (M f (y))|f (y)|dµ(y). (6.26)
λ X
(3) Combine (6.25) and (6.26) and reexamine the proof of Theorem 141 to
have kM2,c f kLp (µ) ≤ p0 kf kLp (µ) .
Exercise 92. [378] Let p ≥ q > 1. Suppose that we have {fj }∞ 1
j=1 ⊂ L (µ) ∩
∞
L (µ) such that fj = 0 for any j  1.
(1) Let p = q. Use Theorem 267 to show Theorem 274.
p
(2) Let p > q. Abbreviate q to r and let r0 be a conjugate exponent of r.
 q ∞
k{M7,c fj }∞ (M7,c fj )q
P
(2 − A) Show that j=1 k`q Lp (µ)
= .
j=1
Lr (µ)
(2 − B) Dualize the right-hand side to find g ∈ M+ (µ) such that kgkLr0 (µ) =
1 such that
 
 q Z ∞
X
∞ q
k{M7,c fj }j=1 k`q Lp (µ) =  M7,c fj (x)  g(x)dµ(x).
X j=1
0
With this in mind, we fix g ∈ Lr (µ) ∩ M+ (µ).
(2 − C) Using Corollary 272, prove that
 q ∞
Z X
k{M7,c fj }∞
j=1 k` q
Lp (µ)
.p,q |fj (x)|q M2,c g(x)dµ(x).
X j=1

(2 − D) Use Hölder’s inequality twice to conclude the proof of Theorem

274.
Exercise 93. Let 1 < p < q ≤ ∞.
 
(1) If q = ∞, then use sup M22,c fj ≤ M22,c sup |fj | to conclude the proof
j∈N j∈N
of Theorem 270. Hence, (6.15) is clear, if q = ∞.
(2) It remains to show when p < q < ∞. Let q > p in what follows and take
pr
another r < q so close to q that s = > 1.
q
(3) Use the duality twice to obtain
  r1
Z ∞
X
 (M22,c fj (x))q  g(x)dµ(x) (6.27)
X j=1
 
Z ∞  
X q
=P sup  M22,c fj (x) hj (x) g(x) dµ(x).
r

hk r0 =1,hk ∈M+ (µ) X j=1

k
288 Morrey Spaces

Below we will write sup instead of sup and we abbreviate

hk r0 =1,hk ∈M+ (µ)
P
k

 
Z ∞  
X q
I ≡ sup  M22,c fj (x) r hj (x) g(x) dµ(x).
X j=1

(4) Use Fubini’s theorem to show

∞ Z  
X q
I = sup M22,c fj (x) r hj (x) g(x)dµ(x)
j=1 X

(5) Use Proposition 273 and the fact that q > r > 1 to have
∞ Z
X q
I .p,q sup |fj (x)| r M7,c [hj g](x)dµ(x)
j=1 X
 
Z ∞
X q
= sup  |fj (x)| r M7,c [hj g](x) dµ(x). (6.28)
X j=1

(6) Use Hölder’s inequality to obtain

 
Z ∞
X q
sup  |fj (x)| r M7,c [hj g](x) dµ(x)
X j=1
  r1   10
r
Z ∞ ∞
X X 0
q  r 
≤ sup  |fj (x)| (M7,c [hj g](x)) dµ(x)
X j=1 j=1

and
 
Z ∞
X q
sup  |fj (x)| r M7,c [hj g](x) dµ(x)
X j=1
  r1   10
r
Z ∞ ∞
X X 0
q  r 
≤ sup  |fj (x)| (M7,c [hj g](x)) dµ(x)
X j=1 j=1
   s00  s10
r

Z ∞ 

X 0
r
× sup  (M7,c [hj g](x))  dµ(x)
 X j=1
 

 General metric measure spaces 289

(7) Note that q > p implies s0 > r0 . Use Theorem 270 with parameters
s0 > r0 to obtain
   s00  s10
r

Z ∞ 

X 0
 (M7,c [hj g](x))r  dµ(x)
 X j=1
 

   s00  s10
r

Z ∞ 

X 0
.p,q  (hj (x)g(x))r  dµ(x) .
 X j=1
 


(8) Show that

   s00  s10
r

Z ∞ 

X 0
sup  (M7,c [hj g](x))r  dµ(x) .p,q 1. (6.29)
 X j=1
 


(9) Use (6.29) to complete the proof.

Exercise 94. [122, Lemma 2.8] The metric space (X, d) has the segment
property if for any x, y ∈ X there exists a continuous curve γ : [0, 1] → X
connecting x and y such that d(x, y) = d(x, γ(t)) + d(γ(t), y) for all t ∈
[0, 1]. Assume that (X, d, µ) is a doubling metric space satisfying the segment
property. Then show that the function r ∈ (0, ∞) 7→ µ(B(x, r)) ∈ (0, ∞) is
continuous for each x ∈ X.
Exercise 95. In Theorem 267, prove that we cannot improve the conclusion
in the following sense: Let 0 < ρ < 1. Then we cannot have
Z
ρ
µ({x ∈ X : M3,uc f (x) > λ}) ≤ |f (x)|dµ(x) (6.30)
λ X

for any f ∈ L0 (µ).

Exercise 96. In Theorem 267, prove that we cannot improve the conclusion
in the following sense: Let 0 < ρ < 1. Then there does not exist a constant
Cρ such that
Z
Cρ
µ({x ∈ X : M3−ρ,uc f (x) > λ}) ≤ |f (x)|dµ(x) (6.31)
λ X

for any f ∈ L0 (µ).

290 Morrey Spaces

6.3 Notes
Section 6.1
General remarks and textbooks in Section 6.1
See standard textbooks [112, Chapter 3], [231], [293] and [459].
See [147, §1.3] for various covering lemmas adapted to quasi-metrics.
Fu, Yang and Yuan considered the commutators in [130, Theorem 1.15].
Futamura, Mizuta and Shimomura consided the Sobolev embedding in the
setting of general metric measure spaces and variable exponents [138, Theo-
rem 4.1]. Garcı́a-Cuerva and Martell obtained the vector-valued estimates for
singular integral operators [143, Section 2] and the existence of the principal
values [144, Theorem 1(f)]. Gorka considered Morrey–Campanato spaces in
the setting of metric measure spaces [154].
Universal estimates for dyadic maximal operators can be found in [223, p.
11, Theorem]; see Theorem 254.

Section 6.1.1
The boundedness of the maximal operator Mκ,uc associated with general
Radon measure goes back to [339, Section 3]. See [338] as well. Later, Sawano
and Terasawa pointed out that the modification rate κ can be decreased [378,
433]. Theorem 247 is [378, Theorem 1.5].
See [68] for the commutators generated by RBMO(µ) and Iα , where Chen
and Sawyer considered a modification constant that differs from the one con-
sidered by Tolsa in [435].

Section 6.1.2
Tolsa pointed out that the modification rate κ in Mκ,uc must satisfy κ > 1
for the boundedness of Mκ,uc ; see [435, p. 127]. This idea was reinvestigated
by Sawano [378].

Section 6.1.3
As we saw in Theorem 253, Tolsa proved that the analogue of the Lebesgue
differentiation theorem is available [435, Remark 2.3]. The covering lemma and
the vector-valued inequality can be found in [378]; see [378, Theorem 1.5] and
[378, Theorem 1.7] for Theorems 247 and 251, respectively.

Section 6.1.4
D(µ)
We considered the universal Lp (µ)-estimate for Mlog defined in (6.13).
See [209, p. 786] for Theorem 256. Theorem 257(3) is recorded as [209, Lemma
 General metric measure spaces 291

2.13]. Lerner and Nazarov proposed to distort dyadic cubes as in Lemma 259;
see [92, Lemma 4.8].

Section 6.1.5
Mauceri and Meda considered the admissible balls in [308]. Proposition
260 is due to Mauceri and Meda [308, Proposition 2.1]. Lemma 262 can be
found in [308, p. 298].
Fabes, Gutiérrez and Scotto dealt with singular integral operators associ-
ated with the Gauss measure [119], where we can find a relation between the
Ornstein–Uhlenbeck process and singular integral operators they handle. See
also [174, 175, 183, 443] for more. See [145, Lemma 2.4] and [308, Lemma 5.1]
for more results related to Lemma 264. The corresponding version in the
locally doubling measure metric spaces was proven in [284, Lemma 2.7].
Lemma 265, which guarantess that a ball is doubling under a certain con-
dition, can be found in [281, Lemma 3.9]. Example 108 together with estimate
(6.25) is [278, Lemma 2.1].
Carbonaro, Mauceri and Meda considered the Hardy space H 1 and BMO
associated with the Gauss measure in [62, 63].
We can use the Gauss measure to show the boundedness of the Riesz
transform; see [358].

Section 6.2
General remarks and textbooks in Section 6.2
The textbooks [112, 147, 458] cover the topic exhaustively. A Radon mea-
sure µ satisfying µ(B(x, r)) . rd is called the Frostman measure. This plays
an important role in various aspects in analysis. See [9, 338, 339, 340, 434, 435,
436]. This type of assumption sometimes appears in the analysis of operators
handled in this book. See [112, Chapter 6] for fractional integral operators
on metric measure spaces with general Radon measures, which includes the
analysis of the maximal operator. See [112, Chapter 8, §3] for singular inte-
gral operators. See [233] for fractional integral operators on non-homogeneous
spaces. See [307] for analysis with general Radon measures.

Section 6.2.1
Terasawa proved Theorem 267 when the function (x, r) 7→ µ(B(x, r)),
x ∈ X and r > 0 is continuous in r [433, Theorem 2.4]. Sawano proved
Theorem 267 when X is separable [378, Theorem 2.1] based on Terasawa’s
result. See [378, Corollary 2.1] for Theorem 269.
See [234, 246] for more about the maximal operators on non-homogeneous
spaces.
292 Morrey Spaces

Section 6.2.2
In most cases, the vector-valued extension of the maximal operator associ-
ated with general Radon measures is done. For the modified maximal opera-
tor, see [378, Theorem 1.3] and [378, Theorem 2.1] for Theorems 270 and 274,
respectively. For the dyadic maximal operator associated with general Radon
measures, we refer to the textbook [89, Theorem A.18].

Section 6.2.3
Stempak constructed Examples 109 and 110; see [419, Example 2.1] and
[419, Example 2.2], respectively. See also [378, 418] for more examples showing
that the modification is necessary.
Chapter 7
Weighted Lebesgue spaces

Chapter 7 examines weighted norm inequalities. Weighted function spaces

arise naturally in many contexts of mathematics. Roughly speaking, weighted
measure spaces, which generate weighted function spaces, are useful although
they break some symmetries. We note that a group can be used to determine
the symmetry of a space. In fact, a group action allows the symmetry of
the spaces to be measured. For example, O(n), the group of all orthonormal
matrices, acts naturally on Rn , and Rn acts on itself as a translation. Hence,
we can say that Rn is symmetric enough.
However, when handling something in Rn , often the whole space Rn is
neglected. Occasionally, some open sets in Rn are used instead of Rn itself. For
example, when considering some differential equations, functions are defined
in open sets. Chapter 7 concentrates on weights that break the symmetry.
Another example of breaking the symmetry is listing the probability space.
Since the whole mass 1 cannot be distributed evenly, a density function of
measures or some discrete measures must be considered. The Gaussian dis-
tribution is a natural one appearing in many branches of sciences. This is
clearly not an invariant measure under the translations, although it is one
under O(n). Consequently, some symmetries must be surrendered.
One way to break the symmetry is to consider weighted measures. For the
time being by a weight we mean a measurable function in a measure space
which has a positive value almost everywhere.
As usual, here Euclidean space is considered. However, in different types
of spaces, the measure changes. For example, the Lebesgue measure dx will
be changed into the weighted measure w(x)dx, where w is a weight.
As shown in an earlier chapter, the Hardy–Littlewood maximal operator
can be used to control other important operators. Chapter 7 is interested in
the weights for which the Hardy–Littlewood maximal operator is bounded by
investigating the weighted estimates for Lebesgue spaces, including some sharp
estimates on constants. Weighted Morrey spaces are discussed in Chapter 15.
Weighted Lebesgue spaces arise naturally when considering the change
of variables. Since the Hardy–Littlewood maximal operator is a fundamental
operator, we are interested in the inequality
Z Z
M f (x)p w(x)dx . |f (x)|p w(x)dx.
Rn Rn

293
294 Morrey Spaces

We characterize weights w for which the above inequality holds. Lebesgue

spaces have a suitable characterization for this inequality. It matters that we
have to measure in some sense the property of weights to see that the weights
are close to the constant weight 1.
The Lebesgue measure of the measurable set E ⊂ Rn is denoted by |E|,
as before. For a measurable set E and f ∈ L1 (E), recall that mE (f ) denotes
the average of f over E with respect to the Lebesgue measure:
Z
1
mE (f ) = f (x)dx.
|E| E
See (1.6). Here and below a weight is a locally integrable function with a
positive value almost everywhere on Rn . For a weight w and a measurable set
E, the weight of E is by Z
w(E) ≡ w(x)dx.
E
The weighted Lebesgue space Lp (w) with 0 < p < ∞ represents the set of all
f ∈ L0 (Rn ) for which the norm
Z  p1
kf kLp (w) ≡ |f (x)|p w(x)dx
Rn

is finite. Among others it is quite important to consider power weights. Let

α ∈ R. We set kf kLp (|x|α ) ≡ kf kLp (w) , where w(x) = |x|α , x ∈ Rn . The
power weighted Lebesgue space Lp (|x|α ) is the set of all f ∈ L0 (Rn ) for which
kf kLp (|x|α ) < ∞. If n = 1, then write Lp (|t|α ) instead of Lp (|x|α ). It is also
important to consider the weighted Lebesgue space on the half line. The power
weighted Lebesgue space Lp (tα ) or Lp (rα ) over (0, ∞) is the set of all f ∈
L0 (0, ∞) for which
Z ∞  p1
p α
kf kLp (tα ) = kf kLp (rα ) ≡ |f (t)| t dt < ∞.
0

Although we do not go into the details, the Jacobi hypergroup is the weighted
Lebesgue spaces. Let α ≥ β ≥ −1 with (α, β) 6= (−1, −1). Then we set
µ(t) ≡ (sinh t)1+α (cosh t)1+β dt for t > 0. The space L2 (R+ , ∆) = L2 (R+ , µ)
is the weighted L2 -space called the Jacobi hypergroup.
Section 7.1 focuses on the one-weight (maximal) inequality
Z Z
p
M f (x) w(x)dx . |f (x)|p w(x)dx.
Rn Rn

Generally speaking when the weights on the both sides are the same, the
inequality is called a one-weight inequality or a one-weight norm inequality.
A more general inequality of the form kT f kLq (v) . kf kLp (w) is called a two-
weight inequality or a two-weight norm inequality. Section 7.2 deals with the
two-weight norm inequality of Hardy operators, (fractional) maximal opera-
tors and singular integral operators. We will obtain characterizations different
from those in Section 7.1.
 Weighted Lebesgue spaces 295

7.1 One-weighted norm inequality

Although the following example is not used in the context of weighted
norm inequalities, this example is useful to see that the weighted norms arise
naturally.
Example 111. Let E(Rn ) be a ball Banach function space, so that there
exists a constant ck such that kf kL1 (B(k)) ≤ ck kf kE for all f ∈ E(Rn ) and
∞
(2k ck )−1 χB(k) , we obtain E(Rn ) ,→ L1 (w). We have a
P
k ∈ N. Setting w ≡
k=1
similar embedding for Banach function spaces over a measure space (X, B, µ).
For 1 ≤ p ≤ ∞, we will consider the class Ap . Here, we distinguish three
cases to define and investigate Ap . See Sections 7.1.1, 7.1.2 and 7.1.3 for the
cases of p = 1, 1 < p < ∞ and p = ∞, respectively. Section 7.1.4 is of
different nature compared with Sections 7.1.1, 7.1.2 and 7.1.3; Sections 7.1.1,
7.1.2 and 7.1.3 are oriented to the Hardy–Littlewood maximal operator and
singular integral operators, while Section 7.1.4 is related to fractional integral
operators and fractional maximal operators.
In Sections 7.1.5 and 7.1.6 we consider applications of Ap defined and
investigated in Sections 7.1.1, 7.1.2 and 7.1.3. Section 7.1.5 will provide a tool
to create some inequalities based on the boundedness property of the weighted
Lebesgue spaces. Section 7.1.6 will solve the A2 -conjecture by employing some
key tools such as the Lerner–Hytönen decomposition obtained in Section 4.2.3.

7.1.1 The class A1

We start with the weighted estimate for the case of p = 1. We will introduce
the class A1 for this purpose. We are thus interested in the weighted weak-
(1, 1) maximal inequality

C
w{x ∈ Rn : M f (x) > λ} ≤ kf kL1 (w) . (7.1)
λ
Here, λ > 0 and f ∈ L0 (Rn ). This is the weighted counterpart of the weak-
(1, 1) inequality of the Hardy–Littlewood maximal operator. Before we inves-
tigate the necessary and sufficient condition, we consider an example. Among
the weights, we are particularly interested in the power weights. Here, by a
power weight we mean a weight w of the form w(x) = |x|α , x ∈ Rn . We are
now interested in the case of w(x) ≡ |x|α with α ∈ R.

Example 112. Let w(x) ≡ |x|α with α ∈ R. If w satisfies the weighted

weak-(1, 1) maximal inequality, then α > −n. In fact, we can simply test
the weighted weak-(1, 1) maximal inequality on f = χB((2,2,...,2),1) keeping in
mind that B((2, 2, . . . , 2), 1) is away from the singularity.
296 Morrey Spaces

Example 113. Let α > −n, and let Q be a cube. Then

Z
|x|α dx ∼ max(`(Q), |c(Q)|)α |Q|.
Q

This can be checked by distinguishing two cases: c(Q) is far from the origin,
that is, |c(Q)| > 4n`(Q), and otherwise.
We are now oriented to a general case of weights. We answer the question
of characterizing w for which (7.1) holds.
Definition 70. For a weight w, define its A1 -characteristic by
 
Mw \
[w]A1 ≡ = inf  {α > 0 : mQ (w) ≤ α essinfx∈Q w(x)} .
w L∞
Q∈Q

The quantity is also called the A1 -constant. A weight w is an A1 -weight if

[w]A1 is finite. The class A1 collects all A1 -weights.
Roughly speaking, the class A1 is close to the constant weight 1. We con-
sider a concrete but important example.
Example 114. Let w(x) ≡ |x|−n+δ , x ∈ Rn with δ ∈ (0, n]. Then [w]A1 ∼
δ −1 . In fact, similar to Example 113, we have the following two-sided estimate:
Z !
1 1
M w(x) ∼ max w(x), |y|−n+δ dy ∼ w(x),
|B(3|x|)| B(3|x|) δ

as is easily seen from the observation that 0 ∈ B(x, 3r) or 0 ∈

/ B(x, 3r) for
any ball B(x, r).
Although WL1 (w) is not a Banach space and M is not a linear operator,
we still define
kM f kWL1 (w)
kM kL1 (w)→WL1 (w) ≡ sup .
f ∈L1 (w)\{0} kf kL1 (w)

The weak-(1, 1) constant of the maximal operator is comparable to the

A1 -constant of w.
Theorem 280. For any weight w, the maximal operator M is bounded from
L1 (w) to WL1 (w) if and only if w ∈ A1 . Furthermore, if this is the case,
kM kL1 (w)→WL1 (w) ∼ [w]A1 .
Proof The “if” part is easy; simply use Corollary 140 and M w ≤ [w]A1 w.
For the “only if” part, consider the inequality χQ mQ (|f |) ≤ M f and λ =
mQ (|f |) for given a cube. Then w(Q) ≤ w{x ∈ Rn : M f (x) ≥ mQ (|f |)} ≤
CmQ (|f |)−1 kf kL1 (w) , or equivalently, w(Q)mQ (|g|w−1 ) ≤ CkgkL1 . Then we
 Weighted Lebesgue spaces 297

are in the position of using the duality L1 (Rn )-L∞ (Rn ). The above obser-
vations include the proof of kM kL1 (w)→WL1 (w) ∼ [w]A1 . Thus, the proof is
complete.
We now present some examples of A1 -weights. Roughly speaking there are
two ways to create A1 -weights. Theorem 281 below, which uses the maximal
operator of measures, can be used to create some concrete examples, while
Theorem 284 can be used for the theoretical aspects.
Theorem 281. Assume that µ is a Radon measure on Rn such that M µ(x)
is finite for a.e. x ∈ Rn . Then for 0 < δ < 1, w ≡ (M µ)δ is an A1 -weight,
where M µ is given by (4.7). Furthermore, A1 (w) .δ 1.
It matters that the implicit constant above does not depend on µ.
Proof We have to establish mQ (w) . w(x) whenever Q is a cube con-
taining x. We split µ according to 10Q. Write µ|A ≡ µ(A ∩ ·), the restriction
of µ to a measurable subset A. Split µ by µ = µ1 + µ2 with µ1 ≡ µ|10Q and
µ2 ≡ µ|(Rn \ 10Q). We will prove

mQ (M µ1 δ ) . w(x), mQ (M µ2 δ ) . w(x)

for all x ∈ Q. To deal with µ1 we use Example 66, the Kolmogorov inequality,
to obtain
 δ
δ δ µ(10Q)
mQ (M µ1 ) . m10Q (M µ1 ) . . w(x).
|10Q|

To handle µ2 , we will employ a pointwise estimate, which we encounter fre-

quently. Let y ∈ Q. First we write out M µ2 (y) in full:
 
µ(R \ 10Q)
M µ2 (y) = sup : R ∈ Q, y ∈ R .
|R|

A geometric observation readily shows `(R) ≥ 3`(Q) whenever R is a cube

intersecting both Q and Rn \ 10Q. Therefore such an R engulfs Q, if we triple
R. In view of this fact, we deduce
 
n µ(S)
M µ2 (y) ≤ 3 sup : S ∈ Q, Q ⊂ S ≤ 3n M µ(x).
|S|

This pointwise estimate readily gives mQ (M µ2 δ ) . w(x). Hence, the theorem

is now proven.
The following proposition gives us a concrete example of A1 -weights.
Proposition 282. Let 0 ≤ a < n. Then w(x) ≡ |x|−a , x ∈ Rn belongs to A1 .
Proof This simply paraphrases Example 114. Alternatively, we argue as
follows: Let a ∈ (0, n). In this case it suffices to take µ = δ0 , the point measure
298 Morrey Spaces

massed at the origin and δ = na , which is a special case of Theorem 281. Then
w equals M µ modulo multiplicative constants.
We completely characterize the power weights in A1 .
Theorem 283. Let α ∈ R. Then | · |α ∈ A1 if and only if −n < α ≤ 0.
Proof Let α > 0. Then M [| · |α ] ≡ ∞. Thus, this simply paraphrases
Example 114 and Proposition 282 again.
We now invoke the Rubio de Francia iteration algorithm to create more
examples of A1 -weights: As usual let L0 (Rn ) denote the set of all measurable
functions f . Of course, we identify two functions if they are different by a set of
measure zero. Denote by M k the k-fold composition of the Hardy–Littlewood
maximal operator M .
Theorem 284. Let (E(Rn ), k · kE ) be a Banach space of L0 (Rn ) such that
M f ∈ E(Rn ) together with the estimate
kM f kE ≤ Dkf kE (7.2)
n
holds whenever f ∈ E(R ), where D is independent of f . Assume in addition
that any function in E(Rn ) is finite almost everywhere. Let g ∈ E(Rn ) with
∞
Mkg
. Then G belongs to E(Rn ) and satisfies
P
norm 1. Define G ≡ |g| + (2D)k
k=1
the following properties:
kGkE ≤ 2kgkE = 2 (7.3)
and
[G]A1 ≤ 2D. (7.4)
Proof Since the series definining G converges absolutely in the topology
of E(Rn ), G ∈ E(Rn ). This also implies (7.3). Hence, G is finite almost every-
where by our assumption. Note that G is everywhere positive because M g > 0.
Finally (7.4) is easy to check using (7.2) and the sublinearity of f , G ∈ A1 .
We end Section 7.1.1 with the characterization of Morrey spaces in terms
of A1 -weights. The following proposition can be used to express the Morrey
norm in terms of the weighted Lebesgue spaces:
q
Proposition 285. Let 1 < q ≤ p < ∞, and let 0 < θ < p. Then for all
1 1
0 n
f ∈ L (R ), kf k ∼ sup |Q| p−q kf k .
Mp
q ( 1 )
Q∈Q Lq (M 1−θ χ )
Q

1 1
Proof Since M χQ ≥ χQ , kf kMpq ≤ sup |Q| p − q kf k ( 1 ) . To show
Q∈Q Lq (M 1−θ χ )
Q

the opposite inequality, we decompose

∞
1 1 X 1 1 n(1−θ)
|Q| p − q kf k q ( 1−θ
1 ) . |Q| p − q 2−j q kf kLq (2j Q) . kf kMpq .
L (M χQ )
j=0
q
Note that the series is convergent since θ < .
p
 Weighted Lebesgue spaces 299

Example 115. Suppose that we have an operator T defined on Lq (w) for

any w ∈ A1 . Then T can be restricted to Mpq (Rn ) due to Proposition 285.
In Theorem 306, we will see that singular integral operators fall under this
scope.

7.1.2 The class Ap

Having set down the necessary and sufficient condition for the weighted
weak-(1, 1) maximal inequality to hold, we are now oriented to the weak-(p, p)
maximal inequality
1
λ[w{x ∈ Rn : M f (x) > λ}] p . kf kLp (w) (f ∈ L0 (Rn ), λ > 0)

and the strong-(p, p) maximal inequality

kM f kLp (w) . kf kLp (w) (f ∈ L0 (Rn )).

Here, 1 < p < ∞. We will introduce the class Ap . It should be noted that the
weighted weak-(1, 1) maximal inequality is equivalent to

λ[w{x ∈ Rn : M f (x) ≥ λ}] ≤ Ckf kL1 (w) (f ∈ L1 (w), λ > 0).

Needless to say, by the Chebyshev inequaltiy, the strong-(p, p) maximal

inequality implies the weak-(p, p) maximal inequality. Furthermore, as it turns
out, for a weight w these two inequalities occur at the same time. However,
to trace the behavior of the constants C, we will consider these two inequali-
ties separately. Before we obtain the necessary and sufficient condition for the
weighted weak/strong-(1, 1) maximal inequality to hold, we give an example.
Example 116. Let 1 < p < ∞ and α ∈ Rn . As is easily seen by mimicking
Example 112, if w satisfies the weighted weak-(p, p) maximal inequality, then
α > −n. Furthermore, we need to have α < n(p − 1). In fact, if α ≥ n(p − 1),
χB(1) (2x)
then f (x) ≡ , x ∈ Rn belongs to Lp (|x|α ). However, M f ≡ ∞,
|x|n log |x|
/ L1loc (Rn ).
since f ∈
For the time being, we seek in Section 7.1.2 to show that the following class
of weights is the solution to the problem of characterizing the weak/strong-
(p, p) maximal inequalities.
Definition 71. Let 1 < p < ∞. A locally integrable weight w is an Ap -
weight or belongs to Ap -class, if 0 < w < ∞ almost everywhere, and [w]Ap ≡
1
( )
sup mQ (w)mQp−1 (w−1 ) < ∞. The quantity [w]Ap is referred to as the Ap -
Q∈Q
constant or the Ap -characteristic. The class Ap collects all Ap -weights.
Remark 13. It is easy to check A1 ⊂ Ap ⊂ Aq ⊂ L1loc (Rn ) by Hölder’s
inequality whenever 1 ≤ p ≤ q < ∞.
300 Morrey Spaces

Before we give some examples of the Ap -weights, we justify the definition,

which shows that the weighted weak-(1, 1) maximal inequality is sufficient for
the weight to belong to the class Ap . As we will see, it is not so hard to see
that w ∈ Ap if the weak-(p, p) maximal inequality holds. In fact we have the
following:
Proposition 286. Let 1 < p < ∞ and N > 0. If a weight w satisfies the
weak-(p, p) maximal inequality
1
λ[w{x ∈ Rn : M f (x) > λ}] p ≤ N kf kLp (w) (f ∈ Lp (w), λ > 0),
1
then N ≥ ([w]Ap ) p .
Proof Let f ∈ L0 (Rn ). If we fix a cube Q, then M f (x) ≥ mQ (|f |) for all
x ∈ Q. Consequently,
1
mQ (|f |)w(Q) p ≤ N kf kLp (w) (f ∈ Lp (w)),

or equivalently,
1 1
mQ (w− p |g|)w(Q) p ≤ N kgkLp (g ∈ Lp (Rn )).
0 1 p−1 1
By the duality Lp (Rn )-Lp (Rn ), we have mQ (w− p−1 ) p w(Q) p ≤ N. Thus,
w ∈ Ap together with the estimate N p ≥ [w]Ap since Q is arbitrary.
We polish Example 116 having somewhat clarified how the class Ap is
natural.
Example 117. Let w(x) = |x|−n+δ , x ∈ Rn with δ ∈ (0, n). Then [w]Ap ∼
1
( )
δ −1 . We calculate mQ (w)mQp−1 (w−1 ) according to the position of Q. If 0 ∈
/
1 1
( ) ( )
2Q, then mQ (w)mQp−1 (w−1 ) ∼ mQ (w(c(Q)))mQp−1 (w(c(Q))−1 ) = 1. If 0 ∈
1
( p−1 )
2Q, then mQ (w)mQ (w−1 ) ∼ δ −1 `(Q)−δ `(Q)δ = δ −1 .
0
Since Lp (Rn ) and Lp (Rn ) are dual to each other, it is natural to ask
ourselves whether the classes Ap and Ap0 are strongly related. Actually, the
following equality, whose easy proof we omit (see Exercise 97), follows from
the definition.
1
Lemma 287. If w ∈ Ap with 1 < p < ∞, then we have w− p−1 ∈ Ap0 with
1 0
[w− p−1 ]Ap0 = ([w]Ap )p −1 .
1
Motivated by Lemma 287, we occasionally call w− p−1 the dual weight of
w ∈ Ap , 1 < p < ∞.
We apply Lemma 287 to the power weights. Example 118 concludes our
calculation on the power weights.
 Weighted Lebesgue spaces 301

Example 118. Let 1 < p < ∞, and let w(x) ≡ |x|−n+δ , x ∈ Rn , with
p0 0
0 < δ < n. Then v(x) ≡ w(x)− p = |x|(p −1)(n−δ) , x ∈ Rn , is an Ap0 -weight
0
with [v]Ap0 ∼ δ −(p −1) . As a result, W (x) ≡ v(x) = |x|(p−1)(n−δ) , x ∈ Rn , is
0 1
an Ap -weight with [W ]Ap ∼ δ −p +1 = δ − p−1 .
We completely characterize the power weights in Ap for 1 < p < ∞. Pay
attention to the endpoint. The condition with p = 1 obtained in Theorem 288
differs from the one in Theorem 283.
Theorem 288. Let α ∈ R, and let 1 < p < ∞. Then | · |α ∈ Ap if and only
if −n < α < n(p − 1).
Proof The necessity follows from Example 116. The sufficiency follows
from Examples 117 and 118 together with the fact that 1 ∈ Ap .
Here, we show that the Ap -condition is sufficient for the weak-(p, p) max-
imal inequality to hold. We need the weak-type estimate later:
Theorem 289. Let w ∈ Ap with p ∈ [1, ∞). Then

sup λp w{x ∈ Rn : M f (x) > λ} . ([w]Ap kf kLp (w) )p

λ>0

for all f ∈ L0 (Rn ). Furthermore, if N is a constant satisfying

sup λp w{x ∈ Rn : M f (x) > λ} ≤ (N kf kLp (w) )p

λ>0

for all f ∈ L0 (Rn ), then N & [w]Ap .

Proof If p = 1, this is already proven earlier in Theorem 280. If p > 1,
carefully examine the proof of the case of p = 1; see Exercise 99.
We check that the power here is optimal including the case of p = 1.
Example 119. Let 1 < p < ∞. Let w(x) ≡ |x|(p−1)(n−δ) , x ∈ Rn with
1
0 < δ < 1. Then for f ≡ χB(1) | · |−n+δ , we have kf kLp (w) ∼ δ − p . Meanwhile,
if we choose κ > 0 appropriately, then

w{x ∈ Rn : M f (x) > κδ −1 } = w(B(1)) ∼ 1.

Thus, if the estimate λw{x ∈ Rn : M f (x) > λ} ≤ D([w]Ap )A kf kLp (w) holds,
1 1
then δ −1 ≤ Dδ −A(p−1)− p for all δ > 0, indicating that −1 ≥ −A(p − 1) − ,
p
or equivalently, Ap ≥ 1. Suppose we have

λw{x ∈ Rn : M f (x) > λ} ≤ D([w]A1 )A kf kL1 (w)

for any λ > 0. If we interpolate between this inequality and

kM f kL∞ (w) ≤ kf kL∞ (w) ≤ D[w]A1 kf kL∞ (w) ,

302 Morrey Spaces

then 1 A
λ[w{x ∈ Rn : M f (x) > λ}] p ≤ D0 ([w]A1 ) p kf kLp (w) .
Going through a similar argument using Example 114, we obtain that the
power in the weighted weak-(1, 1) maximal inequality is sharp.
We now consider the weighted strong-(p, p) maximal inequality. As is
announced before, the power of [w]Ap differs from the one in the weighted
weak-(p, p) maximal inequality.
Theorem 290. Let w ∈ Ap with 1 < p < ∞. Then for all f ∈ Lp (w),
1
kM f kLp (w) .p,n ([w]Ap ) p−1 kf kLp (w) .
The proof hinges on the universal estimate. Let W : Rn → (0, ∞) be a
weight. Denote by Mc,W the centered weighted maximal operator given by
Z
Q 1
Mc,W f (x) = Mc,W f (x) ≡ sup |f (y)| W (y)dy. (7.5)
Q∈Q, c(Q)=x W (5Q) Q

By virtue of Theorem 133, it is easy to see that Mc,W is Lp (W )-bounded.

1
Let σ ≡ w− p−1 be the dual weight.
Proof Let f ∈ L0 (Rn ). Let Q be a cube containing a point x ∈ Rn . By
 p−1
w(5Q) σ(5Q)
the definition of the Ap -weights we have ≤ [w]Ap . Hence,
|5Q| |5Q|
(without using Hölder’s inequality) by the definition of the maximal operator
we deduce
1
 p−1 ! p−1
|Q|
Z
1
p−1 1
mQ (|f |) . [w]Ap |f (y)|dy
w(5Q) σ(5Q) Q
1
 p−1 ! p−1
|Q|
Z
1
1
= [w]Ap−1 |f (y)|σ(y)−1 σ(y)dy .
p
w(5Q) σ(5Q) Q
(It is noteworthy that we do not use w ∈ Ap elsewhere.) Using the definition
of the weighted maximal operators, we obtain
1
 p−1 ! p−1
1
p−1 |Q| −1
mQ (|f |) ≤ [w]Ap inf Mc,σ [f · σ ](z)
w(5Q) z∈Q
1
Z  p−1 ! p−1
1
1
= [w]Ap−1 inf Mc,σ [f · σ −1 ](Z) dz
p
w(5Q) Q Z∈Q
 Z 1
 p−1
1
1 −1 p−1
≤ [w]Ap−1 M c,σ [f · σ ](z) dz
p
w(5Q) Q
 Z 1
 p−1
1
1 −1 p−1 −1
= [w]Ap−1 M c,σ [f · σ ](z) w(z) w(z)dz
p
w(5Q) Q

1
≤ [w]Ap Mc,w Mc,σ [f · σ −1 ]p−1 w−1 (x) p−1
 
 Weighted Lebesgue spaces 303

for all x ∈ Q. Or equivalently, since Q is chosen arbitrarily,

1
M f (x) .p,n [w]Ap Mc,w Mc,σ [f · σ −1 ]p−1 w−1 (x) p−1 (x ∈ Rn ).
 

Inserting this pointwise estimate into kM f kLp (w) and then using the bound-
edness of Mc,w , we obtain
1 1
kM f kLp (w) .p,n [w]Ap−1
p
kMc,w [(Mc,σ [f · σ −1 ])p−1 w−1 ] p−1 kLp (w)
1
≤ ([w]Ap kMc,w [(Mc,σ [f · σ −1 ])p−1 w−1 ]kLp0 (w) ) p−1
1
.p ([w]Ap k(Mc,σ [f · σ −1 ])p−1 w−1 kLp0 (w) ) p−1
1
= ([w]Ap k(Mc,σ [f · σ −1 ])p−1 kLp0 (σ) ) p−1
1
= [w]Ap−1
p
kMc,σ [f · σ −1 ]kLp (σ) .

If we use the boundedness of Mc,σ , then we have

1 1
kM f kLp (w) .p,n [w]Ap−1
p
kMc,σ [f · σ −1 ]kLp (σ) .p [w]Ap−1
p
kf · σ −1 kLp (σ) = kf kLp (w)

for all f ∈ L0 (Rn ). Thus, we obtain the desired Lp (w)-estimate.

Example 120. We dualize Theorem 290. Let 1 < p < ∞ and w ∈ Ap . Then
1 1
kw−1 M [wh]kLp0 (w) = kM [wh]k p0 − p−1
1 . ([w− p−1 ]Ap0 ) p0 −1 khkLp0 (w) =
L (w )
[w]Ap khkLp0 (w) for all h ∈ L0 (Rn ).
As the following example shows, the power in [w]Ap is best possible:
Example 121. Let W (x) ≡ |x|(p−1)(n−δ) , x ∈ Rn , where 1 < p < ∞ and
0 < δ ≤ n. Then f ≡ χB(1) | · |−n+δ ∈ Lp (W ). Meanwhile M f & δ −1 f . Thus,
we cannot improve the power a in the bound kM f kLp (W ) . ([W ]Ap )a kf kLp (W )
1
with a = .
p−1
A doubling weight w is a weight whose corresponding weighted measure
wdx = w(x)dx is a doubling measure. We will use the strong doubling property
of Ap weights. For the proof we use the weighted weak maximal inequality.
Corollary p Let w ∈ Ap , p ≥ 1, and let Q ∈ Q. Then the strong doubling
 291.
|S|
property w(Q) . [w]Ap w(S) holds for any measurable set S ⊂ Q.
|Q|
Proof Using Theorem 289, we have

w(Q) ≤ w({x ∈ Rn : M χS (x) ≥ |Q|−1 |S|}) . [w]Ap (|S|−1 |Q|)p w(S).

Let a ∈ {0, 1/3}n be fixed. Suppose that we have a family Sj = {Qjk }k∈Kj
∞
S
of disjoint cubes in Da for each j ∈ N0 . Recall that S ≡ Sj is an η-sparse
j=0
304 Morrey Spaces
[ [ j X
family with a level structure if Qj+1
k ⊂ Qk and |Qj+1
k ∩R| ≤
k∈Kj+1 k∈Kj k∈Kj+1
η|R| for all R ∈ Sj , j ∈ N. We also remark that a weight w belongs to A2
if and only if w−1 belongs to A2 . Hence, the class A2 is special. Due to this
special property we can prove the following boundedness. We note that the
estimate [w]A2 is optimal in view of the A2 -theorem, which we consider later.
Lemma 292. Let a ∈ {0, 1/3}n . Let {Qjk }j∈N,k∈Kj ⊂ Da be a sparse family
of generalized dyadic cubes with a level structure. As long as the sum makes
sense, define
∞ X
X
Af ≡ mQj (f )χQj
k k
j=1 k∈Kj

for f ∈ L1loc (Rn ). Then kAf kL2 (w) . [w]A2 kf kL2 (w) for all f ∈ L2 (w) and
w ∈ A2 .
Before the proof, a helpful remark may be in order. As is easily seen from
the inequality kf χQ kL1 ≤ kf kL2 (w) kχQ kL2 (w−1 ) , if f ∈ L2 (w) and w ∈ A2 ,
then f ∈ L1loc (Rn ). Hence, at least the definition of mQkj (f ) in Af makes
sense if f ∈ L2 (w).
For the proof we use the following notation. For a weight v, f ∈ L0 (Rn )
and a ∈ {0, 1/3}n , we define
Z
χQ (x)
Da
Mv f (x) ≡ sup |f (y)|v(y)dy (x ∈ Rn ).
Q∈Da v(Q) Q

Qjk and Ekj ≡ Qjk \ Ωj+1 , so that |Ekj | & |Qjk |.

S
Set Ωj ≡
k∈Kj

Proof Write σ ≡ w−1 . We dualize the left-hand side: Matters are reduced
to showing
Z
I≡ Af (x)g(x)dx . [w]A2 kf kL2 (w) kgkL2 (σ)
Rn

for all f ∈ L2 (w) ∩ M+ (Rn ) and g ∈ L2 (σ) ∩ M+ (Rn ). By the truncation, we

may also assume that f, g ∈ L∞ n
c (R ). If we insert the definition of A into the
∞ P
mQj (f )mQj (g)|Qjk |. Thus,
P
left-hand side, then we obtain I =
k k
j=1 k∈Kj

∞ X
X
I. mQj (f )mQj (g)|Ekj |. (7.6)
k k
j=1 k∈Kj

Since w ∈ A2 , we have
mQj (f )mQj (g)
k k
Z Z
1 1
≤ [w]A2 · f (x)w(x)σ(x)dx · g(x)σ(x)w(x)dx.
σ(Qjk ) Qjk w(Qjk ) Qjk
 Weighted Lebesgue spaces 305

By the definition of the maximal operator MσDa and MwDa , we obtain

mQj (f )mQj (g) ≤ [w]A2 inf j MσDa [f · w](x) · inf j MwDa [g · σ](x).
k k
x∈Qk x∈Qk

If we insert this estimate into (7.6), then

I . [w]A2 kMσDa [f · w]kL2 (σ) kMwDa [g · σ]kL2 (w) .
It remains to use the universal estimate, Theorem 257.
We end Section 7.1.2 with the following two observations on the classes
A1 , Ap and Ar with 1 < r ≤ p < ∞ and with 1 ≤ p ≤ r.
Lemma 293. Let 1 < r ≤ p < ∞. If w ∈ Ap and W ∈ M+ (Rn ) satisfy
p−r p−r r−1
W w ∈ A1 , then [W p−1 w]Ar ≤ ([W w]A1 ) p−1 ([w]Ap ) p−1 .
If p = ∞, Lemma 293 formally says that A1 ⊂ Ar . Although the proof is
simple, we need monkish patience.
Proof Let Q ∈ Q be fixed. We have to show that
 Z r−1
p−r 1 p−r 1
− r−1
I ≡ mQ (W p−1 w) (W (x) p−1 w(x)) dx
|Q| Q
p−r r−1
≤ ([W w]A1 ) p−1 ([w]Ap ) p−1 .
p−r p−r r−1
Since W p−1 w = (W w) p−1 w p−1 , we have
 Z r−1
p−r 1 p−r
− (p−1)(r−1) − pp
0
I = mQ (W p−1 w) (W (x)w(x)) w(x) dx
|Q| Q
! p−r
p−1
p−r 1 1
≤ mQ (W p−1 w)mQ (w− p−1 )r−1 .
W w L∞ (Q)
p−r p−r r−1
By Hölder’s inequality, mQ (W p−1 w) ≤ mQ (W w) p−1 mQ (w) p−1 . Thus,
p−r 1 r−1 p−r r−1
I ≤ ([W w]A1 ) p−1 mQ (w− p−1 )r−1 mQ (w) p−1 ≤ ([W w]A1 ) p−1 ([w]Ap ) p−1 .
This completes the proof.
Lemma 294. Let 1 ≤ p ≤ r < ∞. Also let w ∈ A1 and v ∈ Ap . Then we
have v −r+p w ∈ Ar . More precisely, [v −r+p w]Ar ≤ [v]A1 r−p [w]Ar .
Once again we need to prove Lemma 294 with monkish patience.
Proof Let Q ∈ Q be fixed. Then,
 Z 
1 1 p−r
v(x)−r+p w(x)dx mQ (w− r−1 v r−1 )r−1
|Q| Q
1 p−r
≤ kv −r+p kL∞ (Q) mQ (w)mQ (w− r−1 v r−1 )r−1
306 Morrey Spaces
p−1 r−p
by the definition of the L∞ (Q)-norm. We remark that + = 1. By
r−1 r−1
Hölder’s inequality
1 p−r 1 p−1 p−r
mQ (w− r−1 v r−1 ) ≤ mQ (w− p−1 ) r−1 mQ (v) r−1 .

If we insert this estimate into the above expression, we obtain the desired
result.
Example 122. Let 1 < q ≤ p < ∞. Then M is bounded on Mpq (Rn ). In fact,
q
for f ∈ Mpq (Rn ), letting θ = 2p , we obtain
1 1
kM f kMpq ∼ sup |Q| p − q kM f k ( 1 )
Q∈Q Lq (M 1−θ χ )
Q

1
thanks to Proposition 285. Meanwhile, since M ( 1−θ ) χQ ∈ A1 , we have
1 1 1 1
sup |Q| p − q kM f k ( 1 ) . sup |Q| p − q kf k ( 1 ) .
Q∈Q Lq (M 1−θ χQ ) Q∈Q Lq (M 1−θ χ )
Q

If we use Proposition 285 once again, kM f kMpq . kf kMpq . This gives another
proof of Theorem 382 to follow. This simple proof seems to be applicable to
many things. However, as it turns out, in some generalized cases, we need
different discussions.
To conclude this section, we remark that we can consider the local coun-
terpart of the results in this section. The details are omitted.

7.1.3 The class A∞

Since M is always bounded on L∞ (w), it does not seem to make sense to
introduce the class A∞ . However, since [w]Ap ≥ [w]Aq if 1 ≤ p ≤ q < ∞, it
seems important that we take the limit as p → ∞ of [w]Ap to define A∞ . This
attempt makes the following definition natural:

Definition 72. A weight w is an A∞ -weight, if 0 < w < ∞ almost every-

where, and [w]A∞ ≡ sup mQ (w) exp(−mQ (log w)) < ∞. The quantity [w]A∞
Q∈Q
is referred to as the A∞ -constant or the A∞ -characteristic. The class A∞
collects all weights w for which it satisfies 0 < w < ∞ and [w]A∞ is finite.
We can extend Remark 13 as follows:

Example 123. It is easy to check A1 ⊂ Ap ⊂ Aq ⊂ A∞ ⊂S L1loc (Rn ) by

Hölder’s inequality whenever 1 ≤ p ≤ q ≤ ∞. Thus A∞ ⊃ Ap . Later
1≤p<∞
we will check that the last inclusion is actually an equality.
 Weighted Lebesgue spaces 307

For the time being we investigate the property of the class A∞ . Let w be a
(1+ε)
weight. If we use Hölder’s inequality, then mQ (w) ≥ mQ (w) for all weights
w. However, since w ∈ L1loc (Rn ) does not imply w ∈ L1+ε n
loc (R ) for any ε > 0,
(1+ε)
the reverse inequality mQ (w) ≤ mQ (w) and its weaker version
(1+ε)
mQ (w) . mQ (w) (7.7)
fails.
However, if the weight belongs to the class A∞ , then the reverse estimate
can be obtained in the following sense:
Theorem 295 (Reverse Hölder inequality). Let w ∈ A∞ . We abbreviate
1 (1+ε)
ε ≡ n+3 > 0. Then mQ (w) ≤ 2mQ (w) for all cubes Q ∈ Q.
2 [w]A∞
Proof By the dilation and the translation, we may assume that Q ∈
D(Rn ). We will use cubes in D(Q). Thus, we may assume
P that w ∈ L∞ (Rn )
by approximating w with functions of the form mR (w)χR for some
R∈Dj (Rn )
D(Q)
j ∈ Z. Let EQ,λ ≡ {x ∈ Q : M [χQ w](x) > λ} for λ > 0. We use the
Layer-Cake formula to obtain
Z Z ∞
I≡ M D(Q) [χQ w](x)ε w(x)dx = ε λε−1 w(EQ,λ )dλ.
Q 0

Let λ > mQ (w). Consider the set {Qj }j∈J(λ) ⊂ D(Q) of all maximal dyadic
cubes in the set EQ,λ whose average of w exceeds λ. Since λ > mQ (w), each
Qj is a proper subset of Q. So thanks to the maximality of each Qj , we have
X X
w(EQ,λ ) = w(Qj ) ≤ 2n λ |Qj | = 2n λ|EQ,λ |.
j∈J(λ) j∈J(λ)

Thus
Z mQ (w) Z ∞
I≤ε λε−1 w(EQ,λ )dλ + 2n ε λε |EQ,λ |dλ
0 mQ (w)
2n ε
Z
≤ mQ (w)1+ε |Q| + M D(Q) [χQ w](x)1+ε dx.
1+ε Q

Since w ∈ A∞ , mR (w) ≤ [w]A∞ exp (mR (log w)) for all R ∈ D(Rn ). This
implies that
D(Q)
M D(Q) w(x) ≤ [w]A∞ Mlog w(x) = [w]A∞ sup χR (x) exp (mR (log w))
R∈D(Q)

for all x ∈ Q. Thus

2n ε([w]A∞ )1+ε
Z
1+ε D(Q)
I ≤ mQ (w) |Q| + Mlog [χQ w](x)1+ε dx.
1+ε Q
308 Morrey Spaces

Due to the Lebesgue differentiation theorem, we have

2n ε([w]A∞ )1+ε
Z Z
D(Q)
w(x)1+ε dx ≤ mQ (w)1+ε |Q| + Mlog [χQ w](x)1+ε dx.
Q 1+ε Q

By virtue of (6.16), we have

2n eε
Z Z
w(x)1+ε dx ≤ mQ (w)1+ε |Q| + ([w]A∞ )1+ε w(x)1+ε dx.
Q 1+ε Q

Note that
1
n+3 [w]
2n eε[w]A∞ 2n e[w]A∞ ([w]A∞ ) 2
 
A∞ e 1 1
([w]A∞ )ε = ≤ exp ≤ ,
1+ε 2n+3 [w]A∞ + 1 8 16e 2
since ea ≤ 1 + 2a for 0 < a < 0.1. Thus, it follows that
Z Z
1
w(x)1+ε dx ≤ mQ (w)1+ε |Q| + w(x)1+ε dx.
Q 2 Q
Since w is assumed to be bounded, we obtain the desired result.
The following corollary can be located as the weak version of Corollary
291.
Lemma 296. Let E be a Lebesgue measurable set contained in a cube Q.
n+4
Also let w ∈ A∞ and a > 2n+1+2 [w]A∞ . Assume that |Q| ≤ 2n a−1 |E|.
Then w(Q) .[w]A∞ w(E).
Proof Since w(Q) = w(E) + w(Q \ E), it suffices to show that
  1 2n+3 [w]A
w(Q \ E) 1 1+2n+3 [w]A
∞ −
1+2n+4 [w]A
∞
≤2 n+4 [w] =2 ∞ (< 1). (7.8)
w(Q) 21+2 A∞

1
Let ε ≡ as before. First, observe that |Q \ E| ≤ 2n a−1 |Q|. As a
2n+3 [w]A∞
result, by Theorem 295 we obtain
Z 1
 1+ε
ε
1+ε
w(Q \ E) ≤ w(x) dx |Q \ E| 1+ε
Q
1  n ε
 1+ε
|Q| 1+ε 2
≤ 2 mQ (w) |Q| w(Q)
w(Q) a
 n  1+εε
2
= 2 w(Q)
a
  n+31
1 1+2 [w]A
∞
≤ 2 1+2 n+4 [w] .
2 A ∞

Thus we obtain (7.8).

 Weighted Lebesgue spaces 309

Another direct consequence of Theorem 295 is that the class Ap with

1 < p < ∞ enjoys the openness property. We will characterize the class Ap as
follows:
Theorem 297. If w ∈SAp with 1 < p ≤ ∞, then w ∈ Aq for some 1 ≤ q < p.
In other words Ap = Aq .
q∈[1,p)

Proof According to Remark 13, we see that Av ⊂ Au ⊂ A∞ whenever

1 ≤ v ≤ u ≤ ∞. So, one inclusion is clear.
We now distinguish two cases: p < ∞ and p = ∞.
1
Suppose w ∈ Ap and p < ∞. Then w− p−1 ∈ Ap0 ⊂ A∞ . Let ε be the
(1+ε) 1 1
constant satisfying mQ (w− p−1 ) ≤ 2mQ (w− p−1 ) for all cubes Q, whose
p+ε
existence is guaranteed by Theorem 295. Thus if we set q ≡ , then we
1+ε
1 1
have mQ (w− q−1 )q−1 ≤ 2p−1 mQ (w− p−1 )p−1 , implying that w ∈ Aq .
Suppose instead that w ∈ A∞ (and p = ∞). Use Theorem 295 once
  ε
w(E) |E| 1+ε
again. Then according to reverse Hölder’s inequality, ≤ 2 ,
w(Q) |Q|
whenever E is εa subset of a cube Q. Consequently, if we let 0 < α0  1 and
0 < β0 ≡ 2α0 1+ε , then w(E) ≤ β0 w(Q) for all measurable sets E and Q such
that Q is a cube containing E and that |E| ≤ α0 |Q|. Set α ≡ 1 − β0 ∈ (0, 1)
and β ≡ 1 − α0 ∈ (0, 1). If we contrapose this fact, then

|E| < β|Q| (7.9)

whenever measurable sets E and Q satisfy that Q is a cube containing E

w(E) < αw(Q). This in particular implies that w(B(x, 2r)) . w(B(x, r)) for
all x ∈ Rn and r > 0.
k
To prove w ∈ Au for some u ∈ (1, ∞), fix a cube Q and set λk ≡ D α λ0
|Q| D(Q)
for each k ∈ N0 , where D  1 and λ0 ≡ . Denote by Mw the dyadic
w(Q)
maximal operator with respect to Q and the weight w:
Z
D(Q) χR (x)
Mw f (x) ≡ sup |f (y)|w(y)dy (x ∈ Rn ).
R∈D(Q) w(R) R

Keeping in mind the fact that D  1, we split Ωk ≡ {x ∈ Q :

D(Q)
Mw [w−1 ](x) > λk } into a disjoint union of maximal dyadic cubes {Qjk }j
such that
|Qjk |
λk < < Dλk = αλk+1 . (7.10)
w(Qjk )
1
Fix k ∈ N0 . By the Lebesgue differentiation theorem, w(x) ≥ for a.e.
λk
x ∈ Q \ Ωk .
310 Morrey Spaces

Let k, j0 be fixed. Then

X
w(Ωk+1 ∩ Qk,j0 ) ≤ w(Qk+1,j ).
j;Qk+1,j ∈D(Qk,j0 )

Recall that two dyadic cubes are disjoint unless one is contained in the other.
Therefore we have
X |Qk+1,j | |Qk,j0 |
w(Ωk+1 ∩ Qk,j0 ) ≤ ≤ .
λk+1 λk+1
j;Qk+1,j ∈D(Qk,j0 )

|Qk,j0 |
If we use (7.10) once again, then w(Ωk+1 ∩ Qk,j0 ) ≤ ≤ αw(Qk,j0 ). We
λk+1
can pass the above inequality to the unweighted one: |Ωk+1 ∩Qk,j0 | ≤ β|Qk,j0 |.
Adding the above inequality over j0 , we obtain

|Ωk+1 | ≤ β|Ωk |, (7.11)

∞
T
yielding that Ωk = 0 is almost equal to an empty set. We write Ω−1 ≡ Q.
k=0
Z ∞ Z
dx X dx
Then τ
= τ
. Invoking (7.10)–(7.11), we obtain
Q w(x) k=−1Ωk \Ωk+1 w(x)

Z ∞
dx X
≤ λk+1 τ |Ωk |
Q w(x)τ
k=−1
∞  kτ
τ
X D
. λ0 β k |Ω0 |
α
k=−1
∞  kτ
τ
X D
≤ λ0 |Q| βk.
α
k=−1

Therefore, if τ > 0 is sufficientlyZsmall, then the series in the above inequality

1 dx |Q|τ
converges. Thus, we obtain . . Hence, w ∈ A1+τ −1 .
|Q| Q w(x)τ w(Q)τ

Example 124. From (7.9), we obtain w(2k Q) ≥ κw(2k−1 Q) for all k ∈ N

and for all cubes Q, where κ > 1 is independent of Q and k. In particular,
∞
w(2k Q)−θ . w(Q)−θ for any cube Q and θ > 0.
P
k=1
 Weighted Lebesgue spaces 311

Let 0 < p < ∞ and 0 < q ≤ ∞. Given a sequence of measurable functions

{fj }∞
j=0 , we define

  q1
X∞ 
k{fj }∞
j=0 k`q (Lp (w)) ≡ (kfj kLp (w) )q ,
 
j=0
  q1
X∞
k{fj }∞
j=0 kLp (w,`q ) ≡  |fj |q  .
j=0
Lp (w)

Define `q (Lp (w)) and Lp (w, `q ), called weighted vector-valued Lebesgue spaces,
to be the collection of all sequences {fj }∞ j=0 of measurable functions for which
the norms k{fj }∞j=0 k` q (Lp (w)) and k{fj }
∞
j=0 kLp (w,`q ) are finite, respectively. As
applications of Theorem 270, we can prove the following inequality:
Theorem 298 (Weighted vector-valued inequality for M ). Let 1 < p, q < ∞,
1 < r ≤ ∞ and w ∈ Ap . Then k{M fj }∞ ∞
j=0 kLp (w,`q ) .p,q,r k{fj }j=0 kLp (w,`q )
∞ 0 n
for all {fj }j=1 ⊂ L (R ).
Proof Since w ∈ Ap̃ for some p̃ ∈ (1, p) by virtue of Theorem 295, we see
p p̃
that M fj . M22,c,µ [|fj | p̃ ] p , where µ ≡ wdx and M22,c,µ is defined in Section
6.2.1 with k = 22. It remains to combine Theorem 270 with this pointwise
estimate.

