Theory of Sobolev Multipliers: With Applications to Differential and Integral Operators [1 ed.] 3540694900, 9783540694908

Citation preview

Grundlehren der mathematischen Wissenschaften A Series of Comprehensive Studies in Mathematics

Series editors M. Berger B. Eckmann P. de la Harpe F. Hirzebruch N. Hitchin L. Hörmander A. Kupiainen G. Lebeau M. Ratner D. Serre Ya. G. Sinai N.J.A. Sloane A. Vershik M. Waldschmidt Editor-in-Chief A. Chenciner J. Coates

S.R.S. Varadhan

337

, Vladimir G. Maz ya Tatyana O. Shaposhnikova

Theory of Sobolev Multipliers With Applications to Differential and Integral Operators

ABC

, Vladimir Maz ya

Tatyana Shaposhnikova

Department of Mathematical Sciences M&O Building University of Liverpool Liverpool L69 3BX UK

Department of Mathematics Linkö ping University SE-581 83 Linkö ping Sweden [email protected]

and

Department of Mathematics Linkö ping University SE-581 83 Linkö ping Sweden [email protected]

ISBN: 978-3-540-69490-8

e-ISBN: 978-3-540-69492-2

Library of Congress Control Number: 2008 932182 Mathematics Subject Classification Numbers (2000): 26D10, 46E25, 42B25 c 2009 Springer-Verlag Berlin Heidelberg
This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under the German Copyright Law. The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Cover design: WMX Design GmbH, Heidelberg Printed on acid-free paper springer.com

Contents

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1

Part I Description and Properties of Multipliers 1

2

Trace Inequalities for Functions in Sobolev Spaces . . . . . . . . . 1.1 Trace Inequalities for Functions in w1m and W1m . . . . . . . . . . . . . 1.1.1 The Case m = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.2 The Case m ≥ 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Trace Inequalities for Functions in wpm and Wpm , p > 1 . . . . . . . 1.2.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.2 The (p, m)-Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.3 Estimate for the Integral of Capacity of a Set Bounded by a Level Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.4 Estimates for Constants in Trace Inequalities . . . . . . . . . 1.2.5 Other Criteria for the Trace Inequality (1.2.29) with p > 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.6 The Fefferman and Phong Sufficient Condition . . . . . . . . 1.3 Estimate for the Lq -Norm with respect to an Arbitrary Measure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.1 The case 1 ≤ p < q . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.2 The case q < p ≤ n/m . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

7 7 7 12 14 14 16

Multipliers in Pairs of Sobolev Spaces . . . . . . . . . . . . . . . . . . . . . 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Characterization of the Space M (W1m → W1l ) . . . . . . . . . . . . . . . 2.3 Characterization of the Space M (Wpm → Wpl ) for p > 1 . . . . . . . 2.3.1 Another Characterization of the Space M (Wpm → Wpl ) for 0 < l < m, pm ≤ n, p > 1 . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Characterization of the Space M (Wpm → Wpl ) for pm > n, p > 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

33 33 35 38

19 22 25 28 29 30 30

43 47 V

VI

Contents

2.4

2.5 2.6 2.7 2.8 2.9 3

2.3.3 One-Sided Estimates for Norms of Multipliers in the Case pm ≤ n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.4 Examples of Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Space M (Wpm (Rn+ ) → Wpl (Rn+ )) . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Extension from a Half-Space . . . . . . . . . . . . . . . . . . . . . . . . 2.4.2 The Case p > 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.3 The Case p = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Space M (Wpm → Wp−k ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Space M (Wpm → Wql ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Certain Properties of Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . The Space M (wpm → wpl ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Multipliers in Spaces of Functions with Bounded Variation . . . . 2.9.1 The Spaces M bv and M BV . . . . . . . . . . . . . . . . . . . . . . . .

48 49 50 50 51 53 54 57 58 60 63 66

Multipliers in Pairs of Potential Spaces . . . . . . . . . . . . . . . . . . . . 69 3.1 Trace Inequality for Bessel and Riesz Potential Spaces . . . . . . . 69 3.1.1 Properties of Bessel Potential Spaces . . . . . . . . . . . . . . . . . 70 3.1.2 Properties of the (p, m)-Capacity . . . . . . . . . . . . . . . . . . . . 71 3.1.3 Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 3.2 Description of M (Hpm → Hpl ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 3.2.1 Auxiliary Assertions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 3.2.2 Imbedding of M (Hpm → Hpl ) into M (Hpm−l → Lp ) . . . . . 76 3.2.3 Estimates for Derivatives of a Multiplier . . . . . . . . . . . . . . 78 3.2.4 Multiplicative Inequality for the Strichartz Function . . . 79 3.2.5 Auxiliary Properties of the Bessel Kernel Gl . . . . . . . . . . 80 3.2.6 Upper Bound for the Norm of a Multiplier . . . . . . . . . . . . 81 3.2.7 Lower Bound for the Norm of a Multiplier . . . . . . . . . . . . 85 3.2.8 Description of the Space M (Hpm → Hpl ) . . . . . . . . . . . . . . 86 3.2.9 Equivalent Norm in M (Hpm → Hpl ) Involving the Norm in Lmp/(m−l) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 3.2.10 Characterization of M (Hpm → Hpl ), m > l, Involving the Norm in L1,unif . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 3.2.11 The Space M (Hpm → Hpl ) for mp > n . . . . . . . . . . . . . . . . 95 3.3 One-Sided Estimates for the Norm in M (Hpm → Hpl ) . . . . . . . . . 95 3.3.1 Lower Estimate for the Norm in M (Hpm → Hpl ) Involving Morrey Type Norms . . . . . . . . . . . . . . . . . . . . . . 96 3.3.2 Upper Estimate for the Norm in M (Hpm → Hpl ) Involving Marcinkiewicz Type Norms . . . . . . . . . . . . . . . . 96 3.3.3 Upper Estimates for the Norm in M (Hpm → Hpl ) l . . . . . . . . . . . . . . . . . . . . . . . . . . 98 Involving Norms in Hn/m 3.4 Upper Estimates for the Norm in M (Hpm → Hpl ) by Norms in Besov Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 3.4.1 Auxiliary Assertions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 µ . . . . . . . . . . . . . . . . . . . . . . . 103 3.4.2 Properties of the Space Bq,∞

Contents

3.5 3.6

3.7 3.8

4

VII

3.4.3 Estimates for the Norm in M (Hpm → Hpl ) µ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 by the Norm in Bq,∞ 3.4.4 Estimate for the Norm of a Multiplier in M Hpl (R1 ) by the q-Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 Miscellaneous Properties of Multipliers in M (Hpm → Hpl ) . . . . . 111 Spectrum of Multipliers in Hpl and Hp−l . . . . . . . . . . . . . . . . . . . . 115
3.6.1 Preliminary Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 3.6.2 Facts from Nonlinear Potential Theory . . . . . . . . . . . . . . . 117 3.6.3 Main Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 3.6.4 Proof of Theorem 3.6.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120 l The Space M (hm p → hp ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122 Positive Homogeneous Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . 125 3.8.1 The Space M (Hpm (∂B1 ) → Hpl (∂B1 )) . . . . . . . . . . . . . . . . 125 m 3.8.2 Other Normalizations of the Spaces hm p and Hp . . . . . . . 127 3.8.3 Positive Homogeneous Elements of the Spaces l m l M (hm p → hp ) and M (Hp → Hp ) . . . . . . . . . . . . . . . . . . . . 130

The Space M (Bpm → Bpl ) with p > 1 . . . . . . . . . . . . . . . . . . . . . . . 133 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 4.2 Properties of Besov Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 4.2.1 Survey of Known Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 4.2.2 Properties of the Operators Dp,l and Dp,l . . . . . . . . . . . . 136 4.2.3 Pointwise Estimate for Bessel Potentials . . . . . . . . . . . . . . 138 4.3 Proof of Theorem 4.1.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 4.3.1 Estimate for the Product of First Differences . . . . . . . . . . 141 4.3.2 Trace Inequality for Bpk , p > 1 . . . . . . . . . . . . . . . . . . . . . . 143 4.3.3 Auxiliary Assertions Concerning M (Bpm → Bpl ) . . . . . . . 145 4.3.4 Lower Estimates for the Norm in M (Bpm → Bpl ) . . . . . . . 146 4.3.5 Proof of Necessity in Theorem 4.1.1 . . . . . . . . . . . . . . . . . . 149 4.3.6 Proof of Sufficiency in Theorem 4.1.1 . . . . . . . . . . . . . . . . 155 4.3.7 The Case mp > n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164 4.3.8 Lower and Upper Estimates for the Norm in M (Bpm → Bpl ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 4.4 Sufficient Conditions for Inclusion into M (Wpm → Wpl ) with Noninteger m and l . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166 µ . . . . . . . . . . . . . . . . 166 4.4.1 Conditions Involving the Space Bq,∞ 4.4.2 Conditions Involving the Fourier Transform . . . . . . . . . . . 168 l . . . . . . . . . . . . . . . . . 170 4.4.3 Conditions Involving the Space Bq,p l 4.5 Conditions Involving the Space Hn/m . . . . . . . . . . . . . . . . . . . . . . 173 4.6 Composition Operator on M (Wpm → Wpl ) . . . . . . . . . . . . . . . . . . . 174

VIII

Contents

5

The Space M (B1m → B1l ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179 5.1 Trace Inequality for Functions in B1l (Rn ) . . . . . . . . . . . . . . . . . . . 179 5.1.1 Auxiliary Facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180 5.1.2 Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183 5.2 Properties of Functions in the Space B1k (Rn ) . . . . . . . . . . . . . . . . 185 5.2.1 Trace and Imbedding Properties . . . . . . . . . . . . . . . . . . . . . 185 5.2.2 Auxiliary Estimates for the Poisson Operator . . . . . . . . . 189 5.3 Descriptions of M (B1m → B1l ) with Integer l . . . . . . . . . . . . . . . . 193 5.3.1 A Norm in M (B1m → B1l ) . . . . . . . . . . . . . . . . . . . . . . . . . . 194 5.3.2 Description of M (B1m → B1l ) Involving D1,l . . . . . . . . . . . 199 5.3.3 M (B1m (Rn ) → B1l (Rn )) as the Space of Traces . . . . . . . . 201 5.3.4 Interpolation Inequality for Multipliers . . . . . . . . . . . . . . . 202 5.4 Description of the Space M (B1m → B1l ) with Noninteger l . . . . 203 5.5 Further Results on Multipliers in Besov and Other Function Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206 5.5.1 Peetre’s Imbedding Theorem . . . . . . . . . . . . . . . . . . . . . . . . 206 5.5.2 Related Results on Multipliers in Besov and Triebel-Lizorkin Spaces . . . . . . . . . . . . . . . . . . . . . . . . . 208 5.5.3 Multipliers in BM O . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210

6

Maximal Algebras in Spaces of Multipliers . . . . . . . . . . . . . . . . 213 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 6.2 Pointwise Interpolation Inequalities for Derivatives . . . . . . . . . . . 214 6.2.1 Inequalities Involving Derivatives of Integer Order . . . . . 214 6.2.2 Inequalities Involving Derivatives of Fractional Order . . 215 6.3 Maximal Banach Algebra in M (Wpm → Wpl ) . . . . . . . . . . . . . . . . 220 6.3.1 The Case p > 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220 6.3.2 Maximal Banach Algebra in M (W1m → W1l ) . . . . . . . . . . 224 6.4 Maximal Algebra in Spaces of Bessel Potentials . . . . . . . . . . . . . 227 6.4.1 Pointwise Inequalities Involving the Strichartz Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231 6.4.2 Banach Algebra Am,l p 6.5 Imbeddings of Maximal Algebras . . . . . . . . . . . . . . . . . . . . . . . . . . 233

7

Essential Norm and Compactness of Multipliers . . . . . . . . . . . 241 7.1 Auxiliary Assertions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243 7.2 Two-Sided Estimates for the Essential Norm. The Case m > l . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248 7.2.1 Estimates Involving Cutoff Functions . . . . . . . . . . . . . . . . 248 7.2.2 Estimate Involving Capacity (The Case mp < n, p > 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250 7.2.3 Estimates Involving Capacity (The Case mp = n, p > 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257 7.2.4 Proof of Theorem 7.0.3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261 7.2.5 Sharpening of the Lower Bound for the Essential Norm in the Case m > l, mp ≤ n, p > 1 . . . . . . . . . . . . . . 262

Contents

IX

7.2.6 Estimates of the Essential Norm for mp > n, p > 1 and for p = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263 7.2.7 One-Sided Estimates for the Essential Norm . . . . . . . . . . 266 7.2.8 The Space of Compact Multipliers . . . . . . . . . . . . . . . . . . . 267 7.3 Two-Sided Estimates for the Essential Norm in the Case m = l . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270 7.3.1 Estimate for the Maximum Modulus of a Multiplier in Wpl by its Essential Norm . . . . . . . . . . . . . . . . . . . . . . . . 270 7.3.2 Estimates for the Essential Norm Involving Cutoff Functions (The Case lp ≤ n, p > 1) . . . . . . . . . . . . . . . . . . 272 7.3.3 Estimates for the Essential Norm Involving Capacity (The Case lp ≤ n, p > 1) . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 7.3.4 Two-Sided Estimates for the Essential Norm in the Cases lp > n, p > 1, and p = 1 . . . . . . . . . . . . . . . . 278 ˚ Wpl . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281 7.3.5 Essential Norm in M 8

Traces and Extensions of Multipliers . . . . . . . . . . . . . . . . . . . . . . . 285 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285 8.2 Multipliers in Pairs of Weighted Sobolev Spaces in Rn+ . . . . . . . 285 8.3 Characterization of M (Wpt,β → Wps,α ) . . . . . . . . . . . . . . . . . . . . . . 288 8.4 Auxiliary Estimates for an Extension Operator . . . . . . . . . . . . . . 292 8.4.1 Pointwise Estimates for T γ and ∇T γ . . . . . . . . . . . . . . . . 292 8.4.2 Weighted Lp -Estimates for T γ and ∇T γ . . . . . . . . . . . . . 294 8.5 Trace Theorem for the Space M (Wpt,β → Wps,α ) . . . . . . . . . . . . . 297 8.5.1 The Case l < 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298 8.5.2 The Case l > 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301 8.5.3 Proof of Theorem 8.5.1 for l > 1 . . . . . . . . . . . . . . . . . . . . . 303 8.6 Traces of Multipliers on the Smooth Boundary of a Domain . . . 304 8.7 M Wpl (Rn ) as the Space of Traces of Multipliers in the k (Rn+m ) . . . . . . . . . . . . . . . . . . . . . . 305 Weighted Sobolev Space Wp,β 8.7.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305 8.7.2 A Property of Extension Operator . . . . . . . . . . . . . . . . . . 306 8.7.3 Trace and Extension Theorem for Multipliers . . . . . . . . . 308 . . . . . . . . . . . . 311 8.7.4 Extension of Multipliers from Rn to Rn+1 + 8.7.5 Application to the First Boundary Value Problem in a Half-Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311 8.8 Traces of Functions in M Wpl (Rn+m ) on Rn . . . . . . . . . . . . . . . . . 312 8.8.1 Auxiliary Assertions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313 8.8.2 Trace and Extension Theorem . . . . . . . . . . . . . . . . . . . . . . 315 8.9 Multipliers in the Space of Bessel Potentials as Traces of Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319 8.9.1 Bessel Potentials as Traces . . . . . . . . . . . . . . . . . . . . . . . . . 319 8.9.2 An Auxiliary Estimate for the Extension Operator T . . 320 8.9.3 M Hpl as a Space of Traces . . . . . . . . . . . . . . . . . . . . . . . . . . 322

X

9

Contents

Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325 9.1 Multipliers in a Special Lipschitz Domain . . . . . . . . . . . . . . . . . . . 326 9.1.1 Special Lipschitz Domains . . . . . . . . . . . . . . . . . . . . . . . . . . 326 9.1.2 Auxiliary Assertions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326 9.1.3 Description of the Space of Multipliers . . . . . . . . . . . . . . . 329 9.2 Extension of Multipliers to the Complement of a Special Lipschitz Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332 9.3 Multipliers in a Bounded Domain . . . . . . . . . . . . . . . . . . . . . . . . . . 336 9.3.1 Domains with Boundary in the Class C 0,1 . . . . . . . . . . . . 336 9.3.2 Auxiliary Assertions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337 9.3.3 Description of Spaces of Multipliers in a Bounded Domain with Boundary in the Class C 0,1 . . . . . . . . . . . . . 339 9.3.4 Essential Norm and Compact Multipliers in a Bounded Lipschitz Domain . . . . . . . . . . . . . . . . . . . . . 340 9.3.5 The Space M L1p (Ω) for an Arbitrary Bounded Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346 9.4 Change of Variables in Norms of Sobolev Spaces . . . . . . . . . . . . . 350 9.4.1 (p, l)-Diffeomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350 9.4.2 More on (p, l)-Diffeomorphisms . . . . . . . . . . . . . . . . . . . . . . 352 9.4.3 A Particular (p, l)-Diffeomorphism . . . . . . . . . . . . . . . . . . . 353 9.4.4 (p, l)-Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356 9.4.5 Mappings Tpm,l of One Sobolev Space into Another . . . . 357 9.5 Implicit Function Theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364 ˚ m (Ω) → W l (Ω)) . . . . . . . . . . . . . . . . . . . . . . . . . 367 9.6 The Space M (W p p 9.6.1 Auxiliary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367 ˚ m (Ω) → W l (Ω)) . . . . . . . 369 9.6.2 Description of the Space M (W p p

Part II Applications of Multipliers to Differential and Integral Operators 10 Differential Operators in Pairs of Sobolev Spaces . . . . . . . . . . 373 10.1 The Norm of a Differential Operator: Wph → Wph−k . . . . . . . . . . 373 10.1.1 Coefficients of Operators Mapping Wph into Wph−k as Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374 10.1.2 A Counterexample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 378 10.1.3 Operators with Coefficients Independent of Some Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379 10.1.4 Differential Operators on a Domain . . . . . . . . . . . . . . . . . . 382 10.2 Essential Norm of a Differential Operator . . . . . . . . . . . . . . . . . . . 384 10.3 Fredholm Property of the Schr¨ odinger Operator . . . . . . . . . . . . . 386 10.4 Domination of Differential Operators in Rn . . . . . . . . . . . . . . . . . 387

Contents

XI

11 Schr¨ odinger Operator and M (w21 → w2−1 ) . . . . . . . . . . . . . . . . . 391 11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391 odinger 11.2 Characterization of M (w21 → w2−1 ) and the Schr¨ Operator on w21 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393 11.3 A Compactness Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407 11.4 Characterization of M (W21 → W2−1 ) . . . . . . . . . . . . . . . . . . . . . . . 411 11.5 Characterization of the Space M (˚ w21 (Ω) → w2−1 (Ω)) . . . . . . . . . 416 11.6 Second-Order Differential Operators Acting from w21 to w2−1 . . 421 12 Relativistic Schr¨ odinger Operator 1/2 −1/2 and M (W2 → W2 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427 12.1 Auxiliary Assertions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427 12.1.1 Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436 12.2 Corollaries of the Form Boundedness Criterion and Related Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 441 13 Multipliers as Solutions to Elliptic Equations . . . . . . . . . . . . . . 445 13.1 The Dirichlet Problem for the Linear Second-Order Elliptic Equation in the Space of Multipliers . . . . . . . . . . . . . . . . . . . . . . . 445 13.2 Bounded Solutions of Linear Elliptic Equations as Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447 13.2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447 13.2.2 The Case β > 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448 13.2.3 The Case β = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452 1 1 (Ω) into W2,1 (Ω) . 454 13.2.4 Solutions as Multipliers from W2,w(ρ) 13.3 Solvability of Quasilinear Elliptic Equations in Spaces of Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456 13.3.1 Scalar Equations in Divergence Form . . . . . . . . . . . . . . . . 457 13.3.2 Systems in Divergence Form . . . . . . . . . . . . . . . . . . . . . . . . 458 13.3.3 Dirichlet Problem for Quasilinear Equations in Divergence Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461 13.3.4 Dirichlet Problem for Quasilinear Equations in Nondivergence Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463 13.4 Coercive Estimates for Solutions of Elliptic equations in Spaces of Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467 13.4.1 The Case of Operators in Rn . . . . . . . . . . . . . . . . . . . . . . . 467 13.4.2 Boundary Value Problem in a Half-Space . . . . . . . . . . . . . 469 13.4.3 On the L∞ -Norm in the Coercive Estimate . . . . . . . . . . . 473 13.5 Smoothness of Solutions to Higher Order Elliptic Semilinear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474 13.5.1 Composition Operator in Classes of Multipliers . . . . . . . 474 13.5.2 Improvement of Smoothness of Solutions to Elliptic Semilinear Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477

XII

Contents

14 Regularity of the Boundary in Lp -Theory of Elliptic Boundary Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479 14.1 Description of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479 14.2 Change of Variables in Differential Operators . . . . . . . . . . . . . . . 481 14.3 Fredholm Property of the Elliptic Boundary Value Problem . . . 483 l−1/p l−1/p , Wp , 14.3.1 Boundaries in the Classes Mp l−1/p and Mp (δ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483 14.3.2 A Priori Lp -Estimate for Solutions and Other Properties of the Elliptic Boundary Value Problem . . . . 484 14.4 Auxiliary Assertions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489 14.4.1 Some Properties of the Operator T . . . . . . . . . . . . . . . . . . 489 14.4.2 Properties of the Mappings λ and κ . . . . . . . . . . . . . . . . . 490 ˚ph Under a Change 14.4.3 Invariance of the Space Wpl ∩ W of Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 492 14.4.4 The Space Wp−k for a Special Lipschitz Domain . . . . . . . 496 14.4.5 Auxiliary Assertions on Differential Operators in Divergence Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 498 14.5 Solvability of the Dirichlet Problem in Wpl (Ω) . . . . . . . . . . . . . . . 502 14.5.1 Generalized Formulation of the Dirichlet Problem . . . . . 502 14.5.2 A Priori Estimate for Solutions of the Generalized Dirichlet Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 502 14.5.3 Solvability of the Generalized Dirichlet Problem . . . . . . . 503 14.5.4 The Dirichlet Problem Formulated in Terms of Traces . . 504 14.6 Necessity of Assumptions on the Domain . . . . . . . . . . . . . . . . . . . 507 3/2 14.6.1 A Domain Whose Boundary is in M2 ∩ C 1 3/2 but does not Belong to M2 (δ) . . . . . . . . . . . . . . . . . . . . . 507 14.6.2 Necessary Conditions for Solvability of the Dirichlet Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 509 l−1/p (δ) . . . . . . . . . . . . . . . . . . . 510 14.6.3 Boundaries of the Class Mp l−1/p 14.7 Local Characterization of Mp (δ) . . . . . . . . . . . . . . . . . . . . . . . 513 14.7.1 Estimates for a Cutoff Function . . . . . . . . . . . . . . . . . . . . . 513 l−1/p (δ) Involving a Cutoff Function . . 515 14.7.2 Description of Mp 14.7.3 Estimate for s1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516 14.7.4 Estimate for s2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 520 14.7.5 Estimate for s3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523 15 Multipliers in the Classical Layer Potential Theory for Lipschitz Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531 15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531 15.2 Solvability of Boundary Value Problems in Weighted Sobolev Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 537 15.2.1 (p, k, α)-Diffeomorphisms . . . . . . . . . . . . . . . . . . . . . . . . . . . 537 15.2.2 Weak Solvability of the Dirichlet Problem . . . . . . . . . . . . 539 15.2.3 Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 542

Contents

XIII

15.3 Continuity Properties of Boundary Integral Operators . . . . . . . . 547 15.4 Proof of Theorems 15.1.1 and 15.1.2 . . . . . . . . . . . . . . . . . . . . . . . 551 15.4.1 Proof of Theorem 15.1.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 551 15.4.2 Proof of Theorem 15.1.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 557 15.5 Properties of Surfaces in the Class Mp (δ) . . . . . . . . . . . . . . . . . . . 559 15.6 Sharpness of Conditions Imposed on ∂Ω . . . . . . . . . . . . . . . . . . . . 562 15.6.1 Necessity of the Inclusion ∂Ω ∈ Wp in Theorem 15.2.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 562  . . . . . . . . . . . . . . . 563 15.6.2 Sharpness of the Condition ∂Ω ∈ B∞,p  15.6.3 Sharpness of the Condition ∂Ω ∈ Mp (δ) in Theorem 15.2.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 564 15.6.4 Sharpness of the Condition ∂Ω ∈ Mp (δ) in Theorem 15.1.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 566 15.7 Extension to Boundary Integral Equations of Elasticity . . . . . . . 568 16 Applications of Multipliers to the Theory of Integral Operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 573 16.1 Convolution Operator in Weighted L2 -Spaces . . . . . . . . . . . . . . . 573 16.2 Calculus of Singular Integral Operators with Symbols in Spaces of Multipliers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 575 16.3 Continuity in Sobolev Spaces of Singular Integral Operators with Symbols Depending on x . . . . . . . . . . . . . . . . . . . . . . . . . . . . 579 16.3.1 Function Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 580 16.3.2 Description of the Space M (H m,µ → H l,µ ) . . . . . . . . . . . 582 16.3.3 Main Result . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 585 16.3.4 Corollaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 588 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 591 List of Symbols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 605 Author and Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 607

Introduction

‘I never heard of “Uglification,” Alice ventured to say. ‘What is it?’ ’ Lewis Carroll, “Alice in Wonderland”

Subject and motivation. The present book is devoted to a theory of multipliers in spaces of differentiable functions and its applications to analysis, partial differential and integral equations. By a multiplier acting from one function space S1 into another S2 , we mean a function which defines a bounded linear mapping of S1 into S2 by pointwise multiplication. Thus with any pair of spaces S1 , S2 we associate a third one, the space of multipliers M (S1 → S2 ) endowed with the norm of the operator of multiplication. In what follows, the role of the spaces S1 and S2 is played by Sobolev spaces, Bessel potential spaces, Besov spaces, and the like. The Fourier multipliers are not dealt with in this book. In order to emphasize the difference between them and the multipliers under consideration, we attach Sobolev’s name to the latter. By coining the term Sobolev multipliers we just hint at various spaces of differentiable functions of Sobolev’s type, being fully aware that Sobolev never worked on multipliers. After all, Fourier never did either. Sobolev multipliers arise in many problems of analysis and theories of partial differential and integral equations. Coefficients of differential operators can be naturally considered as multipliers. The same is true for symbols of more general pseudo-differential operators. Multipliers also appear in the theory of differentiable mappings preserving Sobolev spaces. Solutions of boundary value problems can be sought in classes of multipliers. Because of their algebraic properties, multipliers are suitable objects for generalizations of the basic facts of calculus (theorems on implicit functions, traces and extensions, point mappings and their compositions etc.) Moreover, some basic operators V.G. Maz’ya, T.O. Shaposhnikova, Theory of Sobolev Multipliers, Grundlehren der mathematischen Wissenschaften 337, c Springer-Verlag Berlin Hiedelberg 2009


1

2

Introduction

of harmonic analysis, like the classical maximal and singular integral operators, act in certain classes of multipliers. We believe that the calculus of Sobolev multipliers provides an adequate language for future work in the theory of linear and nonlinear differential and pseudodifferential equations under minimal restrictions on the coefficients, domains, and other data. Before the 1970s, the word multiplier was usually associated with the name of Fourier, and a deep theory of Lp -Fourier multipliers created by Marcinkiewicz, Mikhlin, H¨ ormander et al was quite popular. As for the multipliers preserving a space of differentiable functions, only a few isolated results were known (Devinatz and Hirschman [DH], Hirschman [Hi1], [Hi2], Strichartz [Str], Polking [Pol1], Peetre [Pe2]), while the multipliers in pairs of such spaces were not considered at all. The first (and the only one for the time being) attempt to work out a more or less comprehensive theory of multipliers acting either in one or in a pair of spaces of Sobolev type was undertaken by the authors in the late 1970s and early 1980s [Maz10], [Maz12], [MSh1]–[MSh16]. Results of that theory were collected in our monograph “Theory of Multipliers in Spaces of Differentiable Functions” (Pitman, 1985) [MSh16]. During the last two decades, we continued to work in the area, adding new results and developing further applications [Sh2]–[Sh14], [MSh17]–[MSh23]. We wish to reflect the present state of our theory in this book. An essential part of the aforementioned monograph is also included here. No results concerning multipliers in spaces of analytic functions are mentioned in what follows, in contrast to [MSh16]. To describe progress in this area achieved during the last twenty five years would require a disproportionate growth of the book. Structure of the book. The book consists of two parts. Part I is devoted to the theory of multipliers and covers the following topics: • • • • • • •

Trace inequalities Analytic characterization of multipliers Relations between spaces of Sobolev multipliers and other function spaces Maximal subalgebras of multiplier spaces Traces and extensions of multipliers Essential norm and compactness of multipliers Miscellaneous properties of multipliers (spectrum, composition and implicit function theorems, point mappings preserving Sobolev spaces, etc.)

In Part II we dwell upon several applications of this theory. Their list is as follows: • •

Continuity and compactness of differential operators in pairs of Sobolev spaces Multipliers as solutions to linear and quasilinear elliptic equations

Introduction

• • •

3

Higher regularity in the single and double layer potential theory for Lipschitz domains Regularity of the boundary in Lp -theory of elliptic boundary value problems Singular integral operators in Sobolev spaces

Each chapter starts with a short introductory outline of the included material. Readership. The volume is addressed to mathematicians working in functional analysis and in the theories of partial differential, integral, and pseudodifferential operators. Prerequisites for reading this book are undergraduate courses in these subjects. Acknowledgments. V. Maz’ya was partially supported by the National Science Foundation (Grant DMS-0500029, USA) and EPSRC (Grant EP/F005563/1, UK). T. Shaposhnikova gratefully acknowledges support from the Swedish Natural Science Research Council (VR).

1 Trace Inequalities for Functions in Sobolev Spaces

In this chapter we characterize the best constant C in the so-called trace inequality
 1/p ≤ C uWpm , u ∈ C0∞ , (1.0.1) |∇l u|p dµ with an arbitrary measure µ on the left-hand side. When the domain is not indicated in the notation of a space or a norm, then it is assumed to be Rn . Another variant of (1.0.1) will be with Wpm replaced by wpm . Here Wpk and wpk are completions of the space C0∞ with respect to the norms ∇k uLp + uLp αn 1 and ∇k uLp , ∇k = {∂ k /∂xα 1 · · · ∂xn }. Two-sided estimates for C are given in different terms for p = 1 (Sect. 1.1) and for p ∈ (1, ∞) (Sect. 1.2). The last Sect. 1.3 concerns the case of p on the left-hand side of (1.0.1) replaced by q = p. In what follows, we denote by c, c1 , c2 various positive constants which depend only on m, l, p, n and similar parameters. The values a and b are called equivalent (a ∼ b) if c1 a ≤ b ≤ c2 a. Here and henceforth Ω is an open set in Rn and C0∞ (Ω) is the space of infinitely differentiable functions with compact supports in Ω, and Br (x) = {y ∈ Rn : |y − x| < r}, Br = Br (0).

1.1 Trace Inequalities for Functions in w1m and W1m We start with the results concerning p = 1 obtained in [Maz11], see also [Maz14], Sect. 1.4. 1.1.1 The Case m = 1 The following lemma gives a representation of the n-dimensional variation of a function as an integral of the area of a level surface. V.G. Maz’ya, T.O. Shaposhnikova, Theory of Sobolev Multipliers, Grundlehren der mathematischen Wissenschaften 337, c Springer-Verlag Berlin Hiedelberg 2009


7

8

1 Trace Inequalities for Functions in Sobolev Spaces

Lemma 1.1.1. Let u ∈ C ∞ (Ω) and let Nt = {x ∈ Ω : |u(x)| ≥ t}. 



Then

∞

|∇u(x)|dx =

s(Ω ∩ ∂Nt )dt,

(1.1.1)

0

Ω

where s is the (n − 1)-dimensional area (It is well known that ∂Nt is a smooth (n − 1)-dimensional manifold for almost all t > 0, see [Mo].) For more general classes of functions, (1.1.1) was proved in [Kr] for n = 2 and in [Fe1]. We give a simple proof of (1.1.1) for u ∈ C ∞ (Ω). Proof. Let w = (w1 , · · · , wn ), wj ∈ C0∞ (Ω). Integrating by parts, we obtain     w∇u dx = u div w dx = − u div w dx − u div w dx. Ω

Ω

u>0

u0

For almost all t > 0   div w dx = − u>t

u>t

 wν ds = −

u=t

u=t

w∇u ds, |∇u|

where ν is the inward normal to {x : u(x) ≥ t}. Therefore   ∞ w∇u ds dt. u div w dx = − 0 u>0 u=t |∇u| The transformation of the integral  u div w dx uρ ∞



s({x ∈ ∂Nt : Φ(x) > ρ})dρ dt

= 

0

0

∞



dt

= 0

Φ(x)dsx . Ω∩∂Nt

 Formula (1.1.1) leads to a relation between the estimate  |u| dµ ≤ C∇uL1 , u ∈ C0∞ ,

(1.1.2)

and an isoperimetric inequality. Namely, we have the following assertion. Lemma 1.1.2. The exact constant C in (1.1.2) is equal to sup g

µ(g) , s(∂g)

(1.1.3)

where g is any open set in Rn with compact closure and smooth boundary. Proof. We have   |u| dµ = 0

∞

µ(g) µ(Nt ) dt ≤ sup s(∂g) g



∞

s(∂Nt ) dt, 0

which, together with Lemma 1.1.1, gives the upper bound for C. Let δ(x) = dist(x, g) and gt = {x : δ(x) < t}. It is well known that there exists a small > 0 such that the surface ∂gt is smooth for t ≤ . We substitute

10

1 Trace Inequalities for Functions in Sobolev Spaces

the function u (x) = α[δ(x)], where α ∈ C ∞ ([0, ∞)) and α(0) = 1, α(t) = 0 for t > , into (1.1.2). According to Corollary 1.1.1,    α (t)s(∂gt ) dt. |∇u | dx = 0

Since s(∂gt ) → s(∂g) as t → +0, it follows that  |∇u | dx → s(∂g).

(1.1.4)



Also,

|u | dµ ≥ µ(g).

(1.1.5)

Combining (1.1.4), (1.1.5) and (1.1.2), we obtain µ(g) ≤ Cs(∂g). The following more general assertion, which will be used in Sect. 5.1.1, is proved in the same way. Proposition 1.1.1. The best constant C in  |u| dµ ≤ CΦ ∇uL1 , where Φ ∈ C(Rn ) and u is an arbitrary function in C0∞ (Rn ), is equal to sup  g

µ(g)

.

Φ(x)dsx ∂g

Here g is any open set in Rn with compact closure, bounded by a smooth surface, as in Lemma 1.1.2. Further, we prove that sup g

µ(g) ∼ sup r1−n µ(Br (x)). s(∂g) x∈Rn ,r>0

(1.1.6)

With this aim in view, we present certain known auxiliary assertions. We start with the formulation of the classical Besicovitch covering theorem (see [Guz]). Lemma 1.1.3. Let E be a bounded set in Rn and let Br(x) (x) be a ball with r(x) > 0 and x ∈ E. By L we denote the totality of these balls. Then one can choose a sequence of balls {B (m) } from L such that  (m) (i) E ⊂ m B ; (ii) there exists a number N , depending only on the dimension of the space, such that every point of the space belongs to at most N balls from {B (m) }; (iii) the balls (1/3)B (m) are disjoint. We present one more well-known geometric lemma (see [Fe2]).

1.1 Trace Inequalities for Functions in w1m and W1m

11

Lemma 1.1.4. Let g be an open subset of Rn with a smooth boundary such that 2 mesn (Br ∩ g) = mesn (Br ). Then s(Br ∩ ∂g) ≥ cn rn−1 , where cn is a positive constant depending only on n. Proof. Let χ and ψ be characteristic functions of the sets g ∩ Br and Br \g. For any vector z = 0 we introduce the projection mapping pz onto a (n − 1)dimensional subspace orthogonal to z. By Fubini’s theorem, 2  (1/4) mesn (B1 )rn = mesn (g ∩ Br ) mesn (Br \g)     = χ(x)ψ(y)dxdy = χ(x)ψ(x + z)dzdx  = mesn ({x : x ∈ Br ∩ g, (x + z) ∈ Br \g})dz. |z|≤2r

Since any segment which joins x ∈ g ∩ Br with (x + z) ∈ Br \g intersects Br ∩ ∂g, the last integral does not exceed  2r mesn−1 [ pz (Br ∩ ∂g)]dz ≤ (2r)n+1 mesn (B1 ) s(Br ∩ ∂g). |z|≤2r



The result follows.

The following covering lemma is due to Gustin [Gus]. We give here the proof found by Federer [Fe2]. Lemma 1.1.5. Let g be a bounded open subset of Rn with smooth boundary. There exists a covering of g by a sequence of balls with radii ρi , i = 1, 2, · · · , such that  ρn−1 ≤ c s(∂g), (1.1.7) j j

where c is a constant which depends only on n. Proof. Each point x ∈ g is the center of a ball Br (x) for which 1 mesn (Br (x) ∩ g) = . mesn (Br (x)) 2

(1.1.8)

(This ratio is a continuous function of r equal to 1 for small r and tending to zero as r → ∞.) By Lemma 1.1.3, there exists a sequence of disjoint balls Brj (xj ) for which ∞  g⊂ B3rj (xj ). j=1

12

1 Trace Inequalities for Functions in Sobolev Spaces

From Lemma 1.1.4 and (1.1.8) we get s(Brj (xj ) ∩ ∂g) ≥ cn rjn−1 . Therefore s(∂g) ≥



s(Brj (xj ) ∩ ∂g) ≥ 31−n cn

j



(3rj )n−1 .

j



Thus, {Brj (xj )} is the required covering. Corollary 1.1.2. The best constant in (1.1.2) is equivalent to K=

sup x∈Rn ,r>0

r1−n µ(Br (x)).

Proof. By Lemma 1.1.2, it is sufficient to show that µ(g) ≤ c Ks(∂g) for any admissible set g. Let {Bρj (xj )} be a covering of g constructed in Lemma 1.1.5. It is clear that   µ(Bρj (xj )) ≤ K ρn−1 ≤ c K s(∂g). µ(g) ≤ j j

j



The proof is complete.

1.1.2 The Case m ≥ 1 Theorem 1.1.1. Let m and l be integers with m ≥ l ≥ 0. Then the best constant in  |∇l u| dµ ≤ Cuw1m , u ∈ C0∞ , (1.1.9) is equivalent to K=

sup x∈Rn ,r>0

rm−l−n µ(Br (x)).

(1.1.10)

Proof. (i) We start with the estimate C ≥ c K, setting u(ξ) = (x1 − ξ1 )l ϕ(r−1 (x − ξ)) in (1.1.9), where ϕ ∈ C0∞ (B2 ) and ϕ = 1 on B1 . Since  |∇l u|dµ ≥ l! µ(Br (x)), ∇m uL1 = c rn−m+l , it follows that C ≥ c K. (ii) Now we establish the estimate C ≤ c K. Let us start with the case l = 0. We have

1.1 Trace Inequalities for Functions in w1m and W1m

   (ξ − x)∇ξ u(ξ) dξ dµ(x) ≤ c |∇u| g dx, |u| dµ(x) = c |ξ − x|n





where

13

(1.1.11)

dµ(y) . |x − y|n−1

g(x) =

We argue by induction on m. For m = 1 the result is contained in Corollary 1.1.2. The last integral in (1.1.11) does not exceed 
 m−n−1 r g(ξ)dξ ∇uwm−1 . c sup 1

Br (x)

x∈Rn ,r>0

Clearly, 





g(ξ)dξ = Br (x)

dξ Br (x)

B2r (x)

 +

|ξ − σ|1−n dµ(σ)

 dξ

Br (x)

Rn \B2r (x)

|ξ − σ|1−n dµ(σ).

The first integral on the right-hand side is majorized by c rµ(B2r (x)) and the second one is not greater than   ∞ n 1−n n |x − σ| dµ(σ) = c (n − 1)r µ{σ : 2r ≤ |x − σ| < t}t−n dt. cr Rn \B2r (x)

2r



Thus rm−n−1

Br (x)

g(ξ) dξ ≤ c

sup x∈Rn ,r>0

rm−n µ(Br (x)).

For l ≥ 1 the result follows by induction.



Remark 1.1.1. It is clear that for m − l > n the finiteness of (1.1.10) means that µ = 0. In the case m − l = n, the value (1.1.10) is equal to µ(Rn ). We give an analogue of Theorem 1.1.1 for the space W1m . Theorem 1.1.2. Let m and l be integers, m ≥ l ≥ 0. Then the best constant in  (1.1.12) |∇l u| dµ ≤ CuW1m , u ∈ C0∞ , is equivalent to K=

sup

rm−l−n µ(Br (x)).

(1.1.13)

x∈Rn ,r∈(0,1]

Proof. The estimate C ≥ c K is obtained in precisely the same way as the analogous one in Theorem 1.1.1. To prove the converse inequality, we introduce a partition of unity {ϕj }i≥1 subordinate to the covering of Rn by unit balls

14

1 Trace Inequalities for Functions in Sobolev Spaces

with centers in nodes of a sufficiently small coordinate grid. We apply Theorem 1.1.1 to the integral  |∇l (ϕj u)|dµj , where µj is the restriction of µ to the support of the function ϕj . Then    ϕj uw1m |∇l (ϕj u)|dµj ≤ c K |∇l u|dµ ≤ j

j

≤ c KuW1m .  Remark 1.1.2. Obviously, in the case m−l ≥ n the value K defined in (1.1.13) is equal to sup µ(B1 (x)). x∈Rn

1.2 Trace Inequalities for Functions in wpm and Wpm, p>1 1.2.1 Preliminaries In this and subsequent chapters we often use operators of the form k(D) = F −1 k(ξ)F, where F is the Fourier transform in Rn and k is a function or a vectorvalued function which is called the symbol. In particular, D = −i∇ and αn 1 Dα = (−i)|α| ∂xα 1 · · · ∂xn . The following assertion is a variant of the Mikhlin theorem on Fourier multipliers (see [Liz]). Lemma 1.2.1. Let the function k and its derivatives ∂ m k/∂ξj1 · · · ∂ξjm , where 0 ≤ j1 + · · · + jm = m ≤ n and j1 , · · · , jm are distinct, be continuous on the set {ξ ∈ Rn : ξ1 · · · ξn = 0} and let ∂mk ξj · · · ξj ≤ const. (1.2.1) m 1 ∂ξj1 · · · ∂ξjm Then the operator k(D) is continuous in Lp , p ∈ (1, ∞). In particular, the singular integral operator with a symbol k ∈ C n (Rn \0) is continuous in Lp . In what follows, the operators (−∆)r/2 and (1 − ∆)s/2 with the symbols |ξ|r and (1 + |ξ|2 )s/2 , r > −n, s ∈ R1 , play an important role.

1.2 Trace Inequalities for Functions in wpm and Wpm , p > 1

15

Lemma 1.2.2. Let 1 = 1, 2, · · · , p ∈ (1, ∞). There exists a constant c > 1, depending only on n, p, l, such that c−1 (−∆)l/2 uLp ≤ ∇l uLp ≤ c (−∆)l/2 uLp

(1.2.2)

for all u ∈ C0∞ . Proof. Let α be a multi-index of order l. Then F −1 ξ α F u = F −1 ξ α |ξ|−l |ξ|l F u. The function ξ α |ξ|−l satisfies the condition of Lemma 1.2.1 which implies the right inequality in (1.2.2). Also,   cα ξ α ξ α |ξ|−l , |ξ|l = |ξ|2l |ξ|−l = |α|=l

where cα = l!/α!. Therefore, F −1 |ξ|l F u =

 |α|=l

cα F −1

ξα α ξ F u. |ξ|l

Applying Lemma 1.2.1 once again, we obtain the left part of (1.2.2).



The following assertion has a similar proof. Lemma 1.2.3. Let l = 1, 2, · · · , p ∈ (1, ∞). There exists a constant c > 1, depending only on n, p, l, such that c−1 (1 − ∆)l/2 uLp ≤ uWpl ≤ c (1 − ∆)l/2 uLp

(1.2.3)

for all u ∈ C0∞ . The operator Il := (−∆)−l/2 is the integral convolution operator with the kernel c |x|l−n , c = const., for l ∈ (0, n). It is usually called the Riesz potential of order l. For l > 0 the operator (1 − ∆)−l/2 has the representation (1 − ∆)−l/2 f = Gl ∗ f, where Gl is the function with the Fourier transform (1 + |ξ|2 )−l/2 . The function Gl can be written in the form  ∞ 2 e−t−x /4t t−n/2+l/2−1 dt Gl (x) = c 0

or in the form Gl (x) = c K(n−l)/2 (|x|)|x|(l−n)/2 , where Kγ is the modified Bessel function of the third kind.

16

1 Trace Inequalities for Functions in Sobolev Spaces

The function Gl is positive and decreases with the growth of |x|. It satisfies the following asymptotic estimates. For |x| → 0, ⎧ l−n ⎨ |x| , 0 < l < n, (1.2.4) Gl (x) ∼ log |x|−1 , l = n, ⎩ 1, l > n. For |x| → ∞ the following relation holds: Gl (x) ∼ |x|(l−n−1)/2 e−|x| .

(1.2.5)

The integral operator J

F →l Gl ∗ f is called the Bessel potential of order l. For properties of Riesz and Bessel potentials see [AMS], [St2], [Str]. We introduce the maximal Hardy–Littlewood operator M defined by  1 |f (y)|dy. (Mf )(x) = sup r>0 mesn Br Br (x) By the Hardy–Littlewood theorem (see [St2]), the operator M is bounded in Lp , p ∈ (1, ∞). 1.2.2 The (p, m)-Capacity We define the (p, m)-capacity of a compact set e ⊂ Rn by Cp,m (e) = inf{f pLp : f ∈ Lp , f ≥ 0, Jm f ≥ 1 on e},

(1.2.6)

where Jm is the Bessel potential of order m. This capacity satisfies Cp,m (e) = inf{(1 − ∆)m/2 upLp : u ∈ C0∞ , u ≥ 1 on e}

(1.2.7)

(see Meyers [Me]). In view of the boundedness of the singular integral operator in Lp , (1.2.8) Cp,m (e) ∼ inf{upWpm : u ∈ C0∞ , u ≥ 1 on e}. Replacing the Bessel potential Jm in (1.2.6) by the Riesz potential Im , we obtain the definition of the capacity cp,m (e). These and analogous set functions used in this book have been the subject of active study (see Maz’ya [Maz7], Maz’ya, Havin [MH1], [MH2], Meyers [Me], Hedberg [Hed2], Adams, Meyers [AM], Sj¨ odin [Sj], Adams, Hedberg [AH]). We describe certain simple properties of the capacities Cp,m (e) and cp,m (e) which will be used in this chapter. Proposition 1.2.1. The capacities Cp,m (e) and cp,m (e) are non-decreasing functions of the set e. The proof is obvious.

1.2 Trace Inequalities for Functions in wpm and Wpm , p > 1

17

Proposition 1.2.2. If mp > n, then Cp,m (e) ∼ 1 for all compact sets e = ∅ with diameter less than one. Proof. Obviously, Cp,m (e) ≤ Cp,m (B1 ). On the other hand by Sobolev’s theorem on the imbedding Wpm ⊂ L∞ , we have c uWpm ≥ uL∞ ≥ 1 for any function u ∈ C0∞ which exceeds one on e. Consequently, Cp,m (e) ≥ c−p .  Proposition 1.2.3. If mp < n, p ∈ (1, ∞), then cp,m (e) ≥ c (mesn e)(n−mp)/n .

(1.2.9)

The proof follows from the definition of the capacity and from Sobolev’s  inequality uLq ≤ c uwpm , where q = pn/(n − mp), u ∈ C0∞ . To prove an estimate similar to (1.2.9) in the case mp = n we need the following known assertion (see Yudoviˇc [Yu], Pohozhaev [Poh1], Trudinger [Tru]) which is given here with the proof for the reader’s convenience. Lemma 1.2.4. If mp = n and p ∈ (1, ∞), then ⎞ ⎛  p
|u| ⎠ dx ≤ 1 Φ ⎝c
upWpm

(1.2.10)

for all u ∈ C0∞ , where c is a constant independent of u, p + p = pp , and Φ(t) = et −

[p] 

tj /j!.

j=0

Proof. Let u = Jm f = Gm ∗ f . By Lemma 1.2.3, it suffices to give the proof under the assumption f Lp = 1. Obviously, 




Φ(c |u|p )dx =

∞ 
cj upLpj
j . j!

(1.2.11)

j=[p]+1

By Young’s inequality for q ≥ p, uLq ≤ Gm Ls f Lp , s =

qp , p + p = pp . q + p

(1.2.12)

18

1 Trace Inequalities for Functions in Sobolev Spaces

Using the estimates (1.2.4) and (1.2.5) for the function Gm , one can show that Gm sLs ≤ c0 q,

(1.2.13)

where c0 = c0 (p, n). From (1.2.12), (1.2.13), where q = p j and s = p j/(j +1), it follows that the right-hand side of (1.2.11) does not exceed ∞ 

cj (c0 p j)j+1 /j!.

(1.2.14)

j=[p]+1

This series converges if c c0 p e < 1. Diminishing c, one can make the sum (1.2.14) arbitrarily small.  Proposition 1.2.4. If mp = n, p ∈ (1, ∞) and d(e) ≤ 1, then  1−p 2n . Cp,m (e) ≥ c log mesn e

(1.2.15)

Proof. Let u ∈ C0∞ , u ≥ 1 on e. It follows from (1.2.10) that 
 Φ c u−p Wpm mesn e ≤ 1. Hence

Φ(c [Cp,m (e)]1/(1−p) ) ≤ (mesn e)−1 .

(1.2.16)

Since the argument of the function Φ in (1.2.16) is bounded away from zero, we have exp(c [Cp,m (e)]1/(1−p) ) ≤ c0 (mesn e)−1 .  Proposition 1.2.5. If mp < n, then cp,m (Br ) = c rn−mp . Proof. Using the dilation, we obtain cp,m (Br ) = rn−mp cp,m (B1 ).  Proposition 1.2.6. If mp < n and 0 < r ≤ 1, then Cp,m (Br ) ∼ rn−mp . Proof. The lower bound for the capacity follows from (1.2.9). The upper one is obtained after the substitution of the function x → η(x/r) into the norm  uWpm , where η ∈ C0∞ and η = 1 on the ball B1 .

1.2 Trace Inequalities for Functions in wpm and Wpm , p > 1

19

Proposition 1.2.7. If mp = n, p ∈ (1, ∞), and 0 < r ≤ 1, then Cp,m (Br ) ∼ (log 2/r)1−p . Proof. The lower bound for the capacity follows from (1.2.15). Let us justify the upper bound. We introduce the function v(x) = (log 2/r)−1 log 2/|x| and by α denote a function in the space C ∞ (R1 ) such that α(t) = 0 for t < 0, α(t) = 1 for t > 1. Further, let u(x) = α[v(x)]. Clearly, u ∈ C0∞ (B2 ) and u = 1 on Br . Moreover, one can check that |∇m u(x)| ≤ c (log 2/r)−1 |x|−m on B2 \Br . This implies that Cp,m (Br ) ≤ c ∇m u; B2 pLp  ≤ c (log 2/r)−p

B2 \Br

|x|−mp dx = c (log 2/r)1−p .

1.2.3 Estimate for the Integral of Capacity of a Set Bounded by a Level Surface The following assertion is proved in [Hed3]. Lemma 1.2.5. Let 0 < θ < 1, 0 < r < n and let Ir f be the Riesz potential of order r with a nonnegative density f . Then 1−θ  θ  (Irθ f )(x) ≤ c (Ir f )(x) (Mf )(x) ,

(1.2.17)

where M is the Hardy–Littlewood maximal operator. Proof. Let t be an arbitrary positive number to be chosen later. We use the equality   t f (y)dy ds = (n − rθ) f (y) dy n−rθ+1 n−rθ |x − y| s 0 Bt (x) Bs (x)  +trθ−n f (y) dy (1.2.18) Bt (x)

which is checked by changing the order of integration on the right-hand side. Hence  f (y)dy ≤ c trθ (Mf )(x). (1.2.19) n−rθ |x − y| Bt (x)

20

1 Trace Inequalities for Functions in Sobolev Spaces

Clearly we have  Rn \B

t (x)

f (y)dy ≤ tr(θ−1) |x − y|n−rθ

 Rn \B

t (x)

f (y)dy |x − y|n−r

≤ tr(θ−1) (Ir f )(x).

(1.2.20)

Adding this inequality to (1.2.19), we obtain   Irθ f (x) ≤ c trθ (Mf )(x) + tr(θ−1) (Ir f )(x). 

Minimization of the right-hand side in t completes the proof.

Corollary 1.2.1. Let m be an integer, 0 < m < n, Im f = |x|m−n ∗ f with f ≥ 0 and let F be a function in C m (0, ∞) such that tk−1 |F (k) (t)| ≤ Q,

k = 0, 1, · · · , m,

Q = const.

Then |∇m F (Im f )| ≤ c Q(Mf + |∇m Im f |)

(1.2.21)

almost everywhere in R . n

Proof. Let u = Im f . One can verify by induction that |∇m F (u)| ≤ c

m 



|F (k) (u)|

|∇j1 u| · · · |∇jk u|.

(1.2.22)

j1 +···+jk =m

k=1

Consequently, |∇m F (u)| ≤ c Q

m 



k=1 j1 +···+jk

|∇j1 u| |∇j u| · · · 1−jk /m . 1−j1 /m k u u =m

(1.2.23)

Since |∇s u| ≤ c Im−s f , it follows from (1.2.23) that m
 |∇m F (u)| ≤ c Q |∇m Im f | +





k=1 j1 +···+jk

 Im−j1 f · · · Im−jk f , (Im f )1−j1 /m · · · (Im f )1−jk /m =m



where the sum is taken over the collections of numbers j1 , · · · , jk each less than m. Applying Lemma 1.2.5, we complete the proof.  Our aim is the following assertion. Theorem 1.2.1. Let p ∈ (1, ∞), m = 1, 2, · · · and mp < n. Then, for any function u ∈ C0∞ ,  ∞ cp,m (Nt ) tp−1 dt ≤ c upwpm , (1.2.24) 0

where Nt = {x : |u(x)| ≥ t} and c is a constant depending only on n, p, m.

1.2 Trace Inequalities for Functions in wpm and Wpm , p > 1

21

Proof. Let u = Im f and v = Im |f |. It is easily seen that v ∈ C m (Rn ) and v(x) = O(|x|m−n ) as |x| → ∞. Thus the set {x : v(x) ≥ t} is compact for any t > 0. Putting tj = 2j , j = 0, ±1, · · · , and using the inequality v(x) ≥ |u(x)|, we obtain  ∞ cp,m (Nt ) tp−1 dt

0

≤c

+∞ 

(tj+1 − tj )p cp,m ({x : v(x) ≥ tj }).

(1.2.25)

j=−∞

Let γ ∈ C ∞ (R1 ), γ(τ ) = 0 for τ < , γ(τ ) = 1 for τ > 1, where > 0. We introduce the function v → F ∈ C ∞ (0, ∞) equal to Fj (v) = tj + (tj+1 − tj )γ((v − tj )(tj+1 − tj )−1 ) on the segment [tj , tj+1 ]. According to the definition of capacity cp,m , the sum on the right-hand side of (1.2.25) does not exceed ∞ 

Fj (v)pwpm = F (v)pwpm .

j=−∞

By Corollary 1.2.1, the last norm is majorized by c (M|f |Lp + ∇m Im |f |Lp ).

(1.2.26)

Since the operator M and the singular integral operator ∇m Im are continuous in Lp , the sum (1.2.26) does not exceed c f Lp . By Lemma 1.2.2, f Lp ∼ uwpm . 

The theorem is proved. Together with (1.2.24), in this chapter we use the inequality  ∞ Cp,m (Nt ) tp−1 dt ≤ c upWpm ,

(1.2.27)

0

where p ∈ (1, ∞) and m = 1, 2, · · · . The proof of (1.2.27) is similar to that of (1.2.24), the role of (1.2.17) being played by 1−θ  θ  (Jrθ f )(x) ≤ c (Jr f )(x) (Mf )(x) ,

(1.2.28)

where 0 < θ < 1, r > 0 and Jr f is the Bessel potential of order r with a nonnegative density f . We do not dwell on a similar though more cumbersome proof of (1.2.28). A more general inequality will be proved in Lemma 4.2.3.

22

1 Trace Inequalities for Functions in Sobolev Spaces

The corollary and its proof remain valid if Im is replaced by Jm . To obtain (1.2.27), it is sufficient to use the following chain of inequalities: 

∞

Cp,m (Nt ) tp−1 dt ≤ c

0

∞ 

(tj+1 − tj )p Cp,m ({x : |v(x)| ≥ tj })

j=−∞ ∞ 

≤c

  Fj (v) − tj pWpm ≤ c ∇m (F (v))pLp + vpLp

j=−∞

and to duplicate the end of the proof of Theorem 1.2.1. Remark 1.2.1. The existence of inequalities of the type (1.2.24) was demonstrated in [Maz3], where (1.2.24) (and for m = 1 even a stronger inequality, in which the capacity of the condenser Nt \N2t plays the role of the capacity of the set Nt ) was obtained only for m = 1 and m = 2. In the more difficult case m = 2 the proof was based on the ‘smooth truncation’ of a potential near equipotential surfaces. Unifying this procedure with Hedberg’s inequality (1.2.17), Adams [Ad3] obtained the above proof for all integers m. Further references can be found in [Maz18]. 1.2.4 Estimates for Constants in Trace Inequalities A simple though important corollary of inequalities (1.2.27) and (1.2.24) is: Theorem 1.2.2. Let p ∈ (1, ∞), m = 1, 2, · · · and let µ be a measure in Rn . (i) The best constant C in  |u|p dµ ≤ C upWpm , u ∈ C0∞ , (1.2.29) is equivalent to sup e

µ(e) , Cp,m (e)

(1.2.30)

where e is an arbitrary compact set in Rn of positive capacity Cp,m (e). (Expressions similar to (1.2.30) often occur in this book. In what follows we do not mention the positivity of capacities in denominators.) (ii) If mp < n, then the best constant C in  |u|p dµ ≤ C upwpm , u ∈ C0∞ , (1.2.31) is equivalent to sup e

µ(e) . cp,m (e)

(1.2.32)

1.2 Trace Inequalities for Functions in wpm and Wpm , p > 1

23

Proof. (i) From the definition of Lebesgue integral we obtain  ∞  µ(Nt ) d(tp ). |u|p dµ = 0

Therefore



µ(e) |u| dµ ≤ sup C e p,m (e)



p

∞

Cp,m (Nt ) d(tp ).

0

Now (1.2.29) follows from (1.2.27). Minimizing the right-hand side of (1.2.29) over the set {u ∈ C0∞ : u ≥ 1 on e}, we get C ≥ sup e

µ(e) . Cp,m (e)

Case (ii) is treated in the same way. Lemma 1.2.6. The best constants C0 and C in the inequalities  |∇l u|p dµ ≤ C0 upwpm ,

(1.2.33)

 |u|p dµ ≤ Cupwm−l ,

(1.2.34)

p

where m > l and u ∈ C0∞ , are equivalent. Proof. The estimate C0 ≤ c C is obvious. We show that C0 ≥ c C. It is clear that  (l!/α!)D2α (−∆)−l u. u= |α|=l

From (1.2.33) and Lemma 1.2.2 we get  |D2α (−∆)−l u|p dµ ≤ C0 Dα (−∆)−l upwpm ≤ c C0 upwm−l . p



Hence (1.2.34) holds with C ≤ c C0 . Lemma 1.2.7. The best constants C0 and C in the inequalities  (|∇l u|p + |u|p )dµ ≤ C0 upWpm ,

(1.2.35)

 |u|p dµ ≤ CupW m−l , p

where m > l and u ∈ C0∞ , are equivalent.

(1.2.36)

24

1 Trace Inequalities for Functions in Sobolev Spaces

Proof. The estimate C0 ≤ c C is obvious. We prove the converse. Let x → σ be a smooth positive function on [0, ∞), equal to x for x > 1. For any u ∈ C0∞ we have the representation u = (−∆)l [σ(−∆)]−l u + T (−∆), where T is a function from C0∞ ([0, ∞)). Since  (−∆)l = (−1)l (l!/α!)D2α , |α|=l

it follows from (1.2.35) and Lemma 1.2.3 that    |u|p dµ ≤ c C0 ∇l [σ(−∆)]−l upWpm + T upWpm ≤ c1 C0 upW m−l . p



The proof is complete.

We give one more expression equivalent to the best constant C in (1.2.29). Corollary 1.2.2. The exact constant C in (1.2.29) is equivalent to µ(e) , C {e:d(e)≤1} p,m (e) sup

where d(e) is the diameter of e. Proof. The lower bound for C follows from Theorem 1.2.2. We prove the upper bound. Let κ be an arbitrary compact set in Rn . Further, let closed cubes Qj form the coordinate grid with step n−1/2 and let 2Qj be homothetic open cubes with double edge length. By u we denote a function in C0∞ such that u ≥ 1 on κ. Let ηj be a function in C0∞ (2Qj ), equal to one on Qj . Since the multiplicity of intersection of 2Qj is finite and depends only on n, we have   Cp,m (κ ∩ Qj ) ≤ c1 ηj u; 2Qj pWpm j

j

≤ c2



u; 2Qj pWpm ≤ c3 upWpm .

j

Minimizing the last norm, we get Cp,m (κ) ≥ c



Cp,m (κ ∩ Qj ).

j

Clearly, µ(κ ∩ Qj ) ≤

µ(e) Cp,m (κ ∩ Qj ). {e:d(e)≤1} Cp,m (e) sup

(1.2.37)

1.2 Trace Inequalities for Functions in wpm and Wpm , p > 1

25

Summing over j and using (1.2.37), we arrive at the inequality µ(κ) ≤ c

µ(e) Cp,m (κ). C (e) p,m {e:d(e)≤1} sup



The result follows.

Corollary 1.2.3. If mp > n, then the best constant in (1.2.29) is equivalent to sup µ(B1 (x)).

x∈Rn

Proof. By Proposition 1.2.2, Cp,m (e) ∼ 1 for any non-empty compact set e with d(e) ≤ 1. It remains to refer to Corollary 1.2.2.  Using the estimates for capacity by the Lebesgue measure obtained in Propositions 1.2.3 and 1.2.4, we immediately obtain: Proposition 1.2.8. The following inequalities hold ⎧ µ(e) ⎪ ⎪ c sup ⎪ ⎨ (n−pm)/n µ(e) {e:d(e)≤1} (mesn e)  p−1 ≤ sup 4n ⎪ {e:d(e)≤1} Cp,m (e) ⎪ ⎪ c sup µ(e) log ⎩ mesn e {e:d(e)≤1} sup e

µ(e) µ(e) ≤ c sup (n−pm)/n cp,m (e) e (mesn e)

for mp < n, for mp = n;

for mp < n.

A direct corollary of Propositions 1.2.5 - 1.2.7 is Proposition 1.2.9. The following inequalities hold ⎧ ρmp−n µ(Bρ (x)) ⎪ ⎪ c nsup ⎨ x∈R ,ρ∈(0,1) µ(e)  p−1 sup ≥ 2 ⎪ {e:d(e)≤1} Cp,m (e) ⎪ µ(Bρ (x)) log ⎩ c nsup ρ x∈R ,ρ∈(0,1) sup e

µ(e) ≥c cp,m (e)

sup x∈Rn ,ρ>0

ρmp−n µ(Bρ (x))

for mp < n, for mp = n;

for mp < n.

1.2.5 Other Criteria for the Trace Inequality (1.2.29) with p > 1 Now we overview several other conditions which are necessary and sufficient for (1.2.29).

26

1 Trace Inequalities for Functions in Sobolev Spaces

We start with a remark due to D. R. Adams [Ad3] stating that ⎡ ⎤p−1 p
⎢ (Jm µe (x)) dx ⎥ µ(e) ⎥ ∼ sup ⎢ sup ⎣ ⎦ µ(e) e Cp,m (e) e

(1.2.38)

where the suprema are taken either over arbitrary compact sets e ⊂ Rn or over compact sets whose diameters do not exceed one, and µe stands for the restriction of µ to e. In fact, let the left and right-hand sides of (1.2.38) be denoted by A and B, respectively. Further, let u be an arbitrary function in C0∞ with u ≥ 1 on e. We have  µ(e) ≤ u(x) dµe (x) ≤ (1 − ∆)−m/2 µe Lp
(1 − ∆)m/2 uLp which can be rewritten as µ(e) ≤ c Jm µe Lp
uWpm . Minimizing the right-hand side over all functions u, we obtain


µ(e) ≤ cB 1/p µ(e)1/p [Cp,m (e)]1/p , i.e. A ≤ cB. Now we check the converse estimate. According to part (i) of Theorem 1.2.2,  |u|p dµ ≤ c AupWpm for all u ∈ C0∞ . Consequently,  p u dµe ≤ c A µ(e)p−1 (1 − ∆)m/2 up , Lp and therefore




Jm µe Lp
≤ c A1/p µ(e)1/p . Thus B ≤ c A.



The relation ⎡ sup e

⎢ µ(e) ∼ sup ⎢ ⎣ cp,m (e) e

p


⎤p−1

(Im µe (x)) dx ⎥ ⎥ ⎦ µ(e)

,

(1.2.39)

where e is an arbitrary compact set in Rn , mp < n, p ∈ (1, ∞), can be established in precisely the same manner. 

1.2 Trace Inequalities for Functions in wpm and Wpm , p > 1

27

By (1.2.38), the upper estimate sup e

µ(e) ≤ c sup Jm (Jm µ)1/(p−1) Cp,m (e) Rn

(1.2.40)

holds if mp ≤ n. In particular, for p = 2 it becomes sup e

µ(e) ≤ c sup J2m µ. C2,m (e) Rn

(1.2.41)

Upper estimates similar to (1.2.40) and (1.2.41), with cp,m and I instead  of Cp,m and J, stem from (1.2.39) in the case mp < n. One cannot allow arbitrary sets e in the capacitary upper bounds on the left-hand sides of (1.2.38) and (1.2.39). This is in contrast with the suprema on the right-hand sides of (1.2.38) and (1.2.39). In fact, it turned out, as shown by Kerman and Sawyer in [KeS], that the role of e on the right-hand side of (1.2.39) can be played by an arbitrary cube Q, i.e.   (I µ (x))p
dx p−1 m Q µ(e) ∼ sup , (1.2.42) sup µ(Q) e cp,m (e) Q With minor technical changes in the proof given in [KeS], one verifies that the set {e} on the right-hand side of (1.2.38) can be reduced to a set of cubes Q, i.e.   (J µ (x))p
dx p−1 m Q µ(e) ∼ sup , (1.2.43) sup µ(Q) e Cp,m (e) {Q:d(Q)≤1} where d(Q) is the diameter of Q. Other conditions, necessary and sufficient for (1.2.29) and (1.2.31), which do not involve arbitrary sets and even cubes and which are of a purely pointwise nature, were found in [MV1]. It is shown in [MV1] that
 J J µp (x) p−1 µ(e) m m ∼ sup , sup Jm µ(x) e Cp,m (e) x∈Rn

(1.2.44)

where mp ≤ n, and
 I I µp (x) p−1 µ(e) m m sup ∼ sup , Im µ(x) e cp,m (e) x∈Rn

where mp < n.

(1.2.45) 

28

1 Trace Inequalities for Functions in Sobolev Spaces

We note that the following three criteria for (1.2.31) which result from (1.2.32), (1.2.42), and (1.2.45), respectively, µ(e) ≤ C cp,m (e), 

 p
Im µQ (x) dx ≤ C µ(Q),


Im (Im µ)p (x) ≤ C Im µ(x)

(1.2.46) (1.2.47) (1.2.48)

have been obtained independently of each other. These criteria lead to different simpler conditions, either necessary or sufficient, for (1.2.31), and each criterion (1.2.46)–(1.2.48) has its own range of applications. For example, the sufficient condition µ(e) ≤ C(mesn e)(n−pm)/n follows readily from (1.2.46) (see Proposition 1.2.8) but its direct derivation from either (1.2.47) or (1.2.48) has not been obtained so far. We finish this subsection with one more condition, necessary and sufficient for the trace inequality (1.2.31) to hold, which was obtained by Verbitsky [Ver1]: for every dyadic cube P0 in Rn  P ⊆P0

p µ(P ) mesn P ≤ C µ(P0 ), (mesn P )1−m/n


where the sum is taken over all dyadic cubes P contained in P0 and the constant does not depend on P0 . Adding the restriction that the side length of P0 does not exceed 1, we have a necessary and sufficient condition for the trace inequality (1.2.29). 1.2.6 The Fefferman and Phong Sufficient Condition It was shown by Fefferman and Phong [F2] that the trace inequality (1.2.31) is true for p > 1, mp < n and for the measure µ, absolutely continuous with respect to the Lebesgue measure: dµ(x) = g(x) dx,

(1.2.49)

if there exists t > 1 such that  Br (x)

[g(y)]t dy ≤ c rn−mpt .

In order to prove this we make use of the inequality    p ν(Br (x)) (Mmt f )(y) dν(y) ≤ c sup n−mpt |f (y)|p dy r x∈Rn , Rn Rn r>0

(1.2.50)

(1.2.51)

1.3 Estimate for the Lq -Norm with respect to an Arbitrary Measure

29

(see [SW]), where Ml f is the fractional maximal function defined by    |f (y)|dy. Ml f (x) = sup rl−n r>0

Obviously, for any δ > 0



∞

Im f (x) = (n − m)

Br (x)

 rm−n−1

f (y)dy dr Br (x)

0


 ≤ c δ m (Mf )(x) + δ m(1−t) (Mmt f )(x) , where M is the Hardy–Littlewood maximal operator. Minimizing the righthand side in δ, we arrive at the inequality 1−1/t  1/t  Im f (x) ≤ c (Mmt f )(x) (Mf )(x) (see [AH], Sect. 3.1). Hence 1/t

1−1/t

Im f Lp (gdx) ≤ c Mmt f Lp (gt dx) Mf Lp

.

Therefore, by (1.2.51) and the boundedness of M in Lp , we find that 
1 1/pt t (g(y)) dy f Lp Im f Lp (gdx) ≤ c sup rn−mpt Br (x) x∈Rn , r>0



which implies the above mentioned result in [F2].

In the case p = 2 condition (1.2.50) was improved in [ChWW], where it is shown that, if ϕ is an increasing function: [0, ∞) → [1, ∞) subject to  ∞ (τ ϕ(τ ))−1 dτ < ∞, (1.2.52) 1

then the condition

 sup B

B

  g(x)ϕ g(x)(diamB)2 dx (mesn B)1−2/n

1 and [Maz11] for p = 1). Lemma 1.3.1. Let 1 ≤ p < q and let mp < n. Then the best constant in (1.3.1) is equivalent to  1/q ρm−n/p µ(Bρ (x) .

sup

(1.3.2)

x∈Rn ,ρ∈(0,1)

The case mp = n considered in the next lemma was treated in [MP], see also [Maz14], Sect. 8.6. Lemma 1.3.2. Let 1 < p < q and let mp = n. Then the best constant in (1.3.1) is equivalent to  1/q (log 2/ρ)(p−1)/p µ(Bρ (x) .

sup

(1.3.3)

x∈Rn ,ρ∈(0,1)

The next lemma is an obvious corollary of Sobolev’s theorem on the imbedding Wpm ⊂ L∞ which holds for mp > n. Lemma 1.3.3. Let 1 < p < q and let mp > n or 1 = p < q, m ≥ n. Then the best constant in (1.3.1) is equivalent to  1/q sup µ(B1 (x) .

(1.3.4)

x∈Rn

1.3.2 The case q < p ≤ n/m To state the next assertion, proved in [MN], we use the following function of one variable (0, ∞)  s → νp,m (µ; s) =

inf

{e:µ(e)>s}

Cp,m (e).

(1.3.5)

Lemma 1.3.4. Let 1 < p < ∞ and let 0 < q < p. Then the best constant in (1.3.1) is equivalent to  ∞
1  p−q sp ds . (1.3.6) q νp,m (µ; s) s 0 Another completely different characterization of the trace inequality (1.3.1) can be found in [COV1] for q > 1 and [COV2] for q > 0.

1.3 Estimate for the Lq -Norm with respect to an Arbitrary Measure

31

Lemma 1.3.5. Let 1 < p < ∞ and let 0 < q < p. Then the best constant in (1.3.1) is equivalent to    q(p−1) Wm,p µ(x) p−q dx, (1.3.7) Rn

where (Wm,p µ)(x) =

 1
1 µ(Br (x))  p−1 dr n−mp r r 0

is the so-called nonlinear Wolff potential.

(1.3.8)

2 Multipliers in Pairs of Sobolev Spaces

2.1 Introduction In the present chapter we study multipliers acting in pairs of spaces Wpk and wpk , where k is a nonnegative integer. The concepts of this chapter prove to be prototypes for the subsequent study of multipliers in other pairs of spaces. Using the result of Sects. 1.2 and 1.1, we derive necessary and sufficient conditions for a function to belong to the space of multipliers M (Wpm → Wpl ) and M (wpm → wpl ), where m ≥ l ≥ 0 and p ∈ [1, ∞) (Sects. 2.2, 2.3, and 2.8). The case of the half-space Rn+ is treated in Sect. 2.4. Section 2.5 contains conditions for the inclusion γ ∈ M (Wpm → Wp−k ), k > 0. In Sect. 2.6 we present a brief description of the space M (Wpm → Wql ). Section 2.7 deals with certain properties of multipliers. In the concluding Sect. 2.9 we give a description of multipliers preserving spaces of functions with bounded variation. As usual, we omit Rn in notations of spaces, norms, and integrals. Let γ ∈ M (Wpm → Wpl ), un → u in Wpm and γun → v in Wpl . Then there exists a sequence {nk }k≥1 , such that for almost all x unk (x) → u(x),

γ(x)unk (x) → v(x).

Hence v = γu almost everywhere in Rn , and therefore the operator Wpm  u → γu ∈ Wpl is closed. Since this operator is defined on the whole of Wpm , it is bounded by the Banach theorem. The norm in M (Wpm → Wpl ) is defined as the norm of the operator of multiplication γM (Wpm →Wpl ) = sup{γuWpl : uWpm ≤ 1}. We use the notation M Wpl instead of M (Wpl → Wpl ). V.G. Maz’ya, T.O. Shaposhnikova, Theory of Sobolev Multipliers, Grundlehren der mathematischen Wissenschaften 337, c Springer-Verlag Berlin Hiedelberg 2009


33

34

2 Multipliers in Pairs of Sobolev Spaces

It is worth noting that we can always assume that m ≥ l, since in the opposite case M (Wpm → Wpl ) = {0}.1 In fact, let m < l and let γ ∈ M (Wpm → Wpl ). We have γM (Wpm →Wpl ) ≥

γ eitx1 ηWpl eitx1 ηWpm

,

where t > 0 and η is an arbitrary non-zero function in C0∞ . Therefore,
γη  Lp + o(1) γM (Wpm →Wpl ) ≥ tl−m ηLp as t → ∞ and this is possible only if γ = 0. l we denote the space endowed with the norm By Wp,unif uWp,unif = sup ηz uWpl , l z∈Rn

where ηz (x) = η(x − z), η ∈ We also need the space

C0∞ ,

and η = 1 on B1 .

l = {u : ηu ∈ Wpl for all η ∈ C0∞ }. Wp,loc

Throughout this book similar notations Sunif and Sloc will be used for other Banach spaces S of functions defined on Rn . The following is the main result of this chapter. Theorem 2.1.1. (i) Let p > 1, mp > n or p = 1, m ≥ n. Then γM (Wpm →Wpl ) ∼ sup γ; B1 (x)Wpl . x∈Rn

(ii) Let p = 1, m < n. Then γM (W1m →W1l ) ∼ sup rm−n ∇l γ; Br (x)L1 x∈Rn r∈(0,1)

+

⎧ ⎨ sup γ; B1 (x)L1 , m > l, x∈Rn

⎩γ; Rn  , L∞

m = l.

(iii) Let mp ≤ n, p > 1. Then γM (Wpm →Wpl ) ∼ sup

e⊂Rn d(e)≤1

+

∇l γ; eLp (Cp,m (e))1/p

⎧ ⎨ sup γ; B1 (x)L1 , m > l, x∈Rn

⎩γ , L∞

m = l,

where d(e) is the diameter of e. 1

In other words, the multipliers, acting between two different spaces, uglify their domain, in full correspondence with Mock Turtle’s terminology in the epigraph to the present book.

2.2 Characterization of the Space M (W1m → W1l )

35

2.2 Characterization of the Space M (W1m → W1l) Theorem 2.2.1. For any m > 0 γM (W1m →L1 ) ∼ sup rm−n γ; Br (x)L1 .

(2.2.1)

x∈Rn r∈(0,1)

Proof. The result follows by Theorem 1.1.2. Theorem 2.2.2. For any 0 < l ≤ m the relation   γM (W1m →W1l ) ∼ sup rm−n ∇l γ; Br (x)L1 + r−l γ; Br (x)L1

(2.2.2)

x∈Rn r∈(0,1)

holds. Proof. First we note that   rj−l ∇j γ; Br (x)L1 ≤ c ∇l γ; Br (x)L1 + r−l γ; Br (x)L1 . Hence the equivalence relation (2.2.2) can be written as γM (W1m →W1l ) ∼

sup

rm−n

x∈Rn ,r∈(0,1)

l 

rj−l ∇j γ; Br (x)L1 .

(2.2.3)

j=0

We start with the lower estimate for the multiplier norm of γ. Let u(y) = ϕ((y − x)/r), where r ∈ (0, 1) for m < n and r = 1 for m ≥ n, ϕ ∈ C0∞ (B2 ), ϕ = 1 on B1 . We set this u into γuW1m ≤ γM (W1m →W1l ) uW1m and use the inequality rj−l ∇j (γu); B2r (x)L1 ≤ c γuW1l ,

j = 0, 1, . . . , l,

valid because supp γu ⊂ B2r (x). Hence rj−l ∇j γ; Br (x)L1 ≤ c γM (W1m →W1l ) rn−m which gives the lower estimate for γM (W1m →W1l ) . To obtain the upper estimate, we combine the obvious inequality ∇l (γu)L1 ≤ c

l 

 |∇j γ| |∇l−j u| L1

j=0

with the estimate  |∇j γ| |∇l−j u| L1 ≤ c

sup x∈Rn ,r∈(0,1)

rm−l+j−n ∇j γ; Br (x)L1 ∇l−j uW m−l+j 1

36

2 Multipliers in Pairs of Sobolev Spaces

which holds by Theorem 1.1.2. Hence ∇l (γu)L1 ≤ c

rm−n

sup x∈Rn ,r∈(0,1)

l 

rj−l ∇j γ; Br (x)L1 uW1m .

j=0

Also, by the same Theorem 1.1.2, γuL1 ≤ c

sup x∈Rn ,r∈(0,1)

rm−l−n γ; Br (x)L1 uW m−l . 1

Adding together the last two inequalities and noting (2.2.3), we complete the proof.  Corollary 2.2.1. Let 0 ≤ l < m. Then γM (W m−l →L1 ) ≤ c γM (W1m →W1l ) . 1

(2.2.4)

Proof. The result follows directly from Theorems 2.2.2 and 2.2.1. The equivalence relation (2.2.2) is modified in the following assertion. Theorem 2.2.3. (i) If m ≥ n, and m ≥ l, then γM (W1m →W1l ) ∼ sup γ; B1 (x)W1l .

(2.2.5)

x∈Rn

(ii) If l < n, then γM W1l ∼

sup x∈Rn ,r∈(0,1)

rl−n ∇l γ; Br (x)L1 + γL∞ .

(2.2.6)

(iii) If l < m < n, then γM (W1m →W1l ) ∼

sup x∈Rn ,r∈(0,1)

rm−n ∇l γ; Br (x)L1

+ sup γ; B1 (x)L1 .

(2.2.7)

x∈Rn

Proof. Relations (2.2.5) and (2.2.6) follow from Theorem 2.2.2. Let l < m < n. The lower estimate is a direct consequence of Theorem 2.2.2. To derive the upper estimate we show that   rm−n−l γ; Br (x)L1 ≤ c γ; B1 (x)L1 + sup ρm−n ∇l γ; Bρ (x)L1 (2.2.8) ρ∈(0,1)

for any x ∈ Rn and r ∈ (0, 1). By the Sobolev integral representation (see, for instance, [Maz14], Subsect. 1.1.10),  
|∇l γ(y)| dy . (2.2.9) |γ(z)| ≤ c γ; B(x)L1 + n−l B(x) |z − y|

2.2 Characterization of the Space M (W1m → W1l )

Hence




γ; Br (x)L1 ≤ c r γ; B1 (x)L1 +



n


≤ c rn γ; B1 (x)L1 +rl ∇l γ; B2r (x)L1 +

Br (x)



Br (x)

B1 (x)

37

 |∇l γ(y)| dydz n−l |z − y| |∇l γ(y)|

B2 (x)\B2r (x)

dydz  . |y|n−l

Combining this fact with the obvious inequality   dydz |∇l γ(y)| n−l ≤ c rl−m+n sup ρm−n ∇l γ; Bρ (x)L1 , |y| ρ∈(0,1) Br (x) B1 (x)\B2r (x) 

we complete the proof.

Theorem 2.2.3 implies an interpolation inequality for elements of the space M (W1m → W1l ). Corollary 2.2.2. Let γ ∈ M (W1m → W1l ). Then 1−j/l

j/l

γM (W m−j →W l−j ) ≤ c γM (W m →W l ) γ||M (W m−l →L ) , 1

1

1

1

(2.2.10)

1

1

where j = 0, 1, . . . , l. Proof. Making the dilation in the well-known inequality 1−j/l

∇l−j γ; B1 L1 ≤ c γ; B1 W l 1

j/l

γ; B1 L1 ,

we obtain 1−j/l  j/l rl−j ∇l−j γ; Br (x)L1 ≤ c rl ∇l γ; Br (x)L1 +γ; Br (x)L1 γ; Br (x)L1 . Hence rm−j−n ∇l−j γ; Br (x)L1 + rm−n−l γ; Br (x)L1 1−j/l  m−n−l j/l  r ≤ c rm−n ∇l γ; Br (x)L1 +rm−n−l γ; Br (x)L1 γ; Br (x)L1 . 

Reference to Theorem 2.2.2 completes the proof. Corollary 2.2.3. Let 0 < l < m. Then γM (W1m →W1l ) ∼

l 

∇l−j γM (W m−j →L1 ) 1

(2.2.11)

j=0

and γM (W1m →W1l ) ∼ ∇l γM (W1m →L1 ) + γL1 ,unif . For m = l the norm γL1 ,unif should be replaced by γL∞ .

(2.2.12)

38

2 Multipliers in Pairs of Sobolev Spaces

Proof. Estimate 2.2.12 results from Theorems 2.2.3 and 2.2.1. By Theorems 2.2.1 and 2.2.3 ∇l−j γM (W m−j →L1 ) ≤ c 1

sup x∈Rn ,r∈(0,1)

rm−j−n ∇l−j γ; Br (x)L1

≤ c γM (W m−j →W l−j ) 1

1

which is dominated by γM (W1m →W1l ) in view of Corollary 2.2.2. The lower estimate 2.2.11 follows. 

2.3 Characterization of the Space M (Wpm → Wpl ) for p > 1 Here we derive necessary and sufficient conditions for a function to belong to the space M (Wpm → Wpl ) for p > 1. The next assertion contains an inequality between multipliers and their mollifiers. Lemma 2.3.1. Let γρ denote a mollifier of a function γ which is defined as  γρ (x) = ρ−n K(ρ−1 (x − ξ))γ(ξ)dξ, where K ∈ C0∞ (B1 ), K ≥ 0, and KL1 = 1. Then γρ M (Wpm →Wpl ) ≤ γM (Wpm →Wpl ) ≤ lim inf γρ M (Wpm →Wpl ) . ρ→0

(2.3.1)

Proof. Let u ∈ C0∞ . By Minkowski’s inequality  p 1/p
 ∇j,x ρ−n K(ξ/ρ)γ(x − ξ)u(x)dξ dx  ≤


 1/p   ∇j,y γ(y)u(y + ξ) p dy ρ−n K(ξ/ρ) dξ,

where j = 0, l. Therefore, γρ uWpl ≤ γM (Wpm →Wpl )



−n

ρ



 1/p K(ξ/ρ) |∇m,y u(y + ξ)|p dy


 1/p  + |u(y + ξ)|p dy dξ ≤ γM (Wpm →Wpl ) uWpm . This gives the left inequality (2.3.1). The right inequality (2.3.1) follows from γuWpl = lim inf γρ uWpl ≤ lim inf γρ M (Wpm →Wpl ) uWpm . ρ→0

The proof is complete.

ρ→0



2.3 Characterization of M (Wpm → Wpl ) for p > 1

39

The following assertion is a particular case of Lemma 1.2.7. Lemma 2.3.2. Let γ ∈ Lp,loc , p ∈ (1, ∞), and let u be an arbitrary function in C0∞ . The best constant in the inequality γ∇l uLp + γuLp ≤ C uWpm is equivalent to the norm γM (Wpm−l →Lp ) . The next lemma concerns derivatives of multipliers. Lemma 2.3.3. Suppose that γ ∈ M (Wpm → Wpl ) ∩ M (Wpm−l → Lp ),

p ∈ (1, ∞).

Then, for any multi-index α of order |α| ≤ l, Dα γ ∈ M (Wpm → Wpl−|α| ) and Dα γM (W m →Wpl−|α| ) p

≤ ε γM (Wpm−l →Lp ) + c(ε) γM (Wpm →Wpl ) ,

(2.3.2)

where ε is an arbitrary positive number. Proof. Let u ∈ Wpl and let ϕ be an arbitrary function in C0∞ . Applying the Leibniz formula
α! Dβ ϕ Dα−β u, Dα (ϕu) = β!(α − β)! {β:α≥β≥0}

we obtain   ϕu(−D)α γdx = γDα (ϕu)dx =


{β:α≥β≥0}

 =




ϕ

{β:α≥β≥0}

α! γDβ ϕ Dα−β u dx β!(α − β)!

α! (−D)β (γDα−β u)dx. β!(α − β)!

Therefore,


uDα γ =

{β:α≥β≥0}

  α! Dβ γ(−D)α−β u , β!(α − β)!

which implies the estimate uDα γWpl−|α| ≤ c


{β:α≥β≥0}

γDα−β uWpl−|α|+|β| .

40

2 Multipliers in Pairs of Sobolev Spaces

Hence, it suffices to prove (2.3.2) for |α| = 1, l ≥ 1. We have u∇γWpl−1 ≤ uγWpl + γ∇uWpl−1   ≤ γM (Wpm →Wpl ) + γM (Wpm−1 →Wpl−1 ) uWpm . Estimating the norm γM (Wpm−1 →Wpl−1 ) by (2.3.8), we arrive at (2.3.2).



The equivalent representation for the norm in M (Wpm → Lp ) is a direct consequence of Theorem 1.2.2. Namely, γM (Wpm →Lp ) ∼ sup e

γ; eLp . (Cp,m (e))1/p

(2.3.3)

By Corollary 1.2.2 this can also be written in the form γM (Wpm →Lp ) ∼

γ; eLp , 1/p e:d(e)≤1 (Cp,m (e)) sup

(2.3.4)

where d(e) is the diameter of e. Now we pass to two-sided estimates for the norms in M (Wpm → Wpl ), p ∈ (1, ∞), given in terms of the spaces M (Wpk → Lp ). We start with lower estimates. Lemma 2.3.4. Let γ ∈ M (Wpm → Wpl ). Then ∇l γM (Wpm →Lp ) + γM (Wpm−l →Lp ) ≤ c γM (Wpm →Wpl ) .

(2.3.5)

Proof. Suppose first that γ ∈ M (Wpm−l → Lp ). We have γ∇l uLp ≤ γM (Wpm →Wpl ) uWpm + c




Dα uDβ γLp

|α|+|β|=l, β
=0

l  
≤ γM (Wpm →Wpl ) + c ∇j γM (Wpm−l+j →Lp ) uWpm .

(2.3.6)

j=1

Lemma 2.3.3 implies ∇j γM (Wpm−l+j →Lp ) ≤ ε γM (Wpm−l →Lp ) + c(ε) γM (Wpm−l+j →Wpj ) .

(2.3.7)

By an interpolation property of Sobolev spaces (see [Tr4]) we have (l−j)/l

j/l

γM (Wpm−j →Wpl−j ) ≤ c γM (W m →W l ) γM (W m−l →L ) , p

p

p

p

(2.3.8)

2.3 Characterization of M (Wpm → Wpl ) for p > 1

41

where 0 ≤ j ≤ l. Estimating the last norm in (2.3.7) by (2.3.8), we obtain ∇j γM (Wpm−l+j →Lp ) ≤ ε γM (Wpm−l →Lp ) + c(ε) γM (Wpm →Wpl ) . Substitution of this inequality into (2.3.6) gives   γ∇l Lp ≤ ε γM (Wpm−l →Lp ) + c(ε) γM (Wpm →Wpl ) uWpm .

(2.3.9)

Also, γuLp ≤ γM (Wpm →Wpl ) uWpm .

(2.3.10)

Combining the last two estimates and applying Lemma 2.3.2, we arrive at γM (Wpm−l →Lp ) ≤ ε γM (Wpm−l →Lp ) + c(ε) γM (Wpm →Wpl ) . Hence, γM (Wpm−l →Lp ) ≤ c γM (Wpm →Wpl ) .

(2.3.11)

Next we remove the assumption γ ∈ M (Wpm−l → Lp ). Since γ ∈ M (Wpm → Wpl ), then γηLp ≤ c ηWpm , where η ∈ C0∞ (B2 (x)), η = 1 on B1 (x), and x is an arbitrary point in Rn . Hence sup γ; B1 (x)Lp < ∞ x

and for any k = 0, 1, . . . there exists a constant cρ such that |∇k γρ | ≤ cρ . Since the function γρ and all its derivatives are bounded, it follows that γρ is a multiplier in Wpk for any k = 1, 2, . . . , and thus γρ ∈ M (Wpm−l → Lp ). By (2.3.11) γρ M (Wpm−l →Lp ) ≤ c γρ M (Wpm →Wpl ) . Letting ρ → 0 and using Lemma 2.3.1 we arrive at (2.3.11) for all γ ∈ M (Wpm → Wpl ). To estimate the first term on the right-hand side of (2.3.5) we combine (2.3.11) with (2.3.7) for j = l.

The estimate inverse to (2.3.5) is contained in the following lemma. Lemma 2.3.5. Let γ ∈ M (Wpm−l → Lp ) and let ∇l γ ∈ M (Wpm → Lp ). Then γ ∈ M (Wpm → Wpl ) and the estimate   γM (Wpm →Wpl ) ≤ c ∇l γM (Wpm →Lp ) + γM (Wpm−l →Lp ) holds.

(2.3.12)

42

2 Multipliers in Pairs of Sobolev Spaces

Proof. Inequality (2.3.8) along with Lemma 2.3.3 gives j/l

1−j/l

∇j γM (Wpm−l+j →Lp ) ≤ c γM (W m →W l ) γM (W m−l →L ) , p

p

p

(2.3.13)

p

where j = 1, . . . , l − 1. For any u ∈ C0∞ , ∇l (γu)Lp ≤ c

l


  |∇j γ| |∇l−j u| Lp ≤ c ∇l γM (Wpm →Lp )

j=0

+γM (Wpm−l →Lp ) +

l−1


 ∇j γM (Wpm−l+j →Lp ) uWpm .

j=1

Then it follows from (2.3.13) that   ∇l (γu)Lp ≤ c ∇l γM (Wpm →Lp ) + γM (Wpm−l →Lp ) uWpm . It remains to note that γuLp ≤ γM (Wpm−l →Lp ) uWpm−l .

Unifying Lemmas 2.3.4 and 2.3.5, we arrive at the following assertion. Theorem 2.3.1. Let m and l be integers, and let p ∈ (1, ∞). A function γ l belongs to the space M (Wpm → Wpl ) if and only if γ ∈ Wp,loc , γ ∈ M (Wpm−l → m Lp ), and ∇l γ ∈ M (Wp → Lp ). Moreover, γM (Wpm →Wpl ) ∼ ∇l γM (Wpm →Lp ) + γM (Wpm−l →Lp ) . Relation (2.3.3) leads to the following reformulation of Theorem 2.3.1. Theorem 2.3.2. Let m and l be integers, and let p ∈ (1, ∞). A function γ l belongs to the space M (Wpm → Wpl ) if and only if γ ∈ Wp,loc and, for any n compact set e ⊂ R , ∇l γ; epLp ≤ c Cp,m (e) and γ; epLp ≤ c Cp,m−l (e). Moreover,  ∇ γ; e  γ; eLp l Lp . γM (Wpm →Wpl ) ∼ sup + (Cp,m (e))1/p (Cp,m−l (e))1/p e An important particular case of Theorem 2.3.2 is m = l.

(2.3.14)

2.3 Characterization of M (Wpm → Wpl ) for p > 1

43

Corollary 2.3.1. Let l be an integer and let p ∈ (1, ∞). A function γ belongs l and, for any compact set e ⊂ Rn , to the space M Wpl if and only if γ ∈ Wp,loc ∇l γ; epLp ≤ c Cp,l (e). Moreover, γM Wpl ∼ sup e

∇l γ; eLp + γL∞ . (Cp,l (e))1/p

(2.3.15)

By (2.3.4), one can use only compact sets with diameters not exceeding 1 in Theorem 2.3.2 and Corollary 2.3.1 and rewrite (2.3.14) and (2.3.15) as follows:  ∇ γ; e  γ; eLp l Lp γM (Wpm →Wpl ) ∼ sup + , (2.3.16) 1/p 1/p (Cp,m−l (e)) e:d(e)≤1 (Cp,m (e)) γM Wpl ∼

∇l γ; eLp + γL∞ . 1/p (C p,l (e)) e:d(e)≤1 sup

(2.3.17)

2.3.1 Another Characterization of the Space M (Wpm → Wpl ) for 0 < l < m, pm ≤ n, p > 1 Lemma 2.3.6. Let p ∈ (1, ∞), 0 < ν < µ, and let ϕ be a nonnegative function in Lpµ,loc . Then, for any compact set e of positive measure,    ϕνp dx 1/ν  ϕµp dx 1/µ sup e ≤ sup e . (2.3.18) Cp,ν (e) Cp,µ (e) e e The same assertion holds for compact sets e of positive measure with diameter not exceeding 1. Proof. Let u ∈ C0∞ and let f = Jν u. By (1.2.17) and H¨ older’s inequality    (µ−ν)p/µ νp/µ  ϕνp |u|p dx ≤ c Mf ϕνp Jµ |f | dx    p ν/µ   p (µ−ν)/µ µp . ≤c Mf dx ϕ Jµ |f | dx Using the continuity of the Hardy–Littlewood operator M in Lp , we find   ϕνp dx ν/µ  νp/µ (µ−ν)p/µ ϕνp |u|p dx ≤ c sup e Jµ |f | Wpµ f Lp . Cp,µ (e) e

44

2 Multipliers in Pairs of Sobolev Spaces

Since Jµ |f | Wpµ ≤ c f Lp , it follows that   ϕνp dx ν/µ  ϕνp |u|p dx ≤ c sup e upWpµ . Cp,µ (e) e



The result follows by Theorem 1.2.2. Corollary 2.3.2. For any m > l ≥ 0 and p ∈ (1, ∞), sup e

γ; eLpm/(m−l) γ; eLp ≤ c sup . 1/p (m−l)/pm (Cp,m−l (e)) e (Cp,m (e))

(2.3.19)

The same assertion holds for compact sets e of positive measure with diameter not exceeding 1. Proof. The result follows by Lemma 2.3.6 with ϕ = |γ|1/(m−l) , ν = m − l, and µ = m.

The following assertion was obtained in [Ad2]. Lemma 2.3.7. Let 0 < l < m ≤ n/p, p ∈ (1, ∞), and let Il f be the Riesz potential with a nonnegative density f ∈ Lp,loc . Then l/m  (m−l)/m  (Mf )(x) . (Il f )(x) ≤ c sup rm−n/p f ; Br (x)Lp r>0

Proof. It is enough to put x = 0. For any δ > 0 we have   f (z)dz f (z)dz (Il f )(0) = + . n−l n−l |z| n Bδ R \Bδ |z| Clearly,

 Bδ

f (z)dz ≤ c δ l (Mf )(0). |z|n−l

(2.3.20)

(2.3.21)

The second integral in (2.3.20) can be written as   ∞  f (z)dz dr l−n = (n − l) f (ξ)dξ n−l+1 − δ f (ξ)dξ. n−l r δ Rn \Bδ |z| Br Bδ By H¨older’s inequality the right-hand side does not exceed  1/p 1/p    ∞  dr p l−n/p p . f (ξ) dξ + δ f (ξ) dξ c r1−l+n/p δ Br Bδ

2.3 Characterization of M (Wpm → Wpl ) for p > 1

Hence

 Rn \B

δ

45

f (z)dz ≤ c δ l−m sup rm−n/p f ; Br (x)Lp |z|n−l r>0

which together with (2.3.21) implies that   (Il f )(0) ≤ c δ l (Mf )(0) + δ l−m sup rm−n/p f ; Br (x)Lp . r>0



The result follows by minimizing the right-hand side in δ.

The next lemma is due to Verbitsky. For its proof see [MSh16], Sect. 2.6 and [MV1], Sect. 3. Lemma 2.3.8. Let p ∈ (1, ∞) and let 0 < m ≤ n/p. Then sup e

M f ; eLp f ; eLp ≤ c sup . 1/p 1/p (Cp,m (e)) e (Cp,m (e))

(2.3.22)

This inequality plays a crucial role in the next assertion, also proved by Verbitsky, see [MSh16], Sect. 2.6. Lemma 2.3.9. Let m and l be integers, 0 < l < m ≤ n/p, and let p ∈ (1, ∞). Then γ; eLpm/(m−l) sup (m−l)/pm e (Cp,m (e))   ∇l γ; eLp . ≤ c sup + sup γ; B (x) 1 L 1 1/p e (Cp,m (e)) x∈Rn

(2.3.23)

Proof. By the Sobolev integral representation [Sob] (see, for instance, [Maz14], Subsect. 1.1.10 and [Bur], Sect. 3.4)  |∇ γ(x + z)|  l |γ(x)| ≤ c dz + sup γ; B (x) 1 L1 . |z|n−l x∈Rn B2 Combining this inequality with Lemma 2.3.7, we obtain  |γ(x)| ≤ c (

sup

z∈Rn ,r∈(0,1)

rm−n/p ∇l γ; Br (z)Lp )l/m ((M∇l γ)(x))1−l/m + sup γ; B1 (x)L1



x∈Rn

for almost all x ∈ Rn . Adopting the notation K=

sup z∈Rn ,r∈(0,1)

rm−n/p ∇l γ; Br (z)Lp ,

(2.3.24)

46

2 Multipliers in Pairs of Sobolev Spaces

we deduce from (2.3.24) that

 |γ|pm/(m−l) dx e

   ≤ c Kpl/(m−l) |(M∇l γ)(x)|p dx + ( sup γ; B1 (x)L1 )pm/(m−l) mesn e . x∈Rn

e

Using here the obvious estimate mesn e ≤ Cp,m (e), we find that   |γ(x)|pm/(m−l) dx (m−l)/pm e

Cp,m (e)    M∇l γ; eLp 1−l/m . ≤ c Kl/m sup + sup γ; B (x) 1 L 1 (Cp,m (e))1/p e x∈Rn Reference to Lemma 2.3.8 and the inequality K ≤ sup e

∇l γ; eLp , (Cp,m (e))1/p



completes the proof.

The following assertion gives one more representation of a norm in M (Wpm → Wpl ). Theorem 2.3.3. Let m and l be integers, 0 < l < m ≤ n/p, and let p ∈ (1, ∞). A function γ belongs to the space M (Wpm → Wpl ) if and only if γ ∈ l and for any compact set e ⊂ Rn Wp,loc ∇l γ; epLp ≤ c Cp,m (e). Moreover, γM (Wpm →Wpl ) ∼ sup e

∇l γ; eLp + sup γ; B1 (x)L1 . (Cp,m (e))1/p x∈Rn

(2.3.25)

Here one can use only compact sets with diameter not exceeding 1. Proof. We start with the lower estimate. By Lemma 2.3.4, γηLp ≤ γM (Wpm−l →Lp ) ηWpm−l for any η ∈ C0∞ (B2 (x)) with η = 1 on B1 (x), where x is an arbitrary point in Rn . Therefore, sup γ; B1 (x)Lp ≤ c γM (Wpm−l →Lp ) .

x∈Rn

The upper estimate is a direct corollary of (2.3.14), Corollary 2.3.2, and Verbitsky’s Lemma 2.3.9.



2.3 Characterization of M (Wpm → Wpl ) for p > 1

47

Corollary 2.3.3. Let m and l be integers, 0 < l < m and let p ∈ (1, ∞). Then l
γM (Wpm →Wpl ) ∼ ∇l−j γM (Wpm−j →Lp ) (2.3.26) j=0

and γM (Wpm →Wpl ) ∼ ∇l γM (Wpm →Lp ) + γL1,unif .

(2.3.27)

For m = l the norm γL1 ,unif should be replaced by γL∞ . Proof. The upper estimates follow from (2.3.25) and (2.3.3). The lower estimate in (2.3.26) results from ∇l−j γM (Wpm−j →Lp ) ≤ c γM (Wpm−j →Wpl−j ) ≤ c γM (Wpm →Wpl ) .

We finish this subsection with one more two-sided estimate for the norm in M (Wpm → Wpl ). Corollary 2.3.4. Let m and l be integers, 0 < l < m and let p ∈ (1, ∞). Then l
∇j γ; eLp c sup ≤ γM (Wpm →Wpl ) 1/p e [Cp,m−l+j (e)] j=0   ∇l γ; eLp (2.3.28) ≤ c sup + γ L1,unif . 1/p e [Cp,m (e)] For m = l the norm γL1,unif should be replaced by γL∞ . Proof. The lower estimate follows from (2.3.26) and (2.3.3). The upper one is contained in (2.3.25).

2.3.2 Characterization of the Space M (Wpm → Wpl ) for pm > n, p>1 For pm > n the space M (Wpm → Wpl ) has a simple description which is contained in the next assertion. Theorem 2.3.4. If pm > n, p ∈ (1, ∞), then γM (Wpm →Wpl ) ∼ sup γ; B1 (x)Wpl .

(2.3.29)

x∈Rn

Proof. Since for compact sets e with diameter less than 1 the equivalence relation Cp,m (e) ∼ 1 holds, the result follows from (2.3.25).



Remark 2.3.1. For pm > n, p ∈ (1, ∞), the relation (2.3.29) can be written as . γM (Wpm →Wpl ) ∼ γWp,unif l

48

2 Multipliers in Pairs of Sobolev Spaces

2.3.3 One-Sided Estimates for Norms of Multipliers in the Case pm ≤ n For mp ≤ n we can give different upper and lower bounds for norms in M (Wpm → Wpl ) which do not involve capacity. In other words, we obtain separate non-capacitary necessary and sufficient conditions for a function to belong to this space of multipliers. Using the estimates of the capacity of a ball given in Proposition 1.2.9 and Theorem 2.3.3 we immediately arrive at the following lower estimates for norms of multipliers. Proposition 2.3.1. (i) If pm < n, p ∈ (1, ∞), then  sup rm−n/p ∇l γ; Br (x)Lp γM (Wpm →Wpl ) ≥ c x∈Rn ,r∈(0,1)

 + sup γ; B1 (x)L1 .

(2.3.30)

x∈Rn

(ii) If pm = n, p ∈ (1, ∞), then  γM (Wpm →Wpl ) ≥ c sup

x∈Rn ,r∈(0,1)

((log(2/r))1−1/p ∇l γ; Br (x)Lp

 + sup γ; B1 (x)L1 .

(2.3.31)

x∈Rn

Next, we give upper estimates for the norm in M (Wpm → Wpl ) which result directly from Theorem 2.3.3 and Proposition 1.2.8. Proposition 2.3.2. (i) If pm < n, p ∈ (1, ∞), then   ∇l γ; eLp , (2.3.32) γM (Wpm →Wpl ) ≤ c sup + sup γ; B (x) 1 L 1 1/p−m/n x∈Rn e:d(e)≤1 (mesn e) where d(e) is the diameter of e. (ii) If pm = n, p ∈ (1, ∞), then  γM (Wpm →Wpl ) ≤ c sup (log(2n /mesn e))1−1/p ∇l γ; eLp e:d(e)≤1

 + sup γ; B1 (x)L1 . x∈Rn

For m = l one should replace sup γ; B1 (x)L1 by γL∞ . x∈Rn

The next assertion follows immediately from Proposition 2.3.2. Corollary 2.3.5. (i) If pm < n, l < m, and p ∈ (1, ∞), then γM (Wpm →Wpl ) ≤ c γWn/m,unif . l (ii) If lp < n, p ∈ (1, ∞), then   γM Wpl ≤ c sup ∇l γ; B1 (x)Ln/l + γL∞ . x∈Rn

(2.3.33)

2.3 Characterization of M (Wpm → Wpl ) for p > 1

49

2.3.4 Examples of Multipliers Propositions 2.3.1, 2.3.1 and Theorem 2.3.4 enable one to verify conditions for inclusion of individual functions into spaces of multipliers. We give three examples of this kind. Example 2.3.1. Let γ(x) = η(x)|x|α+iβ , where η ∈ C0∞ , η(0) = 1, α ∈ R, and β ∈ R\{0}. l = M (Wpm → Wpl ) if and only if α > Let mp > n. Clearly, γ ∈ Wp,unif l − n/p. l and hence γ ∈ / Suppose that mp ≤ n. If α ≤ l − n/p, then γ ∈ / Wp,loc m l M (Wp → Wp ) because of (2.3.30) and (2.3.31). Consider the case mp ≤ n, α > l − n/p. We obtain ∇l γ; eLp ∼ |x|α−l ; eLp ≤ c (mesn e)α−l+n/p for all compact sets e with d(e) ≤ 1. Using (2.3.32) and (2.3.33), we conclude that γ ∈ M (Wpm → Wpl ). Summarizing, we have γ ∈ M (Wpm → Wpl ) ⇐⇒ α > l − n/p. Example 2.3.2. Let µ > 0 and let γ(x) = η(x) exp(i|x|−µ ), where η ∈ C0∞ , η(0) = 1. Clearly, |∇l γ(x)| ∼ |x|−l(µ+1)

as x → 0.

Therefore, γ ∈ Wpl (Rn ) is equivalent to n > pl(µ + 1). Let us find a criterion for γ ∈ M (Wpm → Wpl ). In view of Theorem 2.3.4, the same inequality n > pl(µ + 1) is necessary and sufficient for γ to belong to M (Wpm (Rn ) → Wpl (Rn )) for mp > n. Suppose that mp < n. We have ∇l γ; Br L∞ ∼ |x|−l(µ+1) ; Br Lp and lim rm−n/p ∇l γ; Br Lp = ∞

r→0

for m < l(µ + 1). According to Proposition 2.3.1, this means that γ ∈ / M (Wpm → Wpl ) for m < l(µ + 1). If m ≥ l(µ + 1), then ∇l γ; eLp ≤ c|x|−l(µ+1) ; eLp ≤ c (mesn e)−l(µ+1)/n+1/p for any compact set e with diameter d(e) ≤ 1. This, together with Proposition 2.3.2, implies that γ ∈ M (Wpm → Wpl ). Thus, for mp < n,

50

2 Multipliers in Pairs of Sobolev Spaces

γ ∈ M (Wpm → Wpl ) ⇐⇒ m ≥ l(µ + 1). In the same way we verify that γ ∈ M (Wpm → Wpl ) ⇐⇒ m > l(µ + 1) for mp = n. Example 2.3.3. Let µ, ν > 0, η ∈ C0∞ (B1 (0)), η(0) = 1 and γ(x) = η(x)(log |x|−1 )−ν exp(i (log |x|−1 )µ ). Clearly,

|∇l γ(x)| ∼ c |x|−l (log |x|−1 )l(µ−1)−ν .

Using the same arguments as in Example 2.3.2, from this relation and Propositions 2.3.1 and 2.3.2 we obtain λ ∈ Wpl (Rn ) ⇐⇒ l(µ − 1) < ν − 1/p, λ ∈ M Wpl (Rn ) ⇐⇒ l(µ − 1) ≤ ν − 1 

for lp = n.

l n 2.4 The Space M (Wpm(Rn + ) → Wp(R+ ))

2.4.1 Extension from a Half-Space Let Rn+ = {z = (x, xn ) : x ∈ Rn−1 , xn > 0}. The classical extension operator π is defined for functions given on Rn+ by ⎧ ⎪ for xn > 0, ⎪ ⎨v(z) l  π(v)(z) = ⎪ αj v(x, −jxn ) for xn < 0, ⎪ ⎩ j=1

where αj satisfy the conditions l 

(−1)k j k αj = 1,

0 ≤ k ≤ l − 1.

j=1

Lemma 2.4.1. Suppose that γ ∈ M (Wpm (Rn+ ) → Wpl (Rn+ )), where 0 ≤ l ≤ m and p ∈ [1, ∞). Then π(γ) ∈ M (Wpm (Rn ) → Wpm (Rn )) and π(γ); Rn M (Wpm →Wpl ) ≤ c γ; Rn+ M (Wpm →Wpl ) .

(2.4.1)

l n 2.4 The Space M (Wpm (Rn + ) → Wp (R+ ))

51

l Proof. Since γ ∈ Wp,loc (Rn+ ), it follows by the well-known property of the l operator π that π(γ) ∈ Wp,loc (Rn ). Hence π(γ)u ∈ Wpl (Rn ) for any u ∈ ∞ n C0 (R ). We have

π(γ); Rn pW l = γu; Rn+ Wpl + π(γ)u; Rn− Wpl p

≤ γu; Rn+ Wpl + c

l 

γ uj ; Rn+ pW l , p

j=1

where uj (x, xn ) = u(x, −xn /j) and Rn− = {z = (x, xn ) : x ∈ Rn−1 , xn < 0}. Therefore, l    π(γ); Rn pW l ≤ c γ; Rn+ M (Wpm →Wpl ) u; Rn+ Wpm + uj ; Rn+ Wpm p

j=1

and, since uj ; Rn+ Wpm ≤ c u; Rn− Wpm , it follows that π(γ); Rn pW l ≤ c γ; Rn+ M (Wpm →Wpl ) u; Rn Wpm . p



The lemma is proved.

2.4.2 The Case p > 1 In this subsection and elsewhere we use the notation Br± (Y ) = Br (Y ) ∩ Rn± . Theorem 2.4.1. A function γ belongs to M (Wpm (Rn+ ) → Wpm (Rn+ )) with p ∈ l (1, ∞) if and only if γ ∈ Wp,loc (Rn+ ) and, for any compact set e ⊂ Rn+ , ∇l γ; epLp ≤ c Cp,m (e).

(2.4.2)

Moreover, c1 sup e

l 

∇k γ; eLp ≤ γ; Rn+ M (Wpm →Wpl ) 1/p (C (e)) p,m−l+k k=0


 ∇l γ; eLp + . ≤ c2 sup + sup γ; B (X) L 1 1 1/p e (Cp,m (e)) X∈Rn +

(2.4.3)

52

2 Multipliers in Pairs of Sobolev Spaces

Proof. By Theorem 2.3.3 sup e⊂Rn +

∇k γ; eLp ∇k π(γ); eLp ≤ sup ≤ c π(γ); Rn M (Wpm →Wpl ) . 1/p n (Cp,m−l+k (e))1/p (C e⊂R p,m−l+k (e))

Reference to Lemma 2.4.1 implies the lower estimate for γ; Rn+ M (Wpm →Wpl ) . We turn to the proof of the upper estimate in (2.4.3). Let u ∈ C0∞ (Rn+ ) and let γ ∈ M (Wpm (Rn+ ) → Wpl (Rn+ )). We have γu; Rn+ Wpl ≤ π(γ) π(u); Rn Wpl .

(2.4.4)

By Theorem 2.3.3 the right-hand side does not exceed
 ∇l π(γ); eLp c sup π(u); Rn Wpm . + sup π(γ); B (z) 1 L 1 1/p e⊂Rn (Cp,m (e)) z∈Rn Since π(u); Rn Wpm ≤ c u; Rn+ Wpm

(2.4.5)

sup π(γ); B1 (z)L1 ≤ c sup γ; B1+ (X)L1 ,

(2.4.6)

and X∈Rn +

z∈Rn

it remains to prove the inequality sup e⊂Rn

∇l π(γ); eLp ∇l γ; eLp ≤ c sup . 1/p (Cp,m (e))1/p (C p,m (e)) e⊂Rn +

(2.4.7)

Let e± = e ∩ Rn± . The quotient on the left-hand side of (2.4.7) does not exceed the sum ∇l π(γ); e− Lp ∇l γ; e+ Lp + . (2.4.8) 1/p (Cp,m (e+ )) (Cp,m (e− ))1/p We put ej = {z : (x, −xn /j) ∈ e− },

j = 1, . . . , l.

Using the inequality ∇l π(γ); e− Lp ≤ c ∇l γ; ej Lp along with the equivalence relation Cp,m (e− ) ∼ Cp,m (ej ), we conclude that the second term in (2.4.8) is majorized by c sup e⊂Rn +

The proof is complete.

∇l γ; eLp . (Cp,m (e))1/p 

l n 2.4 The Space M (Wpm (Rn + ) → Wp (R+ ))

53

The following assertion follows directly from the last theorem. Corollary 2.4.1. If γ ∈ M (Wpm (Rn+ ) → Wpl (Rn+ )), then Dα γ ∈ M (Wpm (Rn+ ) → Wpl−|α| (Rn+ )) for any multi-index α of order |α| ≤ l. The estimate Dα γ; Rn+ M (W m →Wpl−|α| ) ≤ c γ; Rn+ M (Wpm →Wpl ) p

holds. 2.4.3 The Case p = 1 Theorem 2.4.2. A function γ belongs to M (W1m (Rn+ ) → W1l (Rn+ )) if and l (Rn+ ) and only if γ ∈ W1,loc ∇l γ; Br+ (X)L1 ≤ c rn−m for any X ∈

Rn+

(2.4.9)

and r ∈ (0, 1). Moreover, γ; Rn+ M (W1m →W1l ) ∼

sup rm−n ∇l γ; Br+ (X)L1 + sup γ; B1+ (X)L1 .

X∈Rn +

(2.4.10)

X∈Rn +

r∈(0,1)

Proof. By Theorem 2.2.3 sup rm−n ∇l γ; Br+ (X)L1 ≤ sup rm−n ∇l π(γ); Br (Y )L1

X∈Rn +

Y ∈Rn + r∈(0,1)

r∈(0,1)

≤ c π(γ); Rn M (W1m →W1l ) . Reference to Lemma 2.4.1 implies the lower estimate for γ; Rn+ M (W1m →W1l ) . To obtain the upper estimate, we take U ∈ C0∞ (Rn+ ) and γ ∈ M (W1m (Rn+ ) → W1l (Rn+ )). In view of (2.4.4), which is also valid for p = 1, the norm γu; Rn+ W1l is dominated by   c sup rm−n ∇l π(γ); Br (Y )L1 + sup π(γ); B1 (Y )L1 π(U ); Rn W1m . Y ∈Rn

Y ∈Rn r∈(0,1)

By (2.4.5), which holds also for p = 1, and (2.4.4), it remains to obtain the majorant for ∇l π(γ); Br (Y )L1 . We have ∇l π(γ); Br (Y )L1 ≤ ∇l π(γ); Br+ (Y )L1 + ∇l π(γ); Br− (Y )L1 . Putting

Br,j = {z : (x, −xn /j) ∈ Br− (Y )}

and using the inequalities ∇l π(γ); Br− (Y )L1 ≤ c ∇l π(γ); Br,j L1 ≤ c sup ∇γ; Br+ (X)L1 , X∈Rn +

we complete the proof.



54

2 Multipliers in Pairs of Sobolev Spaces

2.5 The Space M (Wpm → Wp−k) Let m and k be positive integers, and let Wp−k stand for the dual space (Wpk
) , where p + p = pp . The following assertion contains a sufficient condition for inclusion into the distribution space M (Wpm → Wp−k ). Theorem 2.5.1. (i) Let p ∈ (1, ∞), 0 < m ≤ k. If  γ= Dα γα

(2.5.1)

|α|≤k

with ) ∩ M (Wpm → Lp ), γα ∈ M (Wpk
→ Wpk−m


(2.5.2)

then γ ∈ M (Wpm → Wp−k ). (ii) Let p ∈ (1, ∞), m ≥ k > 0. If  Dα γα γ= |α|≤m

with γα ∈ M (Wpm → Wpm−k ) ∩ M (Wpk
→ Lp
), then γ ∈ M (Wpm → Wp−k ). Proof. It suffices to prove only (i), since (ii) follows from (i) by duality. Let u ∈ Wpm , m ≤ k. Since  cλα Dλ (γα Dα−λ u), cλα = const, uDα γα = λ≤α

we have γuWp−k ≤ c ≤c

 |λ|≤|α|≤k

 |λ|≤|α|≤k

γα Dα−λ uWp|λ|−k

γα M (Wpm−k+|λ| →Wp|λ|−k ) uWpm+|α|−k .

(2.5.3)

Applying the interpolation inequality |λ|/k

(k−|λ|)/k

γα M (Wpm−k+|λ| →Wp|λ|−k ) ≤ c γα M (W m−k →W −k ) γα M (W m →Lp ) , p

p

p

which results from the interpolation property of Sobolev spaces (see [Tr4], Sect. 2.4), we obtain from (2.5.3) γuWp−k ≤ c (γα M (Wpm−k →Wp−k ) + γα M (Wpm →Lp ) )uWpm . It remains to note that γα M (Wpm−k →Wp−k ) = γα M (W k
→W k−m ).
p

The proof is complete.

p



2.5 The Space M (Wpm → Wp−k )

55

The next assertion shows that Theorem 2.5.1 provides a complete characterization of M (Wpm → Wp−k ) which holds under some conditions involving k, m, p, and n. Theorem 2.5.2. Let k and m be positive integers and let either k ≥ m > 0 and k > n/p or m ≥ k > 0 and m > n/p. Then γ ∈ M (Wpm → Wp−k ) if and only if −k ∩ Wp−m (2.5.4) γ ∈ Wp,unif
,unif . In particular, if max{k, m} > n/2 then M (W2m → W2−k ) is isomorphic to .

−min{m,k} W2

Proof. It suffices to consider the case k ≥ m > 0, k > n/p , because the case m ≥ k > 0, m > n/p results by duality. Necessity. It follows from the inclusion γ ∈ M (Wpm → Wp−k ) that γ ∈ −k Wp,unif . Since M (Wpm → Wp−k ) is isomorphic to M (Wpk
→ Wp−m ), we have
−m γ ∈ Wp
,unif as well. Sufficiency. It is standard and easily proved that −k γ ∈ Wp,unif ∩ Wp−m
,unif

if and only if (2.5.1) holds with γα ∈ Lp,unif ∩ Wpk−m
,unif .  ) is isomorphic to Wpk−m Since M (Wpk
→ Wpk−m

,unif for p k > n, it follows that k−m k γα ∈ M (Wp
→ Wp
).

It remains to show that γα ∈ M (Wpm → Lp ). We choose q and r to satisfy 1/q > max{0, 1/p − m/n} > −ε + 1/q, 1/r > max{0, 1/p − (k − m)/n} > −ε + 1/r with a sufficiently small ε. Since 1/p > 1 − k/n, we have 1/p > 1/q + 1/r. By H¨older’s inequality γα uLp,unif ≤ c γα Lr,unif uLq,unif and by Sobolev’s imbedding theorem m γα uLp,unif ≤ c γα W k−m uWp,unif .
p ,unif

This means that γα ∈ M (Wpm → Lp ). The proof is completed by reference to assertion (i) of Theorem 2.5.1. 

56

2 Multipliers in Pairs of Sobolev Spaces

Remark 2.5.1. Note that by Sobolev’s imbedding theorem −k Wp−m
,unif ⊂ Wp,unif ,

k ≥ m,

if and only if either n ≤ (k − m)p or k−m 2−p ≥ . n p

n > (k − m)p,

 Under these conditions, M (Wpm → Wp−k ) is isomorphic to Wp−m
,unif if kp > n.  Analogously, if m ≥ k, mp > n and either n ≤ (m − k)p or

p−2 m−k ≥ , n p

n > (m − k)p ,

−k . then M (Wpm → Wp−k ) is isomorphic to Wp,unif

We now state a direct application of Theorem 2.5.2 to the theory of differential operators. Corollary 2.5.1. Let k and m be integers and let L(D) denote a differential operator of order m+k with constant coefficients. If either k ≥ m and kp > n, or m ≥ k > 0 and mp > n, then the operator Wpm  u → L(D)u + γ(x)u ∈ Wp−k is continuous if and only if −k γ ∈ Wp,unif ∩ Wp−m
,unif .

We conclude this subsection with a simple description of nonnegative elements of the space M (W2m → W2−m ). Theorem 2.5.3. Let γ ≥ 0. Then γ ∈ M (W2m → W2−m ) if and only if γ 1/2 ∈ M (W2m → L2 ). Moreover, γM (W m →W −m ) = γ 1/2 2M (W2m →L2 ) . 2

2

Proof. Let u ∈ v∈ We have  v dx ≤ γ 1/2 uL2 γ 1/2 vL2 ≤ γ 1/2 2M (W2m →L2 ) uW2m vW2m . γu¯ W2m ,

W2m .

Hence γM (W m →W −m ) ≤ γ 1/2 2M (W2m →L2 ) . 2

2

To obtain the converse inequality, we first note that  v dx ≤ γM (W m →W −m ) uW2m vW2m . γu¯ 2

2

2.6 The Space M (Wpm → Wql )

57

Putting here u = v, we obtain  |γ 1/2 u|2 dx ≤ γM (W m →W −m ) u2W2m . 2

2

Thus, γM (W m →W −m ) ≥ γ 1/2 2M (W2m →L2 ) . 2

2



2.6 The Space M (Wpm → Wql) In this book consideration is limited to classes of the type M (Spm → Spl ), i.e. to multipliers acting in one scale of spaces preserving the integrability degree p. Some generalizations to the pairs (Spm , Sql ) are known. This is true, in particular, for the class M (Wpm → Wql ) with nonnegative and integer m and l, which is briefly described in the present subsection. Using the same arguments as in the proof of Theorem 2.3.1, we obtain γM (Wpm →Wql ) ∼ ∇l γM (Wpm →Lq ) + γM (Wpm−l →Lq ) ,

(2.6.1)

where p, q ∈ (1, ∞). Thus, the problem is reduced to a description of the class M (Wpm → Lq ). This can be obtained from Lemmas 1.3.1 – 1.3.5. In particular, by (2.6.1) and the same lemmas we have the following assertions concerning the norm in M (Wpm → Wql ) with p < q. Theorem 2.6.1. If 1 < p < q, then ⎧ sup rm−n/p ∇l γ; Br (x)Lq + sup γ; B1 (x)Lq ⎪ ⎪ n ,r∈(0,1) ⎪ x∈Rn x∈R ⎪ ⎪ ⎪ ⎪ for mp < n, ⎪ ⎪ ⎪
⎪ ⎨ (log 2/r)1/p ∇l γ; Br (x)Lq + sup γ; B1 (x)Lq sup x∈Rn γM (Wpm →Wql ) ∼ x∈Rn ,r∈(0,1) ⎪ ⎪ ⎪for mp = n, ⎪ ⎪ ⎪ ⎪ sup γ; B1 (x)W l ⎪ q ⎪ n ⎪ ⎪ ⎩x∈R for mp > n. Applying the same arguments as in Sect. 2.2, one can derive the following result from relations (1.3.2) and (1.3.4) with p = 1. Theorem 2.6.2. If q > 1, then ⎧ sup rm−n ∇l γ; Br (x)Lq + sup γ; B1 (x)Lq ⎪ ⎪ ⎪ x∈Rn x∈Rn ,r∈(0,1) ⎪ ⎪ ⎨ for m < n, γM (W1m →Wql ) ∼ ⎪ sup γ; B1 (x)Wql ⎪ ⎪ ⎪ x∈Rn ⎪ ⎩ for m ≥ n.

58

2 Multipliers in Pairs of Sobolev Spaces

Choosing

 |∇l γ(x)|q dx

µ(e) = e

in Lemmas 1.3.4, 1.3.5 and using again the equivalence relation (2.6.1), we can obtain descriptions of M (Wpm → Wql ) with p > q, p > 1.

2.7 Certain Properties of Multipliers In this section we study some simple properties of elements of the space M (Wpm → Wpl ) with p ∈ [1, ∞). Proposition 2.7.1. The space M (Wpm → Wpl ) is contained in M (Wpm−j → Wpl−j ), j = 1, . . . , l, and γM (Wpm−j →Wpl−j ) ≤ c γM (Wpm →Wpl ) . Proof. The inequality γM (Wpm−l →Lp ) ≤ c γM (Wpm →Wpl ) is proved in Lemma 2.2.2 for p = 1 and in Lemma 2.3.4 for p > 1. It remains to use interpolation inequalities (2.3.8) and (2.2.10).  Proposition 2.7.2. If a function γ depends only on variables x1 , . . . , xs , s < n, and γ ∈ M (Wpm (Rs ) → Wpl (Rs )), then γ ∈ M (Wpm (Rn ) → Wpl (Rn )) and γ; Rn M (Wpm →Wpl ) ≤ c γ; Rs M (Wpm →Wpl ) . The proof is obvious. Proposition 2.7.3. If γ ∈ M (Wpm → Wpl ) and k is a positive integer satisfying k ≤ m/(m − l), then γ k ∈ M (Wpm → Wpm−k(m−l) ) and γ k M (W m →Wpm−k(m−l) ) ≤ c γkM (Wpm →Wpl ) . p

The proof is obvious.

2.7 Certain Properties of Multipliers

59

Proposition 2.7.4. The estimate γL∞ ≤ γM Wpl

(2.7.1)

holds. Proof. For any N = 1, 2, . . . , and arbitrary u ∈ C0∞ we have 1/N

1/N

1/N

γ N uLp ≤ γ N uW l ≤ γM Wpl uW l . p

p

Passing to the limit as N → ∞, we obtain (2.7.1).



Proposition 2.7.5. Let γ ∈ M Wpl and let σ be a segment on the real axis such that γ(x) ∈ σ for almost all x ∈ Rn . Further let f ∈ C l−1,1 (σ). Then f (γ) ∈ M Wpl and l 

f (γ)M Wpl ≤ c

f (j) ; σL∞ γjM W l . p

j=0

Proof. The assertion is obvious for l = 1. Suppose it is true for l − 1. For all u ∈ C0∞ we have u f (γ)Wpl ≤ f (γ)∇uWpl−1 + u f  (γ)∇γWpl−1 + u f (γ)Lp .

(2.7.2)

By the induction assumption, the first term on the right-hand side of (2.7.2) does not exceed c ∇uWpl−1

l−1 

f (j) ; σL∞ γjM W l−1 . p

j=0

For the same reason, the second term on the right-hand side of (2.7.2) is dominated by c u∇γWpl−1

l 

f (j+1) ; σL∞ γjM W l−1 . p

j=0

From (2.3.2) and Proposition 2.7.1 it follows that ∇γM (W l →Wpl−1 ) ≤ c γM Wpl , p

γM Wpl−1 ≤ c γM Wpl .

Hence the right-hand side of (2.7.2) is dominated by c uWpl

l 

f (j) ; σL∞ γjM W l . p

j=0

The proof is complete.



60

2 Multipliers in Pairs of Sobolev Spaces

Corollary 2.7.1. If γ ∈ M Wpl and γ −1 L∞ < ∞, then γ −1 ∈ M Wpl and l γ −1 M Wpl ≤ c γ −1 l+1 L∞ γM Wpl .

Proof. The result follows from Proposition 2.7.5 with f (γ) = γ −1 and from the inequality γ −1 L∞ γM Wpl ≥ 1 

which is a consequence of Proposition 2.7.4.

Remark 2.7.1. All the assertions of the present section can be reformulated for the space M (wpm → wpl ).

2.8 The Space M (wpm → wpl ) In this section we assume that mp < n, p ∈ (1, ∞) or m ≤ n, p = 1. Lemma 2.8.1. (i) The inequality γM (Wpm →Lp ) ≤ γM (wpm →Lp )

(2.8.1)

holds. (ii) Let ρ > 0 and let γ ∈ M (wpm → Lp ). Then lim ρ−m γ(·/ρ)M (Wpm →Lp ) = γM (wpm →Lp ) .

ρ→0

(2.8.2)

(iii) The function γ satisfies γM (wpm →Lp ) ≥ c sup rm−n/p γ; Br (x)Lp .

(2.8.3)

x∈Rn r>0

Proof. (i) We have γM (Wpm →Lp ) =

sup

u∈C0∞ (Rn )

γuLp uWpm

and (2.8.1) follows from the inequality uWpm ≥ uwpm . (ii) Clearly, ρ−m γ(·/ρ)M (Wpm →Lp ) = =

ρ−m γ(·/ρ)U Lp U Wpm U ∈C0∞ (Rn ) sup

ρ−m γ(·/ρ)u(·/ρ)Lp p p 1/p . u∈C0∞ (Rn ) (∇m u(·/ρ)Lp + u(·/ρ)Lp ) sup

(2.8.4)

2.8 The Space M (wpm → wpl )

61

The right-hand side majorizes γuLp . + ρmp upLp )1/p

(∇m upLp Hence

lim inf ρ−m γ(·/ρ)M (Wpm →Lp ) ≥ γM (wpm →Lp ) . ρ→0

Noting that

(2.8.5)

ρ−m γ(·/ρ)M (wpm →Lp ) = γM (wpm →Lp )

and using (2.8.1), we obtain lim sup ρ−m γ(·/ρ)M (Wpm →Lp ) ≤ γM (wpm →Lp ) .

(2.8.6)

ρ→0

The result follows by combining (2.8.5) and (2.8.6). (iii) Let η ∈ C0∞ (B2 ), η = 1 on B1 . The estimate (2.8.3) follows from (2.8.4) by choosing the test function u(ξ) = η((ξ − x)/r).  Lemma 2.8.2. (i) Let m ≥ l and let γ ∈ M (wpm → wpl ). Then γM (Wpm →Wpl ) ≤ c γM (wpm →wpl ) .

(2.8.7)

(ii) The inequality lim inf ||ρl−m γ(·/ρ)M (Wpm →Wpl ) ≥ γM (wpm →wpl ) ρ→0

(2.8.8)

holds. Proof. (i) Let η be the same as in the proof of Lemma 2.8.1 (iii), and let ηx (ξ) = η(x − ξ). We use the inequality γM (Wpm →Wpl ) ≤ c sup ||ηx γM (Wpm →Wpl ) , x∈Rn

and observe that the norm on the right-hand side is equal to sup

u∈C0∞

ηx γuWpl uWpm

.

In view of the imbedding wpl (Rn ) ⊂ Lpn/(n−lp) (Rn ), the norm in the numerator is equivalent to the norm ηx γuwpl . Hence γM (Wpm →Wpl ) ≤ c sup x,u

≤ c γM (wpm →wpl ) sup x,u

ηx γuwpl uWpm ηx uwpm uWpm

.

62

2 Multipliers in Pairs of Sobolev Spaces

The result follows. (ii) Obviously, for any u ∈ C0∞ (Rn ), ρl−m γ(·/ρ)M (Wpm →Wpl ) ≥

=

ρl−m γ(·/ρ)u(·/ρ)Wpl u(·/ρ)Wpm


∇l (γu)p + ρpl γup 1/p Lp Lp ∇m upLp + ρpm upLp

.

Passing to the limit as ρ → 0 we complete the proof.



Now we give a description of the space M (wpm → wpl ) which follows essentially from Corollary 2.3.3. Theorem 2.8.1. Let mp < n, m ≥ l, p ∈ (1, ∞) or m ≤ n, p = 1. Then l γ ∈ M (wpm → wpl ) if and only if γ ∈ wp,loc , ∇l γ ∈ M (wpm → Lp ), and γ ∈ L∞ (Rn ) lim r

r→∞

−n

for m = l,

γ; Br L1 = 0

for m > l.

(2.8.9)

The norm in the space M (wpm → wpl ), m > l, satisfies the equivalence relation (2.8.10) γM (wpm →wpl ) ∼ ∇l γM (wpm →Lp ) . For m = l the norm γL∞ should be added to the right-hand side of this relation. The equivalence relation γM (wpm →wpl ) ∼

l 

∇l−j γM (wpm−j →Lp )

(2.8.11)

j=0

holds. Proof. We replace γ by ρl−m γ(·/ρ) in the equivalence relation (2.3.26). Then (2.8.11) follows from Lemmas 2.8.1 and 2.8.2 as ρ → 0. We put ρl−m γ(·/ρ) as γ in (2.3.27) to obtain ρl−m γ(·/ρ)M (Wpm →Wpl )   ≤ c ∇l (ρl−m γ(·/ρ))M (Wpm →Lp ) + sup Rm−l−n ||ρl−m γ(·/ρ)BR (x)L1 . x∈Rn , R>0

2.9 Multipliers in Spaces of Functions with Bounded Variation

63

Since the second term on the right is equal to sup rm−l−n γ; Br (x)L1 , x,r

and the first term tends to ∇l γM (wpm →Lp ) as ρ → 0 by (2.8.2), the reference to (2.8.8) gives   γM (wpm →wpl ) ≤ c ∇l γM (wpm →Lp ) + sup rm−l−n γ; Br (x)L1 . (2.8.12) x∈Rn , r>0

It remains to remove the second term on the right-hand side in the case m > l. Consider the case p ∈ (1, ∞). We use the inequality  l/m 1−l/m  |γ(ξ)| ≤ c (M∇l γ)(ξ) sup rm−n/p ∇l γ; Br (x)Lp (2.8.13) x∈Rn , r>0

which follows from (2.3.24) with the term γL1,unif on the right-hand side omitted due to condition (2.8.9). Integrating (2.8.13) over an arbitrary ball Br (x), we arrive at l/m γ; Br (x)L1 ≤ c ( sup rm−n/p ∇l γ; Br (x)Lp (M∇l γ)1−l/m ; Br (x)L1 . x∈Rn r>0

By H¨older’s inequality rm−l−n γ; Br (x)||L1 l 1− ml  n n ≤ c sup rm− p ∇l γ; Br (x)Lp m rm− p M∇l γ; Br (x)Lp .

(2.8.14)

x∈Rn r>0

In view of Lemma 2.3.8, rm−n/p M∇l γ; Br (x)Lp ≤ c sup e

M∇l γ; eLp ≤ c ∇l γM (wpm →Lp ) , (Cp,m (e))1/p

which along with (2.8.14) leads to sup rm−l−n γ; Br (x)L1 ≤ c ∇l γM (wpm →Lp ) .

x∈Rn , r>0

The result follows by (2.8.12). For p = 1 the second term on the right-hand side of (2.8.12) is dominated  by c ∇l γM (w1m →L1 ) by Theorem 1.1.1.

2.9 Multipliers in Spaces of Functions with Bounded Variation In the 1960s, the family of differentiable functions was complemented by the space bv of functions with bounded variation which turned to be useful in geometric measure theory, the calculus of variations and the theory of quasilinear partial differential equations.

64

2 Multipliers in Pairs of Sobolev Spaces

A function u locally integrable on Rn has bounded variation if its gradient, understood in the sense of generalized functions, is a vector charge. In other words, the functional (C0∞ )n  g → (u, div g) satisfies the estimate |(u, div g)| ≤ C max |g|, where C is a constant independent of g. Let 0 < ρ < R and let u ¯(r) stand for the mean value of u on the sphere ∂Br . We introduce gε (x) = x |x|−n ηε (x), where ηε is a smooth function equal to one on BR−ε \Bρ+ε and to zero on BR \Bρ , 0 ≤ ηε ≤ 1. We have |(u, div gε )| ≤ (var ∇u)(BR \Bρ ). Making simple calculations and passing to the limit as ε → 0, we find that |¯ u(R) − u ¯(ρ)| ≤ c (var ∇u)(BR \Bρ ) for almost all ρ and R with ρ < R. Therefore, any function with finite variation has the limit u∞ = lim u ¯(r). r→∞

The set of functions with finite variation for which u∞ = 0 is called the space bv. The norm in bv is defined as the variation of the charge ∇u. Endowed with this norm, bv becomes a Banach space. Not every function in bv can be approximated by functions in C0∞ in the semi-norm ubv , because the completion of C0∞ in this semi-norm is the Sobolev space w11 . However, functions in bv can be approximated by functions in C0∞ in the following weak sense. If u ∈ bv, then there exists a sequence {um }m≥1 of functions in C0∞ such that um → u in L1,loc and  (2.9.1) |∇um | dx = ubv . lim m→∞

The Banach space BV = bv ∩ L1 , endowered with the norm uBV = ubv + uL1 , possesses a similar property. The existence of the sequence {um } and the classical isoperimetric inequality  (n−1)/n −1/n  ≤ n−1 mesn (B1 ) s(∂g), (2.9.2) mesn (g) where g is an arbitrary open subset of Rn with compact closure and smooth boundary, imply the inequality

2.9 Multipliers in Spaces of Functions with Bounded Variation

 −1/n uLn/(n−1) ≤ n−1 mesn (B1 ) ubv ,

u ∈ bv.

By (2.9.1), we may assume that u ∈ C0∞ . We have  ∞ n/(n−1) uLn/(n−1) = mesn (Nt ) d(tq ),

65

(2.9.3)

(2.9.4)

0

where q = n/(n − 1) and Nt = {x : |u(x)| ≥ t}. The last integral has the estimate  ∞
 t  ∞ q−1 q−1 mesn (Nt )t dt ≤ q mesn (Nτ )1/q dτ mesn (Nt )1/q dt q 0

0




0

∞

=

1/q q mesn (Nt ) dt .

0

This fact and (2.9.2) imply the estimate −1

uLn/(n−1) ≤ n

 −1/n mesn (B1 )



∞

s(∂Nt )dt, 0

which is equivalent to (2.9.3) in view of (1.1.1). Setting the characteristic function of a ball into (2.9.3), we conclude that the constant in (2.9.3) is sharp. This inequality with a non-sharp constant was proved first in [Gag1] by a different method. The space bv is closely related to the notion of the perimeter of a set which, to a large extent, is the reason for its importance in analysis. The perimeter P (E), in the sense of Cacciopoli and De Giorgi, of a Lebesgue measurable set E ⊂ Rn is defined by P (E) = inf lim inf s(∂Πm ), Πm m→∞

where {Πm } is a sequence of polyhedra converging in volume to E. This means that the volume of the symmetric difference (Πm \E) ∪ (E\Πm ) tends to zero. This definition, combined with (2.9.2) implies the isoperimetric inequality min

!

(n−1)/n  (n−1)/n " , mesn (Rn \E) mesn (E)  −1/n P (E). ≤ n−1 mesn (B1 )

The characteristic function χE of a set E belongs to bv if and only if P (E) < ∞. Moreover, P (E) = χE bv (see [Fe3]).