7.1.4 The class Ap,q

We are now oriented to the class of (couples of) weights adapted to frac-
tional integral operators or fractional maximal operators. Weak type estimates
are easy to obtain. Hence, we formulate the results in full generality.
Definition 73 (Aα α
p,q ). Given 1 < p ≤ q < ∞ and 0 ≤ α < n, Ap,q collects all
α 1 1 1 1
the pairs (u, σ) of weights if [u, σ]Aαp,q ≡ sup |Q| n + q − p mQ (u) q mQ (σ) p0 < ∞.
Q∈Q
The class Aα
p,q collects all pairs (u, σ) of weights for which [u, σ]Aα
p,q
is finite.
Example 125. Let 1 < p ≤ q < ∞. Also let w be a weight. Write α = np − nq .

−q
  1 1
(1) We claim that (u, σ) ≡ (M w) p0 , w ∈ Aα
p,q . In fact, mQ (u) mQ (σ)
q p0 =

− pq0 1 1
mQ ((M w) ) q mQ (w) p0 . If x ∈ Q, then M w(x) ≥ mQ (w). Therefore,
we have
1
1 − p10 1
mQ (u) q mQ (σ) p0 ≤ mQ (w) mQ (w) p0 = 1.

p0
(2) Likewise, we can check that (u, σ) = (w, (M w)− q ) ∈ Aα
p,q .
312 Morrey Spaces

If the weights are radial and monotone, we can calculate the quantity with
ease.
Example 126. Let p, q > 1 and 0 ≤ α < n. Also let v, w be weights. Assume
that there exist ŵ ∈ M↑ [0, ∞) and v̂ ∈ M↓ [0, ∞) such that w(x) = ŵ(|x|),
1 1
v(x) = v̂(|x|) for all x ∈ Rn . Define σ ≡ w− p−1 and σ̂ ≡ w− p−1 . We compare

1
1
I ≡ sup Rα−n kσ̂χ(0,R) kL1 (tn−1 ) q0 kv̂χ(0,R) kL1 (tn−1 ) p and
R>0
1 1
J ≡ sup Rα−n σ(B(x, R)) q0 v(B(x, R)) p .
x∈Rn ,R>0

Since v̂ ∈ M↓ (0, ∞), v(B(x, r)) ≤ v(B(r)) for any x ∈ Rn and r > 0 and,
since ŵ ∈ M↑ (0, ∞), we have σ(B(x, r)) ≤ σ(B(r)). Hence, J ∼ I.
For the class Aα
p,q , we have a weak type estimate.

Theorem 299. Let 1 < p ≤ q < ∞ and 0 ≤ α < n. Then for any pair of
weights (u, σ) ∈ Aα
p,q we have
1
sup λ(u{x ∈ Rn : Mα f (x) > λ}) q . [u, σ]Aαp,q kf kLp (σ1−p ) (7.12)
λ>0

for all f ∈ Lp (σ 1−p ).

Note that (7.12) is necessary for (u, σ) ∈ Aα p,q since Mα f (x) ≥
χQ (x)`(Q)α mQ (|f |), x ∈ Q for all cubes Q and f ∈ L1 (Rn ). Note also that
1
[u, σ]Aαp,q = inf{D > 0 : `(Q)α mQ (|f |)u(Q) q ≤ Dkf kLp (σ1−p ) }.

Proof We use Theorem 179. Let Q0 be a fixed dyadic cube. By the mono-
tone convergence theorem, we have only to work in Q0 . Set
 
 X 
E ≡ x ∈ Q0 : `(Q)α m3Q (|f |)χEQ (x) > λ
 
Q∈S
S
for λ > 0 when we have a sparse family S, where EQ ≡ Q \ R
R∈S∩D(Q)\{Q}
for each Q ∈ S. We use Hölder’s inequality to have

1
sup `(Q)α mQ (|f |) u(Q0 ) q . [u, σ]Aαp,q kf kLp (σ1−p ) .
Q∈Q] (Q0 )

It thus remains to show that

1
sup λ (u(E)) q . [u, σ]Aαp,q kf kLp (σ1−p ) .
λ>0

We concentrate on the first term. We fix λ > 0 and calculate

X
u(E) = χ(λ,∞) (`(Q)α mQ (|f |))u(EQ ).
Q∈S
 Weighted Lebesgue spaces 313

We let S ∗ be the maximal disjoint family of cubes Q ∈ S satisfying

`(Q)α m3Q (|f |) > λ. Then
X
λq u(E) ≤ (`(Q)α m3Q (|f |))q u(Q).
Q∈S ∗

1−p 1
(p)
By Hölder’s inequality we have m3Q (|f |) ≤ m3Q (f · σ p )m3Q (σ) p0 . As a
result, we obtain
X h iq
α+ n (p) 1−p 1
λq u(E) . `(Q) p0 m3Q (f · σ p )σ(3Q) p0 u(Q)
Q∈S ∗
X  1−p
q
(p)
. [u, σ]Aαp,q q |Q|m3Q (f · σ p )
Q∈S ∗
  pq
X (p) 1−p
. [u, σ]Aαp,q q  |Q|m3Q (f · σ p )p  .
Q∈S ∗

Here, for the last line, we used q ≥ p. Since S ∗ is disjoint, we are in

1
the position of adding this inequality over Q ∈ S ∗ . Thus, λ (u(E)) q .
[u, σ]Aαp,q kf kLp (σ1−p ) .
We consider a special case in the light of the classical Hardy–Littlewood–
Sobolev theorem.
Definition 74 (Ap,q ). Let 1 < p ≤ q < ∞. Let w be a weight. One says that
(q) (p0 )
w ∈ Ap,q if [w]Ap,q ≡ sup mQ (w)mQ (w−1 ) < ∞. The class Ap,q collects all
Q∈Q
weights w for which [w]Ap,q is finite.
We consider the strong type inequalites.
Example 127. Let 1 < p, q < ∞ and α ∈ R. Then | · |α ∈ Ap,q if and only if
| · |αq ∈ A1+ pq0 , or equivalently − nq < α < pn0 .

We characterize the class of weights for which the operator f ∈ Lp (Rn ) 7→

wIα [w−1 f ] ∈ Lq (Rn ) is bounded.
Theorem 300. Let 1 < q ≤ p < ∞ and 0 < α < n. If w ∈ Aα
p,q , then
kIα f kLq (wq ) . kf kLp (wp ) for all f ∈ Lp (wp ).
Note that there is no other condition on p, q and α.
For the proof we use the following notation: We define
Z
D χQ (x)
Mw f (x) ≡ sup |f (y)|w(y)dy (x ∈ Rn )
Q∈D(Rn ) w(Q) Q

for a weight w and f ∈ L0 (Rn ).

314 Morrey Spaces
p
Proof Let β ≡ np − nq . We also set σ ≡ w− p−1 and v ≡ wq . By duality
and a sparse version of Iα (see Corollary 189), we have only to show
X Z
I≡ `(Q)α mQ (f ) g(x)dx . kf kLp (wp ) kgkLq0 (w−q0 )
Q∈S Q

0 0
for all f ∈ Lp (wp ) ∩ M+ (Rn ) and g ∈ Lq (w−q ) ∩ M+ (Rn ) and all sparse fam-
ilies S. Although we need to take into account the vector ~e, due to similarity,
we assume ~e = 0 and hence S ⊂ D(Rn ).
By Hölder’s inequality, we obtain
Z Z
X β 1 β 1
I= `(Q)α−n σ(Q)v(Q)1− n f (x)dx · v(Q) n g(x)dx
σ(Q) Q v(Q) Q
Q∈S
X β
≤ `(Q)α−n σ(Q)v(Q)1− n inf MσD [f · σ −1 ](x) inf Mv,β
D
[g · v −1 ](x).
x∈Q x∈Q
Q∈S

Arithmetic shows that 1 − nβ = p10 + 1q . If we insert the definition of [w]α Ap,q

into the above expression, then
X 1 1
I. σ(Q) p v(Q) p0 inf MσD [f · σ −1 ](x) inf Mv,β
D
[g · v −1 ](x).
x∈Q x∈Q
Q∈S

Since S is disjoint, we have

Z  p1 Z  10
p
D −1 p D −1 p0
I. Mσ [f · σ ](x) σ(x)dx Mv,β [g ·v ](x) v(x)dx .
Rn Rn

D
By Theorem 257, the universal estimates of MσD and Mv,β , we obtain
Z  p1 Z  10
0 0 q
I. f (x)p σ(x)−p σ(x)dx g(x)q v(x)−q v(x)dx
Rn Rn
= kf kLp (wp ) kgkLq0 (w−q0 ) .
Since Mα . Iα , we can transplant Theorem 300 to Mα .
Theorem 301. Let 1 < q ≤ p < ∞ and 0 < α < n. If w ∈ Aα
p,q , then
kMα f kLq (wq ) . kf kLp (wp ) for all f ∈ L0 (Rn ).

7.1.5 Extrapolation
We have established and will establish that the fundamental operators in
harmonic analysis behave well with respect to these classes of weights; see
Theorems 280, 289, 290 and 298 for the Hardy–Littlewood maximal operator
and Theorem 305 to follow for singular integral operators. When we consider
estimates of Ap -weights, we face the following type of estimate:
kT hkLp (W ) ≤ N ([W ]Ap )khkLp (W ) (h ∈ Lp (W ), W ∈ Ap ), (7.13)
 Weighted Lebesgue spaces 315

where T is a mapping from Lp (W ) to L0 (Rn ) and N ∈ M↑ [1, ∞). Extrapola-

tion is a technique to expand the validity of (7.13) for all 1 < p < ∞ based
on the validity of (7.13) for some p0 .
The following proposition describes more precisely how large the constant
should be when N (t) = tθ , t ≥ 0. We obtain a technique to create new
inequalities with the ingredient of weighted estimates.
Theorem 302. Let 1 < r < ∞ and θ ≥ 0. Let F ⊂ M+ (Rn )2 be the set of all
pairs (f, g) for which the inequality kgkLr (v) ≤ ([v]Ar )θ kf kLr (v) holds for all
v ∈ Ar . Let w ∈ Ap with 1 < p < ∞.
r−1
(1) For 1 < p ≤ r and (f, g) ∈ F, kgkLp (w) ≤ Cp,r ([w]Ap ) p−1 θ kf kLp (w) .
(2) For r ≤ p < ∞ and (f, g) ∈ F, kgkLp (w) ≤ Cp,r ([w]Ap )θ kf kLp (w) .
Proof If f = 0, then g = 0, so that there is nothing to prove in this case.
Thus, we may assume f 6= 0. We may also suppose p 6= r since the conclusion
is nothing but the assumption when p = r.
(1) We use Theorem 284. If we choose
∞
X M kf
v ≡ |f | + ,
2k (kM kLp (w)→Lp (w) )k
k=1

then M v ≤ 2kM kLp (w)→Lp (w) v thanks to the sublinearity of M , so that

[v]A1 ≤ 2kM kLp (w)→Lp (w) . Using Hölder’s inequality and the decompo-
p p
sition g = gv −1+ r · v 1− r , we have
Z  p1 − r1
p
kgkLp (w) ≤ v(x) w(x)dx kgkLr (v−r+p w) .
Rn

We will estimate each term. Since M is assumed on Lp (w),

Z  p1 − r1
p
p
v(x) w(x)dx ≤ (2kf kLp (w) )1− r .
Rn

Meanwhile, if w ∈ Ap and v ∈ A1 , then v −r+p w ∈ Ar with the estimate

[v −r+p w]Ar ≤ [v]A1 r−p [w]Ar thanks to Lemma 294. By the property of
F and Theorem 290,
kgkLr (v−r+p w) ≤ ([v −r+p w]Ar )θ kf kLr (v−r+p w)
. (kM kLp (w)→Lp (w) )(r−p)θ ([w]Ap )θ kf kLr (v−r+p w)
r−p
. ([w]Ap ) p−1 θ ([w]Ap )θ kf kLr (v−r+p w)
r−1
= ([w]Ap ) p−1 θ kf kLr (v−r+p w) .
p
Since |v| ≥ |f |, we have kf kLr (v−r+p w) ≤ (kf kLp (w) ) r . Thus,
r−1 p
kgkLr (v−r+p w) . ([w]Ap ) p−1 θ (kf kLp (w) ) r .
If we combine these estimates, then we obtain the desired result.
316 Morrey Spaces

(2) It suffices to show that

Z
g(x)h(x)w(x)dx ≤ Cp,r ([w]Ap )θ kf kLp (w) khkLp0 (w)
Rn
0
for all h ∈ (Lp (w) ∩ M+ (Rn )) \ {0}. With Example 120 in mind, let us
set
∞
X w−1 M k [wh]
H ≡ |h| + .
(D[w]Ap )k
k=0
Here, D p 1. Since h ≤ H and
p0 p0 p0 p(r − 1) p−r
H = H 1− r0 H r0 , r−r = r − p0 (r − 1) = r − = ,
r0 p−1 p−1
we have
Z
g(x)h(x)w(x)dx
Rn
Z
p0 p0
≤ g(x)H(x)1− r0 H(x) r0 w(x)dx (7.14)
Rn
Z  r1 Z  10
p−r r
r p0
≤ g(x) H(x) p−1 w(x)dx H(x) w(x)dx .
Rn Rn
Thanks to Theorem 290
1 1
[H · w]A1 . kM k − 1 − 1 ∼ ([w− p−1 ]Ap0 ) p0 −1 = [w]Ap .
Lp0 (w p−1 )→Lp0 (w p−1 )

Consequently, using Lemma 293, we deduce

p−r p−r r−1
[H p−1 w]Ar . ([H · w]A1 ) p−1 ([w]Ap ) p−1 . [w]Ap .
As a result, if we use the fact that (f, g) ∈ F, then
Z  r1
p−r
r
g(x) H(x) p−1 w(x)dx
Rn
Z  r1
p−r p−r
. ([H p−1 w]Ar )θ f (x)r H(x) p−1 w(x)dx (7.15)
Rn
Z  r1
p−r
. ([w]Ap )θ f (x)r H(x) p−1 w(x)dx .
Rn
If we use Hölder’s inequality, we obtain
Z  r1
p−r
f (x)r H(x) p−1 w(x)dx
Rn
Z r  r1
p−r

= f (x)H(x) r(p−1 ) w(x)dx (7.16)
Rn
Z  r1 − p1
p0
≤ kf kLp (w) H(x) w(x)dx .
Rn
 Weighted Lebesgue spaces 317

Since D  1,
Z  10
0 r p0
H(x)p w(x)dx . (khkLp0 (w) ) r0 . (7.17)
Rn

Combining (7.14)–(7.17) gives the desired result.

We now apply Theorem 302 to obtain the vector-valued inequalities for

Lp (w, `q ) once again.
Theorem 303. Let 1 < p < ∞, 1 < q ≤ ∞ and let w ∈ Ap . Then
k{M fj }∞ ∞ ∞ p q
j=1 kLp (w,`q ) . k{fj }j=1 kLp (w,`q ) for all {fj }j=1 ∈ L (w, ` ).

Proof When p = q or q = ∞, the result is clear from the Fatou lemma:

We may assume 1 < p, q < ∞. Let D  1. Thus, setting
  q1   q1  
 J J 
 X 1 X 
FJ ≡  |fj |q  ,  M fj q   : f1 , f2 , . . . , fJ ∈ L∞ (Rn
) ,
 
c

 j=1
D j=1 


we apply Theorem 302 to F ≡ FJ for each J ∈ N and then apply the monotone
convergence theorem. This is possible, because we have kgkLq (w) ≤ kf kLq (w)
for all (f, g) ∈ FJ and w ∈ Aq as long as D  1.
We present another application. As we did in Lemma 292, define
∞ X
X
Af ≡ mQj (f )χQj
k k
j=1 k∈Kj

for f ∈ L1loc (Rn ) as long as the sum makes sense.

Corollary 304. Let {Qjk }j∈N,k∈Kj be a sparse family of generalized dyadic

cubes with a level structure. Also let 1 < p < ∞ and w ∈ Ap . Then for all
. ([w] )max(1, p−1 ) kf k p .
1
f ∈ L∞ (Rn ), kAf k p
c L (w) Ap L (w)

Proof Simply combine Lemma 292 and Theorem 302.

7.1.6 A2 -theorem
So far we have seen how the constant behaves in the weighted maximal
estimate in Theorems 280, 289 and 290. We will consider the boundedness
property of singular integral operators. The behavior of the constant had been
unknown and it is known as the A2 -problem/A2 -conjecture. This was settled
down by Petermichl [355, 356] and then was expanded by Hytonen and Lerner
[207, 270]. Nowadays, this theorem is called the A2 -theorem. We establish the
following A2 -theorem here.
318 Morrey Spaces

Theorem 305 (A2 -theorem ). Let T be a singular integral operator. Then

kT f kL2 (w) . [w]A2 kf kL2 (w) for all f ∈ L∞ n
c (R ).

We will prove Theorem 305 in a couple of steps.

Combining Theorems 302 and 305, we obtain the Ap -version.
Theorem 306 (Ap -theorem). Let 1 < p < ∞. Let T be a singular integral
operator. Then kT f kLp (w) . [w]Ap max(1, p−1 ) kf kLp (w) for all f ∈ L∞
1
n
c (R ).

Earlier it was known that kT f kLp (w) .[w]Ap kf kLp (w) for all f ∈ L∞ n
c (R )
[205].
For the proof of Theorem 305, we need to set up notation. For a sparse
n
family S ≡ {Ukj }j∈N,k∈Kj ⊂ D ≡ D
Q

[aj , bj ) with a level structure and
j=1
∞ X
X
f ∈ L1loc (Rn ) write AD,S f ≡ mU j (f )χU j . To prove Theorem 305, we
k k
j=1 k∈Kj
use the following estimate together with Lemma 218:
Lemma 307. Let m, N ∈ N. Suppose that L ≡ {Qjk }j∈N,k∈Kj ⊂ D(Rn ) is
a sparse family with a level structure and that K0 = {k} is a singleton with
Q0k = [−2N , 2N )n for k ∈ K0 . Let L0 ≡ {Rkj }j∈N,k∈Kj ⊂ D(Rn ) be a family
such that Qjk ⊂ 3Rkj and that Rkj ⊂ 2m Qjk for all j ∈ N and k ∈ Kj . Write
L0 ≡ Rkj if L = Qjk . Fix f ∈ L∞ n + n
c (R ) ∩ M (R ). Define
X
AL,L0 ;m f ≡ mL0 (f )χL .
L∈L

(1) We have kAL,L0 ;m f kL2 . kf kL2 and kA∗L,L0 ;m f kL2 . kf kL2 , where the
implicit constants are independent of f and m.
(2) There exists a constant c(n) that depends only on n such that

λ|{x ∈ Rn : A∗L,L0 ;m f (x) > λ}| ≤ c(n)mkf kL1

for all λ > 0.

(3) For all cubes Q ∈ D(Rn ), ω2−n−2 (A∗L,L0 ;m f ; Q) . m · mQ (f ), where the
implicit constants are independent of f and m.
(4) For any N ∈ N and real-valued f ∈ L∞ n
c (R ) there exists another
j
sparse family S ≡ {Uk }j∈N,k∈Kj ⊂ D(Rn ) which depends on
f and N such that the pointwise estimate χ[−2N ,2N )n |A∗L,L0 ;m f −
Med(A∗L,L0 ;m f ; [−2N , 2N )n )| . M f + mAD,S f holds.
Proof The most serious part of the proof is to obtain a reasonable control
of the weak-(1, 1) constant of the adjoint operator.
 Weighted Lebesgue spaces 319

(1) We concentrate on the first formula, since the second one can be handled
completely in a similar manner or by the duality argument. To prove
the first formula, we dualize the left-hand side: Matters are reduced to
the inequality
Z
AL,L0 ;m f (x) · g(x)dx . kf kL2 kgkL2
Rn

for all f, g ∈ L∞ n + n
c (R ) ∩ M (R ). Written out in full, the left-hand side
becomes
Z X
AL,L0 ;m f (x) · g(x)dx = mL0 (f )mL (g)|L|.
Rn L∈L

[
We define EL ≡ L \ Qjk0 , if L = Qjk . Then keeping in mind that
k0 ∈Kj+1
{EL }L∈L is disjoint and that 2|EL | ≥ |L|, we have
Z X
AL,L0 ;m f (x) · g(x)dx ≤ 2 m2m L (f )mL (g)|EL |
Rn L∈L
Z
≤2 M f (x)M g(x)dx
Rn
≤ 2kM f kL2 kM gkL2 .

It remains to utilize the L2 (Rn )-boundedness of M .

(2) We ignore the contribution of [−2N , 2N )n ∈ {Q0k }k∈K0 because this term
can be handled with ease. Form the Calderón–Zygmund decomposition
of f as in Example 95. In particular, we obtain a family N ⊂ D(Rn )
satisfying
[ 1
Q . kf kL1 .
λ
Q∈N

Then it is easy to see that the good part g can be handled easily because
we already know that A∗L,L0 ;m is bounded on L2 (Rn ). To handle the bad
part b, we have only to show
 
 [ 
λ x ∈ Rn \ 15Q : |A∗L,L0 ;m b(x)| > λ ≤ c(n)mkf kL1 .
 
Q∈N

We decompose
X X X 1 Z 
A∗L,L0 ;m b = A∗L,L0 ;m bQ = bQ (y)dy χL0 .
|L0 | L
Q∈N Q∈N L∈L
320 Morrey Spaces
[
Here and below we fix x ∈ Rn \ 15Q. Let Q ∈ N and L ∈ D(Rn ).
Q∈N
If L ∩ Q = ∅ or L ⊃ Q, then mL (bQ ) = 0. Thus
Z 
X X 1
A∗L,L0 ;m b(x) = bQ (y)dy χL0 (x).
|L0 | L
Q∈N L∈L∩D(Q)

Suppose that x ∈ L0 and that L ∈ L and Q ∈ N satisfy L ( Q. Observe

that L0 is not contained in 15Q; otherwise
[
x ∈ L0 ⊂ 15Q, Q ∈ N , x ∈ Rn \ 15U,
U ∈N

which is a contradiction. Note also that both 3L0 and Q meet at L by

assumption. Thus, as is seen from the covering of 3L0 by 3n cubes in
D(Rn ) with the same size as L0 , L ( Q ⊂ 3L0 since L0 is not contained
in 15Q. Since `(L0 ) ≤ 2m `(L), we have
Z 
∗
X X 1
|AL,L0 ;m b(x)| ≤ |bQ (y)|dy χL0 (x).
|L0 | L
Q∈N L∈L∩D(Q)
|L|≥2−(m+3)n |Q|

Consequently, by considering cubes L0 satisfying the condition in the

summand
X X Z
kA∗L,L0 ;m bkL1 (Rn \ S 15Q) ≤ |bQ (y)|dy
Q∈N
Q∈N L∈L∩D(Q) L
|L|≥2−(m+3)n |Q|
X Z
≤ (m + 3) |bQ (y)|dy
Q∈N Q

≤ 2(m + 3)kf kL1 .

It remains to use the Chebyshev inequality.

(3) Since χQ (x)|A∗L,L0 ;m f (x) − c| ≤ A∗L,L0 ;m [|f |χQ ](x), we may assume
1
P R
mQ (|f |) > 0. Let c ≡ |L0 | f (z)dz. Note that the sum defin-
L∈L,Q⊂L L
ing c is convergent since f ∈ L∞ n
c (R ). Let λ0 > be a constant that is
specified shortly. Then we have
m
|{x ∈ Q : |A∗L,L0 ;m f (x) − c| > λ0 }| . kf χQ kL1 ,
λ0
as is seen from the above pointwise estimate and the previous part.
Let D  1, which will be specified shortly. If we take λ0 =
Dm · mQ (|f |), then λχQ |A∗ 0 f −c| (λ0 ) ≤ 2−n−2 |Q|. As a result,
L,L ;m
ω2−n−2 (A∗L,L0 ;m f ; Q) ≤ 2λ0 ∼ m · mQ (|f |).
 Weighted Lebesgue spaces 321

(4) Let Q be a fixed cube. By the Lerner–Hytönen formula, we have a sparse

family S ≡ {[−2N , 2N )n } ∪ {Ukj }j∈N,k∈Kj ⊂ D ∪ {[−2N , 2N )n } which
depends on f such that

χ[−2N ,2N )n |A∗L,L0 ;m f − Med(A∗L,L0 ;m f ; [−2N , 2N )n )|

X∞ X
. Mf + ω2−n−2 (A∗L,L0 ;m f ; Ukj )χU j
k
j=1 k∈Kj
∞ X
X
. Mf + m mU j (f )χU j
k k
j=1 k∈Kj

= M f + mAD,S f.

By the use of Lemmas 218, we obtain the weighted estimate of the operator
A∗L,L0 ;m .
Proposition 308. Maintain the notation in Lemma 307. Let w ∈ A2 . Then

kAL,L0 ;m f kL2 (w) . m[w]A2 kf kL2 (w) , kA∗L,L0 ;m f kL2 (w) . m[w]A2 kf kL2 (w)

for all f ∈ L∞ n
c (R ).

Proof We may assume that L is a finite set by the use of the monotone
convergence theorem and that f ∈ M+ (Rn ). It suffices to prove the second
formula thanks to the duality L2 (w−1 )-L2 (w) and the fact that w−1 ∈ A2
with [w−1 ]A2 = [w]A2 . Then we calculate

kχ[−2N ,2N )n (A∗L,L0 ;m f − Med(A∗L,L0 ;m f ; [−2N , 2N )n ))kL2 (w)

. kM f kL2 (w) + mkAD,S f kL2 (w)
. m[w]A2 kf kL2 (w)

thanks to Lemmas 218 and 307. We note Med(A∗L,L0 ;m f ; [−2N , 2N )n ) → 0 as

N → ∞, since L is finite. Thus, we are in the position of using Fatou’s lemma.
We transform Proposition 308 into the form which we use.
Corollary 309. Let D be a generalized dyadic grid. Also let L be a sparse
family and m ∈ N. Then for all f ∈ L∞ n
c (R ),

X
m2m L (f )χL . m[w]A2 kf kL2 (w) .
L∈L L2 (w)

n
Proof For each L ∈ L, there exist disjoint 2n -dyadic cubes L1 , L2 , . . . , L2
[2n
m
k m n
such that `(L ) = 2 `(L) for each k = 1, 2, . . . , 2 and that 2 L ⊂ Lk .
k=1
322 Morrey Spaces

Thanks to Proposition 308 we have

X
mL(k) (f )χL . m[w]A2 kf kL2 (w)
L∈L L2 (w)

for each k = 1, 2, . . . , 2n . If we add this estimate over k, we obtain the desired

result.
Before we prove the A2 -conjecture/theorem, we prove that the exponent
is sharp using the first Riesz transform R1 .
 
1
Proposition 310. Let 1 < p < ∞. Then the exponent max 1, in the
p−1
inequality kR1 f kLp (w) .n ([w]Ap )max(1, p−1 ) kf kLp (w) , f ∈ L∞
1
n
c (R ), is sharp.

Proof Let 1 < p ≤ 2. We use the weight w(x) = |x|(p−1)(n−δ) , x ∈ Rn .

Define f (x) ≡ |x|−n+δ χB(1) (x), x ∈ Rn . If x1 ≥ max(2, x2 , x3 , . . . , xn ), then

x1 − y1
Z
1
R1 f (x) = n+1
f (y)dy & .
Rn |x − y| δ|x|n

As a result,
! p1
|x|(p−1)(n−δ)
Z
1
kR1 f kLp (w) & dx & δ − p −1 .
x1 ≥max(2,x2 ,x3 ,...,xn ) |x|pn

Meanwhile ! p1
Z
1
kf kLp (w) = |x|−n+δ
dx ∼ δ− p .
B(1)

Since [w]Ap ∼ δ −p+1 , it follows that the exponent 1

p−1 is sharp.
Let 2 < p < ∞ and assume that we have

kR1 f kLp (w) .n ([w]Ap )A kf kLp (w) (f ∈ L∞ n

c (R ), w ∈ Ap ).

By duality

kR1 gk − 1 .n ([w]Ap )A kgk − 1 (g ∈ L∞ n

c (R ), w ∈ Ap ),
Lp0 (w p−1 ) Lp0 (w p−1 )

or equivalently,
A
kR1 gkLp0 (σ) .n ([σ]Ap0 ) p0 −1 kgkLp0 (σ) (g ∈ L∞ n
c (R ), σ ∈ Ap0 ).

This forces A ≥ 1, since p0 ∈ (1, 2).

 Weighted Lebesgue spaces 323

We now conclude the proof of Theorem 305. We may assume that T f is

real-valued because we can consider Re(T f ) and Im(T f ) separately. We have
only to establish

kχQ0 (T f − Med(T f ; Q0 ))kL2 (w) . [w]A2 kf kL2 (w)

for all cubes Q0 , where the implicit constant is independent of Q0 , since

Med(T f ; [−2N , 2N )) → 0 as N → ∞. At this step, we obtained a good can-
cellation of the function T f .
By the Lerner formula, we have
∞ X
X
χQ0 (x)|T f (x) − Med(T f ; Q0 )| . ω(T f ; Qjk )χQj (x)
k
j=1 k∈Kj

for almost every x ∈ Rn , where {Qjk }j∈N,k∈Kj is a sparse family with a level
structure. By Lemma 217, we have
∞ X
X ∞ X
χQ0 (x)|T f (x) − Med(T f ; Q0 )| . 2−l m2l Q0 (|f |)χQj (x)
k
l=0 j=1 k∈Kj

for almost every x ∈ Rn . At this step, the singular integral operator T was
controlled by a positive operator. Thus, by Corollary 309,
∞ X
X ∞ X
kχQ0 (T f − Med(T f ; Q0 ))kL2 (w) . 2−l m2l Q0 (|f |)χQj
k
l=0 j=1 k∈Kj
∞
X
. l · 2−l [w]A2 kf kL2 (w)
l=0
∼ [w]A2 kf kL2 (w) .

Thus, the A2 -conjecture is solved.

We end this section with the boundedness of singular integral operators in
the vector-valued setting:
Theorem 311 (Weighted vector-valued boundedness of singular integral
operators). Let 1 < q, r < ∞ and suppose w ∈ Aq . Also let T be a singu-
lar integral operator. Then for all {fj }∞ ∞ n
j=1 ⊂ Lc (R )),

k{T fj }∞ ∞
j=0 kLp (w,`q ) .q,r k{fj }j=0 kLp (w,`q ) .

Here to ignore completely the definability of T f with f ∈ Lq (w), we sup-

pose f ∈ L∞ n
c (R ).

Proof Simply apply Theorems 303 and 306.

A direct consequence of the Ap -theorem is that we can extend T to a
bounded linear operator on Lp (w).
324 Morrey Spaces

7.1.7 Exercises
Exercise 97. Go back to the definition of the Ap /Ap0 -characteristics to show
Lemma 287.
Exercise 98. Let w be a weight on Rn . Consider the uncentered maximal
operator M generated by cubes.
(1) Let f ∈ L0 (Rn ). Then show that f ∈ L1 (w)0 if and only if w−1 f ∈ L∞ .
Furthermore, in this case, kw−1 f kL1 (w)0 = kw−1 f kL∞ .
(2) Show that M maps L1 (w)0 boundedly if and only if w ∈ A1 . Further-
more, in this case, find the operator norm of M .
Exercise 99. Consider the uncentered maximal operator M generated by
cubes.
(1) Let f ∈ L0 (Rn ). Let λ > 0 and define Ω ≡ {x ∈ Rn : M f (x) > λ}.
Show that, for any compact set K contained in Ω, there exists a finite
collection {Qj }Jj=1 of cubes such that mQj (|f |) > λ for all j = 1, 2, . . . , J
SJ
and that K ⊂ Qj .
j=1

(2) Prove Theorem 289 for p > 1 using Hölder’s inequality and the 5r-
covering lemma.
ExerciseZ 100. [135, 450] Let w be a weight. Then show that w ∈ A∞ if and
only if M [wχQ ](x)dx . w(Q) for any cube Q.
Q

7.2 Two-weight norm inequality

So far, we considered weighted norm inequalities of operators whose range
and domains are the same. Here, we generalize the situation in this direction:
we consider two different weights or two different measures and consider the
boundedness of operators from a weighted Lebesgue space to another weighted
Lebesgue space. Of course it is impossible to do so only for a limited class of
operators. The Hardy operator H defined by (1.1) is one of such operators.
Section 7.2.1 deals with the Hardy operator. Section 7.2.2 considers frac-
tional maximal operators, while we consider singular integral operators in
Section 7.2.3.
 Weighted Lebesgue spaces 325

7.2.1 Weighted estimates for the Hardy operator

We recall that the Hardy operator is given by
1 t
Z
Hf (t) = f (s)ds (t > 0).
t 0
See Section 1.4.3. We will consider the weighted norm estimates for this oper-
ator. Since we can consider a more general setting of Radon measures instead
of weighted measures, we let µ be a Borel measure on (0, ∞). To avoid some
technical difficulty, we consider continuous Borel measures. We are now inter-
ested in the following type of estimate:
(Z Z q ) q1
∞ t
f (s)ds dµ(t) ≤ Ckf · wkLp (0,∞) ,
0 0

or equivalently
kHf kLp (tp dµ) ≤ Ckf kLp (wp ) ,
+
where f ∈ M (0, ∞). If p = q, thanks to Theorem 38 we have rich examples
of monomial weights. In the upper triangle case, 1 < p ≤ q < ∞, we can
completely characterize the boundedness of the integral operators.
Theorem 312. Let 1 < p ≤ q < ∞. Suppose that w ∈ M+ (0, ∞) and that µ
is a continuous Borel measure on (0, ∞). Then
Z ∞ Z v q  q1
f (t)dt dµ(v) . kf · wkLp (0,∞)
0 0

holds for any f ∈ M+ (0, ∞) if and only if

Z r  10
p 0 p
K ≡ sup q µ(r, ∞) w(v)−p dv < ∞.
r>0 0

We prove Theorem 312 after we prove a lemma. The following lemma can
be located as a sort of integration by parts.
Lemma 313. Let 1 < p < ∞ and µ be a continuous Borel measure. Then
Z
1 p
p0
p dµ(s) ≤ p p µ(t, ∞).
(t,∞)∩supp(µ) µ(s, ∞)
h s i
Proof Let s > t > 0 and N ∈ N. Then define tN (s) ≡ 2N t N . By the
2 t
monotone convergence theorem,
Z Z ∞
1 1
p0
p dµ(s) = lim lim p0
p dµ(s)
(t,∞)∩supp(µ) µ(s, ∞) ε↓0 N →∞ t µ(−ε + tN (s), ∞)
∞
X µ(t + 2−N (k − 1)t, t + 2−N kt)
≤ lim p0
p ,
N →∞
k=1
µ[t + 2−N kt, ∞)
326 Morrey Spaces

for each fixed ε > 0 since if ε > 2−N for N large enough. By the mean-value
theorem,
Z
1
p0
p dµ(s)
(t,∞)∩supp(µ) µ(s, ∞)
X∞ q q 
≤ lim p p µ(t + 2−N kt, ∞) − p µ(t + 2−N (k − 1)t, ∞)
N →∞
k=1
p
= p p µ(t, ∞).
We prove Theorem 312.
Proof The “only if” part is easy to check; simply use that
Z r Z ∞ Z r q  q1
p
q
f (t)dt · µ(r, ∞) = f (t)dt dµ(v) . kf · wkLp (0,∞)
0 r 0

for all r > 0 and resort to duality. Consequently, we concentrate on the “if”
part.
Z t  p10 p
−p0
Let h(t) ≡ w(s) ds for t ∈ (0, ∞). Then for all s > 0,
0

1 1 d 0 0 0 d
= h(s)pp = p0 ph0 (s)h(s)p p−p −1 = p0 h(s)p .
h(s)p0 w(s)p0 p 0
h(s) ds ds
Thus,
t t  p10
Kp0
Z Z
1 −p0
ds = p0 h(t)p = p0 w(s) ds ≤ p .
0 h(s) w(s)p0
p 0
0
q
µ(t, ∞)
Meanwhile, by Hölder’s inequality, we have
Z ∞ Z v q
f (t)dt dµ(v)
0 0
Z ∞ Z v q
= f (t)h(t)w(h)h(t)−1 w(t)−1 dt dµ(v)
0 0
Z ∞ Z s  pq Z s  q0
p p p 1 p
≤ f (t) h(t) w(t) dt dt dµ(s).
0 0 0 h(t)p0 w(t)p0
Recall that we are assuming that p ≤ q. By virtue of Minkowski’s inequality
Z s  q0
dt p
which is applied to the weighted measure p 0 p 0 dµ(s), we have
0 h(t) w(t)
Z ∞ Z s q  q1
f (t)dt dµ(s)
0 0
 # pq  p1
Z ∞
"Z
∞ Z s  q0
dv p 
≤ f (t)p h(t)p w(t)p dµ(s) dt .
 0 t 0 h(v)p0 w(v)p0 
 Weighted Lebesgue spaces 327

As a result,
Z ∞ Z s q  q1
f (t)dt dµ(s)
0 0
 ! pq  p1
Z ∞ Z
1 dµ(s) 
≤K p0 f (t)p h(t)p w(t)p p0
p dt .
 0 (t,∞)∩supp(µ) µ(s, ∞) 

Thus, from the definition of K and Lemma 313,

Z
dµ(s) p q
p 0 . p
µ(t, ∞) ≤ K p h(t)−q .
(t,∞)∩supp(µ)
p
µ(s, ∞)

Consequently,
(Z q ) q1
∞ Z t
f (s)ds dµ(t) . Kkf · wkLp (0,∞) .
0 0

We compare Theorem 38 with Theorem 312.

Example 128. Let 1 < p < ∞, and let w be a weight on (0, ∞). Then thanks
to Theorem 312
(Z Z p ) q1
∞ t
f (s)ds w(t)p dt . kf · wkLp (0,∞)
0 0

for all f ∈ M+ (0, ∞), if and only if

sup kwkLp (0,r) kw−1 kLp0 (r,∞) < ∞.

r>0

If furthermore w(t) = tα , t > 0, this inequality holds if and only if − p1 < α.

It should be noted that we can handle the operator H. By a change of
variables we can handle another integral operator. Theorem 314 can be located
as the dual of Theorem 312.
Theorem 314. Let 1 < p ≤ q < ∞. Suppose that w ∈ M+ (0, ∞) and that µ
is a continuous Borel measure on (0, ∞). Then
Z ∞ Z ∞ q  q1
f (t)dt dµ(t) . kf · wkLp (0,∞) (7.1)
0 t

holds for f ∈ M+ (0, ∞) if and only if

Z ∞  10
p 0 p
K ≡ sup q
µ(0, r) w(t)−p dt < ∞.
r>0 r
328 Morrey Spaces

As is seen by the proof below, Theorems 312 and 314 are equivalent.
Proof Let ν be a measure satisfying ν(E) = µ{t > 0 : t−1 ∈ E} for all
Borel sets E. Then
Z ∞ Z ∞ q Z ∞ Z ∞ q
f (s)ds dµ(t) = f (s)ds dν(t)
0 t 0 t−1
Z ∞ Z t q
−2 −1
= s f (s )ds dν(t).
0 0

Meanwhile the assumption reads as

Z ∞  10
p
−p0
p
K = sup q ν(r−1 , ∞) w(t) dt .
r>0 r

Consequently, we have
(Z q ) q1  p1
∞ Z t Z ∞ 2
g(s)ds dν(t) . (g(t)t p0 w(t−1 ))p dt
0 0 0

for all g ∈ M+ (0, ∞). In particular,

(Z q ) q1  p1
∞ Z t Z ∞
2
−2
s f (s −1
)ds dν(t) . (f (t−1 )t− p w(t−1 ))p dt .
0 0 0

By the change of variables once again, we conclude

Z ∞ Z ∞ q  q1
f (s)ds dµ(t) . kf · wkLp (0,∞) .
0 t

We have a complete charecterization in the case of 0 < p ≤ 1 and 0 < p ≤

q ≤ ∞ once we give up considering all non-negative measurable functions:
We can drop the assumption 1 < p. For f ∈ L0 (0, ∞), we write kf kLp (0,∞;µ)
for the Lp (µ)-norm. It might be strange to restrict monotone functions when
we consider the boundendness of operators. However, there are some cases
where the norm is a composition of another norm and a monotone function.
For example, the local Morrey-type space, which we are going to define later
in this book, is made up of the norm k · kΦλ,q (0,∞) and a monotone function.
We state the result in full generality.
Theorem 315. Suppose 0 < p ≤ 1 and 0 < p ≤ q < ∞. Let v, w :
(0, ∞) → [0, ∞) be locally integrable functions. Let µ, ν be σ-finite Borel
measures on (0, ∞) charging no point in (0, ∞). Then kHf kLq (0,∞;µ) .
kf kLp (0,∞;ν) for all f ∈ M↓ (0, ∞) with lim f (t) = 0 if and only if A ≡
t→∞
Z ∞  q1 Z t − p1
q −q
sup s max(s, t) dµ(t) dν(t) < ∞.
s>0 0 0
 Weighted Lebesgue spaces 329

The proof of Theorem 315 will be given after we prove two lemmas. First,
under our assumption on the function f , we estimate the Hardy operator H
so that we can understand why max in the definition of A comes into play.
Lemma 316. Let 0 < p ≤ 1. Also let f ∈ C 1 (0, ∞) ∩ M↓ (0, ∞) satisfy
 Z t p
1
lim f (t) = 0. Then Hf (t) . f (t) + − up f (u)p−1 f 0 (u)du for all
t→∞ t 0
t > 0.
Proof Thanks to the fundamental theorem of calculus and
 p1the fact that
Z ∞
1 t
Z
p−1 0
lim f (t) = 0, we have Hf (t) ' − f (u) f (u)du ds. We split
t→∞ t 0 s
the integral into two parts:

Hf (t)
 p1 Z  Z ∞  p1
1 t
Z  Z t
p−1 0 1 t p−1 0
. − f (u) f (u)du ds + − f (u) f (u)du ds.
t 0 s t 0 t

We calculate the second integral in the right-hand side to have

Z t Z t  p1
1 p−1 0
Hf (t) ∼ − f (u) f (u)du ds + f (t)
t 0 s
 p1
1 t
Z  Z t
p−1 0
= − χ(0,u) (s)f (u) f (u)du ds + f (t). (7.2)
t 0 0

We consider
(Z  Z
t t  p1 )p
p−1 0
I≡ − χ(0,u) (s)f (u) f (u)du ds .
0 0

Recall that 0 < p ≤ 1. By Minkowski’s inequality, we have

Z t Z t p
p−1 0

p1
I≤ −χ(0,u) (s)f (u) f (u) ds du
0 0
Z t Z t p
=− f (u)p−1 f 0 (u) χ(0,u) (s)ds du
0 0
Z t
=− up f (u)p−1 f 0 (u)du.
0

It remains to insert this inequality into (7.2).

Proof of Theorem 315 Once we assume (7.1), we readily have A < ∞;
simply let f = χ(0,t) for each t > 0 and take the supremum. We consider the
sufficiency part. By the montone convergence theorem we may assume that
330 Morrey Spaces

f ∈ L∞ (0, ∞) and that f (t) = 0 for t  1. By mollification, we can also

suppose that f ∈ C ∞ (0, ∞). Consequently from Lemma 316, we have
Z ∞  q1
q
Hf (t) dµ(t)
0
 q1 (Z  pq ) q1
Z ∞ ∞  Z t
q p−1 0 p −q
. f (t) dµ(t) + − f (s) f (s)s ds t dµ(t) .
0 0 0

For the first term, we use Lemma 13. Recall that 0 < p ≤ q < ∞. Apply-
ing Minkowski’s inequality to the weighted measures f (s)p−1 f 0 (s)sp ds and
t−q dµ(t), we find
(Z  Z q ) pq
∞ t p

− f (s)p−1 f 0 (s)sp ds t−q dµ(t)

0 0
Z ∞ Z ∞  pq
p−1 0 p −q
≤− f (s) f (s)s t dµ(t) ds.
0 s

If we use the definition of A and Fubini’s theorem once again, then

(Z  Z  pq ) pq
∞ t
p−1 0 p −q
− f (s) f (s)s ds t dµ(t)
0 0
Z ∞
≤ −Ap f (t)p−1 f 0 (t)ν(0, t)dt
0
Z ∞ Z t 
= −Ap f (t)p−1 f 0 (t) dν(s) dt
Z ∞0 0

∼ Ap f (t)p dν(t).
0

Thus, the estimate of the second term is valid.

The following theorem supplements Theorem 315. We can consider q = ∞.
Theorem 317. Let 0 < p ≤ 1, v ∈ L1loc ([0, ∞)) ∩ M+ ([0, ∞)), and let w :
(0, ∞) → (0, ∞) be a continuous function. Then sup w(t)Hf (t) . kf kLp (v) for
t>0
w(T ) min(T, t)
all f ∈ M↓ (0, ∞) with lim f (t) = 0 if and only if A ≡ sup 1 <
t→∞ T,t>0 T v(0, t) p
∞.
Proof The necessity part is trivial again: Simply test the conclusion on
f = χ(0,T ) , T > 0.
We concentrate on the sufficiency part. We can assume that f is smooth
by mollification and that f is bounded. Thanks to Lemma 316
 Z t  p1
w(t) p p−1 0
sup w(t)Hf (t) . sup f (t)w(t) + sup − u f (u) f (u)du .
t>0 t>0 t>0 t 0
 Weighted Lebesgue spaces 331

For the first term, we use (1.8). Applying the assumption A < ∞, we find
 Z t  p1
w(t) p−1 0 p
sup − f (u) f (p)(u)u du
t>0 t 0
 Z t  p  p1
p−1 0 p −1
≤ sup − f (u) f (p)(u)u sup s w(s) du
t>0 0 s≥u
 Z ∞ Z y   p1
p−1 0
≤A − f (y) f (y) v(s)ds dy
0 0
. Akf kLp (v) .

Finally we will list some sufficient conditions for which the Hardy operator
is bounded.
Theorem 318. Let 0 < θ2 ≤ ∞, and let v1 , v2 ∈ M+ (0, ∞).
(1) Let 1 < θ1 ≤ ∞. Then
(Z
∞  Z t θ1 ) θ11
v2 (t) f (s)v1 (s)ds dt . kf kLθ1 (0,∞) (7.3)
0 0

Z ∞ Z t  θθ20
+ θ2 θ10 1
for all f ∈ M (0, ∞) if A ≡ v2 (t) v1 (s) ds dt < ∞.
0 0

(2) Let θ1 = 1. Then inequality (7.3) holds if

!
Z ∞
θ2 θ2
v2 (t) sup v1 (s) dt < ∞. (7.4)
0 s∈[0,1]

!
∞
v1 (s)θ2
Z
θ2
(3) Let 0 < θ1 ≤ min(1, θ2 ) and suppose v2 (t) sup 1−θ1 dt <
0 s∈[0,t] sθ2 θ1

∞. Then inequality (7.3) holds for all f ∈ M↑ (0, ∞).

(4) If θ1 = ∞, then condition (7.4) is also necessary for (7.3).
Proof Let f ∈ M+ (0, ∞).
(1) By Hölder’s inequality, we have
Z ∞  Z t θ 2
v2 (t) v1 (s)f (s)ds dt
0 0
Z ∞ Z t  θθ2 Z t  θθ20
1
θ2 θ1 θ10 1
≤ v2 (t) f (s) ds v1 (s) ds dt
0 0 0
≤ A(kf kLθ1 (0,∞) )θ2 .
332 Morrey Spaces

(2) Simply mimic the previous case; we leave the proof for interested readers
as Exercise 101.
(3) Thanks to Lemma 14,
Z ∞  Z t θ 2 Z ∞ Z t  θθ2
1
θ2 θ1 −1 θ1
v2 (t) g(s)ds dt ≤ v2 (t) s g(s) ds dt.
0 0 0 0

So, we apply (2).

(4) Simply test the boundedness on g = v1 θ1 .
We apply the one dimensional results to the boundedness of operators
acting on functions defined in Rn . We consider the following examples:
Definition 75 (Generalized Hardy operator). Let ϕ ∈ M+ (0, ∞) and 0 <
α
α < n. The generalized Hardy operators Hϕ and Hα ϕ are defined by
Z
α f (t)
Hϕ f (x) ≡ |x|α−n ϕ(|x|) dt (x ∈ Rn )
B(|x|) ϕ(|t|)
and Z
f (t)
Hα
ϕ f (x) ≡ |x| α−n
ϕ(|x|) dt (x ∈ Rn ).
Rn \B(|x|) ϕ(|t|)
We do not deal with these operators, since these operators go beyond the
coverage of this book, although there are many research papers on them.

7.2.2 Two-weight norm inequality for fractional maximal

operators
Let 1 < p ≤ q < ∞, and let v, w be weights. We now seek a characterization
for the weighted inequality kMα f kLq (v) . kf kLp (w) , f ∈ L0 (Rn ) to hold. We
can reduce matters to the estimate kMαDa f kLq (v) . kf kLp (w) , f ∈ L0 (Rn ) and
a ∈ {0, 1/3}n . For simplicity we suppose a = (0, 0, . . . , 0), since other cases
can be handled in a similar manner.
Define Q] (Q0 ) ≡ {Q ∈ Q : Q ⊃ Q0 } for a cube Q0 ∈ D(Rn ). We recall
that thanks to Remark 7 for any f ∈ L∞ n + n
c (R ) ∩ M (R ), there exists a sparse
0 0
family S ⊂ D(Q ) such that, for all x ∈ Q ,
MαD f (x) . LD;S D
α f (x) + c∞,α (f ),
[
where EQ ≡ Q \ R,
R∈S,R(Q

α α
X
LD;S
α f ≡ |Q| n mQ (f )χEQ and cD
∞,α (f ) ≡ sup |Q| n mQ (f ).
Q∈S Q∈D ] (Q0 )

Although we are mainly concerned with the case a = (0, 0, . . . , 0), we state
our main result in terms of Mα instead of MαD .
 Weighted Lebesgue spaces 333

Theorem 319 (Cube testing for Mα ). Let v, w be weights, and let 0 ≤ α < n
1
and 1 < p ≤ q < ∞. Define σ ≡ w− p−1 , the dual weight of w. Then

kMα f kLq (v) . kf kLp (w) (7.5)

holds for all f ∈ L∞ n

c (R ) if and only if

1
kMα [σχQ0 ]kLq (v) . σ(Q0 ) p (7.6)

for all cubes Q0 .

Furthermore, conditions (7.5) and (7.6) both imply
α 1 1
|Q| n −1 σ(Q) p0 · v(Q) q . 1. (7.7)

In particular, condition (7.5) implies that w(x) 6= 0 for almost all x ∈ Rn and
that σ, v ∈ L1loc (Rn ).
Here, we tolerate the case α = 0. It is important that the fundamental
operator Mα is essentially replaced by a somewhat strange operator LSα , whose
role in other fields of mathematics is yet unknown. Also, it matters that LSα
depends on f .
Proof The “only if” part is easy to check. Simply test (7.5) on f = χQ0 σ.
Once we have (7.5), then (7.7) follows immediately from `(Q)α mQ (|f |)χQ ≤
Mα f for all f ∈ L0 (Rn ).
Let us check the “if” part. We will use the fractional maximal operator MαD
defined in Remark 7. Fix a cube Q0 ∈ D(Rn ) to show that kχQ0 MαD f kLq (v) .
kf kLp (w) with the constant independent of Q0 . As we mentioned, although we
need to consider all families Da with a ∈ {0, 1/3}n , we consider D = D(0,0,...,0)
only for simplicity.
P We αmay as well assume that f ∈ M+ (Rn ). We consider
LD;S
α : LD;S
α f = |Q| n m (f )χ
Q EQ and c D
∞,α based on Theorem 179, where
Q∈S
S ⊂ D(Rn ) is a sparse family. We write
X q
V ≡ `(Q)αq (mQ (σ)) χEQ v.
Q∈S

Note that V is almost everywhere positive on Q0 , so that we can suppose

Lq (V ) is a linear subspace of L0 (Q0 ). Thus, it suffices to show the boundedness
from Lp (σ) to Lq (V ) of the operator given by
X 1 Z 
S
m f≡ f (y)σ(y)dy χEQ . (7.8)
σ(Q) Q
Q∈S

It is easy to check that mS is bounded from L∞ (σ) = L∞ (Rn ) to L∞ (V ) =

L∞ (Rn ); see Exercise 102. Consequently, it remains to show that mS is
334 Morrey Spaces
q
bounded from L1 (σ) to WL p (V ) according to Example 67. To this end, we
will fix λ > 0 and h ∈ L∞ n
c (R ). If we set Sλ ≡ {Q ∈ S : |mQ (h)| > λ}, then
[
{x ∈ Rn : |mS h(x)| > λ} = EQ .
Q∈Sλ

Recall that S and Sλ are subsets of D(Rn ). We define Sλmax to be the set of
all maximal cubes in Sλ keeping this in mind. Then
[
{x ∈ Rn : |mS h(x)| > λ} ⊂ Q.
max
Q∈Sλ

Here, the right-hand side is actually a disjoint union, Thus, we have

X q
V {x ∈ Rn : |mS h(x)| > λ} ≤ `(Q)αq (mQ (σ)) v(Q)
max
Q∈Sλ
X Z
≤ Mα (χQ σ)(x)q v(x)dx
max
Q∈Sλ Q

≤ (kMα [χQ σ]kLq (v) )q .

We now invoke the assumption to obtain
   pq
X q [
V {x ∈ Rn : |mS h(x)| > λ} . σ(Q) ≤ σ 
p Q ,
max
Q∈Sλ max
Q∈Sλ

where we used q ≥ p for the second inequality. Since we know that λσ(Q) ≤
khχQ kL1 (σ) for all Q ∈ Sλmax by the definition of Sλ , it follows from the
disjointness of Sλmax that
q
khkL1 (σ) p

n S
V {x ∈ R : |m h(x)| > λ} . .
λ
Next, we will show that cD
∞,α (f )kχQ0 kLq (v) . kf kLp (w) . To this end, we
need to choose a cube Q ∈ D which contains Q0 and we need to show that
α
|Q| n m3Q (|f |)kχQ0 kLq (v) . kf kLp (w) .
α
To this end, since Q0 ⊂ Q, it is enough to show that |3Q| n m3Q (|f |)kχ3Q kLq (v) .
kf kLp (w) . However, this is already done by the previous estimation. In fact,
the family {3Q} is a sparse family.
Thus,
cD S
∞,α (f )kχQ0 kLq (v) + kχQ0 m [f · σ
−1
]kLq (V ) . kf kLp (w) + kf · σ −1 kLp (σ)
∼ kf kLp (w) .
Thus, the proof is complete.
Concerning condition (7.7), we have a clarifying remark.
 Weighted Lebesgue spaces 335

Remark 14. Assume that w, v ∈ M+ (Rn ) \ {0}. Then by the Lebesgue dif-
ferentiation theorem there exists x0 ∈ Rn such that
lim mB(x0 ,r) (σ) > 0, lim mB(x0 ,r) (v) > 0.
r↓0 r↓0

α 1 1 1 1
By condition (7.7), lim |B(x0 , r)| n − p + q mB(x0 ,r) (σ) p0 mB(x0 ,r) (v) q ≤ J < ∞.
r↓0
n n
Hence, the condition α ≥ − is necessary for the validity of (7.5).
p q

7.2.3 Two-weight norm inequality for singular integral

operators
We next investigate the necessary and sufficient conditions on W1 and W2
ensuring the validity of the inequality
kT f kLq (W2 q ) . kf kLq (W1 q ) , (7.9)
where W1 and W2 are radial functions non-negative and measurable on Rn and
the implicit constant is independent of f ∈ Lqloc (Rn ). Under some conditions
(7.10) or (7.11), we can also investigate the boundedness of singular integral
operators.
Theorem 320. Let 1 < q < ∞. Let W1 , W2 be radial weights satisfying
sup W2 (y) . W1 (x) for x ∈ Rn (7.10)
y∈B(4|x|)\B( 14 |x|)

or
W2 (x) . inf W1 (y) for x ∈ Rn . (7.11)
y∈B(4|x|)\B( 41 |x|)

Then
kT f kLq (W2 q ) ≤ Dkf kLq (W1 q ) (f ∈ L∞ n
c (R ))

holds if the following two conditions are satisfied:

J1 ≡ sup r−n kW2 kLq (B(r)) kM χB(r) W1 −1 kLq0 (Rn \B(2r)) < ∞ (7.12)
r>0

and
J2 ≡ sup r−n kW1 −1 kLq0 (B(r)) kM χB(r) W2 kLq (Rn \B(2r)) < ∞. (7.13)
r>0

Moreover, the sharp constant D∗ in the above satisfies D∗ . J1 + J2 , where

the implicit constants are independent of W1 and W2 .
Proof We may assume that f ∈ Lq (W1 q ) satisfies kf kLq (W1 q ) = 1 by
normalization. We decompose
Z ∞ Z
X
|T f (x)|q W2 (x)q dx = |T f (x)|q W2 (x)q dx.
Rn j=−∞ B(2j+1 )\B(2j )
336 Morrey Spaces

Thus, it suffices to show that

∞ Z
X
|T [χB(2j+2 )\B(2j−1 ) f ](x)|q W2 (x)q dx . 1, (7.14)
j=−∞ B(2j+1 )\B(2j )

∞ Z
X
|T [χRn \B(2j+2 ) f ](x)|q W2 (x)q dx . 1 (7.15)
j=−∞ B(2j+1 )\B(2j )

and
∞ Z
X
|T [χB(2j−1 ) f ](x)|q W2 (x)q dx . 1. (7.16)
j=−∞ B(2j+1 )\B(2j )

Assume for the time being (7.10). As for (7.14), we have

∞ Z
X
|T [χB(2j+2 )\B(2j−1 ) f ](x)|q W2 (x)q dx
j=−∞ B(2j+1 )\B(2j )

X∞ Z
≤ sup W2 (x)q · |T [χB(2j+2 )\B(2j−1 ) f ](x)|q dx
j+1 )\B(2j ) B(2j+1 )\B(2j )
j=−∞ x∈B(2
∞
X Z
. sup W2 (x)q · |f (x)|q dx
j+1 )\B(2j ) B(2j+2 )\B(2j−1 )
j=−∞ x∈B(2
∞ Z
X
≤ |f (x)|q sup W2 (z)q dx
j+2 )\B(2j−1 ) 1
j=−∞ B(2 4 |x|≤|z|≤4|x|

X∞ Z
. |f (x)|q W1 (x)q dx = 3.
j=−∞ B(2j+2 )\B(2j−1 )

Next we assume (7.11). Then, we have

∞ Z
X
|T [χB(2j+2 )\B(2j−1 ) f ](x)|q W2 (x)q dx
j=−∞ B(2j+1 )\B(2j )

X∞ Z
. inf W1 (x)q · |T [χB(2j+2 )\B(2j−1 ) f ](x)|q dx
x∈B(2j+1 )\B(2j ) B(2j+1 )\B(2j )
j=−∞
X∞ Z
. inf W1 (x)q · |f (x)|q dx
x∈B(2j+1 )\B(2j ) B(2j+2 )\B(2j−1 )
j=−∞
X∞ Z
= |f (x)|q inf W1 (z)q dx
B(2j+2 )\B(2j−1 ) z∈B(2j+1 )\B(2j )
j=−∞

≤ 3.

Thus, if either (7.10) or (7.11) holds, then we have (7.14).

 Weighted Lebesgue spaces 337

We establish (7.15). By the size condition it suffices to show that

∞ Z Z !q
X dy
|f (y)| n W2 (x)q dx . 1. (7.17)
j=−∞ B(2j+1 )\B(2j ) R n \B(2j+2 ) |y|

We set v1 (r) ≡ W1 (r, 0, . . . , 0) and v2 (r) ≡ W2 (r, 0, . . . , 0) for r ≥ 0. Let µ be

a measure on (0, ∞) given by
∞
X
µ(E) ≡ χE (2j )W2 q (B(2j+1 ) \ B(2j )).
j=−∞

Then (7.17) reads

Z ∞ Z ∞ Z  q
1
|f (Rω)|dσ(ω) dR dµ(t)
0 2r R
S n−1
Z ∞ Z 
n−1 q q
. r v1 (r) |f (rω)| dσ(ω) dr.
0 S n−1

Assumption (7.12) reads

1 n−1
−n
sup µ((0, r]) q k| · | q0 v1 −1 kLq0 (2r,∞) < ∞.
r>0

Hence we have
Z ∞ Z ∞ q Z ∞
F (s)ds dµ(s) . sq+n−1 F (2−1 s)q v1 (s)ds
0 r 0

for all F ∈ M+ (Rn ). Or equivalently we have

Z ∞ Z ∞ q Z ∞
−1
s F (s)ds dµ(s) . sn−1 F (s)q v1 (s)ds
0 2r 0

for all F ∈ M+ (Rn ). We note that (7.15) is a special case of

Z
F (r) ≡ |f (rω)|q dσ(ω) (r > 0).
S n−1

Thus, we obtain (7.15).

The proof of (7.16) is almost the same. By the size condition it suffices to
show that
∞ Z Z !q
X 1
n
|f (y)|dy W2 (x)q dx . 1. (7.18)
j=−∞ B(2 j+1 )\B(2j ) |x| B(2 j−1 )

We define the measue µ∗ by

∞
X Z
µ∗ (E) ≡ χE (2j ) |x|−q W2 (x)q dx.
j=−∞ B(2j+1 )\B(2j )
338 Morrey Spaces

Then (7.18) reads

!q
Z ∞ Z
|f (y)|dy dµ∗ (t) . 1.
0 B( 21 t)

By the polar coordinate we can rephrase (7.18) as

Z ∞ (Z 12 t  Z  )q
r n−1
|f (rω)|dσ(ω) dr dµ∗ (t)
0 0 S n−1
Z ∞ Z 
. |f (rω)| dσ(ω) rn−1 v1 (r)q dr.
q
0 S n−1

Hence, it suffices to show that

Z (Z 1  )q
∞ 2t
Z 
r n−1
|f (rω)|dσ(ω) dr dµ∗ (t)
0 0 S n−1
Z ∞ Z q
. |f (rω)|dσ(ω) rn−1 v1 (r)q dr.
0 S n−1

Note that our assumption (7.13) reads

Z ∞ Z 12 t !q  q1
Z ∞ n−1
q q q0
F (s)ds dµ∗ (t) . F (t) t q
v1 (t) dt
0 0 0

for all F ∈ M+ (Rn ), or equivalently,

Z 1 !q  q1
Z ∞ 2t
Z ∞ n−1
−n+1 ∗ q q q0 −q(n−1) q
s F (s)ds dµ (t) . F (t) t v1 (t) dt ,
0 0 0

which yields (7.18).

7.2.4 Exercises
Exercise 101. Prove Theorem 318(2) using kf kL1 (0,t) ≤ kf kL1 (0,∞) for all
f ∈ M+ (Rn ).

Exercise 102. Using the disjointness of {EQ }Q∈S , prove that mS , given by
(7.8), is bounded on L∞ (Rn ).
 Weighted Lebesgue spaces 339

7.3 Notes
Section 7.1
General remarks and textbooks in Section 7.1
The Ap condition initially appeared, in a somewhat different form, in
[369]. On the real line, Muckenhoupt characterized the characterization of
the weights for which the maximal operator M is bounded on Lp (w) if p > 1
and is weak Lp (w)-bounded if p ≥ 1 [324]. For singular integral operators,
Muckenhoupt, Rochberg and Wheenden obtained the same characterization
[205]. The same attempt was done for fractional integral operators in [326].
See standard textbooks [104, Chapter 7], [112], [146, Chapter IV], [147],
[156, Chapter 7], [231], [293, §1.4], [416, Chapter V], [382, §6.1.2] and [432] for
the class Ap . See [293, §3.4] for the class Ap,q .
See the textbooks [104, Chapter 7, §3], [146, Chapter IV, §5], [89] and [382,
§6.1.2.6] for the extrapolation. Especially we can find an exhaustive account
for the recent extrapolation theory in [89].

Section 7.1.1
Muckenhoupt initialized the theory of the A1 class [324]. Example 113 is
from [322, Remark 2.2] and [332, Example 2.3].
Theorem 280, which gives the information on the growth of the operator
norm, is from [206]. Theorem 284 is recorded in [206, Lemma 2.1].
Finally, Example 115 comes from [105, Corollary 4.4].

Section 7.1.2
Muckenhoupt initialized the theory of the Ap class [324]. Theorem 289 can
be found in [39, (2.6)]. See also the proof of [39, Theorem 2.5], where Buckley
applies the weak estimate to obtain the sharp strong estimate, which in turn
proves the sharpness of his weak estimate. See [39, 206, 324] for Theorem 290.
The proof of Theorem 290 in this book hinges on the idea of Lerner [268].
Example 120 is due to Lerner; see [274, Lemma 2.3].
Soria and Weiss showed that more general operators are bounded in [412].
This is a generalization of Theorem 288.

Section 7.1.3
Muckenhoupt initialized the theory of the A∞ class [324]. Theorem 298 is
due to Andersen and John [19, Theorem 3.1].
The reverse Hölder inequality for Ap , 1 < p < ∞ is proved by Coifman
and Fefferman in [82, Theorem IV].
The openness property for Ap , Theorem 297, is shown by Coifman and
Fefferman in [82, Lemma II] (the case of 1 < p < ∞) and [82, Theorem V]
(the case of p = ∞).
340 Morrey Spaces

We refer to [209, Theorem 2.3] for Theorem 295, the reverse Hölder inequal-
ity with precise constant.

Section 7.1.4
Muckenhoupt and Wheeden considered the boundedness of Iα and Mα
from Lp (v p ) to Lq (v q ). See [326] for Definition 74 and Theorem 300. Cruz-
Uribe and Perez considered the weak Lp (v p )-Lp (up ) boundedness of fractional
integral operators in [93, Theorem 1.1]. See Theorem 299.

Section 7.1.5
Curbera, Garcı́a-Cuerva, Martell and Pérez proved the extrapolation the-
orem, Theorem 302 [94].

Section 7.1.6
Coifman and Fefferman showed that kT f kLp (w) .[w]Ap kf kLp (w) for all f ∈
L∞ n
c (R ); see [82, Theorems I and III]. Later on, the estimate was improved.
The solution of this conjecture motivated the study of sharp dependence on
the Ap constants. Petermichl studied the Hilbert transform [355, 356]. Later
on, Hytönen and Lerner considered the decomposition of measurable functions
based on Fujii’s idea in [136]. See [86, 210, 262, 272, 273, 275] for more recent
advances for this field.
We followed [270] for the proof of Lemma 307. Lemma 307(1),(2) and (3)
correspond to [270, Proposition 3.1], [270, Lemma 3.2] and [270, Lemma 3.3].
Based on these steps, Lerner concluded his proof using Lemma 307(4) as we
did in the proof of Theorem 305. See [208] for more about this direction. See
[19, Theorem 5.2] for Theorem 311. See [262] for a more recent approach for
the A2 -theorem.
In Section 7.1.6, for 1 < p < ∞ we showed that w ∈ Ap is sufficient for
singular integral operators to be Lp (Rn )-bounded. See [82, Theorem II] for
the converse.

Section 7.2
General remarks and textbooks in Section 7.2
See the standard textbooks and lecture notes [112, Chapter 2], [146, Chap-
ter VI, §6], [147, Chapters 3, 4 and 9], [223, Chapter 2, §I], [231] and [235,
Chapters 4 and 6].

Section 7.2.1
Muckenhoupt considered the weighted inequality for Hardy operators; see
[325]. Theorems 312 and 314 are well known; see [35, 230, 309, 422, 437] for
Theorems 312 and 314, the case 1 < p ≤ q < ∞. Our formulation of Theorem
 Weighted Lebesgue spaces 341

312 is exactly [112, Lemma 2.7.1]. See [47, Lemma 5] for Theorem 318(1), (2),
(3) and [45, Lemma 6] for Theorem 318(4).

Section 7.2.2
Kerman and Sawyer obtained the testing condition in [227, 385].
See [227, Theorem 2.3] and [400, Theorem 1] for the counterpart to
fractional integral operators of Theorem 319. Hunt, Kurtz and Neugebauer
pointed out that this theorem characterizes the Ap -class in [204].

Section 7.2.3
See [47, Theorem 3] for Theorem 320, where they also proved the necessity
of the conditions for the Hilbert transform to be bounded.
We list the definition of some related operators.
Example 129. For suitable f ∈ M+ (Rn ), the weighted Hardy operator Uψ
is defined by Z 1
Uψ f (x) ≡ f (sx)ψ(s)ds (x ∈ Rn ). (7.1)
0

Example 130. For suitable f ∈ M+ (Rn ), the weighted Cesàro operator Vψ

is defined by
Z 1  
t
Vψ f (t) ≡ f s−n ψ(s)ds (x ∈ Rn ). (7.2)
0 s

Note that when ψ ≡ 1 and n = 1, the operator is reduced to the classical

Cesàro operator.
We can consider the commutator generated by Vψ and a function b ∈
BMO(Rn ):
Vψb f ≡ Vψ (b · f ) − bVψ f. (7.3)
Likewise, we can consider the commutator generated by Uψ and a function
b ∈ BMO(Rn ):
Uψb f ≡ Uψ (b · f ) − bUψ f. (7.4)
for the operator Uψ given by (7.1).
Chapter 8
Approximations in Morrey spaces

Chapter 8 defines several closed subspaces, two of which coincide. This implies
that Morrey functions are not easy to approximate with nice functions.
Section 8.1 shows that closure in a metric space generally amounts to
measuring the size of the sets which can be approximated in the topology
in question. Section 8.2 gives closed subspaces in terms of closure for actual
approximations in Morrey spaces as well as canonical approximations.

8.1 Various closed subspaces of Morrey spaces

Next, we consider various subspaces. Considering the subspaces makes
sense unlike Lebesgue spaces, since there are many non-dense subspaces. It
could not be better if we can approximate all Morrey functions as we did for
the functions in Lebesgue spaces. Unfortunately, this is not the case. Here, we
classify Morrey functions based on various approximations. More and more
attention has been paid to the closed subspaces of the Morrey space Mpq (Rn )

with 1 ≤ q < p < ∞. In particular, the closed subspaces M cpq (Rn ) and Mpq (Rn )
arise naturally. We define some closed subspaces in Section 8.1.1, all of which
enjoy the lattice property. Section 8.1.2 considers a special subspace which fails
to enjoy the lattice property. Section 8.1.3 investigates the mutual relationship
between them.

8.1.1 Closed subspaces generated by linear lattices

We deal with closed subspaces of Morrey spaces. As we will see, we can
generate closed subspaces starting from linear subspaces satisfying the lattice
property. The following is the key ingredient for us to create closed subspaces.

Definition 76 (U Mpq (Rn )).

(1) A linear subspace U (Rn ) ⊂ L0 (Rn ) enjoys the lattice property if g ∈ U
if f ∈ U and |g| ≤ |f |.

343
344 Morrey Spaces

(2) Let U (Rn ) ⊂ L0 (Rn ) be a linear space with lattice property. For 0 <
Mp n
q (R )
q ≤ p < ∞, define U Mpq (Rn ) ≡ U (Rn ) ∩ Mpq (Rn ) .
The spaces of U we envisage are the following. Due to its importance we
state them as the definitions.
∗
Definition 77 ( Mpq (Rn ), Mpq (Rn ), M
fpq (Rn ), Mpq (Ω) ). Let 0 < q ≤ p < ∞.

(1) The case where U (Rn ) = L∞ (Rn ): The bar subspace Mpq (Rn ) stands for
the closure of L∞ (Rn ) ∩ Mpq (Rn ) with respect to Mpq (Rn ).
(2) The case where U (Rn ) = L0c (Rn ) ≡ {f ∈ L0 (Rn ) : supp(f ) is compact }:
∗
The star subspace Mpq (Rn ) stands for the closure in Mpq (Rn ) of L0c (Rn )∩
Mpq (Rn ).
(3) The case where U (Rn ) = L∞ n ∞ n 0 n
c (R ) = L (R ) ∩ Lc (R ): Denote by
Mq (R ), the tilde subspace, the Mq (R )-closure of Lc (R ) = L∞
f p n p n ∞ n n
c (R )∩
p n
Mq (R ).
(4) Let Ω ⊂ Rn be a measurable set. Define L0 (Ω) to be the subset of
L0 (Rn ) which consists of all measurable functions which vanish almost
everywhere outside Ω. Let U (Rn ) = L0 (Ω). Then we obtain the closed
subspace Mpq (Ω) of all f ∈ Mpq (Rn ) for which f vanishes almost every-
where outside Ω.
Needless to say, L∞ n p n
c (R ) is dense in L (R ) for 0 < p < ∞. Hence, the
above definition is significant only when 0 < q < p < ∞.
Example 131. Let 0 < q ≤ q̃ ≤ p < ∞. We may take U (Rn ) = Mpq̃ (Rn ).
Note that U Mpq (Rn ) differs from Mpq (Rn ) thanks to Theorem 20.
Recall that the Morrey space Mpq (Rn ) with 0 < q ≤ p < ∞ is defined as
the set of all f ∈ Lqloc (Rn ) for which kf kMpq ≡ sup m(f, p, q; r) < ∞, where
r>0

Z ! q1
1 1
m(f, p, q; r) ≡ sup |B(x, r)| p−q q
|f (y)| dy (r > 0).
x∈Rn B(x,r)

Using m(f, p, q; r), we can consider the following subspaces.

(∗)
Definition 78 (V0 Mpq (Rn ), V∞ Mpq (Rn ), V (∗) Mpq (Rn ), V0,∞ Mpq (Rn )). Let
0 < q ≤ p < ∞.
(1) The vanishing Morrey space V0 Mpq (Rn ) at 0 is defined by
V0 Mpq (Rn ) ≡ {f ∈ Mpq (Rn ) : lim m(f, p, q; r) = 0}.
r↓0

(2) The vanishing Morrey space V∞ Mpq (Rn ) at infinity is defined by

n o
V∞ Mpq (Rn ) ≡ f ∈ Mpq (Rn ) : lim m(f, p, q; r) = 0 ,
r→∞
 Approximations in Morrey spaces 345

(3) The space V (∗) Mpq (Rn ) is defined as the set of all functions f ∈ Mpq (Rn )
Z !
q
such that lim sup |f (y)| χRn \B(N ) (y)dy = 0.
N →∞ x∈Rn B(x,1)

(∗)
(4) The space V0,∞ Mpq (Rn ) is defined as the intersection of these three
spaces.
(∗)
We will show that V0,∞ Mpq (Rn ) coincides with M
fpq (Rn ) in Theorem 326.
There are many closed subspaces. Here, we see that Lp (Rn ) is contained
in any of them.
Proposition 321. Let 0 < q ≤ p < ∞. Then

Lp (Rn ) ⊂ V0 Mpq (Rn ) ∩ V∞ Mpq (Rn ) ∩ V (∗) Mpq (Rn ).

∗
(∗) fp (Rn ) = Mp (Rn ) ∩ Mp (Rn ).
In particular, Lp (Rn ) ⊂ V0,∞ Mpq (Rn ) = Mq q q

Proof For example, to check that Lp (Rn ) ⊂ V0 Mpq (Rn ), we argue as fol-
lows: Let f ∈ Lp (Rn ). Then m(f, p, q; r) ≤ m(f, p, p; r) by Hölder’s inequality.
Since L∞ n p n
c (R ) is dense in L (R ), we have

m(f, p, p; r) ≤ m(f − g, p, p; r) + m(g, p, p; r)

≤ kf − gkLp + m(g, p, p; r)
1
≤ kf − gkLp + |B(x, r)| p kgkL∞ .

If we let r ↓ 0, then we obtain lim sup m(f, p, p; r) ≤ kf − gkLp . Since g ∈

r↓0
L∞ n
c (R ) is arbitrary, we obtain lim m(f, p, p; r) = 0.
r↓0

The closed subspaces of local Morrey spaces can be defined analogously to

global Morrey spaces. Here, we content ourselves with definitions.
∗
fpq (Rn )). Let 0 < q ≤ p < ∞.
Definition 79 (LMpq (Rn ), LMpq (Rn ), LM

(1) The local bar subspace LMpq (Rn ) denotes the closure of L∞ (Rn ) ∩
LMpq (Rn ) in LMpq (Rn ).
∗
(2) The local star subspace LMpq (Rn ) stands for the closure in LMpq (Rn ) of
L0c (Rn ) ∩ LMpq (Rn ).

fp (Rn ), the local tilde subspace, the LMp (Rn )-closure of

(3) Denote by LMq q
∞ n
Lc (R ).
What we can say for the closed subspaces in Morrey spaces can be applied
to the closed subspaces in local Morrey spaces.
n
Example 132. Let 0 < q < p < ∞, and let f (x) = |x|− p , x ∈ Rn .
346 Morrey Spaces

(1) Let E be the set defined in Example 11. Then χE ∈ LMpq (Rn ) \
∗
LMpq (Rn ).
∗
(2) χB(1) f ∈ LMpq (Rn ) \ LMpq (Rn ).
∗
(3) f ∈ Mpq (Rn ) \ (LMpq (Rn ) ∪ LMpq (Rn )).

8.1.2 Closed subspaces generated by the translation

As we will see, all these spaces above have the lattice property. Here, we list
an important closed subspace which does not enjoy the lattice property. The

Zorko subspace Mpq (Rn ) is one of such spaces and is used for the boundary
value problem in partial differential equations.

Definition 80 (Mpq (Rn )). Let 0 < q ≤ p < ∞. Then define the Zorko

subspace Mpq (Rn ) by
 

p n p n p n
Mq (R ) ≡ f ∈ Mq (R ) : lim f (· + y) = f in Mq (R ) .
y→0

The Zorko subspace is also called the diamond subspace.

As before, we are interested in the case of p 6= q. If q ≥ 1, then we have
the following characterization:

Theorem 322. Let 1 ≤ q ≤ p < ∞. Then f ∈ Mpq (Rn ) belongs to Mpq (Rn )
if and only if f belongs to the closure with respect to Mpq (Rn ) of the set of all
functions g ∈ C ∞ (Rn ) ∩ L∞ (Rn ) such that ∂ α g ∈ Mpq (Rn ) for all α ∈ N0 n .
Proof We start with a setup: Choose ψ ∈ Cc∞ (B(1)) ∩ M+ (Rn ) with
kψkL1 = 1 and define ψj (y) ≡ j n ψ(jy) for y ∈ Rn .
If f ∈ C ∞ (Rn ) ∩ L∞ (Rn ) satisfies ∂ α f ∈ Mpq (Rn ) for all multi-indexes α
with |α| = 1, then by the mean-value theorem
Z 1
f (x + y) − f (x) = y · gradf (x + ty)dt (x, y ∈ Rn ).
0
Hence kf (· + y) − f kMpq ≤ |y| · kgradf kMpq . Thus, if f is a member in the
 
closure of the set above, then f ∈ Mpq (Rn ). Conversely if f ∈ Mpq (Rn ), then
Z
kf − f ∗ ψj kMpq = f − f (· − y)ψj (y)dy
Rn Mp
q
Z Z
= f ψj (y)dy − f (· − y)ψj (y)dy .
Rn Rn Mp
q

Thus, kf − f ∗ ψj kMpq ≤ sup kf (· + y) − f kMpq → 0. As a result, f

y∈B(j −1 )
belongs to the closure with respect to Mpq (Rn ) of the set of all functions
ϕ ∈ C ∞ (Rn ) ∩ L∞ (Rn ) such that ∂ α ϕ ∈ Mpq (Rn ) for all multi-indexes α.
 Approximations in Morrey spaces 347

8.1.3 Inclusions in closed subspaces of Morrey spaces

Knowing that Mpq (Rn ) does not contain L∞ n
c (R ) as a dense subspace, we
are oriented to the difference between Mpq (Rn ) and Mfpq (Rn ). We now discuss
the difference of these subspaces we defined in Sections 8.1.1 and 8.1.2. Here
is an example of what we want to do.

fpq (Rn ) ⊂ Mpq (Rn ).
Theorem 323. Let 0 < q ≤ p < ∞. Then M
Proof Let f ∈ L∞ n p n p n
c (R ) ⊂ L (R ). Then f (· + y) → f in L (R ), as
p n ∞ n
y → 0. Hence, f (· + y) → f in Mq (R ) as y → 0, since Cc (R ) is dense in

Lp (Rn ). Consequently, we deduce f ∈ Mpq (Rn ).

fpq (Rn ) is a proper subspace of Mpq (Rn ). We now
Later we show that M
compare Mq (R ) with V0 Mpq (Rn ).
p n

Proposition 324. Let 0 < q ≤ p < ∞. Then Mpq (Rn ) ⊂ V0 Mpq (Rn ).
Proof Simply observe that L∞ (Rn ) ∩ Mpq (Rn ) ⊂ V0 Mpq (Rn ).
∗
We compare Mpq (Rn ) with V (∗) Mpq (Rn ).
∗
Proposition 325. Let 0 < q ≤ p < ∞. Then Mpq (Rn ) ⊂ V (∗) Mpq (Rn ).

Proof Simply observe that L0c (Rn ) ∩ Mpq (Rn ) ⊂ V (∗) Mpq (Rn ).
fpq (Rn ) = V (∗) Mpq (Rn ).
We will show that M 0,∞

fp (Rn ) = V (∗) Mp (Rn ).

Theorem 326. Let 0 < q ≤ p < ∞. Then Mq 0,∞ q

(∗) (∗)
Proof Since L∞ n p n fp n
c (R ) ⊂ V0,∞ Mq (R ), we have Mq (R ) ⊂ V0,∞ Mq (R )
p n
p n
by taking the closure with respect to Mq (R ). To verify the converse inclusion,
(∗)
we let f ∈ V0,∞ Mpq (Rn ). Fix R1 , R2 , S > 0 so that R1 < R2 . Then

kf − χ[0,R] (|f |)χB(R) f kMpq

≤ sup m(f, p, q; r) + sup m(f, p, q; r)
x∈Rn ,r≥R2 x∈Rn ,r≤R1

+ sup m(f, p, q; r)
x∈Rn \B(S),R1 ≤r≤R2

+ sup m(f − χ[0,R] (|f |)χB(R) f, p, q; r).

x∈B(S),R1 ≤r≤R2

We note that

sup m(f − χ[0,R] (|f |)χB(R) f, p, q; r)

x∈B(S),R1 ≤r≤R2

≤ C(R1 , R2 , S, r)kf − χ[0,R] (|f |)χB(R) f kLq (B(S+R2 )) .

348 Morrey Spaces

Thus, by the Lebesgue convergence theorem

lim sup kf − χ[0,R] (|f |)χB(R) f kMpq

R→∞
≤ sup m(f, p, q; r) + sup m(f, p, q; r) + sup m(f, p, q; r).
x∈Rn x∈Rn x∈Rn \B(S)
r≥R2 r≤R1 R1 ≤r≤R2

If we let R1 ↓ 0, R2 → ∞ and S → ∞, we have

lim sup kf − χ[0,R] (|f |)χB(R) f kMpq = 0.

R→∞

Since χ[0,R] (|f |)χB(R) f ∈ L∞

c (R n), we obtain the desired result.

8.1.4 Exercises
n
Exercise 103. Let 0 < q < p < ∞, and let f (x) ≡ |x|− p , x ∈ Rn .
(1) Disprove that f = lim χB(k) f holds in Mpq (Rn ).
k→∞

(2) Disprove that lim χB(k−1 ) f = 0 holds in Mpq (Rn ).

k→∞

(3) Conclude that an analogue of the Lebesgue convergence theorem fails

for Morrey spaces.
∗
Exercise 104. Let 0 < q < p < ∞. Show that LMpq (Rn ) and LMpq (Rn )
are different in the nsense that one is not contained in another.
n
Hint: Let
f (x) ≡ χB(1) (x)|x|− p , x ∈ Rn and g(x) ≡ χRn \B(1) (x)|x|− p , x ∈ Rn . Show
∗ ∗
that f ∈ LMpq (Rn ) \ LMpq (Rn ) and that g ∈ LMpq (Rn ) \ LMpq (Rn ).

8.2 Approximation in Morrey spaces

Here, we consider approximation in Morrey spaces. There are several stan-
dard approximations of functions; truncation of functions using compact sets,
truncation of functions using level sets, mollifications and so on. Here, we
characterize our closed subspaces using these approximations.
fpq (Rn ) defined in Definition 77. Section 8.2.2
Section 8.2.1 characterizes M
is what we want to do here. We investigate how we can approximate Morrey
functions by using other function spaces. With concrete examples of functions
Section 8.2.3 shows that the subspaces defined in Sections 8.2.1 and 8.2.2 are
all different.
 Approximations in Morrey spaces 349

8.2.1 fpq (Rn )

Characterization of M
fpq (Rn ) for 0 <
It is convenient to have equivalent characterizations of M
q ≤ p < ∞ keeping in mind that

f = lim χ(0,R) (|f |)χB(R) f (8.1)

R→∞

takes place in Lp (Rn ) and hence in Mpq (Rn ) for any f ∈ Lp (Rn ). When we
consider the boundedness of operators, the following observation is useful.

Proposition 327. Let 0 < q ≤ p < ∞. Then Lp (Rn ) is dense in M

fpq (Rn ).

Proof Since L∞ n fp n ∞ n
c (R ) is dense in Mq (R ) and Lc (R ) ⊂ L (R ) ⊂
p n

fp (Rn ), it follows that Lp (Rn ) is densely contained in M

M fp (Rn ).
q q

We may replace L∞ n ∞ n
c (R ) with Cc (R ).

Proposition 328. Let 0 < q ≤ p < ∞. Then, Cc∞ (Rn ) is densely contained
fpq (Rn ).
in M
Proof We have only to show that any function in L∞ n
c (R ) lies in the
∞ ∞ ∞
Mpq (Rn )-closureof Cc (R ). Let f ∈ Lc (R ). Then since Cc (Rn ) is dense
n n

in L (R ), f can be approximated by functions in Cc∞ (Rn ) in the topology of

p n

Lp (Rn ) and hence Mpq (Rn ).

We also have the following characterization of Mfp (Rn ), namely any func-
q
p n
tion in M
fq (R ) can be approximated by a linear combination of characteristic
functions over sets of finite measure. Denote by Simple(Rn ) the set of all such
functions.
fpq (Rn ).
Proposition 329. For 0 < q ≤ p < ∞, Simple(Rn ) is dense in M

Proof Let f ∈ M fp (Rn ), and let ε > 0. Then, from the definition of
q
Mq (R ) there exists g ∈ L∞
f p n n
c (R ) such that kf − gkMq < ε. Let E = supp(g).
p
∞
Choose a sequence {gk }k=1 of simple functions such that lim kg −gk kL∞ = 0.
k→∞
k
P k
P
Fix k ∈ N, and write gk = al χEl . Define fk ≡ gk χE = al χE∩El . Since
l=1 l=1
supp(fk ) ⊆ E, we have kg − fk kL∞ = k(g − gk )χE kL∞ ≤ kg − gk kL∞ . By using
the inequality
1
kg − fk kMpq ≤ kg − gk kL∞ kχE kMpq ≤ kg − gk kL∞ |E| p

and taking k → ∞, we obtain lim kg − fk kMpq = 0. Since

k→∞

kf − fk kMpq . kf − gkMpq + kg − fk kMpq < ε + kg − fk kMpq ,

350 Morrey Spaces

we have lim sup kf − fk kMpq . ε. Since ε > 0 is arbitrary, we conclude that

k→∞
lim kf − fk kMpq = 0, as desired.
k→∞

We characterize the space M fpq (Rn ) for 0 < q ≤ p < ∞. We will use
∞
{Ek }k=1 to denote an arbitrary sequence of measurable subsets of Rn . We will
write Ek → ∅ a.e., if the characteristic functions χEk converge to 0 pointwise
a.e. Namely, Ek → ∅ a.e. if and only if lim χEk (x) = 0 for a.e. x ∈ Rn . Notice
k→∞
that the sets Ek are not required to have finite measure.
Definition 81 (Absolutely continuous norm). Let 0 < q ≤ p < ∞. A function
f in Mpq (Rn ) has absolutely continuous norm in Mpq (Rn ) if kf χEk kMpq → 0
for every sequence {Ek }∞ k=1 satisfying Ek → ∅ a.e., in the sense that
lim χEk (x) = 0 for almost all x ∈ Rn .
k→∞

Here are functions having or not having the absolutely continuous norm.
Example 133. Let 0 < q < p < ∞.
(1) The function f = χ[0,1]n has the absolutely continuous norm. In fact
simply notice that kχEj f kMpq ≤ kχEj f kLp for any decreasing sequence
{Ej }∞
j=1 of measurable sets such that χEj ↓ 0 almost everywhere.
n
(2) The function f ≡ | · |− p χB(1) fails to have the absolutely continuous
norm. This can be checked by using the sequence {B(j −1 )}∞ j=1 .
n
(3) The function f ≡ | · |− p χRn \B(1) , x ∈ Rn , does not have the absolutely
continuous norm. This can be checked by using the sequence {Rn \
B(j)}∞j=1 .

fpq (Rn ) in terms of the absolute continuity of

We can also characterize M
the norm.
Theorem 330. Let 0 < q ≤ p < ∞ and f ∈ Mpq (Rn ). Then f has the
fpq (Rn ).
absolutely continuous norm if and only if f ∈ M
Proof We begin with a preliminary observation. Let {Fk }∞
k=1 be an arbi-
trary sequence for which Fk → ∅ a.e. Then it follows from the Lebesgue
dominated convergence theorem that

kχE χFk kMpq ≤ kχE χFk kLp → 0 as k → ∞

as long as E is a set of finite measure.

Let f ∈ Mfpq (Rn ) now. Then we can find a sequence of simple functions
{fk }k=1 such that lim kfk −f kMpq = 0. Let {Fk }∞
∞
k=1 be an arbitrary sequence
k→∞
for which Fk → ∅ a.e. Then

kχFj f kMpq ≤ kfk − f kMpq + kχFj fk kMpq ≤ kfk − f kMpq + kχFj fk kLp .
 Approximations in Morrey spaces 351

Letting j → ∞, we obtain

lim sup kχFj f kMpq ≤ kfk − f kMpq .

j→∞

Since k is arbitrary, we have lim kχFj f kMpq = 0. To prove the reverse inclu-
j→∞
sion, we assume that f ∈ Mpq (Rn ) has the absolutely continuous norm. From
the definition of the absolute continuity and the fact that

{|f | > R} ∪ (Rn \ B(R)) → ∅ (R → ∞),

we have lim kf − χ{|f |≤R}∩B(R) f kMpq = lim kχ{|f |>R}∪(Rn \B(R)) f kMpq = 0.
R→∞ R→∞
fp (Rn ).
We thus conclude f ∈ Mq

8.2.2 Approximations and closed subspaces

Given a Morrey function, we discuss whether the function is (canonically)
approximated in the topology of Morrey spaces. To discuss such a kind of
thing, we take up once again several subspaces. Here, we consider the relation
between various approximations and closed subspaces. The following proposi-
∗
tion relates the truncation by balls centered at the origin and Mpq (Rn ).
Proposition 331. Let 0 < q ≤ p < ∞. Then
∗ n o
Mpq (Rn ) = f ∈ Mpq (Rn ) : f = lim χB(R) f in Mpq (Rn ) .
R→∞

Proof Assume that f ∈ Mpq (Rn ) and lim kχRn \B(R) f kMpq = 0. Define
R→∞
fR ≡ χB(R) f . Then fR ∈ L0c (Rn ) and that lim kf − fR kMpq = 0. This shows
R→∞
∗
that f ∈ Mpq (Rn ).
∗
Conversely, assume that f ∈ Mpq (Rn ). Given ε ∈ (0, 1), there exist gε ∈
Lc (Rn ) ∩ Mpq (Rn ) such that kf − gε kMpq < ε. Choose Rε > 0 such that
0

supp(gε ) ⊂ B(Rε ). For any R > Rε , we have

|χRn \B(R) f | ≤ |χRn \B(R) gε | + |χRn \B(R) (f − gε )| ≤ |f − gε |.

Consequently, for all R > Rε , kχRn \B(R) f kMpq ≤ kf − gε kMpq < ε. This shows
that lim kχRn \B(R) f kMpq = 0.
R→∞

The following proposition relates the truncation by the level set of func-
tions and Mpq (Rn ).
Proposition 332. Let 0 < q ≤ p < ∞. Then
n o
Mpq (Rn ) = f ∈ Mpq (Rn ) : f = lim χ[0,R] (|f |)f in Mpq (Rn ) .
R→∞
352 Morrey Spaces

Proof One inclusion is clear from the definition of Mpq (Rn ). Let us
concentrate on another inclusion. Let f ∈ Mpq (Rn ). Choose {fj }∞j=1 ⊆
Mpq (Rn ) ∩ L∞ (Rn ) such that lim kf − fj kMpq = 0. Then, there exists a
j→∞
subsequence {fjm }∞ ∞
m=1 ⊆ {fj }j=1 such that

lim fjm (x) = f (x) (8.2)

m→∞
for a.e. x ∈ Rn .
We may assume f ≥ 0 by decomposing f suitably. Accordingly, we may
assume that fj ≥ 0. By replacing fj with min(fj , f ), we may assume fj ≤
f . If sup kfj kL∞ < ∞, then f ∈ L∞ (Rn ) and there is nothing to prove.
j∈N
Assume otherwise; sup kfj kL∞ = ∞. Since fj ≤ min(f, kfj kL∞ ) ≤ f , we have
j∈N
min(f, kfj kL∞ ) → f in Mpq (Rn ). Consequently, since f − χ[0,2kfj kL∞ ] (|f |)f ≤
2(f − min(f, kfj kL∞ )), we have χ[0,2kfj kL∞ ] (|f |)f → f in Mpq (Rn ).
fpq (Rn ).
We have another expression of M
Proposition 333. Let 0 < q ≤ p < ∞. Then
n o
fpq (Rn ) = f ∈ Mpq (Rn ) : f = lim χB(R)∩{|f |≤R} f in Mpq (Rn ) . (8.3)
M
R→∞

gpq (Rn )
Proof Denote by A the right-hand side of (8.3). As before A ⊂ M
gpq (Rn ).
from the definition of M
Conversely, suppose that f ∈ Mgpq (Rn ). Then we can find a sequence
∞ ∞
{fj }j=1 ⊂ Lc (R ) such that fj → f in Mpq (Rn ). Fix j ∈ N. By the quasi-
n

triangle inequality, we have

kf − χ{|f |≤R}∩B(R) f kMpq .q kf − fj kMpq + kfj − χ{|f |≤R}∩B(R) fj kMpq
.q kf − fj kMpq + kfj − χ{|f |≤R}∩B(R) fj kLp .
Thanks to the dominated convergence theorem kfj − χ{|f |≤R}∩B(R) fj kLp ↓ 0,
as R → ∞. We thus obtain f ∈ A.
By combining Propositions 331–333, we obtain a simple but useful char-
acterization.
∗
fp (Rn ) = Mp (Rn ) ∩ Mpq (Rn ).
Corollary 334. Let 0 < q ≤ p < ∞. Then Mq q

The diamond subspace Mpq (Rn ) is important because we can obtain the
natural initial condition of the partial differential equation if we use the func-

tions in Mpq (Rn ) as initial data.
Theorem 335. Let 1 ≤ q ≤ p < ∞ and f ∈ Mpq (Rn ). Then
| · −y|2
Z  
t∆ 1
e f≡ n exp − f (y)dy
(4πt) 2 Rn 4t

converges back to f in Mpq (Rn ) as t ↓ 0 if and only if f ∈ Mpq (Rn ).
 Approximations in Morrey spaces 353

Proof The “if” part is clear because et∆ f is smooth and satisfies
∂ [et∆ f ] ∈ Mpq (Rn ) for any multi-index α thanks to the property of con-
α

volution. Mimic the proof of Theorem 322 for the “only if” part. See Exercise
108 for more details.
The first part of the following corollary, whose proof we omit, is an analogy
to
Corollary 336. Let 1 ≤ q ≤ p < ∞.

(1) Let ψ ∈ Cc∞ (Rn ) and f ∈ Mpq (Rn ), then f ∗ ψ ∈ Mpq (Rn ).
(2) Let ψ ∈ Cc∞ (Rn ) ∩ M+ (Rn ) with kψkL1 = 1. Then f ∈ Mpq (Rn ) belongs

to Mpq (Rn ) if and only if ψj ∗ f → f in Mpq (Rn ) as j → ∞, where
ψj (x) ≡ j n ψ(jx) for x ∈ Rn and j ∈ N.

(3) If f ∈ Mpq (Rn ) and g ∈ L1 (Rn ), then f ∗ g ∈ Mpq (Rn ).
Proof
(1) Simply use ∂ α (f ∗ ψ) = f ∗ ∂ α ψ.
(2) Go through a similar argument to Theorem 335 using (1).
(3) Since kf ∗ gkMpq ≤ kf kMpq kgkL1 , we may assume that g ∈ Cc∞ (Rn ). In
this case, the assertion follows immediately from (1).

8.2.3 Examples of functions in closed subspaces

We have defined many closed subspaces in Morrey spaces. Here, we aim to
detect the difference between them using many functions constructed earlier.
We start with a function which belongs to all of our closed subspaces.
Example 134. Let 0 < q ≤ p < ∞. Then since χQ(1) ∈ Lp (Rn ),

χQ(1) ∈ V0 Mpq (Rn ) ∩ V∞ Mpq (Rn ) ∩ V (∗) Mpq (Rn ) ∩ Mpq (Rn ).
∗
fpq (Rn ) = V (∗) Mpq (Rn ) = Mpq (Rn ) ∩ Mp (Rn ).
In particular, χQ(1) ∈ M 0,∞ q
n
Next, we check which subspace | · |− p belongs to.
n
Example 135. Let 0 < q < p < ∞, and let f ≡ | · |− p ∈ Mpq (Rn ).
∗
(1) We check f ∈/ Mpq (Rn ). Let g ∈ L0c (Rn )∩Mpq (Rn ). Suppose that supp(g)
is contained in B(R). Then from the definition of f we have
1 1
kf kMpq = lim |B(r)| p − q kf kLq (B(r)\B(R))
r→ ∞

and hence
1 1
kf kMpq ≤ lim |B(r)| p − q kf − gkLq (B(r)) ≤ kf − gkMpq .
r→ ∞
354 Morrey Spaces

/ Mpq (Rn ). Let g ∈ L∞ (Rn ) ∩ Mpq (Rn ). Then

(2) We check f ∈
1 1 1
|B(r)| p − q kgkLq (B(r)) ≤ |B(r)| p kgkL∞ .
1 1
Thus, it follows that lim |B(r)| p − q kgkLq (B(r)) = 0. If we mimic the
r↓0
argument above, then we obtain
Z ! q1
1 1
sup |B(x, r)| p−q q
|f (y) − g(y)| dy ≥ kf kMpq .
(x,r)∈Rn+1
+
B(x,r)

This implies that f ∈ Mpq (Rn ) \ Mpq (Rn ).

(3) Since f ∈ Mpq (Rn ) \ Mpq (Rn ), f ∈

/Mfp (Rn ).
q
1 1 1 1
(4) Since |B(r)| p − q kf kLq (B(r)) = |B(1)| p − q kf kLq (B(1)) for r > 0, f ∈
/
V0 Mpq (Rn ).
/ V∞ Mpq (Rn ).
(5) Likewise f ∈

(6) Since f (x) → 0 as x → ∞, f ∈ V (∗) Mpq (Rn ).


/ Mpq (Rn ).
(7) Since kf (y + ·) − f kMpq & 1 for any y ∈ Rn \ {0}, f ∈
n
In Example 135 we considered the power function | · |− p . Let us see what
happens if we truncated the function by the unit ball B(1).
n
Example 136. Let 0 < q < p < ∞. Define f (x) ≡ |x|− p χB(1) (x) for x ∈ Rn .
∗
(1) f ∈ Mpq (Rn ) since f is compactly supported.
n
/ Mpq (Rn ) since f behaves like | · |− p near the origin.
(2) f ∈

(3) f ∈
/M / Mpq (Rn ).
fpq (Rn ) since f ∈
n
/ V0 Mpq (Rn ) since f behaves like | · |− p near the origin.
(4) f ∈
(5) f ∈ V∞ Mpq (Rn ) since f is compactly supported.

(6) f ∈ V (∗) Mpq (Rn ) since f is compactly supported.


/ Mpq (Rn ) since f ∈
(7) f ∈ / Mpq (Rn ).
n
In Example 135 we considered the power function | · |− p . Let us see what
happens if we remove the singularity at x = 0. Firstly, we consider the smooth
truncation.
n
Example 137. Let 0 < q ≤ p < ∞, and let f (x) ≡ (|x|2 + 1)− 2p for x ∈ Rn .
 Approximations in Morrey spaces 355
n ∗
(1) Since f behaves like | · |− p away from the origin, f ∈
/ Mpq (Rn ).

(2) Since f ∈ L∞ (Rn ) ∩ Mpq (Rn ), f ∈ Mpq (Rn ).

∗
/ Mpq (Rn ), f ∈
(3) Since f ∈ /Mfpq (Rn ).

(4) Since f ∈ Mpq (Rn ), f ∈ V0 Mpq (Rn ).

n
(5) Since f behaves like | · |− p away from the origin, f ∈
/ V∞ Mpq (Rn ).

(6) Since lim f (x) = 0, f ∈ V (∗) Mpq (Rn ).

x→∞

(7) Since f ∈ C ∞ (Rn ) ∩ L∞ (Rn ) and satisfies ∂ α f ∈ Mpq (Rn ) for all multi-

indexes α, f ∈ Mpq (Rn ).
n
We also consider the truncation of | · |− p using the indicator function
χRn \B(1) .
n
Example 138. Let 0 < q < p < ∞. Define f (x) ≡ |x|− p χRn \B(1) (x) for
x ∈ Rn . Then f ∈ Mpq (Rn ).
n ∗
(1) Since f behaves like | · |− p away from the origin, f ∈
/ Mpq (Rn ).

(2) Since f ∈ L∞ (Rn ) ∩ Mpq (Rn ), f ∈ Mpq (Rn ).

∗
/ Mpq (Rn ), f ∈
(3) Since f ∈ /Mfpq (Rn ).

(4) Since f ∈ Mpq (Rn ), f ∈ V0 Mpq (Rn ).

n
(5) Since f behaves like | · |− p away from the origin, f ∈
/ V∞ Mpq (Rn ).

(6) Since f (x) → 0 as x → ∞, f ∈ V (∗) Mpq (Rn ).

(7) If ψ ∈ Cc∞ (Rn ) satisfies χB(1) ≤ ψ ≤ χB(2) , then f − k ∈ Lp (Rn ) ⊂
 n 
Mpq (Rn ), where k(x) ≡ |x|− p (1 − ψ(x)) for x ∈ Rn . Since k ∈ Mpq (Rn ),

we have f ∈ Mpq (Rn ).
∞
S p
Let 0 < q < p < ∞. The indicator function of G ≡ (k − 1 + (k!) p−q , k +
k=1
p
(k!) p−q ) belongs to Mpq (R) similar to Example 12. Let us see which subspace
χG belongs to.
Example 139. Let 0 < q < p < ∞. Let G ⊂ R be as in Example 12.
According to Example 12, we learn that χGn belongs to Mpq (Rn ) ∩ L∞ (Rn ).
356 Morrey Spaces

(1) Using the fact that Gn is made up of infinitely many cubes of volume
fpq (Rn ), showing that
1, we will establish that χGn does not belong to M
∞
Mq (R ) is similar to L (R ) in some sense. In fact, let f ∈ L0c (R) ∩
p n n

Mpq (Rn ). Then kf − χGn kMpq ≥ 1, since

p p
supp(f ) ∩ (k − 1 + (k!) p−q , k + (k!) p−q )n 6= ∅

for large k ∈ N. In fact, taking such large k, we have

|f − χGn | ≥ χ p p .
(k−1+(k!) p−q ,k+(k!) p−q )n

Hence kf − χGn kMpq ≥ kχ p p kMpq = 1, which shows

(k−1+(k!) p−q ,k+(k!) p−q )n
that χGn ∈
/Mfpq (Rn ).

(2) Since χGn ∈ L∞ (Rn ) ∩ Mpq (Rn ), χGn ∈ Mpq (Rn ).

∗
/ Mpq (Rn ), f ∈ M
(3) Since χGn ∈ fpq (Rn ).

(4) Since χGn ∈ L∞ (Rn ) ∩ Mpq (Rn ), f ∈ V0 Mpq (Rn ).

(5) Since k! grows faster than any polynomial as k → ∞, χGn ∈ V∞ Mpq (Rn ).

/ V (∗) Mpq (Rn ) because Gn is the disjoint union of infinitely many

(6) χGn ∈
cubes of volume 1.
(7) A geometric observation shows that lim χGn (· + y) = χGn in Mpq (Rn ).
y→0

Thus, χGn ∈ Mpq (Rn ).
Example 139 is a positive function but here we consider the case of real
functions.
Example 140. Let
∞
X p p
H(t) = χ[0,1] (t − k + 1 − k p−q ) sin3 (2k+1 π(t − k + 1 − k p−q )) (t ∈ R).
k=1

Let G ⊂ R be the set given by (1.14).

∗
/ Mpq (R), since |H| and χG behave almost similarly.
(1) H ∈

(2) Since |H| ≤ χG ∈ Mpq (R), we see that H ∈ Mpq (R).

∗
/ Mpq (R), H ∈
(3) Since H ∈ /Mfp (R).
q

(4) Since H ∈ Mpq (R), H ∈ V0 Mpq (R).

/ V∞ Mpq (R).
(5) Since |H| and χG behave almost similarly, H ∈
 Approximations in Morrey spaces 357

/ V (∗) Mpq (R).

(6) Since |H| and χG behave almost similarly, H ∈
Z t
(7) Set S(t) ≡ sin3 sds for t ∈ R Then
0
∞
d X p p
H(t) = 2−k χ[0,1] (t−k +1−k p−q )S(2k (t−k +1−k p−q )) (t ∈ R).
dt
k=1

If ψ ∈ Cc∞ (R) satisfy χ(−1,1) ≤ ψ ≤ χ(−2,2) , then we learn |ψj ∗ H(t)| ≤

p p
O(2j−k ) using integration by parts if t ∈ [k − 1 + k p−q , k + k p−q ] and

/ Mpq (R).
k > j. Thus kH − ψj ∗ HkMpq & 1 for all j ∈ N. Thus H ∈
We use the self-similar increasing sequence for Mpq (Rn ) considered in
Example 11.
Example 141. Let 0 < q < p < ∞. Let
F ≡ {y + Ra1 + R2 a2 + · · · : {aj }∞ N 1 n
j=1 ∈ {0, 1} ∩ ` (N), y ∈ [0, 1] },
n n n
where R > 1 solves (1 + R) p − q 2 q = 1, so that each connected component of
F is a closed cube with volume 1.
∗
/ Mpq (Rn ), because F is the disjoint union of cubes of volume 1.
(1) χF ∈

(2) χF ∈ Mpq (Rn ) since χF ∈ L∞ (Rn ) ∩ Mpq (Rn ).

∗
(3) χF ∈
/M / Mpq (Rn ).
fpq (Rn ) since χF ∈

(4) χF ∈ V0 Mpq (Rn ) since χF ∈ Mpq (Rn ).

/ V∞ Mpq (Rn ) thanks to the choice of R.
(5) χF ∈

/ V (∗) Mpq (Rn ), since F is the disjoint union of cubes of volume 1.

(6) χF ∈

(7) f ∈ Mpq (Rn ). In fact, for |y| < min(1, R − 1), by Hölder’s inequality

kχF (· + y) − χF kMpq
1 1
. sup kχF (· + y) − χF kLq (Q) + sup |Q| p − q kχF (· + y) − χF kLq (Q) .
Q∈Q Q∈Q
|Q|=1 |Q|≤1

Let |y|  1. If |Q| = 1, then

kf (· + y) − f kLq (Q) ≤ kχ[0,1]n (· + y)f (· + y) − χ[0,1]n f kLq = o(1)
as y → 0. If |Q| ≤ 1, then
1 1 1
kχF (· + y) − χF kLq (Q) . |Q| q |(y + [0, 1]n ) \ [0, 1]n | q = |Q| q o(1).
Thus, lim χF (· + y) = χF in Mpq (Rn ).
y→0
358 Morrey Spaces

8.2.4 Exercises
Exercise 105. See what happens if we use Example 15. Let 0 < q ! < p <
  N
q q
. Also let f = ⊗n
P
∞, and let a ∈ − ,− χ(j−1,j−1+j a ) be a
p−q p j=1
function in Example 15.
∗
/ Mpq (Rn ).
(1) Show that f ∈

(2) Show that f ∈ Mpq (Rn ).

(3) Show that f ∈

/Mfpq (Rn ).
1
(4) By proving kf χQ kMpq ≤ N − p for all cubes Q with `(Q) ≤ N −1 , show
that f ∈ V0 Mpq (Rn ).
/ V∞ Mpq (Rn ).
(5) Show that f ∈

/ V (∗) Mpq (Rn ).

(6) Prove that f ∈

(7) Prove that f ∈ Mpq (Rn ).
∗
/ Mpq (Rn )
Exercise 106. In Example 135 alternatively we can verify that f ∈
as follows: Fix R > 0. Define B ≡ B((2R, 0, . . . , 0), R).
(1) Show that B ⊆ Rn \ B(R) and that |x| ≤ 3R whenever x ∈ B.
Z  q1
1 1
(2) Establish that |B| p−q q
|χRn \B(R) (x)f (x)| dx & 1 for all R > 0
B
and hence lim inf kχRn \B(R) f kMϕq & 1
R→∞

∗
/ Mϕ
(3) Use Proposition 331 to conclude that f ∈ n
q (R ).

Exercise 107. Let 0 < q < p < ∞, and let ψ ∈ Cc∞ (Rn ) be such that
−n
χB(1) ≤ ψ ≤ χB(2) . Let f (x) ≡ (1 − ψ(x))|x| p for x ∈ Rn .
∗ n
/ Mpq (Rn ): Hint: f behaves like | · |− p .
(1) Show that f ∈

(2) Show that f ∈ Mpq (Rn ): Hint: Check that f ∈ L∞ (Rn ) ∩ Mpq (Rn ).

(3) Show that f ∈

/Mfp (Rn ). Hint: We showed f ∈ Mp (Rn ) in part (1).
q q

(4) Show that f ∈ V0 Mpq (Rn ) using f ∈ L∞ (Rn ) ∩ Mpq (Rn ).

n
/ V∞ Mpq (Rn ). Hint: f behaves like | · |− p .
(5) Show that f ∈
n
(6) Show that f ∈ V (∗) Mpq (Rn ) since f (x) = O(|x|−|α|− p ) as |x| → ∞.
 Approximations in Morrey spaces 359
 n
(7) Show that f ∈ Mpq (Rn ). Hint: f is smooth and ∂ α f (x) = O(|x|−|α|− p )
for all α ∈ N0 as |x| → ∞.
Exercise 108. Let 1 ≤ q ≤ p < ∞ and f ∈ Mpq (Rn ). Assume ∂ γ f ∈ Mpq (Rn )
for all multi-indexes γ. Let 0 < t ≤ 1 for the purpose of considering the limit
as t ↓ 0.
(1) Assume for the time being f ∈ C ∞ (Rn ).
Z t
(a) Show that 1 − et∆ f (x) = es∆ ∆f (x)ds.


0
| · |2
 
1
(b) Verify that p exp − ∈ L1 (Rn ) for all t > 0.
(4πt)n t
Z t
(c) Show that es∆ ∆f ds ≤ t k∆f kMpq using the fact that es∆
0 Mp
q
is the convolution operator.
(d) Conclude that et∆ f converges back to f in Mpq (Rn ) as t ↓ 0.

(2) For general f ∈ Mpq (Rn ) and for any k ∈ N, we have fk ∈ C ∞ such that
for all multi-indexes γ, ∂ γ fk ∈ Mpq (Rn ) and that

kf − fk kMpq < 2−k . (8.4)

(a) Find a small number T (k) > 0 such that for all t ∈ (0, T (k)),

fk − et∆ fk Mp
< 2−k (8.5)
q

holds.
(b) Show that f − et∆ f Mp
≤ 2kf − fk kMpq + fk − et∆ fk Mp
q q

(c) By combining this fact with (8.4) and (8.5), show that

f − et∆ f Mp
< 22−k ,
q

which implies that et∆ f goes to f in Mpq (Rn ) as t tends to zero.

8.3 Notes
Section 8.1
General remarks and textbooks in Section 8.1
As we have seen, there are many non-trivial and important non-dense
subspaces. This aspect seems to have been untouched.
360 Morrey Spaces

Section 8.1.1
The definition of vanishing Morrey spaces at 0 goes back to the work of
Vitanza [444]; see Definition 78. We refer to [361] for commutators acting on
these spaces. See [177, Definition 1.1], [467, Definition 2.23] and [398, Defini-
tion 4.5] for bar subspaces, star subspaces and tilde subspaces, respectively.
Density of simple functions in M fp (Rn ) can be found in [65, Remark 3.2]; see
q
Proposition 329. See [65, Lemmas 3.3 and 6.4] for different characterizations
of Mfpq (Rn ). As an application of the theory of closed subspaces in Morrey
spaces, we can obtain some characterization of closed subpsaces; see [65, Lem-
mas 4.1, 4.2 and 6.2]. We refer to [467, Theorem 2.29] for more. The Morrey
space Mpq (Rn ) does not have Cc∞ (Rn ) as a dense closed subspace; see [438,
Proposition 2.16].
Vanishing Morrey spaces at 0 are studied from various aspects. See [14] and
[15] for fractional integral operators and Marcinkiewicz intergrals associated
with Schrödinger operators, respectively. Cao and Chen handled Toeplitz-type
operators in [61].