66

2 Multipliers in Pairs of Sobolev Spaces

The perimeter P (E) does not exceed the Hausdorff measure s(∂E); in particular, the inequality P (E) < s(∂E) is not excluded. The following generalization of the notion of the normal to a smooth surface is useful for revealing a deeper relation between the perimeter and the measure s. A unit vector ν is called the exterior normal to a set E at a point x in the sense of Federer if ! " lim ρ−n mesn y : y ∈ E ∩ Bρ (x), (y − x)ν > 0 = 0, ρ→0

! " lim ρ−n mesn y : y ∈ Bρ (x)\E, (y − x)ν < 0 = 0.

ρ→0

The set of all points x ∈ ∂E for which the normal to E exists is called the reduced boundary of the set E and is denoted by ∂ ∗ E. A set E is said to be a set with a local finite perimeter if P (E ∩ Br (x)) < ∞ for all balls Br (x). The following assertion is a crucial result in the theory of perimeter (see [Fe3]). Theorem 2.9.1. If E is a set with a locally finite perimeter, then its reduced boundary ∂ ∗ E is measurable with respect to s and var ∇χE . Moreover, var ∇χE (Rn \∂ ∗ E) = 0, and for any set L ⊂ ∂ ∗ E



(∇χE )(L) = −

ν(x) s(dx). L

This implies that P (E) = s(∂ ∗ E). Note that for any u ∈ bv the generalization of (1.1.1) holds:  ∞ ubv = P (Mt ) dt, −∞

where Mt = {x : u(t) > t} (see [FR]). 2.9.1 The Spaces M bv and M BV The next assertion contains a description of the space M bv.

(2.9.5)

2.9 Multipliers in Spaces of Functions with Bounded Variation

67

Theorem 2.9.2. A function γ belongs to the space M bv if and only if γ ∈ bvloc ∩ L∞ and for any ball Br (x) var ∇γ(Br (x)) ≤ c rn−1 .

(2.9.6)

The relation γM bv ∼ sup r1−n var ∇γ(Br (x)) x∈Rn , r>0

holds. Proof. Sufficiency. Let u ∈ bv and let γ be a function in bvloc ∩ L∞ subject to (2.9.6). By {um } we denote a sequence of functions in C0∞ convergent in L1,loc for which (2.9.1) holds. Then, for any ϕ ∈ C0∞ , |(∇ϕ, γu)| = | lim (∇ϕ, γum )| m→∞

≤ | lim sup(ϕ, um ∇γ)| + | lim sup(ϕ, γ∇um )|. m→∞

m→∞



Hence |(∇ϕ, γu)| ≤ ϕL∞ lim sup m→∞

|um | var ∇γ(dx) 

+ ϕL∞ γL∞ lim sup

m→∞

|∇um |dx.

Applying Corollary 1.1.2 to the first integral, we find that   |(∇ϕ, γu)| ≤ ϕL∞ c sup r1−n var ∇γ(Br (x)) + γL∞ ubv . x∈Rn , r>0

Thus the sufficiency and the upper estimate for the norm γM bv are proved. Necessity. If γ ∈ M bv, then for any N = 1, 2, . . . γ N vbv ≤ γN M bv vbv . Applying inequality (2.9.3), we find that 1/N

1/N

γ N vLn/(n−1) ≤ γM bv vbv . Passing to the limit as N → ∞, we conclude that γL∞ ≤ γM bv . Note that a mollification vh of the characteristic function of the ball B(1+ε)r (y) obeys the inequality   |vh |var ∇γ(dx) ≤ γvh bv + |γ| |∇vh | dx.

68

2 Multipliers in Pairs of Sobolev Spaces

Passing to the limit as ε → 0, we obtain   var ∇γ(Br (y)) ≤ γM bv + γL∞ rn−1 . Thus, r1−n var ∇γ(Br (y)) ≤ 2 γM bv . 

The proof is complete.

We formulate a similar result for the space M BV which is proved in the same way. Theorem 2.9.3. A function γ belongs to the space M BV if and only if γ ∈ bvloc ∩ L∞ and, for any ball Br with r ∈ (0, 1), (2.9.6) holds. Moreover, γM BV ∼ sup r1−n var ∇γ(Br (x)). x∈Rn , r∈(0,1)

Theorems 2.9.2 and 2.9.3 imply the following necessary and sufficient conditions for inclusion of the characteristic function χE of a set E ⊂ Rn into M bv and M BV . Corollary 2.9.1. (i) The function χE belongs to the space M bv if and only if s(Br (x) ∩ ∂ ∗ E) ≤ c rn−1

(2.9.7)

for any set E with a locally finite perimeter and any ball Br (x). (ii) Adding to this statement the condition r ∈ (0, 1), one gets a necessary and sufficient condition for the inclusion χE ∈ M BV . Proof. It is enough to refer to Theorems 2.9.2 and 2.9.3, noting additionally that (2.9.7) is equivalent to var ∇χE (Br (x)) ≤ c rn−1 . Indeed, by Theorem 2.9.1 s(Br (x) ∩ ∂ ∗ E) = var ∇χE (Br (x) ∩ ∂ ∗ E) = var ∇χE (Br (x)). The proof is complete.



3 Multipliers in Pairs of Potential Spaces

In this chapter we study the space of multipliers M (Hpm → Hpl ) and M (hm p → hlp ), m ≥ l ≥ 0, where Hps and hsp are the space of Bessel and Riesz potentials of order s with densities in Lp . (The case m < l is not interesting, since then M (Hpm → Hpl ) = {0}, as can be shown by the argument used for Sobolev spaces in Sect. 2.1.) The introductory Sect. 3.1 gives information on Bessel linear and nonlinear potentials, on capacity and on imbedding theorems, which is used in subsequent sections. A characterization of the spaces M (Hpm → Hpl ) and l M (hm p → hp ) is given in Sects. 3.2 and 3.7. In Sects. 3.3 and 3.4 we obtain either necessary or sufficient conditions for a function to belong to M (Hpm → Hpl ), formulated in terms of different classes of functions. In Sect. 3.5 certain properties of elements of M (Hpm → Hpl ) are studied. In particular, we consider the imbedding of M (Hpm → Hpl ) into M (Hqm−j → Hql−j ). Descriptions of the point, residual, and continuous spectra of multipliers in Hpl and Hp−l
are given in 3.6. Finally, Sect. 3.8 contains a characterization of posl itive homogeneous elements of the spaces M (Hpm → Hpl ) and M (hm p → hp ).

3.1 Trace Inequality for Bessel and Riesz Potential Spaces Let µ be a Radon measure in Rn and let Spl be a certain space of Sobolev type with p and l being the integrability and smoothness parameters, p ≥ 1, l > 0. As in Chap. 2, characterizations of the space M (Spm → Spl ), m ≥ l, to be obtained in the sequel are based on necessary and sufficient conditions ensuring the trace inequality
 1/p |u|p dµ ≤ c uSpl , Rn

V.G. Maz’ya, T.O. Shaposhnikova, Theory of Sobolev Multipliers, Grundlehren der mathematischen Wissenschaften 337, c Springer-Verlag Berlin Hiedelberg 2009


69

70

3 Multipliers in Pairs of Potential Spaces

where u is an arbitrary function in C0∞ (Rn ). In this section we present such conditions for the spaces of Bessel and Riesz potentials. 3.1.1 Properties of Bessel Potential Spaces Here we survey some known facts on Bessel potential spaces to be used in the sequel. Given any real µ, we put Λµ = (−∆ + 1)µ/2 = F −1 (1 + |ξ|2 )µ/2 F , where F is the Fourier transform in Rn . Let 1 < p < ∞, m ≥ 0. We introduce the space Hpm of Bessel potentials as the completion of C0∞ with respect to the norm uHpm = Λm uLp . If m is integer then, according to Lemma 1.2.3, the spaces Wpm and Hpm are isomorphic. It follows from the definition of Hpm that the function u belongs to Hpm if and only if u = Λ−m f , where f ∈ Lp . Replacing Λm in the definition of Hpm by the operator (−∆)m/2 , we arrive m at the definition of the space hm p . It is known that the space hp with mp < n is isomorphic to the space of Riesz potentials of order m with density in Lp (see [MH2]). Definition 3.1.1. We define (Sm u)(x) = |∇m u(x)| for integer m ≥ 0 and
 ∞  2 1/2 (Sm u)(x) = |∇[m] u(x+θy)−∇[m] u(x)|dθ y −1−2{m} dy (3.1.1) 0

B1

for noninteger m > 0. We present without proof a normalization of potential spaces due to Strichartz [Str]. Theorem 3.1.1. The equivalence relations hold: ∼ Sm uLp , uhm p

(3.1.2)

uHpm ∼ Sm uLp + uLp .

(3.1.3)

The last formula implies the following uniform localization property for the space Hpm (see [Str]). Theorem 3.1.2. Let {B (j) }j≥0 be a covering of Rn by balls with unit diameter. Let this covering have a finite multiplicity, depending only on n. Further, let O(j) be the centre of B (j) , O(0) = 0 and ηj (x) = η(x − O(j) ), where η ∈ C0∞ (2B (0) ) and η = 1 on B (0) . Then
 1/p uHpm ∼ uηj pHpm . (3.1.4) j≥0

We formulate the Sobolev imbedding theorem for the space Hpm .

3.1 Trace Inequality for Bessel and Riesz Potential Spaces

71

Theorem 3.1.3. (i) If mp < n, p ≤ q ≤ np/(n−mp) or mp = n, p ≤ q < ∞, then for all u ∈ Hpm uLq ≤ c uHpm . (ii) If mp > n, then for all u ∈ Hpm uL∞ ≤ c uHpm . The following generalization of the function Sl u is the nonlinear operator of fractional differentiation defined by Polking [Pol1]:
 ∞
 θ/q dρ 1/θ l u)(x) = |u(x + ρy) − u(x)|q dy . (Sq,θ ρ1+lθ 0 B1 Lemma 3.1.1. [Pol1] Let 0 < l < 1, 1 ≤ q ≤ θ, 2 ≤ θ < ∞. If 1 < p < ∞ and p > nq/(n + lq), then there exists a constant c depending only on l, q, p, θ, n such that l u≤ c uHpl Sq,θ for all u ∈ Hpl . 3.1.2 Properties of the (p, m)-Capacity For positive noninteger m the (p, m)-capacity is defined by (1.2.6), i.e. Cp,m (e) = inf{f pLp : f ∈ Lp , f ≥ 0, Jm f ≥ 1 on e}. Similarly to (1.2.7), one has Cp,m (e) ∼ inf{upHpm : u ∈ C0∞ , u ≥ 1 on e}. Another capacity, introduced for any positive and noninteger m, is cp,m (e) = inf{f pLp : f ∈ Lp , f ≥ 0, Im f ≥ 1 on e}. We list certain ‘metric’ properties of these capacities which will be used later. For integer m > 0 they were proved in Propositions 1.2.2, 1.2.4, and 1.2.6. Proposition 3.1.1. If mp > n, then for all compact sets e = ∅ with d(e) ≤ 1 the relation Cp,m (e) ∼ 1 holds. The proof follows from part (ii) of Theorem 3.1.3. Proposition 3.1.2. If mp < n, then Cp,m (e) ≥ c (mesn e)(n−mp)/n . The proof follows from part (i) of Theorem 3.1.3.

(3.1.5)

72

3 Multipliers in Pairs of Potential Spaces

Proposition 3.1.3. If mp = n and d(e) ≤ 1, then
Cp,m (e) ≥ c log

2n 1−p . mesn e

(3.1.6)

Proof. See Proposition 1.2.4. Proposition 3.1.4. (i) If mp < n and 0 < r ≤ 1, then Cp,m (Br ) ∼ rn−mp . (ii) If mp = n, 0 < r ≤ 1, then Cp,m (Br ) ∼ (log 2/r)1−p . (iii) If r > 1, then Cp,m (Br ) ∼ rn . For the proof of these relations see [AH], Sect. 5.1. In the next proposition {B (j) } is the same covering as in Theorem 3.1.2. Proposition 3.1.5. For any compact e  Cp,m (e) ∼ Cp,m (e ∩ B (j) ). j≥0

Proof. Let u ∈ C0∞ , u ≥ 1 on e. From the definition of Cp,m it follows that   Cp,m (e ∩ B (j) ) ≤ uηj pHpm , j≥0

j≥0

where {ηj }j≥0 is the sequence defined in Theorem 3.1.2. By (3.1.4), the righthand side is dominated by c upHpm . Minimizing this value, we obtain the lower bound for Cp,m (e). The required upper bound is a direct corollary of the semi-additivity of the capacity. Proposition 3.1.6. Let mp < n and let e be a compact subset of Rn of diameter d(e) ≤ 1. Then (3.1.7) Cp,m (e) ∼ cp,m (e). Proof. The estimate Cp,m (e) ≥ cp,m (e) is obvious. We prove that Cp,m (e) ≤ c cp,m (e). By definition of the capacity cp,m (e), for any ε > 0 there exists a function u ∈ C0∞ such that u ≥ 1 on e and Sm upLp ≤ cp,m (e) + ε.

(3.1.8)

3.1 Trace Inequality for Bessel and Riesz Potential Spaces

73

We introduce a function η ∈ C0∞ (B2 ) such that η ≥ 1 on e. Then    ≤ c Sm upLp +  |x|k−m ∇k upLp [m]

Sm (ηu)pLp

ηupLp

+

k=0

c Sm upLp

by Hardy’s inequality. Reference to estiwhich does not exceed mate (3.1.8) completes the proof.  Corollary 3.1.1. Let mp < n and let e be a compact subset of Rn of diameter d(e) ≤ 1. Then, for any δ ∈ (0, 1), Cp,m (e) ∼ δ n−mp Cp,m (δ e).

(3.1.9)

Proof. On the one hand, Cp,m (e) ≤ δ n−mp Cp,m (δ e) by dilation. On the other hand, Cp,m (e) ≥ cp,m (e) = δ n−mp cp,m (δ e). 

Reference to Proposition 3.1.6 completes the proof.

We give an estimate for the integral of the capacity of a set bounded by a level surface which contains (1.2.27) as a particular case. For the history and proof see [AH], Ch. 7. Proposition 3.1.7. (i) Let u ∈ Hpm , p ∈ (1, ∞), m > 0, and let Nt = {x : |u(x)| ≥ t}. Then



∞

Cp,m (Nt )tp−1 dt ≤ c upHpm .

(3.1.10)

(ii) If u ∈ hm p , p ∈ (1, n/m), m > 0, then  ∞ cp,m (Nt )tp−1 dt ≤ c uphm .

(3.1.11)

0

0

p

3.1.3 Main Result Here we present a generalization to Hpm of Theorem 1.2.2 on integrability of functions in Wpm with respect to a measure µ. Using (3.1.10) and (3.1.11), we can prove the following theorem in the same way as Theorem 1.2.2.

74

3 Multipliers in Pairs of Potential Spaces

Theorem 3.1.4. (i) The best constant C in  |u|p dµ ≤ C upHpm ,

u ∈ C0∞ ,

(3.1.12)

is equivalent to sup e

µ(e) , Cp,m (e)

where e is an arbitrary compact set of positive capacity Cp,m (e). (ii) Let p > 1 and mp < n. The best constant C in  |u|p dµ ≤ C uphm , u ∈ C0∞ , p

(3.1.13)

is equivalent to sup e

µ(e) , cp,m (e)

where e is an arbitrary compact set of positive capacity cp,m (e). Remark 3.1.1. From Proposition 3.1.5 it follows that sup e

µ(e) µ(e) ∼ sup . Cp,m (e) {e:d(e)≤1} Cp,m (e)

(3.1.14)

Remark 3.1.2. Obviously,the constantC in Theorem3.1.4satisfies the inequality C≥

sup x∈Rn ,ρ∈(0,1/2)

µ(Bρ (x)) . Cp,m (Bρ (x))

(3.1.15)

If the converse estimate (up to a factor c = c(n, p, l)) were valid we could avoid the notion of capacity in this book (see Proposition 3.1.4/4). According to Proposition 3.1.1, this is really the case for mp > n when the right-hand and left-hand sides of (3.1.15) are equivalent to sup{µ(B1 (x)) : x ∈ Rn }. However, as was noted by Adams [Ad3], the finiteness of the right-hand side of (3.1.15) for n ≥ mp does not imply C < ∞. Before we prove the last assertion, we recall the definition of the Hausdorff ϕ-measure of a set E ⊂ Rn , where ϕ is a non-decreasing positive function on [0, 1]: namely,  H(E, ϕ) = lim inf ϕ(ri ) . ε→+0 {B(i) }

i

Here {B (i) } is any covering of E by open balls B (i) with radii ri < ε. We put ϕ(t) = tn−mp for n > mp and ϕ(t) = | log t|1−p for n = mp. Let E be a Borel set in Rn such that its diameter d(E) satisfies d(E) < 1 and 0 < H(E, ϕ) < ∞.

3.2 Description of M (Hpm → Hpl )

75

We may assume E to be closed and bounded, since any Borel set of positive Hausdorff measure contains a closed subset with the same property. By Frostman’s theorem (see Carleson [Car], Theorem 1, Ch. 2) there exists a non-zero measure µ with support in E such that µ(B (x)) ≤ c ϕ(ρ) with a constant c independent of x and ρ. By virtue of Proposition 3.1.4, this means that the right-hand side of (3.1.15) is finite. On the other hand, by the theorem of Meyers [Me] and Havin and Maz’ya [MH1], [MH2], the finiteness of the measure H(E, ϕ) implies Cp,m (E) = 0. So C = ∞ (see Theorem 3.1.4), although the right-hand side of (3.1.15) is finite. Remark 3.1.3. According to [KeS], [MV1], and [Ver1], the criteria for (1.2.31) formulated at the end of Sect. 1.2.5 for integer m hold for all positive m. The same concerns the Fefferman-Phong sufficient condition dealt with in Sect. 1.2.6. For surveys of these and related results see [Ver2] and [Ver3].

3.2 Description of M (Hpm → Hpl ) 3.2.1 Auxiliary Assertions We formulate the Calderon interpolation theorem [Ca1] for spaces of Bessel potentials, of which a particular case is the inequality (2.3.8) used in the study of the space M (Wpm → Wpl ). Proposition 3.2.1. Let p0 , p1 ∈ (1, ∞), θ ∈ (0, 1), µ ∈ R1 and 1 θ 1−θ = + , p p0 p1

l = θl0 + (1 − θ)l1 .

Further, let L be a linear operator mapping Hpl00+µ ∩ Hpl11+µ

into

Hpl00 ∩ Hpl11

and admitting an extension to continuous operators: Hpl00+µ → Hpl00

and

Hpl11+µ → Hpl11 .

Then L can be extended to a continuous operator: Hpl+µ → Hpl , and the interpolation inequality LHpl+µ →H l ≤ c LθH l0 +µ →H l0 L1−θ l1 +µ p

holds.

p0

p0

Hp1

l

→Hp11

76

3 Multipliers in Pairs of Potential Spaces

Setting L = γ in this proposition, we obtain γM (Hpl+µ →H l ) ≤ c γθM (H l0 +µ →H l0 ) γ1−θ p

p0

p0

l +µ

M (Hp11

l

→Hp11 )

.

(3.2.1)

l Lemma 3.2.1. Let γρ be a mollification of γ ∈ Hp,loc (Rn ), 1 < p < ∞, with kernel K ≥ 0 and radius ρ. Then

γρ M (Hpm →Hpl ) ≤ γθM (Hpm →Hpl ) ≤ lim inf γρ M (Hpm →Hpl ) . ρ→0

(3.2.2)

Proof. Let u ∈ C0∞ and {l} > 0. By (3.1.3) #  ∞ $  ρ−n K(ξ/ρ) γρ uHpl ≤ c 0

B1

%2 p/2 &1/p −1−2{l} ×∇[l],x (Q(x + θy, ξ) − Q(x, ξ)) dξ dθ y dy dx p &1/p #  −n +c , ρ K(ξ/ρ)Q(x, ξ) dξ dx where Q(x, ξ) = γ(x − ξ)u(x). By the Minkowski inequality   −n γρ uHpl ≤ ρ K(ξ/ρ)Sl Q(·, ξ)Lp dξ + ρ−n K(ξ/ρ)Q(·, ξ)Lp dξ . Since Q(·, ξ)Hpl ≤ γM (Hpm →Hpl ) uHpm , the left estimate (3.2.2) follows. The right inequality (3.2.2) results from γuHpl = lim γρ uHpl ≤ lim inf γρ M (Hpm →Hpl ) uHpm . ρ→0

ρ→0

For the case {l} = 0 see Lemma 2.3.1.



3.2.2 Imbedding of M (Hpm → Hpl ) into M (Hpm−l → Lp ) In Theorem 3.1.4 and Remark 3.1.1 the following description of the space M (Hpk → Lp ) is contained: Lemma 3.2.2. The relations γM (Hpk →Lp ) ∼ sup e

and γM (Hpk →Lp ) ∼

γ; eLp [Cp,k (e)]1/p

γ; eLp 1/p {e:d(e)≤1} [Cp,k (e)]

hold, where d(e) is the diameter of e.

sup

3.2 Description of M (Hpm → Hpl )

77

Lemma 3.2.3. The inequality γM (Hpm−l →Lp ) ≤ c γM (Hpm →Hpl )

(3.2.3)

holds. Proof. Let γ ∈ M (Hpm → Hpl ) and let γρ be a mollification of γ with radius ρ. Since M (Hpm → Hpl ) ⊂ M (Hpm → Lp ), then γ ∈ Lp,unif . Therefore γρ ∈ L∞ and consequently γρ ∈ M (Hpm−l → Lp ). This property of mollifications will be used in what follows. 1. The case m ≥ 2l. Let u = Jm−l f , f ∈ Lp . By Lemma 1.2.5, |u| ≤ c (Jm |f |)1−l/m (Mf )l/m . Hence

l/m

1−l/m

γρ uLp ≤ c f Lp γρl/(l−m) γρ Jm |f |Lp

.

This inequality and Lemma 3.2.2 imply that ⎛

γρ uLp

⎞(m−l)/pm pl/(m−l) |γ | dx ⎜ e ρ ⎟ l/m 1−l/m ⎟ ≤ c f Lp γρ Jm |f |H l sup ⎜ . ⎝ ⎠ p C (e) e p,l

We use Lemma 2.3.6, which is valid for all ν and µ, 0 < ν < µ, with ϕ = |γρ |1/(m−l) , ν = l, µ = m − l. Then the last supremum does not exceed  c sup e

γρ ; eLp [Cp,m−l (e)]1/p

l/m

which, together with Lemma 3.2.2, gives l/m

1−l/m

l/m

1−l/m

γρ uLp ≤ c γρ M (H m−l →L ) γρ M (H m →H l ) f Lp Jm |f |Hpm p

p

p

l/m

p

1−l/m

≤ c γρ M (H m−l →L ) γρ M (H m →H l ) uHpm−l . p

p

p

p

Then reference to Lemma 3.2.1 yields (3.2.3) for m ≥ 2l. 2. Suppose that m = l. For any positive integer N we have 1/N

1/N

1/N

p

p

γ N uLp ≤ γ N uH l ≤ γM Hpl uH l . Consequently, γM Lp = γL∞ ≤ γM Hpl . 3. Now let 2l > m > l. By ε we denote a positive number such that ε < m − l. Since m − l + ε > 2ε, it follows from the first part of the proof that γρ M (Hpm−l →Lp ) ≤ c γρ M (Hpm−l+ε →H ε ) . p

78

3 Multipliers in Pairs of Potential Spaces

By (3.2.1) we have 1−ε/l

ε/l

γρ M (Hpm−l+ε →H ε ) ≤ c γρ M (H m−l →L ) γρ M (H m →H l ) p

p

p

p

p

which, together with the preceding estimate and Lemma 3.2.1, implies (3.2.3). 

3.2.3 Estimates for Derivatives of a Multiplier l−|α|

Lemma 3.2.4. If γ ∈ M (Hpm → Hpl ), then Dα γ ∈ M (Hpm → Hp multi-index α with |α| ≤ l. The estimate

) for any

Dα γM (H m →Hpl−|α| ) ≤ c γM (Hpm →Hpl )

(3.2.4)

p

holds. Proof. It suffices to consider the case |α| = 1, l ≥ 1. Obviously, u∇γHpl−1 ≤ uγHpl + γ∇uHpl−1   ≤ γM (Hpm →Hpl ) + γM (Hpm−1 →Hpl−1 ) uHpm . Using (3.2.1) and (3.2.3), we obtain 1−1/l

1/l

γM (Hpm−l →Hpl−1 ) ≤ c1 γM (H m−l →L ) γM (H m →H l ) p

p

p

p

≤ c2 γM (Hpm →Hpl ) . Consequently, u∇γHpl−1 ≤ c γM (Hpm →Hpl ) uHpm .  This result and Lemma 3.2.3 give: m−l+|α|

Corollary 3.2.1. If γ ∈ M (Hpm → Hpl ), then Dα γ ∈ M (Hp any multi-index α of order |α| ≤ l. The inequality Dα γM (Hpm−l+|α| →Lp ) ≤ c γM (Hpm →Hpl ) holds.

→ Lp ) for

3.2 Description of M (Hpm → Hpl )

79

3.2.4 Multiplicative Inequality for the Strichartz Function Lemma 3.2.5. Let m ≥ l, 0 < δ < l < 1. Then 1−δ/m  δ/m Sl−δ γ ≤ c Sl γ + γM (Hpm−l →Lp ) γM (H m−l →L

p)

p

.

(3.2.5)

Proof. For any R > 0
 (Sl−δ γ)(x) ≤ c Rδ

R

∞
 B1

R

|γ(x + θy) − γ(x)| dθ

B1

0

 +

2



|γ(x + θy)| dθ

2

y −1−2(l−δ) dy

c |γ(x)| ≤

B1

1/2

1/2

 + |γ(x)|Rδ−l .





Since

y −1−2l dy

|γ(x + θy) − γ(x)| dθ +

B1

|γ(x + θy)| dθ ,

it follows that c R−l |γ(x)| ≤




∞


R


 +

B1

|γ(x + θy) − γ(x)| dθ

∞


B1

R

|γ(x + θy)| dθ

2

2

y −1−2l dy

y −1−2l dy

1/2

1/2 .

Henceforth we assume that R ≤ 1. We have 
 1
 2 1/2 (Sl−δ γ)(x) ≤ c Rδ (Sl γ)(x) + |γ(x + θy)| dθ y −1−2(l−δ) dy
 +

∞


y −n

1

 By

R

B1

|γ(x + s)| ds

 ≤ c Rδ (Sl γ)(x) + Rδ−m

2

y −1−2(l−δ) dy

1/2 

γ; Bρ (x)Lp n/p−m+l x∈Rn ,ρ∈(0,1) ρ sup

 + sup γ; B1 (x)Lp . x∈Rn

It is clear that the last term can be thrown away by changing the constant c. By Lemma 3.2.2, sup x∈Rn ,ρ∈(0,1)

γ; Bρ (x)Lp ≤ c γM (Hpm−l →Lp ) . ρn/p−m+l

Thus, for all R ∈ (0, 1]   (Sl−δ γ)(x) ≤ c Rδ (Sl γ)(x) + R−m γM (Hpm−l →Lp ) .

(3.2.6)

80

3 Multipliers in Pairs of Potential Spaces

If (Sl−δ )γ(x) ≤ γM (Hpm−l →Lp ) , we arrive at (3.2.5) by putting R = 1 in (3.2.6). In the opposite case, (3.2.5) follows from (3.2.6), with Rm =

γM (Hpm−l →Lp ) (Sl γ)(x)

. 

The lemma is proved.

3.2.5 Auxiliary Properties of the Bessel Kernel Gl We formulate some asymptotic properties of the kernel Gl of the Bessel potential which will be used later (see [AMS]): (i) For |x| → 0, Gl (x) ∼ |x|l−n

if

0 δ. Proof. It suffices to consider the case |y| > |x|. If 2|x − y| < |x|, then by (3.2.10), for |x| < 2, # c |x − y||x|l−n−1 , l ≤ n + 1 , |Gl (x) − Gl (y)| ≤ c |x − y| , l ≥n+1, and for |x| ≥ 2 we deduce from (3.2.11) that |Gl (x) − Gl (y)| ≤ c |x − y||x|(l−n−1)/2 e−|x|/2 . These estimates and (3.2.12) yield |Gl (x) − Gl (y)| ≤ c |x − y|δ Gl−δ (x/4)

3.2 Description of M (Hpm → Hpl )

81

for 2|x − y| < |x|. Now let 2|x − y| > |x|. If l > n and |y| < 2, then by (3.2.9) we have |Gl (x) − Gl (y)| ≤ c |y|min(l−n,1) which, together with |y| ≤ 3 |x − y|, gives |Gl (x) − Gl (y)| ≤ c |x − y|δ |y|min(l−n,1)−δ . Combining this estimate with (3.2.7)–(3.2.9), we obtain (3.2.13) in the case l > n. For the same values of x and y we have |Gl (x) − Gl (y)| ≤ c1 | log(|x|/|y|)| + c2 ≤ c (|y|/|x|)δ ≤ c |x − y|δ |x|−δ if l = n, and |Gl (x) − Gl (y)| ≤ c (|x|l−n + |y|l−n ) ≤ c |x − y|δ (|x|l−n−δ + |y|l−n−δ ) if l < n. Using (3.2.8) again, we arrive at (3.2.13). It remains to deal with the case 2|x − y| ≥ |x|,

|y| ≥ 2.

By (3.2.7)–(3.2.9) and (3.2.11) we obtain |Gl (x) − Gl (y)| ≤ c Gl (x) ≤ c |x − y|δ |x|−δ Gl (x) ≤ c |x − y|δ Gl−δ (x/4) for |x| > 1. If |x| < 1, then |x − y| ≥ 1 and therefore |Gl (x) − Gl (y)| ≤ c Gl (x) ≤ c |x − y|δ Gl (x) ≤ c |x − y|δ Gl−δ (x/4) . The proof is complete.



3.2.6 Upper Bound for the Norm of a Multiplier We obtain a sufficient condition for a function to belong to the space M (Hpm → Hpl ). l Lemma 3.2.7. Let γ ∈ Hp,loc , Sl γ ∈ M (Hpm → Lp ), and γ ∈ M (Hpm−l → Lp ). Then γ ∈ M (Hpm → Hpl ) and

γM (Hpm →Hpl ) ≤ c (Sl γM (Hpm →Lp ) + γM (Hpm−l →Lp ) ) . (The function Sl was introduced in Definition 3.1.1.)

(3.2.14)

82

3 Multipliers in Pairs of Potential Spaces

Proof. By Theorem 3.1.1 one can easily verify that a function in L∞ with uniformly bounded derivatives of any order belongs to the space M (Hpm → Hpl ). Let γρ be a mollification of γ with nonnegative kernel and radius ρ. Since M (Hpm → Hpl ) ⊂ M (Hpm → Lp ), it follows that γ ∈ Lp,unif . Therefore ∇j γρ ∈ L∞ , j = 0, 1, . . . , and γρ ∈ M (Hpm → Hpl ). The assertion was proved for the case {l} = 0 in Lemma 2.3.5. Suppose that {l} > 0. For any u ∈ C0∞ , γρ uLp ≤ γρ M (Hpm−l →Lp ) uHpm−l . Clearly,



Sl (γρ u)Lp ≤ c

(3.2.15)

S{l} (Dβ γρ Dα−β u)Lp .

0≤β≤α |α|=[l]

First consider the terms corresponding to multi-indices of order |β| = j < [l]. By Lemma 3.2.4 and interpolation inequality (3.2.1), for any k ∈ (0, l − j) we have ∇j γρ M (Hpm−k →Hpl−k−j ) ≤ c γρ M (Hpm−k →Hpl−k ) 1−k/l

k/l

≤ c γρ M (H m →H l ) γρ M (H m−l →L ) .(3.2.16) p

p

p

p

Consequently, for |β| = j < [l], S{l} (Dβ γρ Dα−β u)Lp ({l}+j)/l

([l]−j)/l

≤ c γρ M (H m →H l ) γρ M (H m−l →L ) uHpm . p

p

p

(3.2.17)

p

This inequality together with (3.2.15) implies that γρ uHpl ≤ (εγρ M (Hpm →Hpl ) + c(ε)γρ M (Hpm−l →Lp ) )uHpm +S{l} (u∇[l] γρ )Lp

(3.2.18)

for any ε > 0. It remains to estimate the last term on the right-hand side. We have S{l} (u∇[l] γρ )Lp ≤ uSl γρ Lp + |∇[l] γρ |S{l} uLp + )
 ∞  2 p/2 *1/p |∆yθ u(x)||∆yθ (∇[l] γρ )(x)| dθ y −1−2{l} dy dx , (3.2.19) 0

B1

where ∆z u(x) = u(x + z) − u(x). Obviously, uSl γρ Lp ≤ c Sl γρ M (Hpm →Lp ) uHpm .

(3.2.20)

Consider the second norm on the right-hand side of (3.2.19). Applying Minkow- ski’s inequality, we obtain

3.2 Description of M (Hpm → Hpl )

83

S{l} u ≤ Λ−[l] S{l} Λ[l] u . This and (3.2.16) yield |∇[l] γρ |S{l} uLp ≤ ∇[l] γρ M (Hpm−{l} →Lp ) Λ{l}−m S{l} Λm−{l} uHpm−{l} {l}/l

[l]/l

≤ c γρ M (H m →H l ) γρ M (H m−l →L ) uHpm . p

p

p

(3.2.21)

p

Now we estimate the third term on the right-hand side of (3.2.19). Let u = Λ−m f and let δ be a sufficiently small positive number. We notice that  |u(x + yθ) − u(x)| ≤ Gm (x − ξ + yθ) − Gm (x − ξ) |f (ξ)| dξ and use Lemma 3.2.6. Then

$

|u(x + yθ) − u(x)| ≤ c y

δ

(Λ

δ−m


x + yθ 
x % δ−m + (Λ |f |) |f |) . 4 4

Thus, the third term in (3.2.19) does not exceed
x+yθ  2  p2 * p1 )
 ∞  |∆yθ (∇[l] γρ )(x)|dθ y −1−2({l}−δ) dy dx (Λδ−m |f |) c 4 0 B1 ) 
x p * p1 +c (Λδ−m |f |) (Sl−δ γρ )p dx . 4 After simple calculations we obtain that this sum is majorized by + +
+  
·  
· + + + + + c + Λδ−m |f | ∇[l] γρ + {l}−δ + +|∇[l] γρ |S{l}−δ Λδ−m |f | + 4 4 Hp Lp + + 
· + δ−m + . S{l}−δ (∇[l] γρ )+ + +(Λ |f |) 4 Lp Let the last norms be denoted by N1 , N2 and N3 respectively. Using (3.2.16), we get +
· + + + N1 ≤ ∇[l] γρ M (Hpm−δ →Hp{l}−δ ) +(Λδ−m |f |) + 4 Hpm−δ 1−δ/l

δ/l

≤ c γρ M (H m →H l ) γρ M (H m−l →L ) f Lp . p

p

p

(3.2.22)

p

Similarly, +  
· + + + N2 ≤ ∇[l] γρ M (Hpm−{l} →Lp ) +S{l}−δ Λδ−m |f | + m−{l} 4 Hp + +
+ δ−m  · + ≤ c ∇[l] γρ M (Hpm−{l} →Lp ) + Λ |f | + 4 Hpm−δ {l}/l

[l]/l

≤ c γρ M (H m →H l ) γρ M (H m−l →L ) f Lp . p

p

p

p

(3.2.23)

84

3 Multipliers in Pairs of Potential Spaces

Now we estimate the norm N3 . According to Lemma 3.2.5, δ/m

S{l}−δ (∇[l] γρ ) ≤ c (Sl γρ )1−δ/m ∇[l] γρ 

m−{l}

M (Hp

→Lp )

+c ∇[l] γρ M (Hpm−{l} →Lp ) . Further we notice that Λδ−m |f | ≤ c (Λ−m |f |)1−δ/m (Mf )δ/m (see Lemma 1.2.5). Thus we have proved the estimate + 1−δ/m 
· δ/m + 
·  + + −m Λ (Mf ) S |f | γ + + l ρ m−{l} M (Hp →Lp ) 4 4 Lp δ/m

N3 ≤ c ∇[l] γρ 

+c ∇[l] γρ M (Hpm−{l} →Lp ) Λδ−m |f |Lp . Consequently, + +1−δ/m 
·  + δ/m + −m Λ S Mf  |f | γ + + l ρ m−{l} Lp M (Hp →Lp ) 4 Lp δ/m

N3 ≤ c ∇[l] γρ 

+c ∇[l] γρ M (Hpm−{l} →Lp ) Λδ−m |f |Hpm−δ . Applying (3.2.16), we finally obtain  δ[l]/ml δ{l}/ml N3 ≤ c γρ M (H m →Lp ) γρ M (H m →Lp ) Sl γρ M (Hpm →Lp ) p p 1−δ/m [l]/l {l}/l +γρ M (H m →H l ) γρ M (H m−l →L ) f Lp . p

p

p

(3.2.24)

p

Adding the estimates (3.2.22)–(3.2.24), we conclude that )
 ∞  2 p/2 *1/p |∆yθ u(x)||∆yθ (∇[l] γρ )(x)| dθ y −1−2{l}dy dx 0

B1

 ≤ εγρ M (Hpm →Hpl ) + ε Sl γρ M (Hpm →Lp )  + c(ε) γρ M (Hpm−l →Lp ) uHpm .

(3.2.25)

This together with (3.2.20) and (3.2.21) makes it possible to deduce from (3.2.19) that  S{l} (u∇[l] γρ )Lp ≤ ε γρ M (Hpm →Hpl ) + c Sl γρ M (Hpm →Lp )  +c(ε)γρ M (Hpm−l →Lp ) uHpm . Substitution of this estimate into (3.2.18) leads to (3.2.14) for γρ . Reference to Lemma 3.2.1 completes the proof. 

3.2 Description of M (Hpm → Hpl )

85

3.2.7 Lower Bound for the Norm of a Multiplier We prove the assertion converse to Lemma 3.2.7. l Lemma 3.2.8. If γ ∈ M (Hpm → Hpl ), then γ ∈ Hp,loc , Sl γ ∈ M (Hpm → Lp ), and γ ∈ M (Hpm−l → Lp ). The following inequality holds

Sl γM (Hpm →Lp ) + γM (Hpm−l →Lp ) ≤ c γM (Hpm →Hpl ) .

(3.2.26)

l Proof. It is clear that γ ∈ Hp,loc . The upper estimate of γM (Hpm−l →Lp ) is contained in Lemma 3.2.3. The assertion was proved for integer l in Lemma 2.3.4. Suppose that {l} > 0. For any u ∈ C0∞ ,  S{l} (Dβ γρ Dα−β u)Lp S{l} (u∇[l] γρ )Lp ≤ Sl (γρ u)Lp + 0≤β l ≥ 0. Proof. The upper estimate for the norm in M (Hpm → Hpl ) follows from Lemmas 3.2.7 and 2.3.6 which implies that sup e

γ; eLmp/(m−l) γ; eLp ≤ c sup . (m−l)/mp [Cp,m−l (e)]1/p [C e p,m (e)]

The lower bound is deduced from Corollary 3.2.3 and Lemma 3.2.8. From this theorem and (3.1.14) we obtain Corollary 3.2.4. The relation γM (Hpm →Hpl ) ∼

sup


γ; eL mp/(m−l)

{e:d(e)≤1}

holds with p ∈ (1, ∞), m > l ≥ 0.

[Cp,m (e)](m−l)/mp

+

Sl γ; eLp  [Cp,m (e)]1/p



3.2 Description of M (Hpm → Hpl )

89

3.2.10 Characterization of M (Hpm → Hpl ), m > l, Involving the Norm in L1,unif In the case m > l, the second term on the right-hand sides of (3.2.28) and (3.2.29) can be replaced by the norm of γ in L1,unif as shown by the following assertion. Theorem 3.2.4. Let 0 < l < m ≤ n/p, p ∈ (1, ∞). Then γM (Hpm →Hpl ) ∼ sup e

Sl γ; eLp + γL1,unif . [Cp,m (e)]1/p

(3.2.36)

This relation also holds if e is any compact subset of Rn with diameter less than 1. The lower estimate for the norm in M (Hpm → Hpl ) is a direct corollary of (3.2.28), (3.2.29), and the obvious inequality sup e

γ; eLp ≥ c sup γ; B1 (x)L1 . [Cp,m−l (e)]1/p x

The upper estimate follows from Theorem 3.2.3 and the next assertion obtained by Verbitsky (see Sect. 2.6 in [MSh16]). This assertion is similar in nature to Lemma 2.3.9. l . Then Lemma 3.2.10. Let 0 < l < m ≤ n/p, p ∈ (1, ∞), and let γ ∈ Hp,loc

sup e

γ; eLmp/(m−l) (Cp,m (e))(m−l)/mp


Sl γ; eLp , ≤ c sup + γ L 1,unif 1/p e (Cp,m (e))

(3.2.37)

The proof is based on several auxiliary assertions. In the first three of them we use the Poisson operator which is defined for functions γ ∈ L1,unif by  1 y γ(ξ)dξ (T γ)(x, y) = , (x, y) ∈ Rn+1 + , (3.2.38) (n+1) 2 |∂B | Rn (y + |x − ξ|2 )(n+1)/2 where |∂B (n+1) | is the area of the (n + 1)-dimensional unit ball. Lemmas 3.2.11–3.2.14 below are proved by Verbitsky, see Sect. 2.6 [MSh16]. Lemma 3.2.11. For any k = 0, 1, . . . there holds the inequality  1 k+1
(T γ)(x, y) k  ∂ |γ(x)| ≤ c γL1,unif + y dy . ∂y k+1 0

(3.2.39)

Proof. The following equality is readily checked by integration by parts γ(x) = (T γ)(x, 1) −

∂(T γ)(x, 1) 1 ∂ 2 (T γ)(x, 1) (−1)k ∂ k (T γ)(x, 1) + − . . . + ∂y 2 ∂y 2 k! ∂y k

90

3 Multipliers in Pairs of Potential Spaces

(−1)k+1 + k!

 0

1

∂ k+1 (T γ)(x, y) k y dy. ∂y k+1

(3.2.40)

Now we show that for any k = 0, 1, . . . the inequality + ∂ k (T γ)(·, 1) + + + ≤ c γL1,unif + + ∂y k L∞

(3.2.41)

holds. Let k > 0. The Poisson kernel P (x, y) = y(|x|2 + y 2 )−(n+1)/2 obeys the estimates ∂ k P (x, y) ≤ c (|x| + y)−n−k ∂y k

(3.2.42)

(see [St2], Ch. 5, Sect. 4). Hence ∂ k (T γ)(x, 1)  ∂ k P (t, 1) = γ(x − t)dt ∂y k ∂y k  
 |γ(x − t)|dt |γ(x − t)|dt  ≤c ≤ c |γ(x − t)|dt + (|t| + 1)n+k |t|n+k |t|≤1 |t|≥1   ∞
 ≤ c γL1,unif + r−n−k−1 dr |γ(x − t)|dt . 1

Note that for r ≥ 1

|t|≤r

 |t|≤r

|γ(x − t)|dt ≤ c rn γL1,unif .

Therefore,  ∞ ∂ k (T γ)(x, 1)
 −k−1 ≤ c γ 1 + r dr . L 1,unif ∂y k 1 Combining (3.2.40) and (3.2.41), we complete the proof for k > 0. The case k = 0 is treated in a similar way with the help of the estimate  P (x, 1) ≤ c (|x| + 1)−n−1 . [l]

Lemma 3.2.12. Let γ ∈ W1,loc and let k = [l] + 1. Then
 0

∞

1/2 ∂ k (T γ)(x, y) 2 1−2{l} y dy ≤ c (Sl γ)(x). ∂y k

3.2 Description of M (Hpm → Hpl )

Proof. First, consider the case k = 1, l ∈ (0, 1). The identity  ∂P (t, y) dt = 0 ∂y (see [St2], Ch. 5, Sect. 4) implies that    ∂(T γ)(x, y) ∂P (t, y) ∂P (t, y)  = γ(x − t)dt = γ(x − t) − γ(x) dt. ∂y ∂y ∂y Using (3.2.42), we get  ∂(T γ)(x, y) |γ(x − t) − γ(x)| dt. ≤c ∂y (|t| + y)n+1 Consequently,  ∞  ∞
 |γ(x − t) − γ(x)| 2 ∂(T γ)(x, y) 2 1−2l 1−2l dy ≤ c y dt dy y ∂y (|t| + y)n+1 0 0  ∞
 2 ≤c y −(1+2l+2n) dy |γ(x − t) − γ(x)|dt 0



∞

+c

y 1−2l

|t|≤y


 |t|≥y

0

We have



∞

A1 = c

y −(1+2l+2n)

∞

=c

y


 |τ |≤1

0



|γ(x − t) − γ(x)| 2 dt dy = A1 + A2 . (|t| + y)n+1

−(1+2l)


 |τ |≤1

0

|γ(x − τ y) − γ(x)|y n dτ

|γ(x − τ y) − γ(x)| dτ

2

2 dy

 2 dy ≤ c (Sl γ)(x) .

To find a majorant for A2 , we rewrite it as follows  ∞ 
 2 y 1−2l r−n−2 dr |γ(x − t) − γ(x)|dt dy. A2 = c 0

r≥y

y≤|t|≤r

Applying the Hardy inequality, we get  ∞
 −2n−2l−1 A2 ≤ c r

|t|≤r

0



∞

=c

r

−2l−1


 |τ |≤1

0

Thus, for l ∈ (0, 1)  0

|γ(x − t) − γ(x)|dt

|γ(x − τ r) − γ(x)|dτ

2

dr

 2 dr ≤ c (Sl γ)(x) .

 2 ∂(T γ)(x, y) 2 1−2l dy ≤ c (Sl γ)(x) . y ∂y

∞

2

91

92

3 Multipliers in Pairs of Potential Spaces

Next, let k = 2m, m = 1, 2, . . .. Since T γ is a harmonic function in Rn+1 + , n  ∂ 2 (T γ)(x, y) ∂ 2 (T γ)(x, y) = − . ∂y 2 ∂x2j j=1

Therefore n  ∂ 2m (T γ)(x, y) ∂ k (T γ)(x, y) m = (−1) . k ∂y ∂x2i1 . . . ∂x2im i ,...,i =1 1

(3.2.43)

m

Using the identity 

and the estimate

∂P (t, y) dt = 0, ∂ti

i = 1, . . . , n,

∂P (x, y) ≤ c (|x| + y)−n−1 ∂xi

(see [St2], Ch. 5, Sect. 4), we get  n ∂ k (T γ)(x, y)   ∂ [l] γ(x) ∂P (t, y)
∂ [l] γ(x−t) dt − = ∂y k ∂ti1 ∂xi1 ∂x2i2 . . . ∂x2im ∂xi1 ∂x2i2 . . . ∂x2im i ,...,i =1 1

m

 ≤c

|∇[l] γ(x − t) − ∇[l] γ(x)| dt. (|t| + y)n+1

We complete the proof in the same way as for l ∈ (0, 1). For k = 2m + 1, m = 1, 2, . . ., we use the identity n  ∂ k (T γ)(x, y) ∂ 2m+1 (T γ)(x, y) m = (−1) ∂y k ∂y∂x2i1 . . . ∂x2im i ,...,i =1 1



n 

= (−1)m

i1 ,...,im =1

m

∂ [l] γ(x)  ∂P (t, y)
∂ [l] γ(x − t) dt − ∂y ∂x2i1 . . . ∂x2im ∂x2i1 . . . ∂x2im

which implies that  ∂ k (T γ)(x, y) |∇[l] γ(x − t) − ∇[l] γ(x)| dt. ≤c ∂y k (|t| + y)n+1 Again we complete the proof in the same way as for l ∈ (0, 1).



In the next two assertions we use the notation K=

sup x∈Rn ,r∈(0,1]

rm−n/p Sl γ; Br (x)Lp .

(3.2.44)

3.2 Description of M (Hpm → Hpl )

93

[l]

Lemma 3.2.13. Let γ ∈ W1,loc , y ∈ (0, 1], and let k = [l] + 1. Then ∂ k (T γ)(x, y) ≤ c Ky {l}−m−1 . ∂y k Proof. Let r/2 < y ≤ r, r ∈ (0, 1]. By Lemma 3.2.12, 
 ∞ ∂ k (T γ)(t, y) 2 p/2 1−2{l} y dy dt ≤ c K p rn−mp . ∂y k 0 Br (x)

(3.2.45)

Applying the mean value theorem for harmonic functions and then the Cauchy inequality, we obtain   r k ∂ k (T γ)(x, y) ∂ (T γ)(t, η) −n−1 ≤ c r dη dt ∂y k ∂η k Br (x) r/2 
 r ∂ k (T γ)(t, η) 2 1/2 ≤ c r−n−1/2 dt dη ∂η k r/2 Br (x) 
 r ∂ k (T γ)(t, η) 2 1/2 1−2{l} η dη dt. ≤ c r−n−1+{l} ∂η k r/2 Br (x) Using (3.2.45) and the H¨ older inequality, we find that ∂ k (T γ)(x, y) ≤ c r−n−1+{l} rn(p−1)/p K r(n−mp)/p ≤ c K y {l}−m−1 . ∂y k 

The proof is complete. [l]

Lemma 3.2.14. Let γ ∈ W1,loc , 0 < l < m ≤ n/p. Then, for almost all x ∈ Rn ,   (3.2.46) |γ(x)| ≤ c K l/m ((Sl γ)(x))(m−l)/m + γL1,unif , where K is introduced in (3.2.44). Proof. We use Lemma 3.2.11 with k = [l] + 1. Let , |∂ k+1 (T γ)(x, y)/∂y k+1 | for 0 < y ≤ 1, ϕ(y) = 0 for y ≥ 1. Then, for any R > 0,  1 k+1  ∞ (T γ)(x, y) k ∂ ϕ(y) y k dy y dy = ∂y k+1 0 0  R  ∞ = ϕ(y) y k dy + ϕ(y) y k dy. 0

R

94

3 Multipliers in Pairs of Potential Spaces

Applying the Cauchy inequality to the first term and using Lemma 3.2.13 to estimate the second one, we find that  ∞ ϕ(y)y k dy 0



 ≤c

R

ϕ(y)2 y 1−2{l} dy

0

1/2




 ≤c

R

y 2k+2{l}−1 dy

0 ∞

ϕ(y)2 y 1−2{l}

1/2



1/2

∞

+K

y k+{l}−1−m dy



R



Rl + KRl−m .

0

Putting here R = K 1/m




∞

ϕ(y)2 y 1−2{l} dy

−1/2m

,

0

we arrive at 

∞

ϕ(y)y dy ≤ c K k

l/m




0

∞

ϕ(y)2 y 1−2{l} dy

(m−l)/2m .

0

Consequently,
 1 ∂ k+1 (T γ)(x, y) 2 
(m−l)/2m 1−2{l} |γ(x)| ≤ c K l/m dy + γL1,unif . y k+1 ∂y 0 

Reference to Lemma 3.2.12 completes the proof.



Proof of Lemma 3.2.10. Let K be defined by (3.2.44). By Lemma 3.2.14, 
 mp/(m−l) mp/(m−l) lp/(m−l) |γ(x)| dx ≤ c K |(Sl γ)(x)|p dx + γL1,unif mesn e ,

e

e

which together with the obvious estimate mesn e ≤ Cp,m (e) implies that γ; eLmp/(m−l) (Cp,m

(e))(m−l)/mp

  ≤ c K l/m Q(m−l)/m + γL1,unif ,

where Q = sup e

Sl γ; eLp . (Cp,m (e))1/p

Since K ≤ Q, we complete the proof.



Corollary 3.2.5. Let 0 < l < m and let p ∈ (1, ∞). Then γM (Hpm →Hpl ) ∼ Sl γM (Hpm →Lp ) + γL1 ,unif . For m = l the norm γL1 ,unif should be replaced by γL∞ . Proof. The result follows by Theorem 3.2.4 and Lemma 3.2.2.

(3.2.47)

3.3 One-Sided Estimates for the Norm in M (Hpm → Hpl )

95

3.2.11 The Space M (Hpm → Hpl ) for mp > n Theorem 3.2.5. If mp > n, p ∈ (1, ∞), then γM (Hpm →Hpl ) ∼ γHp,unif . l Proof. The required lower bound for the norm in M (Hpm → Hpl ) follows from the inequality γηz Hpl ≤ c γM (Hpm →Hpl ) ηz Hpm . Let us obtain the upper bound. Since Hpm is imbedded into L∞ , we have Cp,m (e) ∼ 1 for any compact set e with diameter d(e) not exceeding 1. Therefore, it follows from Theorem 3.2.4 that γM (Hpm →Hpl ) ∼ sup Sl γ; B1/2 (z)Lp + γL1,unif . z∈Rn

We have Sl γ; B1/2 (z)Lp ≤ Sl (γηz ); B1/2 (z)Lp + Sl [γ(1 − ηz )]; B1/2 (z)Lp . The first norm on the right-hand side does not exceed γηz Hpl . By Theorem 3.1.1 the second one is not greater than
 c

∞

sup x∈B1/2 (z)

1/2

y −n

 |s| n is a result by Strichartz [Str].

3.3 One-Sided Estimates for the Norm in M (Hpm → Hpl ) We present here some lower and, separately, upper bounds for the norm in M (Hpm → Hpl ), mp ≤ n, which do not contain the capacity and which follow from the characterization of multipliers in M (Hpm → Hpl ) obtained in the previous section.

96

3 Multipliers in Pairs of Potential Spaces

3.3.1 Lower Estimate for the Norm in M (Hpm → Hpl ) Involving Morrey Type Norms The next assertion follows directly from Proposition 3.1.4 and Theorem 3.2.2. Proposition 3.3.1. Let 0 < l < m. If mp < n, then γM (Hpm →Hpl )  m−n/p  ≥ c sup r Sl γ; Br (x)Lp + rm−l−n/p γ; Br (x)Lp

(3.3.1)

x∈Rn r∈(0,1)

and, if mp = n, then γM (Hpm →Hpl )   ≥ c sup (log 2r−1 )1−1/p Sl γ; Br (x)Lp + r−l γ; Br (x)Lp .

(3.3.2)

x∈Rn r∈(0,1)

For m = l the second term on the right-hand sides of (3.3.1) and (3.3.2) should be replaced by γL∞ . Remark 3.3.1. Let p ∈ (1, ∞) and let λ ∈ (0, n). By the Morrey space Lp,λ one means the set of functions in Rn with the finite norm f Lp,λ =

sup x∈Rn ,r>0

r−λ/p f ; Br (x)Lp .

(3.3.3)

Using this norm, we can rewrite estimate (3.3.1), where mp < n, as   γM (Hpm →Hpl ) ≥ c Sl γLp,n−mp + γL1,unif .

(3.3.4)

3.3.2 Upper Estimate for the Norm in M (Hpm → Hpl ) Involving Marcinkiewicz Type Norms The next assertion is a corollary of Theorem 3.2.4 and estimates (3.1.5), (3.1.6). Proposition 3.3.2. If mp < n and m > l, then  ≤c

γM (Hpm →Hpl ) sup

(mesn e)m/n−1/p Sl γ; eLp + γL1,unif



(3.3.5)

{e:d(e)≤1}

and, if mp = n, m > l, then  ≤c

γM (Hpm →Hpl )

sup

 (log(2n /mesn e))(p−1)/p Sl γ; eLp + γL1,unif ,

(3.3.6)

{e:d(e)≤1}

where d(e) is the diameter of e. In case m = l the norm in L1,unif on the right-hand sides of (3.3.5) and (3.3.6) should be replaced by γL∞ . In connection with (3.3.5) we make the following remark.

3.3 One-Sided Estimates for the Norm in M (Hpm → Hpl )

97

Remark 3.3.2. By the Marcinkiewicz space Mα one means the linear set of functions in a domain Ω ⊂ Rn for which sup t [m(t)]α < ∞ ,

(3.3.7)

0 0. We define the space Bq,∞ of functions in Rn with the norm µ = sup |h|−{µ} ∆h ∇[µ] vLq + vWq[µ] . vBq,∞

(3.4.1)

h∈Rn

In the present section we find sufficient conditions for the inclusion γ ∈ µ M (Hpm → Hpl ) formulated in terms of the space Bq,∞ (see Theorem 3.4.1). 3.4.1 Auxiliary Assertions [µ]

Lemma 3.4.1. Let q ≥ 1, {µ} > 0, µq < n. Further, let v ∈ Wq (B1 ). Then   ∇[µ] v; eLq v; eLq sup ≤ c sup + v; B  . (3.4.2) 1 Lq µ/n {µ}/n e⊂B1 (mesn e) e⊂B1 (mesn e) Proof. By (2.2.8), for any integer l < m we have (mesn e)(m−l)/n−1/q γ; eLq


 ≤ c γ; B1 Lq + (mesn e)(m−l)/n−1/q e

|∇l γ(x + z)| q 1/q  dz dx . |z|n−l |z| 0, {µ} > 0, v ∈ Wq (B2 ). Then sup (mesn e)−{µ}/n ∇[µ] v; eLq $ ≤ c sup |h|−{µ} ∆h ∇[µ] v; B1 Lq + e⊂B1

h∈B1

sup

r

−µ

x∈B1 ,0 µ, 1 ≤ q < ∞ and ∆stei is the difference of order s in the direction of the unit vector ei .

104

3 Multipliers in Pairs of Potential Spaces

µ Proof. Let v ∈ Bq,∞ . Putting k = [µ] and m = s − [µ] in the formula  1  1 ∂kv k+m k ··· ∆m (x + t(ξ1 + · · · + ξk )ei ) dξ1 . . . , dξk ∆tei v(x) = t tei ∂xki . 0 /0 0 1 k

and applying Minkowski’s inequality, we get + + + s−[µ] ∂ [µ] + |t|−µ ∆stei vLq ≤ |t|−{µ} +∆tei v + [µ] Lq ∂xi + [µ] + ∂ + + ≤ 2s−[µ]−1 |t|−{µ} +∆tei [µ] v + . Lq ∂xi (1)

µ This implies that vBq,∞ ≤ c vBq,∞ . µ Now we derive the converse inequality. We show that the finiteness of the (1) [µ] norm vBq,∞ implies the inclusion v ∈ Wq . Let vσ be the mean value of v µ defined in the proof of Lemma 3.4.3. We have  σ ∂ vσ (x) − vε (x) = vτ (x) dτ , 0 < ε < σ, x ∈ Rn . ε ∂τ

Using the expression for ∂vτ (x)/∂τ borrowed from the just mentioned proof, we arrive at the identity vε (x) = vσ (x)  σ  n 
y 
 dτ t dt ∆stei v(x + y + tei )dy. M Ω (3.4.7) + i n+2 τ τ n ε τ R1 R i=1 Differentiating (3.4.7) with k ≤ [µ], we find that  σ 
t dτ k dt M ∇k vε = ∇k vσ (x)(t − 1) n+2+k τ ε τ R1 n 
y  ∆stei v(x + y + tei ) dy × (∇k Ωi ) τ n R i=1

(3.4.8)

which together with Minkowski’s inequality yields  n  σ
t  ∇k vε − ∇k vσ Lq ≤ c ∆stei vLq dt τ −2−k dτ M τ 1 ε R i=1  
 n    σ −2−k t µ ≤ sup |t|−µ ∆stei vLq τ dτ |t| dt M 1 τ 1 0 R i=1 t∈R (1)

. ≤ c σ µ−k vBq,∞ µ The latter and the Lq -convergence vσ → v as σ → +0 imply the existence of any distributional derivative Dα v ∈ Lq , |α| ≤ [µ]. Also,

3.4 Upper Estimates for the Norm in M (Hpm → Hpl ) by Norms in Besov Spaces

105

(1)

∇k vε Lq ≤ ∇k vσ Lq + c σ µ−k vBq,∞ . µ Passing here to the limit as ε → +0, we get ∇k vLq ≤ ∇k vσ Lq + c σ µ−k vBq,∞ ≤ c σ −k vLq + c σ µ−k vBq,∞ . µ µ (1)

(1)

Putting σ = 1, we arrive at the estimate (1)

vWq[µ] ≤ c vBq,∞ . µ Since ∇k v ∈ Lq , it follows that ∇k vσ (x) → 0 as σ → ∞. We also have ∇k vε (x) → ∇k v(x) as ε → 0 for almost all x ∈ Rn . Therefore, by (3.4.7) we find that  ∞  n 
t 
y dτ k dt ∆stei (x+y+tei )dy M (∇ Ω ) ∇k v(x) = (−1) k i n+2+k τ τ n 0 τ R1 R i=1 for almost all x ∈ Rn . Hence, for any λ > 0 and k = [µ], we have ∆λej ∇k v(x) k

= (−1)



n   i=1

+(−1)k+1 λ

∆λej ∆tei v(x + y + tei )dy

Rn

n   i=1

(3.4.9)



1

dξ Rn

0



λ

dt

M

R1

0

  s ∆tei v(x+y +tei +ξej ) dt R1

where ψ(z) = ∇k


t  
y  dτ ∇k Ωi τ τ τ n+2+k ∞

M

λ


t 
y  dτ ψ , τ τ τ n+3+k

∂ Ωi (z). ∂zj

To derive equality (3.4.9), we made the change of variables z = x + y in the second summand and then applied the difference ∆λej to the kernel   , using the formula (∇k Ωi ) z−x τ  1

z − x  
z − x − λej  z − x − λξej  ∇k Ωi − (∇k Ωi ) = −λ dξ. ψ τ τ τ 0 After that we made the reverse change of variables y = z − x − λξej . Since M ∈ C0∞ (0, 1) and Ωi ∈ C0∞ ((0, 1)n ), i = 1, . . . , n, it follows for α = max{|y1 |, . . . , |yn |, t} and 0 < t < λ, that 

λ

0

 ≤c

∞

α

τ −n−2−k M


t τ

∇k Ωi


y τ

dτ

τ −n−2−k dτ ≤ c (|y| + t)−n−1−k .

(3.4.10)

106

3 Multipliers in Pairs of Potential Spaces

If t > λ then the integral on the left-hand side of (3.4.10) is equal to zero. In a similar way we get  ∞
t 
y  −n−3−k τ M (3.4.11) ψ dτ ≤ c (|y| + t + λ)−n−2−k τ τ λ

provided that t > δλ with some δ > 0. In the case t ≤ δλ, the integral on the left-hand side of (3.4.10) is equal to zero. Using (3.4.8)–(3.4.11), we find that  n  λ  |∆stei ∆λej v(x + y + tei ) |∆λej ∇k v(x)| ≤ c dt dy (|y| + t)n+1+k Rn i=1 0  ∞  21 s n  |∆tei v(x + y + tei + λξej )|dξ 0 +c λ dt dy . (|y| + t)n+2+k δλ Rn i=1 Applying Minkowski’s inequality we obtain λ−{µ} ∆λej ∇k vLq ≤ c

+c

n 

 λ1−{µ}

n 

λ−{µ}



∆tei vLq t−2−k dt ≤ c

δλ

i=1

∆tei vLq t−1−k dt

0

i=1 ∞

λ

n  i=1

sup t−µ ∆stei vLq . t>0

Thus sup λ−{µ} ∆λej ∇k vLq ≤ c vBq,∞ , µ (1)

j = 1, . . . , n .

∆|ηj |ej vLq ,

η ∈ Rn .

(3.4.12)

λ>0

Next we note that ∆η vLq ≤

n  j=1

Therefore |η|−{µ} ∆η ∇k vLq ≤

n 

|ηj |−{µ} ∆|ηj |ej ∇k vLq .

j=1

By (3.4.12), the last sum does not exceed cv(1) . The result follows.



The proposition just proved and the definition of the norms · and ·(1) µ imply that the norm in Bq,∞,unif is equivalent to either of the following two norms: sup x∈Rn ,h∈B1 n 

|h|−{µ} ∆h ∇[µ] v; B1 (x)Lq + sup v; B1 (x)Lq ,

sup

n i=1 x∈R ,|t| 0, µq < n. Then r−µ v; Br (x)Lq ≤ c

sup x∈Rn ,r>0

n 

sup |t|−µ ∆stei vLq .

(3.4.13)

1 i=1 t∈R

Proof. It follows from (3.4.3) that |v(z)| ≤

n   i=1

0

∞

dσ σ n+2


 
y − z  t M dt |∆stei v(y + tei )| Ω dy, σ σ 1 R

where M ∈ C0∞ (0, 1) and Ω ∈ C0∞ ((0, 1)n ). By Ui we denote the i-th term on the right-hand side. Let us represent Ui as the sum Vi + Wi of two integrals over σ so that the integration in Vi is taken over σ ∈ [0, r]. By Minkowski’s inequality, 

r

dσ σ n+2

Vi ; Br (x)Lq ≤ 0


 t M dt σ R1

+
y − z  + + + ×+ |∆stei v(y + tei )| Ω dy; Br (x)+ . σ Lq Applying Minkowski’s inequality once more, we obtain +
y − z  + + + + |∆stei v(y + tei )| Ω dy; Br (x)+ ≤ c σ n ∆stei vLq . σ Lq Therefore,



Vi ; Br (x)Lq ≤ c 0

r

dσ n σ σ n+2

 0

σ

τ µ dτ sup |t|−µ ∆stei vLq t∈R1

≤ c rµ sup |t|−µ ∆stei vLq .

(3.4.14)

t∈R1

By H¨older’s inequality,



sup |Wi (z)| ≤ c1 z∈Br (x)

r

∞

dσ σ n+2



σ

σ n/q




$

0

% sup |t|−µ ∆stei vLq τ µ dτ

t∈R1

= c rµ−n/q sup |t|−µ ∆stei vLq . t∈R1

Consequently, Wi ; Br (x)Lq ≤ c rn/q sup |Wi (z)| ≤ c rµ sup |t|−µ ∆stei vLq . z∈Br (x)

This, together with (3.4.14), implies (3.4.13).

t∈R1



108

3 Multipliers in Pairs of Potential Spaces

The next assertion follows immediately from Lemmas 3.4.1 - 3.4.4. [µ]

Corollary 3.4.1. Let {µ} > 0, q ≥ 1, µq ≤ n and let v ∈ Wq,loc . Then sup

  (mesn e)−µ/n (mesn e)[µ]/n ∇[µ] v; eLq + v; eLq

{e:d(e)≤1}

, ≤

µ c vBq,∞,unif

for µq < n ,

µ + vL∞ ) for µq = n , c (vBq,∞,unif

where d(e) is the diameter of a compact set e. 3.4.3 Estimates for the Norm in M (Hpm → Hpl ) by the Norm µ in Bq,∞ The following assertion is the main result of the present section. Theorem 3.4.1. Let q ≥ p, µ = n/q − m + l, µ > l, {µ} > 0. µ (i) If γ ∈ Bq,∞,unif ∩ L∞ , then γ ∈ M Hpl and 
γM Hpl ≤ c sup |h|−{µ} ∆h ∇[µ] γ; B1 (x)Lq + γL∞ .

(3.4.15)

x∈Rn ,h∈B1

µ (ii) If γ ∈ Bq,∞,unif , then γ ∈ M (Hpm → Hpl ) and

γM (Hpm →Hpl )   ≤c sup |h|−{µ} ∆h ∇[µ] γ; B1 (x)Lq + sup γ; B1 (x)Lq .(3.4.16) x∈Rn

x∈Rn ,h∈B1

Proof. Let m ≥ l. It is sufficient to assume that the difference ε = n/q − m is small, since the general case follows by interpolation between the pairs n/q−ε {Hpm−l , Lp } and {Hp , Hpµ−ε } (cf. (3.2.1)). Thus we assume that 1 − {l} > n/q − m > 0. We introduce the function Φx on (0, ∞) with the values  Φx (y) = |∇[l] γ(x + z) − ∇[l] γ(x)| dz . By

Clearly, Sl γ; epLp =


 e

∞

[Φx (y)]2 y −1−2({l}+n) dy

p/2 dx,

0

where e is a compact set with d(e) ≤ 1. Since Φx is an increasing function, the internal integral on the right-hand side is dominated by  ∞   2 Φx (y) dy ∞ dt {l} + n
∞ dy ({l} + n) Φ (t) = Φ (y) . x x 2 y 1+{l}+n y t1+{l}+n y 1+{l}+n 0 0

3.4 Upper Estimates for the Norm in M (Hpm → Hpl ) by Norms in Besov Spaces

109

We write the last integral as the sum of two integrals i1 (x) + i2 (x), of which the first is over the semi-axis y > |e|1/n , where |e| = mesn e. We have   ) ∞
  dy *p i1 (x)p dx ≤ |∇[l] γ(x + z)| dz + |∇[l] γ(x)| 1+{l} dx. y −n y |e|1/n e e By This and Minkowski’s inequality imply that   ) ∞
  p 1/p dy *p i1 (x)p dx ≤ . y −n |∇[l] γ(x+z)|dz+|∇[l] γ(x)| dx y 1+{l} |e|1/n e e By Therefore  ) p i1 (x) dx ≤ c

y

|e|1/n

e




 

∞

≤ c |e|

1−p/q

−n

e

)

∞

 1/p dy *p |∇[l] γ(x+z)|pdz+|∇[l] γ(x)|p dx y 1+{l} By

 y

|e|1/n

−n

 By




|∇[l] γ(x + z)|q dx

1/q dz

e

1/q  dy *p
 |∇[l] γ(x)|q dx . + y 1+{l} e By N (γ) we denote the right-hand sides of (3.4.15) and (3.4.16). According to Corollary 3.4.1, where µ = n/q − m + l,
 1/q |∇[l] γ(x + z)|q dx ≤ c |e|1/q−m/n N (γ) . e

Consequently,

 i1 (x)p dx ≤ c |e|1−mp/n N (γ)p .

(3.4.17)

e

Applying Minkowski’s inequality, we obtain  p




|e|1/n



i2 (x) dx = e

|∇[l] γ(x + z) − ∇[l] γ(x)| dz

dy

p

dx y 1+{l}+n *p ) |e|1/n
 1/p dy ≤ |∇[l] γ(x + z) − ∇[l] γ(x)|p dx dz 1+{l}+n . y 0 By e e

By

0

By H¨older’s inequality the last expression does not exceed |e|

1−p/q

)

|e|1/n

0






By

e

|∇[l] γ(x + z) − ∇[l] γ(x)|q dx

1/q dz

dy y 1+{l}+n



Since

|∇[l] γ(x + z) − ∇[l] γ(x)|q dx ≤ c N (γ)q ,

|z|

q(m−{l})−n e

*p .

110

3 Multipliers in Pairs of Potential Spaces

it follows that 
 p 1−p/q p i2 (x) dx ≤ c |e| N (γ) e

0

|e|1/n

 By

|z|−m+{l}+n/q dz

p

dy y 1+{l}+n

= c |e|1−pm/n N (γ)p .

(3.4.18)

Adding together (3.4.17) and (3.4.18), we arrive at Sl γ; epLp ≤ c |e|1−pm/n N (γ)p . By H¨older’s inequality and Corollary 3.4.1 with µ = n/q − m + l we obtain γ; eLp ≤ |e|1/p−1/q γ; eLq ≤ c |e|1/p−1/q |e|1/q−(m−l)/n N (γ) . Reference to Proposition 3.3.2 completes the proof.

(3.4.19) 

3.4.4 Estimate for the Norm of a Multiplier in M Hpl (R1 ) by the q-Variation Hirschman [Hi2] obtained the following sufficient conditions for a function γ to belong to the class M W2l on a unit circumference C: γ is bounded and has a finite q-variation Varq (γ) for some q, 2 < q < 1/l. Here q-variation is defined by 1/q
m−1  Varq (γ) = sup |γ(tj+1 ) − γ(tj )|q ,

(3.4.20)

j=0

with the supremum taken over all partitions of the circumference C by points tj . Using Theorem 3.4.1, one may easily derive a sufficient condition for a function to belong to M Hpl (R1 ) which, for p = 2, coincides with Hirschman’s condition up to the change of R1 for C. We define the local q-variation of a function γ defined on R1 by (3.4.20), with the supremum taken over all choices of a finite number of points t0 < t1 < · · · < tm on any interval σ of unit length. Since  |γ(t + h) − γ(t)|q dt ≤ c |h|[Varq (γ)]q , σ

we arrive at the following assertion. Corollary 3.4.2. Let n = 1, q ≥ p, lq < 1. If γ ∈ L∞ and Varq (γ) < ∞, then γ ∈ M Hpl and γM Hpl ≤ c (γL∞ + Varq (γ)) .

3.5 Miscellaneous Properties of Multipliers in M (Hpm → Hpl )

111

3.5 Miscellaneous Properties of Multipliers in M (Hpm → Hpl ) The following assertion is a generalization of Proposition 2.7.1. Proposition 3.5.1. If k ∈ [0, l], then M (Hpm → Hpl ) ⊂ M (Hpm−l+k → Hpk ) and γM (Hpm−l+k →H k ) ≤ c γM (Hpm →Hpl ) . p

Proof. The imbedding M (Hpm → Hpl ) ⊂ M (Hpm−l → Lp ) and the corresponding inequality for the norms were proved in Lemma 3.2.3. It remains to use the interpolation inequality k/l

(l−k)/l

γM (Hpm−l+k →H k ) ≤ c γM (H m →H l ) γM (H m−l →L p

p

p

p

p)



which is a particular case of estimate (3.2.1).

Proposition 3.5.2. (i) If mp > n, 1 < q < ∞, 0 ≤ k < l and k ≤ l + n(1/q − 1/p), then M (Hpm → Hpl ) ⊂ M (Hqm−l+k → Hqk ) and γM (Hqm−l+k →H k ) ≤ c γM (Hpm →Hpl ) . q

(3.5.1)

(ii) If mp = n, 0 ≤ k ≤ l, q > 1 and k < l − m + n/q, then M (Hpm → Hpl ) ⊂ M (Hqm−l+k → Hqk ) and inequality (3.5.1) holds. l Proof. (i) According to Theorem 3.2.5, M (Hpm → Hpl ) = Hp,unif . Since mp > n, it follows that l M (Hpm → Hpl ) ⊂ Hn/m,unif

for m > l

and m ∩ L∞ M Hpm ⊂ Hn/m,unif

for m = l.

This, together with Propostion 3.3.1, implies that M (Hpm → Hpl ) ⊂ M (Hqm → Hql ) for any q ∈ (1, n/m). Applying Proposition 3.5.1 and interpolating with respect to q between 1 + ε and p (cf. (3.2.1)), we obtain the inclusion M (Hpm → Hpl ) ⊂ M (Hqm−l+k → Hqk ) for all k ∈ [0, l], q ∈ (1, p]. For q ∈ (p, ∞) we put s = l + n(1/q − 1/p). It is clear that s < l and q(s + m − l) > n. By Theorem 3.2.5 and the Sobolev imbedding theorem, l s ⊂ Hq,unif . M (Hpm → Hpl ) = Hp,unif

Since q(s + m − l) > n, we can apply Theorem 3.2.5 once more to obtain M (Hpm → Hpl ) ⊂ M (Hqm−l+s → Hqs ). Further, by Proposition 3.5.1, M (Hpm → Hpl ) ⊂ M (Hqm−l+k → Hqk ) for all k ≤ s, q ∈ [p, ∞).

112

3 Multipliers in Pairs of Potential Spaces

l (ii) By Theorem 3.2.5, M (Hpm → Hpl ) ⊂ Hn/m,unif for m > l and M Hpm ⊂ m ∩ L∞ . According to the Sobolev imbedding theorem, Hn/m,unif l k Hn/m,unif ⊂ Hn/r,unif ,

where

r = m − l + k, k < l.

This and Proposition 3.3.1 imply that M (Hpm → Hpl ) ⊂ M (Hqm−l+k → Hqk ) for any q ∈ (1, n/r).



Next we present an imbedding theorem for the space M Hpm . Proposition 3.5.3. The space M Hpm is imbedded into M Hqk , k ≤ m, 1 < q < ∞, provided that (i) mp > n, k ≤ m + n(1/q − 1/p), (ii) mp ≤ n, k < mp/q. Proof. For mp ≥ n the assertion is proved in Propostion 3.5.2. Let mp < n. We start with the case q ≥ p. If γ ∈ M Hpm then γ ∈ L∞ and hence γ ∈ M Lr for any r ∈ (1, ∞). The interpolation between Hpm and Lr with r ∈ [q, ∞) (cf. inequality (3.2.1)) implies the imbedding M Hpm ⊂ M Hqk with 0 < k < mp/q, q ≥ p. Now suppose that q < p, k ≤ m and γ ∈ M Hpm . Then by Theorem 3.2.2 Sm γ; epLp 0 one can find an open set ω such that Cp,l (ω) < ε and u is continuous on Rn \ω. For proofs of the next assertions see [MH2].

3.6 Spectrum of Multipliers in Hpl and Hp−l


119

l Proposition 3.6.3. For any u ∈ Hp,loc there exists a (p, l)-refined Borel function which coincides with u almost everywhere.

Proposition 3.6.4. If two (p, l)-refined functions u1 and u2 are equal almost everywhere then they are equal (p, l)-quasi everywhere. Henceforth in this section all the functions are assumed to be (p, l)-refined and Borel. The following assertion is proved for integer l in [Maz14] and for fractional l in [APo] for compact sets. The passage to arbitrary sets does not need new arguments if one uses Proposition 3.6.2. Proposition 3.6.5. Let E ⊂ Rn . The capacity Cp,l (E) is equivalent to the set function inf{vpH l : v ∈ M(E)}, p

where M(E) is the collection of (p, l)-refined functions equal to one (p, l)-quasi everywhere on E and satisfying the inequalities 0 ≤ v ≤ 1. Definition 3.6.9. A set E ⊂ Rn is called the set of uniqueness for the space Hpl if the conditions u ∈ Hpl , u(x) = 0 for (p, l)-quasi all x ∈ Rn \E imply that u = 0. A description of sets of uniqueness for Hpl is given in [Hed3] and [Pol2]. 1/2

The first result of such a kind for H2 and Beurling [AB].

on a circumference is due to Ahlfors

Proposition 3.6.6. (see [Hed3]). Let E be a Borel subset of Rn . The following conditions are equivalent: (i) E is the set of uniqueness for Hpl ; (ii) Cp,l (G\E) = Cp,l (G) for any open set G; (iii) for almost all x lim ρ−n Cp,l (Bρ (x)\E) > 0.

ρ→0

If lp > n then E is the set of uniqueness if and only if it has no interior points. Now we state a theorem which gives a characteristic of the sets σp (γ), σr (γ) and σc (γ) for a multiplier γ in Hpl or Hp−l
. Theorem 3.6.1. (i) Let γ ∈ M Hpl and λ ∈ σ(γ). 1. λ ∈ σp (γ) if and only if the set Zλ = {x : γ(x) = λ} does not satisfy any of conditions (i)–(iii) of Propostion 3.6.6. 2. λ ∈ σr (γ) if and only if the set Zλ satisfies any one of the conditions of Proposition 3.6.6 and Cp,l (Zλ ) > 0. 3. λ ∈ σc (γ) if and only if Cp,l (Zλ ) = 0. (ii) Let γ ∈ M Hp−l
and λ ∈ σ(γ).

120

3 Multipliers in Pairs of Potential Spaces

1. λ ∈ σp (γ) if and only if Cp,l (Zλ ) > 0. 2. λ ∈ σc (γ) if and only if Cp,l (Zλ ) = 0 (hence the set σr (γ) is empty). An obvious corollary of Propositions 3.6.3 and 3.6.4 is the following assertion. Lemma 3.6.1. Let γ be a (p, l)-refined function in M Hpl . The equation (γ − λ)u = 0 has a nontrivial solution in Hpl if and only if there exists a (p, l)refined non-zero function in Hpl vanishing (p, l)-quasi everywhere outside Zλ . This lemma shows that part (i)1 of the theorem immediately follows from Proposition 3.6.6. The proof of the other assertions of the theorem is contained in the next subsection. 3.6.4 Proof of Theorem 3.6.1 Below we shall use the following assertion. Lemma 3.6.2. Let γ be a (p, l)-refined function in M Hpl and let Z0 = {x : γ(x) = 0}. If Cp,l (Z0 ) = 0, then the set γHpl is dense in Hpl . Proof. Let f ∈ C0∞ and let Nτ = {x ∈ supp f : |γ(x)| ≤ τ }. By ε we denote a small positive number and by ω we mean an open set with Cp,l (ω) < ε and such that γ is continuous on Rn \ω. Let G stand for a neighborhood of the set N0 \ω with Cp,l (G) < ε. We note that Nτ \ω ⊂ G for small enough τ > 0. In fact, if for any τ > 0 there exists a point xτ ∈ Nτ \ω which is not contained in G then, by continuity of γ outside ω, the limit point x0 of the family {xτ } is in N0 \ω, contrary to the definition of G. Consequently, Cp,l (Nτ \ω) < ε for small values of τ and Cp,l (Nτ ) ≤ Cp,l (Nτ \ω) + Cp,l (ω) < 2ε . Thus, Cp,l (Nτ ) → 0 as τ → 0. By {wτ }τ >0 we denote a family of functions in M(Nτ ) such that lim wτ Hpl = 0

τ →0

(see Proposition 3.6.5). Further, we put uτ,δ = (1 − wτ )¯ γ f /(γ¯ γ + δ), where δ > 0. Since (1 − wτ )f ∈ Hpl , γ¯ ∈ M Hpl , γ¯ γ ∈ M Hpl , and γ¯ γ + δ ≥ δ,

3.6 Spectrum of Multipliers in Hpl and Hp−l


121

it follows that uτ,δ ∈ Hpl . We have f − γuτ,δ = wτ f + δ(1 − wτ )f /(γ¯ γ + δ) . Let ϕ be a smooth increasing function on [0, ∞), ϕ(0) = τ 2 /4, ϕ(τ ) = t for t > τ 2 /2. Since 1 − wτ = 0 (p, l)-quasi everywhere on Nτ , f − γuτ,δ = wτ f + δ(1 − wτ )f /[ϕ(γ¯ γ ) + δ] . Using the inequality ϕ(γ¯ γ ) + δ > τ 2 /4, we obtain from Proposition 2.7.5 and Corollary 3.5.2 that the norm [ϕ(γ¯ γ ) + δ]−1 M Hpl is uniformly bounded with respect to δ. Therefore f − γuτ,δ Hpl ≤ wτ f Hpl + δk(τ ), where k(τ ) does not depend on δ. We put δ(τ ) = τ /k(τ ). Then f − γuτ,δ(τ ) Hpl ≤ c wτ Hpl + τ and hence γuτ,δ(τ ) → f as τ → 0 in Hpl .



In the next three propositions γ is a (p, l)-refined function from M Hpl . Proposition 3.6.7. A number λ is contained in the pointwise spectrum of a multiplier γ in Hp−l
if and only if Cp,l (Zλ ) > 0. Proof. Sufficiency. Let R be so large that Cp,l (Zλ ∩ BR ) > 0 and let µ be the capacitary measure of Zλ ∩ BR . Note that, whatever the (p, l)-refined function u ∈ Hpl , we have u(x)(γ(x) − λ) = 0 for (p, l)-quasi all x ∈ Zλ ∩ BR . By Proposition 3.6.2, the last equality holds µ-almost everywhere. Therefore,  u(γ − λ) dµ = 0. In other words, (γ − λ)µ = 0. Since





µpH −l = Jl µpLp
= Cp,l (Zλ ∩ BR ) < ∞ , p


we conclude that λ ∈ σp (γ). Necessity. Let λ ∈ σp (γ). Then there exists a distribution T ∈ Hp−l
, T = 0 such that (γ − λ)T = 0. Therefore (T, (γ − λ)u) = 0 for all u ∈ Hpl and the set (γ − λ)Hpl is not dense in Hpl . The result follows by application of Lemma 3.6.2. 

122

3 Multipliers in Pairs of Potential Spaces

Proposition 3.6.8. A number λ is contained in the residual spectrum of a / σp (γ) and Cp,l (Zλ ) > 0. multiplier in Hpl if and only if λ ∈ ¯ is Proof. Sufficiency. Since Cp,l (Zλ ) > 0 it follows by Proposition 3.6.7 that λ −l an eigenvalue of the multiplier γ¯ in Hp
. This fact and (3.6.3) imply that λ ∈ σr (γ) ∪ σp (γ) = σr (γ). ¯ is an eigenvalue of the multiplier γ¯ Necessity. Let λ ∈ σr (γ). By (3.6.3), λ −l  in Hp
. Hence, according to Proposition 3.6.7, Cp,l (Zλ ) > 0. Proposition 3.6.9. The multiplier γ in Hp−l
has no residual spectrum. ¯ is an eigenvalue of γ¯ in H l . This and the Proof. Let λ ∈ σr (γ). By (3.6.3), λ p first assertion in Theorem 3.6.1 part (i) imply that Cp,l (G\Zλ ) < Cp,l (G) for some open set G ⊂ Rn . Since Cp,l (Zλ ) > Cp,l (G) − Cp,l (G\Zλ ), we have Cp,l (Zλ ) > 0. According to Proposition 3.6.7, this means that  λ ∈ σp (γ). Thus we arrive at a contradiction. Thus the statements of Theorem 3.6.1 concerning the pointwise and residual spectrum are proved. The characterization of the continuous spectrum obviously follows from these criteria and the relation (3.6.2). l 3.7 The Space M (hm p → hp)

In this section we assume that mp < n, p ∈ (1, ∞). Using Sobolev’s theorem on the imbedding hm p ⊂ Lq with q = mp/(n − mp), n > mp, one can easily prove that any function ψ ∈ C0∞ belongs to the m space M hm p of multipliers in hp . This immediately implies that the norm in m M hp of the function ψr with values ψr (x) = ψ(x/r), r > 0, does not depend on r. This enables one to proceed as in the proofs of Lemmas 3.2.7 and 3.2.8, with γρ replaced by ψr γρ , where ψ ∈ C0∞ , ψ = 1 on B1 , passing to the limit as ρ → 0, r → ∞ at the final step. l Thus we are led to the relation for the norm in M (hm p → hp ) with 0 ≤ l ≤ m < n/p:   γ; eLp Sl γ; eLp ∼ sup + . (3.7.1) γM (hm l p →hp ) [cp,m (e)]1/p [cp,m−l (e)]1/p e The next two assertions are proved in the same way as Lemmas 2.8.1 and 2.8.2 with W replaced by H and w replaced by h.

l 3.7 The Space M (hm p → hp )

123

Lemma 3.7.1. (i) The inequality γM (Hpm →Lp ) ≤ γM (hm p →Lp )

(3.7.2)

holds. (ii) Let ρ > 0 and let γ ∈ M (hm p → Lp ). Then lim ρ−m γ(·/ρ)M (Hpm →Lp ) = γM (hm . p →Lp )

(3.7.3)

ρ→0

(iii) The function γ ∈ M (hm p → Lp ) satisfies ≥ c sup rm−n/p γ; Br (x)Lp . γM (hm p →Lp )

(3.7.4)

x∈Rn r>0

l Lemma 3.7.2. (i) Let m ≥ l and let γ ∈ M (hm p → hp ). Then

γM (Hpm →Hpl ) ≤ c γM (hm l . p →hp )

(3.7.5)

lim inf ρl−m γ(·/ρ)M (Hpm →Hpl ) ≥ γM (hm l p →hp )

(3.7.6)

(ii) The inequality ρ→0

holds. l Now we give a description of the space M (hm p → hp ). l Theorem 3.7.1. Let mp < n, m ≥ l, p ∈ (1, ∞). Then γ ∈ M (hm p → hp ) if l and only if γ ∈ hp,loc , (3.7.7) Sl γ ∈ M (hm p → Lp ),

and γ ∈ L∞ (Rn ) lim r

r→∞

−n

for m = l,

γ; Br L1 = 0

for m > l.

(3.7.8)

l The norm in the space M (hm p → hp ), m > l, is subject to the equivalence relation ∼ Sl γM (hm . (3.7.9) γM (hm l p →Lp ) p →hp )

For m = l the norm γL∞ should be added to the right-hand side of this relation. The equivalence relation γM (hm l p →hp ) ∼

[l]   j=0

holds.

Sl−j γM (hm−j m−{l}−j →Lp ) + ∇[l]−j γM (hp →Lp ) p



(3.7.10)

124

3 Multipliers in Pairs of Potential Spaces

Proof. We replace γ by ρl−m γ(·/ρ) in (3.2.27). Then (3.7.10) follows from Lemmas 3.7.1 and 3.7.2 as ρ → 0. We take γ as ρl−m γ(·/ρ) in (3.2.47) to obtain ρl−m γ(·/ρ)M (Hpm →Hpl )   ≤ c Sl (ρl−m γ(·/ρ))M (Hpm →Lp ) + sup Rm−l−n ||ρl−m γ(·/ρ); BR (x)L1 . x∈Rn R>0

Since the second term on the right-hand side is equal to sup rm−l−n γ; Br (x)L1 ,

x∈Rn r>0

and the first term tends to Sl γM (hm as ρ → 0 by (3.7.3), reference to p →Lp ) (3.7.6) gives   γM (hm ≤ c Sl γM (hm + sup rm−l−n γ; Br (x)L1 . (3.7.11) l p →Lp ) p →hp ) x∈Rn r>0

It remains to remove the second term on the right-hand side in the case m > l. We use the inequality |γ(x)| ≤ c



sup x∈Rn ,r∈(0,1)

n

rm− p Sl γ; Br (x)Lp

 ml

l

((Sl γ)(x))1− m ,

(3.7.12)

which follows from (3.2.46) with the term γL1,unif omitted on the right-hand side due to condition (3.7.8). Integrating (3.7.12) over an arbitrary ball Br (x), we arrive at l/m Sl γ 1−l/m ; Br (x)L1 . γ; Br (x)L1 ≤ c ( sup rm−n/p Sl γ; Br (x)Lp x∈Rn r∈(0,1)

By H¨older’s inequality rm−l−n γ; Br (x)||L1 l 1− ml  n n ≤ c sup rm− p Sl γ; Br (x)Lp m rm− p Sl γ; Br (x)Lp .

(3.7.13)

x∈Rn r∈(0,1)

The result follows by (3.7.11).



Remark 3.7.1. By Theorem 3.1.4, condition (3.7.7) can be formulated in four ways as follows: (i) for all compact sets e ⊂ Rn Sl γ; epLp ≤ C cp,m (e),

3.8 Positive Homogeneous Multipliers

125

(ii) for all cubes Q in Rn   p
 p (Sl γ)p dx, Im (χQ Sl γ) (x) dx ≤ C Q

(iii) for almost all x ∈ R  p
Im Im (Sl γ)p (x) ≤ C Im (Sl γ)p (x), n

(iv) for every dyadic cube P0 in Rn   p
 (mesn P )(m−n)/n (Sl γ(x))p dx mesn P ≤ C (Sl γ(x))p dx, P

P ⊆P0

P0

where the sum is taken over all dyadic cubes P contained in P0 .

3.8 Positive Homogeneous Multipliers l In this section we give a description of elements in the spaces M (hm p → hp ) m l l−m f (x/|x|). and M (Hp → Hp ) which have the form |x|

3.8.1 The Space M (Hpm (∂B1 ) → Hpl (∂B1 )) Let {Ui } be a finite covering of ∂B1 by open sets with small diameters, and let {ϕi } be a family of diffeomorphisms Ui → Rn−1 which form a smooth structure on ∂B1 . Further, let {νi } be a smooth partition of unity on ∂B1 subordinate to the covering {Ui }. We say that a function v on ∂B1 is in the space Hpl (∂B1 ) if ∈ Hpl (Rn−1 ) (νi ◦ v) ◦ ϕ−1 i

for all i.

We equip Hpl (∂B1 ) with the norm v; ∂B1 Hpl =




n−1 p (νi v) ◦ ϕ−1 H l i ;R

1/p .

p

i

It is well known that the passage from one collection {Ui , ϕi , νi } to another leads to an equivalent norm. Proposition 3.8.1. A function f is contained in the space M (Hpm (∂B1 ) → Hpl (∂B1 )) if and only if ∈ M (Hpm (Rn−1 ) → Hpl (Rn−1 )) (νi ◦ f ) ◦ ϕ−1 i for all i. Moreover, n−1 M (Hpm →Hpl ) . f ; ∂B1 M (Hpm →Hpl ) ∼ max (νi f ) ◦ ϕ−1 i ;R i

(3.8.1)

126

3 Multipliers in Pairs of Potential Spaces

Proof. For any v ∈ Hpm (∂B1 ) we have f v; ∂B1 Hpl ≤




n−1 p (νi f νj ζi v) ◦ ϕ−1 H l i ;R

1/p

p

i,j n−1 ≤ sup (νi f ) ◦ ϕ−1 M (Hpm →Hpl ) i ;R i




n−1 p (νj ζi v) ◦ ϕ−1 Hpm i ;R

1/p ,

i,j

where ζi ∈ C0∞ (Ui ) and ζi νi = νi . Since the mapping ϕj ϕ−1 : ϕj (Uj ∩ Ui ) → ϕi (Uj ∩ Ui ) i is infinitely differentiable, we have −1 n−1 n−1 (νj ζi v) ◦ ϕ−1 Hpm = (νj ζi v) ◦ ϕ−1 Hpm i ;R j ϕj ϕi ; R n−1 ≤ c (νj v) ◦ ϕ−1 Hpm , j ;R

and thus the required upper estimate for the norm in M (Hpm (∂B1 ) → Hpl (∂B1 )) follows. On the other hand, for any w ∈ Hpm (Rn−1 ) n−1 n−1 [(νi f ) ◦ ϕ−1 Hpl = [νi f (w ◦ ϕi )] ◦ ϕ−1 Hpl i ]w; R i ;R

≤ νi f (w ◦ ϕi ); ∂B1 Hpl ≤ f ; ∂B1 M (Hpm →Hpl ) νi (w ◦ ϕi ); ∂B1 Hpm . It is clear that the last norm does not exceed
 1/p n−1 p [νj νi (w ◦ ϕi )] ◦ ϕ−1 ; R  m j Hp j

1/p
 n−1 p ≤c [νj νi (w ◦ ϕ)] ◦ ϕ−1 ; R  ≤ c w; Rn−1 Hpm m i Hp j

and the lower estimate for the norm in M (Hpm (∂B1 ) → Hpl (∂B1 )) follows.  We note that Hpl (∂B1 ) can be supplied with an equivalent norm using the operator (1 − δ)l/2 where δ is the Beltrami operator on the sphere. Namely, v; ∂B1 Hpl ∼ (1 − δ)l/2 v; ∂B1 Lp .

(3.8.2)

It is essentially a consequence of the following property established in [Se]: (1 − δ)l/2 is the pseudo-differential operator with symbol |ξ|l (see also [Sh]).

3.8 Positive Homogeneous Multipliers

127

m 3.8.2 Other Normalizations of the Spaces hm p and Hp

The normalizations in the title of this subsection are given below in Lemma 3.8.1 and Corollary 3.8.1. Let 0 ≤ l ≤ m, p > 1, mp < n. Then the well known Hardy type inequality + u + + + (3.8.3) + m−l + l ≤ c uhm p |x| hp l is valid. It means that |x|l−m ∈ M (hm p → hp ) and it follows, for instance, from (3.7.1) together with the inequalities  dx ≤ c (mesn e)1−mp/n , mp |x| e



 Sl |y|l−m (x) ≤ c |x|−m .

Here, the second inequality results from the easily verified estimate  | |x + θy|l−m − |x|l−m | dθ ≤ c min{y, |x|}|x|l−m−1 . B1

Using (3.8.3), we may equip the space hm p , mp < n, with an equivalent norm. Let Gk = {x : 2k−1 < |x| < 2k+1 },

k = 0, ±1, . . . ,

and let {ψk } be a partition of unity on Rn \{0} subject to the covering {Gk }. We suppose that |Dα ψk | ≤ cα 2−k|α| . Lemma 3.8.1. Let p > 1, mp < n. The relation uhm ∼ p

∞


ψk uphm

1/p

p

(3.8.4)

k=−∞

holds. Proof. First we show that the proof of (3.8.4) easily reduces to the case 0 ≤ m ≤ 1. Suppose that m ≥ 1 and that the assertion is proved for m − 1. One can readily check that, for integer s, the norms of the functions |x|α Dα ψk in M hsp are uniformly bounded with respect to k. The same is true for fractional s by interpolation. Therefore, ψk ∇u m−1 − ∇(ψk u) m−1 hp hp ≤ σk u∇ψk hm−1 ≤ c |x|−1 σk uhm−1 , p p where

σk ∈ C0∞ (Gk ),

σ k ψk = ψk ,

|Dα σk | ≤ cα 2−k|α| .

(3.8.5)

128

3 Multipliers in Pairs of Potential Spaces

Consequently, ∇uphm−1 ≤ c

∞ 

ψk ∇uphm−1

p

≤c

p

k=−∞ ∞ 

  ∇(ψk u)phm−1 + |x|−1 σk uphm−1 . p

p

k=−∞

This and (3.8.3) imply uphm ≤ c p

∞    ψm uphm + σk uphm . p

p

k=−∞

Since {ψk } is a partition of unity and σk M hm ≤ const, the upper estimate p follows. for the norm uhm p Next we derive the lower bound. By (3.8.5) we have +∞ 

ψk uphm p

k=−∞

∞    ψk ∇uphm−1 + |x|−1 σk uphm−1 . ≤c p

p

k=−∞

Replacing σk by ψk in the last norm and using the induction hypothesis, we obtain that the right-hand side does not exceed   c uphm + |x|−1 uphm−1 . p

p

It remains to make use of (3.8.3). In the case m = 0 the relation (3.8.4) is trivial. Let 0 < m < 1. It is clear that ∞  Sm upLp ∼ ψk Sm upLp . (3.8.6) k=−∞

The definition of Sm and Minkowski’s inequality imply that ψk (x)(Sm u)(x) − (Sm (ψk u))(x) ≤


 0

 ≤c

By

2 1/2 |u(x + h)||ψk (x + h) − ψk (x)| dh y −1−2n−2m dy

|u(z)| |ψk (z) − ψk (x)| dz 

≤c

∞ 




∞

y −1−2n−2m dy

1/2

|x−z|

|ψk (z) − ψk (x)| |u(z)| dz . |z − x|n+m

By A(x) we denote the right-hand side and put gk = Gk−1 ∪ Gk ∪ Gk+1 . Since supp ψk ⊂ Gk , it follows that

3.8 Positive Homogeneous Multipliers

129

 ⎧ −k(n+m) ⎪ ⎪ c 2 |u(z)| dz if |x| < 2k−2 , ⎪ ⎪ ⎪ Gk ⎪ ⎪ ⎪ ⎪  ⎪ ⎨ −(n+m) |u(z)| dz if |x| > 2k+2 , A(x) ≤ c |x| ⎪ Gk ⎪ ⎪ ⎪ ⎪   ⎪
⎪ ⎪ |u(z)| dz |u(z)| dz  −k ⎪ ⎪ c 2 + if x ∈ gk . ⎩ n+m−1 n+m B2|x| |z − x| Rn \B2|x| |z| Therefore, ψk Sm u − Sm (ψk u)pLp ≤ c 2k(n−pm−pn) +c 2−kp


 gk

B2|x|

|u(z)| dz |z − x|n+m−1







p

|u(z)| dz

Gk

dx + c gk

Rn \B2|x|

p

|u(z)|dz p dx. |z|n+m

The first term on the right-hand side does not exceed c 2−kpm u; Gk pLp . The second one is majorized by  
p dζ p −kpm |u(z)| dz ≤ c 2 |u(z)|p dz. c 2−kp max n+m−1 x |ζ − x| B2k+3 B2k+3 B2k+3 The third term can be written as  2k+2
 rn−1 c 2k−2

∞

2r

p v(ρ) dρ dr, ρm+1

where v(ρ) is the mean value of |u| on ∂Bρ . Summing over k and using the one-dimensional Hardy inequality, we arrive at +∞ 

ψk Sm u − Sm (ψk u)pLp

k=−∞ −m

≤ c  |x|

upLp

∼

+∞ 

 |x|−m ψk upLp .

(3.8.7)

k=−∞

Now, (3.8.7) and (3.8.3) imply that Sm upLp

≤c

+∞ 

Sm (ψk u)pLp .

k=−∞

Thus the upper bound for the norm in hm p follows. Also, by (3.8.7) we have +∞ 

Sm (ψk u)pLp ≤ Sm upLp +  |x|−m upLp .

k=−∞

Applying (3.8.3) once again, we complete the proof.



130

3 Multipliers in Pairs of Potential Spaces

Relation (3.8.4) and the equivalence uHpm ∼ uhm + uLp p imply the result similar to the last lemma for the space Hpm . Corollary 3.8.1. Let p > 1 and mp < n. Then u

Hpm

∼

+∞


ψk upHpm

1/p .

k=−∞

A direct corollary of Lemma 3.8.1 is the following assertion on the norl malization of the space M (hm p → hp ), pm < n, which will be used in Subsect. 3.8.3. Corollary 3.8.2. If p > 1 and pm < n, then γM (hm ∼ l p →hp )

sup

−∞ 1

which follows directly from the identity 2[u(x + h) − u(x)] = −[u(x + 2h) − 2u(x + h) + u(x)] + [u(x + 2h) − u(x)]. Similarly to the case of Sobolev spaces in Sect. 2.1, one can show that the space M (Bpm → Bpl ) is trivial provided that m < l. We formulate the main result of this chapter. l Theorem 4.1.1. Let 0 < l ≤ m, p ∈ (1, ∞), and let γ ∈ Bp,loc . A function γ m l l belongs to M (Bp → Bp ) if and only if γ ∈ Bp,loc , Dp,l γ ∈ M (Bpm → Lp ), and either γ ∈ L1,unif for m > l or γ ∈ L∞ for m = l. The equivalence relation , Dp,l γ; eLp γL1,unif , m > l, γM (Bpm →Bpl ) ∼ sup + (4.1.4) 1/p m=l γL∞ , e [Cp,m (e)]

holds, where e is an arbitrary compact set in Rn . The relation (4.1.4) remains valid if the condition d(e) ≤ 1 is added, where d(e) is the diameter of e. For mp > n the statement of the above theorem can be simplified. Namely, the relation (4.1.4) is equivalent to γM (Bpm →Bpl ) ∼ γBp,unif l

for m ≥ l,

(4.1.5)

and for lp > n γM Bpl ∼ Dp,l γLp,unif + γL∞ .

(4.1.6)

Various upper estimates for the norm in M (Bpm → Bpl ) are derived in Sects. 4.4 and 4.5. The following assertion concerning the composition ϕ(γ), where γ ∈ M (Wpm → Wpl ), 0 < l < 1, p > 1, is proved in the last Sect. 4.6. If a function ϕ satisfies the H¨older condition |ϕ(t + τ ) − ϕ(t)| ≤ A |τ |ρ ,

|τ | < 1,

with ρ ∈ (0, 1), then ϕ(γ) ∈ M (Wpm−l+r → Wpr ) for any r ∈ (0, lρ).

4.2 Properties of Besov Spaces 4.2.1 Survey of Known Results We start with three classical properties of the space Bpl . Proposition 4.2.1. (see [St2], Sect. 5.1) The equivalence relation uBpk ∼ Λα uBpk−α , holds, where p ∈ (1, ∞), α ∈ (0, k), and Λ = (1 − ∆)1/2 .

(4.2.1)

4.2 Properties of Besov Spaces

135

Proposition 4.2.2. (see [St2], Sect. 5.3) The inequalities uHpl ≤ c uBpl

for 1 < p ≤ 2

uBpl ≤ c uHpl

for 2 ≤ p < ∞

and hold. Proposition 4.2.3. (see [Bes]) Let Rn = Rm ×Rk = {x = (y, z) : y ∈ Rm , z ∈ Rk }. The relation
  dη dz 1/p m p ∆(2) uBpl ∼ uLp + η ∇k u(·, z); R Lp |η|m+pα Rk Rm
  dζ dy 1/p (2) + ∆ζ ∇k u(y, ·); Rk pLp k+pα (4.2.2) |ζ| Rm Rk holds, where l = k + α > 0 and p ∈ [1, ∞). Similarly, for noninteger l,
  dη dz 1/p uWpl ∼ uLp + ∆η ∇[l] u(·, z); Rm pLp m+p{l} |η| Rk Rm
  dζ dy 1/p + ∆ζ ∇[l] u(y, ·); Rk pLp k+p{l} . (4.2.3) |ζ| Rm Rk Analogous norms appear under decomposition of Rn into more than two factors. Now we recall well known trace properties of Besov spaces (see [Usp] or [Bur], Sect. 5.4). Proposition 4.2.4. Let U ∈ C0∞ (Rn+s ) be any extension of u ∈ C0∞ (Rn ) onto the space Rn+s = {z = (x, y) : x ∈ Rn , y ∈ Rs }. (i) If p ∈ (1, ∞) and l > 0, then u; Rn Bpl ∼ inf U ; Rn+s Hpl+s/p . {U }

(4.2.4)

(ii) If p ∈ [1, ∞) and l > 0, then u; Rn Bpl ∼ inf U ; Rn+s Bpl+s/p . {U }

(iii) If p ∈ [1, ∞) and l > 0, then
   1/p u; Rn Bpl ∼ inf |y|p(1−{l})−s |∇[l]+1 U |p + |U |p dz . {U }

(4.2.5)

(4.2.6)

Rn+s

Another classical property of Bpl is the following Sobolev type imbedding result.

4 The Space M (Bpm → Bpl ) with p > 1

136

Proposition 4.2.5. (see [Bes], [Tr3], Sect. 2.8). The inequality uLq ≤ c uBpl holds with p > 1 and q = pn/(n − pl) q ∈ [p, ∞)

if n > pl, if n = pl,

q ∈ [p, ∞]

if n < pl.

The next assertion follows from (4.2.4) and Theorem 3.1.2. Proposition 4.2.6. Let {ηj }j≥0 be the same sequence as in Theorem 3.1.2. Then 1/p
 uBpl ∼ uηj pB l . p

j≥0

One can prove either by the Banach isomorphism theorem or directly that the equivalence relation (4.2.7) uBpl ∼ Dp,l uLp holds for functions u ∈ Bpl with supports in the unit ball. This gives Proposition 4.2.7. Let u ∈ Bpm and supp u ⊂ Bδ . Then for any k ∈ (0, l] δ k Dp,l uLp + uLp ≤ c δ l Dp,l uLp .

(4.2.8)

This and Proposition 4.2.6 imply (j)

Corollary 4.2.1. Let {Bδ } be a covering of Rn by open balls of radius δ ∈ (0, 1) with finite multiplicity depending only on n. Further, let u(j) ∈ Wpl and (j)

supp u(j) ⊂ Bδ . Then 



u(j) pW l ≤ c



p

j

Dp,l u(j) pLp .

j

4.2.2 Properties of the Operators Dp,l and Dp,l We start this section with a composition property of the operator Dp,l . Lemma 4.2.1. For any α, β > 0 with α + β < 1 the inequality Dp,α Dp,β uLp ≤ c Dp,α+β uLp holds.

4.2 Properties of Besov Spaces

Proof. Let t ∈ Rn and let ut (x) = u(x + t). We have
   |(Dp,β u)(x) − (Dp,β ut )(x)| ≤ |∆h u(x) − ut (x) |p

137

dh 1/p . |h|n+pβ

Therefore,  Dp,α Dp,β upLp ≤

  |∆h u(x) − ut (x) |p

dhdtdx |h|n+pβ |t|n+pα

.

We write the integral over R3n on the right-hand side as the sum of two integrals, one of which is taken over |h| ≤ |t| and does not exceed    dt |∆h u(x)|p dh dx n+pα |h|n+pβ |t| |t|≥|h|    dt dh |∆h ut (x)|p dx. + |h|n+pβ |t|≥|h| |t|n+pα Clearly, the second term coincides with the first one which in its turn is equal to   dh c dx |∆h u(x)|p n+(α+β)p = c Dp,α+β upLp . |h| The integral over the set |h| > |t| is estimated in the same way. The result follows.  We proceed with elementary upper pointwise estimates for Dp,l and Dp,l . Lemma 4.2.2. For any positive l > 0 and m > 0 the inequalities   (Dp,l u)(x) ≤ Jm Dp,l Λm u (x),   (Dp,l u)(x) ≤ Jm Dp,l Λm u (x)

(4.2.9) (4.2.10)

hold with Jm and Λm defined in Sects. 1.2.1 and 3.1, respectively. Proof. Let f = Λm u and let l = k + α, α ∈ (0, 1]. Clearly, (Dp,l u)(x) = (Dp,l Jm f )(x)   p
  = Gm (x−ξ+2h)−2Gm (x−ξ+h)+Gm (x−ξ) ∇k f (ξ)dξ

dh 1/p |h|n+pα  
  p dh 1/p = . Gm (x − ξ) ∇k f (ξ + 2h) − 2∇k f (ξ + h) + ∇k f (ξ) dξ |h|n+pα Using Minkowski’s inequality, we obtain Dp,l u ≤ Jm Dp,l f . An analogous esti mate for Dp,l u with {l} > 0 does not need a separate proof in view of (4.1.3). 

138

4 The Space M (Bpm → Bpl ) with p > 1

4.2.3 Pointwise Estimate for Bessel Potentials Henceforth we need the following modification of Hedberg’s inequality (1.2.17). Lemma 4.2.3. Let M be the Hardy–Littlewood maximal operator in Rn and (n+s) let Jr denote the Bessel potential in Rn+s , s ≥ 1. Then, for any nonnegative function f ∈ Lp (Rn+s ), p > 1, and for almost all x ∈ Rn 1−θ  (n+s) θ  (n+s) (Jrθ+s/p f )(x, 0) ≤ c (Jr+s/p f )(x, 0) MF (x) , (4.2.11) where F (x) = f (x, ·); Rs Lp and 0 < θ < 1. Proof. Let δ ∈ (0, 1] and let Eδ (x) = {ζ = (ξ, η) : ξ ∈ Rn , η ∈ Rs , (x − ξ)2 + η 2 > δ 2 }. We express the potential on the left-hand side of (4.2.11) as the sum of two integrals one of which is over Eδ (x). Let rθ < n + s/p . Then in view of (1.2.4) we have  Grθ+s/p (x − ξ, η) f (ξ, η) dξ dη  ≤c

Rn+s \Eδ (x)


 F (ξ)

Bδ (x)

|η| n + s/p .

Next we estimate the integral over Eδ (x). In the case rθ < n + s/p we have   f (ξ, η) dξ dη Grθ+s/p (x − ξ, η) f (ξ, η) dξ dη ≤ c (|x − ξ| + |η|)n−rθ+s/p
Eδ (x) Eδ (x)\E1 (x)  √ f (ξ, η) dξ dη 2 2 +c e− (x−ξ) +η (|x − ξ| + |η|)(n+1−rθ+s/p
)/2 E1 (x)  f (ξ, η) dξ dη ≤ c δ −r(1−θ) n−r+s/p
Eδ (x)\E1 (x) (|x − ξ| + |η|)  √ f (ξ, η) dξ dη 2 2 −r(1−θ)/2 +cδ e− (x−ξ) +η (|x − ξ| + |η|)(n+1−r+s/p
)/2 E1 (x) ≤ c δ −r(1−θ) (Jr+s/p f )(x, 0). (n+s)

The case rθ ≥ n + s/p is treated the same way. As a result we have  Grθ+s/p (x − ξ, η) f (ξ, η) dξ dη Eδ (x)

⎧ (n+s) ⎪ c δ −r(1−θ) (Jr+s/p f )(x, 0), ⎪ ⎪ ⎨ (n+s) ≤ c (1 + | log δ|) (Jr+s/p f )(x, 0), ⎪ ⎪ ⎪ (n+s) ⎩ c (Jr+s/p f )(x, 0),

rθ < n + s/p , rθ = n + s/p ,

(4.2.15)

rθ > n + s/p .

Further, we recall that Gr (z) = O(e−c|z| ) for |z| > 1 to obtain   Grθ+s/p (x − ξ, η) f (ξ, η) dξ dη ≤ c F (ξ)e−c|x−ξ| dξ E1 (x)

=c

∞   j=0

2j n + s/p we have  Grθ+s/p (x − ξ, η) f (ξ, η) dξ dη  ≤

Rn+s \E1 (x)

(n+s)

Rn+s \E1 (x)

f (ξ, η) dξ dη ≤ c (Jrθ+s/p f )(x, 0)

which together with (4.2.15) yields  (n+s)   (n+s)  Jrθ+s/p f (x, 0) ≤ c Jr+s/p f (x, 0).

(4.2.17)

For rθ < n + s/p estimates (4.2.14) –(4.2.16) imply that  (n+s)    (n+s) Jrθ+s/p f (x, 0) ≤ c δ rθ (MF )(x) + δ −r(1−θ) (Jr+s/p f )(x, 0) for all δ ∈ (0, ∞). Minimizing the right-hand side with respect to δ, we arrive at (4.2.11). If rθ = n + s/p , then by (4.2.14) and (4.2.15)   (n+s) (n+s) (Jrθ+s/p f )(x, 0) ≤ c (1 + | log δ|) δ rθ (MF )(x) + (Jr+s/p f )(x, 0) (4.2.18) for all δ ∈ (0, 1). In the case (n+s)

(MF )(x) ≤ (Jr+s/p f )(x, 0), by (4.2.16) we have  (n+s)   (n+s)  Jrθ+s/p f (x, 0) ≤ c Jr+s/p f (x, 0)

(4.2.19)

and (4.2.11) follows from (4.2.16) and (4.2.19). Let (n+s)

(MF )(x) > (Jr+s/p f )(x, 0). We put (n+s)

δ rθ = (Jr+s/p f )(x, 0)/(MF )(x) in (4.2.18). Then  (n+s)  (n+s) Jrθ+s/p f (x, 0) ≤ c (1 + | log δ|)(Jr+s/p )f (x, 0)  (n+s) 1−θ θ  = c (1 + | log δ|)δ rθ(1−θ) (Jr+s/p f )(x, 0) (MF )(x) . Since δ ∈ (0, 1], inequality (4.2.11) is proved. In the case rθ > n + s/p estimate (4.2.11) results from (4.2.16) and (4.2.17). 

4.3 Proof of Theorem 4.1.1

141

4.3 Proof of Theorem 4.1.1 4.3.1 Estimate for the Product of First Differences While deriving estimates involving the second difference of the product of two functions, we need to estimate the product of their first differences, which is based on the next lemma. Lemma 4.3.1. For p ∈ (1, ∞), δ ∈ (0, 1) and any integer k ≥ 1, we have
  Dp,δ γ; eLp dhdx 1/p ≤ c sup uBpk . (4.3.1) |∆h γ(x)∆h u(x)|p n+p 1/p |h| e [Cp,k−1+δ (e)] Proof. Let U ∈ C0∞ (Rn+s ) be an extension of a function u ∈ Bpk (Rn ) to Rn+s with k subject to k < n + s/p . By f we denote the function Λk+s/p U . Then  u(x) = Gk+s/p (x − ξ, η)f (ξ, η)dξdη. Rn+s

We shall use the properties of the function Gr listed in Sect. 1.2.1. Let us introduce the sets N1 = {(ξ, η) : 4|h| < |x − ξ| + |η| < 1}, N2 = {(ξ, η) : |x − ξ| + |η| < min(1, 4|h|)}, N3 = {(ξ, η) : |x − ξ| + |η| > max(1, 4|h|)}, N4 = {(ξ, η) : 4|h| > |ξ − x| + |η| > 1}. It is clear that  |∆h u(x)| ≤

Rn+s

Gk+s/p (x − ξ + h, η) − Gk+s/p (x − ξ, η) |f (ξ, η)|dξdη.

We represent the last integral as the sum of four integrals over N1 , . . . , N4 and estimate each of them. First,   −(n−k+1+s/p
) ≤ c |h| tθ |f (ξ, η)|dξdη, N1

N1

  where t2θ = (x + θh − ξ)2 + η 2 , θ ∈ (0, 1). Since t2θ ≥ c (x − ξ)2 + η 2 on N1 , it follows that   −(n−k+1+s/p
−δ) 1−δ ≤ c |h| t0 |f (ξ, η)|dξdη N1

N1

≤ c |h|

1−δ

 −(k−1+s/p+δ)  Λ |f | (x, 0).

(4.3.2)

4 The Space M (Bpm → Bpl ) with p > 1

142

The integral over N2 is dominated by   −(n−k+s/p
) −(n−k+s/p
)  t1 |f (ξ, η)|dξdη + t0 N2

 ≤ c |h|1−δ

N2

 −(n−k+s/p
+1−δ) −(n−k+s/p
+1−δ) t1 + t0 |f (ξ, η)|dξdη.

Consequently,    ≤ c|h|1−δ (Λ−(k−1+s/p+δ) |f |)(x+h, 0)+(Λ−(k−1+s/p+δ) |f |)(x, 0) . (4.3.3) N2

Using (2.8.5), we obtain  N3

Hence,

 ≤ c |h| 

 N3

≤ c |h|1−δ

N3

e−t0 /2 |f (ξ, η)|dξdη.

(k−n−s/p
−1+δ)/2 −t0 /4

N3

t0

e

|f (ξ, η)|dξdη

x  ≤ c |h|1−δ (Λ−(k−1+s/p+δ) |F |) , 0 , 4 where F (ξ, η) = f (4ξ, 4η). In a similar way we find that   ≤c (e−t1 /2 + e−t0 /2 )|f (ξ, η)|dξdη  ≤ c |h|1−δ

N4

N4

(4.3.4)

N4

 (k−n−s/p
−1+δ)/2 −t1 /4 (k−n−s/p
−1+δ)/2 −t0 /4  t1 |f (ξ, η)|dξdη, e +t0 e

and consequently 
x + h   x  , 0 + (Λ−(k−1+s/p+δ) |F |) , 0 . ≤ c |h|1−δ (Λ−(k−1+s/p+δ) |F |) 4 4 N4 Adding the last inequality to (4.3.2)–(4.3.4) and noting that Gr (az) ≥ Gr (z) for any constant a < 1, we conclude that
 x+h   x  , 0 +(Λ−(k−1+s/p+δ) |F |) , 0 . |∆h u(x)| ≤ c |h|1−δ (Λ−(k−1+s/p+δ) |F |) 4 4 Hence,

 

dhdx |∆h γ(x)∆h u(x)|p n+p |h| 
  p   dhdx x (Λ−(k−1+s/p+δ) |F |) , 0 ≤c |∆h γ(x)|p n+pδ 4 |h|   p  −(k−1+s/p+δ) dhdx = c1 |F |)(x, 0) (Λ |∆h γ(4x)|p n+pδ . |h|

4.3 Proof of Theorem 4.1.1

By (4.3.12),

  |∆h γ(x)∆h u(x)|p

 

143

dhdx |h|n+p

dhdx |h|n+pδ (Λ−(k−1+s/p+δ) |F |)(·, 0)pW k−1+δ . p Cp,k−1+δ (e)

|∆h γ(4x)|p ≤ c sup

e

e

Clearly,   |∆h γ(4x)|p e

dhdx =c |h|n+pδ

  |∆h γ(x)|p 4e

dhdx |h|n+pδ

(4.3.5)

(4.3.6)

and Cp,k−1+δ (e) ≥ c Cp,k−1+δ (4e),

(4.3.7)

where 4e = {x : x/4 ∈ e}. Note also that (Λ−(k−1+s/p+δ) |F |)(·, 0)Wpk−1+δ ≤ c F ; Rn+s Lp = c 4−(n+s) f ; Rn+s Lp = c 4−(n+s) Λk+s/p U ; Rn+s Lp . This, together with (4.3.5)–(4.3.7), implies that   dhdx |∆h γ(x)∆h u(x)|p n+p |h| ≤ c sup e

Dp,δ γ; epLp Cp,k−1+δ (e)

Λk+s/p U ; Rn+s pLp .

Minimizing the last norm over all extensions U , we complete the proof.



4.3.2 Trace Inequality for Bpk , p > 1 Lemma 4.3.2. Let µ be a measure in Rn , p ∈ (1, ∞), k ∈ (0, ∞). The best constant C in the inequality  |u|p dµ ≤ C upB k , u ∈ C0∞ , (4.3.8) p

is equivalent to sup

µ(e) , Cp,k (e)

where e is an arbitrary compact set of positive capacity Cp,k (e).

(4.3.9)

144

4 The Space M (Bpm → Bpl ) with p > 1

Proof. Let U ∈ C0∞ (Rn+1 ) be an arbitrary extension of u ∈ C0∞ (Rn ) to Rn+1 . By Theorem 3.1.4,  µ(e) U ; Rn+1 p k+1/p . |u|p dµ ≤ c sup (n+1) Hp n n e⊂R R Cp,k+1/p (e) (n+1)

In the present proof we use the notation Cp,k+1/p (e) temporarily in order to stress that the functions in the definition (1.2.6) of the capacity are given on Rn+1 instead of Rn . Minimizing the right-hand side over all extensions of u and applying the relation (n+1)

Cp,k (e) ∼ Cp,k+1/p (e),

(4.3.10)

which follows from the definition of capacity and the equivalence u; Rn Bpk ∼ inf U ; Rn+1 Hpk+1/p , {U }

(4.3.11)

we obtain the required upper bound for C. The lower bound is obvious.



Remark 4.3.1. The above result can be found in [Maz9] and [Maz11]. For its generalizations to Besov spaces with three indices see [Wu] and [AX]. By Lemma 4.3.2, γM (Bpk →Lp ) ∼ sup e

γ; eLp [Cp,k (e)]1/p

(4.3.12)

for p ∈ (1, ∞). Hence, in view of Lemma 3.2.2, γM (Bpk →Lp ) ∼ γM (Hpk →Lp ) ∼

γ; eLp . 1/p e,diam(e)≤1 [Cp,k (e)] sup

(4.3.13)

By Proposition 3.1.4 containing the estimates for the capacity of a ball, one obtains the following relations from (4.3.12): if pk > n, then γM (Bpk →Lp ) ∼ γLp,unif ; (4.3.14) if pk < n, then γM (Bpk →Lp ) ≥ c

sup x∈Rn ,r∈(0,1)

rk−n/p γ; Br (x)Lp ;

(4.3.15)

if pk = n, then γM (Bpk →Lp ) ≥ c

 2 (p−1)/p log γ; Br (x)Lp . r x∈Rn ,r∈(0,1) sup

(4.3.16)

4.3 Proof of Theorem 4.1.1

145

By Propositions 3.1.2, 3.1.3 the following upper estimates for the norm in M (Bpk → Lp ) hold: if pk < n, then γM (Bpk →Lp ) ≤ c

(mesn e)k/n−1/p γ; eLp ;

sup

(4.3.17)

e,diam(e)≤1

if pk = n, then γM (Bpk →Lp ) ≤ c



sup

log

e,diam(e)≤1

2n (p−1)/p γ; eLp . mesn e

(4.3.18)

4.3.3 Auxiliary Assertions Concerning M (Bpm → Bpl ) We start with inequalities for mollifiers of multipliers. Lemma 4.3.3. Let γρ denote a mollifier of a function γ which is defined as    −n γρ (x) = ρ K ρ−1 (x − ξ) γ(ξ)dξ, where K ∈ C0∞ (B1 ), K ≥ 0, and KL1 = 1. The inequalities γρ M (Bpm →Bpl ) ≤ γM (Bpm →Bpl ) ≤ lim inf γρ M (Bpm →Bpl ) ,

(4.3.19)

γρ M (Bpm →Lp ) ≤ γM (Bpm →Lp ) ≤ lim inf γρ M (Bpm →Lp ) ,

(4.3.20)

ρ→0

ρ→0

and sup e

Dp,l γρ ; eLp Dp,l γ; eLp ≤ sup 1/p 1/p [Cp,m (e)] e [Cp,m (e)]

(4.3.21)

hold. Proof. Let u ∈ C0∞ . Clearly,  p dh
  1/p
ξ  −n (2) γρ uBpl = ∇l−1,x ∆h (γ(x − ξ)u(x))dξ ρ dx K n+1 ρ |h|  

ξ  1/p p γ(x − ξ)u(x)dξ dx + . (4.3.22) ρ−n K ρ By Minkowski’s inequality, Dp,l (γρ u)Lp ≤ ρ−n

 K


ξ  ρ

  Dp,l γ(· − ξ)u(·) Lp dξ.

This and (4.3.22) imply that −n

γρ uBpl ≤ ρ

 K


ξ  ρ

γ(· − ξ)u(·)Bpl dξ.

4 The Space M (Bpm → Bpl ) with p > 1

146

Since γ(· − ξ)u(·)Bpl ≤ γM (Bpm →Bpl ) uBpm , the left inequality (4.3.19) follows. One can prove the right inequality (4.3.19) duplicating the argument in Lemma 2.3.1. The proof of (4.3.20) is obvious. To derive the estimate (4.3.21) we use Minkowski’s inequality once more to obtain 
  p 1/p Dp,l γ(x − ρz) dx dz K(z) Dp,l γρ ; eLp e ≤ [Cp,m (e)]1/p [Cp,m (e)]1/p 
  p 1/p Dp,l γ(ξ) dξ K(z) dz Dp,l γ; eLp E ≤ B1 ≤ KL1 sup 1/p 1/p [Cp,m (E)] e [Cp,m (e)] where E = {x − ρz : x ∈ e, z ∈ B1 }. The proof is complete. We use the interpolation properties 
Bpm−k = Bpm , Hpm−l 
Bpm−k = Bpm , Bpm−l

and



(4.3.23)

k/l,p

k/l,p

,

(4.3.24)

where l < k < m (see, [T], Th. 2.4.2). In particular, (4.3.24) implies that γM Bpr ≤ c γθM Bpσ γ1−θ M Bpτ ,

(4.3.25)

where p ∈ (1, ∞), σ > τ > 0, 0 < θ < 1, and r = θσ + (1 − θ)τ . It follows from (4.3.12) and (4.3.24) that γ ∈ M (Bpm → Bpl ) ∩ M (Bpm−l → Lp ) implies that γ ∈ M (Bpm−k → Bpl−k ) for 0 < k < l. Moreover, 1−k/l

k/l

γM (Bpm−k →Bpl−k ) ≤ c γM (B m →B l ) γM (B m−l →L p

p

p

p)

(4.3.26)

for 0 < k < l < m and 1−k/l

k/l

γM Bpl−k ≤ c γM B l γL∞ p

for 0 < k < l. 4.3.4 Lower Estimates for the Norm in M (Bpm → Bpl ) The following is the main result of this subsection.

(4.3.27)

4.3 Proof of Theorem 4.1.1

147

Lemma 4.3.4. Let 0 < l ≤ m and p ∈ (1, ∞). Then γL∞ ≤ γM Bpl

f or m = l

(4.3.28)

and γM (Bpm−l →Lp ) ≤ c γM (Bpm →Bpl )

f or m > l.

(4.3.29)

Proof. Let u ∈ Bpl and let N be a positive integer. Clearly, 1/N

1/N

1/N

p

p

γ N uLp ≤ γ N uB l ≤ γM Bpl uB l . Passing to the limit as N → ∞ we arrive at (4.3.28). Now suppose that 0 < l < m. Let γρ be the mollification of γ ∈ M (Bpm → l Bp ). By Lemma 4.3.3, it suffices to prove (4.3.29) for γρ . To simplify the notation we write γ in place of γρ . We consider two cases: m ≥ 2l and 2l > m > l. Assume first that m ≥ 2l. In m−l+1/p (Rn+1 ) which is an extension view of Lemma 4.2.4, there exists U ∈ Hp m−l n n+1 such that of the function u ∈ Bp (R ) to R U ; Rn+1 Hpm−l+1/p ≤ c u; Rn Bpm−l .

(4.3.30)

By the same lemma, the converse estimate u; Rn Bpm−l ≤ c U ; Rn+1 Hpm−l+1/p

(4.3.31) (n+1)

holds for all extensions U . Let us take U as the Bessel potential Jm−l+1/p f with density f ∈ Lp (Rn+1 ). By Lemma 4.2.3, l/m  (n+1) (m−l)/l  MF (x) |u(x)| ≤ c (Jm+1/p |f |)(x, 0) , where F (x) = f (x, ·); R1 Lp . Therefore,  (n+1)  l/m (m−l)/m γuLp ≤ c f ; Rn+1 Lp  |γ|l/(m−l) Jm+1/p |f | (·, 0)Lp . According to Lemma 4.3.2, the right-hand side does not exceed  pl m−l  |γ| m−l dx mp l l   1− m (n+1) n+1 m e c f ; R Lp γ Jm+1/p |f | (·, 0)B l sup . p Cp,l (e) e

(4.3.32)

Setting ϕ = |γ|1/(m−l) , ν = l, µ = m − l in Lemma 2.3.6, which is valid for all ν and µ such that 0 < µ < ν, we find that in the case m ≥ 2l the supremum in (4.3.32) is dominated by   |γ|p dx l/mp l/m c sup e ≤ c γM (B m−l →L ) . p p C (e) e p,m−l

4 The Space M (Bpm → Bpl ) with p > 1

148

Therefore, by (4.3.32) we obtain l

1−

l

l

1−

(n+1)

l

m m m γuLp ≤ cf ; Rn+1 Lmp γM (B . m →B l ) Jm+1/p |f |(·, 0)B m γ M (B m−l →L ) p p

p

p

p

Using first (4.3.31) and then the equality uHpk = Λk uLp and (4.3.30), we obtain (n+1)

(n+1)

Jm+1/p |f |(·, 0)Bpm ≤ c Jm+1/p |f |; Rn+1 Hpm+1/p = c f ; Rn+1 Lp = c U ; Rn+1 Hpm−l+1/p ≤ c u; Rn Bpm−l . Thus, l/m

(m−l)/m

γuLp ≤ c γM (B m−l →L ) γM (B m →B l ) uBpm−l , p

p

p

p

which implies (4.3.29) for m ≥ 2l. Suppose that 2l > m > l. Let µ be an arbitrary positive number less than m − l. By (4.3.26) with k = l − µ, (l−µ)/l

µ/l

γM (Bpm−l+µ →Bpµ ) ≤ c γM (B m−l →L ) γM (B m →B l ) . p

p

p

p

Since m − l + µ > 2µ, it follows from the first part of the proof that (4.3.29) holds with m and l replaced by m − l + µ and µ, respectively, i.e. γM (Bpm−l →Lp ) ≤ c γM (Bpm−l+µ →Bpµ ) . Consequently, (l−µ)/l

µ/l

γM (Bpm−l →Lp ) ≤ c γM (B m−l →L ) γM (B m →B l ) p

p

p

p



and (4.3.29) is proved for 2l > m > l as well. By Lemma 4.3.4 and (4.3.12), the following assertion holds. Corollary 4.3.1. Let γ ∈ M (Bpm → Bpl ), 0 < l < m. Then sup e

γ; eLp ≤ c γM (Bpm →Bpl ) . [Cp,m−l (e)]1/p

Lemma 4.3.4 in combination with (4.3.26) and (4.3.27) implies: Corollary 4.3.2. Let γ ∈ M (Bpm → Bpl ), 0 < l ≤ m. Then γ ∈ M (Bpm−k → Bpl−k ),

0 < k < l,

and γM (Bpm−k →Bpl−k ) ≤ c γM (Bpm →Bpl ) .

4.3 Proof of Theorem 4.1.1

149

The following assertion contains an estimate for derivatives of a multiplier. Lemma 4.3.5. Let γ ∈ M (Bpm → Bpl ), 0 < l ≤ m. Then Dα γ ∈ M (Bpm → Bpl−|α| ) for any multi-index α of order |α| ≤ l and Dα γM (B m →Bpl−|α| ) ≤ c γM (Bpm →Bpl ) . p

Proof. It suffices to consider the case |α| = 1, l ≥ 1. Clearly, u∇γBpl−1 ≤ uγBpl + γ∇uBpl−1   ≤ γM (Bpm →Bpl ) + γM (Bpm−1 →Bpl−1 ) uBpm . Hence, using Corollary 4.3.2, we find that u∇γBpl−1 ≤ c γM (Bpm →Bpl ) uBpm 

which completes the proof.

Corollary 4.3.3. Let γ ∈ M (Bpm → Bpl ), 0 < l ≤ m. Then, for any ε > 0 m−l−|α|

and every multi-index α of order |α| ≤ l, Dα γ ∈ M (Bp inequality

→ Lp ), and the

Dα γM (Bpm−l+|α| →Lp ) ≤ ε γM (Bpm →Bpl ) + c(ε) γM (Bpm →Lp ) holds. Proof. The result follows by Lemma 4.3.5 and inequality (4.3.26). Lemma 4.3.4 and Corollary 4.3.3 imply the following assertion. Corollary 4.3.4. Let γ ∈ M (Bpm → Bpl ), 0 < l ≤ m. Then, for any multim−l−|α|

index α of order |α| ≤ l, Dα γ ∈ M (Bp

→ Lp ), and the inequality

D γM (Bpm−l+|α| →Lp ) ≤ c γM (Bpm →Bpl ) α

holds. 4.3.5 Proof of Necessity in Theorem 4.1.1 In this section we derive the inequalities sup e

Dp,l γ; eLp + sup γ; B1 (x)Lp ≤ c γM (Bpm →Bpl ) , [Cp,m (e)]1/p x∈Rn

m > l, (4.3.33)

and

Dp,l γ; eLp + γL∞ ≤ c γM Bpl . [Cp,l (e)]1/p e The core of the proof is the following assertion. sup

(4.3.34)

150

4 The Space M (Bpm → Bpl ) with p > 1

Lemma 4.3.6. Let γ ∈ M (Bpm → Bpl ), where 0 < l ≤ m and p ∈ (1, ∞). Then Dp,l γ; eLp sup ≤ c γM (Bpm →Bpl ) . (4.3.35) 1/p e [Cp,m (e)] Proof. We use induction on l and start by showing that (4.3.35) holds for l ∈ (0, 1]. (i) Let l ∈ (0, 1). We have   uDp,l γLp ≤ c γuBpl + γDp,l uLp   ≤ c γM (Bpm →Bpl ) uBpl + γDp,l uLp .

(4.3.36)

Consider first the case m = l. Clearly, γDp,l uLp ≤ γL∞ uBpl which together with (4.3.36) and (4.3.28) gives uDp,l γLp ≤ c γM Bpl uBpl . Therefore, Dp,l γM (Bpl →Lp ) ≤ c γM Bpl and, in view of (4.3.12), we obtain (4.3.35). Suppose now that l < m. By (4.2.9), γDp,l uLp ≤ γM (Bpm−l →Lp ) Jm−l Dp,l Λm−l uBpm−l .

(4.3.37)

By Lemma 4.2.1, the last norm does not exceed c Dp,l Λm−l uLp ≤ c Λm−l uBpl ≤ c uBpm which in combination with (4.3.37) implies that γDp,l uLp ≤ c γM (Bpm−l →Lp ) uBpm .

(4.3.38)

Using (4.3.36), (4.3.38) and Lemma 4.3.4, we arrive at uDp,l γLp ≤ c γM (Bpm →Bpl ) uBpm . Thus, Dp,l γM (Bpm →Lp ) ≤ c γM (Bpm →Bpl ) which together with (4.3.12) gives (4.3.35). (ii) Let l = 1. In view of the identity (2)

(2)

(2)

∆h (γu) = γ∆h u + u∆h γ + ∆2h γ∆2h u − 2∆h γ∆h u

(4.3.39)

4.3 Proof of Theorem 4.1.1

151

one has  uDp,1 γLp ≤ γuBp1 + γDp,1 uLp
  1/p +4 |∆h γ(x)∆h u(x)|p |h|−n−p dhdx

(4.3.40)

for any u ∈ C0∞ . We proceed separately for m = 1 and m > 1. Let first m = 1. Using (4.3.1) with k = 1 and δ ∈ (0, 1) together with (4.3.40) and (4.3.28), we find that
Dp,δ γ; eLp  uDp,1 γLp ≤ c γM Bp1 + sup uBp1 . [Cp,δ (e)]1/p e

(4.3.41)

In view of part (i) of this proof, the last supremum is majorized by c γM Bpδ . Hence (4.3.41) leads to the inequality sup e

Dp,1 γ; eLp ≤ c (γM Bp1 + γM Bpδ ). [Cp,1 (e)]1/p

(4.3.42)

Since by Corollary 4.3.2 γM Bpδ ≤ c γM Bp1 , we arrive at (4.3.35) for m = l = 1. Next we estimate the right-hand side of (4.3.40) for m > 1. By (4.2.9), its second term is majorized by γJm−1 Dp,1 Λm−1 uLp ≤ c γM (Bpm−1 →Lp ) Jm−1 Dp,1 Λm−1 uBpm−1 ≤ c γM (Bpm−1 →Lp ) Dp,1 Λm−1 uLp ≤ c γM (Bpm−1 →Lp ) Λm−1 uBp1 ≤ c γM (Bpm →Bp1 ) uBpm .

(4.3.43)

The last inequality in this chain follows from (4.2.1) and (4.3.29). We estimate the third term on the right-hand side of (4.3.40) using (4.3.1) with k = m > 1 and (4.3.35) with l = δ < 1. Then this term does not exceed c sup e

Dp,δ γ; eLp uBpm ≤ c γM (Bpm−1+δ →B δ ) uBpm . p [Cp,m−1+δ (e)]1/p

(4.3.44)

Furthermore, by Corollary 4.3.2 γM (Bpm−1+δ →B δ ) ≤ c γM (Bpm →Bp1 ) . p

Hence the third term on the right-hand side of (4.3.40) is dominated by c γM (Bpm →Bp1 ) uBpm .

152

4 The Space M (Bpm → Bpl ) with p > 1

This along with (4.3.40) and (4.3.43) implies that uDp,1 γLp ≤ c γM (Bpm →Bp1 ) uBpm and thus (4.3.35) holds for l = 1. (iii) Suppose that l is positive and integer, and that the lemma is proved for γ ∈ M (Bpm → Bpk ), where k is any positive integer not exceeding l − 1. Applying (4.3.39), we find that uDp,l γLp ≤ γuBpl + c

l−1 

 |∇j γ|Dp,l−j uLp + c

l−1 

j=0

+c

l−1
  

 |∇j u|Dp,l−j γLp

j=1

|∆h ∇j γ(x)|p |∆h ∇l−1−j u|p |h|−n−p dhdx

1/p .

(4.3.45)

j=0

By (4.2.9) with α = l − j and β = m − l + j, we have (Dp,l−j u)(x) ≤ (Jm−l+j Dp,l−j Λm−l+j u)(x). Therefore, for j = 1, . . . , l − 1 and m ≥ l,  |∇j γ|Dp,l−j uLp ≤ c ∇j γM (Bpm−l+j →Lp ) Jm−l+j Dp,l−j Λm−l+j uBpm−l+j ≤ c ∇j γM (Bpm−l+j →Lp ) Dp,l−j Λm−l+j uLp .

(4.3.46)

According to (4.2.1), Dp,l−j Λm−l+j uLp ≤ Λm−l+j uBpl−j ≤ c uBpm .

(4.3.47)

By Corollary 4.3.4, ∇j γM (Bpm−l+j →Lp ) ≤ c γM (Bpm →Bpl ) ,

j = 1, . . . , l − 1, m ≥ l. (4.3.48)

For j = 0 by Lemma 4.3.4 we obtain γDp,l uLp ≤ γM (Bpm →Bpl ) uBpm .

(4.3.49)

Unifying (4.3.46)–(4.3.49), we find that for all j = 0, . . . , l − 1 and 1 ≤ l ≤ m,  |∇j γ|Dp,l−j uLp ≤ c γM (Bpm →Bpl ) uBpm .