Section 8.1.2
Zorko explained that the set of smooth functions in Morrey spaces cannot
approximate Morrey spaces [472, p. 587]. Actually as in Theorem 322, she
pointed out in [472, Proposition 3] that the translation and the convolution

are closely related. We refer to [467, Definition 2.23] for Mpq (Rn ) and to [398,
Theorem 4.3] for M cp (Rn ). See [65, Lemma 2.1] as well as [180, Theorem 6]
q
for Theorem 322. The space L̃p,α (Rn ) in [34] is nothing but the diamond
subspace. See [33, 56] for more.

Section 8.1.3
We refer to [180, Lemma 3] for Proposition 324, where we showed
Mpq (Rn ) ⊂ V0 Mpq (Rn ).

Section 8.2
General remarks and textbooks in Section 8.2
We can find an exhaustive account for absolutely continuous norms in [29,
Chapter 3]. Adams pointed out that any element in diamond subspaces can
be approximated by the smooth functions in his textbook [6].

Section 8.2.1
See [21, Theorem 1.1] for the approximation of Morrey functions by means
of the Fourier-Jacobi expansions. Note that such an expansion goes back to
the original function in the norm topology of Morrey spaces if and only if
 Approximations in Morrey spaces 361

the original function belongs to closed subspaces of Morrey spaces. See [398,
Theorem 4.6] for Theorem 330.

Section 8.2.2
Recall that we have charecterized Mpq (Rn ) in Theorem 332. See [65,
Lemma 3.1] for a different characterization of Mpq (Rn ).
Israfilov and Tozman considered the approximation by means of Morrey-
Smilnov classes [213].

Section 8.2.3
Example 140 can be found [181, Example 1.4]. See [277] for analytic Morrey
spaces.
Chapter 9
Predual of Morrey spaces

This chapter is devoted to considering the predual. The dual of Morrey spaces
contains an element which cannot be expressed in terms of locally integrable
functions. However, we show that Morrey spaces are realized as the dual of
Banach function spaces. Furthermore these Banach function spaces are again
realized as the dual of Banach function spaces. One of the reasons why such a
situation occurs is that Morrey spaces are wider than Lebesgue spaces. Hence,
there is difficulty in defining some linear operators.
Given a Banach space X , recall that the dual space X ∗ is made up of
all bounded linear functionals. For x∗ ∈ X ∗ , its dual norm is expressed by
kx∗ kX ∗ ≡ sup{|x∗ (x)| ∈ R : x ∈ X , kxkX = 1} by definition. The space X ∗
k · kX ∗ is a Banach space under the dual norm. Thus, the notation X ∗∗ always
makes sense. Given x ∈ X, we can define Q(x) ∈ X ∗∗ by Q(x)(x∗ ) ≡ x∗ (x) for
x∗ ∈ X ∗ . According to the Hahn–Banach theorem, Theorem 87, the canonical
mapping Q : x ∈ X 7→ Q(x) ∈ X ∗∗ is isometrc. If this isometry Q is surjective,
then we say that X is reflexive. See [310, 1.11.6] for example.
Chapter 9 is interested in predual spaces of Morrey spaces. In functional
analysis, duality plays a key role in justifying the convergence of sequences.
However, the dual space of the Morrey spaces Mpq (Rn ) with 1 < q < p < ∞
cannot be described via coupling. The dual space differs from the Köthe dual.
This is a big difference from Lebesgue spaces.
Section 9.1 introduces the predual of Morrey spaces. The predual of Morrey
spaces has some equivalent characterizations, which are investigated in Section
9.2.

9.1 Predual of Morrey spaces

Having investigated functions which belong to Morrey spaces, we are now
oriented to their dual spaces and their reflexivity. As Example 148 shows, the
Morrey space Mpq (Rn ) with 1 < q < p < ∞ is not reflexive. Consequently,
we are more interested in the functionals on Mpq (Rn ) with 1 < q < p < ∞.
Like a certain functional of `∞ (N) known as the Banach limit, we can find a
similar functional in the Morrey space Mpq (Rn ) with 1 < q < p < ∞.

363
364 Morrey Spaces

We start with defining the block spaces in Section 9.1.1 and then we
introduce predual spaces. Section 9.1.2 mainly considers subspaces in predual
spaces. Section 9.1.3 deals with duality. The Fatou property of the function
spaces is one of the important properties. As it turns out, among others it is
difficult to check the Fatou property of predual spaces. We will actually show
that predual spaces enjoy this property in Section 9.1.4. The Fatou prop-
erty of predual spaces has an application. We can specify the Köthe dual of
Morrey spaces in Section 9.1.5. We consider the theory of decomposition in
Section 9.1.6. In particular, as an application of the (pre)duality, we develop
a technique to control the Morrey norm in Section 9.1.6.

9.1.1 Definition of block spaces and examples

Section 9.1.1 seeks to show that the Morrey space Mpq (Rn ) with 1 < q ≤
p < ∞ can be realized as the dual of a certain Banach space. The space will
be called the block space or the Zorko space and we investigate the lattice
property and a standard expression of functions in this space.
Before we come to the precise definition of the Zorko space, let us recall
the norm expression of Lp (µ) for 1 ≤ p ≤ ∞ over a σ-finite measure space
(X, B, µ), which we restate; see Theorem 9 as well.
Theorem 337 (Norm expression of Lp (µ)). Assume that (X, B, µ) is a σ-
finite measure space. Let f be a function that is integrable on any set of finite
measure. Then, for all 1 ≤ p ≤ ∞, we have
Z
1
kf kLp (µ) = sup f (x)g(x)dµ(x) ,
g kgkLp0 (µ) X
0
where g runs over all g ∈ L∞ (µ) ∩ Lp (µ) \ {0} such that µ{g 6= 0} < ∞.
We omit its proof, since there are many textbooks; see [319, Theorem 12.7],
for example.
We state a direct consequence of Theorems 9 and 89 together with the
Banach Alaoglu theorem.
Corollary 338. Let (X, B, µ) be a σ-finite measure space. Let 1 < p < ∞
and {fj }∞ p
j=1 be a bounded sequence in L (µ). Namely, we suppose that there
exists M > 0 such that kfj kLp (µ) ≤ M . Then we can find f ∈ Lp (µ) and a
subsequence {fjk }∞k=1 such that
Z Z
lim fjk (x)g(x)dµ(x) = f (x)g(x)dµ(x)
k→∞ X X
0
p
for all g ∈ L (µ).
We intend to consider the counterpart of Theorems 337 to Morrey spaces.
We have by definition
Z ! q1
1 1
kf kMpq = sup |Q(x, r)| p − q |f (y)|q dy .
(x,r)∈Rn+1
+
Q(x,r)
 Predual of Morrey spaces 365

Of course we can start with the definition using balls. However, for the sake
of simplicity in later discussions, we use cubes. Note that this can be written
as follows: Z
kf kMpq ≡ sup sup |f (x)A(x)|dx,
(x,r)∈Rn+1
+
A Rn
0
where A moves over all Lq (Rn )-functions supported on a cube Q(x, r) and
1 1
satisfying kAkLq0 ≤ |Q(x, r)| p − q . Our observation justifies the following defi-
nition.
Definition 82 ((p, q)-block). Let 1 ≤ q ≤ p < ∞. A function A ∈ L0 (Rn ) is
a (p, q)-block if there exists a compact cube Q that supports A and
1
− p10
kAkLq0 ≤ |Q| q0 . (9.1)
If we need to specify Q, then we say that b is a (p, q)-block supported on Q.
Some prefer to call (p, q)-blocks (p0 , q 0 )-blocks. Here we will need to con-
sider the predual space, so we have to adopt the name “(p, q)-block”.
− p10
Example 142. Let Q be a cube, and let 1 ≤ q ≤ p < ∞. Then |Q| χQ is
a (p, q)-block.
Other examples of (p, q)-blocks, more precisely how to create (p, q)-blocks
in a standard manner, will be presented below.
By definition if Q supports A, then A(x) = 0 for almost every x ∈ Rn \ Q.
Let 1 ≤ q ≤ p < ∞. As is easily verified by Hölder’s inequality, any
0
(p, q)-block has Lp (Rn )-norm less than 1:
kAkLp0 ≤ 1 (9.2)
for all blocks A. The set of constant multiples of all blocks forms a linear
space. However, this set is unnatural due to the freedom of the choice of cubes
in the definition of (p, q)-blocks. We are thus interested in a natural way to
allow us to consider that the set of all (p, q)-blocks is a Banach space.
0
Definition 83 (Hqp0 (Rn )). Let 1 ≤ q ≤ p < ∞ and define the block space (the
0 0
Zorko space) Hqp0 (Rn ) as the set of all f ∈ Lp (Rn ) for which f is realized as
∞
X
the sum f = λj Aj with some {λj }∞ 1 ∞
j=1 ∈ ` (N) and a sequence {Aj }j=1 of
j=1
0
(p, q)-blocks. Define the norm kf kHp0 for f ∈ Hqp0 (Rn ) as
q0

kf kHp0 ≡ inf kλk`1 , (9.3)

q0 λ

where λ = {λj }∞
j=1 runs over all admissible expressions:
∞
X
f= λj Aj , {λj }∞ 1
j=1 ∈ ` (N), Aj is a (p, q)-block for all j ∈ N. (9.4)
j=1
366 Morrey Spaces
∞
P
A direct consequence of (9.2) is that the series f = λj Aj in (9.4)
j=1
0
∞
converges in the topology of Lp (Rn ). Thus, the postulate that f =
P
λj Aj
j=1
0
converges in Lp (Rn ) is natural as we will see.
0 0
Proposition 339. Let 1 < p < ∞. Then Hpp0 (Rn ) = Lp (Rn ) with coincidence
of norms.
0
Proof Let f ∈ Lp (Rn ). Fix ε > 0. Then there is a decomposition
∞
X
f = χQ1 f + χQj \Qj−1 f,
j=2

where {Qj }∞j=1 is an increasing sequence of cubes centered at the origin sati-
fying
ε
kχRn \Qk f kLp0 ≤ k
2
for all k ∈ N. Then using this decomposition, we learn kf kHp0 ≤ kf kLp0 +
p0
∞
X ε 0

k
. Thus, f ∈ Hpp0 (Rn ).
2
k=1
0 0
Meanwhile, if f ∈ Hpp0 (Rn ), then f ∈ Lp (Rn ) as we have seen.
0
We give an example of a function in Hqp0 (Rn ).
Example 143. Let 1 < q ≤ p < ∞. Define
1
f (x) ≡ (x ∈ Rn ).
1 + max(|x1 |, |x2 |, . . . , |xn |)M
0 0
Then f ∈ Hqp0 (Rn ) if and only if M p0 > n. Indeed, if f ∈ Hqp0 (Rn ), then
0
f ∈ Lp (Rn ), which forces M p0 > n. Suppose instead that M p0 > n. Define
1 −j n − n +jM −M
b0 ≡ n
χQ(1) · f, bj ≡ 2 p0 p0 χQ(2j )\Q(2j−1 ) · f (j ∈ N).
2
Then, kb0 kL∞ = 2−n and for each j ∈ N
−j pn0 − pn0 +jM −M
2 −j pn0 − pn0 − p10
kbj k L∞ = <2 ≤ |Q(2j )| .
1 + 2(j−1)M
Hence, each bj , j ∈ N0 is a (p, q)-block modulo a multiplicative constant by
Hölder’s inequality or by Example 142. Observe also that
∞
j pn0 + pn0 −jM +M
X
f = 2n b0 + 2 bj .
j=1
 Predual of Morrey spaces 367
∞
j pn0 + pn0 −jM +M
Assuming that M p0 > n, we have 2n +
P
2 < ∞. Hence, it
j=1
0
follows that f ∈ Hqp0 (Rn ).
Below we collect some fundamental properties of block spaces.
Lemma 340. Let 1 < q ≤ p ≤ ∞. If A is a (p, q)-block, then kAkHp0 ≤ 1.
q0

Proof In (9.4), we choose

A1 = A, A2 = A3 = · · · = 0, λ1 = 1, λ2 = λ3 = · · · = 0.
Then, (9.3) gives the desired result.
The following lemma indicates how to generate blocks.
0
Lemma 341. Let 1 < q ≤ p ≤ ∞. Let A be an Lq (Rn ) function supported
1 1
on a cube Q. Then kAkHp0 ≤ kAkLq0 |Q|− p + q .
q0

1−1
|Q| p q
Proof Set B ≡ kAkLq0 A, assuming that A is not zero almost everywhere.
1 1
Then B is supported on the cube Q and, by virtue of the fact that p + p0 =
1 1
q + q 0 = 1,
1 1 − p10 + q10
kBkLq0 = |Q| p − q = |Q| .
Hence, B is a (p, q)-block. This implies kBkHp0 ≤ 1 according to Lemma 340.
q0
Equating this inequality, we obtain the desired result.
We have the lattice property of block spaces.
Lemma 342 (Lattice property). Let 1 < q ≤ p ≤ ∞. Then, a function f
0 0
belongs to Hqp0 (Rn ) if and only if there exists g ∈ Hqp0 (Rn ) ∩ M+ (Rn ) such that
|f (x)| ≤ g(x) for a.e. x ∈ Rn .
0
Proof Suppose that f ∈ Hqp0 (Rn ). Then there exists a sequence {λk }∞
k=1 ∈
∞ ∞
l1 and a (p, q)-block bk such that f =
P P
λk bk . Letting g ≡ |λk ||bk |,
k=1 k=1
0
we have g ∈ Hqp0 (Rn ) and |f | ≤ g. Conversely, suppose that there exists
0
g ∈ Hqp0 (Rn ) that satisfies |f (x)| ≤ g(x) for a.e. x ∈ Rn . Decompose g as
∞
λ0k b0k where {λ0k }∞ 1 0
P
g= k=1 ∈ ` (N) and bk is a (p, q)-block. Then we see that
k=1
∞
X 1 0
χ{y: g(y)6=0} (x) = λ0k b (x)
g(x) k
k=1
∞ |f (x)|
λ0k fg(x)
(x) 0
bk (x) for almost all x ∈ Rn . Since
P
and, hence, f (x) = ≤1
k=1 g(x)
for a.e. x ∈ Rn , the function (f /g)b0k becomes a (p, q)-block. This proves the
lemma.
368 Morrey Spaces

In the next lemma, we can say that each bk (Q) is supported on 3Q, while, in
Definition 83, each Aj has its support in a cube Qj but there is no information
on the cube Qj .
0
Lemma 343. Let 1 < q ≤ p ≤ ∞ and f ∈ Hqp0 (Rn ) with kf kHp0 < 1. Then
q0
0
λ(Q)b(Q) in the topology of Hqp0 (Rn ),
P
f can be decomposed as f =
Q∈D(Rn )
where {λ(Q)}Q∈D(Rn ) satisfies
X
λ(Q) ≤ 3n (9.5)
Q∈D(Rn )

and b(Q) is a (p, q)-block supported in 3Q.

Proof We may as well assume that f 6= 0. First, decompose f as
X
f= λk bk
k∈K

where K ⊂ N is an index set, the coefficients {λk }k∈K satisfy

X
0< |λk | < 1 (9.6)
k∈K

and each bk is a (p, q)-block. We divide K into the disjoint sets K(Q) ⊂ N,
Q ∈ D(Rn ), as [
K= K(Q)
Q∈D(Rn )

and K(Q) fulfills supp(bk ) ⊂ 3Q and |Qk | ≥ |Q| when k ∈ K(Q). We do this
as follows: Let D = {Q(j) }∞ n
j=1 be an enumeration of D(R ). For each k ∈ K,
we define
jk ≡ min{j : supp(bk ) ⊂ 3Q(j) , |Qk | ≥ |Q(j) |}
for each k. We write K(Q(j) ) ≡ {k ∈ K : jk = j}. We set
 1 P
X  λk bk (λ(Q) 6= 0),
λ(Q) ≡ 3n
|λk |, b(Q) ≡ λ(Q) k∈K(Q)

k∈K(Q) 0 (λ(Q) = 0).

We now rewrite f as
 
X X X
f= λk bk =  λk bk 
k∈K Q∈D(Rn ) k∈K(Q)
   −1 
X  X   X X 

n n
= 3 |λk | · 3 |λk |  λk bk .
   
Q∈D(Rn ) k∈K(Q)  k∈K(Q) k∈K(Q) 
 Predual of Morrey spaces 369

By (9.6), we have
 
X X X X
λ(Q) = 3n  |λk | = 3n |λk | < 3n ,
Q∈D(Rn ) Q∈D(Rn ) k∈K(Q) k∈K

which implies (9.5). Since each bk is a (p, q)-block, we obtain

 −1
X X
3n |λk | λk bk
k∈K(Q) k∈K(Q)
Lq 0
 −1
X X
≤ 3n |λk | |λk | · kbk kLq0
k∈K(Q) k∈K(Q)
 −1
X 1 1 X
≤ 3n |λk | |Q| p − q |λk |
k∈K(Q) k∈K(Q)
1 1
.p,q |3Q| p − q ,

which implies that b(Q) is a (p, q)-block supported in 3Q modulo a multiplica-

tive constant. These complete the proof.
By decomposing b(Q) further, we obtain an equivalent expression of the
0
space Hqp0 (Rn ). Theorem 344 is useful in that we can remove the duplicate of
the cubes; it can happen that the cubes for Aj and Aj 0 are identical for some
1 ≤ j < j 0 < ∞ in Definition 83.
0
Theorem 344. Let 1 < q ≤ p ≤ ∞ and f ∈ Hqp0 (Rn ) with kf kHp0 < 1. Then
P q0
f can be decomposed as f = λ(Q)b(Q), where each λ(Q) ≥ 0 for each
Q∈D(Rn )
λ(Q) ≤ 9n .
P
Q, each b(Q) is a (p, q)-block supported in Q, and
Q∈D(Rn )

0
The block space Hqp0 (Rn ) is designed to study the duality of Mpq (Rn ). To
study a predual space of LMpq (Rn ) we can extend our definition naturally.
For example, the notion of local blocks can be defined in a similar fashion.
Definition 84. Let 1 ≤ p ≤ q < ∞. A function A ∈ L0 (Rn ) is a local (p, q)-
block if there exists a cube Q = Q(r) that is centered at the origin, supports
A and satisfies (9.1).
1 1
Example 144. Let 1 ≤ q ≤ p < ∞. For r > 0, |Q(r)| p − q χQ(r) is a local
(p, q)-block.
We now present the definition of local block spaces.
370 Morrey Spaces
0
Definition 85 (LHqp0 (Rn )). Let 1 ≤ p ≤ q < ∞ and define the local block
0 0
space LHqp0 (Rn ) as the set of all f ∈ Lp (Rn ) for which f is realized as the sum
∞ 0
λj Aj converges in Lp (Rn ) with some {λj }∞ 1
P
f= j=1 ∈ ` (N) and a sequence
j=1
0
p
{Aj }∞ n
j=1 of local (p, q)-blocks. Define the norm kf kLHp0 for f ∈ LHq 0 (R ) by
q0

kf kLHp0 ≡ inf kλk`1 ,

q0 λ

where λ = {λj }∞
j=1 runs over all admissible expressions

∞
X
f= λj Aj , {λj }∞ 1
j=1 ∈ ` (N), Aj is a (p, q)-local block for all j ∈ N.
j=1

We remark that an analogy to usual Morrey spaces is available, whose

details we omit.

9.1.2 Finite decomposition and a dense subspace

0
Here, we consider a dense subspace of Hqp0 (Rn ) for 1 < q ≤ p < ∞ together
with a decomposition result, which will be useful when we consider the bound-
0
edness property of Morrey spaces. It is noteworthy that the space Hqp0 (Rn )
0
has Lqc (Rn ) as a dense space, while the Morrey space Mpq (Rn ) does not.
0 0
Theorem 345. Let 1 < q ≤ p < ∞. Then Lqc (Rn ) is dense in Hqp0 (Rn ). In
particular, by mollification, Cc∞ (Rn ) is dense.
0
Proof Since g ∈ Hqp0 (Rn ), there exists a non-negative summable sequence
∞
{λj }∞ ∞
P
j=1 and a collection {bj } j=1 of (p, q)-blocks such that g = λj bj . Define
j=1
N
P ∞
P
gN ≡ λj bj . Then kg − gN kHp0 ≤ |λj | → 0 as N → ∞.
j=1 q0 j=N +1
0 0
Consequently, when we investigate Hqp0 (Rn ), the space Lqc (Rn ) is a useful
space.
We will be interested in linear operators defined and continuous on Lq (Rn ).
0
Consequently, it will be helpful to have a finite decomposition in Lqc (Rn ).
What differs from Definition 83 is that we are considering finite admissible
expressions. That is, the sum is finite.
0
Theorem 346. Let 1 < q ≤ p < ∞. Then any f ∈ Lqc (Rn ) admits the finite
N
P
decomposition f = λj bj , where λ1 , λ2 , . . . , λN ≥ 0 and each bj is a (p, q)-
j=1
block. Furthermore, kf kHp0 ∼ inf (|λ1 | + |λ2 | + · · · + |λN |), where λ = {λj }N
j=1
q0 λ
 Predual of Morrey spaces 371

runs over all finite admissible expressions:

N
X
f= λj bj , λ1 , λ2 , . . . , λN ≥ 0, bj is a (p, q)-block for all j = 1, 2, . . . , N.
j=1

0 0
Proof Let f ∈ Lqc (Rn ) ⊂ Hqp0 (Rn ). We may assume that kf kHp0 < 1
q0

and that supp(f ) ⊂ Q0 for some dyadic cube Q0 , since we can decompose f
into 2n sums, each of which is supported on a closure of a quadrant. Then
according to Theorem 344 f can be decomposed as
X
f= λ(Q)b(Q)
Q∈D(Rn )

where X
λ(Q) ≤ 9n .
Q∈D(Rn )

We may assume that each b(Q) is non-negative and that λ(Q) is a non-negative
0
real number, since Hqp0 (Rn ) enjoys the lattice property. Furthermore, we may
assume that supp(b(Q)) ⊂ Q0 . Thus f is non-negative. By the monotone
convergence theorem and the fact that supp(f ) ⊂ Q0 , we have
X
f = lim χ(ε,ε−1 ) (|Q|)λ(Q)b(Q)χQ0
ε↓0
Q∈D(Rn )

0
in Lq (Rn ). Thus, if we set
X
g≡ (1 − χ(ε,ε−1 ) (|Q|))λ(Q)b(Q)χQ0 ,
Q∈D(Rn )

0
then since f ∈ Lq (Rn ), we deduce that g is a (p, q)-block as long as ε is small,
so that X
f =g+ χ(ε,ε−1 ) (|Q|)λ(Q)b(Q)χQ0
Q∈D(Rn )

is the desired finite decomposition.

Since the norm of g can be made as small as we wish, we have the desired
norm equivalence: kf kHp0 ∼ inf (|λ1 | + |λ2 | + · · · + |λN |).
q0 λ

9.1.3 Duality–block spaces and Morrey spaces

We show that the Morrey space Mpq (Rn ) is realized by the dual space of
0
Hqp0 (Rn ).
0
Theorem 347 (Duality of Hqp0 (Rn )). Suppose that 1 < q ≤ p < ∞.
372 Morrey Spaces
0
(1) Any f ∈ Mpq (Rn ) defines a continuous functional Lf : g ∈ Hqp0 (Rn ) 7→
Z
f (x)g(x)dx ∈ C.
Rn
0
(2) Conversely, every continuous functional L on Hqp0 (Rn ) can be realized
with f ∈ Mpq (Rn ).
0
7 Lf ∈ (Hqp0 (Rn ))∗ is an isomor-
(3) The correspondence f ∈ Mpq (Rn ) →
phism. Furthermore
Z 
p0 n
kf kMq = sup
p f (x)g(x)dx : g ∈ Hq0 (R ), kgkHp0 = 1 (9.7)
Rn q0

and
Z 
kgkHp0 = max f (x)g(x)dx : f ∈ Mpq (Rn ), kf kMpq = 1 . (9.8)
q0 Rn

Proof The proof of the boundedness of Lf is simple so we omit it. In fact

it is straightforward to show that kLf k∗ ≤ kf kMpq .
0
To prove that every continuous functional L on Hqp0 (Rn ) can be realized
with f ∈ Mpq (Rn ), we let Qj = 2j Q(1). For the sake of the simplicity we
0 0
denote Lq (Qj ) by the set of Lq (Rn )-functions supported on Qj . In view of
0
Lemma 341 the functional g 7→ L(g) is well defined and bounded on Lq (Qj ).
q 0 n
Thus, by Theorem 9 there exists fj ∈ L (Qj )(⊂ L (R )) such that
Z
L(g) = fj (x)g(x)dx
Qj

for all g ∈ L (Qj ). By the uniqueness of this theorem we can find an Lqloc (Rn )
q

function f such that f = fj a.e. on Qj for any j.

We will prove f ∈ Mpq (Rn ). For this purpose, we take an arbitrary Q and
estimate Z 1
1 1 q
I ≡ |Q| p − q |f (x)|q dx . (9.9)
Q

For a fixed cube Q and a fixed function f we set g ≡ χQ sgn(f )|f |q−1 .
1−1
|Q| p q
Notice that the function kgkLq0 g is a (p, q)-block by virtue of Lemma 341.
Thus, we can write
Z  q1
1 1 1 1 1
I = |Q| p−q f (x)g(x)dx = |Q| p − q (L(g)) q . (9.10)
Q
1 1
Hence, we have |L(g)| ≤ kLk∗ |Q|− p + q kgkLq0 . As a result we have I ≤ kLk∗ .
This is the desired result. The proof of (9.7) and (9.8) is included in what we
have proven.
0
As the following example shows, `1 (Z) is embedded into Hqp0 (Rn ).
 Predual of Morrey spaces 373

Example 145. Let 1 < q < p < ∞.

0
(1) Let f ∈ Mpq (Rn ) and g ∈ Hqp0 (Rn ). Then f ∗g ∈ L∞ (Rn ). Since Cc∞ (Rn )
is dense, f ∗ g is continuous.
∞
X jn
(2) Let f ≡ 2− p aj χB(2j+1 )\B(2j ) , where {aj }∞
j=−∞ is a summable
j=−∞
complex sequence. Then
∞
X
kf kHp0 ∼ |aj |.
q0
j=−∞
jn
In fact, since 2− χB(2j+1 )\B(2j ) is a (p, q)-block modulo a multiplicative
p

∞
P
constant, we have kf kHp0 . |aj |. To check the opposite inequality,
q0 j=−∞
n
we use kf · gkL1 ≤ kgk Mp
q
kf kHp0 , where g(x) ≡ |x|− p for x ∈ Rn . If we
q0
∞
P
use this inequality, then kf kHp0 & |aj |.
q0 j=−∞

9.1.4 Fatou property of block spaces

We cannot hope that the block space is a Banach function space because
the Morrey space, which is its dual, is not. However, we do have the Fatou
property as the following theorem shows:
Theorem 348. Let 1 < q ≤ p < ∞. Suppose that f and fk , (k = 1, 2, . . .),
0
are non-negative, that each fk ∈ Hqp0 (Rn ), that kfk kHp0 ≤ 1 and that fk ↑ f
q0
0
a.e.. Then f ∈ Hqp0 (Rn ) and kf k Hp
0 ≤ 1.
q0
0
It is significant that we do not assume f ∈ Hqp0 (Rn ), so that there is no
guarantee that f admits the decomposition into the sum of blocks. Before we
prove Theorem 348, we consider its important corollary.
Corollary 349. Let 1 < q ≤ p < ∞. Suppose that we have g ∈ L0 (Rn ) such
0
that f · g ∈ L1 (Rn ) for all f ∈ Mpq (Rn ). Then g ∈ Hqp0 (Rn ).
Proof By the closed graph theorem, kf · gkL1 ≤ Lkf kMpq for some L > 0
independent of f ∈ Mpq (Rn ). If we set gM ≡ χB(M ) χ[0,M ] (|g|)g for each M ∈
0
N, then gM ∈ Hqp0 (Rn ) according to Lemma 341. If we use Theorem 347, then
0
kgM kHp0 ≤ L. Consequently by Theorem 348, we see that g ∈ Hqp0 (Rn ).
q0

Proof of Theorem 348 We may assume that 1 < q < p < ∞ as we

remarked just below the statement. By Theorem 344 fk can be decomposed
as X
fk = λk (Q)bk (Q),
Q∈D(Rn )
374 Morrey Spaces

where λk (Q) is a non-negative number with

X
λk (Q) ≤ 2 · 9n (9.11)
Q∈D(Rn )

and bk (Q) is a (p, q)-block supported in Q and

1 1
kbk (Q)kLq0 ≤ |Q| p − q . (9.12)

In particular,
0 ≤ λk (Q) ≤ 2 · 9n (9.13)
Noticing (9.12), (9.13) and the weak-compactness of the Lebesgue space
0
Lq (Q) (see Corollary 338), we now apply a diagonalization argument or the
Tikhonov compact theorem and, hence, we can select an infinite subsequence
{fkj }∞ ∞
j=1 ⊂ {fk }k=1 that satisfies the following:
X
fkj = λkj (Q)bkj (Q), lim λkj (Q) = λ(Q),
j→∞
Q∈D(Rn )
0
lim bkj (Q) = b(Q) in the weak-topology of Lq (Q),
j→∞
P
where b(Q) is a (p, q)-block supported in Q. We set f0 ≡ λ(Q)b(Q).
Q∈D(Rn )
Then, by Fatou’s lemma and (9.11),
X X
λ(Q) ≤ lim inf λkj (Q) ≤ 2 · 9n , (9.14)
j→∞
Q∈D(Rn ) Q∈D(Rn )

0
which implies f0 ∈ Hqp0 (Rn ).
We will verify that
Z Z
lim fkj (x)dx = f0 (x)dx (9.15)
j→∞ Q0 Q0

for all Q0 ∈ D(Rn ). Once (9.15) is established, we will see that f = f0 and
0
f ∈ Hqp0 (Rn ) by virtue of the Lebesgue differentiation theorem because at
0
least we know that f0 locally in Lq (Rn ). By the definition of fkj and the fact
0
that the sum defining fkj converges in Lp (Rn ), we have
Z ∞
X X Z
fkj (x)dx = λkj (Q) bkj (Q)(x)dx.
Q0 l=−∞ Q∈D(Q0 )∪D ] (Q0 ), |Q|=2ln |Q0 | Q0

Note that
Z
1 1 1 1
bkj (Q)(x)dx ≤ kbkj (Q)kL1 ≤ kbkj (Q)kLq0 |Q| q ≤ |Q| p − q |Q ∩ Q0 | q
Q0
 Predual of Morrey spaces 375

for any cube Q contained in Q0 . Since

∞
X 1 1 1
|Q| p − q |Q ∩ Q0 | q < ∞,
l=−∞

we are in the position of Lebesgue’s convergence theorem to obtain

Z
lim fkj (x)dx
j→∞ Q
0
 
∞
X X Z
=  lim λkj (Q) bkj (Q)(x)dx .
j→∞ Q0
l=−∞ Q∈D(Q0 )∪D ] (Q0 ), |Q|=2ln |Q0 |

Since X

Q∈D(Q0 )∪D ] (Q0 ), |Q|=2ln |Q0 |

is the symbol of summation over a finite set for each l, we have
X Z X Z
lim λkj (Q) bkj (Q)(x)dx = λ(Q) b(Q)(x)dx. (9.16)
j→∞ Q0 Q0
Q∈D(Rn ) Q∈D(Rn )

Thus, we obtain (9.15). Z Z

Since fk ↑ f a.e., we must have f (x)dx = f0 (x)dx for all Q0 ∈
Q0 Q0
D(Rn ) by (9.15). This together with the Lebesgue differential theorem yields
0 0
f = f0 a.e. and hence f ∈ Hqp0 (Rn ). Since we have verified f ∈ Hqp0 (Rn ), it
follows that
Z 
kf kHp0 = sup fk (x)g(x)dx : k = 1, 2, . . . , kgkMq ≤ 1 ≤ 1.
p
q0 Rn
This completes the proof of the theorem.
Similar to Theorem 348, we can prove the Fatou property of local block
spaces and its consequence.
Theorem 350. Let 1 < q ≤ p ≤ ∞.
0
(1) Let f ∈ M+ (Rn ) and fk ∈ LHqp0 (Rn ) ∩ M+ (Rn ), k ∈ N. Suppose that
0
kfk kLHp0 ≤ 1 and that fk ↑ f a.e.. Then f ∈ LHqp0 (Rn ) and kf kLHp0 ≤ 1.
q0 q0

0
fpq (Rn ) is canonically identified with
(2) The dual of LM LHqp0 (Rn ).

9.1.5 Köthe dual of Morrey spaces

We will characterize the predual of block spaces in terms of the Köthe
dual. The result will be an application of the Fatou property. Let us start
with a general fact. We give a general definition.
376 Morrey Spaces

Definition 86 (ρ0 ). If ρ is a ball Banach function norm, its “associate norm”

ρ0 is defined in M+ (Rn ) by

ρ0 (g) ≡ sup kf · gkL1 : f ∈ M+ (Rn ), ρ(f ) ≤ 1 (g ∈ M+ (Rn )). (9.17)



Likewise we can consider the Köthe dual for Banach lattices.

0
Example 146. Let 1 ≤ p ≤ ∞. Then Lp (Rn )0 = Lp (Rn ). It is worth com-
0
paring with this Lp (Rn )∗ = Lp (Rn ) for 1 ≤ p < ∞.
Let f, g ∈ M+ (Rn ). A trivial consequence of the related definitions is;
Z
f (x)g(x)dx ≤ ρ(f )ρ0 (g). (9.18)
Rn

By no means do we have a guarantee that any functional in Morrey spaces

can be described in terms of an integral like Lp (Rn ) with 1 < p < ∞. Con-
sequently, we here want to consider functionals in Morrey spaces that can be
described in terms of integrals.
The dual of a ball Banach function norm is again a ball Banach function
norm.
Theorem 351. Let ρ be a ball Banach function norm. Then the associate
norm ρ0 is a ball Banach function norm.
Proof We need to verify the properties (P1)–(P5) of Banach function
norms.
By virtue of the Lebesgue differentiation theorem, we have g = 0 a.e.
whenever g ∈ M+ (Rn ) satisfies
Z
f (x)g(x)dx = 0
Rn

for all f ∈ M+ (Rn ) such that ρ(f ) ≤ 1. Other two properties of (P1) are clear
from the corresponding properties of integrals. Consequently, (P1) is verified.
By the monotonicity of integral (P2) is also trivial. Let us check (P3). To
this end, we take a non-negative increasing sequence {gj }∞ j=1 ; 0 ≤ gj ↑ g. If
ρ(g) = 0, then there is nothing to prove. Hence, assume ρ0 (g) > 0. Choose any
non-negative number M ∈ [0, ρ0 (g)).
Then by the definition of ρ0 (g), we can find f ∈ M+ (Rn ) with ρ(f ) ≤ 1
such that Z
M< f (x)g(x)dx ≤ ρ0 (g).
Rn
By virtue of the monotone convergence theorem, if j is large enough
Z
M< f (x)gj (x)dx ≤ ρ0 (gj ) ≤ ρ0 (g).
Rn

Thus, ρ0 (gj ) ↑ ρ0 (g).

 Predual of Morrey spaces 377

The property (P4) follows from (P5) for (X, ρ);

Z
f (x)dx .Q 1
Q

for all f ∈ M+ (Rn ) with ρ(f ) ≤ 1.

Finally the property (P5) follows from (P4) for (X, ρ) and (9.18);
Z
g(x)dx ≤ ρ(χQ )
Q

for all g ∈ M+ (Rn ) with ρ0 (g) ≤ 1.

We define the associated space with respect to ρ. It should be noted that
in general the associate space differs from the dual space as the example of
Morrey spaces.
Definition 87 (X (ρ0 ), X 0 (ρ)). Let ρ be a ball Banach function norm, and let
X = X (ρ) be the ball Banach function space determined by ρ as in Definition
13. Let ρ0 be the associate norm of ρ. The Banach function space X (ρ0 ) =
X 0 (ρ) determined by ρ0 is called the associate space or the Köthe dual of X .
At this moment, we content ourselves with a simple example of Lebesgue
spaces.
Example 147. Let 1 ≤ p ≤ ∞ and p0 be its conjugate, and let (X, B, µ) be
0
a measure space. Then the associate space of Lp (µ) is Lp (µ). It matters that
we tolerate the case p = ∞.
As the example of X = `1 (N) shows, the bidual X ∗∗ of a Banach space X
can be strictly bigger than X itself. However, unlike the dual spaces of Banach
spaces, the associate space of the associate space of any ball Banach function
space E(Rn ) is E(Rn ) itself.
Theorem 352. Every ball Banach function space E(Rn ) coincides with its
second associate space E 00 (Rn ). In other words, a function f belongs to E(Rn )
if and only if it belongs to E 00 (Rn ), and in that case kf kE = kf kE 00 or all
f ∈ E(Rn ) = E 00 (Rn ).
We have an analogy to Banach function spaces; see [29]. Keeping in mind
that Morrey spaces are main function spaces in this book, we consider ball
Banach function spaces.
Proof It is trivial from (9.18) that E(Rn ) ⊂ E 00 (Rn ) and that ρ00 (f ) ≤
ρ(f ) for any f ∈ M+ (Rn ).
To prove the reverse inclusion, we let h ∈ E 00 (Rn ). Since we have (P2),
(P3) and (P4)’, we can suppose that h ∈ L∞ n
c (R ). Set

S ≡ {f ∈ L1 (Rn ) : supp(f ) ⊂ supp(h), ρ(|f |) ≤ 1}.

378 Morrey Spaces

Let us check that S is a closed convex set in L1 (Rn ). Let f1 , f2 ∈ S and

a ∈ (0, 1). Then ρ(|af1 + (1 − a)f2 |) ≤ aρ(|f1 |) + (1 − a)ρ(|f2 |) ≤ 1 by virtue of
(P1) and supp(af1 + (1 − a)f2 ) ⊂ supp(f1 ) ∪ supp(f2 ) ⊂ supp(h). Thus, S is
a convex set. To prove that S is a closed set, we take a sequence {fj }∞ j=1 in S
that converges to f ∈ L1 (Rn ) in the L1 (Rn ) topology. If necessary, by passing
to a subsequence, we can assume that {fj }∞ j=1 converges almost everywhere
to f . Thus, supp(f ) is contained in supp(h). In addition, by (P3), the Fatou
property,
   
ρ(|f |) = ρ lim inf |fj | = lim ρ inf |fj | ≤ 1.
k→∞ j≥k k→∞ j≥k

Thus, S is a closed set.

λ
Let λ > 1. Then g ≡ khk E
h does not belong to S, so that, by the Hahn–
Banach theorem of geometric form (see Theorem 88) and the duality L1 (Rn )-
L∞ (Rn ), we can find ϕ ∈ L∞ (Rn ) such that
Z  Z
λ
Re f (x)ϕ(x)dx < h(x)ϕ(x)dx (∈ R)
Rn khkE Rn
for any f ∈ S. From the definition of E 00 (Rn ), we have
Z 
λ 00
Re f (x)ϕ(x)dx < ρ (h)ρ0 (ϕ).
Rn khkE
If we take supremum over all f ∈ S, then
λ 00
ρ0 (ϕ) ≤ ρ (h)ρ0 (ϕ).
khkE
Since λ > 1 is arbitrary, we see that khkE ≤ ρ00 (h) = khkE 00 .
We will show that ball Banach function spaces are closed under the oper-
ation of taking the Köthe dual.
Proposition 353. Let E(Rn ) be a ball Banach function space. Then its asso-
ciate E 0 (Rn ) is also a ball Banach function space.
Proof To show that E 0 (Rn ) is a ball Banach function space, we need to
show (i)–(v) below:
(i) kf kE 0 = 0 implies that f = 0 almost everywhere;
(ii) |g| ≤ |f | almost everywhere implies that kgkE 0 ≤ kf kE 0 ;
(iii) 0 ≤ fn ↑ f almost everywhere implies that kfn kE 0 ↑ kf kE 0 ;
(iv) B ∈ B implies that χB ∈ E(Rn );
(v) for any B ∈ B, there exists C(B) > 0 such that kf kL1 (B) ≤ C(B) kf kE 0
for all f ∈ E 0 (Rn ).
 Predual of Morrey spaces 379

We concentrate on (iii) since the remaining assertions are proven easily.

By Lebesgue’s differentiation theorem, Theorem 12, we conclude that (i)
holds true. From the definition of E 0 (Rn ), it follows that (ii) holds true. Next,
we obtain (iv). Moreover, from the fact that χB ∈ E(Rn ) for all B ∈ B and
the definition of E 0 (Rn ), we deduce that (v) holds true.
Finally, we prove (iii). Let {fm }∞ 0 n + n
m=1 ⊂ E (R ) ∩ M (R ) and f ∈ E (R )
0 n

satisfy that fm ↑ f almost everywhere, and let A ∈ (0, kf kE ). Then, by the

0

definition of kf kE 0 , we know that there

Z exists a real-valued
Z function g ∈ E(Rn )
with kgkE = 1 such that A < f (x)g(x)dx ≤ f (x)|g(x)|dx, which,
Rn Rn
combined with the monotone
Z convergence theorem, implies that there exists
N ∈ N such that A < fN (x)|g(x)|dx. Thus, kfN kE 0 > A, which, together
Rn
with the arbitrariness of A ∈ (0, kf kE 0 ), implies that (iii) holds true. This
finishes the proof of Proposition 353.
Remark that a similar theory can be developed for Banach lattices. For
example, we can extend Theorem 352 as follows:
Theorem 354. Let E(µ) be a Banach lattice over a measure space (X, B, µ).
Then E(µ) has the Fatou property if and only if E(µ) = E 00 (µ).
Proof If E(µ) has the Fatou property, then we can mimic the proof of
Theorem 352. See Exercise 117. Conversely, suppose E(µ) = E 00 (µ). Then
since F 0 (µ) has the Fatou property for any Banach lattice F (µ), we see that
E(µ) has the Fatou property.
We characterize the Köthe dual of Morrey spaces.
Theorem 355. Let 1 < q ≤ p < ∞. Then the associate space (Mpq )0 (Rn ) as
0
a ball Banach function space coincides with the block space Hqp0 (Rn ).
By no means does Theorem 355 give any complete information on the dual
space of Mpq (Rn ).
Proof The proof is a mere combination of Theorems 348 and 354. Alter-
natively, we can argue directly as follows: As we have seen in Theorem 347,
0
Hqp0 (Rn ) ⊂ (Mpq )0 (Rn ). Hence, we will verify the converse. Suppose that
f ∈ L0 (Rn ) satisfies
n o
sup kf · gkL1 : kgkMpq ≤ 1 ≤ 1. (9.19)

Then we first see that |f (x)| < ∞, (a.e. x ∈ Rn ). Splitting f into its real
and imaginary parts and each of these into its positive and negative parts,
we may assume without loss of generality that f ∈ M+ (Rn ). For k = 1, 2, . . .,
0
set fk ≡ min(f, k)χQ(k) . We notice that fk ∈ Hqp0 (Rn ) and kfk kHp0 ≤ 1 by
q0
0
Lemma 342 and (9.19). Since fk ↑ f a.e., it follows that f ∈ Hqp0 (Rn ) and
kf kHp0 ≤ 1 by virtue of Theorem 348. This proves the theorem.
q0
380 Morrey Spaces
0
Theorems 352 and 355 yield Hqp0 0 (Rn ) = (Mpq )00 (Rn ) = Mpq (Rn ). Further-
more, from the fact that (Mpq )00 (Rn ) = Mpq (Rn ), we are able to characterize
the predual of block spaces.
Recall that we defined M fp (Rn ) in Definition 77. The block spaces are
q
fpq (Rn ).
realized as the dual of M
0
Theorem 356. Let 1 < q ≤ p < ∞. Then a predual space of Hqp0 (Rn ) is
fpq (Rn ) in the following sense:
M
0
If g ∈ Hqp0 (Rn ), then Lg : f 7→ Rn f (x)g(x)dx is an element of M fp (Rn )∗ .
R
q
0
Moreover, for any L ∈ M fp (Rn )∗ , there exists g ∈ Hp0 (Rn ) such that L = Lg .
q q
0
Proof The first assertion is clear, since Mpq (Rn ) is the dual of Hqp0 (Rn ).
0
Hence, we will prove that M fpq (Rn )∗ ⊂ Hp0 (Rn ). Thanks to Theorem 355, we
q
need only show M fpq (Rn )∗ .
fpq (Rn )∗ ⊂ (Mpq )0 (Rn ). Suppose that L belongs to M
p 0 n
We will exhibit a function g in (Mq ) (R ) such that
Z
L(f ) = f (x)g(x)dx (f ∈ M fpq (Rn )). (9.20)
Rn

If we let Qm ≡ m + [0, 1)n , m ∈ Zn , then the sequence {Qm }m∈Zn par-

titions Rn , each of which has measure one and whose union is all of Rn . For
each i ∈ Zn , let Ai denote the Lebesgue measurable subsets of Qi and define
a set-function λi on Ai by

λi (A) ≡ L(χA ), (A ∈ Ai ).

Notice that λi (A) is well defined for all A ∈ Ai because χA belongs to Mfp (Rn ).
q
n
We claim that λm is countably additive on Am for any m ∈ Z . Indeed,
let (Ak )∞
k=1 be a sequence of disjoint sets from Am , and let

l
[ ∞
[ ∞
[
Bl ≡ Ak , (l = 1, 2, . . .), A≡ Ak = Bl .
k=1 k=1 l=1

It follows from the Lebesgue dominated convergence theorem that

kχA − χBl kMpq ≤ kχA − χBl kLp (Qm ) ↓ 0 as l → ∞.

The continuity and linearity of L give

l
X ∞
X
λm (A) = L(χA ) = lim L(χBl ) = lim L(χAk ) = λm (Ak ),
l→∞ l→∞
k=1 k=1

which establishes the claim.

 Predual of Morrey spaces 381

Since |λm (A)| ≤ kLk(M

fp n ∗ kχA kM
q (R ))
fp
q
for all A ∈ Ai and λm (A) = 0 for
all A ∈ Am such that |A| = 0, by the Radon–Nikodym theorem (see [236,
Theorem 15.41], for example), there uniquely exists gm ∈ L0 (Qm ) such that
Z
L(χA ) = λm (A) = χA (x)gm (x)dx, (A ∈ Am ).
Rn

Since the sets Qm are disjoint we may define a function g on all of Rn by

setting g ≡ gm on each Qm . Clearly,
Z
L(χE ) = χE (x)g(x)dx (9.21)
Rn

for all characteristic functions of sets of finite measure χE .

We will show that g belongs to (Mpq )0 (Rn ). Choose and fix f in Mpq (Rn ).
4l
2−l kχFk,l for l = 1, 2, . . ., where
P
Let fl ≡
k=1

Fk,l ≡ x ∈ B(2l ) : 2−l k ≤ |f (x)| < 2−l (k + 1) .



If we suppose for the moment that g is real-valued, then fl · sgn(g) becomes a

finite linear combination of characteristic functions of sets of finite measure.
Hence, we may apply (9.21) and use the linearity of L to obtain
Z
fl (x)|g(x)|dx = L(fl · sgn(g)) ≤ kLkM
fp n ∗ kfl kMq .
q (R )
p

Rn

Letting l → ∞, we have kf · gkL1 ≤ kLkM fp n ∗ kf kMq from the monotone

q (R )
p

convergence theorem and the Fatou property of Morrey norm. This means
that g belongs to (Mpq )0 (Rn ). If g is complex-valued, then the same argument
applied separately to the real and imaginary parts of g shows that each of
these is in (Mpq )0 (Rn ). Hence, that g again belongs to (Mpq )0 (Rn ).
Write, for a function f which can be written as a finite linear combination
of characteristic functions of sets of finite measure,
Z
L(f ) = f (x)g(x)dx
Rn

and observe the continuity of both sides on Mpq (Rn ). Then we conclude from
Theorem 330 that (9.20) holds. This completes the proof of the theorem.
We conclude Section 9.1.5 with an example showing that the dual space
of the Morrey space Mpq (Rn ) with 1 ≤ q < p < ∞ does not coincide with its
Köthe dual. In particular, Mpq (Rn ) with 1 ≤ q < p < ∞ is not reflexive.
Example 148. Let 1 < q < p < ∞ and L : Mpq (Rn ) → C be a bounded
linear functional. Then in view of the embedding Lp (Rn ) ,→ Mpq (Rn ), one has
0
a function g ∈ Lp (Rn ) such that
Z
L(f ) = f (x)g(x)dx, (f ∈ Lp (Rn )).
Rn
382 Morrey Spaces

However, it can happen that L 6= 0 even when g ≡ 0; one can show this by an
example. Recall the set G considered in Example 12. Set
p p
Ik ≡ (k − 1 + k p−q , k + k p−q ).

for k ∈ N. Then,
∞
[ Z
G= Ik , lim χG (x)dx = 1.
k→∞ Ik
k=1

With this in mind, let us define a closed subspace H of Mpq (Rn ) by

 Z 
H ≡ f ∈ Mpq (R) : lim f (x)dx exists .
k→∞ Ik

Then, from the definition of the norm, we have

Z
lim f (x)dx ≤ kf kMpq (R) .
k→∞ Ik

Consequently, it follows from the Hahn–Banach theorem, Theorem 86, that

mapping Z
f ∈ H 7→ lim f (x)dx ∈ R
k→∞ Ik

extends to a certain continuous linear functional L on Mpq (R) satisfying

|L(f )| ≤ kf kMpq . Observe that L(χG ) = 1. Hence L 6= 0. Meanwhile, L anni-
hilates any compactly supported function in Mpq (R) because such a function
belongs to H. Therefore, if one considers a function g satisfying
Z
L(f ) = f (x)g(x)dx
Rn

for all f ∈ Mpq (Rn ) and hence all f ∈ Lp (Rn ), then one obtains g ≡ 0 by
virtue of the Lebesgue dominated convergence theorem.

9.1.6 Decomposition and averaging technique in Morrey

spaces
One of the fundamental techniques in the theory of function spaces is
to decompose functions into some elementary pieces. As we witnessed in
the Calderón–Zygmund decomposition, this is useful when we analyze linear
operators. Here we provide a standard decomposition of functions in Morrey
spaces.
Theorem 357. Let 1 < q ≤ p < ∞. Then any f ∈ Mpq (Rn ) admits a
decomposition: X
f= λQ aQ ,
Q∈A
 Predual of Morrey spaces 383

where A is sparse, aQ ∈ PL (Rn )⊥ and |aQ | ≤ χQ for all Q ∈ A and {λQ }Q∈A
satisfies
X
λQ χQ . kf kMpq .
Q∈A
Mp
q

For the proof of Theorem 357, we need to recall some elementary notions.
Let Q be a cube and let L ∈Z N. For all f ∈ Z L1loc (Rn ) there uniquely
exists PQL (f ) ∈ PL (Rn ) such that f (x)g(x)dx = PQL f (x)g(x)dx for all
Q Q
g ∈ PL (Rn ) since {xα }|α|≤k is independent in L2 (Q). The polynomial PQL f is
called the Gram–Schmidt polynomial of order L for Q. We have
Z
L
PQ(1) f (x) = R(x, y)f (y)dy
Q(1)
 
L −n L · − x0
for some polynomial R. The scaling law PQ(x0 ,r)
f = r PQ(1) f
r
holds (see Proposition 117).
Let A > 4n . We can find a set {Qkj }j∈Jk of dyadic cubes and a collection
{g} ∪ {bkj }j∈Jk of countable collections of L1 (Rn )-functions such that:

(1) (Partition of the set {M D f > Ak }) The family {Qkj }j∈Jk is disjoint and
Ωk ≡ { M D f > Ak } =
P k
Qj . Furthermore,
j∈Jk

Ak < mQkj (|f |) ≤ 2n Ak .

(2) f admits the following decomposition: f = g k + bkj . Here

P
j∈Jk
X
g k = f χRn \Ωk + PQkj (f )χQkj .
j∈Jk

(3) (The L∞ (Rn )-condition) The function g k is L∞ (Rn )-bounded: More pre-
cisely, kg k kL∞ ≤ 2n Ak .
(4) (The support condition) The function bkj is supported on the closure of
Qkj ; namely supp(bkj ) ⊂ Qkj .

(5) (The moment condition) bkj ⊥ Pl (Rn ).

In the above collection of cubes, we have the sparseness property.
Proposition 358. If A > 4n , then A = {Qkj }k∈Z,j∈Jk is sparse.
Proof Since the proof is similar to Theorem 172, we omit the detail.
We have the following property:
384 Morrey Spaces

Proposition 359. For all k ∈ Z, |g k | . Ak .

Proof If x ∈ Rn \ Ωk , then this is clear from the Lebesgue differentiation

theorem. If x ∈ Qkj for some j ∈ Jk , we have
!
x − c(Qkj ) y − c(Qkj )
Z
1
|g k (x)| ≤ k R , |f (y)|dy . mQkj (|f |) . Ak .
|Qj | Qk
j
`(Qkj ) `(Qkj )

Corollary 360. Let f ∈ L1loc (Rn ) be such that {M D f > λ} never contains
any quadrant for any λ > 0. Then f = lim (g K − g −K+1 ) in the sense of
K→∞
almost everywhere convergence.
Proof Let K ∈ N. Since f − g K is supported on {M D f > AK }, f =
lim g K . Likewise |g −K | ≤ A−K almost everywhere, f = lim (g K − g −K+1 ).
K→∞ K→∞

We set
akj ≡ χQkj (g k − g k+1 ) (k ∈ Z, j ∈ Jk ).

Proposition 361. Let f ∈ L1loc (Rn ), and let k ∈ Z, j ∈ Jk .

(1) |akj | . Ak χQkj .

(2) akj ∈ PL⊥ (Rn ).

Proof
(1) This is clear from Proposition 359.
(2) We calculate
Z Z Z
xβ akj (x)dx = xβ PQkj f (x)dx − xβ χQkj \Ωk+1 (x)f (x)dx
Rn Qk
j Qk
j
X Z
− xβ PQk+1
0
f (x)dx
k+1 j
j 0 ∈Jk+1 Qk
j ∩Qj 0
Z Z
= xβ f (x)dx − xβ χQkj \Ωk+1 (x)f (x)dx
Qk
j Qk
j
X Z
− xβ f (x)dx
k+1
j 0 ∈J Qk
j ∩Qj 0
k+1

= 0.

Here for the third line we used the fact that Qkj contains Qk+1
j0 or that
Qkj ∩ Qk+1
j 0 = ∅ for any k ∈ Z, j ∈ J k and j 0
∈ Jk+1 .
We prove Theorem 357 after we translate it into the following form:
 Predual of Morrey spaces 385

Theorem 362. Let 1 < q ≤ p < ∞. Then for any f ∈ Mpq (Rn ) there is a
decomposition
∞
X X
f= akj
k=−∞ j∈Jk

satisfying akj ∈ PL (Rn )⊥ ,

∞
X X
|akj | . Ak χQkj (k ∈ Z, j ∈ Jk ), Ak χQkj . kf kMpq
k=−∞ j∈Jk
Mp
q

for some sparse collection {Qkj }k∈Z,j∈Jk of dyadic cubes.

Proof Note that M D is bounded on Mpq (Rn ), so that we are in the posi-
tion of using the previous observation. What is not contained in the previous
observations is the norm estimate. We do this as follows: First, observe that
Ak χΩk ≤ M D f . Consequently, since the Ωk ’s are nested,

∞
X X ∞
X
Ak χQkj = Ak χΩk
k=−∞ j∈Jk k=−∞ Mp
Mp
q q

. sup Ak χΩk
k∈Z Mp
q
D
≤ kM f kMpq
. kf kMpq .

Here for the last line we used the boundedness of M D on Mpq (Rn ).
There is another side in the theory of decomposition. As an application
of the duality, we obtain the averaging technique.
X As we saw, in some cases,
operators can be decomposed into the sum λQ (f )χFQ , where U is a family
Q∈U
of cubes and FQ is a measurable subset of Q. Hence, we are interested in the
general estimate of the function of the form
X
aQ ,
Q∈U

where aQ ∈ L0c (Q). We have partial information on aQ , which we think has

the singularity at least locally. We will try to remove the singularity. The next
theorem, which guarantees that this is possible as long as the size of aj is not
too large, is a standard result for Morrey spaces.
Theorem 363. Suppose that the parameters p, q, s and t satisfy

1 < q ≤ p < ∞, 1 < t ≤ s < ∞, q < t, p < s.

386 Morrey Spaces

Assume that {Qj }∞ n ∞ s n ∞

j=1 ⊂ Q(R ), {aj }j=1 ⊂ Mt (R ) and {λj }j=1 ⊂ [0, ∞)
1
fulfill the support condition supp(aj ) ⊂ Qj , the size condition kaj kMst ≤ |Qj | s
∞
X ∞
P
and the norm condition λj χQj < ∞. Then f = λj aj converges
j=1 j=1
Mp
q
in S 0 (Rn ) ∩ Lqloc (Rn ) and satisfies

∞
X
kf kMpq .p,q,s,t λj χQj . (9.22)
j=1
Mp
q

Proof By decomposing Qj , we may suppose each Qj is dyadic.

To prove this, we resort to the duality. For the time being, we assume
that there exists N ∈ N such that λj = 0 whenever j ≥ N . Let us assume in
0
addition that aj ≥ 0 for each j. Fix a (p, q)-block g ∈ Hqp0 (Rn ) ∩ M+ (Rn ) with
the associated cube Q. Then we claim

∞
X
kf · gk L1 . λj χQj . (9.23)
j=1
Mp
q

By Theorem 344, we may assume that Q is dyadic.

Assume first that each Qj contains Q as a proper subset. If we group j’s
such that Qj are identical, we can assume that |Qj | = 2jn |Q| for each j ∈ N.
Then
X∞ Z ∞
X
kf · gkL1 = λj aj (x)g(x)dx ≤ λj kaj kLq (Q) kgkLq0 (Q)
j=1 Q j=1

∞
P
from f = λj aj . By the size condition of aj and g, we obtain
j=1

∞ ∞
1
1 1 1 − p10 1 1 1
X X
kf · gkL1 ≤ λj |Q| q − s |Qj | s |Q| q0 = λj |Q| p − s |Qj | s .
j=1 j=1

Note that
∞
X 1
λj χQj ≥ λj0 χQj0 Mp
= |Qj0 | p λj0
q
j=1
Mp
q

for each j0 . Consequently, it follows from the condition p < s that

∞ ∞ ∞
X 1 1 1 1 X X
kf · gkL1 ≤ |Q| p − s |Qj | s − p λk χQk . λj χQj .
j=1 k=1 Mp
q
j=1
Mp
q
 Predual of Morrey spaces 387

Conversely, assume that Q contains each Qj . Then

X∞ Z X∞
kf · gkL1 = λj aj (x)g(x)dx ≤ λj kaj kLt (Qj ) kgkLt0 (Qj ) .
j=1 Qj j=1
∞ 1 1 1
λj |Qj | t − s |Qj | s kgkLt0 (Qj ) . Thus,
P
Due to the condition of aj , kf · gkL1 ≤
j=1
in terms of the Hardy–Littlewood maximal operator M , we obtain
∞
X 0 1
kf · gkL1 ≤ λj |Qj | × inf M [|g|t ](y) t0
y∈Qj
j=1
 
Z ∞
X 0 1
≤  λj χQj (y) M [|g|t ](y) t0 dy
Rn j=1
 
Z ∞
X 0 1
≤  λj χQj (y) χQ (y)M [|g|t ](y) t0 dy.
Rn j=1

q0
If we let κ be the operator norm of the maximal operator M on L t0 (Rn ), then
0 0 1
we obtain κ−1/t χQ M [|g|t ] t0 is a (p, q)-block thanks to Lemma 341. Hence,
we obtain (9.23). This is the desired result.
When q = 1, we may assume t = 1 in Theorem 364.
Theorem 364. Suppose that the parameters p and s satisfy
1 < p < ∞, 1 < s < ∞, p < s.
Assume that {Qj }∞ n ∞ s n ∞
j=1 ⊂ Q(R ), {aj }j=1 ⊂ M1 (R ) and {λj }j=1 ⊂ [0, ∞)
1
fulfill the support condition supp(aj ) ⊂ Qj , the size condition kaj kMs1 ≤ |Qj | s
∞
X ∞
P
and the norm condition λj χQj < ∞. Then f = λj aj converges
j=1 j=1
Mp
1
in L1loc (Rn ) and satisfies
∞
X
kf kMp1 .p,s λj χQj .
j=1
Mp
1

Proof We may assume f ∈ M+ (Rn ). The proof resembles that of Theorem

1
363. What differs from Theorem 363 is that we can take g ≡ |Q| p −1 χQ in
Theorem 364. We need to change the proof of Theorem 363 in particular, the
1
case of Qj ⊂ Q since we need to use q < t there. However, since g = |Q| p −1 χQ ,
we have
∞ Z ∞
1 X 1 X
kf · gkL1 = |Q| p −1 λj aj (x)dx ≤ |Q| p −1 λj kaj kL1 (Qj ) .
j=1 Qj j=1
388 Morrey Spaces

If we use the condition on each aj , we obtain

∞ ∞ ∞
|Qj |
Z X
X 1 X
kf · gkL1 ≤ λj = |Q| p −1 λj χQj (x)dx ≤ λj χQj .
1
1− p
j=1 |Q| Q j=1 j=1
Mp
1

9.1.7 Exercises
0 0
Exercise 109. Let 1 < q ≤ p < ∞. Then show that LHqp0 (Rn ) ⊂ Hqp0 (Rn ).
Exercise 110. [472, Proposition 4] Let 1 < q ≤ p < ∞. Then show that
0
Hqp0 (Rn ) is a Banach space by using (9.2) if necessary.
Exercise 111. [218, Example 2.5] Let 1 < q ≤ p < ∞ and a > 0. Define
1
f (x) ≡ (x ∈ Rn ).
a + |x|M
0
Then similar to Example 143, show that f ∈ Hqp0 (Rn ) if and only if M p0 > n.
Exercise 112. Let 1 ≤ p ≤ q < ∞. Let b1 , b2 be (p, q)-blocks supported on
cubes Q1 and Q2 , respectively. Then show that D−1 (b1 + b2 ) is a (p, q)-block
supported on a bigger cube Q as long as D > 0 is large enough.
Exercise 113. Let Q be a fixed cube, and let 1 < p ≤ q < ∞. Suppose
that we have a sequence {bk }∞k=1 of (p, q)-blocks supported on Q. Assume in
addition that {bk }∞
k=1 converges weakly to b. Then show that b is a (p, q)-block.

Exercise 114. Let 1 < q ≤ p < ∞, and let f ∈ Mpq (Rn ). For each j ∈ N
we set fj ≡ f χB(j) χ[0,j] (|f |). Then show that fj → f as j → ∞ in the
weak-* topology. Hint: Although we need to consider the pairing of f and any
0
g ∈ Hqp0 (Rn ), we can justify that g is a (p, q)-block.
Exercise 115. [95, Lemma 1], [282, Lemma 4.6] Let 0 < α < n and 1 <
p0
q ≤p< α n
. Suppose that we have a bounded sequence {fj }∞ n
j=1 in Hq 0 (R )
n ∞
such that Iα fj (x) converges to F (x) for almost all x ∈ R and that {fj }j=1
0
converges to f in the weak-∗ topology of Hqp0 (Rn ) via Theorem 356.
(1) Let gε (x) ≡ ε−α |x|α−n χB(ε) (x) and hε (x) ≡ ε−α |x|α−n χRn \B(ε) (x) for
x ∈ Rn and ε > 0. Use the weak-∗ convergence of {fj }∞ j=1 to show that
F (x) − Iα f (x) = limj→∞ εα gε ∗ (fj − f )(x) for almost all x ∈ Rn .
(2) Prove that sup kgε ∗ (fj − f )kHp0 < ∞.
j∈N q0

(3) Conclude that F (x) = Iα f (x) for almost all x ∈ Rn .

Exercise 116. For a Banach lattice E(µ) over a measure space (X, B, µ) and
a > 0, the a E(µ), a times E(µ), is given by (E(µ), ak · kE(µ) ). Then show that
(aE(µ))0 = a−1 E 0 (µ). Hint:f ∈ E(µ) 7→ af ∈ E(µ) is an isomorphism.
 Predual of Morrey spaces 389

Exercise 117. Let E(µ) be a Banach lattice over a measure space (X, B, µ).
By reexamining the proof of Theorem 352, prove that E(µ) = E 00 (µ) if E(µ)
has the Fatou property.
Exercise 118. [158, Lemma 4.6] Let 0 < p < ∞ and α < max(1, p). Assume
that q ∈ (p, ∞] ∩ [1, ∞]. Suppose that we have a sequence {Qk }∞
k=1 of cubes
and a sequence {Fk }∞ q n + n
k=1 ⊂ L (R ) ∩ M (R ). Then show that

∞
X ∞
X
M (α) χQk · Fk . χQk M (q) Fk .
k=1 Lp k=1 Lp

9.2 Choquet integral and predual spaces

There are couple of integration theories, Lebesgue integral, Riemannian
integral, Henstock integral and so on. Choquet integral is one of the integration
theories. Here, we are interested in other representations of predual spaces of
Morrey spaces.
To solidify our idea, we first introduce the notion of Hausdorff capacity
in Section 9.2.1 and then develop an integration theory in Section 9.2.2. We
obtain some equivalent characterizations in Section 9.2.3.

9.2.1 Hausdorff capacity

0
Here, we aim to rewrite the predual space Hqp0 (Rn ). To this end, we need
to consider integration against the Hausdorff capacity. There are several defi-
nitions of the capacities. However, some of them do not have good properties,
although they are equivalent pointwise. Hence, first we clarify what is cor-
rect for these capacities and then we investigate integration. Based on these
preliminaries, we will consider characterizing the dual spaces.
Since we want to collect some more preliminary facts on measure theory,
recall the theory of capacity briefly.
Definition 88 (Hausdorff capacity). Let 0 ≤ d ≤ n. The d-dimensional
Hausdorff capacity of the set E ⊂ Rn is defined by setting
 
X ∞ ∞ 
d
[
H d (E) ≡ inf |B(xj , rj )| n : E ⊂ B(xj , rj ) ,
 
j=1 j=1

where the infimum takes over all covers {B(xj , rj )}∞

j=1 of E by countable
families of open balls.
Here, E need not be measurable.
390 Morrey Spaces

We can show that the set function H d is monotone, countably subadditive

and vanishes on empty sets. Moreover, the notion of H d can be extended to
d = 0; for any subset E ⊂ Rn , define
 
 N
[ 
H 0 (E) ≡ inf N ∈ N : E ⊂ B(xj , rj ) ,
 
j=1

where the infimum takes over all covers {B(xj , rj )}Nj=1 of E by at most count-
able families of open balls.
As is the case with the Lebesgue integral, it is difficult to calculate H d (E)
for general sets. Here is a special example of E for which we can find H d (E)
exactly.
d
Example 149. Let us show that H d (B(1)) = |B(1)| n for 0 < d ≤ n. It
d
follows from the definition that H d (B(1)) ≤ |B(1)| n . Thus, we need to show
d
H d (B(1)) ≥ |B(1)| n . Let {Bj }∞
j=1 be a ball covering of B(1). Then

  nd
∞ ∞
d d
X X
|Bj | n ≥ |Bj | ≥ |B(1)| n .
j=1 j=1

We want to define a functional by the Hausdorff capacity. This functional

is known to be nonsublinear as Example 150 shows, so occasionally we need to
use an equivalent integral with respect to the d-dimensional modified dyadic
Hausdorff capacity He d.

Definition 89 (Modified dyadic Hausdorff capacity). Let E ⊂ Rn and 0 <

d ≤ n. One defines d-dimensional modified dyadic Hausdorff capacity by
 
X∞ 
H̃0d (E) ≡ inf `(Qj )d ,
 
j=1

where {Qj }∞
j=1 runs over all countable collections of dyadic cubes satisfying
 
∞
[
E ⊂ Int  Qj  . (9.1)
j=1

Example 150. Let n = 2 and a ∈ R. Then H̃01 ([0, 1) × {a}) = 1. In fact,

H̃01 ([0, 1)×{a}) ≤ 1 since [0, 1)×{a} ⊂ Q0m for some m ∈ Z2 . Let us prove the
reverse conclusion. Let {Qλ }λ∈Λ be a covering of [0, 1) × {a}. By Lemma 136
we may assume that {Qλ }λ∈Λ is disjoint. Then denoting by `([x0 , x1 )×{a}) =
x1 − x0 for −∞ < x0 < x1 < ∞, we obtain
X X
`([0, 1) × {a}) = `(([0, 1) × {a}) ∩ Qλ ) ≤ `(Qλ ),
λ∈Λ λ∈Λ
 Predual of Morrey spaces 391

showing H̃01 ([0, 1) × {a}) ≥ 1. Likewise, we can show H̃01 ([0, 1) × {3, π}) = 1(<
2).
Based on a fundamental geometric observation, we see that H d and H̃0d
are not so different.
Proposition 365. Let 0 < d ≤ n. Then H̃0d ∼ H d .
Proof Let E ⊂ Rn . Let {Qj }∞ n
j=1 ⊂ D(R ) be a countable collection
!
∞
S
satisfying (9.1). We write Bj ≡ B(c(Qj ), 2n`(Qj )). Then E ⊂ Int Qj ⊂
j=1
∞ ∞
Bj . As a result, we see that H d . H̃0d . To end the proof, we may
S S
Qj ⊂
j=1 j=1
thus assume H d (E) < ∞.
When we have a ball B ≡ B(x, r), we can find 6n dyadic cubes
Q1 , Q2 , . . . , Q6n such that
n
6
1 `(Qj ) [
≤ ≤ 2, B(x, r) ⊂ Qj . (9.2)
2 r j=1

n
!
6
S
Thus, we see that B(x, r) = Int(B(x, r)) ⊂ Int Qj . With this in mind,
j=1
we suppose that M ∈ (H d (E), ∞). Then we have a covering {B(xj , rj )}∞ j=1
P∞ d
∞
S
such that M > |B(xj , rj )| n and that E ⊂ B(xj , rj ). Then we can find
j=1 j=1
dyadic cubes {Qj,k }j∈N,k∈N∩[1,6n ] such that
6n
!
[ 1 `(Qj,k )
B(xj , rj ) ⊂ Int Qj,k , ≤ ≤ 2, (j ∈ N).
2 rj
k=1

As a result, we have M & H̃0d (E), as was to be shown.

From the definition of the covering, it is easy to see that
H̃0d (E1 ∪ E2 ) ≤ H̃0d (E1 ) + H̃0d (E2 ). (9.3)
According to Example 150, we do not have H̃0d (E1 ∪ E2 ) + H̃0d (E1 ∩ E2 ) =
H̃0d (E1 ) + H̃0d (E2 ). But we have the following weaker version.
Theorem 366 (Strong subadditivity). We have H̃0d (E1 ∪E2 )+ H̃0d (E1 ∩E2 ) ≤
H̃0d (E1 ) + H̃0d (E2 ) for any sets E1 , E2 ⊂ Rn and 0 < d ≤ n.
Proof From (9.3), we may assume that H̃0d (E1 ) + H̃0d (E2 ) < ∞ and that
E1 ∩ E2 6= ∅ without any loss of generality. From the definition of the H̃0d (E1 )
and H̃0d (E2 ), we have only to establish that
∞
X ∞
X
H̃0d (E1 ∪ E2 ) + H̃0d (E1 ∩ E2 ) ≤ `(Q1,j )d + `(Q2,k )d , (9.4)
j=1 k=1
392 Morrey Spaces

whenever we have two collections {Q1,j }∞ ∞

j=1 and {Q2,k }k=1 of dyadic cubes
satisfying  
∞ ∞
!
[ [
E1 ⊂ Int  Q1,j  , E2 ⊂ Int Q2,k .
j=1 k=1

Observe first that

   
∞ ∞ ∞
!
[ [ [
E1 ∩ E2 ⊂ Int  Q1,j  ∩ Int Q2,k = Int  Q1,j ∩ Q2,k  (9.5)
j=1 k=1 j,k=1

and that
   
∞ ∞ ∞
!
[ [ [
E1 ∪E2 ⊂ Int  Q1,j  ∪Int Q2,k = Int  Q1,j ∪ Q2,k  . (9.6)
j=1 k=1 j,k=1

Let

A1 ≡ {(j, k) : Q1,j ( Q2,k }, A2 ≡ {(j, k) : Q1,j ) Q2,k },

3
[
A3 ≡ {(j, k) : Q1,j = Q2,k }, A4 ≡ N × N \ Al .
l=1

Then
∞
[ [ [
Q1,j ∪ Q2,k = Q1,j ∪ Q2,k (9.7)
j,k=1 (j,k)∈A2 ∪A3 ∪A4 (j,k)∈A1 ∪A4

and
∞
[ [ [
Q1,j ∩ Q2,k = Q1,j ∪ Q2,k . (9.8)
j,k=1 (j,k)∈A1 (j,k)∈A2 ∪A3

Consequently, we deduce from (9.6) and (9.7)

 
[ [
E1 ∪ E2 ⊂ Int  Q1,j ∪ Q2,k 
(j,k)∈A2 ∪A3 ∪A4 (j,k)∈A1 ∪A4

and from (9.5) and (9.8)

 
[ [
E1 ∩ E2 ⊂ Int  Q1,j ∪ Q2,k  .
(j,k)∈A1 (j,k)∈A2 ∪A3

Thus,
d d
X X
H̃0d (E1 ∪ E2 ) ≤ |Q1,j | n + |Q2,k | n
(j,k)∈A2 ∪A3 ∪A4 (j,k)∈A1 ∪A4
 Predual of Morrey spaces 393

and
d d
X X
H̃0d (E1 ∩ E2 ) ≤ |Q1,j | n + |Q2,k | n .
(j,k)∈A1 (j,k)∈A2 ∪A3

Therefore, (9.4) follows.

The capacity H̃0d is compatible with decreasing sequences of compact sets.
Theorem 367. Let 0 < d ≤ n. If {Kj }∞ is a sequence of compact sets
j=1 
\∞
decreasing to K, then lim H̃0d (Kj ) = H̃0d  Kj  .
j→∞
j=1

Proof We have only to show that

lim H̃0d (Kj ) ≤ H̃0d (K). (9.9)

j→∞

To this end, we choose a sequence of dyadic cubes {Qj }∞

j=1 arbitrarily so that
!
∞
S
K ⊂ Int Qj . Since K is compact, we can choose ε > 0 so that
j=1

 
∞
[
K ε ≡ {x ∈ Rn : dist(K, x) < ε} ⊂ Int  Qj  . (9.10)
j=1

!
∞
S
Thus, there exists j0 ∈ N such that Kl ⊂ Int Qj for l ≥ j0 . As a result,
j=1
∞
X
lim H̃0d (Kj ) ≤ `(Qj )d . Since the sequence {Qj }∞
j=1 is arbitrary, we obtain
j→∞
j=1
(9.9).
Like the Lebesgue measure, essentially corresponding to the case where
d = n, the compactness of the sets is absolutely necessary as an example
similar to Example 150 shows.
Lemma 368 (Increasing property for open sets). Let 0 < d ≤ n, and let
{Oj }∞ n d
j=1 be a sequence of open sets in R expanding to O. Then lim H̃0 (Oj ) =
j→∞
!
∞
d
S
H̃0 Oj .
j=1

Proof We have only to show that

lim H̃0d (Oj ) ≥ H̃0d (O). (9.11)

j→∞
394 Morrey Spaces

To this end, we may assume that sup H̃0d (Oj ) < ∞; otherwise (9.11) is trivial.
j∈N
Let ε > 0 be fixed. We choose {Qj,k }∞ n
k=1 ⊂ D(R ) so that

∞
!
[
Oj ⊂ Int Qj,k (9.12)
k=1

and
∞
X
`(Qj,k )d ≤ H̃0d (Oj ) + 2−j ε. (9.13)
j=1

Note that sup `(Qj,k )d ≤ ε + sup H̃0d (Oj ). This means that the size of Qj,k is
j,k∈N j∈N
bounded by a constant independent of j and k. Denote by {Qi }i∈I the disjoint
maximal cubes in {Qj,k }j,k∈N . !
∞ ∞ S
∞
 
S S S
Then O = Oj ⊂ Int Qj,k = Int Qi from (9.12). Let
j=1 j=1 k=1 i∈I
m ∈ N be fixed. We write
(1)
{Qi }i∈I ∩ {Q1,j }∞
j=1 = {Qi }i∈I1 . (9.14)

Define !
(1)
[
O1,m ≡ Om ∩ Int Qi (m ∈ N).
i∈I1

(1)
We set {Qm,k }k∈K1,m ≡ {Qm,k : k ∈ N, O1,m ∩ Qm,k 6= ∅}. Then

(1)
[
O1,m ⊂ Qm,k . (9.15)
k∈K1,m

(1) (1) (1)

For each k ∈ K1,m , there exists i ∈ I1 such that Qi ∩ Qm,k 6= ∅. Since Qi is
(1)
maximal in inclusion among {Qj,k }j,k∈N and Qm,k = QJ,K for some J, K ∈ N,
(1) (1)
we have Qm,k ⊂ Qi . As a result, we have

(1) (1)
[ [
Qm,k ⊂ Qi . (9.16)
k∈K1,m i∈I1

Let
(1)
{Q1,k }∞ ∗
k=1 \ {Qi }i∈I1 = {Q1,k,m }k∈J1 . (9.17)
From (9.12) with j = 1, (9.14) and (9.17), we learn
! !
[ (1) [
∗
O1 = O1 ∩ Om ∩ Qi ∪ O1 ∩ Q1,k,m .
k∈I1 k∈J1
 Predual of Morrey spaces 395

Hence
!! !
(1)
[ [
O1 = O1 ∩ Om ∩ Int Qi ∪ O1 ∩ Q∗1,k,m , (9.18)
k∈I1 k∈J1

since we are assuming that O1 is open. From (9.15) and (9.18), we have
(1)
[ [
O1 ⊂ Qm,k ∪ Q∗1,k,m . (9.19)
k∈K1,m k∈J1

As a result, we have
(1)
X X
H̃0d (O1 ) ≤ `(Qm,k )d + `(Q∗1,k,m )d . (9.20)
k∈K1,m k∈J1

Note that (9.13) with j = 1 reads as:

X (1)
X ε
`(Qi )d + `(Q∗1,k,m )d ≤ H̃0d (O1 ) + , (9.21)
2
i∈I1 i∈J1

by virtue of (9.14) and (9.17). By combining (9.20) and (9.21), we obtain

X (1) ε X X (1) ε
`(Qi )d ≤ H̃1d (O1 ) + − `(Q∗1,k,m )d ≤ `(Qm,k )d + . (9.22)
2 2
i∈I1 k∈J1 k∈K1,m

(2)
For j = 2, we mimic the argument above to have a sequence {Qm,k }k∈K2,m
with properties similar to (9.19) and (9.22) above in addition to the property
(1) (2)
{Qm,k }k∈K1,m ∩ {Qm,k }k∈K2,m = ∅. (9.23)
More precisely, we set
(1) (2)
({Qi }i∈I \ {Qm,k }k∈K1,m ) ∩ {Q2,k }∞
k=1 = {Qi,m }i∈I2,m . (9.24)
Define  
(2)
[
O2,m ≡ Om ∩ Int  Qi,m  .
i∈I2,m

(2)
We write {Qm,k }k∈K2,m ≡ {Qm,k : O2,m ∩ Qm,k 6= ∅}. As before, for any
(2) (2)
k ∈ K2,m , Qm,k intersects Qi,m for some i ∈ I2 . In view of (9.24), we have
(9.23). Rearrange the cubes to obtain
(2)
{Q2,k }∞ ∗
k=1 \ {Qi,m }i∈I2,m = {Q2,k }k∈J1 .

(2)
[ [
Going through an argument above, we have O2 ⊂ Qm,k ∪ Q∗2,k
k∈K2,m k∈J2
(2) (2)
`(Qi )d ≤ `(Qm,k )d + 41 ε, corre-
P P
corresponding to (9.19), and
i∈I2,m k∈K2,m
sponding to (9.22).
396 Morrey Spaces
(j)
Continuing this procedure, we can find a collection {Qm,k }k∈Ij,m for j =
(j )
1, 2, . . . , m and a sequence of subsets {Jj }m 1
j=1 of N such that {Qm,k }k∈Ij1 ,m ∩
(j2 ) (j)
[ [
{Qm,k }k∈Ij2 ,m = ∅ for any j1 < j2 ≤ m, that O2 ⊂ Qm,k ∪ Qj,k ,
k∈Kj,m k∈Jj
and that
(j) (j)
X X
`(Qi )d ≤ `(Qm,k )d + 2−j ε.
i∈Ij,m k∈Kj,m
m
(j)
`(Qi )d ≤ 2ε + H̃0d (Om ). Observe that, for any finite
P P
Notice that
j=1 i∈Ij,m
set I0 , there exists an integer M = M (I0 ) such that {Qi }i∈I0 P ⊂ {Qm,k :
m = 1, 2, . . . , M, k = 1, 2, . . .}. Thus, if we let m → ∞, we have `(Qi )d ≤
i∈I
lim H̃0d (Om ) + 2ε, as was to be shown.
m→∞

The capacity H̃0d enjoys the monotone property.

Theorem 369 (Monotone property of H̃0d ). Let 0 < d ≤ n. Whenever we
have an increasing sequence {Ej }∞ n
j=1 of subsets in R , we have
 
∞
[
lim H̃0d (Ej ) = H̃0d  Ej  .
j→∞
j=1

Proof Again, we need to prove

 
∞
[
lim H̃0d (Ej ) ≥ H̃0d  Ej  . (9.25)
j→∞
j=1

To this end, we may assume that sup H̃0d (Ej ) < ∞. We fix ε > 0. Then we can
j∈N
choose an open set Oj such that Ej ⊂ Oj and that H̃ d (Oj ) ≤ H̃0d (Ej ) + 2−j ε.
Since
3
H̃0d (O1 ∪ O2 ) + H̃0d (O1 ∩ O2 ) ≤ H̃0d (O1 ) + H̃0d (O1 ) ≤ H̃0d (E1 ) + H̃0d (E2 ) + ε,
4

we have H̃ d (O1 ∪ O2 ) ≤ H̃0d (E2 ) + 34 ε. Inducting on j, we obtain H̃ d (O1 ∪

O2 ∪ · · · ∪ Oj ) ≤ H̃0d (Ej ) + (1 − 2−j )ε. Letting j → ∞, we obtain
     
[∞ [∞ [J
H̃0d  Ej  ≤ H̃0d  Oj  = lim H̃0d  Oj  ≤ lim H̃0d (EJ ) + ε
J→∞ J→∞
j=1 j=1 j=1

thanks to Lemma 368, proving (9.25).

 Predual of Morrey spaces 397

9.2.2 Choquet integral

Motivated by the Layer-Cake formula, Theorem 5, we define the Choquet
integral against H d [78]. We adopt the idea in Theorem 5 because H d is
not additive. As is the case with the Lebesgue integral, we need to consider
f −1 (λ, ∞] = {x ∈ Rn : f (x) > λ} for f ∈ M+ (Rn ); see Theorem 5.
Definition 90 (Choquet integral against H d ). For any function f : Rn 7→
f (x)dH d (x) against H d is defined by
R
[0, ∞], its Choquet integral
Rn
Z Z ∞
f (x)dH d (x) ≡ H d (f −1 (λ, ∞])dλ,
Rn 0

where the right-hand side is understood asRthe Riemannian integral. Likewise,

for a set E and a function f : E 7→ [0, ∞] f (x)dH d (x) against H d is defined
E
by Z Z ∞
f (x)dH d (x) ≡ H d (f −1 (λ, ∞])dλ,
E 0

where in this case f −1 (λ, ∞] = {x ∈ E : f (x) > λ}.

We do not have to assume that f is measurable in some sense.
Example 151. Let d ∈ (0, n], N ≥ 0 and a > 0. Define f (x) ≡ (a + |x|)−N
for x ∈ Rn . Then f −1 (λ, ∞] = B((λ−1/N − a)+ ). Here, it will be understood
that B(0) = ∅. Hence |f −1 (λ, ∞]| ∼ ((λ−1/N − a)+ )n . If we integrate this over
(0, ∞), we see that
Z Z ∞
(a + |x|)−N dH d (x) = H d ({f > λ})dλ
n
R
Z0 ∞
∼ ((λ−1/N − a)+ )d dλ
0
Z a−N
= (λ−1/N − a)d dλ
0
Z 1
= ad−N (λ−1/N − 1)d dλ
0
d−N
= CN,d a .

Here, CN,d is a constant such that CN,d is finite if and only if d < N . Likewise,
d
we can check that kχB kL1 (H d ) ∼ |B| n for any ball B.

One considers the Choquet integral against H̃0d analogously to that against
H keeping in mind that H d and H̃0d are equivalent according to Proposition
d

365.
398 Morrey Spaces

Definition 91 (Choquet integral against H̃0d ). For any function f : Rn 7→

f (x)dH̃0d (x) against H̃0d is defined by
R
[0, ∞], its Choquet integral
Rn
Z Z ∞
f (x)dH̃0d (x) ≡ H̃0d (f −1 (λ, ∞])dλ,
Rn 0

where the right-hand side is understood as the Riemann R integral. Likewise,

for a set E and a function f : E 7→ [0, ∞] we can define f (x)dH d (x).
E

As before it is difficult to find the precise value of the integral. Here, we

content ourselves with such an example.
Example 152. Let E be a set. Then
Z
χE (x)dH̃0d (x) ≡ H̃0d (E).
Rn

As is the case with the Lebesgue integral, we are not interested in the value
of the integral itself. Let f : Rn 7→ [0, ∞] be any function, and let a > 0, We
mention that we readily obtain
Z Z
d
a · f (x)dH̃0 (x) = a f (x)dH̃0d (x) (9.26)
Rn Rn

by a change of variables.
The following formulas are easy to verify:
Proposition 370. For any increasing sequences {fk }∞
k=1 satisfying 0 ≤ fk ≤
fk+1 , Z Z
lim fk (x)dH̃0d (x) = lim fk (x)dH̃0d (x).
k→∞ Rn Rn k→∞

Proof By the monotone convergence theorem and Theorem 369, we have

Z Z ∞
d
H̃0d fk −1 (λ, ∞] dλ


lim fk (x)dH̃0 (x) = lim
k→∞ Rn k→∞ 0
Z ∞
lim H̃0d fk −1 (λ, ∞] dλ


=
k→∞
Z0 ∞  
= H̃0d x ∈ Rn : lim fk (x) > λ dλ
k→∞
Z0
= lim fk (x)dH̃0d (x),
Rn k→∞

as was to be shown.
One of the important properties of the Choquet integral is the subadditiv-
ity given in the next theorem:
 Predual of Morrey spaces 399

Theorem 371 (Subadditivity). Let 0 < d ≤ n. Then for any functions

f, g : Rn 7→ [0, ∞],
Z Z Z
(f (x) + g(x))dH̃0d (x) ≤ f (x)dH̃0d (x) + g(x)dH̃0d (x). (9.27)
Rn Rn Rn

We prove Theorem 371 step by step. The next lemma, showing that addi-
tivity is available for some special cases, is a key to our observation:
Lemma 372. Let E1 ⊃ E2 ⊃ · · · ⊃ EN be a finite decreasing sequence of sets
N
X N
X −1
in Rn . Set f ≡ χEk and h ≡ χEk = f − χEN . Then
k=1 k=1
Z Z ∞
f (x)dH̃0d (x) = H̃0d (h−1 (λ, ∞])dλ + H̃0d (EN ).
Rn 0

Proof Keeping in mind that f −1 ((λ, ∞]) = ∅ if λ ≥ N , we decompose the

integral in the definition of f (x)dH̃0d (x) into three parts to obtain
R
Rn
Z
f (x)dH̃0d (x)
Rn
Z ∞
= H̃0d (f −1 (λ, ∞])dλ
0
Z N −1 Z N Z ∞
= H̃0d (h−1 (λ, ∞])dλ + H̃0d (EN )dλ + H̃0d (∅)dλ
0 N −1 N
Z ∞
= H̃0d (h−1 (λ, ∞])dλ + H̃0d (EN ),
0

as was to be shown.
Lemma 373. Inequality (9.27) remains true when f and g assume their value
in N0 and sup g ≤ 1.

Proof In view of Proposition 370, we may assume that f is bounded.

N
P
Consequently, we can assume that f = χEk and g = χF1 , where E1 ⊃
k=1
E2 ⊃ · · · ⊃ EN . If N = 1, then we readily have (9.27) thanks to Theorem 366:
Z
(f (x) + g(x))dH̃0d (x) = H̃0d (E1 ∪ F1 ) + H̃0d (E1 ∩ F1 )
Rn
≤ H̃0d (E1 ) + H̃0d (F1 )
Z Z
= f (x)dH̃0d (x) + g(x)dH̃0d (x).
Rn Rn
400 Morrey Spaces
−1
NP
Suppose that (9.27) is true for N = m − 1 ≥ 1. Set f ≡ χEk . Then
k=1
according to Lemma 372, we have
Z Z Z
f (x)dH̃0d (x) + g(x)dH̃0d (x) = h(x)dH̃0d (x) + H̃ d (F1 ) + H̃0d (EN ).
Rn Rn Rn

We have
Z Z
f (x)dH̃0d (x) + g(x)dH̃0d (x)
Rn Rn
Z
= h(x)dH̃0d (x) + H̃ d (F1 ∪ EN ) + H̃0d (F1 ∩ EN )
Rn

by virtue of Theorem 366. Thus,

Z Z
f (x)dH̃0d (x) + g(x)dH̃0d (x)
R n R n
Z
≥ (h(x) + χF1 ∪EN (x))dH̃0d (x) + H̃0d (F1 ∩ EN )
Rn

from the induction assumption to the function h, which satisfies sup h 6= N −1.
Finally, using Lemma 372, we obtain
Z Z
f (x)dH̃0d (x) + g(x)dH̃0d (x)
Rn Rn
Z
≥ (h(x) + χF1 ∪EN (x) + χF1 ∩EN (x))dH̃0d (x)
Rn
Z
≥ (h(x) + χF1 (x) + χEN (x))dH̃0d (x)
n
ZR
≥ (f (x) + g(x))dH̃0d (x),
Rn

which proves (9.27).

We will complete the proof of the subadditivity of the Choquet integral
after another auxiliary step proposed by Akihiko Miyachi.
Lemma 374. Proposition 370 remains true when f and g assume its value
in N0 .
Proof In view of Proposition 370, we may assume that f and g are
bounded. Consequently, we can assume that
N
X N
X
f= χEk , g= χF k , (9.28)
k=1 k=1

where E1 ⊃ E2 ⊃ · · · ⊃ EN and F1 ⊃ F2 ⊃ · · · ⊃ FN .
 Predual of Morrey spaces 401

We prove (9.27) by the induction on N . If N = 1, then (9.27) readily

follows from Lemma 373. Suppose that (9.27) is true for N = m − 1 ≥ 1.
−1
NP −1
NP
Consider a function given by (9.28). Set h ≡ χEk , and k ≡ χFk . Then
k=1 k=1
according to Lemma 372 and Theorem 366, we have
Z Z
f (x)dH̃0d (x) + g(x)dH̃0d (x)
Rn Rn
Z Z
d
= h(x)dH̃0 (x) + k(x)dH̃0d (x) + H̃0d (EN ) + H̃0d (FN )
Rn Rn
Z Z
d
≥ h(x)dH̃0 (x) + k(x)dH̃0d (x) + H̃0d (EN ∪ FN ) + H̃0d (EN ∩ FN ).
Rn Rn
We now invoke the induction assumption to obtain
Z Z
f (x)dH̃0d (x) + g(x)dH̃0d (x)
R n R n
Z
≥ (h(x) + k(x))dH̃0d (x) + H̃0d (EN ∪ FN ) + H̃0d (EN ∩ FN )
Rn
Z Z
d
= (h(x) + k(x))dH̃0 (x) + χEN ∪FN (x)dH̃0d (x) + H̃0d (EN ∩ FN ).
Rn Rn
Next, by using Lemma 373 twice, we obtain
Z Z
d
f (x)dH̃0 (x) + g(x)dH̃0d (x)
Rn Rn
Z
≥ (h(x) + k(x) + χEN ∪FN (x))dH̃0d (x) + H̃0d (EN ∩ FN )
n
ZR
≥ (h(x) + k(x) + χEN ∪FN (x) + χEN ∩FN (x))dH̃0d (x)
Rn
Z
≥ (f (x) + g(x))dH̃0d (x),
Rn
which proves (9.27).
Finally, we prove (9.27) for the general case. Set
fN ≡ 2N min([2−N f ], 2−N N ), gN ≡ 2N min([2−N g], 2−N N ).
Then according to (9.26) and Proposition 370, we have
Z Z
d
(f (x) + g(x))dH̃0 (x) = lim (fN (x) + gN (x))dH̃0d (x)
Rn N →∞ Rn
and
Z Z
f (x)dH̃0d (x) = lim fN (x)dH̃0d (x),
Rn N →∞ Rn
Z Z
g(x)dH̃0d (x) = lim gN (x)dH̃0d (x).
Rn N →∞ Rn
This proves (9.27).
402 Morrey Spaces

9.2.3 Predual spaces of Morrey spaces by way of the

Choquet integral
We present a method of expressing the predual in terms of the Choquet
integral.
Definition 92 (L1 (H̃0d )). Let 0 ≤ d ≤ n. The space LZ1 (H̃0d ) denotes the clo-
∞
sure of Cc (Rn ) with respect to the norm kf kL1 (H̃ d ) ≡ H̃0d ({|f | > λ})dλ.
0
0

Example 153. Let E be a subset of Rn . Then kχE kL1 (H̃ d ) = H̃0d (E) since
0

H̃0d ({χE > λ}) = H̃0d (E)χ(0,1) (λ).

Let µ be a signed measure. We recall Definition 9; µ ∈ Mp1 (Rn ) if and
n− n
only if |µ|(B(x, r)) . r p for all balls B(x, r), where |µ| denotes the total
variation of µ. The following is the fundamental duality result:
Lemma 375. Let 1 ≤ p < ∞. Write d ≡ n − np . Let µ ∈ Mp1 (Rn ) and
f ∈ Cc (Rn ). Then
Z
f (x)dµ(x) . kµkMp1 kf kL1 (H̃ d ) .
0
Rn

Proof By the triangle inequality and the Layer-Cake formula

Z Z Z ∞
f (x)dµ(x) ≤ |f (x)| d|µ|(x) = |µ|({|f | > λ})dλ. (9.29)
Rn Rn 0

Let λ > 0 be fixed. Let {B(xj , rj )}∞

j=1 be a covering of the set {|f | > λ}.
Then
∞
X ∞
X
|µ|({|f | > λ}) ≤ |µ|(B(xj , rj )) ≤ kµkMp1 rj d .
j=1 j=1

As a result, taking the infimum of the ball covers {B(xj , rj )}∞

j=1 of E, we
conclude
|µ|({|f | > λ}) . kµkMp1 H d ({|f | > λ}). (9.30)
Inserting (9.30) to (9.29) and using Proposition 365, we obtain
Z Z ∞
f (x)dµ(x) . kµkMp1 H̃0d ({|f | > λ})dλ = kµkMp1 kf kL1 (H̃ d ) ,
0
Rn 0

as was to be shown.
If we choose d suitably, L1 (H̃0d ) can be considered as the predual of
0
Mp1 (Rn ).
Unlike the duality Hqp0 (Rn )–Mpq (Rn ), the characterization is some-
what indirect because we need to depend on a dense subspace of L1 (H̃0d ).
 Predual of Morrey spaces 403

Proposition 376. Let 1 < p < ∞. Write d ≡ n − np . Then a predual space

of Mp1 (Rn ) is given by L1 (H̃0d ). More precisely, for µ ∈ Mp1 (Rn ), there exists
a unique linear functional Lµ on L1 (H̃0d ) such that
Z
Lµ (ω) = ω(x)dµ(x) (9.31)
Rn

for all ω ∈ Cc (Rn ), and conversely any continuous linear functional L on

L1 (H̃0d ) is realized as L = Lµ for some µ ∈ Mp1 (Rn ).
Proof According to Lemma 375, we see that such Lµ exists. Conversely,
for any linear functional L, we consider the composition ι : Cc (Rn ) → L1 (H̃0d )
and L : L1 (H̃0d ) → C to obtain a linear functional L ◦ ι : Cc (Rn ) → C. By the
Riesz representation theorem [118, §1.8], there exists a Radon measure µ such
that (9.31) holds for all f ∈ Cc (Rn ). Let us show µ ∈ Mp1 (Rn ). Then
kµk(B(x, r)) . sup {|L(f )| : f ∈ Cc (B(x, 2r)), kf kL∞ ≤ 1}
n o
≤ sup kLk(L1 (H̃ d ))∗ kf kH̃ d : f ∈ Cc (B(x, 2r)), kf kL∞ ≤ 1
0 0

d
. kLk(L1 (H̃ d ))∗ r
0

thanks to Example 153, as was to be shown.

We see once again that Mpq (Rn ) is realized as the dual of a Banach space.
 
Theorem 377. Let 1 ≤ q ≤ p < ∞. Set d ≡ n 1 − pq . Then kf kMpq ∼
n o
sup kf kLq (w) : w ∈ M+ (Rn ), kwkL1 (H̃ d ) ≤ 1 for any f ∈ L0 (Rn ).
0

q
Proof For a ball B, by letting w ≡ |B|−1+ p χB , we obtain
n o
kf kMpq . sup kf kLq (w) : w ∈ M+ (Rn ), kwkL1 (H̃ d ) ≤ 1
0

from the definition of the Morrey norm k·kMpq . To prove the converse relation,
Z
we fix w ∈ M+ (Rn ) with kwkL1 (H̃ d ) ≤ 1 and we set µ(E) ≡ |f (x)|q dx for
0
E
a Borel set E. Then µ is a measure. We observe
Z
|f (x)|q w(x)dx . kµkMp1 ∼ (kf kMpq )q
Rn

from Lemma 375. Combining these inequalities completes the proof.

We are now interested in another expression of a predual space of Mpq (Rn )
with 1 < q < p < ∞.
 0

Definition 93. Let 1 < u < v < ∞. Define d ≡ n 1 − uv 0 ∈ (0, n). Let
g ∈ L0 (Rn ). Then define
kgkVvu ≡ inf kgkLv (w1−v ) , (9.32)
w
404 Morrey Spaces

where w runs over all elements in B(Rn ) such that

{w = 0} ⊂ {g = 0}, kwkL1 (H̃ d ) ≤ 1.

0

The space Vvu (Rn ) denotes the linear space of g ∈ L0 (Rn ) for which there
exists w ∈ L1 (H̃0d ) ∩ B+ (Rn ) which does not vanish almost everywhere on
{g 6= 0}.
Example 154. Let 1 < u < v < ∞. If g ∈ Lvc (Rn ), then g ∈ Vvu (Rn ). In fact,
2
we can take w(x) ≡ κe−|x| , x ∈ Rn with some suitable κ > 0 in Definition
93. If in addition g is supported on a ball B, then kgkVvu ≤ kχkχ B gkLv
Bk 1 d
.
L (H̃0 )
d
− nv
|B| 0
kχB gkLv .
We show that Vvu (Rn ) is a ball Banach function space.
Lemma 378. Let 1 < u ≤ v < ∞. Then Vvu (Rn ) is a ball Banach function
space.
Proof We content ourselves with the completeness of Vvu (Rn ): Let
∞
{gj }∞ u n
P
j=1 ⊂ Vv (R ) satisfy kgj kVvu < ∞. Write λj ≡ kgj kVvu for j ∈ N.
j=1
By the definition of λj there exists wj ∈ B+ (Rn ) such that

{wj = 0} ⊂ {gj = 0}, kwj kL1 (H̃ d ) ≤ 1,

0

Z ! v1
|gj (x)|v wj (x)1−v dx . λj .
{wj 6=0}

∞
P
Let W ≡ λj wj .
j=1
By Hölder’s inequality, we have
 q  q−1  
X∞ X∞ ∞
X
 |gj | ≤  λj wj   λj |gj |wj 1−q  ,
j=1 j=1 j=1

so that  q
∞
X ∞
X
 |gj | W 1−q ≤ λj |gj |wj 1−q .
j=1 j=1

If we integrate this inequality over {W 6= 0}, we obtain

 q
Z ∞
X X∞
 |gj (x)| W (x)1−q dx . λj .
Rn j=1 j=1
 Predual of Morrey spaces 405
∞ ∞
|gj | ∈ Vvu (Rn ). Hence gj ∈ Vvu (Rn ). If we
P P
Thus, since kW kL1 (H̃ d ) ≤ 1,
0
j=1 j=1
∞
set W k ≡
P
wj , then
j=k
q
Z ∞
X ∞
X
gj (x) W k (x)1−q dx . λj .
Rn j=k j=k

∞
gj = g in the topology of Vvu (Rn ).
P
Thus,
j=1
0
So far we showed that Vqp0 (Rn ) is a ball Banach function space. We will
0 0
establish that Vqp0 (Rn ) and Hqp0 (Rn ) are the same.
Theorem 379. Let 1 < q ≤ p < ∞. Then a predual space of Mpq (Rn ) is
0
Vqp0 (Rn ) with the equivalence of norms under the following pairing:
Z
0
hf, gi ≡ f (x)g(x)dx (f ∈ Mpq (Rn ), g ∈ Vqp0 (Rn )).
Rn

Moreover, for all f ∈ Mpq (Rn )

 
0
kf kMpq ∼ sup kf · gkL1 : g ∈ Vqp0 (Rn ), kgkV p0 ≤ 1 .
q0

0
Proof Let g ∈ Vqp0 (Rn ) and f ∈ Mpq (Rn ). Then by Hölder’s inequality
Z  q1 Z  10
q
q q0 1−q 0
kf · gkL1 ≤ |f (x)| w(x)dx |g(x)| w(x) dx
Rn Rn

for all Borel functions w that are allowed to vanish only on the set {g = 0}.
Therefore, it follows from Theorem 377 that
Z  10
q
q0 1−q 0
kf · gkL . kf kMq
1 p |g(x)| w(x) dx
Rn

for all such Borel functions w. Thus, from (9.32), we obtain the desired equal-
ity.
0
Conversely, suppose that we have a linear functional L on Vqp0 (Rn ). Then
0
for all balls B and a function g ∈ Lq (Rn ) with support B, we have
q0
 
− q10 1− p 1 1
|L(g)| ≤ kgkV p0 . |B| 0
kgkLq0 = |B| q − p kgkLq0 ,
q0

thanks to Example 154. Thus, we can mimic the proof of Theorem 347.
As a consequence, we obtain an equivalent expression of the Köthe dual
of Mpq (Rn ).
406 Morrey Spaces
0
Corollary 380. Let 1 < q ≤ p < ∞. Then the Köthe dual of Vqp0 (Rn ) is
0
Mpq (Rn ), or equivalently the Köthe dual of Mpq (Rn ) is Vqp0 (Rn ).
0
Thanks to Theorem 355 and Corollary 380, we see that two spaces Hqp0 (Rn )
0
and Vqp0 (Rn ) are the same.
0
Theorem 381. Let 1 < q ≤ p < ∞. Then the block space Hqp0 (Rn ) coincides
0
with Vqp0 (Rn ) with equivalence of norms.
Proof Simply combine Theorem 355 and Corollary 380.

9.2.4 Exercises
Exercise 119. By a change of variables, prove (9.26).
Exercise 120. For all nonempty sets E ⊂ Rn , show that H 0 (E) ≥ 1, and
that H 0 (E) = 1 if and only if E is bounded. Hint: When H 0 (E) = 1, then we
can find a ball that covers E.
Exercise 121. Let 0 < d ≤ n, and let E ⊂ Rn . Then show that H̃0d (E) =
inf{H̃0d (O) : E ⊂ O, O is open}.

9.3 Notes
Section 9.1
General remarks and textbooks in Section 9.1
Zorko [472], Adams [4] and Kalita [224] studied preduals of Morrey spaces.
Motivated by the work [221], Zorko considered the atomic space [472, p. 589]
unlike what we did in this book. Adams and Xiao investigated the relation
between them [8]. See [298] for the duality of central Herz–Morrey–Musielak–
Orlicz spaces of variable exponents.
See [457] for the duality in terms of the heat kernel.

Section 9.1.1
0
The definition of Hqp0 (Rn ) in Definition 83, originally denoted by H q,ϕ in
[472] in the framework of generalized Morrey spaces (see Section 12.1), is due
0
to Zorko. Examples of functions in Hqp0 (Rn ) can be found in [218]; see [218,
Example 2.4] for Example 143. Example 142, which can be used to calculate
the norm kχQ kHp0 , is due to Komori and Mizuhara [252, Lemma 1]. Although
q0
the position of the cubes in the definition of blocks is not specified, we can
 Predual of Morrey spaces 407

use the dyadic grid D(Rn ) for this purpose; see [217, Lemma 2.4] and [397,
Lemma 2.2] for Theorem 344. Note that Theorem 348 goes back to a work
by Izumi, Sato and Yabuta [217, Lemma 2.5] as well as the one by Sawano
and Tanaka [398, Theorem 1.3]. The lattice property, Lemma 342, is [396,
Proposition 3.3].
Theorem 351 is [29, Theorem 2.2] when E(µ) is a Banach function space;
let ρ be a function norm. Then the associate norm ρ0 is itself a function norm.
Theorem 352 is [29, Theorem 2.7] when E(µ) is a Banach function space.
More precisely, every Banach function space E(µ) coincides with its second
associate space E 00 (µ). In other words, a function f belongs to E(µ) if and
only if it belongs to E 00 (µ), and in that case

kf kE (µ) = kf kE 00 (µ) , (f ∈ E(µ) = E 00 (µ)).

Theorem 347 was investigated by Zorko [472]; see [224] as well. We refer to
[8] and [151] for more recent characterizations.

Section 9.1.2
0 0
We showed in Theorem 345 that Lqc (Rn ) is dense in Hqp0 (Rn ); see [341,
Lemma 2.6]. If the function is compactly supported, then we have an equiva-
0
lent expression of the norm in Hqp0 (Rn ); see [398, Proposition 5.3] for Theorem
346.

Section 9.1.3
Zorko characterized the predual in [472, Proposition 5]; see Theorem 347.

Section 9.1.4
Kantorovich and Akilov proved that the Fatou property of Banach lattices
is a key for the “bi”-Köthe dual to be back to the original space. [226]; see
Theorem 354.

Section 9.1.5
Köthe and Toeplitz began the study of certain pairs of subspaces of the
space of all real sequences [239, 240, 242]. Their theory has been generalized
by Dieudonné, Cooper and Lorentz and Wertheim [87, 103, 290]. Theorem 352
can be located as a counterpart to ball Banach function spaces of the result
by Lorentz and Luxemberg. We essentially followed the argument used in [29,
0
Theorem 4.1] to specify a predual space of Hqp0 (Rn ). Yang and Yuan showed
that the Morrey space Mpq (Rn ) is not reflexive. We followed [398, Example
5.2] for the proof. See [438, Corollary 2.20] for Theorem 356. The case where
q = 1 can be covered; see [303, Theorem 4.1].
408 Morrey Spaces

Section 9.1.6
We refer to [420, Chapter 8, Lemma 5] for the averaging technique for
Lebesgue spaces. We followed the averaging technique used in [212]; see The-
orem 363. See [388] for the applications to elliptic differential operators and
[180] for the applications to bilinear fractional integral operators. See [315] for
the case of Herz spaces.
See [255] for another type of decomposition of Morrey spaces.

Section 9.2
General remarks and textbooks in Section 9.2
See standard textbooks [6, Chapter 3], [112], [231], [382, §1.1.4] for the
Choquet integral. In particular, we can find Proposition 9.31 in [6, Theorem
5.1]. See [6, Theorem 5.2] for Lemma 375. In [6, Section 5.2] Adams showed
that the various predual spaces defined in this book coincide.
See [156, Appendix J], [415, Chapter I, §3] for the Whitney decomposi-
tion of the domain. Using the capacity, we can consider the notion of quasi-
continuity.
Definition 94 (H̃0d -quasi continuous). A function f is H̃0d -quasi continuous,
if for all ε > 0 there exists an open set such that H̃0d (O) < ε and that f is
continuous outside O.

Section 9.2.1
Yang and Yuan showed that H̃ is subadditive in [462]; see Theorem 366.

Section 9.2.2
The theory of the Choquet integral goes back to Choquet [78]. Orobitg
and Verdera applied the Choquet integral to the boundedness of the Hardy–
Littlewood maximal operators [349]. See [426, Theorem 1] for a generaliza-
tion to fractional maximal operators. Kuznetsov considered a vector-valued
inequality in [261]. Essoh, Fofana and Koua passed this result to Morrey spaces
[114]. Example 151 comes from [349, Lemma. 1]. See the lecture note [4] for
more. It may be interesting that the Hardy-Littlewood maximal operator is
L1 -bounded if the integral is considered in the sense of Choquet; see [4, The-
orem A].

Section 9.2.3
Adams gave a characterization of a predual space of Mp1 (Rn ); see [4, Propo-
sition 1]. The definition of the space Vvu (Rn ) is due to Adams and Xiao [8,
p.1632]; see Definition 93. Adams and Xiao proved Theorems 377, 379 and
381 in [8, Proposition 1 and Theorem 2.2], [8, Theorem 2.2] and [8, Theorem
3.3], respectively.
Chapter 10
Linear and sublinear operators in
Morrey spaces

One of the important problems in the theory of function spaces is the bound-
edness property of operators. This is important because it yields many appli-
cations in various fields of mathematics such as partial differential equations
and potential theory. There is a standard technique which is a local/global
strategy. However, unlike Lebesgue spaces, the operators need to be care-
fully defined in some cases. For example, singular integral operators, dealt
with in Section 10.4, must be handled with care because they are defined by
an approximation of integral kernels. Chapter 10 presents methods to define
operators in Morrey spaces. Likewise, care must be taken when handling com-
mutators in Section 10.5. The Hardy–Littlewood maximal operator (including
the sharp maximal operator) and fractional integral operators are discussed
in Sections 10.1 and 10.3, respectively.
Basically, the boundedness property of operators can be proved by the local
estimate initiated by Burenkov and Guliyev. However, to make the proof self-
contained we consider a direct proof which still recasts the flavor of the local
estimates.

10.1 Maximal operators in Morrey spaces

Having set down the structure of Morrey spaces and block spaces in Chap-
ter 9, we are now oriented to the boundedness property of operators. We con-
tinue to work on Rn equipped with the Lebesgue measure dx as before. We
will consider the boundedness properties of Morrey spaces and local Morrey
spaces in Sections 10.1.1 and 10.2.2, respectively.

10.1.1 Maximal operator in Morrey spaces

Our main interest here is the Hardy–Littlewood maximal operator M .
We now present a typical argument about the proof of the boundedness of
operators in Morrey spaces. As the proof shows, we depend heavily upon the

409
410 Morrey Spaces

so called local/global strategy. We fix a cube or a ball and decompose functions

according to its five times expansion.
Theorem 382. Let 1 < q ≤ p < ∞. Then kM f kMpq .q kf kMpq whenever
f ∈ Mpq (Rn ).
Before we come to the proof of Theorem 382, it is noted that Lemma 130
is a key tool for the proof. The proof will be a model case of the proof of the
boundedness of operators acting on Morrey spaces.
Proof We will take the local/global strategy. We have only to show, from
the definition, that
Z  q1
1
− q1 q
|Q| p M f (y) dy . kf kMpq (Q ∈ Q). (10.1)
Q

Write f = f1 + f2 , where f1 = f on 5Q and f2 = f outside 5Q. The estimate

of (10.1) can be split into the local estimate
Z  q1
1
− q1 q
|Q| p M f1 (y) dy . kf kMpq (10.2)
Q

and the global estimate

Z  q1
1 1
|Q| p − q M f2 (y)q dy . kf kMpq . (10.3)
Q

We know that M is Lq (Rn )-bounded. Thus, by expanding the integration

domain and inserting the definition of f1 into what we have obtained from the
Lp (Rn )-boundedness, we obtain
Z  q1 Z  q1 Z  q1
M f1 (y)q dy ≤ M f1 (y)q dy . |f (y)|q dy .
Q Rn 5Q

From the definition of the Morrey norm k · kMpq ,

Z  q1 Z  q1
1
− q1 q 1
− q1 q
|Q| p M f1 (y) dy . |5Q| p |f (y)| dy ≤ kf kMpq .
Q 5Q

Thus (10.2) is proved.

It remains to prove (10.3). If we use Lemma 130, then we obtain
Z  q1
1
− q1 q 1
|Q| p M f2 (y) dy . |Q| p sup mR (|f |).
Q R∈Q] (Q)

Taking into account Mpq (Rn ) ,→ Mp1 (Rn ), we see that

Z  q1 Z
1 1 1
|Q| p − q M f2 (y)q dy . sup |R| p −1 |f (y)|dy = kf kMp1 ≤ kf kMpq .
Q R∈Q R

Consequently (10.3) is proven.

 Linear and sublinear operators in Morrey spaces 411

We obtain the vector-valued Morrey-boundedness of the Hardy–Littlewood

maximal operator by taking the local/global strategy again. We will write

{fj }∞
j=1 Mp u ≡ k{fj }∞
j=1 k`u Mp
q (` ) q

for every sequence of {fj }∞ 0 n

j=1 ⊂ L (R ).

Theorem 383. Suppose that the parameters p, q and u satisfy 1 < q ≤ p <
∞ and 1 < u ≤ ∞. Then {M fj }∞ ∞
j=1 Mp (`u ) . {fj }j=1 Mp (`u ) for every
q q

sequence of {fj }∞ 0 n
j=1 ⊂ L (R ).