(4.3.50)

For j = 1, . . . , l − 1 we have  |∇j u|Dp,l−j γLp ≤ c sup e

Dp,l−j γ; eLp uBpm . [Cp,m−j (e)]1/p

(4.3.51)

4.3 Proof of Theorem 4.1.1

153

From the induction assumption and Corollary 4.3.2 it follows that for m ≥ l one has sup e

Dp,l−j γ; eLp ≤ c γM (Bpm−j →Bpl−j ) ≤ c γM (Bpm →Bpl ) [Cp,m−j (e)]1/p

(4.3.52)

which together with (4.3.51) implies that  |∇j u|Dp,l−j γLp ≤ c γM (Bpm →Bpl ) uBpm ,

j = 1, . . . , l − 1.

(4.3.53)

Next we estimate the last sum in (4.3.45). Let δ ∈ (0, 1) be such that m+δ is a noninteger. By (4.3.1) with γ replaced by ∇j γ, u replaced by ∇l−1−j u, and k = m − l + j + 1, each term of the last sum in (4.3.45) does not exceed c sup e

Dp,j+δ γ; eLp ∇l−1−j uBpm−l+j+1 . [Cp,m−l+j+δ (e)]1/p

(4.3.54)

By the induction assumption and Corollary 4.3.2 this implies that
  1/p |∆h ∇j γ(x)|p |∆h ∇l−1−j u|p |h|−n−p dhdx ≤ c γM (Bpm−l+j+δ →Bpj+δ ) uBpm ≤ c γM (Bpm →Bpl ) uBpm .

(4.3.55)

Combining this with (4.3.53) and (4.3.51), we obtain from (4.3.45) uDp,l γLp ≤ c γM (Bpm →Bpl ) uBpm

(4.3.56)

and thus (4.3.35) follows for all integer l. (iv) Now let l be a noninteger. Suppose that sup e

Dp,l γ; eLp ≤ c γM (Bpm →Bpl ) [Cp,m (e)]1/p

for all noninteger l ∈ (0, N ), where N is an integer. Let N < l < N + 1. In view of the equivalence Dp,l γ ∼ Dp,l γ we have uDp,l γLp ≤ γuBpl + c

N 

 |∇j γ|Dp,l−j uLp

j=0

+c

N 

 |∇j u|Dp,l−j γLp .

(4.3.57)

j=1

Let t ∈ (0, m − l + j) if m > l or m = l, j > 0 and let t = 0 if m = l and j = 0. By (4.2.10) with α = l − j and β = t one has     Dp,l−j u (x) ≤ Jt Dp,l−j Λt u (x).

154

4 The Space M (Bpm → Bpl ) with p > 1

Using (4.2.1), we find  |∇j γ|Dp,l−j uLp ≤ ∇j γM (Wpm−l+j →Lp ) Jt Dp,l−j Λt uWpm−l+j ≤ c ∇j γM (Bpm−l+j →Lp ) Dp,l−j Λt uWpm−l+j−t .

(4.3.58)

By definition of the operator Dp,l and the space Wpl , Dp,l−j vWpm−l+j−t = Dp,m−l+j−t Dp,{l} ∇[l−j] vLp + Dp,l−j vLp . We use Lemma 4.2.1 with α = m − l + j − t, β = {l} assuming t to be so close to m − l + j that 0 < m − t − [l] + j < 1. Then  Dp,m−l+j−t Dp,{l} ∇[l]−j vLp ≤ c Dp,m−t−[l]−j ∇[l]−j vLp ≤ c vWpm−t .

(4.3.59)

We may also choose t in such a way that m−t is a noninteger so that Wpm−t = Bpm−t . Then (4.3.57) together with (4.3.58) and (4.3.59), where v = Λt u, and Corollary 4.3.4 imply that  |∇j γ|Dp,l−j uLp ≤ c ∇j γM (Bpm−l+j →Lp ) Λt uBpm−t ≤ c γM (Bpm →Bpl ) uBpm .

(4.3.60)

By the induction hypothesis, we have Dp,l−j γ; eLp ∇j uBpm−j 1/p e [Cp,m−j (e)] ≤ c γM (Bpm−j →Bpl−j ) uBpm

 |∇j u|Dp,l−j γLp ≤ c sup

(4.3.61)

for j = 1, . . . , N which, together with Corollary 4.3.2, implies that  |∇j u|Dp,l−j γLp ≤ c γM (Bpm →Bpl ) uBpm . This together with (4.3.60) and (4.3.57) leads to uDp,l γLp ≤ c γM (Bpm →Bpl ) uBpm . The proof is complete.



The following simple corollary contains the required lower estimate of the norm in M (Bpm → Bpl ) in Theorem 4.1.1. It also finishes the proof of necessity in Theorem 4.1.1. Corollary 4.3.5. Let γ ∈ M (Bpm → Bpl ), where 0 < l ≤ m and p ∈ (1, ∞). Then (4.3.33) and (4.3.34) hold.

4.3 Proof of Theorem 4.1.1

155

Proof. Since γ ∈ M (Bpm → Bpl ), it follows that γηLp ≤ γM (Bpm →Bpl ) ηBpm for any η ∈ C0∞ (B2 (x)), η = 1 on B1 (x), where x is an arbitrary point of Rn . Therefore, sup γ; B1 (x)Lp ≤ c γM (Bpm →Bpl ) . x∈Rn

The result follows by combining this inequality with Lemma 4.3.6.



The next corollary contains one more lower estimate for the norm in the space M (Bpm → Bpl ). Corollary 4.3.6. Let γ ∈ M (Bpm → Bpl ), where 0 < l ≤ m, p ∈ (1, ∞). Then for any k = 0, . . . , [l], if l is a noninteger, and for any k = 0, . . . , l − 1, if l is an integer, the inclusion Dp,l−k γ ∈ M (Bpm−k → Lp ) holds and Dp,l−k γM (Bpm−k →Lp ) ≤ c γM (Bpm →Bpl ) . Proof. By Corollaries 4.3.5 and 4.3.2, sup e

Dp,l−k γ; eLp ≤ c γM (Bpm−k →Bpl−k ) ≤ c γM (Bpm →Bpl ) . [Cp,m−k (e)]1/p

(4.3.62) 

It remains to make use of (4.3.12).

4.3.6 Proof of Sufficiency in Theorem 4.1.1 The aim of this section is to prove the upper estimate for γM (Bpm →Bpl ) in (4.1.4). l Lemma 4.3.7. Let γ ∈ Bp,loc , p ∈ (1, ∞). Then for m > l
D γ; e  γ; eLp p,l Lp γM (Bpm →Bpl ) ≤ c sup + . 1/p 1/p [Cp,m−l (e)] e,diam(e)≤1 [Cp,m (e)]

(4.3.63)

For m = l the second term should be replaced by γL∞ . Proof. It follows from the finiteness of the right-hand side of (4.3.63) that γ ∈ L1,unif . Let γρ denote a mollifier of γ with radius ρ. Since γ ∈ L1,unif we see that all derivatives of γρ are bounded. Hence γρ ∈ M (Bpm → Bpl ). For integer l we find by (4.3.45) that l−1 l−1
  γρ uBpl ≤ c  |∇j γρ |Dp,l−j uLp +  |∇j u|Dp,l−j γρ Lp j=0

+

l−1
   j=0

j=0

|∆h ∇j γρ (x)|p |∆h ∇l−1−j u|p |h|−n−p dhdx

1/p  .

(4.3.64)

156

4 The Space M (Bpm → Bpl ) with p > 1

By Corollary 4.3.4, ∇j γρ M (Bpm−l+j →Lp ) ≤ c γρ M (Bpm−l+j+α →B α ) p

(4.3.65)

for any α ∈ (0, 1). In view of (4.3.26), for m > l the right-hand side in (4.3.65) does not exceed (l−α)/l α/l c γρ M (B m−l →L ) γρ M (B m →B l ) . p

p

p

p

Combining this fact with (4.3.46) and (4.3.47), we obtain  |∇j γρ |Dp,l−j uLp   ≤ εγρ M (Bpm →Bpl ) + c(ε)γρ M (Bpm−l →Lp ) uBpm ,

(4.3.66)

where j = 0, . . . , l − 1, and ε is an arbitrary positive number. In case m = l inequalities (4.3.65) and (4.3.27) imply that (l−j)/l

∇j γρ M (Bpj →Lp ) ≤ c γρ L∞

j/l

γρ M B l . p

Unifying this estimate with (4.3.46) and (4.3.47) for m = l, we obtain    |∇j γρ |Dp,l−j uLp ≤ εγρ M Bpl + c(ε)γρ L∞ uBpl . (4.3.67) It follows from (4.3.51), (4.3.52), and (4.3.26), (4.3.27) that for j > 0  |∇j u|Dp,l−j γρ Lp   ≤ εγρ M (Bpm →Bpl ) + c(ε)γρ M (Bpm−l →Lp ) uBpm ,

(4.3.68)

if m > l, and    |∇j u|Dp,l−j γρ Lp ≤ εγρ M Bpl + c(ε)γρ L∞ uBpl ,

(4.3.69)

if m = l. The third sum on the right-hand side of (4.3.64) is estimated by using (4.3.55) and (4.3.26), (4.3.27). It has the same majorant as the right-hand side of (4.3.68) for m > l or (4.3.69) for m = l. Thus, for m > l we find that
γρ uBpl ≤ εγρ M (Bpm →Bpl ) + c(ε)γρ M (Bpm−l →Lp ) +c

Dp,l γρ ; eLp  uBpm . 1/p e,diam(e)≤1 [Cp,m (e)] sup

(4.3.70)

4.3 Proof of Theorem 4.1.1

157

Similarly, for m = l,
γρ uBpl ≤ εγρ M Bpl + c(ε)γρ L∞ +c

Dp,l γρ ; eLp  uBpl . 1/p e,diam(e)≤1 [Cp,l (e)] sup

(4.3.71)

For noninteger l the following estimate, simpler than (4.3.64), holds: [l] [l] 
   |∇j γρ |Dp,l−j uLp +  |∇j u|Dp,l−j γρ Lp γρ uBpl ≤ c j=0

j=0

Combining (4.3.60) with Corollary 4.3.4 and (4.3.26), (4.3.27), we arrive at (4.3.66) and (4.3.67) in the same way as for integer l. We also note that (4.3.61) and (4.3.26) for m > l and (4.3.27) for m = l imply (4.3.68) and (4.3.69) for noninteger l. Reference to (4.3.12) and Lemma 4.3.3 completes the proof.  The required upper estimate of γM Bpl in (4.1.4) is obtained in Lemma 4.3.7. In order to show that the second term in the right-hand side of (4.3.63) can be replaced by γL1,unif for m > l, we need several auxiliary assertions. Let (T γ)(x, y) denote the Poisson integral of a function γ ∈ L1,unif defined by (3.2.38). [l]

Lemma 4.3.8. Let l be a noninteger and let γ ∈ W1,loc . Then


∞

0

1/p ∂ [l]+1 (T γ)(x, y) p p−1−p{l} y dy ≤ c (Dp,l γ)(x). ∂y [l]+1

Proof. Our argument is similar to that used in the proof of Lemma 3.2.12. We start with the inequality  ∂ [l]+1 (T γ)(x, y) |∇[l] γ(x − ξ) − ∇[l] γ(x)| ≤ c dξ (4.3.72) [l]+1 (|ξ| + y)n+1 ∂y derived in the proof of Lemma 3.2.12. Hence  ∞ [l]+1 γ(x, y) p p−1−p{l} ∂ dy y ∂y [l]+1 0 

∞

≤c

y 0

 =c 0

∞


p(1−{l})


 |∇ γ(x − ξ) − ∇ γ(x)|  |ξ| −n−1 p dy [l] [l] 1 + dξ y n+1 y y

|∇[l] γ(x − ξ)−∇[l] γ(x)|  |ξ| n+{l}  |ξ| −n−1 p dy 1+ . (4.3.73) dξ y y y |ξ|n+{l}

4 The Space M (Bpm → Bpl ) with p > 1

158

Introducing spherical coordinates, we write the last expression as  ∞
 ∞ t dt p dy c g(t, x) , f y t y 0 0 where

f (s) = sn+{l} (1 + s)−n−1

and g(t, x) = t−{l}

 |∇[l] γ(x + tθ) − ∇[l] γ(x)|dθ. ∂B1

Clearly,  ∞
 0

∞

0

t dt p dy g(t, x) = f y t y



∞
 ∞

f (s)g(sy, x) 0

0

ds p dy . s y

(4.3.74)

By Minkowski’s inequality the expression on the right-hand side does not exceed
 ∞
 ∞ dy 1/p ds p (f (s))p (g(sy, x))p y s 0 0  
 ∞
 ∞ dτ 1/p ds p
∞ dτ ds p ∞ = f (s) (g(τ, x))p = f (s) (g(τ, x))p . τ s s τ 0 0 0 0 We deduce from the definition of f that  ∞  1  ∞ ds ds n+{l} ds ≤ + < ∞. f (s) s s{l}−1 s s s 0 0 1

(4.3.75)

Therefore, (4.3.73)–(4.3.75) imply the estimate  0

 ∞ dτ ∂ [l]+1 (T γ)(x, y) p p−1−p{l} y dy ≤ c (g(τ, x))p . [l]+1 τ ∂y 0

∞

It remains to note that  ∞  ∞
 p dτ dτ = (g(τ, x))p τ −p{l} |∇[l] γ(τ θ + x) − ∇[l] γ(x)|dθ τ τ 0 0 ∂B1  ≤

∞



0

|∇[l] γ(τ θ + x) − ∇[l] γ(x)|p dθ ∂B1

 ≤c The proof is complete.

dτ τ 1+p{l}

|∇[l] γ(x + h) − ∇[l] γ(x)|p dh. |h|n+p{l} 

4.3 Proof of Theorem 4.1.1

159

The following two lemmas are similar to Lemmas 3.2.13 and 3.2.14. [l]

Lemma 4.3.9. Let γ ∈ W1,loc , y ∈ (0, 1]. Then ∂ [l]+1 (T γ)(x, y) sup rm−n/p Dp,l γ; Br (x)Lp . ≤ c y {l}−m−1 ∂y [l]+1 x∈Rn ,r∈(0,1) Proof. We introduce the notation K=

sup x∈Rn ,r∈(0,1)

rm−n/p Dp,l γ; Br (x)Lp .

Let r ∈ (0, 1]. By Lemma 4.3.8,   ∞ [l]+1 (T γ)(t, y) p p−1−p{l} ∂ dy dt ≤ c K p rn−mp . y ∂y [l]+1 Br (x) 0

(4.3.76)

(4.3.77)

Applying the mean value theorem for harmonic functions, we find for r/2 < y < 2r/3 that   r [l]+1 ∂ [l]+1 (T γ)(x, y) (T γ)(t, η) ∂ −n−1 ≤ c r dηdt. ∂y [l]+1 ∂η [l]+1 Br (x) r/4 By H¨older’s inequality the right-hand side is dominated by  r [l]+1
 1/p (T γ)(t, η) p p−1−p{l} ∂ c r{l}−1−n/p dηdt η [l]+1 ∂η Br (x) r/4 which by (4.3.77) does not exceed c r{l}−m−1 K. The result follows.



[l]

Lemma 4.3.10. Let γ ∈ W1,loc . Then for almost all x ∈ Rn the inequality  l/m  |γ(x)| ≤ c sup rm−n/p Dp,l γ; Br (x)Lp (Dp,l γ(x))(m−l)/m + γL1,unif x∈Rn , r∈(0,1)

holds. Proof. We put

⎧ ⎪ ∂ [l]+1 (T γ)(x, y) ⎪ ⎪ ⎨ ∂y [l]+1 v(y) = ⎪ ⎪ ⎪ ⎩0

for 0 < y ≤ 1, for y > 1.

Then, for any R > 0  1 [l]+1  ∞  R  ∞ (T γ)(x, y) [l] ∂ [l] [l] dy = v(y)y dy = v(y)y dy+ v(y)y [l] dy. y [l]+1 ∂y 0 0 0 R

160

4 The Space M (Bpm → Bpl ) with p > 1

Applying H¨ older’s inequality, we find that 

R

v(y)y dy ≤ c R [l]

l




0

R

(v(y))p y p−p{l}−1 dy

1/p .

0

By Lemma 4.3.9, ∂ [l]+1 (T γ)(x, y) ≤ c Ky {l}−m−1 , ∂y [l]+1 where K is defined by (4.3.76). Hence  ∞

 ∞ 1/p  v(y)y [l] dy ≤ c Rl (v(y))p y p−p{l}−1 dy + Rl−m K . 0

0

Putting here R = K 1/m




∞

v(y)p y p−p{l}−1 dy

−1/pm

,

0

we arrive at  ∞

v(y)y [l] dy ≤ c K l/m

0




∞

v(y)p y p−p{l}−1 dy

(m−l)/pm .

0

Combining this inequality with (3.2.39) for k = [l] we arrive at
 ∞ 
(m−l)/pm l/m |γ(x)| ≤ c K v(y)p y p−p{l}−1 dy + γL1,unif . 0



Reference to Lemma 4.3.8 completes the proof. Now, we are in a position to prove the main result of this section. Lemma 4.3.11. Let 0 < l < m, p ∈ (1, ∞). Then
γM (Bpm →Bpl ) ≤ c

 Dp,l γ; eLp . + γ L 1,unif 1/p e,diam(e)≤1 [Cp,m (e)] sup

(4.3.78)

Proof. By (2.3.18) with ϕ = |γρ |1/(m−l) , λ = m − l, and µ = m − ε, where ε is a positive number less than l such that both l − ε and m − ε are nonintegers, we find that   m−ε m−l p  |γρ | (x)dx |γρ | m−l p (x)dx m−ε ≤ c sup e . (4.3.79) sup e Cp,m−l (e) Cp,m−ε (e) e e Using Lemma 4.3.10 with l replaced by l − ε and m replaced by m − ε, we obtain

4.3 Proof of Theorem 4.1.1

 |γρ |

(m−ε)p m−l



dx ≤ c

e

 × e

n

sup x∈Rn ,r∈(0,1)

rm−ε− p Dp,l−ε γρ ; Br (x)Lp

161

 (l−ε)p m−l

 (m−ε)p m−l |(Dp,l−ε γρ )(x)|p dx + γρ L1,unif mesn e .

Hence   |γ | (m−ε)p ρ m−l (x)dx

m−l (m−ε)p

e

Cp,m−ε (e)

l−ε #  m−ε m−ε− n p D ≤c sup r p,l−ε γρ ; Br (x)Lp x∈Rn , r∈(0,1)

  m−l & Dp,l−ε γρ ; eLp m−ε × sup + γ  ρ L1,unif . [Cp,m−ε (e)]1/p e

(4.3.80)

By Corollary 4.3.2, sup e

Dp,l−ε γρ ; eLp ≤ c γρ M (Wpm−ε →Wpl−ε ) [Cp,m−ε (e)]1/p

= c γρ M (Bpm−ε →Bpl−ε ) ≤ c γρ M (Bpm →Bpl ) . Thus, the left-hand side of (4.3.80) has the majorant 
 m−l  l−ε n m−ε c sup rm−ε− p Dp,l−ε γρ ; Br (x)Lp m−ε γρ M + γ  ρ L m l 1,unif (B →B ) p

x∈Rn , r∈(0,1)

p

which together with (4.3.79) implies the inequality   |γ |p (x)dx 1/p ρ n sup e ≤ c(δ) sup rm−ε− p Dp,l−ε γρ ; Br (x)Lp Cp,m−l (e) x∈Rn , e r∈(0,1)

+ δγρ M (Bpm →Bpl ) + c γρ L1,unif ,

(4.3.81)

where δ is an arbitrary positive number. Next we show that n

sup x∈Rn ,r∈(0,1)

≤ c(σ)

sup x∈Rn ,r∈(0,1)

n

rm−ε− p Dp,l−ε γρ ; Br (x)Lp

rm− p Dp,l γρ ; Br (x)Lp + σ γρ M (Bpm →Bpl )

(4.3.82)

where σ is an arbitrary positive number. We note that by (4.1.3) Dp,l−ε γρ can be replaced by Dp,l−ε γρ . Let ω denote a positive number to be chosen later. Further, let k = l − 1 and λ = 1 for integer l, and let k = [l] and λ = {l} for noninteger l. We then have

4 The Space M (Bpm → Bpl ) with p > 1

162





|∇k γρ (y + 2h) − 2∇k γρ (y + h) + ∇k γρ (y)|p dh |h|n+p(λ−ε) Bωr   |∇k γρ (y + 2h) − 2∇k γρ (y + h) + ∇k γρ (y)|p pε dy dh ≤ (ωr) |h|n+pλ Br (x) Bωr

dy Br (x)

≤ (ωr)pε Dp,l γρ ; Br (x)pLp .

(4.3.83)

Also, 

 dy Br (x)

Rn \Bωr




 ≤c

dy

Br (x)

Rn \Bωr

  |∇k γρ (y + 2h)|p |∇k γρ (y + h)|p dh+ dy dh |h|n+p(λ−ε) |h|n+p(λ−ε) Br (x) Rn \Bωr  (4.3.84) +(ωr)p(ε−λ) ∇k γρ ; Br (x)pLp .

Further, we have   dy Br (x)

Rn \Bωr

|∇k γρ (y + 2h)|p dh |h|n+p(λ−ε)

 ≤

|∇k γρ (y + 2h) − 2∇k γρ (y + h) + ∇k γρ (y)|p dh |h|n+p(λ−ε)



dh

Rn \Bωr

|h|n+p(λ−ε)

≤ c ω p(ε−λ) rn−pm+pε

Br (x+2h)

sup x∈Rn ,r∈(0,1)

|∇k γρ (z)|p dz rp(m−λ)−n ∇k γρ ; Br (x)pLp .

By (4.3.14)–(4.3.16) the last supremum is dominated by c ∇k γρ pM (W m−λ →L

p)

p

which by Corollary 4.3.4 does not exceed c γρ pM (B m →B l ) . p

p

Clearly, the second term on the right-hand side of (4.3.84) is estimated in the same way. Similarly, the third term does not exceed c ω p(ε−λ) rn−pm+pε γρ pM (B m →B l ) . p

Hence



 dy Br (x)

Rn \Bωr

p

|∇k γρ (y + 2h) − 2∇k γρ (y + h) + ∇k γρ (y)|p dh |h|n+p(λ−ε)

≤ c ω p(ε−λ) rn−pm+pε γρ pM (B m →B l ) . p

p

(4.3.85)

4.3 Proof of Theorem 4.1.1

163

From (4.3.83) and (4.3.85) we obtain   rm−ε−n/p Dp,l−ε γρ Lp ≤ c ω ε rm−n/p Dp,l γρ ; Br (x)Lp+ω ε−λ γρ M (Bpm →Bpl ) . Setting σ = c ω ε−λ , we arrive at (4.3.82). By (4.3.14)–(4.3.16) and (4.3.82), n

sup x∈Rn ,r∈(0,1)

≤ c(σ) sup e

rm−ε− p Dp,l−ε γρ ; Br (x)Lp

Dp,l γρ ; eLp + σγρ M (Bpm →Bpl ) , [Cp,m (e)]1/p

which together with (4.3.81) and Lemma 4.3.7 gives 
Dp,l γρ ; eLp . γρ M (Bpm →Bpl ) ≤ c sup + γ  ρ L 1,unif [Cp,m (e)]1/p e

(4.3.86)

It remains to estimate the right-hand side of (4.3.86) by Lemma 4.3.3, and to use the equivalence relation  Cp,m (e) ∼ Cp,m (e ∩ B (j) ), j≥1

where {B (j) }j≥0 is a covering of Rn by balls of diameter one with multiplicity depending only on n (see Proposition 3.1.5). The result follows.  Combining the statements of Lemmas 4.3.6 and 4.3.11, we complete the proof of Theorem 4.1.1. The next assertion contains a modified version of Theorem 4.1.1. Corollary 4.3.7. Let 0 < l < m and let p ∈ (1, ∞). Then for noninteger l γM (Bpm →Bpl ) ∼

[l]    Dp,l−j γM (Bpm−j →Lp ) + ∇[l]−j γM (Bpm−j−{l} →Lp ) ,

(4.3.87)

j=0

and for integer l γM (Bpm →Bpl ) ∼

l−1  j=0

Dp,l−j γM (Bpm−j →Lp ) +

l 

∇l−j γM (Bpm−j →Lp ) .

(4.3.88)

j=1

Also, γM (Bpm →Bpl ) ∼ Dp,l γM (Bpm →Lp ) + γL1 ,unif . For m = l the norm γL1 ,unif should be replaced by γL∞ .

(4.3.89)

4 The Space M (Bpm → Bpl ) with p > 1

164

Proof. The upper estimate in (4.3.87) follows from Theorem 4.1.1 and the equivalence relation (4.3.12). By Corollaries 4.3.6 and 4.3.4, the lower bound in (4.3.87) results from Dp,l−j γM (Bpm−j →Lp ) ≤ c γM (Bpm−j →Bpl−j ) ≤ c γM (Bpm →Hpl ) and ∇[l]−j γM (Bpm−j−{l} →Lp ) ≤ c γM (Bpm−j−{l} →Bp[l]−j ) ≤ c γM (Bpm →Bpl ) .  Remark 4.3.2. It follows from Remark 3.1.3 that the supremum on the righthand side of (4.1.4) is equivalent to each of the suprema sup

Jm χQ (Dp,l γ)p ; QLp/(p−1)

{Q}

Dp,l γ; Qp−1 Lp

,

(4.3.90)

where {Q} is the collection of all cubes, χQ is the characteristic function of Q, and Jm (Jm (Dp,l γ)p )p/(p−1) (x) sup . (4.3.91) Jm (Dp,l γ)p (x) x∈Rn Adding to (4.3.90) and (4.3.91) the norms γL1,unif for m > l and γL∞ for m = l, we arrive at two non-capacitary necessary and sufficient conditions for γ ∈ M (Bpm → Bpl ). 4.3.7 The Case mp > n For mp > n Theorem 4.1.1 has a simpler formulation. Corollary 4.3.8. Let 0 < l ≤ m, mp > n, and p ∈ (1, ∞). Then   γM (Bpm →Bpl ) ∼ sup Dp,l γ; B1 (x)Lp + γ; B1 (x)Lp .

(4.3.92)

x∈Rn

For m = l the second term on the right-hand side can be replaced by γL∞ . Proof. The lower estimate of γM (Bpm →Bpl ) follows from the relation Cp,m (e) ∼ 1

(4.3.93)

wich holds for mp > n and e with diam(e) ≤ 1, and from Corollary 4.3.5. The upper estimate results from   γM (Bpm →Bpl ) ≤ γM Bpl ≤ c sup Dp,l γ; eLp + γL∞ e,diam(e)≤1

  ≤ c sup Dp,l γ; B1 (x) + γ; B1 (x)Lp . x∈Rn

The proof is complete.



4.3 Proof of Theorem 4.1.1

165

Remark 4.3.3. One can easily verify that the right-hand side of (4.3.92) is l . Hence M (Bpm → Bpl ) is isomorphic to equivalent to the norm of γ in Bp,unif l Bp,unif for 0 < l ≤ m, mp > n, p ∈ (1, ∞). 4.3.8 Lower and Upper Estimates for the Norm in M (Bpm → Bpl ) Here we present some lower and, separately, upper bounds for the norm in M (Bpm → Bpl ), mp ≤ n, which do not involve the capacity and which follow from the characterization of multipliers in M (Bpm → Bpl ). The next assertion stems directly from Proposition 3.1.4 and Theorem 4.1.1. Proposition 4.3.1. Let 0 < l < m. If mp < n, then   γM (Bpm →Bpl ) ≥ c sup rm−n/p Dp,l γ; Br (x)Lp + γL1,unif

(4.3.94)

x∈Rn r∈(0,1)

and, if mp = n, then γM (Bpm →Bpl )   ≥ c sup (log 2r−1 )1−1/p Dp,l γ; Br (x)Lp +γL1,unif .

(4.3.95)

x∈Rn r∈(0,1)

For m = l the second term on the right-hand side of (4.3.94) and (4.3.95) should be replaced by γL∞ . Finally we formulate a corollary of Theorem 4.1.1 and Propositions 3.1.2 and 3.1.3. Proposition 4.3.2. Let 0 < l < m. If mp < n, then γM (Bpm →Bpl )  ≤c

(mesn e)m/n−1/p Dp,l γ; eLp + γL1,unif

sup



(4.3.96)

{e:d(e)≤1}

and, if mp = n, then γM (Bpm →Bpl )  ≤c

sup


(log(2n /mesn e))1/p Dp,l γ; eLp + γL1,unif .

(4.3.97)

{e:d(e)≤1}

For m = l the second term on the right-hand side of (4.3.96) and (4.3.97) should be replaced by γL∞ .

166

4 The Space M (Bpm → Bpl ) with p > 1

4.4 Sufficient Conditions for Inclusion into M (Wpm → Wpl ) with Noninteger m and l It may be of use to compare the contents of this section with sufficient conditions for inclusion into the class M (Hpm → Hpl ) obtained in 3.4. Here similar conditions are found for M (Wpm → Wpl ), {m} > 0, {l} > 0. They are forµ l l (cf. (4.4.2)), Bq,p (cf. (4.4.3)) and Hn/m mulated in terms of the spaces Bq,∞ (cf. (4.4.4). s we denote the space of functions in Rn having the finite norm By Bq,θ s uBq,θ =




∆h ∇[s] uθLq |h|−n−θ{s} dh

1/θ

+ uWq[s] ,

(4.4.1)

where {s} > 0, q, θ ≥ 1. µ 4.4.1 Conditions Involving the Space Bq,∞

We prove an assertion analogous to Theorem 3.4.1 and formulated in terms s . of the space Bq,θ,unif It is clear that 
 1/θ s ∼ sup ∆h ∇[s] u; B1 (x)θLq |h|−n−θ{s} dh uBq,θ,unif x∈Rn B1  +u; B1 (x)Wq[s] . Theorem 4.4.1. Let q ≥ p > 1, {m} > 0, {l} > 0, µ = n/q − m + l, µ > l, and {µ} > 0. µ (i) If γ ∈ Bq,∞,unif ∩ L∞ then γ ∈ M Wpl and µ γM Wpl ≤ c (γBq,∞,unif + γL∞ ).

(4.4.2)

µ , then γ ∈ M (Wpm → Wpl ) and (ii) If γ ∈ Bq,∞,unif µ γM (Wpm →Wpl ) ≤ c γBq,∞,unif .

(4.4.3)

Proof. Let m ≥ l. It suffices to assume that the difference ε = µ − l is small, since the general case follows by interpolation between the pairs {Wpm−l , Lp } n/q−ε

and {Wp , Wpµ−ε } (see (4.3.26)). Thus we assume that 1 + [l] > µ > l. Let e be a compact set in Rn , d(e) ≤ 1, and let |e| = mesn e. We have   dh p |∇[l] γ(x + h) − ∇[l] γ(x)|p dx . (4.4.4) Dp,l γ; eLp = |h|n+p{l} e We express the integral over Rn as the sum of two integrals i1 + i2 , the first being taken over the exterior of the ball {h : |h| < |e|1/n }. Obviously,

4.4 Sufficient Conditions for Inclusion into M (Wpm → Wpl )

 i1 ≤ c

dh |h|>|e|1/n



≤ c |e|




|h|n+p{l}

e

dh

1−p/q |h|>|e|1/n



|∇[l] γ(x + h)|p dx +


|h|n+p{l}

|∇[l] γ(x)|p dx e





|∇[l] γ(x + h)| dx +

|∇[l] γ(x)|q dx

q

e

167

p/q .

e

Hence, using Corollary 3.4.1, we obtain  i1 ≤ c |e|1−p/q+{µ}p/n

|h|>|e|1/n

dh N (γ)p , |h|n+p{l}

(4.4.5)

where N (γ) is the right-hand side of either (4.4.2) or (4.4.3). Since [µ] = [l], we have {µ} − {l} = µ − l and consequently i1 ≤ c |e|1−mp/n N (γ)p . By H¨older’s inequality,   p/q dh
1−p/q q i2 ≤ |e| |∇ γ(x + h) − ∇ γ(x)| dx [l] [l] n+p{l} |h| l. It remains to refer to Proposition 4.3.2.



Using the same arguments as in the proof of Corollary 3.4.2, we obtain: Corollary 4.4.1. Let n = 1, q ≥ p and lq < 1. If γ ∈ L∞ and Varq (γ) < ∞, then γ ∈ M Wpl and γM Wpl ≤ c (γL∞ + Varq (γ)) . We give one more sufficient condition for a function to belong to the space M (Bpm → Bpl ) for noninteger m and l in the case mp = n. We introduce the semi-norm γ = sup sup |h|−{l} log(1/|h|)∆h ∇[l] γ; B1 (y)Lp . y∈Rn h∈B1/2

168

4 The Space M (Bpm → Bpl ) with p > 1

Theorem 4.4.2. Let {m} > 0, {l} > 0, p > 1. (i) If lp = n, γ ∈ L∞ and γ < ∞, then γ ∈ M Wpl and γM Wpl ≤ c (γ + γL∞ ) .

(4.4.7)

(ii) If mp = n, γ ∈ Lp,unif and γ < ∞, then γ ∈ M (Wpm → Wpl ) for l < m and (4.4.8) γM (Wpm →Wpl ) ≤ c (γ + γLp,unif ) . Proof. By Q(γ) we denote either of the right-hand sides of (4.4.7) and (4.4.8). We express the integral over Rn in (4.4.4) as the sum i1 + i2 of two integrals, the first being taken over the exterior of the ball K = {h : |h| < c0 |e|1/n (log 2n /|e|)(p−1)/p{l} } , where |e| = mesn e and c0 is a small positive constant depending on n and p. By Corollary 3.4.1 with q = p and µ = n/p = l,     dh
p p |∇ γ(x + h)| dx + |∇ γ(x)| dx i1 ≤ c [l] [l] n+p{l} Rn \K |h| e e  dh Q(γ)p ≤ c(log 2n /|e|)1−p Q(γ)p . (4.4.9) ≤ c |e|{l}p/n n+p{l} Rn \K |h| Moreover,

 i2 ≤ γ

p K

dh |h|n (log 1/|h|)p

.

The integral on the right-hand side does not exceed c (log 2n /|e|)1−p . This estimate, together with (4.4.4) and (4.4.9), yields (log 2n /|e|)(p−1)/p Dp,l γ; eLp ≤ c Q(γ). In addition, by (4.4.6) |e|−l/n γ; eLp ≤ c Q(γ). Reference to Proposition 4.3.1 completes the proof.



4.4.2 Conditions Involving the Fourier Transform s We start with the following known characterization of the space B2,∞ (see, for example, [Tr3]).

Lemma 4.4.1. The relation s uB2,∞ ∼ sup Rs F u; B2R \BR L2 + uL2

R>1

holds.

(4.4.10)

4.4 Sufficient Conditions for Inclusion into M (Wpm → Wpl )

169

s Proof. The lower bound for the norm in B2,∞ is obtained as follows:  ∆ρθ ∇[s] u2L2 dσθ sup |h|−2{s} ∆h ∇[s] u2L2 = c sup ρ−2{s} ρ>0

h

∂B1

= c sup ρ−2{s}



|ξ|2[s] |F u(ξ)|2 sin2

ρ>0

∂B1



≥ c sup ρ2(1−{s}) ρ>0  ≥ c sup R2s

ρ−1 >|ξ|>(2ρ)−1

B2R \BR

ρ>0

sup |h| h

∆h ∇[s] u2L2

−2{s}

≤ c sup |h|



+ c sup |h|

2(1−{l}) |ξ||h|1

h

|ξ|2[s]+1 |F u(ξ)|2 dξ

|F u(ξ)|2 dξ .

On the other hand, −2{s}

ρ(θ, ξ) dσθ dξ 2

|ξ|2[s] |F u(ξ)|2 dξ

|ξ|2[s]+2 |F u(ξ)|2 dξ .

(4.4.11)

The first term on the right-hand side does not exceed  ∞  c sup |h|−2s 4j[s] |F u(ξ)|2 dξ h

≤c

∞ 

2j >|ξ||h|>2j−1

j=1

4−j{s} sup R2s

 B2R \BR

R

j=1

|F u(ξ)|2 dξ .

The second term on the right-hand side of (4.4.11) is majorized by  ∞  4−j([s]+1) |F u(ξ)|2 dξ c sup |h|−2s h

≤c

∞ 

2−j−1 1

Theorems 4.4.1, 4.4.2 and Lemmas 4.4.1, 4.4.2 imply: Theorem 4.4.3. (i) If 1 < p ≤ 2, n/2 > m, m, l are noninteger, m > l and (F γ)(ξ) = O((1 + |ξ|)m−l−n ), then γ ∈ M (Wpm → Wpl ). (ii) If 1 < p ≤ 2, n/2 > l, l is a noninteger, γ ∈ L∞ and (F γ)(ξ) = O((1 + |ξ|)−n ), then γ ∈ M Wpl . (iii) If n is odd, 2l = n, γ ∈ L∞ and (F γ)(ξ) = O(|ξ|−n (log |ξ|)−1 ) for |ξ| ≥ 2, then γ ∈ M W2l . l 4.4.3 Conditions Involving the Space Bq,p µ The condition γ ∈ Bq,∞,unif in Theorem 4.4.1 requires the ‘number of derivatives’ µ to exceed l. In this subsection we obtain sufficient conditions for a l . function to belong to the class M (Wpm → Wpl ) in terms of the space Bq,p,unif µ We recall that diminishing the exponent θ leads to narrowing of Bq,θ , and diminishing of µ leads to expansion of this space. So new sufficient conditions are not comparable with the conditions of Theorem 4.4.1.

Theorem 4.4.4. Let {m} > 0, {l} > 0, p > 1. l ∩ (i) Let q ∈ [n/l, ∞] for pl < n and q ∈ (p, ∞] for lp = n. If γ ∈ Bq,p,unif l L∞ , then γ ∈ M Wp and γM Wpl 

1/p ≤ c sup ∆h ∇[l] γ; B1 (x)pLq |h|−n−p{l} dh + γL∞ . (4.4.12) x∈Rn

B1

(ii) Let m > l, q ∈ [n/m, ∞] for mp < n and q ∈ (p, ∞) for mp = n. If l γ ∈ Bq,p,unif , then γ ∈ M (Wpm → Wpl ) and γM (Wpm →Wpl ) ≤ c sup

x∈Rn



 B1

∆h ∇[l] γ; B1 (x)pLq |h|−n−p{l} dh

1/p

 +γ; B1 (x)Lp . (4.4.13)

Proof. Proposition 4.2.6 implies that γM (Wpm →Wpl ) ≤ c sup ηx γM (Bpm →Bpl ) , x∈Rn

4.4 Sufficient Conditions for Inclusion into M (Wpm → Wpl )

171

where η ∈ C0∞ (B1 ), η = 1 on B1/2 and ηx (y) = η(x − y). Therefore it suffices to obtain (4.4.12), (4.4.13) under the assumption that the diameter of supp γ does not exceed 1. Let e ⊂ Rn , d(e) ≤ 1. We have  dh Dp,l γ; epLp ≤ ∆h ∇[l] γ; epLp n+p{l} |h|  dh 1−p/q ≤ (mesn e) ∆h ∇[l] γ; epLq n+p{l} |h|  dh ≤ (mesn e)1−p/q sup ∆h ∇[l] γ; B1 (x)pLq n+p{l} . |h| x∈Rn Reference to Proposition 4.3.2 completes the proof.



Putting q = ∞ in (4.4.12), we obtain a simple condition for a function γ to belong to the class M Wpl (and hence to M (Wpm → Wpl )) formulated in terms of the modulus of continuity ω of the vector-function ∇[l] γ:  $ 0

ω(t) t{l}+1/p

%p dt < ∞ .

(4.4.14)

The last theorem contains the condition lp ≤ n. Nevertheless, (4.4.14) ensures l l ⊃ B∞,p . that γ ∈ M Wpl for lp > n, since in that case M Wpl = Wp,unif We show that even a rough condition (4.4.14) is the best possible in some sense. Example 4.4.1. Let ω be a continuous increasing function on [0, 1] satisfying the inequalities 

1

δ 0

ω(t) dt + t2



δ

0

Further, let

ω(t) dt ≤ c ω(δ) , t



1

1 > ω(δ) ≥ c δ.

(4.4.15)

[ω(t)t−{l}−1/p ]p dt = ∞.

0

We construct a function γ on Rn such that 1. the modulus of continuity of the vector-function ∇[l] γ does not exceed c ω, where c = const; l and hence γ ∈ / M (Wpm → Wpl ). 2. γ ∈ / Wp,unif We put n ∞  η(xi ) e−[l]k ω(e−k ) sin(ek x1 ) , (4.4.16) γ(x) = i=1

where η ∈

C0∞ (−2π, 2π),

k=1

η = 1 on (−π, π), 0 < η ≤ 1.

4 The Space M (Bpm → Bpl ) with p > 1

172

For small enough |h| we have |∇[l] γ(x + h) − ∇[l] γ(x)| ∞
 ≤ c |h| ω(e−k ) + |h|

 k≤log |h|−1

k=1

 ω(ek )



ω(e−k )ek +

k>log |h|−1

which, together with (4.4.15), gives |∇[l] γ(x + h) − ∇[l] γ(x)| ≤ c ω(|h|) . Further,



 γpW l p

≥c

(4.4.17)

Rn

We set f (x1 ) =

∂ [l] γ p dt ∆te1 [l] 1+p{l} . R1 ∂x1 t

(4.4.18)

e−[l]k ω(e−k )eie

(4.4.19)

∞ 

k

x1

.

k=1

By virtue of (4.4.18) we have f pW l ≥ c Imf [l] ; (−π, π)p

{l}

Wp

p

≥ c f [l] ; (−π, π)p

{l}

Wp

.

It is clear that ∆t f [l] (x1 ) =

∞ 

e−[l]k ω(e−k )(eie

k

t

− 1)eie

k

x1

.

k=1

According to a known property of lacunary trigonometric series (see [Zy], v. 1, Th. 8.20), ∆t f [l] ; (−π, π)2Lp ∼

∞ 

e−2[l]k (ω(e−k ) sin(ek t/2))2 .

k=1

Therefore

 γpW l p

π

≥c

[ω(e−k(t) ) sin(ek(t) t/2)]p

0

dt t1+p{l}

where k(t) = [log 2t−1 ]. Finally,  γpW l p

1

≥c 0

 ω(t) p dt = ∞ . t{l}+1/p

,

l 4.5 Conditions Involving the Space Hn/m

173

l 4.5 Conditions Involving the Space Hn/m

Here we obtain an upper bound for the norm in M (Wpm → Wpl ) with noninteger m and l, p ≥ 2, and mp < n, by the norm in the Bessel potential space l . Hn/m,unif Theorem 4.5.1. The estimates hold: , γM (Wpm →Wpl ) ≤ c γHn/m,unif l

(4.5.1)

where m > l, {m} > 0, {l} > 0, mp < n and p ≥ 2, and γM Wpl ≤ c (γHn/l,unif + γL∞ ) , l

(4.5.2)

where {l} > 0, lp < n and p ≥ 2. Proof. It suffices to derive (4.5.1) and (4.5.2) under the assumption that the diameter of supp γ does not exceed 1 (cf. the beginning of the proof of Theorem 4.4.4). We use the inequality , Dp,l γLn/m ≤ c γHn/m l where p ≥ 2, n > p(m − {l}), see Polking [Pol1]. Moreover, by Theorem 3.1.3, γLn/(m−l) ≤ c γHn/m , l where m > l. It remains to apply the estimate γM (Bpm →Bpl ) ≤ c sup (Dp,l γ; B1 (x)Ln/m + γ; B1 (x)Ln/(m−l) ) x∈Rn

which follows from (4.3.96).



Possibly, it is of interest to compare the last theorem with Theorem 3.3.1, according to which the right-hand sides of (4.5.1) and (4.5.2) majorize the norms of M Hpl and M (Hpm → Hpl ) for any p ∈ (1, ∞). We show that the condition p ≥ 2 in the theorem of this subsection cannot be omitted. Example 4.5.1. Let us consider the function γ defined by (4.4.16). Since, for q ∈ (1, ∞), n +
 ∂ 2 l/2 + + + γ+ γHql ∼ + 1− 2 ∂x Lq j j=1

174

4 The Space M (Bpm → Bpl ) with p > 1

(see the proof of Proposition 3.5.4), it follows that γHql ∼ η Im f ; R1 Hql , where η and f are functions defined in Example 4.4.1. It is known that η Im f ∈ Hql (R1 ) if and only if the function Im f belongs to the space Hql on the unit circumference C and q 1/q
 π
d2 l/2 Im f (θ) dθ . η Im f ; R1 Hql ∼ Im f ; CHql = 1− 2 dθ −π (We omit a standard but rather tedious proof of this fact.) Therefore γHql ∼

∞


e−2[l]k (1 + e2k )l/2 [ω(e−k )]2

1/2

k=1

and consequently γ ∈

Hql

if and only if  1
ω(t) 2 dt 1. Further, let ϕ be a function defined on R1 if Im γ = 0 or on C1 if γ is complex-valued. Suppose that ϕ(0) = 0 and for all t and τ , |τ | < 1, the inequality |ϕ(t + τ ) − ϕ(t)| ≤ A |τ |ρ

with ρ ∈ (0, 1]

4.6 Composition Operator on M (Wpm → Wpl )

175

is valid. Then ϕ(γ) ∈ M (Wpm−l+r → Wpr ), where r ∈ (0, lρ) if ρ < 1 and r = l if ρ = 1. The following estimate holds: ϕ(γ)M (Wpm−l+r →W r ) ≤ c A(γρM (W m →W l ) + γM (Wpm →Wpl ) ) . p

p

p

Proof. First we note that for all t and τ |ϕ(t + τ ) − ϕ(t)| ≤ A (|τ |ρ + |τ |) .

(4.6.1)

By (4.6.1), the inclusions ϕ(γ) ∈ L1,unif and ϕ(γ) ∈ L∞ follow from γ ∈ L1,unif and γ ∈ L∞ , respectively. Consider the case ρ = 1. We have ϕ(γ)uWpl = Dp,l [ϕ(γ)u]Lp + ϕ(γ)uLp ≤ uDp,l ϕ(γ)Lp + ϕ(γ)Dp,l uLp + ϕ(γ)uLp . Using (4.6.1), we see that the sum on the right-hand side does not exceed 2A(uDp,l γLp + γDp,l uLp + γuLp ). It is clear that γDp,l uLp ≤ Dp,l (γu)Lp + uDp,l γLp . Hence ϕ(γ)uWpl ≤ 2A(2uDp,l γLp + γuWpl ) . Applying (4.3.12), we get 
Dp,l γ; eLp ϕ(γ)uWpl ≤ 2A c sup + γ m →W l ) uW m M (W p p p 1/p e [Cp,m (e)] which together with Theorem 4.1.1, gives ϕ(γ)M (Wpm →Wpl ) ≤ c A γM (Wpm →Wpl ) . Now let 0 < ρ < 1. Let us write the integral  |ϕ(γ(x))u(x) − ϕ(γ(y))u(y)|p |x − y|−n−pr dydx as the sum of two integrals of which one is taken over the set M = {(x, y) : |γ(y)| ≤ |γ(x)|}.

176

4 The Space M (Bpm → Bpl ) with p > 1

It is sufficient to estimate the integral over M which obviously does not exceed
 c |u(x)|p |ϕ(γ(x)) − ϕ(γ(y))|p |x − y|−n−pr dydx M

 +

|ϕ(γ(y))|p |u(x) − u(y)|p |x − y|−n−pr dydx .

(4.6.2)

M

We introduce two sets M1 (x) = {y : |γ(y)| ≤ |γ(x)|, |γ(x) − γ(y)| ≤ 1} and M2 (x) = {y : |γ(y)| ≤ |γ(x)|, |γ(x) − γ(y)| > 1}. It follows from (4.6.1) that  |ϕ(γ(x)) − ϕ(γ(y))|p |x − y|−n−pr dy M1 (x)  p ≤ cA |γ(x) − γ(y)|pρ |x − y|−n−pr dy M1 (x) 
≤ c Ap 2pρ |γ(x)|pρ |x − y|−n−pr dy |x−y|≥δx   + |γ(x) − γ(y)|pρ |x − y|−n−pr dy ,

(4.6.3)

|x−y| −s. Then  (n+s) Br (z)∩∂g

|η|α dσ(ζ) ≥ c rn+s−1 (r + |y|)α ,

(5.1.5)

where σ is the (n + s − 1)-dimensional area. The proof is based on the following lemma. Lemma 5.1.2. Let α > −s for s > 1 and 0 ≥ α > −1 for s = 1. Then for any v ∈ C ∞ (Br 

(n+s)

) there exists a constant V such that  |v(ζ) − V ||η|α dζ ≤ c r |η|α |∇v(ζ)|dζ.

(n+s)

Br

(5.1.6)

(n+s)

Br

(s)

Proof. It suffices to derive (5.1.6) for r = 1. We adopt the notation B1 ×B1 = Q. By R(ζ) we denote the distance from the point ζ ∈ ∂Q to the origin i.e. R(ζ) = (1 + |ζ|2 )1/2 for |η| = 1, |ξ| < 1, and R(ζ) = (1 + |η|2 )1/2 for (n+s) |ξ| = 1, |η| < 1. Since B1 is the bi-Lipschitz image of Q under the mapping ζ → ζ/R(ζ), we can deduce (5.1.6) from the inequality   |v(ζ) − V ||η|α dζ ≤ c |∇v(ζ)||η|α dζ. (5.1.7) Q

Q

Let us show that (5.1.7) holds. Since (s + α)|η|α = div(|η|α η), we find by integrating by parts on the left-hand side of (5.1.7) that it does not exceed  
  −1 α+1 |∇v||η| dζ + dξ |v(ζ) − V |ds(η) . (5.1.8) (s + α) (n)

(s)

B1

Q (s)

∂B1

(s)

We put T = B1 × (B1 \B1/2 ). Let s > 1. The second term in (5.1.8) is not greater than   c |∇v|dζ + c |v − V |dζ. T

T

Hence, taking the mean value of v in T as V , we get (5.1.7) from (5.1.8). If s = 1 then the set T has two components: T+ = B1 × (1/2, 1) and T− = B1 × (−1, −1/2). The same arguments as for the case s > 1 lead to

5.1 Trace Inequality for Functions in B1l (Rn )



 B1

|v(ξ, ±1) − V± | dξ ≤ c

181



|∇v(ζ)|dζ ≤ c T±

|∇v(ζ)||η|α dζ, Q

where V± is the mean value of v in T± . It remains to be noted that    1 ∂v |V+ − V− | ≤ c dξ |∇v(ζ)||η|α dζ dη ≤ c −1 ∂η B1 Q for α ≤ 0. Thus, for s = 1, inequality (5.1.7) follows with V+ or V− in place of V . (n+s)

Proof of Lemma 5.1.1. For the sake of brevity, let B = Br (z). We let v in (5.1.6) be the mollification χρ of the characteristic function of g. Then the left-hand side is bounded from below by the sum   α |1 − V | |η| dζ + |V | |η|α dζ, e1

e0

where ei = {z ∈ B : χρ (z) = i}, i = 0, 1. Let be an arbitrarily small positive number. By (5.1.4),   1 α ( − )(|1 − V | + |V |) |η| dζ ≤ c r |η|α |∇χρ (ζ)|dζ 2 B B for sufficiently small ρ. Consequently    1 |y|α dζ ≤ c r lim sup |η|α |∇ζ χρ (ζ)|dζ = c r |η|α dσ(ζ). 2 B ρ→+0 B B∩∂g (The last equality can be derived from Corollary 1.1.1.) It remains to note that  |y|α dζ ≥ c rn+s (r + |y|)α . B


Lemma 5.1.3. Let ν be a measure in Rn+s and let α > −s. The best constant K1 in   |U |dν ≤ K1 |y|α |∇z U |dz, U ∈ C0∞ (Rn+s ), (5.1.9) Rn+s

Rn+s

is equivalent to Q1 = sup (ρ + |y|)−α ρ1−n+s ν(Bρ(n+s) (z)).

(5.1.10)

z,ρ>0

Proof. 1. First let m > 1 or 0 ≥ α > −1, m = 1. According to Proposition 1.1.1,

182

5 The Space M (B1m → B1l )

K1 = sup 

ν(g)

g

,

|y|α dσ ∂g

where g is an arbitrary open subset of Rn+s with compact closure and smooth boundary. We show that for any g there exists a covering of g by a sequence (n+s) (zi ), i = 1, 2, · · · , such that of balls Bρi   α ρn+s−1 (ρ + |y |) ≤ c |y|α dσ. (5.1.11) i i i ∂g

i (n+s)

Every point z ∈ g is the centre of a ball Br (z) for which (5.1.4) holds. In fact, the ratio on the left-hand side of (5.1.4) is a continuous function in r, equal to one for small values of r and tending to zero as r → ∞. By Lemma (n+s) (zi ) such that 1.1.3 there exists a sequence of disjoint balls Bri g⊂

∞ 

(n+s)

B3ri

(zi ).

i=1

Lemma 5.1.1 implies that  (n+s)

Bri

(n+s)

Consequently {B3ri ν(g) ≤

(zi )∩∂g

|y|α dσ ≥ c rin+s−1 (ri + |yi |)α .

(zi )}i≥1 is the required covering. Obviously,



(n+s)

ν(B3ri

(zi )) ≤ Q1

i



rin+s−1 (ri + |yi |α )

i

 ≤ c Q1

|y|α dσ. ∂g

Thus K1 ≤ cQ1 . 2. Let s = 1 and α > 0. We construct a covering of the set {ζ : η = 0} by balls B (j) such that the radius ρj of B (j) is equal to the distance from B (j) to the hyperplane {ζ : η = 0}. By {ϕj } we denote a partition of unity (see [St2], Ch. VI, §1). subordinate to the covering {B (j) } with |∇ϕj | ≤ c ρ−1 j Using the present assertion for α = 0, we get  |ϕj u|dν ≤ c sup ρ−n νj (Bρ(n+1) (z))∇(ϕj u); Rn+1 L1 , Rn+1

ρ,z

where νj is the restriction of the measure ν to B (j) . It is clear that sup ρ−n νj (Bρ(n+1) (z)) ≤ c ρ,z

sup ρ≤ρj ,z∈B(j)

ρ−n ν(Bρ(n+1) (z)).

5.1 Trace Inequality for Functions in B1l (Rn )

183

Therefore,  Rn+1

≤c

|ϕj u|dν −α −n

(ρ + ρj )

sup

ρ

ρ≤ρj ,z∈B(j)

Summing over j, we find that 
 |u|dν ≤ c K1 Rn+1

Rn+1



Since

Rn+1

 ν(Bρ(n+1) (z))

Rn+1

|∇(ϕj u)||η|α dζ.

 |∇u||η|α dζ +

|u||η|α−1 dζ ≤ α−1

Rn+1

 |u||η|α dζ .

 Rn+1

|∇u||η|α dζ

for α > 0, we also have K1 ≤ c Q1 for s = 1, α > 0. 3. To obtain the converse estimate, we put U (ξ) = ϕ(ρ−1 (ζ − z)) into (n+s) (n+s) (5.1.9), where ϕ ∈ C0∞ (B2 ), ϕ = 1 on B1 . We notice that   |η|α |∇ζ U |dζ ≤ c ρ−1 |η|α dζ ≤ c ρn+s−1 (ρ + |y|)α . (n+s)

B2ρ

(n+s)

B2ρ

(z)

(z)



The proof is complete.

Corollary 5.1.1. Let ν be a measure in Rn and let α > −s. The best constant in (5.1.9) is equivalent to sup x∈Rn ,ρ>0

ρ1−n−s−α ν(Bρ(n+s) (z)).

To prove this assertion, it suffices to note that the value Q1 defined in (5.1.10) is equivalent to the last supremum if supp ν ⊂ Rn . 5.1.2 Main Result Now we are in a position to prove Theorem 5.1.1. The estimates K1 ≥ c Q and K2 ≥ c Q can be be obtained quite simply. It suffices to put u(ξ) = ϕ(ρ−1 (x − ξ)) into (5.1.3), where ϕ ∈ C0∞ (B2 ) and ϕ = 1 on B1 , and to note that  |u|dµ ≥ µ(Bρ (x)), D1,l u; Rn L1 = c ρn−l . Rn

Now we obtain the estimates K ≤ c Q. Let l ∈ (0, 1). According to Corollary 5.1.1,   |u|dµ ≤ c Q |y|−l |∇U |dz, (5.1.12) Rn

Rn+1

184

5 The Space M (B1m → B1l )

where U ∈ C0∞ (Rn+1 ) is an arbitrary extension of u ∈ C0∞ (Rn ) to Rn+1 . If l = 1, then by Theorem 1.1.1   |u|dµ ≤ c Q |∇2 U |dz. (5.1.13) Rn

Rn+1

It is known (see [Usp]) that  D1,l u; R L1 ∼ inf n

U

Rn+1

|y|−l |∇U |dz,

l ∈ (0, 1),



D1,1 u; Rn L1 ∼ inf U

Rn+1

|∇2 U |dz.

Hence, minimizing the right-hand sides of (5.1.12) and (5.1.13) over all extensions U , we arrive at  |u|dµ ≤ c QD1,l u; Rn L1 , l ∈ (0, 1]. Rn

Suppose that the estimate K ≤ c Q is proved under the condition l ∈ (m − 2, m − 1], where m is an integer, m ≥ 2. Duplicating the argument used in part (ii) of the proof of Theorem 1.1.1, we obtain the required estimate for l ∈ (m − 1, m].  Remark 5.1.1. It follows from Theorem 5.1.1 that (5.1.3) with l > n is valid only in the trivial case µ = 0, and for l = n if and only if the measure µ is finite. Theorem 5.1.2. The best constant K0 in  |u|dµ ≤ K0 u; Rn B1l Rn

is equivalent to Q0 =

sup

ρl−n µ(Bρ (x)).

x∈Rn ,ρ∈(0,1)

Proof. The estimate K0 ≥ c Q0 can be obtained in the same way as the estimate K ≥ c Q in Theorem 5.1.1. To prove the converse inequality, we use the sequence {ηj }j≥0 defined in Theorem 3.1.2. We apply Theorem 5.1.1 to the integral  |ηj u|dµj , Rn

where µj is the restriction of µ to the support of ηj . Then    |u|dµ ≤ c |ηj u|dµj ≤ c Q0 D1,l (ηj u)L1 . Rn

j

Rn

j

5.2 Properties of Functions in the Space B1k (Rn )

Since u; Rn B1l ∼



185

ηj u; Rn B1l ,

j

the last sum does not exceed c u; Rn B1l .



Remark 5.1.2. It is clear that Q0 = sup µ(B1 (x)) x∈Rn

for l ≥ n.

5.2 Properties of Functions in the Space B1k(Rn) 5.2.1 Trace and Imbedding Properties We start with the statement of a well-known trace and extension result. Lemma 5.2.1. [Usp] Suppose that m ≥ 1. (i) Let U be an arbitrary function in the space W1l+1 (Rn+1 + ). Then the limit u(x) = lim U (x, ρ) ρ→0

exists for almost all x ∈ Rn , the function u belongs to the space B1l (Rn ), and u; Rn B1l ≤ c U ; Rn+1 + W l+1 . 1

(ii) Let T u denote the action of the Poisson operator on the function u ∈ L1,unif (Rn ) defined by (3.2.38). Then n T u; Rn+1 + W l+1 ≤ c u; R B1l . 1

The next lemma contains an interpolation inequality for functions in B1l (Rn ). Lemma 5.2.2. Let u ∈ B1l (Rn ), where l is an integer, l ≥ 1. Then for j = 0, . . . , l − 1 (l−j)/l j/l u; Rn L1 . (5.2.1) u; Rn B l−j ≤ c u; Rn B l 1

1

Proof. We introduce the function  (Ds(q) u)(x)

= Rn

(q)

|∆h u(x)| dh |h|n+s

(5.2.2)

5 The Space M (B1m → B1l )

186

(q)

with any integer q > s, where ∆h u(x) is the difference of order q defined by (q)

∆h u(x) =

q    q i=0

i

(−1)i u(x + (q − i)h).

Given s > 0, the equivalence relation u; Rn B1s ∼ Ds(q) u; Rn L1 + u; Rn L1

(5.2.3)

holds for all values of q greater than s (see [Tr4], Sect. 3.5.3). Let q > l. Adding together the two inequalities 

 Rn

and

(q)

|∆h u(x)| dh dx ≤ |h|n+l−j

B





Rn



Rn

(q)

B

|∆h u(x)| dh dx |h|n+l

(q)

Rn \B

we find that



|∆h u(x)| dh dx ≤ c u; Rn L1 (Rn ) , |h|n+l−j 



(q)

|∆h u(x)| dh dx |h|n+l−j n n R R   (q) ≤ c Dl u; Rn L1 + u; Rn L1 .

(q)

Dl−j u; Rn L1 =

(5.2.4)

By (5.2.3),   (q) D1,l−j u; Rn L1 ≤ c Dl−j u; Rn L1 + u; Rn L1 . This, together with (5.2.4) and (5.2.3), leads to   D1,l−j u; Rn L1 ≤ c D1,l u; Rn L1 + u; Rn L1 .

(5.2.5) 

The interpolation inequality (5.2.1) follows from (5.2.5) by dilation.

Let B denote the unit ball centered at the origin and let l be a positive integer. We introduce the space B1l (B) of functions on B with finite norm u; BB1l =

l−1 

∇j u; BL1 +

j=0

l−1    j=0

B

B

|(∆(2) y ∇j u)(x)|

dx dy . |x − y|n+1

A local variant of inequality (5.2.1) is contained in the next statement. Corollary 5.2.1. Let u ∈ B1l (B). Then, for any j = 0, . . . , l − 1 (l−j)/l

u; BB l−j ≤ c u; BB l 1

1

j/l

u; BL1 .

(5.2.6)

5.2 Properties of Functions in the Space B1k (Rn )

187

Proof. It is well known (see [Tr4], Sect. 4.5) that u can be extended onto Rn so that (5.2.7) u; Rn B1l ≤ c u; BB1l and u; Rn L1 ≤ c u; BL1 . These inequalities, combined with Lemma 5.2.2, give (5.2.6).

(5.2.8) 

We need the following Hardy-type inequality. Lemma 5.2.3. Let u ∈ B1l (Rn ), where l is an integer, 1 ≤ l < n. Then  |x|−l |u(x)| dx ≤ c u; Rn B1l . (5.2.9) Rn

Proof. Let U ∈ W1l+1 (Rn+1 + ) be an arbitrary extension of u. We have    2l − 1 ∞ dr |x|−l |u(x)| dx = |u(x)| dx. (5.2.10) l rl+1 B2r \Br 0 Rn To estimate the right-hand side of (5.2.10), we use the standard trace inequality    −1  r |U (z)| + |∇U (z)| dz, |u(x)| dx ≤ c B2r \Br

G2r \Gr

(n+1)

where Gr = Br ∩Rn+1 + . Together with (5.2.10), this inequality implies that  
|U (z)|  dz −l + |∇U (z)| |x| |u(x)| dx ≤ c . (5.2.11) |z| |z|l Rn Rn+1 + Iterating the Hardy-type inequality   dz dz |∇j U (z)| l+1−j ≤ c |∇j+1 U (z)| l−j n+1 n+1 |z| |z| R+ R+ with j = 0, . . . , l − 1, we find that the right-hand side in (5.2.11) is dominated by  c Rn+1 +

|∇l+1 U (z)|dz.

Taking into account that u is the trace of U on Rn and using part (i) of Lemma 5.2.1, we complete the proof.  The next lemma contains two more inequalities for intermediate derivatives of functions given on the ball Br . The integral over (Br )2 stands for the double integral over Br .

5 The Space M (B1m → B1l )

188

Lemma 5.2.4. Let l be a positive integer and let j = 0, . . . , l − 1. Then for any r ∈ (0, 1] rj−l ∇j u; Br L1  
  dx dy −l (5.2.12) ≤c | ∆(2) ∇l−1 u (x, y)| + r u; B  r L 1 |x − y|n+1 (Br )2 

and

|(∆(2) ∇j u)(x, y)|

rj+1−l ≤c




(Br )2

|(∆(2) ∇l−1 u)(x, y)|

(Br )2

dx dy |x − y|n+1

 dx dy + r−l u; Br L1 , n+1 |x − y|

where (∆(2) v)(x, y) = v(x) − 2v

(5.2.13)


x + y 

+ v(y). 2 Proof. By dilation, the proof reduces to the case r = 1. It is well known that for j = 1, . . . , l − 2 ∇j u; BL1 ≤ c (∇l−1 u; BL1 + u; BL1 ) .

(5.2.14)

Hence, it suffices to prove (5.2.12) for j = l − 1. We introduce the function  ϕ ∈ C0∞ (B) subject to ϕ(y)dy = 1. Rn



We have

∇l−1 u(x) = ϕ(y)∇l−1 u(x)dy  B  x + y   − ∇l−1 u(y) dy. ϕ(y)∆(2) ∇l−1 u(x, y)dy + ϕ(y) 2∇l−1 u = 2 B B Integrating by parts in the last integral, we obtain  ∇l−1 u(x) = ϕ(y)∆(2) ∇l−1 u(x, y)dy 

B

 +(−1)l−1

  2−l  x + y  − u(y) (∇l−1 ϕ)(y)dy. 2 u 2 B

(5.2.15)

Therefore,      |∇l−1 u(x)|dx ≤ ϕ(y) ∆(2) ∇l−1 u (x, y)dy dx + c u; BL1 . B

B

B

Since the right-hand side does not exceed 
    ∆(2) ∇l−1 u (x, y) dydx + u; BL , c 1 n+1 |x − y| B B we arrive at (5.2.12) with j = l − 1. The proof of (5.2.12) is complete. Finally, (5.2.13) results from the definition of the space B1l (B) and inequalities (5.2.6) and (5.2.12). 

5.2 Properties of Functions in the Space B1k (Rn )

189

5.2.2 Auxiliary Estimates for the Poisson Operator We deal with the Poisson operator T defined by (3.2.38). l−1 (Rn ). Then Lemma 5.2.5. Let γ ∈ W1,loc



∞

0

∂ l+1 (T γ) dy ≤ c (D1,l γ)(x), ∂y l+1

(5.2.16)

where (Dl γ)(x) is defined by (5.1.1). Proof. For every n-dimensional multi-index α with |α| = 2, 
ξ − x Dxα (T γ)(x, y) = y −n−2 (Dα ζ) γ(ξ)dξ y Rn  
h
h (2) −n−2 α −n−2 γ(x+h)dh = y ∆h γ(x)dh, (5.2.17) (D ζ) ζ0,α =y y y Rn Rn where

1 α (D ζ)(ξ). 2 The last equality in (5.2.17) holds because Dα ζ is even and satisfies  Dα ζ(t)dt = 0. ζ0,α =

Rn

Since 1 (T γ)(x, y) = y −n 2

 ζ Rn


h y

(2)

∆h γ(x)dh + γ(x),

(5.2.18)

it follows for every n-dimensional multi-index β with |β| = 1 that  ∂ β ∂
−n−1  β 
h  (2) 1 Dx (T γ)(x, y) = y ∆h γ(x)dh Dy ζ ∂y 2 Rn ∂y y =y

−n−2

 Rn

where

ζ0,β


h y

(2)

∆h γ(x)dh,

(5.2.19)

 −1  (n + 1 + ξ, ∇)Dβ ζ (ξ). 2 Suppose that l ≥ 2. Let τ = α + δ, where |τ | = l + 1, |α| = 2, |δ| = l − 1. By (5.2.17),   h  (2) δ Dxτ (T γ)(x, y) = y −n−2 ∆ (D γ)(x)dh. ζ0,α (5.2.20) y h Rn ζ0,β =

5 The Space M (B1m → B1l )

190

Next, let τ = β + δ, where |τ | = l, |β| = 1, |δ| = l − 1. By (5.2.19),   h  (2) δ ∂ τ Dx (T γ)(x, y) = y −n−2 ∆ (D γ)(x)dh. ζ0,β ∂y y h Rn

(5.2.21)

Suppose that l + 1 is even, then the harmonicity of T γ implies that ∂ l+1 (T γ)(x, y) = (−∆x )(l+1)/2 (T γ)(x, y). ∂y l+1 Hence, by (5.2.20),  ∂ l+1 h (2) |(∆h ∇l−1 γ)(x)|dh, ζ1 l+1 (T γ)(x, y) ≤ c y −n−2 ∂y y Rn where

0 < ζ1 (ξ) ≤ c (1 + |ξ|)−n−3 .

(5.2.22)

(5.2.23)

If l + 1 is odd, then we have by harmonicity of T γ ∂ l+1 ∂ (−∆x )l/2 (T γ)(x, y). (T γ)(x, y) = ∂y l+1 ∂y This, together with (5.2.21), gives  ∂ l+1 h (2) −n−2 ≤ c y |(∆h ∇l−1 γ)(x)|dh, (T γ)(x, y) ζ2 l+1 ∂y y n R where Hence,

0 < ζ2 (ξ) ≤ c (1 + |ξ|)−n−2 .  0

(5.2.24)

(5.2.25)

 ∞  (2) |(∆h ∇l−1 γ)(x)| ∂ l+1 (T γ) dy ≤ c dy dh ∂y l+1 (y + |h|)n+2 Rn 0  (2) |(∆h ∇l−1 γ)(x)| c = dh n + 1 Rn |h|n+1

∞



which completes the proof. l−1 Lemma 5.2.6. Suppose that γ ∈ W1,loc (Rn ) and let

N = sup rm−n D1,l γ; Br (x)L1 . x∈Rn r∈(0,1)

Then, for any y ∈ (0, 1] ∂ l+1 (T γ)(x, y) ≤ c N y −m−1 . ∂y l+1

(5.2.26)

5.2 Properties of Functions in the Space B1k (Rn )

Proof. By Lemma 5.2.5,   dz Br (x)

0

∂ l+1 (T γ)(z, y) dy ≤ c N rn−m l+1 ∂y

191

∞

(5.2.27)

for r ∈ (0, 1). Let r/2 < y ≤ r. Applying the mean value theorem for harmonic functions, we find that  r l+1  ∂ l+1 (T γ)(x, y) c ∂ (T γ)(z, η) ≤ dz dη. ∂y l+1 rn+1 Br (x) ∂η l+1 r/2 By (5.2.27), the right-hand side is dominated by c N r−1−m . The proof is complete.  The next assertion is based mainly on two previous lemmas. l−1 Corollary 5.2.2. Let 0 < l < m ≤ n and let γ ∈ W1,loc (Rn ). Then, for all n x∈R    (m−l)/m |γ(x)| ≤ c N l/m (D1,l γ)(x) + γL1,unif .

Proof. Introducing the notation , |∂ l+1 (T γ)(x, y)/∂y l+1 | ϕ(y) = 0

for 0 < y ≤ 1, for y > 1,

for any R > 0, we have  1 l+1  ∞  ∞  R ∂ (T γ)(x, y) l l l y dy = ϕ(y) y dy ≤ R ϕ(y)dy + ϕ(y)y l dy. l+1 ∂y 0 0 0 R By Lemma 5.2.5, the first term on the right-hand side is majorized by cRl (D1,l γ)(x) and, by Lemma 5.2.6,  ∞  ∞ ϕ(y)y l dy ≤ c N y l−m−1 dy = c N Rl−m . R

R

Choosing R as

 −1/m R = N 1/m (D1,l γ)(x) ,

we arrive at the inequality  ∞  (m−l)/m ϕ(y)y l dy ≤ c N l/m (D1,l γ)(x) 0

which together with (3.2.39) completes the proof.



5 The Space M (B1m → B1l )

192

l−1 Corollary 5.2.3. Suppose that γ ∈ W1,loc (Rn ). For any integer l ≥ 1 and any z ∈ Rn
 rm−n−l γ; Br (z)L1 ≤ c sup ρm−n D1,l γ; Br (z)L1 +γL1,unif . (5.2.28) ρ∈(0,1)

Proof. By Corollary 5.2.2, rm−n−l γ; Br (z)L1 
≤ c N l/m rm−n−l

Br (z)

((D1,l γ)(x))(m−l)/m dx + rm−l γ; Rn L1,unif

which does not exceed 

l/m m−n c N r

(m−l)/m

Br (z)

(D1,l γ)(x)dx

+ γ; Rn L1,unif

by H¨ older’s inequality. Using (5.2.26), we complete the proof.







In the next two lemmas, we return to the Poisson operator T . Lemma 5.2.7. Let r < y < 1. Then, for any integer k ≥ 1 |∇k (T γ)(x, y)| ≤ c rl−m−k sup ρm−n−l γ; Bρ (z)L1 .

(5.2.29)

z∈Rn ρ∈(0,1)

Proof. By (3.2.38),  |∇k (T γ)(x, y)| ≤ c 

We have

Br (x)

Rn

|γ(ξ)| dξ. (|x − ξ| + y)n+k

|γ(ξ)| dξ ≤ c y −n−k (|x − ξ| + y)n+k

(5.2.30)

 Br (x)

|γ(ξ)|dξ

≤ c rl−m−k sup ρm−n−l γ; Bρ (z)L1 .

(5.2.31)

z∈Rn ρ∈(0,1)

Also,



|γ(ξ)| dξ ≤ n+k Rn \Br (x) (|x − ξ| + y)   ≤ c r−n dz Rn \B2r (x)

 Rn \Br (x)

Br (z)

|γ(ξ)| dξ |x − ξ|n+k

|γ(ξ)| dξ. |x − ξ|n+k

Since |ξ−x| > |z−x|/2, it follows that the right-hand side of the last inequality does not exceed

5.3 Descriptions of M (B1m → B1l ) with Integer l

 c rl−m Rn \B

Therefore,  Rn \B

r (x)

2r (x)

193

dt sup ρm−n−l γ; Bρ (x)L1 . |t − x|n+k x∈Rn ρ∈(0,1)

|γ(ξ)| dξ ≤ c rl−m−k sup ρm−n−l γ; Bρ (z)L1 . (5.2.32) (|x − ξ|+y)n+k z∈Rn ρ∈(0,1)

Now, (5.2.29) results by combining inequalities (5.2.31), (5.2.32), and (5.2.30).  Lemma 5.2.8. Let y > 1 and let k ≥ 0. Then |∇k (T γ)(x, y)| ≤ c y −k γL1,unif .

(5.2.33)

Proof. First, observe that   |γ(ξ)| dξ −n−k ≤ cy |γ(ξ)|dξ ≤ cy −k γL1,unif . (5.2.34) n+k By (x) (|x − ξ| + y) By (x) Clearly,  Rn \By (x)

|γ(ξ)| dξ ≤ c (|x − ξ| + y)n+k



 dz

Rn \B2 (x)

B1 (z)

|γ(ξ)| dξ. (5.2.35) |x − ξ|n+k

Since |ξ − x| > |z − x|/2, the right-hand side of (5.2.35) is dominated by  dz c γL1,unif . n+k Rn \B2 (x) |z − x| Therefore,  Rn \By (x)

|γ(ξ)| dξ ≤ c y −k γL1,unif . (|x − ξ| + y)n+k

(5.2.36)

Combining the inequalities (5.2.34) and (5.2.36), and then using (5.2.30), we complete the proof. 

5.3 Descriptions of M (B1m → B1l ) with Integer l In this section we give two different characterizations of the space M (B1m → B1l ) with integer l.

194

5 The Space M (B1m → B1l )

5.3.1 A Norm in M (B1m → B1l ) (n)

As before, in the following theorem the integral over (Br (z))2 stands for the (n) double integral over Br (z). Theorem 5.3.1. Let l be an integer and let m ≥ l ≥ 1. The equivalence relation holds: γM (B1m →B1l ) ∼
    dx dy −l sup rm−n | ∆(2) ∇l−1 γ (x, y)| +r γ; B (z) r L1 . (5.3.1) |x−y|n+1 z∈Rn (Br(z))2 r∈(0,1)

Proof. We use the norm |||v; Br |||B1l =

l−1 

rj−l ∇j v; Br L1

j=0

+

l−1 

 rj+1−l (Br

j=0

)2

|∆(2) y ∇j v(x)|

dx dy |x − y|n+1

(5.3.2)

defined for a positive integer l and r ∈ (0, 1). Lemma 5.2.4 implies that  dx dy |||v; Br |||B1l ∼ |(∆(2) + r−l v; Br L1 . (5.3.3) y ∇l−1 v)(x)| |x − y|n+1 (Br )2 By dilation in (5.2.6) we obtain 1−j/l

|||v; Br |||B l−j ≤ c |||v; Br |||B l 1

1

j/l

v; Br L1

(5.3.4)

for any j = 0, . . . , l − 1. By (5.3.3), the required relation (5.3.1) can be written as γM (B1m →B1l ) ∼ sup rm−n |||γ; Br (z)|||B1l .

(5.3.5)

z∈Rn r∈(0,1)

From Lemma 5.2.3 and (5.3.3), we see that |||v; Br |||B1l ≤ c v; Rn B1l

(5.3.6)

 y−x 

for l < n. Let u(y) = η r , where r ∈ (0, 1) for m < n, and r = 1 for m ≥ n, and η ∈ C0∞ (B2 ), η = 1 on B1 . Setting this u into the inequality γ uB1l ≤ γM (B1m →B1l ) uB1m

(5.3.7)

and using (5.3.6) with v = γu, we have |||γ; Br (x)|||B1l ≤ c rn−m γ; Rn M (B1m →B1l )

(5.3.8)

5.3 Descriptions of M (B1m → B1l ) with Integer l

195

for any x ∈ Rn . The required lower estimate for the norm γ; Rn M (B1m →B1l ) follows from (5.3.3). Now we obtain the upper estimate for the norm γ; Rn M (B1m →B1l ) . As before, let T γ stand for the Poisson integral of γ. For any U ∈ W1m+1 (Rn+1 + ), we have by Lemma 5.2.1 that γ u; Rn B1l ≤ c (T γ)U ; Rn+1 + W l+1 , 1

(5.3.9)

where u(x) = U (x, 0). Let X = (x, y) ∈ Rn+1 and let + Gr (X) = Br(n+1) (X) ∩ Rn+1 + . By Theorem 2.4.2, for any integer l ∈ [0, m), Γ ; Rn+1 + M (W m+1 →W l+1 ) ∼ 1

sup rm−n ∇l+1 Γ ; Gr (X)L1 +

n+1 X∈R+

1

sup Γ ; G1 (X)L1 .

(5.3.10)

X∈Rn+1 +

r∈(0,1)

The first supremum in (5.3.10) can be replaced by sup ∇l+1 Γ ; G1 (X)L1

X∈Rn+1 +

in the case m ≥ n. Furthermore, Γ ; Rn+1 + M W l+1 ∼ 1

sup rl−n ∇l+1 Γ ; Gr (X)L1 + Γ ; Rn+1 + L∞ . (5.3.11)

n+1 X∈R+ r∈(0,1)

This relation and (5.3.9) give γ u; Rn B1l ≤ c Km,l U ; Rn+1 + W m+1 , 1

(5.3.12)

where Km,l =

sup rm−n ∇l+1 (T γ); Gr (X)L1 + sup T γ; G1 (X)L1 . (5.3.13) X∈Rn+1 +

n+1 X∈R+ r∈(0,1)

We introduce one more notation  m−n km,l := sup r z∈Rn r∈(0,1)

|∆(2) ∇l−1 γ(x, y)|

(Br (z))2

dx dy |x − y|n+1

(5.3.14)

and intend to show that   (n) Km,l ≤ c km,l + sup γ; B1 (z)L1 . z∈Rn

(5.3.15)

196

5 The Space M (B1m → B1l )

Then the upper estimate for γ; Rn M (B1m →B1l ) follows from (5.3.12) by Lemma 5.2.1 and the arbitrariness of U . Let us justify (5.3.15). When estimating ∇l+1 (T γ); Gr (X0 )L1 , where X0 ∈ Rn+1 + , it suffices to take X0 = (0, y0 ). Suppose first that y0 > 2. Then, by Lemma 5.2.8, rm−n ∇l+1 (T γ); Gr (X0 )L1 ≤ c γ; Rn L1,unif . For 2 > y0 ≥ 2r, from Lemma 5.2.7 we have rm−n ∇l+1 (T γ); Gr (X0 )L1 ≤ c sup ρm−n−l γ; Bρ(n) (x)L1 . x∈Rn ρ∈(0,1)

Given any r ∈ (0, 1), it remains to estimate the norm ∇l+1 (T γ); Gr (X0 )L1 for y0 < 2r. For any even k ≥ 2 and |σ| = l + 1 − k, the harmonicity of T γ in Rn+1 + implies that ∂k σ D (T γ)(x, y) = Dxσ (−∆x )k/2 (T γ)(x, y). ∂y k x This together with (5.2.20) gives  ∂k
h (2) |(∆h ∇l−1 γ)(x)|dh, ζ1 k Dxσ (T γ)(x, y) ≤ c y −n−2 ∂y y Rn

(5.3.16)

where ζ1 obeys (5.2.23). Similarly, for any odd k ≥ 3 ∂ σ ∂k σ D (−∆x )(k−1)/2 (T γ)(x, y). D (T γ)(x, y) = ∂y k x ∂y x Using (5.2.21), we have  ∂k
h (2) σ −n−2 ≤ c y |(∆h ∇l−1 γ)(x)|dh D (T γ)(x, y) ζ2 k x ∂y y n R with ζ2 satisfying (5.2.25). Introducing the notation   J1 := y −n−2 dxdy Gr (X0 )

By

ζ2


h

(2)

|∆h (∇l−1 γ)(x)|dh,

y

we deduce from (5.2.25) that for y0 < 2r  3r   −n−2 J1 ≤ c y dy dx 0

 =c B3r

B2r

|h|−n−1 dh



B2r

(2)

By

|∆h ∇l−1 γ)(x)|dh

(2)

|(∆h ∇l−1 γ)(x)|dx.

(5.3.17)

(5.3.18)

5.3 Descriptions of M (B1m → B1l ) with Integer l

197

Therefore, J1 ≤ c rn−m km,l

(5.3.19)

with km,l given by (5.3.14). Let   h (2) J2 := |(∆h ∇l−1 γ)(x)|dh y −n−2 dxdy. ζ2 y Gr (X0 ) Rn \By

(5.3.20) (n)

By (5.3.17) and (5.2.25), we have for the inner integral over Rn \By 

h (2) |(∆h ∇l−1 γ)(x)|dh ≤ c y n+2 ζ2 y n R \By



(2)

Rn \By

|(∆h ∇l−1 γ)(x)|dh . |h|n+2

We write the integral on the right-hand side as the sum of two integrals, one taken over B2r \By and another over Rn \B2r . We see that 



3r

Br

0

 ≤



|(∆h ∇l−1 γ)(x)|dx

B2r \By

dh |h|n+2

(2)

|(∆h ∇l−1 γ)(x)|dh ≤ c rn−m km,l . |h|n+1

dx Br



(2)

dy

B2r

(5.3.21)

Also, 



3r

dy 0

 (n)

(2)

dx

Br

Rn \B2r

where





3r

I=

|(∆h ∇l−1 γ)(x)|dh ≤ c (I + I+ + I− ), |h|n+2

dy 0

and



(n) Br

|∇l−1 γ(x)| dx 



3r

I± =



dy Br

0

Clearly, I ≤ cr

−1

Rn \B

2r

Rn \B

2r

(5.3.22)

dh |h|n+2

|∇l−1 γ(x±h)|dh dx. |h|n+2

 Br

|∇l−1 γ(x)|dx.

(5.3.23)

Hence, I ≤ c rn−m




 sup ρm−n−1 z∈Rn ρ∈(0,1)

Bρ (z)

 |∇l−1 γ(x)|dx .

(5.3.24)

198

5 The Space M (B1m → B1l )

Obviously,  I± ≤ c r1−n



(n) Br



Rn \B

|∇l−1 γ(x±h)|dh dξ dx. |h|n+2

Br (ξ)

2r

In view of the estimate |ξ| ≤ r + |h| < 12 |ξ| + |h|, this implies that    dξ 1−n I± ≤ c r |∇l−1 γ(x±h)| dx dh n+2 , |ξ| Rn \B2r Br (ξ) Br and therefore I± ≤ c rn−m






 |∇l−1 γ(x)|dx .

sup ρm−n−1 z∈Rn ρ∈(0,1)

Bρ (z)

(5.3.25)

(5.3.26)

Combining (5.3.21), (5.3.23), and (5.3.26), we conclude that J2 defined by (5.3.20) is subject to the inequality  
J2 ≤ c rn−m kl,m + sup ρm−n−1 |∇l−1 γ(x)|dx . (5.3.27) z∈Rn ρ∈(0,1)

Bρ (z)

Together with (5.3.19), this leads to rm−n ∇l+1 (T γ); Gr (X0 )L1 
 m−n−1 ≤ c kl,m + sup ρ |∇l−1 γ(x)|dx . z∈Rn ρ∈(0,1)

(5.3.28)

Bρ (z)

It remains to show that sup X0 ∈Rn+1 +

T γ; G1 (X0 )L1 ≤ c sup γ; B1 (x)L1 .

(5.3.29)

x∈Rn

If y0 ≥ 2, this inequality stems directly from (5.2.33). Let y0 < 2. Clearly, 

3

 

T γ; G1 (X0 )L1 ≤

ζ 0

 +



3

dy 0

B



(n) B1

ζ Rn \B

y (x)


ξ − x y

By (x)


ξ − x y

|γ(ξ)|dξ dx

|γ(ξ)|dξ dx

dy . yn

The first term on the right-hand side does not exceed   3 ζ(t) dt |γ(x + ty)| dxdy ≤ c sup γ; B1 (z)L1 . B

0

B

z∈Rn

dy yn (5.3.30)

(5.3.31)

5.3 Descriptions of M (B1m → B1l ) with Integer l

199

Since ζ is the Poisson kernel, the second term in (5.3.30) is dominated by 

3

 

c 0



B

Rn \By



 

3

≤c

|γ(x + h)| dh dx ydy (y + |h|)n+1

dξ B

0

Rn \B2y

By (ξ)

|γ(x + h)| dh dy dx n−1 . n+1 |h| y

(5.3.32)

In view of the inequality |h| > |ξ|/2 which is valid since |ξ| ≤ r+|h| < 12 |ξ|+|h|, the right-hand side in (5.3.32) is majorized by  3





c 0

Rn \B2y By(ξ)

B

|γ(x + h)|dxdh

dy dξ ≤ c sup γ; (B1 (z))L1 . n+1 |ξ| y n−1 z∈Rn

Combining the last estimate with (5.3.31) and (5.3.32), we arrive at (5.3.29). Now, adding the inequalities (5.3.19), (5.3.27), and (5.3.29), we conclude that the value Kl,m defined by (5.3.13) satisfies  
|∇l−1 γ(x)|dx+sup γ; B1(z)L1 . (5.3.33) Kl,m ≤ c kl,m+ sup rm−n−1 z∈Rn r∈(0,1)

Br (z)

z∈Rn

Estimating the second term on the right-hand side by Lemma 5.2.4, we arrive at (5.3.15). The result follows for l < m. For m = l, instead of (5.3.29), we use the maximum principle n T γ; Rn+1 + L∞ ≤ γ; R L∞ .



The proof of Theorem 5.3.1 is complete.

Remark 5.3.1. It is obvious that for m = l the relation (5.3.1) can be written as γM B1l ∼ 

  | ∆(2) ∇l−1 γ (x, y)|

sup rm−n z∈Rn r∈(0,1)

(Br

(z))2

dx dy + γL∞ . |x − y|n+1

(5.3.34)

Corollary 5.3.1. Suppose that γ ∈ M (B1m (Rn ) → B1l (Rn )). Then ∇j γ ∈ M (B1m−j (Rn ) → B1l−j (Rn )). Proof. This follows directly from Theorem 5.3.1 and Lemma 5.2.4. 5.3.2 Description of M (B1m → B1l ) Involving D1,l Now we give another description of the space M (B1m (Rn ) → B1l (Rn )).

200

5 The Space M (B1m → B1l )

Theorem 5.3.2. Let l be an integer and let m ≥ l ≥ 1. Then the equivalence relation γM (B1m →B1l ) ∼ sup rm−n D1,l γ; Br (z)L1 + γL1,unif

(5.3.35)

z∈Rn r∈(0,1)

holds. If m = l, then γM B1l ∼ sup rm−n D1,l γ; Br (z)L1 + γL∞ .

(5.3.36)

z∈Rn r∈(0,1)

For m ≥ n and m > l,

  γM (B1m →B1l ) ∼ sup D1,l γ; B1 (z)L1 + γ; (B1 (z))L1

(5.3.37)

z∈Rn

which, in its turn, is equivalent to γ; Rn B1,unif . l Proof. The desired lower estimate for the norm γM (B1m →B1l ) follows from (5.3.8) and the estimate D1,l γ; Br (z)L1 ≤ c sup |||γ; Br (ξ)|||B1l ,

(5.3.38)

ξ∈Rn

which holds for all z ∈ Rn and r ∈ (0, 1]. In order to justify (5.3.38), it suffices to check that    (2)  dh | ∆h ∇l−1 γ (x)| n+1 dx |h| Br (z) Rn \Br ≤ c r−1 sup ∇l−1 γ; Br (ξ)L1 .

(5.3.39)

ξ∈Rn

Clearly,



 Rn \B

Br (z)

≤ cr Also,



c ≤ n r



r

dh dx |h|n+1

sup ∇l−1 γ; Br (ξ)L1 .



(5.3.40)

ξ∈Rn



Br (z)

Br (z)

−1

|∇l−1 γ(x)|

Rn \B

Rn \B2r

|∇l−1 γ(x±h)| r

 Br (ξ)

dh dx |h|n+1

|∇l−1 γ(x±h)|

dh dξ dx. |h|n+1

Since |ξ| < r + |h|, it follows that |h| > |ξ|/2 and, therefore, the right-hand side of the last inequality is dominated by   dξ c |∇l−1 γ(x±h)|dh n+1 ≤ c r−1 sup ∇l−1 γ; Br (z)L1 |ξ| n z∈Rn R \B2r Br (ξ) which together with (5.3.40) implies (5.3.39).

5.3 Descriptions of M (B1m → B1l ) with Integer l

201

To get the required upper estimate for γM (B1m →B1l ) , we combine (5.3.33) with Lemma 5.2.4 and Corollary 5.2.3 to conclude that   Km,l ≤ c sup rm−n D1,l γ; Br (z)L1 + γL1,unif . z∈Rn r∈(0,1)

Using this inequality in (5.3.12), the result follows. For m ≥ n, the right-hand side in (5.3.5) is obviously equivalent to l B1,unif (Rn ). The proof is complete. 

5.3.3 M (B1m (Rn ) → B1l (Rn )) as the Space of Traces = {(x, y) : x ∈ Rn , y > 0} and Rn = ∂Rn+1 We use the notation Rn+1 + + . By n+1 k W1 (R+ ) with integer k we mean the space of functions defined on Rn+1 + with finite norm n+1 n+1 U ; Rn+1 + W1k = ∇k U ; R+ L1 + U ; R+ L1 .

The next theorem shows that M (B1m (Rn ) → B1l (Rn )), with integer m and l, l+1 (Rn+1 is the space of traces on Rn of functions in M (W1m+1 (Rn+1 + ) → W1 + )). Theorem 5.3.3. Let m and l be integers, m ≥ l ≥ 1. (i) Suppose that γ ∈ M (B1m (Rn ) → B1l (Rn )). Then the Dirichlet problem ∆Γ = 0 on Rn+1 + ,

Γ |Rn = γ

l+1 has a unique solution in M (W1m+1 (Rn+1 (Rn+1 + ) → W1 + )) and the estimate n Γ ; Rn+1 + M (W m+1 →W l+1 ) ≤ c γ; R M (B1m →B1l ) 1

1

(5.3.41)

holds. (ii) Suppose that l+1 Γ ∈ M (W1m+1 (Rn+1 (Rn+1 + ) → W1 + )).

If γ is the trace of Γ on Rn , then γ ∈ M (B1m (Rn ) → B1l (Rn )) and the estimate γ; Rn M (B1m →B1l ) ≤ c Γ ; Rn+1 + M (W m+1 →W l+1 ) 1

holds.

1

(5.3.42)

5 The Space M (B1m → B1l )

202

Proof. (i) Suppose that γ ∈ M (B1m (Rn ) → B1l (Rn )). Then by Theorem 5.3.1, the right-hand side in (5.3.1) is finite. Taking into account (5.3.15), we conclude that Km,l defined in (5.3.13) is finite. Then reference to the equivalence relation (5.3.10) completes the proof of part (i). (ii) Let U ∈ W1m+1 (Rn+1 + ) and U (x, 0) = u(x). Clearly, by part (i) of Lemma 5.2.1, γ u; Rn B1l ≤ c Γ U ; Rn+1 + W l+1 1

≤

n+1 c Γ ; Rn+1 + M (W1m+1 →W1l+1 ) U ; R+ W1m+1 .

Minimizing the right-hand side over all extensions U of u and using part (ii) of Lemma 5.2.1, we complete the proof. 5.3.4 Interpolation Inequality for Multipliers From Theorem 5.1.2 one readily obtains Corollary 5.3.2. Let 0 < s < n, then γ; Rn M (B1s →L1 ) ∼ sup rs−n γ; Br (x)L1 .

(5.3.43)

γ; Rn M (B1s →L1 ) ∼ sup γ; B1 (x)L1 .

(5.3.44)

x∈Rn r∈(0,1)

Let s ≥ n, then x∈Rn

Theorem 5.3.4. Let m and l be integers, m ≥ l > 0, and let j = 0, . . . , l − 1. Then γM (B m−j →B l−j ) 1

≤

1

1−j/l j/l c γM (B m →B l ) γM (B m−l →L ) . 1 1 1 1

(5.3.45)

Proof. By (5.3.4), 1−j/l

|||u; Br (x)|||B l−j ≤ c |||u; Br (x)|||B l 1

1

j/l

γ; Br (x)L1 .

Hence, sup rm−j−n |||u; Br (x)|||B l−j 1

x∈Rn r∈(0,1)

≤c



sup rm−n |||u; Br (x)|||B l−j

x∈Rn r∈(0,1)

1

1−j/l 

sup rm−l−n γ; Br (x)L1

x∈Rn r∈(0,1)

It remains to apply Theorem 5.3.1 and Corollary 5.3.2.

j/l

.

5.4 Description of the Space M (B1m → B1l ) with Noninteger l

203

5.4 Description of the Space M (B1m → B1l ) with Noninteger l Before we pass to a complete characterization of the space of multipliers M (W1m (Rn ) → W1l (Rn )) for noninteger l, we prove an assertion similar to Lemma 5.2.4. Lemma 5.4.1. Let l be a positive noninteger, and let j = 0, . . . , [l]. Then for any r ∈ (0, 1], rj−l ∇j u; Br L1
  dx dy −l ≤c |∇[l] u(x) − ∇[l] u(y)| + r u; B  (5.4.1) r L1 |x − y|n+{l} (Br )2 

and r ≤c

|∇j u(x) − ∇j u(y)|

j−[l] (Br




)2

|∇[l] u(x) − ∇[l] u(y)|

(Br )2

dx dy |x − y|n+{l}

 dx dy −l . + r u; B  r L 1 |x − y|n+{l}

(5.4.2)

Proof. We apply the same argument as in the proof of Lemma 5.2.4 with (5.2.15) replaced by the identity    ∇[l] u(x) = ϕ(y) ∇[l] u(x) − ∇[l] u(y) dy B



+(−1)[l] B

u(y)(∇[l] ϕ)(y)dy.

(5.4.3) 

Theorem 5.4.1. Let l be a noninteger and let m ≥ l ≥ 0. The equivalence relation γM (W1m →W1l ) ∼
  dx dy −l sup rm−n |∇[l] γ(x) − ∇[l] γ(y)| + r γ; B (z) r L1 |x − y|n+{l} z∈Rn (Br(z))2 r∈(0,1)

holds. For m ≥ n

. γM (W1m →W1l ) ∼ γW1,unif l

Proof. We use the norm |||v; Br |||W1l =

[l] 

rj−l ∇j v; Br L1

j=0

+

[l]  j=0

 r

|∇j v(x) − ∇j v(y)|

j−[l] (Br

)2

dx dy |x − y|n+{l}

(5.4.4)

204

5 The Space M (B1m → B1l )

defined for positive and noninteger l, and r ∈ (0, 1]. Lemma 5.4.1 implies that  dx dy |∇[l] v(x) − ∇[l] v(y)| + r−l v; Br L1 . (5.4.5) |||v; Br |||W1l ∼ |x − y|n+{l} (Br )2 Making dilation in (5.2.6) with noninteger l, we obtain 1−j/l

|||v; Br |||W l−j ≤ c |||v; Br |||W l 1

1

j/l

v; Br L1

(5.4.6)

for any j = 0, . . . , l − 1, and hence |||v; Br |||W l−j ≤ c |||v; Br |||W1l .

(5.4.7)

1

By (5.4.5), the equivalence required in the theorem can be written as γ; Rn M (W1m →W1l ) ∼ sup |||γ; B1 (z)|||W1l .

(5.4.8)

z∈Rn

For m ≥ n the right-hand side of (5.4.8) becomes sup rm−n |||γ; Br (z)|||W1l ∼ γ; Rn W1,unif . l

z∈Rn r∈(0,1)

From Lemma 5.2.3 and (5.4.5), we obtain |||v; Br |||W1l ≤ c v; Rn W1l (5.4.9)   , where r ∈ (0, 1) for m < n and r = 1 for for l < n. Let u(y) = η y−x r m ≥ n, and η ∈ C0∞ (B2 ), η = 1 on B1 . Setting this u into the inequality γ uW1l ≤ γM (W1m →W1l ) uW1m

(5.4.10)

and using (5.4.9) with v = γu, we have |||γ; Br (x)|||W1l ≤ c rn−m γ; M (W1m →W1l )

(5.4.11)

for any x ∈ Rn . The required lower estimate for the norm γM (W1m →W1l ) follows from (5.4.5). Now we obtain the upper estimate for the norm γM (W1m →W1l ) . We have D1,l (γu)L1 ≤c

[l]     |∇j u| D1,l−j γL1 +  |∇j γ| D1,l−j uL1 .

(5.4.12)

j=0

By Theorem 5.1.2  |∇j u| D1,l−j γL1 ≤ c sup r x∈Rn r∈(0,1)

m−j−n

D1,l−j γ; Br (x)L1 ∇j uW m−j . 1

(5.4.13)

5.4 Description of the Space M (B1m → B1l ) with Noninteger l

205

Note that the estimate D1,l γ; Br (z)L1 ≤ c sup |||γ; Br (ξ)|||W1l ,

(5.4.14)

ξ∈Rn

holds for all z ∈ Rn and r ∈ (0, 1]. To justify this estimate, it suffices to check that   dh |∇[l] γ(x + h) − ∇[l] γ(x)| n+{l} dx (n) |h| n Br (z) R \Br ≤ c r−{l} sup ∇[l] γ; Br (ξ)L1

(5.4.15)

ξ∈Rn

which is proved in the same way as (5.3.39) with obvious changes. By (5.4.15), r−j D1,l−j γ; Br (x)L1 ≤ c r−j−{l} sup ∇[l]−j γ; Br (ξ)L1 ξ∈Rn

≤ c sup |||γ; Br (ξ)|||W1l . ξ∈Rn

This together with (5.4.13) implies that  |∇j u| D1,l−j γL1 ≤ c sup rm−n |||γ; Br (ξ)|||W1l uW1m

(5.4.16)

ξ∈Rn r∈(0,1)

for m < n, and  |∇j u| D1,l−j γL1 ≤ c sup |||γ; B1 (z)|||W1l uW1m

(5.4.17)

z∈Rn

for m ≥ n. Since the proof of (4.3.60) is also valid for p = 1, we have  |∇j γ| D1,l−j uL1 ≤ c ∇j γM (W m−l+j →L1 ) uW1m 1

which together with Theorem 5.1.2 gives  |∇j γ| D1,l−j uL1 ≤ c sup rm−l+j−n ∇j γ; Br (x)L1 uW1m . x∈Rn r∈(0,1)

This estimate along with Lemma 5.4.1 implies that  |∇j γ| D1,l−j uL1 ≤ c sup rm−n |||γ; Br (z)|||W1l uW1m

(5.4.18)

z∈Rn r∈(0,1)

for m < n, and  |∇j γ| D1,l−j uL1 ≤ c sup |||γ; B1 (z)|||W1l uW1m

(5.4.19)

z∈Rn

for m ≥ n. Substituting (5.4.16)–(5.4.19) into (5.4.12), we complete the proof. 

206

5 The Space M (B1m → B1l )

5.5 Further Results on Multipliers in Besov and Other Function Spaces 5.5.1 Peetre’s Imbedding Theorem As early as in 1976 Peetre showed that, in order for a function γ be a multiplier l l in Bp,θ , it suffices that γ belongs to B∞,θ (see [Pe2]). He used a LittlewoodPaley decomposition of a function u as the sum of elementary functions un such that the Fourier transform of un is supported by the dyadic annulus of width ∼2n . In this section we reproduce Peetre’s result. l with p ≥ 1 defined in Sect. 4.4 can be supplied with other The space Bp,θ norms (see [Pe2], [Tr3]). For example, for l ∈ (0, 1) one may put ⎧
 ∞  −1 θ dt 1/θ ⎪ ⎨ t ωp (t, u) + uLp for θ < ∞, t 0 = uBp,θ l ⎪ for θ = ∞, ⎩sup t−1 ωp (t, u) + uLp t>0




where ωp (t, u) = sup

|h|≤t

Rn

|u(x + h) − u(x)|p dx

1/p .

l Another frequently used norm in Bp,θ introduced by Peetre has the following definition. Let ϕ ∈ S, where S is the space of rapidly decreasing functions. Further let (1) supp ϕ = {ξ : 2−1 ≤ |ξ| ≤ 2}, (2) ϕ(ξ) > 0 for 2−1 < |ξ| < 2, ∞  ϕ(2−k ξ) = 1, ξ = 0. (3) k=−∞

Let Ψ and ϕk be given by  F Ψ (ξ) = 1 − ϕ(2−k ξ);

F ϕk (ξ) = ϕ(2−k ξ),

−∞ < k < +∞, (5.5.1)

k≥1

where F is the Fourier transform. Then the norm (the quasi-norm for 0 < θ < 1) l in Bp,θ is equivalent to ⎧
  kl θ 1/θ ⎪ ⎪ 2 u ∗ ϕk Lp + u ∗ Ψ Lp for θ < ∞, ⎨ (1) k≥1 uB l = p,θ ⎪ ⎪ for θ = ∞ ⎩sup 2kl u ∗ ϕk Lp + u ∗ Ψ Lp k≥1

(see, for example, [Pe2], [Tr3]). Using the right-hand sides of these equalities, l for all l ∈ R1 and for p ∈ (0, 1). one can define the spaces Bp,θ l Theorem 5.5.1. [Pe2] If l > 0, p ∈ [1, ∞], and θ ∈ (0, ∞], then B∞,θ ⊂ l M Bp,θ .

5.5 Further Results on Multipliers in Besov and Other Function Spaces

207

l l Proof. Let u ∈ Bp,θ and let γ ∈ B∞,θ . We put U = u ∗ Ψ , uk = u ∗ ϕk , and similarly Γ = γ ∗ Ψ , γk = γ ∗ ϕk . Then   u=U+ uk , γ=Γ + γk . k≥1

k≥1

For the product g = γ u we have the decomposition    g = ΓU + Γ uj + U γk + γk uj . j≥1

k≥1

(5.5.2)

k≥1 j≥1

Clearly, g ∗ Ψ Lp ≤ c gLp . In order to estimate the norm
 θ 1/θ 2ml g ∗ ϕm Lp , m≥1

we consider only the functions (γk uj ) ∗ ϕm because other terms in (5.5.2) can be estimated in a similar way. Let us introduce the spherical layer Gm = {ξ : 2m−1 < |ξ| < 2m+1 }. Note that

    F −1 (γk uj ) ∗ ϕm = F −1 γk ∗ F −1 uj F −1 ϕm

and supp F −1 γk ⊂ Gk ,

supp F −1 uj ⊂ Gj ,

supp F −1 ϕm ⊂ Gm .

Therefore, if (γk uj ) ∗ ϕm is not identical to zero, we have 2m−1 ≤ 2j+1 + 2k+1 . Hence it is sufficient to estimate the norms +   + (γk uj ) ∗ ϕm +L Mm := + p

j≥m−3 k≥1

and

+   + (γk uj ) ∗ ϕm +L . Lm := + p

k≥m−3 j≥1

We have

+     + uj Mm = + γk ∗ ϕm +L

p

j≥m−3

+ + ≤ c+ γk +L



k≥1

j≥m−3

∞

k≥1

uj Lp ≤ c γL∞



uj Lp .

j≥m−3

This implies the estimate Mθm ≤ c γθL∞ sup



j≥m−3

 2sθj uj θLp 2−sθm

(5.5.3)

208

5 The Space M (B1m → B1l )

for any s ∈ (0, l). Hence    2lθm Mθm ≤ c γθL∞ 2(l−s)θm 2sθj uj θLp m≥4

m≥4



≤ c γθL∞

j≥m−3

2lθm um θLp .

m≥1

The norm Lm satisfies +     + γk Lm = + uj ∗ ϕm +L

p

k≥m−3

j≥1

  + + ≤ c+ uj +Lp γk L∞ ≤ c uLp γk L∞ . j≥1

k≥m−3

k≥m−3

Hence, by the same argument as in the case of the norm Mm , we conclude that   2lθm Mθm ≤ c uθLp 2lθm γm θL∞ . m≥4

m≥1

Finally, we obtain   γuBp,θ . ≤ c γB∞,θ uLp + γL∞ uBp,θ l l l 

The proof is complete.

5.5.2 Related Results on Multipliers in Besov and Triebel-Lizorkin Spaces We begin an overview of further results with Triebel’s theorem, similar to l l Theorem 5.5.1, on multipliers in Bp,θ and in the Triebel-Lizorkin space Fp,θ with l ∈ R1 and positive p and θ. l is defined by The norm (quasi-norm) in Fp,θ +
 θ 1/θ + + + uFp,θ 2kl |u ∗ ϕk | =+ l + k≥1

Lp

+ u ∗ Ψ Lp .

(For properties of these spaces see [Tr4] and [RS].) Theorem 5.5.2. [Tr4] (i) If l ∈ R1 , p ∈ (0, ∞], θ ∈ (0, ∞], and ρ > ρ l ⊂ M Bp,θ . max{l, −l + n/p}, then B∞,∞ (ii) If l ∈ R1 , p ∈ (0, ∞], θ ∈ (0, ∞], and ρ > max{l, −l + n/ min(p, θ)}, ρ l ⊂ M Fp,θ . then B∞,∞

5.5 Further Results on Multipliers in Besov and Other Function Spaces

209

l Note that Fp,2 coincides with the space of Bessel potentials Hpl for p > 1 and in this case part (ii) of Theorem 5.5.2 is contained in part (i) of Theorem 3.4.1. Peetre’s approach, based on the decomposition (5.5.2) and sometimes called the paraproduct algorithm, was used in the study of multipliers in Besov and Triebel-Lizorkin spaces in [Sic1]–[Sic3], [Tr4], Sect. 2.8, [Jo], [RS], [Mar1]–[Mar3], [Yam], [Yu], [SS], [KoS], [Ger]. In [SS] Sickel and Smirnov showed that l l = Bp,θ,unif M Bp,θ

for any p, θ such that 1 ≤ p ≤ θ and l > n/p, whereas Bourdaud [Bo] demonstrated that l l = Bp,θ,unif M Bp,θ for 1 ≤ θ < p ≤ ∞, l > n/p. 0 0 and B∞,∞ were characterMultipliers preserving the Besov spaces B∞,1 ized by Koch and Sickel [KoS]. In particular, they found the equivalence relation 0 0 ∼ γL∞ + γF∞,1 + sup(k + 1) γ ∗ ϕk L∞ γM B∞,∞ k≥0

0 with ϕk defined in (5.5.1). We also mention that M Bp,θ = L∞ unless p = θ = 2, according to Frazier and Jawerth [FrJ]. s s , Fp,θ ⊂ M (X), where X = Bps00 ,θ0 or The imbeddings of the form Bp,θ X = Fps00,θ0 were studied in [RS], Ch. 4. A description of M (w2m → w2−k ), where m and k belong to (−n/2, n/2) and m = k, can be found in [Ger]. Netrusov [Net] gave a characterization of multipliers in Triebel-Lizorkin l l , p ≤ θ ≤ ∞, and Besov spaces Bp,∞ , where 0 < p ≤ 1. Different spaces Fp,θ l characterizations of M Fp,θ were obtained by Sickel [Sic3] in the case

0 < p < ∞,

0 < θ ≤ ∞,

l > n max{0, p−1 − 1, θ−1 − 1}.

1 We note that, since B1,∞ = BV (see [Pe2], p. 164, and [Gu1]), it follows from Theorem 2.9.3 that 1 γM B1,∞ ∼ sup r1−n var ∇γ(Br (x)).

(5.5.4)

x∈Rn , r∈(0,1)

l l → Bp,∞ ), where p ∈ (1, ∞) The following description of the space M (Bp,1 and 0 < l ≤ 1/p, was given by Gulisashvili [Gu1], [Gu2]. l l → Bp,∞ ), p ∈ Theorem 5.5.3. A function γ belongs to the space M (Bp,1 (1, ∞), 0 < l ≤ 1/p, if and only if γ ∈ L∞ and  |∆h γ(t)|p dt ≤ c |h|pl rn−pl Br (x)

for all balls Br (x), r ∈ (0, 1) and any h ∈ Rn .

5 The Space M (B1m → B1l )

210

The relation ∼ γM (Bp,1 l →B l p,∞ )


sup x∈Rn ,r∈(0,1) h∈Rn \{0}

1 |h|pl rn−pl

 Br (x)

|∆h γ(t)|p dt

1/p

+ uL∞

holds. As a corollary of this result one obtains the following condition for the in1/p 1/p clusion of χE into M (Bp,1 → Bp,∞ ), where χE is the characteristic function of a Lebesgue measurable set E in Rn . (See Sect. 2.9 for the notations s and ∂ ∗ .) 1/p

1/p

Corollary 5.5.1. [Gu1] The inclusion χE ∈ M (Bp,1 → Bp,∞ ) holds if and only if E is a set with local finite perimeter and s(Br (x) ∩ ∂ ∗ E) ≤ c rn−1 for all balls Br (x) with r ∈ (0, 1). Comparing this assertion with Corollary 2.9.1 and (5.5.4), we see that χE 1/p 1/p 1 1 belongs to M (Bp,1 → Bp,∞ ) and M (B1,1 → B1,∞ ) simultaneously. 5.5.3 Multipliers in BM O The space BM O of functions with bounded mean oscillation (see [JN], [Cam], [F1], [Ste2], [Ja1], [Ja2], and elsewhere) plays an important role in modern 0 0 and B∞,∞ . This space is defined as analysis. It is situated between B∞,1 n follows. Let Qr (x) be the cube in R with side length r centered at x whose sides are parallel to coordinate axes. By f (Q) we denote the mean value of f on a cube Q, that is,  1 f (Q) = f (x)dx. mesn Q Q Further, we introduce the mean oscillation of f on Q by  1 O(f, Q) = |f (x) − f (Q)|dx. mesn Q Q By BM O we denote the space of functions integrable on Rn and such that sup x∈Rn ,r∈(0,1)

O(f, Qr (x)) < ∞.

Endowed with the norm f BM O = f L1 + BM O becomes a Banach space.

sup x∈Rn ,r∈(0,1/2)

O(f, Qr (x)),

5.5 Further Results on Multipliers in Besov and Other Function Spaces

211

We can include BM O into the family of spaces BM Oϕ of locally integrable functions with the finite norm f L1 +

sup x∈Rn ,r∈(0,1/2)

O(f, Qr (x)) , ϕ(r)

where ϕ ia a positive nondecreasing function on (0, 1/2). The following theorem, containing a description of the space M (BM O) of multipliers in BM O, is due to Stegenga [Ste1] and Janson [Ja1]. Theorem 5.5.4. The space M (BM O) coincides with BM O| log r|−1 ∩ L∞ . Proof. We begin by deriving the following useful estimate for functions in BM O: |f (Qr ) − f (Q1/2 )| ≤ c f BM O | log r|, (5.5.5) where Qρ = Qρ (x). It is clear that the left-hand side of this inequality does not exceed  1/2   1/2 dρ d |f (y) − f (Qρ )|dsy f (Qρ ) dρ ≤ dρ ρn ∂Qρ r r  ≤ c1 r

1/2

dρ ρn+1





ρ

|f (y) − f (Qρ )|dsy .

dr ∂Qρ

ρ/2

Therefore, 

1/2

|f (Qr ) − f (Q1/2 )| ≤ c r

dρ ρn+1

 |f (y) − f (Qρ )|dy, Qρ

which implies (5.5.5). By (5.5.5), |f (Qr )| ≤ 2n f L1 + c f BM O | log r|.

(5.5.6)

Let γ ∈ BM O| log r|−1 ∩ L∞ . Then  |(γf )(y) − γ(Qr )f (Qr )|dy r−n Qr

≤ r−n

 Qr

|γ(x)| |f (x) − f (Qr )|dy + r−n

 |f (Qr )| |γ(y) − γ(Qr )|dy Qr

≤ γL∞ O(f, Qr ) + |f (Qr )| O(γ, Qr ). From (5.5.6) it follows that the last sum is dominated by   c | log r| γL∞ + γBM O| log r|−1 f BM O .

5 The Space M (B1m → B1l )

212

It remains to note that 2 O(f, Qr ) ≤ mesn Qr

 |f (y) − a|dy Qr

for any number a. Hence γ ∈ M (BM O). Now let us show that M (BM O) ⊂ BM O| log r|−1 ∩ L∞ . Obviously, 1/N

1/N

γL∞ = lim γ N f L1 ≤ lim inf γ N f BM O ≤ γM (BM O) . N →∞

N →∞

(5.5.7)

We see that  1 |γ(y) − γ(Qr )| |f (y)|dy ≤ O(γf, Qr ) + 2γL∞ O(f, Qr ). mesn Qr Qr This together with (5.5.7) and the inequality O(γf, Qr ) ≤ γM (BM O) f BM O implies that 1 mesn Qr

 |γ(y) − γ(Qr )| |f (y)|dy ≤ 3 γM (BM O) f BM O . Qr

Setting here f (y) = η(y) log[x − y], where η ∈

C0∞ (Qr )

and η = 1 on Q1/2 , we obtain O(γ, Qr ) ≤ c | log r|−1 . 

The proof is complete.

The following general result has a slightly more complicated proof (see [Ja1]). Theorem 5.5.5. Let the function ϕ(r) r−1 be almost decreasing in the sense that for ρ ≥ r. ϕ(ρ) ρ−1 ≤ c ϕ(r) r−1 Then M (BM Oϕ ) = BM Oψ ∩ L∞ , where


 ψ(r) = ϕ(r) r

1

ϕ(t)

dt −1 . t

Concerning the space M (BM Oϕ ) for Rn and for general domains, see the series of papers by Bloom [Blo], Nakai [Na1], [Na2], Nakai and Yabuta [NY1], [NY2], Yabuta [Ya]. Using the duality of the Hardy space H 1 and BM O (see [St2]), Janson [Ja1] proved the coincidence of spaces of multipliers in H 1 and BM O. In other words, M H 1 = BM O| log r|−1 ∩ L∞ .

6 Maximal Algebras in Spaces of Multipliers

6.1 Introduction Let A be a subset of a Banach function space. Then A is called a multiplication algebra if for all u and v in A their product uv belongs to A and there exists a constant c such that uv ≤ c u v. Let l be an integer. For lp ≤ n and p ∈ (1, ∞), or for l < n and p = 1, the space Wpl contains unbounded functions which are certainly not multipliers in Wpl (see, for example, (2.7.1)). Hence the space Wpl is not a multiplication algebra for the values of p, l, and n given above. It is not difficult to describe the maximal algebra contained in Wpl . If u ∈ A, then for any N = 1, 2, . . . 1/N

1/N

uN Lp ≤ uN W l ≤ c uWpl . p

Consequently, A ⊂ Wpl ∩ L∞ . On the other hand, it is well known that the intersection Wpl ∩ L∞ is a multiplication algebra. In fact, for all u and v in Wpl ∩ L∞ , ∇l (uv)Lp ≤ c

l 

 |∇k u| |∇l−k v| Lp ≤ c

k=0

≤c

l 

∇k uLpl/k ∇l−k vLpl/(l−k)

k=0 l 

(l−k)/l

uL∞

k/l

k/l

(l−k)/l

uW l vL∞ vW l p

.

p

k=0

Here we have used the Gagliardo - Nirenberg inequality (l−j)/l

∇j uLpl/j ≤ c uL∞

j/l

uW l , p

j = 1, . . . , l − 1.

(see [Gag2] and [Nir]). Thus the space Wpl ∩ L∞ is the maximal algebra contained in Wpl . V.G. Maz’ya, T.O. Shaposhnikova, Theory of Sobolev Multipliers, Grundlehren der mathematischen Wissenschaften 337, c Springer-Verlag Berlin Hiedelberg 2009


213

214

6 Maximal Algebras in Spaces of Multipliers

Since by Sobolev’s theorem Wpl ⊂ L∞ for lp > n, p ∈ (1, ∞) or for l ≥ n, p = 1, it follows that Wpl is a multiplication algebra for the indicated values of p and l. Obviously, this known assertion follows from (2.2.5) and (2.3.29) as well. Let us turn to the question of Banach algebras in spaces of multipliers M (Wpm → Wpl ). Obviously, this question is trivial for M Wpl which is a Banach algebra itself, but this is not the case for spaces of multipliers mapping one Sobolev space into a different one. In Sect. 6.3 we show that the maximal Banach algebra Am,l p , imbedded in the space of multipliers M (Wpm → Wpl ) which map the Sobolev space Wpm to Wpl with noninteger m and l, m > l and p ∈ [1, ∞), is isomorphic to M (Wpm → Wpl ) ∩ L∞ . It is proved in Sect. 6.4 that the maximal Banach algebra Am,l p , imbedded in the space of multipliers acting between Bessel potential spaces M (Hpm → Hpl ), is isomorphic to M (Hpm → Hpl ) ∩ L∞ . Precise descriptions of the imbed⊂ Aµ,λ and Am,l ⊂ Aµ,λ are given in Sect. 6.5. dings Am,l p p p p

6.2 Pointwise Interpolation Inequalities for Derivatives 6.2.1 Inequalities Involving Derivatives of Integer Order The aim of this subsection is the following inequality (see [Kal] and [MSh17]). Lemma 6.2.1. Let 0 < k < m. Then |∇k u(x)| ≤ c (Mu(x))

m−k m

k

(M∇m u(x)) m .

(6.2.1)

Proof. Let η be a function in the ball B1 with Lipschitz derivatives of order m − 2 and which vanishes on ∂B1 together with all these derivatives. Also let  η(y) dy = 1. (6.2.2) B1

We need the Sobolev integral representation: v(0) =



t

−n

|β| 1 and {s} < {l}, then even the rougher inequality l−s   s  l Dp,l u(x) l Dp,s u (x) ≤ c uL∞

does not hold. Proof. (i) From (6.2.17) we deduce that 

   l−s  s (r) (r) Dp,s u (x) ≤ c Mu(x) l r−l Mu(x) + Dp,l u(x) l .

The result follows by passing to the limit as r → ∞. (ii) Let s > 1 and {s} < {l}. Suppose that the inequality {l}−{s}   s Dp,[s]+{l} u(x) [s]+{l} (Dp,s u)(x) ≤ c uL[s]+{l} ∞

(6.2.27)

holds. According to part (i), [l]−[s]

(Dp,[s]+{l} u)(x) ≤ c uL∞l

  [s]+{l} l Dp,l u(x)

which together with (6.2.27) results in (6.2.17). Hence it is enough to disprove (6.2.27).

220

6 Maximal Algebras in Spaces of Multipliers

We set x = 0 and assume that supp u ⊂ B2 \B1 . Then (6.2.27) implies that
 1/p |∇[s] u(h)|p dh ≤ c uL∞ B2 \B1



which, obviously, does not hold for all u. This completes the proof. Remark 6.2.2. For p = ∞ the inequality (6.2.17) becomes sup y


  l−s |∇[s] u(y) − ∇[s] u(x)| |∇[l] u(y) − ∇[l] u(x)|  sl l sup ≤ c Mu(x) , |y − x|{s} |y − x|{l} y

where s < l and either {s} ≤ {l} or 0 < s < 1.

6.3 Maximal Banach Algebra in M (Wpm → Wpl ) 6.3.1 The Case p > 1 Theorem 6.3.1. Let m ≥ l ≥ 0, and let p ∈ (1, ∞). The maximal Banach imbedded into algebra Am,l p M (Wpm → Wpl ) is isomorphic to the space M (Wpm → Wpl ) ∩ L∞ .

(6.3.1)

The estimate γ1 γ2 M (Wpm →Wpl )   ≤ c γ1 L∞ γ2 M (Wpm →Wpl ) + γ2 L∞ γ1 M (Wpm →Wpl ) holds. Proof. The statement is trivial for m = l, since M Wpl is an algebra and is imbedded into L∞ (see (2.3.15) and (4.3.28)). be a Banach subalgebra of M (Wpm → Wpl ) and let c be a constant Let Am,l p such that γM (Wpm →Wpl ) ≤ c γAm,l p m,l m for all γ ∈ Am,l p . For every N = 1, 2 . . . and all γ ∈ Ap , u ∈ Wp we have 1/N

1/N

1/N

1/N

γ N uLp ≤ γ N uW l ≤ γ N M (W m →W l ) uWpm p

p

1/N

p

1/N

≤ (c γ N Am,l )1/N uWpm ≤ c1/N γAm,l uWpm . p p

6.3 Maximal Banach Algebra in M (Wpm → Wpl )

221

Passing to the limit as N → ∞, we obtain that γ ∈ L∞ and γL∞ ≤ γAm,l . p

(6.3.2)

Consider first the case of integer l. Suppose that γ1 and γ2 belong to (6.3.1). Then, for all u ∈ Wpm l
 ∇l (γ1 γ2 u)Lp ≤ c γ1 L∞ ∇l (γ2 u)Lp + γ2 L∞  |∇h γ1 | |∇l−h u| Lp h=1

+

l−1  l−h 

  |∇h γ1 | |∇k γ2 | |∇l−h−k u| Lp .

(6.3.3)

h=1 k=1

The first term on the right-hand side is majorized by c γ1 L∞ γ2 M (W m →W l ) uW m . p

p

p

To estimate the second term, we recall that, if γ ∈ M (Wpm → Wpl ), then for any j = 0, . . . , l, ∇j γ ∈ M (Wpm → Wpl−j ) ⊂ M (Wpm−l+j → Lp ) and the estimate ∇j γM (Wpm−l+j →Lp ) ≤ c γM (W m →W l ) p

p

(6.3.4)

holds (see (2.3.13)). Therefore, the second term on the right-hand side of (6.3.3) is not greater than c γ2 L∞ γ1 M (W m →W l ) uWpm . p

p

To estimate the remaining terms on the right-hand side of (6.3.3), we use the inequality k

h

(M∇h+k γ(x)) k+h |∇h γ(x)| ≤ c γLh+k ∞ stemming from (6.2.1). Hence  |∇h γ1 | |∇k γ2 | |∇l−h−k u| Lp k

h

h

k

≤ c γ1 Lh+k γ2 Lh+k  (M∇h+k γ1 ) h+k (M∇h+k γ2 ) h+k |∇l−h−k u| Lp ∞ ∞ k

h

h

≤ cγ1 Lh+k γ2 Lh+k (M∇h+k γ1 )|∇l−h−k u| Lh+k (M∇h+k γ2 )|∇l−h−k u| Lp . ∞ ∞ p By (2.3.22) we have MγM (W s →Lp ) ≤ c γM (W s →Lp ) . p

p

(6.3.5)

222

6 Maximal Algebras in Spaces of Multipliers

This inequality and (6.3.4) give  |∇h γ1 | |∇k γ2 | |∇l−h−k u| Lp k

h

k

h

h

k

≤ c γ1 Lh+k γ2 Lh+k  |∇h+k γ1 | |∇l−h−k u| Lh+k  |∇h+k γ2 | |∇l−h−k u| Lh+k ∞ ∞ p p h

k

h+k ≤ c γ1 Lh+k γ2 Lh+k γ1 M γ  h+k uW m ∞ ∞ p (Wpm →Wpl ) 2 M (Wpm →Wpl )

which completes the proof for integer l. Let l be a positive noninteger. Suppose that γ1 and γ2 belong to (6.3.1). Then, for all u ∈ Wpm Dp,l (γ1 γ2 u)Lp = Dp,{l} ∇[l] (γ1 γ2 u)Lp    Dp,{l} Dα γ1 Dβ γ2 Dσ u Lp ≤c

(6.3.6)

|α|+|β|+|σ|=[l]

≤c



(Aα,β,σ + Bα,β,σ + Cα,β,σ ),

|α|+|β|+|σ|=[l]

where Aα,β,σ = (Dα γ1 ) (Dβ γ2 ) Dp,{l} Dσ uLp , Bα,β,σ = (Dα γ1 ) (Dp,{l} Dβ γ2 ) Dσ uLp , and Cα,β,σ is given by
  dhdx 1/p . |Dα γ1 (x)|p |Dβ γ2 (x+h)|p |Dσ u(x+h)−Dσ u(x)|p n+p{l} |h| To estimate Aα,β,σ , we use the inequality k−i  i |∇i ϕ(x)| ≤ c ϕLk∞ M∇k ϕ(x) k a.e. in Rn

(6.3.7)

(6.3.8)

with 0 ≤ i ≤ k, which follows directly from (6.2.1). We have 1−

|α|

1−

|β|

Aα,β,σ ≤ cγ1 L∞[l]−|σ| γ2 L∞[l]−|σ|
    p|α|   p|β| × M∇[l]−|σ| γ1 (x) [l]−|σ| M∇[l]−|σ| γ2 (x) [l]−|σ| ×|Dσ u(x + h)−Dσ u(x)|p

dhdx 1/p . |h|n+p{l}

By H¨older’s inequality, the double integral on the right-hand side is dominated by |α|
   p dhdx  ([l]−[σ|)p c M∇[l]−|σ| γ1 (x) |Dσ u(x + h) − Dσ u(x)|p n+p{l} |h| |β|
   p σ dhdx  ([l]−|σ|)p × M∇[l]−|σ| γ2 (x) |D u(x + h) − Dσ u(x)|p n+p{l} |h|

6.3 Maximal Banach Algebra in M (Wpm → Wpl )

which equals

223

|α|   c  M∇[l]−|σ| γ1 Dp,|σ|+{l} uL[l]−|σ| p |β|   × M∇[l]−|σ| γ2 Dp,|σ|+{l} uL[l]−|σ| . p

(6.3.9)

This expression does not exceed |α|

c M∇[l]−|σ| γ1  [l]−|σ|m−|σ|−{l} M (Wp

|α|

→Lp )

|β|

×M∇[l]−|σ| γ2  [l]−|σ|m−|σ|−{l} M (Wp

[l]−|σ| uW m p |β|

→Lp )

[l]−|σ| uW . m p

(6.3.10)

We estimate (6.3.10) using (6.3.5) combined with the inequality ∇j γM (Wpm−l+j →Lp ) ≤ c γM (Wpm →Wpl )

(6.3.11)

(see (2.2.11)). Since |α| + |β| + |σ| = [l], we obtain from (6.3.9)– (6.3.11) |β|

|α|

Aα,β,σ ≤ c γ1 L|α|+|β| γ2 L|α|+|β| ∞ ∞ |α|

|β|

|α|+|β| |α|+|β| ×γ1 M γ2 M uWpm . (W m →W l ) (W m →W l ) p

p

p

p

Now, by H¨ older’s inequality  Aα,β,σ ≤ c γ1 L∞ γ2 M (Wpm →Wpl )  +γ2 L∞ γ1 M (Wpm →Wpl ) uWpm .

(6.3.12)

To estimate Bα,β,σ defined by (6.3.7), we use (6.3.8) with ϕ = Dα γ1 and apply Lemma 6.2.3 with ϕ = γ2 and s = {l} + |β|. Then 1−

|α|

1− |β|+{l}

Bα,β,σ ≤ c γ1 L∞l−|σ| γ2 L∞ l−|σ|

 |α|   |β|+{l}  × Dp,l−|σ| γ1 l−|σ| Dp,l−|σ| γ2 l−|σ| Dσ uLp . By H¨older’s inequality, the last norm is dominated by |α|

|β|+{l}

Dp,l−|σ| γ2 Dσ uLl−|σ| . Dp,l−|σ| γ1 Dσ uLl−|σ| p p Since Dp,l−j γM (Wpm−j →Lp ) ≤ c γM (Wpm →Wpl ) (see (4.3.87)), it follows for i = 1, 2 that Dp,l−|σ| γi Dσ uLp ≤ Dp,l−|σ| γi M (Wpm−|σ| →Lp ) uWpm ≤ c γi M (Wpm →Wpl ) uWpm .

224

6 Maximal Algebras in Spaces of Multipliers

Hence, |β|+{l}

|α|+{l}

Bα,β,σ ≤ c γ1 L|α|+|β|+{l} γ2 L|α|+|β|+{l} ∞ ∞ |α|

|β|+{l}

|α|+|β|+{l} |α|+|β|+{l} × γ1 M γ2 M uWpm . (W m →W l ) (W m →W l ) p

p

p

p

Therefore, by H¨ older’s inequality, Bα,β,σ has the same majorant (6.3.12) as Aα,β,σ . In order to estimate Cα,β,σ , we use (6.3.8). Then 1−

|α|

1−

|β|

Cα,β,σ ≤ c γ1 L∞[l]−|σ| γ2 L∞[l]−|σ| × ×


    p|α|   p|β| M∇[l]−|σ| γ1 (x) [l]−|σ| M∇[l]−|σ| γ2 (x + h) [l]−|σ| 1/p |Dσ u(x + h) − Dσ u(x)|p dhdx . n+p{l} |h|

By H¨older’s inequality, the double integral on the right-hand side is not greater than |α|
   p dhdx  ([l]−|σ|)p c M∇[l]−|σ| γ1 (x) |Dσ u(x + h) − Dσ u(x)|p n+p{l} |h| |β|
   p dhdx  ([l]−|σ|)p × M∇[l]−|σ| γ2 (x + h) |Dσ u(x + h) − Dσ u(x)|p n+p{l} |h| which coincides with (6.3.9). The above estimate of (6.3.9) implies (6.3.12)  with Aα,β,σ replaced by Cα,β,σ . This completes the proof. Theorems 6.3.1 and 4.1.1 imply Corollary 6.3.1. The maximal Banach algebra in M (Wpm → Wpl ), m ≥ l, l with finite norm p ∈ (1, ∞), consists of functions γ ∈ Wp,loc sup e⊂Rn diam(e)≤1

Dp,l γ; eLp + γL∞ . (Cp,m (e))1/p

(6.3.13)

In the case mp > n the norm (6.3.13) can be simplified as Dp,l γLp,unif + γL∞ . 6.3.2 Maximal Banach Algebra in M (W1m → W1l ) Theorem 6.3.2. Let m ≥ l ≥ 0. The maximal Banach algebra Am,l imbedded 1 into M (W1m → W1l ) is isomorphic to the space

6.3 Maximal Banach Algebra in M (Wpm → Wpl )

M (W1m → W1l ) ∩ L∞ .

225

(6.3.14)

The estimate γ1 γ2 M (W1m →W1l )



≤ c γ1 L∞ γ2 M (W1m →W1l ) + γ2 L∞ γ1 M (W1m →W1l )



holds. Proof. The imbedding Am,l ⊂ L∞ is proved in the same way as in Theorem 1 6.3.1, where the case p > 1 is considered. (i) Let l be integer. Suppose that γ1 and γ2 belong to (6.3.14). Then γ1 γ2 uW1l ≤ c

l 

 |∇j (γ1 γ2 )| |∇l−j u|L1 .

j=0

We show that ∇j (γ1 γ2 ) ∈ M (W1m−l+j → L1 ), j = 0, . . . , l,

(6.3.15)

which implies the result. Since the intersection W1j (B1 ) ∩ L∞ (B1 ) is an algebra, we have   γ1 γ2 ; B1 W j ≤ c γ1 ; B1 L∞ γ2 ; B1 W j + γ2 ; B1 L∞ γ1 ; B1 W j . 1

1

1

Hence, for any r > 0   rj |∇j (γ1 γ2 )(x)|dx ≤ c γ1 ; Br L∞ (rj ∇j γ2 ; Br L1 + rn γ2 ; Br L1 ) Br  + γ2 ; Br L∞ (rj ∇j γ1 ; Br L1 + rn γ1 ; Br L1 ) . This inequality, taken for r ∈ (0, 1), along with (2.2.3) gives  rm−l+j−n ∇j (γ1 γ2 ); Br L1 ≤ c γ1 ; Br L∞ γ2 M (W1m →W1l ) +γ2 ; Br L∞ γ1 M (W1m →W1l ) which implies (6.3.15). (ii) Let l be a noninteger. We have       |∇[l] γ1 (x)γ2 (x) − ∇[l] γ1 (y)γ2 (y) | B1

B1


 [l]



dxdy |x − y|n+{l}



dxdy |x − y|n+{l} B1 B1 k=0    dxdy . (6.3.16) + |∇[l]−k γ2 (y)| |∇k γ1 (x) − ∇k γ1 (y)| |x − y|n+{l} B1 B1 ≤

|∇k γ1 (x)| |∇[l]−k γ2 (x) − ∇[l]−k γ2 (y)|

226

6 Maximal Algebras in Spaces of Multipliers

By H¨older’s inequality, the right-hand side does not exceed c

[l] 1  

∇k γ1+i ; B1 L l k

i=0 k=0



 × B1

B1

|∇[l]−k γ2−i (x) − ∇[l]−k γ2−i (y)|

l  l−k  dy dx n+{l} |x − y|

l−k l

. (6.3.17)

Duplicating the argument leading from (6.2.25) to (6.2.17), we conclude that (6.2.15) implies the estimate 1− kl  l  k (t) t u; Bt L∞ + Dp,l u(0) l . |∇k u(0)| ≤ c u; Bt L∞

(6.3.18)

For t = 1 and u replaced by γ it becomes |∇k γ(x)| 1− k

≤ c γ; B1 L∞l


 B1

|∇[l] γ(x)−∇[l] γ(y)|

 kl dy +γ; B  . (6.3.19) 1 L ∞ |x − y|n+{l}

Furthermore, by Lemma 6.2.3  |∇[l]−k γ(x) − ∇[l]−k γ(y)| B1

k

≤ cγ; B1 Ll ∞


 B1

|∇[l] γ(x) − ∇[l] γ(y)|

dy |x − y|n+{l}

 l−k dy l +γ; B  . (6.3.20) 1 L 1 n+{l} |x − y|

Using (6.3.17)–(6.3.20) in (6.3.16), we find that       |∇[l] γ1 (x)γ2 (x) − ∇[l] γ1 (y)γ2 (y) | B1

≤c

[l] 1  

B1

1− k γ1+i ; Rn L∞l

i=0 k=0


 B1

 B1

|∇[l] γ1+i (x) − ∇[l] γ1+i (y)|

+γ1+i ; B1 L1
 ×

B1

 B1

dxdy |x − y|n+{l}

 kl

k

γ2−i ; Rn Ll ∞

|∇[l] γ2−i (x) − ∇[l] γ2−i (y)|

1− kl dy + γ ; B  . 2−i 1 L 1 |x − y|n+{l}

By the dilation x → x/r we obtain       m−n |∇[l] γ1 (x)γ2 (x) − ∇[l] γ1 (y)γ2 (y) | r Br

Br

dy |x − y|n+{l}

dxdy |x − y|n+{l}

6.4 Maximal Algebra in Spaces of Bessel Potentials

≤c

[l] 1  

1− k γ1+i L∞l

 


r

m−n Br Br

i=0 k=0

+ rm−l γ1+i ; Br L1 
× rm−n

k l

|∇[l] γ1+i (x) − ∇[l] γ1+i (y)|

Br

dy |x − y|n+{l}

k

γ2−i Ll ∞



Br

227

|∇[l] γ2−i (x) − ∇[l] γ2−i (y)|

1− k dy l m−l + r γ ; B  . 2−i r L 1 |x − y|n+{l}

Using this estimate for r ∈ (0, 1) together with Theorem 5.4.1, we arrive at (r)

rm−n D1,l (γ1 γ2 ); Br L1 ≤ c

[l] 1  

1− k

k

k

1− k

l l γ1+i L∞l γ1+i M γ2−i Ll ∞ γ2−i M (W m →W l ) (W m →W l ) 1

i=0 k=0

1

1

1

≤ γ1 L∞ γ2 M (W1m →W1l ) + γ2 L∞ γ1 M (W1m →W1l ) which, along with Theorem 5.4.1 and the obvious inequality rm−l−n γ1 γ2 ; Br (x)L1 ≤ c γ1 L∞ γ2 M (W1m →W1m ) , 

completes the proof.

The following assertion resulting from Theorem 6.3.2 involves the norm ||| · ||| defined by (5.4.4) and possessing the property (5.4.5). Corollary 6.3.2. Let m ≥ l ≥ 0. The maximal Banach algebra in M (W1m → l with finite norm W1l ), consists of functions γ ∈ W1,loc sup rm−n |||γ; Br (x)|||W1l + γL∞ .

(6.3.21)

x∈Rn r∈(0,1)

In the case m ≥ n the norm (6.3.21) can be simplified as sup |||γ; B1 (x)|||W1l + γL∞ .

x∈Rn

6.4 Maximal Algebra in Spaces of Bessel Potentials 6.4.1 Pointwise Inequalities Involving the Strichartz Function We start with an inequality similar to (6.2.16).

228

6 Maximal Algebras in Spaces of Multipliers

Lemma 6.4.1. Let k and r be an integer and noninteger, respectively, with [r] 0 < k < r and let ϕ ∈ Wp,loc . There exists a positive constant c = c(k, r, n) such that r−k k (6.4.1) |∇k ϕ(x)| ≤ c (Mϕ(x)) r (Sr ϕ(x)) r for almost all x ∈ Rn . Proof. We use the inequality 
 |∇k ϕ(0)| ≤ c |ϕ(y)|dy + B1

B1

|∇[r] ϕ(y)|



dy |y|n−[r]+k

(see (6.2.5)). Clearly, the right-hand side is majorized by 
 dy c Mϕ(0)+ |∇[r] ϕ(y) − ∇[r] ϕ(0)| n−[r]+k + |∇[r] ϕ(0)| . |y| B1

(6.4.2)

(6.4.3)

Using the notation ψ(y) = |∇[r] ϕ(y) − ∇[r] ϕ(0)|, we find that the second term in (6.4.3) is equal to  1   −n+[r]−k−1 (n − [r] − k) t ψ(z)dzdt + Bt

0

ψ(z)dz.

B1

This sum is dominated by 
 c Sr ϕ(0) + (2 − |z|)ψ(z)dz . B2

We have

 B2


 ≤c

 (2 − |z|)ψ(z)dz =

Bt

2 1/2 ψ(z)dz dt ≤ c Sr ϕ(0).

(6.4.4)

  |∇k ϕ(0)| ≤ c Mϕ(0) + Sr ϕ(0) + |∇[r] ϕ(0)| .

(6.4.5)

2

t−2n−1−2{r}



Bt

0

Therefore,

 ψ(z)dzdt

0



2

In order to estimate the third term in (6.4.5), we use the identity  η(y)∇[r] ϕ(y)dy ∇[r] ϕ(0) = B1    η(y) ∇[r] ϕ(0) − ∇[r] ϕ(y) dy, (6.4.6) + B1

where η ∈ C0∞ (B1 ) is such that  η(y)dy = 1. B1

6.4 Maximal Algebra in Spaces of Bessel Potentials

229

Clearly, the first term on the right-hand side of (6.4.6) is majorized by c Mϕ(0). The second term is estimated by  c (2 − |z|) ψ(z)dz B2

and does not exceed c Sr ϕ(0) by (6.4.4). Therefore,   |∇[r] ϕ(0)| ≤ c Mϕ(0) + Sr ϕ(0) .

(6.4.7)

Combining (6.4.7) with (6.4.5), we arrive at   |∇k ϕ(0)| ≤ c Mϕ(0) + Sr ϕ(0) . Now we obtain by dilation x → x/ρ that   |∇k ϕ(0)| ≤ c ρ−k Mϕ(0) + ρr−k Sr ϕ(0) . 

The result follows by minimization of the right-hand side in ρ.

Lemma 6.4.2. Let k and l be an integer and noninteger, respectively, with [l] 0 < k < l, and let ϕ ∈ Wp,loc . There exists a positive constant c = c(k, l, n) such that k l−k (6.4.8) Sl−k ϕ(x) ≤ c ϕLl ∞ (Sl ϕ(x)) l for almost all x ∈ Rn . Proof. Clearly,  2
 0

B1

2     | ∇[l]−k ϕ (θy) − ∇[l]−k ϕ (0)|dθ

 2
   ∇[l]−k ϕ (θy) − ≤ 0

B1

 {α:|α|≤k−1}

+

k−1 

dy y 1+2{l}

 2 dy yα θα  α D ∇[l]−k ϕ (0) dθ α! y 1+2{l}

|∇[l]−i ϕ(0)|2 .

(6.4.9)

i=0

The difference in the integral over B1 on the right-hand side is equal to  1      y α θ α
 α D ∇[l]−k ϕ (τ θy) − Dα ∇[l]−k ϕ (0) (1 − τ )k−1 dτ. k α! 0 {α:|α|=k}

Using this and Minkowski’s inequality, we see that the first term on the righthand side of (6.4.9) is dominated by 2 dy 1/2 2
 1
 ∞
 |(∇[l] ϕ)(τ θy) − (∇[l] ϕ)(0)|dθ dτ c y 1+2{l} 0 0 B1 = c (Sl ϕ(0))2 .

(6.4.10)

230

6 Maximal Algebras in Spaces of Multipliers

By Lemma 6.4.1, {l}+i

|∇[l]−i ϕ(0)| ≤ c ϕL∞l (Sl ϕ(0))

[l]−i l

, i = 0, . . . , [l],

(6.4.11)

which together with (6.4.9) and (6.4.10) gives  2
 2 dy |(∇[l]−k ϕ)(θy) − −(∇[l]−k ϕ)(0)|dθ y 1+{l} 0 B1  2 ≤ c ϕL∞ + Sl ϕ(0) . Obviously, 

∞
 B1

2

|(∇[l]−k ϕ)(θy) − (∇[l]−k ϕ)(0)|dθ


≤ c |(∇[l]−k ϕ)(0)|2 +

∞


2

|∇[l]−k ϕ(z)|dz

2 dy y 1+{l}

2

By

(6.4.12)



dy y 1+2{l}+2n

.

(6.4.13)

The second term on the right-hand side does not exceed  ∞
  2 dy dξ |∇[l]−k ϕ(z)|dz . 1+2{l}+2n y 2 By+1 B1 (ξ) Since by (6.2.5) 



B1 (ξ)

|∇[l]−k ϕ(z)|dz ≤ c

B1 (ξ)



 |∇[l] ϕ(z)| + |ϕ(z)| dz,

the expression (6.4.14) is dominated by 
 ∞
 2 c dξ |∇[l] ϕ(z)|dz 2


 ≤c

By+1 ∞


B1 (ξ)

|∇[l] ϕ(z)|dz

2

dy y 1+2{l}+2n

dy

y 1+2{l}+2n 2  ≤ c Sl ϕ(0) + |∇[l] ϕ(0)| + ϕL∞ . 2

By+2

+ ϕ2L∞

+ ϕ2L∞

Hence, we find by (6.4.13) and (6.4.11) that  ∞
 2 |(∇[l]−k ϕ)(θy) − (∇[l]−k ϕ)(0)|dθ 2

(6.4.14)

B1





dy y 1+2{l}

 2 ≤ c ϕL∞ + Sl ϕ(0) . Using (6.4.12), we obtain   Sl−k ϕ(x) ≤ c ϕL∞ + Sl ϕ(x) . The result follows by dilation as in Lemma 6.4.1.



6.4 Maximal Algebra in Spaces of Bessel Potentials

231

6.4.2 Banach Algebra Am,l p Theorem 6.4.1. The maximal Banach algebra Am,l imbedded into M (Hpm → p l Hp ), m ≥ l, is isomorphic to the space M (Hpm → Hpl ) ∩ L∞ .

(6.4.15)

The estimate  γ1 γ2 M (Hpm →Hpl ) ≤ c γ1 L∞ γ2 M (Hpm →Hpl )  + γ2 L∞ γ1 M (Hpm →Hpl ) holds. Proof. Let Am,l be a subset of M (Hpm → Hpl ) and let γ ∈ Am,l p p . The inequality γL∞ ≤ γAm,l p

(6.4.16)

is proved in the same way as (6.3.2). Suppose that γ1 and γ2 belong to the space (6.4.15). For any u ∈ Hpm , 

Sl (γ1 γ2 u)Lp = S{l} ∇[l] (γ1 γ2 u)Lp ≤ c

  S{l} Dα γ1 Dβ γ2 Dσ u Lp

|α|+|β|+|σ|=[l]

≤c





 Aα,β,σ + Bα,β,σ + Cα,β,σ ,

|α|+|β|+|σ|=[l]

where Aα,β,σ = (Dα γ1 ) (Dβ γ2 ) S{l} Dσ uLp ,

Cα,β,σ =




(6.4.17)

Bα,β,σ = (Dα γ1 ) (S{l} Dβ γ2 ) Dσ uLp , (6.4.18)
 ∞  |Dβ γ2 (x + θy)| |Dσ u(x + θy) |Dα γ1 (x)|p B1

0

−Dσ u(x)|dθ

2

dy y 1+2{l}

1/p

p/2 dx

.

(6.4.19)

Applying Lemma 6.4.1, we obtain 1−

|α|

1−

|β|

Aα,β,σ ≤ cγ1 L∞[l]−|σ| γ2 L∞[l]−|σ|  |α|   |β|  × M∇[l]−|σ| γ1 [l]−|σ| M∇[l]−|σ| γ2 [l]−|σ| S{l}+|σ| uLp .

(6.4.20)

By H¨older’s inequality the last norm is dominated by |α| |β|     c  M∇[l]−|σ| γ1 S{l}+|σ| uL[l]−|σ|  M∇[l]−|σ| γ2 S{l}+|σ| uL[l]−|σ| . p p

(6.4.21)

232

6 Maximal Algebras in Spaces of Multipliers

Since by Minkowski’s inequality S{l} v ≤ Λ|σ|+{l}−m S{l} Λm−|σ|−{l} v, m−|σ|−{l}

it follows that for γ ∈ M (Hp

→ Lp )

γS{l} vLp ≤ γM (Hpm−|σ|−{l} →Lp ) Λ|σ|+{l}−m S{l} Λm−|σ|−{l} v||Hpm−|σ|−{l} ≤ c γM (Hpm−|σ|−{l} →Lp ) vHpm−|σ| . Putting here γ = M∇[l]−|σ| γi , i = 1, 2, and v = ∇|σ| u, we find that (6.4.21) does not exceed |α|

c M∇[l]−|σ| γ1  [l]−|σ|m−|σ|−{l} M (Hp

|α|

→Lp )

[l]−|σ| uH m p

→Lp )

[l]−|σ| uH . m p

|β|

×M∇[l]−|σ| γ2  [l]−|σ|m−|σ|−{l} M (Hp

|β|

(6.4.22)

We use (6.3.5) and the estimate ∇j γM (Hpm−l+j →Lp ) ≤ c γM (Hpm →Hpl )

(6.4.23)

(see Corollary 3.2.1). Then, using the equality |α| + |β| + |σ| = [l], we obtain that (6.4.22) does not exceed |α|

|β|

|α|+|β| |α|+|β| c γ1 M γ2 M uHpm . (H m →H l ) (H m →H l ) p

p

p

(6.4.24)

p

Now, by (6.4.20) and H¨ older’s inequality,   Aα,β,σ ≤ c γ1 L∞ γ2 M (Hpm →Hpl ) +γ2 L∞ γ1 M (Hpm →Hpl ) uHpm . (6.4.25) To estimate Bα,β,σ , defined by (6.4.18), we apply Lemma 6.4.1 to the older’s function Dα γ1 and Lemma 6.4.2 to the function S{l} Dβ γ2 . Then, by H¨ inequality,  |α|   |β|+{l} 1− |α| 1− |β|+{l}  Bα,β,σ ≤ c γ1 L∞l−|σ| γ2 L∞ l−|σ|  Sl−|σ| γ1 l−|σ| Sl−|σ| γ2 l−|σ| Dσ uLp |α|   σ |β|+{l}  1− |α| 1− |β|+{l}  D uLl−|σ| ≤ c γ1 L∞l−|σ| γ2 L∞ l−|σ|  Sl−|σ| γ1 Dσ uLl−|σ|  S γ . 2 l−|σ| p p

By Lemma 3.2.8, we have for i = 1, 2    Sl−|σ| γi Dσ uLp ≤ Sl−|σ| γi M (Hpm−|σ| →Lp ) uHpm ≤ c γi M (Hpm →Hpl ) uHpm . Hence Bα,β,σ has the same majorant (6.4.25) as Aα,β,σ .

6.5 Imbeddings of Maximal Algebras

233

In order to estimate Cα,β,σ defined by (6.4.19), we use Lemma 6.4.1 and get 1−

|α|

1−

|β|

Cα,β,σ ≤ c γ1 L∞[l]−|σ| γ2 L∞[l]−|σ| Kα,β,σ ,


where Kα,β,σ =

p|α|

(M∇[l]−|σ| γ1 (x)) [l]−|σ|

∞
   p|β| 2 M∇[l]−|σ| γ2 (x+θy) [l]−|σ| |∇|σ| u(x+θy)−∇|σ| u(x)|dθ × 0

B1

 p2  p1 dx . 1+2{l} dy

y

By H¨older’s inequality, Kα,β,σ is dominated by (6.4.22) which, as was shown above, has the majorant (6.4.24). Therefore, (6.4.25) holds with Aα,β,σ replaced by Cα,β,σ . This completes the proof.  Corollary 6.4.1. The maximal Banach algebra Am,l in M (Hpm → Hpl ), m ≥ p l l, 1 < p < ∞, is the set of functions γ ∈ Hp,loc such that sup e⊂Rn diam(e)≤1

Sl γ; eLp (Cp,m (e))

1/p

+ γL∞ < ∞.

(6.4.26)

In the case mp > n this condition can be simplified as Sl γLp,unif + γL∞ < ∞.

6.5 Imbeddings of Maximal Algebras In this section we deal with the imbeddings Am,l ⊂ Aµ,λ and Am,l ⊂ Aµ,λ p p p p . We fix an arbitrary µ and find the maximum value of λ for which the imbeddings hold. Since the best value of λ is equal to l when µ ≥ m, we can restrict ourselves to µ < m. The next theorem contains a complete characterization of the imbedding m,l ⊂ Aµ,λ is stated Am,l p p . The corresponding assertion relating the algebras Ap and proved in exactly the same way with Dp,l replaced by Sl in the proof. Theorem 6.5.1. Let m, l, mθ, lθ be nonintegers for any θ ∈ (0, 1), m ≥ l, and let p ∈ (1, ∞). The following imbeddings hold: ⊂ Amθ,lθ , (i) if pm ≤ n, then Am,l p p mθ,min{mθ,l}

(ii) if pmθ > n, then Am,l ⊂ Ap p (iii) if pmθ = n, then

,

Am,l ⊂ Amθ,mθ for mθ < l p p

234

6 Maximal Algebras in Spaces of Multipliers

and ⊂ Amθ,l− for mθ ≥ l Am,l p p with an arbitrary small > 0, (iv) if pmθ < n < pm, then ⊂ Amθ,mθ Am,l p p

for pl > n

and ⊂ Amθ,mθlp/n for pl ≤ n. Am,l p p All these imbeddings are best possible. continuously maps Lp to Lp Proof. (i) Since the multiplication by γ ∈ Am,l p m l m,l mθ,lθ and Wp to Wp , the imbedding Ap ⊂ Ap results by complex interpolation (see [Tr4], Sec. 2.4.7). In the cases (ii)–(iv) we have pm > n. Thus, by Corollary 6.2.2, l = (Wp,unif ∩ L∞ ). Am,l p λ (ii) Since pmθ > n, Corollary 6.2.2 implies that Amθ,λ = (Wp,unif ∩ L∞ ) p min{mθ,l}

l for λ ≤ mθ. The result follows from the imbedding Wp,unif ⊂ Wp,unif . (iii) If pmθ = n and mθ < l, then pl > n and by Remark 4.3.3 we have l = M Wpl . This last space is imbedded into M Wpmθ = Amθ,mθ by the Wp,unif p mθ,mθ complex interpolation between Wpl and Lp . Hence Am,l ⊂ A . p p can be Let pmθ = n and mθ ≥ l. By Corollary 6.2.2, a norm in Amθ,l− p given by Dp,l− γ; eLp + γL∞ . sup n (Cp,mθ (e))1/p e⊂R diam(e)≤1

Since (3.1.6) shows for pmθ = n that Cp,mθ (e) ≥ c (mesn e)/n with an arbitrary > 0, it follows that   ≤ c γW l− + γL∞ γAmθ,l− p q,unif

l with q = pn/(n − p ). It remains to use the Sobolev imbedding Wp,unif ⊂ l− Wq,unif . l = Wp,unif = M Wpl . Since (iv) Let pmθ < n < pm and pl > n. Then Am,l p l > mθ, we have . M Wpl ⊂ M Wpmθ = Amθ,mθ p

The result follows. Now let pmθ < n < pm and lp ≤ n. By (3.1.6), we obtain from Corollary 6.2.2 that

6.5 Imbeddings of Maximal Algebras

235

 Dp,mθlp/n γ; eLp + γ L ∞ (mesn e)(n−pmθ)/np e⊂Rn diam(e)≤1   (6.5.1) ≤c γW mθlp/n + γL∞ .

γ; Rn Amθ,mθlp/n ≤c p




sup

n/mθ,unif

Suppose first that {mθlp/n} > 0. Let Hpt (Rn+1 + ) denote the Bessel potential n+1 space of functions defined on R+ = {(x, y) : x ∈ Rn , y > 0}. The space t+1/p Wpt (Rn ) is the space of traces on Rn of functions in Hp (Rn+1 + ), where {t} > 0 and p ∈ (1, ∞). This and the Gagliardo-Nirenberg type inequality pmθ/n

n+1 Γ ; Rn+1 + H mθ(lp+1)/n ≤ c Γ ; R+ 

l+1/p

Hp,unif

n/mθ,unif

1−pmθ/n

Γ ; Rn+1 + L∞

(see [AF], Lemma 3.4) imply that the right-hand side of (6.5.1) is dominated by   pmθ/n 1−pmθ/n γL∞ + γL∞ ≤ c γAm,l . c γW l p p,unif

mθ,mθlp/n

Hence the imbedding Am,l ⊂ Ap is valid. p To obtain the same imbedding for integer mθlp/n we use H¨older’s inequality and the estimate 1−k/l

|∇k u(x)| ≤ c uL∞ (Dp,l u(x))k/l , k < l,

(6.5.2)

valid by Corollary 6.2.1. Then the right-hand side of (6.5.1) is dominated by pmθ/n

γW l

p,unif

1−pmθ/n

γ; L∞

.

⊂ Amθ,λ with λ given in (i)–(iv) We now show that the imbedding Am,l p p cannot be improved. Let γµ (x) = exp(i|x|−µ )

(6.5.3)

with µ > 0. From the equivalence relations |∇[l] γµ (x)| ∼ |x|−[l](µ+1) and |∇[l] γµ (x + h) − ∇[l] γµ (x)| ∼

min{|h|, |x|1+µ } , |x|([l]+1)(1+µ)

where |x| is sufficiently small, it follows that Dp,l γµ (x) ∼ |x|−l(µ+1)

(6.5.4)

for |x| < 1. Furthermore, Dp,l γµ (x) is bounded for |x| ≥ 1. Now, by Corollary 6.3.1,

236

6 Maximal Algebras in Spaces of Multipliers

γµ ∈ Am,l p

⎧ ⎪ ⎨(µ + 1)l ≤ m if pm < n, ⇐⇒ (µ + 1)l < m if pm = n, ⎪ ⎩ p(µ + 1)l < n if pm > n.

(6.5.5)

We conclude by (6.5.5) that in the case pm ≤ n γ ∈ Am,l ⇐⇒ γ ∈ Amθ,lθ for θ ∈ (0, 1) p p which shows that the imbedding (i) is sharp. Since the imbedding (ii) is equivalent to min{mθ,l}

l Wp,unif ∩ L∞ ⊂ Wp,unif

∩ L∞

and, obviously, λ ≤ mθ in Am,l ⊂ Amθ,λ , it follows that (ii) cannot be imp p proved. We turn to the imbeddings (iii). The optimality of the first one (coris not responding to mθ < l) is obvious. Let mθ ≥ l. We show that Am,l p . imbedded into Amθ,l p Let µ ≥ 0 and pl(µ + 1) = n. We introduce the function   Γµ,δ (x) = η(x)exp i|x|−µ (log |x|−1 )−δ , where δ > −1 and η is a function in C0∞ (Rn ) with support in a small neighbourhood of the origin, equal to 1 near the origin. For µ > 0 direct calculations imply that |∇[l] Γµ,δ (x)| ∼ |x|−[l](µ+1) (log |x|−1 )−[l]δ and |∇[l] Γµ,δ (x + h) − ∇[l] Γµ,δ (x)| ∼

min{|h|, |x|1+µ (log |x|−1 )δ } , |x|([l]+1)(1+µ) (log |x|−1 )([l]+1)δ

where |x| is sufficiently small. Therefore, for small |x| Dp,l Γµ,δ (x) ∼ |x|−l(µ+1) (log |x|−1 )−lδ .

(6.5.6)

Analogously, |∇[l] Γ0,δ (x)| ∼ |x|−[l] (log |x|−1 )−[l](δ+1) and |∇[l] Γ0,δ (x + h) − ∇[l] Γ0,δ (x)| ∼

min{|h|, |x|(log |x|−1 )δ+1 } |x|[l]+1 (log |x|−1 )[l]+1)(δ+1)

for small |x|. Hence, Dp,l Γ0,δ (x) ∼ |x|−l (log |x|−1 )−l(δ+1) .

(6.5.7)

6.5 Imbeddings of Maximal Algebras

237

Now it is straightforward that l Γµ,δ ∈ Am,l = (Wp,unif ∩ L∞ ) p

if and only if plδ > 1 for µ > 0, and pl(δ + 1) > 1 for µ = 0. On the other hand, by Corollary 6.2.2 and Proposition 3.1.4, Γµ,δ Amθ,l ≥ c (log ρ−1 )(p−1)/p Dp,l Γµ,δ ; Bρ Lp p for small ρ > 0. Applying (6.5.6) and (6.5.7), we obtain Γµ,δ Amθ,l ≥ c (log ρ−1 )(p−1)/p+(1−plδ)/p p and

for µ > 0

≥ c (log ρ−1 )(p−1)/p+(1−pl(δ+1))/p . Γ0,δ Amθ,l p

This, obviously, implies that Γµ,δ ∈ Am,l and Γµ,δ ∈ Amθ,l if 1 > lδ > 1/p for p p m,l mθ,l if 1 > l(δ + 1) > 1/p. The result µ > 0, and Γ0,δ ∈ Ap and Γ0,δ ∈ Ap follows. We pass to (iv). It suffices to consider only the case pl < n. Assume that Am,l ⊂ Amθ,λ p p

with

λ = mθpl/n(1 − )

for some > 0. We choose µ to satisfy pl(µ + 1) > n(1 − ). Then the function l γµ , introduced by (6.5.3), belongs to Wp,unif = Am,l p . On the other hand, by Corollary 6.2.2 and (6.5.4), γµ Amθ,λ ≥ c ρmθ−n/p Dp,λ γµ ; Bρ Lp ≥ c ρmθ−λ(µ+1) p for ρ < 1. Since mθ − λ(µ + 1) = mθ(1 − pl(µ + 1)/n(1 − )) < 0, we have γµ ∈ Amθ,λ . The proof is complete. p



The following assertion is an analogue of Theorem 6.5.1 for p = 1. Theorem 6.5.2. Let m, l, mθ, lθ be nonintegers for any θ ∈ (0, 1), m ≥ l. The following imbeddings are valid: (i) if m < n, then Am,l ⊂ Amθ,lθ , 1 1 mθ,min{mθ,l} m,l , (ii) if mθ > n, then A1 ⊂ A1 (iii) if mθ < n < m, then Am,l ⊂ Amθ,mθ 1 1 and

mθ,mθl/n

⊂ A1 Am,l 1

All these imbeddings are best possible.

for l > n for l < n.

238

6 Maximal Algebras in Spaces of Multipliers

Proof. (i) By (6.2.17) and H¨ older’s inequality, for all θ ∈ (0, 1) (r)

sup rmθ−n D1,lθ γ; Br (x)L1

x∈Rn r∈(0,1)

(r)

m−n ≤ c γ1−θ D1,l γ; Br (x)L1 )θ . L∞ sup (r x∈ Rn r∈(0,1)

Hence Am,l ⊂ Amθ,lθ and 1 1 θ γAmθ,lθ ≤ c γ1−θ L∞ γAm,l . 1

1

In the cases (ii), (iii) we have mθ ≥ n. Therefore, by Corollary 6.2.2, l Am,l = W1,unif ∩ L∞ . 1 (ii) Since mθ > n, Corollary 6.2.2 implies that λ = W1,unif ∩ L∞ Amθ,λ 1

for λ ≤ mθ.

The result follows from the imbedding min{mθ,l}

l W1,unif ⊂ W1,unif

.

(iii) Let mθ < n < m and l > n. Then l Am,l = W1,unif = M W1l . 1

Since l > mθ, we have M W1l ⊂ M W1mθ = Amθ,mθ . 1 The result follows. Now let mθ < n < m and l < n. By Corollary 6.3.2, 
(r) γAmθ,mθl/n ≤ c sup rmθ−n D1,mθl/n γ; Br (x)L1 + γL∞ . 1

(6.5.8)

x∈Rn r∈(0,1)

mθ,mθl/n

If mθl/n is integer, then the imbedding Am,l ⊂ A1 follows by the same 1 argument as for p > 1 (see part (iv) of Theorem 6.5.1). Suppose that mθl/n is a noninteger and {mθl/n} ≥ {l}. Then, by Lemma 6.2.3 and H¨ older’s inequality, (6.5.8) is dominated by mθ/n

c γW l

1,unif

1−mθ/n

γL∞

,

mθ,mθl/n

and we arrive at the imbedding Am,l ⊂ A1 1 Let {mθl/n} < {l}. Then γW mθl/n

n/mθ,unif

mθ/n

≤ c γW l

1,unif

. 1−mθ/n

γL∞

.

6.5 Imbeddings of Maximal Algebras

239

One can show that the imbedding Am,l ⊂ Amθ,λ with λ given in (i)–(iii) 1 1 cannot be improved by using the same argument as in the proof of the sharpness of (i), (ii), and (iv) in Theorem 6.5.1 with p = 1.  Remark 6.5.1. The conditions on parameters of integrability and smoothness for concrete function spaces to be multiplication algebras, were studied by many authors, see Strichartz [Str], Peetre [Pe2], Hertz [Her], Bennet and Gilbert [BG], Johnson [Jo], Triebel [Tr1], [Tr2], [Tr4], Zolesio [Zo], Bliev [Bl1], [Bl2], Kalyabin [K1]–[K3], Marschall [Mar3], Benchekroun and Benkirane [BB], Ali Mehmeti and Nicaise [AN], Runst [Ru]. Considerable attention was paid to the description of the range of a product of two or more functions in Sobolev type spaces (see Maz’ya [Maz6], Amann [Am], Hanouzet [Ha], Valent [Va], Franke [Fr] Johnsen [Jo], Miyachi [Mi], Runst and Sickel [RS], Sickel [Sic1], Sickel and Triebel [ST], Sickel and Youssfi [SY], Dacorogna and Moser [DM1], Ye [Ye], Runst and Youssfi [RY], Youssfi [Yo], Drihem and Moussai [DM2] et al.

7 Essential Norm and Compactness of Multipliers

In this chapter we study elements of the space M (Wpm → Wpl ), where p ∈ [1, ∞), and m and l are arbitrary, integer and noninteger, with m ≥ l ≥ 0. As usual, we omit Rn in notations of spaces, norms, and integrals. By ess γM (Wpm →Wpl ) we denote the essential norm of the operator of multiplication by γ ∈ M (Wpm → Wpl ), that is, inf γ − T Wpm →Wpl ,

{T }

where {T } is the set of compact operators Wpm → Wpl . As before, (Dp,l γ)(x) = |∇l γ(x)| for integer l and
 |∆ ∇ u(x)|p 1/p h [l] (Dp,l γ)(x) = dh |h|n+p{l} for noninteger l. The main results are sharp two-sided estimates for ess γM (Wpm →Wpl ) . We formulate the main result concerning m > l. Theorem 7.0.3. Let γ ∈ M (Wpm → Wpl ), where m > l ≥ 0. (i) If p ∈ (1, ∞) and mp ≤ n, then ess γM (Wpm →Wpl )  Dp,l γ; eLp m−l− n p γ; B (x) ρ Lp 1 + sup ρ   δ→0 {e:d(e)≤δ} x∈Rn Cp,m (e) p ρ≤δ
 Dp,l γ; eLp sup + lim + sup γ; B1 (x)Lp , 1 r→∞ e⊂Rn \Br (C p x∈Rn \Br p,m (e)) d(e)≤1 ∼ lim




sup

V.G. Maz’ya, T.O. Shaposhnikova, Theory of Sobolev Multipliers, Grundlehren der mathematischen Wissenschaften 337, c Springer-Verlag Berlin Hiedelberg 2009


(7.0.1) 241

242

7 Essential Norm and Compactness of Multipliers

where d(e) is the diameter of a compact set e ⊂ Rn . (ii) If m < n, then ess γM (W1m →W1l )   ∼ lim sup δ m−n sup D1,l γ; Bδ (x)L1 + δ −l γ; Bδ (x)L1 x∈Rn

δ→0

+ lim sup |x|→∞



 sup rm−n D1,l γ; Br (x)L1 + γ; B1 (x)L1 .

(7.0.2)

r∈(0,1)

(iii) If mp > n, p ∈ (1, ∞), then ess γM (Wpm →Wpl ) ∼ lim sup γ; B1 (x)Wpl .

(7.0.3)

|x|→∞

Theorem 7.0.4. Let γ ∈ M Wpl . (i) If p ∈ (1, ∞) and lp ≤ n, then ess γM Wpl ∼ lim

sup

δ→0 {e:d(e)≤δ}

+ lim

r→∞

sup e⊂Rn \Br d(e)≤1

Dp,l γ; eLp  1 Cp,l (e) p

Dp,l γ; eLp 1

(Cp,l (e)) p

+ γL∞ .

(7.0.4)

(ii) If l < n, then ess γM W1l ∼ lim sup δ l−n sup D1,l γ; Bδ (x)L1 δ→0

x∈Rn

+ lim sup sup rl−n D1,l γ; Br (x)L1 + γL∞ . |x|→∞

r∈(0,1)

(iii) If m ≥ n, then ess γM W1l ∼ lim sup γ; B1 (x)Wpl + γL∞ .

(7.0.5)

|x|→∞

Clearly, if multipliers have compact supports, the above equivalence relations for the essential norm are simplified, since all terms containing either r → ∞ or |x| → ∞ vanish. ˚ (Wpm → As simple corollaries we obtain characterizations of the space M Wpl ), m > l, of compact multipliers. We note also that Sect. 7.2.7 contains one-sided estimates for the essential norm of a multiplier which do not involve capacities.

7.1 Auxiliary Assertions

243

7.1 Auxiliary Assertions In this chapter we use the following cutoff functions. Definition 7.1.1. Let x ∈ Rn , δ ∈ (0, 1), and let η be a function in C0∞ [0, 2), equal to 1 on [0, 1]. Furthermore, we assume that 0 ≤ η ≤ 1. We set Rn  y → ηδ,x (y) = η


|y − x|  δ

.

Definition 7.1.2. Let η be the same as in Definition 7.1.1. We put
2 log δ  Rn  y → µδ,x (y) = η , δ ∈ (0, 1/2). log |y − x| We also adopt the notation ηδ = ηδ,0 and µδ = µδ,0 . Definition 7.1.3. Let ζr (y) = ζ(y/r), where r > 1, ζ ∈ C ∞ (Rn ), ζ = 0 for y ∈ B1 and ζ(y) = 1 for y ∈ Rn \B2 . Furthermore, let 0 ≤ ζ ≤ 1. As usual, by Wpl (Br ) we denote the space of functions with the finite norm ⎧ l  ⎪ ⎪ ⎪ ∇j u; BLp for {l} = 0, ⎪ ⎪ ⎪ ⎪ ⎨j=0 [l]
  1/p u; Br Wpl =  dxdy p ⎪ |∇ u(x)−∇ u(y)| +u; Br Wp[l] ⎪ j j ⎪ ⎪ |x − y|n+p{l} Br Br ⎪ j=0 ⎪ ⎪ ⎩ for {l} > 0. We introduce one more norm in Wpl (Br ) depending on r ∈ (0, 1). Namely, we set ⎧ l  ⎪ ⎪ ⎪ rj−l ∇j u; Br Lp for {l} = 0, ⎪ ⎪ ⎪ ⎪ j=0 ⎪ ⎪ ⎪ [l] ⎨
  1/p dxdy rj−[l] |∇j u(x)−∇j u(y)|p |||u; Br |||Wpl = ⎪ |x − y|n+p{l} Br Br ⎪ j=0 ⎪ ⎪ ⎪ [l] ⎪  ⎪ ⎪ ⎪ + rj−l ∇j u; Br Lp for {l} > 0. ⎪ ⎩ j=0

It is clear that the last norm is invariant under dilation. We present some properties of the norm |||u; Br |||Wpl which will be used henceforth. Lemma 7.1.1. If l is a positive noninteger, then Dp,l u; Br Lp ≤ c sup |||u; Br |||Wpl . x∈Rn

244

7 Essential Norm and Compactness of Multipliers

Proof. It suffices to estimate 
 Br/2 (z)

Rn \Br/2 (z)

|∇[l] u(x) − ∇[l] u(y)|p

1/p dxdy , |x − y|n+p{l}

where z is an arbitrary point of the ball Br . This value does not exceed 
 1/p dx |∇[l] u(y)|p dy n+p{l} Br/2 (z) Rn \Br/2 (z) |x − y| 


 +

|∇[l] u(x)| dx p

Rn \Br/2 (z)

≤ c r−{l} ∇[l] u; Br (z)Lp


 +c

Br/2 (z)

1/p dy |x − y|n+p{l}

Rn \Br/2 (z)

|∇[l] u(x)|p dx 1/p . |x − z|n+p{l}

The second term on the right-hand side does not exceed  1/p
 1 p |∇ u(x)| dxdξ , c [l] n+p{l} Rn \Br/2 (z) |x − ξ| Br (ξ) which is dominated by

c r−{l} sup |||u; Br |||Wpl . x∈Rn



The proof is complete.

Next we formulate three well-known properties of the norm |||u; Br |||Wpm , leaving their proof to the reader as an exercise. Lemma 7.1.2. If ϕ ∈ C0∞ (Br ) and |∇k ϕ| ≤ c r−k , k = 0, 1, . . . , m, then for all u ∈ Wpm (Br ) ϕ u; Rn Wpm ≤ c |||u; Br |||Wpm

for

r≤1

ϕ u; Rn Wpm ≤ c u; Br Wpk

for

r > 1.

and Lemma 7.1.3. Let u ∈ Wpm (Bδ ). There exists a polynomial P of degree [m] of the form 
y 
x β −n δ ϕβ P (u; x) = u(y) dy, δ δ Bδ β

where ϕ ∈

C0∞ (B1 ),

and such that   |∇[m] u(x)−∇[m] u(y)|p |||u − P (u; ·); Bδ |||Wpm ≤ c Bδ Bδ

dxdy . |x − y|n+p{m}

7.1 Auxiliary Assertions

245

Lemma 7.1.4. The inequality s/k

1−s/k

|||u; Br |||Wps ≤ c |||u; Br |||W k u; Br Lp

,

0 < s < k,

(7.1.1)

p

holds. From (7.1.1) we immediately obtain Corollary 7.1.1. If k is a noninteger, then 1/p
  dxdy |||u; Br |||Wpk ∼ |∇[k] u(x)−∇[k] u(y)|p + r−k u; Br Lp . |x − y|n+p{k} Br Br If k is an integer, then |||u; Br |||Wpk ∼ ∇k u; Br Lp + r−k u; Br Lp . Using the Hardy-type inequality u |x|−k Lp ≤ c uWpk ,

kp < n,

(which, in particular, easily follows from (4.3.17)), we deduce Lemma 7.1.5. If kp < n, then |||u; Br |||Wpk ≤ c uWpk with a constant c independent of r. Next we prove some technical lemmas. Lemma 7.1.6. Let ϕ ∈ C0∞ (Bδ ) with δ < 1, and let |∇k ϕ| ≤ c δ −k , k = 0, 1, . . . , [l] + 1. Then ϕM Wpl ≤ c, where lp < n and p ∈ [1, ∞), or l = n and p = 1. Proof. Let u ∈ C0∞ (Rn ). According to Lemma 7.1.2 and Corollary 7.1.1, ϕ uWpl ≤ c |||u; B2δ |||Wpl

 ≤c



B2δ B2δ

|∇[l] u(x)−∇[l] u(y)|p

 p1  dx dy +u; B2δ L pn . (7.1.2) n+p{l} n−lp |x − y|

Now the result follows from Proposition 4.2.5.



Lemma 7.1.7. If u ∈ C0∞ and lp < n, then sup ηδ,x uWpl → 0

x∈Rn

as δ → 0,

where ηδ,x is the function in Definition 7.1.1.

(7.1.3)

246

7 Essential Norm and Compactness of Multipliers

Proof. This assertion follows from (7.1.2). Lemma 7.1.8. Let {l} > 0. Then |∇j µδ (z)| ≤ c | log |z| |−1 |z|−j and

 Bδ

|∇j µδ (z) − ∇j µδ (y)|p dy ≤ cj | log |z| |−p |z|−p({l}+j) , |z − y|n+p{l}

(7.1.4)

(7.1.5)

where z ∈ Bδ , j = 0, 1, . . . Proof. Estimate (7.1.4) is obvious. We prove (7.1.5). Since |Dα µδ (z)| ≤ |z|−|α|

|α|  k=1

σk


2 log δ  log |z|

(2 log δ)−k ,

with σk ∈ C0∞ (−1, 1), it follows that ⎧ ⎪ c | log δ|−1 |z − y| |z|−j−1 ⎪ ⎪ ⎪ ⎪ ⎪ if |z|/2 ≤ |y| ≤ |z|; ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨c | log δ|−1 max{|z|, |y|}−j if j > 0 |∇j µδ (z) − ∇j µδ (y)| ≤ and either |y| < |z|/2 or |z| < |y|/2; ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ |z| −1 ⎪ log if j = 0 c | log δ| ⎪ ⎪ |y| ⎪ ⎪ ⎩and either |y| < |z|/2 or |z| < |y|/2.

(7.1.6)

These estimates imply that  |∇j µδ (z) − ∇j µδ (y)|p dy ≤ c | log δ|−p |z|−p({l}+j) |z − y|n+p{l} Bδ which is equivalent to (7.1.4) for z ∈ Bδ \Bδ3 by the second and the third estimates in (7.1.6). Let z ∈ Bδ3 . Then   |∇j µδ (z) − ∇j µδ (y)|p |δ0j − ∇j µδ (y)|p dy ≤ c dy |z − y|n+p{l} |y|n+{l} Bδ Bδ \Bδ2 ≤ cj | log δ|−p

 Bδ \Bδ2

|y|−jp dy = cj | log δ|−p δ −3p(j+{l}) , |y|n+{l}

where δ0j is the Kroneker delta. Putting here |z| = δ 3 and noting that  t3({l}+j) | log t| increases near t = 0, we arrive at (7.1.5) for z ∈ Bδ3 .

7.1 Auxiliary Assertions

247

Lemma 7.1.9. Let ∇ψ ∈ C0∞ (B1 ) and ψr (y) = ψ(y/r), r > 1. Then ψr M Wpl ≤ 1 + c r−σ , where σ = l if 0 < l < 1 and σ = 1 if l ≥ 1. Proof. The assertion is obvious for integer l. Let {l} > 0. Then ψr uWpl ≤ ψr Dp,l uLp + ψr uLp +



Dα ψr Dβ uWp{l} + ∇[l] u Dp,{l} ψr Lp .

(7.1.7)

|α|+|β|=[l], |α|>0

Note that Dα ψr = r−|α| (Dα ψ)r

and Dp,{l} ψr = r−{l} (Dp,{l} ψ)r .

Since the function (Dα ψ)r is bounded together with all its derivatives and since (Dp,{l} ψ)r is uniformly bounded with respect to r, it follows that Dα ψr Dβ uWp{l} ≤ c r−|α| uWp|β|+{l} and

∇[l] u Dp,{l} ψr Lp ≤ c r−{l} uWpl .

Combining (7.1.7) and (7.1.8), we complete the proof.

(7.1.8) 

Lemma 7.1.10. Let ψ and ψr be the functions defined in Lemma 7.1.9. Further, let ψ = 0 in the ball B1/2 . Then, for any u ∈ Wpl , lim ψr uWpl = 0.

r→∞

Proof. The result follows from the inequality ψr uWpl ≤ ψr Dp,l uLp + ψr uLp + c r−σ uWpl established in the proof of Lemma 7.1.9.



Lemma 7.1.11. Let mp < n, k ∈ [0, m]. Further, let e be a compact subset of the ball Br , r ∈ (0, 1). Then Cp,k (e) ≤ c r(m−k)p Cp,m (e). Proof. Let u ∈ C0∞ with u ≥ 1 on e. We have [Cp,k (e)]1/p ≤ ηr uWpk = Dp,k (ηr u)Lp + ηr uLp .

(7.1.9)

248

7 Essential Norm and Compactness of Multipliers

By inequality (4.2.8), ηr uLp ≤ c rm Dp,m (ηr u)Lp and therefore [Cp,k (e)]1/p ≤ c rm−k ηr uWpm . This and Lemmas 7.1.6 and 7.1.9 imply that [Cp,k (e)]1/p ≤ c1 rm−k uWpm . 

Minimizing the right-hand side, we arrive at (7.1.9).

7.2 Two-Sided Estimates for the Essential Norm. The Case m > l 7.2.1 Estimates Involving Cutoff Functions Lemma 7.2.1. The estimate lim sup ζr γM (Wpm →Wpl ) ≤ c ess γM (Wpm →Wpl ) r→∞

holds, where m ≥ l, p ≥ 1. Proof. Let ε > 0 and let T = T (γ, ε) be a compact operator such that γ − T  ≤ ess γM (Wpm →Wpl ) + ε . Then, for all u ∈ Wpm , γu − T uWpl ≤ ( ess γM (Wpm →Wpl ) + ε)uWpm .

(7.2.1)

Let S be the unit ball in Wpm centered at the origin and let {vk } be a finite ε-net in T S. Without loss of generality we assume that vk ∈ C0∞ . It is clear that ζr vk = 0 for sufficiently large r. Therefore, ζr γuWpl = ζr (γu − vk )Wpl ≤ ζr (γu − T u)Wpl + ζr (T u − vk )Wpl . From this inequality and from Lemma 7.1.9 we obtain ζr γuWpl ≤ c ( ess γM (Wpm →Wpl ) + ε)uWpm . The result follows.



7.2 Two-Sided Estimates for the Essential Norm. The Case m > l

249

Theorem 7.2.1. Let m > 1 and let either lp < n and p > 1, or l ≤ n and p = 1. Then the equivalence relation ess γM (Wpm →Wpl ) ∼ lim sup sup ηδ,x γM (Wpm →Wpl ) δ→0

x∈Rn

+ lim sup ζr γM (Wpm →Wpl )

(7.2.2)

r→∞

holds. Proof. (i) The lower bound for the essential norm. We use the notation introduced in the proof of Lemma 7.2.1. For any u ∈ S, ηδ,x γuWpl ≤ ηδ,x (γu − vk )Wpl + ηδ,x vk Wpl ≤ ηδ,x (γu − T u)Wpl + ηδ,x (T u − vk )Wpl + ηδ,x vk Wpl . From this inequality and Lemma 7.1.6 we get ηδ,x γuWpl ≤ c ( ess γM (Wpm →Wpl ) + 2ε) + ε which, together with reference to Lemma 7.2.1, completes the proof of the lower bound for ess γM (Wpm →Wpl ) . (ii) The upper bound for the essential norm. We choose δ and r so that the following estimates hold: sup η2δ,x γM (Wpm →Wpl ) ≤ lim sup sup ηδ,x γM (Wpm →Wpl ) + ε,

x∈Rn

δ→0

x∈Rn

(7.2.3) ζr γM (Wpm →Wpl ) ≤ lim sup ζr γM (Wpm →Wpl ) + ε . r→∞

(j)

By {Kδ } we denote a finite covering of the ball B2r by open balls with (j) radius δ and centers xj . We can choose the balls Kδ so that the multiplicity (j) of the covering of B2r by the balls K2δ depends only on n. Let {ϕ(j) } be a (j) smooth partition of unity subordinate to {Kδ } and such that |∇k ϕ(j) | ≤ c δ −k ,

k = 0, 1, . . . .

Given any u ∈ Wpm (Bδ ), we use the polynomials introduced in Lemma 7.1.3. (j)

Let P (j) = P (j) (u; ·) be such polynomials constructed for the balls K2δ . Further, let Γ = (1 − ζr )γ and let T∗ be the finite-dimensional operator defined by  ϕ(j) (x)P (j) (u; x) . (7.2.4) (T∗ u)(x) = Γ (x) j

250

7 Essential Norm and Compactness of Multipliers

We have (γ − T∗ )uWpl ≤ (Γ − T∗ )uWpl + ζr γuWpl . Since (Γ − T∗ )u =



(7.2.5)

Γ η2δ,xj ϕ(j) (u − P (j) ),

j

it follows from Corollary 4.2.1 that  Γ η2δ,xj ϕ(j) (u − P (j) )pW l (Γ − T∗ )upW l ≤ p

p

j

≤ c sup Γ η2δ,xj pM (W m →W l ) p

j



p

ϕ(j) (u − P (j) )pWpm .

(7.2.6)

j

Using Lemmas 7.1.2 and 7.1.3, we obtain that the last sum does not exceed  (j) |||u − P (j) ; K2δ |||pWpm ≤ c1 Dp,m upLp . (7.2.7) c j

We further note that, by Lemma 7.1.9, Γ η2δ,xj M (Wpm →Wpl ) ≤ c γ η2δ,xj M (Wpm →Wpl ) . This and inequalities (7.2.3), (7.2.5)–(7.2.7) imply that γ − T∗ Wpm →Wpl

(7.2.8)

  ≤ c lim sup sup ηδ,x γM (Wpm →Wpl ) + lim sup ζr γM (Wpm →Wpl ) + ε . δ→0

x∈Rn

r→∞

 Remark 7.2.1. Relation (7.2.2) fails for lp > n. In fact, let γ = 1. It is clear that 1 ∈ M (Wpm → Wpl ). On the other hand, Remark 4.3.3 implies that lim ηδ,x M (Wpm →Wpl ) = ∞ .

δ→0

7.2.2 Estimate Involving Capacity (The Case mp < n, p > 1) The following theorem presents one more relation for the essential norm. Theorem 7.2.2. If mp < n and p ∈ (1, ∞), then
γ; e Dp,l γ; eLp  Lp ess γM (Wpm →Wpl ) ∼ lim sup + δ→0 {e : d(e)≤δ} [Cp,m−l (e)]1/p [Cp,m (e)]1/p
+ lim

sup

r→∞ {e⊂Rn \B : d(e)≤1} r

Dp,l γ; eLp  γ; eLp . + [Cp,m−l (e)]1/p [Cp,m (e)]1/p

(7.2.9)

7.2 Two-Sided Estimates for the Essential Norm. The Case m > l

251

Proof. We limit consideration to the case of noninteger l, since for integer l the proof is analogous and slightly simpler. (i) The lower bound for the essential norm. We introduce the notation fk (γ; e) =

Dp,k γ; eLp [Cp,m−l+k (e)]1/p

for noninteger k, 0 < k ≤ l, and fk (γ; e) =

∇k γ; eLp [Cp,m−l+k (e)]1/p

for integer k, 0 ≤ k ≤ l. Clearly, for any compact set e with d(e) ≤ δ, f0 (γ; e) ≤ sup

sup f0 (γ; e) ≤ sup

x∈Rn e⊂Bδ (x)

sup f0 (ηδ,x γ; e) .

x∈Rn e⊂Bδ (x)

This, together with Corollary 4.3.1, implies that f0 (γ; e) ≤ c sup ηδ,x γM (Wpm →Wpl ) . x∈Rn

Applying Theorem 7.2.1, we obtain f0 (γ; e) ≤ c ( ess γM (Wpm →Wpl ) + ε).

(7.2.10)

Now we turn to the bound for fl (γ; e). For e ⊂ Bδ (x) we have   |∇[l] (γ(y)η2δ,x (y)) − ∇[l] (γ(z)η2δ,x (z))|p dz Dp,l γ; epLp = dy |y − z|n+p{l} e B2δ (x) 



|∇[l] γ(y) − ∇[l] γ(z)|p

dy

+

Rn \B

e

2δ (x)

dz . |y − z|n+p{l}

Therefore,  Dp,l γ; eLp ≤ c Dp,l (η2δ,x γ); eLp 
 + |∇[l] γ(y)|p dh e


 + Rn \B2δ (x)

Rn \B2δ (x)

 |∇[l] γ(y)| dy p

e

1/p dz |y − z|n+p{l} 1/p  dz . (7.2.11) |y − z|n+p{l}

The second term on the right-hand side does not exceed c1 δ −{l} ∇[l] γ; eLp ≤ c2 δ −{l} ∇[l] (γη2δ,x ); eLp .

(7.2.12)

252

7 Essential Norm and Compactness of Multipliers

Since e ⊂ Bδ (x), the third term is not greater than
 |∇[l] γ(z)|p dz 1/p c (mesn e)1/p n+p{l} Rn \B2δ (x) |z − x| 
 ≤ c1 (mesn e)1/p |z − x|−n−p{l} δ −n Rn \B2δ (x)

Bδ (z)

|∇[l] γ(ξ)|p dξ dz

≤ c2 (mesn e)1/p sup ∇[l] (η2δ,z γ); Bδ (z)Lp δ −{l}−n/p .

1/p

(7.2.13)

z∈Rn

Therefore, 
C  p,m−{l} (e) 1/p fl (γ; e) ≤ c fl (η2δ,x γ; e) + δ −{l} f[l] (η2δ,x γ; e) Cp,m (e)
δ −mp mes e 1/p  n + sup sup f[l] (η2δ,x γ; e) . (7.2.14) Cp,m (e) x∈Rn e⊂Bδ (x) Now from (7.2.14) and Lemma 7.1.11 we obtain fl (γ; e) ≤ c sup

sup [fl (η2δ,x γ; e) + f[l] (η2δ,x γ; e)]

x∈Rn e⊂Bδ (x)

which, together with Proposition 4.3.1 and Theorem 7.2.1, implies that   fl (γ; e) ≤ sup η2δ,x γM (Wpm →Wpl ) ≤ c1 ess γM (Wpm →Wpl ) + ε . (7.2.15) x∈Rn

Combining (7.2.10) and (7.2.16), we arrive at the inequality
γ; e Dp,l γ; eLp  Lp lim sup + δ→0 {e : d(e)≤δ} [Cp,m−l (e)]1/p [Cp,m (e)]1/p ≤ c ess γM (Wpm →Wpl ) .

(7.2.16)

Let e ⊂ Rn \B3r . It is clear that f0 (γ; e) ≤ f0 (ζr γ; e) and, by Theorems 4.1.1 and 7.2.1, the estimate (7.2.10) holds if r is sufficiently large. Let us estimate fl (γ; e). We have   |∇[l] (γ(y)ζr (y)) − ∇[l] (γ(z)ζr (z))|p Dp,l γ; epLp = dy dz |y − z|n+p{l} e Rn \B2r   |∇[l] γ(y) − ∇[l] γ(z)|p + dy dz . |y − z|n+p{l} e B2r Consequently,

 1/p 
 dz |∇[l] γ(y)|p dy Dp,l γ; eLp ≤ c Dp,l (ζr γ); eLp + n+p{l} e B2r |y − z|   
 1/p dy . (7.2.17) + |∇[l] γ(z)|p dz n+p{l} B2r e |y − z|

7.2 Two-Sided Estimates for the Essential Norm. The Case m > l

253

The second term on the right-hand side of (7.2.17) does not exceed 1/p
 5 61/p c r−{l} |∇[l] γ(y)|p dy ≤ c1 r−{l} f[l] (γ; e) Cp,m−{l} (e) e

which, by Lemma 7.1.11 and Theorem 4.1.1, is not greater than c2 r−{l} γM (Wpm →Wpl ) [Cp,m (e)]1/p . Let us get a similar estimate for the third term on the right-hand side of (7.2.17). We have 
 1/p dy |∇[l] γ(z)|p dz n+p{l} B2r e |y − z| ≤ c r−n/p−{l} (mesn e)1/p ∇[l] γ; B2r Lp ≤ c f[l] (γ; B2r )(mesn e)1/p r−m . Therefore, the third term is majorized by c r−m γM (Wpm →Wpl ) [Cp,m (e)]1/p . Finally,   fl (γ; e) ≤ c fl (ζr γ; e) + (r−{l} + r−m )γM (Wpm →Wpl ) .

(7.2.18)

Taking the supremum with respect to e on both sides of this inequality and making r → ∞, we arrive at lim

fl (γ; e) ≤ c lim sup fl (ζr γ; e) .

sup

r→∞ e⊂Rn \B

r→∞

r

e

Combining this estimate with (7.2.10) and Theorem 7.2.1, we conclude that
lim

sup

r→∞ e⊂Rn \B

r

Dp,l γ; eLp  γ; eLp ≤ c ess γM (Wpm →Wpl ) . (7.2.19) + 1/p [Cp,m−l (e)] [Cp,m (e)]1/p

Adding (7.2.16) and (7.2.19), we obtain the required estimate for the essential norm. (ii) The upper bound for the essential norm. Let e be an arbitrary compact set in Rn . We have [l]  Dp,l (ηδ,x γ); eLp ≤ c |∇j (η2δ,x γ)|Dp,l−j ηδ,x ; e Lp j=0

 [l]
  |∇[l]−j (η2δ,x (y)γ(y)) − ∇[l]−j (η2δ,x (z)γ(z))|p 1/p + |∇j ηδ,x (y)|p dy dz |y − z|n+p{l} Rn e j=0  (7.2.20) +Dp,l (ηδ,x η2δ,x γ); e\B4δ (x)Lp .

254

7 Essential Norm and Compactness of Multipliers

The obvious estimate Dp,l−j ηδ,x ≤ c δ j−l and Lemma 7.1.11 imply that |∇j (η2δ,x γ)|Dp,l−j ηδ,x ; eLp ≤ c δ j−l fj (η2δ,x γ; e)[Cp,m−l+j (e)]1/p ≤ cfj (η2δ,x γ; e)[Cp,m (e)]1/p . By Theorem 4.1.1, fj (η2δ,x γ; e) ≤ c η2δ,x γM (Wpm−l+j →Wpj ) .

(7.2.21)

Hence, from Lemma 7.1.6 and Corollary 4.3.7, we obtain fj (η2δ,x γ; e) ≤ c sup ηδ,ξ η2δ,x γM (Wpm−l+j →Wpj ) ξ∈Rn

≤ c sup ηδ,ξ γM (Wpm−l+j →Wpj ) ξ∈Rn

≤ ε sup ηδ,ξ γM (Wpm →Wpl ) + c(ε) sup ηδ,ξ γM (Wpm−l →Lp ) . ξ∈Rn

ξ∈Rn

Thus the first sum in (7.2.20) does not exceed   ε ηδ,x γM (Wpm →Wpl ) + c(ε)ηδ,x γM (Wpm−l →Lp ) [Cp,m (e)]1/p .

(7.2.22)

The j-th term in the second sum on the right-hand side of (7.2.20) is majorized by c δ −j Dp,l−j (η2δ,x γ); eLp ≤ c δ −j fl−j (η2δ,x γ; e)[Cp,m−j (e)]1/p ≤ c fl−j (η2δ,x γ; e)[Cp,m (e)]1/p . Using the same arguments as when estimating fj (γ; e), we get fl−j (η2δ,x γ; e) ≤ ε sup ηδ,ξ γM (Wpm →Wpl ) + c(ε) sup ηδ,ξ γM (Wpm−l →Lp ) ξ∈Rn

ξ∈Rn

for j = 1, . . . , [l]. Therefore, the second sum on the right-hand side of (7.2.20) does not exceed   εsup ηδ,x γM (Wpm →Wpl ) +c(ε) sup ηδ,ξ γM (Wpm−l →Lp )+cfl (γ; e) [Cp,m (e)]1/p. x∈Rn

ξ∈Rn

Now we give a bound for Dp,l (ηδ,x η2δ,x γ); e\B4δ (x)Lp .

7.2 Two-Sided Estimates for the Essential Norm. The Case m > l

255

By H¨older’s inequality and the estimate for the capacity (3.1.2), we find that for z ∈ B2δ (x)  |y − z|−n−p{l} dy e\B4δ (x)

≤ c (mesn e)

(n−mp)/n




|y − z|−(n+p{l})n/mp dy

mp/n

e\B4δ (x)

≤ c Cp,m (e) δ p(m−{l})−n . Consequently, Dp,l (ηδ,x η2δ,x γ); e\B4δ (x)pLp  [l]   p (j−[l])p ≤c |∇j (η2δ,x (z)γ(z))| dz δ j=0

Bδ (x)

≤ c Cp,m (e)

[l] 

 δ (m−j+l)p−n B2δ (x)

j=0

≤ c Cp,m (e)

|y − z|−p{l}−n dy

e\B4δ (x)

[l] 

|∇j (η2δ,x (z)γ(z))|p dz

fj (η2δ,x γ; B2δ (x)) .

j=0

Following the same lines as when estimating fj (η2δ,x γ; e), we conclude that the third term on the right-hand side of (7.2.20) does not exceed (7.2.22). Substituting the derived estimates into (7.2.20), we arrive at
fl (ηδ,x γ; e) ≤ c ε sup ηδ,ξ γM (Wpm →Wpl ) ξ∈Rn

 + c(ε) sup ηδ,ξ γM (Wpm−l →Lp ) + fl (γ; e) , ξ∈Rn

which implies that ηδ,x γM (Wpm →Wpl )  ≤ c ε sup ηδ,ξ γM (Wpm →Wpl ) + c(ε) sup ηδ,ξ γM (Wpm−l →Lp ) ξ∈Rn

+

sup

ξ∈Rn

f0 (γ; e) +

e⊂B2δ (x)

sup

 fl (γ; e) .

e⊂B4δ (x)

Taking the supremum over x on both sides, we find that sup ηδ,x γM (Wpm →Wpl )

x∈Rn

≤c



sup {e : d(e)≤4δ}

f0 (γ; e) +

sup {e : d(e)≤8δ}

 fl (γ; e) .

(7.2.23)

256

7 Essential Norm and Compactness of Multipliers

Now let e ⊂ Rn with d(e) ≤ 1, and let r be a sufficiently large positive number. We have Dp,l (ζ3r γ); eLp [l]  ≤c |∇j γ|Dp,l−j ζ3r ; e\Br Lp j=0

+



[l]
  Rn

j=0

|∇j ζ3r (y)|p dy e\Br

1/p |∇[l]−j γ(y) − ∇[l]−j γ(z)|p dy dz n+p{l} |y − z|



+ Dp,l (ζ3r γ); e ∩ Br Lp .

(7.2.24)

The first sum on the right-hand side does not exceed c

[l] 

rj−l fj (γ; e\Br )[Cp,m−l+j (e\Br )]1/p

j=0

≤ c r−{l} γM (Wpm →Wpl ) [Cp,m (e\Br )]1/p . The j-th term on the right-hand side of (7.2.24) is majorized by c r−j Dp,l−j γ; e\Br Lp ≤ c r−j fl−j (γ; e\Br )[Cp,m−j (e\Br )]1/p . Hence the sum in question is dominated by c (fl (γ; e\Br ) + r−1 γM (Wpm →Wpl ) )[Cp,m (e)]1/p . Further, we estimate the last term on the right-hand side of (7.2.24). We have Dp,l (ζ3r γ); e ∩ Br pLp ≤

[l]   j=0

Rn \B

 e∩Br

3r

≤ c mesn e

[l]   j=0

We note that  Rn \B3r

|y − z|−n−p{l} dy

|∇j γ(z)|p dz r(j−[l])p

|∇j γ(z)|p

Rn \B3r

dz |z|n+p{l}

|∇j γ(z)|p dz r(j−[l])p . |z|n+p{l}

≤ r−p{l}

sup ξ∈Rn \B2r

∇j γ; B1 (ξ)pLp

≤ c r−p{l} γpM (W m →W l ) . p

p

7.2 Two-Sided Estimates for the Essential Norm. The Case m > l

257

Therefore, Dp,l (ζ3r γ); e ∩ Br Lp ≤ c (mesn e)1/p r−{l} γM (Wpm →Wpl ) . Consequently,   Dp,l (ζ3r γ); eLp ≤ c fl (γ; e\Br ) + r−{l} γM (Wpm →Wpl ) [Cp,m (e)]1/p . Using Theorem 4.1.1, we obtain the estimate ζ3r γM (Wpm →Wpl ) ≤ c1

sup e⊂Rn \Br ,d(e)≤1

  fl (γ; e) + f0 (γ; e)

+ c2 r−{l} γM (Wpm →Wpl ) .

(7.2.25)

This inequality, together with (7.2.23) and Theorem 7.2.1, implies the required upper bound for the essential norm. 

7.2.3 Estimates Involving Capacity (The Case mp = n, p > 1) Theorem 7.2.3. The relation (7.2.9) also holds for mp = n, p > 1. Proof. As in the proof of Theorem 7.2.2, we need to consider only the more difficult case of noninteger l, 0 < l < m. (i) The lower bound for the essential norm. Let e be a compact set in Rn with d(e) ≤ δ ≤ 1/2. The argument leading to (7.2.10) applies equally for obtaining the required estimate of f0 (γ; e). For fl (γ; e) we have the estimates (7.2.11)–(7.2.13). The right-hand side of (7.2.12) does not exceed c δ −{l} (mesn e){l}/n ∇[l] (γη2δ,x ); eLpn/(n−p{l}) ≤c

| log δ|(p−1)/p ∇[l] (γη2δ,x ); eLpn/(n−p{l}) . | log mesn e|(p−1)/p

The expression on the right-hand side here is not greater than c [Cp,m (e)]1/p sup

x∈Rn

Dp,l (γη2δ,x ); B4δ (x)Lp . [Cp,m (B4δ (x))]1/p

(7.2.26)

Similarly, the right-hand side of (7.2.13) does not exceed c (mesn e)1/p | log mesn e|(p−1)/p [Cp,m (e)]1/p δ −n/p sup Dp,l (γη2δ,z ); B4δ (z)Lp z∈Rn

which, in turn, is majorized by (7.2.26). Thus fl (γ; e) ≤ c sup fl (γη2δ,x ; B4δ (x)) . x∈Rn

258

7 Essential Norm and Compactness of Multipliers

Combining this estimate with Theorem 4.1.1 and Theorem 7.2.1, we get (7.2.15). Consequently we arrive at (7.2.16). The proof of (7.2.18) holds for mp = n as well. Thus the required lower bound for the essential norm is obtained. (ii) The upper bound for the essential norm. We take the relation (7.2.2) as the basis of the proof. Since ηδ2 ,x µδ,x = ηδ2 ,x , then, by Lemma 7.1.6, ηδ2 ,x γM (Wpm →Wpl ) ≤ c µδ,x γM (Wpm →Wpl ) . Our aim is to prove the estimate µδ,x γM (Wpm →Wpl ) ≤ c

sup (f0 (γ; e) + fl (γ; e)) .

(7.2.27)

e⊂B6δ (x)

Let e be a compact set in Rn with d(e) < 1/2. We have Dp,l (µδ,x γ); eLp [l]  ≤c |∇j γ|Dp,l−j µδ,x ; e ∩ B2δ (x)Lp j=0 [l]
 



|∇[l]−j γ(y) − ∇[l]−j γ(z)|p 1/p dz |y − z|n+p{l} e∩B2δ (x) j=1  (7.2.28) + Dp,l (µδ,x γ); e\B2δ (x)Lp + µδ,x Dp,l γ; eLp . +

|∇j µδ,x (y)| dy p

Applying Lemma 7.1.8, we find that |∇j γ|Dp,l−j µδ,x ; e ∩ B2δ (x)Lp ≤ c |∇j γ| | log r|−1 rj−l ; e ∩ B2δ (x)Lp ,

(7.2.29)

where r(z) = |z − x|. By H¨ older’s inequality the right-hand side does not exceed c ∇j γ; B2δ (x)Lpn/(n−p(l−j)) | log r|−1 rj−l ; e ∩ B2δ (x)Ln/(l−j) . Since the function | log t|−1 tj−l decreases near t = 0, the maximum value of the integral  dz n n/(l−j) E |z| | log |z|| over all sets E with prescribed small mesn E is attained for the ball centered at z = 0. Consequently, | log r|−1 rj−l ; e∩B2δ (x)Ln/(l−j) ≤ c | log mesn (e∩B2δ (x))|(l−j−n)/n . (7.2.30) We further note that   ∇j γ; B2δ (x)Lnp/(n−p(l−j)) ≤ c Dp,l γ; B2δ (x)Lp + δ −l γ; B2δ (x)Lp .

7.2 Two-Sided Estimates for the Essential Norm. The Case m > l

259

From (7.2.29) and two last inequalities we obtain |∇j γ|Dp,l−j µδ,x ; e ∩ B2δ (x)Lp   ≤ c | log δ|(p(l−j)−n)/np | log δ|(1−p)/p fl (γ; B2δ (x)) + f0 (γ; B2δ (x) × | log mesn (e ∩ B2δ (x))|(1−p)/p . Applying Proposition 4.3.1, we arrive at |∇j γ|Dp,l−j µδ,x ; e ∩ B2δ (x)Lp   ≤ c fl (γ; B2δ (x)) + f0 (γ; B2δ (x)) [Cp,m (e)]1/p .

(7.2.31)

The general term of the second sum on the right-hand side of (7.2.28) is equal to 
 1/p dh p p |∆ ∇ γ(z)| |∇ µ (z + h)| dz . h j δ,x [l]−j |h|n+p{l} e∩B2δ (x) Since supp µδ,x ⊂ Bδ (x), the last expression does not exceed
 |∇j µδ,x ; e ∩ B2δ (x)|Ln/j ∆h ∇[l]−j γ; B2δ (x)pLnp/(n−jp) B3δ (x)

1/p

dh |h|n+p{l}

.

Using Lemma 7.1.8 and applying the same argument as in the proof of (7.2.30), we conclude that (j−n)/n ∇j µδ,x ; e ∩ B2δ (x)Ln/j ≤ c log mesn (e ∩ B2δ (x)) . (7.2.32) It is known that the space Bqs11,p is imbedded continuously into Bqs22,p with s1 −n/q1 = s2 −n/q2 , 1 < q1 < q2 < ∞ (see [Bes]). This, in particular, implies that  dh ∆h ∇[l]−j γ; B2δ (x)pLnp/(n−jp) n+p{l} ≤ c |||γ; B6δ (x)|||pW l p |h| B3δ (x) p  ∼ c Dp,l γ; B6δ (x)Lp + δ −l γ; B6δ (x)Lp . (7.2.33) From (7.2.32) and (7.2.33) we obtain that the general term of the second sum on the right-hand side of (7.2.28) is dominated by  c | log δ|(pj−n)/np | log δ|(1−p)/p fl (γ; B6δ (x))  + f0 (γ; B6δ (x)) | log mesn (e ∩ B2δ (x))|(1−p)/p . This, together with Proposition 3.1.3, yields 
 |∇[l]−j γ(y) − ∇[l]−j γ(z)|p 1/p dz |∇j µδ,x (y)|p dy |y − z|n+p{l} e∩B2δ (x) ≤ c (fl (γ; B6δ (x)) + f0 (γ; B6δ (x)))[Cp,m (e)]1/p .

(7.2.34)

260

7 Essential Norm and Compactness of Multipliers

Let us estimate the norm Dp,l (µδ,x γ); e\B2δ (x)Lp , which is obviously equal to 
 |∇[l] (µδ,x γ)(z)|p 1/p dy dz |y − z|n+p{l} e\B2δ (x) Bδ (x)  1/p
 dy = |∇[l] (µδ,x γ)(z)|p dz . n+p{l} Bδ (x) e\B2δ (x) |y − z| It is clear that  e\B2δ (x)

dy ≤ min{δ −n−p{l} mesn e, δ −p{l} } |y − z|n+p{l} ≤

| log δ|p−1 . | log mesn e|p−1 δ p{l}

Moreover, by Lemma 4.2.7,  |∇[l] (µδ,x γ)(z)|p dz Bδ (x)

≤δ





p{l} Bδ (x)

Bδ (x)

|∇[l] (µδ,x γ)(y) − ∇[l] (µδ,x γ)(z)|p dz dy . |y − z|n+p{l}

Consequently, Dp,l (µδ,x γ); e\B2δ (x)pLp  | log δ|p−1
Dp,l γ; Bδ (x)pLp + |∇j γ|Dp,l−j µδ,x ; Bδ (x)pLp p−1 | log mesn e| j=0 [l]

≤c

+

[l]   j=1

Bδ (x)

 |∇j µδ,x (y)|p dy

Bδ (x)

|∇[l]−j γ(y) − ∇[l]−j γ(z)|p  dz . |y − z|n+p{l}

Putting e = Bδ (x) in (7.2.31) and (7.2.34), we see that either of the two last sums may be estimated from above by  p c | log δ|1−p fl (γ; B6δ (x)) + f0 (γ; B6δ (x)) . Therefore, Dp,l (µδ,x γ); e\B2δ (x)pLp ≤ c (fl (γ; B6δ (x)) + f0 (γ; B6δ (x)))p Cp,m (e) . Setting this, together with (7.2.31) and (7.2.34), into (7.2.28), we arrive at (7.2.27). The estimate lim sup ζr γM (Wpm →Wpl ) ≤ c lim r→∞

sup

r→∞ e⊂Rn \B ,d(e)≤1 r

(fl (γ; e) + f0 (γ; e))

was obtained at the end of the proof of Theorem 7.2.2.



7.2 Two-Sided Estimates for the Essential Norm. The Case m > l

261

7.2.4 Proof of Theorem 7.0.3 To obtain the lower estimate for essγM (Wpm →Wpl ) we use Theorem 7.2.2 and the upper estimates for the capacity of a ball (see Proposition 3.1.4). Then γ; eLp γ; Bρ (x)Lp ≥ sup 1/p 1/p ρ≤δ [Cp,m−l (Bρ (x))] {e:d(e)≤δ} [Cp,m−l (e)] sup

= c sup ρm−l−n/p γ; Bρ (x)Lp ≥ c sup ρm−l−n γ; Bρ (x)L1 , ρ≤δ

ρ≤δ

where x is an arbitrary point of R . Also, n

γ; eLp ≥ sup γ; B1 (x)Lp . 1/p {e⊂Rn \Br :d(e)≤1} [Cp,m−l (e)] x∈Rn \Br sup

To prove the upper estimate in part (i) of Theorem 7.0.3, we show that γ; eLp [C (e)]1/p p,m−l {e:d(e)≤δ} sup


≤c

 Dp,l γ; eLp m−l−n + sup δ γ; B (x) δ L 1 1/p x∈Rn {e:d(e)≤δ} [Cp,m (e)] sup

(7.2.35)

for pm ≤ n. In view of Corollary 4.3.1 and Lemma 4.3.11 we have γ; eLp 1/p [C p,m−l (e)] {e:d(e)≤1} sup


≤c

 Dp,l γ; eLp . + sup γ; B (x) 1 L 1 1/p x∈Rn {e:d(e)≤1} [Cp,m (e)] sup

(7.2.36)

In the last inequality we replace e by δ −1 E where d(E) ≤ δ and introduce Γ (·) = γ(δ −1 ·). Then (7.2.36) becomes δ −n/p Γ ; ELp −1 E)]1/p {E:d(E)≤δ} [Cp,m−l (δ sup


≤c

 δ l−n/p Dp,l Γ ; ELp + sup δ −n Γ ; Bδ (y)L1 . −1 1/p E)] y∈Rn {E:d(E)≤δ} [Cp,m (δ sup

For mp < n, by Corollary 3.1.1 we have Cp,m−l (δ −1 E) ≤ c δ p(m−l)−n Cp,m−l (E) and

Cp,m (δ −1 E) ≥ c δ pm−n Cp,m (E).

(7.2.37)

262

7 Essential Norm and Compactness of Multipliers

For mp = n the last inequality should be replaced by the estimate Cp,m (δ −1 E) ≥ c Cp,m (E), which holds since the capacity is a non-decreasing set function. Thus, for mp ≤ n δ −(m−l) Γ ; ELp sup 1/p {E:d(E)≤δ} [Cp,m−l (E)]
≤c

 δ −(m−l) Dp,l Γ ; ELp + sup δ −n Γ ; Bδ (y)L1 1/p [Cp,m (E)] y∈Rn {E:d(E)≤δ} sup



and (7.2.35) follows.

7.2.5 Sharpening of the Lower Bound for the Essential Norm in the Case m > l, mp ≤ n, p > 1 In addition to Theorems 7.2.2 and 7.2.3 we prove the following result. Theorem 7.2.4. If m > l, mp ≤ n, and p > 1, then ess γM (Wpm →Wpl ) ≥ c lim

sup

[l]
 Dp,l−j γ; eLp

δ→0 {e : d(e)≤δ} j=0

+ c lim

sup

[Cp,m−j (e)]1/p

+

∇j γ; eLp  [Cp,m−l+j (e)]1/p

[l]
 Dp,l−j γ; eLp

r→∞ {e⊂Rn \B : d(e)≤1} r j=0

[Cp,m−j (e)]1/p

+

∇j γ; eLp  . [Cp,m−l+j (e)]1/p

To prove this theorem, we need an auxiliary assertion. Lemma 7.2.2. For any multi-index α with |α| ≤ l < m, ess Dα γM (W m →Wpl−|α| ) ≤ c p

|α| 

ess γM (Wpm−j →Wpl−j ) .

j=0

Proof. It suffices to limit consideration to |α| = 1. By Tj , j = 0, 1, we denote compact operators acting from Wpm−j into Wpl−j and such that γu − Tj uWpl−j ≤ ( ess γM (Wpm−j →Wpl−j ) + ε)uWpm−j . For any function u ∈ Wpm we have γ∇u + u∇γ − ∇T0 uWpl−1 ≤ ( ess γM (Wpm →Wpl ) + ε)uWpm .

7.2 Two-Sided Estimates for the Essential Norm. The Case m > l

263

Therefore, u∇γ − ∇T0 u + T1 ∇uWpl−1 ≤ ( ess γM (Wpm →Wpl ) + ε)uWpm + γ∇u − T1 ∇uWpl−1 ≤ ( ess γM (Wpm →Wpl ) + ess γM (Wpm−l →Wpl−1 ) + 2ε)uWpm . Since T = ∇T0 − T1 ∇ is a compact operator acting from Wpm into Wpl−1 , it follows that ess ∇γM (W m →Wpl−1 ) ≤ ess γM (Wpm →Wpl ) + ess γM (Wpm−1 →Wpl−1 ) . p

 Proof of the Theorem 7.2.4. From the interpolation property (4.3.26) and from Theorem 7.2.1, we find that (l−j)/l

j/l

ess γM (Wpm−j →Wpl−j ) ≤ c ess γM (W m →W l ) ess γM (W m−l →L p

p

p

p)

which, together with Lemmas 2.3.4 and 4.3.4, gives ess γM (Wpm−j →Wpl−j ) ≤ c ess γM (Wpm →Wpl ) . By this inequality and Lemma 7.2.2 we conclude that ess ∇j γM (W m →Wpl−j ) ≤ c ess γM (Wpm →Wpl ) . p

It remains to make use of Theorems 7.2.2 and 7.2.3.



7.2.6 Estimates of the Essential Norm for mp > n, p > 1 and for p = 1 In the two cases mentioned, we do not need a capacity. The simplest formulation is for mp > n, p ≥ 1 and for m = n, p = 1. Theorem 7.2.5. If mp > n and p ≥ 1, or m = n and p = 1, and γ ∈ M (Wpm → Wpl ), m > l, then ess γM (Wpm →Wpl ) ∼ lim sup γ; B1 (x)Wpl . |x|→∞

(7.2.38)

264

7 Essential Norm and Compactness of Multipliers

Proof. Applying Corollary 4.3.8 and Theorem 5.4.1 to the multiplier ζr γ, we obtain from Lemma 7.2.1 that lim sup sup ζr γ; B1 (x)Wpl ≤ c ess γM (Wpm →Wpl ) r→∞

x∈Rn

which is equivalent to lim sup γ; B1 (x)Wpl ≤ c ess γM (Wpm →Wpl ) . |x|→∞

Let us prove the converse estimate. Corollary 4.3.8 and Theorem 5.4.1 imply that γM Wpl ∼ γM (Wpm →Wpl ) < ∞. This and Lemma 7.1.2 yield (1 − ζr )γuWpl ≤ γM Wpl (1 − ζr )uWpl ≤ c u; B4r Wpl . Since any bounded subset of Wpm is compact in Wpl (B4r ), the operator (1 − ζr )γ : Wpm → Wpl is compact. Consequently, for any r > 0, ess γM (Wpm →Wpl ) = ess ζr γM (Wpm →Wpl ) ≤ ζr γM (Wpm →Wpl ) . Estimating the last norm with the help of Corollary 4.3.8 and passing to the limit as r → ∞, we complete the proof.  Theorem 7.2.6. If l < m < n, then ess γM (W1m →W1l ) ∼ lim sup δ m−n sup |||γ; Bδ (x)|||W1l δ→0

x∈Rn m−n

+ lim sup sup r |x|→∞ r∈(0,1)

|||γ; Br (x)|||W1l .

(7.2.39)

Proof. According to Theorems 5.4.1 and 7.2.1, ess γM (W1m →W1l ) ∼ lim sup sup δ→0

sup rm−n |||ηδ,x γ; Br (y)|||W1l

x,y∈Rn r∈(0,1)

+ lim sup sup sup rm−n |||ζρ γ; Br (y)|||W1l . ρ→∞ y∈Rn r∈(0,1)

Let the first term on the right-hand side be denoted by A1 and the second one by A2 . Dealing first with A2 , we have A2 ≥ lim sup sup

sup rm−n |||γ; Br (y)|||W1l

ρ→∞ y ∈B / 2ρ r∈(0,1)

= lim sup sup rm−n |||γ; Br (y)|||W1l . |y|→∞ r∈(0,1)

7.2 Two-Sided Estimates for the Essential Norm. The Case m > l

265

An upper bound for A2 can be obtained as follows: A2 ≤ lim sup sup

sup rm−n |||ζρ γ; Br (y)|||W1l

ρ→∞ y ∈B / ρ/2 r∈(0,1)

≤ c lim sup sup

sup rm−n |||γ; B2r (y)|||W1l

ρ→∞ y ∈B / ρ/2 r∈(0,1)

≤ c1 lim sup sup rm−n |||γ; Br (y)|||W1l . |y|→∞ r∈(0,1)

Now we turn to estimates for A1 . We have A1 ≥ lim sup sup δ m−n |||ηδ,x γ; Bδ (x)|||W1l δ→0

x∈Rn

≥ c lim sup sup δ m−n |||γ; Bδ/2 (x)|||W1l . δ→0

x∈Rn

On the other hand, sup rm−n |||ηδ,x γ; Br (y)|||W1l

r∈(0,1)

≤

sup r∈(0,δ/2)

rm−n |||ηδ,x γ; Br (y)|||W1l + (2δ)m−n

sup r∈(δ/2,1)

|||ηδ,x γ; Br (y)|||W1l .

The first term on the right-hand side does not exceed c

sup r∈(0,δ/2)

rm−n |||γ; B2r (y)|||W1l

and the second one is not greater than c δ m−n |||γ; B2δ (x)|||W1l . Consequently, A1 ≤ c lim sup δ m−n sup |||γ; Bδ (x)|||W1l . δ→0

x∈Rn

 Remark 7.2.2. It follows from Lemma 7.1.1 that (7.2.39) can be written as   ess γM (W1m →W1l ) ∼ lim sup δ m−n sup δ −l γ; Bδ (x)L1 + D1,l γ; Bδ (x)L1 δ→0

x∈Rn

  + lim sup sup rm−n r−l γ; Br (x)L1 + D1,l γ; Br (x)L1 . |x|→∞ r∈(0,1)

Remark 7.2.3. Let T∗ be the operator defined by (7.2.4) for pm ≤ n, p > 1 and m < n, p = 1. For other values of p and m we put T∗ = (1 − ζr )γ. In the proofs of Theorems 7.2.1–7.2.6 we verified in passing the following estimates for the norm of γ − T∗ for fixed δ and r.

266

7 Essential Norm and Compactness of Multipliers

(i) If mp < n, p > 1, m > l, then γ − T∗ Wpm →Wpl ≤ c
+c

sup {e⊂Rn \Br/2 : d(e)≤1}

sup {e : d(e)≤8δ}

(fl (γ; e) + f0 (γ; e))

 (fl (γ; e) + f0 (γ; e)) + r−{l} γM (Wpm →Wpl ) .

(ii) If mp = n, p > 1, m > l, then the last estimate remains valid with {e : d(e) ≤ 8δ} replaced by {e : d(e) ≤ δ 1/2 }. (iii) If l < m < n, then   γ − T∗ W1m →W1l ≤ c δ m−n sup δ −l γ; Bδ (x)L1 + D1,l γ; Bδ (x)L1 x∈Rn

+c

sup x∈Rn \B

sup r/2

 −l  ρ γ; Bρ (x)L1 + D1,l γ; Bρ (x)L1 .

ρ∈(0,1)

(iv) If mp > n, p > 1 or m ≥ n, p = 1, m > l, then γ − T∗ Wpm →Wpl ≤ c

sup x∈Rn \Br/2

γ; B1 (x)Wpl .

From (i)-(iv), together with lower estimates for the essential norm proved in the above theorems, it follows that, given any ε, one can find r so large and δ so small that γ − T∗ Wpm →Wpl ≤ c ( ess γM (Wpm →Wpl ) + ε) .

(7.2.40)

7.2.7 One-Sided Estimates for the Essential Norm Using Theorem 7.0.3 and the lower estimates of the capacity Cp,m (see Propositions 3.1.2 and 3.1.3), one can readily obtain upper estimates for the essential norm in M (Wpm → Wpl ). Theorem 7.2.7. Let γ ∈ M (Wpm (Rn ) → Wpm (Rn )), where m > l ≥ 0. (i) If p ∈ (1, ∞) and mp < n, then ess γM (Wpm →Wpl )  Dp,l γ; eLp m−l− n p γ; B (x) ρ L 1 m + sup ρ p   − δ→0 {e:d(e)≤δ} x∈Rn mesn e p n ρ≤δ
 Dp,l γ; eLp sup . (7.2.41) sup γ; B (x) +c2 lim 1 L 1 m + p r→∞ e⊂Rn \Br (mes e) p − n x∈Rn \Br n d(e)≤1 ≤ c1 lim




sup

(ii) If p ∈ (1, ∞) and mp = n, then ess γM (Wpm →Wpl )

7.2 Two-Sided Estimates for the Essential Norm. The Case m > l

≤ c1 lim




 log

sup

δ→0 {e:d(e)≤δ}


+c2 lim

r→∞

 sup log

e⊂Rn \Br d(e)≤1

267

 2n 1− p1 Dp,l γ; eLp + sup ρ−l γ; Bρ (x)Lp mesn e x∈Rn ρ≤δ

 2n 1− p1 Dp,l γ; eLp+ sup γ; B1 (x)Lp . (7.2.42) mesn e x∈Rn \Br

Restricting the suprema in Theorem 7.0.3 to balls of radii less than one and using the formulas for the capacity of a ball (see Proposition 3.1.4), we arrive at the following lower estimates for the essential norm in M (Wpm → Wpl ). Theorem 7.2.8. Let γ ∈ M (Wpm (Rn ) → Wpm (Rn )), where m > l ≥ 0. (i) If p ∈ (1, ∞) and mp < n, then ess γM (Wpm →Wpl )
 n ≥ c1 lim sup ρm− p Dp,l γ; Bρ (x)Lp + ρ−l γ; Bρ (x)Lp δ→0

x∈Rn ρ≤δ

+c2 lim

γ; B1 (x)Wpl .

sup

r→∞ x∈Rn \B

(7.2.43)

r

(ii) If p ∈ (1, ∞) and mp = n, then ess γM (Wpm →Wpl ) ≥ c1 lim sup δ→0

x∈Rn ρ≤δ


  1− p1 log(2/ρ) Dp,l γ; Bρ (x)Lp + ρ−l γ; Bρ (x)Lp +c2 lim

γ; B1 (x)Wpl .

sup

r→∞ x∈Rn \B

(7.2.44)

r

Remark 7.2.4. Some one-sided estimates for essential norms of multipliers acts s (Rn ), Fp,q (Rn ) were obtained ing in pairs of Besov-Triebel-Lizorkin spaces Bp,q by Edmunds and Shargorodsky in [ES]. These estimates involve norms in the same scales of spaces. The proofs are based on a certain abstract functionalanalytic equivalent representation of the essential norm. 7.2.8 The Space of Compact Multipliers ˚ (Wpm → Wpl ), m > l, we mean the set of functions Definition 7.2.1. By M γ such that the operator of multiplication by γ is a compact operator acting from Wpm into Wpl . Needless to say, ˚ (W m → W l ) if and only if γ∈M p p

ess γM (Wpm →Wpl ) = 0.

Therefore, Theorems 7.0.3, 7.2.5, and 7.2.6 imply the following necessary and sufficient conditions for a function γ ∈ M (Wpm → Wpl ) to belong to the class ˚ (Wpm → Wpl ). M

268

7 Essential Norm and Compactness of Multipliers

˚ (Wpm → Wpl ) if and Theorem 7.2.9. (i) If mp ≤ n and p > 1, then γ ∈ M only if
lim

δ→0

 Dp,l γ; eLp m−l− n p γ; B (x) + sup ρ ρ Lp = 0, (7.2.45)  1 x∈Rn {e:d(e)≤δ} Cp,m (e) p ρ≤δ
 Dp,l γ; eLp sup lim + sup γ; B1 (x)Lp = 0. (7.2.46) 1 r→∞ e⊂Rn \Br (C p x∈Rn \Br p,m (e)) d(e)≤1 sup

(ii) Let either mp > n and p ≥ 1, or m = n and p = 1. Then γ ∈ ˚ (W m → W l ) if and only if γ ∈ W l M p p p,unif and lim γ; B1 (x)Wpl = 0 .

|x|→∞

(7.2.47)

(iii) In the case m < n, a necessary and sufficient condition for γ ∈ ˚ (W m → W l ) is (7.2.47) together with M 1 1 lim δ m−n sup |||γ; Bδ (x)|||W1l = 0 .

δ→0

(7.2.48)

x∈Rn

Remark 7.2.5. From Theorems 7.2.2 and 7.2.3, we obtain another form of the ˚ (W m → W l ) with mp ≤ n, p > 1: compactness criteria for γ ∈ M p p


Dp,l γ; eLp  γ; eLp = 0 , (7.2.49) + δ→0 {e : d(e)≤δ} [Cp,m−l (e)]1/p [Cp,m (e)]1/p
γ; e Dp,l γ; eLp  Lp lim = 0. (7.2.50) sup + r→∞ {e⊂Rn \B : d(e)≤1} [Cp,m−l (e)]1/p [Cp,m (e)]1/p r lim

sup

Theorem 7.2.1 immediately implies: ˚ (W m → W l ) if Theorem 7.2.10. Let lp < n, m > l and p ≥ 1. Then γ ∈ M p p and only if lim sup ηδ,x γM (Wpm →Wpl ) = 0 ,

δ→0 x∈Rn

lim ζr γM (Wpm →Wpl ) = 0 .

r→∞

From this theorem combined with Propositions 4.3.1, 4.3.2 and the results in Sect. 4.4, we can get various necessary or sufficient conditions for ˚ (W m → W l ) which do not contain capacity. γ∈M p p ˚ (Wpm → Wpl ) with The following theorem gives one more description of M m > l. ˚ (Wpm → Wpl ) is the completion of C ∞ with Theorem 7.2.11. The space M 0 respect to the norm in M (Wpm → Wpl ).

7.2 Two-Sided Estimates for the Essential Norm. The Case m > l

269

˚ (Wpm → Wpl ). Therefore, any function in Proof. By Theorem 7.2.9, C0∞ ⊂ M m l M (Wp → Wp ), approximated by a sequence in C0∞ in the norm of M (Wpm → Wpl ), generates a compact operator of multiplication: Wpm → Wpl . ˚ (W m → W l ). AccordFurther, we prove the converse assertion. Let γ ∈ M p p ing to parts (ii) and (iii) of Theorem 7.2.9, it suffices to consider the case mp ≥ n, p > 1. By Theorem 7.2.1 we have lim γ − (1 − ζr )γM (Wpm →Wpl ) = 0 .

(7.2.51)

r→∞

Let Γ = (1 − ζr )γ and let Γρ be a mollification of Γ with radius ρ. By T∗ (ρ) and T∗ we denote the operators given by (7.2.4) for Γ and Γρ respectively. It follows from (7.2.8) that lim γ − T∗ Wpm →Wpl = 0 .

(7.2.52)

r→0

By (7.2.8) and Theorem 7.2.1, (ρ)

Γρ − T∗ Wpm →Wpl ≤ c lim sup sup ηδ,x Γρ M (Wpm →Mpl ) + ε δ→0

x∈Rn

≤ c ess Γρ M (Wpm →Wpl ) + ε = ε .

(7.2.53)

The last equality holds since Γρ ∈ C0∞ . (ρ) From the definitions of the operators T∗ and T∗ we get  (ρ) (Γ − Γρ )ϕ(j) Wpl (T∗ − T∗ )uWpl ≤ c(δ, r)uLp j

and hence (ρ)

T∗ − T∗ Wpm →Wpl ≤ c(δ, r)Γ − Γρ Wpl .

(7.2.54)

The right-hand side of this inequality tends to zero as ρ → 0. Since (ρ)

Γ − Γρ M (Wpm →Wpl ) ≤ Γ − T∗ Wpm →Wpl + Γρ − T∗ Wpm →Wpl (ρ)

+ T∗ − T∗ Wpm →Wpl , it follows by (7.2.52)–(7.2.54) that lim

r→∞,ρ→0

γ − Γρ M (Wpm →Wpl ) = 0 .

Since Γρ ∈ C0∞ , the proof is complete.



270

7 Essential Norm and Compactness of Multipliers

7.3 Two-Sided Estimates for the Essential Norm in the Case m = l 7.3.1 Estimate for the Maximum Modulus of a Multiplier in Wpl by its Essential Norm Theorem 7.3.1. If l > 0 and 1 ≤ p < ∞, then γL∞ ≤ ess γM Wpl .

(7.3.1)

Proof. Let T be a compact operator in Wpl such that (γ − T )uWpl ≤ ( ess γM Wpl + ε)uWpl

(7.3.2)

for all u ∈ Wpl . Let η be an arbitrary function in C0∞ (Qk ), where Qk is the cube {y : |yj | < πk} and k is an integer. We consider the sequence uN (y) = N −l exp(iN y1 )η(y) ,

N = 1, 2, . . . .

Obviously, for integer l we have uN Wpl = ηLp + O(N −1 ). Let l be a noninteger, 0 < l < 1. Then uN Wpl = N −l eiN y1 ηWpl = N −l Dp,l eiN y1 ηLp + O(N −1 ) . Clearly, |Dp,l (eiN y1 η) − |η|Dp,l eiN y1 | ≤ Dp,l η. Since Dp,l eiN y1 = al N l , where al = const > 0, it follows that Dp,l (eiN y1 η) − al N l |η| Lp = O(1) .

(7.3.3)

Let l > 1. Then Dp,l (eiN y1 η) = Dp,{l} (∇[l] eiN y1 ). We have [l]
 ∂ [l]−j η  iN y1 [l] iN y1 η)) − N Dp,{l} (e η) ≤ c N j Dp,{l} eiN y1 [l]−j . Dp,{l} (∇[l] (e ∂y1 j=0

7.3 Two-Sided Estimates for the Essential Norm in the Case m = l

271

This and (7.3.3) imply that Dp,l (eiN y1 η) − N l a{l} Lp ≤ c N [l] . So in the case {l} > 0 we obtain uN Wpl = a{l} ηLp + O(N −{l} ) .

(7.3.4)

˚pl (Qk ). Let f be any We show that {uN } converges weakly to zero in W ˚ l (Qk ). If p ≤ 2, then the restriction of f to W ˚ l (Qk ) is linear functional on W p 2 l ˚ (Qk ). Consequently, a linear functional on W 2  f (uN ) = Λl uN Λl ψ dx, ˚ l (Qk ). Since where ψ ∈ W 2 Λl un − eiN y1 ηL2 = O(N −1 ) and the sequence

 eiN y1 η(y)Λl ψ dy Qk

tends to zero, being a sequence of Fourier coefficients of a function in L2 (Qk ), it follows that f (uN ) −→ 0 as N → ∞. Let p > 2. Taking into account the imbedding of Hpl into Wpl , we get |f (uN )| ≤ c uN Hpl . Therefore,

 f (uN ) =

gΛl uN dy ,

where g ∈ Lp
. Since Λl uN − eiN y1 ηLp = O(N −1 ) , 

we have f (uN ) =

Qk

eiN y1 η(y)g dy + O(N −1 )gLp
.

Applying the Hausdorff-Young theorem (see [Zy], Ch. V.II) to the function ηg ∈ Lp
(Qk ), p < 2, we conclude that f (uN ) −→ 0 as N → ∞. By ϕ we denote a function in C0∞ (Q1 ) which is equal to one on the cube Q1−δ , δ > 0, and we set ϕk (y) = ϕ(y/k). The compactness of the operator ˚ l (Qk ) implies that ϕk T in W p ϕk T uN −→ 0 N →∞

˚pl (Qk ). in W

272

7 Essential Norm and Compactness of Multipliers

Now it follows from Lemma 7.1.9 and (7.3.2) that lim sup ϕk γuN Wpl = lim sup ϕk (γ − T )uN Wpl N →∞

N →∞

≤ (1 + O(k −δ )) lim sup (γ − T )uN Wpl N →∞

≤ (1 + O(k

−δ

)) lim sup uN Wpl ( ess γM Wpl + ε) N →∞

which together with (7.3.4) yields lim sup ϕk γuN Wpl ≤ (1 + O(k −δ ))a{l} ηLp ( ess γM Wpl + ε) . N →∞

With the same arguments as in the proof of (7.3.4), we obtain lim ϕk γuN Wpl = a{l} ϕk γηLp .

N →∞

Thus lim sup ϕk γηLp ≤ ηLp ess γM Wpl . k→∞

Since ϕk η = η for large values of k, and η is an arbitrary function in C0∞ , the result follows. 

7.3.2 Estimates for the Essential Norm Involving Cutoff Functions (The Case lp ≤ n, p > 1) Theorem 7.3.2. For lp < n, p ≥ 1 the relation ess γM Wpl ∼ lim sup sup ηδ,x γM Wpl + lim sup ζr γM Wpl (7.3.5) δ→0

x∈Rn

r→∞

holds. The proof of this relation can be obtained by duplicating the proof of Theorem 7.2.1, where m = l. Theorem 7.3.3. If 0 < l ≤ 1, lp = n, and p > 1, then ess γM Wpl ∼ γL∞ + lim sup sup ηδ,x Dp,l γM (Wpl →Lp ) δ→0

x∈Rn

+ lim sup ζr γM Wpl .

(7.3.6)

r→∞

Proof. (i) The upper bound for the essential norm. We choose δ and r so that sup ηδ,x Dp,l γM (Wpl →Lp ) ≤ lim sup sup ηδ,x Dp,l γM (Wpl →Lp ) + ε ,

x∈Rn

δ→0

x∈Rn

7.3 Two-Sided Estimates for the Essential Norm in the Case m = l

273

ζr γM Wpl ≤ lim sup ζr γM Wpl + ε . r→∞

Let Γ and T∗ be the function and operator introduced in the second part of Theorem 7.2.1. By (7.2.5) it suffices to get the estimate for (Γ − T∗ )uWpl . We have (Γ − T∗ )upW l ≤ A + B + C ,

(7.3.7)

p

where +  (j) +p A = +Γ ϕ (u − P (j) )+L , p

j

+ +p B = + (Dp,l Γ )η2δ,xj ϕ(j) (u − P (j) )+L , p

j

 + +p C = +Γ Dp,l ϕ(j) (u − P (j) )+L . p

j

By Lemma 7.1.2, A ≤ γpL∞



(j)

u − P (j) ; Kδ pLp ≤ c γpL∞ Dp,l upLp .

(7.3.8)

j

It follows from Lemmas 7.1.2 and 7.1.3 that  B≤c (Dp,l Γ )η2δ,xj ϕ(j) (u − P (j) )pLp j

≤ c sup η2δ,xj Dp,l Γ pM (W l →Lp )



p

j

ϕ(j) (u − P (j) )pLp

j

≤ c1 sup η2δ,xj Dp,l Γ pM (W l →Lp ) Dp,l upLp .

(7.3.9)

p

j

Using Lemmas 7.1.1–7.1.3, we deduce that  ϕ(j) (u − P (j) )pW l C ≤ c γpL∞

p

j

≤ c1 γpL∞



(j)

|||u − P (j) ; Kδ |||pW l ≤ c2 γpL∞ upW l p

p

j

which together with (7.3.7)–(7.3.9) implies that   (Γ − T∗ )uWpl ≤ c γL∞ + sup η2δ,xj Dp,l Γ M (Wpl →Lp ) uWpl . j

Lemma 7.1.9 enables one to replace Γ by γ on the right-hand side of the last inequality. The required upper estimate for the essential norm is obtained. (ii) The lower bound for the essential norm. Let T be a compact operator in Wpl such that

274

7 Essential Norm and Compactness of Multipliers

Dp,l (γu) − Dp,l (T u)Lp ≤ ( ess γM Wpl + ε)uWpl . Hence uDp,l γ − Dp,l T uLp ≤ ( ess γM Wpl + ε)uWpl + γDp,l uLp . It follows from the inequality Dp,l v1 − Dp,l v2 Lp ≤ Dp,l (v1 − v2 )Lp that the compactness of the set {T u : u ∈ S} in the space Wpl , where S is the unit ball in Wpl , implies the compactness of the set {Dp,l T u : u ∈ S} in Lp . Let the collection {wk } form a finite ε-net in the last set. Then, for u ∈ S, uηδ,x Dp,l γLp ≤ ηδ,x (Dp,l T u − wk )Lp + ηδ,x wk Lp + ess γM Wpl + ε + γL∞ . Consequently, for any x ∈ Rn and for sufficiently small δ > 0, uηδ,x Dp,l γLp ≤ c ( ess γM Wpl + ε) . (Here we have used Theorem 7.2.1.) This, together with Lemma 7.2.1 and Theorem 7.3.1, implies the required lower bound for the norm ess γM Wpl .  Theorem 7.3.4. If l ≥ 1, lp = n, and p > 1, then ess γM Wpl ∼ γL∞ + lim sup sup ηδ,x ∇k γM (W l →Wpl−k ) δ→0

x∈Rn

p

+ lim sup ζr γM Wpl , r→∞

where k = 1, . . . , [l]. Proof. (i) Upper bound for the essential norm. We have (Γ − T∗ )upW l p  +   (j) +p + ≤ c ∇k Γ ϕ (u − P (j) ) +Wpl−k + (Γ − T∗ )upLp . (7.3.10) j

The second term on the right-hand side does not exceed c γpL∞ upW l

p

(see estimate (7.3.8)). The first term is not greater than  +  +p +Dα Γ c Dβ [ϕ(j) (u − P (j) )]+Wpl−k |α|+|β|=k

j

7.3 Two-Sided Estimates for the Essential Norm in the Case m = l k 

≤c

|α|=0



sup η2δ,xj Dα Γ p

l−k+|α| M (Wp →Wpl−k )

j

275

ϕ(j) (u − P (j) )pW l . (7.3.11) p

j

Since p(l − k) < n, we conclude, using Theorems 7.2.1 and 7.3.2, that the expression on the right-hand side of (7.3.11) is dominated by


c sup η2δ,xj ∇k Γ M (W l →Wpl−k ) +

k−1 

p

j

ess ∇j Γ (Wpl−k+j →Wpl−k )

p

upW l , p

j=0

which by Lemma 7.2.2 does not exceed k p
 c sup η2δ,xj ∇k Γ M (W l →Wpl−k ) + ess Γ M Wpl−i upW l . p

j

p

i=1

We obtain by interpolation that (l−i)/l

i/l

ess Γ M Wpl−i ≤ Γ − T∗ Wpl−i →Wpl−i ≤ c Γ − T∗ W l →W l Γ − T∗ Lp →Lp p

p

≤ εΓ − T∗ Wpl →Wpl + c(ε)γL∞ ,

(7.3.12)

where ε is an arbitrarily small positive number. Therefore, the right-hand side of (7.3.10) does not exceed
p c sup η2δ,xj ∇k Γ M (W l →Wpl−k ) + εΓ − T∗ Wpl →Wpl + c(ε)γL∞ upW l . j

p

p

Choosing ε sufficiently small and applying Lemma 7.1.9, we obtain from the last inequality and from (7.3.10) that   (Γ − T∗ )Wpl →Wpl ≤ c sup η2δ,xj ∇k Γ M (W l →Wpl−k ) + γL∞ (7.3.13) j

p

which, together with (7.2.5) and Lemma 7.1.9, gives the required upper bound for the essential norm. (ii) The lower bound for the essential norm. By Lemma 7.2.1 and Theorem 7.3.1, it suffices to show that ηδ,x ∇k γM (W l →Wpl−k ) ≤ c ( ess γM Wpl + ε) p

(7.3.14)

for all x ∈ Rn and small enough δ > 0. Let T be a compact operator for which (7.3.2) holds. Then, for all u ∈ Wpl , ∇k [(γ − T )u]Wpl−k ≤ ( ess γM Wpl + ε)uWpl . In view of the inequality p(l − k) < n, ηδ,x ∇k [(γ − T )u]Wpl−k ≤ c ( ess γM Wpl + ε)uWpl .

276

7 Essential Norm and Compactness of Multipliers

Let S be the unit ball in Wpl . The set {v = Dα T u, |α| = k : u ∈ S} is compact in Wpl−k . Let {vν } be an ε-net in Wpl−k for the last set. Without loss of generality we may assume that vν ∈ C0∞ . Since p(l − k) < n, we see by Lemma 7.1.7 that for small δ sup ηδ,x vν Wpl−k < ε

x∈Rn

and hence sup ηδ,x ∇k (T u)Wpl−k < c ε .

x∈Rn

Thus, for all u ∈ S, 

 uηδ,x ∇k γWpl−k ≤ c ess γM Wpl +

ηδ,x Dα uDβ γWpl−k + ε



|α|+|β|=k, |α|>0

  ≤ c essγM Wpl + |α|+|β|=k |α|>0

 ηδ,x Dβ γM (Wpl−|α|→Wpl−k ) Dα uWpl−|α| +ε . (7.3.15)

From Theorems 7.2.1 and 7.3.2, it follows that for small δ ηδ,x Dβ γM (Wpl−|α| →Wpl−k ) ≤ c ess Dβ γM (Wpl−|α| →Wpl−k ) . Making use of Theorem 7.3.2 and interpolating (see (7.3.12)), we obtain ess D γM (Wpl−|α| →Wpl−k ) ≤ c β

|β| 

ess γM Wpl−|α|−j

j=0

≤ εγ − T∗ Wpl →Wpl + c(ε)γL∞ . From this inequality, combined with (7.3.15) and the estimate   γ − T∗ Wpl →Wpl ≤ c sup η2δ,xj ∇k γM (W l →Wpl−k ) + γL∞ + ζr γM Wpl j

p

established in the first part of the proof (see (7.3.13)), we get for u ∈ S uηδ,x ∇k γWpl−k ≤ c(ε) ess γM Wpl + ε sup η2δ,xj ∇k γM (W l →Wpl−k ) + ε . j

p

Inequality (7.3.14) follows, which completes the proof of the theorem.



7.3 Two-Sided Estimates for the Essential Norm in the Case m = l

277

7.3.3 Estimates for the Essential Norm Involving Capacity (The Case lp ≤ n, p > 1) For the theorem in this subsection we need the following interpolation inequality. Lemma 7.3.1. If lp ≤ n, p > 1, and 0 < σ < l, then σ/l

1−σ/l

ess γM Wpσ ≤ c ess γM W l γL∞

.

(7.3.16)

p

Proof. When proving any of Theorems 7.3.2–7.3.4, it was shown in passing that for some r and δ we obtain from the definition of T∗ that γ − T∗ Wpl →Wpl ≤ c ess γM Wpl + ε

(7.3.17)

(cf. Remark 7.2.2). Moreover, γ − T∗ Lp →Lp ≤ c γL∞ . Hence, interpolating between Wpl and Lp , we get 1−σ/l

ess γM Wpσ ≤ γ − T∗ Wpσ →Wpσ ≤ c ( ess γM Wpl + ε)σ/l γL∞

. 

Theorem 7.3.5. Let lp ≤ n and p > 1. Then Dp,l γ; eLp δ→0 {e : d(e)≤δ} [Cp,l (e)]1/p

ess γM Wpl ∼ γL∞ + lim

+ lim

sup

sup

r→∞ {e⊂Rn \B : d(e)≤1} r

Dp,l γ; eLp . (7.3.18) [Cp,l (e)]1/p

Proof. For lp < n it suffices to duplicate the proof of Theorem 7.2.2, putting m = l. Inequalities (7.2.18) and (7.2.25), which are also valid for m = l ≤ n/p, imply that lim sup ζr γM Wpl r→∞


∼ lim γ; Rn \Br L∞ + r→∞

Dp,l γ; eLp  . (7.3.19) 1/p {e⊂Rn \Br : d(e)≤1} [Cp,l (e)] sup

Therefore, from Theorem 7.3.3 and Remark 7.2.2, we have (7.3.18) for 0 < l ≤ 1 and lp = n.

278

7 Essential Norm and Compactness of Multipliers

Let lp = n and l > 1. It is shown in the proof of Theorem 7.3.3 that
ηδ2 ,x ∇γM (W l →Wpl−1 ) ≤ c p


ζr γM Wpl ≤ c

sup

∇γ; eLp  Dp,l−1 (∇γ); eLp + , 1/p [Cp,l (e)] [Cp,1 (e)]1/p {e : d(e)≤2δ} sup


D

{e⊂Rn \Br/2 : d(e)≤1}

p,l−1 (∇γ); eLp [Cp,l (e)]1/p

+

 ∇γ; eLp  +γ . L ∞ [Cp,1 (e)]1/p

Hence, using the estimate for ess γM Wpl and Theorem 7.3.4, we get
ess γM Wpl ≤ c γL∞ + lim

Dp,l γ; eLp δ→0 {e : d(e)≤δ} [Cp,l (e)]1/p sup

 Dp,l γ; eLp 1 . + ess γ M W p r→∞ {e⊂Rn \B : d(e)≤1} [Cp,l (e)]1/p r

+ lim

sup

It remains to note that, by Lemma 7.3.1, 1/l

(l−1)/l

ess γM Wp1 ≤ c ess γM W l γL∞

.

p

Let us derive the lower bound for the essential norm. By Lemma 7.3.1 and Theorem 7.3.1, ess γM Wpl ≥ c ess γM Wpl−1 which, together with Lemma 7.2.2, gives the estimate ess γM Wpl ≥ c ess ∇γM (W l →Wpl−1 ) . p

Taking into account Theorem 7.2.3, we obtain that the right-hand side of this inequality is not less than Dp,l γ; eLp . δ→0 {e : d(e)≤δ} [Cp,l (e)]1/p

c lim

sup



It remains to use (7.3.19) and Theorem 7.2.2. The result follows.

7.3.4 Two-Sided Estimates for the Essential Norm in the Cases lp > n, p > 1, and p = 1 Theorem 7.3.6. If lp > n and p > 1, then ess γM Wpl ∼ γL∞ + lim sup γ, B1 (x)Wpl . |x|→∞

(7.3.20)

7.3 Two-Sided Estimates for the Essential Norm in the Case m = l

279

Proof. From Corollary 4.3.8 and Lemma 7.1.9 we obtain lim sup γ; B1 (x)Wpl ∼ lim sup ζr γM Wpl . |x|→∞

(7.3.21)

r→∞

Hence the required lower bound for the essential norm follows by Lemma 7.2.1 and Theorem 7.3.1. Now we establish the upper bound. Let Γ and T∗ be the function and operator specified in the second part of Theorem 4.2.1. With the function Γρ , which stands for a mollification of Γ with radius ρ, we associate the operator (ρ) T∗ by the same rule. Using Corollary 4.3.8 for sufficiently small ρ, we find that Γ − Γρ M Wpl ≤ c Γ − Γρ Wpl < ε .

(7.3.22)

Next we note that the proof of inequality (7.3.13) holds in the case lp > n. Replacing the numbers l and k by an integer s, s > np, in (7.3.13) and using Corollary 4.3.8, we arrive at 
(ρ) Γρ − T∗ Wps →Wps ≤ sup η2δ,xj ∇s Γρ Lp + γL∞ . j

This implies for small δ that (ρ)

Γρ − T∗ Wps →Wps ≤ c γL∞ .

(7.3.23)

The same inequality obviously holds for s = 0. Interpolating between Lp and Wps , we obtain (7.3.23) for s = l which, together with (7.3.22), gives (ρ)

Γ − T∗ Wpl →Wpl ≤ c γL∞ + ε . The result follows from the last inequality and (7.2.5). (ρ) Further, we note that one may replace the operator T∗ by T∗ in the last inequality. In fact, it follows from Lemmas 7.1.2, 7.1.3 and Corollary 4.3.1 that   + +p +u − ϕ(j) P (j) +W l ≤ c ϕ(j) (u − P (j) )pW l ≤ c1 upW l . j

p

p

p

j

Therefore, + (j) (j) + (ρ) ϕ P +W l ≤ c Γ − Γρ M Wpl uWpl (T∗ − T∗ )uWpl ≤ Γ − Γρ M Wpl + j

and it remains to use inequality (7.3.22).

p



280

7 Essential Norm and Compactness of Multipliers

Theorem 7.3.7. If l < n, then ess γM W1l ∼ lim sup δ l−n sup |||γ; Bδ (x)|||W1l x∈Rn

δ→0

+ lim sup sup rl−n |||γ; Br (x)|||W1l .

(7.3.24)

|x|→∞ r∈(0,1)

The proof runs in the same way as that of Theorem 7.2.6, where one should put m = l and use Theorem 7.3.2 instead of Theorem 7.2.1. Remark 7.3.1. By Lemma 7.1.1, the equivalence relation (7.3.24) can be rewritten as ess γM W1l ∼ γL∞ + lim sup δ l−n sup D1,l γ; Bδ (x)L1 x∈Rn

δ→0

+ lim sup sup rl−n D1,l γ; Br (x)L1 . |x|→∞ r∈(0,1)

Theorem 7.3.8. If l ≥ n, then ess γM W1l ∼ γL∞ + lim sup γ; B1 (x)W1l .

(7.3.25)

|x|→∞

Proof. The lower bound for the essential norm follows directly from Lemma 7.2.2 and Theorems 5.4.1 and 7.3.1. Next we obtain the upper bound. Let k = [l] + 1 − n. We have 
+  + ϕ(j) (u − P (j) ) +W l−k + (Γ − T∗ )uL1 . (Γ − T∗ )uW1l ≤ c +∇k 1

j

The second term on the right-hand side does not exceed c γL∞ uW1l (see (7.3.8)). The first one is not greater than   + + +Dα Γ Dβ [ϕ(j) (u − P (j) )]+W l−k c |α|+|β|=k

≤c

1

j k  |α|=0

sup η2δ,xj Dα Γ M (W l−k+|α| →W l−k ) 1

1

j



ϕ(j) (u − P (j) )W1l .

j

With the help of Lemmas 7.1.1–7.1.3 we obtain that the last norm is majorized by c uW1l . Now we show that lim sup sup ηδ,x Γ M W l−k ≤ c Γ L∞ . δ→0

x∈Rn

1

(7.3.26)

Since l − k < n, it follows by Theorems 7.3.2 and 7.3.3 that the left-hand side of (7.3.26) does not exceed

7.3 Two-Sided Estimates for the Essential Norm in the Case m = l

281

c ess Γ M W l−k ∼ lim sup δ l−k−n sup |||Γ ; Bδ (x)|||W l−k 1

1

x∈Rn

δ→0

≤ lim sup δ l−n sup |||Γ ; Bδ (x)|||W1l = Γ L∞ . δ→0

x∈Rn

Thus, (7.3.26) follows. To complete the proof, it suffices to establish the equality lim sup ηδ,x ∇j Γ M (W l−k+j →W l−k ) = 0 ,

δ→0 x∈Rn

1

1

(7.3.27)

where j = 1, . . . , k. By Theorem 7.2.1 the left-hand side is equivalent to the essential norm of ∇j Γ in M (W1l−k+j → W1l−k ). Since ∇j Γ ∈ W1l−j ,

supp∇j Γ ⊂ B2r ,

and l − k + j > n, we have by Theorem 7.2.9, part (ii), that ˚ (W l−k+j → W l−k ). ∇j Γ ∈ M 1 1 

The equality (7.3.27) is proved, and so is the theorem.

Remark 7.3.2. In addition to Lemma 7.3.1 we note that by Theorems 7.3.7, 7.3.8 and estimate (7.1.7), the following interpolation inequality is valid: 1−σ/l

ess γM W1σ ≤ c ( ess γM W1l )σ/l γL∞

,

0 < σ < l.

˚W l 7.3.5 Essential Norm in M p ˚ (Wpm → According to Theorem 7.2.11, the space of compact multipliers M Wpl ), m > l, coincides with the completion of C0∞ with respect to the norm ˚ W l denotes the completion of C ∞ of the space M (Wpm → Wpl ). Similarly, M p 0 with respect to the norm of the space M Wpl . The following theorem shows ˚ Wpl is equivalent to the norm in L∞ . that the essential norm in M ˚ Wpl , l ≥ 0, and p ≥ 1, then Theorem 7.3.9. If γ ∈ M γL∞ ≤ ess γM Wpl ≤ c γL∞ .

(7.3.28)

Proof. The left-hand estimate was obtained in Theorem 7.3.1. Let us establish the upper bound for the essential norm. Without loss of generality, we may assume that γ ∈ C0∞ . Let p > 1 and lp ≥ n. Since Cp,l (e) ≥ c (mesn e)ν ,

282

7 Essential Norm and Compactness of Multipliers

where ν ∈ (0, 1) and d(e) ≤ δ (see Proposition 3.1.2), we have Dp,l γ; eLp ≤ c (mesn e)(1−ν)/p [Cp,l (e)]1/p and

Dp,l γ; eLp = 0. δ→0 {e : d(e)≤δ} [Cp,l (e)]1/p lim

sup

For any compact set e ⊂ Rn \Br with d(e) ≤ 1, Dp,l γ; eLp ≤ c r−{l}−n/p (mesn e)(1−ν)/p . [Cp,l (e)]1/p Therefore, lim

sup

r→∞ {e⊂Rn \B : d(e)≤1} r

Dp,l γ; eLp = 0. [Cp,l (e)]1/p

Now the right-hand inequality in (7.3.28) follows from Theorem 7.3.5. For lp > n, p > 1 and for l ≥ n, p = 1, the result is a consequence of the equality lim γ; B1 (x)Wpl = 0 |x|→∞

and Theorems 7.3.6 and 7.3.8. Finally, for l < n and p = 1, the desired estimate for the essential norm follows immediately from Remark 7.3.1, since for γ ∈ C0∞ lim δ l−n sup D1,l γ; Bδ (x)L1 = 0

δ→0

x∈Rn

and lim

sup rl−n D1,l γ; Br (x)L1 = 0 .

|x|→∞ r∈(0,1)

 ˚ W l without approximation by functions in We can describe the space M p The following assertion, supplementing Theorem 7.2.9, holds.

C0∞ .

˚ Wpl if and only if γ is a continTheorem 7.3.10. A function γ belongs to M uous function vanishing at infinity and satisfying one of the conditions: (i) If lp ≤ n and p > 1, then Dp,l γ; eLp = o(1) 1/p {e : d(e)≤δ} [Cp,l (e)] sup

and

Dp,l γ; eLp = o(1) 1/p {e⊂Rn \Br : d(e)≤1} [Cp,l (e)] sup

as

as

δ→0

(7.3.29)

r → ∞.

(7.3.30)

7.3 Two-Sided Estimates for the Essential Norm in the Case m = l

283

(ii) If l ≤ n, then sup δ l−n D1,l γ; Bδ (x)L1 = o(1)

as

x∈Rn

δ→0

(7.3.31)

and sup rl−n D1,l γ; Br (x)L1 = o(1)

as

|x| → ∞ .

(7.3.32)

r∈(0,1)

(iii) If lp > n and p > 1, or l ≥ n and p = 1, then Dp,l γ; B1 (x)Lp = o(1)

as

|x| → ∞ .

(7.3.33)

˚ W l and let {γj } be a sequence of functions in C ∞ Proof. Necessity. Let γ ∈ M p 0 l approximating γ in M Wp . It follows from the expressions for equivalent norms in M Wpl derived in Chap. 4 that the left-hand sides of (7.3.29)–(7.3.33), with γ replaced by γ − γj , are arbitrarily small for sufficiently large j. On the other hand, it has been shown in the proof of Theorem 7.3.9 that (7.3.29)–(7.3.33) hold for γj ∈ C0∞ . Consequently, they hold for γ as well. Sufficiency. Let a function γ ∈ C ∩ M Wpl satisfy one of the conditions (7.3.29)–(7.3.33) and let γ(x) → 0 as |x| → ∞. For lp > n, p > 1 and for p = 1 the possibility of approximation of γ by mollifications of functions ζr γ, r → ∞, immediately follows from the expressions for the norm in M Wpl derived in Chap. 4. Consider the case lp ≤ n, p > 1. Then ζr γM Wpl → 0

as r → ∞

because of (7.3.19). Therefore, it suffices to approximate the multiplier γ with support in Br/2 for a fixed r by functions from C0∞ . Let γρ be a mollification of γ with nonnegative kernel K and radius ρ. We introduce the operators   (ρ) ϕ(j) P (j) , T∗ = γρ ϕ(j) P (j) . T∗ = γ j

j

Here we retain the same notation as in the definition of the operator T∗ in the proof of Theorem 7.2.1. Obviously, (ρ)

(ρ)

γ − γρ M Wpl ≤ (γ − γρ ) − (T∗ − T∗ )Wpl →Wpl + T∗ − T∗ Wpl →Wpl . For lp < n as well as for lp = n, 0 < l < 1, we have (ρ)

(γ − γρ ) − (T∗ − T∗ )Wpl →Wpl
 Dp,l (γ − γρ ); eLp ≤c sup + γ − γρ L∞ 1/p [Cp,l (e)] {e : d(e)≤δ}

(7.3.34)

(see the proof of Theorem 7.2.2, where the restriction l < m is insignificant, and the proof of Theorem 7.3.3). The right-hand side of (7.3.34) does not exceed

284

7 Essential Norm and Compactness of Multipliers


c


D γ; e  Dp,l γρ ; eLp  p,l Lp + γ − γ . +  ρ L ∞ 1/p [Cp,l (e)]1/p {e : d(e)≤δ} [Cp,l (e)] sup

Replacing here c by 2c, we can omit the second term because of the estimate  Dp,l γρ ; eLp ≤ ρ−n K(ξ/ρ)Dp,l γ; eξ Lp dξ, where eξ = {x : x + ξ ∈ e}. Further, we note that (ρ)

T∗ − T∗ Wpl →Wpl ≤ c(δ, r)γ − γρ ; Br M Wpl . Consequently, lim sup γ − γρ M Wpl ≤ 2c ρ→0

Dp,l γ; eLp 1/p {e : d(e)≤δ} [Cp,l (e)] sup

and it remains to make use of (7.3.29). For lp = n, l > 1, the proof follows the same lines provided that (7.3.34) is replaced by the estimate (ρ)




(γ − γρ ) − (T∗ − T∗ )Wpl →Wpl ≤ c

+

sup {e : d(e)≤δ}

sup {e : d(e)≤δ}

(see the proof of Theorem 7.3.5).

Dp,l (γ − γρ ); eLp [Cp,l (e)]1/p ∇(γ − γρ ); eLp [Cp,1

(e)]1/p

+ γ − γρ L∞





8 Traces and Extensions of Multipliers

8.1 Introduction Let Rn+ denote the upper half-space {z = (x, y) : x ∈ Rn−1 , y > 0}. We introduce the weighted Sobolev space Wps,α (Rn+ ) with the norm (min{1, y})α ∇s U ; Rn+ Lp + (min{1, y})α U ; Rn+ Lp ,

(8.1.1)

where s is nonnegative integer. We always assume that −1 < αp < p − 1. Obviously, the usual Sobolev space Wps (Rn+ ) is included here as Wps,0 (Rn+ ). It is well known that the fractional Sobolev space Wpl (Rn−1 ) is the space of traces on Rn−1 of functions in Wps,α (Rn+ ), where s = [l] + 1, α = 1 − {l} − 1/p, and p ∈ (1, ∞) (see [Usp]). In Sects. 8.2–8.5 we show that a similar trace characterization holds for spaces of multipliers acting in a pair of fractional Sobolev spaces. To be precise, we prove that, for all noninteger m and l with m ≥ l > 0, the multiplier space M (Wpm (Rn−1 ) → Wpl (Rn−1 )) is the space of traces on Rn−1 of functions in M (Wpt,β (Rn+ ) → Wps,α (Rn+ )), where s and α are as above and β = 1 − {m} − 1/p, t = [m] + 1. Sect. 8.6 concerns traces of multipliers on the smooth boundary of a domain. The remaining Sects. 8.7–8.9 are devoted to three trace and extension theorems for multipliers preserving a certain Sobolev-type space.

8.2 Multipliers in Pairs of Weighted Sobolev Spaces in Rn + We introduce the notion of (p, s, α)-capacity of a compact set e ⊂ Rn+ : Cp,s,α (e) = inf{U ; Rn+ pWps,α : U ∈ C0∞ (Rn+ ), U ≥ 1 on e}. The following result is known (see [Maz15], Sects. 8.1, 8.2). V.G. Maz’ya, T.O. Shaposhnikova, Theory of Sobolev Multipliers, Grundlehren der mathematischen Wissenschaften 337, c Springer-Verlag Berlin Hiedelberg 2009


285

286

8 Traces and Extensions of Multipliers

Proposition 8.2.1. Let k be a nonnegative integer, −1 < βp < p − 1, and let 1 < p < ∞. Then Γ ∈ M (Wpk,β (Rn+ ) → Wp0,α (Rn+ )) if and only if sup e⊂Rn + d(e)≤1

(min{1, y})α Γ ; eLp < ∞, (Cp,k,β (e))1/p

where d(e) is the diameter of e. The equivalence relation Γ M (Wpk,β →Wp0,α ) ∼ sup

e⊂Rn +

(min{1, y})α Γ ; eLp (Cp,k,β (e))1/p

(8.2.1)

d(e)≤1

holds. We shall use some general properties of multipliers in weighted Sobolev spaces. We start with the inequality Γ M (Wpt−j,β →Wps−j,α ) (s−j)/s j/s Γ  , M (Wpt,β →Wps,α ) M (Wpt−s,β →Wp0,α )

≤ c Γ 

(8.2.2)

where 0 ≤ j ≤ s, −1 < αp < p − 1, and −1 < βp < p − 1, which follows from the interpolation property of weighted Sobolev spaces (see [Tr4], Sect. 3.4.2). in notations of spaces, In this section and in Sects. 8.3, 8.4 we omit Rn+1 + (d) norms, and integrals, when it causes no ambiguity. The notations Br (x) = (n) {z ∈ Rd : |z − x| < r} and Br (x) = Br (x) will be adopted. The next assertion contains inequalities between multipliers and their mollifiers in x. Lemma 8.2.1. Let Γρ denote a mollifier of a function Γ defined by  K(ρ−1 (x − ξ))Γ (ξ, y)dξ, Γρ (x, y) = ρ−n+1 Rn−1

where K ∈ C0∞ (B1

(n−1)

), K ≥ 0, and K; Rn−1 L1 = 1. Then

Γρ M (Wpt,β →Wps,α ) ≤ Γ M (Wpt,β →Wps,α ) ≤ lim inf Γρ M (Wpt,β →Wps,α ) . ρ→0

(8.2.3)

Proof. Let U ∈ C0∞ . By Minkowski’s inequality 
 p 1/p pα (min{1, y}) ∇j,z ρ−n K(ξ/ρ)Γ (x − ξ, y)U (x, y)dξ dz Rn +

 ≤

Rn−1

−n

ρ Rn−1


 K(ξ/ρ) Rn +

1/p   (min{1, y})pα |∇j,z Γ (x, y)U (x + ξ, y) |p dz dξ,

8.2 Multipliers in Pairs of Weighted Sobolev Spaces in Rn +

287

where either j = 0 or j = s. Therefore, Γρ uWps,α ≤ Γ M (Wpt,β →Wps,α )  )
 1/p −n × ρ K(ξ/ρ) (min{1, y})pβ |∇t,z U (x + ξ, y)|p dz Rn−1


 +

Rn +

Rn +

(min{1, y})pβ |U (x + ξ, y)|p dz

1/p * dξ

≤ Γ M (Wpt,β →Wps,α ) U Wpt,β . This gives the left inequality (8.2.3). The right inequality (8.2.3) follows from Γ uWps,α = lim inf Γρ U Wps,α ≤ lim inf Γρ M (Wpt,β →Wps,α ) U Wpt,β . ρ→0

ρ→0



The proof is complete.

Lemma 8.2.2. Let Γ ∈ Lp,loc , p ∈ (1, ∞), −1 < βp < p − 1, and let U be an arbitrary function in C0∞ (Rn+ ). The best constant in the inequality (min{1, y})α Γ ∇s U Lp + (min{1, y})α Γ U Lp ≤ C U Wpt,β

(8.2.4)

is equivalent to the norm Γ M (Wpt−s,β →Wp0,α ) . Proof. The estimate C ≤ c Γ M (Wpt−s,β →Wp0,α ) is obvious. To derive the converse estimate, we introduce a function x → σ which is positive on [0, ∞) and equal to x for x > 1. For any U ∈ C0∞ (Rn+ ) we have  −[l]−1 u + T (−∆)u, U = (−∆)s σ(−∆) where T is a function in C0∞ ([0, ∞)). Since (−∆)s = (−1)s

 s! D2τ , τ!

|τ |=s

it follows from (8.2.4) and the theorem on the boundedness of convolution operators in weighted Lp spaces (see [And]) that  (min{1, y})pα |Γ (z)U (z)|p dz Rn +

 ≤ c C ∇s (σ(−∆))−s U p

Wpt,β

The proof is complete.

+ T U p

Wpt,β



≤ c CU p

Wpt−s,β

. 

288

8 Traces and Extensions of Multipliers

8.3 Characterization of M (Wpt,β → Wps,α) Here we derive necessary and sufficient conditions for a function to belong to the space M (Wpt,β → Wps,α ) for p ∈ (1, ∞) with α and β satisfying −1 < αp < p − 1,

−1 < βp < p − 1,

t ≥ s.

(8.3.1)

These inequalities will be assumed throughout. We start with an assertion on derivatives of multipliers. Lemma 8.3.1. Suppose that Γ ∈ M (Wpt,β → Wps,α ) ∩ M (Wpt−s,β → Wp0,α ), s−|σ|,α

Then Dσ Γ ∈ M (Wpt,β → Wp and

p ∈ (1, ∞).

) for any multi-index σ of order |σ| ≤ s,

Dσ Γ M (Wpt,β →Wps−|σ|,α ) ≤ ε Γ M (Wpt−s,β →Wp0,α ) + c(ε) Γ M (Wpt,β →Wps,α ) ,

(8.3.2)

where ε is an arbitrary positive number. Proof. Let U ∈ Wps,α and let ϕ be an arbitrary function in C0∞ . Applying the Leibniz formula  σ! Dσ (ϕU ) = Dτ ϕDσ−τ U, τ !(σ − τ )! {τ :σ≥τ ≥0}

we obtain   ϕU (−D)σ Γ dz = Γ Dσ (ϕU )dz =

 {τ :σ≥τ ≥0}

 =



ϕ

{β:σ≥τ ≥0}

σ! Γ Dτ ϕDσ−τ U dz τ !(σ − τ )!

σ! (−D)τ (Γ Dσ−τ U )dz. τ !(σ − τ )!

Therefore, U Dσ Γ =

 {τ :σ≥τ ≥0}

σ! (D)τ (Γ (−D)σ−τ U ), τ !(σ − τ )!

which implies the estimate U Dσ Γ Wps−|σ|,α ≤ c

 {τ :σ≥τ ≥0}

Γ Dσ−τ U Wps−|σ|+|τ |,α .

8.3 Characterization of M (Wpt,β → Wps,α )

289

Hence, it suffices to prove (8.3.2) for |σ| = 1. We have U ∇Γ Wps−1,α ≤ U Γ Wps,α + Γ ∇U Wps−1,α   ≤ Γ M (Wpt,β →Wps,α ) + Γ M (Wpt−1,β →Wps−1,α ) U Wpt,β . Estimating the norm Γ M (Wpt−1,β →Wps−1,α ) by (8.2.2), we arrive at (8.3.2).  Now we pass to two-sided estimates of the norms in M (Wpt,β → Wps,α ), p ∈ (1, ∞), given in terms of the spaces M (Wpk,β → Wp0,α ). We start with lower estimates. Lemma 8.3.2. Let Γ ∈ M (Wpt,β → Wps,α ). Then ∇s Γ M (Wpt,β →Wp0,α ) + Γ M (Wpt−s,β →Wp0,α ) ≤ c Γ M (Wpt,β →Wps,α ) . (8.3.3) Proof. Suppose first that Γ ∈ M (Wpt−s,β → Wp0,α ). We have 

Γ ∇s U Wp0,α ≤ Γ M (Wpt,β →Wps,α ) U Wpt,β + c

Dσ U Dτ Γ Wp0,α

|σ|+|τ |=s, τ =0

s 
 ≤ Γ M (Wpt,β →Wps,α ) + c ∇j Γ M (Wpt−s+j →Wp0,α ) U Wpt,β .

(8.3.4)

j=1

By Lemma 8.3.1, ∇j Γ M (Wpt−s+j,β →Wp0,α ) ≤ ε Γ M (Wpt−s,β →Wp0,α ) + c(ε) Γ M (Wpt−s+j,β →Wpj,α ) .

(8.3.5)

Estimating the last norm by (8.2.2), we obtain ∇j Γ M (Wpt−s+j,β →Wp0,α ) ≤ ε Γ M (Wpt−s,β →Wp0,α ) + c(ε) Γ M (Wpt,β →Wps,α ) . Substitution of this inequality into (8.3.4) gives
Γ ∇s U Wp0,α ≤ ε Γ M (Wpt−s,β →Wp0,α )  + c(ε) Γ M (Wpt,β →Wps,α ) U Wpt,β .

(8.3.6)

Also, Γ U Wp0,α ≤ Γ M (Wpt,β →Wps,α ) U Wpt,β .

(8.3.7)

Adding the last two estimates and applying Lemma 8.2.2, we arrive at Γ M (Wpt−s,β →Wp0,α ) ≤ ε Γ M (Wpt−s,β →Wp0,α ) + c(ε) Γ M (Wpt,β →Wps,α ) .

290

8 Traces and Extensions of Multipliers

Hence, Γ M (Wpt−s,β →Wp0,α ) ≤ c Γ M (Wpt,β →Wps,α ) .

(8.3.8)

Now we remove the assumption Γ ∈ M (Wpt−s,β → Wp0,α ). Since Γ ∈ M (Wpt,β → Wps,α ), it follows that Γ ηWps,α ≤ c ηWpt,β , where η ∈ C0∞ (B2 (z)), η = 1 on B1 (z), and z is an arbitrary point in Rn+ . s,α (Rn+ ), which implies that for any (n − 1)-dimensional multiHence Γ ∈ Wp,unif s,α index τ the derivative Dxτ Γρ belongs to Wp,unif (Rn+ ). Therefore, Γρ ∈ L∞ (Rn+ ) which, in its turn, guarantees that Γρ ∈ M (Wpt−s,β → Wp0,α ). Thus, we may put Γρ into (8.3.8) to obtain (n)

(n)

Γρ M (Wpt−s,β →Wp0,α ) ≤ c Γρ M (Wpt,β →Wps,α ) . Letting ρ → 0 and using Lemma 8.2.1, we arrive at (8.3.8) for all Γ ∈ M (Wpt,β → Wps,α ). To estimate the first term on the right-hand side of (8.3.3), we combine (8.3.8) with (8.3.5) for j = s.  The estimate opposite to (8.3.3) is contained in the following lemma. Lemma 8.3.3. Let Γ ∈ M (Wpt−s,β → Wp0,α ) and let ∇s Γ ∈ M (Wpt,β → Wp0,α ). Then Γ ∈ M (Wpt,β → Wps,α ) and the estimate 
Γ M (Wpt,β →Wps,α ) ≤ c ∇s Γ M (Wpt,β →Wp0,α ) + Γ M (Wpt−s,β →Wp0,α ) (8.3.9) holds. Proof. By Lemma 8.3.2 and (8.2.2) we have  ∇j Γ M (Wpt−s+j,β →Wp0,α ) ≤ c Γ M (Wpt−s+j,β →Wpj,α ) j/s 1−j/s Γ  , M (Wpt,β →Wps,α ) M (Wpt−s,β →Wp0,α )

≤ c Γ 

(8.3.10)

where j = 1, . . . , s. For any U ∈ C0∞ (Rn+ ), (min{1, y})α ∇s (Γ U )Lp ≤ c

s 

(min{1, y})α |∇j Γ | |∇s−j U | Lp

j=0


≤ c ∇s Γ M (Wpt,β →Wp0,α ) + Γ M (Wpt−s,β →Wp0,α ) +

s−1  j=1

 ∇j Γ M (Wpt−s+j,β →Wp0,α ) U Wpt,β .

8.3 Characterization of M (Wpt,β → Wps,α )

291

This and (8.3.10) imply that  (min{1, y})α ∇s (Γ U )Lp 
≤ c ∇s Γ M (Wpt,β →Wp0,α ) + Γ M (Wpt−s,β →Wp0,α ) U Wpt,β . It remains to note that (min{1, y})α Γ U Lp ≤ Γ M (Wpt−s,β →Wp0,α ) U Wpt−s,β . 

The proof is complete.

Using Lemmas 8.3.2 and 8.3.3, we arrive at the following description of the space M (Wpt,β (Rn+ ) → Wps,α (Rn+ )). Theorem 8.3.1. A function Γ belongs to the space M (Wpt,β → Wps,α ) if and s,α , Γ ∈ M (Wpt−s,β → Wp0,α ), and ∇s Γ ∈ M (Wpt,β → Wp0,α ). only if Γ ∈ Wp,loc Moreover, Γ M (Wpt,β →Wps,α ) ∼ ∇s Γ M (Wpt,β →Wp0,α ) + Γ M (Wpt−s,β →Wp0,α ) . The equivalence relation (8.2.1) enables us to reformulate Theorem 8.3.1 as follows. Theorem 8.3.2. A function Γ belongs to the space M (Wpt,β → Wps,α ) if and s,α , and, for any compact set e ⊂ Rn+ , only if Γ ∈ Wp,loc (min{1, y})α ∇s Γ ; epLp ≤ c Cp,t,β (e) and (min{1, y})α Γ ; epLp ≤ c Cp,t−s,β (e). Moreover, Γ M (Wpt,β →Wps,α )

∼ sup

e⊂Rn +


(min{1, y})α ∇ Γ ; e (min{1, y})α Γ ; eLp  s Lp , + (Cp,t,β (e))1/p (Cp,t−s,β (e))1/p

(8.3.11)

d(e)≤1

where d(e) is the diameter of e. We formulate the important particular case of Theorem 8.3.2 when t = s. Corollary 8.3.1. A function Γ belongs to the space M Wps,α if and only if s,α and, for any compact set e ⊂ Rn+ , Γ ∈ Wp,loc (min{1, y})α ∇s Γ ; epLp ≤ c Cp,s,α (e). Moreover, Γ M Wps,α ∼ sup

e⊂Rn +

d(e)≤1

(min{1, y})α ∇s Γ ; eLp + Γ L∞ . (Cp,s,α (e))1/p

(8.3.12)

292

8 Traces and Extensions of Multipliers

8.4 Auxiliary Estimates for an Extension Operator 8.4.1 Pointwise Estimates for T γ and ∇T γ For functions γ ∈ L1,unif (Rn−1 ), we introduce the operator T by 
x − ξ  γ(ξ)dξ, (x, y) ∈ Rn+ , ζ (T γ)(x, y) = y 1−n y n−1 R

(8.4.1)

where ζ is a continuously differentiable function defined on Rn+ . We assume that (8.4.2) (|z| + 1)|∇ζ(z)| + |ζ(z)| ≤ C (|z| + 1)−n 

and that

ζ(z)dz = 1.

(8.4.3)

Rn−1

Lemma 8.4.1. Let γ ∈ M (Wpm−l (Rn−1 ) → Lp (Rn−1 )), where m ≥ l and 1 < p < ∞. Then |T γ(z)| + y|∇(T γ(z))| ≤ c (1 + y l−m )γ; Rn−1 M (Wpm−l →Lp ) . Proof. By (8.4.2), ≤ c y 1−n

|T γ(z)| + y|∇(T γ(z))|  n |γ(ξ)|dξ + y


 (n−1)

By

(n−1)

Rn−1 \By

(x)

(x)

|γ(ξ)|dξ  . |ξ − x|n

By H¨older’s inequality,  |γ(ξ)|dξ ≤ cy (n−1)(p−1)/p γ; By(n−1) (x)Lp . (n−1)

By

(8.4.4)

(8.4.5)

(x)

Let y ∈ (0, 1). The right-hand side in (8.4.5) does not exceed c y −m+l+n−1 sup (1 + rm−l−

n−1 p

r∈(0,1) x∈Rn−1

) γ; Br(n−1) (x)Lp .

This, being combined with (8.4.5) and (4.3.15), implies that  |γ(ξ)|dξ ≤ c y −m+l+n−1 γ; Rn−1 M (Wpm−l →Lp ) (n−1)

By

(8.4.6)

(x)

for y < 1. Suppose that y > 1. By (4.3.12), (n−1)

γ; Rn−1 M (Wpm−l →Lp ) ≥ c sup  y>1

γ; By

≥ c sup y (1−n)/p γ; By(n−1) Lp , y>1

Lp

(n−1) 1/p Cp,m−l (By )

(8.4.7)

8.4 Auxiliary Estimates for an Extension Operator

293

because Cp,m−l (By(n−1) (x)) ∼ y n−1 for y > 1. Combining (8.4.7) with (8.4.5), we have  1−n |γ(ξ)|dξ ≤ c y −(n−1)/p γ; By(n−1) Lp y (n−1)

By

(x)

≤ c γ; Rn−1 M (Wpm−l →Lp ) . Thus, (8.4.6) and (8.4.8) give  1−n |γ(ξ)|dξ ≤ c (1 + y l−m )γ; Rn−1 M (Wpm−l →Lp ) y (n−1)

By

(8.4.8)

(8.4.9)

(x)

for any y > 0. Now we estimate the second integral on the right-hand side of (8.4.4). Clearly,   ∞  |γ(ξ)|dξ dρ ≤n |γ(ξ)|dξ. (8.4.10) n (n−1) ρn+1 Bρ(n−1) (x) y Rn−1 \By (x) |ξ − x| By H¨older’s inequality the right-hand side of (8.4.10) has the majorant  ∞ n−1 ρ−2− p γ; Bρ(n−1) (x)Lp dρ. (8.4.11) c y

Let y > 1. Then by (8.4.7), the function (8.4.11) does not exceed c y −1 γ; Rn−1 M (Wpm−l →Lp ) . Suppose that y < 1. Then  1 n−1 ρ−2− p γ; Bρ(n−1) (x)Lp dρ y

≤ cy −m+l−1 sup (1 + rm−1− r∈(0,1) x∈Rn−1

n−1 p

) γ; Br(n−1) (x)Lp

which is dominated by c y −m+l−1 γ; Rn−1 M (Wpm−l →Lp ) owing to (4.3.15). Furthermore, (8.4.7) implies that  ∞ n−1 ρ−2− p γ; Bρ(n−1) (x)Lp dρ ≤ c γ; Rn−1 M (Wpm−l →Lp ) . 1

(8.4.12)

294

8 Traces and Extensions of Multipliers

Hence, for y < 1  ∞ n−1 ρ−2− p γ; Bρn−1 (x)Lp dρ ≤ c (1 + y −m+l−1 ) γ; Rn−1 M (Wpm−l →Lp ) . 1

This, in combination with the case y > 1, implies that the integral (8.4.11) does not exceed cy −1 (1 + y l−m )γ; Rn−1 M (Wpm−l →Lp ) for all y > 0. Thus, the result follows from (8.4.9), (8.4.10), and (8.4.4).



8.4.2 Weighted Lp -Estimates for T γ and ∇T γ Lemma 8.4.2. Let the extension operator T be defined by (8.4.1). Suppose that γ ∈ M (Wpm−l (Rn−1 ) → Lp (Rn−1 )), where l ∈ (0, 1), [m] ≥ 1, and 1 < p < ∞. Then, for k = 1, . . . , [m],
 1  p 1/p y p(k−l)−1 |T γ(z)| + y|∇(T γ)(z)| dy 0 k−l

m−k

m−l ≤ c γ; Rn−1 M [(Mγ)(x)] m−l , (W m−l →L ) p

(8.4.13)

p

where M is the Hardy–Littlewood maximal operator in Rn−1 . Proof. Let δ be a number in (0,1] to be chosen later. We set  1  δ   p p(k−l)−1 |T γ| + y|∇(T γ)(z)| dy = y . . . dy + 0

0

By (8.4.4),  δ



δ

. . . dy ≤ c 0

y p(k+1−l−n)−1 

0 δ

y p(k+1−l)−1

+c


 (n−1)




By

(x)

(n−1)

. . . dy.