Proof We modify the proof of Theorem 382. Instead of (10.1), we have

to show that
Z  q1
1
− q1 ∞
|Q| p u
q
k{M fj (y)}j=1 k` dy . {fj }∞
j=1 Mp (`u ) (Q ∈ Q).
q
Q

We consider a similar decomposition as before; write fj = f1,j + f2,j , where

fj,1 ≡ f on 5Q and fj,2 ≡ f outside 5Q. Estimates (10.2) and (10.3) corre-
spond to
Z  q1
1 1
I ≡ |Q| p − q k{M fj,1 (y)}∞ q
j=1 k`u dy . {fj }∞
j=1 Mp u , (10.4)
q (` )
Q
Z  q1
1 1
II ≡ |Q| p − q k{M fj,2 (y)}∞ q
j=1 k`u dy . {fj }∞
j=1 Mp u , (10.5)
q (` )
Q

respectively. For (10.4), we use Theorem 270 and we proceed as in (10.2): We

first expand the integration domain.
Z  q1
1 1
I ≤ |Q| p − q k{M fj,1 (y)}∞ q
j=1 k`u dy .
Rn

Next, we use the vector-valued boundedness of M to have

Z  q1 Z  q1
1 1 1 1
I . |Q| p−q k{fj,1 (y)}∞ q p−q k{fj }∞ q
j=1 k`u dy = |Q| j=1 k`u dy .
Rn 5Q

By the definition of the Morrey norm k · kMpq , I . {fj }∞ j=1 Mp u . Thus,

q (` )
the estimate of I is valid.
For the estimate of II, recall Lemma 130. By Lemma 130 with f replaced
∞
P
by fj , we have M fj,2 (y) . m2k Q (|fj |). By Minkowski’s inequality, we have
k=1

∞ Z
X 1
k{M fj (y)}∞
j=1 k`u . k{fj (z)}∞
j=1 k`u dz (y ∈ Q).
|2k Q| 2k Q
k=1
412 Morrey Spaces

Consequently,
Z ∞ Z
1
−1
X 1 1
|Q| p k{M fj (y)}∞
j=1 k`u dy . |Q| p k{fj (z)}∞
j=1 k`u dz.
Q |2k Q| 2k Q
k=1

Going through an argument similar to (10.3), we obtain (10.5).

The space Mp1 (Rn ) is a nasty space as the following example shows:
Example 155. Let 1 < p < ∞. Let 0 < r < 1/2. Define f0 (t) ≡ rt and
f1 (t) ≡ 1 − r + rt so that f0 and f1 are affine mappings from [0, 1] to [0, 1].
Define inductively E0 ≡ [0, 1], Ek+1 ≡ f0 (Ek ) ∪ f1 (Ek ) for k ∈ N. Choose
1 k
r > 0 so that r p = 2r. Then we have kχEk kMp1 (R) ∼ kχ[0,rk ] kLp = r p = (2r)k .
Meanwhile, we have the lower bound of M χEk from below:
M χEk ≥ χEk + 2rχEk−1 \Ek + · · · + (2r)l χEk−l \Ek−l+1 + · · · + (2r)k χE0 \E1 .
As a result,
kM χEk kMp1 (R) ≥ (2r)k + 2r · (2r)k−1 (1 − 2r) + · · · + (2r)k (1 − 2r)
= k(1 − 2r)(2r)k + (2r)k+1 .
This implies that kM χEk kMp1 (R) ≥ k(1 − 2r)kχEk kMp1 (R) . Since k is arbitrary,
this relation shows that M is NOT bounded on Mp1 (R).
A natural question appears: what we can say if f ∈ L0 (Rn ) satisfies M f ∈
p
M1 (Rn )? Since Mp1 (Rn ) contains Mpq (Rn ) and M is bounded on Mpq (Rn )
if 1 < q ≤ p < ∞, M f ∈ Mp1 (Rn ) does not imply f = 0. Meanwhile, if
f ∈ L0 (Rn ) satisfies M f ∈ L1 (Rn ), then f = 0. So one may guess that there
is something different from Lebesgue spaces. We will completely characterize
the functions f ∈ L0 (Rn ) for which M f ∈ Mp1 (Rn ) in the second book.
Using the weak-boundedness of operators, we can prove various results.
For example, we can supplement Theorem 382.
Theorem 384. For 1 ≤ p < ∞, M is bounded from Mp1 (Rn ) to WMp1 (Rn ).
Proof We have only to show, from the definition, that
1 1
t|Q| p −1 λM f (2t) = t|Q| p −1 |{x ∈ Q : M f (x) > 2t}| . kf kMp1
(Q ∈ Q)
(10.6)
for all t > 0. Write f = f1 + f2 , where f1 = f on 5Q and f2 = f outside 5Q.
The estimate of (10.6) can be split into:
1 1
t|Q| p −1 λM f1 (t) = t|Q| p −1 |{x ∈ Q : M f1 (x) > t}| . kf kMp1 , (10.7)
1 1
t|Q| p −1 λM f2 (t) = t|Q| p −1 |{x ∈ Q : M f2 (x) > t}| . kf kMp1 . (10.8)
1 n
We know that M is weak L (R )-bounded. Thus, (10.7) can be shown easily ;
Z Z
1 1 1
−1 −1 −1
t|Q| p λM f1 (t) . |Q| p |f (y)|dy ' |5Q| p |f (y)|dy ≤ kf kMp1 .
5Q 5Q
 Linear and sublinear operators in Morrey spaces 413

It remains to prove (10.8). Note that {x ∈ Q : M f2 (x) > t} = ∅, if

t & mR (|f |) for any cube R. Consequently, we may assume otherwise. Then
we obtain
1 1
t|Q| p −1 λM f2 (t) . |Q| p sup mR (|f |)
R∈Q] (Q)
1 1
from Lemma 130. Thus, t|Q| p −1 λM f2 (t) . sup |R| p mR (|f |) = kf kMp1 . Con-
R∈Q
sequently (10.8) is proven.
The following remark shows that the weak Morrey spaces are not so arti-
ficial.
Remark 15. It is noteworthy that kM f kWMp1 ∼ kf kMp1 for all f ∈ Mp1 (Rn ).
1
In fact, given a cube Q by letting λ0 ≡ mQ (|f |), we obtain λ0 |Q| p −1 |{x ∈ Q :
1
M f (x) ≥ λ0 }| = |Q| p λ0 .
We end this section with the boundedness of the Hardy–Littlewood max-
imal operator in predual spaces.
0
Theorem 385. Let 1 < q ≤ p < ∞. Then M is bounded on Hqp0 (Rn ).
Proof It suffices to show that M maps (p, q)-blocks to a bounded set
0
in Hqp0 (Rn ). Let a be a (p, q)-block supported on Q. This follows from the
∞
P
decomposition M a = χ2Q M a + χ2k Q\2k−1 Q M a.
k=1

10.1.2 Maximal operator in local Morrey spaces

If we mimic the proof above, then we obtain the boundedness of the Hardy–
Littlewood maximal operator in local Morrey spaces.
Theorem 386. Let 1 < q ≤ p < ∞. Then kM f kLMpq .q kf kLMpq for all
f ∈ LMpq (Rn ).
Proof We omit the proof for interested readers; see Exercise 123.
We also have the vector-valued maximal inequality for local Morrey spaces.
We will write
{fj }∞ ∞
j=1 LMp (`u ) ≡ k{fj }j=1 k`u LMp
q q

for every sequence of {fj }∞

j=1
0
⊂ L (R ). n

Theorem 387. Suppose that the parameters p, q and u satisfy 1 < q ≤ p < ∞
and 1 < u ≤ ∞. Then {M fj }∞ ∞
j=1 LMp (`u ) . {fj }j=1 LMp (`u ) for every
q q

sequence of {fj }∞ 0 n
j=1 ⊂ L (R ).
Proof We omit the proof; see Exercise 123.
We avoided defining LMpq (Rn ) for −∞ < p < 0 < q < ∞. One of the
reasons is that M fails to be bounded as the following example shows.
414 Morrey Spaces

Example 156. Let −∞ < p < 0 < q < ∞. Define the local Morrey norm
k · kLMpq formally by (1.1). Then since | · |θ χRn \B(1) ∈ LMpq (Rn ) whenever
0 < θ < − np , the maximal operator M does not map LMpq (Rn ) boundedly.

10.1.3 Exercises
Exercise 122. For a cube Q ∈ Q, we define Q1 (Q) ≡ {R ∈ Q : R ∩ Q 6=
∅, R∩(Rn \8Q) 6= ∅}. For a function f ∈ Mp1 (Rn ), define Λ ≡ sup mR (|f |).
R∈Q1 (Q)
Let us write Eλ ≡ {x ∈ Q : M f (x) > λ} for λ > Λ. Then for any x ∈ Eλ ,
show that we can find a cube Q0 such that Q0 contains x, that mQ (|f |) > λ
1
and that Q0 ⊂ 8Q. Prove also that |Q| p Λ . kf kMp1 .
Exercise 123.
(1) Reexamine the proof of Theorem 382 to prove Theorem 386.
(2) Prove Theorem 387 by mimicking the proof of Theorem 383.

10.2 Sharp maximal operators in Morrey spaces

One of the prominent roles of Morrey spaces is that Morrey spaces can
describe the local regularity and the global regularity more precisely than
Lebesgue spaces. This aspect can be seen from the sharp maximal inequality.
As is seen from the definition of the sharp maximal operator, it annihilates
the constant function 1. This means that it annihilates the global regularity.
To recover the global regularity, we can use Morrey spaces. Section 10.2.2
defines the sharp maximal operator and then formulates the sharp maximal
inequality. As an application, we will prove the boundedness of commutators,
which is left open in Section 5.2.
To control the singularity of operators, we use the sharp maximal opera-
tors following the idea of Lerner and Hytönen. Section 10.2.2. considers the
counterpart to local Morrey spaces

10.2.1 Sharp maximal inequalities for Morrey spaces

As an application of Theorem 174, following Definition 51, we will prove
the following sharp maximal inequality for the operator M2],D
−n−2 given by

M2],D ],D
−n−2 f (x) ≡ sup χQ (x)M2−n−2 ,Q f (x) (x ∈ Rn ).
Q∈Q

Theorem 388 (Sharp-maximal inequality for Morrey spaces). Let 0 < q̄ <
q ≤ p < ∞. Then kM2],D 0 n
−n−2 f kMq + kf kMq̄ ∼ kf kMq for all f ∈ L (R ).
p p p
 Linear and sublinear operators in Morrey spaces 415

Proof Since M2],D

−n−2 f . Mf , where M is the maximal operator given by

Mτ f (x) = Mf (x) ≡ sup χQ (x)(f χQ )∗ (τ |Q|) (x ∈ Rn ), (10.1)

Q∈Q

one inequality . is clear. Let us prove the reverse inequality &.

1 1
Let Q0 be a fixed cube. Then we need to estimate |Q0 | p − q kf kLq (Q0 ) . By
Theorem 174, we have a decomposition: for almost every x ∈ Q0 ,
∞ X
X
|f (x) − Med(f ; Q0 )| ≤ 2 ω2−n−2 (f ; Qjk )χQj (x),
k
j=1 k∈Kj

so that we obtain
1 1 1
|Q0 | p − q kf kLq (Q0 ) ≤ |Q0 | p |Med(f ; Q0 )|
∞ X
1 1 X
+ 2|Q | 0 p−q ω2−n−2 (f ; Qjk )χQj .
k
j=1 k∈Kj
Lq (Q0 )

Set Ekj ≡ Qjk \ Qj+1 j

S
k0 . Then we have χQj (x) ≤ 2M χE j (x), since 2|Ek | ≥
k k
k0 ∈Kj+1
|Qjk |. Consequently, by letting u > (q̄)−1 , we obtain
1 1 1
|Q0 | p − q kf kLq (Q0 ) . |Q0 | p |Med(f ; Q0 )|
∞ X
1 1 X
+ |Q0 | p − q ω2−n−2 (f ; Qjk )(M χE j )u .
k
j=1 k∈Kj
Lq (Q0 )

If we use the Fefferman-Stein vector-valued inequality for Morrey spaces (The-

orem 145), then
1 1 1
|Q0 | p − q kf kLq (Q0 ) . |Q0 | p |Med(f ; Q0 )|
∞ X
1 1 X
+ |Q0 | p − q ω2−n−2 (f ; Qjk )χE j .
k
j=1 k∈Kj
Lq (Q0 )

Thus, we obtain
Z  q1
1 1 1 1 1
0
|Q | p−q 0
kf kLq (Q0 ) . |Q | |Med(f ; Q )| + |Q | 0 0 p−q M2n,] q
−n−2 (x) dx .
p

Q0

From the definition of the Morrey norm, we conclude kf kMpq . kf kMpq̄ +

kM2],D
−n−2 f kMq . Thus, we obtain the desired result.
p

Similar to Theorem 166, we can prove the following result:

Theorem 389 (Sharp maximal inequality for Morrey spaces). Let 1 < q ≤
p < ∞. Then kf kMpq .p kM ] f kMpq for all f ∈ L0 (Rn ) with min(M f, 1) ∈
Mpq (Rn ).
416 Morrey Spaces

We have to be careful since f does not always satisfy (4.2) as the example
of f = χF shows, where F is the set defined in Example 11.
Proof Although f fails (4.2), we will prove kf · gkL1 ≤ 6kM ] f · M D gkL1
for any (p, q)-blocks g arguing as in Theorem 165. Once this is done, we can
0
use the boundedness of M D on a predual space Hqp0 (Rn ), Theorem 385.
We may assume that g ∈ L∞ n
c (R ) by the truncation. We indicate the
change. We use the same idea assuming that f ∈ L∞ (Rn ) ∩ M+ (Rn ), so that
f ∈ Mpq (Rn ) by assumption. The major change from Theorem 165 is the
control of Ik . Note that

|Gk (x)| ≤ M g(x) . 2kθ M g(x)1−θ (x ∈ Rn ).

0
Observe that (M g)1−θ ∈ Hqp0 (Rn ) as long as θ is sufficiently small and note
that f ∈ Mpq (Rn ); see Example 143.
Thus, Ik = O(2kθ ) as k → −∞. If we go through the same argument as
before, we obtain the desired result.

10.2.2 Sharp maximal inequalities for local Morrey spaces

Similar to Theorem 388, we can prove the following theorem:
Theorem 390 (Sharp-maximal inequality for local Morrey spaces). Let 0 <
r < q ≤ p < ∞. Then kM2],D
−n−2 f kLMq + kf kLMr ∼ kf kLMq for all f ∈
p p p

0 n
L (R ).
Proof The proof is omitted; see Exercise 124.

10.2.3 Exercises
Exercise 124. Reexamine the proof of Theorem 388, to prove Theorem 390.
Exercise 125. Let 1 < q ≤ p < ∞.
(1) Show that kf kLMpq .p kM ] f kLMpq for f ∈ L0 (Rn ) with min(M f, 1) ∈
LMpq (Rn ) by reexamining the proof of Theorem 389.

(2) Using (1) and the translation, show that kf kMpq .p kM ] f kMpq for f ∈
L0 (Rn ) with min(M f, 1) ∈ Mpq (Rn ). This will reprove Theorem 389.

10.3 Fractional integral operators in Morrey spaces

One of the important properties in Morrey spaces is that we can describe
the boundedness property of fractional integral operators more precisely than
 Linear and sublinear operators in Morrey spaces 417

Lebesgue spaces because Morrey spaces are equipped with two parameters. We
will consider the boundedness properties of Morrey spaces and local Morrey
spaces in Sections 10.3.1 and 10.3.2. One of the important differences between
the boundedness properties of Morrey spaces and local Morrey spaces is the
difference of the range of the parameter t for which the fractional integral
operator Iα is bounded from Mpq (Rn ) to Mst (Rn ) and Iα is bounded from
LMpq (Rn ) to LMst (Rn ).

10.3.1 Fractional integral operators in Morrey spaces

Recall that the Riesz potential Iα of order α and the fractional maximal
operator Mα of order α are defined by
Z Z
f (y) χQ
Iα f ≡ n−α
dy, Mα f ≡ sup 1−α
|f (y)|dy,
Rn | · −y| Q∈Q |Q| Q

respectively. Define the modified fractional integral operator by

Z  
1 χQ0 c (y)
I˜α f ≡ − f (y)dy,
Rn | · −y|n−α |x0 − y|n−α
where Q0 is a fixed cube centered at x0 .
Morrey spaces, the BMO space and Hölder–Zygmund spaces stand in a
line. This fact can be described by using the (modified) fractional integral
operator I˜α . More precisely, we formulate this as follows:
Theorem 391. Suppose that 1 ≤ q ≤ p < ∞ and 0 < α < n.
n
(1) (Subcritical case) The Adams theorem [2]: Let 1 < q ≤ p < . Assume
α
that the parameters s and t satisfy
1 1 α t q
1 < t ≤ s < ∞, = − , = . (10.1)
s p n s p
Then
kIα f kMst . kf kMpq (10.2)
for every f ∈ Mpq (Rn ) +
∩ M (R ). n

n
(2) (Critical case) Assume that 1 ≤ q ≤ p = . Then
α
kI˜α f k∗ . kf kMpq (10.3)
for every f ∈ Mpq (Rn ), where k · k∗ denotes the BMO(R )-norm. n

n
(3) (Supercritical case) Assume that 0 < α− < 1 and that 1 ≤ q ≤ p < ∞.
p
Then
I˜α f α− n
. kf kMpq (10.4)
Lip p

for every f ∈ Mpq (Rn ).

418 Morrey Spaces

Here, we content ourselves with proving (10.3). We leave the interested

readers for the proof of (10.2) and (10.4) as exercises; see Exercise 126.
Proof We have to prove mQ (|I˜α f − mQ (I˜α f )|) . kf k αn for all cubes
Mq
n
Q. For this purpose, we may assume that q < because we always have
n
α
n
kf k αn ≤ kf k αn = kf kL αn for any f ∈ M αn (Rn ) = L α (Rn ). We decompose
Mq Mn α
α
f according to 2Q as usual. That is, we split f = f1 + f2 with f1 ≡ χ2Q · f and
f2 ≡ f − f1 . By virtue of the triangle inequality our present task is partitioned
into proving:
mQ (|I˜α f1 − mQ (I˜α f1 )|) . kf k αn (10.5)
Mq

and
mQ (|I˜α f2 − mQ (I˜α f2 )|) . kf k n . (10.6)
Mqα

Then the estimate (10.5) for f1 is simple. Indeed, to estimate f1 , we define

1 1 α
an auxiliary index w ∈ (q, ∞) by = − . We may replace I˜α with Iα ,
w q n
since f1 ∈ L1 (Rn ) and we are considering Iα f1 − mQ (Iα f1 ). By the triangle
inequality and Hölder’s inequality, we have
(w)
mQ (|Iα f1 − mQ (Iα f1 )|) ≤ 2mQ (|Iα f1 |) ≤ 2mQ (|Iα f1 |).

By using the Lq (Rn )-Lw (Rn ) boundedness of the fractional integral operator
of Iα , which follows from (10.2), we obtain
1 n 1
mQ (|Iα f1 − mQ (Iα f1 )|) . |Q|− w kIα f1 kLw . |Q| α m2Q (|f |q ) q . kf k n .
Mqα

For the proof of the second inequality, we write the left-hand side out in full:

mQ (|I˜α f2 − mQ (I˜α f2 )|)

Z ZZ  
1 f (z) f (z)
= − dydz dx.
|Q|2 Q Q×(Rn \2Q) |x − z|n−α |y − z|n−α

First, we bound the right-hand side with the triangle inequality. Then the
right-hand side is majorized by
ZZZ
1 1 1
2
|f (z)| · n−α
− dxdydz. (10.7)
|Q| Q×Q×(Rn \2Q) |x − z| |y − z|n−α

By virtue of the mean-value theorem, we have

1 1 |x − y| `(Q)
n−α
− n−α
. n−α+1
. (10.8)
|x − z| |y − z| |z − c(Q)| |z − c(Q)|n−α+1
 Linear and sublinear operators in Morrey spaces 419

for all x, y ∈ Q and z ∈ Rn \ 2Q. Thus inserting (10.8) into (10.7) gives us

|f (z)|
Z
˜ ˜
mQ (|Iα f2 − mQ (Iα f2 )|) . `(Q) dz
n |z − c(Q)| n−α+1
R \2Q
|f (z)|
Z
≤ `(Q) n−α+1
dz.
Rn \B(c(Q),`(Q)) |z − c(Q)|

Lemma 160 yields

∞
χRn \B(c(Q),`) (z)d`
Z
1
= (n − α + 1) .
|z − c(Q)|n−α+1 0 `n−α+2

Hence, it follows that

|f (z)|
Z
n−α+1
dz.
Rn \B(c(Q),`(Q)) |z − c(Q)|
Z ∞ Z !
|f (z)|
= (n − α + 1) n−α+2
d` dz
0 B(c(Q),`)\B(c(Q),`(Q)) `
!
∞
|f (z)|
Z Z
≤ (n − α + 1) n−α+2
d` dz.
`(Q) B(c(Q),`) `

Thus the integral of the right-hand side is bounded by

kf k αn
Z ∞ Z ! Z ∞
d` d` Mq
|f (z)|dz . kf k n ≤ .
`(Q) B(c(Q),`) `n−α+2 M1α `(Q) `2 `(Q)

Thus, the proof of (10.6) is complete. Hence the proof of (10.3) is concluded.
We will discuss certain inequalities for the fractional maximal operator,
which is defined by
rα
Z
Mα f (x) ≡ sup |f (y)|dy.
r>0 |Q(x, r)| Q(x,r)

In particular, when α = 0, Mα = M is the (centered) Hardy–Littlewood

maximal operator. Here, we summarize the boundedness property of Mα in
Morrey spaces.
n
Corollary 392. Let 0 ≤ α < n. Let 1 < q ≤ p < . Assume that the param-
α
eters s and t satisfy (10.1). Then Mα is bounded from Mpq (Rn ) to Mst (Rn ).
Proof If α = 0, then we go back to Theorem 382. If 0 < α < n, then
we use the Adams theorem, Theorem 391, for Iα together with the pointwise
estimate Mα f . Iα [|f |] valid for all f ∈ L0 (Rn ).
We discuss the necessity of the parameter t: It must be small enough.
420 Morrey Spaces

Proposition 393. Let 0 < α < n, 1 ≤ q ≤ p < ∞ and 1 ≤ t ≤ s < ∞.

If there exists a constant C > 0 such that kIα f kWMst ≤ Ckf kMpq for all
f ∈ M+ (Rn ), it is necessary that
1 1 α t q
= − , ≤ . (10.9)
s p n s p
1 1
Proof We prove s = p −α
n . From Theorem 17

n n
kg(λ·)kWMst = λ− s kgkWMst , kf (λ·)kMpq = λ− p kf kMpq ,

for λ > 0, and arithmetic shows Iα [f (λ·)] = λ−α Iα f (λ·) for all f, g ∈ M+ (Rn ).
Consequently, we obtain 1s = p1 − α
n . We may assume that q < p for the purpose
t q 1 1 1
of establishing = . Let R > 1 be the solution of (1 + R)− p = 2 q (1 + R)− q .
s p
n
Then for the set Ej in Example 9, we have kχEj kMpq ∼ (1 + R)−j p .
Let R be any one of the connected components of Ej . Then

Iα χEj (x) ≥ Iα χR (x) ∼ `(R)α χR (x) = (1 + R)−jα χR (x) (x ∈ Rn ).

Thus, Iα χEj (x) & (1+R)−jα χEj (x). Also, from the definition of Mst , we have
jnq
(1 + R)− pt = kχEj kLt ≤ kχEj kMst from the definition of the Morrey norm.
Consequently, it follows that
jnq jn
. (1 + R)jα kχEj kMpq ∼ (1 + R)j (α− p ) = (1 + R)−
n
(1 + R)− pt s .
qs
Since j ∈ N is arbitrary, it follows that t ≤ .
p
Example 157. We work in R. Suppose that the parameters p, s, t and α
satisfy
1 < p < ∞, 0 < α < 1
and
1 1 s
= − α, t= .
s p p
Let Ek ⊂ [0, 1] be the same set as Example 155. As before, choose r > 0 so
1
that r p = 2r. Then the fractional maximal operator Mα , given by
Z
1
Mα f (x) ≡ sup 1−α |f (x − y)|dy
R>0 R [−R,R]

for f ∈ L0 (R), satisfies

k
X
Mα χEk (x) & r(k−l)α (2r)l χEk−l \Ek−l+1 .
l=0
 Linear and sublinear operators in Morrey spaces 421

Consequently,
k
X
t
(kMα χEk k ) &Lt r(k−l)tα (2r)tl (2r)k−l
l=1
k
X
≥ r(k−l)tα rtl/p r(k−l)/p
l=1
kt/p
= kr .
since
   
1 1 1 lt l 1 1 1 1
α+ = α + = , − ltα − = lt −α− = lt −α− = 0.
tp s p p p p pt p s
As before, this shows that Mα is not bounded from Mp1 (R) to Mst (R). Since
Iα |f | & Mα f , it follows that the Hardy–Littlewood-Sobolev inequality fails;
Iα is not bounded from Mp1 (R) to Mst (R).
One important observation is that we did not use the Hardy–Littlewood–
Sobolev theorem, which asserts Iα maps Lu (Rn ) to Lv (Rn ) boundedly as long
as 1 < u < v < ∞ and v1 = u1 − α p
n , to prove the boundedness from Mq (R ) to
n
s n p n
Mt (R ). Instead, we used the Mq (R )-boundedness of the Hardy–Littlewood
maximal operator.
To clarify the terminology which we use in the sequel, we recall some ter-
minologies. Estimate (10.2), the boundedness of fractional integral operators
on the (classical) Morrey spaces Mpq (Rn ) was studied by Spanne, Adams [2],
Chiarenza and Frasca [75] etc. Basically there are two results on the Morrey
boundedness of fractional integral operators. The first result is due to Spanne,
but was communicated by Peetre in [353, Theorem 5.4].
n
Theorem 394. Let 1 < q ≤ p < . Assume that the parameters s and u
α
satisfy
1 1 α 1 1 α
1 < u ≤ s < ∞, = − , = − . (10.10)
s p n u q n
Then
kIα f kMsu . kf kMpq (10.11)
for every f ∈ Mpq (Rn ) ∩ M+ (Rn ).
Before the proof, let us review Theorem 391 due to Adams. In 1975, as we
saw in Theorem 391, Adams [2] proved that
kIα f kMsu . kf kMpq (f ∈ Mpq (Rn )) (10.12)
as long as q satisfies (10.1).
Proof Arithmetic shows that u defined by (10.10) is less than q defined
by (10.1). Therefore, as was mentioned in [75, Corollary, p. 277], in view of the
embedding in Theorem 19 (together with the sharpness proven in Proposition
393) (10.12) is a better estimate than (10.11).
422 Morrey Spaces

10.3.2 Fractional integral operators in local Morrey spaces

We prove the boundedness of the fractional integral operator Iα in local
Morrey spaces. To this end, we need a pointwise estimate different from the
Hedberg inequality used for Morrey spaces.
We will prove the Hardy–Littlewood Sobolev theorem for local Morrey
spaces. We obtain a result of Spanne-type. An outstanding difference between
local Morrey spaces and Morrey spaces lies in the assumption postulated on
the parameter t; compare Theorems 391 and 395.
Theorem 395. Let 0 < α < n, 1 < q ≤ p < ∞ and 1 < t ≤ s < ∞ satisfy
1 1 α 1 1 α
= − , = − . (10.13)
s p n t q n
Then Iα is bounded from LMpq (Rn ) to LMst (Rn ).
Proof Let B = B(x, r) be a ball. Note that our assumption reads p1 + 1t =
1
s + 1q . Thus, thanks to Lemma 187 and the embedding Mpq (Rn ) ,→ Mp1 (Rn ),
Z ∞
n n n n n n n ds
r s − t kIα f kLt (B) . r p − q kf kLq (2B) + r p − q + t sα−n kf kL1 (B(x,s))
s
Zr ∞
n n n n n n ds
. r p − q kf kLq (2B) + r p − q + t s− p +α kf kLMp1
r s
. kf kLMpq .
As a corollary of the above result, we recover the boundedness of Iα of
Spanne-type.
Corollary 396. Let 0 < α < n, 1 < q ≤ p < ∞ and 1 < t ≤ s < ∞ satisfy
(10.13). Then Iα is bounded from Mpq (Rn ) to Mst (Rn ).
Proof Simply observe that Iα commutes with translation. See Exercise
129.
Recall once again that we have the Adams theorem for Morrey spaces;
see Theorem 391. An example shows that the Adams type boundedness of
fractional integral operators in local Morrey spaces fails.
Example 158. Let 1 < q ≤ p < ∞, 1 < t ≤ s < ∞ and 0 < α < n.
Assume (10.13). Let e1 ≡ (1, 0, . . . , 0) be the unit vector in the x1 direction.
Let 0 < u < ∞ and define
n
f (x) ≡ |x − 3e1 |− u χB(3e1 ,2) (x) (x ∈ Rn ).
Then f belongs to LMpq (Rn ) if and only if u < q according to Example 25.
Since Iα maps | · |−n/u to c0 | · |α−n/u , where c0 6= 0, we see that Iα maps f to
c0 | · −3e1 |α−n/u + g, where g is an L∞ (Rn )-function. We also remark that we
can use the scaling argument. Thus, Iα maps LMpq (Rn ) to LMuv (Rn ) if and
only if u = s and v ≤ t.
 Linear and sublinear operators in Morrey spaces 423

10.3.3 Exercises
n
Exercise 126. Let 0 < α < n, 1 ≤ q ≤ p < α.

(1) Define s > 1 by

1 1 α
= − .
s p n
Show that p p
|Iα f (x)| . M f (x) s (kf kMpq )1− s (10.14)
for a.e. x ∈ Rn by mimicking the proof of Lemma 181.
(2) Prove (10.2) and (10.4).
Exercise 127. Let 0 < α < n.
n/α
(1) Show that Mα maps M1 (Rn ) to L∞ (Rn ).
(2) Suppose that a ball Banach space E(Rn ) is mapped by Mα to L∞ (Rn ).
n/α
Then show that E(Rn ) is embedded into M1 (Rn ).
n/α
(3) Disprove that Iα maps M1 (Rn ) to L∞ (Rn ). Hint: Disprove that Iα
maps Ln/α (Rn ) to L∞ (Rn ).
1 1 α
Exercise 128. Let 1 < q ≤ p < ∞ and 0 < α < n. If < − and
S p n
T q
= , then disprove that Iα maps Mpq (Rn ) to LTloc (Rn ) using the proof of
S p
Proposition 393.
Exercise 129. Let 0 < α < n.
(1) Show that Iα [f (· + x0 )] = Iα f (· + x0 ) for x0 ∈ Rn and f ∈ M+ (Rn ).
(2) Prove Corollary 396.
Exercise 130. [390, Proposition 3.8] Let 0 < α < n, 1 < p ≤ p0 < ∞ and
p
1 < r ≤ r0 < ∞. Suppose that pn0 > α, r10 = p10 − α r
n and r0 = p0 .

(1) Show that kIα f k p0 . kf k r0 for all f ∈ L∞ n

c (R ).
Hp00 Hr00

r0
(2) Using the Fatou property of Hr00 (Rn ), show that kIα f k p0 . kf k r0
Hp00 Hr00
r0
for all f ∈ Hr00 (Rn ).
424 Morrey Spaces

10.4 Singular integral operators in Morrey spaces

One of the successful achievements in the theory of Morrey spaces is that
we could nicely loosen the integrability assumption of the gradient to have the
Lipschitz continuity. This fact was initially applied by Morrey. Morrey used
this observation to elliptic differential equations. Later some important formu-
las which can be expressed in terms of singular integral operators came about.
These expressions are used to consider the boundedness of the solution operator
for the elliptic differential operators. Consequently, the action of singular inte-
gral operators in Morrey spaces is interesting in application. However, there is
a problem. Due to the singularity of the integral kernel, it is difficult to define
singular integral operators in Morrey spaces. We present some methods to over-
come this problem here. In principle, Sections 10.4.1 and 10.4.2 are parallel,
where we develop the idea presented above. We will present 4 methods: the use
of the A1 -weighted Lebesgue spaces, the use of the duality, the use of the bi-
duality and the use of the estimate of the integral kernel.

10.4.1 Singular integral operators in Morrey spaces

We will now show that singular integral operators T defined and inves-
tigated in Section 4.5 are bounded in Morrey spaces. Recall that singular
integral operators are bounded on Lp (Rn ) for 1 < p < ∞ and satisfy the
weak-(1, 1) estimate. Note that the Morrey space Mpq (Rn ) contains Lp (Rn ).
Therefore, it is yet for T to be defined in Mpq (Rn ). Namely, since the Morrey
space Mpq (Rn ) is strictly larger than Lp (Rn ), we have to define T f carefully
for f ∈ Mpq (Rn ) with 1 ≤ q ≤ p < ∞ and a singular integral operator T .
Here, we consider four different methods. The first but simplest one is
to resort to Example 115. That is, since L∞ n q
c (R ) is dense in L (w) for any
q
w ∈ A1 , we can extend the domain of operators T up to L (w). Then by
restricting T to Lq ((M χB(1) )θ ) with 1 − pq < θ < 1, we can define T over
Mpq (Rn ) although L∞ n p
c (R ) is not dense in Mq (R ).
n

We use the following lemma to define the operator T on Mpq (Rn ). In many
methods we will present, the following estimate is fundamental:
Lemma 397. Let f ∈ L0 (Rn ). Then for any ball B ≡ B(a, r), we have
Z ∞
X
|K(x, y)f (y)| dy . m2k B (|f |) (10.1)
Rn \2B k=1
for all x ∈ B.
Inequality (10.1) corresponds to Lemma 130 for the maximal operator.
Proof By the size condition and the dyadic decomposition of Rn \ 2B, we
obtain
∞ Z
|f (y)| |f (y)|
Z Z X
|K(x, y)f (y)| dy . n
dy = n
dy.
Rn \2B Rn \2B |x − y| 2k+1 B\2k B |x − y|
k=1
 Linear and sublinear operators in Morrey spaces 425

Observe that for y ∈ 2k+1 B \ 2k B and x ∈ B, we have

|x − y| ≥ |y − a| − |x − a| > (2k − 1)r ≥ 2k−1 r ∼ |2k+1 B|1/n .
Hence, for all x ∈ B, we have
Z ∞
X
|K(x, y)f (y)|dy . m2k+1 B (|f |),
Rn \2B k=1

as was to be shown modulo shifting the index k.

We verify that Morrey spaces fall under the scope of Lemma 397.
Corollary 398. Let 1 ≤ q ≤ p < ∞. If f ∈ Mpq (Rn ), then
Z
1
|K(x, y)f (y)| dy . |B|− p kf kMpq
Rn \2B

for all balls B and x ∈ B.

1
Proof Simply use m2k+1 B (|f |) . |2k B|− p kf kMpq for each k ∈ N and
Lemma 397.
We start with a simple method to define singular integral operators T on
Morrey spaces. A direct consequence is that T , which is defined initially on
Lq (w) for any w ∈ A1 , can be restricted to a linear operator from Mpq (Rn )
to L0 (Rn ). If we argue as we did for the Hardy–Littlewood maximal operator
using Corollary 398, we can show that T is bounded on Mpq (Rn ).
We move on to the second definition of singular integral operators. Keeping
the above auxiliary estimate of singular integral operators above in mind, we
indicate how to define singular integral operators in Morrey spaces. This is
the second method; other techniques will be obtained later in Section 10.4.1.
Definition 95 (Singular integral operators in Morrrey spaces I). Let T be a
singular integral operator. Define T as
Z
T f (x) ≡ T [f χ2B ](x) + K(x, y)f (y)dy, (x ∈ B) (10.2)
Rn \2B

for functions f ∈ L1loc (Rn ) for which the left-hand side of (10.1) converges for
every ball B.
We need to check that (10.2) makes sense despite ambiguity of the choice
of the ball B containing x. We check this as follows:
Proposition 399. Let f ∈ L1loc (Rn ) be such that the left-hand side of (10.1)
converges for every ball B. If B1 and B2 are balls and x ∈ B1 ∩ B2 , then
Z
T [f χ2B1 ](x) + K(x, y)f (y)dy
Rn \2B1
Z
= T [f χ2B2 ](x) + K(x, y)f (y)dy.
Rn \2B2
426 Morrey Spaces

Proof Actually, letting B3 be a ball such that B1 ∪ B2 ⊂ B3 , we have, for

x ∈ B1 ∩ B2 ,
Z
T [f χ2Bj ](x) + K(x, y)f (y)dy
Rn \2Bj
Z !
− T [f χ2B3 ](x) + K(x, y)f (y)dy
Rn \2B3
Z
= T [f χ2Bj − f χ2B3 ](x) + K(x, y)f (y)dy = 0.
2B3 \2Bj

Therefore, for f ∈ L1loc (Rn ), T f (x) is well defined for all x ∈ Rn since the
right-hand side of (10.1) converges for every ball B. In this case we also write
Z
T [f χRn \2B ](x) = K(x, y)f (y)dy (x ∈ B).
Rn \2B

We now extend the domain of singular integral operators to Morrey spaces

via another method: To do this, we use duality. We propose two different meth-
ods here. Although the structure of the dual space of Mpq (Rn ) is complicated,
the predual, whose dual space is Mpq (Rn ), is well understood because it is
0
included in Lp (Rn ). Note that the operation taking dual reverses inclusion:
X ⊂ Y implies X ∗ ⊃ Y ∗ .
Keep in mind that the domain of a singular integral operator T defined in
0 0 0
Definition 57 contains Hqp0 (Rn ); Hqp0 (Rn ) is a subset of Lp (Rn ).
Theorem 400. Let 1 < p ≤ q < ∞. Let T be a singular integral operator.
0
Then T , which is defined a priori on Lp (Rn ), is bounded on Hqp0 (Rn ).
Proof We have only to prove the assertion in the block level. That is, given
1 1
a block a, which is supported on a cube Q and satisfies kakLq ≤ |Q| q − p , we
need only show kT akHp0 . 1.
q0
Note that T is a bounded operator on Lq (Rn ). Hence, χ2Q · T a is a (p, q)-
block modulo a multiplicative constant. Consequently,
kχ2Q · T akHp0 . 1. (10.3)
q0

Furthermore, by forming a dyadic decomposition away from the singularity,

for all x ∈ Rn , we have
1 ∞
|Q|1− p χRn \2Q (x) X −j (n− n ) j 1
−p
|χRn \2Q (x) · T a(x)| . . 2 2 Q (x) · |2 Q|
p χ j ,
|x − c(Q)|n j=1

which together with Lemma 342 and (10.3), yields kT akHp0 . 1.

q0

In (4.7), we defined the adjoint T ∗ . Here, we consider the “formal” adjoint

T ∗ , since Mpq (Rn ) does not carry the structure of a Hilbert space. Here is a
precise definition.
 Linear and sublinear operators in Morrey spaces 427

Definition 96 (Singular integral operators in Morrrey spaces II). Let 1 <

q ≤ p < ∞. Given a singular integral operator T , extend it to Mpq (Rn ) so
that Z Z
T f (x) · g(x)dx = f (x) · T ∗ g(x)dx (10.4)
Rn Rn
0
holds for all g ∈ Hqp0 (Rn ), where T ∗ is a formal adjoint whose kernel is given
by (x, y) 7→ K(y, x).
Example 159. Although Mpq (Rn ) is larger than the Lebesgue space Lp (Rn ),
by using the Lebesgue differentiation theorem, we can verify that the formal
0 0
adjoint Rj∗ : Mpq (Rn ) → Mpq (Rn ) of Rj : Hqp0 (Rn ) → Hqp0 (Rn ) in the above
sense is given by Rj∗ = −Rj : Mpq (Rn ) → Mpq (Rn ).
Consequently, as we will see, T f ∈ Mpq (Rn ) is determined indirectly from
Theorem 347 and (10.4).
Theorem 401. Suppose that 1 < q ≤ p < ∞. Let T be a singular integral
operator. Then T , which is initially defined in Lp (Rn ), can be extended to a
bounded linear operator on Mpq (Rn ).
Proof Let f ∈ Lp (Rn ). Then consider the functional
Z
0
g ∈ Hqp0 (Rn ) 7→ f (x) · T ∗ g(x)dx ∈ C.
Rn
0
Note that this functional is bounded on Hqp0 (Rn ) by Theorem 400. Therefore,
thanks to Theorem 347, there exists a unique element T f ∈ Mpq (Rn ) such
that Z Z
T f (x) · g(x)dx = f (x) · T ∗ g(x)dx (10.5)
Rn Rn
0
for all g ∈ Hqp0 (Rn ).
We claim that the mapping f ∈ Mpq (Rn ) 7→ T f ∈
p n
Mq (R ) is the desired extension.
By the Hahn–Banach theorem, Theorem 87, for all f ∈ Mpq (Rn ), we can
0
find g ∈ Hqp0 (Rn ) with kgkHp0 ≤ 2 such that
q0

Z
kT f kMpq = T f (x) · g(x)dx.
Rn

By (10.5), we obtain kT f kMpq ≤ kf kMpq kT ∗ gkHp0 . Now that T ∗ is bounded on

q0
0
Hqp0 (Rn ),we see, taking into account that g has unit norm, that kT f kMpq .
kf kMpq , which is the desired result.
There is another way to define singular integral operators via duality. Let
1 < q ≤ p < ∞ as before. Suppose that we have a linear operator S :
Lp (Rn ) → Lp (Rn ) such that kSf kMpq ≤ Ckf kMpq for all f ∈ Lp (Rn ). Then
thanks to Theorem 356 by density we can extend S to a bounded linear
428 Morrey Spaces
fpq (Rn ) → M
operator T : M fpq (Rn ) thanks to Proposition 321. By duality we
0 0
have T ∗ : Hqp0 (Rn ) → Hqp0 (Rn ). Again by duality we have T ∗∗ : Mpq (Rn ) →
Mpq (Rn ). We claim that T ∗∗ is the extension of T .

Lemma 402. We have T ∗∗ |M̃pq (Rn ) = T |M̃pq (Rn ).

Proof Let Q be a cube. Then for all f ∈ M fpq (Rn ),

Z Z
∗∗
T f (x)dx = T ∗∗ f (x)χQ (x)dx
Q Rn
Z
= f (x)T ∗ χQ (x)dx
Rn
Z
= T f (x)χQ (x)dx.
Rn

Consequently, for S = T ∗∗ , we have

Z Z
Sf (x)dx = T f (x)dx. (10.6)
Q Q

Thus, by the Lebesgue differentiation theorem, we obtain T f = T ∗∗ f .

Note that Definitions 95 and 96 coincide since (10.6) holds with S replaced
by the operator in Definition 96.
We give another definition of singular integral operators in Morrey spaces.
Definition 97 (Singular integral operators in Morrrey spaces III). Let 1 <
q ≤ p < ∞. Suppose that we have a linear operator S : Lp (Rn ) → Lp (Rn )
such that kSf kMpq . kf kMpq for all f ∈ Lp (Rn ). Then the natural extension
T : Mpq (Rn ) → Mpq (Rn ) of S is defined as S ∗∗ given as above.
Example 160. Let 1 < q ≤ p < ∞. In view of the Lebesgue differentiation
theorem, we see that the formal adjoint of Rj : M fpq (Rn ) is given
fpq (Rn ) → M
by the same formula as the one for Lebesgue spaces:
(xj − yj )f (y)
Z
Rj∗ f (x) = − n+1
dy (x ∈ Rn \ supp(f ))
R n |x − y|
for all f ∈ Mpq (Rn ).
We define T to be T ∗∗ on Mpq (Rn ). Let us see that the kernel expression
is still valid.
Proposition 403. For all f ∈ Mpq (Rn ) ∩ L0c (Rn ), we have (4.1).
Proof Let x ∈ Rn \ supp(f ). Let B(y, r) be a ball that never intersects
0
supp(f ). Then for all g ∈ Hqp0 (Rn ) supported on B(y, r),
Z Z Z Z 
T f (x)g(x)dx = f (x)T ∗ g(x)dx = f (x) K(y, x)g(y)dy dx.
Rn Rn Rn Rn
 Linear and sublinear operators in Morrey spaces 429

Since the kernel of T ∗ is (x, y) 7→ K(y, x), we obtain

Z
T ∗ g(x) = K(y, x)g(y)dy
Rn

for almost all x ∈ Rn \ supp(g). Thus,

Z Z Z 
T f (x)g(x)dx = g(y) K(x, y)f (x)dx dy.
Rn Rn Rn

Consequently, since g is arbitrary, (4.1) follows.

Remark that Definitions 96 and 97 coincide because the adjoint operator
fpq (Rn ) coincides with the restriction of T ∗ from Lp0 (Rn ) to
fpq (Rn ) → M
T :M
0
Hqp0 (Rn ).
Here, we list some examples and a counterexample for which we can or
cannot extend the domain of operators.
Example 161. Let 1 < q ≤ p < ∞. Let T be a linear operator defined
in Lp (Rn ). Assume that there exists a constant c0 > 0 such that, for all
f ∈ L1c (Rn ),
|Ω(x − y)|
Z
|T f (x)| ≤ c0 n
|f (y)|dy, x ∈
/ supp(f ), (10.7)
Rn |x − y|

where Ω ∈ L0 (Rn ) is homogeneous of degree zero, in the sense that Ω(t·) = Ω

for all t > 0 and Ω ∈ Lq̃ (S n−1 ) for some q̃ ∈ [q 0 , ∞]. Assume also that T is
Lq (Rn )-bounded, in the sense that

kT f kLq ≤ Cq kf kLq (10.8)

for all f ∈ Lp (Rn ) ∩ Lq (Rn ). Then kT f kMpq . kf kMpq for all f ∈ Lp (Rn ). To
prove this, we have only to show that
1 1 1 1
|Q| p − q kT [χ3Q f ]χQ kLq + |Q| p − q kT [f − χ3Q f ]χQ kLq . kf kMpq .

The first estimate follows readily from (10.8). Meanwhile, for almost all x ∈ Q,
we deduce from (10.7)

|T f (x)|
∞ Z
X 1
. |Ω(x − y)| |f (y)|dy
|B(c(Q), 2l `(Q))| B(c(Q),2l `(Q))
l=1
∞
! q̃1 Z ! 10
Z q̃
X 1 q̃ q̃ 0
. |Ω(y)| dy |f (y)| dy
|B(2l+n `(Q))| B(2l `(Q)) B(c(Q),2l `(Q))
l=1
∞
! 10
Z q̃
X 1 q̃ 0
. |f (y)| dy
|B(c(Q), 2l+n `(Q))| B(c(Q),2l `(Q))
l=1
430 Morrey Spaces

by Hölder’s inequality. Since q̃ 0 ≥ q, if we integrate this inequality we obtain

1 1
|Q| p − q kT [f − χ3Q f ]χQ kLq . kf kMp0 ≤ kf kMpq .
q̃

Thus, as a result we obtain the desired result.

Example 162. Let T be a sublinear operator defined in L∞ n
c (R ). That is,
the operator T satisfies that, for all f, g ∈ Lc (R ), k ∈ C and for a.e. x ∈ Rn ,
1 n

|T [f + g](x)| ≤ |T f (x)| + |T g(x)| and |T [kf ](x)| = |k||T f (x)|. We also assume
that
|T f (x) − T g(x)| . |T (f − g)(x)| (10.9)
Assume that there exists a constant c0 > 0 such that, for all f ∈ L∞ n
c (R ),
(10.7) holds, where Ω ∈ L0 (Rn ) is homogeneous of degree zero and Ω ∈
Lp̃ (S n−1 ) for some p̃ ∈ [1, ∞]. Assume also that T is Lq (Rn )-bounded, in the
sense that kT f kLp ≤ Cp kf kLp for all f ∈ Lp (Rn ) ∩ Lq (Rn ). Then kT f kMpq .
kf kMpq for all f ∈ L1c (Rn ) = L1 (Rn ) ∩ L0c (Rn ).

Example 163. Let 1 ≤ q < p < ∞. Then the operator T f ≡ eikf kLq M f
does not satisfy (10.9) and cannot be extended to Morrey space Mpq (Rn ).
One important observation is that we used the Lq (Rn )-boundedness of
singular integral operators to prove the Mpq (Rn )-boundedness. So one of the
advantages of using Morrey spaces is that Morrey spaces can describe the
boundedness property more precisely than Lebesgue spaces.

10.4.2 Singular integral operators in local Morrey spaces

We now define singular integral operators in local Morrey spaces. To this
end, it is helpful to note the following fact:
Proposition 404. Let 1 ≤ q ≤ p < ∞. Then Lp (Rn ) is dense in LM
fp (Rn ).
q

Proof Simply mimic the proof of Proposition 327; see Exercise 131.
Lemma 405. Let 1 < q ≤ p < ∞, and let T be a singular integral operator.
Then kT f kLMpq . kf kLMpq for all f ∈ Lp (Rn ).
Proof Reexamine the proof of Theorem 401, where Lemma 397 is used;
see Exercise 132.
We can take the same strategy as the definition of singular integral oper-
ators acting on Morrey spaces. What is useful in Theorem 406 is that we can
define singular integral operators on the function space LMpq (Rn ) much larger
than Mpq (Rn ).
Theorem 406. Let 1 < q ≤ p < ∞. Then any singular integral operator T ,
defined initially on Lp (Rn ), extends to a bounded linear operator on LMpq (Rn ).
Proof Mimic Definition 97 as before. See Exercise 133.
 Linear and sublinear operators in Morrey spaces 431

An important conclusion of Theorem 406 is that the boundedness of sin-

gular integral operators acting on Morrey spaces can be “decomposable” in
the sense that a counterpart to local Morrey spaces together with the transla-
tion readily gives the boundedness on Morrey spaces. This is a good contrast
to the fractional integral operators. Actually, the Adams-type boundedness
is no longer available from the boundedness of Iα on local Morrey spaces as
Example 158 shows.

10.4.3 Exercises
Exercise 131. Mimic the proof of Proposition 327 to prove Proposition 404.
Exercise 132. Use Lemma 397 to prove Lemma 405.
Exercise 133. Prove Theorem 406 based on Definition 97.
Exercise 134. Let 1 < q ≤ p < ∞. Suppose that we have a linear operator
T : Lq (Rn ) → Lq (Rn ) satisfying kT kLq →Lq = sup{kT f kLq : kf kLq = 1} <
∞. Assume in addition that there exists K ∈ L0 (Rn × Rn ) such that
|K(x, y)| . |x − y|−n
and that Z
T f (x) = K(x, y)f (y)dy (a.e. x ∈
/ supp(f ))
Rn
for all f ∈ Lqc (Rn ) ≡ Lq (R ) ∩ L0c (Rn ). Then show that
n

kT f kMpq .p,q (1 + kT kLq →Lq )kf kMpq (f ∈ Mpq (Rn ) ∩ Lq (Rn )).

10.5 Commutators in Morrey spaces

Section 10.5 investigates the boundedness of commutators generated by
singular integral operators and BMO(Rn )-functions. Recall first that b ∈
L1loc (Rn ) belongs to BMO(Rn ), if b satisfies;
Z
1
kbk∗ ≡ sup |b(x) − mQ (b)| dx < ∞.
Q∈Q |Q| Q

Let T be a singular integral operator in Definition 57. In view of the John–

Nirenberg inequality, for any f ∈ L∞ n 1 n
c (R ), we can define [T, b]f ∈ Lloc (R )
by
[T, b]f (x) ≡ T [b · f ](x) − b(x)T f (x). (10.1)
Note that b · f ∈ Lqc (Rn ) for all q ∈ (1, ∞). Hence, the definition (10.1) above
makes sense. We define and investigate commutators acting on Morrey spaces
and local Morrey spaces in Sections 10.5.1 and 10.5.2, respecitvely.
432 Morrey Spaces

10.5.1 Commutators in Morrey spaces

As a continuation of Section 5.2.1, we consider the commutator [a, T ] gen-
erated by a ∈ BMO(Rn ) and a singular integral operator T :
Z
[a, T ]f (x) ≡ lim (a(x) − a(y))K(x, y)f (y)dy (x ∈ Rn ). (10.2)
ε↓0 Rn \B(x,ε)

Keeping in mind the proof of the boundedness of singular integral opera-

tors, we start with the boundedness of commutators on predual spaces.

Proposition 407. Let 1 < q ≤ p < ∞, a ∈ BMO(Rn ) and T be a singular

integral operator. Then, the commutator [a, T ], which is defined initially on
0 0
Lp (Rn ), is Hqp0 (Rn )-bounded if it is restricted to Hqp0 (Rn ).
Proof Mimic the proof of Theorem 400.
As we did in the case of singular integral operators, we define the commu-
tator [a, T ] on Morrey spaces via duality.
Definition 98. Let 1 < q ≤ p < ∞, a ∈ BMO(Rn ) and T be a singular inte-
gral operator. Then, the commutator [a, T ] : Mpq (Rn ) → Mpq (Rn ) is defined
0
by way of the duality Hqp0 (Rn )-Mpq (Rn ); for f ∈ Mpq (Rn ), [a, T ]f ∈ Mpq (Rn )
0
is a unique element h ∈ Mpq (Rn ) such that for all g ∈ Hqp0 (Rn )
Z Z
h(x)g(x)dx = − f (x)[a, T ]g(x)dx.
Rn Rn

Arguing as in Theorem 401, we automatically obtain the boundedness of

commutators.
Theorem 408. Let 1 < q ≤ p < ∞, a ∈ BMO(Rn ) and T be a singular
integral operator. Then, the commutator [a, T ] is bounded on Mpq (Rn ).
Theorem 236 also carries over to Morrey spaces.
Theorem 409. Let f ∈ Mpq (Rn ) with 1 < p < ∞ and a ∈ BMO(Rn ).
Suppose that K ∈ L0 (Rn × Rn ) satisfies the following conditions:
(1) K(x, ·) ∈ C ∞ (Rn \ {0}) for all x ∈ Rn ;

(2) K(x, tz) = t−n K(x, z) for every t > 0 and (x, z) ∈ Rn × S n−1 ;
Z
(3) K(x, z)dσ(z) = 0 for all x ∈ Rn ;
S n−1

∂αK
(4) max < ∞.
|α|≤2n ∂z α L∞ (Rn ×S n−1 )
 Linear and sublinear operators in Morrey spaces 433

For any ε > 0, we set

Z
[a, T ]ε f (x) ≡ K(x, x − y)(a(x) − a(y))f (y)dy (f ∈ Mpq (Rn )).
Rn \B(x,ε)

Then for any f ∈ Mpq (Rn ), there exist [a, T ]f ∈ Mpq (Rn ) such that
k[a, T ]f kMpq . kf kMpq .
Proof This is a general fact on singular integral operators together with
the expansion formula used in the proof of Theorem 236. In fact, since q > 1,
we see that the limit

[a, T ]ε f (x) = [a, T ]ε [χB(2R) f ](x) + [a, T ]ε [χRn \B(2R) f ](x)

= [a, T ]ε [χB(2R) f ](x) + [a, T ][χRn \B(2R) f ](x)

exists as ε ↓ 0 for almost all x ∈ B(R) if R > 0. Since R > 0 is arbitrary, we

obtain the desired result.

10.5.2 Commutators in local Morrey spaces

As we did in the case of commutator operators acting on Morrey spaces,
we define the commutator [a, T ] on local Morrey spaces via duality.
Definition 99. Let 1 < q ≤ p < ∞, a ∈ BMO(Rn ) and T be a singu-
lar integral operator. Then, the commutator [a, T ] : LMpq (Rn ) → LMpq (Rn )
0
is defined by way of the duality LHqp0 (Rn )-LMpq (Rn ); for f ∈ LMpq (Rn ),
[a, T ]f ∈ LMpq (Rn ) is a unique element h ∈ LMpq (Rn ) such that
Z Z
h(x)g(x)dx = − f (x)[a, T ]g(x)dx
Rn Rn
0
for all g ∈ LHqp0 (Rn ).
We can also take the same strategy of Morrey spaces and the related spaces
when we consider the boundedness property of commutators in local Morrey
spaces.
Proposition 410. Let 1 < q ≤ p < ∞, a ∈ BMO(Rn ) and T be a singular
integral operator. Then, the commutator [a, T ], which is defined initially on
0 0
Lp (Rn ), is LHqp0 (Rn )-bounded if it is restricted to LHqp0 (Rn ).
Proof The proof is left as an exercise; see Exercise 135.
Arguing as in Theorem 401, we automatically obtain the following conclu-
sion:
Theorem 411. Let 1 < q ≤ p < ∞, a ∈ BMO(Rn ) and T be a singular
integral operator. Then, the commutator [a, T ] is bounded on LMpq (Rn ).
434 Morrey Spaces

10.5.3 Exercises
Exercise 135. Prove Proposition 410 by mimicking the proof of Theorem
400.
Exercise 136. Prove Theorem 401 by mimicking the proof of Theorem 408.

10.6 Notes
Section 10.1
See the survey [362].

General remarks and textbooks in Section 10.1

See the textbook [6, Chapter 6] for the boundedness property of the Hardy–
Littlewood maximal operator, where we can find two different proofs; one is
the same as our proof of Theorem 382 and the other hinges on Proposition
285 and Theorem 290.