δ

|γ(ξ)|dξ

Rn−1 \By

0

1

(x)

p dy

p |γ(ξ)| dξ dy. |ξ − x|n

From the definition of M it follows that  δ
 p y p(k+1−l−n)−1 |γ(ξ)|dξ dy ≤ c [(Mγ)(x)]p δ p(k−l) . (n−1)

By

0

Using (8.4.10), we obtain  δ
 y p(k+1−l)−1 0

(8.4.14)

(x)

(n−1)

Rn−1 \By

≤ c [(Mγ)(x)] δ

(x)

p |γ(ξ)| dξ dy n |ξ − x|

p p(k−l)

.

(8.4.15)

8.4 Auxiliary Estimates for an Extension Operator

295

Combining (8.4.14) and (8.4.15), we conclude that 

δ

. . . dy ≤ c [(Mγ)(x)]p δ p(k−l) .

(8.4.16)

0

By Lemma 8.4.1, 

1

 p y p(k−l)−1 |T γ| + y|∇(T γ)(z)| dy

δ

≤ c γ; Rn−1 pM (W m−l →L ) δ p(k−m) . p

(8.4.17)

p

Adding (8.4.16) and (8.4.17), we find that 

1

 p y p(k−l)−1 |T γ| + y|∇(T γ)(z)| dy

0

  ≤ c [(Mγ)(x)]p δ p(k−l) + γ; Rn−1 pM (W m−l →L ) δ p(k−m) . p

p

The right-hand side in this inequality attains its minimum value for δ =


γ; Rn−1 

M (Wpm−l →Lp )

1/(m−l) .

(Mγ)(x)



The proof is complete.

Lemma 8.4.3. Let the operator T be defined by (8.4.1) and let 0 < l < 1. Then  1  p y p(1−l)−1 |∇(T γ)(z)|p dy ≤ c (Dp,l γ)(x) . 0

Proof. Let R(ξ, x) = γ(ξ) − γ(x). Using the identity 
ξ − x dξ = const, y −n+1 ζ y Rn−1 we have ∂T γ ∂
−n+1 (x, y) = y ∂y ∂y

 ζ Rn−1


ξ − x y

 R(ξ, x)dξ .

Furthermore, it is clear that ∂T γ (x, y) = y −n+1 ∂xj

 R(ξ, x) Rn−1

∂
ξ − x dξ. ζ ∂xj y

(8.4.18)

296

8 Traces and Extensions of Multipliers

Therefore, |∇(T γ)(x, y)| ≤ c y −n

1   k=0


k
∇k ζ ξ − x 1 + |ξ − x| |R(ξ, x)|dξ. y y Rn−1

This estimate and (8.4.2) imply that 
|ξ − x| −n |R(ξ, x)|dξ |∇(T γ)(x, y)| ≤ c y −n 1+ y Rn−1 
|ξ − x| n−1/p
|ξ − x| −n |R(ξ, x)| −1/p 1+ = cy dξ. y y |ξ − x|n−1/p Rn−1 Consequently,  1 y p(1−l)−1 |∇(T γ)(x, y)|p dy 0

≤ c

 1
 f Rn−1

0


|ξ − x|  |R(ξ, x)| p dy dξ y p(1−l)−1 , n−1/p y y |ξ − x|

where f (η) = η n−1/p (1 + η)−n . We write the last integral over (0, 1) as 




dt p p(1−l)−1 dy y y t y 0 0  1
 ∞ ds p p(1−l)−1 dy , f (s)g(sy, x) y = s y 0 0 1

∞

f


t

g(t, x)

(8.4.19)



with g(t, x) = t1/p−1

(n−1)

|R(tθ + x, x)|dθ.

∂B1

By Minkowski’s inequality, the right-hand side of (8.4.19) does not exceed


∞
 1

0




(f (s))p (g(sy, x))p y p(1−l)−1

0

dy 1/p ds p y s


 s dτ 1/p ds p = f (s) (g(τ, x))p τ p(1−l)−1 τ s2−l−1/p 0 0  
∞ dτ ds p ∞ ≤ f (s) 2−l−1/p (g(τ, x))p τ p(1−l)−1 . τ s 0 0 ∞

Therefore,  1



∞

y p(1−l)−1 |∇(T γ)(x, y)|p dy ≤ c 0

0

(g(τ, x))p τ p(1−l)−1

dτ . τ

(8.4.20)

8.5 Trace Theorem for the Space M (Wpt,β → Wps,α )

297

It remains to note that  ∞   ∞  p dτ p p(1−l)−1 dτ −pl = (g(τ, x)) τ τ |γ(τ θ + x) − γ(x)|dθ (n−1) τ τ ∂B1 0 0 

∞

≤c =c

 |γ(τ θ + x) − γ(x)| dθ p

(n−1)

0

∂B1





dτ τ pl+1

 ≤c

Rn−1

|γ(x + h) − γ(x)|p dh |h|pl+n−1

p Dp,l γ (x) . 

The result follows.

8.5 Trace Theorem for the Space M (Wpt,β → Wps,α) Theorem 8.5.1. (i) Let m and l be positive nonintegers with m ≥ l, and let Γ ∈ M (Wpt,β (Rn+ ) → Wps,α (Rn+ )) where t = [m] + 1, s = [l] + 1, β = 1 − {m} − 1/p, and α = 1 − {l} − 1/p. If γ is the trace of Γ on Rn−1 , then γ ∈ M (Wpm (Rn−1 ) → Wpl (Rn−1 )) and the estimate γ; Rn−1 M (Wpm →Wpl ) ≤ c Γ ; Rn+ M (Wpt,β →Wps,α )

(8.5.1)

holds. (ii) Let γ ∈ M (Wpm (Rn−1 ) → Wpl (Rn−1 )). Then the Dirichlet problem ∆Γ = 0 on Rn+ , Γ |Rn−1 = γ

(8.5.2)

has a unique solution in M (Wpt,β (Rn+ ) → Wps,α (Rn+ )), where t = [m] + 1, s = [l] + 1, β = 1 − {m} − 1/p, and α = 1 − {l} − 1/p. The estimate Γ ; Rn+ M (Wpt,β →Wps,α ) ≤ c γ; Rn−1 M (Wpm →Wpl )

(8.5.3)

holds. Proof. We start with (i). Let U ∈ Wpt,β (Rn+ ) and let u be the trace of U on Rn−1 . By setting Γ U and γu instead of U and u, respectively, in the inequality u; Rn−1 Wpl ≤ c U ; Rn+ Wps,α ,

298

8 Traces and Extensions of Multipliers

we obtain the estimate γu; Rn−1 Wpl ≤ c Γ ; Rn+ M (Wpt,β →Wps,α ) U ; Rn+ Wpt,β . Minimizing the right-hand side over all extensions U of u, we obtain γu; Rn−1 Wpl ≤ c Γ ; Rn+ M (Wpt,β →Wps,α ) u; Rn−1 Wpm which gives (8.5.1). The proof of (ii) will be given separately for l < 1 and for l > 1 in Sects. 8.5.1 and 8.5.2. 

8.5.1 The Case l < 1 Our aim now is to prove that for l < 1 and s = 1 the operator T de[m]+1,β (Rn+ ) → fined by (8.4.1) maps M (Wpm (Rn−1 ) → Wpl (Rn−1 )) into M (Wp 1,α n Wp (R+ )) with α = 1 − l − 1/p, β = 1 − {m} − 1/p, and that T γ; Rn+ M (Wpt,β →Wp1,α ) ≤ c Cγ; Rn−1 M (Wpm →Wpl ) , where C is the constant in (8.4.2). We have  (min{1, y})α ∇(U T γ); Rn+ pLp ≤ c



1

∞

+c 1

Rn−1

 |∇(T γ)|p |U |p +|T γ|p |∇U |p dz

Rn−1

0





y pα

(8.5.4)

  |∇(T γ)|p |U |p + |T γ|p |∇U |p dz



=c

 . . . dz + c

0 1 y|∇(T γ)(z)| + |(T γ)(z)| ≤ c γ; Rn−1 M (Wpm−l →Lp ) . Hence

 y>1

. . . dz ≤ c γ; Rn−1 pM (W m−l →L ) U ; Rn+ pW 1,α . p

p

p

It remains to refer to the estimate U ; Rn+ Wp1,α ≤ c U ; Rn+ Wpt,β which follows from the one-dimensional Hardy inequality.

(8.5.6)

8.5 Trace Theorem for the Space M (Wpt,β → Wps,α )

299

Introducing the notation [m]  ∂k yk R0 U (z) = U (z) − , U (x, 0) ∂y k k! k=0

 ∂k yk ∇U (x, 0) ∂y k k!

[m]−1

R1 U (z) = ∇U (z) −

for m > 1,

k=0

and R1 U (z) = ∇U (z)

for m < 1,

we have 

 . . . dz ≤ c

y p(1−l)−1

0 ϕ(x)}, where ϕ is a function satisfying the Lipschitz condition. We find necessary and sufficient conditions for a function to belong to the space M (Wpm (G) → Wpl (G)), where m and l are integers with 0 ≤ l ≤ m. In Sect. 9.2 we show that the Stein extension operator (see [St2], Ch.6, §3) maps continuously M (Wpm (G) → Wpl (G)) into M (Wpm (Rn ) → Wpl (Rn )) . Analogous results for the space M (Wpm (Ω) → Wpl (Ω)), where Ω is a bounded domain with boundary in the Lipschitz class C 0,1 , are obtained in 9.3. A description of the space M L1p (Ω) is given, where L1p (Ω) = {u ∈ Lp,loc (Ω) : ∇u ∈ Lp (Ω)} and Ω is an arbitrary domain. We show that, in general, the restriction to Ω of a multiplier in Wp1 (Rn ) is not a multiplier in Wp1 (Ω). Further, in Sect. 9.4 we study the influence of a change of variables upon Sobolev spaces. Here we introduce classes of mappings ((p, l)-diffeomorphisms) which preserve the space Wpl , as well as classes of non-smooth manifolds on which the space Wpl is correctly defined. These definitions of mappings and manifolds involve spaces of multipliers. In conclusion, a change of variables Tpm,l acting in the pair of Sobolev spaces Wpm (V ) → Wpl (U ) is defined and investigated. In Sect. 9.5 we give the following modification of the classical implicit function theorem (see, for example, [KP]) which involves multipliers in its statement. We consider a function u in a special Lipschitz domain G and assume that ∇u ∈ M Wpl−1 (G), l ≥ 2, that u vanishes on ∂G, i.e., for y = ϕ(x), and that the trace of ∂u/∂y on ∂G is separated from zero. We show that l−1−1/p ∇ϕ ∈ M Wp (Rn−1 ). ˚pm (Ω) → Finally, in Sect. 9.6.2 we give a description of the space M (W l Wp (Ω)). V.G. Maz’ya, T.O. Shaposhnikova, Theory of Sobolev Multipliers, Grundlehren der mathematischen Wissenschaften 337, c Springer-Verlag Berlin Hiedelberg 2009


325

326

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

9.1 Multipliers in a Special Lipschitz Domain 9.1.1 Special Lipschitz Domains Let z = (x, y), where x ∈ Rn−1 and y ∈ R1 . By a special Lipschitz domain we mean G = {z ∈ Rn : x ∈ Rn−1 , y > ϕ(x)}, where ϕ is a function satisfying the Lipschitz condition |ϕ(x1 ) − ϕ(x2 )| ≤ L |x1 − x2 |. It is shown in [St2] (§3, Ch. 6) that there exists a function z → δ ∗ (z) with the properties: (i) δ ∗ ∈ C ∞ (Rn \∂G) and, for any multi-index α, |Dα δ ∗ | ≤ cα (δ ∗ )1−|α| , where cα are constants depending on L. (ii) For all z ∈ Rn \G, 2 [ϕ(x) − y] ≤ δ ∗ (z) ≤ a [ϕ(x) − y] ,

(9.1.1)

where a = const > 2. We introduce the operator C which performs an extension to the whole of Rn of a function f defined on G. Namely, if z ∈ Rn \G, then we put 

2

(Cf )(z) =

f (x, y + λδ ∗ (z))ψ(λ) dλ ,

(9.1.2)

1

where ψ is a function in C([1, 2]) such that 



2

1

2

λk ψ(λ) dλ = 0 ,

ψ(λ) dλ = 1 ,

k = 1, 2, . . . , l .

(9.1.3)

1

The operator C maps Wpl (G) continuously into Wpl (Rn ) (see [St2], Ch.6, §3). Let Br = {x ∈ Rn : |x| < r} and let K be a nonnegative function in ∞ C0 (B1 ) with support in the cone {z : y > 2L|x|}. With a function f defined on Rn we associate its mollification with radius h,  [K(h)f ](z) = f (z + hζ)K(ζ) dζ . (9.1.4) Rn

It is clear that if z ∈ G then [K(h)f ](z) depends only on the values of f in G. 9.1.2 Auxiliary Assertions Here, as in Sect. 9.1.1, G is a special Lipschitz domain in Rn .

9.1 Multipliers in a Special Lipschitz Domain

327

Lemma 9.1.1. Let w be a measurable nonnegative function defined on G and let m and l be integers with 0 ≤ l < m and 1 < p < ∞. The best constant C in the inequality    |∇l u(z)|p + |u(z)|p w(z) dz ≤ C u; GpWpm , (9.1.5) G

for all u ∈ Wpm , is equivalent to  w(z) dz L = sup

e

Cp,m−l (e)

,

(9.1.6)

where the supremum is taken over all compact subsets e of the domain G. Proof. We extend w by zero to the exterior of G. Then (9.1.5) implies the same inequality with G replaced by Rn . Now the desired lower estimate for the constant C follows immediately from Theorem 1.2.2 and Lemma 1.2.7. Let us obtain the upper bound for C. By wε we denote a function which coincides with w on the set {z ∈ G : dist(z, ∂G) > ε, w(z) < 1/ε},

ε > 0,

and vanishes elsewhere. By Lemma 1.2.7,   Rn

  |∇l v|p + |v|p wε dz ≤ c sup E

wε (z) dz E

Cp,m−l (E)

v; Rn pWpm

(9.1.7)

for all v ∈ C0∞ (Rn ), where the supremum is taken over all compact subsets of Rn . It follows from the definition of wε and the monotonicity of the capacity that this supremum does not exceed (9.1.6). Let u ∈ Wpm (G). Then Cu ∈ Wpm (Rn ). Approximating Cu by C0∞ (Rn )-functions in the norm of the space Wpm (Rn ), we obtain from (9.1.7) that    |∇l Cu|p + |Cu|p wε dz ≤ cL Cu; Rn pWpm . Rn

Since Cu = u in G, wε = 0 in Rn \G and the operator C : Wpm (G) → Wpm (Rn ) is continuous, it follows that  (|∇l u|p + |u|p )wε dz ≤ cL u; GpWpm . Rn

Passing to the limit on the left-hand side as ε → 0, we complete the proof.  The next assertion results directly from Lemma 9.1.1 and (1.2.8).

328

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

Corollary 9.1.1. The equivalence relation γ; eLp 1/p e⊂G [Cp,m (e)]

γ; GM (Wpm →Lp ) ∼ sup holds.

From the existence of the operator C and the interpolation property of Sobolev spaces in Rn it follows that the spaces Wpk (G) have the same interpolation property. In particular, (l−j)/l

j/l

γ; GM (Wpm−j →Wpl−j ) ≤ c γ; GM (W m →W l ) γ; GM (W m−l →L ) . (9.1.8) p

p

p

p

We introduce some notation. Let γ be a function defined on G, whose distributional derivatives of order k are locally integrable with power p. We put fk (γ; e) =

∇k γ; eLp , [Cp,m−l+k (e)]1/p

(9.1.9)

where m ≥ l, 0 ≤ k ≤ l and e is a compact subset of G. Further, let sk (γ) = sup fk (γ; e) ,

(9.1.10)

e⊂G

where the supremum is taken over all compact subsets of G of positive n-dimensional measure. If m = l, then s0 (γ) = γ; GL∞ . We note that the value sk (γ) does not change if e is replaced in its definition by any compact subset of G of positive n-dimensional measure. In fact, for any ε > 0 there exists a compact set E ⊂ G such that sup fk (γ; e) ≤ (1 + ε)fk (γ; E) .

(9.1.11)

e⊂G

Let Eδ = {z ∈ E : y ≥ ϕ(x) + δ}, δ > 0. It is clear that fk (γ; e) ≤

∇k γ; ELp . [Cp.m−l+k (Eδ )]1/p

Since for small δ, ∇k γ; ELp ≤ (1 + ε)∇k γ; Eδ Lp , it follows that fk (γ; E) ≤ (1 + ε)fk (γ; Eδ ) which, together with (9.1.11), yields sup fk (γ; e) ≤ (1 + ε)2 sup fk (γ; e) . e⊂G

e⊂G

It remains to make use of the arbitrariness of ε.



9.1 Multipliers in a Special Lipschitz Domain

329

Lemma 9.1.2. Let K(h)γ be the mollification of γ, given by (9.1.4). Then K(h)γ; GM (Wpm →Wpl ) ≤ γ; GM (Wpm →Wpl ) ≤ lim inf K(h)γ; GM (Wpm →Wpl ) , h→0

(9.1.12)

and lim inf sk (K(h)γ) ≥ sk (γ) ≥ c sk (K(h)γ) .

(9.1.13)

h→0

The proof of (9.1.12) is the same as that of Lemma 2.3.1. Inequality (9.1.13) follows from (9.1.12) with l = 0, and Corollary 9.1.1. 9.1.3 Description of the Space of Multipliers Theorem 9.1.1. Let m and l be integers with m ≥ l ≥ 0 and p ∈ (1, ∞). Then the space M (Wpm (G) → Wpl (G)) consists of functions γ which are locally integrable with power p, along with their distributional derivatives up to order l, and such that sl (γ) + s0 (γ) < ∞. The equivalence relation γ; GM (Wpm →Wpl ) ∼ sl (γ) + s0 (γ)

(9.1.14)

holds. The proof is similar to that of Theorem 2.3.2 except that we use Lemmas 9.1.1 and 9.1.2 as well as the interpolation inequality (9.1.8). From Theorem 9.1.1 and the inequality j/l

1−j/l

∇j γ; GM (Wpm−l+j →Lp ) ≤ c γ; GM (W m →W l ) γ; GM (W m−l →L ) , p

p

p

p

where j = 1, . . . , l − 1 (cf. (2.3.8)), it follows that γ; GM (Wpm →Wpl ) ∼

l 

sj (γ) .

(9.1.15)

j=0

Next we turn to the space M (W1m (G) → W1l (G)). Lemma 9.1.3. Let G be a special Lipschitz domain and let w be a measurable function defined on G. Then the best constant C in wu; GL1 ≤ Cu; GW1m ,

u ∈ W1m (G) ,

is equivalent to N=

sup z∈Rn ,ρ∈(0,1)

ρm−n w; Bρ (z) ∩ GL1 .

(9.1.16)

330

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

Proof. We extend w by zero to the exterior of G. Then, from (9.1.16) we obtain the same inequality with G replaced by Rn . Now the desired lower bound for the constant C follows from Theorem 2.2.3. The same theorem implies that wv; Rn L1 ≤ cN v; Rn W1m for all v ∈ W1m (Rn ). Minimizing the right-hand side over all extensions of u ∈ W1m (G), we arrive at (9.1.16) with the constant cN .  Remark 9.1.1. Obviously, replacing the condition ρ ∈ (0, 1) in the definition of N by ρ ∈ (0, C), where C is an arbitrary positive constant, we obtain an equivalent value. The same is true if z ∈ Rn is replaced by z ∈ G. Theorem 9.1.2. Let G be a special Lipschitz domain and let m and l be integers with 0 ≤ l ≤ m. The space M (W1m (G) → W1l (G)) consists of functions γ which are locally integrable in G together with their distributional derivatives of order l and such that ∇l γ; Bρ (z) ∩ GL1 + ρ−l γ; Bρ (z) ∩ GL1 ≤ c ρ−m+n for all z ∈ Rn , ρ ∈ (0, 1). The relation γ; GM (W1m →W1l ) ∼

sup

ρm−n

z∈Rn ,ρ∈(0,1)

l 

ρj−l ∇j γ; Bρ (z) ∩ GL1

j=0

holds. (Obviously, the same relation holds if we take z ∈ G on the right-hand side.) Proof. Let us substitute the function ζ → u(ζ) = Φ((ζ − z)/ρ), where z ∈ Rn , Φ ∈ C0∞ (B2 ), Φ = 1 on B1 , and ρ ∈ (0, 1), into the inequality γu; GW1l ≤ γ; GM (W1m →W1l ) u; GW1m .

(9.1.17)

Since, for j = 0, 1, . . . , l − 1, ρj−l ∇j (γu); B2ρ (z) ∩ GL1 ≤ c ∇l (γu); B2ρ (z) ∩ GL1 , it follows from (9.1.17) that ρj−l ∇j γ; Bρ (z) ∩ GL1 ≤ γ; GM (W1m →W1l ) ρ−m+n . Thus the required lower estimate for the norm in M (W1m (G) → W1l (G)) is obtained. The upper estimate results from the obvious inequality   |∇j u| |∇k γ|; GL1 γu; GW1l ≤ c 0≤k+j≤l

and the above lemma.



9.1 Multipliers in a Special Lipschitz Domain

331

The following theorem shows that the description of the space M (Wpm (G)→ is especially simple if either mp > n and p > 1, or m ≥ n and p = 1. This theorem can be derived from Theorems 9.1.1, 9.1.2, but we present a direct proof. Wpl (G))

Theorem 9.1.3. Let G be a special Lipschitz domain and let m and l be integers with 0 ≤ l ≤ m. Further, let either mp > n and p > 1, or m ≥ n and p = 1. Then the space M (Wpm (G) → Wpl (G)) consists of functions γ which are locally integrable with power p in G together with their distributional derivatives of order l and such that γ; B1 (z) ∩ GWpl ≤ const for any z ∈ Rn . Moreover, γ; GM (Wpm →Wpl ) ∼ sup γ; B1 (z) ∩ GWpl . z∈Rn

(Obviously, the same relation holds if we take z ∈ G on the right-hand side above.) Proof. Putting the function ζ → u(ζ) = ϕ(ζ − z), where z ∈ Rn , ϕ ∈ C0∞ (B2 ), and ϕ = 1 on B1 , into γu; GWpl ≤ γ; GM (Wpm →Wpl ) u; GWpm , we obtain γ; B1 ∩ GWpl ≤ c γ; GM (Wpm →Wpl ) . By {B (j) }j≥1 we denote a covering of Rn by unit balls with a finite multiplicity which depends only on n. We have γu; B (j) ∩ GWpl ≤ c

l 

 |∇i u| |∇l−i γ|; B (j) ∩ GLp

i=0

≤c

l 

∇i u; B (j) ∩ GLqi ∇l−i γ; B (j) ∩ GLqi p/(qi −p) ,

i=0

where qi = pn/[n − p(m − i)] if n > p(m − i), qi = ∞ if n < p(m − i) and qi is an arbitrary positive number in the case n = p(m − i). According to the Sobolev imbedding theorem, ∇i u; B (j) ∩ GLqi ≤ c u; B (j) ∩ GWpm , ∇l−i γ; B (j) ∩ GLqi p/(qi −p) ≤ c γ; B (j) ∩ GWpl . Consequently, γu; B (j) ∩ GpW l ≤ c γ; B (j) ∩ GpW l u; B (j) ∩ GpWpm . p

Summing over j and applying the inequality   aα aj )α , where aj ≥ 0, j ≤( we complete the proof.

(9.1.18)

p

α ≥ 1, 

332

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

9.2 Extension of Multipliers to the Complement of a Special Lipschitz Domain Theorem 9.2.1. Let γ ∈ M (Wpm (G) → Wpl (G)), where m and l are integers, 0 ≤ l ≤ m and 1 ≤ p < ∞. Then Cγ ∈ M (Wpm (Rn ) → Wpl (Rn )) and Cγ; Rn M (Wpm →Wpl ) ≤ c γ; GM (Wpm →Wpl ) .

(9.2.1)

Proof. Let u ∈ C0∞ (Rn ). We have uCγ; Rn Wpl ≤ uCγ; Rn \GLp + ∇l (uCγ); Rn \GLp + uγ; GWpl . (9.2.2) Let us estimate the first term on the right-hand side. It is clear that 

2

|(Cγ)(z)| ≤ c

∗

∗ −1



|γ(x, y + λδ (z))| dλ = c(δ )

2δ ∗

δ∗

1

|γ(x, y + s)| ds .

Using property (ii) of δ ∗ , we obtain for z ∈ Rn \G |(Cγ)(z)| ≤ c (ϕ(x) − y)−1



a(ϕ(x)−y)

|γ(x, y + s)| ds 2(ϕ(x)−y)



a

|γ(x, y + t(ϕ(x) − y))| dt .

=c

(9.2.3)

2

This and the Minkowski inequality imply that
 a  p 1/p |u(z)|p |γ(x, y + t(ϕ(x) − y))| dt dz uCγ; Rn \GLp ≤ c Rn \G

≤c

2

 a
 Rn \G

2

|u(z)γ(x, y + t(ϕ(x) − y))|p dz

1/p dt .

Let p > 1. In view of Lemma 9.1.1, we have uCγ; Rn \GLp ≤ C sup

2 p and the numbers s and q are sufficiently close to m and p respectively. From the inequality p/q
 |u|q dµ ≤c sup ρps−n [µ(Bρ (z) ∩ Ω]p/q u; Rn pWps , z∈Ω,ρ∈(0,1)

Ω

where µ is a measure in Ω and u ∈ C0∞ (Rn ) (see Lemma 1.3.1), it follows that (9.3.4) is not greater than c

sup

ρs−n/p ∇k γ; Bρ (z)Lq .

z∈Ω,ρ∈(0,1)

Since q is close to p, it follows that ρk−l+n(1/p−1/q) ∇k γ; Bρ (z)Lq ≤ c (∇l γ; Bρ (z)Lp + ρ−l γ; Bρ (z)Lp ). Consequently, (9.3.4) is majorized by c

ρµ (ρm−n/p ∇l γ; Bρ (z)Lp + ρm−l−n/p γ; Bρ (z)Lp ),

sup z∈Ω,ρ∈(0,1)

where µ = l − k + n(1/q − 1/p) + s − m. Since µ > 0, the result follows.



Lemma 9.3.2. The following inequalities hold: c K(h)γ; ΩM (Wpm →Wpl ) ≤ γ; ΩM (Wpm →Wpl ) ≤ lim inf K(h)γ; ΩM (Wpm →Wpl ) , h→0

lim inf sk (K(h)γ) ≥ sk (γ) ,

(9.3.6)

h→0

sk (K(h)γ) ≤ c [sk (γ) + s0 (γ)] , where sk (γ) is the same as in (9.1.10) with G replaced by Ω. Proof. It is clear that K(h)γ; ΩM (Wpm →Wpl ) ≤ c



Ki (h)(ϕi γ); Gi M (Wpm →Wpl ) .

0≤i≤N

This and Lemma 9.1.2 imply that K(h)γ; ΩM (Wpm →Wpl ) ≤ c



ϕi γ; Gi M (Wpm →Wpl )

0≤i≤N

=c



(9.3.5)

ϕi γ; ΩM (Wpm →Wpl ) .

0≤i≤N

Since ϕ ∈ C0∞ (Rn ), the left inequality in (9.3.5) is proved.

(9.3.7)

9.3 Multipliers in a Bounded Domain

339

The right inequality in (9.3.5) follows from uγ; ΩWpl = lim uK(h)γ; ΩWpl ≤ lim inf K(h)γ; ΩM (Wpm →Wpl ) u; ΩWpm . h→0

h→0

Obviously, for any compact set E ⊂ Ω


|∇k (K(h)γ)|p dx

lim

lim inf sk (K(h)γ) ≥

h→0

h→0

1/p

E

.

[Cp,m−l+k (E)]1/p

Since ϕi Ki (h)(ϕi γ) → ϕ2i γ in Wpk (Ω) and



(9.3.8)

ϕ2i = 1 ,

i

the right-hand side of (9.3.8) is equal to fk (γ; E). Thus, (9.3.6) is proved. Next we turn to the estimate (9.3.7). Since ϕ ∈ C0∞ (Rn ), it follows for small enough h that    ∇k [ϕi Ki (h)(ϕi γ)] p dx ≤ c ∇j [Ki (h)(ϕi γ)] p dx . E

0≤j≤k

By Lemma 9.1.2, 



∇j [Ki (h)(ϕi γ)] p dx

e

sup e⊂Gi

E∩Gi

Cp,m−l+k (e)

|∇j (ϕi γ)|p dx ≤ sup

e

Cp,m−l+k (e)

e⊂Gi

Therefore,

.

 

sk (K(h)γ) ≤ c



|∇j γ|p dx sup

0≤i≤n 0≤j≤k e⊂Gi

e

Cp,m−l+k (e)

.

It remains to use Lemma 9.3.1, noting first that the right-hand side of (9.3.3) does not exceed c(sl (γ) + s0 (γ)). 

9.3.3 Description of Spaces of Multipliers in a Bounded Domain with Boundary in the Class C 0,1 Theorem 9.3.1. Let m and l be integers with m ≥ l ≥ 0 and p ∈ (1, ∞). The space M (Wpm (Ω) → Wpl (Ω)) consists of functions γ ∈ Wpl (Ω) such that sup e⊂Ω

∇l γ; eLp < ∞. [Cp,m (e)]1/p

(9.3.9)

340

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

The following inequalities hold: c

l 

sup

j=0 e⊂Ω

∇j γ; eLp ≤ γ; ΩM (Wpm →Wpl ) [Cp,m−l+j (e)]1/p


∇l γ; eLp . ≤ c sup + γ; Ω L 1 1/p e⊂Ω [Cp,m (e)]

(9.3.10)

The proof follows the same lines as that of Corollary 2.3.4 and the theorems preceding it, except that in place of the usual mollification operator we use K(h) given by (9.3.2) and Lemma 9.3.2. The next assertion is proved in the same way as Theorem 9.1.2. Theorem 9.3.2. Let m and l be integers with m ≥ l ≥ 0. The space M (W1m (Ω) → W1l (Ω)) consists of functions γ ∈ W1l (Ω) such that ∇l γ; Bρ (z) ∩ ΩL1 + ρ−l γ; Bρ (z) ∩ ΩL1 ≤ c ρn−m for all z ∈ Ω and ρ ∈ (0, 1). The relation γ; ΩM (W1m →W1l ) ∼

sup

ρm−n

z∈Ω,ρ∈(0,1)

l 

ρj−l ∇j γ; Bρ (z) ∩ ΩL1

j=0

holds. The following theorem results directly from Theorem 9.1.3. Theorem 9.3.3. Let either mp > n and p ∈ (1, ∞), or m ≥ n and p = 1. Then the space M (Wpm (Ω) → Wpl (Ω)) coincides with Wpl (Ω). A corollary of Theorem 9.2.1 and formula (9.3.1) is: Theorem 9.3.4. Let C be the extension operator defined by (9.3.1) and let γ ∈ (Wpm (Ω) → Wpl (Ω)), p ≥ 1. Then Cγ ∈ M (Wpm (Rn ) → Wpl (Rn )) and Cγ; Rn M (Wpm →Wpl ) ≤ c γ; ΩM (Wpm →Wpl ) . 9.3.4 Essential Norm and Compact Multipliers in a Bounded Lipschitz Domain Let Ω be a bounded domain with ∂Ω ∈ C 0,1 . As in the case of the whole space Rn , we associate with any element γ ∈ M (Wpm (Ω) → Wpl (Ω)) the essential norm ess γ; ΩM (Wpm →Wpl ) = inf γ − T ; ΩWpm →Wpl {T }

where {T } is the collection of all compact linear operators: Wpm (Ω) → Wpl (Ω).

9.3 Multipliers in a Bounded Domain

341

To derive two-sided estimates for the essential norm in M (Wpm (Ω) → we do not need new arguments beyond those given in Chap. 7. This is even simpler, since m and l are integers and the domain Ω is bounded now. The role of the operator T∗ used in Chap. 7 is played by the mapping T∗ , defined by  (T∗ u)(x) = γ(x) ϕ(j) (x)P (j) (u; x) Wpl (Ω))

j

(cf. the proof of the second part of Theorem 7.2.1). Here we use the following notation: {ϕ(j) } is a smooth finite partition of unity subordinate to the cov(j) ering of Ω by open balls Kδ with radius δ and with centers xj ∈ Ω; P (j) are polynomials of the form 
y − x  
x − xj β j u(y) dy δ −n ψβ (j) δ δ Kδ ∩Ω |β|≤m−1

where ψβ ∈ C0∞ (B1 ). In the same way as in Chap. 7, majorants for ess γ; ΩWpm →Wpl can be obtained from upper bounds for the norms (γ − T∗ )u; ΩWpl which are collected in the next assertion (cf. Remark 7.2.3). Lemma 9.3.3. Let γ ∈ M (Wpm (Ω) → Wpl (Ω)) where m and l are integers with m ≥ l > 0. (i) If p > 1 and mp < n, then
γ − T∗ ; ΩWpm →Wpl ≤ c

 ∇l γ; e + sup γ; B (x) ∩ Ω δ Lp . 1/p x∈Ω {e⊂Ω : d(e)≤δ} [Cp,m (e)] sup

In particular,
γ − T∗ ; ΩWpl →Wpl ≤ c

 ∇l γ; eLp + γ; ΩL∞ . 1/p {e⊂Ω : d(e)≤δ} [Cp,l (e)] sup

If p > 1 and mp = n, then δ is replaced by δ 1/2 on the right-hand sides of these inequalities. (ii) If m ≤ n, then  γ − T∗ ; ΩW1m →W1l ≤ c δ m−n sup ∇l γ; Bδ (x) ∩ ΩL1 x∈Ω  +δ −l γ; Bδ (x) ∩ ΩL1 . In particular, 
γ − T∗ ; ΩW1l →W1l ≤ c δ l−n sup ∇l γ; Bδ (x) ∩ ΩL1 + γ; ΩL∞ . x∈Ω

Now we state a theorem on two-sided estimates for the essential norm.

342

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

Theorem 9.3.5. Let γ ∈ M (Wpm (Ω) → Wpl (Ω)), where m and l are integers with m ≥ l ≥ 0. (i) If p > 1 and mp ≤ n, then ess γ; ΩM (Wpm →Wpl )
∇ γ; e  γ; eLp l Lp ∼ lim sup + δ→0 {e⊂Ω : d(e)≤δ} [Cp,m (e)]1/p [Cp,m−l (e)]1/p
 ∇l γ; eLp n ∼ lim sup + sup ρm−l− p γ; Bρ (x) ∩ ΩLp . 1/p δ→0 {e⊂Ω : d(e)≤δ} [Cp,m (e)] {x∈Ω,ρ≤δ} In particular, ess γ; ΩM Wpl ∼ lim

sup

δ→0 {e⊂Ω : d(e)≤δ}

∇l γ; eLp + γ; ΩL∞ . [Cp,l (e)]1/p

(ii) If m < n, then  ess γ; ΩM (W1m →W1l ) ∼ lim sup sup ρm−n ∇l γ; Bρ (x) ∩ ΩL1 δ→0

x∈Ω ρ≤δ

 +ρ−l γ; Bρ (x) ∩ ΩL1 . In particular, ess γ; ΩM W1l ∼ lim sup sup ρl−n ∇l γ; Bρ (x) ∩ ΩL1 + γ; ΩL∞ . x∈Ω ρ≤δ

δ→0

(iii) If mp > n and p ∈ (1, ∞), or m ≥ n and p = 1, then ess γ; ΩM (Wpm →Wpl ) = 0

for m > l

and ess γ; ΩM Wpl ∼ γ; ΩL∞

for m = l .

This immediately implies: Proposition 9.3.1. A function γ ∈ M (Wpm (Ω) → Wpl (Ω)), where m and l ˚ (Wpm (Ω) → Wpl (Ω)) of are integers with m > l ≥ 0, belongs to the space M compact multipliers if and only if
lim

sup

δ→0 {e⊂Ω : d(e)≤δ}

∇l γ; eLp  γ; eLp =0 + [Cp,m−l (e)]1/p [Cp,m (e)]1/p

or, equivalently,
lim

δ→0

 ∇l γ; eLp m−l− n p γ; B (x) ∩ Ω =0 + sup ρ ρ L p 1/p {e⊂Ω : d(e)≤δ} [Cp,m (e)] {x∈Ω,ρ≤δ} sup

9.3 Multipliers in a Bounded Domain

343

for p ∈ (1, ∞) and mp ≤ n;   lim δ m−n sup ∇l γ; Bδ (x) ∩ ΩL1 + δ −l γ; Bδ (x) ∩ ΩL1 = 0 δ→0

x∈Ω

for m < n. Finally, ˚ (W m (Ω) → W l (Ω)) = W l (Ω) M p p p if mp > n and p ∈ (1, ∞), or m ≥ n and p = 1. Similarly to Theorem 7.2.11, we can obtain the following description of the space of compact multipliers. ˚ (W m (Ω) → W l (Ω)), where m and l are integers Proposition 9.3.2. M p p with m > l ≥ 0, is the completion of C ∞ (Ω) with respect to the norm in M (Wpm (Ω) → Wpl (Ω)). ˚ W l (Ω) we denote the completion of In concert with this assertion, by M p C ∞ (Ω) with respect to the norm of M Wpl (Ω). The following proposition is an analogue of Theorem 7.3.10. ˚ Wpl (Ω), where l is a positive Proposition 9.3.3. A function γ belongs to M integer, if and only if γ ∈ C(Ω) and one of the following conditions is satisfied: (i) If pl ≤ n and p > 1, then ∇l γ; eLp = o(1) 1/p [C p,l (e)] {e⊂Ω : d(e)≤δ} sup

as

δ → 0.

(9.3.11)

as

δ → 0.

(9.3.12)

(ii) If l < n, then δ l−n sup ∇l γ; Bδ (x)L1 = o(1) x∈Ω

˚ Wpl (Ω) = Wpl (Ω). (iii) If pl > n and p > 1, or l ≥ n and p = 1, then M To conclude this section, we consider a relation between the essential norm and the constant K in the inequality γu; ΩWpl ≤ Ku; ΩWpm + C(γ)u; ΩLp .

(9.3.13)

Theorem 9.3.6. Let γ ∈ M (Wpm (Ω) → Wpl (Ω)), where m and l are integers with m ≥ l ≥ 0, and let inf K be the infimum of those K for which there exists a constant C(γ) such that (9.3.13) holds for all u ∈ Wpm (Ω). Then inf K ≤ ess γ; ΩM (Wpm →Wpl ) ≤ c inf K where c = c(Ω, n, p, l, m).

(9.3.14)

344

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

Proof. We set u−



ϕ(j) P (j)

j

in place of u in (9.3.13). Then   + + + + ϕ(j) P (j) ; Ω +W m +C(γ)+u − ϕ(j) P (j) ; Ω +L . γu−T∗ u; ΩWpl ≤ K +u − p

p

j

j

We have  + +p + (j) +p +u − ϕ(j) P (j) ; Ω +W m = + ϕ (u − P (j) ); Ω +W m j

p

p

j

≤c

m 

δ −kp u − P (j) ; Kδ ∩ ΩpW m−k . (j)

p

j

k=0

Since Ω is Lipschitz, it follows that, for some c ≥ 1, (j)

(j)

u − P (j) ; Kδ ∩ ΩWpm−k ≤ c δ k u; Kcδ ∩ ΩWpm . Therefore, (γ − T∗ )u; ΩWpl ≤ c (K + C(γ)δ m )u; ΩWpm and hence ess γ; ΩM (Wpm →Wpl ) ≤ c (K + C(γ)δ m ) for any small enough δ. The right estimate in (9.3.14) follows. Now we turn to the left estimate of (9.3.14). According to the definition of the essential norm, for some compact operator T : Wpm (Ω) → Wpl (Ω) and for all u ∈ Wpm (Ω), we have γu; ΩWpl ≤ ( ess γ; ΩWpm →Wpl + ε)u; ΩWpm + T u; ΩWpl . We need to show that for any ε > 0 one can find a constant Cε such that T u; ΩWpl ≤ ε u; ΩWpm + Cε u; ΩLp .

(9.3.15)

We assume that this is not the case. Then for some ε > 0 there exist a function sequence {uj } with uj ; ΩWpm = 1 and a number sequence {kj }, kj → +∞, such that T uj ; ΩWpl > ε + kj uj ; ΩLp .

(9.3.16)

Since the operator T : Wpm (Ω) → Wpl (Ω) is bounded and the norm of uj in Wpm (Ω) is equal to one, we see by (9.3.16) that uj → 0 in Lp (Ω). We select a subsequence from {uj } weakly convergent in Wpm (Ω) for which we retain the notation {uj }. Let v be its weak limit. Then, for any g ∈ Lp
(Ω), where p + p = pp , we have

9.3 Multipliers in a Bounded Domain

345



 guj dx → Ω

gv dx Ω

and hence v = 0 because uj → 0 in Lp (Ω). Since T transforms a sequence weakly convergent in Wpm (Ω) into a sequence strongly convergent in Wpl (Ω) we have T uj ; ΩWpl → 0, 

contrary to (9.3.16). ˚ (Wpl (Ω)), then Proposition 9.3.4. If m = l and γ ∈ M inf K = γ; ΩL∞ .

Proof. Let γ1 be a function in C ∞ (Ω) such that γ − γ1 ; ΩM Wpl < ε. Clearly, γ1 u; ΩWpl ≤ (γ; ΩL∞ u; ΩWpl + c

l 

 |∇j γ1 | |∇l−j u|; ΩLp .

j=1

Hence   γu; ΩWpl ≤ γ; ΩL∞ + 2ε u; ΩWpl + c γ1 ; ΩC l u; ΩWpl−1 . Since u; ΩWpl−1 ≤ ε u; ΩWpl + c(ε) u; ΩLp , it follows that inf K ≤ γ; ΩL∞ . Let us estimate inf K from below. We set uδ (x) = δ l−n/p η((x − y)/δ) into (9.3.13), where m = l. We take y ∈ Ω, δ > 0, η ∈ C0∞ (B1 ), η(0) = 1. ˚ W l (Ω), the function γ can be assumed to be smooth in Ω. By definition of M p One can easily see that γuδ ; ΩLp = |γ(y)| uδ ; ΩWpl + o(1) and uδ ; ΩLp = o(1)

as δ → 0.

This, together with (9.3.13) and the inequality lim inf uδ ; ΩWpl > 0, δ→0

gives |γ(y)| ≤ K.



346

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

Remark 9.3.1. The inequality   γu; ΩWpl ≤ γ; ΩL∞ + ε u; ΩWpl + C(γ; ε)u; ΩLp with an arbitrarily small ε > 0 is used in the Lp -theory of elliptic boundary value problems (see, for example, [Tr3], Sect. 5.3.4). In conclusion, we state an obvious corollary of Theorem 9.3.6 and Proposition 9.3.4. ˚ (Wpl (Ω)) then Corollary 9.3.1. If γ ∈ M γ; ΩL∞ ≤ ess γ; ΩM Wpl ≤ c γ; ΩL∞ .

(9.3.17)

Similarly to Theorem 7.3.1, we can prove that the left estimate in (9.3.17) holds for any γ ∈ M Wpl (Ω). 9.3.5 The Space M L1p (Ω) for an Arbitrary Bounded Domain Let Ω be a domain in Rn with compact closure. By L1p (Ω) we denote the space of functions in Lp,loc (Ω) with the first distributional derivatives in Lp (Ω), p ∈ [1, ∞). We supply L1p (Ω) with the norm u; ΩL1p = ∇u; ΩLp + u; ωLp , where ω is a nonempty domain contained in Ω along with its closure. We can check that the change of ω leads to an equivalent norm. If Ω is a domain with boundary in the class C 0,1 , then L1p (Ω) ⊂ Lp (Ω) and therefore L1p (Ω) = Wp1 (Ω). Comparing Theorem 2.1.1 with Theorems 9.3.1, 9.3.2 and using Theorem 9.3.4, we obtain that for Lipschitz domains the space M L1p (Ω) coincides with the space of restrictions to Ω of multipliers in Wp1 (Rn ). The following example shows that this fails for arbitrary domains. Example 9.3.1. Let Ω be the union of the rectangles Am = {x : 21−m − δm < x1 < 21−m , 2/3 < x2 < 1} , Bm = {x : 21−m − εm < x1 < 21−m , 1/3 ≤ x2 ≤ 2/3} , C = {x : 0 < x1 < 1, 0 < x2 < 1/3} , where δm = 2−m−1 , εm = 2−(m+1)β , β ≥ 1, m = 1, 2, . . . (see Fig. 9.1). This domain was proposed by Nikodym in 1933 [Nik] as an example of the failure of the Poincar´e inequality. We show that the function γ(x) = xλ1 is a multiplier in L1p (Ω) if and only 1 (Rn ) if λ ≥ (β + p − 1)/p. It is clear that this function is a multiplier in Wp,loc

9.3 Multipliers in a Bounded Domain

347

y 1 Am

Bm

C

0

1

x

Fig. 9.1. Domain in Example 9.3.1

even for λ > (p − 1)/p. Thus, in the case (p − 1)/p < λ < (β + p − 1)/p, 1 (Rn ) is not a multiplier in the restriction to Ω of the multiplier xλ1 in Wp,loc 1 Lp (Ω). The necessity of the condition λ ≥ (β + p − 1)/p can be checked easily. In fact, let u be a continuous function equal to (2mβ m−2 )1/p in Am and to zero in C, and let u be linear to Bm . Clearly, ∞ 

∇u; ΩpLp =

2mβ m−2 mes2 Bm < ∞ ,

m=1

∇(γu); ΩpLp ≥ c

∞ 

2mβ m−2 2−(λ−1)pm mes2 Am .

m=1

The last series diverges if λ < (β + p − 1)/p. + − Now let λ ≥ (β + p − 1)/p. By Bm and Bm we denote the rectangle Bm raised and lowered by one-third, respectively. We have −1 δm

 Am

|u|p dx − ε−1 m

 + Bm

 p−1 |u|p dx ≤ c δm

|∂u/∂x2 |p dx . Am

348

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

Moreover,    1 |u|p dx − |u|p dx ≤ c |∂u/∂x1 |p dx . + − + − 2 Bm ∪Bm Bm Bm ∪Bm ∪Bm Therefore, 

δm
|u| dx ≤ c εm Am ∪Bm





p

|∇u| dx + p

− Am ∪Bm ∪Bm

− Bm

 |u|p dx . (9.3.18)

Obviously, u∇γ; ΩpLp ≤ c




|u|p dx +

C

∞ 

 (λ−1)p δm

m=1

Am ∪Bm

 |u|p dx .

Taking into account (9.3.18), we conclude that the sum on the right-hand side does not exceed  ∞
   (λ−1)p+1−β p δm |∇u| dx + |u|p dx c − Am ∪Bm ∪Bm

m=1

(λ−1)p+1−β ≤ c sup δm m

Consequently,




Ω



|∇u|p dx +

− Bm

 |u|p dx .

C

  ∇(γu); ΩLp ≤ c ∇u; ΩLp + u; CLp ,

i.e., γ ∈ M L1p (Ω). It can be easily shown that xλ1 ∈ M Wp1 (Ω) if and only if λ ≥ 1. Therefore, 1 does not belong for (p − 1)/p < λ < 1 the restriction to Ω of xλ1 ∈ M Wp,loc 1 to M Wp (Ω). Below we describe the space of multipliers in L1p (Ω), where Ω is an arbitrary bounded domain. This description is obtained as a corollary of theorems in [Maz15] on necessary and sufficient conditions for the validity of imbeddings of spaces of functions with first derivatives in Lp (Ω) (for earlier publications see, for example, [Maz1]–[Maz3]). In what follows, by g and G we denote the so-called admissible subsets of Ω, i.e., bounded sets such that Ω ∩ ∂g and Ω ∩ ∂G are manifolds of the class C ∞ . Further, let closΩ g be the closure of g with respect to Ω. We need the following capacity of the pair of sets g and G: p-capΩ (g, G) = inf{∇u; ΩpLp : u ∈ C ∞ (Ω), u = 0 on G, u = 1 on g} . Theorem 9.3.7. ([Maz15]). Let u be an arbitrary function in C ∞ (Ω) with u = 0 on G.

9.3 Multipliers in a Bounded Domain

349

(i) For p > 1, the inequality γu; ΩLp ≤ C ∇u; ΩLp holds if and only if

(9.3.19)

 |γ|p dx ≤ const p-capΩ (g, G), g

where g is an admissible set with closΩ ⊂ Ω\closΩ G. The best constant C in (9.3.19) for p > 1 satisfies the inequality γ; gpLp (p − 1)p−1 p ≤ Cp . C ≤ sup pp g p-capΩ (g, G) (ii) For p = 1, the inequality (9.3.19) holds if and only if  |γ| dx ≤ const s(Ω ∩ ∂g), g

where g is any admissible set with closΩ g ⊂ Ω\closΩ G and s is the (n − 1)dimensional area. The best constant C in (9.3.19) for p = 1 is given by C = sup g

γ; gL1 . s(Ω ∩ ∂g)

Using Theorem 9.3.7, one easily obtains a description of the space M L1p (Ω). Theorem 9.3.8. A function γ belongs to M L1p (Ω) if and only if γ ∈ L∞ (Ω)∩ 1 M Wp,loc (Ω) and, for some admissible set G with G ⊂ Ω, ∇γ; gpLp

1, the condition γ ∈ M Wp,loc (Ω) can be replaced by ∇γ; gpLp < ∞. sup Cp,1 (e)

Here the supremum is taken over all admissible sets g placed at an arbitrary fixed positive distance from ∂Ω. 1 According to Theorem 9.3.2, γ ∈ M W1,loc (Ω) if and only if sup r1−n ∇γ; Br (x)L1 < ∞ , where the supremum is taken over all balls Br (x), x ∈ Ω, placed at an arbitrary fixed positive distance from ∂Ω.

9.4 Change of Variables in Norms of Sobolev Spaces In this section we introduce and study certain classes of differentiable mappings which are considered as operators in pairs of Sobolev spaces. In Sect. 9.4.1, using spaces of multipliers, we define the so-called (p, l)-diffeomorphisms. We show that these mappings preserve the space Wpl . With the help of (p, l)-diffeomorphisms we introduce in Sect. 9.4.4 the class of (p, l)-manifolds on which Wpl can be properly defined. Sect. 9.4.5 concerns mappings of the class Tpm,l , i.e., the mappings U → V which generate continuous operators: Wpm (V ) → Wpl (U ), where p ≥ 1, m and l are integers, U and V are open sets in Rn . 9.4.1 (p, l)-Diffeomorphisms Let V be an open subset of Rn and let l be a noninteger. By Wpl (V ) we mean the space of functions with the finite norm
  1/p |∇[l] u(x)−∇[l] u(y)|p |x−y|−n−p{l} dxdy . u; V Wpl = ∇[l] u; V Lp + V

V

Let p ∈ (1, ∞), l ≥ 1, and let U and V be open subsets of Rn . A quasiisometric mapping κ : U → V is called a (p, l)-diffeomorphism if all elements of its Jacobi matrix ∂κ belong to the space of multipliers M Wpl (U ). The following four lemmas contain some properties of (p, l)-diffeomorphisms. By ∂κ; U M Wpl−1 we denote the sum of the norms of elements of ∂κ in M Wpl−1 (U ). Lemma 9.4.1. Let u ∈ Wpl (V ), l ∈ [1, ∞), and let κ : U → V be a (p, l)diffeomorphism. Then u ◦ κ ∈ Wpl (U ) and u ◦ κ; U Wpl ≤ c u; V Wpl .

(9.4.1)

9.4 Change of Variables in Norms of Sobolev Spaces

351

Proof. Set λ = inf det ∂κ. Clearly, u ◦ κ; U Wp1 ≤ (∂κ)∗ (∇u) ◦ κ; U Lp + u ◦ κ; U Lp   ≤ λ−1/p ∂κ; U L∞ ∇u; V Lp + u; V ||Lp . In the case {l} > 0 we have
  1/p |u(κ(x)) − u(κ(y))|p |x − y|−n−p{l} dxdy  u ◦ κ; U Wp{l} = U



U

+u ◦ κ; U Lp ≤ λ−2/p ∂κ; U L∞

n/p+{l}

 + λ−1/p u; V Wp{l} .

Suppose that (9.4.1) holds for all l with [l] = 1, . . . , k − 1. Then, for [l] = k,  u ◦ κ; U Wpl = ∇(u ◦ κ); U Wpl−1 + u ◦ κ; U Lp = (∂κ)∗ (∇u) ◦ κ); U Wpl−1 + u ◦ κ; U Lp ≤ ∂κ; U M Wpl−1 (∇u) ◦ κ); U Wpl−1 + u ◦ κ; U Lp . Using the induction assumption, we complete the proof.



Lemma 9.4.2. If κ is a (p, l)-diffeomorphism, then κ −1 is also a (p, l)diffeomorphism. Proof. For l = 1 this assertion is contained in the definition of the (p, l)diffeomorphism. Let 1 < l < 2 and let u be an arbitrary function in Wpl−1 (V ). Since κ is a bi-Lipschitz mapping, it follows that   u ∂(κ −1 ); V Wpl−1 =  (u ◦ κ)(∂κ)−1 ◦ κ −1 ; V Wpl−1 ≤ c (u ◦ κ)(∂κ)−1 ; U Wpl−1 . Hence u ∂(κ −1 ); V Wpl−1 ≤ c (∂κ)−1 ; U M Wpl−1 u ◦ κ; U Wpl−1 . It remains to use Lemma 9.4.1 and the condition ∂κ ∈ M Wpl−1 (U ). We proceed by induction. Suppose that the lemma is proved for [l] = 1, . . . , k − 1. Let [l] = k. The condition ∂κ ∈ M Wpl−1 (U ) implies the inclusion ∂κ ∈ M Wpl−2 (U ) and hence, by the induction assumption, ∂(κ −1 ) ∈ M Wpl−2 (V ). This together with Lemma 9.4.1 implies that the matrix (u ◦ κ)(∂κ)−1 ◦ κ −1 belongs to the space Wpl−1 (V ) provided that (u◦κ)(∂κ)−1 ∈ Wpl−1 (U ). The last inclusion holds since (∂κ)−1 ∈ M Wpl−1 (U )  and u ◦ κ ∈ Wpl−1 (U ) by Lemma 9.4.1.

352

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

Lemma 9.4.3. Let γ ∈ M Wpl (V ) and let κ be a (p, l)-diffeomorphism. Then γ ◦ κ; U M Wpl ≤ c γ; V M Wpl .

(9.4.2)

Proof. By Lemma 9.4.1, for all u ∈ Wpl−1 (U )   (γ ◦ κ)u; U Wpl =  u ◦ κ −1 )γ ◦ κ; U Wpl ≤ c γ; V M Wpl u ◦ κ −1 ; V Wpl .

(9.4.3)

Since by Lemma 9.4.2 κ −1 is a (p, l)-diffeomorphism, Lemma 9.4.1 implies the estimate u ◦ κ −1 ; V Wpl ≤ c u; U Wpl . Combined with (9.4.3), this completes the proof.



Lemma 9.4.4. Let κ1 : U → V and κ2 : V → W be (p, l)-diffeomorphisms. Then their composition κ2 ◦ κ1 : U → W is a (p, l)-diffeomorphism. Proof. Since the matrix ∂(κ2 ◦κ1 ) is equal to the product of matrices (∂κ2 ◦κ1 ) and ∂κ1 , it follows that ∂(κ2 ◦ κ1 ); U M Wpl−1 ≤ ∂κ2 ◦ κ1 ; U M Wpl−1 ∂κ1 ; U M Wpl−1 . We estimate the first factor on the right by (9.4.2) replacing l by l − 1, γ by ∂κ2 and κ by κ1 .  Remark 9.4.1. The above definition of (p, l)-diffeomorphisms can be generalized replacing M Wpl by one of the multiplier algebras Am,l and Am,l dealt p p with in Sect. 6.3 and Sect. 6.4.2. Obviously, all results in this subsection remain true. Remark 9.4.2. Runst and Youssfi [RY] used (p, l)-diffeomorphisms to prove an existence theorem for an equation involving the Jacobian. (For other results in the same area see Dacorogna and Moser [DM1], Ye [Ye], and Sickel and Youssfi [SY].) 9.4.2 More on (p, l)-Diffeomorphisms In the sequel we sometimes use the norm |||·||| in Wpl (V ), invariant with respect to dilations, which is defined for integer l by |||u; V |||Wpl =

l  j=0

dj−l ∇j u; V Lp ,

(9.4.4)

9.4 Change of Variables in Norms of Sobolev Spaces

353

where d is the diameter of V . In the case of noninteger l we put |||u; V |||Wpl =

[l] 

dj−l ∇j u; V Lp

j=0

+

[l]  j=0

d

j−[l]


  V

|∇j u(x) − ∇j u(y)|p

V

1/p dxdy . |x − y|n+p{l}

(9.4.5)

The norm in M (Wpm (V ) → Wpl (V )) generated by the norm ||| · ||| in Wpl (V ) will be denoted by |||γ; Ω|||M (Wpm →Wpl ) .

(9.4.6)

In this subsection we collect some properties of (p, l)-diffeomorphisms related to the norm ||| · |||. (i) Let u ∈ Wpl (V ), l ∈ [1, ∞), and let κ : U → V be a (p, l)diffeomorphism. Then u ◦ κ ∈ Wpl (U ) and |||u ◦ κ; U |||Wpl ≤ c |||u; V |||Wpl .

(9.4.7)

(ii) If κ is a (p, l)-diffeomorphism, then κ −1 is also a (p, l)-diffeomorphism, that is, (9.4.8) |||∂(κ −1 ); U |||M Wpl−1 ≤ c. (iii) Let γ ∈ M Wpl (V ) and let κ be a (p, l)-diffeomorphism. Then |||γ ◦ κ; U |||M Wpl ≤ c |||γ; V |||M Wpl .

(9.4.9)

The constants c in (i)–(iii) depend on inf det ∂κ, p, l, n, and the norm |||∂κ; U |||M Wpl−1 . A similar remark concerns the next property. (iv) Let κ1 : U → V and κ2 : V → W be (p, l)-diffeomorphisms. Then their composition κ2 ◦ κ1 : U → W is a (p, l)-diffeomorphism, i.e. |||∂(κ2 ◦ κ1 ); U |||M Wpl−1 ≤ c.

(9.4.10)

Assertions (i)–(iv) are obtained in Sect. 9.4.1 for the usual norm in Wpl . The passage to the norm (9.4.6) does not change the proof. 9.4.3 A Particular (p, l)-Diffeomorphism Let ϕ : Rn → R be a Lipschitz function with the Lipschitz constant L. We introduce the operator T by the formula  (T f )(x, y) = ζ(t)ϕ(x + ty)dt, y > 0, (9.4.11) Rn

354

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

where ζ ∈ C0∞ (Rn ). By Theorem 8.7.2, if l is an integer, p ∈ [1, ∞) and l−1/p ∇s ϕ ∈ M Wp (Rn ), s = 0, 1, . . . , l, then ∇s (T ϕ) ∈ M Wpl (Rn+1 + ) and n ∇s (T ϕ); Rn+1 + M Wpl ≤ c ∇s ϕ; R M Wpl−1/p .

(9.4.12)

We assume further that ζ ≥ 0 and  ζ(t)dt = 1. Rn

Let, as before, G = {(x, y) : x ∈ Rn , y > ϕ(x)} and let N be a sufficiently large constant depending on L. We introduce the mapping λ : Rn+1  (ξ, η) → (x, y) ∈ G (9.4.13) + by the equalities x = ξ,

y = N η + (T ϕ)(ξ, η).

(9.4.14)

Lemma 9.4.5. For any ξ ∈ Rn the mapping αξ : R+  η → y = N η + (T ϕ)(ξ, η) is one to one, and the inverse is Lipschitz. Moreover, ∂ (αξ−1 (y)) ≤ (N − L)−1 . ∂y and

−1 α (y) − α−1 (y) ≤ c L(N − cL)−1 ξ1 − ξ2 Rn , ξ1 ξ2

where c is a constant depending on n. Proof. We fix points x ∈ Rn and y ∈ R+ . The operator   β : η → N −1 y − (T ϕ)(ξ, η) maps the segment {η : |η| ≤ |y − ϕ(x)|} into itself, because   |β(η)| ≤ N −1 |y − ϕ(x)| + |(T ϕ)(ξ, η) − (T ϕ)(ξ, 0)|   ≤ N −1 |y − ϕ(x)| + L |η| ≤ (1 + L)N −1 |y − ϕ(x)|. Also, β is a contraction operator, since |β(η1 ) − β(η2 )| ≤ LN −1 |η1 − η2 |.

(9.4.15)

(9.4.16)

9.4 Change of Variables in Norms of Sobolev Spaces

355

Therefore, there exists a unique solution η of the equation   N −1 y − (T ϕ)(ξ, η) = η, or, equivalently, of the equation αξ (η) = y. Let y1 and y2 be arbitrary points in R+ and let ηj = αξ−1 (yj ), j = 1, 2. We have   η1 − η2 = N −1 y1 − y2 − (T ϕ)(x, η1 ) + (T ϕ)(x, η2 ) . Hence

  |η1 − η2 | ≤ N −1 |y1 − y2 | + L|η1 − η2 | .

Since N > L, we arrive at (9.4.15). The equalities y = N αξ−1 (y) + (T ϕ)(ξ, αξ−1 (y)), j j

j = 1, 2,

imply that   |αξ−1 (y) − αξ−1 (y)| ≤ c N −1 L ξ1 − ξ2 Rn + |αξ−1 (y) − αξ−1 (y)| . 1 2 1 2 

Hence (9.4.16) follows.

Lemma 9.4.6. Let l be an integer, l > 1, and p ∈ [1, ∞). Further, let ∇ϕ ∈ M W p,l−1−1/p . Then the mapping λ defined by (9.4.14) is a (p, l)diffeomorphism. Proof. By Lemma 9.4.5, the inverse mapping λ−1 exists, is defined by ξ = x,

η = αx−1 (y),

and satisfies the Lipschitz condition. Its Jakobi matrix ∂λ is given by   I 0 (9.4.17) ∇ξ (T f ) N + ∂(T f )/∂η where I is the identity (n − 1) × (n − 1)-matrix. Since |∂(T ϕ)/∂η| ≤ L, it follows that det ∂λ = N + ∂(T ϕ)/∂η ≥ N − L > 0. In view of (9.4.12), the elements of ∂λ belong to the space M Wpl−1 (Rn+ ). Remark 9.4.3. By Lemma 9.4.2, the mapping κ diffeomorphism of the domain

=

G = {(x, y) : x ∈ Rn−1 , y > ϕ(x)},



λ−1 is a (p, l)(9.4.18)

356

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

where ϕ is a Lipschitz function, onto Rn+ . The mapping κ is given by ξ = x,

η = u(x, y),

where u is the unique solution of the equation y = N u + (T ϕ)(x, u).

(9.4.19)

The restrictions of the mappings λ and κ to Rn−1 = ∂Rn+ and ∂G will be also denoted by λ and κ, respectively. l−1/p Let G be a special Lipschitz domain. By Wp (∂G) we denote the space l of traces on ∂G of functions in Wp (G). In a similar way, we define the space l−1/p

Wp

(Γ ), where Γ is a subset of ∂G. Since G can be mapped onto Rn+ by a (p, l)-diffeomorphism, it follows by l−1/p Theorem 8.7.2 and Lemma 9.4.3 that M Wp (∂G) is the space of traces of l functions in M Wp (G). 9.4.4 (p, l)-Manifolds In terms of the (p, l)-diffeomorphisms, we can define in a standard manner (see, for instance, de Rham [dR], H¨ ormander [H1]) a class of non-smooth n-dimensional manifolds on which Sobolev spaces can be properly defined. We recall that a topological space M is called an n-dimensional manifold, if there exists a collection of homeomorphisms {ϕ} of open sets Uϕ ∈ M onto open subsets of Rn with M = ∪Uϕ . The pair (ϕ, Uϕ ) is called a map (or the coordinate system) and the set of maps is called the atlas. We say that two maps (ϕ, Uϕ ) and (ψ, Uψ ) have a (p, l)-overlapping, if the mapping ϕψ −1 : ϕ(Uϕ ∩ Uψ ) → ψ(Uϕ ∩ Uψ ) is a (p, l)-diffeomorphism. By Lemma 9.4.2, the same is true for the inverse mapping. If any two maps have a (p, l)-overlapping, then we have a (p, l)-atlas. The maximal (p, l)-atlas on M is called a (p, l)-structure. By a (p, l)-manifold we mean a manifold with a (p, l)-structure. Since l−1 (Rn ) for p(l − 1) > n, M Wpl−1 (Rn ) = Wp,unit it follows for these values of p and l that the structure on a (p, l)-manifold belongs to the class C 1 , whereas for p(l − 1) ≤ n such manifolds are Lipschitz. For functions defined on a (p, l)-manifold Ω we introduce the space l l l (Ω). Namely, u ∈ Wp,loc (Ω) if u ◦ ϕ−1 belongs to Wp,loc (ϕ(Uϕ )) for Wp,loc each map (ϕ, Uϕ ). With the help of Lemmas 9.4.1 and 9.4.2 we can prove in a standard way l (Ω) it suffices to use (see [H1], Theorem 2.6.2) that to define the space Wp,loc only one arbitrary (p, l)-atlas; i.e., the following assertion holds.

9.4 Change of Variables in Norms of Sobolev Spaces

357

Theorem 9.4.1. If a function u defined on a (p, l)-manifold Ω is such that l u ◦ ϕ−1 ∈ Wp,loc (ϕ(Uϕ )) l for any map of some atlas, then u ∈ Wp,loc (Ω). If ηϕ ∈ C0∞ (ϕ(Uϕ )) and the open sets Vϕ = {x ∈ Uϕ : ηϕ (ϕ(x)) = 0} l cover Ω, then, in order to define a topology in the space Wp,loc (Ω), it suffices to introduce the seminorms

u → ηϕ (u ◦ ϕ−1 Wpl . l If a manifold Ω is compact, then the topology in Wp,loc (Ω) can be induced by the norm  ηϕ (u ◦ ϕ−1 )Wpl , ϕ

where the sum is taken over all maps of a certain atlas. Replacing the space Rn by the closed half-space Rn+ = {ζ ∈ Rn : ζn ≥ 0} in the definition of a (p, l)-manifold M, we obtain the definition of a (p, l)manifold M with the boundary ∂M. Let l be an integer, l ≥ 2, and let M be a (p, l)-manifold. If p(l − 1) ≤ n, we additionally assume that the (p, l)-structure on M belongs to the class C 1 . Then the implicit function theorem 9.5.2, which will be proved in Sect. 9.5, implies that the (p, l)-structure on ∂M induces the (p, l − 1/p)-structure on ∂M. 9.4.5 Mappings Tpm,l of One Sobolev Space into Another Let U and V be domains in Rn . We say that a mapping κ : U → V belongs to the class Tpm,l if u ◦ κ ∈ Wpl (U ) for any u ∈ Wpm (V ) and u ◦ κ; U Wpl ≤ c u; V Wpm .

(9.4.20)

We limit consideration to integer m and l, m ≥ l ≥ 1. For m = l we write Tpl instead of Tpl,l . In this subsection we give sufficient and, for some values of p, m, l, necessary and sufficient conditions for a mapping to belong to the class Tpm,l . In particular, for m = l we obtain a wider set of mappings than the class of (p, l)-diffeomorphisms. In what follows, κ = (κ1 , . . . , κn ) is a one-to-one mapping with ∂κ ∈ W1l−1 (U ) such that det ∂κ does not change its sign and   u(κ(z)) |det ∂κ(z)| dz = u(ζ) dζ (9.4.21) U

V

358

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

for any u ∈ L1 (V ). (For sufficient conditions ensuring (9.4.21) see, for instance, Vodop’yanov, Gol’dshtein, Reshetnyak [VGR], and Mal´ y [Mal]). Since κ is a mapping of the class W1l (U ), it follows that, for any u ∈ C l (V ) and multi-index α with |α| ≤ l,  β ϕα (9.4.22) Dα [u(κ(z))] = β (z)(D u)(κ(z)) 1≤|β|≤|α|

a.e. in U . Here and henceforth, ϕα β =

 s

cs

n -

Dsij κi ,

i=1 j

where the sum is taken over all collections of multi-indices s = (sij ) satisfying   sij = α, |sij | ≥ 1, (|sij | − 1) = |α| − |β|. i,j

i,j

We note that another and more explicit expression for the functions ϕα β was found by Fraenkel [Fra]. Proposition 9.4.1. If |det (∂κ ◦ κ −1 )|−1/p ∈ M (Wpm (V ) → Lp (V )) and

−1/p (ϕα )) ◦ κ −1 ∈ M (Wpm−|β| (V ) → Lp (V )) β (|det ∂κ|

for all multi-indices α and β with l ≥ |α| ≥ |β| ≥ 1, then the mapping κ belongs to the class Tpm,l . Proof. Inequality (9.4.20) for any u ∈ C l (V ) ∩ Wpm (V ) follows directly from (9.4.21) and (9.4.22). The additional assumption u ∈ C l (V ) can be removed by approximation.  We give an example which shows that the conditions of Proposition 9.4.1 are sharp. Example 9.4.1. We consider the domain 2 U = {z : z12 + · · · + zn−1 < zn2γ , 0 < zn < 1},

γ > 0,

and the mapping κ: z → ζ

with ζi = zi , 1 ≤ i ≤ n − 1, and ζn = znγ .

9.4 Change of Variables in Norms of Sobolev Spaces

zn

ζn

U

V

0

359

0

zi

zj

ζi

ζj

Fig. 9.2. Mapping in the class Tpm,l of a cusp to a cone

It is clear that κ transforms U into the cone 2 < ζn2 , V = {ζ : ζ12 + · · · + ζn−1

0 < ζn < 1}.

(See Fig. 9.2) We show that κ ∈ Tpm,l if and only if either

p(m − 1) < n,

or

p(m − 1) ≥ n,

pl − 1 , pm − 1 pl − 1 γ> . p+n−1 γ≥

(9.4.23) (9.4.24)

Let u(ζ) = ζnσ , where σ = 1 in the case p(m − 1) ≥ n and σ is a noninteger in the interval (m − n/p, (pl − 1 + γ − γn)/γp] for p(m − 1) < n. Clearly, u ∈ Wpm (V ). On the other hand, u(κ(z)) = znγσ and therefore  1  znγ p −1 p(γσ−l) ∇l (u ◦ κ ); U Lp ≥ c zn rn−2 dr dzn = ∞ . 0

0

Thus, conditions (9.4.23), (9.4.24) are necessary.

360

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

Now we turn to the proof of sufficiency. By straightforward computation |β|γ−|α| we have ϕα and det ∂κ(z) = γ znγ−1 . Consequently, β (z) = c zn −1 )(ζ) = c ζn|β|−|α|/γ (ϕα β ◦κ

and det(∂κ ◦ κ −1 )(ζ) = γzn1−1/γ . To check the conditions of Proposition 9.4.1, it suffices to verify the inequalities  |ζ|(1−γ)/γ |w|p dζ ≤ c w; V pWpm V

and

 |ζ|p(|β|−|α|/γ)+(1−γ)/γ |w|p dζ ≤ c w; V p

m−|β|

Wp

V

for any w ∈ Wpm (V ). It is known that the Hardy inequality  |w|p dζ ≤ c w; V pWps , pk < n , pk V |ζ| holds if and only if either ps ≥ n or ps < n and k ≤ s. It remains to note that (1 − γ)/γ > −n ,

(1 − γ)/γp ≥ −m

and that, by (9.4.23), p(|β| − |α|/γ) + (1 − γ)/γ ≥ p(1 − l/γ) + (1 − γ)/γ > −n and |β| − |α|/γ + (1 − γ)/γp ≥ |β| − m for p(m − 1) < n.



Remark 9.4.4. For |β| = |α| = l = m, one of the conditions of Proposition 9.4.1 takes the form −1/p ϕα ∈ L∞ (U ) . β |det ∂κ| We have ∂l u(κ(z)) ∂zi1 · · · ∂zil l  
∂lu (κ(z)) ∂κkν /∂ziν + . . . , = ∂ζk1 · · · ∂ζkl ν=1

(9.4.25)

k1 ,...,kl

where the terms involving differentiation with respect to ζ of order less than l are omitted. So the condition mentioned above is equivalent to

9.4 Change of Variables in Norms of Sobolev Spaces

∂κk1 ∂κkl ··· |det ∂κ|−1/p ∈ L∞ (U ) . ∂zi1 ∂zil

361

(9.4.26)

Here k1 , . . . , kl , i1 , . . . , il are numbers with values 1, . . . , n. The inclusion (9.4.26) can be rewritten in the form |∂κ|pl ≤C, |det ∂κ|

(9.4.27)

where C = const and |∂κ| =

n

 ∂κ 2 1/2 i

i,j=1

∂zj

.

According to the Hadamard determinant inequality, |det ∂κ| ≤

n
 n
∂κi 2 1/2 i=1 j=1

Hence

∂zj

.

|det ∂κ| ≤ n−n/2 |∂κ|n .

Consequently, for pl > n, |∂κ| ≤ (Cn−n/2 )1/(pl−n) , i.e. κ is a Lipschitz mapping. In the case pl < n we obtain |det ∂κ| ≥ (npl/2 C −1 )n/(n−pl) .

(9.4.28)

Suppose that the mapping κ satisfies (9.4.27) together with κ −1 . Then κ is Lipschitz for pl < n also. Indeed, replacing κ by κ −1 in (9.4.28), we get |det ∂κ| ≤ (npl/2 C −1 )n/(pl−n) . This estimate and (9.4.27) imply that |∂κ| ≤ nn/2(pl−n) C (2n−pl)/(n−pl)pl . Thus, for pl = n, the mapping κ, which belongs to Tpl together with κ −1 , is bi-Lipschitz. The class of mappings which perform the isomorphism Wp1 (U ) ≈ Wp1 (V ) was studied in [VG], where it is shown that such mappings are bi-Lipschitz for p ≥ n. For pl = n, (9.4.27) means that κ is a mapping with bounded distortion (see Reshetnyak [Re]). Mappings subjected to (9.4.27) with p = l = 1 are called subareal since they either decrease the area of (n − 1)-dimensional surfaces or increase it with a finite coefficient (see [Maz4], [Maz15], Sect. 3.3.1).

362

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

Proposition 9.4.2. Inequality (9.4.27) is necessary for κ ∈ Tpl . The same inequality is equivalent to κ ∈ Tp1 . (Hence, by interpolation, Tpl ⊂ Tpk , k = 1, . . . , l − 1.) Proof. We put

u(ζ) = η(ζ)|λ|−l exp(i(λ, ζ)) ,

where η ∈ C0∞ (V ) and λ ∈ Cn , into (9.4.20). Applying the Cauchy formula   dλ1 dλn Dλγ P (0; z) = γ!(2πi)−n ··· P (λ; z)λ−γ ··· λ1 λn |λ1 |=1 |λn |=1 to the polynomial λ → P (λ; z) = |λ|l Dα [u(κ(z))] , we find that its coefficients belong to Lp (U ). Therefore, we may pass to the limit as |λ| → ∞ in (9.4.20). As a result, for any unit vector θ = (θ1 , . . . , θn ) we obtain + +  + + θγ Dγ P (0; ·) ◦ κ −1 ; V + ≤ c η; V Lp +η |det (∂κ ◦ κ −1 )|−1/p Lp

|γ|=l

which by (9.4.25) can be written as +
∂κ  ∂κkl  −1 + + + k1 ◦κ ; V + ≤ cη; V Lp . θk1 · · · θkl ··· +η |det (∂κ◦κ −1 )|−1/p ∂zi1 ∂zil Lp k1 ,...,kl

Since η is arbitrary, we conclude that the functions |det(∂κ ◦ κ −1 )|−1/p

∂(θ, κ) ∂(θ, κ) ··· ∂zi1 ∂zil 

are bounded. Hence condition (9.4.27) holds.

The next assertion concerning conditions for a mapping to belong to Tpm,l follows directly from Proposition 9.4.1 and Theorem 9.3.1. Proposition 9.4.3. Let V be a bounded domain with boundary in the class C 0,1 and let p ∈ (1, ∞). If, for any compact set e ⊂ V , mesn κ −1 (e) ≤ c Cp,m (e) and, for all multi-indices α, β with l ≥ |α| ≥ |β| ≥ 1, sup e⊂V

then κ ∈ Tpm,l .

−1 ϕα (e)Lp β; κ

[Cp,m−|β| (e)]1/p

< ∞,

9.4 Change of Variables in Norms of Sobolev Spaces

363

Now we present two propositions on necessary and sufficient conditions for a mapping to belong to the class Tpm,l . Proposition 9.4.4. Let V be a bounded domain with boundary in the class C 0,1 . The mapping κ belongs to T1m,l if and only if, for any ball Br (ζ) with ζ ∈ V and r ∈ (0, 1), mesn κ −1 (Br (ζ) ∩ V ) ≤ c rn−m and, for all multi-indices α and β with l ≥ |α| ≥ |β| ≥ 1, −1 rm−|β|−n ϕα (Br (ζ) ∩ V )L1 < ∞ . β; κ

sup ζ∈V ,r∈(0,1)

Proof. Sufficiency is an immediate corollary of Proposition 9.4.1 and Theorem 9.3.2. On the other hand, the inclusion κ ∈ T1m,l is equivalent to + + +

+ u + ; V + + det(∂κ ◦ κ −1 ) L1

 + + +



1≤|α|≤l 1≤|β|≤|α|

+
ϕα  + β ◦ κ −1 Dβ u; V + det ∂κ L1

≤ c u; V W1m . It remains to use Theorem 10.1.1 to be proved in the sequel.



The following two assertions can be proved with the same arguments. Proposition 9.4.5. Let V be a bounded Lipschitz domain and let (m−l)p > n and p ∈ (1, ∞). A mapping κ is an element of Tpm,l if and only if mesn U < ∞ and ϕα β ∈ Lp (U ) for all multi-indices α, β with l ≥ |α| ≥ |β| ≥ 1. Proposition 9.4.6. Let V be a bounded domain with boundary in the class C 0,1 and let p > n. A mapping κ belongs to Tpl if and only if (9.4.27) holds and ϕα β ∈ Lp (U ) for l ≥ |α| ≥ |β| ≥ 1. For instance, for p > n the inclusion κ ∈ Tp2 is equivalent to the conditions |∂κ|2p (|det ∂κ|)−1 ∈ L∞ (U ) and ∇∂κ ∈ Lp (U ) . Similarly, κ ∈ Tp3 for p > n if and only if |∂κ|3p (|det ∂κ|)−1 ∈ L∞ (U ) , and

 1≤ρ,σ,τ ≤n

∂κr ∂ 2 κs ∈ Lp (U ) , ∂zρ ∂zσ ∂zτ

∇2 ∂κ ∈ Lp (U )

r, s = 1, . . . , n .

If both U and V are bounded and belong to C 0,1 , then the conditions of Proposition 9.4.5 can be simplified. Namely, the following assertion holds.

364

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

Proposition 9.4.7. Let U and V be bounded domains with boundaries in C 0,1 and let p > 1 and (m − l)p > n. A mapping κ belongs to Tpm,l if and only if κ ∈ Wpl (U ). Proof. Since

! α "1≤i≤n |β|=1 {ϕα β }||α|=l = D κi |α|=l

and |β|=l

{ϕα β }|α|=l =

l )∂κkν *1≤iν ≤n , ∂ziν 1≤kν ≤n ν=1

the necessity follows from Proposition 9.4.5. By definition of ϕα β we obtain ϕα β ; U Lp ≤ c

n s

Dsij κi ; U Lp|α|/|sij |

i=1 j

which, together with the Gagliardo-Nirenberg inequality |sij |/|α|

Dsij κi ; U Lp|α|/|sij | ≤ c κi ; U 

|α|

Wp

1−|sij |/|α|

κi ; U L∞

(see [Gag2], [Nir]), completes the proof.



9.5 Implicit Function Theorems Using properties of the (p, l)-diffeomorphisms we are in a position to prove the following assertion concerning regularity properties of a function defined implicitly. For the existence of such a function without differentiability assumptions, one may consult Sect. 5.2 in [KP]. Theorem 9.5.1. Let G be the domain (9.4.18) where ϕ ∈ C 0,1 and let u be a function in G satisfying (i) ∇u ∈ M Wpl−1 (G), l ≥ 2, (ii) tr u = 0, where tr stands for the trace on ∂G, (iii) inftr(∂u/∂y) > 0, where l is an integer and p ∈ [1, ∞). Then ∇ϕ ∈ M Wpl−1−1/p (Rn−1 ). Proof. We introduce the bi-Lipschitz mapping τ : G  (x, y) → (ξ, η) ∈ Rn+ by ξ = x,

η = y − ϕ(x)

and put v(ξ, η) = u(ξ, η + ϕ(ξ)). Since ∂z u ∈ L∞ (G), it follows that ∂ξ v ∈ L∞ (Rn+ ) and almost everywhere in Rn+

9.5 Implicit Function Theorems

365

∂ξ v = (∂x u + ∂y u∇ϕ) ◦ τ −1 (see, for example, [GR], p.244). Hence, for any g ∈ C0∞ (Rn ) and almost all η ∈ R+ we have     g(ξ)∂ξ v(ξ, η)dξ = g(ξ) (∂x u + ∂y u∇ϕ) ◦ τ −1 (ξ, η)dξ. (9.5.1) Rn−1

Rn−1

Since ∂z u(x, ·) ∈ Wpl−1 (R1+ ) and l ≥ 2, the function y → ∂z u(x, y) is continuous for almost all x in Rn . Hence, for almost all ξ ∈ Rn−1 the right-hand side of (9.5.1) is continuous in η, and, in particular, the limit     lim g(ξ)∂ξ v(ξ, η)dξ = g(ξ) (∂x u + ∂y u∇ϕ) ◦ τ −1 (ξ, 0)dξ (9.5.2) η→0

Rn−1

Rn−1

exists. The function (ξ, η) → v(ξ, η) is Lipschitz and v(ξ, 0) = 0. Therefore,   lim g(ξ)∂ξ v(ξ, η)dξ = − lim ∂g(ξ)v(ξ, η)dξ = 0. η→0

η→0

Rn−1

Rn−1

This and (9.5.2) imply that (∂x u + ∂y u∇ϕ) ◦ τ −1 (ξ, 0) = 0 for almost all ξ ∈ Rn−1 , or, equivalently, ∂x u(x, ϕ(x)) + ∂y u(x, ϕ(x))∇ϕ(x) = 0 for almost all x ∈ Rn−1 . Thus, the identity  −1 ∇ϕ(x) = − ∂y u(x, ϕ(x)) ∂x u(x, ϕ(x))

(9.5.3)

holds for almost all x ∈ Rn−1 . Since ∇u ∈ M Wp1 (G), we have by Theorem 8.7.2 that tr uxi and tr uy 1−1/p

belong to M Wp

(∂G) or, equivalently, that the functions

x → uxi (x, ϕ(x) + 0)

and x → uy (x, ϕ(x) + 0)

1−1/p

(Rn−1 ). The inequality inf uy (x, ϕ(x) + 0) > 0 and the inare in M Wp 1−1/p 1−1/p (Rn−1 ) imply that 1/uy (·, ϕ(·)) ∈ M Wp clusion uy (·, ϕ(·)) ∈ M Wp n−1 (R ). Hence the function x → uxi (x, ϕ(x) + 0)/uy (x, ϕ(x) + 0) 1−1/p

belongs to the space M Wp

(Rn−1 ). Thus,

ϕxi ∈ M Wp1−1/p (Rn−1 ).

366

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

We proceed by induction. Let k be a positive integer such that 2 ≤ k < l. k−1−1/p Suppose that ϕxi ∈ M Wp (Rn−1 ). By Lemma 9.4.6 the mapping λ : n R+ → G defined by (9.4.14) is a (p, k)-diffeomorphism. This and the inclusion uxi , uy ∈ M Wpk (G) imply that uxi ◦ λ, uy ◦ λ ∈ M Wpk (Rn+ ). By Theorem 8.7.2, tr(uxi ◦ λ), tr(uy ◦ λ) ∈ M Wpk−1/p (Rn−1 ). Since the function tr(uy ◦ λ) is separated from zero, it follows that the ratio k−1/p tr(uxi ◦ λ)/tr(uy ◦ λ) belongs to the space M Wp (Rn−1 ). It remains to note that (9.5.3) can be written as ∇ϕ(x) = −

tr(∇x u ◦ λ) . tr(uy ◦ λ) 

Now we prove a local variant of Theorem 9.5.1. Theorem 9.5.2. Let G be the same domain as in Theorem 9.5.1. Further, let ω be an (n − 1)-dimensional domain and let U be the cylinder {(x, y) : x ∈ ω, y ∈ R}. Suppose that the function u, defined on G ∩ U , satisfies the conditions: p,l−1 (i) ∇u ∈ (M Wloc (U ∩ G))n , where l is an integer, l ≥ 2, and p ∈ [1, ∞). (ii) tr u = 0 on ω ∩ ∂G, (iii) the function tr ∂u ∂y is separated from zero on any compact subset of ω ∩ ∂G. Then l−1−1/p (ω). ∇ϕ ∈ M Wp,loc Proof. Duplicating the beginning of the proof of Theorem 9.5.1 we arrive at (9.5.3) for almost all x ∈ ω. In the rest of the proof we need only to replace the spaces Wps (G), Wps−1/p (∂G), Wps (Rn+ ), Wps−1/p (Rn−1 ) by the spaces s−1/p

ps−1/p

p,s s Wp,loc (U ∩ G), Wp,loc (U ∩ ∂G), Wp,loc (τ (U ∩ G)), Wp,loc

where τ = λ−1 .

(ω),

˚pm (Ω) → Wpl (Ω)) 9.6 The Space M (W

367

Remark 9.5.1. Theorems 9.5.1 and 9.5.2 are sharp in the following sense. If l−1−1/p ∇ϕ ∈ M Wp (Rn−1 ), then there exists a function u in G satisfying the conditions (i)–(iii). The role of such a function can be played by a solution of equation (9.4.19). In fact, (9.4.19) implies that u(x, ϕ(x)) = 0 and ∂u
∂(T ϕ) −1 = N+ ≥ (N + L)−1 . ∂y ∂u Since τ is a (p, l)-diffeomorphism, it follows that ∇u ∈ M Wpl−1 (G). To conclude this section we formulate the implicit mapping theorem analogous to Theorem 9.5.1 and which can be proved in the same way. Theorem 9.5.3. Let l and s be integers, n > s > n − (l − 1)p ≥ 0, z = (x, y), x ∈ Rs , y ∈ Rn−s , and let u ϕ be the mappings Rn → Rn−s and Rs → Rn−s , respectively, with the properties: (i) uz ∈ M Wpl−1 (Rn ), (ii) tru = 0, where tr is the trace on the surface {z : x ∈ Rs , y = ϕ(x)}, (iii) the inverse matrix (tr uy )−1 exists and its norm is uniformly bounded on the surface {z : x ∈ Rs , y = ϕ(x)}. l−1−(n−s)/p Then ϕx ∈ M Wp (Rs ).

˚ m(Ω) → W l (Ω)) 9.6 The Space M (W p p 9.6.1 Auxiliary Results In the present section m and l are integers, Ω is a domain in Rn , p ∈ (1, ∞), ˚ m (Ω) is the completion of C ∞ (Ω) in the norm W m (Ω). and W p p 0 We define two capacities of a compact set e ⊂ Ω by Cp,l (e, Ω) = inf{u; ΩpW l : u ∈ C0∞ (Ω), u ≥ 1 on e} p

and

cp,l (e, Ω) = inf{∇l u; ΩpLp : u ∈ C0∞ (Ω), u ≥ 1 on e}.

Obviously, if e1 ⊂ e2 and Ω1 ⊃ Ω2 , then Cp,l (e1 , Ω1 ) ≤ Cp,l (e2 , Ω2 ) . The capacity cp,l (e, Rn ) has the same property of monotonicity. It is also clear that the capacity cp,l (e, Rn ) acquires the factor dn−pl under the similarity transform with coefficient d. The capacity cp,l (e, Rn ) vanishes for any compact set e if n ≤ lp, p > 1. The Sobolev theorem on the imbedding of Wpl (Rn ) into L∞ (Rn ) for lp > n, p > 1, implies that the capacity Cp,l (e, Rn ) is separated from zero.

368

9 Sobolev Multipliers in a Domain, Multiplier Mappings and Manifolds

We present some other known properties of capacity (see [Maz15], Ch. 9): (i) Let lp < n and let e be a compact subset of the ball Bρ . Then cp,l (e, Bρ ) ≤ c cp,l (e, Rn )

(9.6.1)

where c does not depend on ρ. (ii) For all compact subsets e of the ball B1 , Cp,l (e, B2 ) ∼ Cp,l (e, Rn ) .

(9.6.2)

We recall certain properties of capacity discussed in Sect. 3.1.2: (iii) If ρ ≤ 1, then , ρn−pl if n > pl, p > 1; Cp,l (B ρ ) ∼ 1−p if n = pl, p > 1 . (log 2/ρ) (iv) If ρ > 1, then Cp,l (B ρ ) ∼ ρn . (v) If n > pl, then Cp,l (e) ≥ c (mesn e)(n−pl)/n . (vi) If n = pl and d(e) ≤ 1, then Cp,l (e) ≥ c (log(2n /mesn e))1−p . To reveal the dependence of certain constants upon the diameter of the domain, we use the norms (9.4.4) and (9.4.6) in Wpk (Ω) and M (Wpm (Ω) → Wpl (Ω)), respectively. In the following theorem we present the norms equivalent to the norm (9.4.6). The equivalence means that their ratios are bounded and separated from zero by constants independent of d. Theorem 9.6.1. Let Ω be a domain with ∂Ω ∈ C 0,1 and finite diameter d < ∞. (i) If p ∈ (1, ∞), then |||γ; Ω|||M (Wpm →Wpl ) ∼ sup

e⊂Ω

∇l γ; eLp + γ; ΩL1 , [cp,m (e, Bad )]1/p

(9.6.3)

where a > 1 and Bad is a ball with center 0 ∈ Ω. In the case mp < n we can replace Bad by Rn . If m = l, then the second term is equal to γ; ΩL∞ . (ii) If either pm > n and p > 1, or m ≥ n and p = 1, then the relation |||γ; Ω|||M (Wpm →Wpl ) ∼ dm−n/p |||γ; Ω|||Wpl

(9.6.4)

holds. (iii) If m < n, then |||γ; Ω|||M (W1m →W1l ) ∼

sup x∈Ω , 2ρ n and p > 1, or m ≥ n and p = 1. Then m−n/p

γ; ΩM (W ˚ m →W l ) ∼ sup dj p

p

j

|||γ; Qj |||Wpl .

Corollary 9.6.3. The relation   m−n ∇l γ; Bρ (x)L1 + ρ−l γ; Bρ (x)L1 γ; ΩM (W ˚ m →W l ) ∼ sup ρ 1

1

x∈Ω , 2ρ n and p > 1, then the relation  aα M (Wph−|α| →Wph−k ) P W h →Wph−k ∼

(10.1.3)

p

|α|≤k

p

|α|≤k

holds. Proof. The estimate (10.1.2) is obvious, so we need to prove only (ii). Let x ∈ Rn and let η ∈ C0∞ (B2 ) with η = 1 on B1 . Further, let u(y) = η((x−y)/δ), where δ ∈ (0, 1]. By substituting the function u into the inequality + +  + + aα Dα u+ h−k ≤ c uWph , + Wp

|α|≤k

we obtain sup a0 ; Bδ (x)Wph−k ≤ c δ n/p−h

x∈Rn

which, together with Theorem 2.1.1, Corollary 4.3.8, and Theorem 5.3.1, shows that a0 ∈ M (Wph → Wph−k ). h−|α|

Suppose that aα ∈ M (Wp  |α|≤ν

→ Wph−k ) for |α| ≤ ν, ν ≤ k − 1, and

aα M (Wph−|α| →Wph−k ) ≤ c P W h →Wph−k . p

We show that the same holds with ν is replaced by ν + 1. Clearly, + + +   + + + aα Dα uWph−k ≤ +P u − aα Dα u+ h−k + |α|≥ν+1

|α|≤ν

Wp

≤ c P W h →Wph−k uWph p

10.1 The Norm of a Differential Operator: Wph → Wph−k

375

for all u ∈ Wph . Putting here u(y) = (x − y)α η((x − y)/δ),

|α| = ν + 1,

we obtain sup aα ; Bδ (x)Wph−k ≤ c P W h →Wph−k δ n/p−h+|α| . p

x∈Rn

This, together with Theorem 2.1.1, Corollary 4.3.8, and Theorem 5.3.1 implies h−|α| that aα ∈ M (Wp → Wph−k ).  Now we present an analogous result for matrix operators. Let u(x) = {u1 (x), u2 (x), . . . , uN (x)} be an N -tuple vector-valued function. Consider the operator N *M ) Pu = Pjk (x, Dx )uk , x ∈ Rn , (10.1.4) j=1

k=1

where



Pjk (x, Dx )uk =

(α)

ajk (x)Dxα uk

|α|≤sj +tk

and sj , tk are integers. Theorem 10.1.1. Let h ≥ s = max sj , j = 1, . . . , M . (i) The operator P is a continuous mapping N -

P :

Wptk +h →

M -

Wph−sj ,

(10.1.5)

j=1

k=1

if (α)

ajk ∈ M (Wptk +h−|α| → Wph−sj ). The estimate P  ≤ c

N  M 



(α)

ajk 

k=1 j=1 |α|≤sj +tk

t +h−|α|

M (Wpk

h−sj

→Wp

(10.1.6)

)

holds. (ii) If p(h − s) ≥ n and p > 1, or p = 1, then the relation P  ∼

N  M 



k=1 j=1 |α|≤sj +tk

holds.

(α)

ajk 

t +h−|α|

M (Wpk

h−sj

→Wp

)

376

10 Differential Operators in Pairs of Sobolev Spaces

Proof. Inequality (10.1.6) follows from Lemma 10.1.1. Let M + N +  + + Pjk (x, Dx )uk + + j=1

h−sj

Wp

k=1

≤ c P 

N 

uk W h+tk . p

k=1

We fix i and set uk = 0 provided that k = i. Then M 

Pji (x, Dx )ui 

h−sj

Wp

j=1

≤ c P ui Wph+ti

and in the case p(h − s) > n, p > 1, as well as in the case p = 1, Lemma 10.1.1 gives the estimate M 



(α)

aji 

j=1 |α|≤sj +ti

h−s h+ti −|α| →Wp j )

M (Wp

≤ c P  , 

where i = 1, 2, . . . , N .

Next we show that an equivalence relation similar to (10.1.3) with an arbitrary p ∈ (1, ∞) can be obtained for partial differential operators of a special form. Theorem 10.1.2. Let h and s be positive integers, h ≥ 2s, 1 < p < ∞ and P (x, Dx )u =

s 

bj (x)∆j u,

j=0

where ∆ is the Laplace operator. Then P is a continuous mapping: Wph → Wph−2s if and only if bj ∈ M (Wph−2j → Wph−2s ),

j = 0, . . . , s.

Moreover, the relation P W h →Wph−2s ∼

s 

p

bj M (Wph−2j →Wph−2s )

(10.1.7)

j=0

holds. Proof. The sufficiency as well as the upper bound for the norm of P follows from Lemma 10.1.1. Suppose that, for all u ∈ Wph , P uWph−2s ≤ c uWph .

10.1 The Norm of a Differential Operator: Wph → Wph−k

377

Let u = 1 in a neighborhood of a compact set e with d(e) ≤ 1. Then ∇h−2s b0 ; eLp + b0 ; eLp ≤ c uWph . Consequently, ∇h−2s b0 ; eLp ≤ c [Cp,h (e)]1/p and b0 ; B1 (x)Lp ≤ c for all x ∈ Rn . By Theorem 2.3.3 this means that b0 ∈ M (Wph → Wph−2s ). Therefore, the operator Q∆ with Q=

s−1 

bj+1 (x)∆j

j=0

satisfies the inequality Q∆uWph−2s ≤ cuWph .

(10.1.8)

Let ζ ∈ Wph−2 and let supp ζ ⊂ {x = (x1 , . . . , xn ) : 0 < xi < 1}. We put w(x) = ζ(x) − ζ(−x) and wi (x) = w(x1 , . . . , xi−1 , xi /2, xi+1 , . . . , xn ) for i = 1, . . . , n. Further, let v(x) = w(x) +

n 

αi wi (x − ai )

(10.1.9)

i=1

where ai are fixed points with dist(ai , aj ) > (8n)1/2 and αi are arbitrary constants. It is clear that all the functions on the right-hand side of (10.1.9) are orthogonal to one and have disjoint supports. We show that the coefficients αi can be selected so that  xj v(x) dx = 0 , j = 1, . . . , n, which is equivalent to the algebraic system with respect to α1 , . . . , αn :  2

n
  αi (1 + δij ) = 0 , xj ζ(x) dx 1 + 2

j = 1, . . . , n ,

i=1

where δij is the Kronecker delta. The system is solvable because det 1 + δij ni,j=1 = n + 1. Let u be the harmonic (Newtonian for n > 2 and logarithmic for n = 2) potential with density v. Since v is orthogonal to 1, x1 , . . . , xn and the diameter of its support is bounded by a constant depending only on n, then

378

10 Differential Operators in Pairs of Sobolev Spaces

uWph ≤ c vWph−2 ≤ c1 ζWph−2 . This and (10.1.8) imply that QζWph−2s ≤ c ζWph−2 . By the arbitrariness of the origin, the last inequality holds for all functions ζ ∈ Wph−2 supported by any cube of the coordinate grid. Hence it is valid for all ζ ∈ Wph−2 , i.e. Q is a continuous operator: Wph−2 → Wph−2s . Now the result follows by successive reduction of order of the operator. 

10.1.2 A Counterexample Theorem 10.1.2 and part (ii) of Lemma 10.1.1 suggest the hypothesis: relation (10.1.3) holds for all p, h, k, i.e. the coefficients of any differential operator (10.1.1), mapping Wph into Wph−k , are necessarily multipliers in corresponding Sobolev spaces. The next example disproves this conjecture. Example 10.1.1. Let x = (x , xn ), x = (x1 , . . . , xn−1 ), n ≥ 3. We show that the coefficient a of the continuous operator a(x )

∂ : W22 → L2 ∂xn

need not be an element of M (W21 → L2 ). Suppose that  3−n A= sup r y∈Rn−1 ,r∈(0,1)

(n−1)

Br

(10.1.10)

|a(x )|2 dx < ∞, (y)

(n−1)

where Br (y) is the (n − 1)-dimensional ball with centre y and radius r. By u ˆ(x , λ) we denote the Fourier transform of the function u with respect to xn . According to Theorem 1.1.2,  |a(x )ˆ u(x , λ)|2 dx ≤ A [ˆ u(·, λ)]2 ; Rn−1 W12 Rn−1

 ≤ cA

Rn−1

  |ˆ u(x , λ)| |∇2,x
u ˆ(x , λ)| + |∇x
u ˆ(x , λ)|2 + |ˆ u(x , λ)|2 dx .

Consequently,     2  ∂u 2 |λ u ) dx ≤ c A dλ ˆ(x , λ)| |∇2,x
u ˆ(x , λ)| a(x ∂x n 1 n−1 n R R R    +|λ∇x
u ˆ(x , λ)|2 + |λˆ u(x , λ)|2 dx ≤ c1 A ∇2 u2L2 + ∇u2L2 .

10.1 The Norm of a Differential Operator: Wph → Wph−k

379

Thus the finiteness of the value A is sufficient for the continuity of operator (10.1.10). The necessity of the same condition results from the estimate   sup rh−|α|−n/p ∇h−k aα ; Br (x)Lp P W h →Wph−k ≥ c p

|α|≤k

x;r∈(0,1)

+ rk−|α|−n/p aα ; Br (x)Lp



(10.1.11)

derived in the proof of the second part of Lemma 10.1.1 for all p ∈ [1, ∞). If (10.1.3) is valid for all p, h, and k, then the continuity of the operator (10.1.10) implies that a ∈ M (W21 → L2 ). We choose the coefficient a as follows: a(x ) = ρ−1 | log ρ|ε−1 η(x1 , x2 ) ζ(x3 , . . . , xn−1 ) where ρ2 = x21 + x22 ,

0 < ε < 1/2,

η ∈ C0∞ (B1 ), (2)

ζ ∈ C0∞ (B1

(n−3)

).

It is clear that for any y ∈ Rn−1 and for all r ∈ (0, 1/2) we have    2  n−3 |a(x )| dx ≤ c r ρ−2 | log ρ|2(ε−1) dx1 dx2 (n−1)

Br

(2)

Br

(y)

= c rn−3 | log r|2ε−1 . Therefore A < ∞. Suppose that au; Rn L2 ≤ c u; Rn W21 for any u ∈ C0∞ (Rn ). Then, for all v ∈ C0∞ (R2 ), ηρ−1 | log ρ|ε−1 v; R2 L2 ≤ c v; R2 W21 . This estimate implies that  ρ−2 | log ρ|2(ε−1) dx1 dx2 ≤ c C2,1 (Br(2) ) ρ s − 2(m − l). Then, for all u ∈ C0∞ ,
 1/2 |∇l u|2 dµ ≤ c Ku; Rs τW2m u; Rs 1−τ (10.1.14) L2 , Rs

where τ = (2l + s − σ)/2m and c is a constant independent of u and µ. Moreover, condition (10.1.13) is necessary for the validity of (10.1.14). Proof of Theorem 10.1.3. Sufficiency. First we verify that for |γ| > 0 the operator aβγ (y)Dyβ Dzγ : W2h (Rn ) → W2h−k (Rn ) is continuous. With this aim in view, we show the continuity of the operator (Dyµ aβγ (y))Dyρ+β Dzγ+θ : W2h (Rn ) → L2 (Rn ), where 0 ≤ |µ| + |ρ| + |θ| ≤ h − k. By u ˆ(y, λ) we denote the Fourier transform of the function u with respect to z. Putting l = |ρ| + |β|,

σ = 2|β| + 2|γ| + s − 2k − 2|µ|,

m = k + |µ| + |ρ| + |θ|

10.1 The Norm of a Differential Operator: Wph → Wph−k

381

in Lemma 10.1.2, we conclude that  Rs

|(Dyµ aβγ (y))Dyρ+β u ˆ(y, λ)|2 dy 2(1−τ )

≤ c A2β,γ,µ ˆ u(·, λ); Rs 2τ u(·, λ); Rs L2 W2m ˆ

,

where τ = (|µ| + |ρ| + k − |γ|)/(|µ| + |ρ| + |θ| + k) and Aβ,γ,µ =

sup y∈Rs ,r∈(0,1)

r−σ/2 Dyµ aβγ ; Br(s) (y)L2 .

(We note that the condition σ > s − 2(m − l) in Lemma 10.1.2 is equivalent to |γ| > 0.) Multiplying the last inequality by |λ|2(|γ|+|θ|) and integrating over λ, we get (Dyµ aβγ )Dyρ+β Dzγ+θ u; Rn 2L2  2(1−τ ) m ≤ c A2β,γ,µ ˆ u(·, λ); Rs 2τ ˆ(·, λ); Rs L2 dλ W2m |λ| u Rn−s

≤ c A2β,γ,µ u; Rn 2W2m . Since m ≤ h and Aβ,γ,µ does not exceed the value (10.1.12), the operator P (y, Dy , Dz ) − P (y, Dy , 0) : W2h (Rn ) → W2h−k (Rn )

(10.1.15)

is continuous. In order to derive the continuity of P (y, Dy , 0) in the same pair of spaces we prove the inequality k/h

1−k/h

P (y, Dy , 0)v; Rs L2 ≤ c v; Rs W h v; Rs L2

.

(10.1.16)

2

Applying Lemma 10.1.2 with l = |β| and m = h, for any multi-index β with |β| ≤ k, we obtain aβ0 Dyβ v; Rs L2 ≤c

sup y∈Rn ,r∈(0,1)

k/h

1−k/h

rk−|β|−s/2 aβ0 ; Br(s) (y)L2 v; Rs W h v; Rs L2 2

which entails (10.1.16). Using (10.1.16) together with (10.1.12), we find that P (·, Dy , 0)u; Rn 2W h−k 2

 ≤c



Rn−s

 P (·, Dy , 0)ˆ u(·, λ); Rs 2W h−k + |λ|2(h−k) P (·, Dy , 0)ˆ u(·, λ); Rs 2L2 dλ 2

382

10 Differential Operators in Pairs of Sobolev Spaces

 ≤c

 Rn−s

2k/h

2(1−k/h) 

ˆ u(·, λ); Rs 2W h + ˆ u(·, λ); Rs W h |λ|h u ˆ(·, λ); Rs L2 2

dλ

2

≤ c u; Rn 2W h 2

for all u ∈ The sufficiency follows. Necessity. We note that by replacing s by n in (10.1.12) we obtain an equivalent condition. Then the finiteness of (10.1.12) for all multi-indices β, γ follows from (10.1.11). Thus the first part of this theorem implies the continuity of the operator (10.1.15). Since the mapping W2h (Rn ).

P (y, Dy , Dz ) : W2h (Rn ) → W2h−k (Rn ) is continuous, it follows for all v ∈ W2h (Rn ) that P (y, Dy , 0)v, Rn W h−k ≤ c v; Rn W2h . 2

Substituting here v(x) = η(z)u(y), where η is a fixed function in C0∞ (Rn−s ) and u is an arbitrary function in W2h (Rs ), we complete the proof of Theorem 10.1.3. 

10.1.4 Differential Operators on a Domain Let U and V be open sets in Rn and let  pα (z)Dzα u P (z, Dz )u =

(10.1.17)

|α|≤k

be a differential operator on U . Given any (p, l)-diffeomorphism κ : U → V with l ≥ k, we introduce the differential operator Q on V , defined by Q(u ◦ κ −1 ) = (P u) ◦ κ −1 . In view of Lemmas 9.4.1 and 9.4.2, Q maps Wpl (V ) continuously into Wpl−k (V ) if and only if P maps Wpl (U ) continuously into Wpl−k (U ). l,k By Op,loc (U ) we denote the class of operators of the form (10.1.17) such that l−|α| l−k (U )) pα ∈ M (Wp,loc (U ) → Wp,loc for any multi-index α with |α| ≤ k. l,k Lemma 10.1.3. The operator P belongs to the class Op,loc (U ) if and only if l,k Q ∈ Op,loc (V ).

10.1 The Norm of a Differential Operator: Wph → Wph−k

383

Proof. Let ζ = κ(z). We have 

Dα [v(κ(z))] =

(Dβ v)(κ(z))

 s

1≤|β|≤|α|

cs

n -

Dsij κi (z) ,

where the sum is taken over all multi-indices s = (sij ) such that   sij = α , |sij | ≥ 1, (|sij | − 1) = |α| − |β| . i,j

(10.1.18)

i=1 j

(10.1.19)

i,j

Let Q(ζ, Dζ ) =



qβ (ζ)Dζβ .

|β|≤k

By (10.1.18), 

qβ =

(pα ◦ κ −1 )

|β|≤|α|≤k

 s

cs

n -

(Dsij κi ) ◦ κ −1 .

(10.1.20)

i=1 j

Since l−1 l−r (U ) ⊂ M Wp,loc (U ), ∇κi ∈ M Wp,loc

1 ≤ r ≤ l,

it follows that l−r−|sij |+1

l−r Dsij κi ∈ M (Wp,loc (U ) → Wp,loc

Therefore,

-

l−|β|

l−|β|−

Dsij κi ∈ M (Wp,loc (U ) → Wp,loc

(U )) .



i,j (|sij |−1)

(U )),

i,j

which is the same as l−|β| l−|α| Dsij κi ∈ M (Wp,loc (U ) → Wp,loc (U )) . i,j l−|α|

l−k (U )) together It remains to use the condition pα ∈ M (Wp,loc (U ) → Wp,loc with Lemmas 9.4.1 and 9.4.2.  l,k (U ) is sufficient for the operaBy Lemma 10.1.1, the inclusion P ∈ Op,loc l,k l−k l tor P to map Wp,loc (U ) into Wp,loc (U ). The inclusion P ∈ Op,loc (U ) is also necessary for p = 1 or p(l − k) > n (see Lemma 10.1.1).

Remark 10.1.1. Let Ω be a (p, l)-manifold. Such manifolds were introduced in Sect. 9.4.4. We define a differential operator of order k, k ≤ l, as a linear l−k l (Ω) into the space Wp,loc (Ω) if, for any map mapping P of the space Wp,loc l,k (ϕ, Uϕ ), there exists a differential operator Pϕ in the class Op,loc (ϕ(Uϕ )) such that (P u) ◦ ϕ−1 = Pϕ (u ◦ ϕ−1 ) on ϕ(Uϕ ).

By Lemma 10.1.3, it suffices to restrict oneself to the maps of a certain atlas.

384

10 Differential Operators in Pairs of Sobolev Spaces

10.2 Essential Norm of a Differential Operator In this section we obtain bounds for the essential norm of a differential operator. Let P (x, Dx ) be a differential operator of order k, defined by (10.1.1), let P0 be its principal homogeneous part, and let ess P W h →Wph−k be the p essential norm of the mapping P : Wph → Wph−k ,

h ≥ k.

Lemma 10.2.1. For all θ ∈ ∂B1 , P0 (·, θ)L∞ ≤ ess P W h →Wph−k . p

The proof is quite similar to that of Theorem 7.3.1, so we only outline it. Let η, ϕk and Qk be the same functions and cube as in the proof of Theorem 7.3.1. We put n 
 [ξj ]yj , vξ (y) = |ξ|−h η(y) exp i j=1

where ξ ∈ R \{0} and [ξj ] is the integer part of ξj . By the same argument as in the proof of Theorem 7.3.1, we show that in the first place n

lim vξ Wph = Ah ηLp ,

|ξ|→∞

Ah = const > 0 ,

in the second place lim ϕk P vξ Wph−k = Ah ϕk P0 (·, ξ|ξ|−1 )ηLp ,

|ξ|→∞

and in the third place ϕk T v ξ → 0

as

˚ph (Qk ), |ξ| → ∞ in W

where T is a compact operator in Wph . Then lim sup ϕk P vξ Wph−k = lim sup ϕk (P − T )vξ Wph−k |ξ|→∞

|ξ|→∞

and, by Lemma 7.1.9, for some σ > 0, lim sup ϕk P vξ |Wph−k ≤ (1 + O(k −σ )) lim sup (P − T )vξ Wph−k |ξ|→∞

|ξ|→∞

≤ (1 + O(k

−σ

)) lim sup vξ Wph ( ess P W h →Wph−k + ε) . p

|ξ|→∞

Consequently, ϕk P0 (·, ξ|ξ|−1 )ηLp ≤ (1 + O(k −σ ))ηLp ess P W h →Wph−k p

and finally

P0 (·, ξ|ξ|−1 η)Lp ≤ ηLp ess P W h →Wph−k . p



10.2 Essential Norm of a Differential Operator

385

Lemma 10.2.2. (i) The estimate ess P W h →Wph−k ≤ c p

 |α|≤k

ess aα M (Wph−|α| →Wph−k )

(10.2.1)

holds. (ii) If p = 1 or p(h − k) > n, p > 1 and P maps continuously Wph into h−k Wp , then  ess aα M (Wph−|α| →Wph−k ) . (10.2.2) ess P W h →Wph−k ∼ p

|α|≤k

h−|α|

Proof. (i) Let ε > 0 and let Tα be a compact operator which maps Wp into Wph−k and satisfies aα − Tα Wph−|α| →Wph−k ≤ ess aα M (Wph−|α| →Wph−k ) + ε . Since the operator T =



Tα Dα : Wph → Wph−k

|α|≤k

is compact, we arrive at (10.2.1). (ii) We begin with the case p = 1, h − k ≤ n. Let ε > 0 and let T be a compact operator such that P − T W h →W h−k ≤ ess P W h →W h−k + ε . 1

1

1

1

Further, let ηδ,x and ζr be the cutoff functions introduced at the beginning of Sect. 7.1. Duplicating the proof of the upper estimate in Theorem 7.2.1 with obvious changes, we obtain the estimate ηδ,x P W h →W h−k + ζr P W h →W h−k ≤ c ( ess P W h →W h−k + ε) 1

1

1

1

1

1

(10.2.3)

which, by (10.1.3), is equivalent to   ηδ,x aα M (W h−|α| →W h−k ) + ζr aα M (W h−|α| →W h−k ) 1

|α|≤k

1

1

1

≤ c ( ess P W h →W h−k + ε) . 1

1

It remains to use Theorem 7.2.1. Next let p(h − k) > n and p ≥ 1. According to Lemma 10.1.1, aα ∈ h−|α| M (Wp → Wph−k ). The inequality ζr P W h →Wph−k ≤ c ( ess P W h →Wph−k + ε) , p

p

where ε is sufficiently small and r is sufficiently large, can be obtained in the same way as (10.2.3). This inequality and (10.1.3) give

386

10 Differential Operators in Pairs of Sobolev Spaces

 |α|≤k

ζr aα M (Wph−|α| →Wph−k ) ≤ c ( ess P W h →Wph−k + ε) . p

Therefore, lim sup |x|→∞



aα ; B1 (x)Wph−k ≤ c ess P W h →Wph−k p

|α|≤k

and, by Theorem 7.2.5,  ess aα M (Wph−|α| →Wph−k ) ≤ c ess P W h →Wph−k . p

|α|0,

(10.4.1)

holds for every u ∈ S if and only if R/P ∈ M (W2l → L2 ), which is equivalent to R/P ; eL2 sup n the condition (10.4.2) means that  sup |R(ξ)/P (ξ)|2 dξ < ∞ . (10.4.3) x∈Rn

B1 (x)

Proof. Sufficiency. The left-hand side in (10.4.1) is equal to sup ϕ

|(R(D)u, ϕ)| |(P F u, (R/P )Φ)| = sup , ϕL2 ((1+|x|2 )l/2 ) ΦW2l Φ

where F is the Fourier transform in Rn . The last supremum does not exceed P F uL2 sup Φ

(R/P )ΦL2 = R/P M (W2l →L2 ) P (D)uL2 . ΦW2l

Necessity. Let Qε (ξ) = (|P (ξ)|2 + ε)1/2 ,

where

ε = const > 0,

−1 −1 and let Qε (D) = F −1 Qε F . Since F S = S, the operator Q−1 Qε F ε (D) = F −1 maps S into itself. We set u = Qε (D)f , where f is an arbitrary function in S. By (10.4.1),

R(D)Q−1 ε (D)f L2 ((1+|x|2 )−l/2 ) ≤ C f L2 which is the same as |(Ψ, (R/Qε )Φ)| ≤ C Ψ L2 ΦW2l . Consequently, for all Φ ∈ W2l , we have (R/Qε )ΦL2 ≤ C ΦW2l . Passing to the limit as ε → +0, we complete the proof.



A rough corollary of Theorem 10.4.1 is the sufficiency of (10.4.3) for the validity of R(D)u; KL2 ≤ C(K)P (D)uL2 , (10.4.4) where u ∈ S, K is an arbitrary compact set in Rn and C(K) is a constant independent of u. Namely, the following theorem holds:

10.4 Domination of Differential Operators in Rn

389

Theorem 10.4.2. [Maz8] Inequality (10.4.4) is true if and only if the functions R and P satisfy (10.4.3). Proof. We need to prove only the necessity. Let x be a fixed point in Rn , let χx be the characteristic function of the ball B1 (x), and let K be the cube {x ∈ Rn : |xi | ≤ 1, 1 ≤ i ≤ n}. We define a family of functions {uε,h } by F uε,h =


Rχ  x , |P |2 + ε h

where (ϕ)h is the mollification of ϕ with radius h and ε is a positive number. Clearly, uε,h ∈ S and it can be put into (10.4.4). We then have + PR + + + R + + + + ; B1 (x)+ ≤ + ; B (x) lim P (D)uε,h L2 = + 2 + . 1 2 1/2 h→0 |P | +ε L2 L2 (|P | +ε)

(10.4.5)

On the other hand, R(D)uε,h ; KL2 = c ψ ∗ RF uε,h L2 , where ψ(ξ) =

-

ξi−1 sin ξi .

1≤i≤n

Therefore



R(D)uε,h ; K2L2

≥c

B1 (x)

The right-hand side tends to   c B1 (x)

B1 (x)

 2 ψ(ξ − η)R(η)(F uε,h )(η) dη dξ .

ψ(ξ − η)

|R(η)|2 dη 2 dξ |P (η)|2 + ε

as h → 0. Here |ξ − η| < 2 and consequently ψ(ξ − η) ≥ const > 0. We arrive at the inequality  |R(η)|2 dη . (10.4.6) lim inf R(D)uε,h ; KL2 ≥ c 2 h→0 B1 (x) |P (η)| + ε Now, from (10.4.4) for uε,h and from (10.4.5) and (10.4.6) we obtain R(|P |2 + ε)−1/2 ; B1 (x)L2 ≤ c . Passing to the limit as ε → 0, we get (10.4.3).



11 Schr¨ odinger Operator and M (w21 → w2−1 )

11.1 Introduction The results presented in this chapter were obtained in [MV2]. Here a characterization is given for the class of measurable functions (or, more generally, real- or complex-valued distributions) V such that the Schr¨ odinger operator H = −∆ + V maps the energy space w21 (Rn ) to its dual w2−1 (Rn ). Similar results are obtained for the inhomogeneous Sobolev space W21 (Rn ). In other words, a complete solution is given to the problem of the relative formboundedness of the potential energy operator V with respect to the Laplacian −∆, which is fundamental to quantum mechanics. Relative compactness criteria for the corresponding quadratic forms are established as well. Analogous boundedness and compactness criteria for Sobolev spaces on domains Ω ⊂ Rn are obtained under mild restrictions on ∂Ω. The abovementioned mapping property of H is equivalent to the classical inequality   2 ≤C |u(x)| V (x) dx |∇u(x)|2 dx, u ∈ C0∞ (Rn ), (11.1.1) Rn

Rn

holding. Here the “indefinite weight” V may change sign, or even be a complexvalued distribution on Rn , n ≥ 3. (In the latter case, the left-hand side of (11.1.1) is understood as | < V u, u > |, where < V ·, · > is the quadratic form associated with the corresponding multiplication operator V .) We also characterize an analogous inequality for the inhomogeneous Sobolev space W21 (Rn ), n ≥ 1:  2 n |u(x)| V (x) dx R  ≤C [ |∇u(x)|2 + |u(x)|2 ] dx, u ∈ C0∞ (Rn ). (11.1.2) Rn

Such inequalities are used extensively in the spectral and scattering theory of the Schr¨ odinger operator H = H0 + V , where H0 = −∆ is the Laplacian V.G. Maz’ya, T.O. Shaposhnikova, Theory of Sobolev Multipliers, Grundlehren der mathematischen Wissenschaften 337, c Springer-Verlag Berlin Hiedelberg 2009


391

392

11 Schr¨ odinger Operator and M (w21 → w2−1 )

on Rn . In particular, (11.1.2) is equivalent to the concept of the relative boundedness of V (potential energy operator) with respect to H0 in the sense of quadratic forms. It follows from the polarization identity u ¯v =

 1 |u + v|2 − |u − v|2 − i|u − iv|2 + i|u + iv|2 4

(11.1.3)

that (11.1.1) can be restated equivalently in terms of the sesquilinear form < V u, v > as | < V u, v > | ≤ C ∇uL2 ∇vL2 , (11.1.4) for all u, v ∈ C0∞ (Rn ). In other words, it is equivalent to the boundedness of the operator H = H0 + V , H : w21 (Rn ) → w2−1 (Rn ),

n ≥ 3.

(11.1.5)

Here the energy space w21 (Rn ) is defined as the completion of C0∞ (Rn ) with respect to the norm ∇uL2 , and w2−1 (Rn ) is the dual of w21 (Rn ). Similarly, (11.1.2) means that H is a bounded operator which maps W21 (Rn ) to W2−1 (Rn ), n ≥ 1. Note that (11.1.4) means that the distribution V belongs the space of multipliers M (w21 (Rn ) → w2−1 (Rn )). Before stating the main results we note that we use some expressions involving pseudodifferential operators, e.g. ∇∆−1 or (−∆)−1/2 , which will be defined in the main body of this chapter. As before, Rn will be omitted in notations of norms and integrals. Theorem 11.1.1. Let V be a complex-valued distribution on Rn , n ≥ 3. Then V ∈ M ((w21 → w2−1 ), i.e. (11.1.1) holds, if and only if V is the divergence of a vector field Γ : Rn → Cn such that   |u(x)|2 |Γ(x)|2 dx ≤ C |∇u(x)|2 dx, (11.1.6) where the constant is independent of u ∈ C0∞ . The vector field Γ ∈ L2,loc can be chosen as Γ = ∇∆−1 V . Equivalently, the Schr¨ odinger operator H = H0 +V acting from w21 to w2−1 is bounded if and only if V = div Γ with Γ subject to (11.1.6). Furthermore, the corresponding multiplication operator V : w21 → w2−1 is compact if and only if V = div Γ, where Γ is such that the embedding w21 ⊂ L2 (|Γ|2 ) is compact. Obviously, (11.1.6) means that Γ ∈ M (w21 → L2 ). Recall that the last space was discussed in Sect. 2.8.

11.2 Characterization of M (w21 → w2−1 ) and the Schr¨ odinger Operator on w21

393

We remark that once V is written as V = div Γ, the implication (11.1.6) =⇒ (11.1.1) becomes trivial. It follows using integration by parts and the Schwarz inequality (compare with Sect. 2.5, where a similar argument was used to obtain sufficient conditions for the inclusion into M (Wpm → Wp−k )). On the other hand, the converse statement (11.1.1)=⇒(11.1.6), where Γ = ∇∆−1 V , is rather delicate. Its proof is based on a special factorization δ of the equilibrium harmonic potential of functions in w21 involving powers PK PK associated with an arbitrary compact set K ⊂ Rn of positive Wiener’s δ , where ultimately δ is picked so that capacity. New sharp estimates for PK 1 < 2δ < n/(n − 2), are established in a series of lemmas and propositions in Sect. 11.1.22. One also makes use of the fact that standard Mikhlin-Calderonδ , and the correspondZygmund operators are bounded on L2 with a weight PK ing operator norm bounds do not depend on K. We now briefly outline the contents of this chapter. In Sect. 11.2 we define the Schr¨ odinger operator on the energy space w21 , and characterize the basic inequality (11.1.1). The compactness problem is treated in Sect. 11.3. Analogous results for the Sobolev space W21 are obtained in Sect. 11.4, while Sect. 11.5 is devoted to similar problems on a domain Ω ⊂ Rn for a broad class of Ω, including those with Lipschitz boundaries.

11.2 Characterization of M (w21 → w2−1 ) and the Schr¨ odinger Operator on w21 In this section, we assume that n ≥ 3, since for the homogeneous space w21 (Rn ) our results become vacuous if n = 1 and n = 2, (11.1.1) then implying that V = 0. (Analogous results for inhomogeneous Sobolev spaces W21 (Rn ) are valid for all n ≥ 1; see Sect. 11.4 below.) For V ∈ (C0∞ ) , consider the multiplication operator on C0∞ defined by < V u, v > := < V, u ¯ v >,

u, v ∈ C0∞ ,

(11.2.1)

where < ·, · > represents the usual pairing between C0∞ and its dual (C0∞ ) . Elements of w21 (Rn ), for n ≥ 3, are weakly differentiable functions u ∈ n n 2n (R ) whose first-order weak derivatives lie in L2 (R ). By Hardy’s inL n−2 equality, an equivalent norm on w21 is given by $ uw21 =



−2

|x|

|u(x)| + |∇u(x)| 2

2



% 12 dx

.

If the sesquilinear form < V ·, · > is bounded on w21 × w2−1 : | < V u, v > | ≤ c ∇uL2 ∇vL2 ,

u, v ∈ C0∞ ,

(11.2.2)

394

11 Schr¨ odinger Operator and M (w21 → w2−1 )

where the constant c is independent of u and v, then V u ∈ w2−1 , and the multiplication operator can be extended by continuity onto w21 . As usual, this extension is also denoted by V . Note that the least constant c in (11.2.2) is equal to the multiplier norm: V M (w1 →w−1 ) = sup { V uw−1 : uw21 ≤ 1, 2

2

2

u ∈ C0∞ }.

¯v > defined by the For V ∈ M (w21 → w2−1 ), we extend the form < V, u right-hand side of (11.2.1) to the case where both u and v are in w21 . This can be done by letting < V u, v > := lim < V uN , vN >, N →∞

where u = lim uN , N →∞

and v = lim vN N →∞

in w21

with uN , vN ∈ C0∞ . We now state the main result of this section for arbitrary (complex-valued) distributions V . By L2,loc = L2,loc ⊗Cn we denote the space of vector-functions Γ = (Γ1 , . . . , Γn ) such that Γi ∈ L2,loc , i = 1, . . . , n. Theorem 11.2.1. Let V ∈ (C0∞ ) . Then V ∈ M (w21 → w2−1 ), i.e., the inequality (11.2.3) | < V u, v > | ≤ c uw21 vw21 holds for all u, v ∈ C0∞ , if and only if there is a vector field Γ ∈ L2,loc such that V = div Γ, and (11.1.6) holds for all u ∈ C0∞ . The vector field Γ can be chosen in the form Γ = ∇ ∆−1 V. Proof. Suppose that V = div Γ, where Γ satisfies (11.1.6). Then using integration by parts and the Schwarz inequality we obtain: | < V u, v > | = | < V, u ¯ v > | = | < Γ, v ∇¯ u > + < Γ, u ¯ ∇v > | √ uL2 + ΓuL2 ∇vL2 ≤ 2 C ∇uL2 ∇vL2 , ≤ Γ¯ v L2 ∇¯ where C is the constant in (11.1.6). This completes the proof of the “if” part of Theorem 11.2.1. The proof of the “only if” part of Theorem 11.2.1 is based on several lemmas and propositions. In the next lemma, we show that Γ = ∇ ∆−1 V ∈ L2,loc , and give a crude preliminary estimate of the rate of its decay at infinity. Lemma 11.2.1. Suppose that V ∈ M (w21 → w2−1 ).

(11.2.4)

11.2 Characterization of M (w21 → w2−1 ) and the Schr¨ odinger Operator on w21

Then

395

Γ = ∇ ∆−1 V ∈ L2,loc

and V = div Γ

in

(C0∞ ) .

Moreover, for any ball BR (x0 ) (R > 0) and > 0,  |Γ(x)|2 dx ≤ C(n, ) Rn−2+ V 2M (w1 →w−1 ) , 2

BR (x0 )

(11.2.5)

2

where R ≥ max{1, |x0 |}. Proof. Suppose that V ∈ M (w21 → w2−1 ). Define the vector field Γ ∈ (C0∞ ) by < Γ, φ > = − < V, ∆−1 div φ >,

(11.2.6)

for every φ ∈ C0∞ ⊗ Cn . In particular, < Γ, ∇ψ > = − < V, ψ >,

ψ ∈ C0∞ ,

(11.2.7)

i.e., V = div Γ in (C0∞ ) . We first have to check that the right-hand side of (11.2.6) is well-defined, which a priori is not obvious. For φ ∈ C0∞ ⊗ Cn , let w = ∆−1 div φ, where −∆−1 f = I2 f is the Newtonian potential of f ∈ C0∞ . Clearly, w(x) = O(|x|1−n ) and hence

and |∇w(x)| = O (|x|−n ) as

|x| → ∞,

w = ∆−1 div φ ∈ w21 ∩ C ∞ .

We will show below that w = u v, where u is real-valued, and both u and v are in w21 ∩ C ∞ . Then, since V ∈ M (w21 → w2−1 ), it follows that < V, w > =< V u, v > is defined through the extension of the multiplication operator V as explained above. For our purposes, it is important to note that this extension of < V, w > to the case where w = u ¯v, and u, v ∈ w21 ∩ C ∞ , is independent of the choice of factors u and v. To demonstrate this, we define a real-valued cutoff function ηN (x) = η(N −1 |x|),

where

η ∈ C ∞ (R+ ),

so that η(t) = 1 for 0 ≤ t ≤ 1 and η(t) = 0 for t ≥ 2. Note that ∇ηN is supported in the annulus N ≤ |x| ≤ 2N , and |∇ηN (x)| ≤ c |x|−1 . It follows easily (for instance, from Hardy’s inequality) that lim ηN u − uw21 = 0,

N →∞

u ∈ w21 .

Then letting uN = ηN u

and vN = ηN v,

11 Schr¨ odinger Operator and M (w21 → w2−1 )

396

2 so that uN vN = ηN w, we define < V, w > explicitly by setting: 2 w>. < V, w > := lim < V uN , vN >= lim < V, ηN N →∞

N →∞

This definition is independent of the choice of η and the factors u, v. Moreover, ¯ v; u, v ∈ w21 ∩ C ∞ }, | < V, w > | ≤ C inf{uw21 vw21 : w = u where C = V M (w1 →w−1 ) . 2

2

Now we fix > 0 and factorize: w(x) = ∆−1 div φ(x) = u(x) v(x), where u(x) = (1 + |x|2 )−

n−2+ 4

and v(x) = (1 + |x|2 )

n−2+ 4

∆−1 div φ(x). (11.2.8)

Obviously, u ∈ w21 ∩ C ∞ , and uw21 = c(n, ) < ∞. It is easy to see that v ∈ w21 ∩C ∞ as well. Furthermore, the following statement holds. Proposition 11.2.1. Suppose that φ ∈ C ∞ , and supp φ ⊂ BR (x0 ). Let v be defined by (11.2.8), where 0 < < 2. Then vw21 ≤ c(n, ) R

n−2+ 2

φL2 ,

(11.2.9)

for R ≥ max{1, |x0 |}. Proof. Since φ is compactly supported, it follows that |∆−1 div φ(x)| ≤ c(n) I1 |φ|(x), Hence c(n, ) vw21 ≤ (1 + |x|2 ) +(1 + |x|2 ) −1

Note that ∇ ∆ weight

n−2+ 4

n−4+ 4

x ∈ Rn .

∇ ∆−1 div φ(x)L2

I1 |φ|(x)L2 .

div is a Mikhlin-Calderon-Zygmund operator, and that the w(x) = (1 + |x|2 )

n−2+ 2

belongs to the Muckenhoupt class A2 if 0 < < 2 (see [CF]). Applying the corresponding weighted norm inequality, we have: (1 + |x|2 )

n−2+ 4

∇ ∆−1 div φ(x)L2

11.2 Characterization of M (w21 → w2−1 ) and the Schr¨ odinger Operator on w21

≤ c(n, ) (1 + |x|2 )

n−2+ 4

|φ(x)| L2 .

397

(11.2.10)

The other term is estimated by the weighted Hardy inequality:  n−4+ (I1 |φ|(x))2 (1 + |x|2 ) 2 dx  n−2+ (11.2.11) ≤ c(n, ) |φ(x)|2 (1 + |x|2 ) 2 dx. Clearly, (1 + |x|2 )

n−2+ 4

φ(x)L2 ≤ c(n, ) R

n−2+ 2

φL2 .

Hence, combining (11.2.10), (11.2.11), and the preceding estimate, we obtain the desired inequality (11.2.9). The proof of Proposition 11.2.1 is complete.  Now let us prove (11.2.5). Suppose that φ ∈ C ∞ ⊗ Cn , and supp φ ⊂ BR (x0 ). Then by (11.2.6) and Proposition 11.2.1, | < Γ, φ > | = | < V, u v > | ≤ V M (w1 →w−1 ) uw21 vw21 2

≤ C(n, ) R

n−2+ 2

2

V M (w1 →w−1 ) φL2 . 2

2

(11.2.12)

Taking the supremum over all φ supported in BR (x0 ) with the unit L2 -norm, we arrive at (11.2.5). The proof of Lemma 11.2.1 is complete. It remains to prove the main estimate (11.1.6) of Theorem 11.2.1. For this aim, it suffices to establish the inequality  |Γ(x)|2 dx ≤ c(n) V 2M (w1 →w−1 ) c2,1 (e), (11.2.13) 2

e

2

for every compact set e ⊂ Rn . Notice that in the special case e = BR (x0 ), the preceding estimate gives a sharper version of (11.2.5):  |Γ(x)|2 dx ≤ C(n) Rn−2 V 2M (w1 →w−1 ) , x0 ∈ Rn , R > 0. 2

BR (x0 )

2

Without loss of generality we assume that c2,1 (e) > 0; otherwise mesn e = 0, and (11.2.13) holds. Denote by P (x) = Pe (x) the equilibrium potential on e. It is well known that P is the Newtonian potential of a positive measure which gives a solution to several variational problems. This measure νe is called the equilibrium measure for e. We list some standard properties of νe and its potential Pe (x) = I2 νe (x) which will be used below: (a) (b) (c)

supp νe ⊂ e; Pe (x) = 1 dνe − a.e.; νe (e) = c2,1 (e) > 0;

(d) (e)

∇Pe 2L2 = c2,1 (e); sup Pe (x) ≤ 1.

x∈Rn

(11.2.14)

398

11 Schr¨ odinger Operator and M (w21 → w2−1 )

The rest of the proof of Theorem 11.2.1 is based on some inequalities involving the powers Pe (x)δ which are established below. Proposition 11.2.2. Let δ > 12 and let P = Pe be the equilibrium potential of a compact set e of positive capacity. Then 4 δ ∇P δ L2 = √ c2,1 (e). (11.2.15) 2δ − 1 Proof. Clearly,   |∇P (x)∆ |2 dx = δ 2 |∇P (x)|2 P (x)2δ−2 dx.

(11.2.16)

Using integration by parts, together with the properties −∆P = νe (understood in the distributional sense) and P (x) = 1 dνe -a.e., we have   |∇P (x)|2 P (x)2δ−2 dx = ∇P (x) · ∇P (x) P (x)2δ−2 dx  =

 2δ−1

P (x)

dνe − (2δ − 2)

|∇P (x)|2 P (x)2δ−2 dx

 = c2,1 (e) − (2δ − 2)

|∇P (x)|2 P (x)2δ−2 dx.

The integration by parts above is easily justified for δ > behavior of the potential and its gradient at infinity

1 2

by examining the

c1 |x|2−n ≤ P (x) ≤ c2 |x|2−n , |∇P (x)| = O (|x|1−n ),

as

|x| → ∞.

(11.2.17)

It follows from these calculations that  (2δ − 1) |∇P (x)|2 P (x)2δ−2 dx = c2,1 (e). Combining this with (11.2.16) yields (11.2.15). The proof of Proposition 11.2.2 is complete.  Remark 11.2.1. For δ ≤ 12 , it is easy to see that ∇P δ ∈ L2 . In the next lemma we demonstrate that ∇vL2 is equivalent to the weighted norm P −δ ∇(vP δ )L2 . Lemma 11.2.2. Let δ > 0, and let v ∈ w21 . Then  dx ≤ (δ + 1)(4δ + 1) ∇v2L2 . ∇v2L2 ≤ |∇(v P δ )(x)|2 P (x)2δ

(11.2.18)

11.2 Characterization of M (w21 → w2−1 ) and the Schr¨ odinger Operator on w21

399

Proof. Without loss of generality we may assume that v is real-valued. We first prove (11.2.18) for v ∈ C0∞ . The general case will follow using an approximation argument. Clearly,   dx = |∇v(x) + δ v(x) ∇P (x) P (x)−1 |2 dx |∇(v P δ )(x)|2 2δ P (x)    2 v(x) 2 2 2 |∇P (x)| dx. dx + 2δ ∇v · ∇P (x) v(x) = |∇v(x)| dx + δ 2 P (x) P (x) Integration by parts and the equation −∆P = νe (understood in the distributional sense) give    |∇P (x)|2 v(x) 2 dνe (x) dx = v(x) dx + v(x)2 2 ∇v · ∇P (x) dx. P (x) P (x) P (x)2 Using this identity, we rewrite the preceding equation in the form   dx = |∇v(x)|2 dx |∇(v P δ )(x)|2 2δ P (x)   |∇P (x)|2 dνe (x) . +δ(δ + 1) v(x)2 dx + δ v(x)2 2 P (x) P (x)

(11.2.19)

The lower estimate in (11.2.18) is now obvious provided the last two terms on the right-hand side of the preceding equation are finite. They are estimated in the following proposition, which holds for Newtonian potentials of arbitrary (not necessarily equilibrium) positive measures. Proposition 11.2.3. Let ω be a positive Borel measure on Rn such that P (x) = I2 ω(x) ≡ ∞. Then the inequalities hold:  |∇P (x)|2 v(x)2 dx ≤ 4 ∇v2L2 , v ∈ C0∞ , (11.2.20) P (x)2 and

 v(x)2

dω(x) ≤ ∇v2L2 , P (x)

v ∈ C0∞ .

(11.2.21)

Proof. Suppose that v ∈ C0∞ . Then A = supp v is a compact set, and obviously inf P (x) > 0.

x∈A

Without loss of generality we may assume that ∇P ∈ L2,loc , and hence the left-hand side of (11.2.20) is finite. (Otherwise we replace ω by its convolution with a compactly supported mollifier: ωt = ω ∗ t , and complete the proof by applying the estimates given below to P (x) = I2 ωt (x), and then passing to the limit as t → ∞.)

400

11 Schr¨ odinger Operator and M (w21 → w2−1 )

Using integration by parts together with the equation −∆P = ω as above, and applying the Schwarz inequality, we get    v(x) |∇P (x)|2 2 dω(x) = 2 ∇v(x) · ∇P (x) dx v(x)2 dx + v(x) P (x)2 P (x) P (x)  ≤2

|∇P (x)|2 v(x) dx P (x)2 2

 12 

|∇v(x)| dx 2

 12

for all v ∈ C0∞ . The preceding inequality obviously yields both (11.2.20) and (11.2.21). This completes the proof of Proposition 11.2.3.  Remark 11.2.2. The constants 4 and 1 respectively in (11.2.20) and (11.2.21) are sharp. Indeed, if ω is a point mass at x = 0, it follows that P (x) = c(n) |x|2−n . Hence, (11.2.20) boils down to the classical Hardy inequality   dx 4 (11.2.22) |∇u(x)|2 dx, u ∈ C0∞ , |u(x)|2 2 ≤ |x| (n − 2)2 with the best constant 4/(n − 2)2 . To show that the constant in (11.2.21) is sharp, it suffices to let ω = νe for a compact set e of positive capacity, so that P (x) = 1 dω-a.e. and νe (e) = c2,1 (e), and then minimize the right-hand side over all v ≥ 1 on e, where v ∈ C0∞ . We now complete the proof of Lemma 11.2.2. Combining (11.2.19) with (11.2.20) and (11.2.21) (with νe in place of ω), we arrive at the estimate  dx ≤ (δ + 1)(4δ + 1) ∇v2L2 , ∇v2L2 ≤ |∇(v P δ )(x)|2 P (x)2δ for all v ∈ C0∞ . To verify this inequality for arbitrary v in w21 , let v = lim vN N →∞

both in w21 and dx-a.e. for vN ∈ C0∞ . Now put vN in place of v in (11.2.20) and let N → ∞. Using Fatou’s lemma, we see that (11.2.20) holds for all v ∈ w21 . Hence  |∇P (x)|2 lim dx = 0, |vN (x) − v(x)|2 N →∞ P (x)2 and consequently  lim |∇(vN P δ )(x)|2 N →∞

dx = lim P 2δ (x) N →∞

 |∇vN (x) + δvN (x)

∇P (x) 2 | dx P (x)

11.2 Characterization of M (w21 → w2−1 ) and the Schr¨ odinger Operator on w21



=

|∇v(x) + δv(x)

∇P (x) 2 | dx = P (x)



|∇(v P δ )(x)|2

401

dx . P 2δ (x)

Thus, the proof of the general case is completed by putting vN in place of v in (11.2.18), and letting N → ∞. The proof of Lemma 11.2.2 is complete.  Remark 11.2.3. In what follows only the lower estimate in (11.2.18) will be used, together with the fact that P −δ ∇(vP δ )L2 < ∞ for every v ∈ w21 . In the next proposition, we extend the equation < V, w > = − < Γ, ∇w > to the case where w = u v, where both u and v lie in w21 , are locally bounded, and have a certain decay at infinity. Proposition 11.2.4. Suppose that V ∈ M (w21 → w2−1 ), and Γ = ∇ ∆−1 V ∈ L2,loc is defined as in Lemma 11.2.1. Suppose that w = u v, where u, v ∈ w21 , and |u(x)| ≤ C (1 + |x|2 )−β/2 ,

|v(x)| ≤ C (1 + |x|2 )−β/2 ,

x ∈ Rn , (11.2.23)

for some β > (n − 2)/2. Then Γ · ∇w ¯ is integrable, and  < V, w > = − Γ · ∇w(x) ¯ dx.

(11.2.24)

Proof. Clearly, 



|Γ(x)| |u(x)| dx

|Γ · ∇w(x)| ¯ dx ≤

2

 +

|Γ(x)| |v(x)| dx 2

2

2

 12 

 12 

|∇v(x)| dx 2

Rn

|∇u(x)| dx 2

 12

 12 .

To show that the right-hand side is finite, note that, for every > 0 and R ≥ 1,  |Γ(x)|2 dx ≤ C Rn−2+ , (11.2.25) |x|≤R

by Lemma 11.2.1. It is easy to see that the preceding estimate yields  (11.2.26) |Γ(x)|2 (1 + |x|2 )−β dx < ∞, for β > (n − 2)/2. Indeed, pick ∈ (0, 2β − n + 2), and estimate    |Γ(x)|2 dx + |Γ(x)|2 |x|−2β dx |Γ(x)|2 (1 + |x|2 )−β dx ≤ |x|≤1

|x|>1

402

11 Schr¨ odinger Operator and M (w21 → w2−1 )

 ≤ c1 + c2

∞



1



|x|≤r ∞

≤ c1 + c2

|Γ(x)|2 dx r−2β−1 dx

rn−3−2β dx < ∞.

1

From this and (11.2.23) it follows that   |Γ(x)|2 |u(x)|2 dx < ∞, |Γ(x)|2 |v(x)|2 dx < ∞. Thus Γ · ∇w ¯ is integrable. To prove (11.2.24), we first assume that both u and v lie in w21 ∩ C ∞ , and satisfy (11.2.23). Let ηN (x) be a smooth cutoff function as in the proof of Lemma 11.2.1. Let uN = ηN u and vN = ηN v. Then by (11.2.7),  < V, uN vN >= − Γ · ∇(¯ uN v¯N )(x) dx  =−

 Γ · ∇¯ uN (x) v¯N (x) dx −

Note that 0 ≤ ηN (x) ≤ 1

Γ · ∇¯ vN (x) u ¯N (x) dx.

and |∇ηN (x)| ≤ C |x|−1 ,

which gives  |Γ · ∇¯ uN (x) v¯N (x)| + |Γ · ∇¯ uN (x) v¯N (x)| ≤ C |Γ(x)| |u(x)| |v(x)||x|−1  +|∇u(x)||v(x)| + |∇v(x)||u(x)| . Since v ∈ w21 ), it follows from Hardy’s inequality (or directly from (11.2.23)) that |v(x)||x|−1 ∈ L2 . Applying (11.2.26) and the Schwarz inequality, we conclude that the right-hand side of the preceding inequality is integrable. Thus (11.2.24) follows from the dominated convergence theorem in this case. It remains to show that the C ∞ restriction on u and v can be dropped. We set ur = u  φr , vr = v  φr , where φr (x) = r−n φ(x/r). Here φ ∈ C0∞ is a C ∞ -mollifier supported in B1 such that 0 ≤ φ(x) ≤ 1. It is not difficult to verify that ur and vr satisfy estimates (11.2.23). Obviously, |ur (x)| = |u ∗ φr (x)| ≤ Mu(x), where M is the Hardy–Littlewood maximal operator. We can suppose without loss of generality that (n − 2)/2 < β < n in (11.2.23). Notice that, for 0 < β < n, x ∈ Rn . M(1 + |x|2 )−β/2 ≤ C(1 + |x|2 )−β/2 ,

11.2 Characterization of M (w21 → w2−1 ) and the Schr¨ odinger Operator on w21

403

Hence, |ur (x)| ≤ Mu(x) ≤ C(1 + |x|2 )−β/2 ,

x ∈ Rn ,

(11.2.27)

where C does not depend on r, and a similar estimate holds for v. We also need the estimate |∇ur (x)| = |∇u  φr (x)| ≤ M|∇u|(x).

(11.2.28)

As was shown above,  < V, ur vr >= −

 Γ · ∇¯ ur (x) v¯r (x) dx −

Γ · ∇¯ vr (x) u ¯r (x) dx.

Moreover, by (11.2.27) and (11.2.28) we have |Γ · ∇¯ ur (x)¯ vr (x)| + |Γ · ∇¯ vr (x) u ¯r (x)| ≤ C|Γ(x)|(1 + |x|2 )−β/2 (M|∇u|(x) + M|∇v|(x)). Since u, v ∈ w21 , and M is a bounded operator on L2 , it follows that M|∇u| and M|∇v| lie in L2 . Applying (11.2.26) again, we see that the right-hand side of the preceding inequality is integrable. Thus, letting r → 0, and using the dominated convergence theorem, we obtain  ¯ dx, < V, w >= lim < V, ur vr >= − Γ · ∇w(x) r→0



which completes the proof of Proposition 11.2.4. Now we continue the proof of (11.2.13). Suppose that V ∈ M (w21 → w2−1 ), i.e., the inequality | < V u, v > | ≤ V M (w1 →w−1 ) uw21 vw21 2

2

holds, where u, v ∈ w21 . Let φ = (φ1 , . . . , φn ) be an arbitrary vector field in C0∞ ⊗ Cn , and let w = ∆−1 div φ = −I2 div φ,

(11.2.29)

so that φ = ∇w + s, Note that w ∈

w21

div s = 0.

∞

∩ C , since

w(x) = O(|x|1−n ) and |∇w(x)| = O (|x|−n ) as

|x| → ∞.

(11.2.30)

404

11 Schr¨ odinger Operator and M (w21 → w2−1 )

Now set u(x) = P (x)δ

and

v(x) =

w(x) , P (x)δ

(11.2.31)

where P (x) is the equilibrium potential of a compact set e ⊂ Rn , and 1 < 2δ < n/(n − 2). By (11.2.14) and (11.2.17) we have 0 ≤ P (x) ≤ 1

for all x ∈ Rn

and for |x| large.

P (x) ≤ c |x|2−n Hence

|P (x)|δ ≤ C (1 + |x|2 )−δ(n−2)/2 .

Since β = δ(n − 2) > (n − 2)/2, it follows that u satisfies (11.2.23). To verify that (11.2.23) holds for v = w P −δ , note that inf P (x) > 0 K

for every compact set K, and hence by (11.2.17) P (x)−δ ≤ C (1 + |x|2 )δ(n−2)/2 . Combining this estimate with (11.2.30), we conclude that |v(x)| ≤ C (1 + |x|2 )−β/2 , where β = −δ(n − 2) + n − 1 > (n − 2)/2. By Proposition 11.2.2 and Lemma 11.2.2 both u and v lie in w21 . Now applying Proposition 11.2.4, we obtain  < V u, v > = < V, w > = − Γ · ∇w(x) ¯ dx. Hence,

 Γ · ∇w(x) ¯ dx ≤ V M (w1 →w−1 ) ∇uL2 ∇vL2 . 2 2

By Lemma 11.2.2,  ∇v2L2 ≤ |∇(vP δ )(x)|2

dx = P (x)2δ

 |∇w(x)|2

dx < ∞. P (x)2δ

Applying this together with Proposition 11.2.2, we obtain the estimate  1 Γ · ∇w(x) ¯ dx ≤ C(δ) V M (w1 →w−1 (Rn )) c2,1 (e) 2 2

 ×

dx |∇w(x)| P (x)2δ 2

2

 12 .

(11.2.32)

11.2 Characterization of M (w21 → w2−1 ) and the Schr¨ odinger Operator on w21

405

To complete the proof of Theorem 11.2.1, we need one more estimate which involves powers of equilibrium potentials. Proposition 11.2.5. Let w be defined by (11.2.29) with φ ∈ C0∞ ⊗ Cn . Suppose that 1 < 2δ < n/(n − 2). Then   dx dx ≤ C(n, δ) |φ(x)|2 . (11.2.33) |∇w(x)|2 2δ P (x) P (x)2δ Proof. Note that ∇w is related to φ through the Riesz transforms Rj , j = 1, . . . , n ([St2]): ∇w =

n !

" R j R k φk ,

j = 1, . . . , n.

k=1

Since Rj are bounded operators on L2 (ρ) with a weight ρ in the Muckenhoupt class A2 (Rn ) ([CF], [St3]), we have ∇wL2 (ρ) ≤ C φL2 (ρ) , where the constant C depends only on the Muckenhoupt constant of the weight. Let ρ(x) = P (x)−2δ . It is easily seen that inf P (x) > 0

x∈K

for every compact set K, and hence P (x)−2δ ∈ L1, loc (Rn ). It was proved in [MV1] that P (x)2δ is an A2 -weight, provided that 1 < 2δ < n/(n − 2). Moreover, its Muckenhoupt constant depends only on n and δ, but not on the compact set e. (See [MV1], p. 95, the proof of Lemma 3.1 in the case p = 2.) Clearly, the same is true for ρ(x) = P (x)−2δ . This completes the proof of Proposition 11.2.5.  Now we are in a position to complete the proof of Theorem 11.2.1. Recall that from (11.2.7) and Proposition 11.2.4 it follows that   < V, w > = − Γ · ∇w(x) ¯ dx = − Γ · φ(x) dx. Using (11.2.32) and Proposition 11.2.5, we obtain   12  1 |φ(x)|2 Γ · φ(x) dx ≤ C(n, δ) V  2 dx M (w21 →w2−1 ) c2,1 (e) P (x)2δ for all φ ∈ C0∞ ⊗ Cn , and hence for all φ ∈ L2,loc . Let us pick R > 0 so that e ⊂ BR . Letting φ = χBR P 2δ Γ in the preceding inequality, we conclude that

406

11 Schr¨ odinger Operator and M (w21 → w2−1 )

 BR

|Γ(x)|2 P (x)2δ (x) dx

 12

1

≤ C(n, δ) V M (w1 →w−1 ) c2,1 (e) 2 . 2

2

Since P (x) ≥ 1 dx-a.e. on e (actually P (x) = 1 on e\E, where E is a polar set, i.e., c2,1 (E) = 0), it follows that  |Γ(x)|2 dx ≤ C(n, δ)2 V 2M (w1 →w−1 ) c2,1 (e). 2

e

2

Thus, (11.2.13) holds for every compact set e ⊂ Rn , and hence this yields (11.1.6). The proof of Theorem 11.2.1 is complete.  We now prove an analogue of Theorem 11.2.1 formulated in terms of (−∆)−1/2 V . Theorem 11.2.2. Under the assumptions of Theorem 11.2.1, it follows that V ∈ M (w21 → w2−1 ) if and only if

(−∆)−1/2 V ∈ M (w21 → L2 ).

Proof. By Theorem 11.2.1, ∇ ∆−1 V ∈ L2,loc is well defined in terms of distributions. We now have to show that (−∆)−1/2 V is also well defined. Since ∇ ∆−1 V lies in M (w21 → L2 ) ⊗ Cn , it follows from Corollary 3.2 in [MV1] that the Riesz transforms Rj (j = 1, . . . , n) are bounded operators on M (w21 → L2 ). Hence (−∆)−1/2 ∇ = {Rj }1≤j≤n is a bounded operator from M (w21 → L2 ) to M (w21 → L2 ) ⊗ Cn . Then (−∆)−1/2 V can be defined by (−∆)−1/2 V = (−∆)−1/2 ∇ · ∇ ∆−1 V as an element of M (w21 → L2 ). The proof of Theorem 11.2.2 is complete.



Remark 11.2.4. It is worthwhile to observe that the “na¨ıve” approach is to decompose V into its positive and negative parts: V = V+ − V− , and to apply the criteria in Sect. 1.2 to both V+ and V− . However, this procedure drastically diminishes the class of admissible weights V by ignoring a possible cancellation between V+ and V− . This cancellation phenomenon is evident for strongly oscillating weights considered below. Example 11.2.1. Let us set V (x) = |x|N −2 sin (|x|N ),

(11.2.34)

11.3 A Compactness Criterion

407

where N is an integer, N ≥ 3, which may be arbitrarily large. Obviously, both V+ and V− fail to satisfy (11.1.1) due to the growth of the amplitude at infinity. However, V (x) = div Γ(x) + O (|x|−2 ),

where

Γ(x) =

−1 x cos (|x|N ). (11.2.35) N |x|2

By Hardy’s inequality (11.2.22) with n ≥ 3, the term O (|x|−2 ) in (11.2.35) is harmless, whereas Γ clearly satisfies (11.1.6) since |Γ(x)|2 ≤ |x|−2 . This shows that V is admissible for (11.1.1), while |V | is obviously not. Similar examples of weights with strong local singularities can easily be constructed.

11.3 A Compactness Criterion In this section we give a compactness criterion for V ∈ M (w21 → w2−1 ). Denote ˚ (w1 → w−1 ) the class of compact multiplication operators acting from by M 2 2 1 w2 to w2−1 . Obviously, ˚ (w21 → w−1 ) ⊂ M (w21 → w−1 ), M 2 2 where the latter class was characterized in the preceding section. Theorem 11.3.1. Let V ∈ (C0∞ ) and n ≥ 3. Then ˚ (w1 → w−1 ) V ∈M 2 2 if and only if V = div Γ,

(11.3.1)

where Γ = (Γ1 , . . . , Γn ) is a vector field such that ˚ (w1 → L2 ), Γi ∈ M 2

i = 1, . . . , n.

Moreover, Γ can be represented in the form ∇∆−1 V , as in Theorem 11.2.1. Proof. Let V be given by (11.3.1), and let u belong to the unit ball B in w21 . Then V u = div (u Γ) − Γ · ∇u. (11.3.2) The set {div (u Γ) : u ∈ B} is compact in w2−1 because the set {u Γ : u ∈ B}

408

11 Schr¨ odinger Operator and M (w21 → w2−1 )

is compact in w2−1 . The set {Γ · ∇u : u ∈ B} is also compact in w2−1 since the set {|∇u| : u ∈ B} is bounded in L2 , and the multiplier operators Γ¯i , being adjoint to Γi (i = 1, . . . , n), are compact from L2 to w2−1 . This completes the proof of sufficiency of (11.3.1). We now prove the necessity. Pick F ∈ C ∞ (R+ ), where F (t) = 1 for t ≤ 1 and F (t) = 0 for t ≥ 2. For x0 ∈ Rn , δ > 0, and R > 0, define the cutoff functions κδ,x0 (x) = F (δ −1 |x − x0 |),

and ξR (x) = 1 − F (R−1 |x|).

Lemma 11.3.1. If f ∈ w2−1 , then lim sup κδ,x0 f w−1 = 0,

(11.3.3)

lim ξR f w−1 = 0.

(11.3.4)

2

δ→0 x0 ∈Rn

and R→∞

2

Proof. Let us prove (11.3.3). The distribution f has the form f = div φ, where φ = (φ1 , . . . , φn ) ∈ L2 . Hence, κδ,x0 f = div (κδ,x0 φ) − φ ∇κδ,x0 . Clearly, κδ,x0 f w−1 ≤ κδ,x0 |φ| L2 + c δ ∇κδ,x0 · φL2 2

≤ c  |φ|; B2δ (x0 )L2 . This proves (11.3.3). Since (11.3.4) is derived in a similar way, the proof of Lemma 11.3.1 is complete.  ˚ (w1 → w−1 ), then Lemma 11.3.2. If V ∈ M 2 2 lim sup κδ,x0 V M ˚ (w1 →w−1 ) = 0,

(11.3.5)

lim ξR V M ˚ (w1 →w−1 ) = 0.

(11.3.6)

δ→0 x0 ∈Rn

2

2

and R→∞

2

2

Proof. Fix > 0, and pick a finite number of fk ∈ w2−1 such that V u − fk w−1 < 2

for k = 1, . . . , N ( ), and for all u ∈ B, where B is the unit ball in w21 . Note that by Hardy’s inequality sup

x0 ∈Rn , δ>0

κδ,x0 M (w1 →w−1 ) ≤ c < ∞. 2

2

11.3 A Compactness Criterion

409

Next, κδ,x0 V uw−1 ≤ κδ,x0 (V u − fk )w−1 + κδ,x0 fk w−1 2

2

2

≤ c + κδ,x0 fk w−1 . 2

Hence, κδ,x0 M (w1 →w−1 ) ≤ c + κδ,x0 fk w−1 . 2

2

2

By Lemma 11.3.1, this gives (11.3.5), and the proof of (11.3.6) is quite similar. The proof of Lemma 11.3.2 is complete.  We can now complete the proof of the necessity part of Theorem 11.3.1. Suppose that ˚ (w21 → w−1 ). V ∈M 2 By Theorem 11.2.1, ∇ ∆−1 (ξR V )M (w21 →L2 ) ≤ c ξR V M (w21 →L2 ) . By the preceding estimate and (11.3.6), lim ∇ ∆−1 (ξR V )M (w21 →L2 ) = 0.

R→∞

Hence we can assume without loss of generality that V is compactly supported, e.g., supp V ⊂ B1 . To show that ˚ (w21 → L2 ), Γ = ∇ ∆−1 V ∈ M consider a√covering of the closed unit ball B1 by open balls Bk (k = 1, . . . , n) of radius n δ centered at the nodes xk of the lattice with mesh size δ. We introduce a partition of unity φk subordinate to this covering and satisfying the estimate |∇ φk | ≤ c δ −1 , so that supp φk ⊂ Bk∗ , where Bk∗ is the ball of √ radius 2 n δ concentric to Bk . Also, pick ψk ∈ C0∞ (Bk∗ ), where φk ψ k = φk ,

and |∇ ψk | ≤ c δ −1 .

We have 

N (δ)

∇ ∆V =

∇ ∆(φk V ) =

k=1



N (δ)

=

k=1



N (δ)

∇ ∆(φk ψk V )

k=1



N (δ)

ψk ∇ ∆(φk V ) +

[ ∇ ∆, ψk ] φk V,

k=1

where [ A, B ] = A B − B A is the commutator of the operators A and B. Since the multiplicity of the covering

410

11 Schr¨ odinger Operator and M (w21 → w2−1 )



N (δ)

Bk

k=1

depends only on n, it follows that (δ) + +N + + ψk ∇ ∆(φk V )+ + k=1

M (w21 →L2 )

≤ c(n)

sup 1≤k≤N (δ)

∇ ∆(φk V )M ˚ (w1 →L2 ) . 2

The last supremum is bounded by c φk V M ˚ (w1 →w−1 ) , 2

2

which is made smaller than any > 0 by choosing δ = δ( ) small enough. It remains to check that each function Φk := [ ∇ ∆, ψk ] φk V is a compact multiplier from w21 to L2 , k = 1, . . . , n. Indeed, the kernel of the operator V → [ ∇ ∆, ψk ] φk V is smooth, and hence |Φk (x)| = |( [ ∇ ∆, ψk ] φk V )(x)| ≤ ck (1 + |x|)1−n φk V w−1 2

1−n ≤ ck (1 + |x|)1−n V M , ˚ (w1 →w−1 ) φk w21 ≤ Ck (1 + |x|) 2

2

where the constant Ck does not depend on x. Since n > 2, this means that the multiplier operator Φk : w21 → L2 is compact. The proof of Theorem 11.3.1 is complete.  Remark 11.3.1. The compactness of the multipliers Γi : w21 → L2 , where i = 1, . . . , n, is obviously equivalent to the compactness of the embedding w21 ⊂ L2 (|Γ|2 ).

(11.3.7)

An analytic characterization of this property is equivalent to the inequalities  |Γ|2 dx e = 0, lim sup δ→0 {e:diam(e)≤δ} c2,1 (e)  |Γ|2 dx lim

sup

ρ→∞ e⊂Rn \B

(see [Maz2] and [Maz15], § 2.5).

e ρ

c2,1 (e)

=0

11.4 Characterization of M (W21 → W2−1 )

411

11.4 Characterization of M (W21 → W2−1 ) In this section, we characterize the class of multipliers V : W21 → W2−1 for n ≥ 1. Here W2−1 = (W21 ) , the dual of W21 . Let Jα (0 < α < +∞) denote the Bessel potential of order α. Every u ∈ W21 can be represented in the form u = J1 g, where c1 gL2 ≤ uW21 ≤ c2 gL2 . (See [St2].) Let S denote the Schwarz space of infinitely differentiable functions on Rn and S  its dual. We say that V ∈ S  is a multiplier from W21 to W2−1 if the sesquilinear form defined by < V u, v > := < V, u ¯v > is bounded on W21 × W21 : | < V u, v > | ≤ c uW21 vW21 ,

u, v ∈ S,

(11.4.1)

where the constant c is independent of u and v in Schwartz space S. As in the case of homogeneous spaces, the preceding inequality is equivalent to the boundedness of the corresponding quadratic form; i.e., it suffices to verify (11.4.1) for u = v. If (11.4.1) holds, then V defines a bounded multiplier operator from W21 to W2−1 . (Originally, it is defined on S, but by continuity is extended to W21 .) The corresponding class of multipliers is denoted by M (W21 → W2−1 ). Since I − ∆ : W21 → W2−1 is a bounded operator, it follows that V ∈ M (W21 → W2−1 ) if and only if the operator (I − ∆) + V : W21 → W2−1 is bounded. If V is a locally finite complex-valued measure on Rn , then (11.4.1) can be rewritten in the form  u(x) v(x) dV (x) ≤ c uW 1 vW 1 , (11.4.2) 2 2 where u, v ∈ S. We now characterize (11.4.2) in the general case of distributions V . Theorem 11.4.1. Let V ∈ S  . Then V ∈ M (W21 → W2−1 ) if and only if there exist a vector field Γ = {Γ1 , . . . , Γn } ∈ L2,loc and Γ0 ∈ L2,loc such that V = div Γ + Γ0 ,

(11.4.3)

412

and

11 Schr¨ odinger Operator and M (w21 → w2−1 )

 |u(x)|2 |Γi (x)|2 dx ≤ C u2W 1 ,

i = 0, 1, . . . , n,

2

(11.4.4)

where C does not depend on u ∈ S. In (11.4.3), one can set Γ = −∇ (I − ∆)−1 V,

and

Γ0 = (I − ∆)−1 V.

(11.4.5)

Proof. Suppose that V is represented in the form (11.4.3), and (11.4.4) holds. Then using integration by parts and the Schwarz inequality, we have | < V, u ¯ v > | = | < Γ, v ∇¯ u > + < Γ, u ¯ ∇v > + < Γ0 , u ¯v > | ≤ ΓvL2 ∇uL2 + ΓuL2 ∇vL2 + Γ0 uL2 vL2 √ ≤ 3 C uW21 vW21 , where C is the constant in (11.4.4). This proves the “if” part of Theorem 11.4.1. To prove the “only if” part, define Γ = {Γ1 , . . . , Γn } and Γ0 by (11.4.5). Then, for every j = 0, 1, . . . , n, we have Γj ∈ L2,loc , and the following crude estimate holds:  |Γj (x)|2 dx ≤ C(n, ) Rn−2+ V 2M (W 1 →W −1 ) , (11.4.6) 2

BR (x0 )

2

where R ≥ max{1, |x0 |}. The proof is based on the same argument as the proof of Lemma 11.2.1 in the homogeneous case. Now fix a compact set e ⊂ Rn such that diam (e) ≤ 1, and C2,1 (e) > 0. Denote by P (x) = Pe (x) the equilibrium potential of e which corresponds to the capacity C2,1 . Letting u(x) = P (x)δ

and

v(x) =

w(x) , P (x)δ

where 1 < 2δ < n/(n − 2) and w ∈ S, we have | < V, w > | ≤ V M (W 1 →W −1 ) P δ W21 ∇vW21 . 2

2

Calculations analogous to those of Propositions 11.2.2 - 11.2.5 yield 1

P δ W21 ≤ C(n, δ) C2,1 (e) 2 , and

$ ∇vW21 ≤ C(n, δ)

(|w(x)|2 + |∇w(x)|2 )

dx P (x)2δ

% 12 .

11.4 Characterization of M (W21 → W2−1 )

413

Combining the preceding inequalities, we obtain 1

| < V, w > | ≤ C(n, δ) V M (W 1 →W −1 ) C2,1 (e) 2 2

$ ×

2

dx (|w(x)| + |∇w(x)| ) P (x)2δ 2

2

% 12 .

Set w = (1−∆)−1 div φ, where φ is an arbitrary vector-field with components in S. Then the preceding estimate can be restated in the form $ 1

| < Γ, φ > | ≤ C(n, δ)C2,1 (e) 2

(|w(x)|2 +|∇w(x)|2 )

dx P (x)2δ

% 12 .

(11.4.7)

Unlike in the homogeneous case, P (x)−2δ is not a Muckenhoupt weight for Bessel potentials. To proceed, we need a localized version of the estimates used in Sect. 11.3. Lemma 11.4.1. Let P (x) = Pe (x) be the equilibrium potential of a compact set e of positive capacity C2,1 and such that e ⊂ B, where B = B1 (x0 ) is the ball of radius 1 centered at x0 ∈ Rn . Let w = (I − ∆)−1 ∇ ψ, where ψ ∈ C ∞ and supp ψ ⊂ B. Suppose that 1 < 2δ < n/(n − 2). Then   dx dx 2 2 ≤ C(n, δ) |ψ(x)|2 . (11.4.8) (|w(x)| + |∇w(x)| ) P (x)2δ P (x)2δ Proof. Let ν = νe be the equilibrium measure of the compact set e in the sense of the capacitiy C2,1 , so that P (x) = J2 ν(x) (see [AH]). Suppose first that n ≥ 3. Since both supp ν and supp ψ are contained in B, it follows that  dν(y) , x ∈ 2B, (11.4.9) P (x) = J2 ν(x) ∼ I2 ν(x) = c(n) |x − y|n−2 B where 2B is a concentric ball of radius 2. We set ρ(x) = I2 ν(x)−2δ . Then ρ(x) ∼ P (x)−2δ on 2B, and ρ(x) is an A2 -weight (see the proof of Proposition 11.2.4). Note that ∇w = ∇2 (I − ∆)−1 ψ, where ∇2 (I − ∆)−1 = {−Rj Rk ∆ (I − ∆)−1 },

j, k = 1, . . . , n.

Here Rj , j = 1, . . . , n, are the Riesz transforms which are bounded operators on L2 (ρ) (see [St3]). Since ∆ (I − ∆)−1 = I − (I − ∆)−1 , we have to show that J2 = (I − ∆)−1 is a bounded operator on L2 (ρ), and its norm is bounded by a constant which depends only on the Muckenhoupt

414

11 Schr¨ odinger Operator and M (w21 → w2−1 )

constant of ρ. It is not difficult to see that the same is true for more general α operators Jα = (I − ∆)− 2 , where α > 0. Indeed, denote by Gα (x) the kernel of the Bessel potential Jα . Then clearly, |Jα f (x)| = |Gα  f (x)| ≤ c(n, α) Mf (x)

∞ 

2kn

k=−∞

max

2k ≤|t|≤2k+1

Gα (t),

where Mf (x) is the Hardy–Littlewood maximal function. Standard estimates of Bessel kernels Gα (x) (see Sect. 3.2.5), show that ∞  k=−∞

2kn

max

2k ≤|t|≤2k+1

Gα (t) < ∞,

for every α > 0. Since M is bounded on L2 (ρ) (see [St3]), it follows that Jα f L2 (ρ) ≤ C f L2 (ρ) ,

(11.4.10)

where C depends only on n, α, and the Muckenhoupt constant of ρ. Applying (11.4.10) with α = 2, we get   dx |∇w(x)|2 ≤ C(n, δ) |ψ(x)|2 ρ(x) dx 2δ P (x) 2B  dx ≤ C(n, δ) |ψ(x)|2 . P (x)2δ Similarly,

|w(x)| = |∇ (I − ∆)−1 ψ(x)| ≤ C J1 |ψ|(x)

and, by (11.4.10) with α = 1,   dx |w(x)|2 ≤ C (J1 |ψ|(x))2 ρ(x) dx P (x)2δ 2B 2B   dx |ψ(x)|2 . ≤ C(n, δ) |ψ(x)|2 ρ(x) dx ≤ C(n, δ) P (x)2δ Now suppose that x ∈ (2B)c . Then, by standard estimates of the Bessel kernel as |x| → ∞ (see Sect. 3.2.5),  1−n 2 −|x| 2 e |ψ(y)| dy |∇w(x)| = |∇ J2 ψ(x)| ≤ C(n) |x| B

and |w(x)| ≤ C(n)|∇J2 ψ(x)| ≤ C |x|

−n 2

−|x|



e

B

|ψ(y)| dy.

Also, for x ∈ (2B)c , P (x) = J2 ν(x) ∼ |x| where ν(e) = C2,1 (e) > 0.

1−n 2

e−|x| ν(e),

|x| → ∞,

11.4 Characterization of M (W21 → W2−1 )

415

Now pick δ so that 1 < 2δ < min {2, n/(n − 2)}. Using the above estimates of w(x), ∇w(x), and P (x), and the inequality 2δ < 2, we get 2   dx −2δ (|w(x)|2 + |∇w(x)|2 ) ≤ C(n, δ) ν(e) |ψ(y)| dy . P (x)2δ (2B)c B By the Schwarz inequality,  2  |ψ(y)| dy ≤ |ψ(y)|2 B

B

dy P (y)2δ

 P (x)2δ dx. B

Applying Minkowski’s inequality and the fact that 2δ < n/(n − 2), we obtain   P (x)2δ dx ≤ (I2 ν)2δ dx ≤ C(n, δ) ν(e)2δ . B

Thus,

B

 (|w(x)|2 + |∇w(x)|2 ) (2B)c

dx ≤ C(n, δ) P (x)2δ



dx . P (x)2δ

|ψ(x)|2

This completes the proof of (11.4.8) for n ≥ 3. The cases n = 1 and n = 2 are treated in a similar way with obvious modifications. The proof of Lemma 11.4.1 is complete.  Let

w = (I − ∆)−1 div φ,

φ = {φk } ∈ S.

where

Applying Lemma 11.4.1 with ψ = φk , k = 1, . . . , n, we obtain   dx dx ≤ C(n, δ) |φ(x)|2 . (|w(x)|2 + |∇w(x)|2 ) P (x)2δ P (x)2δ This and (11.4.7) yield $ | < Γ, φ > | ≤ C(n, δ) C2,1 (e)

dx |φ(x)| P (x)2δ

1 2

2

% 12 .

By duality, the preceding inequality is equivalent to  |Γ(x)|2 P (x)2δ dx ≤ C(n, δ) V 2M (W 1 →W −1 ) C2,1 (e). 2

2

Since P (x) ≥ 1 a.e. on e, we obtain the desired estimate  |Γ(x)|2 dx ≤ C(n, δ) V 2M (W 1 →W −1 ) C2,1 (e). e

2

2

The corresponding inequality with Γ0 in place of Γ is verified in a similar way. By (2.3.3) with p = 2 these inequalities are equivalent to (11.4.4). The proof of Theorem 11.4.1 is complete. 

416

11 Schr¨ odinger Operator and M (w21 → w2−1 )

Remark 11.4.1. It is easy to see that in the sufficiency part of Theorem 11.4.1 the restriction on the “lower order” term Γ0 in (11.4.4) can be relaxed. It is enough to assume that Γ0 ∈ L1,loc is such that  (11.4.11) |u(x)|2 |Γ0 (x)| dx ≤ C u2W 1 . 2

Finally, we state a compactness criterion in the case of the space W21 analogous to that of Theorem 11.3.1. Theorem 11.4.2. Let V ∈ S  (Rn ), n ≥ 1. Then ˚ (W21 → W −1 ) V ∈M 2 if and only if V = div Γ + Γ0 , where Γ = (Γ1 , . . . , Γn ), and ˚ (W21 → L2 ), Γi ∈ M

i = 0, . . . , n.

Moreover, one can set Γ = −∇(I − ∆)−1 V

and

Γ0 = (I − ∆)−1 V,

as in Theorem 11.4.1. The proof of Theorem 11.4.2 requires only minor modifications outlined in the proof of Theorem 11.4.1, and is omitted here.

11.5 Characterization of the Space M (w ˚21 (Ω) → w2−1 (Ω)) Using dilation and the description of the space M (W21 → W2−1 ) given in the preceding section, we arrive at the following auxiliary statement. Corollary 11.5.1. Let V ∈ M (W21 → W2−1 ). Suppose that there exists a number d > 0 such that | < V, |u|2 > | ≤ c (∇u2L2 + d−2 u2L2 ),

(11.5.1)

where c does not depend on u ∈ C0∞ . Then V can be represented as V = div Γ + d−1 Γ0 , where Γ0 and Γ = (Γ1 , . . . , Γn ) are in M (W21 → L2 ), and  |Γi u(x)|2 dx ≤ C (∇u2L2 + d−2 u2L2 ), for all i = 0, 1, . . . , n.

(11.5.2)

(11.5.3)

11.5 Characterization of the Space M (˚ w21 (Ω) → w2−1 (Ω))

417

Now let Ω be an open set in Rn such that, for all u ∈ C0∞ (Ω), Hardy’s inequality holds:   dx |u(x)|2 ≤ const |∇u(x)|2 dx. (11.5.4) d∂Ω (x)2 Ω Ω Here d∂Ω (x) = dist (x, ∂Ω). It is well-known that (11.5.4) holds for a wide class of domains including those with Lipschitz boundaries. (See [Dav], [Lew], [MMP] for a discussion of Hardy’s inequality and related questions, including best constants, on domains Ω in Rn .) Let Qj be the cubes with side-length dj forming Whitney’s covering of Ω (see [St2], Sec. 5.1). Denote by Q∗j the open cube obtained from Q by dilation with coefficient 98 dj . The cubes Q∗j form an open covering of Ω of finite multiplicity which depends only on n. By {ηj } (ηj ∈ C0∞ (Q∗j )) we denote a smooth partition of unity subordinate to the covering {Qj } and such that |∇ηj (x)| ≤ c d−1 j . In the proof of the following theorem we also need the functions ζj ∈ C0∞ (Q∗j ) such that ζj (x) ηj (x) = ηj (x),

and |∇ζj (x)| ≤ c d−1 j .

(11.5.5)

In this section we deal with the space w ˚21 (Ω) defined as the completion −1 ∞ w21 (Ω)) is of C0 (Ω) in the norm ∇u; ΩL2 . By w2 (Ω) the dual space (˚ denoted. The next result is a characterization of the space M (˚ w21 (Ω) → w2−1 (Ω)). Theorem 11.5.1. (i) Let d∂Ω (x) = dist (x, ∂Ω), and let V = div Γ + d−1 ∂Ω Γ0 , where Γ = {Γ1 , . . . , Γn } and Γi ∈ M (˚ w21 (Ω) → L2 (Ω)),

i = 0, 1, . . . , n.

Suppose that (11.5.4) holds. Then V ∈ M (˚ w21 (Ω) → w2−1 (Ω)), and  V ; ΩM (˚ Γi ; ΩM (˚ w21 →L2 ) . w1 →w−1 ) ≤ c 2

2

(11.5.6)

0≤i≤n

(ii) Conversely, if V ∈ M (˚ w21 (Ω) → w2−1 (Ω)), then there exist Γ = (Γ1 , . . . , Γn ) and Γ0 such that Γi ∈ M (˚ w21 (Ω) → L2 (Ω)), and Moreover,

i = 0, 1, . . . , n,

V = div Γ + d−1 ∂Ω Γ0 . 

Γi ; ΩM (˚ w21 →L2 ) ≤ C V ; ΩM (˚ w1 →w−1 ) . 2

0≤i≤n

2

(11.5.7)

11 Schr¨ odinger Operator and M (w21 → w2−1 )

418

Proof. The proof of statement (i) is straightforward (see, e.g., the proof of Theorem 11.4.1 above). To prove (ii), note that, for all u, v ∈ C0∞ (Ω), and the functions ζj with the properties (11.5.5), we have | < V ηj , u v > | = | < V ηj , ζj u ζj v > | −1 −1 ≤ V ηj ; ΩM (˚ w1 →w−1 ) (∇uL2 + dj uL2 ) (∇vL2 + dj vL2 ). 2

2

Hence by Corollary 11.5.1, V ηj = div Γ(j) + d−1 j Γ0 , (j)

(j)

where Γ(j) and Γ0

(11.5.8)

satisfy the inequality  (j) |Γi u(x)|2 dx

2 ≤ C V ηj ; Ω2M (˚ (∇u2L2 + d−2 j uL2 ), w1 →w−1 ) 2

(11.5.9)

2

for all i = 0, 1, . . . , n. Multiplying (11.5.8) by ζj , we obtain V ηj = div (ζj Γ(j) ) + d−1 j Γ0

(j)

− Γ(j) ∇ζj .

We set Γ=



ζj Γ(j)

and

Γ0 =



j

(j)

(dj Γ0

− Γ(j) ∇ζj ).

j

If u ∈ C0∞ (Ω), then

 |(|Γ| + |Γ0 |) u|2 dx

≤c

  Ω

j

Ω

|Γ(j) ζj u|2 dx + d−2 j

 (j) |(dj Γ0 ζj − Γ(j) ∇ζj ) κj u|2 dx ,

 Ω

where κj ∈ C0∞ (Q∗j ), and κj = 1 on supp ζj . By (11.5.9), the last sum does not exceed  2 2 sup V ηj ; ΩM (˚ (|∇(κj u)|2 + d−2 j |κj u| ) dx. w1 →w−1 ) j

2

2

j

Ω

By Hardy’s inequality (11.5.4), this is bounded by  c V ; Ω2M (˚ |∇u|2 dx. w1 →w−1 ) 2

2

The proof of Theorem 11.5.1 is complete.

Ω



11.5 Characterization of the Space M (˚ w21 (Ω) → w2−1 (Ω))

419

Remark 11.5.1. In Theorem 11.5.1, one can replace  Γi ; ΩM (˚ w21 →L2 ) 0≤i≤n

by the equivalent norm sup sup j

e⊂Qj

(|Γ| + |Γ0 |); eL2 C2,1 (e, Q∗j ) 2 1

.

(11.5.10)

In the case n > 2, one can use Wiener’s capacity c2,1 in place of C2,1 (·, Q∗j ) (see Sect. 13.2.2). ˚ (˚ We now characterize the class of compact multipliers, M w21 (Ω) → We use the same notation as in the previous section.

w2−1 (Ω)).

Theorem 11.5.2. Under the assumptions of Theorem 11.5.1, a distribution ˚ (˚ V is in M w21 (Ω) → w2−1 (Ω)) if and only if V = div Γ + d−1 ∂Ω Γ0 ,

(11.5.11)

˚ (˚ where Γi ∈ M w21 (Ω) → L2 (Ω)) for i = 0, 1, . . . , n. Proof. Suppose that V is given by (5.11). Let u be an arbitrary function in the unit ball B of w21 (Ω). Then V u = div (u Γ) − Γ + d−1 ∂Ω u Γ0 . The set {div (u Γ) : u ∈ B} is compact in w2−1 (Ω) since the set {u Γ : u ∈ B} is compact in L2 (Ω). The sets {∇u · Γ : u ∈ B} and {d−1 ∂Ω Γ0 u : u ∈ B} are also compact in w2−1 (Ω) since the sets {|∇ u| : u ∈ B} and

{d−1 ∂Ω u : u ∈ B}

are bounded in L2 (Ω), and the multiplier operators Γ¯i : L2 (Ω) → w2−1 (Ω), i = 1, . . . , n are compact, being adjoint to Γi . This completes the proof of the “if” part of Theorem 11.5.2. To prove the “only if” part let us assume that O ∈ Rn \Ω. Then, for any x ∈ Ω, it follows that |x| ≥ d∂Ω (x), and the inequality   |u(x)|2 dx ≤ c |∇u(x)|2 dx (11.5.12) 2 Ω |x| Ω follows from (11.5.4). As in the previous section, we introduce the cutoff functions   d∂Ω κδ (x) = F , δ

420

11 Schr¨ odinger Operator and M (w21 → w2−1 )



and ξR (x) = 1 − F

|x| R

 ,

where F ∈ C ∞ (R+ ) such that F (t) = 1 for t ≤ 1 and F (t) = 0 for t ≥ 2. The proofs of the following two lemmas are similar to those of Lemma 11.3.1 and Lemma 11.3.2. Lemma 11.5.1. If f ∈ w2−1 (Ω), then lim κδ f ; Ωw−1 = 0

(11.5.13)

lim ξR f ; Ωw−1 = 0.

(11.5.14)

2

δ→0

and R→∞

2

˚ (˚ Lemma 11.5.2. If V ∈ M w21 (Ω) → w2−1 (Ω)), then lim κδ V ; ΩM ˚ (˚ w1 →w−1 ) = 0,

(11.5.15)

lim ξR V ; ΩM ˚ (˚ w1 →w−1 ) = 0.

(11.5.16)

δ→0

2

2

and R→∞

2

2

We now complete the proof of the “only if” part of Theorem 11.5.2. Write V in the form V = κδ V + ξR V + (1 − κδ − ξR ) V. By Theorem 11.5.1 (ii), there exist Γδ and Γ (0) such that κδ V = div Γδ + d−1 ∂Ω Γδ , (0)

where



(i)

Γδ ; ΩM (˚ w21 →L2 ) ≤ C κδ V ; ΩM (˚ w1 →w−1 ) . 2

2

0≤i≤n

Analogously, ξR V = div Γ(R) + |x|−1 Γ(R) , (0)

where



(i)

Γ(R) ; ΩM (˚ w21 →L2 ) ≤ C ξR V ; ΩM (˚ w1 →w−1 ) . 2

0≤i≤n

Hence, by Lemma 11.5.2, 

lim

δ→0

0≤i≤n

and lim

(i)

Γδ ; ΩM (˚ w21 →L2 ) = 0,

R→∞

 0≤i≤n

(i)

Γ(R) ; ΩM (˚ w21 →L2 ) = 0.

2

11.6 Second-Order Differential Operators Acting from w21 to w2−1

421

Now we estimate the multiplier Vδ,R := (1 − κδ − ξR ) V. Note that

˚ (˚ Vδ,R ∈ M w21 (Ω) → w2−1 (Ω)).

Since its support is separated from ∞ and from ∂Ω, it follows that ˚ (W 1 (Rn ) → W −1 (Rn )). Vδ,R ∈ M 2 2 By Theorem 11.4.2, Vδ,R = div Γδ,R + Ψδ,R ,

(11.5.17)

˚ (W 1 (Rn ) → where all components of Γδ,R , together with Ψδ,R , are in M 2 n L2 (R )). Multiplying, if necessary, both sides of (11.5.17) by a cutoff function as before, we may assume that the supports of |Γδ,R | and Ψδ,R are in Ω, and are both separated from ∞, and from ∂Ω. Hence, the components of Γδ,R , as well ˚ (˚ w21 (Ω) → L2 (Ω)). Finally, as d∂Ω Ψδ,R , are in M (0) , V = div Γ + d−1 ∂Ω Γ

where Γ = Γδ + Γ(R) + Γδ,R , and

(0)

Γ (0) = Γδ

+ |x|−1 d∂Ω Γ(R) + d∂Ω Γδ,R . (0)

(0)

It remains to note that Γδ ,

Γ(R) ,

(0)

Γδ ,

and |x|−1 d∂Ω Γ(R) (0)

(0)

are small in the corresponding operator norms, while Γδ,R and Γδ,R are compact. This completes the proof of Theorem 11.5.2. 

11.6 Second-Order Differential Operators Acting from w21 to w2−1 Further development of the topic of the present section can be found in [MV4]. Here we survey some results of this article, where explicit necessary and sufficient conditions for the boundedness of the general second-order differential operator n n   aij ∂i ∂j + bj ∂ j + c L= i, j=1

j=1

422

11 Schr¨ odinger Operator and M (w21 → w2−1 )

with real- or complex-valued distributional coefficients aij , bj , and c, acting from the Sobolev space w21 to its dual w2−1 , are found. For the sake of convenience, let us assume that the principal part of L is in the divergence form, i.e., u ∈ C0∞ ,

L u = div (A ∇u) + b · ∇u + q u,

(11.6.1)

where A = (aij )ni, j=1 ∈ (C0∞ ) ,

b = (bj )nj=1 ∈ (C0∞ ) ,

and q ∈ (C0∞ ) .

We present necessary and sufficient conditions on A, b, and q which guarantee the boundedness of the sesquilinear form associated with L: |L u, v| ≤ C uw21 vw21 ,

(11.6.2)

where the constant C does not depend on u, v ∈ C0∞ . Equivalently, we characterize all A, b, and q such that L : w21 → w2−1

(11.6.3)

is a bounded operator. We state this boundedness criterion. For A = (aij ), let At = (aji ) denote the transposed matrix, and let Div : (C0∞ ) → (C0∞ ) be the row divergence operator defined by Div(aij ) =

n


n ∂j aij

j=1

i=1

.

(11.6.4)

We do not differ in notations between spaces of scalar, vector-, and matrixvalued functions. Theorem 11.6.1. [MV4] Let L = div (A ∇·) + b · ∇ + q, where A ∈ (C0∞ ) , b ∈ (C0∞ ) and q ∈ (C0∞ ) , n ≥ 2. Then the following statements hold. (i) The sesquilinear form of L is bounded, i.e., (11.6.2) holds if and only if 1 2

(A + At ) ∈ L∞ ,

and b and q can be represented respectively in the form b = c + Div F,

q = div h,

(11.6.5)

where F is a skew-symmetric matrix field such that F−

1 2

(A − At ) ∈ BMO,

whereas c and h belong to M (w21 → L2 ).

(11.6.6)

11.6 Second-Order Differential Operators Acting from w21 to w2−1

423

(ii) If the sesquilinear form of L is bounded, then c, F , and h in the decomposition (11.6.5) can be determined explicitly by c = ∇(∆−1 div b), F = ∆−1 curl [b − where and

∆−1 curl [b −

1 2

h = ∇(∆−1 q), 1 2

Div (A − At )] + 12 (A − At ).

Div (A − At )] ∈ BMO

∇(∆−1 div b)

and

(11.6.7) (11.6.8) (11.6.9)

∇(∆−1 q)

belong to M (w21 → L2 ). Remark 11.6.1. Condition (11.6.9) in statement (ii) of Theorem 11.6.1 may be replaced by (11.6.10) b − Div 12 (A − At ) ∈ BMO−1 , which ensures that the decomposition (11.6.5) holds. Here BMO−1 stands for the well-known space of distributions that can be represented in the form f = div g where g ∈ BMO. Remark 11.6.2. In the case n = 2, it is shown in [MV4] that (11.6.2) holds if and only if 1 2

b−

1 2

(A + At ) ∈ L∞ (R2 ),

Div (A − At ) ∈ BMO−1 (R2 ),

and q = div b = 0. Remark 11.6.3. Expressions like ∇(∆−1 div b),

Div(∆−1 curl b),

and ∇(∆−1 q)

used above, which involve nonlocal operators, are defined in the sense of distributions. This is possible since ∆−1 div b,

∆−1 curl b,

and ∆−1 q

can be understood in terms of convergence in the weak-∗ topology of BMO of, respectively, ∆−1 div (ψN b),

∆−1 curl (ψN b),

and

∆−1 (ψN q)

as N → +∞. Here ψN is a smooth cutoff function supported on {x : |x| < N }, and the limits above do not depend on the choice of ψN .

11 Schr¨ odinger Operator and M (w21 → w2−1 )

424

It follows from Theorem 11.6.1 that L : w21 → w2−1 is bounded if and only if the symmetric part of A is essentially bounded, i.e., 1 (A + At ) ∈ L∞ 2 and

b1 · ∇ + q : w21 → w2−1

is bounded, where b1 = b −

1 2

Div(A − At ).

(11.6.11)

In particular, the principal part Pu = div(A ∇u) : w21 → w2−1 is bounded if and only if 1 2 (A

and

+ At ) ∈ L∞

Div 12 (A − At ) ∈ BMO−1 .

(11.6.12) (11.6.13)

A simpler condition with 1 2 (A

− At ) ∈ BMO

in place of (11.6.13) is sufficient, but generally not necessary, unless n ≤ 2. Thus, the boundedness problem for the general second order differential operator in the divergence form (11.6.1) is reduced to the special case L = b · ∇ + q,

b ∈ (C0∞ ) ,

q ∈ (C0∞ ) .

(11.6.14)

As a corollary of Theorem 11.6.1, one obtains that, if b · ∇ + q : w21 → w2−1

(11.6.15)

is bounded, then the Hodge decomposition b = ∇(∆−1 div b) + Div (∆−1 curl b) holds, where and  |x−y| 0, x ∈ Rn , in the case n ≥ 3; in two dimensions div b = q = 0. The condition (11.6.17) is generally stronger than ∆−1 div b ∈ BMO

and

∆−1 q ∈ BMO,

11.6 Second-Order Differential Operators Acting from w21 to w2−1

425

while the divergence-free part of b is characterized by ∆−1 curl b ∈ BMO, for all n ≥ 2. A close sufficient condition of the Fefferman–Phong type can be stated in the following form:  [ |∇(∆−1 div b)|2 + |∇(∆−1 q)|2 ]1+ dy ≤ const rn−2(1+) , (11.6.18) |x−y| 0 and all r > 0, x ∈ Rn . This is a consequence of Theorem 11.6.1 coupled with (1.2.50), where |(∆−1 div b)|2 + |∇(∆−1 q)|2 is used in place of g, p = 2, m = 1, and t = 1 + . It is worth mentioning that the class of potentials obeying (11.6.18) is substantially broader than its subclass  (|b|2 + |q|)1+ dy ≤ const rn−2(1+) . (11.6.19) |x−y| 0. By definition this means that Q is relatively form bounded with respect to |D|. In particular, if Q is real-valued, and 0 < a < 1 in the preceding inequality, then by the so-called KLMN Theorem ([RS1], Theorem X.17), H = |D| + Q is defined as a unique self-adjoint operator such that Hu, v = |D| u, v + Q u, v,

∀u ∈ C0∞ .

For the complex-valued Q such that (12.1.3) holds with 0 < a < 1/2, it follows that H = |D| + Q, understood in a similar sense, is an m-sectorial operator ([EE], Theorem IV.4.2). In the case of Q ∈ L1, loc , (12.1.3) is equivalent to the inequality  |u(x)|2 Q(x) dx ≤ const u2 1/2 , ∀u ∈ C0∞ , (12.1.4) W 2

and hence to the boundedness of the corresponding sesquilinear form:  u(x) v(x) Q(x) dx ≤ const u 1/2 v 1/2 , W2 W2 where the constant is independent of u, v ∈ C0∞ .

12.1 Auxiliary Assertions 1/2

429 −1/2

The characterization of potentials Q such that H : W2 → W2 is based on a series of lemmas and propositions presented below, and the results for the nonrelativistic Schr¨ odinger operator obtained in the previous chapter. By L2,unif , we denote the class of f ∈ L2,loc such that f L2,unif = sup χB1 (x) f L2 < ∞.

(12.1.5)

x∈Rn

Lemma 12.1.1. Let 0 < l < 1, and m > l. Then γ ∈ M (W2m → W2l ) if and only if and |D|l γ ∈ M (W2m → L2 ). γ ∈ W2m−l → L2 Moreover, + + γM (W2m →W2l ) ∼ +|D|l γ +M (W m →L ) + γM (W m−l →L2 ) . 2

2

2

(12.1.6)

Proof. We first prove the lower estimate for γM (W2m →W2l ) + l + +|D| γ + + γM (W m−l →L2 ) ≤ c γM (W2m →W2l ) . M (W m →L ) 2

2

2

(12.1.7)

Here and below c denotes a constant which depends only on l, m, and n. Let u ∈ C0∞ . Using the integral representation  u(x) − u(y) |D|l u(x) = c(n, l) dy (12.1.8) |x − y|n+l which follows by inspecting the Fourier transforms of both sides, we obtain

Hence,

|D|l (γ u)(x) − γ(x) |D|l u(x) − u(x) |D|l γ(x)  (u(x) − u(y))(γ(x) − γ(y)) = −c(n, l) dy. |x − y|n+l |D|l (γ u) − γ |D|l u − u |D|l γ ≤ c D2,l/2 u · D2,l/2 γ,

(12.1.9)

where, as in Chap. 4,  D2,s u(x) =

|u(x) − u(y)|2 dy |x − y|n+2s

 12 ,

s > 0.

Next, we estimate + + + + u · |D|l γL2 ≤ + |D|l (γ u)+L + +γ |D|l u+L + c D2,l/2 u · D2,l/2 γL2 2

≤ γuW2l + γM (W m−l →L2 ) 2

2

+ + + |D|l u+

W2m−l

+ c D2,l/2 u · D2,l/2 γL2

≤ γM (W2m →W2l ) uW2m + γM (W m−l →L2 ) uW2m + c D2,l/2 u · D2,l/2 γL2 2

≤ c γM (W2m →W2l ) uW2m + c D2,l/2 u · D2,l/2 γL2 .

(12.1.10)

430

−1/2

1/2

→ W2

12 Relativistic Schr¨ odinger Operator and M (W2

)

In the last line we have used the inequality γM (W m−l →L2 ) ≤ c γM (W2m →W2l )

(12.1.11)

2

(see (4.3.29)). To estimate the term D2,l/2 u · D2,l/2 γL2 , we apply the pointwise estimate (4.2.10) D2,l/2 u ≤ Js D2,l/2 ((−∆ + 1)s/2 u), with s = m − l/2, where Js = (−∆ + 1)−s/2 is the Bessel potential of order s. Hence D2,l/2 u · D2,l/2 γL2 ≤ Jm−l/2 D2,l/2 ((−∆ + 1)m/2−l/4 u)) · D2,l/2 γL2 ≤ c D2,l/2 γM (W m−l/2 →L2 ) Jm−l/2 D2,l/2 ((−∆ + 1)m/2−l/4 u))W m−l/2 2

2

≤ c D2,l/2 γM (W m−l/2 →L2 ) D2,l/2 (−∆ + 1)

m/2−l/4

2

uL2

≤ c D2,l/2 γM (W m−l/2 →L2 ) uW2m .

(12.1.12)

2

We next notice that, by Corollary 4.3.6, D2,l/2 γM (W m−l/2 →L2 ) ≤ c γM (W2m →W2l ) .

(12.1.13)

2

Combining the estimates (12.1.10)–(12.1.13), we obtain + + +u · |D|l γ + ≤ c γM (W m →W l ) uW m , 2 L 2 2 2

which is equivalent to the inequality + l + +|D| γ + ≤ c γM (W2m →W2l ) . M (W m →L ) 2

2

This, together with (12.1.11), completes the proof of (12.1.7). We now prove the upper estimate 
+ + γM (W2m →W2l ) ≤ c +|D|l γ +M (W m →L ) + γM (W m−l →L2 ) . 2

2

2

(12.1.14)

By (12.1.9), + l + + + + + +|D| (γu)+ ≤ +γ|D|l u+ + +|D|l γ · u+ + c D2,l/2 u · D2,l/2 γL . 2 L2 L2 L2 Using the elementary estimate uW m−l ≤ c uW2m , 2

12.1 Auxiliary Assertions

431

we have γuL2 ≤ γM (W m−l →L2 ) uW m−l ≤ c γM (W m−l →L2 ) uW2m . 2

2

2

From these inequalities, combined with the estimate D2,l/2 u · D2,l/2 γL2 ≤ c γM (W m−l/2 →W l/2 ) uW2m 2

2

established above, it follows that + +  γuW2l ≤ c γM (W m−l →L2 ) uW2m + +|D|l γ +M (W m →L 2

2)

2

uW2m



+c γM (W m−l/2 →W l/2 ) uW2m . 2

2

Thus, + + γM (W2m →W2l ) ≤ c +|D|l γ +M (W m →L2 ) + γM (W m−l →L2 ) 2

2

 1/2 1/2 +γM (W m−l →L ) γM (W m →W l ) . 2

2

2

2

Combining this with (12.1.11), we find that 
+ + γM (W2m →W2l ) ≤ c +|D|l γ +M (W m →L ) + γM (W m−l →L2 ) . 2

2

2

This completes the proof of Lemma 12.1.1.



Lemma 12.1.2. Let 0 < l < 1 and n2 ≥ m > l. Then γ ∈ M (W2m → W2l ) if and only if (−∆ + 1)l/2 γ ∈ M (W2m → L2 ) and γM (W2m →W2l ) ∼ (−∆ + 1)l/2 γM (W2m →L2 ) .

(12.1.15)

Proof. Recall that a nonnegative weight w ∈ L1,loc is said to be in the Muckenhoupt class A1 if Mw(x) ≤ c w(x)

for almost all x ∈ Rn ,

where M is the Hardy–Littlewood maximal operator. The least constant on the right-hand side of the preceding inequality is called the A1 -bound of w. We need the following statement established in [MV1], Lemma 3.1 (see also [MSh16], Sec. 2.6.3) for the homogeneous Sobolev space wpm (Rn ).

432

1/2

12 Relativistic Schr¨ odinger Operator and M (W2

−1/2

→ W2

)

Lemma 12.1.3. Let γ ∈ M (wpm → Lp ), where 1 < p < ∞ and 0 < m < np . Suppose that T is a bounded operator on the weighted space Lp (w) for every w ∈ A1 . Suppose additionally that, for all f ∈ Lp (w), the inequality T f Lp (w) ≤ C f Lp (w) holds with a constant C which depends only on the A1 -bound of the weight w. Then T γ ∈ M (wpm → Lp ), and T γM (wpm →Lp ) ≤ C1 γM (wpm →Lp ) , where the constant C1 does not depend on γ. We also need a Fourier multiplier theorem of Mikhlin type for Lp spaces with weights. Let m ∈ L∞ . Then the Fourier multiplier operator with symbol m is defined on L2 by Tm = F −1 m F, where F and F −1 are respectively the direct and inverse Fourier transforms. The next lemma follows from the results of Kurtz and Wheeden [KWh], Theorem 1. Lemma 12.1.4. Suppose that 1 < p < ∞ and w ∈ A1 . Suppose also that m ∈ C ∞ (Rn \ {0}) satisfies the Mikhlin multiplier condition: |Dα m(x)| ≤ Cα |x|−|α| ,

x ∈ Rn \ {0},

(12.1.16)

for every multi-index α such that 0 ≤ |α| ≤ n. Then the inequality Tm f Lp (w) ≤ C f Lp (w) ,

f ∈ Lp (w) ∩ L2 ,

holds, where C depends only on p, n, the A1 -bound of w, and the constant Cα in (12.1.16). Corollary 12.1.1. Suppose that 1 < p < ∞ and w ∈ A1 . Suppose also that 0 < l ≤ 2. Define (12.1.17) ml (x) = (1 + |x|2 )l/2 − |x|l . Then Tml f Lp (w) ≤ C f Lp (w) ,

f ∈ Lp (w) ∩ L2 ,

(12.1.18)

where the constant C depends only on l, p, n, and the A1 -constant of w. Proof. Clearly, 0 ≤ ml (x) ≤ C (1 + |x|)l−2 ,

x ∈ Rn .

Furthermore, it is easy to see by induction that, for any multi-index α with |α| ≥ 1, we have the following estimates:

12.1 Auxiliary Assertions

|Dα ml (x)| ≤ Cα,l |x|l−2−|α| ,

433

|x| → ∞,

and |Dα ml (x)| ≤ Cα,l |x|l−|α| ,

|x| → 0.

Since 0 < l ≤ 2, it follows from this that ml satisfies (12.1.16), and hence by Lemma 12.1.4 the inequality Tml f Lp (w) ≤ C f Lp (w) holds with a constant that depends only on l, p, and the A1 -bound of w.



Now we are in a position to complete the proof of Lemma 12.1.2. Suppose that γ ∈ M (W2m → W2l ), where n2 ≥ m > l and 0 < l < 1. By Corollary 12.1.1, the operator Tml = (1 − ∆)l/2 − |D|l is bounded on L2 (w) for every w ∈ A1 , and its norm is bounded by a constant which depends only on l, n, and the A1 -bound of w. Hence by Lemma 12.1.3 it follows that γ ∈ M (w2m → L2 ) yields 
Tml γ = (1 − ∆)l/2 − |D|l γ ∈ M (w2m → L2 ), and Tml γM (w2m →L2 ) ≤ c γM (w2m →L2 ) , where c depends only on l, m, and n. We need to replace w2m in the preceding inequality by W2m . To this end, let B = B1 (x0 ) denote a ball of radius 1 in Rn , and 2B = B2 (x0 ). Suppose that m < n2 (the case m = n2 requires usual modifications). Using Theorem 3.1.2, we obtain that γ ∈ M (W2m → L2 ) if and only if sup χB γM (w2m →L2 ) < +∞, B

and γM (W2m →L2 ) ∼ sup χB γM (w2m →L2 ) . B

Hence, Tml γM (W2m →L2 ) ≤ c sup χB Tml γM (w2m →L2 ) . B

We set γ = χ2B γ + χ(2B)c γ and estimate each term separately. By Lemma 12.1.3, χB Tml (χ2B γ)M (w2m →L2 ) ≤ c sup χ2B γM (w2m →L2 ) ≤ c γM (W2m →L2 ) . B

To estimate the second term, notice that Tml (χ(2B)c γ) ∈ L∞ (B), and hence χB Tml (χ(2B)c γ)M (w2m →L2 ) ≤ c Tml (χ(2B)c γ)(x); BL∞ ≤ c γM (W2m →L2 ) .

1/2

434

12 Relativistic Schr¨ odinger Operator and M (W2

−1/2

→ W2

Indeed, for x ∈ B,

)

 

|Tml (χ(2B)c γ)(x)| ≤ c

|x−y|≥1

|γ(y)| dy ≤ c |x − y|n+l



+∞

1

|γ(y)|dy

Br (x) rn+l+1

dr.

Since γ ∈ M (W2m → L2 ), it follows that γ ∈ L2,unif , and hence  |γ(y)|2 dy ≤ c rn γ2M (W2m →L2 ) , r ≥ 1. Br (x)

Consequently,  |γ(y)| dy ≤ c rn/2 γ; Br (x)L2 ≤ c rn γM (W2m →L2 ) , Br (x)

r ≥ 1.

Hence, Tml (χ(2B)c γ); BL∞ ≤ c γM (W2m →L2 ) . Thus, we have proved the inequality +
 + + + + (1 − ∆)l/2 − |D|l γ +

M (W2m →L2 )

≤ c γM (W2m →L2 ) .

Using this estimate, inequality (12.1.11), and Lemma 12.1.1, we arrive at +  + (1 − ∆)l/2 γM (W2m →L2 ) ≤ c +|D|l γ +M (W m →L + γM (W2m →L2 ) 2

)

≤ c γM (W2m →W2l ) . Conversely, suppose that (1 − ∆)l/2 γ ∈ M (W2m → L2 ). It follows from the above estimate of +
 + + + + (1 − ∆)l/2 − |D|l γ + that + l + +|D| γ + M (W m →L 2

2)

M (W2m →L2 )


≤ c (1 − ∆)l/2 γM (W2m →L2 ) + γM (W2m →L2 ) .

Obviously, γM (W2m →L2 ) ≤ c γM (W m−l →L2 ) . 2

Applying again Lemma 12.1.1 together with the preceding estimates, we have 
+ + γM (W2m →W2l ) ≤ c + |D|l γ +M (W m →L ) + γM (W m−l →L2 ) 2

2

2

12.1 Auxiliary Assertions




435



≤ c (1 − ∆)l/2 γM (W2m →L2 ) + γM (W m−l →L2 ) . 2

It remains to obtain the estimate γM (W m−l →L2 ) ≤ c (1 − ∆)l/2 γM (W2m →L2 ) . 2

Since (1 − ∆)l/2 γ ∈ M (W2m → L2 ), it follows that  |(1 − ∆)l/2 γ|2 dx ≤ (1 − ∆)l/2 γ2M (W2m →L2 ) C2,m (e), e

for every compact set e ⊂ Rn . Hence, for every ball Br (a),  |(1 − ∆)l/2 γ|2 dx ≤ c (1 − ∆)l/2 γ2M (W2m →L2 ) rn−2m , Br (a)

0 < r ≤ 1,

and in particular (1 − ∆)l/2 γL2 ,unif ≤ c (1 − ∆)l/2 γM (W2m →L2 ) . In view of the properties (1.2.4), (1.2.5) of the Bessel function Gl , it is easy to derive the pointwise estimate  |γ(x)| ≤ Gl (x − t) |(1 − ∆)l/2 γ(t)| dt  ≤c

|z|≤1

|(1 − ∆)l/2 γ(x + z)| dz + (1 − ∆)l/2 γL2 ,unif |z|n−l

.

Using Lemma 2.3.7 together with the preceding pointwise estimate, we deduce that  l  |(1 − ∆)l/2 γ|2 dy 2m l B (a) r |γ(x)| ≤ c (M (1 − ∆)l/2 γ(x))1− m sup rn−2m 0 0,

for the Bessel potential of order s on Rn+1 ; here ∆n+1 denotes the Laplacian on Rn+1 . Now by Theorem 11.4.1 we obtain that γ ⊗ δ ∈ M (W21 (Rn+1 ) → W2−1 (Rn+1 )) if and only if (n+1)

J1

(γ ⊗ δ) ∈ M (W21 (Rn+1 ) → L2 (Rn+1 )),

and (n+1)

J1

(γ ⊗ δ); Rn+1 M (W21 →L2 ) ≤ c γ ⊗ δ; Rn+1 M (W 1 →W −1 ) 2

2

≤ c1 γ; Rn M (W 1/2 →W −1/2 ) . 2

2

Next, pick 0 < < 1/2 and observe that (n+1)

J1

(n+1)

= (−1 + ∆n+1 )1/4+/2 J+3/2 . (n+1)

Using Lemma 12.1.2 with l = 1/2 + , m = 1, and J+3/2 (γ ⊗ δ) in place of γ, we deduce that (n+1)

J1

(n+1)

(γ ⊗ δ); Rn+1 M (W21 →L2 ) ∼ J+3/2 (γ ⊗ δ); Rn+1 M (W 1 →W 1/2+ ) . 2

2

As was proved above, the left-hand side of the preceding relation is bounded by a constant multiple of γ; Rn M (W 1/2 →W −1/2 ) . 2 2 Thus, (n+1)

J+3/2 (γ ⊗ δ); Rn+1 M (W 1 →W 1/2+ ) ≤ c γ; Rn M (W 1/2 →W −1/2 ) . 2

2

2

2

1/2

440

12 Relativistic Schr¨ odinger Operator and M (W2

−1/2

→ W2

)

Passing to the trace on Rn = {xn+1 = 0} in the multiplier norm on the left-hand side, we obtain (n+1)

Trace J+3/2 (γ ⊗ δ); Rn M (W 1/2 →W ) ≤ c γ; Rn M (W 1/2 →W −1/2 ) . 2

2

2

2

We now observe that (n+1)

(n)

Trace J+3/2 (γ ⊗ δ) = const J+1/2 (γ), which follows immediately by inspecting the corresponding Fourier transforms. In other words, (n)

J+1/2 γ; Rn M (W 1/2 →W ) ≤ c γ; Rn M (W 1/2 →W −1/2 ) . 2

2

2

(12.1.21)

2

From this estimate and Lemma 12.1.2 with l = , m = 1/2, and with γ (n) replaced by J+1/2 γ, it follows that (n)

(n)

J1/2 γM (W 1/2 →L2 ) = (−∆ + 1)/2 J+1/2 γM (W 1/2 →L2 ) 2

2

(n)

≤ c J+1/2 γM (W 1/2 →W ) ≤ C γM (W 1/2 →W −1/2 ) . 2

2

2

2

Thus, (n)

1/2

Φ = J1/2 γ ∈ M (W2

→ L2 )

and ΦM (W 1/2 →L2 ) ≤ C γM (W 1/2 →W −1/2 ) . 2

2

2



The proof of Theorem 12.1.1 is complete. The theorem just proved can be reformulated as follows.

Theorem 12.1.2. Let Q ∈ (C0∞ ) , n ≥ 1. The following statements are equivalent: √ (i) The relativistic Schr¨ odinger operator H = −∆ + Q is bounded from 1/2 −1/2 W2 to W2 . (ii) The inequality (12.1.22) |Qu, u| ≤ C u2W 1/2 2

C0∞ .

holds for all u ∈ (iii) Φ = (−∆ + 1)−1/4 Q ∈ L2, loc , and the inequality  |u(x)|2 |Φ(x)|2 dx ≤ C u2W 1/2 2

holds for all u ∈ C0∞ .

(12.1.23)

12.2 Corollaries of the Form Boundedness Criterion and Related Results

441

12.2 Corollaries of the Form Boundedness Criterion and Related Results Theorem 12.1.1 combined with the criteria for the trace inequality with a nonnegative measure (see Theorem 3.1.4 and Remark 3.1.3) implies √ Theorem 12.2.1. Let Q ∈ (C0∞ ) , n ≥ 1, and let H = −∆ + Q. Then 1/2 −1/2 H : W2 → W2 is bounded if and only if Φ = (−∆ + 1)−1/4 Q ∈ L2, loc and any one of the following equivalent conditions holds: (i) For every compact set e ⊂ Rn ,  |Φ(x)|2 dx ≤ C C2,1/2 (e),

(12.2.1)

e

where the constant does not depend on e. (ii) The function J1/2 |Φ|2 is finite a.e., and  2 J1/2 J1/2 |Φ|2 (x) ≤ C J1/2 |Φ|2 (x)

a.e.

Here J1/2 = (−∆ + 1)−1/4 is the Bessel potential of order 1/2. (iii) For every dyadic cube P0 in Rn of side length !(P0 ) ≤ 1, ⎡  ⎤2 2  |Φ(x)| dx  ⎢ ⎥ ⎢ P ⎥ mesn P ≤ C |Φ(x)|2 dx, ⎣ (mesn P )1−1/2n ⎦ P 0 P ⊆P

(12.2.2)

(12.2.3)

0

where the sum is taken over all dyadic cubes P contained in P0 , and the constant does not depend on P0 . Some simpler either necessary or sufficient conditions which do not involve capacities are discussed in this section. The following necessary condition is immediate from (12.2.1) and the known estimates of the capacity of the ball in Rn (see Sect. 1.2.2). Corollary 12.2.1. Suppose that Q ∈ (C0∞ ) , n ≥ 1. Suppose also that H = √ 1/2 −1/2 −∆ + Q : W2 → W2 is a bounded operator. Then, for every ball Br (a) n in R ,  |Φ(x)|2 dx ≤ c rn−1 , 0 < r ≤ 1, n ≥ 2, (12.2.4) Br (a)

and

 Br (a)

|Φ(x)|2 dx ≤

c , log 2r

0 < r ≤ 1,

n = 1,

where the constant c does not depend on a ∈ Rn and r.

(12.2.5)

1/2

442

12 Relativistic Schr¨ odinger Operator and M (W2

−1/2

→ W2

)

We notice that the class of distributions Q such that Φ = (−∆ + 1)−1/4 Q satisfies (12.2.4) can be regarded as a Morrey space of order −1/2. Combining Theorem 12.2.1 with the Fefferman-Phong condition (see Sect. 1.2.6) applied to |Φ|2 , we arrive at sufficient conditions involving Morrey spaces of negative order. (Strictly speaking, the Fefferman-Phong condition [F2] was originally established for estimates in the homogeneous Sobolev space w21 of order m = 1. However, it can be carried over to Sobolev spaces W2m for all 0 < m ≤ n/2. See, e.g., [KeS] or [MV1], p. 98.) Corollary 12.2.2. Suppose that Q ∈ (C0∞ ) , n ≥ 2. Suppose also that Φ = 1/2 (−∆ + 1)−1/4 Q, and t > 1. Then H is a bounded operator from W2 to −1/2 if W2  Br (a)

|Φ(x)|2t dx ≤ C rn−t ,

0 < r ≤ 1,

(12.2.6)

where the constant does not depend on a ∈ Rn and r. Remark 12.2.1. It is worth mentioning that the condition (12.2.6) defines a class of potentials which is strictly broader than the (relativistic) FeffermanPhong class of Q such that  |Q(x)|t dx ≤ const rn−t , 0 < r ≤ 1, n ≥ 2, (12.2.7) Br (a)

for some t > 1. This follows from the observation that if one replaces Q by |Q| in (12.2.6), then obviously the resulting class defined by  (J1/2 |Q|)2t dx ≤ const rn−t , 0 < r ≤ 1, n ≥ 2, (12.2.8) Br (a)

becomes smaller, but still contains some singular measures, together with all functions in the Fefferman-Phong class (12.2.7). (This was noticed earlier in [MV1], Proposition 3.5.) A smaller but more conventional class of admissible potentials appears when one replaces C2,1/2 (e) on the right-hand side of (12.2.1) by its lower bound involving the Lebesgue measure of e ⊂ Rn , as shown by the following result. Corollary 12.2.3. Suppose that Q ∈ (C0∞ ) , n ≥ 1. Suppose also that Φ = √ 1/2 (−∆ + 1)−1/4 Q. Then H = −∆ + Q is a bounded operator from W2 to −1/2 if, for every measurable set e ⊂ Rn , W2  |Φ(x)|2 dx ≤ c (mesn e)(n−1)/n , diam (e) ≤ 1, n ≥ 2, (12.2.9) e

12.2 Corollaries of the Form Boundedness Criterion and Related Results

or

 |Φ(x)|2 dx ≤ e

c log

2 mesn e

diam (e) ≤ 1,

,

n = 1,

443

(12.2.10)

where the constant c does not depend on e. We remark that (12.2.9), without the extra assumption diam (e) ≤ 1, is equivalent to Φ ∈ L2n, ∞ , where Lp, ∞ is the Lorentz (weak Lp ) space of functions f such that |{x ∈ Rn : |f (x)| > t}| ≤

C , tp

t > 0. −1/2

In particular, (12.2.9) holds if Φ ∈ L2n or, equivalently, Q ∈ W2n . Furthermore, if Φ ∈ L∞ , then obviously (12.2.9) holds as well, since C2,1/2 (e) ≥ C mesn e, if diam (e) ≤ 1. This leads to the sufficient condition Φ ∈ L2n + L∞ , n ≥ 2. It is worth noting that (12.2.9) defines a substantially broader class of admissible potentials than the standard (in the relativistic case) class Q ∈ Ln + L∞ , n ≥ 2 ([LL], Sec. 11.3). This is a consequence of the imbedding −1/2

Ln ⊂ W2n

,

n ≥ 2, 1/2

which follows from the classical Sobolev imbedding Wp ⊂ Lr for p = 2n/(2n − 1) and r = n/(n − 1), n ≥ 2. Indeed, by duality, the latter is equivalent to −1/2 Ln = (Lr ) ⊂ (Wp1/2 ) = W2n . Similarly, in the one-dimensional case, the class of potentials defined by (12.2.10) is wider than the standard class L1+ (R1 ) + L∞ (R1 ), > 0. It is easy to see that actually Q ∈ Ln (Rn ) + L∞ (Rn ) if n ≥ 2, or Q ∈ L1+ (R1 ) + L∞ (R1 ) if n = 1, is sufficient for the inequality  |u(x)|2 |Q(x)| dx ≤ C u2W 1/2 , u ∈ C0∞ , 2

which is a “na¨ıve” version of (12.1.4), where Q is replaced by |Q|. We conclude this chapter with mentioning the article by Frank and Seiringer [FrS] which is mostly devoted to sharp Hardy type inequalities involving Besov type seminorms and their generalizations. In particular, the authors found the optimal value of the constant Cn,s,p in the inequality    |u(x) − u(y)|p |u(x)|p dx dy ≥ C dx, (12.2.11) n,s,p n+ps ps Rn Rn |x − y| Rn |x|

444

1/2

12 Relativistic Schr¨ odinger Operator and M (W2

−1/2

→ W2

)

where 0 < s < 1 and 1 ≤ p < n/s. Moreover, they extend this result to functionals on the left-hand side of (12.2.11) with |x − y|−n−ps replaced by an arbitrary symmetric and nonnegative, but not necessarily translation invariant, kernel k(x, y):   E[u] := |u(x) − u(y)|p k(x, y) dx dy. Rn Rn

In [FrS], a sufficient condition for the following general version of Hardy’s inequality  Q(x) |u(x)|p dx ≤ E[u], u ∈ C0∞ (Rn ), Rn

is found. The function Q is assumed to be of the form    1−p Q(x) = 2 ω(x) ω(x) − ω(y) |ω(x) − ω(y)|p−2 k(x, y) dy

(12.2.12)

Rn

with a certain positive function ω. The integral on the right-hand side of (12.2.12) might be divergent and some regularization of the principal value type is needed for its definition. In particular, the representation 
dy ω(x)  (12.2.13) Q(x) = 2 1− ω(y) |x − y|n+1 Rn with ω > 0 proves to be sufficient for the inequality    |u(x) − u(y)|2 2 Q(x)|u(x)| dx ≤ dx dy, u ∈ C0∞ (Rn ). (12.2.14) |x − y|n+1 Rn Rn Rn 1/2

−1/2

Note that (12.2.14) does not imply that Q ∈ M (w2 → w2 ) if Q is not nonnegative. However, for Q ≥ 0 the representation (12.2.13) with a positive 1/2 −1/2 factor in place of 2 is sufficient for the inclusion Q ∈ M (w2 → w2 ) (and 1/2 −1/2 )). It is of interest to investigate the question hence for Q ∈ M (W2 → W2 of necessity.

13 Multipliers as Solutions to Elliptic Equations

In Sects. 13.1–13.3 of this chapter, solutions of second-order linear and quasilinear elliptic differential equations and systems are considered as multipliers in certain spaces of differentiable functions in a domain Ω. On one hand, this can be of interest for the theory of functions, since it leads to new characterizations of multipliers and, on the other hand, for the theory of partial differential equations, since it allows us to obtain a priori information about the solutions in spaces different from the usual ones. In Sect. 13.4 we obtain coercive estimates in multiplier spaces for solutions of linear elliptic systems in a half-space. The last Sect. 13.5 is devoted to regularity of solutions to higher order semilinear elliptic equations.

13.1 The Dirichlet Problem for the Linear Second-Order Elliptic Equation in the Space of Multipliers Let us start with a multiplier analogue of the classical unique solvability of a linear second-order equation in the variational sense. By Ω we denote a bounded domain in Rn with ∂Ω ∈ C 0,1 . Let Lu = −

n  ∂
∂u  aij (x) . ∂xi ∂xj i,j=1

(13.1.1)

Suppose that the coefficients aij are real measurable bounded functions on Ω and that the matrix aij  is symmetric and uniformly positive definite. We consider the Dirichlet problem Lu = 0 in Ω,

˚21 (Ω), u−g ∈W

where g ∈ W21 (Ω). This problem is uniquely solvable.

V.G. Maz’ya, T.O. Shaposhnikova, Theory of Sobolev Multipliers, Grundlehren der mathematischen Wissenschaften 337, c Springer-Verlag Berlin Hiedelberg 2009


445

446

13 Multipliers as Solutions to Elliptic Equations

Theorem 13.1.1. If g ∈ M W21 (Ω), then u ∈ M W21 (Ω). Moreover, ˚21 (Ω)) u − g ∈ M (W21 (Ω) → W and u; ΩM W21 ≤ c g; ΩM W21 .

(13.1.2)

Proof. By the maximum principle for variational solutions of the equation Lu = 0 we have u; ΩL∞ ≤ g; ΩL∞ .

(13.1.3)

˚ 1 (Ω) ∩ L∞ (Ω). The definition of Hence the function γ = u − g belongs to W 2 a variational solution yields   ∂ϕ ∂γ ∂ϕ ∂g aij dx = − aij dx (13.1.4) ∂x ∂x ∂x i j i ∂xj Ω Ω ˚ 1 (Ω). Let v be an arbitrary function in W 1 (Ω) ∩ L∞ (Ω). It for any ϕ ∈ W 2 2 ˚ 1 (Ω). We set ϕ = γv 2 in (13.1.4). is easily seen that γv and γv 2 belong to W 2 Then   ∂(γv) ∂(γv) ∂v ∂v 2 aij dx = aij γ dx ∂xi ∂xj ∂xi ∂xj Ω Ω   ∂(γv) ∂g ∂v ∂g aij v dx − γ v aij dx . − ∂xi ∂xj ∂xi ∂xj Ω Ω Consequently, c ∇(γv); Ω2L2 ≤ γ; Ω|2L∞ ∇v; Ω2L2 +∇(γv); ΩL2 v∇g; ΩL2 + γ; ΩL∞ ∇v; ΩL2 v∇g; ΩL2 .

(13.1.5)

Clearly, v∇g; ΩL2 ≤ vg; ΩW21 + g; ΩL∞ v; ΩW21 . Since g; ΩL∞ ≤ g; ΩM W21 , it follows that v∇g; ΩL2 ≤ 2 g; ΩM W21 v; ΩW21 . This inequality, together with (13.1.3) and (13.1.5), implies that c ∇(γv); Ω2L2 ≤ 8 g; Ω2M W 1 v; Ω2W 1 + 2 ∇(γv); ΩL2 g; ΩM W21 v; ΩW21 . 2

2

Hence γv; ΩW21 ≤ c g; ΩM W21 v; ΩW21

13.2 Bounded Solutions of Linear Elliptic Equations as Multipliers

447

or, which is the same, uv; ΩW21 ≤ c g; ΩM W21 v; ΩW21 .

(13.1.6)

˚ 1 (Ω) for all Since W21 (Ω) ∩ L∞ (Ω) is dense in W21 (Ω) and γv ∈ W 2 1 1 ˚ v ∈ W2 (Ω)∩L∞ (Ω), we have (13.1.6) and γv ∈ W2 (Ω) for any v ∈ W21 (Ω).  Remark 13.1.1. Let Ω be a bounded domain in Rn with ∂Ω ∈ C 0,1 . By 1/2 Theorem 8.7.2, M W2 (Rn−1 ) is the space of traces on Rn−1 of functions 1/2 in M W21 (Rn+ ). This clearly implies that ϕ ∈ M W2 (∂Ω) has an extension 1 g ∈ M W2 (Ω) for which g; ΩM W21 ∼ ϕ; ∂ΩM W 1/2 . 2

This, together with Theorem 13.1.1, proves the unique solvability of the Dirichlet problem Lu = 0 in Ω ,

1/2

u|∂Ω = ϕ ∈ M W2 (∂Ω)

in M W21 (Ω). Remark 13.1.2. Let Ω be a bounded domain in Rn with ∂Ω ∈ C 0,1 . In ˚ l (Ω)). Let us show that Theorem 13.1.1 we used the space M (Wpm (Ω) → W p ˚pl (Ω)) = W ˚pl (Ω) ∩ M (Wpm (Ω) → Wpl (Ω)) . M (Wpm (Ω) → W We denote the left-hand side of this equality by A and the right-hand side by B. Since 1 ∈ Wpm (Ω), it follows that A ⊂ B. Let u ∈ Wpm (Ω), γ ∈ B and let {uν }ν≥1 be a sequence of functions in C ∞ (Ω) such that uν → u in ˚ l (Ω) and Wpm (Ω). Then γuν ∈ W p γu − γuν ; ΩWpl ≤ γ; ΩM (Wpm →Wpl ) u − uν ; ΩWpm = o(1) . ˚ l (Ω), that is, γ ∈ A. Consequently γu ∈ W p

13.2 Bounded Solutions of Linear Elliptic Equations as Multipliers 13.2.1 Introduction In this section we study bounded solutions of a linear elliptic second-order equation without any requirements on their boundary values. It is shown that, under some conditions on the right-hand side of the equation, such solutions are multipliers in certain function spaces.

448

13 Multipliers as Solutions to Elliptic Equations

Let Ω be a domain in Rn with compact closure and sufficiently smooth 1 (Ω), β ∈ R1 we denote the space of functions boundary ∂Ω. By W2,β 1 u ∈ W2,loc (Ω), having the finite norm  1 u; ΩW2,β =

1/2

Ω

ρ(x)β | ∇u|2 dx + u; Ω2L2

,

where ρ(x) = dist (x, ∂Ω) . Theorem 13.2.1 below states that bounded solutions of the above men1 (Ω) for β > 1. For β < 1 tioned equations are multipliers in the space W2,β 1 (Ω) with this fact is not true since the space of traces of functions from W2,β (1−β)/2

β < 1 is W2 (∂Ω), which does not contain all bounded functions on ∂Ω. The case β = 1 is special. It is considered in Theorems 13.2.2-13.2.4, in which solutions from L∞ (Ω) appear as multipliers acting into the space 1 1 (Ω) from some function spaces more narrow than W2,1 (Ω). In Theorems W2,1 1 (Ω) similar 13.2.3 and 13.2.4 we deal with the weighted Hilbert space W2,w(ρ) 1 to W2,β (Ω), where the role of ρβ is played by a weight w(ρ). Here it is shown 1 1 (Ω) → W2,1 (Ω)) if that all bounded solutions belong to the class M (W2,w(ρ) and only if 1/w ∈ L(0, 1). 13.2.2 The Case β > 1 Let L be the uniformly elliptic operator (13.1.1) with sufficiently smooth co¯ and let aij = aji . We consider the equation efficients in Ω Lγ = f + div g,

(13.2.1)

where f is a scalar function while g is a vector-valued function from L2,loc (Ω). 1 By a variational solution of (13.2.1) we mean a function γ ∈ W2,loc (Ω), satisfying   ∂γ ∂η aij dx = (f η − g ∇η)dx, (13.2.2) ∂x j ∂xi Ω Ω where η is an arbitrary function from W21 (Ω) with compact support in Ω. Theorem 13.2.1. Let β > 1 and let the functions f and g satisfy the condition 1 (Ω) → L2 (Ω)). ρβ/2 (|f |1/2 + |g|) ∈ M (W2,β Then any variational solution γ ∈ L∞ (Ω) of equation (13.2.1) belongs to the 1 (Ω), and the estimate space M W2,β  1 ≤ c ρβ/2 |f |1/2 ; Ω2M (W 1 →L2 ) γ; ΩM W2,β 2,β

1 →L ) + γ; ΩL + ρβ/2 g; ΩM (W2,β ∞ 2

holds with a constant c independent of f, g and γ.



(13.2.3)

13.2 Bounded Solutions of Linear Elliptic Equations as Multipliers

449

Proof. Let R denote a solution of the Dirichlet problem LR = 1 in Ω,

R = 0 on ∂Ω.

From Giraud’s theorem on the sign of the normal derivative (see[Mir], Sect. 3.5) and the boundedness of the gradient of R it follows that c1 ρ ≤ R ≤ c2 ρ on Ω. Choosing a sufficiently small δ > 0, we introduce a family of functions {ζτ } from C0∞ (Ω), 0 < τ < δ, such that ζτ = 1 on the set Ωτ = {x ∈ Ω : ρ(x) > τ }, and 0 ≤ ζτ ≤ 1, |Dα ζτ | ≤ c τ −|α| for all multi-indices α. We set η = Rβ ζτ2 u2 γ 1 in (13.2.2), where u is an arbitrary function from W2,β (Ω). Then the left-hand side of (13.2.2) can be written as    ∂γ ∂η ∂(γu) ∂(γu) β 2 ∂u ∂u 2 β 2 aij dx = aij R ζτ dx − aij γ R ζτ dx ∂x ∂x ∂x ∂x ∂x j i j i j ∂xi Ω Ω Ω

 +

aij Ω

∂(γu) ∂(Rβ ζτ2 ) γu dx − ∂xj ∂xi

 aij Ω

∂u ∂(Rβ ζτ2 ) 2 γ u dx. ∂xj ∂xi

(13.2.4)

Since |∇(Rβ ζτ2 )| ≤ c Rβ−1 ζτ , the third term on the right-hand side of (13.2.4) does not exceed  c γ; ΩL∞ ζτ |∇(γu)| |u| ρβ−1 dx Ω

≤ c γ; ΩL∞ ζτ |∇(γu)|ρβ/2 ; ΩL2 u ρ(β−2)/2 ; ΩL2 .

(13.2.5)

By the Hardy inequality, 1 . u ρ(β−2)/2 ; ΩL2 ≤ c u; ΩW2,β

(13.2.6)

We estimate the fourth term on the right-hand side of (13.2.4) using (13.2.6):   ∂u ∂(Rβ ζτ2 ) 2 γ u dx ≤ c γ; Ω2L∞ |∇u| |u| ρβ−1 dx aij ∂xj ∂xi Ω Ω ≤ c γ; Ω2L∞ u; Ω2W 1 . 2,β

450

13 Multipliers as Solutions to Elliptic Equations

Consequently, (13.2.4) implies that  ∂(γu) ∂(γu) β 2 aij R ζτ dx ≤ c γ; Ω2L∞ u; Ω2W 1 2,β ∂xj ∂xi Ω    f Rβ ζτ2 u2 γ − g ∇(Rβ ζτ2 u2 γ) dx . +

(13.2.7)

Ω

 f Rβ ζτ2 u2 γ dx

We have

Ω

+2 + ≤ c γ; ΩL∞ +ρβ/2 |f |1/2 ; Ω +M (W 1

2,β →L2 )

Also,

u; Ω2W 1 .

(13.2.8)

2,β

  |g| ρβ ζτ |u| |∇(γu)| dx g∇(Rβ ζτ2 u2 γ)dx ≤ c Ω Ω  |gu| |∇(Rβ ζτ u)|dx ≤ + c γ; ΩL∞ Ω

β/2 1 →L ) ζτ ρ 1 g; ΩM (W2,β ∇(γu); ΩL2 u; ΩW2,β 2 + + +c γ; ΩL∞ ρβ/2 g u; ΩL2 +ρ−β/2 ∇(Rβ ζτ2 u); Ω +L .

≤ c ρ

β/2

2

1 , it follows that Since by (13.2.6) the last norm does not exceed c u; ΩW2,β   β/2 1 →L ) ζτ ρ ∇(γu); ΩL2 g ∇(Rβ ζτ2 u2 γ)dx ≤ c ρβ/2 g; ΩM (W2,β 2

Ω

 1 1 . u; ΩW2,β + γ; ΩL∞ u; ΩW2,β

Combining this estimate with (13.2.8) and (13.2.7), we find that  ζτ ρβ/2 ∇(γu); ΩL2 ≤ c γ; ΩL∞ + ρβ/2 |f |1/2 ; Ω2M (W 1  1 →L ) u; ΩW 1 . +ρβ/2 g; ΩM (W2,β 2 2,β

2,β →L2 )

Passing to the limit as τ → 0, we complete the proof.



Remark 13.2.1. The same proof shows that for 0 ≤ β < 1 the assertion of Theorem 13.2.1 remains valid for bounded solutions γ ∈ W21 (Ω) of equation (13.2.1) satisfying the Dirichlet condition γ = 0 on ∂Ω. 1 (Ω) for β > 1. By Qj we denote the We next describe the space M W2,β cubes with edge lengths dj forming a Whitney covering of Ω. Let Q∗j be the cube in Ω, concentric to Qj and with edge length 9dj /8. The cubes Q∗j form a covering of Ω of finite multiplicity, depending only on n. We introduce the relative capacity of a compact set e ⊂ Ω,

cap(e, Ω) = inf{∇u; Ω2L2 : u ∈ C0∞ (Ω), u = 1 on e}.

13.2 Bounded Solutions of Linear Elliptic Equations as Multipliers

451

Proposition 13.2.1. For β > 1 1 ∼ sup sup γ; ΩM W2,β

j

e⊂Qj

∇γ; eL2 + γ; ΩL∞ . [cap(e, Q∗j )]1/2

Proof. First of all we note that for β > 1 the Hardy inequality (13.2.6) leads to the relation    u; Ω2W 1 ∼ ρ(x)β |∇u|2 + ρ(x)−2 u2 dx. 2,β

Ω

By {ηj } we denote a partition of unity, subordinate to the covering {Qj } and such that |∇ηj | ≤ c d−1 j . Assume also that {ζj } is a sequence of smooth functions with supp ζj ⊂ Q∗j ,

ζj = 1 on Qj

and |∇ζj | ≤ c d−1 j .

By the last equivalence relation and the Poincar´e inequality, we have  β u; Ω2W 1 ∼ dj ∇(ηj u); Ω2L2 . 2,β

j

The same holds if we replace ηj by ζj on the right-hand side. This relation and Theorem 1.2.2 give  β γu; Ω2W 1 ≤ c dj ∇(γηj ζj u); Ω2L2 2,β

j

≤ c sup sup j

e⊂Qj

∇(γηj ); e2L2  β dj ∇(ζj u); Ω2L2 cap(e, Q∗j ) j


∇γ; e2L2 mesn e  + γ; Ω2L∞ d−2 u; Ω2W 1 . ≤ c sup sup j 2,β cap(e, Qj ) j e⊂Qj cap(e, Qj ) Since the Poincar´e inequality implies the estimate cap(e, Q∗j ) ≥ c d2j mesn e, 1 we obtain the required upper estimate for the norm of γ in M W2,β (Ω). The inequality 1 γ; ΩL∞ ≤ γ; ΩM W2,β

is obtained by standard arguments (see Proposition 2.7.4). Setting u ∈ C0∞ (Q∗j ) , u = 1 on a compact e ⊂ Qj , in  ρ(x)β |∇u|2 dx ≤ γ; Ω2M W 1 u; Ω2W 1 , 2,ρ

Ω

we obtain

2,β

∇γ; e2L2 ≤ c cap(e, Q∗j )γ; Ω2M W 1 . 2,β

The proposition is proved.



452

13 Multipliers as Solutions to Elliptic Equations

In a similar manner we can derive the relation 1 →L ) ∼ sup sup h ρβ/2 ; ΩM (W2,β 2

j

e⊂Qj

h; eL2 , [cap(e, Q∗j )]1/2

where β > 1. Therefore, the estimate of the solution γ obtained in Theorem 13.2.1 can be written in the form
∇γ; eL2 f ; eL1 + g; eL2  . sup sup ≤ c γ; ΩL∞ + sup sup ∗ 1/2 [cap(e, Q∗j )]1/2 j e⊂Qj [cap(e, Qj )] j e⊆Qj In the case n > 2 we can replace cap(e, Q∗j ) by the Wiener capacity c2,1 , and the supremum with respect to j and with respect to e ⊂ Qj , by the supremum over compact sets e with d(e) ≤ ρ(e), where d(e) is the diameter of e and ρ(e) is the distance from e to ∂Ω. 1 (Ω) It follows from Proposition 13.2.1 that the spaces of multipliers in W2,β are isomorphic for all β > 1. Since the proof of the lower estimate for 1 (Ω) remains valid for β = 1 as well, it follows that in the case γM W2,β 1 1 β > 1 we have the imbedding M W2,β (Ω) ⊂ M W2,1 (Ω). We show that the last imbedding is strict. Indeed, let and u(x) = |log ρ(x)|1/2−ε , ε > 0.

γ(x) = sin log ρ(x)

1 1 One checks directly that u ∈ W2,1 (Ω) whereas γu ∈ W2,1 (Ω).

13.2.3 The Case β = 1 1 By S(Ω) we mean the space of functions from W2,loc (Ω) with the finite norm

u; ΩS =




2

ρ(x)|∇u| dx + Ω

u; Ω2L2



1/2

δ

∇u; Γτ L2 dτ,

+ 0

where δ is a small positive number and Γτ is the boundary of the domain Ωτ . Theorem 13.2.2. Suppose that the functions f and g are subject to ρ1/2 (|f |1/2 + |g|) + |g|1/2 ∈ M (S(Ω) → L2 (Ω)). Then any variational solution γ ∈ L∞ (Ω) of (13.2.1) belongs to the space 1 (Ω)) and M (S(Ω) → W2,1 + +2  1 ) ≤ c γ; ΩL γ; ΩM (S→W2,1 + +(ρ |f | + |g|)1/2 ; Ω +M (S→L ∞

2)

+ρ1/2 g; ΩM (S→L2 ) with a constant c, independent of f, g and γ.



(13.2.9)

13.2 Bounded Solutions of Linear Elliptic Equations as Multipliers

453

Proof. We use the notations R and ζτ , introduced in the proof of Theorem 13.2.1. We set η = Rζτ2 u2 γ in (13.2.2), where u is an arbitrary function from S(Ω). Then the left-hand side in (13.2.2) takes the form   ∂(γu) ∂(γu) 2 ∂u ∂u 2 2 aij Rζτ dx − aij γ Rζτ dx ∂x ∂x ∂x j i j ∂xi Ω Ω   ∂(γu) ∂(Rζτ2 ) ∂u ∂(Rζτ2 ) 2 aij γu dx − aij γ u dx. (13.2.10) + ∂xj ∂xi ∂xj ∂xi Ω Ω The third term on the right-hand side of (13.2.10) is equal to    
∂R ∂ζ 2 1 1 1 τ (γu)2 L(Rζτ2 ) dx = (γu)2 ζτ2 dx+ (γu)2 aij + RL(ζτ2 ) dx. 2 Ω 2 Ω ∂xi ∂xj 2 Ω The absolute value of the last integral does not exceed  u2 dx ≤ c1 γ; Ω2L∞ sup u; Γτ 2L2 . c γ; Ω2L∞ τ −1 0 ε; for ε2 ≤ r ≤ ε,

where ε is a small positive number. Then the right-hand side of (13.3.12) is O((log ε−1 )−1 ), while the left-hand side majorizes the expression  c 0

ε2


 ∞ loglog r−1 2 (log t)2 (loglog ε−1 )2 rdr ≥ c dt ≥ c . −1 2 rlog r t logε−1 logε−2

Thus, (13.3.12) is false and γ ∈ M W21 (Ω). It remains to note that the vectorvalued function γ = (γ1 , γ2 ) given by (13.3.11) satisfies a system for which (13.3.6) is valid, but that k is not sufficiently small. 13.3.3 Dirichlet Problem for Quasilinear Equations in Divergence Form In this subsection we extend Theorem 13.1.1 to a class of quasilinear equations. As above, we assume Ω to be an open bounded subset of Rn . Let the functions Ai (x, ξ) be measurable with respect to x for all ξ = (ξ1 , ..., ξn ), and continuous for almost all x ∈ Ω, and let the inequalities Ai (x, ξ)ξi ≥ c1 |ξ|p ,

n 

|Ai (x, ξ)| ≤ c2 |ξ|p−1

(13.3.13)

i=1

be satisfied for any ξ, where c1 , c2 are positive constants and p > 1. Further suppose that the monotonicity condition   Ai (x, v) − Ai (x, w) (vi − wi ) > 0 is satisfied for v = w. By the solution to the Dirichlet problem for the equation ∂Ai (x, ∇u) =0 ∂xi

462

13 Multipliers as Solutions to Elliptic Equations

we mean a function u ∈ Wp1 (Ω), satisfying  ∂v ˚ 1 (Ω), Ai (x, ∇u) dx = 0, u − g ∈ W p ∂x i Ω

(13.3.14)

˚p1 (Ω). where g is a given function from Wp1 (Ω), and v is any function from W It is known (see [LeL])that, under the above conditions on Ai , the problem (13.3.14) has a unique solution from the space Wp1 (Ω). If g is bounded, then the solution u is also bounded by the maximum principle. The following theorem on the unique solvability of the problem (13.3.14) in a multiplier space basically follows from Theorem 13.3.1. Theorem 13.3.3. If g ∈ M Wp1 (Ω), then the solution u of (13.3.14) belongs to the space M Wp1 (Ω). Proof. Set γ = u − g and introduce the notation ai (x, ∇γ(x)) := Ai (x, ∇γ(x) + ∇g(x)). By (13.3.13) we have   ai (x, ξ)ξi = Ai (x, ξ + ∇g(x))ξi ≥ c1 |ξ + ∇g(x)|p ≥ c1 21−p |ξ|p − |∇g(x)|p , n 

|ai (x, ξ)| =

i=1

n 

 p−1 |Ai (x, ξ + ∇g(x))| ≤ c2 |ξ| + |∇g(x)| .

i=1

Since ∇g ∈ M (Wp1 (Ω) → Lp (Ω)), the functions ai satisfy (13.3.3) and (13.3.4). It remains to use Theorem 13.3.1.  Let Ω be a bounded domain in Rn with ∂Ω ∈ C 0,1 (that is, a Lipschitz 1−1/p graph domain). In Sect. 8.7.4 it is shown that M Wp (Rn−1 ) is the space n−1 1 n of functions from M Wp (R+ ). It readily follows that any of traces on R 1−1/p

function ϕ from the space M Wp

(∂Ω) has an extension g onto Ω such that

c−1 g; ΩM Wp1 ≤ ϕ; ∂ΩM Wp1−1/p ≤ c g; ΩM Wp1 , where c is a constant depending only on Ω and c > 1. The next assertion follows from Theorem 13.3.3. Corollary 13.3.1. The Dirichlet problem ∂Ai (x, ∇u) = 0, ∂xi 1−1/p

where ϕ ∈ M Wp

u ∈ Wp1 (Ω), u ∂Ω = ϕ,

(∂Ω), has a unique solution in the space M Wp1 (Ω).

13.3 Solvability of Quasilinear Elliptic Equations in Spaces of Multipliers

463

13.3.4 Dirichlet Problem for Quasilinear Equations in Nondivergence Form We next prove a theorem showing that solutions to the Dirichlet problem for nondivergence quasilinear elliptic equations belong to the space M W22 (Ω). Let Ω be a domain with C 2 boundary and let Q be a nonnegative constant. Suppose that the functions F (x, ξ0 , ξ) and aij (x, ξ0 , ξ), i, j = 1, ..., n, are measurable with respect to x for all ξ0 ∈ (−Q, Q), ξ = (ξ1 , ..., ξn ) ∈ Rn , and continuous with respect to ξ0 , ξ for almost all x ∈ Ω. We assume that the functions aij are bounded and that there exists a positive constant c such that aij ξi ξj ≥ c |ξ|2 for the same ξ0 , ξ. Further, for some ε > 0, let the coefficients aij satisfy the Cordes condition


2n − 4 +

n n n 2 
 2
 3   2 ≤ (1 − ε) n aij − aii aii n+1 i,j=1 i=1 i=1

(see [Cor]), which restricts the dispersion of eigenvalues of the matrix aij ni,j=1 . Suppose that |F (x, ξ0 , ξ)| ≤ k |ξ|2 + g(x) for |ξ0 | ≤ Q, where

g ∈ M (W22 (Ω) → L2 (Ω))

and k is a small constant. Consider the Dirichlet problem aij (x, u, ∇u)

∂2u = F (x, u, ∇u) in Ω, ∂xi ∂xj

u ∂Ω = ϕ,

(13.3.15)

where u ∈ W22 (Ω), and ϕ ∈

3/2 M W2 (∂Ω).

|u(x)| ≤ Q almost everywhere in Ω,

We set v = u − Φ,

(13.3.16)

where Φ is an extension of ϕ onto Rn belonging to the class M W22 (Rn ) (see Theorem 8.6.1). Then the problem (13.3.15) is equivalent to the problem bij (x, v, ∇v)

∂2v = G(x, v, ∇v) in Ω, ∂xi ∂xj

v|∂Ω = 0,

where bij (x, ξ0 , ξ) = aij (x, ξ0 + Φ(x), ξ + ∇Φ(x))

464

13 Multipliers as Solutions to Elliptic Equations

and G(x, ξ0 , ξ) = F (x, ξ0 + Φ(x), ξ + ∇Φ(x)) + bij (x, ξ0 , ξ)

∂2Φ . ∂xi ∂xj

It is clear that the coefficients bij satisfy the same conditions as aij . Moreover, |G(x, ξ0 , ξ)| ≤ 2 k |ξ|2 + h(x),

(13.3.17)

where h = 2k|∇Φ|2 + c |∇2 Φ|. Lemma 13.3.1. The function h belongs to the space M (W22 (Ω) → L2 (Ω)). Proof. For any function w ∈ C0∞ (Rn ) we have     |∇Φ|4 w2 dx = − Φ div |∇Φ|2 (∇Φ)w2 dx Rn

≤ c Φ; R L∞ n

Rn


 Rn



|∇Φ| |∇2 Φ|w dx + 2

2




Rn

|∇Φ|3 |w| |∇w| dx



 ≤ c Φ; Rn L∞  |∇Φ|2 w; Rn L2  |∇2 Φ|w; Rn L2 +  |∇Φ| |∇w| ; Rn L2 . Since ∇2 Φ; Rn M (W22 →L2 ) + ∇Φ; Rn M (W21 →L2 ) ≤ c Φ; Rn M W22 ,

(13.3.18)

it follows that  |∇Φ|2 w; Rn L2 ≤ c Φ; Rn L∞ Φ; Rn M W22 w; Rn W22 . It is clear that the last inequality is valid for all w ∈ W22 (Rn ). Let w be any function in W22 (Ω), extended onto Rn and such that w; Rn W22 ≤ c w; ΩW22 , where c is independent of w. Then  |∇Φ|2 w; ΩL2 ≤ c w; ΩW22 , which means that |∇Φ|2 ∈ M (W22 (Ω) → L2 (Ω)). Using the inclusion g ∈ M (W22 (Ω) → L2 (Ω)), by (13.3.18) we find that the function |∇2 Φ| belongs to class M (W22 (Ω) →  L2 (Ω)). The proof is complete.

13.3 Solvability of Quasilinear Elliptic Equations in Spaces of Multipliers

465

3/2

Theorem 13.3.4. If ϕ ∈ M W2 (∂Ω) and u is a solution of problem (13.3.15) in the space W22 (Ω) such that |u| ≤ Q, then u belongs to the space M W22 (Ω). Proof. It is sufficient to assume that n ≥ 4, since otherwise M W22 (Ω) = ¯ and let σe be the C2,2 W22 (Ω). Let e be an arbitrary compact subset of Ω capacitary Bessel potential of e (see Sect. 3.6.2). We show that the function v, given by (13.3.16), belongs to the space M W22 (Ω). Since σe ∈ L∞ (Rn ) and W22 (Ω) ∩ L∞ (Ω) is an algebra with respect to multiplication, the product σe v is a function from W22 (Ω). We have bij

∂ 2 (σe v) ∂σe ∂v ∂ 2 σe = σe G − 2bij − v bij . ∂xi ∂xj ∂xi ∂xj ∂xi ∂xj

According to [Cor], the estimate   σe v; ΩW22 ≤ c σe G; ΩL2 +  |∇σe | |∇v| ; ΩL2 + v∇2 σe ; ΩL2 is valid. Hence we find from (13.3.17) and Lemma 13.3.1 that
σe v; ΩW22 ≤ c k σe |∇v|2 ; ΩL2 +  |∇σe ||∇v| ; ΩL2    + h; ΩM (W22 →L2 ) + v; ΩL∞ σe ; Rn W22 . We have

(13.3.19)

σe = (1 − ∆)−2 µ and (1 − ∆)σe = (1 − ∆)−1 µ,

where µ is a nonnegative measure. By G4 and G2 we denote the kernels of the integral operators (1 − ∆)−2 and (1 − ∆)−1 . In view of the relations given in Subsect. 13.3.4, , O(|x|3−n ) for |x| ≤ 1, |∇G4 | = for |x| > 1. O(|x|(3−n)/2 e−|x| Also,

, |G2 (x)| ≥ c

and

|x|2−n |x|(1−n)/2 e−|x|

⎧ 4−n ⎪ ⎨|x| |G4 (x)| ≥ c log|x|−1 ⎪ ⎩ (3−n)/2 −|x| e |x|

for |x| ≤ 1, for |x| > 1 for |x| ≤ 1, n > 4, for |x| < 1/2, n = 4, for |x| > 1.

Therefore, 1/2

1/2

|∇G4 | = O(G4 G2 ) and |∇σe (x)|2 ≤ c σe (x)(1 − ∆)σe (x).

(13.3.20)

466

13 Multipliers as Solutions to Elliptic Equations

We use this inequality to estimate the second norm on the right-hand side of (13.3.19): 1/2

1/2

 |∇σe | |∇v| ; ΩL2 ≤ c σe |∇v|2 ; ΩL2 (1 − ∆)σe ; ΩL2

≤ k σe |∇v|2 ; ΩL2 + c k −1 σe ; Rn W22 .

(13.3.21)

Now, (13.3.21) and (13.3.19) imply that
 σe v; ΩW22 ≤ c 2kσe |∇v|2 ; ΩL2

 + (h; ΩM (W22 →L2 ) + v; ΩL∞ + k −1 )σe ; Rn W22 .

(13.3.22)

˚ 2 (Ω) ∩ We estimate the first norm on the right-hand side. Since v ∈ W 2 integration by parts yields  I:= σe2 |∇v|4 dx Ω 
  σe |∇σe | |∇v|3 dx + σe2 |∇v|2 |∇2 v|dx , ≤ c v; ΩL∞

W21 (Ω),

Ω

Ω

which together with (13.3.20) gives the estimate
  3/2  1/2 I ≤ c v; ΩL∞ σe |∇v|2 (1 − ∆)σe dx Ω  + I 1/2 σe |∇2 v| ; ΩL2 . Consequently,
 1/2 I ≤ c v; ΩL∞ I 3/4 σe ; Rn W 2 + I 1/2 σe |∇2 v| ; ΩL∞ 2

and hence  1/2 I 1/2 ≤ c v; ΩL∞ I 1/4 σe ; Rn W 2 + σe v; ΩW22 2

 + |∇σe | |∆v| ; ΩL2 + v∇2 σe ; ΩL2 . Applying (13.3.20), we conclude that
I 1/2 ≤ c v; ΩL∞ σe v; ΩW22 + v; ΩL∞ σe ; Rn W22  + kI 1/2 + c k −1 σe ; Rn W22 . Since k v; Rn L∞ is small, it follows that
 σe |∇v|2 ; ΩL2 = I 1/2 ≤ c v; ΩL∞ σe v; Rn W22 + c k −1 σe ; Rn W22 .

13.4 Coercive Estimates for Solutions of Elliptic Equations

467

The required estimate is obtained for the first norm on the right-hand side of (13.3.22). We find from (13.3.22) that   σe v; ΩW22 ≤ c h; ΩM (W22 →L2 ) + k −1 σe ; Rn W22 . Therefore,  e

2  |∇2 v|2 dx ≤ c h; ΩM (W22 →L2 ) + k −1 C2,2 (e).

Using the equivalent norm for v; ΩM W22 , we complete the proof.



13.4 Coercive Estimates for Solutions of Elliptic equations in Spaces of Multipliers It is well known that solutions of elliptic boundary value problems satisfy coercive estimates in Sobolev spaces (see [ADN2]). The purpose of this section is to show that similar estimates are valid for norms in classes of multipliers acting in a Sobolev space or in a pair of Sobolev spaces. 13.4.1 The Case of Operators in Rn Theorem 13.4.1. Let P be an elliptic (in the sense of Douglis and Nirenberg) operator (10.1.4), where M = N . Let the coefficients of P be constant. Further let γ = {γ 1 , . . . , γ N } be a vector-valued function in the space - h+t 7 Wp,lock M (Wpr−h−tk → Lp ) k

k

and let Pγ ∈

-

M (Wpr → Wph−sj ),

j

where r ≥ h + tk ≥ 0, r ≥ h − sj ≥ 0, 1 ≤ j, k ≤ N . Then γ∈ M (Wpr → Wph+tk ) k

and the estimate γ8  ≤ C P γ8 holds.

k

h+tk )

M (Wpr →Wp

h−sj r ) j M (Wp →Wp

+ γ8k M (W r−h+tk →Lp )



(13.4.1)

468

13 Multipliers as Solutions to Elliptic Equations

Proof. It is known that for all u ∈ u8

t +h

k k Wp

8 k

Wptk +h

 ≤ C1 P u8

h−sj

Wp

j

 + uLp .

Consequently, for all ϕ ∈ C0∞ , γρ ϕ8  ≤ C1 ϕP γρ 8

h−sj

j Wp

tk +h k Wp

+ γρ ϕLp + [ϕ, P ]γρ 8

 h−sj

j Wp

,

(13.4.2)

where [ϕ, P ] is the commutator of P and the operator of multiplication by ϕ. As usual, by γρ we denote a mollification of γ with radius ρ. It is clear that ϕP γρ 8

≤ P γρ 8

h−sj j Wp

h−sj r ) j M (Wp →Wp

ϕWpr

(13.4.3)

and γρ ϕLp ≤ γρ 8

r−h−tk →Lp ) k M (Wp

ϕW r−h−mink tk .

(13.4.4)

p

It remains to estimate the third term in (13.4.2). For any multi-index α, |α| ≤ sj + tk , we have  Dβ ϕ Dα−β γρ 8 W h−|α|+tk [ϕ, Dα ]γρ 8 h−sj ≤ c j Wp

k

0 n. In the second, stronger, formulation, the boundary data are prescribed by means of some differential operators Pj , 1 ≤ j ≤ h. We prove that such a problem is solvable for h > 1 if l−1/p l−1−1/p ∂Ω belongs to the class Mp defined by the condition ∇ϕ ∈ M Wp , where ϕ is the same as in (14.1.3), and if the Lipschitz constant of ϕ is small. l−1/p In the case p(l − 1) > n, this condition is equivalent to ∂Ω ∈ Wp . l−1/p The inclusion ∂Ω ∈ Wp for p(l − 1) > n is not only sufficient but also necessary for solvability of the Dirichlet problem in the second formulation (see Sect. 14.6). l−1/p (δ) inIn Sect. 14.6.3 we give an analytic description of the class Mp volving a capacity and obtain some simpler conditions for the inclusion of l−1/p (δ). For instance, if the norm ∇ϕ; Rn−1 L∞ is small and ϕ ∂Ω into Mp

14.2 Change of Variables in Differential Operators

481

l−1/p

belongs to the Besov space Bq,p (Rn−1 ) with q ∈ [p(n − 1)/(p(l − 1) − 1), ∞] l−1/p for p(l −1) < n and q ∈ (p, ∞] for p(l −1) = n, then ∂Ω belongs to Mp (δ). Putting q = ∞ we obtain that this inclusion follows from the convergence of the integral  [ωl−1 (t)/t]p dt, 0

where ωl−1 is the modulus of continuity of the vector-function ∇l−1 ϕ. l−1/p l−1/p , the last condition is also In view of the imbedding B∞,p ⊂ Wp l−1/p sufficient for ∂Ω ∈ Wp . Using this fact, one can immediately derive the following assertions from our theorems. The inclusion ωl−1 (t)/t ∈ Lp (0, 1) provides the Fredholm property of the operator {P ; trPj } as well as the unique solvability of the Dirichlet problem in the second formulation. Moreover, the unique solvability in Wpl (Ω) of the Dirichlet problem in the first (generalized) formulation is obtained under the assumption ωl−h (t)/t ∈ Lp (0, 1). In Sect. 14.6.3 we note that even these, the roughest of our sufficient conditions, are precise in a sense. The proof of a theorem in 14.6.3, which contains a local characterization l−1/p (δ), is given in 14.7. of the class Mp

14.2 Change of Variables in Differential Operators Consider the domain G = {z = (x, y) ∈ Rn : x ∈ Rn−1 , y > ϕ(x)}, where ϕ is a function satisfying the Lipschitz condition |ϕ(x1 ) − ϕ(x2 )| ≤ L |x1 − x2 |. The following assertion characterizes coefficients of a differential operator under a change of variables. Proposition 14.2.1. Let G be a special Lipschitz domain and let λ be an arbitrary (p, l)-diffeomorphism Rn+ → G, κ = λ−1 . Further, let R(z, Dz ) =



aα (z)Dzα ,

z∈G,

0≤|α|≤h

and S(ζ, Dζ ) =

 0≤|β|≤h

bβ (ζ)Dζβ ,

ζ ∈ Rn+ ,

482

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

be differential operators in G and Rn+ such that Sv = [R(v ◦ κ)] ◦ λ .

(14.2.1)

If aα ∈ M (Wpl−|α| (G) → Wpl−h (G)) for all multi-indices α, then bβ ∈ M (Wpl−|β| (Rn+ ) → Wpl−h (Rn+ )) and bβ ; Rn+ M (Wpl−|β| →Wpl−h ) ≤ c

 |β|≤|α|≤h

aα ; GM (Wpl−|α| →Wpl−h ) .

The proof follows the same lines as that of Lemma 10.1.3. We note that according to (10.1.20) the equality   bβ = (aα ◦ λ) cs (Dzsij κi ) ◦ λ |β|≤|α|≤h

(14.2.2)

(14.2.3)

i,j

holds. Lemma 14.2.1. Let G denote a special Lipschitz domain and let λ be an arbitrary (p, l)-diffeomorphism: Rn+ → G. Further, let R be a homogeneous differential operator of order h with constant coefficients and let S be an operator defined by (14.2.1). Then S − R; Rn+ M (W l →Wpl−h ) ≤ c I − ∂λ; Rn+ M Wpl−1 , p

(14.2.4)

where c is a continuous function of the norm of ∂λ in M Wpl−1 (Rn+ ). (Here and henceforth by the norm of a matrix we mean the sum of the norms of its elements.) Proof. We put κ = λ−1 and a = I − ∂λ; Rn+ M Wpl−1 . Let S1 (ζ, Dζ ) denote the principal homogeneous part of the operator S. Since S1 (ζ, ρ) = S((∂κ)∗ ρ) ◦ λ for any vector ρ ∈ Rn , it follows that every coefficient of S1 differs from the corresponding coefficient of R by O(a) in the norm of M Wpl−h (Rn+ ). Hence, S1 − R; Rn+ M (W l →Wpl−h ) ≤ c a. p

Consider the coefficients of S which multiply the derivatives of order |β| < h. Let formula (14.2.3) relate the coefficients aα and bβ of the operators R

14.3 Fredholm Property of the Elliptic Boundary Value Problem

483

and S. Since R is homogeneous, we have |α| = h in (14.2.3). Hence by (14.2.2) every term in (14.2.3) with |β| < h contains at least one factor Dsij κi (z) for which |sij | > 1. Noting that such a factor is equal to Dsij [κi (z) − zi ], we obtain bβ ; Rn+ M (Wpl−|β| →Wpl−h ) ≤ c I − ∂κ; GM Wpl−1 ≤ c a (see the proof of Lemma 10.1.3). Therefore, S − S1 ; Rn+ M (W l →Wpl−h ) ≤ c a. p

 Duplicating the proof of Lemma 14.2.1 with obvious changes and using the properties of (p, l)-diffeomorphisms given in Sect. 9.4.1, we obtain the following local variant of Lemma 14.2.1. Lemma 14.2.2. Let all conditions of Proposition 14.2.1 be satisfied. Then for each v ∈ Wpl (Rn+ ) with support in Br ∩ Rn+ , (S − R)v; Rn+ Wpl−h ≤ c |||I − ∂λ; Br ∩ Rn+ |||M Wpl−1 v; Rn+ Wpl ,

(14.2.5)

where c is a constant independent of r ∈ (0, 1). For p(l − 1) > n it follows from (9.6.4) that (14.2.5) is equivalent to (S − R)v; Rn+ Wpl−h ≤ c rl−1−n/p |||I − ∂λ; Br ∩ Rn+ |||Wpl−1 v; Rn+ Wpl . (14.2.6)

14.3 Fredholm Property of the Elliptic Boundary Value Problem 14.3.1 Boundaries in the Classes Mpl−1/p , Wpl−1/p , and Mpl−1/p (δ) l−1/p

Let Ω be a bounded domain with ∂Ω ∈ C 0,1 . We introduce the class Mp (l = 2, 3, . . . ) of boundaries ∂Ω, satisfying the following condition. For every point of ∂Ω there exists an n-dimensional neighborhood in which ∂Ω is specified (in a certain Cartesian coordinate system) by a function ϕ such that ∇ϕ ∈ M Wpl−1−1/p (Rn−1 ). 1−1/p

Furthermore, by definition, Mp = C 0,1 . l−1/p We say that ∂Ω belongs to the class Wp if ∂Ω can be locally specified l−1/p n−1 by a function ϕ ∈ Wp (R ). Since l−1−1/p

M Wpl−1−1/p (Rn−1 ) ⊂ Wp,loc

(Rn−1 ),

l ≥ 2,

484

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems l−1/p

and C 0,1 (Rn−1 ) ⊂ Wp,loc (Rn−1 ), it follows that any bounded domain Ω with l−1/p

l−1/p

∂Ω ∈ Mp satisfies ∂Ω ∈ Wp . According to Corollary 4.3.8, for p(l − 1) > n we have ∇ϕ; Rn−1 M Wpl−1−1/p ∼ sup ∇ϕ; B1 (x)Wpl−1−1/p . x∈Rn−1

l−1/p

l−1/p

Therefore, the classes Mp and Wp coincide for p(l − 1) > n. l−l/p 0,1 For a bounded domain Ω with ∂Ω ∈ C , by Wp (∂Ω) we denote the l space of traces on ∂Ω of functions in Wp (Ω). Taking into account the anall−1/p

ogous fact for special Lipschitz domains of the class Mp (see Sect. 9.4.3), l−1/p we obtain that M Wp (∂Ω) is the space of traces of functions in M Wpl (Ω). ¯ of orders 2h, k1 , . . . , kh , Let P, P1 , . . . , Pk be differential operators in Ω respectively, where 2h ≤ l and kj < l. Suppose that the coefficients of P and ¯ and C l−kj (Ω), ¯ respectively. (This restriction can be Pj belong to C l−2h (Ω) removed by the use of spaces of multipliers, but we do not want to complicate the formulations.) We assume that the operators P , tr P1 , . . . , tr Pn form an elliptic boundary value problem at every point O ∈ ∂Ω with respect to the hyperplane y = 0 and that P is an elliptic operator in Ω. In our subsequent exposition the following additional condition on Ω will play an important role. l−1/p l−1/p (δ). We say that ∂Ω belongs to the class Mp (δ) if The class Mp for each point O ∈ ∂Ω there exists a neighborhood U and a special Lipschitz domain G = {z = (x, y) : x ∈ Rn−1 , y > ϕ(x)} such that U ∩ Ω = U ∩ G and ∇ϕ; Rn−1 M Wpl−1−1/p ≤ δ . Here p(l − 1) ≤ n and δ is a constant which depends on the coefficients of the principal homogeneous parts of P , P1 , . . . , Ph calculated at the point O in the coordinate system (x, y). For l = 1 the role of the last inequality is played by ∇ϕ; Rn−1 L∞ ≤ δ. l−1/p

l−1/p

Obviously, the boundaries in Mp (δ) belong to the class Mp and, l−1/p therefore, to the class Wp . In 14.6.3 we give an equivalent description of l−1/p Mp (δ) and discuss sufficient conditions for the inclusion into this class. 14.3.2 A Priori Lp -Estimate for Solutions and Other Properties of the Elliptic Boundary Value Problem In the next two theorems we consider separately the cases p(l − 1) ≤ n and p(l − 1) > n. Theorem 14.3.1. If p(l − 1) ≤ n, 1 < p < ∞, and if ∂Ω belongs to the class l−1/p Mp (δ), then (14.1.2) holds for any u ∈ Wpl (Ω).

14.3 Fredholm Property of the Elliptic Boundary Value Problem

485

l−1/p

Proof. We retain the notation used in the definition of Mp (δ). Let U be an open ball with a small radius, σ ∈ C0∞ (U ), and R and Rj be the principal homogeneous parts of the operators P and Pj with “frozen” coefficients at the point O. Clearly, (P −R)(σu); U ∩ΩWpl−2h ≤ ε σu; U ∩ΩWpl +c σu; U ∩ΩWpl−1 , (14.3.1) where ε is a small positive number (the required smallness is defined by the coefficients of the operators R, R1 , . . . , Rh ). An analogous estimate holds for l−k the norm of (Pj − Rj )(σu) in Wp j (U ∩ Ω). l−1/p (δ), the Lipschitz constant of ϕ is small, By definition of the class Mp so we can put N = 1 in the definition (9.4.14) of the mapping λ : Rn+ → G. Then, from (9.4.12), we obtain that ∂λ differs from the identity matrix by O(δ) in the norm of M Wpl−1 (Rn+ ). It is well known (see, for instance, [ADN1], [Tr3], Sect. 5.3.3) that, for all v ∈ Wpl (Rn+ ) with supports in B1 ∩ Rn+ , 

h
 trRj v; Rn−1  v; Rn+ Wpl ≤ c Rv; Rn+ Wpl−2h +

l−k −1/p Wp j

j=1

.

Since δ is small, we can replace here R by Rj and S by Sj (see Lemma 14.2.2). From the estimate obtained after this change, it follows that
σu; U ∩ ΩWpl ≤ c R(σu); U ∩ ΩWpl−2h +

h 



trRj (σu); U ∩ ∂Ω

l−k −1/p Wp j

j=1

.

This inequality and (14.3.1) entail h
 σu; ΩWpl ≤ c P (σu); ΩWpl−2h + trPj (σu); ∂Ω

l−kj −1/p

Wp

j=1

 +σu; ΩWpl−1 . Summing over all sufficiently small neighborhoods U which generate a covering ¯ we arrive at of Ω, h
 trPj u; ∂Ω u; ΩWpl ≤ c P u; ΩWpl−2h + j=1

l−k −1/p Wp j

 + u; ΩWpl−1 .

It remains to use the known inequality u; ΩWpl−1 ≤ ε u; ΩWpl + c(ε)u; ΩL1 , where ε is any positive number. Estimate (14.1.2) is proved.



486

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems l−1/p

Theorem 14.3.2. If p(l − 1) > n, 1 < p < ∞, and if ∂Ω ∈ Wp conclusion of Theorem 14.3.1 holds.

, then the

l−1/p

Proof. From the condition ∂Ω ∈ Wp and the Sobolev embedding theorem, it follows that ∂Ω ∈ C 1 . We place the origin at the point O ∈ ∂Ω and direct the axis Oy along the interior normal to ∂Ω. Let U be the neighborhood of O l−1/p , i.e. U ∩ Ω = U ∩ G, where G = {z : x ∈ in the definition of the class Wp l−1/p Rn−1 , y > ϕ(x)} and ϕ ∈ Wp (Rn−1 ). Let ε be a small positive number, which will be specified later, and let Bρ = {z ∈ Rn : |z| < ρ}. We choose a small number ρ such that ∇ϕ; Bρ ∩ Rn−1 L∞ < ε and B2ρ ⊂ U . Let τ ∈ C0∞ (B2 ), τ = 1 on B1 , and τρ (z) = τ (z/ρ). We introduce the function ϕ∗ = ϕτρ on Rn−1 and note that ∇ϕ∗ ; Rn−1 L∞ < c ε. We also define the extension Φ of ϕ∗ onto Rn+ by Φ = T ϕ∗ . According to (9.4.12), where ϕ is replaced by ϕ∗ , ∇Φ; Rn+ L∞ ≤ c ε . Since

(14.3.2)

ϕ∗ ; Rn−1 Wpl−1/p ≤ c(ρ)ϕ; Rn−1 Wpl−1/p ,

we have Φ; Rn+ Wpl ≤ c(ρ)ϕ; Rn−1 Wpl−1/p . Now let r be a small positive number such that r < ρ and rl−1−n/p Φ; Rn+ Wpl < ε .

(14.3.3)

It follows from the inequality   |||∇Φ; Br ∩ Rn+ |||Wpl−1 ≤ c ∇l Φ; Br ∩ Rn+ Lp + r1−l+n/p ∇Φ; Br ∩ Rn+ L∞ and the estimates (14.3.2) and (14.3.3) that rl−1−n/p |||∇Φ; Br ∩ Rn+ |||Wpl−1 ≤ c ε . According to (9.6.4), this means that |||∇Φ; Br ∩ Rn+ |||M Wpl−1 ≤ c ε . Using the function Φ, we define the mapping λ by (9.4.14) with N = 1. By the last inequality, |||I − ∂λ; Br ∩ Rn+ |||M Wpl−1 ≤ c ε . Now it suffices to duplicate the arguments we have already used in Theorem 14.3.1, with Br in place of the ball U , and estimate (14.2.6) instead of (14.2.5).

14.3 Fredholm Property of the Elliptic Boundary Value Problem

487

The following assertion can be deduced in a standard way from the a priori estimate (14.1.2) (see, for instance [H1], §10.5; [Tr3], Sect. 5.4.3). Proposition 14.3.1. Let the domain Ω satisfy the conditions of either Theorem 14.3.1 or Theorem 14.3.2. (i) If the kernel of the operator (14.1.1) is trivial, then the norm u; ΩL1 in (14.1.2) can be omitted. (ii) The kernel of the operator (14.1.1) is finite-dimensional. (iii) The range of the operator (14.1.1) is closed. Proof. (i) Suppose that the assertion is not true. Then there exists a sequence of functions {vm }m≥1 in Wpl (Ω) such that vm ; ΩWpl = 1, P vm ; ΩWpl−2h +

(14.3.4) h 

trPj vm ; ∂Ω

l−kj −1/p

Wp

j=1

→0.

(14.3.5)

We can select a subsequence of {vm }, also denoted by {vm }, which weakly converges in Wpl (Ω) to a function v ∈ Wpl (Ω). Since the imbedding operator Wpl (Ω) → L1 (Ω) is compact, we can assume that vm → v in L1 (Ω). Substituting vm − vk into (14.1.2), we have vm → v in Wpl (Ω). This and (14.3.5) imply that v ∈ ker{P ; trPj }, i.e. v = 0, which contradicts (14.3.4). (ii) It follows from (14.1.2) that, for all v ∈ ker{P ; trPj }, v; ΩWpl ≤ c v; ΩL1 . Therefore a unit sphere in ker{P ; trPj }, considered as a subspace of Wpl (Ω), is compact and the dimension of the kernel is finite. (iii) Since dim ker{P ; trPj } < ∞, there exists a projection operator Π which acts parallel to ker {P ; trPj }. Duplicating the arguments used in part (i) of the present proof, we obtain h
 v; ΩWpl ≤ c P v; ΩWpl−2h + trPj v; ∂Ω j=1



l−k −1/p Wp j

for all v ∈ Wpl (Ω), which implies that the range of the operator {P ; tr Pj } is closed.  Next we derive a local a priori estimate for solutions of the elliptic boundary value problem (cf. [ADN1], Sect. 15). Proposition 14.3.2. Let the domain Ω satisfy the condition of either Theorem 14.3.1 or Theorem 14.3.2. Further, let U and V be open subsets of Rn with U ⊂ V , and let u ∈ Wpl (V ∩ Ω). Then

488

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

u; U ∩ ΩWpl

h
 ≤ c P u; V ∩ ΩWpl−2h + trPj u; V ∩ ∂Ω

 +u; V ∩ ΩL1 .

l−kj −1/p

Wp

j=1

(14.3.6)

Proof. Let U and V be concentric balls with radii r and ρ, r < ρ. Further, let either V¯ ⊂ Ω or the centre of the balls be placed on ∂Ω. It suffices to prove the proposition under this additional assumption. We introduce the sets C0 = U ∩ Ω,

Ck = {x ∈ Ω : δk < ρ − |x| < δk−1 },

where k = 1, 2, . . . and δk = (ρ − r)2−k . Let D0 = C0 ∪ Cl ,

Dk = Ck−1 ∪ Ck ∪ Ck+1 , k = 1, 2, . . . .

We construct a C ∞ -partition of unity {σk }k≥0 subordinate to the covering {Dk }k≥0 of V ∩ Ω and satisfying −|α|

|Dα σk | = O(δk

)

for any multi-index α. Applying the a priori estimate (14.1.2) to σk v, we obtain
σk v; ΩWpl ≤ c P (σk v); ΩWpl−2h +

h 

trPj (σk v); ∂Ω

l−k −1/p Wp j

j=1

 + σk v; ΩL1 ,

which implies that


σk v; ΩWpl ≤ c σk P v; ΩWpl−2h +

h 

σk trPj v; ∂Ω

l−k −1/p Wp j

j=1

 + δk−l v; Dk Wpl−1 .

Let M be a sufficiently large positive number. We have ∞ 

∞
 δkM v; Ck Wpl ≤ c δkM σk P v; ΩWpl−2h

k=0

+

∞  k=0

k=0

δkM

h 

σk trPj v; ∂Ω

j=1

l−k −1/p Wp j

+

∞ 

δkM −l v; Dk Wpl−1

k=0

h
 ≤ c P v; V ∩ ∂ΩWpl−2h + trPj v; V ∩ ∂Ω j=1

+

∞  k=0

 δkM −l v; Dk Wpl−1 .

l−kj −1/p

Wp



14.4 Auxiliary Assertions

489

Note that, for any ε > 0 and for some positive N , v; Dk Wpl−1 ≤ ε δkl v; Dk Wpl + c(ε)δk−N v; Dk L1 . Consequently, ∞ 

h
 δkM v; Ck Wpl ≤ c P v; V ∩ ΩWpl−2h + trPj v; V ∩ ∂Ω

k=0

+c(ε)v; V ∩ ΩL1 + ε

j=1 ∞ 

l−kj −1/p

Wp

 δkM v; Dk Wpl .

k=0

Clearly, the last sum can be removed by changing c. The result follows.



Using the same properties of P , Pj , λ, κ as those used in the proof of Theorems 14.3.1 and 14.3.2, one can establish the existence of a right regularizer by the same argument as in, for instance, [Wl], Sect. 13. Proposition 14.3.3. Let the domain Ω satisfy the condition of either Theorem 14.3.1 or Theorem 14.3.2. Then there exists a linear bounded operator R : Wpl−2h (Ω) ×

h -

Wpl−kj −1/p (∂Ω) → Wpl (Ω)

j=1

such that {P ; trPj }R = I + K. Here I and K are the identity and compact operators respectively. A direct corollary of Proposition 14.3.1 and 14.3.3 is: l−1/p

l−1/p

Theorem 14.3.3. Let ∂Ω belong to Mp (δ) for p(l−1) ≤ n and to Wp for p(l − 1) > n. Then the operator (14.1.1) is Fredholm, that is, its null space is finite-dimensional and its range is closed and has a finite codimension. In the following sections we consider the Dirichlet problem in more detail.

14.4 Auxiliary Assertions 14.4.1 Some Properties of the Operator T In this subsection, T is the operator defined by (9.4.11). Lemma 14.4.1. Let α be an n-tuple multi-index and let k, r be nonnegative integers with k ≥ |α| − r ≥ 0. Then the operator M (Wpk−1/p (Rn−1 ))  γ → η r (Dα T γ)(ζ) ∈ M (Wpk (Rn+ ) → Wpk−|α|+r (Rn+ )) is continuous.

490

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

Proof. Clearly, η r (Dα T γ)(ζ) = Dβ



cν η |ν| (Dν T γ)(ζ) ,

0≤|ν|≤r

where β is a multi-index of order |α| − r, cν = const. The operator T

ν η |ν| (Dν T γ)(ζ) γ −→

k−1/p

has the same form as T and therefore it maps M Wp M (Wpk (Rn+ )). Hence, the continuity of the operator

(Rn−1 ) into

M (Wpk−1/p (Rn−1 ))  γ → Dβ Tν γ ∈ M (Wpk (Rn+ ) → Wpk−|α|+r (Rn+ )) 

follows from Corollary 2.4.1. The next assertion follows directly from the lemma just proved.

Corollary 14.4.1. Let G be a special Lipschitz domain, let α be a positive n-tuple multi-index and let r be a nonnegative integer with l ≥ |α| − r > 0. Then the function ζ → η r (Dα T ϕ)(ζ) belongs to the space M (Wpl−1 (Rn+ ) → l−|α|+r

Wp

(Rn+ )) and

η r Dα T ϕ; Rn+ M (Wpl−1 →Wpl−|α|+r ) ≤ c ∇ϕ; Rn−1 M Wpl−1−1/p . 14.4.2 Properties of the Mappings λ and κ Let G be a special Lipschitz domain and let λ be the mapping (9.4.13) defined by (9.4.14). As in Sect. 9.4.3, we denote by κ the inverse mapping to λ. We shall assume that L < 1 and N = 1. From Corollary 14.4.1 we deduce Corollary 14.4.2. Let α be a multi-index. Let r be a nonnegative integer with l ≥ |α| − r + 1 > 0, and write λ(ζ) = {λ1 (ζ), . . . , λn (ζ)}. Then the function ζ → η r (Dα ∂λi )(ζ) l−1−|α|+r

belongs to M (Wpl−1 (Rn+ ) → Wp

(Rn+ )) and

η r Dα (∂λ − I); Rn+ M (Wpl−1 →Wpl−1−|α|+r ) ≤ c ∇ϕ; Rn−1 M Wpl−1−1/p . A similar assertion concerning the mapping κ needs a separate proof.

14.4 Auxiliary Assertions

491

Lemma 14.4.2. Let α be a multi-index, let r be a nonnegative integer with l ≥ |α| − r + 1 > 0, and write κ(z) = {κ1 (z), . . . , κn (z)}. Then the function z → (η r Dα ∂κi )(z) l−1−|α|+r

belongs to M (Wpl−1 (G) → Wp

(G)) and

η r Dα (∂κ − I); GM (Wpl−1 →Wpl−|α|+r−1 ) ≤ c ∇ϕ; Rn−1 M Wpl−1−1/p . Proof. For |α| = 0 the result follows from the definition of a (p, l)diffeomorphism. Suppose that the lemma is proved for |α| < N . Let |α| = N , r < N . For any multi-index δ of order N − 1, (Dδ ∂κ)(z) = Dδ [∂λ(κ(z))]−1 =



[Dβ (∂λ)−1 ](κ(z))



cs

n -

Dsij κi (z) ,

i=1 j

1≤|β|≤|δ|

where the summation is taken over all collections of multi-indices s = (sij ) such that   (|sij | − 1) = |δ| − |β|. sij = δ, |sij | ≥ 1, Therefore, the expression (η r Dδ ∂κ)(z) is the sum of the products of two factors 5 6 π1 (z) = cβ η |β| Dβ (∂λ)−1 (κ(z)) and π2 (z) =

n -

(η r−|β| Dsij κi )(z) .

i=1 j

Corollary 14.4.2 implies that the function ζ → η |β| Dβ (∂λ)−1 belongs to the space M Wpl−1 (Rn+ ), and since κ is a (p, l)-diffeomorphism, it follows that π1 ∈ M Wpl−1 (G). We introduce positive integers σij such that  σij ≤ |sij |, (σij − 1) = r − |β|. Then π2 (z) =

n i=1 j

(η σij −1 Dsij κi )(z) .