Section 10.1.1
Chiarenza and Frasca showed that the Hardy–Littlewood maximal opera-
tor is weakly and strongly bounded on the Morrey space Mpq (Rn ) [75, Theorem
1(1)] and [75, Theorem 1(2)] when 1 ≤ q ≤ p and when 1 < q ≤ p < ∞, respec-
tively. See [75, Theorem 1] for Theorem 382 as well as the survey paper [391,
Theorem 3.1]. The proof hinges upon the Stein inequality for the Lebesgue
measure Corollary 272 and Proposition 271. Remark 15, concerning the weak-
boundedness of M acting on Morrey spaces, comes from [381].
Theorem 145 is from [394, Theorem 2.2] and [428, Lemma 2.5]. Example
155, asserting that M fails to be Mp1 (Rn )-bounded, comes from [328, Corollary
2.5]. The proof used here comes from [304, Lemma 4.6(i) and Lemma 4.7]. See
[194] for the action of Hardy operators on Morrey spaces.
See [202, Lemma 3.5] for the sharp maximal inequality as well as the
application to singular integral operators.
Di Fazio and Ragusa considered the boundedness of the sharp maximal
operators in [123]. See [247, Theorem 12], [283, Lemma 2.5] and [395, Theo-
rems 1.3 and 1.4] as well. See [383] for weak Morrey spaces.

Section 10.1.2
See [22] for the case of Carleson curves.
 Linear and sublinear operators in Morrey spaces 435

Section 10.2
General remarks and textbooks in Section 10.2
It seems that there is no textbook which specilizes in sharp maximal
inequality on Morrey spaces.

Section 10.2.1
Di Fazio and Ragusa initially considered the sharp maximal inequality
for Morrey spaces in connection with elliptic differential equations [123]. See
[247, 333, 395, 440, 441] for more.

Section 10.2.2
See [169, Theorem 4.2] for the sharp maximal inequality for local Morrey
spaces.

Section 10.3
See the textbook [6, Chapter 7], where we can also find an application
of closed subspaces and preduals to the Morrey-boundedness of fractional
integral operators. See also the survey [362].
We refer to [6, Lemma 9.1] for the critical case of Theorem 391. In the
supercritical case, we considered the difference of the kernel and the constant
(average). We can consider the difference of the kernel and the Taylor polyno-
mial; see [282] together with its application to fractional Laplace equations.

General remarks and textbooks in Section 10.3

See standard textbooks [112], [147], [231], [293] and [319, Chapter 2].

Section 10.3.1
The boundedness of fractional integral operators on the (classical) Morrey
spaces Mpq (Rn ) was studied by Spanne, Adams [2], Chiarenza and Frasca [75]
etc. See [2] for the subcritical case of Theorem 391. See [3, Theorem 4.1 and
Remark 4.1] for the critical case of Theorem 391. See [10] for the supercritical
case of Theorem 391. We refer to [402, Theorem 1.1] for the case of bounded
domains, where Serrin gave an explicit bound for the operator norms. We
followed the survey paper [391, Theorem 3.3] for the proof of Theorem 391.
Olsen showed that we can no longer improve the Adams theorem for the
Morrey boundedness of fractional integral operators. See [342, Theorem 10]
for Proposition 393 with n = 3. His proof works for general dimensions. See
also the survey paper [391, Proposition 2.2].
436 Morrey Spaces

Adams used the fractional maximal operator Mα to investigate the bound-

edness of the fractional integral operator Iα [2, Proposition 3.1 and Theorem
3.1]. Adams employed a technique obtained by Hedberg in 1972; see [2, p.
768]. Hedberg pointed out that
p 1− p
Iα f . M f s kf kLp s (f ∈ M+ (Rn )) (10.1)

in [190, (3)]. Chiarenza and Frasca improved the method; in fact Chiarenza
and Frasca refined (10.1) to:
p p
Iα f . M f s (kf kMpq )1− s (f ∈ M+ (Rn ))

in [75, p. 277]. Chiarenza and Frasca reproved the weak/strong boundedness

of Iα , which was initially obtained in [2, Theorems 5.1 and 3.1], respectively;
see [75, Theorem 2].
Let 1 < q ≤ p < ∞, 1 < t ≤ s < ∞ satisfy st = pq and 1s = p1 − α n . We can
extend result (10.12) to the vector-valued setting:
  r1   r1
∞
X X∞
 |Iα fj |r  .  |fj |r  ({fj }∞ p r
j=1 ∈ Mq (` )).
j=1 j=1
Msu Mp
q

We can assume that 1 ≤ r ≤ ∞, due to the positivity of the integral kernel;

see [379, Theorem 2] for this technique. Note that the vector-valued inequality
of the Hardy–Littlewood maximal operator fails for r = 1 [416].
Adams used the fractional maximal operator Mα to investigate the bound-
edness of the fractional integral operator Iα [2, Proposition 3.1 and Theorem
3.1]. Adams employed a technique obtained by Hedberg in 1972. More pre-
cisely, he obtained an estimate in Lemma 183 and its generalization with
α1 = 0; see [2, p. 768]. Hedberg pointed out Lemma 181 in [190, (3)]. It seems
that the method was improved by Chiarenza and Frasca; in fact Chiarenza and
Frasca pointed out (10.14) in [75, p. 277]. Chiarenza and Frasca reproved the
weak/strong boundedness of Iα , which was initially obtained in [2, Theorems
5.1 and 3.1], respectively; see [75, Theorem 2].
We can pass to this result (10.12) to the vector valued setting;
  r1   r1
X∞ X∞
 I α fj r  .  |fj |r  .
j=1 j=1
Mst Mp
q

It counts that we can assume that 1 ≤ r ≤ ∞ due to the positivity of the

integral kernel; see [379, Theorem 2] for this technique.
Komori-Furuya and Sato gave an example showing that the Adams type
boundedness of fractional integral operators in local Morrey spaces fails. See
[253] for Example 158.
 Linear and sublinear operators in Morrey spaces 437

Section 10.3.2
Fu, Li and Lu [133] and Li and Yang [278] proved the boundedness of
fractional integral operators in Herz spaces. See also [306].

Section 10.4
General remarks and textbooks in Section 10.4
We refer to the textbook [6, Chapter 8], where Adams used Proposition 285
as well as Theorem 209. See the survey [264] for applications of Morrey spaces
to Navier–Stokes equations, where singular integral operators come into play.
The strategy taken to define singular integrals via duality is due to Rosenthal
and Schmeisser [370].

Section 10.4.1
Let us start to recall the results on singular integral operators acting on Mor-
rey spaces. Peetre considered the boundedness of singular integral operators in
[353]; see Theorem 401. There are several methods of the definition of singular
integral operators but the definition given here is close to that in [327, (3.1)] and
[198, §3]. There is another definition by means of the predual [396, p. 483]. We
refer to [438, Proposition 2.25] for the boundedness of singular integral opera-
tors. In particular, the boundedness of the Riesz transform is discussed in [438,
Corollary 2.27]; see [371, 438] for more recent information. The boundedness
of singular integrals is also proven there [75, Theorem 3], where Chiarenza and
Frasca employed the A1 -weight technique for the boundedness.
We move on to recall the results of singular integral operators acting on
related function spaces. Komori-Furuya investigated the boundedness of sin-
gular integral operators acting on block spaces in [245, Theorem]; see Theorem
400. See [372] for the concrete case of singular integral operators; they inves-
tigated the Cauchy integral in the predual of Morrey spaces.
Finally, we consider operators related to singular integral operators. Fu
and Lu calculated the operator norm in Morrey spaces of the operators given
by (7.1) and (7.2) [134, Theorem 1.3]. Fu and Lu also characterized a condition
under which Uψb is bounded in Morrey spaces for all b ∈ BMO in terms of ψ
[134, Theorem 1.6]. A similar result for Vψb is obtained in [134, Theorem 1.6].
A passage to the higher order commutator is done in [134, Theorem 3.1].
See [286, Theorem 1.3] for the boundendness of the Bochner–Riesz opera-
tor.
The Schrödinger operator generates singular integral operators; see [75].
Kurata and Sugano investigated such operators acting on Morrey spaces in
[259, 260]; see [285, Theorem 1] for its subsequent result and [285, Theorem
2] for a passage to commutators.
Chen and Jin obtained the boundedness of Marcinkiewicz integrals asso-
ciated with the Schrödinger operator on Morrey spaces [67].
438 Morrey Spaces

Ho dealt with singular integral operators with rough kernel [197].

We can consider “more singular” integral operators in Morrey spaces as
well. The boundedness of the Riesz transform with respect to elliptic differen-
tial operators with non-smooth coefficients can be found in [153, 388]. Nakai
and Sobukawa constructed an example showing that the Lp (Rn )-boundedness
is not enough to extend the operators to Morrey spaces [330, Remark 5]; see
Example 163.
There are not so many works dealing with the convolution as an extension
of Young’s inequality. See the paper by Peetre [352].

Section 10.4.2
Hu, Lu and Yang obtained the boundedness of singular integral operators
on Herz spaces [200]. Later on, Balakishiyev, Burenkov, Guliyev, Gurbuz,
Tararykova and Serbetchi considered singular integral operators in local Mor-
rey spaces [23, 48, 47]. Deringoz, Guliyev and Samko essentially considered the
boundedness of singular integral operators in generalized local Morrey spaces
[99, Theorem 5.5]. See also [132, 301] as well.

Section 10.5
See the survey [362]. Many authors investigated the boundedness property
of commutators on Morrey spaces; see [11, 97, 123, 395].

General remarks and textbooks in Section 10.5

See the textbook [6, Chapter 12] for the boundedness property of commu-
tators handled in this book.

Section 10.5.1
Many authors investigated the boundedness property of commutators
given by (10.2); see [102, 252, 317, 318, 389, 395] for more details.
Di Fazio and Ragusa considered commutators generated by BMO functions
and singular integral operators. See [123, Theorem 2.1] for Theorems 408 and
409. Shirai considered commutators generated by BMO functions and the
Riesz potential [408]. See [469] for the application of Theorem 409 to parabolic
differential equations. Zhang, Jiang, Sheng and Zhou discussed the uniqueness
and existence of solution.
There are many generalizations. A passage to higher orders is done by Pen
and Liu in [351]. See [201, 286] for commutators generated by the Bochner–
Riesz operator and Lipschitz functions.
The compactness of commutators is also investigated. See [389] for the
sufficient conditions and [71, 72, 73, 74] for the necessary and sufficient con-
ditions. See [20, 341, 343, 430] for more generalization of [71, 72, 73, 74].
Concerning the commutator of the form [b, T ], based on the result by Janso
 Linear and sublinear operators in Morrey spaces 439

and Paluszynski [219, 350], Shi and Lu considered the case when b is a Lips-
chitz function or a function in Morrey–Campanato spaces [406, 407]. We can
consider commutators generated by BMO and the maximal operators; see
[97, 456].

Section 10.5.2
Yu and Tao obtained the boundedness of commutators in local Morrey
spaces [468]. See [98, Theorem 3.1] for the boundedness of commutators in
Morrey-type spaces. Guliyev, Rahimova and Omarova considered the bound-
edness of commutators generated by the Littlewood–Paley functions and
BMO [171, Theorem 1–3]. Guliyev, Omarova, Ragusa and Scapellato worked
in the setting of generalized Morrey spaces. [169, Theorem 4.7]. The work
by Hasanov, Muradova and Guliyev [188] and the one by Eroglu, Hasanov
Omarova [113] considered the commutators in the multilinear and generalized
setting. See also [25, 96, 165, 166, 170, 403].
Bibliography

[1] D.R. Adams, Traces of potential arising from translation invariant oper-
ators, Ann. Scuola Norm. Sup. Pisa 25 (1971), 203–217. <80>
[2] D.R. Adams, A note on Riesz potentials, Duke Math. J., 42 (1975),
765–778. <205, xi, 81, 417, 421, 435, 436>
[3] D.R. Adams, Lectures on Lp -potential theory, Volume 2, Departement
of Mathematics, University of Umea, 1981. <240, 435>
[4] D.R. Adams, A note on Choquet integrals with respect to Hausdorff
capacity, in: Function spaces and applications (Lund, 1986), 115–124,
Lecture Notes in Math. 1302, Springer, Berlin, 1988. <217, 406, 408>
[5] D.R. Adams, My love affair with the Sobolev inequality. Sobolev spaces
in mathematics. I, 1–23, Int. Math. Ser. (N. Y.), 8, Springer, New York,
2009. <85>
[6] D.R. Adams, Morrey spaces, Lecture Notes in Applied and Numeri-
cal Harmonic Analysis. Birkhäuser/Springer, Cham, 2015. xv+121 pp.
<143, 288, 290, 324, 364, 365, 360, 408, 434, 435, 437, 438>
[7] D.R. Adams and L.I. Hedberg, Function spaces and potential theory,
Grundlehren der Mathematischen Wissenschaften [Fundamental Prin-
ciples of Mathematical Sciences], 314. Springer-Verlag, Berlin, 1996.
<289, 216, 217>
[8] D.R. Adams and J. Xiao, Nonlinear potential analysis on Morrey spaces
and their capacities, Indiana Univ. Math. J. 53 (2004), 1629–1663. <72,
80, 216, 406, 407, 408>
[9] D.R. Adams and J. Xiao, Morrey potentials and harmonic maps, Comm.
Math. Phys., 308, (2011), 439–456. <291>
[10] D.R. Adams and J. Xiao, Morrey spaces in harmonic analysis, Ark.
Mat., 50 (2012), 201–230. <212, 435>
[11] D.R. Adams and J. Xiao, Regularity of Morrey commutators, Trans.
Amer. Math. Soc., 364 (2012), 4801–4818. <438>
[12] R.A. Adams, Sobolev spaces, Pure and Applied Mathematics, 65, Aca-
demic Press, New York-London, 1975. <85>

441
442 Bibliography

[13] R.A. Adams and J.J.F. Fournier, Sobolev spaces. Volume 140 of Pure
and Applied Mathematics,V Elsevier, Academic Press, New York, 2nd
edition, 2003. <85>
[14] A. Akbulut, V.S. Guliyev, S. Celik and M.N. Omarova, Fractional inte-
gral associated with Schrödinger operator on vanishing generalized Mor-
rey spaces, J. Math. Inequal. 12 (2018), no. 3, 789–805. <360>
[15] A. Akbulut and O. Kuzu, Marcinkiewicz integrals with rough kernel
associated with Schrödinger operator on vanishing generalized Morrey
spaces, Azerb. J. Math. 4 (2014), no. 1, 40–54. <360>
[16] R.M. Alireza, An embedding theorem for Sobolev type functions with
gradients in a Lorentz space, Studia Math. 191 (2009), no. 1, 1–9.
<217>
[17] J. Alvarez, M. Guzmán-Partida and J. Lakey, Spaces of bounded λ-
central mean oscillation, Morrey spaces, and λ-central Carleson mea-
sures, Collect. Math., 51 (2000), 1–47. <82>
[18] J. Alvarez, The distribution function in the Morrey spaces, Proc. Amer.
Math. Soc. 83 (1981), no. 4, 693–699. <31>
[19] K.F. Andersen and R.T. John, Weighted inequalities for vecter-valued
maximal functions and singular integrals, Studia Math., 69 (1980), 19–
31. <339, 340>
[20] R. Arai and E. Nakai, An extension of the characterization of CMO and
its application to compact commutators on Morrey spaces, J. Math. Soc.
Japan 72 (2020), no. 2, 507–539. <438>
[21] A. Arenas and O. Ciaurri, Fourier-Jacobi expansions in Morrey spaces,
J. Math. Anal. Appl. 415 (2014), no. 1, 302–313. <360>
[22] H. Armutcu, A. Eroglu and F. Isayev, Characterizations for the frac-
tional maximal operators on Carleson curves in local generalized Morrey
spaces, Tbilisi Math. J. 13 (2020), no. 1, 23–38. <434>
[23] A.S. Balakishiyev, V.S. Guliyev, F. Gurbuz and A. Serbetci, Sublinear
operators with rough kernel generated by Calderón-Zygmund operators
and their commutators on generalized local Morrey spaces, J. Inequal.
Appl. 2015, 2015:61, 18 pp. <438>
[24] R.J. Bagby, An extended inequality for the maximal function, Proc.
Amer. Math. Soc. 48 (1975), 419–422. <80>
[25] A.S. Balakishiyev, E.A. Gadjieva, F. Gurbuz and A. Serbetci, Bounded-
ness of some sublinear operators and their commutators on generalized
local Morrey spaces, Complex Var. Elliptic Equ. 63 (2018), no. 11, 1620–
1641. <73, 439>
 Bibliography 443

[26] M. Beigböck, W. Schachermayer and B. Veliyev, A short proof of the

Doob-Meyer theorem, Stochastic Processes and their applications, 122,
Issue 4 (2012), 1204–1209. <108>
[27] A. Benedek and R. Panzone, The spaces LP , with mixed norm, Duke
Math. J. 28 (1961), 301–324. <80>
[28] A. Benedek, A. Calderón and R. Panzone, Convolution operators on
Banach space valued functions, Proc. Nat. Acad. Sci. U. S. A. 48 (1962),
356–365. <218>
[29] C. Bennett and R. Sharpley, Interpolation of Operators, Academic
Press, 1988. <289, 323, 324, 363, 79, 80, 83, 84, 215, 244, 360, 377,
407>
[30] J. Bergh and J. Löfström, Interpolation spaces. An introduction.
Grundlehren der Mathematischen Wissenschaften, No. 223. Springer-
Verlag, Berlin-New York, 1976. <323, 324, 363, 364, 83>
[31] A. Beurling, Construction and analysis of some convolution algebra,
Ann. Inst. Fourier, 14 (1964), 1–32. <82>
[32] N. Bez, S. Lee, S. Nakamura and Y. Sawano, Sharpness of the Brascamp-
Lieb inequality in Lorentz spaces, Electron. Res. Announc. Math. Sci.
24 (2017), 53–63. <84>
[33] B.T. Bilalov, A.A, Huseynli and S.R. El-Shabrawy, Basis properties of
trigonometric systems in weighted Morrey spaces, Azerb. J. Math. 9
(2019), no. 2, 200–226. <360>
[34] B.T. Bilalov and A.A. Quliyeva, On basicity of exponential systems in
Morrey-type spaces, Internat. J. Math. 25 (2014), no. 6, 1450054, 10
pp. <144, 360>
[35] J.S. Bradley, Hardy inequalities with mixed norms, Canad. Math. Bull.
21 (1978), 405–408. <340>
[36] M. Bramanti, Commutators of integral operators with positive kernels,
Le Matematiche (Catania) 49 (1994), no. 1, 149–168. <141, 245>
[37] O. Blasco, A. Ruiz and L. Vega, Non-interpolation in Morrey–
Campanato and block spaces, Ann. Scuola Norm. Sup. Pisa Cl. Sci.
(4) 28 (1999), no. 1, 31–40. <325, 365, xii>
[38] S. Bochner, Integration von Funktionen, deren Werte die Elemente eines
Vektorraumes sind, Fundamenta Mathematicae, 20 (1933), 262–276.
<108>
[39] S.M. Buckley, Estimates for operator norms on weighted spaces and
reverse Jensen inequalities, Trans. Amer. Math. Soc., 340(1) 253–272,
1993. <339>
444 Bibliography

[40] V.I. Burenkov, On partition of unity. Proceeding of the Steklov Institute

of Mathematics, Issue 4, 25–31 (1981) <85>
[41] V.I. Burenkov, Sobolev spaces on domains, Teubner-Texte Zur Mathe-
matik, 137, (1998) Springer. <141, 142, 85>
[42] V.I. Burenkov, Recent progress in the problem of the boundedness of
classical operators of real analysis, Eurasian Mathematical Journal 3
(2012), no. 3, 8–27. <223, 241, 80>
[43] V.I. Burenkov, A. Ghorbanalizadeh and Y. Sawano, Uniform bounded-
ness of Kantorovich operators in Morrey spaces, Positivity 22 (2018),
no. 4, 1097–1107. <215>
[44] V.I. Burenkov, A. Gogatishvili, V.S. Guliyev and R.Ch. Mustafayev,
Boundedness of the fractional maximal operator in local Morrey-type
spaces, Complex Var. Elliptic Equ. 55 (8-10) (2010), 739–758. <240,
241, 217>
[45] V.I. Burenkov and H.V. Guliyev, Necessary and sufficient conditions for
boundedness of the maximal operator in the local Morrey-type spaces,
Studia Math. 163, (2) (2004), 157–176. <224, 240, 241, xi, 341>
[46] V.I. Burenkov and V.S. Guliyev, Necessary and sufficient conditions
for the boundedness of the Riesz potential in local Morrey-type spaces,
Potential Anal. 30 (3) (2009), 211–249. <68, 70, 241, 215, 217>
[47] V.I. Burenkov, V.S. Guliyev, A. Serbetci and T.V. Tararykova, Neces-
sary and sufficient conditions for the boundedness of genuine singular
integral operators in local Morrey type spaces, Eurasian Math. J. 1
(2010), no. 1, 32–53. <241, 218, 219, 341, 438>
[48] V.I. Burenkov, V.S. Guliyev, T.V. Tararykova and A. Serbetchi, A. Nec-
essary and sufficient conditions for the boundedness of genuine singu-
lar integral operators in Morrey-type local spaces (Russian), Dokl. Akad.
Nauk 422 (2008), no. 1, 11–14; translation in Dokl. Math. 78 (2008), no. 2,
651–654. <240, 218, 438>
[49] V.I. Burenkov and E.D. Nursultanov, Description of interpolation spaces
for local Morrey-type spaces. (Russian), Tr. Mat. Inst. Steklova 269
(2010), Teoriya Funktsii i Differentsialnye Uravneniya, 52–62; translation
in Proc. Steklov Inst. Math. 269 (2010), no. 1, 46–56. <325, 365, xi>
[50] V.I. Burenkov, E.D. Nursultanov and D.K. Chigambayeva, Description
of the Interpolation Spaces for a Pair of Local Morrey-Type Spaces and
Their Generalizations, Proceedings of the Steklov Institute of Mathe-
matics, 2014, 284, 97–128. <324, 325, 83>
[51] A.P. Calderón, On the theorems of M. Riesz and Zygmund, Proc. Amer.
Math. Soc. 1 (1950), 533–535. <218>
 Bibliography 445

[52] A.P. Calderón, Intermediate spaces and interpolation, the complex

method, Studia Math. 24 (1964), 113–190. no. 1, 46–56. <364, 365, 108>
[53] A.P. Calderón, Singular integrals, Bull. Amer. Math. Soc. 72 (1966),
427–465. <218>
[54] A.P. Calderón and A. Zygmund, On the existence of certain singular
integrals, Acta Math. 88, 85–139 (1952) <363, 219>
[55] A.P. Calderón and A. Zygmund, On singular integrals. Amer. J. Math.,
78 (1956), 289–309. <218>
[56] Z. Cakir, C. Aykol, D. Soylemez and A. Serbetci, Approximation by
trigonometric polynomials in weighted Morrey spaces, Tbilisi Math. J.
13 (2020), no. 1, 123–138. <360>
[57] S. Campanato, Proprietá di inclusione per spazi di Morrey, Ricerche
Mat. 12 (1963), 67–86. <141, 245>

[58] S. Campanato, Proprietà di h´’olderianità di alcune classi di funzioni,

Ann. Scuola Norm. Sup. Pisa (3) 17 1963 175–188. <245>
[59] S. Campanato, Proprietà di una famiglia di spazi funzionali, (Italian)
Ann. Scuola Norm. Sup. Pisa (3) 18 (1964), 137–160. <142, 127, 245>
[60] S. Campanato, Equazioni ellittiche del II◦ ordine e spazi L(2,λ) , Ann.
Pura. Mat. Appl. (4) 69 (1965), 321–381. <217, 245>
[61] X. Cao and D. Chen, The boundedness of Toeplitz-type operators on
vanishing-Morrey spaces, Anal. Theory Appl. 27 (2011), no. 4, 309–319.
<360>
[62] A. Carbonaro, G. Mauceri and S. Meda, H 1 and BMO for certain non-
doubling metric measure spaces, Ann. Sc. Norm. Super. Pisa Cl. Sci. (5)
8 (2009), 543–582. <291>
[63] A. Carbonaro, G. Mauceri and S. Meda, H 1 and BMO for certain locally
doubling metric measure spaces of finite measure, Colloq. Math. 118
(2010), 13–41. <291>
[64] M.J. Carro, C. Pérez, F. Soria and J. Soria, Maximal functions and
the control of weighted inequalities for the fractional integral operator,
Indiana Univ. Math. J., 54 (2005), no. 3, 627–644. <81, 84, 215>
[65] L. Caso, R. D’Ambrosio and S. Monsurro, Some remarks on spaces of
Morrey type, Abstr. Appl. Anal. 2010, Art. ID 242079, 22 pp. <287,
288, 360, 361>
[66] S. Chanillo, A note on commutators, Indiana Univ. Math. J. 31 (1982),
7–16. <xi, 245>
446 Bibliography

[67] D.X. Chen and F.T. Jin, The boundedness of Marcinkiewicz integrals
associated with Schrödinger operator on Morrey spaces, J. Funct. Spaces
2014, Art. ID 901267, 11 pp. <437>
[68] W.G. Chen and E. Sawyer, A note on commutators of fractional integrals
with RBMO(µ) functions, Illinois J. Math. 46 (2002), no. 4, 1287–1298.
<290>

[69] Y.M. Chen, S. Levine and M. Rao, Variable exponent, linear growth
functionals in image restoration, SIAM J. Appl. Math. 66 (2006), 1383–
1406. <84>
[70] Y. Chen and K. Lau, Some new classes of Hardy spaces, J. Funct. Anal.,
84 (1989), 255–278. <246>
[71] Y.P. Chen, Y. Ding and X. Wang, Compactness of commutators of Riesz
potential on Morrey spaces, Potential Anal. 30 (2009), no. 4, 301–313.
<438>
[72] Y.P. Chen, Y. Ding and X.X. Wang, Commutators of Littlewood-Paley
operators on the generalized Morrey space, J. Inequal. Appl. 2010, Art.
ID 961502, 20 pp. <438>
[73] Y.P. Chen, Y. Ding and X. Wang, Compactness for commutators of
Marcinkiewicz integrals in Morrey spaces, Taiwanese J. Math. 15 (2011),
no. 2, 633–658. <438>
[74] Y.P. Chen, Y. Ding and X. Wang, Compactness of commutators for
singular integrals on Morrey spaces, Canad. J. Math. 64 (2012), no. 2,
257–281. <438>
[75] F. Chiarenza and M. Frasca, Morrey spaces and Hardy–Littlewood max-
imal function, Rend. Mat., 7 (1987), 273–279. <141, xi, 421, 434, 435,
436, 437>

[76] F. Chiarenza, M. Frasca and P. Longo, Interior W 2,p estimates for non
divergence elliptic equations with discontinuous coefficients, Ricerche di
Matematica, 15, fasc. 1◦ , (1991), 149–168. <141, 127, 219, 245>
[77] F. Chiarenza, M. Frasca and P. Longo, W 2,p -solvability of the Dirich-
let problem for nondivergence elliptic equations with VMO coefficients,
Trans. Amer. Math. Soc. 336(2) (1993), 841–853. <83>
[78] G. Choquet, Theory of capacities. Ann. Inst. Fourier Grenoble 5, 1953–
1954, 131–295 (1955) <397, 408>
[79] A. Cianchi and L. Pick, Sobolev embeddings into spaces of Campanato,
Morrey, and Hölder type, J. Math. Anal. Appl. 282 (2003), no. 1, 128–
150. <218>
 Bibliography 447

[80] F. Cobos, J. Peetre and L. E. Persson. On the connection between real

and complex interpolation of quasi-Banach spaces, Bull. Sci. Math. 122
(1998) 17–37. <365, xii>
[81] D.I. Cohn, Second edition. Birkhäuser Advanced Texts: Basler
Lehrbücher. [Birkhäuser Advanced Texts: Basel Textbooks]
Birkhäuser/Springer, New York, 2013. xxi+457 pp. <7, 108>
[82] R.R. Coifman and C. Fefferman, Weighted norm inequalities for maxi-
mal functions and singular integrals, Studia Math., 51 (1974), 241–250,
<339, 340>
[83] R.R. Coifman and Y. Meyer, Au dela des operateurs pseudo-differentiels,
Asterisque 57 (1979). <218>
[84] R.R. Coifman, R. Rochberg, and G. Weiss, Factorization theorems for
Hardy spaces in several variables, Ann. of Math. 103 (1976), 611–635.
<xi, 244, 245>
[85] R.R. Coifman and G. Weiss, Analyse Harmonique Non-commutative sur
Certains Espaces Homogènes, Lecture Notes in Math. 242, Springer,
Berlin, 1971. <126>
[86] J.M. Conde-Alonso and G. Rey, A pointwise estimate for positive dyadic
shifts and some applications, Math. Ann. 365 (2016), 1111–1135. <340>
[87] M.J. Cooper, Coordinated linear spaces. Proc. London Math. Soc. (3)
3 (1953), 305–327. <407>
[88] M. Cotlar, A unified theory of Hilbert transforms and ergodic theorems,
Rev. Mat. Cuyana, 1 (1955), 105–167. <108>
[89] D. Cruz-Uribe, J.M. Martell and C. Pérez, Weights, Extrapolation and
the Theory of Rubio de Francia, Weights, extrapolation and the theory
of Rubio de Francia. Operator Theory: Advances and Applications, 215.
Birkhäuser/Springer Basel AG, Basel, 2011. xiv+280 pp. <292, 339>
[90] D. Cruz-Uribe and K. Moen, A fractional Muckenhoupt-Wheeden the-
orem and its consequences, Integral Equations Operator Theory 76
(2013), no. 3, 421–446. <217>
[91] D. Cruz-Uribe and K. Moen, One and two weight norm inequalities for
Riesz potentials, Illinois J. Math. 57 (2013), no. 1, 295–323. <217>
[92] D. Cruz-Uribe, K. Moen and S. Rodney, Matrix Ap weights, degenerate
Sobolev spaces, and mappings of finite distortion, J. Geom. Anal. (2016)
26:2797–2830. <291>
[93] D. Cruz-Uribe and C. Pérez, Two-weight, weak-type norm inequalities
for fractional integrals, Calderón–Zygmund operators and commutators,
Indiana Univ. Math. J. 49 (2000), 697–721. <340>
448 Bibliography

[94] G.P. Curbera, J. Garcı́a-Cuerva, J. M. Martell and C. Pérez, Extrapo-

lation with weights, rearrangement-invariant function spaces, modular
inequalities and applications to singular integrals, Adv. Math. 20 (2006),
203(1), 256–318. <340>
[95] B. Dahlberg, Regularity properties of Riesz potentials, Indiana Univ.
Math. J. 28 (1979), 257–268. <289, 388>
[96] F. Deringoz, V.S. Guliyev and G. Hasanov, Characterizations for the
Riesz potential and its commutators on generalized Orlicz-Morrey
spaces, J. Inequal. Appl. 2016, Paper No. 248, 22 pp. <439>
[97] F. Deringoz, V.S. Guliyev and G. Hasanov, Maximal operator and its
commutators on generalized weighted Orlicz-Morrey spaces, Tokyo J.
Math. 41 (2018), no. 2, 347–369. <290, 438, 439>
[98] F. Deringoz, V.S. Guliyev and G. Hasanov, Commutators of fractional
maximal operator on generalized Orlicz-Morrey spaces, Positivity 22
(2018), no. 1, 141–158. <439>
[99] F. Deringoz, V.S. Guliyev and S. Samko, Boundedness of the maximal
and singular operators on generalized Orlicz–Morrey spaces, Operator
theory, operator algebras and applications, 139–158, Oper. Theory Adv.
Appl., 242, Birkhäuser/Springer, Basel, 2014. <101, 283, 290, 438>
[100] R.A. DeVore and B. Sharpely, Maximal functions measuring smooth-
ness, Mem. Amer. Math. Soc. 47:293 (1984), viii+115. <246>
[101] Y. Ding and S. Lu, Weighted norm inequalities for fractional integral
operators with rough kernel, Canad. J. Math. 50, 29–39. <217>
[102] Y. Ding, D. Yang and Z. Zhow, Boundedness of sublinear operators
and commutators on Lp,w (Rn ). Yokohama Math. J., 46 (1998), 15–26.
<438>
[103] J. Dieudonné, Sur les espaces de K´’othe. (French) J. Anal. Math. 1
(1951), 81–115. <407>
[104] J. Duoandikoetxea, Fourier Analysis. Translated and revised from the
1995 Spanish original by D. Cruz-Uribe. Graduate Studies in Mathemat-
ics, 29. American Mathematical Society, Providence, RI, 2001. <214,
215, 216, 218, 244, 339>
[105] J. Duoandikoetxea and M. Rosenthal, Extension and boundedness of
operators on Morrey spaces from extrapolation techniques and embed-
dings, J. Geom. Anal. 28 (2018), no. 4, 3081–3108. <339>
[106] J. Duoandikoetxea and M. Rosenthal, Boundedness of operators on
certain weighted Morrey spaces beyond Muckenhoupt range, Potential
Analysis, online. <151, 156, 193, 208, 213, xi>
 Bibliography 449

[107] J. Duoandikoetxea and M. Rosenthal, Boundedness properties in a fam-

ily of weighted Morrey spaces with emphasis on power weights, J. Funct.
Anal. 279 (2020), no. 8, 108687. <151, 210>
[108] X.T. Duong and L.X. Yan, New Morrey-Campanato spaces associated
with operators and applications, East J. Approx. 15 (2009), no. 3, 345–
374. <144, 245>
[109] X.T. Duong, J. Xiao and L.X. Yan, Old and new Morrey spaces with
heat kernel bounds, J. Fourier Anal. Appl. 13 (2007), 87–111. <245>
[110] M.S. Dzhabrailov and S.Z. Khaligova, Anisotropic fractional maximal
operator in anisotropic generalized Morrey spaces, J. Math. Res. 4
(2012), no. 6, 109–120. <83>
[111] M.S. Dzhabrailov and S.Z. Khaligova, Some embeddings into the
anisotropic Morrey and modified anisotropic Morrey spaces. Some appli-
cations, An. –tiin–. Univ. Al. I. Cuza Ia–i. Mat. (N.S.) 60 (2014), no. 1,
245–257. <83>
[112] D. Edmunds, V.M. Kokilashvili and A. Meskhi, Bounded and com-
pact integral operators. Kluwer Academic Publishers, Dordrecht, Boston
London, 2002. <83, 290, 291, 339, 340, 341, 408, 435>
[113] A. Eroglu, A.A. Hasanov and M.N. Omarova, (p, q)-admissible multi-
linear fractional integral operators and their commutators in product
generalized local Morrey spaces, Trans. Natl. Acad. Sci. Azerb. Ser.
Phys.-Tech. Math. Sci. 36 (2016), no. 4, Mathematics, 77–98. <439>
[114] M. Essoh, I. Fofana and K. Koua, Inégalités de type faible pour
l’opérateur maximal fractionnaire dans les espaces de Morrey par rap-
port à la capacite de Hausdorff (French), Ital. J. Pure Appl. Math. No.
28 (2011), 81–92. <408>
[115] Eridani and H. Gunawan, Stummel class and Morrey spaces, Southeast
Asian Bull. Math. 29 (2005), 1053–1056. <141, 81>
[116] Eridani, V.M. Kokilashvili and A. Meskhi Morrey spaces and fractional
integral operators, Expo. Math. 27 (2009), no. 3, 227–239. <143, 81>
[117] L.C. Evans, Partial differential equations. Second edition. Graduate
Studies in Mathematics, 19. American Mathematical Society, Provi-
dence, RI, 2010. xxii+749 pp. <65, 85, 217>
[118] L.C. Evans and R.F. Gariepy, Measure theory and fine properties of
functions / Lawrence C. Evans and Ronald F. Gariepy. – CRC Press,
1999. – (Studies in Advanced Mathematics) <85, 214, 403>
[119] E.B. Fabes, C. Gutiérrez and R. Scotto, Weak type estimates for the
Riesz transforms associated with the Gaussian measure, Rev. Mat.
Iberoam. 10 (1994), 229–281. <291>
450 Bibliography

[120] Y. Fan and G. Gao, Some estimates of rough singular bilinear fractional
integral, J. Funct. Spaces. Appl. 2012, Article ID 406540, 17 pages.
<16, 83>
[121] X. Fan and D. Zhao, On the spaces Lp(x) (Ω) and W m,p(x) (Ω), J. Math.
Anal. Appl. 263 (2001), no. 2, 424–446. <84>
[122] G. Di Fazio, C.E. Gutierez and E. Lanconelli, Covering theorems,
inequalites on metric spaces and applications to PDE’s, Math. Ann.
341 2 (2008), 255–291. <289>
[123] G. Di Fazio and M.A.G. Ragusa, Commutators and Morrey spaces, Boll.
Un. Mat. Ital. A (7) 5 (1991), no. 3, 323–332. <xi, 434, 435, 438>
[124] C. Fefferman and E.M. Stein, Some maximal inequalities, Amer. J.
Math, 93 (1971), 107–115. <215>
[125] C. Fefferman and E. Stein, H p spaces of several variables, Acta Math.
129 (1972), 137–193. <215, 244>
[126] H. Feichtinger, An elementary approach to Wiener’s third Tauberian
theorem on Euclidean n-space, In: Proceedings of Conference at Cortona
1984, Symposia Mathematica, 29 (1987), 267–301, Academic Press, New
York. <35, 82>
[127] G.B. Folland, Introduction to Partial Differential Equations. Second
Edition. Princeton University Press, Princeton, NJ, 1995. xii+324 pp.
<127>
[128] J. Fournier, Mixed norms and rearrangements: Sobolev’s inequality and
Littlewood’s inequality, Ann. Math. Pura Appl. 148 (1987), 51–76.
<80>
[129] M. Frazier and B. Jawerth, Decomposition of Besov spaces. Indiana
Univ. Math. J. 34 (1985), no. 4, 777–799. <126>
[130] X. Fu, D. Yang and W. Yuan, Generalized fractional integrals and their
commutators over non-homogeneous metric measure spaces, Taiwanese
J. Math. 18 (2014), no. 2, 509–557. <290>
[131] Z.W. Fu, λ-central BMO estimates for commutators of N -dimensional
Hardy operators, J. Inequal. Pure Appl. Math. 9 (2008), no. 4, Article
111, 5 pp. <246>
[132] Z.W. Fu, S.Z. Lu, H.B. Wang and L.G. Wang, Singular integral opera-
tors with rough kernels on central Morrey spaces with variable exponent,
Ann. Acad. Sci. Fenn. Math. 44 (2019), no. 1, 505–522. <438>
[133] Z.W. Fu, Y. Lin and S.Z. Lu, λ-central BMO estimates for commutators
of singular integral operators with rough kernels, Acta Math. Sin. (Engl.
Ser.) 24 (2008), no. 3, 373–386. <437>
 Bibliography 451

[134] Z.W. Fu and S.Z. Lu, Weighted Hardy operators and commutators on
Morrey spaces, Front. Math. China 5 (2010), no. 3, 531–539. <437>
[135] N. Fujii, Weighted bounded mean oscillation and singular integrals,
Math. Japon. 22 (1977/1978), 529–534. <324>
[136] N. Fujii, A proof of the Fefferman-Stein-Strömberg inequality for the
sharp maximal functions, Proc. Amer. Soc. 106 (1991), no.2, 371–377.
<216, 340>
[137] Y. Furusho, On inclusion property for certain L(p,λ) spaces of strong
type, Funkcial Ekvac. 23 (1980), no. 2,197–205. <273, 81>
[138] T. Futamura, Y. Mizuta and T. Shimomura, Sobolev embedding for
variable exponent Riesz potentials on metric spaces, Ann. Acad. Sci.
Fenn. Math. 31 (2006), 495–522. <290>
[139] S. Gala On regularity criteria for the three-dimensional micropolar fluid
equations in the critical Morrey-Campanato space, Nonlinear Anal. Real
World Appl. 12 (2011), no. 4, 2142–2150. <81>
[140] G. Gao, Boundedness for commutators of n-dimensional rough Hardy
operators on Morrey-Herz spaces, Comput. Math. Appl. 64 (2012), no.
4, 544–549. <83>
[141] J. Garcı́a-Cuerva, Hardy spaces and Beurling algebras, J. London Math.
Soc. (2), 39, 499–513 (1989) <246>
[142] J. Garcı́a-Cuerva and M.J. L. Herrero, A theory of Hardy spaces assosi-
ated to the Herz spaces, Proc. London Math. Soc. (1994), 69, 605–628.
<82>
[143] J. Garcı́a-Cuerva and J.M. Martell, Weighted inequalities and vector-
valued Carderón–Zygmund operators on nonhomogeneous spaces, Publ.
Mat. 44 (2000), no. 2, 613–640. <290>
[144] J. Garcı́a-Cuerva and J.M. Martell, On the existence of principal values
for the Cauchy integral on weighted Lebesgue spaces for non-doubling
measures, J. Fourier Anal. Appl. 7 (2001), no. 5, 469–487. <290>
[145] J. Garcı́a-Cuerva, G. Mauceri, P. Sjögren and J.L. Torrea, Higher-order
Riesz operators for the Ornstein-Uhlenbeck semigroup, Potential Anal.
10 (1999), 379–407. <291>
[146] J. Garcı́a-Cuerva and J.L. Rubio de Francia, Weighted norm inequalities
and related topics, North-Holland Math. Stud., 116 (1985). <xiii, 143,
214, 215, 218, 244, 339, 340>
[147] I. Genebashvili, A. Gogatishvili, V.M. Kokilashvili and M. Krbec,
Weight theory for integral transforms on spaces of homogeneous type,
Pitman Monographs and Surveys in Pure and Applied Mathematics,
92. Longman, Harlow, 1998. <289, 83, 84, 218, 244, 290, 291, 339, 340,
435>
452 Bibliography

[148] M. Giaquinta, Multiple Integrals in the Calculus of Variations and Non-

linear Elliptic Systems, Ann. of Math. Stud. 105, Princeton Univ. Press,
Princeton, NJ, 1983. <142, 81>
[149] Y. Giga and T. Miyakawa, Navier-Stokes flow in R3 with measures as
initial vorticity and Morrey spaces, Comm. Partial Differential Equa-
tions 14 (1989), no. 5, 577–618. <80, 81>

[150] D. Gilberg and N.S. Trudinger, Elliptic Partial Differential Equations

of Second Order, 2nd ed., Springer-Verlag, Berlin, (1983). <217>
[151] A. Gogatishvili and R.Ch. Mustafayev, New pre-dual space of Morrey
space, J. Math. Anal. Appl. 397 (2013), 678–692. <407>
[152] A. Gogatishvili, R.Ch. Mustafayev and M. Agcayazi, Weak-type esti-
mates in Morrey spaces for maximal commutator and commutator of
maximal function, Tokyo J. Math. 41(1):193–218. <215>
[153] R. Gong, The area integral associated to self-adjoint operators on
weighted Morrey spaces, J. Math. Sci. Adv. Appl. 16 (2012), no. 1-2,
47–59. <210, 438>
[154] P. Gorka, Campanato theorem on metric measure spaces, Ann. Acad.
Sci. Fenn. Math. 34 (2009), no. 2, 523–528. <142, 290>
[155] F.Q. Gouvêa, p-adic Numbers, An Introduction (2nd ed.). Springer
(2000). <143, 266>
[156] L. Grafakos, Classical Fourier analysis, Graduate texts in mathematics,
249, Springer, New York, 2008. <323, xii, xiii, 49, 79, 83, 143, 214, 216,
339, 408>
[157] L. Grafakos, Modern Fourier analysis, Graduate Texts in Mathematics,
250, Springer, New York, 2009. <xii, 1, 15, 289, 363, xiii, 143, 215,
244>
[158] L. Grafakos, H.V. Nguyen, S. Nakamura and Y. Sawano, Conditions
for boundedness into Hardy spaces, Math. Nachr. 292 (2019), no. 11,
2383–2410. <389>
[159] B. Grevholm, On the structure of the spaces Lp,λ
k , Math. Scand. 26
(1970), 241–254. <246>
[160] H. Greenwald, On the theory of homogeneous Lipschitz spaces and Cam-
panato spaces, Pacific J. Math. 106 (1983), no. 1, 87–93. <246>
[161] G. Gu and M. Cai, Sharp maximal function inequalities and bounded-
ness for commutators related to generalized fractional singular integral
operators, J. Inequal. Appl. 2014, 2014:211, 13 pp. <245>
 Bibliography 453

[162] V.S. Guliyev, Integral operators on function spaces on the homogeneous

groups and on domains in Rn (Russian), Doctor degree dissertation,
Mat. Inst. Steklov, Moscow 1994. <67, 70, 219>
[163] V.S. Guliyev, Function spaces, Integral operators and two weighted
inequalities on homogeneous groups. Some Applications (Russian), Baku
1999. <70, 240, 219>
[164] V. S. Guliyev, Boundedness of the maximal, potential and singular oper-
ators in the generalized Morrey spaces, J. Inequal. Appl. 2009, Art. ID
503948, 20 pp. <70, 71, 217>
[165] V.S. Guliyev, F. Deringoz and J.J. Hasanov, (Φ, Ψ)-admissible singu-
lar operators and their commutators on vanishing generalized Orlicz-
Morrey spaces, J. Inequal. Appl.2014, 2014:143, 18 pp. <290, 439>
[166] V.S. Guliyev, F. Deringoz and J.J. Hasanov, (Φ, Ψ)-admissible potential
operators and their commutators on vanishing Orlicz–Morrey spaces,
Collect. Math. 67 (2016), no. 1, 133–153. <290, 439>
[167] V.S. Guliyev, J.J. Hasanov and Y. Sawano, Decompositions of local
Morrey-type spaces, Positivity 21 (2017), no. 3, 1223–1252. <83>
[168] V.S. Guliyev and R.Ch. Mustafayev, Boundedness of the anisotropic
maximal and anisotropic singular integral operators in generalized Mor-
rey spaces, Acta Math. Sin. 27 (2011), no. 12, 2361–2370. <144, xi>
[169] V.S. Guliyev, M.N. Omarova, M.A. Ragusa and A. Scapellato, Commu-
tators and generalized local Morrey spaces, J. Math. Anal. Appl. 457
(2018), no. 2, 1388–1402. <435, 439>
[170] G.S. Guliyev, M.N. Omarova and Y. Sawano, Boundedness of intrinsic
square functions and their commutators on generalized weighted Orlicz–
Morrey spaces, Banach J. Math. Anal. 9 (2015), no. 2, 44–62. <213,
216, 439>
[171] V.S. Guliyev, K.R. Rahimova and M.N. Omarova, Commutators of
vector-valued intrinsic square functions on vector-valued generalized
Morrey spaces, Trans. Natl. Acad. Sci. Azerb. Ser. Phys.-Tech. Math.
Sci. 34 (2014), no. 1, Mathematics and Mechanics, 35–50. <439>
[172] H. Gunawan, D.I. Hakim, K.M. Limanta and A.A. Masta, Inclusion
properties of generalized Morrey spaces, Math. Nachr. 290 (2017), no.
2-3, 332–340. <81>
[173] H. Gunawan, D.I. Hakim, E. Nakai and Y. Sawano, On inclusion relation
between weak Morrey spaces and Morrey spaces, Nonlinear Analysis,
168 (2018), 27–31. <81>
[174] C.E. Gutiérrez, On the Riesz transforms for Gaussian measures, J.
Funct. Anal. 120 (1994), 107–134. <291>
454 Bibliography

[175] C.E. Gutiérrez, C. Segovia and J.L. Torrea, On higher order Riesz trans-
forms for Gaussian measures, J. Fourier Anal. Appl. 2 (1996), 583–596.
<291>
[176] De Guzman, Real Variable Methods in Fourier Analysis, North-Holland
Mathematical Studies 46, Notas de Matematica (75) (1981). <xiii, 143>
[177] D.I. Hakim, Complex interpolation of certain closed subspaces of Morrey
spaces, Tokyo J. Math. 41 (2018), no. 2, 487–514. <367, 360>
[178] D.I. Hakim and Y. Sawano, Interpolation of generalized Morrey spaces,
Rev. Mat. Complut. 29 (2016), no. 2, 295–340. <70, 72, 365, 366, 367,
81>
[179] D.I. Hakim and Y. Sawano, Calderón’s first and second complex inter-
polation of closed subspaces of Morrey spaces, J. Fourier Analysis and
Applications, 23 (2017), no. 5, 1195–1226. <363, 365, 367, xii>
[180] D.I. Hakim and Y. Sawano, Complex interpolation of vanishing Morrey
spaces, Annals of Functional Analysis, 11 (2020), 643–661. <16, 360,
408>
[181] D.I. Hakim and Y. Sawano, Complex interpolation of various subspaces
of Morrey spaces, Sci. China Math., 63 (2020), 937–964. <365, 361>
[182] G.H. Hardy and J.E. Littlewood, A maximal theorem with function-
theoretic applications, Acta Math. 54 (1930), no. 1, 81–116. <214>
[183] E. Harboure, J.L. Torrea and B. Viviani, Vector-valued extensions of
operators related to the Ornstein–Uhlenbeck semigroup, J. Anal. Math.
91 (2003), 1–29. <291>
[184] G.H. Hardy and J. Littlewood, Some properties of fractional integrals
I, Math. Z. 27 (1927), 565–606. <216>
[185] G.H. Hardy and J. Littlewood, Some properties of fractional integrals
II, Math. Z. 34 (1932), 403–439. <216>
[186] D.D. Haroske and L. Skrzypczak, Embeddings of weighted Morrey
spaces, Math. Nachr. 290 (2017), no. 7, 1066–1086. <210, 81>
[187] P. Harjulehto, P. Hästö, V. Latvala and O. Toivanen, Critical variable
exponent functionals in image restoration, Appl. Math. Lett. 26 (2013),
no. 1, 56–60. <84>
[188] A.A. Hasanov, Sh.A. Muradova and E.A. Gadjieva, Commutators of
multilinear singular integral operators on generalized local Morrey
spaces, Trans. Natl. Acad. Sci. Azerb. Ser. Phys.-Tech. Math. Sci. 35
(2015), no. 4, Mathematics, 61–74. <439>
[189] P. Halmos, Measure Theory, Springer Verlag, New York, (1974). <79>
 Bibliography 455

[190] L.I. Hedberg, On certain convolution inequalities, Proc. Amer. Math.

Soc. 36 (1972), 505–510. <180, 216, 436>
[191] J. Heinonen, Lectures on Analysis on Metric Spaces, Springer-Verlag,
New York, 2001. <2>
[192] S. Hencl and J. Kleprlik, Composition of q-quasiconformal mappings
and functions in Orlicz–Sobolev spaces, Illinois J. Math. 56(3) (2012),
931–955. <64>
[193] C. Herz, Lipschitz spaces and Bernstein’s theorem on absolutely conver-
gent Fourier transforms, J. Math. Mech., 18 (1968), 283–324. <82>
[194] K.-P. Ho, Hardy’s inequality and Hausdorff operators on rearrangement-
invariant Morrey spaces, Publ. Math. Debrecen 88 (2016), no. 1-2, 201–
215. <434>
[195] K.-P. Ho, Strong maximal operator on mixed-norm spaces, Ann. Univ.
Ferrara 62 (2016), 275–291. <80>
[196] K.-P. Ho, Mixed norm Lebesgue spaces with variable exponents and
applications, Riv. Mat. Univ. Parma 9 (2018), 21–44. <80>
[197] K.P. Ho, Singular integral operators with rough kernels on Morrey type
spaces, Studia Math. 244 (2019), no. 3, 217–243. <438>
[198] K.P. Ho, Definability of singular integral operators on Morrey-Banach
spaces, J. Math. Soc. Japan 72 (2020), no. 1, 155–170. <437>
[199] P. Honzik, J.J. Jaye, On the good-λ inequality for nonlinear potentials,
Proc. Amer. Math. Soc. 140 (2012), no. 12, 4167–4180. <218>
[200] G. Hu, S. Lu and D. Yang, Boundedness of rough singular integral oper-
ators on homogeneous Herz spaces. J. Austral. Math. Soc. Ser. A66
(1999), no. 2, 201–223. <438>
[201] L.L. Hu and D.X. Chen, Some estimates of maximal commutators
for Bochner-Riesz operator on Morrey spaces, Anal. Theory Appl. 26
(2010), no. 1, 76–83. <438>
[202] C.X. Huang and L.Z. Liu, Sharp function inequalities and boundness
for Toeplitz type operator related to general fractional singular integral
operator, Publ. Inst. Math. (Beograd) (N.S.) 92 (106) (2012), 165–176.
<434>
[203] L. Huang and D. Yang, Function spaces with mixed norms, in prepara-
tion. <79>
[204] R.A. Hunt, D. Kurtz and C.J. Neugebauer, A note on the equivalence of
Ap and Sawyer’s condition for equal weights, Conference on Harmonic
Analysis in Honor of Antoni Zygmund, W. Beckner et al., eds., vol. 1,
pp. 156-158, Wadsworth, Belmont, 1983. <341>
456 Bibliography

[205] R. Hunt B. Muckenhoupt, and R. Wheeden, Weighted norm inequali-

ties for the conjugate function and the Hilbert transform, Trans. Amer.
Math, Soc. 176 (1973), 227–251. <318, 339>
[206] T.P. Hytönen, Weighted norm inequalities, Lecture notes of a
course at the University of Helsinki, Winter 2011, http://wiki-
app.it.helsinki.fi/download/attachments/64424417/weighted.pdf.
<339>
[207] T.P. Hytönen, The sharp weighted bound for general Calderón–
Zygmund operators, Ann. of Math. (2) 175 (2012), 1473–1506. <317>
[208] T.P. Hytönen, The A2 theorem: remarks and complements. Harmonic
analysis and partial differential equations, Vol. 612. Contemp. Math.
Amer. Math. Soc., Providence, RI, 2014, 91–106. <340>
[209] T. Hytönen and C. Perez, Sharp weighted bounds involving A∞ , Anal.
PDE 6 (2013), no. 4, 777–818. <290, 340>
[210] T.P. Hytönen, L. Roncal and O. Tapiola, Quantitative weighted esti-
mates for rough homogeneous singular integrals, Israel J. Math. 218
(2017), 133–164. <340>
[211] S. Igari, Interpolation of operators in Lebesgue spaces with mixed norm
and its applications to Fourier analysis, Tohoku Math. J. 38 (1986),
469–490. <80>
[212] T. Iida, Y. Sawano and H. Tanaka, Atomic Decomposition for Morrey
Spaces, Z. Anal. Anwend. 33 (2014), no. 2, 149–170. <408>
[213] D.M. Israfilov and N.P. Tozman, Approximation in Morrey-Smirnov
classes, Azerb. J. Math. 1 (2011), no. 1, 99–113. <361>
[214] M. Izuki, Boundedness of vector-valued sublinear operators on Herz–
Morrey spaces with variable exponent, Math. Sci. Res. J. 13 (2009), no.
10, 243–253. <83>
[215] M. Izuki, Fractional integrals on Herz–Morrey spaces with variable expo-
nent, Hiroshima Math. J. 40 (2010), no. 3, 343–355. <83>
[216] Izuki, M., Nakai, E., Sawano, Y.: Function spaces with variable
exponents–an introduction–, Sci. Math. Jpn. 77, no. 2, 187–315 (2014)
<84, 85>
[217] T. Izumi, E. Sato and K. Yabuta, Remarks on a subspace of Morrey
spaces, Tokyo J. Math. 37 (2014), no. 1, 185–197. <71, 407>
[218] T. Izumi, Y. Sawano and H. Tanaka, Littlewood-Paley theory for Mor-
rey spaces and their preduals, Rev. Mat. Complut. 28 (2015), 411–447.
<388, 406>
 Bibliography 457

[219] S. Janson, Mean oscillation and commutators of singular integral oper-

ators, Ark. Math. 16:2 (1978), 263–270. <439>
[220] S. Janson, M. Taibleson and G. Weiss, Elementary characterizations
of the Morrey–Campanato spaces, Harmonic analysis (Cortona 1982),
Springer, Berlin, 1983, pp. 101–114. <126, 245, 246>
[221] F. John and L. Nirenberg, On functions of bounded mean oscillation,
Comm. Pure Appl. Math. 14 (1961), 415–426. <244, 406>
[222] A. Jonsson, P. Sjögen and H. Wallin, Hardy and Lipschitz spaces on
subsets of Rn , University of Umea, Sweeden, Report No. 8, 1981, 38pp.
<246>
[223] J. Journé, Calderón–Zygmund operators, pseudo differential operators
and the Cauchy integral of Calderón, Lecture Notes in Mathematics,
994. Springer-Verlag, Berlin, 1983. <83, 108, 214, 218, 244, 290, 340>
[224] E.A. Kalita, Dual Morrey spaces, Dokl. Akad. Nauk 361 (1998), no. 4,
447–449. <406, 407>
[225] N. Kalton and M. Mitrea, Stability results on interpolation scales
of quasi-Banach spaces and applications, Trans. Amer. Math. Soc.
350 (1998), no. 10, 3903–3922. <91>
[226] L.V. Kantorovich and G.P. Akilov, Functional analysis, 2nd ed., Perga-
mon Press, Oxford, 1982. <407>
[227] R. Kerman and E. T. Sawyer, The trace inequality and eigenvalue esti-
mates for Schrödinger operators, Annales de l’institute Fourier 36 no. 4
(1986), 207–228. <289, 341>
[228] H. Kita, On maximal functions in Orlicz spaces, Proc. Amer. Math.
Soc., 124 (1996), 3019–3025. <84>
[229] H. Kita, On Hardy–Littlewood maximal functions in Orlicz spaces,
Math. Nachr., 183 (1997), 135–155. <84>
[230] V.M. Kokilashvili, On Hardy’s inequalities in weighted spaces, (Russian)
Soobsch, Akad. Nauk. Gruz, SSR 96 (1979), 37–40. <340>
[231] V.M. Kokilashvili, Weighted estimates for classical integral operators,
Nonlinear analysis, function spaces and applications, 4 (Roudnice nad
Labem, 1990), 86–103, Teubner-Texte Math., 119, Teubner, Leipzig,
(1990). <289, 323, 217, 244, 290, 339, 340, 408, 435>
[232] V.M. Kokilashvili and M. Krbec, Weighted Inqualities in Lorentz and
Orlicz Spaces, vol 57. World Scientific Publishing, River Edge, NJ
(1991). <101, 84>
[233] V.M. Kokilashvili and A. Meskhi, Fractional integrals on measure
spaces, Fract. Calc. Appl. Anal. 4 (2001), no. 1, 1–24. <291>
458 Bibliography

[234] V.M. Kokilashvili and A. Meskhi, Maximal and potential operators in

variable Morrey spaces defined on nondoubling quasimetric measure
spaces, Bull. Georgian Natl. Acad. Sci. (N.S.) 2 (2008), no. 3, 18–21.
<291>
[235] V.M. Kokilashvili, A. Meskhi, H. Rafeiro and S. Samko, Integral Oper-
ators in Non-standard Function Spaces, Volume 1: Variable exponent
Lebesgue and amalgam spaces, Operator Theory: Advances and Appli-
cations, 248. Birkhäuser/Springer, [Cham], 2016. xx+567 pp. <80,
340>
[236] J.J. Koliha, Metrics, Norms and Integrals, An Introduction to Contem-
porary Analysis. World Scientific Publishing Co. Pte. Ltd., Hackensack,
NJ, 2008. xviii+408 pp. <3, 4, 6, 381>
[237] Kolmogoroff, A.: Sur les fonctions harmoniques conjuguées et les séries
de Fourier. Fund. Math. 7 (1925), 24–29. <218>
[238] S. Koshi and T. Shimogaki, On F -norms of quasi-modular spaces, J.
Fac. Sci. Hokkaido Univ. 15 (1961), 202–218. <84>

[239] G. Köthe, Stufenr´’aume, eine einfache Klasse linearer vollkommener

R´’aume. (German), Math. Z. 15 (1948), 317–345. <107, 407>

[240] G. Köthe, Neubegr´’undung der Theorie der vollkommenen R´’aume.