492

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

Since |β| ≥ 1, we have |sij | ≤ N − 1 and by the induction hypothesis the function z → (η σij −1 Dsij κi )(z) belongs to M (Wpq (G) → Wpq−1−|sij |+σij (G)),

q = |sij | − σij , . . . , l − 1.

Hence 

π2 ∈ M (Wpl−1 (G) → Wpl−2−

(|sij |−σij )

(G)) = M (Wpl−1 (G) → Wpl−2−|δ|+r (G)).

Noting that |α| = 1 + |δ| and π1 ∈ M Wpl−1 (G), we obtain π1 π2 ∈ M (Wpl−1 (G) → Wpl−1−|α|−r (G)). Since the space Wpl is invariant under the (p, l)-diffeomorphisms, a function l−1/p

u on ∂G belongs to Wp

l−1/p

(∂G) if and only if u◦tr λ ∈ Wp

(Rn−1 ). We put

u; ∂GWpl−1/p = u ◦ trλ; Rn−1 Wpl−1/p . Taking here an arbitrary (p, l)-diffeomorphism: Rn+ → G instead of λ, we obtain an equivalent norm (see Lemmas 9.4.1, 9.4.2 and 9.4.4). ˚ h Under a Change 14.4.3 Invariance of the Space Wpl ∩ W p of Variables In this subsection we present auxiliary assertions which will be used later in the study of conditions for solvability of the Dirichlet problem in Wpl (Ω). As before, let G = {z = (x, y) : x ∈ Rn−1 , y > ϕ(x)} and let ∂G belong l+1−h−1/p to the class Mp , where l and h are integers with l ≥ h ≥ 1. In other words, if l > h ∇ϕ ∈ M Wpl−h−1/p (Rn ) and ϕ ∈ C 0,1 (Rn−1 )

if

l = h.

Let λ be the mapping Rn+  (ξ, η) → (x, y) ∈ G, defined by (9.4.13), and let κ = λ−1 . ˚pk ∩ Wpt+k )(G), where 0 ≤ t ≤ l − h, and let Lemma 14.4.3. Let v ∈ (W l+1−h−1/p

∂G ∈ Mp

. Then η −k v; GWpt ≤ c v; GWpt+k .

(14.4.1)

14.4 Auxiliary Assertions

493

Proof. Since η(z) is equivalent to y − ϕ(x), the inequality (14.4.1) with t = 0 follows from the Hardy inequality,  ∞  ∞ k p dy ∂ v |v(x, y)|p ≤ c (14.4.2) k (x, y) dy pk (y − ϕ(x)) ∂y ϕ(x) ϕ(x) a.e. in Rn−1 . Let the lemma be proved for all t < T and k > K. We have η −K v; GWpT ≤ ∇(η −K v); GWpT −1 + η −K v; GLp . The second term on the right-hand side is estimated by (14.4.2) and the first one does not exceed η −K ∇v; GWpT −1 + Kη −K−1 v∇η; GWpT −1 . l+1−h−1/p

Since ∂G ∈ Mp

(14.4.3)

, it follows that

∇η ∈ M Wpl−h (G) ⊂ M WpT −1 (G). Hence the sum (14.4.3) is dominated by η −K ∇v; GWpT −1 + c η −K−1 v; GWpT −1 . Using the induction hypothesis, we complete the proof.



˚ph (G), the inequality Lemma 14.4.4. For all u ∈ Wpl (G) ∩ W u ◦ λ; Rn+ Wpl ≤ c u; GWpl

(14.4.4)

holds. Proof. We have  Rn +

≤c

|Dζα [u(λ(ζ))]|p dζ

 

1≤|β|≤l

Rn +

p  - s β cs Dζ ij λi (ζ) dζ , (D u)(λ(ζ)) s

i,j

where α is an arbitrary positive multi-index of order l and s = (sij ) is the set of multi-indices satisfying (10.1.19). Hence,  |Dζα [u(λ(ζ))]|p dζ Rn +

≤c

p    cs (Dsij λi )(κ(z)) dz ∂κ; GpL∞ . (14.4.5) (Dβ u)(z) 1≤|β|≤l

G

s

i,j

494

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

Let |β| ≥ h or, what is the same, l − |β| ≤ l − h. Since ∇λi ∈ M Wpk (Rn+ ) for k ≤ l − h, we have Dsij λi ∈ M (Wpk (Rn+ ) → Wpk−|sij |+1 (Rn+ )) for |sij | − 1 ≤ k ≤ l − h. Consequently, -

  (|sij |−1) n  Dsij λi ∈ M Wp i,j (R+ ) → Lp (Rn+ ) = M (Wpl−|β| (Rn+ ) → Lp (Rn+ )).

i,j

Since κ is a (p, l − h)-diffeomorphism, we obtain Dsij λi ◦ κ ∈ M (Wpl−|β| (G) → Lp (G)) . i,j

Therefore, the terms on the right-hand side of (14.4.5) which correspond to multi-indices β of order |β| ≥ h are majorized by c u; GpW l . Now suppose that |β| ≤ h − 1. By (14.4.1) the function

p

z → η(z)|β|−h (Dβ u)(z) belongs to the space Wpl−h (G). By σij we denote integers subject to 1 ≤ σij ≤ |sij |,



(σij − 1) = h − |β|.

i,j

Such numbers exist, since  (|sij | − 1) = l − |β|

and

l ≥ h.

i,j

By Corollary 14.4.2, the function ζ → η σij −1 (Dsij λi )(ζ) k−|sij |+σij

belongs to M (Wpk (Rn+ ) → Wp -

(Dsij λi )(ζ) = η |β|−h

i,j

(Rn+ )). Using the identity -

η σij −1 (Dsij λi )(ζ) ,

i,j

we observe that the function ζ → η h−|β|

(Dsij ◦ λi )(ζ)

belongs to M (Wpl−h (Rn+ ) → Lp (Rn+ )). Since κ is a (p, l − h)-diffeomorphism, the function z → η(z)h−|β| (Dsij λi )(κ(z))

14.4 Auxiliary Assertions

495

is an element of M (Wpl−h (G) → Lp (G)). Therefore, the terms with |β| ≤ h−1 on the right-hand side of (14.4.5) are bounded by c u; GpW l . We also have p

u; GLp ∼ u ◦ λ; Rn+ Lp because λ is a bi-Lipschitz mapping. This relation and (14.4.5) give the estimate (14.4.4).  Now we prove an analogous assertion concerning the mapping κ. ˚ h (Rn ), the inequality Lemma 14.4.5. For each v ∈ Wpl (Rn+ ) ∩ W p + v ◦ κ; GWpl ≤ c v; Rn+ Wpl

(14.4.6)

holds. Proof. It is sufficient to derive the estimate  |Dzα [v(κ(z))]|p dz ≤ c v; Rn+ pW l . p

Ω

The left-hand   c 1≤|β|≤l

side is dominated by

Rn +

p  β cs (Dsij κi )(λ(ζ)) dζ ∂λ; Rn+ pL∞ (Dζ v)(ζ) s

(14.4.7)

i,j

(cf. (14.4.5)). Let |β| ≥ h. Repeating the same arguments as in Lemma 14.4.4 with Rn+ replaced by G, λ by κ, u by v and vice versa, we obtain that the terms in (14.4.7) with |β| ≥ h do not exceed c v; Rn+ pW l . p ˚ h (Rn ), it follows that Now let |β| ≤ h − 1. Since v ∈ W l (Rn ) ∩ W p

+

p

+

η h−|β| (Dβ v)(ζ) belongs to Wpl−h (Rn+ ) and its norm is dominated by c v; Rn+ Wpl . According to Lemma 14.4.2, η |sij |−1 (Dsij κi )(λ(ζ)) is a multiplier in Wpl−h (Rn+ ). Hence η |β|−h

(Dsij κi )(λ(ζ))

is a multiplier in the same space.

496

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

14.4.4 The Space Wp−k for a Special Lipschitz Domain We assume G to be a special Lipschitz domain. In other words, we put l = h in the conjectures of the preceding subsection. Let us retain the notation of Sect. 14.4.3. ˚pk (G), where We introduce the space Wp−k (G) of linear functionals on W   p + p = pp , k = 0, 1, . . . . One can immediately check the continuity of the operator Dα : Wps (G) → s−|α|

(G) for any s = 0, ±1, . . . . By Lemmas 14.4.4 and 14.4.5, the mapping λ performs an isomorphism ˚ k
(Rn ). Therefore, λ maps W −k (G) onto W −k (Rn ) ˚ k
(G) and W between W + p p + p p isomorphically. The following assertion, which is proved in a standard way, gives one of possible realizations of Wp−k (G). Wp

˚ k
(G) can be identified with Proposition 14.4.1. Any linear functional on W p ∞  a distribution f ∈ (C0 (G)) of the form  f (z) = Dα fα (z) , (14.4.8) |α|≤k

where fα is a function such that η k−|α| fα ∈ Lp (G). The norm of this functional is equivalent to the norm +
 1/2 + + + f  = inf + η 2(k−|α|) fα2 ; G+ |α|≤k

Lp

,

the infimum being taken over all collections {fα }|α|≤k in (14.4.8). ˚ k
(G) can be supplied with the norm Proof. By Lemma 14.4.1, the space W p +
 1/2 + + + η 2(k−|α|) (Dα u)2 ; G+ + |α|≤k

Lp


.

˚ k
(G), and Therefore the right-hand side of (14.4.8) is a linear functional on W p +
 1/2 + + + f  ≤ + η 2(k−|α|) fα2 ; G+ |α|≤k

Lp

.

˚ k
in the form (14.4.8), we To express an arbitrary linear functional on W p consider the space Lp
(G) of vectors v = {vα }|α|≤k with components in Lp
(G) endowed with the norm

14.4 Auxiliary Assertions

(



497

vα2 )1/2 ; GLp
.

|α|≤k

Further, let

Λk = {(−1)|α| η |α|−k Dα }|α|≤k .

˚ k
(G) is complete, the range of the operator Λk : W ˚ k
(G) → Since the space W p p ˚ k
(G) and let f (u) be the Lp
(G) is a closed subspace of Lp
(G). Let u ∈ W p value of the functional f ∈ Wp−k (G) on u. We define the functional Φ by Φ(v) = f (u) on the set of vectors v which can be expressed in the form Λk u. Then Φ = f  and, by the Hahn-Banach theorem, Φ can be extended to a linear functional on Lp
(G) with the same norm. Consequently,   Φ(Λk u) = gα (−1)|α| η |α|−k Dα u dz , |α|≤k

G

where gα ∈ Lp (G). It remains to put fα = η |α|−k gα .



˚ t+k (G), where t < 0, k is a nonnegative integer, Lemma 14.4.6. Let v ∈ W p and t + k ≥ 0. Then (14.4.1) holds. Proof. We have η −k v; GWpt = sup

˚ −t w∈W p


(η −k v, w) . wW −t
p

By Lemma 14.4.1, . (η −k v, w) ≤ η −k−t v; GLp η t wLp
≤ c v; GWpt+k wW −t
p

 Corollary 14.4.3. Let η k v ∈ Wpt+k (G), where t and k are the same numbers as in Lemma 14.4.6. Then v; GWpt ≤ c η k v; GWpt+k . ˚ −t Proof. Let w ∈ W p
. We have (v, w) ≤ η k v; GWpt+k η −k w; GW −t−k .
p

Using Lemma 14.4.6 with p and t replaced by p and −t − k, we obtain ≤ c w; GW −t . η −k w; GW −t−k

p

The result follows.

p



498

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

14.4.5 Auxiliary Assertions on Differential Operators in Divergence Form l+1−h−1/p

Lemma 14.4.7. Let ∂G ∈ Mp , l ≥ h, and let  Pu = (−1)|α| Dα (aαβ (z)Dβ u) .

(14.4.9)

|α|,|β|≤h

If η h−|β| aαβ ∈ M (Wpl−h (G) → Wpl−2h+|α| (G))

for l − 2h + |α| ≥ 0

and η 2h−|α|−|β| aαβ ∈ M Wpl−h (G) for l − 2h + |α| < 0, then P is a continuous operator: ˚ph )(G) → Wpl−2h (G) (Wpl ∩ W and its norm does not exceed c A, where 
 η h−|β| aαβ ; GM (Wpl−h →Wpl−2h+|α| ) A= |β|≤h

+

|α|≥2h−l

 η 2h−|α|−|β| aαβ ; GM Wpl−h .



|α|0

it follows that S(ζ, Dζ ) =



(−1)|γ| Dζγ (bγδ (ζ)Dζδ ),

|γ|,|δ|≤h

where bγδ (ζ) =

 µ≥γ>0


(−1)|µ|−|γ| cµγ Dζµ−γ

 1 det ∂λ(ζ)fµδ (ζ) . det ∂λ(ζ)

500

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

In particular, if |γ| = |δ| = h, we have  bγδ (ζ) = aαβ Pαβγδ (∂κ ◦ λ) , |α|=|β|=h

where Pαβγδ is a polynomial of elements of the matrix ∂κ. Also Pαβγδ (I) = 1 if α = γ, β = δ, and Pαβγδ (I) = 0 if α = γ or β = δ. Let  (−1)h Dζγ (bγδ (ζ)Dζδ ) , S1 = S − S 0 . S0 (ζ, Dζ ) = |γ|=|δ|=h

It is clear that (S0 − R)v; Rn+ Wpl−2h ≤



(Pαβγδ (∂κ ◦ λ) − Pαβγδ (I))Dζδ v; Rn+ Wpl−h

α,β,γ,δ

≤ c I − ∂λ; Rn+ M Wpl−h v; Rn+ Wpl ≤ c ∇ϕ; Rn−1 M Wpl+1−h−1/p v; Rn+ Wpl . Next we derive an analogous estimate for the norm (S − S0 )v; Rn+ Wpl−2h . According to Lemma 14.4.7, is suffices to prove the two inequalities η h−|δ| bγδ ; Rn+ M (Wpl−h →Wpl−2h+|γ| ) ≤ c ∇ϕ; Rn−1 M Wpl+1−h−1/p

(14.4.10)

for |γ| ≤ h, |δ| ≤ h, |γ| + |δ| < 2h, l − 2h + |γ| ≥ 0, and η 2h−|γ|−|δ| bγδ ; Rn+ M Wpl−h ≤ c ∇ϕ; Rn−1 M Wpl+1−h−1/p .

(14.4.11)

for |γ| ≤ h, |δ| ≤ h, |γ| + |δ| < 2h, l − 2h + |γ| < 0. By Corollary 14.4.2, Dζµ−γ (1/ det ∂λ(ζ)) ∈ M (Wpl−h (Rn+ ) → Wpl−h−|µ|+|γ| (Rn+ ))

(14.4.12)

and, for µ > γ, Dζµ−γ (1/ det ∂λ(ζ)); Rn+ M (Wpl−h →Wpl−h−|µ|+|γ| ) ≤ c ∇ϕ; Rn−1 M Wpl+1−h−1/p .

(14.4.13)

We show that η h−|δ| fµδ ∈ M (Wpl−h−|µ|+|γ| (Rn+ ) → Wpl−2h+|γ| (Rn+ )) .

(14.4.14)

14.4 Auxiliary Assertions

501

Applying Corollary 14.4.2 once more, we obtain η h−|δ| (Dtij κi ) ◦ λ ∈ M Wpl−h (Rn+ ) , i,j

(Dsij κi ) ◦ λ ∈ M (Wpl−h (Rn+ ) → Wpl−2h+|µ| (Rn+ )) i,j

⊂ M (Wpl−h−|µ|+|γ| (Rn+ ) → Wpl−2h+|γ| (Rn+ ))

(14.4.15)

and therefore the inclusion (14.4.14) holds. Since, for µ = γ, at least one of the exponents tij and sij is greater than 1, it follows that η h−|δ| fγδ ; Rn+ M (Wpl−h →Wpl−2h+|γ| ) ≤ c ∇ϕ; Rn−1 M Wpl+1−h−1/p . (14.4.16) Now (14.4.10) results directly from (14.4.12)–(14.4.14), and (14.4.16). Next we turn to the proof of (14.4.11). By virtue of Corollary 14.4.2, the inclusion (14.4.15) holds. Moreover, η |µ|−|γ| Dζµ−γ (1/ det ∂λ(ζ)) ∈ M Wpl−h (Rn+ ) , and η h−|µ|

(Dsij κi ) ◦ λ ∈ M Wpl−h (Rn+ ) . i,j

To obtain (14.4.11), it remains to note that we always have either µ > γ or one of the exponents tij , sij is greater than one, and then to apply Corollary 14.4.2 once more.  With minor modifications in the above proof and using the properties of (p, l)-diffeomorphisms given in Sect. 9.4.2, we arrive at the following local variant of Lemma 14.4.8. ˚ h )(Rn ), l ≥ h, with supports in Br ∩Rn , Lemma 14.4.9. For all v ∈ (Wpl ∩ W p + + the inequality (S − R)v; Rn+ Wpl−2h ≤ c |||I − ∂λ; Br ∩ Rn+ |||M Wpl−h v; Rn+ Wpl

(14.4.17)

holds. For p(l − h) > n it follows from (9.6.4) that (14.4.17) is equivalent to (S − R)v; Rn+ Wpl−2h ≤ c rl−h−n/p |||I − ∂λ; Br ∩ Rn+ |||Wpl−h v; Rn+ Wpl .

(14.4.18)

502

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

14.5 Solvability of the Dirichlet Problem in Wpl (Ω) 14.5.1 Generalized Formulation of the Dirichlet Problem Let Ω be open subset of Rn and let P be the operator (14.4.9), where aαβ ∈ ¯ l ≥ h. Further, let the G˚ arding inequality C l−h (Ω),   Re aαβ Dα uDβ u dz ≥ c u; Ω2W h (14.5.1) Ω |α|=|β|=h

2

hold for u ∈ C0∞ (Ω). We say that u ∈ Wpl (Ω) is a solution of the generalized Dirichlet problem in Wpl (Ω) if ˚ h (Ω) , Pu = f , u − g ∈ Wpl (Ω) ∩ W (14.5.2) p where f and g are given functions in the spaces Wpl−2h (Ω) and Wpl (Ω) respectively. By Wp−k (Ω), k = 1, 2, . . . , we mean the space of linear continuous func˚ k
(Ω). tionals in W p 14.5.2 A Priori Estimate for Solutions of the Generalized Dirichlet Problem Following the proof of Theorem 14.3.1 and using Lemma 14.4.8 in place of Lemma 14.2.1, we arrive at: Theorem 14.5.1. If p(l − h) ≤ n, 1 < p < ∞, and if ∂Ω belongs to the class l+1−h−1/p (δ), then Mp u; ΩWpl ≤ c (P u; ΩWpl−2h + u; ΩL1 )

(14.5.3)

˚ph )(Ω). for all u ∈ (Wpl ∩ W Duplicating the proof of Theorem 14.3.2 and using (14.4.18) instead of (14.2.6), we obtain: l+1−h−1/p

Theorem 14.5.2. If p(l − h) > n, 1 < p < ∞, and ∂Ω ∈ Wp Theorem 14.5.1 holds.

, then

Next we state two corollaries of (14.5.3) which are analogous to Propositions 14.3.1 and 14.3.2. Proposition 14.5.1. Let Ω satisfy the conditions of either Theorem 14.5.1 or Theorem 14.5.2. (i) If the kernel of the operator ˚ph )(Ω) → Wpl−2h (Ω) P : (Wpl ∩ W

(14.5.4)

is trivial, then the norm u; ΩL1 in (14.5.3) can be omitted. (ii) The kernel of the operator (14.5.4) is finite-dimensional and the range of this operator is closed.

14.5 Solvability of the Dirichlet Problem in Wpl (Ω)

503

Proposition 14.5.2. Let Ω satisfy the conditions of either Theorem 14.5.1 or Theorem 14.5.2. Further, let U and V be open bounded subsets of Rn with ¯ ⊂ V , and let u ∈ (Wpl ∩ W ˚ph )(Ω). Then U u; U ∩ ΩWpl ≤ c (P u; V ∩ ΩWpl−2h + u; V ∩ ΩL1 ) . 14.5.3 Solvability of the Generalized Dirichlet Problem Let the G˚ arding inequality (14.5.1) hold for all u ∈ C0∞ (Ω). Then, as is well ˚ h (Ω). known, the equation P u = f with f ∈ W2−h (Ω) is uniquely solvable in W 2 l+1−h−1/p

Theorem 14.5.3. Let ∂Ω ∈ Mp for p(l − h) ≤ n and let ∂Ω belong l+1−h−1/p to the class Wp for p(l − h) > n. (i) If f ∈ Wpl−2h (Ω) ∩ W2−h (Ω) and g ∈ Wpl (Ω) ∩ W2h (Ω), 1 < p < ∞, ˚ h (Ω), then u ∈ Wpl (Ω) and if u ∈ W2h (Ω) is such that P u = f and u − g ∈ W 2 h ˚ and u − g ∈ Wp (Ω). (ii) The problem (14.5.2) has one and only one solution u ∈ Wpl (Ω). Proof. It is sufficient to assume that g = 0. (i) First let p(l − h) ≤ n. We put ϕε (x) = ε + Φ(x, ε), where Φ is an extension of ϕ defined by Φ = T ϕ (cf. (9.4.11)). We introduce the domain Gε = {z = (x, y) : x ∈ Rn−1 , y > ϕε (x)}. Since 1 + ∂Φ/∂η > 0, it follows that {Gε } is an increasing family, Gε ⊂ G, and Gε → G as ε → +0. By part (i) of Theorem 8.7.2 we have ∇ϕε ; Rn−1 M Wpl+1−h−1/p ≤ c ∇Φ; Rn+ M Wpl−h+1 . This inequality and (9.4.12) imply that ∇ϕε ; Rn−1 M Wpl+1−h−1/p ≤ c ∇ϕ; Rn−1 M Wpl+1−h−1/p ,

(14.5.5)

¯ \Gε ) and let uε be a with a constant c independent of ε. Let Ωε = Ω\(U ˚ h (Ωε ). It is known (see, for instance, Neˇcas [Ne], 6.6, solution of P u = f in W 2 ˚ h (Ω). By U1 we denote an open set such that U1 ⊂ U . Ch. 3), that uε → u in W 2 Since ϕε ∈ C ∞ (Rn−1 ), we have uε ∈ Wpl (U1 ∩ Ωε ) by the known theorem on the regularity of weak solutions of elliptic boundary value problems near a smooth part of a boundary. This and Proposition 14.5.2 give the estimate uε ; U2 ∩ Ωε Wpl ≤ c (f ; U1 ∩ Ωε Wpl−2h + uε ; U1 ∩ Ωε L1 ) , where U2 is an open set, U2 ⊂ U1 , and c does not depend on ε. Hence the left-hand side is uniformly bounded with respect to ε. Now, if we fix a domain ω such that ω ¯ ⊂ Ω, then the upper limit lim sup uε ; ωWpl ε→+0

504

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

is majorized by a constant independent of ω. From {uε } we select a sequence which is weakly convergent in Wpl (ω). This sequence converges in W2h (ω), and hence its weak limit in Wpl (ω) coincides with u. Therefore, u ∈ Wpl (ω), and the Wpl (ω)-norm of u is uniformly bounded with respect to ω. Thus u ∈ Wpl (Ω). ˚ h (Ω) and W ˚ph (Ω) for Lipschitz domains The identity of the spaces Wph (Ω) ∩ W 2 Ω is known. The case p(l − 1) > n can be treated in the same way, with (14.5.5) replaced by ϕε ; Rn−1 Wpl+1−h−1/p ≤ c ϕ; Rn−1 Wpl+1−h−1/p . (ii) For p ≥ 2, the assertion follows from the unique solvability of the ˚ h (Ω), together with the first part of the theorem. problem in W 2 Consider the case p < 2. By P t we denote the operator formally conjugate ¯ and G˚ arding’s inequality to P . The coefficients of P t belong to C l−2h (Ω), 1−1/p t (14.5.1) holds for P too. Recall that the property ∂Ω ∈ Mp implies that the Lipschitz constants of the functions ϕ which locally specify ∂Ω are small. Hence the equation ˚ph
(Ω) with v ∈ W P t v = F ∈ Wp−h
(Ω) ˚ h
(Ω). Let u be a solution of the homogeneous probis uniquely solvable in W p lem (14.5.2) and let {vm }m≥1 be a sequence of functions in C0∞ (Ω), which ˚ h
(Ω). Then converges to v in W p 0 = lim



(aαβ Dα u, Dβ vm ) =

|α|,|β|≤h



(aαβ Dα u, Dβ v) = (u, F ) ,

|α|,|β|≤h

and the uniqueness property of problem (14.5.2) follows. Let ¯ m = 1, 2, . . . , fm → f in Wpl−2h (Ω). fm ∈ C l (Ω), ˚ h (Ω) satisfying P um = fm . According to the By um we denote a function in W 2 ˚ph (Ω). By part (i) of Proposition first part of the theorem, um ∈ Wpl (Ω) ∩ W 14.5.1, um − uk ; ΩWpl ≤ c fm − fk ; ΩWpl−2h . ˚ph (Ω) and its limit satisfies P u = f . Thus {um } converges in Wpl (Ω) ∩ W



14.5.4 The Dirichlet Problem Formulated in Terms of Traces The first boundary value problem (14.5.2) is not a particular case of the general boundary value problem formulated in Sect. 14.3.1. In the present subsection we study the Dirichlet problem in another formulation which is analogous to that considered in 14.3.1

14.5 Solvability of the Dirichlet Problem in Wpl (Ω)

505

¯ Let P be the elliptic operator (14.4.9) with coefficients aαβ in C l−h (Ω), l ≥ h, for which the inequality (14.5.1) holds. We assume ∂Ω to be in the class C 0,1 . ¯ and a We introduce a sufficiently small finite open covering {U } of Ω corresponding partition of unity {ζU }. Let if U ∩ ∂Ω = ∅

PjU = ∂ j−1 /∂y j−1 , j = 1, . . . , h, and PjU = 0

if U ∩ ∂Ω = ∅.

The Dirichlet boundary conditions will be prescribed by the operators  Pj = ζU PjU . U

We give a formulation of the Dirichlet problem. Let us look for a function u ∈ Wpl (Ω) such that Pu = f

in

Ω,

trPj u = fj

on

∂Ω,

j = 1, . . . , h ,

(14.5.6)

l+1−j−1/p

(∂Ω) respectively. where f and fj are functions in Wpl−2h (Ω) and Wp It is clear that any solution of the problem (14.5.2) is a solution of (14.5.6) with fj = trPj g. The following lemma shows that an opposite statement holds l−1/p if ∂Ω ∈ Mp . Lemma 14.5.1. Let G = {z = (x, y) : x ∈ Rn−1 , y > ϕ(x)} be a dol−1/p main with ∂G in the class Mp and let f1 , . . . , fh be arbitrary functions in l−1/p l+1−h−1/p Wp (∂G), . . . , Wp (∂G). Then there exists a function g ∈ Wpl (G) such that tr(∂ j−1 g/∂y j−1 ) = fj , j = 1, . . . , h. Proof. We use the notation λ, κ, and T ϕ introduced in Sect. 9.4.3. We have [(∂/∂y)j−1 g] ◦ λ = [(N + ∂(T ϕ)/∂η)−1 (∂/∂η)]j−1 (g ◦ λ) . Since ∇(T ϕ) ∈ M Wpl (Rn+ ), it follows that [(∂/∂y)j−1 g] ◦ λ =

j 

aνj (∂/∂η)ν−1 (g ◦ λ) ,

j = 1, . . . , h ,

(14.5.7)

ν=1

where aνj ∈ M (Wpl−ν+1 (Rn+ ) → Wpl−j+1 (Rn+ )),

ajj = (N + ∂(T ϕ)/∂η)1−j .

We note that (14.5.7) is a triangular algebraic system with respect to (∂/∂η)ν−1 (u ◦ λ). Hence

506

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

(∂/∂η)ν−1 (g ◦ λ) =

ν 

bjν [(∂/∂y)j−1 g] ◦ λ ,

ν = 1, . . . , h ,

j

where bjν ∈ M (Wpl−j+1 (Rn+ ) → Wpl−ν+1 (Rn+ )). Taking into account that tr bjν ∈ M (Wpl+1−j−1/p (Rn−1 ) → Wpl+1−ν−1/p (Rn−1 )), we obtain (tr bjν )fj ◦ λ ∈ Wpl+1−ν−1/p (Rn−1 ). Therefore, there exists a function H ∈ Wpl (Rn+ ) such that tr (∂/∂η)ν−1 H =

j 

(trbjν )fj ◦ λ .

ν=1

Setting g = H ◦ κ, we complete the proof.



Since both formulations (14.5.2) and (14.5.6) of the Dirichlet problem l−1/p , then, from are equivalent for domains with boundary in the class Mp Theorem 14.5.3, we obtain: Theorem 14.5.4. Let any of the following conditions hold: l−1/p (α) h = 1, p(l − 1) ≤ n; ∂Ω ∈ Mp (δ); l−1/p (β) h = 1, p(l − 1) > n; ∂Ω ∈ Wp ; l−1/p (γ) h > 1, ∂Ω ∈ Mp and ∂Ω is locally defined by equations of the form y = ϕ(x), where ϕ is a function with a small Lipschitz constant (for l−1/p p(l − 1) > n, this is equivalent to ∂Ω ∈ Wp ). Then the operator {P ; trPj } :

Wpl (Ω)

→

Wpl−2h (Ω)

×

h -

Wpl+1−j−1/p (∂Ω)

j=1

is an isomorphism. Proof. For h = 1 the result follows from Theorem 14.5.3. Let h > 1. According to (2.3.8), n−1 1−α ∇ϕ; Rn−1 M Wpl−h−1/p ≤ c ∇ϕ; Rn−1 α L∞ l−1−1/p ∇ϕ; R MW p

with α = (p(l − h) − 1)/(p(l − 1) − 1). Consequently, it follows from (γ) that ∂Ω ∈ Mpl+1−h−1/p (δ)

if p(l − h) ≤ n

and ∂Ω ∈ Wpl+1−h−1/p By Theorem 14.5.3 the proof is complete.

if p(l − h) > n. 

14.6 Necessity of Assumptions on the Domain

507

Thus, by changing the formulation of the Dirichlet problem, we have obtained its solvability in Wpl (Ω) under stricter assumptions on Ω (compare the last theorem with Theorem 14.5.3). The exception is the second-order operator P , i.e., h = 1, when admissible classes of domains coincide for both formulations. In the following section we discuss the necessity of the conditions in the last theorem.

14.6 Necessity of Assumptions on the Domain 3/2

14.6.1 A Domain Whose Boundary is in M2 3/2 Belong to M2 (δ)

∩ C 1 but does not

In this subsection we give an example which shows that for p(l − 1) ≤ n and l−1/p for h = 1 the condition ∂Ω ∈ Mp (δ) in part (α) of Theorem 14.5.4 cannot l−1/p be replaced by the assumption that ∂Ω belongs to the class Mp ∩ C l−1 . 3/2 To be precise, we construct a domain Ω with ∂Ω ∈ M2 ∩ C 1 for which the problem −∆u = f in Ω , tr u = 0 on ∂Ω (14.6.1) is not solvable in W22 (Ω) for all f ∈ L2 (Ω). This means that the smallness of 3/2 ∇ϕ; Rn−1 M W 1/2 in the definition of M2 (δ) is essential for the solvability 2

of the problem (14.6.1) in W22 (Ω). Let the domain Ω be specified in a neighborhood of O by the inequality y > Cϕ(x), where C is a positive constant and ϕ(x) = η(x, 0)|x1 |/ log(1/|x1 |). Here and henceforth η is a function in C0∞ (B1/2 ) with η = 1 on the ball B1/4 . We introduce the domain ω = {ξ : x1 + ix2 : |ξ| < 1/2, x2 > C|x1 |/ log(1/|x1 |)} and we denote by ζ(ξ) the conformal mapping of ω onto the half-disc {ζ : Im ζ > 0, |ζ| < 1} with the fixed point ξ = 0. Let ξ = iρ exp(iθ) and let ω be given in polar coordinates (ρ, θ) as ω = {ξ : ρ < 1/2, |θ| < π/2 + ϕ(ρ)}. It is easily checked that ϕ(ρ) = C(log 1/ρ)−1 + O((log 1/ρ)−3 ) . According to an asymptotic formula due to Warschawski [Wa],

508

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems



Im ζ(ξ) = c exp − π

 
dr πθ + o(1) cos r(π + 2ϕ(r)) π + 2ϕ(ρ) ρ
 πθ = c ρ(log 1/ρ)2C/π cos + o(1) as ρ → +0. π + 2ϕ(ρ) 1/2

It is clear that, for C ≥ π/4,  (Im ζ(ξ))2 dx1 dx2 = ∞ . 4 2 ω ρ (log ρ)

(14.6.2)

Next we need the following assertion. ˚ 1 (ω), then Lemma 14.6.1. If h ∈ W22 (ω) ∩ W 2  2 h dx1 dx2 ≤ c h; ω2W 2 . 4 (log ρ)2 2 ρ ω

(14.6.3)

Proof. First we show that, for any g ∈ W21 (ω),  dx1 dx2 g2 2 ≤ c g; ω2W 1 . 2 ρ (log ρ)2 ω

(14.6.4)

Clearly, to prove (14.6.4) it suffices to assume that ω is the half-disc {ξ : ρ < 1/2, |θ| < π/2}. After integration of the Hardy inequality 

1/2

dρ g ≤4 ρ(log ρ)2



1/2

2

0


∂g 2

0

∂ρ

ρ dρ

with respect to the variable θ, we arrive at (14.6.4). Putting g = ∂h/∂x1 and g = ∂h/∂x2 into (14.6.4), we obtain  dx1 dx2 (∇h)2 2 ≤ c h; ω2W 2 . 2 2 ρ (log ρ) ω Let Cρ = {ξ : |ξ| = ρ}. Since h = 0 on ∂ω, it follows that for almost all ρ>0    h2 dθ ≤ c (∂h/∂θ)2 dθ ≤ c ρ2 (∇h)2 dθ , ω∩Cρ

which leads to

ω∩Cρ



dx1 dx2 h 4 ≤c ρ (log ρ)2 ω

ω∩Cρ



2

(∇h)2 ω

dx1 dx2 . ρ2 (log ρ)2 

14.6 Necessity of Assumptions on the Domain

509

By (14.6.2) and (14.6.3), we see that the function u defined on Ω by u(z) = η(2z)Im ζ(x1 + ix2 ) ˚ 1 (Ω) and satisfies does not belong to W22 (Ω). On the other hand, u is in W 2 the equation −∆u = f, f ∈ L2 (Ω). ˚ 2 (Ω) if Consequently, the boundary value problem (14.6.1) is solvable in W 2 and only if C < π/4. 3/2 Obviously, ∂Ω is in the class C 1 . We show that ∂Ω ∈ M2 , i.e. ∇ϕ ∈ 1/2 M W2 (Rn−1 ). With this aim in view, we verify that the gradient of the function ψ defined as ψ(z) = η(z)r log r ,

where r = (x21 + y 2 )1/2 ,

belongs to the space M W21 (Rn ) (cf. Theorem 8.7.1). Clearly, ∇ψ ∈ L∞ (Rn ) and it remains to prove that ∇2 ψ ∈ M (W21 (Rn ) → L2 (Rn )). In fact, for all u ∈ W21 (Rn ),  u∇2 ψ; Rn 2L2

≤c

B1/2



u 2 dz ≤ c u; B1 2Wp1 . r log r

3/2

Thus, ∂Ω ∈ M2 . 14.6.2 Necessary Conditions for Solvability of the Dirichlet Problem The next assertion, which follows directly from the Implicit Function Theorem l−1/p 9.5.2, shows that the condition ∂Ω ∈ Wp with p(l −1) > n is necessary for the solvability of the problem (14.5.6) in Wpl (Ω) for an operator P of higher than the second order. Theorem 14.6.1. Let Ω be a bounded Lipschitz domain and let l be an integer, l ≥ 2h, p(l − 1) > n, 1 < p < ∞, and h > 1. If there exists a solution u ∈ Wpl (Ω) of the problem P u = 0 in Ω, tr u = 0, l−1/p

then ∂Ω ∈ Wp

tr P2 u = 1, tr Pj u = 0, j = 3, . . . , h ,

(14.6.5)

.

Under the additional assumption ∂Ω ∈ C l−2,1 , we can prove the necessity l−1/p of the inclusion ∂Ω ∈ Wp for p(l − 1) ≤ n.

510

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

Theorem 14.6.2. Let ∂Ω be in the class C l−2,1 and let l be an integer, l ≥ 2h, p(l − 1) ≤ n, 1 < p < ∞, and h > 1. If there exists a solution u ∈ Wpl (Ω) of l−1/p

problem (14.6.5), then ∂Ω ∈ Wp

.

Proof. We use the same notation as in the formulation of Theorem 9.5.2. Since l−1 ¯ and ϕ ∈ C l−2,1 (Rn−1 ), (U ∩ G) ∇x u, uy ∈ Wp.loc

it follows that ∇x u ◦ λ,

l−1 ¯ uy ◦ λ ∈ Wp.loc (κ(U ∩ G)). l−1−1/p

(ω). Now we note Therefore, tr(∇x u ◦ λ) and tr(uy ◦ λ) belong to Wp,loc h that ∂Ω is in the class C and so, by the known coercive estimate for solutions of the elliptic boundary value problem in the variational form (cf. [ADN1], ¯ Sect. 15), we have u ∈ Wqh (Ω) for any q < ∞. In particular, ∇u ∈ C(Ω). l−1−1/p

Since the space (Wp,loc function

∩ L∞ )(ω) is a multiplication algebra, the vector

∇ϕ = tr (∇x u ◦ λ)/tr (uy ◦ λ) l−1−1/p

belongs to Wp,loc



(ω).

We consider the case of the second-order operator P . Theorem 14.6.3. Let l be an integer, l ≥ 2, 1 < p < ∞, h = 1, and P (1) ≤ 0. Let Ω be a domain with ∂Ω ∈ C 1 and let the normal to ∂Ω satisfy the Dini condition. If, for a nonpositive function f ∈ C0∞ (Ω), there exists a solution u ∈ Wpl (Ω) of the problem Pu = f l−1/p

then ∂Ω ∈ Wp

in

Ω,

tr u = 0 ,

(14.6.6)

.

Proof. It is sufficient to note that the interior normal derivative at any point of ∂Ω is positive by Giraud’s theorem (see, for example, [Mir], Sect. 3.5) and then to duplicate the proof of Theorem 14.6.2. 

14.6.3 Boundaries of the Class Mpl−1/p (δ) Let Ω be a bounded Lipschitz domain. We denote by O an arbitrary point of ∂Ω and introduce a neighborhood U of O such that Ω ∩ U = G ∩ U with G = {(x, y) : x ∈ Rn−1 , y > ϕ(x)}. l−1−1/p By Theorem 4.1.1, the norm of ∇ϕ in M Wp (Rn−1 ) is equivalent to sup e⊂Rn−1

Dp,l−1/p ϕ; eLp + ∇ϕ; Rn−1 L∞ . [Cp,l−1−1/p (e)]1/p

(14.6.7)

14.6 Necessity of Assumptions on the Domain

511

(Here we can restrict ourselves to compact sets e with d(e) ≤ 1.) Thus the l−1/p definition of the class Mp (δ) means that the sum (14.6.7) is sufficiently small. The following assertion, whose proof is postponed to Sect. 14.7, gives a l−1/p (δ). local characterization of the class Mp l−1/p

(δ) has the following Theorem 14.6.4. Let p(l − 1) ≤ n. The class Mp equivalent description. For any point O ∈ ∂Ω there exists a neighborhood U and a special Lipschitz domain G = {z = (x, y) : x ∈ Rn−1 , y > ϕ(x)} such that U ∩ Ω = U ∩ G and
 Dl−1/p (ϕ; Bε ); eLp (14.6.8) + ∇ϕ; Bε L∞ ≤ c δ . lim sup 1/p ε→0 e⊂Bε [Cp,l−1−1/p (e)] Here Bε is the ball with centre at O and radius ε, c is a constant which depends only on l, p, n, and
 1/p |∇j−1 ϕ(x) − ∇j−1 ϕ(y)|p |x − y|−n+2−p dy . Dj−1/p (ϕ; Bε )(x) = Bε

Theorem 14.6.4 and properties (v), (vi) of the capacity formulated in Sect. 9.6.1 lead to: Corollary 14.6.1. (i) If n > p(l − 1) and
lim

ε→0

sup e⊂Bε

 Dl−1/p (ϕ; Bε ); eLp < cδ , + ∇ϕ; B  ε L ∞ (mesn−1 e)[n−p(l−1)]/(n−1)p

l−1/p

then ∂Ω ∈ Mp (δ). (ii) If n = p(l − 1) and
 lim sup Dl−1/p (ϕ; Bε ); eLp | log(mesn−1 e)|(p−1)/p + ∇ϕ; Bε L∞ < c δ , ε→0

e⊂Bε

l−1/p

then ∂Ω ∈ Mp

(δ). l−1/p

Now we derive another test for the inclusion into Mp m Besov space Bq,p (cf. Corollary 14.6.2 below).

(δ) involving the l−1/p

We say that the boundary of the Lipschitz domain Ω belongs to Bq,p (l = 1, 2, . . . , ) if, for any point of ∂Ω, there exists a neighborhood in which ∂Ω is specified in Cartesian coordinates by a function ϕ satisfying 
 p/q |∇l−1 ϕ(x + h) − ∇l−1 ϕ(x)|q dx |h|2−n−p dh < ∞ . Rn−1

Rn−1

512

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

Corollary 14.6.2. Let p(l − 1) ≤ n and let Ω be a bounded Lipschitz domain l−1/p with ∂Ω ∈ Bq,p where q ∈ [p(n − 1)/(p(l − 1) − 1), ∞]

if p(l − 1) < n

and q ∈ (p, ∞]

if p(l − 1) = n.

Further, let ∂Ω be locally defined in Cartesian coordinates by y = ϕ(x), where l−1/p ϕ is a function with a Lipschitz constant less than c δ. Then ∂Ω ∈ Mp (δ). Proof. We have  Dl−1/p (ϕ; Bε ); epLp

≤

Bε

−n+2−p

|h|

 dh 

|∇l−1 ϕ(x + h) − ∇l−1 ϕ(x)|p dx e

|h|−n+2+p dh ≤ (mesn−1 e)1−p/q Bε
 p/q × |∇l−1 ϕ(x + h) − ∇l−1 ϕ(x)|q dx . Bε

Then the result follows from part (i) of Corollary 14.6.1.



Corollary 14.6.2 can be made sharper in the case p(l − 1) = n, by virtue of part (ii) of Corollary 14.6.1, if one uses the Orlicz space Ltp (log+ t)p−1 instead of Lq , but we shall not go into this. Setting q = ∞ in Corollary 14.6.2, we obtain a simple sufficient condil−1/p (δ) formulated in terms of the modulus of tion for the inclusion into Mp continuity ωl−1 (t) of ∇l−1 ϕ: 
ωl−1 (t) p dt < ∞ . (14.6.9) t 0 l−1/p

l−1/p

l−1/p

Since B∞,p ⊂ Wp , it follows that (14.6.9) is sufficient for ∂Ω ∈ Wp . We show that (14.6.9) is in a sense a sharp condition for solvability of the Dirichlet problems (14.6.5) and (14.6.6) in Wpl (Ω). We have shown in 4.4.3 that, for any increasing function ω ∈ C[0, 1] satisfying the inequalities (4.4.15) as well as the condition  1
ω(t) p dt = ∞, t 0 one can construct a function ϕ on Rn−1 such that (i) the continuity modulus of ∇l−1 ϕ does not exceed c ω with c = const; (ii) supp ϕ ⊂ Q2π , where Qd = {x ∈ Rn−1 : |xi | < d}; l−1/p (iii) ϕ ∈ / Wp (Rn−1 ).

l−1/p

14.7 Local Characterization of Mp

(δ)

513

By Ω we denote a bounded domain in Rn such that Ω ∩ {z : x ∈ Q3π , |y| < 1} = {z : x ∈ Q3π , ϕ(x) < y < 1} . Further, we assume that ∂Ω is a surface of the class C ∞ in the exterior of the set {z : x ∈ Q2π , y = ϕ(x)}. By Theorems 14.6.1–14.6.3 the problems (14.6.5), (14.6.6) for this Ω have no solutions in Wpl (Ω). Remark 14.6.1. Suppose that, for any point O ∈ ∂Ω, there exists a neighborhood U such that U ∩ Ω is C l -diffeomorphic to the domain {(x, y) : y > ϕ(x1 , . . . , xn−s )},

2 ≤ s ≤ n − 1,

i.e. ‘the dimensions of singularities of ∂Ω are not less than s − 1’. Then all l−1/p properties of domains with boundaries in Mp (δ) remain valid after the change from n − 1 to n − s. This follows from the definition of the class l−1/p (δ), from Theorem 8.7.1 and from the fact that, if ψ is defined in Rn Mp and ψ depends on n − s + 1 variables only, the norms ψ; Rn M Wpl ,

ψ; Rn−s+1 M Wpl

are equivalent (see Proposition 2.7.2).

14.7 Local Characterization of Mpl−1/p(δ) In this section we prove Theorem 14.6.4 which gives a local characterization l−1/p (δ). By B we mean the ball in Rn−1 of radius of surfaces in the class Mp centered at the origin. 14.7.1 Estimates for a Cutoff Function Let η be an even function in C0∞ (−1, 1) with η = 1 on (−1/2, 1/2). For z ∈ Rn we define ⎧ η(|z|/ ), if p(l − 1) < n, ⎨ η (z) = ⎩ η(log / log |z|), if p(l − 1) = n. Clearly, supp η ⊂ Bε and ⎧ c −j , if p(l − 1) < n, ⎨ |∇j η (z)| ≤ ⎩ c | log |z||−1 |z|−j , if p(l − 1) = n.

514

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

Lemma 14.7.1. Let p(l − 1) = n. Then  |∇j η (x) − ∇j η (y)|p dy ≤ cj | log |x||−p |x|1−p−pj , n−2+p |x − y| B

(14.7.1)

where x ∈ B , j = 0, 1, .... Proof. Since −|α|

D η (x) = |x| α

|α| 

σk (log / log |x|)(log )−k ,

k=1

where σk ∈ C0∞ (−1, 1), it follows that ⎧ cj | log |−1 |x − y||x|−j−1 ⎪ ⎪ ⎪ ⎪ if |x|/2 ≤ |y| ≤ 2|x|, ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ cj | log |−1 (max{|x|, |y|})−j |∇j η (x) − ∇j η (y)| ≤ if j > 0 and |y| < |x|/2 or |x| < |y|/2, ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ cj | log |−1 | log(|x|/|y|)| ⎪ ⎪ ⎩ if j = 0 and |y| < |x|/2 or |x| < |y|/2. These inequalities imply that  |∇j η (x) − ∇j η (y)|p dy ≤ cj | log |−p |x|1−p−pj , |x − y|n−2+p B

(14.7.2)

which is equivalent to (14.7.1) for x ∈ B \B3 . Let x ∈ B3 . We then have   |δ0,j − ∇j η (y)|p |∇j η (x) − ∇j η (y)|p dy ≤ c dy, n−2+p |y| |x − y|n−2+p B \B 2 B where δ0,j = 1 for j = 0 and δ0,j = 0 for j > 0. Consequently,  |δ0,j − ∇j η (y)|p dy ≤ c | log |−p |x|1−p−pj . n−2+p |y| B \B 2 Setting |x| = 3 in the last inequality and observing that t3(p+pj−1) | log t|p increases near t = 0, we obtain (14.7.1).



l−1/p

14.7 Local Characterization of Mp

(δ)

515

14.7.2 Description of Mpl−1/p (δ) Involving a Cutoff Function The aim of this subsection is to prove the following assertion on a local charl−1/p acterization of Mp (δ) involving a cutoff function. Hereafter, without loss of generality, we assume that ϕ(0) = 0. l−1/p

(δ) if and only if Lemma 14.7.2. A surface ∂Ω belongs to the class Mp for any O ∈ ∂Ω there exists a neighborhood U and a special Lipschitz domain G = {z = (x, y) : x ∈ Rn−1 , y > ϕ(x)} such that U ∩ Ω = U ∩ G and lim sup ∇(η ϕ); Rn−1 M Wpl−1−1/p ≤ c δ,

(14.7.3)

→0

where c is a constant which depends on l, p, n, and η is the function introduced in the previous subsection. l−1/p

(δ). In order to obtain the Proof. Clearly, (14.7.3) implies that ∂Ω ∈ Mp converse assertion, it is sufficient to derive the estimate ∇(η ϕ); Rn−1 M Wpl−1−1/p ≤ c ∇ϕ; Rn−1 M Wpl−1−1/p .

(14.7.4)

Let Φ = T ϕ be an extension of ϕ, defined in Sect. 8.7.2. By Theorem 8.7.1, ∇(η ϕ); Rn−1 M Wpl−1−1/p ≤ c ∇(η Φ); Rn+ M Wpl−1 .

(14.7.5)

For any function u ∈ Wpl−1 (Rn+ ), we have u∇(η Φ); Rn+ Wpl−1 ≤ c

l−1


Φ |∇j+1 η | |∇l−1−j u|; Rn+ Lp

j=0

+

j l−1  

 |∇j+1−k Φ| |∇k η | |∇l−1−j u|; Rn+ Lp .

(14.7.6)

j=0 k=0

Let p(l − 1) < n. The first sum on the right-hand side of (14.7.6) is dominated by l−1  n c ∇Φ; R+ L∞ r−j ∇l−1−j u; Rn+ Lp j=0

and the second one is not greater than j l−1  

∇j+1−k Φ; Rn+ M (Wpj−k →Lp ) r−k ∇l−1−j u; Rn+ Wpj−k .

j=0 k=1

From Corollary 2.4.1 and the inclusion M Wps (Rn+ ) ⊂ M Wpt (Rn+ ), s > t, it follows that

516

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

∇j+1−k Φ; Rn+ M (Wpj−k →Lp ) ≤ c ∇Φ; Rn+ M Wpj−k ≤ c ∇Φ; Rn+ M Wpl−1 .

(14.7.7)

Moreover, by Hardy’s inequality r−k ∇l−1−j u; Rn+ Wpj−k ≤ c u; Rn+ Wpl−1 with p(l − 1) < n and l − 1 ≥ j ≥ k, we obtain ∇(η Φ); Rn+ M Wpl−1 ≤ c ∇Φ; Rn+ M Wpl−1 .

(14.7.8)

This, together with Theorem 8.7.2 and (14.7.5), implies (14.7.4) for p(l−1) < n. Now let p(l − 1) = n. The first sum on the right-hand side of (14.7.6) does not exceed l−2
 r−j ∇l−1−j u; Rn+ Lp c ∇Φ; Rn+ L∞ | log |−1 j=0

 + rl−1 (log r)−1 u; B1/2 ∩ Rn+ Lp , and the second one is not greater than j l−1  

∇j+1−k Φ; Rn+ M (Wpj−k →Lp ) r−k (log r)−1 η1/2 ∇l−1−j u; Rn+ Wpj−k .

j=0 k=0

Using Hardy’s inequality r−k (log r)−1 v; Rn+ Wpj−k ≤ c v; Rn+ Wpj , where v is a function supported by B1/2 , we obtain r−k (log r)−1 η1/2 ∇l−1−j u; Rn+ Wpj−k ≤ c u; Rn+ Wpl−1 , which, together with (14.7.7), gives (14.7.8). This, in combination with Theorem 8.7.2 and (14.7.5), leads to (14.7.4) for p(l − 1) = n. The result follows. 

14.7.3 Estimate for s1 Clearly, if sup e⊂Rn−1

Dp,l−1/p ϕ; eLp + ∇ϕ, Rn−1 L∞ ≤ c δ, [Cp,l−1−1/p (e)]1/p

then (14.6.8) holds. So we must prove the sufficiency of (14.6.8).

l−1/p

14.7 Local Characterization of Mp

(δ)

517

According to Theorem 4.1.1, the condition (14.7.3) means that Dp,l−1/p (η ϕ); eLp + ∇(η ϕ); B L∞ < c δ [Cp,l−1−1/p (e)]1/p

sup e

(14.7.9)

for sufficiently small > 0. Hereafter, e is a compact set in Rn−1 with d(e) < 1. Since ϕ(0) = 0, we derive from (14.7.1) that ∇(η ϕ); B L∞ ≤ c ∇ϕ; B L∞ . The first term in (14.7.9) is majorized by sup


D

e⊂Rn−1

p,l−1/p (η ϕ); e\B Lp [Cp,l−1−1/p (e\B )]1/p

+

Dp,l−1/p (η ϕ); e ∩ B Lp  . [Cp,l−1−1/p (e ∩ B )]1/p

(14.7.10)

(If either e\B or e ∩ B has zero capacity, then the corresponding term is equal to zero). Consequently, the supremum in (14.7.9) is not greater than s1 + s2 + s3 , where Dp,l−1/p (η ϕ); eLp , 1/p e⊂Rn−1 \B [Cp,l−1−1/p (e)] 
 |∇l−1 (η ϕ)(x) − ∇l−1 (η ϕ)(y)|p 1/p dx dy |x − y|n−2+p e Rn−1 \B s2 = sup , [Cp,l−1−1/p (e)]1/p e⊂B

s1 =

sup

s3 = sup e⊂B

Dl−1/p (η ϕ; B ); eLp . [Cp,l−1−1/p (e)]1/p

The goal of this subsection is to give an estimate for s1 . Lemma 14.7.3. If (14.6.8) holds, then s1 ≤ c δ for sufficiently small . Proof. We have 

 |∇l−1 (η ϕ)(y)| dy

|x − y|2−n−p dx

p

sp1

=

sup e⊂Rn−1 \B

B

e

Cp,l−1−1/p (e)

.

Let q = (n − 1)/(p(l − 1) − 1) if p(l − 1) < n, and let q ∈ [1, ∞) if p(l − 1) = n. Since y ∈ supp η ⊂ B/2 , it follows that |x − y|2−n−p dx ≤ (mesn−1 e)1−1/q e

1/q





Rn−1 \B

|x − y|(2−n−p)q dx

≤ c (mesn−1 e)1−1/q 2−n−p+(n−1)/q .

(14.7.11)

518

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

We use the inequalities , Cp,l−1−1/p (e) ≥

if p(l − 1) < n,

c (mesn−1 e)1−1/q n

1−p

c (log(2 /mesn−1 e))

if p(l − 1) = n.

(14.7.12)

Then, for p(l − 1) < n,  |x − y|2−n−p dx ≤ c p(l−2)+1−n (mesn−1 e)1−1/q . e

In the case p(l − 1) = n, we have  n p−1 |x − y|2−n−p dx (log(2 /mesn−1 e)) e

≤ c (mesn−1 e)

1−1/q

n

(log(2 /mesn−1 e))p−1 2−n−p+(n−1)/q .

(14.7.13)

If mesn−1 e ≤ n−1 , then the right-hand side is dominated by c 1−p | log |p−1 . If mesn−1 e > n−1 , then, setting q = 1 in (14.7.13), we obtain the same majorant c 1−p | log |p−1 . Thus ⎧  ⎪ p(l−2)+1−n ⎪ |∇l−1 (η ϕ)|p dy if p(l − 1) < n, ⎨ c p B  s1 ≤ ⎪ 1−p p−1 ⎪ |∇l−1 (η ϕ)(y)|p dy if p(l − 1) = n. ⎩c | log | B

In the case p(l − 1) < n, this implies that 
sp1 ≤ c p(l−2)+1−n (1−l)p

B

|ϕ|p dy +

l−2 

−jp

 B

j=0

 |∇l−1−j ϕ|p dy .

We introduce the notation v; B p,l−1−1/p = Dl−1−1/p (v; B ); B Lp and use the inequality

 B

|∇l−2−j v|p dy


≤ c p(j+1)−1 v; B pp,l−1−1/p + p(j+2−l)

B

Then sp1




−p+1−n

≤c



(14.7.14)

 |ϕ| dy + p

B

 |v|p dy .

1−n B

|∇ϕ|p dy + p(l−1)−n ϕ; B pp,l−1/p

 ≤c p(l−1)−n ϕ; B pp,l−1/p + ∇ϕ; B pL∞ , 

which, together with (14.6.8), gives s1 ≤ cδ.



l−1/p

14.7 Local Characterization of Mp

Now let p(l − 1) = n. In the case l = 2, we have 
 sp1 ≤ c 1−p | log |p−1 ∇ϕ; B pL∞ |η |p dy + B



Since

 |η | dy ≤ c p

B

we obtain

B

B

(δ)

519

 |y|p |∇η |p dy .

|y|p |∇η |p dy ≤ c p−1 | log |−p ,

sp1 ≤ c | log |−1 ∇ϕ; B pL∞ .

Suppose that l > 2 and p(l − 1) = n. We write 
sp1 ≤ c 1−p | log |p−1 | log |−p |y|(1−l)p |ϕ|p dy B

+

l−2  j=1

−p

| log |



−jp

B

|y|

 |∇l−1−j ϕ| dy + p

B

 |η ∇l−1 ϕ|p dy .

(14.7.15)

The first term in brackets does not exceed c p−1 | log |−p ∇ϕ; B pL∞ . Using the inequality 
 |y|−jp |∇l−2−j v|p dy ≤ c B

B

(14.7.16) 

|∇l−2 v|p dy + p(2−l)

B

|v|p dy



with v = ∂ϕ/∂yi , we conclude that the sum with respect to j in (14.7.15) is majorized by 
 p−1 −p 1−p (14.7.17) c | log | |∇l−1 ϕ|p dy + ∇ϕ; B pL∞ . B

We apply (14.7.14) with j = 0 to the vector function v = ∇ϕ. Then (14.7.17) is not greater than   c p−1 | log |−p ϕ : B pp,l−1/p + ∇ϕ; B pL∞ . (14.7.18) Now we turn to an estimate for the last integral in (14.7.15). The inequality w; B Lp ≤ c 1−1/p w; B p,1−1/p holds for all w defined on B and vanishing outside B(1−c) , c ∈ (0, 1). Hence,  1−p |η ∇l−1 ϕ|p dy ≤ c ∇l−1 (η ϕ); B pp,1−1/p B

520

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

 
|η (x) − η (y)|p  p p ≤ c ∇l−1 ϕ; B p,1−1/p + |∇l−1 ϕ(x)| dx dy . (14.7.19) |x − y|n−2+p B According to Lemma 14.7.1, 

|η (x) − η (y)|p dy ≤ c | log |−p |x|1−p . |x − y|n−2+p

Therefore, ∇l−1 (η ϕ); B pp,1−1/p 
≤ c ∇l−1 ϕ; B pp,1−1/p + | log |−p

 |∇l−1 ϕ(y)|p |y|1−p dy .

B

Since p(l − 1) = n and l > 2, we see that p < n and Hardy’s inequality  
 p p 1−p 1−p |v| |y| dy ≤ c v; B p,1−1/p + |v|p dy B

B

holds. Setting here ∇l−1 ϕ as v, we get ∇l−1 (η ϕ); B pp,1−1/p 
p 1−p −p ≤ c ϕ; B p,l−1/p + | log |

B

 |∇l−1 ϕ(y)|p dy .

By (14.7.14), the last integral does not exceed p−1 (ϕ; B pp,l−1/p + ∇ϕ; B pL∞ ) and, therefore,   ∇l−1 (η ϕ); B pp,1−1/p ≤ c ϕ; B pp,l−1/p + | log |−p ϕ; B pL∞ . (14.7.20) Consequently,    |η ∇l−1 ϕ|p dy ≤ c p−1 ϕ; B pp,l−1/p + | log |−p ∇ϕ; B pL∞ . (14.7.21) B

Substituting (14.7.14), (14.7.18) and (14.7.21) into (14.7.15), we derive   sp1 ≤ c | log |p−1 ϕ; B pp,l−1/p + ∇ϕ; B pL∞ which, together with (14.6.8), gives s1 ≤ cδ. 14.7.4 Estimate for s2 Lemma 14.7.4. If (14.6.8) holds, then s2 ≤ c δ for sufficiently small .

l−1/p

14.7 Local Characterization of Mp

(δ)

521

Proof. Clearly, 

 |∇l−1 (η ϕ)|p dx e

sp2 = sup

Rn−1 \B

|x − y|2−n−p dy .

Cp,l−1−1/p (e)

e⊂B

Let p(l − 1) < n. By q we denote any number sufficiently close to p, such that q > p. We have  |∇l−1−j ϕ|p dx l−1  p −(j+1)p+1 e s2 ≤ c sup Cp,l−1−1/p (e) e⊂B j=0 
p/q q |∇ ϕ| dx l−1 l−1−j  ≤c . (14.7.22) −(j+1)p+1 sup (mesn−1 e)1−p/q e Cp,l−1−1/p (e) e⊂B j=0 From Lemma 1.3.1 and the equality Wpl−1 (Rn )|Rn−1 = Wpl−1−1/p (Rn−1 ), it follows that
 p/q |u|q dµ ≤c Rn−1

[µ(Bρ (x))]p/q u; Rn−1 p l−1−1/p , Wp C (B ) ρ x∈B ,ρ∈(0,) p,l−1−1/p sup

where µ is a measure with supp µ ⊂ B . Therefore sup e⊂B

[µ(e)]p/q [µ(Bρ (x))]p/q ≤c sup . Cp,l−1−1/p (e) x∈B ,ρ∈(0,) Cp,l−1−1/p (Bρ )

(14.7.23)

This estimate and (14.7.22) imply that sp2 ≤ c

l−1 

−(j+1)p+1+(n−1)(1−p/q)

j=0

×





sup x∈B ,ρ∈(0,)

Bρ (x)

|∇l−1−j ϕ|q dy

p/q

 ρp(l−1)−n .

Since −(j + 1)p + 1 + (n − 1)(1 − p/q) ≤ (n − 1)(1 − p/q) + 1 − p < 0 for any q sufficiently close to p, we obtain sp2

≤c

l−2 


 sup

j=0 x∈B ,ρ∈(0,)

Bρ (x)

|∇l−1−j ϕ|q dyρq(l−j−2)+1−n

p/q

+ c ∇ϕ; B pL∞ .

522

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

We use the inequality ρl−j−2−(n−1)/q


 Bρ

≤ cρ

|∇l−2−j v|q dy

l−1−1/p−(n−1)/p

1/q


1−n v; Bρ p,l−1−1/p + c ρ

Bρ

|v|p dy

1/p , (14.7.24)

j = 0, · · ·, l−2, which follows by dilation from the continuity of the embedding l−1−1/p of Wp (B1 ) into Wql−2−j (B1 ). Then sp2 ≤ c

sup

∇ϕ; Bρ (x)pp,l−1−1/p ρn−1−p(l−1−1/p)

x∈B ,ρ∈(0,)

+ c ∇ϕ; B pL∞ .

(14.7.25)

Let p(l − 1) = n. By H¨ older’s inequality and (14.7.23 ) we have
 p/q |∇l−1 (η ϕ)|q dx sp2 ≤ c 1−p sup (mesn−1 e)1−p/q e Cp,l−1−1/p (e) e⊂B
 p/q 1−p+(n−1)(1−p/q) p−1 sup | log ρ| |∇l−1 (η ϕ)|q dy . ≤ c Bρ (x)

x∈B ,ρ∈(0,)

Hence sp2 ≤ c ×

 sup

ρ1−p+(n−1)(1−p/q) | log ρ|−1

x∈B ,ρ∈(0,)

l−2
  j=1

Bρ (x)

|y|−jq |∇l−1−j ϕ(y)|q dy

+ ρ1−p+(n−1)(1−p/q) | log ρ|−1

p/q




+ 1−p+(n−1)(1−p/q) | log ρ|p−1

Bρ (x)

|y|−(l−1)q |ϕ(y)|q dy




Bρ (x)

|η ∇l−1 ϕ|q dy

p/q

p/q  . (14.7.26)

We apply the following variant of Hardy’s inequality:
 1/q |y|−jq |∇l−2−j v|q dy Bρ

≤ cρ

v; Bρ p,l−1−1/p 1/p
 + c ρ2−l+(n−1)(1/q−1/p) |v|p dy , j = 1, · · · , l − 2, 1−1/p+(n−1)(1/q−1/p)

Bρ

with ∇ϕ instead of v. Then the first term on the right-hand side of (14.7.26) is dominated by

l−1/p

14.7 Local Characterization of Mp

(δ)

523

  c | log ρ|−1 ϕ; Bρ (x)pp,l−1/p + ∇ϕ; Bρ (x)pL∞   ≤ c ϕ; B2 pp,l−1/p + ∇ϕ; B2 pL∞ . Obviously, the second term on the right-hand side of (14.7.26 ) does not exceed
 p/q 1−p+(n−1)(1−p/q) −1 cρ | log ρ| |y|−(l−2)q dy ∇ϕ; Bρ (x)pL∞ Bρ (x)

−1

≤ c | log ρ|

∇ϕ; Bρ (x)pL∞ .

We turn to an estimate of the third term on the right-hand side of (14.7.26). Let s be a number sufficiently close to q with s > q. By H¨ older’s inequality this term is majorized by p/s
 |η ∇l−1 ϕ|s dy 1−p+(n−1)(1−p/q) ρ(n−1)p(1/q−1/s) | log ρ|p−1 ≤ 1−p+(n−1)(1−p/s) | log |p−1


 B

Bρ (x)

|η ∇l−1 ϕ|s dy

p/s

.

(14.7.27)

The inequality w; B Ls ≤ c (n−1)(1/s−1/p)+1−1/p w; B p,1−1/p holds for all w, defined on B and vanishing outside B(1−c) , c ∈ (0, 1). Hence the right-hand side of (14.7.27) is not greater than c | log |p−1 η ∇l−1 ϕ; B pp,1−1/p . By (14.7.21) this last term does not exceed   c | log |p−1 ϕ; B pp,1−1/p + | log |−p ∇ϕ; B pL∞ . Using the estimates which were obtained for three terms in (14.7.26), we arrive at 
sp2 ≤ c sup | log ρ|p−1 ϕ; Bρ (x)pp,l−1/p + ∇ϕ; B2 pL∞ . x∈B ,ρ∈(0,)

The result follows from (14.6.8).

14.7.5 Estimate for s3 Lemma 14.7.5. If (14.6.8) holds, then s3 ≤ c δ for sufficiently small .



524

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

Proof. We have Dl−1/p (η ϕ; B ); epLp ≤ c






D1−1/p (Dα η Dβ ϕ; B ); epLp

|α|+|β|=l−1

|α|>0

 +Dl−1/p (ϕ; B ); epLρ +

 |∇l−1 ϕ(x)|p dx

e

B

|η (x) − η (y)|p  dy . (14.7.28) |x − y|n−2+p

Clearly, D1−1/p (Dα η Dβ ϕ; B ); epLp


≤c

 |Dβ ϕ(x)|p dx

B

e





+ B

|Dα η (y)|p dy e

|Dα η (x) − Dα η (y)|p dy |x − y|n−2+p

|Dβ ϕ(x) − Dβ ϕ(y)|p  dx . |x − y|n−2+p

(14.7.29)

Let p(l − 1) < n. Using (14.7.2), we obtain  |∇l−1−j ϕ(x)|p dx l−1  p −(j+1)p+1 e s3 ≤ c sup Cp,l−1−1/p (e) e⊂B j=0 

 + c sup

l−1 

dx −jp

B

e

e⊂B j=0

|∇l−1−j ϕ(x) − ∇l−1−j ϕ(y)|p dy |x − y|n−2+p . Cp,l−1−1/p (e)

(14.7.30)

The first supremum is bounded by the right-hand side of (14.7.25) (see the beginning of the proof of Lemma 14.7.4). Let q denote a number sufficiently close to p and such that q > p. By H¨older’s inequality and (14.7.23), the second supremum on the right-hand side of (14.7.30) does not exceed c

l−1 

sup

−jp+n(1−p/q) ρp(l−1)−n

ξ∈B ,ρ∈(0,) j=1

  × (p−q)/p

Bρ (ξ)

+ c sup e⊂B


 B

|∇l−1−j ϕ(x) − ∇l−1−j ϕ(y)|p q/p p/q dy dx |x − y|n−2+p

Dl−1/p (ϕ; B ); epLp Cp,l−1−1/p (e)

.

We note that for j = 1, · · · , l − 2 
 |∇ q/p p l−1−j ϕ(x) − ∇l−1−j ϕ(y)| (p−q)/p dx dy |x − y|n−2+p Bρ (ξ) B

(14.7.31)

l−1/p

14.7 Local Characterization of Mp





(δ)

525




(q−p)/q |∇l−1−j ϕ(x) − ∇l−1−j ϕ(y)|q dy dy n−2 |x − y|n−2+q Bρ (ξ) B B |x − y|  
|∇l−1−j ϕ(x) − ∇l−1−j ϕ(y)|q ≤c dx dy |x − y|n−2+q Bρ (ξ) B2ρ (ξ)    |∇l−1−j ϕ(x) − ∇l−1−j ϕ(y)|q dx dy . (14.7.32) + n−2+q |x − y| B \B2ρ (ξ) Bρ (ξ)

≤ (p−q)/p

dx

The first term on the right-hand side of (14.7.32) is dominated by c ∇ϕ; B2ρ (ξ)qq,l−1−j−1/q which does not exceed q  1 1 n c ρ( q − p )n+j ∇ϕ; B2ρ (ξ)p,l−1−1/p +ρ−l+1+j+ q ∇ϕ; B2ρ (ξ)L∞ . (14.7.33) Since |x − y| > ρ in the second term on the right-hand side of (14.7.32), this term is majorized by     dz c |∇l−1−j ϕ(x + z)|q + |∇l−1−j ϕ(x)|q dx n−2+q B2 \Bρ |z| Bρ (ξ)  ≤c


∇l−1−j ; Bρ (ξ + z)q

Lq

|z|n−2+q

B2 \Bρ

 + c ρ1−q ∇l−1−j ; Bρ (ξ + z)qLq dz

which does not exceed   c sup ρq[n(1/q−1/p)+j] ϕ; Bρ (ξ)qp,l−1/p + ρ(−l+1+j)q+n ∇ϕ; Bρ (ξ)qL∞ . ξ∈B3

Note that n(1/q − 1/p) + j > 0 for j = 1, . . . , l − 2, because q is close to p. Hence, the terms in the second sum in (14.7.30) with j = 1, . . . , l − 2 are estimated by  −jp+n(1−p/q) ρq[n(1/q−1/p)+j] ρp(l−1)−n ϕ; Bρ (ξ)pp,l−1/p c sup ξ∈B3 ,ρ∈(0,)

≤c

sup ξ∈B3 ,ρ∈(0,)

 +ρjp+n(p/q−1) ∇ϕ; Bρ (ξ)pL∞  p(l−1)−n  ρ ϕ; Bρ (ξ)pp,l−1/p + ∇ϕ; Bρ (ξ)pL∞ .

The term with j = l − 1 in (14.7.31) has the majorant 
 −(l−1)p+(n−1)(1−p/q) ρp(l−1)−n ∇ϕ; B2 pL∞ dx Bρ (ξ)

B

q/p p/q dy n−2 |x − y|

≤ c 1−(l−1)p+(n−1)(1−p/q) ρp(l−1)−n+(n−1)p/q ∇ϕ; B2 pL∞ ≤ c ∇ϕ; B2 pL∞ .

526

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

Finally, the second supremum in (14.7.30) is dominated by c

((ρp(l−1)−n ϕ; B2ρ (ξ)pp,l−1/p + ∇ϕ; B2ρ (ξ)pL∞ )

sup

ξ∈B ,ρ∈(0,)

Dl−1/p (ϕ; B ); epLp

+ c sup

Cp,l−1−1/p (e)

e⊂B

.

Taking into account the estimate for the first supremum in (14.7.30), obtained previously, we complete the proof for the case p(l − 1) < n. Let p(l − 1) = n. We have Dl−1/p (η ϕ; B ); epLp ≤ σ1 + σ2 + σ3 + c Dl−1/p (ϕ; B ); epLp ,

(14.7.34)

where 

l−2  

σ1 = c

|∇l−1−j ϕ(x)|p dx

e

j=0

l−2  

σ2 = c



B

j=1



|∇j η (y)|p dy e



σ3 = c

dx B

e

By (14.7.2),  B

B

|∇j η (x) − ∇j η (y)|p dy, |x − y|n−2+p

|∇l−1−j ϕ(x) − ∇l−1−j ϕ(y)|p dx, |x − y|n−2+p

|ϕ(x)∇l−1 η (x) − ϕ(y)∇l−1 η (y)|p dy. |x − y|n−2+p

|∇j η (x) − ∇j η (y)|p dy ≤ c | log |x||−p |x|1−p(j+1) . |x − y|n−2+p

Therefore, σ1 ≤ c

l−2   j=0

≤c

e

l−3
  j=0

×

|∇l−1−j ϕ(x)|p |x|1−p(j+1) | log |x||−p dx




e

B

|∇l−1−j ϕ(x)|p(n−1)/(n−p(j+1)) dx dx

(n−p(j+1))/(n−1)

(p(j+1)−1)/(n−1)

|x|n−1 | log |x||p(n−1)/(p(j+1)−1)

+ c ∇ϕ; B pL∞



|x|1−n | log |x||−p dx.

e

The function tn−1 | log t|α increases near t = 0. Hence, among all sets e with a fixed mesn−1 , the integral  |x|1−n | log ||−α dx e

l−1/p

14.7 Local Characterization of Mp

(δ)

527

attains its maximum at a ball with centre x = 0. Consequently, for α > 1, we have  |x|1−n | log |x||−α dx ≤ c | log mesn−1 e|1−α (14.7.35) e

and hence, σ1 ≤ c

l−3 

| log mesn−1 e|1−p−(n−p(j+1))/(n−1) ∇l−1−j ϕ; B pLp(n−1)/(n−p(j+1))

j=0

+ c | log mesn−1 e|1−p ∇ϕ; B pL∞ .

(14.7.36)

This and (14.7.24) with q = p(n − 1)/[n − p(j + 1)], ρ = and ∇ϕ as v yield   σ1 ≤ c | log mesn−1 e|1−p ϕ; B pp,l−1/p + ∇ϕ; B pL∞ . We now estimate the sum σ2 . Note that  |∇l−1−j ϕ(x) − ∇l−1−j ϕ(y)|p dx |x − y|n−2+p e
 dx ≤c |∇l−1−j ϕ(x)|p n−2+p |x| {x∈e:|x|>2|y|}  + |∇l−1−j ϕ(y)|

p

 + {x∈e:|x|≤2|y|} (1)

(2)

{x∈e:|x|>2|y|}

dx |x|n−2+p

|∇l−1−j ϕ(x) − ∇l−1−j ϕ(y)|p  dx . |x − y|n−2+p (3)

Therefore, σ2 ≤ σ2 + σ2 + σ2 , where (1)

σ2

=c

l−2   j=1

(2)

σ2

=c

(3) σ2

=c

e

l−2   j=1

|∇l−1−j ϕ(x)|p

e

dx |x|n−2+p

l−2   j=1

B

dx |x|n−2+p

 B|x|/2

|∇j η (y)|p dy,

 B|x|/2

|∇l−1−j ϕ(y)|p |∇j η (y)|p dy,

 |∇j η (y)| dy p

{x∈e:|x|≤2|y|}

|∇l−1−j ϕ(x) − ∇l−1−j ϕ(y)|p dx. |x − y|n−2+p

Using the definition of η given in Sect. 14.7.1, we obtain  l−2   |∇l−1−j ϕ(x)|p (1) dx | log |y||−p |y|−jp dy σ2 ≤ c n−2+p |x| e B |x|/2 j=1 ≤c

l−2   j=1

e

|∇l−1−j ϕ(x)|p dx . |x|(j+1)p−1 | log |x||p

528

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

A majorant for this sum was found when estimating σ1 . It is equal to   (14.7.37) c | log mesn−1 e|1−p ϕ; B2 pp,l−1/p + ∇ϕ; B2 pL∞ . Clearly, (2)

≤c

σ2

l−1   e

j=1

≤c

l−2   e

j=1



dx |x|n−2+p

B|x|/2

|∇l−1−j ϕ(y)|p |y|−jp | log |y||−p dy

dx |x|n−2+p | log |x||p

 B|x|/2

|∇l−1−j ϕ(y)|p |y|−jp dy.

In view of the inequality    |∇l−1−j ϕ(y)|p |y|−jp dy ≤ c |x|p−1 ϕ; B|x|/2 pp,l−1/p + ∇ϕ; B|x|/2 pL∞ B|x|/2

we have



(2)

  ϕ; B pp,l−1/p + ∇ϕ; B pL∞ .

dx

σ2 ≤ c e

|x|n−1 | log |y||p

(2)

This inequality, together with (14.7.35), gives (14.7.37) as a majorant of σ2 . (3) The value σ2 does not exceed c

l−2   j=1

B

dx |y|jp | log |y||p

 {x∈e:|x|≤2|y|}

|∇l−1−j ϕ(x) − ∇l−1−j ϕ(y)|p dx. |x − y|n−2+p

Changing the order of integration and using the monotonicity of tjp | log t|p near t = 0, we find that (3) σ2

≤c

l−1   e

j=1

dx |x|jp | log |x||p

 B

|∇l−1−j ϕ(x) − ∇l−1−j ϕ(y)|p dy. |x − y|n−2+p

We apply H¨ older’s inequality and (14.7.35). Then, for qj = (n−1)/(n−1−jp), (3)

σ2

≤c ×

l−2
  j=1

e

B

B

 


≤c

l−2 

|x|1−n | log |x||(1−n)/j dx

jp/(n−1)

|∇l−1−j ϕ(x) − ∇l−1−j ϕ(y)|p qj 1/qj dy dx |x − y|n−2+p

| log mesn−1 e|1−p−1/qj

j=1

×

 
 B

B

|∇l−1−j ϕ(x + h) − ∇l−1−j ϕ(x)|p

dh |h|n−2+p

qj

dx

1/qj

.

l−1/p

14.7 Local Characterization of Mp

(δ)

529

Now, by Minkowski’s inequality, (3)

σ2

≤ c | log mesn−1 e|1−p ×

l−1 
  j=1

B

B

|∇l−1−j ϕ(x + h) − ∇l−1−j ϕ(x)|pqj dx

1/qj

dh . |h|n−2+p

Since l−j−1−1/p Wpl−1−1/p (Rn−1 ) ⊂ Bpq (Rn−1 ) j ,p

with

j = 1, · · · , l − 2,

(see [Bes] or [Tr3], Sect. 2.8.1), it follows that 
 1/qj |∇l−2−j v(x + h) − ∇l−2−j v(x)|pqj dx B

B

dh |h|n−2−p

≤ c (v; B3 pp,l−1−1/p + 1−n v; B3 pL∞ ). Therefore,   (3) σ2 ≤ c | log mesn−1 e|1−p ϕ; B3 pp,l−1/p + ∇ϕ; B3 pL∞ . (1)

(2)

Taking into account the estimates for σ2 and σ2 which were obtained above, we find that σ2 is bounded from above by the right-hand side of the last inequality. To obtain an estimate for σ3 we note that   dy dx |ϕ(x)∇l−1 η (x) − ϕ(y)∇l−1 η (y)|p n−2+p |x − y| e B2|x| \B|x|/2   dx dy ≤ c ∇ϕ; B pL∞ (l−1)p | log |x||p |x − y|n−2 |x| e B2|x|  dx ≤ c ∇ϕ; B pL∞ . (14.7.38) n−1 | log |x||p e |x| Moreover,   dx e

B|x|/2

|ϕ(x)∇l−1 η (x) − ϕ(y)∇l−1 η (y)|p 



dy |x − y|n−2+p

(|x|2−l | log |x||−1 + |y|2−l | log |y||−1 )p dy |x − y|n−2+p e B|x|/2   dx dy ≤ c ∇ϕ; B pL∞ n−2+p n−p | log |y||p e |x| B|x|/2 |y|  dx ≤ c ∇ϕ; B pL∞ . (14.7.39) n−1 | log |x||p |x| e ≤

c ∇ϕ; B pL∞

dx

530

14 Regularity of the Boundary in Lp -Theory of Elliptic Problems

In the same way we obtain   dx |ϕ(x)∇l−1 η (x) − ϕ(y)∇l−1 η (y)|p Rn−1 \B2|x|

e





dy |x − y|n−2+p

(|x|2−l | log |x||−1 + |y|2−l | log |y||−1 )p dy |x − y|n−2+p e Rn−1 \B2|x|   dx dy ≤ c ∇ϕ; B pL∞ n−2+p (l−2)p p | log |x|| Rn−1 \B2|x| |y| e |x|  dx ≤ c ∇ϕ; B pL∞ . (14.7.40) n−1 | log |x||p |x| e ≤c

∇ϕ; B pL∞

dx

Adding the estimates (14.7.38) – (14.7.40) and applying (14.7.35), we arrive at σ3 ≤ c | log mesn−1 e|1−p ∇ϕ; B pL∞ . Now from (14.7.34) and the estimates for σ1 , σ2 , σ3 , it follows that  Dl−1/p (η ϕ; B ); epLp ≤ c | log mesn−1 e|1−p ϕ; B3 pp,l−1/p  + ∇ϕ; B3 pL∞ + c Dl−1/p (ϕ; B ); epLp . This, together with (14.7.31), gives the estimate sp3

≤c




ϕ; B3 pp,l−1/p

+

∇ϕ; B3 pL∞

+ sup

which together with (14.6.8) gives s3 ≤ c δ.

e⊂B

Dl−1/p (ϕ; B ); epLp  Cp,l−1−1/p (e) 

Proof of Theorem 14.6.4. Lemmas 14.7.3–14.7.5 and the inequality (14.7.10) imply that (14.7.9) holds for sufficiently small . As was pointed out at the beginning of Sect. 14.7.3, the estimate (14.7.9) is equivalent to the l−1/p inclusion ∂Ω ∈ Mp (δ). The result follows. Remark 14.7.1. The results of the present chapter were obtained in [MSh10]. Filonov applied these results to the study of the Maxwell operator in Lipschitz domains [Fil].

15 Multipliers in the Classical Layer Potential Theory for Lipschitz Domains

In this chapter we give applications of Sobolev multipliers to the question of higher regularity in fractional Sobolev spaces of solutions to boundary integral equations generated by the classical boundary value problems for the Laplace equation in and outside a Lipschitz domain. Since the sole Lipschitz graph property of ∂Ω does not guarantee higher regularity of solutions, we are forced to select an appropriate subclass of Lipschitz domains whose description involves a space of multipliers. For domains of this subclass we develop a solvability and regularity theory analogous to the classical one for smooth domains. We also show that the chosen subclass of Lipschitz domains proves to be the best possible in a certain sense. We end the chapter with a brief discussion of boundary integral equations of linear elastostatics.

15.1 Introduction We study the internal and external Dirichlet problems ∆u+ = 0 in Ω,

tr u+ = Φ+ on ∂Ω,

(D+ )

and ∆u− = 0 in Rn \Ω,

tr u− = Φ− on ∂Ω,

u− (x) = O(|x|2−n )

as |x| → ∞,

(D− )

where the boundary trace is denoted by tr, as well as the internal and external Neumann problems ∆v+ = 0 in Ω, and ∆v− = 0 in Rn \Ω,

∂v+ = Ψ+ on ∂Ω, ∂ν

(N+ )

∂v− = Ψ− on ∂Ω, ∂ν

V.G. Maz’ya, T.O. Shaposhnikova, Theory of Sobolev Multipliers, Grundlehren der mathematischen Wissenschaften 337, c Springer-Verlag Berlin Hiedelberg 2009


531

532

15 Higher Regularity in the Classical Layer Potential Theory

v− (x) = O(|x|2−n ) as |x| → ∞,

(N− )

where ν stands for the outer normal with respect to Ω. In what follows, we exclude the case n = 2, which will simplify the presentation. The changes required in formulations, in comparison with dimensions n > 2, are the same as in the logarithmic potential theory for smooth contours. Our proofs, given for n > 2, apply to the two-dimensional case after minor changes. A classical method for solving the problems (D± ) and (N± ) is representation of their solutions using the double layer potential  ∂ Dσ(z) = Γ (ζ − z) σ(ζ)dsζ , z ∈ Rn \∂Ω, ∂νζ ∂Ω

and the single layer potential  Γ (ζ − z) ρ(ζ)dsζ , Sρ(z) =

z ∈ Rn \∂Ω,

∂Ω

where Γ is the fundamental solution of ∆ with singularity at the origin. Putting u± = Dσ± and v± = Sρ± , one arrives at the boundary integral equations   1 (2± ) ± 2 I + D σ± = Φ± , and



 ∓ 21 I + D∗ ρ± = Ψ± ,

(3± )

∗

where D is the adjoint of D given by  ∂ ∗ D ρ(z) = Γ (ζ − z) ρ(ζ)dsζ . ∂νz ∂Ω

Looking for solutions of the problems (D± ) and (N± ) with boundary data Φ± = Φ and Ψ± = Ψ in the form u± = Sρ and v± = Dσ, one obtains the integral equations on ∂Ω Sρ = Φ, (15.1.1) and

∂ Dσ = Ψ. (15.1.2) ∂ν Let ! be positive and noninteger. In the case p(! − 1) > n − 1, our sole restriction on Ω is the inclusion of its boundary in the class Wp , which means that every function ϕ in (14.1.3) belongs to Wp (Rn−1 ). In the opposite case p(! − 1) ≤ n − 1, we assume that ∂Ω belongs to the class Mp if every point O ∈ ∂Ω has a neighborhood U such that Ω ∩ U is given by (14.1.3) with ϕ ∈ C 0,1 (Rn−1 ) subject to ∇ϕ ∈ M Wp−1 (Rn−1 )

15.1 Introduction

533

(here and elsewhere we do not differ between spaces of scalar and vectorvalued functions in our notation). Furthermore, the surface ∂Ω is said to be in the class Mp (δ) if ∇ϕ; Rn−1 M Wp−1 ≤ δ, (15.1.3) where δ is a positive number. Obviously,  Mp (δ). Mp = δ>0

These definitions are in accordance with those in Sect. 14.3.1, where we dealt with the particular case ! = l − 1/p with integer l. Several conditions, either necessary or sufficient for ∂Ω ∈ Mp (δ), will be discussed in Sect. 15.5. In particular, the inclusion ∂Ω ∈ Mp (0) := ∩ Mp (δ) δ>0

is guaranteed by the condition p  1 ωq (∇[] ϕ, t) dt < ∞, {} t t 0

(15.1.4)

where ωq (∇k ϕ, t) is the Lq continuity modulus of the vector α

n−1 1 ∇k ϕ = {∂ α ϕ/∂xα 1 . . . , ∂xn−1 }

with α1 +· · ·+αn−1 = |α| = k, and q is any number satisfying (n−1)/(!−1) ≤ q ≤ ∞ for p(! − 1) < n − 1 and p < q ≤ ∞ for p(! − 1) = n − 1. Clearly, any surface in the class C + , > 0, belongs to Mp (0). However, there are surfaces in C  which are not in Mp . Note that ∂Ω ∈ Mp may have vertices and edges on ∂Ω in the case p(! − 1) < n − 1. We formulate our main result concerning the boundary integral equations (2± )-(15.1.2). In the statement of this result and in the sequel, the notation Wps (∂Ω)  g with g ∈ (Wps (∂Ω))∗ stands for the subspace of functions ψ ∈ Wps (∂Ω) such that  ψgds = 0. ∂Ω

Theorem 15.1.1. Let n > 2, p ∈ (1, ∞), and let ! be a noninteger with ! > 1. Suppose that ∂Ω is connected, ∂Ω ∈ Wp for p(! − 1) > n − 1 and ∂Ω ∈ Mp (δ) with some δ = δ(n, p, !) > 0 for p(!−1) ≤ n−1. Then the following assertions hold. (i) The operator 12 I + D is an isomorphism of Wp (∂Ω). (ii) The operator 12 I + D∗ is an isomorphism of Wp−1 (∂Ω). (iii) The operator S maps Wp−1 (∂Ω) isomorphically onto Wp (∂Ω). (iv) The operator (∂/∂ν)D maps Wp (∂Ω) continuously into Wp (∂Ω)  1.

534

15 Higher Regularity in the Classical Layer Potential Theory

There exists a continuous inverse −1  ∂ D : Wp−1 (∂Ω)  1 → Wp (∂Ω)  1. ∂ν (v) There exists a continuous inverse  1 −1 ∂P −2I + D , : Wp (∂Ω)  1 → Wp (∂Ω)  ∂ν where P is the harmonic capacitary potential of Ω and ∂P/∂ν ∈ Wp−1 (∂Ω) ∩(Wp (∂Ω))∗ .   The equality − 12 I + D 1 = 0 holds. (vi) There exists a continuous inverse −1  1 : Wp−1 (∂Ω)  1 → Wp−1 (∂Ω)  1. − 2 I + D∗ The equality



 1 − I + D∗ ∂P/∂ν = 0 2

holds. Counterexamples in Sect. 15.6 show that Theorem 15.1.1 fails if Mp (δ) is replaced by Mp . The invertibility properties of the operators 1 ± I + D, 2

1 ± I + D∗ , 2

S,

and

(∂/∂ν)D

in Theorem 15.1.1 result from solvability properties of the problems (D± ) and (N± ) collected in the next Theorem 15.1.2 which is of independent interest. The continuity properties of D, D∗ , S, and (∂/∂ν)D stated in Theorem 15.1.1 are deduced from the part of Theorem 15.1.2 concerning the transmission problem ∆w+ = 0 in Ω, ∆w− = 0 in Rn \Ω, tr w+ − tr w− = Φ,

∂w− ∂w+ − = Ψ on ∂Ω, ∂ν ∂ν

w− (x) = O(|x|2−n ) as |x| → ∞.

(T )

In the formulation of Theorem 15.1.2 and in the sequel, we use the weighted Sobolev space Wpk,α (Ω) endowed with the norm  u; ΩWpk,α =

1/p (dist(z, ∂Ω))pα |∇k u(z)|p dz

Ω

+ u, ΩLp .

(15.1.5)

15.1 Introduction

535

k,α Also, Wp,loc (Rn \Ω) stands for the space of functions subject to

u; B\ΩWpk,α < ∞ for an arbitrary open ball B containing Ω. Theorem 15.1.2. Let n > 2, p ∈ (1, ∞), and α = 1 − {!} − 1/p, where ! is a noninteger with ! > 1. Suppose that ∂Ω ∈ Wp for p(! − 1) > n − 1, and ∂Ω ∈ Mp (δ) with some δ = δ(n, p, !) for p(! − 1) ≤ n − 1. Then (i) For every Φ+ ∈ Wp (∂Ω) there exists a unique solution of the problem []+1,α

(D+ ), u+ ∈ Wp

(Ω), and u+ ; ΩWp[]+1,α ≤ c Φ+ ; ∂ΩWp .

(15.1.6)

This solution is represented uniquely as (Dσ+ )+ with σ+ ∈ Wp (∂Ω) subject to equation (2+ ). Moreover, u+ can be represented uniquely in the form Sρ with ρ ∈ Wp−1 (∂Ω) subject to equation (15.1.1). (ii) For every Φ− ∈ Wp (∂Ω) there exists a unique solution of the problem []+1,α

(D− ), u− ∈ Wp,loc

(Rn \Ω) and, for every ball B with B ⊃ Ω,

u− ; B\ΩWp[]+1,α ≤ c(B) Φ− ; ∂ΩWp .

(15.1.7)

This solution is represented uniquely in the form z ∈ Rn \Ω,

u− (z) = (Dσ− )(z) + C Γ (z),

where C is a constant, the singularity of the fundamental solution Γ is situated in Ω, and σ− ∈ Wp (∂Ω)  1 is a solution of the equation  1  − 2 I + D σ− = Φ− − CΓ on ∂Ω. (15.1.8) Moreover, u− can be represented uniquely in the form Sρ with ρ ∈ Wp−1 (∂Ω) subject to equation (15.1.1). (iii) For every Ψ+ ∈ Wp−1 (∂Ω)  1 there exists a unique solution of the []+1,α

problem (N+ ), v+ ∈ Wp

(Ω), subject to v+ ⊥ 1 on Ω and satisfying

v+ ; ΩWp[]+1,α ≤ c Ψ+ ; ∂ΩWp−1 .

(15.1.9)

This solution is represented uniquely in the form v+ (z) = (Sρ+ )(z) + C,

z ∈ Ω,

where C is a constant, ρ+ ∈ Wp−1 (∂Ω)  1 and ρ+ satisfies (3+ ). Moreover, v+ can be represented uniquely as v+ (z) = (Dσ)(z) + C,

z ∈ Ω,

where C is a constant and σ ∈ Wp (∂Ω)  1 satisfies (15.1.2).

536

15 Higher Regularity in the Classical Layer Potential Theory

(iv) For every Ψ− ∈ Wp−1 (∂Ω) there exists a unique solution of the prob[]+1,α

lem (N− ), v− ∈ Wp,loc

(Rn \Ω) and, for every ball B with B ⊃ Ω,

v− ; B\ΩWp[]+1,α ≤ c(B)Ψ− ; ∂ΩWp−1 .

(15.1.10)

The solution is represented uniquely in the form (Sρ− )− with ρ− ∈ Wp−1 (∂Ω) subject to equation (3− ). Moreover, v− can be represented uniquely as v− (z) = (Dσ)(z) + CΓ (z), z ∈ Rn \Ω, 

where C=−

Ψ− ds,

σ ∈ Wp (∂Ω)  1

∂Ω

and σ satisfies the equation ∂ ∂ (Dσ)− = Ψ− − C Γ− . ∂ν ∂ν

(15.1.11)

(v) For every (Φ, Ψ ) ∈ Wp (∂Ω) × Wp−1 (∂Ω) there exists a unique solution of the problem (T ) []+1,α

(w+ , w− ) ∈ Wp[]+1,α (Ω) × Wp,loc

(Rn \Ω)

and, for every ball B with B ⊃ Ω, w+ ; ΩWp[]+1,α + w− ; B\ΩWp[]+1,α   ≤ c(B) Φ; ∂ΩWp + Ψ ; ∂ΩWp−1 .

(15.1.12)

This solution is given explicitly by w± = (SΨ )± + (DΦ)±

on Rn \∂Ω.

(15.1.13)

This theorem follows essentially from Theorem 15.2.1 in Sect. 15.2 con[]+1,α cerning the Wp -solvability of the Dirichlet, Neumann, and transmission problems for equations with nonzero right-hand sides. A typical statement, contained in Theorem 15.2.1, runs as follows. Let n ≥ 2, 1 < p < ∞, ! > 1, and {!} > 0. If ∂Ω ∈ Wp for p(!−1) > n−1 and ∂Ω ∈ Mp (δ) with some δ = δ(n, p, !), for p(! − 1) ≤ n − 1, then the mapping Wp[]+1,α (Ω)  u → {∆u, tr u} ∈ Wp[]−1,α (Ω) × Wp (∂Ω)

(15.1.14)

is isomorphic. In the case p(!−1) > n−1 this last assertion can be inverted for a subclass of Lipschitz domains: the isomorphism property of the mapping (15.1.14) implies that ∂Ω ∈ Wp (Theorem 15.6.1).

15.2 Solvability of Boundary Value Problems

537

Note that this implication fails for the whole class of Lipschitz domains. As for the case p(! − 1) ≤ n − 1, several examples in Sects. 15.5 and 15.6 illustrate the sharpness of the condition ∂Ω ∈ Mp (δ) in formulations of Theorems 15.1.1-15.2.1. In particular, Example 15.6.6 shows that in general the condition ∂Ω ∈ Mp (δ) in Theorem 15.2.1 cannot be improved by ∂Ω ∈ Mp ∩ C [] . We outline the structure of this chapter. In Sect. 15.2.1 we introduce and study a class of mappings, the so-called (p, k, α)-diffeomorphisms, preserving Wpk,α and which play a crucial role in the subsequent study of the boundary value problems. Properties of the problems (D± ), (N± ), and (T ) are obtained in Sect. 15.2 (Proposition 15.3.1). The next section deals with continuity properties of the potentials and their normal derivatives. Here, in particular, definitions of all integral operators involved in Theorem 15.1.1 are given. Proof of Theorems 15.1.1 and 15.1.2 can be found in Sect. 15.4. The short Sect. 15.5 is devoted to a discussion of the class Mp (δ). In Sect. 15.6 we give a number of examples of domains which demonstrate the sharpness of our solvability results for the Dirichlet and Neumann problems as well as for the corresponding integral equations.

15.2 Solvability of Boundary Value Problems in Weighted Sobolev Spaces 15.2.1 (p, k, α)-Diffeomorphisms In this section U and V are open subsets of Rn+ = {z = (x, y) : x ∈ Rn−1 , y > 0}. By Wpk,α (V ) we denote the space of functions with the finite norm 

1/p (min{1, y}) (|∇k v(x, y)| + |v(x, y)| )dz pα

p

p

,

V

where k is a positive integer, −1 < pα < p−1, and 1 ≤ p ≤ ∞. By analogy with the definition of the (p, l)-diffeomorphism given in Sect. 9.4.1, a bi-Lipschitz homeomorphism κ : U → V will be called a (p, k, α)-diffeomorphism if the elements of its Jacobi matrix ∂κ belong to the space of multipliers M Wpk−1,α (U ). The next two propositions contain basic properties of (p, k, α)-diffeomorphisms, verified in the same way as the corresponding properties of (p, l)diffeomorphisms in Chap. 9. By ∂κ, U M Wpk−1,α we denote the sum of the norms of the elements of ∂κ in the space M Wpk−1,α (U ). Proposition 15.2.1. (i) If u ∈ Wpk,α (V ) and κ is a (p, k, α)-diffeomorphism: U → V , then u ◦ κ ∈ Wpk,α (U ) and u ◦ κ; U Wpk,α ≤ c u; V Wpk,α .

538

15 Higher Regularity in the Classical Layer Potential Theory

(ii) If κ is a (p, k, α)-diffeomorphism, then κ −1 is also a (p, k, α)-diffeomor phism. (iii) If γ ∈ M Wpk,α (V ) and κ is a (p, k, α)-diffeomorphism, then γ ◦ κ ∈ M Wpk,α (U ) and γ ◦ κ; U M Wpk,α ≤ c γ; V M Wpk,α . (iv) If κ1 : U → V and κ2 : V → W are (p, k, α)-diffeomorphisms then their composition κ2 ◦ κ1 : U → W is a (p, k, α)-diffeomorphism. Let T denote the extension operator defined by (9.4.11), where ζ(τ ) = 0 for |τ | ≥ 1. Consider a domain G = {(x, y) : x ∈ Rn−1 , y > ϕ(x)},

(15.2.1)

where ϕ is a Lipschitz function such that ϕ(0) = 0 and |∇ϕ(x)| ≤ L for almost all x ∈ Rn−1 . We introduce the mapping κ : Rn+  (ξ, η) → (x, y) ∈ G by the equalities x = ξ, y = N η + (T f )(ξ, η),

(15.2.2)

where N is a sufficiently large constant depending on L. Proposition 15.2.2. Let ! be a noninteger with ! > 1, and let p ∈ (1, ∞). If ∇ϕ ∈ M Wp−1 (Rn−1 ), then κ is a (p, [!] + 1, α)-diffeomorphism. We say that a function f defined on ∂G belongs to the space Wp (∂G) if the function Rn−1  x → f (x, ϕ(x)) belongs to Wp (Rn−1 ). This can be written as (15.2.3) f ∈ Wp (∂G) ⇔ f ◦ κ|Rn−1 ∈ Wp (Rn−1 ). []+1,α

By (15.2.3) and Proposition 15.2.1 (i), the inclusion u ∈ Wp (G) implies that tr u ∈ Wp (∂G) and there exists a linear extension operator: Wp (∂G) → []+1,α

Wp

(G). Note that (15.2.2) gives an extension of κ to the lower half-space Rn− = {(x, y) : x ∈ Rn−1 , y < 0}: Rn−  (ξ, η) → (x, y) ∈ Rn \G

and this extension has the same properties as the original mapping κ. We preserve the same notation κ for the extended mapping so that, now, κ is a quasi-isometric mapping of Rn onto Rn and a (p, [!] + 1, α)-diffeomorphic mapping of Rn+ and Rn− onto G and Rn \G, respectively.

15.2 Solvability of Boundary Value Problems

539

15.2.2 Weak Solvability of the Dirichlet Problem We need the following assertion which is similar in flavor to [GG] and Sect. 5.7.2 in [Tr4]. Lemma 15.2.1. Let p ∈ (1, ∞), and let 0 < α + 1/p. Suppose that the Lipschitz constant of the function ϕ in (15.1.3) does not exceed a sufficiently small constant depending on n, p, and α. Then the mapping Wp1,α (Ω)  u → {∆u, tr u} ∈ Wp−1,α (Ω) × Wp1−α−1/p (∂Ω) is an isomorphism. k−α−1/p

(Rn−1 ) Proof. It is well known that the fractional Sobolev space Wp is the space of traces on Rn−1 of functions in the space Wpk,α (Rn+ ), where p ∈ (1, ∞) (see [Usp]). Since ∂Ω ∈ C 0,1 , it follows from this result that the Dirichlet problem ∆u = F in Ω, u = Φ on ∂Ω (15.2.4) []−1,α

with F ∈ Wp (Ω) and Φ ∈ Wp (∂Ω) can be reduced to the case Φ = 0. ˚ 1 (Ω) be the completion of C ∞ (Ω) in the norm of W 1 (Ω) and let Let W q q 0 −1 Wq
(Ω) stand for the dual of Wq1 (Ω), where q+q  = qq  . We choose s = s(p, α) so that the imbeddings ˚ 1,α (Ω) ⊂ W p ˚ Ws1
(Ω) ⊂ Wp−1,α (Ω)

⊂

˚ 1 (Ω), W s ˚ Wp1,α (Ω),

(15.2.5)

Ws−1 (Ω),

(15.2.7)

(15.2.6)

hold. By H¨ older’s inequality these imbeddings follow from p , 1 + αp s ≥ p, s ≤ p, p , s > 1 + αp s ≥ p, s
0, for α = 0, for α < 0.

We can put, for example, " ! 1
p s= ,p , 1 + min 2 p − 1 − αp and s=

! p " 1
1 + min ,p , 2 1 + αp

for α ≤ 0

for α > 0.

Since s > 2, the operator ˚s1
(Ω)  u → ∆u ∈ W −1 W s
(Ω)

(15.2.8)

540

15 Higher Regularity in the Classical Layer Potential Theory

is a monomorphism. We now show the existence of a bounded inverse to (15.2.8) defined on Ws−1
(Ω). (Ω), and let u ∈ W21 (Ω) be a solution to the problem (15.2.4) Let F ∈ Ws−1
with Φ = 0. We denote by U a small coordinate neighborhood of a point O ∈ ∂Ω and by V an open set such that O ∈ V and V ⊂ U . We take a function χ ∈ C0∞ (U ) with χ = 1 on V . Then ∆(χu) = [∆, χ]u + χF. Let κ be the bi-Lipschitz diffeomorphism: Rn+  (ξ, η) → (x, y) ∈ G defined by x = ξ, y = η + ϕ(ξ), (15.2.9) and let σ denote its inverse. Clearly, σ maps U ∩ ∂Ω onto an open subset of the hyperplane η = 0. Now, (χu) ◦ κ satisfies the boundary value problem div (A∇((χu) ◦ κ)) = (χF ) ◦ κ + ([∆, χ]u) ◦ κ

on Rn+ , (15.2.10)

(χu) ◦ κ Rn−1 = (χΦ) ◦ (κ Rn−1 ), where

A = (∂σ ◦ κ)∗ (∂σ ◦ κ).

(15.2.11) (15.2.12)

Obviously, the right-hand side of (15.2.10) belongs to Ws−1
(Ω). Therefore, the function v := (χu) ◦ κ ∈ Ws1
(Rn+ ) is a solution of the problem v Rn−1 = 0,

(15.2.13)

H = (χF ) ◦ κ + ([∆, χ]u) ◦ κ − (χu) ◦ κ.

(15.2.14)

div(A∇v) − v = H on Rn+ , where Clearly,

I − ∂κ; Rn+ L∞ ≤ c ∇ϕ; Rn−1 L∞ , which implies that I − A; Rn+ L∞ < ε,

(15.2.15)

where ε is sufficiently small. It is a classical fact that the Dirichlet problem −∆w + w = g0 + div g

on Rv+ , w|Rn−1 = 0,

(15.2.16)

with g0 ∈ Lq (Rn+ ) and g ∈ (Lq (Rn+ ))n , 1 < q < ∞, is uniquely solvable ˚ 1 (Rn ). (This follows from the explicit representation of w by Green’s in W q + function and the continuity of a singular integral operator in Lq (Rn ).) Let

15.2 Solvability of Boundary Value Problems

541

(I − ∆)−1 stand for the inverse operator of the problem (15.2.16). We write (15.2.13) in the form v − (I − ∆)−1 Sv = (∆ − I)−1 H,

(15.2.17)

with H given by (15.2.14), and Sv = div ((A − I)∇v).

(15.2.18)

This leads to the Neumann series v=

∞ 

((I − ∆)−1 S)j (∆ − I)−1 H,

(15.2.19)

j=0

where the operator (I − ∆)−1 S has a small norm in Wq1 (Rn+ ) for every q ∈ (2, s ], by (15.2.15). Hence, v; Rn+ Wq1 ≤ c (∆ − I)−1 H; Rn+ Wq1 . Using (15.2.14) and the arbitrariness of the point O ∈ ∂Ω, we obtain u; ΩWq1 ≤ c (F ; ΩWq−1 + u; ΩLq ).

(15.2.20)

By Sobolev’s imbedding theorem, u ∈ L2n/(n−2) (Ω) if n > 2. Thus, 1 (Ω) u ∈ W2n/(n−2)

by (15.2.20). Using Sobolev’s theorem again, we see that u ∈ L2n/(n−4) (Ω)

if

n>4

and u ∈ Ws1
(Ω)

if

n ≤ 4.

Repeating this argument m times, where m > n(s − 2)/2s , we conclude that u ∈ Ws1
(Ω) and arrive at the estimate u; ΩW 1
≤ c (F ; ΩW −1 + u; ΩL2n/(n−2) ).
s

s

This implies that u; ΩW 1
≤ c (F ; ΩW −1 + u; ΩW21 )
s

s

+ F ; ΩW −1 ) ≤ c F ; ΩW −1 . ≤ c (F ; ΩW −1

s

2

s

Hence, the operator (15.2.8) is isomorphic. By duality, the operator ˚s1 (Ω)  u → ∆u ∈ Ws−1 (Ω) W

(15.2.21)

542

15 Higher Regularity in the Classical Layer Potential Theory

is isomorphic as well. This fact, combined with (15.2.5), shows that the operator ˚p1,α (Ω) → Wp−1,α (Ω) ∆:W is a monomorphism. 1 Let F ∈ Ws−1
(Ω), and let u ∈ Ws
(Ω) be a solution of (15.2.4) with Φ = 0. 1,α ˚p (Ω). It remains to prove the estimate By (15.2.6), we see that u ∈ W u; ΩWp1,α ≤ c F ; ΩWp−1,α .

(15.2.22)

It is well known that there exists a bounded inverse (I − ∆)−1 of the operator I − ∆ in Rn+ with zero Dirichlet data on Rn−1 , acting from Wpk−2,α (Rn+ ) into Wpk,α (Rn+ ), k = 1, 2, . . . (cfr. [GG], [Tr4], Sect. 5.3.2). Using this inverse, we write (15.2.13) in the form (15.2.17) and arrive at the Neumann series (15.2.19), where the operator (I − ∆)−1 S has a small norm in Wp1,α (Rn+ ), by (15.2.15). Hence, v; Rn+ Wp1,α ≤ c (∆ − I)−1 H; Rn+ Wp1,α . Using the arbitrariness of the point O ∈ ∂Ω and (15.2.14), we obtain u; ΩWp1,α ≤ c (F ; ΩWp−1,α + u; ΩWp0,α ).

(15.2.23)

It follows from the one-dimensional Hardy inequality that u; ΩWp0,α ≤ ε0 u; ΩWp1,α + C(ε0 )u; ΩL1 .

(15.2.24)

for any sufficiently small ε0 > 0. Since the operator (15.2.21) is isomorphic and the imbedding (15.2.7) holds, we have u; ΩL1 ≤ c1 u; ΩWs1 ≤ c2 F ; ΩWs−1 ≤ c3 F ; ΩWp−1,α , which together with (15.2.23) and (15.2.24) completes the proof of Lemma 15.2.1. 

15.2.3 Main Result Let Wpk,α (Ω) be the weighted Sobolev space endowed with the norm (8.1.1). We also need the weighted Sobolev space Wpk,α (Rn \Ω) supplied with the norm   v; Rn \ΩWpk,α =

1/p (min{dist(x, ∂Ω), 1})pα (|∇k v(x)|p + |v(x)|p )dx

.

Rn \Ω

Using a partition of unity and properties of the special Lipschitz domain (15.2.1) mentioned at the end of Sect. 15.2.1, we can introduce the

15.2 Solvability of Boundary Value Problems

543

[]+1,α

space Wp (∂Ω) and show that it is the trace space for both Wp (Ω) and []+1,α ˚ 1,α (Ω) obtained by completion of Wp (Rn \Ω). We also need the space W p C0∞ (Ω) in the norm of Wp1,α (Ω). By Wp−1,α (Ω) we denote the space of distributions F = g0 + div g, with g0 ∈ Wp0,α (Ω) and g ∈ (Wp0,α (Ω))n . We supply Wp−1,α (Ω) with the norm   F ; ΩWp−1,α = inf g0 ; ΩWp0,α + g; Ω(Wp0,α )n , where the infimum is taken over all representations F = g0 + div g. The next theorem contains all the information on auxiliary boundary value problems (D± ), (N± ), and (I) to be used in the sequel. Theorem 15.2.1. Let p ∈ (1, ∞), and let α = 1 − {!} − 1/p, where ! is a noninteger with ! > 1. Suppose that ∂Ω ∈ Wp for p(! − 1) > n − 1 and ∂Ω ∈ Mp (δ) with some δ = δ(n, p, !) for p(! − 1) ≤ n − 1. The five mappings []+1,α

Wp

(Ω)

→ []+1,α

Wp

u

(15.2.25)

{∆u, tr u} ∈ Wp[]−1,α (Ω)×Wp (∂Ω),

(Rn \Ω)  u

→ []+1,α

Wp

(15.2.26)

{∆u − u, tr u} ∈ Wp[]−1,α (Rn \Ω)×Wp (∂Ω), (Ω)

→

u

(15.2.27)

{∆u − u, ∂u/∂ν} ∈ Wp[]−1,α (Ω) × Wp−1 (∂Ω),

Wp[]+1,α (Rn \Ω)  u → {∆u − u, ∂u/∂ν} ∈

(15.2.28) Wp[]−1,α (Rn \Ω)

×

Wp−1 (∂Ω),

Wp[]+1,α (Ω) × Wp[]+1,α (Rn \Ω)  (u+ , u− ) (15.2.29) " ! ∂u− ∂u+ − → ∆u+ , ∆u− − u− , tr (u+ − u− ), ∂ν ∂ν ∈ {Wp[]−1,α (Ω) × Wp[]−1,α (Rn \Ω) × Wp (∂Ω) × Wp−1 (∂Ω)} are all isomorphisms. Proof. The continuity of the mappings (15.2.25)–(15.2.29) is obvious. Dealing with their invertibility, we restrict ourselves to a detailed treatment of (15.2.25), since the analysis of (15.2.26)–(15.2.29) is essentially the same.

544

15 Higher Regularity in the Classical Layer Potential Theory []−1,α

Let us show that the Dirichlet problem (15.2.4) with F ∈ Wp []+1,α Φ ∈ Wp (∂Ω) is uniquely solvable in Wp (Ω), and that   u; ΩWp[]+1,α ≤ c F ; ΩWp[]−1,α + Φ; ∂ΩWp .

(Ω) and

By Lemma 15.2.1, the problem (15.2.4) has a unique solution u ∈ Wp1,α (Ω). Therefore, in order to prove Theorem 15.2.1 we only need to show that the []+1,α (Ω) and to estimate its norm in this space. solution u belongs to Wp Let U be a coordinate neighborhood of a point O ∈ ∂Ω and let V denote an open set such that O ∈ V and V ⊂ U . We take a function χ ∈ C0∞ (U ) with χ = 1 on V . Then ∆(χu) = [∆, χ]u + χF. Let κ be the (p, [!] + 1, α)-diffeomorphism defined by (15.2.2), where N = 1, and let σ denote its inverse. Clearly, σ maps U ∩ ∂Ω onto an open subset of the hyperplane η = 0. Now, (χu) ◦ κ satisfies the boundary value problem (χF ) ◦ κ + ([∆, χ]u) ◦ κ div (A∇((χu) ◦ κ)) = det(∂σ ◦ κ) (χu) ◦ κ Rn−1 = (χΦ) ◦ (κ Rn−1 ), where A=

on Rn+ , (15.2.30)

(∂σ ◦ κ)∗ (∂σ ◦ κ) . det(∂σ ◦ κ)

(15.2.31)

(15.2.32)

By Proposition 15.2.1 (i), (iii), the right-hand side of (15.2.30) belongs to []−1,α Wp (Rn+ ) and the Dirichlet data (15.2.31) are in Wp (Rn−1 ). These data []+1,α

have an extension Θ ∈ Wp

(Rn+ ). Therefore, the function

v := (χu) ◦ κ − Θ ∈ Wp1,α (Rn+ ) is a solution of the problem div(A∇v) − v = H on Rn+ ,

v Rn−1 = 0,

(15.2.33)

where H=

(χF ) ◦ κ + ([∆, χ]u) ◦ κ − div(A∇Θ) + Θ − (χu) ◦ κ. det(∂σ ◦ κ)

(15.2.34)

We shall consider the cases p(!−1) ≤ n−1 and p(!−1) > n−1 separately. The case p(! − 1) ≤ n − 1. Let ∂Ω ∈ Mp (δ). By (9.4.17) and Theorem 8.7.1, I − ∂κ; Rn+ M Wp[],α ≤ c ∇ϕ; Rn−1 M Wp−1 . This along with (15.1.3) implies that I − A; Rn+ M Wp[],α ≤ c ∇ϕ; Rn−1 M Wp−1 ≤ c δ.

(15.2.35)

15.2 Solvability of Boundary Value Problems

545

We can replace [!] on the left-hand side of (15.2.35) by any k = 0, 1, . . . [!] be[],α cause of the imbedding M Wp (Rn+ ) ⊂ M Wpk,α (Rn+ ). This imbedding follows from M Wp0,α (Rn+ ) = L∞ (Rn+ ) ⊃ M Wpk,α (Rn+ ) [],α

by interpolation between Wp (Rn+ ) and Wp0,α (Rn+ ) (see [Tr4], Sect. 3.4.2). It is standard that there exists a bounded inverse (I −∆)−1 to the operator I − ∆ in Rn+ with zero Dirichlet data on Rn−1 , acting from Wpk,α (Rn+ ) into Wpk−2,α (Rn+ ), k = 0, 1, . . . (see [Tr4], Sect. 5.3.2). We write (15.2.33) in the form v − (I − ∆)−1 Sv = (∆ − I)−1 H

(15.2.36)

with H given by (15.2.34) and Sv = div((A − I)∇v). This leads to the Neumann series v=

∞ 

((I − ∆)−1 S)j (∆ − I)−1 H

j=0

where the operator (I −∆)−1 S has a small norm in Wpk+1,α (Rn+ ), k = 0, 1, . . . , by (15.2.35). Since H ∈ Wp0,α (Rn+ ) and (∆ − I)−1 H ∈ Wp2,α (Rn+ ), it follows that v ∈ 2,α Wp (Rn+ ) and therefore, χu ∈ Wp2,α (Ω). Using the arbitrariness of the point O ∈ ∂Ω we derive that u ∈ Wp2,α (Ω) which completes the proof for ! < 2. Let ! > 2. Using Proposition 15.2.1 and u ∈ Wp2,α (Ω), we obtain H ∈ 1,α Wp (Rn+ ) which implies that v ∈ Wp3,α (Rn+ ) by (15.2.36). Repeating this []+1,α

argument several times if necessary, we conclude that u ∈ Wp is the required result for p(! − 1) ≤ n − 1. The case p(! − 1) > n − 1. We have A; Rn+ Wp[],α ≤ c ϕ; Rn−1 Wp .

(Ω). This

(15.2.37)

Without loss of generality we may assume that ∇ϕ; Rn−1 L∞ < δ, where δ is sufficiently small. Then I − A; Rn+ L∞ ≤ cδ.

(15.2.38)

We introduce a cutoff function ζ ∈ C0∞ (B2 ) with ζ = 1 on B1 , and set ζε (ξ, η) = ζ(ξ/ε, η/ε), where ε is a small positive number. By (15.2.33) div(A∇(ζε v)) − ε−2 ζε v = K on Rn+ , ζε v Rn−1 = 0, (15.2.39) where

K = ζε H + ∇ζε A∇v + div(vA∇ζε ) − ε−2 ζε v

546

15 Higher Regularity in the Classical Layer Potential Theory

with H and v defined as in the case p(! − 1) ≤ n − 1. We know that u ∈ Wp1,α (Ω). Let us suppose that u ∈ Wpk,α (Ω), 1 < k ≤ [!]. Then v ∈ Wpk,α (Rn+ ) and H ∈ Wpk−1,α (Rn+ ), which implies K ∈ Wpk−1,α (Rn+ ). We introduce the new coordinates (ξ/ε, η/ε) and use the nota˜ v˜ and K ˜ for A, v, and K as functions of (ξ/ε, η/ε). Written in these tions A, dilated variables, the problem (15.2.39) becomes ˜ on Rn+ , ζ v˜ n−1 = 0. (I − ∆)(ζ v˜) − div((A˜ − I)∇(ζ v˜)) = ε2 K R By (15.2.37) ˜ Rn  0,α ≤ c ε−1−(n−1)/p ϕ; Rn−1 W  . ∇[] A; + Wp p ˜ Therefore, A˜ − I; Rn  Also, (15.2.38) holds with A replaced by A. + M Wp[],α is sufficiently small. This implies that the operator P given by P w = (I − ∆)−1 div((A˜ − I)∇w) is contractive in Wpk+1,α (Rn+ ). Hence, ζ v˜ ∈ Wpk+1,α (Rn+ ) which implies that u ∈ Wpk+1,α (Ω). This gives the result for the mapping (15.2.25). Only trivial changes in the above argument are needed in order to treat the mappings (15.2.26)–(15.2.29).  Now we deduce certain properties of the problems (D± ), (N± ) and (I) from Theorem 15.2.1. Proposition 15.2.3. Let Ω satisfy the conditions in Theorem 15.2.1. Then: (i) For every Φ+ ∈ Wp (∂Ω) there exists a unique solution u+ ∈ []+1,α

Wp

(Ω) of (D+ ) subject to (15.1.6). (ii) For every Φ− ∈ Wp (∂Ω) there exists a unique solution u− ∈

[]+1,α

Wp,loc (Rn \Ω) of (D− ) subject to (15.1.7). (iii) For every Ψ+ ∈ Wp−1 (∂Ω)  1 there exists a unique solution v+ ∈ []+1,α

Wp

(Ω) of (N+ ) subject to v+ ⊥ 1 on Ω and (15.1.9). (iv) For every Ψ− ∈ Wp−1 (∂Ω) there exists a unique solution v− ∈

[]+1,α

Wp,loc

(Rn \Ω) of (N− ) subject to (15.1.10). []+1,α

(v) For every (Φ, Ψ ) ∈ Wp (∂Ω) × Wp lution (w+ , w− ) ∈

[]+1,α Wp (Ω)

×

(∂Ω) there exists a unique so-

[]+1,α Wp,loc (Rn \Ω)

of (I) subject to (15.1.12).

Proof. Assertion (i) was justified in Theorem 15.2.1. Let us prove (ii). Since the local Lipschitz constant of ∂Ω is small, the 1,α (Rn \Ω) is standard. It suffices to prove unique solvability of (N− ) in Wp,loc []+1,α

1,α that the solution u ∈ Wp,loc (Rn \Ω) belongs to Wp,loc C0∞ (Rn ), χ = 1 on Ω. Clearly,

(Rn \Ω). Let χ ∈

15.3 Continuity Properties of Boundary Integral Operators

(I − ∆)(χu) = −χu − [∆, χ]u on Rn \Ω,

547

tr(χu) = 0 on ∂Ω.

Since χu + [∆, χ]u ∈ Wpk−1,α (Rn \Ω)

for

u ∈ Wpk,α (Rn \Ω),

it follows from Lemma 15.2.1 with [!] replaced by k that u ∈ Wpk+1,α (Rn \Ω). Letting k = 1, . . . , [!], we arrive at (ii). Proofs of (iii)–(v) require only obvious changes in the argument just used. 

15.3 Continuity Properties of Boundary Integral Operators We collect basic properties of the potentials Dσ and Sρ with σ ∈ Wp (∂Ω) and ρ ∈ Wp−1 (∂Ω) where, as usual, p ∈ (1, ∞), ! > 1, and {!} > 0. Proposition 15.3.1. Let the notations Dσ and Sρ refer to the double and single layer potentials defined on Rn \∂Ω. For almost all Q ∈ ∂Ω there exist the seven limits  1 (ζ − Q, ν(ζ)) (Dσ)(Q) := lim σ(ζ)dsζ , ε→0 |∂B1 | |ζ − Q|n ∂Ω\Bε (Q)

1 (D σ)(Q) := lim ε→0 |∂B1 | ∗



(ζ − Q, ν(Q)) σ(ζ)dsζ , |ζ − Q|n

∂Ω\Bε (Q)

lim (Dσ)(z) =

z→Q z∈Ω

1

2I

+ D)σ(Q),

  lim (Dσ)(z) = − 12 I + D σ(Q),

z→Q z∈Rn \Ω

−1 (Sρ)(Q) := lim (Sρ)(z) = z→Q |∂B1 |(n − 2) z∈Rn \∂Ω

 ∂Ω

ρ(ζ)dsζ , |ζ − Q|n−2

  ∂ (Sρ)+ (Q) := lim (ν(Q), (∇Sρ)(z)) = − 12 I + D∗ ρ(Q), z→Q ∂ν z∈Ω ∂ (Sρ)− (Q) := ∂ν

lim (ν(Q), (∇Sρ)(z)) =

z→Q z∈Rn \Ω

1

2I

 + D∗ ρ(Q),

where (Sρ)+ and (Sρ)− are the restrictions of Sρ to Ω and Rn \Ω.

(15.3.1) (15.3.2)

(15.3.3)

(15.3.4) (15.3.5)

548

15 Higher Regularity in the Classical Layer Potential Theory

These classical properties of the layer potentials can be found in [Verc] for σ and ρ in Lp (∂Ω), where z → Q means a nontangential approach. As a justification, a reference is given in [Verc] to the methods developed in [CMM], [Ca3], and [FJR]. However, for our more regular σ and ρ, the above identities can be deduced directly by using the convergence of the integral  |σ(ζ) − σ(z)|p + |ρ(ζ) − ρ(z)|p dsζ |ζ − z|n−1+p{} ∂Ω

for almost every z ∈ ∂Ω. Proposition 15.3.2. The operators D, D∗ , and S satisfy  Dσ; ∂ΩWp ≤ c σ, ∂ΩWp

(15.3.6)

 (Dσ)+ ; ΩWp[]+1,α ≤ c σ; ∂ΩWp

(15.3.7)

 (Dσ)− ; B\ΩWp[]+1,α ≤ c(B) σ, ∂ΩWp

(15.3.8)

 Sρ; ∂ΩWp ≤ c ρ; ∂ΩWp−1

(15.3.9)

 (Sρ)+ ; ΩWp[]+1,α ≤ c ρ; ∂ΩWp−1

(15.3.10)

 (Sρ)− ; B\ΩWp[]+1,α ≤ c(B) ρ, ∂ΩWp−1

(15.3.11)

 D∗ ρ; ∂ΩWp−1 ≤ c ρ; ∂ΩWp−1 ,

(15.3.12)

where (Dρ)± and (Sρ)± are the restrictions of Dσ and Sρ to Ω and Rn \Ω, respectively, and B is an arbitrary ball containing Ω. Proof. Let us prove (15.3.6)–(15.3.8). Suppose that σ ∈ Wp (∂Ω). By Proposition 15.2.3 (v), the transmission problem (I) with Φ = σ and Ψ = 0 has a unique solution []+1,α

(w+ , w− ) ∈ Wp[]+1,α (Ω) × Wp,loc

(Rn \Ω)

subject to w+ ; ΩWp[]+1,α ≤ c σ; ∂ΩWp ,

(15.3.13)

w− ; B\ΩWp[]+1,α ≤ c(B) σ; ∂ΩWp .

(15.3.14)

By Green’s formula, w± = D(w+ − w− ) = Dσ on Rn \∂Ω which implies (15.3.7), (15.3.8), and tr w+ , ∂ΩWp ≤ c σ, ∂ΩWp . Since Dσ = tr w+ − σ/2 by (15.3.1), this last inequality leads to (15.3.6).

15.3 Continuity Properties of Boundary Integral Operators

549

Combining (15.3.6) with (15.3.1) and (15.3.2), we see that tr (Dσ)+ and tr (Dσ)− belong to Wp (∂Ω). This together with Theorem 15.1.2 (i), (ii) lead to (15.3.7) and (15.3.8). We turn to the proof of (15.3.9)–(15.3.12). Let ρ ∈ Wp−1 (∂Ω). By Proposition 15.2.3 (v) the transmission problem (T ) with Φ = 0 and Ψ = ρ has a unique solution []+1,α

(w+ , w− ) ∈ Wp[]+1,α (Ω) × Wp,loc

(Rn \Ω)

subject to w+ ; ΩWp[]+1,α ≤ c Ψ ; ∂ΩWp−1 and w− ; B\ΩWp[]+1,α ≤ c(B) Ψ, ∂ΩWp−1 . By Green’s formula,  w± = S

∂w+ ∂w− − ∂ν ∂ν

 (15.3.15)

= (SΨ )±

which implies (15.3.10), (15.3.11), and tr w− ; ∂ΩWp + 

∂w− ; ∂ΩWp−1 ≤ c ρ; ∂ΩWp−1 . ∂ν

(15.3.16)

Since

1 D∗ ρ = ∂w− /∂ν − ρ 2 by (15.3.5), we arrive at (15.3.12). Finally, (15.3.9) follows from (15.3.15) and (15.3.16). 

We finish this section with a discussion of properties of the normal derivatives of the double layer potential with density in Wp (∂Ω). By (15.3.7), the trace of ∇(Dσ)+ belongs to Wp−1 (∂Ω) and defines a continuous operator: Wp (∂Ω) → Wp−1 (∂Ω). We need the following weighted extension of Proposition 2.7.5 which is proved in the same way. Proposition 15.3.3. Let Γ ∈ M Wpk,α (Rn+ ) and let Γ0 := Γ ; Rn+ L∞ . If g ∈ C k−1 ([−Γ0 , Γ0 ]), then g(Γ ) ∈ M Wpk,α (Rn+ ) and g(Γ ); Rn+ M Wpk,α ≤ c

k  j=0

g (j) ; [−Γ0 , Γ0 ]L∞ Γ ; Rn+ j

M Wpk,α

.

550

15 Higher Regularity in the Classical Layer Potential Theory

Corollary 15.3.1. Let γ ∈ Wp (Rn−1 ) and let γ0 := γ; Rn−1 L∞ . Suppose that g ∈ C [],1 ([−γ0 , γ0 ]). Then g(γ) ∈ M Wp (Rn−1 ) and 

[]+1

g(γ); Rn−1 M Wp ≤ c

g (j) ; [−γ0 , γ0 ]L∞ γ; Rn−1 jM W  . p

j=0

Proof. The result follows from Proposition 15.3.3 by putting Γ = T γ, where T is defined by (9.4.11), and using Theorem 8.7.1. Proposition 15.3.4. Let σ ∈ Wp (∂Ω). The operator defined by  ∂ (Dσ)+ (P ) := ν(P ), tr ∇(Dσ)+ ) ∂ν

(15.3.17)

maps Wp (∂Ω) into Wp−1 (∂Ω)  1 continuously, and ∂ ∂ (Dσ)+ = (Dσ)− a.e. on ∂Ω. ∂ν ∂ν

(15.3.18)

Proof. The components of ν, expressed in a local cartesian system (x, y), depend smoothly on ∇ϕ, where ϕ is the function in (14.1.3). Since ∇ϕ ∈ M Wp−1 (Rn−1 ), we conclude by Proposition 15.3.3 that ν ∈ M Wp−1 (∂Ω).

(15.3.19)

Hence the operator Wp (∂Ω)  σ →

∂ (Dσ)+ (P ) ∈ Wp−1 (∂Ω) ∂ν

is continuous. Let us consider the solution (w+ , w− ) of problem (T ) with the boundary conditions ∂w− ∂w+ − = 0 a.e. on ∂Ω. ∂ν ∂ν

(15.3.20)

∂w− ∂w+ and S = D tr w− on Ω. ∂ν ∂ν

(15.3.21)

tr w+ − tr w− = σ and By Green’s formula, w+ = D tr w+ − S Analogously, w− = S

∂w− ∂w+ − D tr w− and S = D tr w+ on Rn \Ω. ∂ν ∂ν

(15.3.22)

15.4 Proof of Theorems 15.1.1 and 15.1.2

551

Hence, w+ = D(tr w+ − tr w− ) = Dσ on Ω

(15.3.23)

and w− = D(tr w+ − tr w− ) = Dσ on Rn \Ω. Now, equality (15.3.18) is a consequence of (15.3.20), and ∂(Dσ)+ /∂ν ⊥ 1 follows from (15.3.23).  The proposition just proved enables us to introduce the operator (∂/∂ν)D by ∂  ∂ D σ := (Dσ)± ∂ν ∂ν

(15.3.24)

and to conclude that (∂/∂ν)D maps Wp (∂Ω) into Wp−1 (∂Ω)  1.

15.4 Proof of Theorems 15.1.1 and 15.1.2 15.4.1 Proof of Theorem 15.1.1 The continuity of the operators D : Wp (∂Ω) → Wp (∂Ω) D∗ : Wp−1 (∂Ω) → Wp−1 (∂Ω) S : Wp−1 (∂Ω) → Wp (∂Ω) ∂ D : Wp (∂Ω) → Wp−1 (∂Ω)  1 ∂ν was established in Propositions 15.3.2 and 15.3.4. []+1,α (Ω) solve (D+ ) with Solvability of equation (2+ ). Let u+ ∈ Wp Φ+ ∈ Wp (∂Ω). Then ∂u+ /∂ν ∈ Wp−1 (∂Ω). We find a solution v− ∈ []+1,α

Wp,loc

(Rn \Ω) of problem (N− ) with Ψ− := ∂u+ /∂ν. By Green’s formula, u+ = D tr u+ − S

∂v− ∂u+ and S = D tr v− on Ω. ∂ν ∂ν

Hence, u+ = D(tr u+ − tr v− ) on Ω. This together with (15.3.1) shows that σ+ := tr u+ − tr v− ∈ Wp (∂Ω) is a solution of (2+ ). We have σ+ ; ∂ΩWp ≤ tr u+ ; ∂ΩWp + tr v− ; ∂ΩWp   ≤ c u+ ; ΩWp[]+1,α + v− ; B\ΩWp[]+1,α .

(15.4.1)

552

15 Higher Regularity in the Classical Layer Potential Theory

By Proposition 15.2.3 (iv) and (15.3.19) v− ; B\ΩWp[]+1,α ≤ c 

∂u+ ; ∂ΩWp−1 ≤ c u+ ; ΩWp[]+1,α . ∂ν

The last norm does not exceed c Φ+ , ∂ΩWp by Proposition 15.3.1 (i) which together with (15.4.1) leads to the estimate σ+ ; ∂ΩWp ≤ c Φ+ ; ∂ΩWp .

(15.4.2)

Uniqueness for equation (2+ ). Let 1 ( I + D)σ = 0 with 2

σ ∈ Wp (∂Ω).

By Proposition 15.2.3 (v) we can find a solution []+1,α

(w+ , w− ) ∈ Wp[]+1,α (Ω) × Wp,loc

(Rn \Ω)

of the transmission problem for the Laplace equation on Rn \∂Ω with boundary conditions (15.3.20). By (15.3.21), w+ = (Dσ)+ . It follows from (15.3.1) and the definition of σ that tr w+ = 0. In view of Proposition 15.2.3 (i), w+ = 0 which together with (15.3.20) implies that ∂w− /∂ν = 0. Proposition 15.2.3 (iv) gives w− = 0 and hence σ = tr w+ − tr w− = 0. This completes the proof of (i). []+1,α Solvability of equation (3− ). Let v− ∈ Wp,loc (Rn \Ω) solve (N− ) with []+1,α

Ψ− ∈ Wp−1 (∂Ω). Then tr v− ∈ Wp (∂Ω). We find a solution u+ ∈ Wp of (D+ ) with Φ+ := tr v− . By Green’s formula,

(Ω)

v− = S(∂v− /∂ν − ∂u+ /∂ν) which implies that ρ− = ∂v− /∂ν − ∂u+ /∂ν ∈ Wp−1 (∂Ω) satisfies (3− ). By (15.3.19), ρ− ; ∂ΩWp−1 ≤ c (tr ∇v− ; ∂ΩWp−1 + tr ∇u+ ; ∂ΩWp−1 ) ≤ c (v− ; B\ΩWp[]+1,α + u+ ; ΩWp[]+1,α ). Using Proposition 15.2.3 (i), we see that the last norm does not exceed c tr v− ; ∂ΩWp which is majorized by v− ; B\ΩWp[]+1,α . Hence, ρ− ; ∂ΩWp−1 ≤ c v− ; B\ΩWp[]+1,α .

15.4 Proof of Theorems 15.1.1 and 15.1.2

553

Reference to Proposition 15.2.3 (iv) yields the estimate ρ− , ∂ΩWp−1 ≤ c Ψ− , ∂ΩWp−1 . Uniqueness for equation (3− ). Let 1 ( I + D∗ )ρ− = 0, 2

where

ρ− ∈ Wp−1 (∂Ω). []+1,α

By Proposition 15.2.3 (v) we can find a solution (w+ , w− ) ∈ Wp (Ω) × []+1,α Wp,loc (Rn \Ω) of the transmission problem for the Laplace equation on Rn \∂Ω with boundary conditions tr w+ − tr w− = 0 and

∂w+ ∂w− − = ρ− on ∂Ω. ∂ν ∂ν

(15.4.3)

By Green’s formula, w− = S

∂w− ∂w+ − D w− and S − D w+ = 0 on Rn \Ω. ∂ν ∂ν

Hence w− = S

(15.4.4)

 ∂w− ∂w+  − = Sρ− on Rn \Ω. ∂ν ∂ν

By (15.3.5),

  ∂w− = 12 I + D∗ ρ− ∂ν which implies that ∂w− /∂ν = 0 on ∂Ω. Using Theorem 15.1.2 (iv), we see that w− = 0 on Rn \Ω. This and (15.4.3) gives tr w+ = 0. Proposition 15.2.3 (i) shows that w+ = 0. Therefore, ρ− = 0 by (15.4.3). This completes the proof of assertion (ii). We turn to assertion (iii). []+1,α Solvability of equation (15.1.1). Let u+ ∈ Wp (Ω) be a solution of  (D+ ) with Φ+ := Φ ∈ Wp (∂Ω). By u− we denote a solution of (D− ) with []+1,α

Φ− := Φ, u− ∈ Wp,loc

(Rn \Ω). Using Green’s formula we obtain u+ = S(∂u− /∂ν − ∂u+ /∂ν)

which together with (15.3.19) implies that ρ = ∂u− /∂ν − ∂u+ /∂ν ∈ Wp−1 (∂Ω). Hence, ρ is a solution of (15.1.1). We have + + + ∂u+ + ∂u− ; ∂Ω +Wp−1 + + ; ∂Ω +Wp−1 ρ; ∂ΩWp−1 ≤ + ∂ν ∂ν 
≤ c u+ ; ΩWp[]+1,α + u− ; B\ΩWp[]+1,α

554

15 Higher Regularity in the Classical Layer Potential Theory

and, in view of Proposition 15.2.3 (i), (ii), we obtain ρ; ∂ΩWp−1 ≤ c Φ; ∂ΩWp . Uniqueness for equation (15.1.1). Let ρ ∈ Wp−1 (∂Ω) and Sρ = 0 on ∂Ω. By (15.3.3), tr (Sρ)± = 0 which together with Proposition 15.2.3 (i), (ii) implies that (Sρ)± = 0. Since ρ = ∂(Sρ)− /∂ν − ∂(Sρ)+ /∂ν by (15.3.4) and (15.3.5), it follows that ρ = 0. Our next goal is assertion (iv). Solvability of equation (15.1.2). Let Ψ ∈ Wp−1 (∂Ω)  1. By Proposition []+1,α

15.3.1 (iii) there exists a solution v+ ∈ Wp (Ω) of (N+ ) with boundary data Ψ , unique up to an arbitrary constant term. By v− we denote a unique []+1,α Wp,loc (Rn \Ω)-solution of (N− ) with the same boundary data ψ which exists by Proposition 15.2.3 (iv). Let σ = tr v+ − tr v− . Then (15.3.20) holds and, by (15.3.21), v+ = D σ. This together with (15.3.24) gives (15.1.2). Choosing the value of an arbitrary constant term in v+ , we obtain σ ⊥ 1. We have σ; ∂ΩWp ≤ tr v+ − tr v+ ; ∂ΩWp + tr v− − tr v− ; ∂ΩWp , where the bar over a function stands for its mean value. Hence, σ; ∂ΩWp ≤ v+ − tr v+ ; ΩWp[]+1,α + v− ; B\ΩWp[]+1,α , where B is a ball containing Ω. Using Proposition 15.2.3 (iii), (iv), we obtain σ; ∂ΩWp ≤ c Ψ ; ∂ΩWp−1 . Uniqueness for equation (15.1.2). Let σ ∈ Wp−1 (∂Ω) and let ∂(Dσ)/∂ν = 0 on ∂Ω. By (15.3.24), ∂(Dσ)± /∂ν = 0 and therefore, by Proposition 15.2.3 (ii), (iv), (Dσ)− = 0. (Dσ)+ = const, It follows from σ = tr (Dσ)+ − tr (Dσ)− that σ = const. Solvability of equation (2− ). We recall that the capacitary potential P of Ω is a unique solution of (D− ) with the Dirichlet data 1 and that  ∂P ds = cap Ω > 0. − ∂Ω ∂ν Suppose that Φ− ∈ Wp (∂Ω)  ∂P/∂ν.

15.4 Proof of Theorems 15.1.1 and 15.1.2 []+1,α

Let u− ∈ Wp,loc

555

(Rn \Ω) satisfy problem (D− ). Then ∂u− /∂ν ∈ Wp−1 (∂Ω)

and

 ∂Ω

∂u− ds = ∂ν

 ∂Ω

∂u− tr P ds = ∂ν

 Φ− ∂Ω

∂P ds = 0. ∂ν []+1,α

By Proposition 15.2.3 (iii), there exists a solution v+ ∈ Wp with Ψ+ = ∂u− /∂ν and v+ ⊥ 1 on Ω. By Green’s formula, u− = S

(Ω) of (N+ )

∂u− ∂v+ − D tr u− and S = D tr v+ on Rn \Ω. ∂ν ∂ν

Hence, u− = D(tr v+ − tr u− ). This together with (15.3.2) shows that tr v+ − tr u− ∈ Wp (∂Ω) is a solution of (2− ). From (D1)− = 0 and (15.3.2) we find (− 12 I + D)1 = 0. Therefore, the function σ− := tr v+ − tr u− − tr v+ + tr u− satisfies (2− ). Clearly,

  σ− ; ∂ΩWp ≤ c tr v+ ; ∂ΩWp + tr u− ; ∂ΩWp   ≤ c v+ ; ΩWp[]+1,α + u− ; B\ΩWp[]+1,α .