(German) Math. Nachr. 4 (1951), 70–80. <407>
[241] G. Köthe, Topological Vector Spaces, I, Springer-Verlag, 1969. <107,
247>
[242] G. Köthe, Lineare R´’aume mit unendlich vielen Koorinaten J. Reine
Angew. Math. 171 (1934), 193–226. <407>
[243] O. Kováčik and J. Rákosnı́k, On spaces Lp(x) and W k,p(x) , Czechoslovak
Math. 41 (116) (1991), 592–618. <84>
[244] J. Komlós, A generalization of a problem of Steinhaus, Acta Math. Acad.
Sci. Hungar., 18, 217–229. <108>
[245] Y. Komori, Calderón–Zygmund operators on the predual of a Morrey
space, Acta Mathematica Sinica, 19 (2) (2003), 297–302. <437>
[246] Y. Komori, Weighted estimates for operators generated by maximal
functions on nonhomogeneous space, Goergian Math. J. 12 (2005), no.1,
121–130. <291>
[247] Y. Komori-Furuya, Local good-λ estimate for the sharp max-
imal function and weighted Morrey space, Journal of Func-
tion Spaces Volume 2015 (2015), Article ID 651825, 4 pages
http://dx.doi.org/10.1155/2015/651825 <212, 213, 434, 435>
 Bibliography 459

[248] Y. Komori-Furuya and K. Matsuoka, Strong and weak estimates for

fractional integral operators on some Herz-type function spaces, Pro-
ceedings of the Maratea Conference FAAT 2009, Rendiconti del Circolo
Mathematico di Palermo, Serie II, Suppl., 82 (2010), 375–385. <82>
[249] Y. Komori-Furuya and K. Matsuoka, Weighted estimates for singular
integral operators on CMO spaces, Sci. Math. Jpn. 72 (2010), no. 1,
21–28. <82>
[250] Y. Komori-Furuya, K. Matsuoka, E. Nakai and Y. Sawano, Integral
operators on Bσ -Morrey-Campanato spaces, Revista Mathematica Com-
plutense, 26 (2013), no. 1, 1–32. <82>
[251] Y. Komori-Furuya, K. Matsuoka, E. Nakai and Y. Sawano, Integral
operators on B σ spaces, J. Funct. Spaces Appl. 2013, Art. ID 859402,
21 pp. <82>
[252] Y. Komori and T. Mizuhara, Notes on commutators and Morrey spaces,
Hokkaido Math. J. 32 (2003), no. 2, 345–353. <80, 406, 438>
[253] Y. Komori-Furuya and E. Sato, Fractional integral operators on central
Morrey spaces, Math. Inequal. Appl. 20(3) (2017), 801–813. <436>
[254] Y. Komori and S. Shirai, Weighted Morrey spaces and a singular integral
operator, Math. Nachr. 282 (2009), no. 2, 219–231. <157, 160, 194, 210,
211, 213, 215, xi>
[255] N. Kruglyak and E.A. Kuznetsov, Smooth and nonsmooth Calderón–
Zygmund type decompositions for Morrey spaces, J. Fourier Anal. Appl.
11 (2005), no. 6, 697–714. <408>
[256] J.C. Kuang, Generalized Hilbert operators on weighted Morrey-Herz
spaces, J. Math. Inequal. 6 (2012), no. 1, 69–77. <83>
[257] J.C. Kuang, Generalized Hausdorff operators on weighted Morrey-Herz
spaces (Chinese), Acta Math. Sinica (Chin. Ser.) 55 (2012), no. 5, 895–
902. <83>
[258] A. Kufner, O. John and S. Fucik, Function spaces. Monographs and
Textbooks on Mechanics of Solids and Fluids; Mechanics: Analysis.
Noordhoff International Publishing, Leyden; Academia, Prague, 2013.
<7, 65, 79, 80, 83, 84, 245>
[259] K. Kurata and S. Sugano, A remark on estimates for uniformly elliptic
operators on weighted Lp spaces and Morrey spaces, Math. Nachr., 209
(2000), 137–150. <437>
[260] K. Kurata and S. Sugano, Estimates of the fundamental solution for
magnetic Schrödinger operators and their applications, Tohoku Math.
J. (2) 52 (2000), no. 3, 367–382. <437>
460 Bibliography

[261] E.A. Kuznetsov, Maximal Theorem and Calderón–Zygmund Type

Decompositions for the Fractional Maximal Function, Doctoral Thesis,
May 2005, Department of Mathematic Lule◦a; University of Technology
SE-97187 Lule◦a, Sweden. <408>
[262] M.T. Lacey, An elementary proof of the A2 bound, Israel Journal of
Mathematics, 217 (2017), 181–195. <340>
[263] P.G. Lemarié-Rieusset, The Navier-Stokes equations in the critical
Morrey-Campanato space, Rev. Mat. Iberoam. 23 (2007), no. 3, 897–
930. <289, xii>
[264] P.G. Lemarié-Rieusset, The role of Morrey spaces in the study of Navier-
Stokes and Euler equations, Eurasian Math. J. 3 (2012), no. 3, 62–93.
<437>
[265] P.G. Lemarié-Rieusset, Multipliers and Morrey spaces, Potential Anal.
38 (2013), no. 3, 741–752. <278, 287, 288, 325, 364, 365, 366, xii, 30>
[266] P.G. Lemarié-Rieusset, Erratum to: Multipliers and Morrey spaces,
Potential Anal. 41 (2014), no. 3, 1359–1362. <364, 365, 366, xii>
[267] A.K. Lerner, Weighted norm inequalities for the local sharp maximal
function, J. Fourier Analysis and Applications 10, Issue 5 (2004), 465–
474. <215>
[268] A.K. Lerner, An elementary approach to several results on the Hardy–
Littlewood maximal operator. Proc. Am. Math. Soc. 136, no. 8, 2829–
2833 (2008) <339>
[269] A.K. Lerner, Sharp weighted norm inequalities for Littlewood-Paley
operators and singular integrals, Adv. Math. 226 (2011), no. 5, 3912–
3926. <216>

[270] A.K. Lerner, On an estimate of Calderón–Zygmund operators by dyadic

positive operators, J. Anal. Math. 121 (2013), 141–161. <216, 317, 340>
[271] A.K. Lerner, A simple proof of the A2 conjecture, Int. Math. Res. Not.,
2013 (14), 3159–3170. <219>
[272] A.K. Lerner, On pointwise estimates involving sparse operators, New
York J. Math. 22 (2016), 341–349. <340>
[273] A.K. Lerner and F. Nazarov, Intuitive dyadic calculus: the basics, Expo.
Math.37 (2019), no. 3, 225–265. <15, 173, 340>
[274] A.K. Lerner, S. Ombrosi and C. Peréz, A1 bounds for Calderón–
Zygmund related to a problem of Muckenhoupt and Wheeden, Math.
Res. Lett. 16 (2009), 149–156. <339>
 Bibliography 461

[275] A.K. Lerner, S. Ombrosi and P. Rivera-Rios, On pointwise and weighted

estimates for commutators of Calderon–Zygmund operators, Adv. Math.
319 (2017), 153–181. <340>
[276] F. Li, Z. Li and L. Pi, Variable exponent functionals in image restoration.
Appl. Math. Comput. 216 (2010), no. 3, 870–882. <84>
[277] P.T. Li, J.M. Liu and Z.J. Lou, Integral operators on analytic Morrey
spaces, Sci. China Math. 57 (2014), no. 9, 1961–1974. <361>
[278] X. Li and D. Yang, Boundedness of some sublinear operators on Herz
spaces, Illinois J. Math. 40 (1996), no. 3, 484–501. <437>
[279] W.M. Li, John–Nirenberg type inequalities for the Morrey–Campanato
spaces, J. Ineq. Appl. Vol 2008, Article ID 239414, 5 pages. <244>
[280] J.L. Lions and J. Peetre, Sur une classe d’epaces d’interpolation, Inst.
Hautes Études Sci. Publ. Math. 19, 5–68 (1964) <323, 108>
[281] L.G. Liu, Y. Sawano and D. Yang, Morrey-type spaces on Gauss measure
spaces and boundedness of singular integrals, J. Geom. Anal. 24 (2014),
no. 2, 1007–1051. <143, 144, 291>
[282] L.G. Liu, S.Q. Wu, D. Yang and W. Yuan, New characterizations of
Morrey spaces and their preduals with applications to fractional Laplace
equations, J. Diff. Equ. 266 (2019), no. 8, 5118–5167. <388, 435>
[283] L.Z. Liu, Sharp maximal function inequalities and boundedness of
Toeplitz type operator related to general fractional integral operators,
Banach J. Math. Anal. 7 (2013), no. 1, 142–159. <434>
[284] L. Liu and D. Yang, Boundedness of Littlewood-Paley operators asso-
ciated with Gauss measures, J. Inequal. Appl. 2010, Art. ID 643948, 41
pp. <144, 291>
[285] Y. Liu and L.J. Wang, Boundedness for Riesz transform associated
with Schrödinger operators and its commutator on weighted Morrey
spaces related to certain nonnegative potentials, J. Inequal. Appl. 2014,
2014:194, 16 pp. <210, 437>
[286] Y.Q. Liu and D.X. Chen, The boundedness of maximal Bochner-Riesz
operator and maximal commutator on Morrey type spaces, Anal. Theory
Appl. 24 (2008), no. 4, 321–329. <71, 437, 438>
[287] L.H. Loomis, A note on Hilbert’s transform, Bull. Amer. Math. Soc. 52
(1946), 1082–1086. <218>
[288] G.G. Lorentz, Some new functional spaces, Ann. of Math. (2) 51 (1950),
37–55. <83>
[289] G.G. Lorentz, On the theory of spaces Λ, Pac. J. Math. 1 (1951), 411–
429. <83>
462 Bibliography

[290] G.G. Lorentz and D.G. Wertheim, Representation of linear functionals

on Köthe spaces, Canad. J. Math. 5 (1953), 568–575. <407>

[291] S.S. Lu and D. Yang, The Littlewood-Paley function and φ-transform

characterizations of a new Hardy space HK2 associated with the Herz
space, Studia Math. 101, no. 3 (1992), 285–298. <246>
[292] S.S. Lu and D. Yang, The central BMO spaces and Littlewood-Paley
operators, Approx. Theory Appl. 11 (1995), 72–94. <246>
[293] S.S. Lu, Y. Ding and D. Yan, Singular integrals and related topics, World
Scientific, 2007. <15, 289, 323, 214, 215, 216, 217, 218, 244, 290, 339,
435>
[294] Y.F. Lu, D. Yang and W. Yuan, Interpolation of Morrey spaces on metric
measure spaces, Canad. Math. Bull. 57 (2014), no. 3, 598–608. <365,
366, xii>
[295] S. Lu and L. Xu, Boundedness of rough singular integral operators on
the homogeneous Morrey-Herz spaces, Hokkaido Math. J. 34 (2005), no.
2, 299–314. <83>
[296] W.A.J. Luxemburg, Banach function spaces, Assen, 1955. <79, 83>
[297] F.Y. Maeda, Y. Mizuta, T. Ohno and T. Shimomua, Duality of
non-homogeneous central Herz-Morrey-Musielak-Orlicz spaces, Poten-
tial Anal. 47 (2017), no. 4, 447–460. <84>
[298] F.Y. Maeda, Y. Mizuta, T. Ohno and T. Shimomua, Duality of cen-
tral Herz-Morrey-Musielak-Orlicz spaces of variable exponents, Collect.
Math. 70 (2019), no. 1, 117–151. <406>
[299] L. Maligranda, Orlicz spaces and interpolation, Seminários de Mateática
[Seminars in Mathematics], 5. Universidade Estadual de Campinas,
Departamento de Matemática, Campinas, 1989. iii+206 pp. <363, 84>
[300] L. Maligranda, Simple norm inequalities, Amer. Math. Monthly 113
(2006), 256–260. <107>
[301] L. Maligranda and K. Matsuoka, Calderón–Zygmund singular integrals
in central Morrey–Orlicz spaces, Tohoku Math. J. (2) 72 (2020), no. 2,
235–259. <438>
[302] Marcinkiewicz, J.: Sur l’interpolation d’operations, C. R. Acad. Sci.
Paris 208, 1272–1273 (1939) <80, 214>
[303] M. Mastylo, Y. Sawano and H. Tanaka, Morrey type space and its Köthe
dual space, Bull. Malaysian Mathematical Society 41 (2018), 1181–1198.
<71, 407>
 Bibliography 463

[304] M. Mastylo and Y. Sawano, Complex interpolation and Calderón–

Mityagin couples of Morrey spaces, Anal. PDE. 12, no. 7 (2019), 1711–
1740. <323, 325, 363, 364, 80, 81, 434>
[305] M. Mastylo and Y. Sawano, Applications of interpolation methods and
Morrey spaces to elliptic PDEs (Interpolation methods and Morrey
spaces), to appear in Ann. Scuola Norm. Sup. Pisa. <366, xii>

[306] K. Matsuoka and E. Nakai, Fractional integral operators on B p,λ with

Morrey-Campanato norms, Function spaces IX, Banach Centre Publ.,
92, 249–264, Polish Acad. Sci. Inst. Math., Warsaw (2011). <82, 437>
[307] P. Mattila, Geometry of sets and measures in Euclidean spaces. Frac-
tals and rectifiability, Cambridge Studies in Advanced Mathematics, 44,
1995. <291>
[308] G. Mauceri and S. Meda, BMO and H 1 for the Ornstein-Uhlenbeck
operator, J. Funct. Anal. 252 (2007), 278–313. <291>
[309] V.G. Maz’ya, Sobolev spaces, Springer, Berlin, 1985. <288, 85, 340>
[310] R. Megginson, An introduction to Banach space theory. Graduate Texts
in Mathematics, 183. Springer-Verlag, New York, 1998. <xiii, 88, 106,
107, 363>
[311] S. Meng and Y. Chen, Boundedness of homogeneous fractional integral
operator on Morrey space, J. Inequal. Appl. 2016, 2016:61, 11 pp. <217>
[312] N.G. Meyers, Mean oscillation over cubes and Hölder continuity, Proc.
Amer. Math. Soc. 15 1964, 717–721. <245>
[313] N.G. Meyers, A theory of capacities for potentials of functions in
Lebesgue classes, Math. Scand. 26 (1970), 255–292. <245>
[314] C.X. Miao and B.Q. Yuan, Weak Morrey spaces and strong solutions
to the Navier-Stokes equations, Sci. China Ser. A 50 (2007), no. 10,
1401–1417. <81>
[315] A. Miyachi, Remarks on Herz–Type Hardy Spaces Acta Mathe. Sinica,
(2001), 17, No.2, 339–360. <408>
[316] T. Mizuhara, Boundedness of some classical operators on generalized
Morrey spaces, Harmonic analysis (Sendai, 1990), 183–189, ICM-90
Satell. Conf. Proc., Springer, Tokyo, 1991. <19, 67, 68, 70, xi>
[317] T. Mizuhara, Commutators of singular integral operators on Morrey
spaces with general growth functions, Harmonic analysis and nonlinear
partial differential equations (Kyoto, 1998), Suurikaiseki Kôkyûroku No.
1102 (1999), 49–63. <438>
464 Bibliography

[318] T. Mizuhara, Commutators of singular integral operators on Morrey

spaces with some growth functions, Proceedings of the Second ISAAC
Congress, Vol. 1 (Fukuoka, 1999), 65–73, Int. Soc. Anal. Appl. Comput.,
7, Kluwer Acad. Publ., Dordrecht, 2000. <438>
[319] Y. Mizuta, Potential theory in Euclidean spaces, Gakkotosho, Tokyo,
1996. <2, 3, 4, 6, 7, 79, 91, 214, 217, 364, 435>
[320] Y. Mizuta, E. Nakai, T. Ohno, and T. Shimomura, An elementary proof
of Sobolev embeddings for Riesz potentials of functions in Morrey spaces
L1,ν,β (G), Hiroshima Math. J. 38 (2008), no. 3, 425–436. <69, 217>
[321] Y. Mizuta and T. Ohno, Sobolev’s theorem and duality for Herz–Morrey
spaces of variable exponent, Ann. Acad. Sci. Fenn. Math. 39 (2014), no.
1, 389–416. <83>
[322] Y. Mizuta and T. Shimomura, Weighted Morrey spaces of variable expo-
nent and Riesz potentials. Math. Nachr. 288 (2015), no. 8–9, 984–1002.
<339>
[323] C.B. Morrey, Jr., On the solutions of quasi-linear elliptic partial differ-
ential equations. Trans. Amer. Math. Soc. 43 (1938), no. 1, 126–166.
<105, xi, 80, 218, 221>
[324] B. Muckenhoupt, Weighted norm inequalities for the Hardy maximal
function. Trans. Amer. Math. Soc., 165 (1972), 207–226. <339>
[325] B. Muckenhoupt, Hardy’s inequality with weights, Studia Math. 44
(1972), 31–38. <340>
[326] B. Muckenhoupt and R. Wheeden, Weighted norm inequalities for frac-
tional integrals, Trans. Amer. Math. Soc., 192 (1974), 261–274. <288,
339, 340>
[327] E. Nakai, Hardy–Littlewood maximal operator, singular integral opera-
tors and the Riesz potentials on generalized Morrey spaces, Math. Nachr.
166 (1994), 95–103. <18, 19, 67, 68, 69, 70, 71, 72, 211, xi, 437>
[328] E. Nakai, Orlicz–Morrey spaces and the Hardy–Littlewood maximal
function, Studia Math. 188 (2008), no. 3, 193–221. <61, 68, 69, 70,
72, 101, 102, 103, 434>
[329] E. Nakai, A generalization of Hardy spaces H p by using atoms, Acta
Math. Sin. 24 (2008), no. 8, 1243–1268. <245>
[330] E. Nakai and T. Sobukawa, Buw -function spaces and their interpolation,
Tokyo Mathematical Journal, 39 (2016), No. 2, 483–517. <325, 365,
438>
[331] E. Nakai and K. Yabuta, Pointwise multipliers for functions of bounded
mean oscillation, J. Math. Soc. Japan 37 (1985), no. 2, 207–218. <67,
72, 73, 273, 287, 227>
 Bibliography 465

[332] S. Nakamura, Generalized weighted Morrey spaces and classical opera-

tors. Math. Nachr. 289 (2016), no. 17-18, 2235–2262. <193, 212, 213,
214, 216, 339>
[333] S. Nakamura and Y. Sawano, The singular integral operator and its
commutator on weighted Morrey spaces, Collect. Math. 68 (2017), no.
2, 145–174. <212, 216, 435>
[334] S. Nakamura, Y. Sawano and H. Tanaka, The fractional operators on
weighted Morrey spaces, J. Geom. Anal. 28 (2018), no. 2, 1502–1524.
<214, 215, 216, xi, 217, 218>
[335] S. Nakamura, Y. Sawano and H. Tanaka, Weighted local Morrey spaces,
Annal. Acad. Sci. Fen. Math., 45 (2020), 67–93. <211, 214, 215, 216,
xi>
[336] H. Nakano, Modulared Semi-Ordered Linear Spaces, Maruzen Co., Ltd.,
Tokyo, 1950. i+288 pp. <84, 85>
[337] H. Nakano, Topology of Linear Topological Spaces, Maruzen Co., Ltd.,
Tokyo, 1951, viii+281 pp. <84, 85>
[338] F. Nazarov, S. Treil and A. Volberg, Cauchy integral and Calderón–
Zygmund operators on nonhomogeneous spaces, Internat. Math. Res.
Notices (1997), no. 15, 703–726. <290, 291>
[339] F. Nazarov, S. Treil and A. Volberg, Weak type estimates and Cot-
lar inequalities for Calderön–Zygmund operators on nonhomogeneous
spaces, Internat. Math. Res. Notices no. 9 (1998), 463–487. <290, 291>
[340] F. Nazarov, S. Treil and A. Volberg, The T b-theorem on non-
homogeneous spaces, Acta Math. (2003) 190, 151–239. <291>
[341] T. Nogayama, Boundedness of commutators of fractional integral oper-
ators on mixed Morrey spaces, Integral Transforms Spec. Funct. 30
(2019), no. 10, 790–816. <407, 438>
[342] P. Olsen, Fractional integration, Morrey spaces and Schrödinger equa-
tion, Comm. Partial Differential Equations, 20 (1995), 2005–2055. <274,
276, 286, 288, 80, 435>
[343] M.N. Omarova, Characterizations for the commutator of parabolic non-
singular integral operator on parabolic generalized Orlicz-Morrey spaces,
Tbilisi Math. J. 13 (2020), no. 1, 97–111. <438>
[344] A. Ono, On imbedding between strong Hölder spaces and Lipschitz
spaces in Lp . Mem. Fac. Sci. Kyushu Univ. Ser. A 24 (1970), 76–93.
<227>
[345] A. Ono, On imbedding theorems in strong L(p,λ) spaces, Rend. Ist. Mat.
Univ. Triste 4 (1972), 53–65. <81>
466 Bibliography

[346] A. Ono, On isomorphism between certain strong L(p,λ) spaces and the
Lipschitz spaces and its applications, Funkcial Ekvac. 21 (1978), no. 3,
261–270. <81>
[347] A. Ono and Y. Furusho, On isomorphism and inclusion property for cer-
tain L(p,λ) spaces of strong type Ann. Mat. Pura Appl. (4) 114 (1977),
289–304. <81>
[348] W. Orlicz, Über konjugierte Exponentenfolgen. Studia Math. 3, 200–212
(1931) <84>
[349] J. Orobitg and J. Verdera, Choquet integrals, Hausdorff content and
the Hardy–Littlewood maximal operator, Bull. London Math. Soc. 30
(1998) 145–150. <408>
[350] M. Paluszynski, Characterization of the Besov spaces via the commuta-
tor operator of Coifman, Rochberg and Weiss, Indiana Univ. Math. J.
44 (1995), 1–17. <439>
[351] M. Peng and L.Z. Liu, Sharp estimates for multilinear integral operators,
Southeast Asian Bull. Math. 34 (2010), no. 1, 163–176. <438>
[352] J. Peetre, On convolution operators leaving Lp,λ spaces invariant, Ann.
Mat. Pura Appl. 72 (1966), 295–304. <218, 438>
[353] J. Peetre, On the theory of Lp,λ spaces, J. Funct. Anal. 4 (1969), 71–87.
<325, 365, 15, 80, 245, 421, 437>
[354] C. Pérez, Sharp Lp -weighted Sobolev inequalities, Ann. Inst. Fourier
(Grenoble) 45 (1995), 809–824. <218>
[355] S. Petermichl, The sharp bound for the Hilbert transform on weighted
Lebesgue spaces in terms of the classical Ap characteristic, Amer. J.
Math. 129 (2007), 1355–1375. <317, 340>
[356] S. Petermichl, The sharp weighted bound for the Riesz transforms, Proc.
Amer. Math. Soc. 136 (2008), 1237–1249. <317, 340>
[357] D. Pettis, On integration in vector spaces, Trans. Amer. Math. Soc. 44
(1938), no. 2, 277–304. <108>
[358] G. Pisier, Riesz transforms: a simpler analytic proof of P.A. Meyer’s
inequality, Lecture Notes in Math. 1321 (1988), 485–501. <291>
[359] H. Rafeiro, N. Samko and S. Samko, Morrey-Campanato spaces:
an overview. Operator theory, pseudo-differential equations, and
mathematical physics, 293–323, Oper. Theory Adv. Appl., 228,
Birkhauser/Springer Basel AG, Basel, 2013. <17, 245>
[360] M.A. Ragusa, Regularity for weak solutions to the Dirichlet problem
in Morrey space, Riv. Mat. Univ. Parma (5) 3 (1995), no. 2, 355–369.
<81>
 Bibliography 467

[361] M.A. Ragusa, Commutators of fractional integral operators on

Vanishing-Morrey spaces, J. Global Optim., 40 (1-3) (2008), 361–368.
<360>
[362] M.A. Ragusa, Operators in Morrey type spaces and applications,
Eurasian Math. J. 3 (2012), no. 3, 94–109. <434, 435, 438>
[363] M.A. Ragusa, Necessary and sufficient condition for a VMO function,
Appl. Math. Comput. 218 (2012), no. 24, 11952–11958. <244>
[364] M.A. Ragusa and P. Zamboni, Local regularity of solutions to quasilinear
elliptic equations with general structure, Commun. Appl. Anal. 3 (1999),
no. 1, 131–147. <84>
[365] M.A. Ragusa and P. Zamboni, A potential theoretic inequality,
Czechoslovak Math. J. 51, (2001), no. 1, 55–65. <141, 81>
[366] M.M. Rao and Z.D. Ren, Theory of Orlicz Spaces, Dekker 1991. <84>
[367] M. Riesz, Les fonctions conjuguées et les séries de Fourier, C. R. Acad.
Sci. Paris 178 (1924), 1464–1467. <218>
[368] M. Riesz, Sur les fonctions conjuguées, Math. Zeit. 27 (1927), 218–244.
<218>
[369] M. Rosenblum, Summability of Fourier series in Lp (µ), Trans. Amer.
Math. Soc. 105 (1962), 32–42. <339>
[370] M. Rosenthal and H.J. Schmeisser, On the boundedness of singular inte-
grals in Morrey spaces and its preduals, J. Fourier Anal. Appl. 22 (2016),
no. 2, 462–490. <437>
[371] M. Rosenthal and H. Triebel, Calderon–Zygmund operators in Morrey
spaces, Rev. Mat. Complut. 27 (2014), no. 1, 1–11. <211, 437>
[372] J. Ruan and J. Chen, The Cauchy integral operator on the predual of a
Morrey space, J. Inequal. Appl. 2010, Art. ID 729824, 14 pp. <437>
[373] M. Ruzhansky and D. Suragan, Hardy Inequalities on Homogeneous
Groups, Progress in Mathematics 327 (2019), Birkhäuser. <67, 288,
83>
[374] M. Růžička, Electrorheological fluids: modeling and mathematical the-
ory, Lecture Notes in Math. 1748, Springer, 2000. <84>
[375] D. Salim, Y. Soeharyadi, W.S. Setya Budhi, Rough fractional inte-
gral operators and beyond Adams inequalities, Math. Inequal. Appl.
22 (2019), no. 2, 747–760. <217>
[376] N. Samko, Weighted Hardy and singular operators in Morrey spaces, J.
Math. Anal. Appl. 350 (2009), no. 1, 56–72. <73, 143, 194, 211, 212,
xi>
468 Bibliography

[377] N. Samko, Weighted Hardy and potential operators in Morrey spaces,

J. Funct. Spaces Appl. 2012, Art. ID 678171, 21 pp. <143, 147, xi>

[378] Y. Sawano, Sharp estimates of the modified Hardy–Littlewood maximal

operator on the nonhomogeneous space via covering lemmas, Hokkaido
Math. J. 34 (2005), 435–458. <215, 287, 290, 291, 292>
[379] Y. Sawano, lq -valued extension of the fractional maximal operators for
non-doubling measures via potential operators, Int. J. Pure Appl. Math.
26 (2006), no. 4, 505–523. <68, 128, 143, 191, 436>
[380] Y. Sawano, A non-dense subspace in Mpq with 1 < q < p < ∞, Trans.
A. Razmadze Math. Inst. 171 (2017), no. 3, 379–380. <81>
[381] Y. Sawano, Morrey spaces from various points of view, In: Rodino L,
Toft J. (eds) Mathematical Analysis and Applications–Plenary Lectures.
ISAAC 2017, Springer Proceedings in Mathematics & Statistics, 262.
Springer, Cham, 139–160 (2017) <434>
[382] Y. Sawano, Theory of Besov spaces, Developments in Mathematics, 56.
Springer, Singapore, 2018. xxiii+945 pp. <xii, 279, 280, 289, 323, 363,
xiii, 80, 84, 143, 214, 218, 244, 339, 408>
[383] Y. Sawano, Morrey spaces from various points of view, Mathemati-
cal analysis and applications–plenary lectures, 139–160, Springer Proc.
Math. Stat., 262, Springer, Cham, 2018. <434>
[384] Y. Sawano, Homogeneous Besov spaces–survey, Kyoto J. Math. 60
(2020), no. 1, 1–43. <85>
[385] E. Sawyer, A characterization of a two-weight norm inequality for max-
imal operators, Studia Math. 75 (1982), 1–11. <341>
[386] Y. Sawano, H. Gunawan, V. Guliyev and H. Tanaka, Morrey spaces
and related function spaces [Editorial], J. Funct. Spaces 2014, Art. ID
867192, 2 pp. <xiii>
[387] Y. Sawano, D.I. Hakim and H. Gunawan, Non-smooth atomic decompo-
sition for generalized Orlicz-Morrey spaces, Math. Nachr. 288 (2015),
no. 14–15, 1741–1775. <84, 214>
[388] Y. Sawano, D.I. Hakim and D. Salim, Riesz transform and fractional
integral operators generated by nondegenerate elliptic differential oper-
ators, Adv. Oper. Theory 4 (2019), no. 4, 750–766. <408, 438>
[389] Y. Sawano and S. Shirai, Compact commutators on Morrey spaces with
non-doubling measures, Georgian Math. J., 15 (2008), no. 2, 353–376.
<68, 69, 438>
 Bibliography 469

[390] Y. Sawano, S. Sugano and H. Tanaka, Generalized fractional integral

operators and fractional maximal operators in the framework of Morrey
spaces, Trans. Amer. Math. Soc., 363 (2011), no. 12, 6481–6503. <68,
72, 103, 288, 80, 81, 423>
[391] Y. Sawano, S. Sugano and H. Tanaka, Olsen’s inequality and its applica-
tions to Schrödinger equations, Suurikaiseki Kôkyûroku Bessatsu B26,
(2011) 51–80. <288, 81, 215, 434, 435>

[392] Y. Sawano, S. Sugano and H. Tanaka, Orlicz–Morrey spaces and frac-

tional operators, Potential Anal., 36, No. 4 (2012), 517–556. <18, 68,
73, 97, 101, 102, 103, 288, 215>
[393] Y. Sawano and S.M. Tabatabaie, Speciability on Morrey spaces, to
appear in Iranian Journal of Mathematical Sciences and Informatics.
<30, 81>
[394] Y. Sawano and H. Tanaka, Morrey spaces for non-doubling measures,
Acta Math. Sinica, 21 (2005), no. 6, 1535–1544. <68, 73, 142, 143, 144,
214, xi, 217, 434>
[395] Y. Sawano and H. Tanaka, Sharp maximal inequalities and commutators
on Morrey spaces with non-doubling measures, Taiwanese J. Math. 11
(2007), no. 4, 1091–1112. <68, 213, 434, 435, 438>
[396] Y. Sawano and H. Tanaka, Predual spaces of Morrey spaces with non-
doubling measures, Tokyo J. Math. 32 (2009), 471–486. <68, 143, 193,
211, 407, 437>
[397] Y. Sawano and H. Tanaka, Fatou property of predual Morrey spaces
with non-doubling measures, Int. J. Appl. Math. 27 (2014), no. 3, 283–
296. <129, 407>
[398] Y. Sawano and H. Tanaka. The Fatou property of block spaces, J. Math.
Sci. Univ. Tokyo. 22 (2015), 663–683. <71, 81, 360, 361, 407>
[399] Y. Sawano and H. Yoshida, A predual of a predual of Bσ and its appli-
cations to commutators, Sci. China Math., 61 (2018), 1437–1472. <82>
[400] E. Sawyer and R. Wheeden, Weighted inequalities for fractional integrals
on euclidean and homogeneous spaces, Amer. J. Math, 114 (1992), 813–
874. <341>
[401] C. Segovia and J.L. Torrea, Vector-valued commutators and applica-
tions, Indiana Univ. Math. J. 38 no. 4 (1989), 959–970. <245>
[402] J. Serrin, A remark on the Morrey potential. Control methods in PDE-
dynamical systems, 307–315, Contemp. Math., 426, Amer. Math. Soc.,
Providence, RI, 2007. <435>
470 Bibliography

[403] Y. Shen and L.Z. Liu, λ-central BMO estimates for multilinear com-
mutators of Littlewood-Paley operator on central Morrey spaces, Bull.
TICMI 15 (2011), 35–46. <439>
[404] S. Spanne, Some function spaces defined using the mean oscillation over
cubes, Ann. Scuola Norm. Sup. Pisa 19 (3) (1965), 593–608. <72, 244>
(p,Φ)
[405] S. Spanne, Sur l’interpolation entres les espaces Lk , Ann. Scuola
Norm. Sup. Pisa 20 (1966), 625–648. <365, 81, 218>
[406] S. Shi and S. Lu, Some characterizations of Campanato spaces via com-
mutators on Morrey spaces, Pacific J. Math. 264 (2013), no. 1, 221–234.
<439>
[407] S.G. Shi and S.Z. Lu, A characterization of Campanato space via com-
mutator of fractional integral, J. Math. Anal. Appl. 419 (2014), no. 1,
123–137. <439>
[408] S. Shirai, Necessary and sufficient conditions for boundedness of com-
mutators of fractional integral operators on classical Morrey spaces,
Hokkaido Math. J. 35 (2006), no. 3, 683–696. <438>
[409] B. Simon, Real analysis. With a 68 page companion booklet. A Com-
prehensive Course in Analysis, Part 1. American Mathematical Society,
Providence, RI, 2015. xx+789 pp. <93, 107>
[410] B. Simon, Harmonic analysis. A Comprehensive Course in Analysis, Part
3. American Mathematical Society, Providence, RI, 2015. xviii+759 pp.
<85, 127, 214, 215, 218>
[411] S. Sobolev, On a theorem in functional analysis (Russian), Math. Sob.
46 (1938), 471–490. <85, 216>
[412] F. Soria, and G. Weiss, A remark on singular integrals and power
weights, Indiana Univ. Math. J., 43 (1994), 187–204. <339>
[413] G. Stampaccia, L(p,λ) -spaces and interpolation, Comm. Pure Appl.
Math. 17 (1964), 293–306. <365, xii, 80>
[414] E.M. Stein, Intégrales singulières et fonctions différentiable de plusieurs
variables. Lecture Notes, Faculté des Sciences d’Orsay, 1967. <141, 218>
[415] E.M. Stein, Singular Integrals and Differentiability Properties of Func-
tions, Princeton University Press, Princeton, New Jersey, 1970. <141,
289, xiii, 85, 127, 143, 217, 218, 408>
[416] E.M. Stein, Harmonic Analysis, Real-variable Methods, Orthogonality,
and Oscillatory Integrals, Princeton University Press, Princeton, NJ,
1993. <289, xiii, 85, 143, 214, 215, 217, 218, 244, 339, 436>
 Bibliography 471

[417] E.M. Stein and G. Weiss, Introduction to Fourier Analysis on Euclidean

Spaces, Princeton Mathematical Series, No. 32. Princeton University
Press, Princeton, N.J., 1971. x+297 pp. <69, 323, 83, 127, 214, 218>
[418] K. Stempak, Modified Hardy-Littlewood maximal operators on nondou-
bling metric measure spaces, Ann. Acad. Sci. Fenn. Math. 40 (2015),
no. 1, 443–448. <292>
[419] K. Stempak, Examples of metric measure spaces related to modified
Hardy-Littlewood maximal operators, Ann. Acad. Sci. Fenn. Math. 41
(2016), no. 1, 313–314. <292>
[420] J.O. Strömberg and A. Torchinsky, Weighted Hardy Spaces, Lecture
Notes in Math., 1381, Springer-Verlag, 1989. <408>
[421] S. Sugano, Estimates for the operators V α (−∆ + V )−β and V α ∇(−∆ +
V )−β with certain nonnegative potentials V , Tokyo J. Math. 21 (1998),
no. 2, 441–452. <216>
[422] G. Talenti, Osservazioni sopra una classe di disuguaglianze, (Italian.)
Rend. Sem. Mat. Fis. Milano 39 1969 171–185. <340>
[423] M.H. Taibleson and G. Weiss, The molecular characterization of Hardy
spaces. In: Harmonic analysis in Euclidean spaces (Proc. Sympos. Pure
Math., Williams Coll., Williamstown, Mass., 1978), Part 1, Proc. Sym-
pos. Pure Math., XXXV, Part, pp. 281–287. Amer. Math. Soc., Provi-
dence, R.I., 1979. <245>
[424] H. Tanaka, Morrey spaces and fractional operators, J. Aust. Math. Soc.
88 (2010), no. 2, 247–259. <214, 288, 192, 216>
[425] H. Tanaka, Two-weight norm inequalities on Morrey spaces, Ann. Acad.
Sci. Fenn. Math. 40 (2015), no. 2, 773–791. <213, xi>
[426] L. Tang, Choquet integrals, weighted Hausdorff content and maximal
operators, Georgian Math. J. 18 (2011), no. 3, 587–596. <408>
[427] L. Tang, Some characterizations for weighted Morrey-Campanato
spaces. Math. Nachr. 284 (2011), no. 8–9, 1185–1198. <210, 215, 245>
[428] L. Tang and J. Xu, Some properties of Morrey type Besov-Triebel spaces,
Math. Nachr. 278 (2005), no. 7–8, 904–917. <xii, 434>
[429] C.Q. Tang, F. Xue and Y. Zhou, Commutators of weighted Hardy opera-
tors on Herz-type spaces, Ann. Polon. Math. 101 (2011), no. 3, 267–273.
<83>
[430] J. Tao, Da. Yang and Do. Yang, Boundedness and compactness char-
acterizations of Cauchy integral commutators on Morrey spaces, Math.
Methods Appl. Sci. 42 (2019), no. 5, 1631–1651. <438>
472 Bibliography

[431] X.X. Tao and H.H. Zhang, On the boundedness of multilinear operators
on weighted Herz–Morrey spaces, Taiwanese J. Math. 15 (2011), no. 4,
1527–1543. <83>
[432] A. Torchinsky, Real variable methods in harmonic analysis. Academic
Press, 1986. <244, 339>
[433] Y. Terasawa, Outer measures and weak type (1,1) estimates of Hardy–
Littlewood maximal operators, Journal of Inequalities and Applications,
Volume 2006 (2006), Article ID 15063, 13 pages. <290, 291>
[434] X. Tolsa, Littlewood-Paley theory and the T (1) theorem with non-
doubling measures, Adv. Math. 164 (2001), no. 1, 57–116. <143, 291>
[435] X. Tolsa, BMO, H 1 , and Calderón–Zygmund operators for non doubling
measures, Math. Ann. 319 (2001), 89–149. <290, 291>
[436] X. Tolsa, Painlevé’s problem and the semiadditivity of analytic capacity,
Acta Math. 190 (2003), no. 1, 105–149. <291>
[437] G. Tomaselli, A class of inequalities, Boll. Un. Mat. Ital. (4) 2 (1969),
622–631. <340>
[438] H. Triebel, Hybrid function spaces, heat and Navier-Stokes equa-
tions, EMS Tracts in Mathematics, 24. European Mathematical Society
(EMS), Zürich, 2014. x+185 pp. <xii, 211, 80, 81, 360, 407, 437>
[439] S. Trudinger, On imbeddings into Orlicz spaces and some applications,
J. Math. Mech. 17 (1967) 473–483. <84, 217>
[440] Y. Tsutsui, Sharp maximal inequalities and bilinear estimates. Har-
monic analysis and nonlinear partial differential equations, 133–146,
Suurikaiseki Kôkyûroku Bessatsu B18, Res. Inst. Math. Sci. (RIMS),
Kyoto, 2010. <81, 435>
[441] Y. Tsutsui, Sharp maximal inequalities and its application to some bilin-
ear estimates. J. Fourier Anal. Appl. 17 (2011), no. 2, 265–289. <213,
435>
[442] N.K. Tumalun, D.I. Hakim, and H. Gunawan, Inclusion between gen-
eralized Stummel Classes and Other Function Spaces, Math. Inequal.
Appl. 23 (2020), no. 2, 547–562. <81>
[443] W. Urbina, On singular integrals with respect to the Gaussian measure,
Ann. Sc. Norm. Sup. Pisa Cl. Sci. 18 (1990), 531–567. <291>
[444] C. Vitanza, Functions with vanishing Morrey norm and elliptic par-
tial differential equations. In: Proceedings of Methods of Real Analysis
and Partial Differential Equations, Capri, pp. 147–150. Springer, 1990.
<360>
 Bibliography 473

[445] L.W. Wang and L.S. Shu, Higher order commutators of Marcinkiewicz
integral operator on Herz–Morrey spaces with variable exponent, J.
Math. Res. Appl. 34 (2014), no. 2, 175–186. <83>
[446] G.V. Welland, Weighted norm inequalities for fractional integrals, Proc.
Amer. Math. Soc. 51 (1975), 143–148. <216>
[447] N. Wiener, Generalized harmonic analysis, Acta Math., 55, 117–258
(1930) <82>
[448] N. Wiener, Tauberian theorems, Ann. Math., 33 (1932), 1–100 <82>
[449] N. Wiener, The ergodic theorem. Duke Math. J. 5 (1939), 1–18 <214>
[450] Weighted inequalities for the dyadic square function without dyadic A∞ ,
Duke Math. J. 55 (1987), 19–50. <324>
[451] J.L. Wu, Boundedness of commutators on homogeneous Morrey-Herz
spaces over locally compact Vilenkin groups, Anal. Theory Appl. 25
(2009), no. 3, 283–296. <83>
[452] J.L. Wu, Boundedness for commutators of fractional integrals on Herz–
Morrey spaces with variable exponent, Kyoto J. Math. 54 (2014), no.
3, 483–495. <83>
[453] J.L. Wu, Boundedness of some sublinear operators on Herz–Morrey
spaces with variable exponent, Georgian Math. J. 21 (2014), no. 1, 101–
111. <83>
[454] J.L. Wu, Boundedness for fractional Hardy-type operator on Herz–
Morrey spaces with variable exponent, Bull. Korean Math. Soc. 51
(2014), no. 2, 423–435. <83>
[455] J.L. Wu and Q.G. Liu, Boundedness of some singular integral operators
on Morrey-Herz spaces over locally compact Vilenkin groups, Commun.
Math. Res. 27 (2011), no. 2, 147–155. <83>
[456] C.P. Xie, Some estimates of commutators, Real Anal. Exchange 36
(2010/11), no. 2, 405–415. <439>
[457] M. Xu, The duality between Morrey spaces and its predual spaces char-
acterized by heat kernel, Anal. Theory Appl. 25 (2009), no. 1, 71–78.
<406>
[458] D. Yang, G. Hu and S. Lu, Herz type spaces and their applications,
Mathematics Monograph Series 10. <82, 291>
[459] Da. Yang, Do. Yang and G. Hu, The Hardy space H 1 with non-doubling
measures and their applications, Lecture Notes in Mathematics, 2084,
Springer, Cham, 2013. <290>
474 Bibliography

[460] D. Yang and S. Yang, Elementary characterizations of generalized

weighted Morrey-Campanato spaces, Appl. Math. J. Chinese Univ. Ser.
B 25 (2010), no. 2, 162–176. <210, 245>
[461] D. Yang and S. Yang, New characterizations of weighted Morrey-
Campanato spaces. Taiwanese J. Math. 15 (2011), no. 1, 141–163. <210,
245>
[462] D. Yang and W. Yuan, A note on dyadic Hausdorff capacities. Bull. Sci.
Math. 132, no. 6, 500–509 (2008) <408>
[463] Y. Yang and S.Q. Tao, Weak estimates of singular integrals with variable
kernel and fractional differentiation on Morrey-Herz spaces, J. Funct.
Spaces 2017, Art. ID 4340805, 10 pp. <219>
[464] X.F. Ye, Grand Morrey spaces estimates for nondivergence elliptic equa-
tions with VMO, Georgian Math. J. 20 (2013), no. 1, 195–211. <245>
[465] K. Yoshida, Functional Analysis, Classics in Mathematics, 123, Springer.
<108>
[466] W.H. Young, On classes of summable functions and their Fourier series,
Proc. Royal Soc. London 87 (1912), 225–229. <84>
[467] W. Yuan, W. Sickel, and D. Yang, Interpolation of Morrey-Campanato
and related smoothness spaces, Sci. Math. China. 58 (2015), 1835–1908.
<xii, 72, 365, 360>
[468] X. Yu and X.X. Tao, Boundedness for a class of generalized commutators
on λ-central Morrey space, Acta Math. Sin. (Engl. Ser.) 29 (2013), no.
10, 1917–1926. <439>
[469] L. Zhang, Y. Jiang, Y. Sheng and J. Zhou, Parabolic equations with
VMO coefficients in generalized Morrey spaces, Acta Math. Sin. 26
(2010), no. 1, 117–130. <438>
[470] Y. Zhang, S. Wang, D. Yang and W. Yuan, Weak Hardy-type spaces
associated with ball quasi-Banach function Spaces I: Decompositions
with applications to boundedness of Calderón–Zygmund operators,
arXiv:1905.02097. <30, 31>
[471] Y. Zhou and S. Gala, Regularity criteria in terms of the pressure for
the Navier-Stokes equations in the critical Morrey-Campanato space, Z.
Anal. Anwend. 30 (2011), no. 1, 83–93. <289, 81>
[472] C.T. Zorko, Morrey space, Proc. Amer. Math. Soc. 98 (1986), 586–592.
<67, 70, 71, 141, 80, 360, 388, 406, 407>
[473] A. Zygmund, On a theorem of Marcinkiewicz concerning interpolation
of operators. J. Math. Pures Appl. (9) 35 (1956), 223–248. <214>
Index

(µ ⊗ ν)∗ , 4 M D , 149
D(µ;Q)
Aαp,q , 311 Mlog , 262
A∞ , 306 MΦ , 147
Ap , 299 MΩ,α , 216
A1 , 296 Pk (·; x0 , ρ, f ), 239
B(r), xvii Q(j), 229
B(x, r), xvii Q(x, r), xvii
Bp , 150 S n−1 , 122, 196
Bσ (E)(Rn ), 33 U Mpq (Rn ), 343
Ej f , 200 Vvu (Rn ), 403
Gs , 188 (∗)
V0,∞ Mpq (Rn ), 344
IΩ,α , 216
α V (∗) Mpq (Rn ), 344
Kpq (Rn ), 34
0 V0 Mpq (Rn ), 344
L (µ), 2
V∞ Mpq (Rn ), 344
L0 (Rn ), 2
X(ρ), 28
LΦ (µ), 58
[a, T ], 228, 229
Lp (`q , Rn ), 144
[w]A∞ , 306
Lp (µ), 5
[w]Ap , 299
Lp (|t|α ), 294
[w]A1 , 296
Lp (|x|α ), 294
∆2 , 56
Lp (rα ), 294
Φλ,q (0, ∞), 42
Lp (tα ), 294
χ, xviii
Lp (w, `q ), 311
Ḃσ (E)(Rn ), 33
Lploc (Rn ), 7 α
K̇pq , 34
L1 (H̃0d ), 402 ` (Lp (w)), 311
q
Lp,∞ (Rn ), 135 `q (Lp , Rn ), 144
Lp,q (µ), 42 η-function, 133
M , 130 ηj,m , 133
],D
Mλ;Q 0
, 163 λf , 5
k
M , 298 ∇2 , 56
D(Q)
Mα , 262 Mpq (Rn ), 344
D(Q) ∗
Mlog , 262
Mpq (Rn ), 344
D(µ)
Mα , 264 
D(µ) Mpq (Rn ), 346
Mlog , 264
D(µ;Q)
ρ0 , 376
Mα , 262 T̃ , 205

475
476 Index

ϕ∗ , 52 LMpq (Rn ), 32
fpq (Rn ), 344
M Med(f ; Q), 164
aα (x0 ; ρ, f ), 239 WLp (µ), 8
f ⊗ g, 22 WL∞ (µ), 8
f ∗ , 39, 164 WLp (Rn ), 135
w ∈ B, 9 WLMpq , 32
w(E), 294 p.v., 197
Mp1 (Rn ), 16 5r-covering lemma, 135
M↓ (I), 38
M↑ (0, a), 50 A∞ -characteristic, 306
M↑ (I), 38 A∞ -constant, 306
Rn+1
+ , 13
A∞ -weight, 306
C γ (Rn ), 78 A1 -characteristic, 296
D(Q), 136, 151 A1 -constant, 296
D(Q0 ), 163 A1 -weight, 296
D(Rn ), 14 Ap -characteristic, 299
D1 (Q), 136 Ap -constant, 299
0
Hqp0 (Rn ), 365 Ap -theorem, 318
Ap -weight, 299
Lλ,q n
k (R ), 238 absolutely continuous norm, 350
p
Mq (Ω), 344 Adams’ theorem, 417
Mpq (Rn ), 13 admissible, 272
P(Rn ), 109 associate norm, 376
PL⊥ (Rn ), 111
Pk (Rn ), 109 ball Banach function norm, 29
Q(Q), xvii ball Banach function space, 29
Q] (Q), xvii Banach function norm, 27
X 0 (ρ), 377 Banach function space, 28
X (ρ0 ), 377 Bessel kernel, 188
X0 ∩ X1 , 89 best approximation, 118, 239
X0 + X1 , 89 Beurling algebra, 82
D+ , 168 [a, T ], 432, 433
D0 , 168
Da , 168 Calderón–Zygmund decomposition,
BCα (Rn ), 78 202
BMO(Rn ), 222 Calderon–Zygmund operator, 195
BMO+ (Rn ), 222 centered modified maximal operator,
Lip(R), 106 275
Lip(Rn ), 74 Choquet integral, 397, 398
LMpq (Rn ), 345 classical Morrey space, 13
∗
LMpq (Rn ), 345 commutator, 228, 432, 433
conjugate function, 52
LMfpq (Rn ), 345
Cube testing for Mα , 333
L logq L(µ), 58
p

Lp logq L, 58 d-dimensional Hausdorff capacity,

0
LHqp0 (Rn ), 370 389
 Index 477

decreasing rearrangement, 39 Hardy–Littlewood maximal

decreasing rearrangement, 164 inequality, 135
∆2 condition, 56 Hardy–Littlewood maximal operator,
diamond subspace, 346 130
distribution function, 5 harmonic polynomial, 120
distributional dyadic sharp maximal Herz–Morrey space, 36
operator in cubes, 163 Hörmander’s condition, 196
doubling, 274 homogeneous Hölder–Zygmund
doubling constant, 56 space, 74
doubling function, 56 homogeneous Herz space, 34
doubling weight, 303 homogeneous polynomial space, 110
dual inequality of Stein-type, 138, homogeneous smooth
139 Calderón–Zygmund kernel,
dual weight, 300 196
duality theorem, 90
dyadic average operator, 200 inner regular, 2
dyadic Morrey norm, 14
Jacobi hypergroup, 294
edge, 247 John–Nirenberg inequality, 225
α,λ
Fatou property, 27, 29 Kpq (Rn ), 36
Fefferman–Stein vector-valued Kolmogorov inequality, 137
maximal inequality, 144
5r-covering lemma, 134 L1 (Rn )-condition, 201
fractional integral operator, 131 L∞ (Rn )-condition, 201, 383
fractional integral operators with lattice property, 343
rough kernel, 216 Lebesgue differentiation theorem, 8
fractional maximal operator, 131 Lebesgue space, 5
fractional maximal operators with Legendre conjugate function, 52
rough kernel, 216 Lipschitz space, 73
local block space, 370
(generalized) Calderón–Zygmund local cube, 35
operator, 195 local distributional dyadic sharp
generalized dyadic grid, 168 maximal operator, 163
generalized Hardy operator, 332 local dyadic maximal operator, 163
genuine Calderon–Zygmund local estimate, 410
operator, 198 local Morrey norm, 32
global estimate, 410 local Morrey space, 32
Gram–Schmidt polynomial, 119 local (p, q)-block, 369
graph, 247 local/global strategy, 410
locally doubling condition, 272
Hϕα , 332 locally doubling measure metric
H1 , 283 space, 272
Hölder–Zygmund space, 73 locally reverse doubling condition,
Hardy operator, 45 272
Hardy’s inequality, 45 Lorentz space, 42
478 Index

lower half space, xviii regular (measure), 2

Lp (Rn )-inequality of the reverse Hölder inequality, 307
Hardy–Littlewood maximal Riesz potential, 131
operator, 140 Riesz transform, 196
Luxemburg–Nakano norm, 58 Rubio de Francia iteration
algorithm, 298
Marcinkiewicz space, 8
maximal, 272 self-similar increasing sequence for
maximal admissible balls, 266 Mpq (Rn ), 20
measure, 2 sharp-maximal inequality, 162
median, 164 sharp-maximal inequality for Morrey
metric measure space, 274 spaces, 414
Minkovski sum, xx σ-algebra, 2
modification rate, 255 singular integral operator of
modified dyadic Hausdorff capacity, convolution type, 196
390 singular integral operator on
Morrey norm, 13 Mpq (Rn ), 427
Morrey space, 13 size condition, 196, 386, 387
space of homogeneous type, 248
∇2 condition, 56 Spanne-type, 422
Nakano–Luxenburg norm, 64 sparse family, 169
nonhomogeneous Herz space, 34 strong doubling property, 303
nonhomogeneous polynomial space, strong subadditivity, 391
110 sunrise lemma, 151
norm condition, 386, 387 support condition, 201, 383, 386,
normalized Young function, 61 387
one-weight inequality, 294 H̃0d -quasi continuous, 408
one-weight maximal inequality, truncated maximal singular integral
294 operator, 205
one-weight norm inequality, 294 two-weight inequality, 294
openness property, 309 two-weight norm inequality, 294
outer regular, 2
uncentered modified maximal
p-convexity, 7 operator, 275
(p, q)-block, 365 universal estimate, 262, 264
Φ-average, 61 upper half space, xviii
Φ-dyadic maximal operator, 149
Φ-maximal operator, 147 variable exponent Lebesgue space,
powered Hardy–Littlewood maximal 64
operator, xix vector-valued norm, 144
vector-valued norms, 144
radial and decreasing function, vertex, 247
133
reflected singular integral operator, weak-L1 (Rn ) estimate, 203
46 weak Lp space, 135
 Index 479

weak Lp -space, 8 weighted vector-valued boundedness

weak local Morrey norm, 32 of singular integral
weak local Morrey space, 32 operators, 323
WMpq (Rn ), 25 Weighted vector-valued inequality
weak Morrey norm, 25 for M , 311
weak Morrey space, 25
weight, 294 X0 + X1 , 89
weight of E, 294
weighted measure, 293 Zorko subspace, 346