(15.4.5)

In view of Proposition 15.2.3 (ii) and (15.3.19) + + ∂u− , ∂Ω +Wp−1 ≤ c u− ; B\ΩWp[]+1,α v+ ; ΩWp[]+1,α ≤ c + ∂ν for an arbitrary ball B ⊃ Ω. The last norm does not exceed c Φ− ; ∂ΩWp , by Proposition 15.2.3 (ii), which together with (15.4.5) leads to σ− ; ∂ΩWp ≤ c Φ− ; ∂ΩWp . Uniqueness for equation (2− ). Suppose that σ ∈ Wp (∂Ω) and (− 12 I + D)−1 σ = 0. By Proposition 15.2.3 (v) we can find a solution []+1,α

(w+ , w− ) ∈ Wp[]+1,α (Ω) × Wp,loc

(Rn \Ω)

556

15 Higher Regularity in the Classical Layer Potential Theory

of the transmission problem for the Laplace equation on Rn \∂Ω with boundary conditions (15.3.20). In view of (15.3.22), w− = (Dσ)− . It follows from (15.3.2) and the definition of σ that tr w− = 0. By Proposition 15.3.1 (ii), w− = 0 which together with (15.3.20) implies that ∂w+ /∂ν = 0. Proposition 15.2.3 (iii) gives w+ = const and hence σ = tr w+ − tr w− = const. The result follows since σ ⊥ 1. []+1,α Solvability of equation (3+ ). Let v+ ∈ Wp (Ω) solve (N+ ) with Ψ+ ∈ −1 Wp (∂Ω)  1. We assume that v+ ⊥ 1 on Ω. We find a solution u− ∈ []+1,α

Wp,loc

(Rn \Ω) of (D− ) with Φ− := tr v+ ∈ Wp (∂Ω). By Green’s formula, v+ = D tr v+ − S

∂u− ∂v+ and S = D tr u− on Ω. ∂ν ∂ν

Hence,

 ∂u− ∂v+  − . ∂ν ∂ν This together with (15.3.4) shows that v+ = S

∂u− /∂ν − ∂v+ /∂ν ∈ Wp−1 (∂Ω) is a solution of (3+ ). Since S∂P/∂ν = 1 on Ω, it follows from (15.3.4) that (− 12 I + D∗ )∂P/∂ν = 0. Therefore, the function ρ+ :=

∂v+ ∂P ∂u− − +C , C = const, ∂ν ∂ν ∂ν

(15.4.6)

satisfies (3+ ). The constant C can be chosen so that ρ+ ⊥ 1 on ∂Ω. By (15.4.6) and (15.3.19), ρ+ ; ∂ΩWp−1 ≤ c (tr ∇v+ ; ∂ΩWp−1 + tr ∇u− ; ∂ΩWp−1 )   ≤ c v+ ; ΩWp[]+1,α + u− ; B\ΩWp[]+1,α . By Proposition 15.2.3 (iii), the last norm does not exceed c tr v+ ; ∂ΩWp which is majorized by c v+ ; ΩWp[]+1,α . Hence, ρ+ ; ∂ΩWp−1 ≤ c v+ ; ΩWp[]+1,α . Reference to Proposition 15.2.3 (iii) yields in the estimate ρ+ ; ∂ΩWp−1 ≤ c Ψ+ ; ∂ΩWp−1 .

15.4 Proof of Theorems 15.1.1 and 15.1.2

557

Uniqueness for equation (3+ ). Let (− 12 I + D∗ )ρ+ = 0,

where ρ+ ∈ Wp−1 (∂Ω)  1.

By Proposition 15.2.3 (v) we can find a solution []+1,α

(w+ , w− ) ∈ Wp[]+1,α (Ω) × Wp,loc

(Rn \Ω)

of the transmission problem for the Laplace equation on Rn \∂Ω with the boundary conditions tr w+ − tr w− = 0 and

∂w+ ∂w− − = ρ+ on ∂Ω. ∂ν ∂ν

(15.4.7)

By Green’s formula, w+ = D tr w+ − S

∂w− ∂w+ and S = D tr w− on Ω. ∂ν ∂ν

Hence, w+ = S

(15.4.8)

 ∂w− ∂w+  − = S ρ+ on Ω. ∂ν ∂ν

Using (15.3.4), we have ∂w+ = (− 12 I + D∗ )ρ+ ∂ν which implies that ∂w+ /∂ν = 0 on ∂Ω. Using Proposition 15.2.3 (iii), we see that w+ = const on Ω. This and (15.4.7) gives tr w− = const which implies that w− = constP . Using (15.4.7) again, we obtain ρ+ = const ∂P/∂ν. This  together with ρ+ ⊥ 1 completes the proof of assertion (v).

15.4.2 Proof of Theorem 15.1.2 Now, we are in a position to prove Theorem 15.1.2 stated in Introduction. All assertions concerning the solvability of the problems (D± ), (N± ), and (T ), as well as the estimates (15.1.6)–(15.1.12) have been proved in Proposition 15.2.3. We need to justify the representations of solutions to these problems by layer potentials. (i) By Theorem 15.1.1 (i), there exists a unique solution σ+ ∈ Wp (∂Ω) to equation (2+ ). By (15.3.1) and (15.3.7), (Dσ+ )+ is a solution of (D+ ) in []+1,α Wp (Ω). Hence, u+ = (Dσ+ )+ by Proposition 15.2.3 (i). Theorem 15.1.1 (iii) implies the existence of a unique solution ρ ∈ Wp−1 (∂Ω) of (15.1.1). From (15.3.3) and (15.3.10) we obtain that (Sρ)+ is a []+1,α

solution of (D+ ) in Wp (Ω). Hence, u+ = (Sρ)+ by Proposition 15.2.3 (i). (ii) By Theorem 15.1.1 (v), (15.1.8) has a solution σ− ∈ Wp (∂Ω)  1 if and only if

558

15 Higher Regularity in the Classical Layer Potential Theory

 (Φ− − C Γ ) ∂Ω

which is equivalent to

∂P ds = 0, ∂ν

 C=

Φ− ∂Ω

∂P ds. ∂ν

By (15.3.2) and (15.3.8), the function (Dσ− )− + C Γ− is a solution in []+1,α Wp,loc (Rn \Ω) to (D− ). Hence, u− = (Dσ− )− + C Γ− by Proposition 15.2.3 (ii). According to Theorem 15.1.1 (iii), there exists a unique solution ρ ∈ Wp−1 (∂Ω) of (15.1.1). Using (15.3.3) and (15.3.11), we find that (Sρ)− is []+1,α

a solution of (D− ) in Wp,loc (Rn \Ω). Hence, u− = (Sρ)− by Proposition 15.2.3 (ii). (iii) Theorem 15.1.1 (vi) implies the existence of a unique solution ρ+ ∈ Wp−1 (∂Ω)  1 of equation (3+ ). From (15.3.4) and (15.3.10) we obtain that []+1,α

(Sρ)+ is a solution of (N+ ) in Wp

(Ω). Therefore,

v+ = (Sρ)+ + C. The constant C can be chosen to ensure that v+ ⊥ 1 on Ω. By Theorem 15.1.1 (iv), there exists a unique solution σ ∈ Wp (∂Ω)  1 of (15.1.2). From (15.3.17) and (15.3.7) we find that (Dσ)+ + C is a solution of []+1,α (N+ ) in Wp (Ω). Choosing C to ensure the orthogonality of (Dσ)+ + C and 1 on Ω, we conclude that v+ = (Dσ)+ + C. (iv) Theorem 15.1.1 (ii) implies the existence of a unique solution ρ− ∈ Wp−1 (∂Ω) of equation (3− ). It follows from (15.3.5) and (15.3.11) that (Sρ− )− []+1,α

is a solution of (N− ) in Wp,loc (Rn \Ω). Hence, v− = (Sρ− )− by Proposition 15.2.3 (iv). By Theorem 15.1.1 (iv), there exists a unique solution σ ∈ Wp (∂Ω)  1 of (15.1.11) provided that  C=−

Ψ− ds. ∂Ω

It follows from (15.3.24) and (15.3.8) that (Dσ)− + CΓ− is a solution of (N− ) []+1,α in the space Wp,loc (Rn \Ω). Therefore, v− = (Dσ)− + CΓ− by Proposition 15.2.3 (iv). []+1,α (v) We note that (SΨ )+ + (DΦ)+ belongs to Wp (Ω) and that []+1,α (SΨ )− +(DΦ)− belongs to Wp,loc (Rn \Ω) by (15.3.7), (15.3.10) and (15.3.8),

15.5 Properties of Surfaces in the Class Mp (δ)

559

(15.3.11), respectively. Furthermore, (SΨ )± + (DΦ)± satisfies the boundary conditions of problem (T ) by (15.3.1), (15.3.2), (15.3.3), and (15.3.5). The equality w± = (SΨ )± + (DΦ)± results from Proposition 15.2.3 (v). The proof of Theorem 15.1.2 is complete. 

15.5 Properties of Surfaces in the Class Mp(δ) Let p(! − 1) ≤ n − 1. According to Theorem 4.1.1, the condition ∂Ω ∈ Mp (δ) is equivalent to the inequality Dp, ϕ; Rn−1 M (Wp−1 →Lp ) + ∇ϕ; Rn−1 L∞ < c δ, where


 Dp, ϕ(x) =

Rn−1

|∇[] ϕ(x + h) − ∇[] ϕ(x)|p |h|1−n−p{} dh

(15.5.1)

1/p .

The following local characterization of Mp (δ) is obtained in the same way as Lemma 14.7.2, where ! = l − 1/p with integer l. Let η be an even function in C0∞ (−1, 1) with η = 1 on (−1/2, 1/2). We put ⎧ ⎪ if p(! − 1) < n, ⎨η(|z|/ε) ηε (z) = ⎪ ⎩ η(log ε/log |z|) if p(! − 1) = n. Lemma 15.5.1. A surface ∂Ω belongs to the class Mp (δ) if and only if for any O ∈ ∂Ω there exists a neighborhood U and a special Lipschitz domain G = {z = (x, y) : x ∈ Rn−1 , y > ϕ(x)} such that U ∩ Ω = U ∩ G and lim sup ∇(ηε ϕ); Rn−1 M Wp−1 ≤ c δ,

(15.5.2)

ε→0

where c is a constant which depends on !, p, n, and η is introduced above. A proof of the next local condition, equivalent to ∂Ω ∈ Mp (δ), follows the same lines as that of Theorem 14.6.4. Theorem 15.5.1. A surface ∂Ω belongs to the class Mp (δ) if and only if for every point O ∈ ∂Ω there exists a neighborhood U such that (14.1.3) holds with ϕ satisfying  Dp, (ϕ, Bε ); eLp lim sup  (15.5.3) 1/p + ∇ ϕ; Bε L∞ ≤ c δ, ε→0 e⊂Bε C−1,p (e)

560

15 Higher Regularity in the Classical Layer Potential Theory

where Bε = {ζ ∈ Rn−1 , |ζ| < ε},  Dp, (ϕ, Bε )(x) =

Bε

|∇[] ϕ(x) − ∇[] ϕ(ζ)|p dζ |x − ζ|n−1+p{}

1/p ,

and c is a constant depending on n, p, and !. Simpler conditions sufficient for ∂Ω ∈ Mp (δ) can be derived from (15.5.3) combined with the well-known inequalities between the capacity and the Lebesgue measure (see Propositions 3.1.2, 3.1.3): We have ∂Ω ∈ Mp (δ) if either (a) p(! − 1) < n − 1 and   Dp, (ϕ, Bε ); eLp + ∇ ϕ; B  lim sup  ≤ c δ, ε L∞  n−1−p(−1) ε→0 e⊂Bε mesn−1 (e) (n−1)p or p(! − 1) = n − 1 and   lim sup |log (mesn−1 (e))|(p−1)/p Dp, (ϕ, Bε ); eLp + ∇ ϕ; Bε L∞ ≤ c δ.

ε→0

e⊂Bε

This leads to the following condition, sufficient for ∂Ω ∈ Mp (0):  ∂Ω ∈ Bq,p and ∇ϕ; Rn−1 L∞ < δ.  The condition ∂Ω ∈ Bq,p can be improved for p(! − 1) = n − 1, if the Orlicz space Ltp (log+ t)p−1 is used instead of Lq with an arbitrary q, but we shall not  means that the continuity modulus ω[] of go into this. Note that ∂Ω ∈ B∞,p ∇[] ϕ satisfies   1 ω[] (t) p dt < ∞, (15.5.4) t t{} 0

which implies, in particular, that any surface ∂Ω in the class C [],{}+ε with  an arbitrary ε > 0 belongs to Mp (0). The next example shows that the condition (15.5.4), sufficient for ∂Ω ∈ Mp (0), is sharp. It demonstrates, in particular, that there exist surfaces in C [],{} which do not belong to Mp . Example 15.5.1. Let T denote a domain in R2 with compact closure and (2) boundary ∂T . By Br we denote the open disk of a sufficiently small radius r centered at an arbitrary point O ∈ ∂T . We assume that Br(2) ∩ T = {(x1 , y) ∈ Br(2) : x1 ∈ R1 , y > F (x1 )}.

15.5 Properties of Surfaces in the Class Mp (δ)

561

Let Bρ = {x ∈ Rn−2 : |x | < ρ}, where x = (x2 , . . . , xn ) and let η ∈ (n−2) (n−2) ∞ C0 (B2 ) with η = 1 on B1 . Also let ϕ(x) = F (x1 ) η(x ) and U = (2) (n−2) . We construct a bounded domain Ω ⊂ Rn satisfying (14.1.3) Br × B2 whose boundary is smooth outside U . According to Example 4.4.1, for any increasing function ω ∈ C[0, 1] satisfying the inequality (n−2)

 δ δ

1

dt ω(t) 2 + t



δ

ω(t) 0

dt ≤ c ω(δ), t

as well as the condition  1 0

ω(t) t{}

p

dt = ∞, t

(15.5.5)

one can construct a function ϕ of the above form such that the continuity modulus of ∇[] ϕ does not exceed c ω with c = const, and ϕ∈ / Wp (Rn−1 ).

(15.5.6)

Therefore, ∂Ω ∈ / Mp . In the case ∂Ω ∈ C [],{} we have ω(t) = t{} which implies (15.5.5). Hence, the last inclusion is not sufficient for ∂Ω to be in Mp .  Now we show that surfaces in the class Mp (δ) with p(! − 1) < n − 1 may have conic vertices and s-dimensional edges if s < n − 1 − p(! − 1). Example 15.5.2. Let s be an integer, 1 ≤ s ≤ n − 1, and let x = (x1 , . . . , xn−1 ) ∈ Rn−1 . We use the notations ξ = (x1 , . . . , xs ) and η = (xs+1 , . . . , xn−1 ). Consider the domain G = Kn−s × Rs , where Kn−s is the (n − s)-dimensional cone {(η, y) : y > −A |η|},

A = const > 0.

(15.5.7)

The well-known equivalence relation  v; R

n−1

Wp−1 ∼

Rs

1/p v(ξ, ·); Rn−1−s pW  p

dξ

 + Rn−1−s

1/p v(·, η); Rs pW  dη p

(15.5.8)

(see (4.2.3)) implies that the Hardy-type inequality  Rn−1

|v|p dx ≤ c v; Rn−1 pW −1 p |η|p(−1)

(15.5.9)

562

15 Higher Regularity in the Classical Layer Potential Theory

holds for all v ∈ C0∞ (Rn−1 ) if and only if  |w|p dη ≤ c w; Rn−1−s pW −1 p(−1) p |η| n−1−s R holds for all w ∈ C0∞ (Rn−1−s ). It is standard that the last inequality is valid if and only if p(! − 1) < n − 1 − s. One can easily check that Dp, |η| = c |η|1− . Hence, (15.5.9) is equivalent to Dp, |η| ∈ M (Wp−1 (Rn−1 ) → Lp (Rn−1 )). By (4.3.89), the last inclusion can be written as ∇|η| ∈ M Wp−1 (Rn−1 ). Thus, the domain G belongs to Mp ∩ C 0,1 if and only if s < n − 1 − p(! − 1). Under  this restriction on the dimension of the edge, ∂G ∈ Mp (c A). Remark 15.5.1. Suppose that for any point O ∈ ∂Ω there exists a neighborhood U such that U ∩ Ω is C ∞ -diffeomorphic to the domain Rs × {(x, y) : y > f (xs+1 , . . . , xn−1 )},

0 ≤ s ≤ n − 2,

i.e. the dimensions of boundary singularities are at most n − 1 − s. Then (15.5.8) shows that (15.1.3) is equivalent to ∇ϕ; Rn−1−s M Wp−1 ≤ c δ and, in particular, it takes the form ∇ϕ; Rn−1−s W −1 ≤ c δ, p,unif

if n − 1 − s < p(! − 1) ≤ n − 1. In other words, ∂Ω ∈ Mp (δ) if and only if the (n − 1 − s)-dimensional domain {(x, y) : y > ϕ(xs+1 , . . . , xn−1 )} belongs to Mp (c δ).

15.6 Sharpness of Conditions Imposed on ∂Ω 15.6.1 Necessity of the Inclusion ∂Ω ∈ Wp in Theorem 15.2.1 We start by showing that the condition ∂Ω ∈ Wp is necessary for the solv[]+1,α

ability in Wp

(Ω) of the Dirichlet problem

∆ u = g ∈ Wp[]−1,α (Ω),

u|∂Ω = Φ ∈ Wp (∂Ω)

(15.6.1)

provided that Ω is subject to some regularity assumptions. It is worth noting that certain additional conditions on ∂Ω should be imposed to guarantee the above statement. For example, it is well known that the problem ∆ u = g ∈ L2 (Ω),

u|∂Ω = 0

is uniquely solvable in W22 (Ω) for any convex domain which is not necessarily 3/2 in W2 (p = 2, α = 0, ! = 3/2).

15.6 Sharpness of Conditions Imposed on ∂Ω

563

Theorem 15.6.1. Let one of the following conditions hold: Either ! ∈ (1, 2), ∂Ω ∈ C 1 , and the continuity modulus ω of the normal to ∂Ω satisfies the Dini condition  1 dt (15.6.2) ω(t) < ∞, t 0 or ! > 2 and ∂Ω ∈ C []−1,1 . Then ∂Ω ∈ Wp if, for every Φ ∈ Wp (∂Ω), the problem (15.6.1) has a []+1,α

solution u ∈ Wp

(Ω), where α = 1 − {!} − 1/p.

Proof. Let Φ be a nonnegative function vanishing on U ∩ ∂Ω, where U is an arbitrary coordinate neighborhood. It is well known that (15.6.2) guarantees that u ∈ C 1 (Ω) and the outer normal derivative at any point of U ∩ ∂Ω is positive. Let us use the mapping λ = κ −1 with κ introduced in Sect. 9.4.3. Since ϕ ∈ C []−1,1 (Rn−1 ) and ∇x u, uy ∈ W [],α (V ∩ Ω) for any set V with V ⊂ U , it follows that ∇x u ◦ λ and uy ◦ λ belong to W [],α (κ(V ∩ Ω)). Therefore, tr (∇x u ◦ λ) and tr (uy ◦ λ) are in Wp (κ(V ∩ ∂Ω)). Observing that (Wp ∩ L∞ )(κ(V ∩ ∂Ω)) is a multiplication algebra, we conclude that ∇ϕ = −tr (∇x u ◦ λ)/tr (uy ◦ λ) ∈ Wp (κ(V ∩ ∂Ω)). The result follows from the arbitrariness of V and U .



Remark 15.6.1. By the above proof we have shown that the inclusion ∂Ω ∈ Wp is necessary for the solvability of (D+ ) in W []+1,α (Ω) for all Φ ∈ Wp (∂Ω) under the conditions imposed on ∂Ω in Theorem 15.6.1. Note that ∂Ω ∈ Wp is also sufficient in the case p(! − 1) > n − 1 by Theorem 15.1.2.  15.6.2 Sharpness of the Condition ∂Ω ∈ B∞,p

Using Remark 15.6.1 we show in the following example that no condition on  (condition (15.5.4)) can give the solvability of ∂Ω weaker than ∂Ω ∈ B∞,p []+1,α problem (D+ ) in W (Ω) for all Φ+ ∈ Wp (∂Ω). We recall that ∂Ω ∈  is sufficient for ∂Ω ∈ Mp (δ) and hence for this solvability in the case B∞,p p(! − 1) ≤ n − 1 (see Theorem 15.1.2 and Sect. 15.4). Example 15.6.1. Let Ω be the domain described in Example 15.5.1. By (15.5.6) and Theorem 15.6.1, problem (D+ ) for Ω is not generally solvable []+1,α in Wp (Ω) if Φ+ ∈ Wp (∂Ω).

564

15 Higher Regularity in the Classical Layer Potential Theory

The next example of the same nature demonstrates the sharpness of the  for the solvability of the Neumann conditions ∂Ω ∈ Wp and ∂Ω ∈ B∞,p problem. Example 15.6.2. We use the domains T and Ω from Example 15.5.1. Let ∂T (2) be a simple contour and let (α, β) denote the arc Br ∩ ∂T . We choose an arbitrary point τ ∈ ∂T \(α, β) and introduce a function γ ∈ Wp−1 (∂T ) equal to zero on (α, β) and at the point τ , negative on (τ, α) and positive on (β, τ ). We require also that γ is orthogonal to one on ∂T . Since ∂T ∈ C []−1,1 , the problem ∆h = 0 in R2 \T , ∂h/∂ν = γ on ∂T []

has a solution h ∈ (Wp,loc ∩ L∞ )(R2 \T ). Let ζ ∈ C0∞ (R2 ) with ζ = 1 on T , and let η be the cutoff function from Example 15.5.1. The function v(x, y) = h(x1 , y) ζ(x1 , y) η(x ) satisfies the Neumann problem ∆v − v = g in Rn \Ω,

∂v/∂ν = Ψ on ∂Ω

(15.6.3)

with g = h∆η + 2η∇h∇ζ ∈ Wp[] (Rn \Ω) ⊂ Wp[]−1,α (Rn \Ω) and ψ = η∂h/∂ν + h∂η/∂x ∈ Wp−1 (∂Ω). []+1,α

If the problem (15.6.3) is solvable in the space Wp g ∈ Wp[]−1,α (Rn \Ω) []+1,α

(Rn \Ω) for all

and Ψ ∈ Wp−1 (∂Ω), []+1,α

then v ∈ Wp (Rn \Ω) and hence h ∈ Wp,loc (R2 \T ). By χ we denote a conjugate harmonic function of h such that h(α) = 0. Clearly, []+1,α (R2 \T ). Since the first derivative of χ|∂T is equal to γ, it folχ ∈ Wp (2) lows that χ = 0 on Br ∩ ∂T and χ ≥ 0 on ∂T . Repeating the proof of Theorem 15.6.1 with R2 \T and χ|∂T instead of Ω and ϕ, respectively, we obtain ∂T ∈ Wp which implies that ∂Ω ∈ Wp . However, this is not true in view of (15.5.6), and therefore (15.6.3) is not solvable []+1,α in Wp (Rn \Ω), in general. Thus, no matter how weak the violation of the  be, it may lead to the breakdown of the solvability in inclusion ∂Ω ∈ B∞,p []+1,α

Wp

(Ω), p(! − 1) ≤ n, for the Neumann problem (15.6.3).

15.6.3 Sharpness of the Condition ∂Ω ∈ Mp (δ) in Theorem 15.2.1 It was mentioned preceding Theorem 15.6.1, that the inclusion ∂Ω ∈ Wp []+1,α

is not necessary for the solvability of the Dirichlet problem in Wp (Ω). Hence, there is no necessity of the condition Mp (δ). However, we show in this

15.6 Sharpness of Conditions Imposed on ∂Ω

565

section that the inclusion ∂Ω ∈ Mp (δ) is best possible in a certain sense. In fact, the following two examples demonstrate that the inequality (15.1.3), where p(! − 1) < n − 1 and δ is not small, is not sufficient, in general, for the []+1,α -solvability of the Dirichlet and Neumann problems. Wp Example 15.6.3. Let a domain Ω coincide with the domain G in Example 15.5.2 in a neighborhood of the origin. We adopt the same notations as in Example 15.5.2. Let u be a positive harmonic function in Ω, satisfying tr u = Φ+ ∈ Wpl (∂Ω) with Φ+ vanishing on U ∩∂Ω. It is well known that for small r = (|η|2 +y 2 )1/2 u(x) = C(ξ)rλ Θ(ω) + O(rλ1 ),

(15.6.4)

where 1 > λ1 > λ > 0, ω = (η/r, y/r), Θ is smooth on {(η, y) ∈ Kn−s : r = 1}, and C is smooth and positive near the origin of Rs . Moreover, the asymptotic []+1,α relation (15.6.4) is infinitely differentiable and therefore u ∈ Wp (Ω) if and only if n − 1 − s > p(! − λ). If s < n − 2, λ can be made arbitrarily small by choosing sufficiently large A in (15.5.7). In the case s = n − 2, we have λ > 1/2, and λ − 1/2 can be made arbitrarily small by increasing the value of A. According to Example 15.5.2, ∂Ω ∈ Mp if and only if n − 1 − s > p(! − 1). []+1,α

At the same time, one can choose A to have u ∈ / Wp (Ω) if and only if n−1−s < p! for s < n−2, and 1 < p(!−1/2) for s = n−2. Thus, the inclusion []+1,α ∂Ω ∈ Mp ∩ C 0,1 does not imply the solvability of (D+ ) in Wp (Ω) for all Φ+ ∈ Wp (∂Ω) if p! > n − 1 − s > p(! − 1)

for s < n − 2

p(! − 1/2) > 1 > p(! − 1)

for s = n − 2.

and  In the next example we demonstrate that the inclusion ∂Ω ∈ Mp (δ) in Theorem 15.2.1 cannot be replaced by ∂Ω ∈ Mp ∩ C [] , for a particular choice of p and !. Example 15.6.4. Let the domain Ω be described in a neighborhood of O by the inequality y > ϕ(x), where ϕ(x) = C η(x, 0)|x1 |/log(1/|x1 |) with C ≥ π/4 and η ∈ C0∞ (B1/2 ), η = 1 on B1/4 . By ζ(t) we denote the conformal mapping of the domain {t = x1 + ix2 : |t| < 1/2, x2 > C|x1 |/log(1/|x1 |)}

566

15 Higher Regularity in the Classical Layer Potential Theory

into the half-disk {ζ : Im ζ > 0, |ζ| < 1} with ζ(0) = 0. By Sect. 14.6.1, the function u(z) = η(2z)Im ζ(x1 + ix2 ) does not belong to W22 (Ω) and satisfies the Dirichlet problem ∆ u = f ∈ L2 (Ω),

3/2

tr u = Φ ∈ W2 (∂Ω).

(15.6.5)

Replacing Ω by Rn \Ω and using the function v(z) = η(2z)Re ζ(x1 + ix2 ), we arrive at a solution of the Neumann problem ∆ v − v = f ∈ L2 (Ω),

1/2

∂v/∂ν = Ψ ∈ W2 (∂Ω),

(15.6.6)

which does not belong to W22 (Ω). Thus, there is no solvability in W22 (Ω) and in W22 (Rn \Ω) of problems (15.6.5) and (15.6.6) in spite of the inclusion 3/2 ∂Ω ∈ M2 ∩ C 1 . The same result can be obtained for the problem (N− ) by making small changes in the above argument. We require that Rn \Ω coincides with the domain G from Example 15.5.2 near the origin O. We note that there exists a harmonic function in Rn \Ω satisfying ∂u/∂ν = Ψ− ∈ Wp−1 (∂Ω) with Ψ− = 0 in a neighborhood of O such that the asymptotic representation (15.6.4) holds with C(0) = 0. The rest of the argument is literally the same as for (D+ ). 15.6.4 Sharpness of the Condition ∂Ω ∈ Mp (δ) in Theorem 15.1.1 Here we give counterexamples concerning the solutions σ and ρ of the integral equations (2+ ) and (3− ). First we show that the solvability properties of (2+ ) and (3− ) proved in Theorem 15.1.1 may fail if ∂Ω ∈ Mp ∩ C 0,1 and ∂Ω ∈ / Mp (δ). Example 15.6.5. Let us consider the domain Ω at the beginning of Example 15.5.2 with n = 3 and s = 0. Now we deal with the three-dimensional conic singularity {z = (r, θ, ω) : r > 0, 0 ≤ θ < π − ε, 0 ≤ ω < 2π}, where ε > 0 and θ is the angle between y-axis and z. We asume that the functions Φ+ and Ψ− in (2+ ) and (3− ) vanish near the vertex of this cone. It was proved in [LeM] that solutions of (2+ ) and (3− ) have the asymptotic representations σ+ (z) = σ(0) + c1 |z|λ + O(|z|1+ε ), ρ− (z) = c2 |z|λ−1 (1 + O(|z|µ )) with µ > 0, 0 < λ < 1, and nonzero c1 and c2 . The exponent λ can be made arbitrarily small by diminishing the value of ε. Also note that these asymptotic formulae can be differentiated. According to Example 15.5.2, ∂Ω ∈ Mp if and

15.6 Sharpness of Conditions Imposed on ∂Ω

567

only if p(!−1) < 2. However, for p! > 2, we can choose A in the cone (15.3.22) / Wp (∂Ω) and ρ− ∈ / Wp−1 (∂Ω). so large that σ+ ∈ −1 Now, suppose that ρ− ∈ Wp (∂Ω) is a solution of (15.1.1) with Φ = 1 near O, 0 ≤ Φ ≤ 1 on ∂Ω. Let us denote the solutions of the interior and exterior Dirichlet problems for the Laplace equation with the same boundary data Φ by u+ and u− . It is well known that ∇k u− (z) = o(|z|k−1 )

as |z| → 0 for k = 1, 2

and u+ (z) = c |z|λ α(θ) (1 + o(|z|µ ))

as |z| → 0,



where µ > 0, α is smooth, α (π − ε) = 0, and λ > 0 can be made arbitrarily small by choosing a sufficiently small ε > 0. The above asymptotics of u+ can be differentiated. Hence, ρ = ∂u− /∂ν − ∂u+ /∂ν has the differentiable representation ρ(z) = c3 |z|λ−1 (1 + o(|z|µ )) 

which contradicts the inclusion ρ ∈ Wp−1 (∂Ω).

Next, we give an example demonstrating that, in general, the condition ∂Ω ∈ Mp (δ) in Theorem 15.1.1 (iii) cannot be improved by ∂Ω ∈ Mp ∩ C [] . Example 15.6.6. Consider the same domain Ω as in Example 15.6.4. Let 1/2 ρ ∈ W2 (∂Ω) be a solution of (15.1.1) with Φ = 1 near O and 0 ≤ Φ ≤ 1 on ∂Ω. By u+ and u− we mean the solution of the interior and exterior Dirichlet problems for the Laplace equation with tr u± = Φ. Using the conformal mapping t → ζ(t) one can show that u+ has the differentiable asymptotic representation u+ (z) = H(ξ) Im ζ(t) (1 + | log |t| |−1 )

as |t| → 0,

where ξ = (x3 , . . . , xn−1 , y) and H is a smooth function with H(0) = 0. We also have ∇k u− (z) = o(|z|k−1 ) as |z| → 0 for k = 1, 2. Hence ρ = ∂u− /∂ν − ∂u+ /∂ν has the differentiable representation ρ(z) = c H(ξ) | log |t| |2C/π (1 + | log |t| |−1 ) for sufficiently small |ξ| and |t| → 0. One can check directly that the function 1/2 on the right-hand side does not belong to W2 in any neighborhood of O for 3/2 C ≥ π/4. If the condition ∂Ω ∈ W2 (δ) in Theorem 15.1.1 (iii) could be 3/2 replaced by ∂Ω ∈ W2 ∩ C 1 , one would have a contradiction.

568

15 Higher Regularity in the Classical Layer Potential Theory

Remark 15.6.2. For the history of boundary integral equations generated by elliptic boundary value problems in domains with nonsmooth boundaries see [Ke2], [Maz17]. In particular, a comprehensive theory of integral equations on the boundaries of Lipschitz graph domains was developed in [JK1], [JK2], [CMM], [Verc], [Ke1], [Ca4], [Fab], [FKV], [DKV], [Cos], [MT1]–[MT5], and [MM]. All these works concern solvability and regularity properties either in Lp (∂Ω) or in fractional Sobolev spaces Wp (∂Ω), 0 < ! < 1. Under the assumption that ∂Ω is sufficiently smooth, one can apply such powerful tools as pseudodifferential calculus to equations (2± )–(15.1.2), which results in a comprehensive theory of their solvability in various spaces of differentiable functions. The regularity theory for equations (2± )–(15.1.2) with respect to the scale of the fractional Sobolev spaces Wp (∂Ω) is developed in this chapter under weak smoothness assumptions on ∂Ω, when the corresponding results in the theory of pseudodifferential operators on ∂Ω are unavailable at the present time. As a substitute, we rely upon an approach proposed in [Maz13], [Maz14], [Maz16], [Maz17], which reduces the study of boundary integral equations to the study of the inverse operators of auxiliary boundary value problems. The exposition of Sects. 15.1–15.6 follows the paper [MSh23].

15.7 Extension to Boundary Integral Equations of Elasticity In principle, the Laplace operator we dealt with in this chapter can be replaced by the operator  ∂2 Aij ∂zi ∂zj 1≤i,j≤n

m with constant matrix coefficients Aij = Ars ij r,s=1 , subject to the symmetry rs sr condition Aij = Aji and the Legendre-Hadamard strong ellipticity condition

(Aij η, η)ξi ξj ≥ c|ξ|2 |η|2 ,

c = const > 0,

for all vectors ξ ∈ Rn and η ∈ Rm . The statement of the interior and exterior Dirichlet problems for a bounded Lipschitz domain Ω ∈ Rn does not change, whereas the Neumann condition is replaced by  1≤i,j≤n

νi Aij tr

∂u± = Ψ± ∂zj

with ν = (ν1 , . . . , νn ) standing for the outer unit normal with respect to Ω. In particular, one may include the Dirichlet and traction problems for the Lam´e system of linear elastostatics. We preserve the same notations for boundary value problems and elastic potentials as in the harmonic potential

15.7 Extension to Boundary Integral Equations of Elasticity

569

theory developed previously. Also, we make no difference in notations of spaces of scalar and vector-valued functions. Let Ω be a domain in R3 with compact closure and boundary ∂Ω. We study the internal and external Dirichlet problems for the Lam´e system µ ∆u+ + (λ + µ)∇div u+ = 0 in Ω, tr u+ = Φ+ on ∂Ω,

(D+ )

and µ ∆u− + (λ + µ)∇div u− = 0 in R3 \Ω, tr u− = Φ− on ∂Ω, u− (x) = O(|x|−1 ) as |x| → ∞,

(D− )

where the boundary trace is denoted by tr and µ > 0,

3λ + 2µ > 0,

as well as the internal and external Neumann problems µ ∆v+ + (λ + µ)∇div v+ = 0 in Ω, J v+ = Ψ+ on ∂Ω,

(N+ )

and µ ∆v− + (λ + µ)∇div v− = 0 in R3 \Ω, J v− = Ψ− on ∂Ω, v− (x) = O(|x|−1 ) as |x| → ∞,

(N− )

where J is the traction operator given by J


∂  ∂u , νx u = 2 µ + λνx · div u + µ νx × rot u. ∂x ∂νx

We also need the transmission problem µ ∆w+ + (λ + µ)∇div w+ = 0 in Ω, µ ∆w− + (λ + µ)∇div w− = 0 in R3 \Ω, tr w+ − tr w− = Φ and J w+ − J w− = Ψ on ∂Ω, w− (x) = O(|x|−1 ) as |x| → ∞.

(T )

We collect properties of the problems (D± ), (N± ), and (T ) in the following statement.

570

15 Higher Regularity in the Classical Layer Potential Theory

Theorem 15.7.1. Let p ∈ (1, ∞) and α = 1 − {!} − 1/p, where ! is a noninteger with ! > 1. Suppose that ∂Ω ∈ Wp for p(! − 1) > 2 and ∂Ω ∈ Mp (δ) with some δ = δ(p, !) for p(! − 1) ≤ 2. Then (i) For every Φ+ ∈ Wp (∂Ω) there exists a unique solution u+ of (D+ ) in []+1,α

Wp

(Ω). (ii) For every Φ− ∈ Wp (∂Ω) there exists a unique solution u− of (D− ) in

[]+1,α

Wp,loc (Rn \Ω). (iii) For every Ψ+ ∈ Wp−1 (∂Ω)  1 there exists a unique solution v+ of []+1,α

(N+ ) in Wp (Ω), subject to v+ ⊥ 1 on Ω. (iv) For every Ψ− ∈ Wp−1 (∂Ω) there exists a unique solution v− of (N− ) []+1,α

in Wp,loc (Rn \Ω). (v) For every (Φ, Ψ ) ∈ Wp (∂Ω) × Wp−1 (∂Ω) there exists a unique solution []+1,α

(w+ , w− ) ∈ Wp[]+1,α (Ω) × Wp,loc

(Rn \Ω)

of (T ). The proof is essentially the same as that of Theorem 15.1.2. The following two results in the theory of elastic potentials are parallel to Proposition 15.3.2 and Theorem 15.1.1, and can be proved in a similar way. We restrict ourselves to the solution of (D± ) by means of the single layer potential as well as (D± ) and (N± ) by means of the double layer potential. We recall that the Kelvin-Somigliana tensor Γ = Γij 3i,j=1 , where Γij (x) = −


λ + 3µ δ j xi xj  λ+µ i , + 8πµ(λ + 3µ) λ + µ |x| |x|3

is a fundamental solution of the Lam´e system, and we introduce the elastic single layer potential  (Sρ)(x) = Γ (x − ξ) ρ(ξ)ds. ∂Ω

Theorem 15.7.2. Let ∂Ω satisfy the conditions in Theorem 15.7.1. Then  Sρ; ∂ΩWp ≤ c ρ; ∂ΩWp−1 ,

(15.7.1)

 (Sρ)+ ; ΩWp[]+1,α ≤ c ρ; ∂ΩWp−1 ,

(15.7.2)

 (Sρ)− ; B\ΩWp[]+1,α ≤ c(B) ρ, ∂ΩWp−1 ,

(15.7.3)

where (Sρ)± are the restrictions of Sρ to Ω and Rn \Ω, respectively, and B is an arbitrary ball containing Ω.

15.7 Extension to Boundary Integral Equations of Elasticity

571

Let D be the elastic double layer potential defined by  
∂  / ∂Ω. , νξ Γ (x − ξ) χ(ξ) ds, x ∈ J (Dχ)(x) = ∂ξ ∂Ω If u+ = Dχ+ , then χ+ satisfies the integral equation 1 2 χ+

+ Dχ+ = Φ+ on ∂Ω,

(15.7.4)

which is understood in the same sense as in [Ke1] and [Fab]. The solution of (D− ) may be represented as the sum (Dχ)(x) + a Γ (x, 0) + b rot Γ (x, 0), where a and b are unknown constant vectors. The triple (χ, a, b) satisfies the equation − 12 χ− + Dχ− + a Γ (·, 0) + b rot Γ (·, 0) = −Φ− .

(15.7.5)

Representing solutions of problems (N± ) in the form Sχ± , one arrives at the equations − 12 χ+ + D∗ χ+ = Ψ+ , ∗ 1 2 χ− + D χ− = Ψ − ,

(15.7.6) (15.7.7)

where D∗ is the adjoint of D. Here is the main result concerning the integral equations (15.7.4)–(15.7.7). Theorem 15.7.3. Let ∂Ω satisfy the conditions in Theorem 15.7.1. Then (i) the operators D and D∗ are bounded on Wp (∂Ω) and Wp−1 (∂Ω), respectively; (ii) for Φ± ∈ Wp (∂Ω) the equations (15.7.4) and (15.7.5) are uniquely solvable in Wp (∂Ω) and Wp (∂Ω) × R3 × R3 ; (iii) there exists a continuous inverse of 12 I + D∗ on the space Wp−1 (∂Ω). Equation (15.7.6) has a solution in Wp−1 (∂Ω) for an arbitrary Ψ+ orthogonal to all rigid motions. Remark 15.7.1. A straightforward modification of our arguments used in the harmonic potential theory developed in this chapter leads to analogous higher regularity results in the theory of hydrodynamic potentials related to the Stokes system ν∆u − ∇p = 0, div u = 0, (for background, see [Lad], [Ke1], [Fab], and Sects. 2.2–2.4 in [Maz17]).

16 Applications of Multipliers to the Theory of Integral Operators

In this chapter it is shown that Sobolev multipliers are useful for the study of integral operators. First, in Sect. 16.1 we consider an arbitrary convolution operator acting in a pair of weighted L2 -spaces and collect corollaries of the theory of multipliers providing criteria of boundedness and compactness of the convolutions and a characterization of their spectra. Next we turn to classical singular integral operators acting in Sobolev spaces. In Sect. 16.2 a calculus of these operators is developed under the assumption that their symbols belong to classes of multipliers in Sobolev spaces. Finally, in Sect. 16.3 sharp conditions for continuity of the singular integral operators acting from W2m to W2l are found. These conditions are formulated in terms of certain classes of multipliers.

16.1 Convolution Operator in Weighted L2 -Spaces Let K :u→k∗u be a convolution operator with the kernel k. The results of Sects. 3.6 and 4.6 for the case p = 2 can be interpreted as theorems on properties of K considered as a mapping K : L2 ((1 + |x|2 )m/2 ) → L2 ((1 + |x|2 )l/2 ), where uL2 ((1+|x|2 )k/2 ) =




m ≥ l ≥ 0,

|u|2 (1 + |x|2 )k dx

(16.1.1)

1/2 .

For example, the operator K is continuous if and only if its symbol, i.e. the Fourier transform F k, belongs to M (W2m → W2l ). By Theorem 4.1.1 this is l and, for every compact equivalent to the following properties: F k ∈ W2,unif n set e ⊂ R , V.G. Maz’ya, T.O. Shaposhnikova, Theory of Sobolev Multipliers, Grundlehren der mathematischen Wissenschaften 337, c Springer-Verlag Berlin Hiedelberg 2009


573

574

16 Applications of Multipliers to the Theory of Integral Operators

 [D2,l (F k)]2 dx ≤ const C2,m (e) , e

where
 (D2,l u)(x) = if {l} > 0, and

|∇[l],x (u(x + h) − u(x))|2 |h|−n−2{l} dh

1/2

(D2,l u)(x) = |∇l u(x)|

if {l} = 0. Moreover,
D2,l (F k); eL2  . K ∼ sup F kL2,unif + [C2,m (e)]1/2 e In the case 2m > n, K ∼ F kW2,unif . l Theorem 4.6.1 describes properties of a function of the operator K considered as the mapping (16.1.1). Namely, let 0 < l < 1 and let ϕ be a complexvalued function of a complex argument with ϕ(0) = 0. By ϕ(K) we denote the convolution operator with the symbol ϕ(F k). If |ϕ(t + τ ) − ϕ(t)| ≤ A|τ |ρ , where |τ | < 1 and ρ ∈ (0, 1), we derive the following assertion from Theorem 4.6.1. If the operator (16.1.1) is continuous, then the operator ϕ(K) : L2 ((1 + |x|2 )(m−l+r)/2 ) → L2 ((1 + |x|2 )r/2 ) with r ∈ (0, lρ) is continuous as well. Results of Chap. 7 imply two-sided estimates for the essential norm and conditions for compactness of the convolution K: (i) If m > l and 2m ≤ n, then ess K ∼ lim

δ→0




 D2,l (F k); eL2 m−l− n 2 F k; B (x) + sup ρ ρ L 2  1 x∈Rn {e:d(e)≤δ} C2,m (e) 2 ρ≤δ


+ lim

r→∞

sup e⊂Rn \Br d(e)≤1

sup

D2,l (F k); eL2 1

(C2,m (e)) 2

+

sup x∈Rn \Br

 F k; B1 (x)L2 ,

(16.1.2)

where d(e) is the diameter of a compact set e ⊂ Rn . (ii) If m > l and 2m > n, then ess K ∼ lim sup F k; B1 (x)W2l . |x|→∞

Hence K is compact if and only if either m > l, 2m ≤ n, and

(16.1.3)

16.2 Calculus of Singular Integral Operators


lim

δ→0

575

 n D2,l (F k); eL2 sup ρm−l− 2 F k; Bρ (x)L2 = 0,   12 + x∈R n {e:d(e)≤δ} C2,m (e) ρ≤δ sup


lim

r→∞

sup e⊂Rn \Br d(e)≤1

D2,l F k; eL2 (C2,m (e))

1 2

+

sup x∈Rn \Br

 F k; B1 (x)L2 = 0;

or m > l, 2m > n and lim F k; B1 (x)W2l = 0.

|x|→∞

According to Corollary 3.6.1, a complex number λ belongs to the spectrum σ(K) of an operator K, continuous in L2 ((1 + |x|2 )l/2 ), l > 0, if and only if (F k − λ)−1 ∈ / L∞ . Let λ ∈ σ(K). By Theorem 3.6.1, λ is an eigenvalue of K if and only if lim ρ−n C2,l (Bρ (x)\Zλ ) = 0

ρ→0

for all x in a set of positive measure, where Zλ = {ξ ∈ Rn : (F k)(ξ) = λ}. This condition is equivalent to C2,l (G\Zλ ) < C2,l (G)

for some open set G.

(16.1.4)

By the same Theorem 3.6.1, λ belongs to the residual spectrum σr (K) if and only if (16.1.4) does not hold and C2,l (Zλ ) > 0. Furthermore, λ belongs to the continuous spectrum of K if and only if C2,l (Zλ ) = 0. If λ is a point of the spectrum of an operator K, continuous in L2 ((1 + |x|2 )−l/2 ), then Theorem 3.6.1 implies that λ ∈ σp (K) ⇐⇒ C2,l (Zλ ) > 0 and λ ∈ σc (K) ⇐⇒ C2,l (Zλ ) = 0 . Consequently, the convolution acting in L2 ((1 + |x|2 )−l/2 ) has no residual spectrum.

16.2 Calculus of Singular Integral Operators with Symbols in Spaces of Multipliers ˚ Wpl are useful in In this section we demonstrate that the spaces M Wpl and M construction of a calculus of singular integral operators acting in Wpl , 1 < p < ∞, l = 0, 1, . . . . First we quote basic definitions of the theory of singular integrals (see Mikhlin and Pr¨ ossdorf [MiP]).

576

16 Applications of Multipliers to the Theory of Integral Operators

Let α be a bounded measurable function defined on Rn × ∂B1 , orthogonal (0) to one on ∂B1 , and let α0 ∈ L∞ (Rn ). The singular integral operator is defined by  α(x, θ) (0) u(y) dy , x ∈ Rn , (16.2.1) (Au)(x) = α0 (x)u(x) + rn n R where r = |y − x|, θ = (y − x)/r and the integral is interpreted in the sense of the Cauchy principal value. We express α as a series in spherical harmonics α(x, θ) =

km ∞  

(k) αm (x)Ym(k) (θ) ,

(16.2.2)

m=1 k=1 (k)

where km is the number of spherical harmonics Ym of order m. Then (16.2.1) and (16.2.2) imply the formal expansion (Au)(x) =

(0) α0 (x)u(x)

+

km ∞  

 (k) αm (x)

m=1 k=1

(k)

Rn

Ym (θ) u(y) dy . (16.2.3) rn

(0)

We put (S0 u)(x) = u(x), k0 = 0 and  (k) (Sm u)(x) =

(k)

Rn

Ym (θ) u(y) dy . rn

It is known that Sm = µm F −1 Ym F , where F is the Fourier transform in Rn , µ0 = 1 and (k)

µm = i−m π n/2

(k)

Γ (m/2) , Γ ((n + m)/2)

|µm | ∼ m−n/2 ,

(16.2.4)

for m ≥ 1 with Γ standing for the Gamma function. Hence the operator A can be written in the form −1 (Au)(x) = Fξ→x [a(x, ξ/|ξ|)(F u)(ξ)] ,

where a is defined by a(x, θ) =

km ∞  

(k) µm αm (x)Ym(k) (θ)

(16.2.5)

m=0 k=1

and is called the symbol of the singular integral operator A. Next we introduce the space C ∞ (M Wpl , ∂B1 ) of infinitely differentiable ˚ Wpl , ∂B1 ) is functions defined on ∂B1 with range in M Wpl . The space C ∞ (M defined in a similar way.

16.2 Calculus of Singular Integral Operators

577

Lemma 16.2.1. If a ∈ C ∞ (M Wpl , ∂B1 ), then the singular integral operator A with the symbol a is continuous in Wpl and can be expressed as the series km ∞  

(k) (k) αm Sm

m=0 k=1

which converges in the operator norm in Wpl . Proof. From (16.2.4), (16.2.5), and the definition of the space C ∞ (M Wpl , ∂B1 ), it follows that for any positive integer N there exists a constant CN such that (k) ; Rn M Wpl ≤ CN m−N , αm

m ≥ 1. (k)

It remains to make use of the fact that the singular convolution operator Sm is continuous in Wpl and its norm increases no faster than a certain degree of m as m → ∞.  Henceforth, A, B, and C are singular integral operators in Rn with the symbols a(x, θ), b(x, θ), and c(x, θ), where x ∈ Rn and θ ∈ ∂B1 . Theorem 16.2.1. Let AB be a singular operator with the symbol ab and let A ◦ B be the composition of operators A and B. If a ∈ C ∞ (M Wpl , ∂B1 ) and there exists a function b∞ ∈ C ∞ (∂B1 ) such ˚ W l , ∂B1 ), then the operator AB − A ◦ B is compact in that b − b∞ ∈ C ∞ (M p l Wp . Proof. By Lemma 16.2.1, it is sufficient to consider the operators A and B expressed in the form of finite sums   (k) (k) αm Sm , βq(r) Sq(r) . q,r

m,k

It is clear that A ◦ B = F −1 =






 (k) µm αm µq βq(r) Ym(k) Yq(r) F

m,k,q,r (k) (r) αm βq µm F −1 Ym(k) F µq F −1 Yq(r) F

m,k,q,r

=



(k) (r) (k) (r) αm βq Sm Sq .

m,k,q,r

On the other hand,  (k) (k) (r) (r) AB = αm Sm βq Sq m,k,q,r

=



m,k,q,r

(k) (r) (k) (r) αm βq Sm Sq +

 m,k,q,r

(k) (k) αm [Sm , βq(r) ]Sq(r) ,

578

16 Applications of Multipliers to the Theory of Integral Operators

where [X, Y ] = XY − Y X. Therefore,  (k) (k) αm [Sm , βq(r) ]Sq(r) , AB − A ◦ B =

(16.2.6)

m,k,q,r (k)

(r)

and it remains to show that the commutator [Sm , βq ] is compact in Wpl . This assertion is contained in the following lemma. ˚ W l , and let A be a singular integral operator with Lemma 16.2.2. Let γ ∈ M p the symbol a(θ), where a ∈ C ∞ (∂B1 ). Then the commutator [γ, A] is compact in Wpl . Proof. Let {γj } be a sequence of functions, γj ∈ C0∞ , and let γj converge to γ in M Wpl . Then the operators (γ − γj )A and A(γ − γj ) tend to zero in the operator norm in Wpl . The compactness of the mapping [γj , A] in Wpl is well known and can be easily verified.  The following theorem contains conditions for the operator AB − A ◦ B to be of order −1 in Wpl (cf. [KN]). Theorem 16.2.2. If a ∈ C ∞ (M Wpl+1 , ∂B1 ) and ∇x b ∈ C ∞ (M Wpl , ∂B1 ), then the operator AB − A ◦ B maps Wpl continuously into Wpl+1 . Here AB is a singular operator with the symbol ab, and A ◦ B is the composition of operators A and B. This assertion follows from (16.2.6) and the next lemma. Lemma 16.2.3. Let a function γ satisfy the Lipschitz condition and let ∇γ ∈ M Wpl . Further, let A be a singular integral operator with the symbol a(ξ), where a ∈ C ∞ (∂B1 ). Then the commutator [γ, A] satisfies the inequality [γ, A]W l →Wpl+1 ≤ c ∇γM Wpl . p

Proof. For l = 0 the assertion is a known result due to Calderon [Ca2]. Let the lemma be proved for all l = 0, 1, . . . , k − 1. Then, for all u ∈ Wpk+1 , [γ, A]uWpk+1 ≤

n + +  + + ∂ [γ, A]u+ k + [γ, A]uWpk . + ∂x Wp j j=1

(16.2.7)

In view of the imbedding M Wpk ⊂ M Wpk−1 , the last term in (16.2.7) is estimated by the induction hypothesis. Since (∂/∂xj )[γ, A] = (∂γ/∂xj )A − A(∂γ/∂xj ) + [γ, A](∂/∂xj ), it follows that

16.3 Continuity of Singular Integral Operators

579

+ ∂ + ∂γ + + + ∂u + + + + + + + [γ, A]u+ ≤ 2AWpk →Wpk + uWpk + +[γ, A] . + + + ∂xj ∂xj M Wpk ∂xj Wpk Wpk Applying the induction hypothesis to the last norm, we complete the proof.  To conclude this section we formulate two corollaries on the regularization of a singular integral operator which follow from Theorems 16.2.1 and 16.2.2. Corollary 16.2.1. Let there exist a function a∞ ∈ C ∞ (∂B1 ) such that ˚ Wpl , ∂B1 ). a − a∞ ∈ C ∞ (M Further, let c = 1/a ∈ L∞ (Rn × ∂B1 ). Then c ∈ C ∞ (M Wpl , ∂B1 ) and

˚ W l , ∂B1 ), c − c∞ ∈ C ∞ (M p

where c∞ = 1/α∞ . Moreover, the operators A ◦ C − I and C ◦ A − I are compact in Wpl . Corollary 16.2.2. Let a ∈ L∞ (Rn × ∂B1 ) and let ∇x a ∈ C ∞ (M Wpl , ∂B1 ). Further, let c = 1/a ∈ L∞ (Rn × ∂B1 ). Then ∇x c ∈ C ∞ (M Wpl , ∂B1 ) and the operators A ◦ C − I and C ◦ A − I map Wpl continuously into Wpl+1 . Remark 16.2.1. The condition of infinite differentiability of the symbols on ∂B1 can be replaced everywhere in this section by the condition of their sufficient smoothness.

16.3 Continuity in Sobolev Spaces of Singular Integral Operators with Symbols Depending on x Here we give conditions for the boundedness of a singular integral operator, acting from the Sobolev class W2m (Rn ) into W2l (Rn ) with m ≥ l ≥ 0. The symbol may depend not only on the angular variable θ ∈ ∂B but also on the space variable x ∈ Rn . Here ∂B stands for the unit sphere in Rn centered at the origin. It will be shown that the conditions, which are stated in terms of a certain space of multipliers, are sharp.

580

16 Applications of Multipliers to the Theory of Integral Operators

16.3.1 Function Spaces Let µ be a measurable function defined on Rn−1 and satisfying the inequalities   µ(ξ) ≥ c and µ(ξ + η) ≤ 1 + c |ξ|Q µ(η), where c and Q are positive constants. By Hµ (Rn−1 ) we denote the completion of C0∞ (Rn−1 ) in the norm
 1/2 n−1 v; R Hµ = |µ(ξ)(F v)(ξ)|2 dξ , (16.3.1) Rn−1

where F is the Fourier transform in Rn−1 . We obtain the space of Bessel potentials H2l (Rn−1 ), l ∈ R1 , by setting µ(ξ) = (1 + |ξ|2 )l . The space Hµ was introduced and studied in [H2], [VP]. In particular, it was shown in [H1], [VP] that Hµ (Rn−1 ) is embedded into the space C(Rn−1 ) of continuous and bounded functions on Rn−1 if and only if  dξ < ∞. (16.3.2) 2 Rn−1 µ(ξ) Everywhere in this section we assume that (16.3.2) holds. We suppose that µ is weakly subadditive, that is, µ(ξ + η) ≤ c (µ(ξ) + µ(η)),

c = const.

An easy modification of the proof of a similar result for H2l given in [Pe1] shows that the space Hµ (Rn−1 ) is an algebra with respect to pointwise multiplication if µ satisfies (16.3.2). The converse assertion also holds. In fact, since µ(ξ) ≥ c > 0, we have n−1 N Hµ c uN ; Rn−1 L2 ≤ uN ; Rn−1 Hµ ≤ cN 1 u; R

for all u ∈ Hµ (Rn−1 ), where N = 1, 2, . . . and the constants c and c1 do not depend on N . Taking the N -th root and passing to the limit as N → ∞, we arrive at u; Rn−1 L∞ ≤ c1 u; Rn−1 Hµ . Consequently, Hµ (Rn−1 ) ⊂ C(Rn−1 ), which is equivalent to (16.3.2). We supply the sphere ∂B with a structure of the class C ∞ by introducing a family of coordinate neighborhoods {Uk } and a family of diffeomorphisms ϕk : Uk → Rn−1 . Further, let {νk } be a smooth partition of unity on ∂B subordinate to the covering {Uk }. A function σ defined on ∂B belongs to the space Hµ (∂B) if n−1 ) (νk σ) ◦ ϕ−1 k ∈ Hµ (R

for all k. The norm in Hµ (∂B) is introduced by

16.3 Continuity of Singular Integral Operators

σ; ∂BHµ =



n−1 2 (νk σ) ◦ ϕ−1 Hµ k ; R

1/2

581

.

k

Similarly to Hµ (Rn−1 ), the space Hµ (∂B) is an algebra with respect to multiplication if and only if (16.3.2) holds. The same condition is equivalent to the embedding Hµ (∂B) ⊂ C(∂B). Let B(x) denote the unit ball in Rn centered at x, and let B = B(0). We need the space H l,µ (B × ∂B) of functions B × ∂B  (x, θ) → u(x, θ) with the finite norm
 1/2 (∇l u(x, ·); ∂B2Hµ + u(x, ·); ∂B2Hµ )dx B

for integer l ≥ 0, and with the finite norm
  ∇[l],x u(x, ·) − ∇[l],y u(y, ·); ∂B2Hµ B

B

 + B

u(y, ·); ∂B2Hµ dy

dxdy |x − y|n+2{l}

1/2

for noninteger l > 0. Further, we introduce the space H l,µ (Rn × ∂B) of functions Rn × ∂B  (x, θ) → u(x, θ) endowed with the norm u; R × ∂BH l,µ = n


 Rn



2  2  1/2 Dl,µ u(x) + D0,µ u(x) dx ,

where Dl,µ u(x) = ∇l,x u(x, ·); ∂BHµ for {l} = 0, and
 Dl,µ u(x) = ∇[l],x u(x + h, ·)−∇[l],x u(x, ·); ∂B2Hµ Rn

(16.3.3)

dh

1/2

|h|n+2{l}

(16.3.4)

for {l} > 0. We say that a function γ defined on Rn × ∂B belongs to the space of multipliers M (H m,µ → H l,µ ) if γu ∈ H l,µ (Rn ×∂B) for all u ∈ H m,µ (Rn ×∂B). Since the operator H m,µ (Rn × ∂B)  u → γu ∈ H l,µ (Rn × ∂B) is closed, it is bounded. As a norm in M (H m,µ → H l,µ ) we take the norm of the multiplication operator: γ; Rn × ∂BM (H m,µ →H l,µ ) = sup{γu; Rn × ∂BH l,µ : u; Rn × ∂BH m,µ ≤ 1}. We use the notation M H l,µ instead of M (H l,µ → H l,µ ).

582

16 Applications of Multipliers to the Theory of Integral Operators

16.3.2 Description of the Space M (H m,µ → H l,µ ) Here we characterize the space M (H m,µ → H l,µ ). Consider first the case l = 0. Lemma 16.3.1. A function γ defined on Rn × ∂B belongs to the space M (H m,µ → H 0,µ ) if and only if γ ∈ H 0,µ (B(x) × ∂B) for an arbitrary unit ball B(x), and for any compact set e ⊂ Rn γ; e × ∂B2H 0,µ ≤ c C2,m (e), where c is a constant which does not depend upon e. Moreover, the equivalence relation   γ(x, ·); ∂B2 dx 1/2 Hµ e γ; Rn × ∂BM (H m,µ →H 0,µ ) ∼ sup (16.3.5) C2,m (e) e⊂Rn holds. Proof. Necessity. We substitute u(x, θ) := u(x), where u ∈ W2m (Rn ), into the inequality
 1/2 γ(x, ·)u(x, ·); ∂B2Hµ dx ≤ c u; Rn × ∂BH m,µ . Rn




Then

γ(x, ·); ∂B2Hµ |u(x)|2 dx

Rn

1/2

≤ c uW2m .

By Theorems 1.2.2 and 3.1.4, the exact constant in this inequality is equivalent to the right-hand side of (16.3.5). Sufficiency. Since the space Hµ (∂B) is an algebra under the condition (16.3.2), it follows that  γu; Rn × ∂B2H 0,µ ≤ c γ; ∂B2Hµ u; ∂B2Hµ dx =c





Rn−1

j

|µ(ξ)|2

Rn

Rn

−1 2 γ; ∂B2Hµ |F [νj (φ−1 j (ξ))u(x, φj (ξ))]| dxdξ.

Applying Lemma 16.3.1 to the internal integral, one obtains  γ(x, ·); ∂B2Hµ dx n 2 e γu; R × ∂BH 0,µ ≤ c sup C2,m (e) e⊂Rn   
  2 −1 |µ(ξ)|2 |F ∆h ∇[m],x νj (φ−1 × j (ξ))u(x, φj (ξ)) | dξdx Rn

Rn

j



Rn−1



+ Rn

j

Rn−1

  2 −1 |µ(ξ)|2 |F νj (φ−1 j (ξ))u(x, φj (ξ)) | dξdx ,

dh |h|n+2{m}

16.3 Continuity of Singular Integral Operators

583

where ∆h v(x, θ) = v(x + h, θ) − v(x, θ). Hence, using the definition of Hµ (∂B),we arrive at  γ; ∂B2Hµ n 2 e γu; R × ∂BH 0,µ ≤ c sup u; Rn × ∂B2H m,µ . C2,m (e) e⊂Rn 

The proof is complete.

Remark 16.3.1. According to (3.1.14), in Lemma 16.3.1 we may restrict ourselves to compact sets e satisfying diam(e) ≤ 1. In order to obtain sharp two-sided estimates for the norm in M (H m,µ → H l,µ) for m ≥ l > 0 we should prove some auxiliary assertions which are derived in the same way as the corresponding assertions on multipliers in the classes M (H2m (Rn ) → H2l (Rn )) (see Sects. 2.3 and 3.2). When doing this we should replace |γ(x)| by γ(x, ·); ∂B1 Hµ and replace Sl u(x) by Dl,µ u(x) defined by (16.3.3) and (16.3.4). As a result we arrive at the following description of the class M (H m,µ → H l,µ ). Theorem 16.3.1. The equivalence relation holds:  2
(Dl,µ γ(x)) dx 1/2 n e γ; R × ∂BM (H m,µ →H l,µ ) ∼ sup C2,m (e) e:d(e)≤1 ⎧
 ⎪ ⎪ ⎪ ⎨ supn +

x∈R

B(x)

γ(y, ·); ∂B2Hµ dy

1/2 m > l, (16.3.6)

⎪ ⎪ ⎪ ⎩ lim sup γ(x, ·); ∂BHµ

m = l.

x∈Rn

The restriction d(e) ≤ 1 can be omitted. Remark 16.3.2. In the same way as for the space M (W2m → W2l ) (cf. Sect. 4.3.4) we can check that M (H m,µ → H l,µ ) is continuously embedded into M (H m−l,µ → H 0,µ ). Since the spaces H m,µ (Rn × ∂B) form an interpolation scale in m (see, for instance, [Tr3], Sec.1.18.5), we have for any j ∈ [0, l] γ; Rn × ∂BM (H m−j,µ →H l−j,µ ) (l−j)/l

j/l

≤ c γ; Rn × ∂BM (H m,µ →H l,µ ) γ; Rn × ∂BM (H m−l,µ →H 0,µ ) .

(16.3.7)

584

16 Applications of Multipliers to the Theory of Integral Operators

The embedding M (H m,µ → H l,µ ) ⊂ M (H m−l,µ → H 0,µ ) together with (16.3.7) implies that the space M (H m,µ → H l,µ ) is continuously embedded into M (H m−j,µ → H l−j,µ ). From this and Theorem 16.3.1 it follows that (16.3.6) is equivalent to γ; Rn × ∂BM (H m,µ →H l,µ )   2 [l] [l] (Dl−j,µ γ(x)) dx  (Dj,µ γ(x))2 dx 1/2
 e e + ∼ sup . C2,m−j (e) C2,m−l+j (e) e:d(e)≤1 j=0 j=0

(16.3.8)

For m = l the term corresponding to j = 0 in the second sum should be replaced by (16.3.9) ess sup γ(x, ·); ∂B2Hµ . x∈Rn

Clearly, for integer l both sums in (16.3.8) coincide. The restriction d(e) ≤ 1 can be omitted. Duplicating the proof of Corollary 4.3.8, we arrive at the following assertion. Corollary 16.3.1. For 2m > n

∼ sup

x∈Rn


 B(x)

γ; Rn × ∂BM (H m,µ →H l,µ )  1/2 (Dl,µ γ(y))2 dy + γ(y, ·); ∂B2Hµ dy .

(16.3.10)

B(x)

One can verify directly that the right-hand side of (16.3.10) is equivalent to the norm γ; B × ∂BH l,µ . From Theorem 16.3.1 upper estimates for the norm in M (H m,µ → H l,µ ) can be obtained, using the lower estimates for the capacity of a compact set stated in terms of its Lebesgue measure mesn (see Propositions 3.1.2 and 3.1.3). Corollary 16.3.2. For 2m < n c γ; Rn × ∂BM (H m,µ →H l,µ )
 1/2 (Dl,µ γ(x))2 dx e ≤ sup 1 m (mesn e) 2 − n e:d(e)≤1
 1/2 + sup γ(y, ·); ∂B2Hµ dy . x∈Rn

B(x)

For 2m = n c γ; Rn × ∂BM (H m,µ →H l,µ )

(16.3.11)

16.3 Continuity of Singular Integral Operators

≤




1/2 2n (Dl,µ γ(x))2 dx mesn e e:d(e)≤1 e 
1/2 + sup γ(y, ·); ∂B2Hµ dy . sup

585

1/2


log

(16.3.12)

B(x)

x∈Rn

In the case m = l the second term on the right-hand sides of (16.3.11) and (16.3.12) should be replaced by (16.3.9). 16.3.3 Main Result Let σ be a measurable function on Rn with values in L2 (∂B). For any u ∈ C0∞ (Rn ) we define the singular integral operator S with symbol σ by the equality −1 [σ(x, ξ/|ξ|)(Fu)(ξ)], (16.3.13) Su(x) = Fξ→x where F is the Fourier transform in Rn and F −1 is its inverse. In what follows we use the notation
 dτ 1/2 . K= 2 Rn−1 µ(τ )

(16.3.14)

As before, we omit Rn in notations of spaces and norms. Theorem 16.3.2. Let K < ∞ and let σ ∈ M (H m,µ → H l,µ ),

m ≥ l ≥ 0.

(16.3.15)

Then the operator (16.3.13) maps W2m continuously into W2l . Moreover, the estimate (16.3.16) SW2m →W2l ≤ c KσM (H m,µ →H l,µ ) holds. Proof. Let x, ξ ∈ Rn , θ = ξ/|ξ| and let u be an arbitrary function from C0∞ (Rn ). We write the operator S as   ∞ Su(x) = e2πixξ σ(x, θ)Fu(ξ)|ξ|n−1 d|ξ|dθ ∂B

0

or, briefly,

 Su(x) =

σ(x, θ)v(x, θ)dθ,

(16.3.17)

e2πixξ Fu(ξ)|ξ|n−1 d|ξ|

(16.3.18)

∂B



where

∞

v(x, θ) = 0



and Fu(ξ) =

Rn

e−2πiyξ u(y)dy.

586

16 Applications of Multipliers to the Theory of Integral Operators

Using the C ∞ structure on ∂B introduced above, we have  −1 −1 Su(x) = νk (ϕ−1 k (t))σ(x, ϕk (t))v(x, ϕk (t))|Jk (t)|dt, k

Rn−1

∞ where Jk is the Jacobian of the mapping ϕ−1 k . Let ηk ∈ C0 (Uk ) be such that ηk νk = νk . We put −1 σk (x, t) = νk (ϕ−1 k (t))σ(x, ϕk (t))

and

−1 vk (x, t) = ηk (ϕ−1 k (t))v(x, ϕk (t))|Jk (t)|.

By Parseval’s theorem, 

Su(x) =

k

=

 Rn−1

k

Rn−1

σk (x, t)vk (x, t)dt

F σk (x, τ )F −1 vk (x, τ )dτ,

(16.3.19)

where F is the Fourier transform in Rn−1 . By (16.3.18), we obtain  −1 F −1 vk (x, τ ) = e−2πiτ t ηk (ϕ−1 k (t))v(x, ϕk (t))|Jk (t)|dt Rn−1



e−2πiτ ϕk (θ) ηk (θ)v(x, θ)dθ

=  = Rn

∂B

e2πixξ ηk (θ)e−2πiτ ϕk (θ) Fu(ξ)dξ.

(16.3.20)

The last integral can be interpreted as a family of singular integral convolution operators Ek (τ ), depending on a parameter τ ∈ Rn−1 , with symbols ηk (θ)e−2πiτ ϕk (θ) ,

k = 1, 2, . . .

Now, it follows from (16.3.19) and (16.3.20) that S can be represented in the form  F σk (x, τ )Ek (τ )u(x)dτ. (16.3.21) Su(x) = k

Rn−1

Let l be a noninteger and let
 |∆h ∇[l] w(x)|2 Dl w(x) = Rn

dh |h|n+2{l}

1/2 .

16.3 Continuity of Singular Integral Operators

587

We have |Dl Su(x)|2 ≤c

[l]  
  Rn

j=0 k





+

Rn−1

|F ∇j,x σk (x+h, τ )| |∆h ∇[l]−j,x Ek (τ )u(x)|dτ



Rn

|F ∆h ∇[l]−j,x σk (x, τ )| |∇j,x Ek (τ )u(x)|dτ

Rn−1

2

2

dh |h|n+2{l} 

dh |h|n+2{l}

.

The right-hand side does not exceed 

[l]  


c

Rn

j=0 k

Rn−1

|µ(τ )F ∇j,x σk (x + h, τ )|2 dτ

 ×

|∆h ∇[l]−j,x Ek (λ)u(x)|2

Rn−1





+ Rn

 ×

Rn−1

Rn−1

dλ dh µ(λ)2 |h|n+2{l}

|µ(τ )F ∆h ∇[l]−j,x σk (x, τ )|2 dτ

|(∇j,x Ek (λ)u(x)|2

dλ dh  . µ(λ)2 |h|n+2{l}

(16.3.22)

Consequently, Dl Su2L2 ≤c



[l]  
 Rn

j=0 k

∇j,x σk (x, ·); Rn−1 2Hµ






+ Rn

Rn


 ×

Rn−1

|(Dl−j Ek (λ)u)(x)|2

Rn−1

|(∇j Ek (λ)u)(x)|2

dλ  µ(λ)

2

2 dx

µ(λ)



dh

∆h ∇[l]−j,x σk (x, ·); Rn−1 2Hµ

dλ

|h|n+2{l}

 dx .

This inequality and Lemma 16.3.1 imply that Dl Su2L2   
[l]

≤c

sup e

j=0 k

e

∇j,x σ(x, ·); Rn−1 2Hµ dx  C2,m−l+j (e)

Rn−1



+ sup e

e

Dl−j σk (x, ·); Rn−1 2Hµ dx  C2,m−j (e)

Rn−1

Ek (λ)u2W2m

Ek (λ)u2W2m

dλ µ(λ)2

dλ  . µ(λ)2

Since the operators Ek (λ) are uniformly bounded in W2m , it follows that Dl SuL2 does not exceed

588

16 Applications of Multipliers to the Theory of Integral Operators


 K sup

 (Dj,µ γ(x))2 dx

[l]

e

j=0

e

C2,m−l+j (e)

 [l]

+

e

(Dl−j,µ γ(x))2 dx 1/2 C2,m−j (e)

j=0

uW2m ,

which together with Remark 16.3.2 gives Dl SuL2 ≤ c KσM (H m,µ →H l,µ ) uW2m .

(16.3.23)

For integer l the proof is similar and somewhat easier. In particular, the counterpart of (16.3.22) is c

l   j=0 k

Rn−1

 |µ(τ )F ∇j,x σk (x, τ )|2 dτ

Rn−1

|∇l−j,x Ek (λ)u(x)|2

dλ . µ(λ)2

Duplicating the above arguments, we arrive at an analogue of (16.3.23) with Dl replaced by ∇l on the left-hand side. This together with the inequality SuL2 ≤ c KσM (H m,µ →H 0,µ ) uW2m , corresponding to l = 0, completes the proof. Remark 16.3.3. We show that Theorem 16.3.2 is sharp in a sense. Let the symbol of S have the form a(x)b(θ), where x ∈ Rn and θ ∈ ∂B, and let b ∈ Hµ (∂B) and |b(θ)| ≥ const > 0. Clearly, the mapping S : W2m → W2l is continuous if and only if the operator of multiplication by a is a continuous operator from W2m into W2l . In other words, (16.3.15) follows from the continuity of S. Now let S be the operator (16.3.13) with symbol b(θ), where θ ∈ ∂B. Its continuity from W2m into W2l is equivalent to the inequality |b(θ)|(1 + |ξ|2 )(l−m)/2 ≤ const which gives the boundedness of b. Therefore, if the operator S : W2m → W2l is continuous for any b ∈ Hµ (∂B), then Hµ (∂B) ⊂ L∞ (∂B), which is equivalent to K < ∞. 16.3.4 Corollaries Now, we give sufficient conditions for the continuity of the operator S : W2m → W2l , which follow from Theorem 16.3.2 and from either necessary and sufficient or sufficient conditions for a function to belong to M (H m,µ → H l,µ ) (see Sect. 16.3.2). The next assertion is a direct corollary of Theorems 16.3.1 and 16.3.2.

16.3 Continuity of Singular Integral Operators

589

Corollary 16.3.3. The estimate (16.3.16) is equivalent to   2 Dl,µ σ(x) dx
e SW2m →W2l ≤ c K sup n C2,m (e) {e⊂R :d(e)≤1}  + sup B(x)

x∈Rn

σ(y, ·); ∂B2Hµ dy

1/2 (16.3.24)

for m > l ≥ 0. For m = l the second term on the right-hand side of (16.3.24) should be replaced by (16.3.25) lim sup σ(x, ·); ∂B2Hµ . x∈Rn

Theorem 16.3.2 and Corollary 16.3.1 imply the following assertion. Corollary 16.3.4. Let 2m > n. The inequality (16.3.16) is equivalent to SW2m →W2l ≤ c K sup σ; B(x) × ∂BH l,µ .

(16.3.26)

x∈Rn

Combining Theorem 16.3.2 with Corollary 16.3.2 we can remove the capacity from inequality (16.3.24) as follows. Corollary 16.3.5. Let 2m < n. Then  SW2m →W2l ≤ c K


sup e⊂Rn :diam(e)≤1

 2 Dl,µ σ(x) dx

e

(mesn e)1−2m/n

 + sup x∈Rn

B(x)

σ(y, ·); ∂B2Hµ dy

1/2 .

(16.3.27)

For 2m = n the expression (mesn e)1−2m/n should be replaced by 

−1 log(2n /mesn e) .

In the case m = l the second term on the right-hand side of (16.3.27) should be replaced by (16.3.25). Remark 16.3.4. Theorem 16.3.2 and its corollaries can be directly extended to classical pseudo-differential operators with symbols of the form ζ(ξ)

N 

σk (x, ξ/|ξ|)|ξ|rk ,

k=1

where r1 > · · · > rN and ζ ∈ C ∞ (Rn−1 ) with ζ(ξ) = 1 for |ξ| > 2 and ζ(ξ) = 0 for |ξ| < 1 (see [KN] for a theory of these operators).

References

[Ad1] [Ad2] [Ad3] [AF] [AH] [AM]

[APo] [AX] [ADN1]

[ADN2]

[AB] [AN] [Am] [And] [AMS]

D.R. Adams, A trace inequality for generalized potentials, Studia Math., 48:1 (1973), 99-105. D.R. Adams, A note on Riesz potentials, Duke Math. J., 42:4 (1975), 99-105. D.R. Adams, On the existence of capacitary strong type estimates in Rn , Ark. Mat., 14 (1976), 125-140. D.R. Adams and M. Frazier, Composition operators on potential spaces, Proc. Amer. Math. Soc., 114 (1992), 155-165. D.R. Adams and L.-I. Hedberg, Function Spaces and Potential Theory, Springer, 1996. D.R. Adams and N.G. Meyers, Bessel potentials. Inclusion relations among classes of exceptional sets, Indiana Univ. Math. J., 22:9 (1973), 873-905. D.R. Adams and J.C. Polking, The equivalence of two definitions of capacity, Proc. Amer. Math. Soc., 37 (1973), 529-534. D.R. Adams and J. Xiao, Strong type estimates for homogeneous Besov capacities, Math. Ann., 325 (2003), 695-709. S. Agmon, A. Douglis, and L. Nirenberg, Estimates near the boundary for the solutions of elliptic equations satisfying general boundary values, I., Comm. Pure Appl. Math., 12 (1959), 623-727. S. Agmon, A. Douglis, and L. Nirenberg, Estimates near the boundary for the solutions of elliptic equations satisfying general boundary values, II., Comm. Pure Appl. Math., 17 (1964), 35-92. L. Ahlfors and A. Beurling, Conformal invariants and function-theoretic null-sets, Acta Math. 83 (1950), 623-727. F. Ali Mehmeti and S. Nicaise, Banach algebras of functions on nonsmooth domains, Oper. Theory Adv. Appl., 102, Birkh¨ auser, Basel, 1998, 11-20. H. Amann, Multiplication in Sobolev and Besov spaces, Nonlinear Analysis, Scuola Normale Superiore, Pisa, 1991, 27-50. K.F. Andersen, Weighted inequalities for convolutions, Proc. AMS, 123:4 (1995), 1129-1136. N. Aronszajn, F. Mulla, and P. Szeptycki, On spaces of potentials connected with Lp -spaces, Ann. Inst. Fourier, 13 (1963), 211-306.

591

592

References

B.M. Benchekroun, A. Benkirane, Sur l’alg`ebre d’Orlicz-Sobolev, Bull. Belg. Math. Soc., 2:4 (1995), 463-476. [BG] C. Bennet and J.E. Gilbert, Homogeneous algebras on the circle: II. Multipliers, Ditkin conditions, Ann. Inst. Fourier, 22:3 (1972), 21-50. [Bes] O.V. Besov, Investigation of a family of function spaces in connection with imbedding and extension theorems, Trudy Mat. Inst. Steklov, 60 (1961), 42-81. [BIN] O.V. Besov, V.P. Il’in, and S.M. Nikol’skii, Integral Representations of Functions and Imbedding Theorems, Vol I, 1978, and Vol. II, 1979, John Wiley & Sons, New York-Toronto-London. [Beu] A. Beurling, Construction and analysis of some convolution algebras, Ann. Inst. Fourier (Grenoble), 14 (1964), 1-32. [Bl1] N.K. Bliev, On products of functions in Nikolskii-Besov spaces, Izv. AN Kazach. SSR, Ser. Phis.-Mat., no. 5 (1979), 69-71. [Bl2] N.K. Bliev, Homeomorphisms of Beltrami equation in fractional spaces, Differential and integral equations. Boundary value problems, Tbilisi, 1979, 33-43. [Blo] S. Bloom, Pointwise multipliers of weighted BMO spaces, Proc. Amer. Math. Soc., 105 (1989), 950-960. [Bo] G. Bourdaud, Localizations des espaces de Besov, Studia Math., 90 (1988), 153-163. [Bur] V. Burenkov, Sobolev Spaces on Domains, Teubner-Texte zur Mathematik, 137. B. G. Teubner, Stuttgart, Leipzig, 1998. [Ca1] A.P. Calderon, Lebesgue spaces of differentiable functions and distributions, Proc. Sympos. Pure Math., 4 (1961), 33-49. [Ca2] A.P. Calderon, Commutators of singular integral operators, Proc. Nat. Acad. Sci. USA, 53 (1965), 1092-1099. [Ca3] A.P. Calderon, Algebra of singular integral operators, Proc. Symp. Pure Math., 10, AMS, Providence, R.I., 1967. [Ca4] A.P. Calderon, Boundary value problems for the Laplace equation in Lipschitz domains, Recent progress in Fourier Analysis, Sci. Publ., Amsterdam, 1985, 33-48. [Cam] S. Campanato, Propriet` a di h¨ olderianit` a di alcune classi di funzioni, Ann. Scuola Norm. Sup. Pisa, 17 (1963), 175-188. [Car] L. Carleson, Interpolation by bounded analytic functions and the corona problem, Ann. Math., 76 (1962), 547-559. [COV1] C. Cascante, J.M. Ortega, and I.E. Verbitsky, Nonlinear potentials and two weight trace inequalities for general dyadic and radial kernels, Indiana Univ. Math. J., 53 (2004), 845-882. [COV2] C. Cascante, J.M. Ortega, and I.E. Verbitsky, On Lp –Lq trace inequalities, J. London Math. Soc., 74:2 (2006), 497-511. [ChWW] S.-Y. A. Chang, J. M. Wilson, and T. H. Wolff, Some weighted norm inequalities concerning the Schr¨ odinger operators, Comment. Math. Helv., 60 (1985), 217-246. [CF] R. R. Coifman and C. Fefferman, Weighted norm inequalities for maximal functions and singular integrals, Studia Math., 51 (1974), 241-250. [CMM] R. R. Coifman, A. McIntosh, and I. Meyer, L’ int´egrale de Cauchy d´efinit un op´erateur born´e sur L2 pour les courbes Lipschitziennes, Ann. of Math., 116 (1982), 361-387. [BB]

References [Cor]

[Cos] [DM1] [DKV]

[Dav]

[dR] [DH] [DM2]

[DS] [EE] [ES]

[Fab]

[FJR] [FKV] [Fe1] [Fe2] [Fe3] [F1] [F2] [Fil]

[FR] [Fra] [FrS]

593

H.O. Cordes, Die erste Randwertaufgabe bei Differentialgleichungen zweiter Ordnung in mehr als zwei Variabeln, Math. Ann., 131:3 (1956), 278-312. M. Costabel, Boundary integral operators on Lipschitz domains: elementary results, SIAM J. Math. Anal., 19:3 (1988), 613-623. B. Dacorogna and J. Moser, On a partial differential equation involving the Jacobian determinant, Ann. Inst. Henri Poincar´e, 7 (1991), 1-26. B.E.J. Dahlberg, C.E. Kenig, and G.C. Verchota, Boundary value problems for the systems of elastostatics in Lipschitz domains, Duke Math. J., 57:3 (1988), 795-818. E.B. Davies, A review of Hardy inequalities, The Maz’ya Anniversary Collection, Eds. J. Rossmann, P. Tak´ aˇc, and G. Wildenhain, Operator Theory: Advances and Applications, Vol. 110, Birkh¨ auser, 1999, 55-67, Basel–Boston–Berlin. G. de Rham, Vari´et´es Diff´erentiables, Hermann, Paris, 1960. A. Devinatz and I.I. Hirschman, Multiplier transformations on l2,α , Annals of Math., 69:3 (1959), 575-587. D. Drihem and M. Moussai, On the pointwise multiplication in Besov and Lizorkin-Triebel spaces, Int. J. Math. Math. Sci. Art. ID 76182 (2006), 1-18. N. Dunford and J.T. Schwartz, Linear Operators. Part I: General Theory, Interscience Publishers, 1967. D.E. Edmunds and W.D. Evans, Spectral Theory and Differential Operators, Clarendon Press, Oxford, 1987. D.E. Edmunds and E. Shargorodsky, The inner variation of an operator and the essential norm of pointwise multipliers in function spaces, Houston J. Math., 31:3 (2005), 841-855. E.B. Fabes, Boundary value problems of linear elastostatics and hydrostatics on Lipschitz domains, Proc. Cent. Math. Anal. Aust. Nat. Univ., 9 (1985), 27-45. E.B. Fabes, M. Jodeit, and N.M. Riviere, Potential techniques for boundary value problems in C 1 domains, Acta Math., 141:3-4 (1978), 165-186. E.B. Fabes, C.E. Kenig, and G.C. Verchota, The Dirichlet problem for the Stokes system on Lipschitz domains, Duke Math. J., 57:3 (1988), 769-793. H. Federer, Curvature measures, Trans. AMS, 93:3 (1959), 418-491. H. Federer, The area of nonparametric surface, Proc. AMS, 11:3 (1960), 436-439. H. Federer, Geometric Measure Theory, Springer, 1969. C. Fefferman, Characterizations of bounded mean oscillation, Bull. AMS, 77 (1971), 587-588. C. Fefferman, The uncertainty principle, Bull. AMS, 9 (1983), 129-206. N. Filonov, Principal singularities of the magnetic field component in resonators with boundary of a given class of smoothness, Algebra i Analiz, 9:2 (1997), 241-255. W.H. Fleming and R.W. Rishel, An integral formula for total gradient variation, Arch. Math., 11:3 (1960), 218-222. L.E. Fraenkel, Formulae for high derivatives of composite functions, Math. Proc. Camb. Soc., 77 (1971), 587-588. R.L. Frank and R. Seiringer, Non-linear ground state representations and sharp Hardy inequalities, arXiv:0803.0503.

594 [Fr] [FrJ] [Gag1] [Gag2] [GSh] [Ger]

[GG] [GM] [GR]

[Gu1] [Gu2]

[Gus] [Guz]

[Ha]

[Hed1] [Hed2] [Hed3] [Her] [Hi1] [Hi2] [H1] [H2] [Ja1]

References s J. Franke, On the spaces Fp,q of Triebel-Lizorkin type: Pointwise multipliers and spaces on domains, Math. Nachr., 125 (1986), 29-68. M. Frazier and B. Jawerth, A discrete transform and decompositions of distribution spaces, J. Funct. Analysis, 93 (1990), 34-170. E. Gagliardo, Propriet` a di alcune classi di funzioni in pi` u variabili, Ric. Mat., 7 (1958), 102-137. E. Gagliardo, Ulteriori propriet` a di alcune classi di funzioni in pi` u variabili, Ric. Mat., 8:1 (1959), 24-51. I.M. Gelfand and G.E. Shilov, Generalized Functions, Vol. 1, Operators on them, Academic Press, NY, 1964. P. Germain, Multipliers, paramultipliers, and weak-strong uniqueness for the Navier-Stokes equations, J. Differential Equations, 226:2 (2006), 373-428. G. Geymonat and P. Grisvard, Problemi ai limiti lineari ellittici negli spazi di Sobolev con peso, Matematiche (Catania), 22 (1967), 212-249. L. Grafakos and C. Morpurgo, A Selberg integral formula and applications, Pacific J. Math., 191:1 (1999), 85-94. V. Gol’dshtein, and Yu.G. Reshetnyak, Quasiconformal Mappings and Sobolev Spaces, Translated and revised from the 1983 Russian original. Mathematics and its Applications (Soviet Series), vol. 54. Kluwer Academic Publishers, Dordrecht, 1990. A. Gulisashvili, Multipliers in Besov spaces, Zapiski Nauchn. Sem. LOMI, 135 (1984), 36-50. A. Gulisashvili, Multipliers in Besov spaces and traces of functions on subspaces of Euclidean spaces, Dokl. Akad. Nauk SSSR, 281:4 (1985), 777-781; English translation: Soviet Math. Dokl., 31:2 (1985), 332-336. W. Gustin, Boxing inequalities, J. Math. Mech., 9 (1960), 229-239. M. de Guzman, Covering lemma with applications to differentiability of measures and singular integral operators, Studia Math., 34:3 (1970), 299-317. B. Hanouzet, Applications bilin´eaires compatibles avec un syst`eme ` a coefficients variables. Continuit´e dans les ´espaces de Besov, Comm. Partial. Diff. Eq., 10:4 (1985), 433-465. L.-I. Hedberg, On certain convolution inequalities, Proc. AMS, 36 (1972), 505-510. L.-I. Hedberg, Nonlinear potentials and approximation in the mean by analytic functions, Math. Zeitschr., 129 (1972), 299-319. L.-I. Hedberg, Approximation in the mean by solutions of elliptic equations, Duke Math. J., 40 (1973):1, 9-16. C.S. Herz, Lipschitz spaces and Bernstein’s theorem on absolutely convergent Fourier transforms, J. Math. Mech., 18:4 (1968), 283-323. I.I. Hirschman, On multiplier transformations, II, Duke Math. J., 28 (1961), 45-56. I.I. Hirschman, On multiplier transformations, III, Proc. AMS, 13 (1962), 851-857. L. H¨ ormander, Linear Partial Differential Operators, Springer, 1963. L.H¨ ormander, The Analysis of Linear Partial Differential Operators, vol.2, Springer, 1983. S. Janson, On functions with conditions on the mean oscillation, Ark. Mat., 14:2 (1976), 189-196.

References [Ja2] [JK1] [JK2] [JN] [Jo] [Kal] [K1]

[K2] [K3]

[Ka1] [Ke1]

[Ke2]

[KeS]

[KoS]

[KN] [KZPS] [KP] [Kr] [KWh] [Lad] [Las]

595

S. Janson, Mean oscillation and commutators of singular integral operators, Ark. Mat. 16:2 (1978), 263-270. D.S. Jerison and C.E. Kenig, The Dirichlet problem in nonsmooth domains, Ann. of Math., 113 (1981), 367-382. D.S. Jerison and C.E. Kenig, The Neumann problem on Lipschitz domains, Bull. AMS, 4 (1981), 203-207. F. John and L. Nirenberg, On functions of bounded mean oscillation, Comm. Pure Appl. Math., 14 (1961), 415-426. J. Johnsen, Pointwise multiplication of Besov and Triebel-Lizorkin spaces, Math. Nachr., 175 (1995), 85-133. A. Kalamajska, Pointwise interpolative inequalities and Nirenberg type estimates in weighted Sobolev spaces, Studia Math., 108:3 (1994), 275-290. G.A. Kalyabin, Conditions for multiplicative property of Besov and Lizorkin-Triebel function spaces, Dokl. Akad. Nauk SSSR, 251:1 (1980), 25-26. G.A. Kalyabin, Descriptions of functions in classes of Besov-TriebelLizorkin type, Trudy Math. Inst. Steklov, 156 (1980), 82-109. G.A. Kalyabin, Criteria of the multiplication property and the embedding in C of spaces of Besov-Triebel-Lizorkin type, Mat. Zametki, 30 (1981), 517-526. T. Kato, Schr¨ odinger operators with singular potentials, Israel J. Math., 13 (1972), 135-148. C.E. Kenig, Boundary value problems of linear elastostatics and hydrostatics on Lipschitz domains, Semin. Goulaouic-Meyer-Schwartz, Equation Deriv. Partielles 1983-1984, Exp. N 21, 1-12. C.E. Kenig, Harmonic analysis techniques for second order elliptic boundary value problems, CBMS Regional Conference Series in Mathematics, 83, AMS, Providence, 1994. R. Kerman and E. Sawyer, The trace inequality and eigenvalue estimates for Schr¨ odinger operators, Ann. Inst. Fourier (Grenoble), 36 (1986), 207-228. H. Koch and W. Sickel, Pointwise multipliers of Besov spaces of smoothness zero and spaces of continuous functions, Rev. Mat. Iberoamericana, 18 (2002), 587-626. J.J. Kohn and L. Nirenberg, An algebra of pseudo-differential operators, Comm. Pure Appl. Math., 18:1-2 (1965), 269-305. M.A. Krasnoselskii, P.P. Zabreyko, E.I. Pustylnik, P.E. Sobolevskii, Integral Operators in Spaces of Summable Functions, Noordhoff, Leiden, 1976. S.G. Krantz and H.R. Parks, The Implicit Function Theorem. History, Theory, and Applications, Birkh¨ auser, 2002. A.S. Kronrod, On functions of two variables, Usp. Mat. Nauk, 5:1 (1950), 24-134. D.S. Kurtz and R.L. Wheeden, Results on weighted norm inequalities for multipliers, Trans. AMS, 255 (1979), 343-362. O.A. Ladyzhenskaya, The Mathematical Theory of Viscous Incompressible Flow, Gordon and Breach, 1969. I. Lasiecka, Finite-dimensional attractors of weak solutions to von Karman plate model, J. Math. Systems, Estimation, and Control, 7:3 (1997), 251275.

596 [LR] [LRM]

[LeL]

[LeM] [Lew] [LL] [LiM1] [LiM2]

[Liz] [Mal]

[MaMi] [MMP] [MaPa]

[Mar1] [Mar2] [Mar3] [MM] [Maz1] [Maz2] [Maz3]

[Maz4]

References P.G. Lemari´e-Rieusset, Recent Developments in the Navier-Stokes Problem, Chapman and Hall, Research Notes in Math. 431 (2002). P.G. Lemari´e-Rieusset and R. May, Uniqueness for the Navier-Stokes equations and multipliers between Sobolev spaces, Nonlinear Anal., 66:4 (2007), 819-838. J. Leray and J.-L. Lions, Quelques r´esultats de Viˇsik sur les probl`emes elliptiques non-lineaires par les m´ethodes de Minty-Browder, Bull. Math. Soc. France, 93 (1965), 97-107. A.V. Levin and V. Maz’ya, Asymptotics of densities of harmonic potentials near the vertex of a cone, Z. Anal. Anwend., 8:6 (1989), 501-514. J. Lewis, Uniformly fat sets, Trans. AMS, 308 (1988), 177-196. E.H. Lieb and M. Loss, Analysis, Second Edition, AMS, Providence, RI, 2001. J.-L. Lions and E. Magenes, Probl`emes aux limites non homog` enes, IV, Ann. Scuola Norm. Sup. Pisa, 15 (1961), 311-326. J.-L. Lions and E. Magenes, Non-homogeneous Boundary Value Problems and Applications, Vol. I. Die Grundlehren der Mathematischen Wissenschaften, Band 181. Springer, 1972. P.I. Lizorkin, On function characteristics of interpolation spaces (Lp (Ω), Wp1 (Ω))θ,p , Trudy Mosk. Matem. Inst., 134 (1975), 180-203. J. Mal´ y, Sufficient conditions for change of variables in integral, Proceedings on Analysis and Geometry. International conference in honor of the 70th birthday of Professor Yu. G. Reshetnyak, Novosibirsk, Russia, August 30-September 3, 1999. Novosibirsk: Izdatel’stvo Instituta Matematiki Im. S. L. Soboleva SO RAN. 370-386 (2000). M. Marcus and V. Mizel, Absolute continuity on tracks and mappings of Sobolev spaces, Arch. Rat. Mech. Anal., 45:4 (1972) 294-320. M. Marcus, V. Mizel, and Y. Pinchover, On the best constant for Hardy’s inequality in Rn , Trans. AMS, 350 (1998), 3237-3255. F. Marchand and M. Paicu, Remarques sur l’unicit´e pour le syst´eme de Navier-Stokes tridimensionnel, C. R. Math. Acad. Sci. Paris, 344:6 (2007), 363-366. J. Marschall, Some remarks on Triebel spaces, Studia Math., 87 (1987), 79-92. J. Marschall, On the boundedness and compactness of nonregular pseudodifferential operators, Math. Nachr., 175 (1995), 231-262. J. Marschall, Remarks on nonregular pseudo-differential operators, Z. Anal. Anwendungen, 15 (1996), 109-148. S. Mayboroda and M. Mitrea, Sharp estimates for Green potentials on non-smooth domains, Math. Res. Lett., 11:4 (2004), 481-492. V.G. Maz’ya, Classes of domains and embedding theorems for functional spaces, Dokl. Akad. Nauk SSSR, 133 (1960), 527-530. V.G. Maz’ya, On the theory of the n-dimensional Schr¨ odinger operator, Izv. Akad. Nauk SSSR, ser. Matem., 28 (1964), 1145-1172 (Russian). V.G. Maz’ya, On certain integral inequalities for functions of many variables, Probl. Math. Anal., 3, Leningrad Univ. (1972), 33-68. English translation: J. Soviet Math., 1 (1973), 205-234. V.G. Maz’ya, Weak solutions of the Dirichlet and Neumann problems, Trudy Mosk. Matem. Obsh., 20 (1969), 137-172.

References [Maz5] [Maz6]

[Maz7]

[Maz8] [Maz9] [Maz10]

[Maz11]

[Maz12] [Maz13] [Maz14]

[Maz15] [Maz16]

[Maz17] [Maz18] [MH1]

[MH2] [MN] [MP] [MSh1] [MSh2] [MSh3]

597

V.G. Maz’ya, The degenerate problem with oblique derivative, Mat. Sb., 87 (1972), 417-454. V.G. Maz’ya, The removable singularities of bounded solutions of quasilinear elliptic equations of arbitrary order, Zap. Nauchn. Sem. LOMI, 27 (1972), 116-130. English translation: J. Math. Sci., 3:4 (1975), 480-492. V.G. Maz’ya, The (p, l)-capacity, embedding theorems, and the spectrum of a selfadjoint elliptic operator, Izv. Akad. Nauk SSSR, ser. Matem., 37 (1973), 356-385. V.G. Maz’ya, On the local square summability of convolution, Zap. Nauchn. Sem. LOMI, 73 (1977), 211-216. V.G. Maz’ya, On capacitary strong type estimates for fractional norms, Zap. Nauchn. Sem. LOMI, 73 (1977), 161-168. V.G. Maz’ya, Multipliers in Sobolev spaces. In the book: Application of function theory and functional analysis methods to problems of mathematˇ ical physics. Pjatoe Sovetso-Cehoslovackoe Soveˇsˇcanie, 1976, Novosibirsk, 1978, 181-189. V.G. Maz’ya, On summability with respect to an arbitrary measure of functions in Sobolev-Slobodezkii spaces, Zap. Nauch. Sem. LOMI, 92 (1979), 192-202. V.G. Maz’ya, An imbedding theorem and multipliers in pairs of Sobolev spaces, Trudy Tbilis. Mat. Inst., 66 (1980), 59-69. V.G. Maz’ya, The integral equations of potential theory in domains with piecewise smooth boundary, Usp. Mat. Nauk, 36;4 (1981), 229-230. V.G. Maz’ya, Boundary integral equations of elasticity in domains with piecewise smooth boundaries, Equadiff 6, Proc. Int. Conf., Brno/Czech., Lect. Notes Math., 1192, (1985), 235-242. V.G. Maz’ya, Sobolev Spaces, Springer, 1985. V.G. Maz’ya, Potential theory for the Lam´ e equations in domains with piecewise smooth boundary, In: Proc. All-Union Symp., Tbilisi, April 21-23 (1982), Metsniereba: Tbilisi, 1986, 123-129. (Russian) V.G. Maz’ya, Boundary integral equations, Encyclopaedia of Mathematical Sciences, 27, Springer, 1991, 127-233. V.G. Maz’ya, Conductor and capacitary inequalities for functions on topological spaces and their applications, J. Funct. Anal., 224 (2005), 408-430. V.G. Maz’ya and V. Havin, Nonlinear analogue of Newton potential and metric properties of (p, l)-capacity, Dokl. Akad. Nauk SSSR, 194:4 (1970), 770-773. V.G. Maz’ya and V. Havin, Nonlinear potential theory, Usp. Mat. Nauk, 27:6 (1972), 67-138. V.G. Maz’ya and Y. Netrusov, Some counterexamples for the theory of Sobolev spaces on bad domains, Potential Analysis, 4 (1995), 47-65. V.G. Maz’ya and S.P. Preobrazhenski, Estimates for capacities and traces of potentials, Internat. J. Math. Math. Sci., 7:1 (1984), 41-63. V.G. Maz’ya and T.O. Shaposhnikova, Multipliers in function spaces with fractional derivatives, Dokl. Akad. Nauk SSSR, 244:5 (1979), 1065-1067. V.G. Maz’ya and T.O. Shaposhnikova, Multipliers in Sobolev spaces, Vestnik Leningrad. Univ. Mat. Mekh. Astr., no. 2 (1979), 33-40. V.G. Maz’ya and T.O. Shaposhnikova, On traces and extensions of multipliers in the space Wpl , Usp. Mat. Nauk, 34:2 (1979), 205-206.

598

References

[MSh4]

[MSh5]

[MSh6] [MSh7]

[MSh8]

[MSh9] [MSh10]

[MSh11] [MSh12]

[MSh13] [MSh14]

[MSh15] [MSh16]

[MSh17] [MSh18]

[MSh19]

[MSh20]

V.G. Maz’ya and T.O. Shaposhnikova, Multipliers in spaces of differentiable functions, Trudy Sem. S.L. Soboleva, Novosibirsk, no. 1 (1979), 37-90. V.G. Maz’ya and T.O. Shaposhnikova, On conditions for the boundary in the Lp -theory of elliptic boundary value problems, Dokl. Akad. Nauk SSSR, 251:5 (1980), 1055-1059. V.G. Maz’ya and T.O. Shaposhnikova, Multipliers of Sobolev spaces in a domain, Math. Nachr., 99 (1980), 165-183. V.G. Maz’ya and T.O. Shaposhnikova, A coercive estimate for solutions of elliptic equations in spaces of multipliers, Vestnik Leningrad. Univ. ser. Mat. Mekh. Astr., no. 1 (1980), 41-51. V.G. Maz’ya and T.O. Shaposhnikova, Theory of multipliers in spaces of differentiable functions and their applications, Theory of cubature formulas and numerical mathematics (Proc. Conf. , Novosibirsk, 1978), Nauka, Novosibirsk, 1980, 225-233. V.G. Maz’ya and T.O. Shaposhnikova, Multipliers in spaces of Bessel potentials, Math. Nachr., 99 (1980), 363-379. V.G. Maz’ya and T.O. Shaposhnikova, On the regularity of the boundary in Lp -theory of elliptic boundary value problems, Part I: Trudy Sem. S.L. Soboleva, Novosibirsk, no. 2 (1980), 39-56; Part II: Trudy Sem. S.L. Soboleva, Novosibirsk, no. 1 (1981), 57-102. V.G. Maz’ya and T.O. Shaposhnikova, Multipliers in pairs of spaces of differentiable functions, Trudy Moskov. Mat. Obsh., 43 (1981), 37-80. ˚pm and V.G. Maz’ya and T.O. Shaposhnikova, Multipliers on the space W their applications, Vestnik Leningrad. Univ., ser. Mat. Mekh. Astr. no. 1 (1981), 42-47. V.G. Maz’ya and T.O. Shaposhnikova, Sufficient conditions for belonging to classes of multipliers, Math. Nachr., 100 (1981), 151-162. V.G. Maz’ya and T.O. Shaposhnikova, Change of variables as an operator on a pair of Sobolev spaces, Vestnik Leningrad. Univ., ser. Mat. Mekh. Astr., no. 1 (1982), 43-48. V.G. Maz’ya and T.O. Shaposhnikova, Theory of multipliers in spaces of differentiable functions, Uspekhi Mat. Nauk, 38:3 (1983), 23-86. V.G. Maz’ya and T.O. Shaposhnikova, Theory of Multipliers in Spaces of Differentiable Functions, Monographs and Studies in Mathematics, 23, Pitman, Boston–London, 1985. V.G. Maz’ya and T.O. Shaposhnikova, On pointwise interpolation inequalities for derivatives, Math. Bohemica, 124:2-3 (1999), 131-148. V.G. Maz’ya and T.O. Shaposhnikova, Maximal algebra of multipliers between fractional Sobolev spaces, Proceedings of Analysis and Geometry, S.K. Vodop’yanov (Ed.), Sobolev Institute Press, Novosibirsk, 2000, pp. 387-400. V.G. Maz’ya and T.O. Shaposhnikova, Pointwise interpolation inequalities for Riesz and Bessel potentials, Analytical and Computational Methods in Scattering and Applied mathematics, Chapman and Hall, London, 2000, pp. 217-229. V.G. Maz’ya and T.O. Shaposhnikova, Maximal Banach algebra of multipliers between Bessel potential spaces, Problems and Methods in Mathematical Physics, The Siegfried Pr¨ ossdorf Memorial Volume, J.

References

599

Elschner, I. Gohberg, B. Silbermann (Eds.), Operator Theory: Advances and Application, Vol. 121, Birkh¨ auser, 2001, pp. 352-365. [MSh21] V.G. Maz’ya and T.O. Shaposhnikova, Characterization of multipliers in pairs of Besov spaces, Operator Theory. Advances and Applications, Vol. 147 (2004), 365-386. [MSh22] V.G. Maz’ya and T.O. Shaposhnikova, Traces of multipliers in pairs of weighted Sobolev spaces, J. Function Spaces Appl., 3 (2005), 91-115. [MSh23] V.G. Maz’ya and T.O. Shaposhnikova, Higher regularity in the classical layer potential theory for Lipschitz domains, Indiana Univ. Math. J., 54:1 (2005), 99-142. [MV1] V.G. Maz’ya and I.E. Verbitsky, Capacitary estimates for fractional integrals, with applications to partial differential equations and Sobolev multipliers, Arkiv f¨ or Matem., 33 (1995), 81-115. [MV2] V.G. Maz’ya and I.E. Verbitsky, The Schr¨ odinger operator on the energy space: boundedness and compactness criteria, Acta Math., 188 (2002), 263-302. [MV3] V.G. Maz’ya and I.E. Verbitsky, The form boundedness criterion for the relativistic Schr¨ odinger operator, Ann. Inst. Fourier (Grenoble), 54 (2004), 317-339. [MV4] V.G. Maz’ya and I.E. Verbitsky, Form boundedness of the general second order differential operator, Comm. Pure Appl. Math., 59:9 (2006), 1286-1329. [Me] N.G. Meyers, A theory of capacities for potentials of functions in Lebesgue classes, Math. Scand., 26 (1970), 255-292. [MiP] S.G. Mikhlin and S. Pr¨ ossdorf, Singul¨ are Integraloperatoren, Berlin, Akademie-Verlag, 1980. [Mir] C. Miranda, Partial Differential Equations of Elliptic Type, Springer, 1970. [MT1] M. Mitrea and M. Taylor, Boundary layer methods for Lipschitz domains in Riemannian manifolds, J. Funct. Anal., 163 (1999), 181-251. [MT2] M. Mitrea and M. Taylor, Potential theory on Lipschitz domains in Rieolder space results, Comm. Anal. mannian manifolds: Lp , Hardy, and H¨ Geom., 9 (2001), 369-421. [MT3] M. Mitrea and M. Taylor, Potential theory on Lipschitz domains in Riemannian manifolds: Sobolev-Besov space results and the Poisson problem, J. Funct. Anal., 176 (2000), 1-79. [MT4] M. Mitrea and M. Taylor, Potential theory on Lipschitz domains in Riemannian manifolds: H¨ older continuous metric tensors, Comm. PDE, 25 (2000), 1487-1536. [MT5] M. Mitrea and M. Taylor, Potential theory on Lipschitz domains in Riemannian manifolds: the case of Dini metric tensors, TAMS, 355:5 (2002), 1961-1985. [Mi] A. Miyachi, Multiplication and factorization of functions in Sobolev spaces and in Cpα spaces on general domains, Math. Nachr., 176 (1995), 209-241. [Mo] A.P. Morse, The behavior of a function on its critical set, Ann. Math., 40 (1939), 62-70. [Na1] E. Nakai, Pointwise multipliers for functions of weighted bounded mean oscillation, Studia Math., 105 (1993), 105-119. [Na2] E. Nakai, Pointwise multipliers on weighted BMO spaces, Studia Math., 125:1 (1997), 35-56.

600 [NY1] [NY2]

[Ne] [Net]

[Nik] [Nir] [Pa] [Pe1] [Pe2] [Poh1] [Poh2] [Pol1] [Pol2] [RS1] [Re] [Ru] [RS]

[RY] [SW]

[Sch] [Se] [Sha]

References E. Nakai and K. Yabuta, Pointwise multipliers for functions of bounded mean oscillation, J. Math. Soc. Japan, 37 (1985), 207-218. E. Nakai and K. Yabuta, Pointwise multipliers for functions of weighted bounded mean oscillation on spaces of homogeneous type, Math. Japon., 46:1 (1997), 15-28. ´ J. Neˇcas, Les M´ethodes Directes en Th´ eorie des Equations Elliptiques, Academia, Prague, 1967. Yu. Netrusov, Theorems on traces and multipliers for functions in Lizorkin-Triebel spaces, Zap. Nauchn. Sem. St.-Petersburg. Otdel. Mat. Inst. Steklov. (POMI), 200:24 (1992), 132-138. English translation: J. Math. Sci. 77:3 (1995), 3221-3224. O. Nykodim, Sur une classe de fonctions consid´er´ees dans le probl`eme de Dirichlet, Fundam. Mat., 21 (1933), 129-150. L. Nirenberg, On elliptic partial differential equations: Lecture 2, Ann. Sc. Norm. Sup. Pisa, Ser. 3, 13 (1959), 115-162. R. Palais, Seminar on the Atiyah-Singer Index Theorem, Princeton University Press, Princeton, 1965. J. Peetre, On the differentiability of the solutions of quasilinear partial differential equations, Trans. Amer. Math. Soc., 104:3 (1962), 476-482. J. Peetre, New Thoughts on Besov Spaces, Duke Univ. Math. Ser., Durham, 1976. S.I. Pohozhaev, On eigenfunctions of the equation ∆u + λf (u) = 0, Dokl. Akad. Nauk SSSR, 165:1 (1965), 36-39. S.I. Pohozhaev, On higher order quasi-linear elliptic equations, Diff. Uravneniya, 17:1 (1981), 115-128. J.C. Polking, A Leibniz formula for some differential operators of fractional order, Indiana Univ. Math. J., 27:11 (1972), 1019-1029. J.C. Polking, Approximation in Lp by solutions of elliptic differential equations, Amer. Math. J., 94 (1972), 1231-1244. M. Reed and B. Simon, Methods of Modern Mathematical Physics. I: Functional Analysis, Academic Press, New York–London, 1980. Yu.G. Reshetnyak, Spatial mappings with bounded distortion, Sib. Mat. ˇ 8:3 (1967), 629-658. Z., T. Runst, Mapping properties of non-linear operators in spaces of TriebelLizorkin and Besov type, Anal. Math., 12 (1986), 313-346. T. Runst and W. Sickel, Sobolev Spaces of Fractional Order, Nemytskij Operators, and Nonlinear Partial Differential Equations, Walter de Gruyter, Berlin-New York, 1996. T. Runst and A. Youssfi, The Jacobian-determinant equation on Besov and Triebel-Lizorkin spaces, Nonlinear World, 4 (1997), 267-282. E.T. Sawyer and R.L. Wheeden, Weighted norm inequalities for fractional integrals on Euclidean and homogeneous spaces, Amer. J. Math., 114 (1992), 813-874. M. Schechter, Hamiltonians for singular potentials, Indiana Univ. Math. J., 22 (1972), 483-503. R.T. Seeley, Complex powers of an elliptic operator, Proc. Symp. AMS, Jan. 1967, Boston, 1967, 288-307. E. Shamir, Une propri´et´e des espaces H s,p , C.R. Acad. Sci. Paris, Ser. A-B, 255 (1962), A448-A449.

References [Sh1] [Sh2] [Sh3]

[Sh4]

[Sh5]

[Sh6] [Sh7]

[Sh8]

[Sh9]

[Sh10]

[Sh11] [Sh12]

[Sh13]

[Sh14] [Sh] [Sic1]

[Sic2] [Sic3]

601

T. Shaposhnikova, Equivalent norms in spaces with fractional or funcˇ 21 (1980), 184-196. tional smoothness, Sibir. Mat. Z., T. Shaposhnikova, On the spectrum of multipliers in Bessel potential ˇ spaces. Casopis Pˇest. Mat., 110:2 (1985), 197-206. T. Shaposhnikova, Bounded solutions of linear elliptic equations as multipliers in spaces of differentiable functions, Zapiski Nauchn. Semin. LOMI, 149 (1986), 165-176. T. Shaposhnikova, An implicit mapping theorem for multipliers in spaces of Bessel potentials, Izv. Akad. Nauk Azerba˘ıdzhan. SSR Ser. Fiz.-Tekhn. Mat. Nauk, 8:1 (1987), 14-18. T. Shaposhnikova, The superposition operator in classes of multipliers of S. L. Sobolev spaces, Seminar Analysis (Berlin, 1986/87), 181-190, Akad. Wiss. DDR, Berlin, 1987. T. Shaposhnikova, Solvability of quasilinear elliptic equations in spaces of multipliers, Izv. Vissh. Uchebn. Zaved. Math., no. 8 (1987), 74-81. T. Shaposhnikova, Applications of multipliers in S. L. Sobolev spaces to Lp -coercivity of the Neumann problem, Dokl. Akad. Nauk SSSR 305:4 (1989), 786-789; translation in Soviet Math. Dokl. 39:2 (1989), 344-347. T. Shaposhnikova, Multipliers in the space of Bessel potentials as traces of multipliers in weighted classes, Trudy Tbiliss. Mat. Inst. Razmadze Akad. Nauk Gruzin. SSR, 88 (1989), 59-63. T. Shaposhnikova, Traces of multipliers in the space of Bessel potentials, Mat. Zametki, 46:3 (1989), 100-109. English translation: Math. Notes, 46:3-4 (1990), 743-749. T. Shaposhnikova, Applications of multipliers to the problem of coercivity in Wpl of the Neumann problem., Translated in J. Soviet Math., 64:6 (1993), 1381-1388. Probl. Mat. Anal., 11, Nonlinear equations and variational inequalities. Linear operators and spectral theory (Russian), 237-248, Leningrad. Univ., Leningrad, 1990. T. Shaposhnikova, On continuity of singular integral operators in Sobolev spaces, Math. Scand., 76 (1995), 85-97. T. Shaposhnikova, Sobolev multipliers in the theory of integral convolution operators, Mathematical aspects of boundary element methods (Palaiseau, 1998), 285-295, Chapman and Hall/CRC Res. Notes Math., 414, Chapman and Hall/CRC, Boca Raton, FL, 2000. T. Shaposhnikova, Sobolev multipliers in the Lp theory of boundary integral equations of elasticity on non-smooth surfaces, Problemi Attuali dell’ Analisi e della Fisica Matematica, Gaetano Fichera memorial volume, Aracne, Rome, 2000, 161-166. T. Shaposhnikova, Description of pointwise multipliers in pairs of Besov spaces B1k (Rn ), Z. Anal. Anwend., 28:1 (2009). M.A. Shubin, Pseudodifferential Operators and Spectral Theory, Second edition, Springer, 2001. W. Sickel, On pointwise multipliers in Besov-Triebel-Lizorkin spaces, Seminar Analysis of the Karl-Weierstrass-Institute 1985/1986, Teubner-Texte Math. Vol. 96, Teubner, Leipzig, 1987. W. Sickel, Pointwise multiplication in Triebel-Lizorkin spaces, Forum Math., 5 (1993), 73-91. s W. Sickel, On pointwise multipliers for Fp,q (Rn ) in case σp,q < s < n/p, Ann. Mat. Pura Appl. 76 (1999), 209-250.

602 [SS] [ST] [SY]

[Sj] [Sob]

[Ste1]

[Ste2] [St1] [St2] [St3]

[Str] [Tr1] [Tr2] [Tr3] [Tr4] [Tru] [Yu]

[Usp]

[Va] [Ver1] [Ver2]

References W. Sickel and I. Smirnow, Localization properties of Besov spaces and its associated multiplier spaces, Jenaer Schriften Math/Inf 21/99, Jena, 1999. W. Sickel and H. Triebel, H¨ older inequalities and sharp embeddings in s s and Fp,q type, J. Anal. Appl., 14:1 (1995), 105-140. function spaces of Bp,q W. Sickel and A. Youssfi, The characterization of the regularity of the Jacobian determinant in the framework of potential spaces, J. London Math. Soc., 59:1 (1999), 287-310. T. Sj¨ odin, Capacities of compact sets in linear subspaces of Rn , Pacif. J. of Math., 78:1 (1978), 261-266. S.L. Sobolev, Some Applications of Functional Analysis to Mathematical Physics, Translations of Mathematical Monographs, 90. AMS, Providence, RI, 1991. D.A. Stegenga, Bounded Toeplitz operators on H 1 and applications of duality between H 1 and the functions of bounded mean oscillations, Amer. J. Math., 98 (1976), 573-589. D.A. Stegenga, Multipliers on the Dirichlet space, Illinois J. of Math., 24 (1980), 113-139. E.M. Stein, The characterization of functions arising as potentials, Bull. AMS, 67 (1961), 102-104. E.M. Stein, Singular Integrals and Differentiability properties of Functions, Princeton University Press, Princeton, 1970. E.M. Stein, Harmonic Analysis: Real-Variable Methods, Orthogonality, and Oscillatory Integrals, Princeton University Press, Princeton , New Jersey, 1983. R.S. Strichartz, Multipliers on fractional Sobolev spaces, J. Math. and Mech., 16:9 (1967), 1031-1060. s s and Fp,q . QuasiH. Triebel, Multiplication properties of the spaces Bp,q Banach algebras of functions, Ann. Mat. Pura Appl., 113:4 (1997), 33-42. H. Triebel, Multiplication properties of Besov spaces, Ann. Mat. Pura Appl., 114:4 (1997), 87-102. H. Triebel, Interpolation Theory. Function Spaces. Differential Operators, Berlin, VEB Deutscher Verlag der Wissenschaften, 1978. H. Triebel, Theory of Function Spaces. II, Monographs in Mathematics, 84, Birkh¨ auser, 1992. N.S. Trudinger, On imbeddings into Orlicz spaces and some applications, J. Math. Mech., 17 (1967), 473-483. V.I.Yudovich, On certain estimates connected with integral operators and solutions of elliptic equations, Dokl. Akad. Nauk SSSR, 138:4 (1961), 805808. S.V. Uspenskii, Imbedding theorems for classes with weights, Tr. Mat. Inst. Steklova, 60 (1961), 282-303 (Russian), English translation: AMS Transl., 87 (1970), 121-145. V. Valent, A property of multiplication in Sobolev spaces. Some applications, Rend. Sem. Mat. Univ. Padova, 74 (1985), 63-73. I.E. Verbitsky, Imbedding and multiplier theorems for discrete Littlewood– Paley spaces, Pacific J. Math., 176 (1996), 529-556. I.E. Verbitsky, Superlinear equations, potential theory, and weighted norm inequalities, Nonlinear Analysis, Function Spaces and Applications, Vol. 6 (Prague, 1998), Acad. Sci. Czech Repub., Prague, 1999, 223-269.

References [Ver3]

[Verc]

[VG]

[VGR]

[VP] [Wa] [Wl] [Wu] [Ya] [Yam]

[Ye] [Yo] [Zo] [Zy]

603

I.E. Verbitsky, Nonlinear potentials and trace inequalities, The Maz’ya Anniversary Collection, Eds. J. Rossmann, P. Tak´ aˇc, G. Wildenhain, Operator Theory: Advances and Applications, Vol. 110, Birkh¨ auser, 1999, 323-343. G. Verchota, Layer potentials and regularity for the Dirichlet problem for Laplace’s equation in Lipschitz domains, J. Funct. Anal., 59:3 (1984), 572611. S.K. Vodop’yanov and V.M. Gol’dshtein, Quasi-conformal mappings and spaces of functions with the first generalized derivatives, Sib. Mat. Z., 16:3 (1976), 515-531. S.K. Vodop’yanov, V.M. Gol’dshtein, and Yu.G. Reshetnyak, The geometric properties of functions with generalized first derivatives, Uspehi Matem. Nauk, 34:1 (1979), 17-65. L.R. Volevich and B.P. Paneyah, Some spaces of generalized functions and embedding theorems, Usp. Mat. Nauk, 20 (1965), 3-74. S.E. Warschawski, On conformal mapping of infinite strips, Trans. AMS, 51 (1942), 280-335. J. Wloka, Partial Differential Equations, Cambridge Univ. Press, 1987. Z. Wu, Strong type estimate and Carleson measures for Lipschitz spaces, Proc. Amer. Math. Soc., 127 (1991), 3243-3249. K. Yabuta, Pointwise multipliers of weighted BMO spaces, Proc. AMS, 117 (1993), 737-744. M. Yamazaki, A quasi-homogeneous version of paradifferential operators I: Boundedness on spaces of Besov type, J. Fac. Sci. Univ. Tokyo Sect. IA Math. 33 (1986), 131-174. A quasi-homogeneous version of paradifferential operators II: A symbolic calculus. Ibidem 33 (1986), 311-345. D. Ye, Prescribing the Jacobian determinant in Sobolev spaces, Ann. Inst. Henri Poincar´e, 11 (1994), 275-296. A. Youssfi, Commutators on Besov spaces and factorization of the paraproduct, Bull. Sci. Math., 119 (1995), 157-186. J.L. Zolesio, Multiplication dans les espaces de Besov, Proc. Roy. Soc. Edinburgh, 78:1-2 (1977), 113-117. A. Zygmund, Trigonometric Series, Cambridge, 1959.

List of Symbols

Classes of boundaries: C 0,1 , 336 3/2 M2 ∩ C 1 , 507 l−1/p , 480 Mp l−1/p Mp (δ), 513

Operators: (−∆)−1/2 , 392 (−∆)r/2 , 14 (1 − ∆)s/2 , 14 D∗ , 532 Dp,l , 133 (r) Dp,s , 217 Jl , 16 P (x, Dx ), 374 P0 (·, θ), 384 Sm , 70 l , 71 Sq,θ Tml , 432 ∆−1 , 392 ∂ ∂ν D, 550 C, 326 √ H = −∆ + Q, 427 Dp,l , 133 ∇, 8 ∇l , 7 π, 50 ϕ(K), 574 |D|l , 429 Dl,µ , 581 L, 445 M, 16 T , 313 Div, 422

curl, 423 div, 8 tr, 364

Set functions: Cp,m (e), 16 Cp,m (E), 117 Cp,m (E), 117 cp,m (e), 16 d(e), 18 p-capΩ (g, G), 348 Cp,s,α (e), 285 cap(e, Ω), 450 mesn , 11

Spaces: (C0∞ ) , 393 (Wpk
) , 54 BM O, 210 BM O−1 , 423 BM Oϕ , 211, 212 BV , 64 µ , 106 Bq,∞,unif µ , 99 Bq,∞ l , 166 Bq,θ s , 166 Bq,θ,unif Bpl , 133 C0∞ ([0, ∞)), 24 C 0,1 , 325 C l−1,1 (σ), 59 C0∞ (Ω), 539 l , 208 Fp,θ Hpm , 70 Hp−l
, 115, 118 Hpl (∂B), 125 605

606

List of Symbols

L2 (|Γ |2 ), 392 L1p (Ω), 325 Lp,unif , 55 Lp,loc , 44 M (˚ w21 (Ω) → w2−1 (Ω)), 416 M (BM O), 211 M (B1m → B1l ), 179 M (Bpm → Bpl ), 134 l l → Bp,∞ ), 209 M (Bp,1 m l M (Hp → Hp ), 69 M (Hpm (∂B1 ) → Hpl (∂B1 )), 125 M (H m,µ → H l,µ ), 581 M (Hpm−l → Lp ), 76 1 (Ω)), 452 M (S(Ω) → W2,1 1 ˚ 1 (Ω)), 446 M (W2 (Ω) → W 2 1 → L2 ), 448 M (W2,β 1 1 (Ω) → W2,1 (Ω)), M (W2,w(ρ) 448, 455 M (Wpm → Wpl ), 33 M (Wpm → Wql ), 33 M (Wpm (G) → Wpl (G)), 329 M (Wpm (Ω) → Wpl (Ω)), 339 M (Wpm → Wp−k ), 33 M (W1m → W1l ), 35 1/2 −1/2 M (W2 → W2 ), 427 M (Wpm (Rn+ ) → Wpl (Rn+ )), 50 M (Wpm → Lp ), 60 M (Wpt,β → Wps,α ), 298 ˚ m (Ω) → W l (Ω)), 325, 369 M (W p p l M (hm p → hp ), 69 l M (hm p → hp ), 122 m M (wp → wpl ), 33 M (w21 → w2−1 ), 391 M (wpm → wpl ), 60–62 M BV , 66, 68 M L1p (Ω), 349 M L1p (Ω), 346 M Wpl , 59 M bv, 66 S(Ω), 452 Sloc , 34

Sunif , 34 1 (Ω), 448 W2,β 1 W2,w(ρ) (Ω), 454 Wpk , 7 k , 305 Wp,β W1l (Br ), 205 Wpl (Ω), 325 Wpm (G), 327 Wp−k , 54 Wp−k (G), 496 Wpk,α (Rn \Ω), 542 k,α Wp,loc (Rn \Ω), 535 l Wp,loc , 42 Wps,α (Rn+1 + ), 285 Wpl (Br ), 243 l−1/p

Wp (∂G), 356 l , 234 Wp,unif −k , 55 γ ∈ Wp,unif ∗ Mα , 97 Mα , 97 8 h−s M (Wpr → Wp j ), 467 8j r−h−tk → Lp ), 467 k M (Wp −1 1 ˚ M (˚ w2 (Ω) → w2 (Ω)), 419 ˚ (w1 → L2 ), 407 M 2 ˚p1,α (Ω), 543 W ˚ 1 (Ω), 539 W s ˚ W l , 281 M p ˚ (Wpm (Ω) → Wpl (Ω)), M 342 ˚ Wpl (Ω), 343 M ˚ (w1 → w−1 ), 407 M 2 2 bv, 64 hm p , 70 wpk , 7 w11 , 64 w2−1 (Ω), 417 Hµ , 580 Lp,λ , 96 S, 387 S  , 411

Author and Subject Index (p, l)-refined function, 118, 120, 121 (p, k, α)-diffeomorphism, 537, 538, 544 (p, l)-diffeomorphism, 350–353, 355–357, 364, 366, 367, 382, 481–483, 491, 492, 494, 501, 537 (p, l)-manifold, 350, 356, 357 Tpm,l -mapping, 357, 358, 361–364 q-variation, 110 Adams, 16, 22, 112 Ahlfors, 119 Ali Mehmeti, 239 Amann, 239

368, 370, 393, 398, 400, 412, 413, 419, 450, 452, 480, 511, 517, 560, 584, 589 capacity of a ball, 48, 144, 261, 267, 441 Carath´eodory conditions, 474, 477 Carleson, 75 compact multipliers, 267 compact multipliers in a bounded Lipschitz domain, 342 continuous spectrum, 116, 122, 575 convolution operator, 15, 287, 573, 586 Cordes condition, 463 covering lemma, 11

Benchekroun, 239 Benkirane, 239 Bennet, 239 Besicovitch, 10 Besov space, 1, 99, 133–135, 144, 209, 481, 511 Bessel potential, 16, 21, 80, 117, 138, 147, 319, 413, 414, 430, 439, 441, 465 Bessel potential space, 1, 69, 70, 75, 173, 214, 227, 319, 580 Beurling, 119, 174 Bliev, 239 Bloom, 212 Bourdaud, 209

Dacorogna, 239, 352 De Giorgi, 65 de Rham, 356 Devinatz, 2 Dirichlet Problem, 445 Dirichlet problem, 201, 297, 312, 445, 447, 449, 461–463, 480, 481, 489, 502, 504–507, 512, 536, 537, 539, 540, 544, 562, 564, 566, 567 Dirichlet problem in the halfspace, 311 domain of the Lipschitz class, 3, 332, 337, 509, 511, 512, 530, 531, 536–538 Drihem, 239

Cacciopoli, 65 Calderon interpolation theorem, 75 capacity, 16–19, 21, 22, 25, 48, 69, 71–74, 95, 117–119, 143, 144, 165, 250, 255, 262, 263, 266, 268, 285, 327, 337, 348, 367,

Edmunds, 267 elastic double layer potential, 571 elastic single layer potential, 570 elliptic semilinear systems, 474, 477 equilibrium potential, 397 essential norm of a multiplier, 241

607

608

Author and Subject Index

essential norm of multipliers in a bounded Lipschitz domain, 341 Fatou theorem, 476 Federer, 11, 66 Fefferman, 28, 75, 425, 442 Filonov, 530 Frank, 443 Franke, 114, 239 Frazier, 114, 209 Fredholm operator, 480 Fubini, 11 functions with bounded variation, 33, 63 G˚ arding inequality, 502 Gagliardo-Nirenberg inequality, 213, 235, 364 Gelfand, 116 Gilbert, 239 Giraud theorem, 449, 510 Gol’dshtein, 358 Gulisashvili, 114, 209 Gustin, 11 Hanouzet, 239 Hardy’s inequality, 73, 129, 299, 301, 302, 319, 393, 395, 402, 407, 408, 417, 418, 443, 449, 516, 520, 522 Hardy–Littlewood maximal operator, 29, 138, 402, 431, 435 Hausdorff measure, 66, 75 Hausdorff-Young theorem, 271 Havin, 16, 75 Hedberg, 16 Hedberg’s inequality, 22, 138 Hertz, 239 Hirschman, 2, 98, 110, 114, 174 Il’in, 106 inner capacity, 117 interpolation inequality, 37, 54, 75, 82, 87, 111, 185, 202, 277, 281 isoperimetric inequality, 9, 65

Janson, 211 Jawerth, 114, 209 Johnsen, 239 Johnson, 239 Kalyabin, 239 Kelvin-Somigliana tensor, 570 Kerman, 27 Koch, 209 Kurtz, 432 Lam´e system, 569 Lebesgue measure, 28 left regularizer, 387 Legendre-Hadamard strong ellipticity condition, 568 Lemari´e-Rieusset, 478 Lions, J.-L., 114 Lipschitz class, 353, 356, 455, 480, 538 Littlewood-Paley decomposition, 206 Lorentz space, 443 Magenes, 114 Mal´ y, 358 Marcinkiewicz space, 97, 98 Marcinkiewicz-Sobolev space, 98 Marschall, 239 Maxwell operator, 530 Meyers, 16, 75, 112 Mikhlin, 14, 432, 575 Mikhlin-Calderon-Zygmund operator, 393, 396 Miyachi, 239 mollification, 67, 76, 77, 82, 147, 181, 269, 279, 283, 326, 329, 337, 340, 389, 468–470, 473– 475 Morrey space, 96, 442 Morrey-Sobolev space, 98 Moser, 239, 352 Moussai, 239 Muckenhoupt class, 396, 405, 431 Muckenhoupt constant, 405

Author and Subject Index

Nakai, 212 Navier-Stokes system, 478 Netrusov, 209 Neumann problem, 531, 536, 537, 564, 566, 569 Nicaise, 239 Nikodym, 346 nondivergence equation, 456, 463 nonlinear Wolff potential, 31 operator elliptic in the sense of Douglis-Nirenberg, 467, 473 outer capacity, 117 paraproduct algorithm, 209 Peetre, 209, 239 Peetre imbedding theorem, 206 Phong, 28, 75, 425, 442 Pohozhaev, 17 Poincar´e inequality, 346 pointwise interpolation inequality, 214 pointwise spectrum, 116, 121 Poisson integral, 157, 195, 303 Poisson kernel, 90, 199, 301 Poisson operator, 89, 185, 189, 192 Polking, 2, 71, 99, 173 positive homogeneous multipliers, 125 Pr¨ ossdorf, 575 quasilinear second-order equation, 2, 456 Reshetnyak, 358, 361 residual spectrum, 116, 122, 575 resolvent set, 116 Riesz potential, 15, 16, 19 Riesz potential space, 69, 70 Riesz transforms, 405, 406, 413 right regularizer, 387 Runst, 114, 239 Sawyer, 27 Seiringer, 443

609

sesquilinear form, 392, 393, 411, 422, 423, 427 Shamir, 114 Shargorodsky, 267 Sickel, 114, 209, 239, 352 singular integral operator, 2, 3, 14, 16, 21, 540, 573, 575–577, 579 Sj¨ odin, 16 Smirnov, 209 Sobolev imbedding theorem, 56, 70, 99, 111, 112, 331 Sobolev integral representation, 36, 45, 214 special Lipschitz domain, 325, 326, 329–331, 336, 337, 356, 481, 482, 484, 490, 496, 511, 515, 542, 559 spectrum, 2, 115, 116, 118, 575 Stegenga, 211 Stein, 87 Strichartz, 2, 79, 98, 113, 114, 239 trace inequality, 7, 25, 28, 30, 69, 143, 179, 187, 441 transmission problem, 534, 536, 548, 549, 552, 553, 556, 557 Triebel, 114, 208, 239 Triebel-Lizorkin space, 208, 209, 267 Trudinger, 17 Valent, 239 Verbitsky, 28, 45, 89, 113 Vodop’yanov, 358 Wheeden, 432 Whitney covering, 417 Yabuta, 212 Ye, 239, 352 Young’s inequality, 17 Youssfi, 239, 352 Yudoviˇc, 17 Zolesio, 239

Grundlehren der mathematischen Wissenschaften A Series of Comprehensive Studies in Mathematics A Selection 248. Suzuki: Group Theory II 249. Chung: Lectures from Markov Processes to Brownian Motion 250. Arnold: Geometrical Methods in the Theory of Ordinary Differential Equations 251. Chow/Hale: Methods of Bifurcation Theory 252. Aubin: Nonlinear Analysis on Manifolds. Monge-Ampère Equations 253. Dwork: Lectures on ρ -adic Differential Equations 254. Freitag: Siegelsche Modulfunktionen 255. Lang: Complex Multiplication 256. Hörmander: The Analysis of Linear Partial Differential Operators I 257. Hörmander: The Analysis of Linear Partial Differential Operators II 258. Smoller: Shock Waves and Reaction-Diffusion Equations 259. Duren: Univalent Functions 260. Freidlin/Wentzell: Random Perturbations of Dynamical Systems 261. Bosch/Güntzer/Remmert: Non Archimedian Analysis – A System Approach to Rigid Analytic Geometry 262. Doob: Classical Potential Theory and Its Probabilistic Counterpart 263. Krasnosel’skiˇı/Zabreˇıko: Geometrical Methods of Nonlinear Analysis 264. Aubin/Cellina: Differential Inclusions 265. Grauert/Remmert: Coherent Analytic Sheaves 266. de Rham: Differentiable Manifolds 267. Arbarello/Cornalba/Griffiths/Harris: Geometry of Algebraic Curves, Vol. I 268. Arbarello/Cornalba/Griffiths/Harris: Geometry of Algebraic Curves, Vol. II 269. Schapira: Microdifferential Systems in the Complex Domain 270. Scharlau: Quadratic and Hermitian Forms 271. Ellis: Entropy, Large Deviations, and Statistical Mechanics 272. Elliott: Arithmetic Functions and Integer Products 273. Nikol’skiˇı: Treatise on the shift Operator 274. Hörmander: The Analysis of Linear Partial Differential Operators III 275. Hörmander: The Analysis of Linear Partial Differential Operators IV 276. Liggett: Interacting Particle Systems 277. Fulton/Lang: Riemann-Roch Algebra 278. Barr/Wells: Toposes, Triples and Theories 279. Bishop/Bridges: Constructive Analysis 280. Neukirch: Class Field Theory 281. Chandrasekharan: Elliptic Functions 282. Lelong/Gruman: Entire Functions of Several Complex Variables 283. Kodaira: Complex Manifolds and Deformation of Complex Structures 284. Finn: Equilibrium Capillary Surfaces 285. Burago/Zalgaller: Geometric Inequalities 286. Andrianaov: Quadratic Forms and Hecke Operators 287. Maskit: Kleinian Groups 288. Jacod/Shiryaev: Limit Theorems for Stochastic Processes 289. Manin: Gauge Field Theory and Complex Geometry 290. Conway/Sloane: Sphere Packings, Lattices and Groups

291. Hahn/O’Meara: The Classical Groups and K-Theory 292. Kashiwara/Schapira: Sheaves on Manifolds 293. Revuz/Yor: Continuous Martingales and Brownian Motion 294. Knus: Quadratic and Hermitian Forms over Rings 295. Dierkes/Hildebrandt/Küster/Wohlrab: Minimal Surfaces I 296. Dierkes/Hildebrandt/Küster/Wohlrab: Minimal Surfaces II 297. Pastur/Figotin: Spectra of Random and Almost-Periodic Operators 298. Berline/Getzler/Vergne: Heat Kernels and Dirac Operators 299. Pommerenke: Boundary Behaviour of Conformal Maps 300. Orlik/Terao: Arrangements of Hyperplanes 301. Loday: Cyclic Homology 302. Lange/Birkenhake: Complex Abelian Varieties 303. DeVore/Lorentz: Constructive Approximation 304. Lorentz/v. Golitschek/Makovoz: Construcitve Approximation. Advanced Problems 305. Hiriart-Urruty/Lemaréchal: Convex Analysis and Minimization Algorithms I. Fundamentals 306. Hiriart-Urruty/Lemaréchal: Convex Analysis and Minimization Algorithms II. Advanced Theory and Bundle Methods 307. Schwarz: Quantum Field Theory and Topology 308. Schwarz: Topology for Physicists 309. Adem/Milgram: Cohomology of Finite Groups 310. Giaquinta/Hildebrandt: Calculus of Variations I: The Lagrangian Formalism 311. Giaquinta/Hildebrandt: Calculus of Variations II: The Hamiltonian Formalism 312. Chung/Zhao: From Brownian Motion to Schrödinger’s Equation 313. Malliavin: Stochastic Analysis 314. Adams/Hedberg: Function spaces and Potential Theory 315. Bürgisser/Clausen/Shokrollahi: Algebraic Complexity Theory 316. Saff/Totik: Logarithmic Potentials with External Fields 317. Rockafellar/Wets: Variational Analysis 318. Kobayashi: Hyperbolic Complex Spaces 319. Bridson/Haefliger: Metric Spaces of Non-Positive Curvature 320. Kipnis/Landim: Scaling Limits of Interacting Particle Systems 321. Grimmett: Percolation 322. Neukirch: Algebraic Number Theory 323. Neukirch/Schmidt/Wingberg: Cohomology of Number Fields 324. Liggett: Stochastic Interacting Systems: Contact, Voter and Exclusion Processes 325. Dafermos: Hyperbolic Conservation Laws in Continuum Physics 326. Waldschmidt: Diophantine Approximation on Linear Algebraic Groups 327. Martinet: Perfect Lattices in Euclidean Spaces 328. Van der Put/Singer: Galois Theory of Linear Differential Equations 329. Korevaar: Tauberian Theory. A Century of Developments 330. Mordukhovich: Variational Analysis and Generalized Differentiation I: Basic Theory 331. Mordukhovich: Variational Analysis and Generalized Differentiation II: Applications 332. Kashiwara/Schapira: Categories and Sheaves. An Introduction to Ind-Objects and Derived Categories 333. Grimmett: The Random-Cluster Model 334. Sernesi: Deformations of Algebraic Schemes 335. Bushnell/Henniart: The Local Langlands Conjecture for GL(2) 336. Gruber: Convex and Discrete Geometry , 337. Maz ya/Shaposhnikova: Theory of Sobolev Multipliers. With Applications to Differential and Integral Operators 338. Villani: Optimal Transport: Old and New